xref: /linux/fs/xfs/libxfs/xfs_attr_leaf.c (revision c94cd9508b1335b949fd13ebd269313c65492df0)
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 #include "xfs_health.h"
33 
34 
35 /*
36  * xfs_attr_leaf.c
37  *
38  * Routines to implement leaf blocks of attributes as Btrees of hashed names.
39  */
40 
41 /*========================================================================
42  * Function prototypes for the kernel.
43  *========================================================================*/
44 
45 /*
46  * Routines used for growing the Btree.
47  */
48 STATIC int xfs_attr3_leaf_create(struct xfs_da_args *args,
49 				 xfs_dablk_t which_block, struct xfs_buf **bpp);
50 STATIC int xfs_attr3_leaf_add_work(struct xfs_buf *leaf_buffer,
51 				   struct xfs_attr3_icleaf_hdr *ichdr,
52 				   struct xfs_da_args *args, int freemap_index);
53 STATIC void xfs_attr3_leaf_compact(struct xfs_da_args *args,
54 				   struct xfs_attr3_icleaf_hdr *ichdr,
55 				   struct xfs_buf *leaf_buffer);
56 STATIC void xfs_attr3_leaf_rebalance(xfs_da_state_t *state,
57 						   xfs_da_state_blk_t *blk1,
58 						   xfs_da_state_blk_t *blk2);
59 STATIC int xfs_attr3_leaf_figure_balance(xfs_da_state_t *state,
60 			xfs_da_state_blk_t *leaf_blk_1,
61 			struct xfs_attr3_icleaf_hdr *ichdr1,
62 			xfs_da_state_blk_t *leaf_blk_2,
63 			struct xfs_attr3_icleaf_hdr *ichdr2,
64 			int *number_entries_in_blk1,
65 			int *number_usedbytes_in_blk1);
66 
67 /*
68  * Utility routines.
69  */
70 STATIC void xfs_attr3_leaf_moveents(struct xfs_da_args *args,
71 			struct xfs_attr_leafblock *src_leaf,
72 			struct xfs_attr3_icleaf_hdr *src_ichdr, int src_start,
73 			struct xfs_attr_leafblock *dst_leaf,
74 			struct xfs_attr3_icleaf_hdr *dst_ichdr, int dst_start,
75 			int move_count);
76 STATIC int xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index);
77 
78 /*
79  * attr3 block 'firstused' conversion helpers.
80  *
81  * firstused refers to the offset of the first used byte of the nameval region
82  * of an attr leaf block. The region starts at the tail of the block and expands
83  * backwards towards the middle. As such, firstused is initialized to the block
84  * size for an empty leaf block and is reduced from there.
85  *
86  * The attr3 block size is pegged to the fsb size and the maximum fsb is 64k.
87  * The in-core firstused field is 32-bit and thus supports the maximum fsb size.
88  * The on-disk field is only 16-bit, however, and overflows at 64k. Since this
89  * only occurs at exactly 64k, we use zero as a magic on-disk value to represent
90  * the attr block size. The following helpers manage the conversion between the
91  * in-core and on-disk formats.
92  */
93 
94 static void
95 xfs_attr3_leaf_firstused_from_disk(
96 	struct xfs_da_geometry		*geo,
97 	struct xfs_attr3_icleaf_hdr	*to,
98 	struct xfs_attr_leafblock	*from)
99 {
100 	struct xfs_attr3_leaf_hdr	*hdr3;
101 
102 	if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) {
103 		hdr3 = (struct xfs_attr3_leaf_hdr *) from;
104 		to->firstused = be16_to_cpu(hdr3->firstused);
105 	} else {
106 		to->firstused = be16_to_cpu(from->hdr.firstused);
107 	}
108 
109 	/*
110 	 * Convert from the magic fsb size value to actual blocksize. This
111 	 * should only occur for empty blocks when the block size overflows
112 	 * 16-bits.
113 	 */
114 	if (to->firstused == XFS_ATTR3_LEAF_NULLOFF) {
115 		ASSERT(!to->count && !to->usedbytes);
116 		ASSERT(geo->blksize > USHRT_MAX);
117 		to->firstused = geo->blksize;
118 	}
119 }
120 
121 static void
122 xfs_attr3_leaf_firstused_to_disk(
123 	struct xfs_da_geometry		*geo,
124 	struct xfs_attr_leafblock	*to,
125 	struct xfs_attr3_icleaf_hdr	*from)
126 {
127 	struct xfs_attr3_leaf_hdr	*hdr3;
128 	uint32_t			firstused;
129 
130 	/* magic value should only be seen on disk */
131 	ASSERT(from->firstused != XFS_ATTR3_LEAF_NULLOFF);
132 
133 	/*
134 	 * Scale down the 32-bit in-core firstused value to the 16-bit on-disk
135 	 * value. This only overflows at the max supported value of 64k. Use the
136 	 * magic on-disk value to represent block size in this case.
137 	 */
138 	firstused = from->firstused;
139 	if (firstused > USHRT_MAX) {
140 		ASSERT(from->firstused == geo->blksize);
141 		firstused = XFS_ATTR3_LEAF_NULLOFF;
142 	}
143 
144 	if (from->magic == XFS_ATTR3_LEAF_MAGIC) {
145 		hdr3 = (struct xfs_attr3_leaf_hdr *) to;
146 		hdr3->firstused = cpu_to_be16(firstused);
147 	} else {
148 		to->hdr.firstused = cpu_to_be16(firstused);
149 	}
150 }
151 
152 void
153 xfs_attr3_leaf_hdr_from_disk(
154 	struct xfs_da_geometry		*geo,
155 	struct xfs_attr3_icleaf_hdr	*to,
156 	struct xfs_attr_leafblock	*from)
157 {
158 	int	i;
159 
160 	ASSERT(from->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
161 	       from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
162 
163 	if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) {
164 		struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)from;
165 
166 		to->forw = be32_to_cpu(hdr3->info.hdr.forw);
167 		to->back = be32_to_cpu(hdr3->info.hdr.back);
168 		to->magic = be16_to_cpu(hdr3->info.hdr.magic);
169 		to->count = be16_to_cpu(hdr3->count);
170 		to->usedbytes = be16_to_cpu(hdr3->usedbytes);
171 		xfs_attr3_leaf_firstused_from_disk(geo, to, from);
172 		to->holes = hdr3->holes;
173 
174 		for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
175 			to->freemap[i].base = be16_to_cpu(hdr3->freemap[i].base);
176 			to->freemap[i].size = be16_to_cpu(hdr3->freemap[i].size);
177 		}
178 		return;
179 	}
180 	to->forw = be32_to_cpu(from->hdr.info.forw);
181 	to->back = be32_to_cpu(from->hdr.info.back);
182 	to->magic = be16_to_cpu(from->hdr.info.magic);
183 	to->count = be16_to_cpu(from->hdr.count);
184 	to->usedbytes = be16_to_cpu(from->hdr.usedbytes);
185 	xfs_attr3_leaf_firstused_from_disk(geo, to, from);
186 	to->holes = from->hdr.holes;
187 
188 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
189 		to->freemap[i].base = be16_to_cpu(from->hdr.freemap[i].base);
190 		to->freemap[i].size = be16_to_cpu(from->hdr.freemap[i].size);
191 	}
192 }
193 
194 void
195 xfs_attr3_leaf_hdr_to_disk(
196 	struct xfs_da_geometry		*geo,
197 	struct xfs_attr_leafblock	*to,
198 	struct xfs_attr3_icleaf_hdr	*from)
199 {
200 	int				i;
201 
202 	ASSERT(from->magic == XFS_ATTR_LEAF_MAGIC ||
203 	       from->magic == XFS_ATTR3_LEAF_MAGIC);
204 
205 	if (from->magic == XFS_ATTR3_LEAF_MAGIC) {
206 		struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)to;
207 
208 		hdr3->info.hdr.forw = cpu_to_be32(from->forw);
209 		hdr3->info.hdr.back = cpu_to_be32(from->back);
210 		hdr3->info.hdr.magic = cpu_to_be16(from->magic);
211 		hdr3->count = cpu_to_be16(from->count);
212 		hdr3->usedbytes = cpu_to_be16(from->usedbytes);
213 		xfs_attr3_leaf_firstused_to_disk(geo, to, from);
214 		hdr3->holes = from->holes;
215 		hdr3->pad1 = 0;
216 
217 		for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
218 			hdr3->freemap[i].base = cpu_to_be16(from->freemap[i].base);
219 			hdr3->freemap[i].size = cpu_to_be16(from->freemap[i].size);
220 		}
221 		return;
222 	}
223 	to->hdr.info.forw = cpu_to_be32(from->forw);
224 	to->hdr.info.back = cpu_to_be32(from->back);
225 	to->hdr.info.magic = cpu_to_be16(from->magic);
226 	to->hdr.count = cpu_to_be16(from->count);
227 	to->hdr.usedbytes = cpu_to_be16(from->usedbytes);
228 	xfs_attr3_leaf_firstused_to_disk(geo, to, from);
229 	to->hdr.holes = from->holes;
230 	to->hdr.pad1 = 0;
231 
232 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
233 		to->hdr.freemap[i].base = cpu_to_be16(from->freemap[i].base);
234 		to->hdr.freemap[i].size = cpu_to_be16(from->freemap[i].size);
235 	}
236 }
237 
238 static xfs_failaddr_t
239 xfs_attr3_leaf_verify_entry(
240 	struct xfs_mount			*mp,
241 	char					*buf_end,
242 	struct xfs_attr_leafblock		*leaf,
243 	struct xfs_attr3_icleaf_hdr		*leafhdr,
244 	struct xfs_attr_leaf_entry		*ent,
245 	int					idx,
246 	__u32					*last_hashval)
247 {
248 	struct xfs_attr_leaf_name_local		*lentry;
249 	struct xfs_attr_leaf_name_remote	*rentry;
250 	char					*name_end;
251 	unsigned int				nameidx;
252 	unsigned int				namesize;
253 	__u32					hashval;
254 
255 	/* hash order check */
256 	hashval = be32_to_cpu(ent->hashval);
257 	if (hashval < *last_hashval)
258 		return __this_address;
259 	*last_hashval = hashval;
260 
261 	nameidx = be16_to_cpu(ent->nameidx);
262 	if (nameidx < leafhdr->firstused || nameidx >= mp->m_attr_geo->blksize)
263 		return __this_address;
264 
265 	/*
266 	 * Check the name information.  The namelen fields are u8 so we can't
267 	 * possibly exceed the maximum name length of 255 bytes.
268 	 */
269 	if (ent->flags & XFS_ATTR_LOCAL) {
270 		lentry = xfs_attr3_leaf_name_local(leaf, idx);
271 		namesize = xfs_attr_leaf_entsize_local(lentry->namelen,
272 				be16_to_cpu(lentry->valuelen));
273 		name_end = (char *)lentry + namesize;
274 		if (lentry->namelen == 0)
275 			return __this_address;
276 	} else {
277 		rentry = xfs_attr3_leaf_name_remote(leaf, idx);
278 		namesize = xfs_attr_leaf_entsize_remote(rentry->namelen);
279 		name_end = (char *)rentry + namesize;
280 		if (rentry->namelen == 0)
281 			return __this_address;
282 		if (!(ent->flags & XFS_ATTR_INCOMPLETE) &&
283 		    rentry->valueblk == 0)
284 			return __this_address;
285 	}
286 
287 	if (name_end > buf_end)
288 		return __this_address;
289 
290 	return NULL;
291 }
292 
293 /*
294  * Validate an attribute leaf block.
295  *
296  * Empty leaf blocks can occur under the following circumstances:
297  *
298  * 1. setxattr adds a new extended attribute to a file;
299  * 2. The file has zero existing attributes;
300  * 3. The attribute is too large to fit in the attribute fork;
301  * 4. The attribute is small enough to fit in a leaf block;
302  * 5. A log flush occurs after committing the transaction that creates
303  *    the (empty) leaf block; and
304  * 6. The filesystem goes down after the log flush but before the new
305  *    attribute can be committed to the leaf block.
306  *
307  * Hence we need to ensure that we don't fail the validation purely
308  * because the leaf is empty.
309  */
310 static xfs_failaddr_t
311 xfs_attr3_leaf_verify(
312 	struct xfs_buf			*bp)
313 {
314 	struct xfs_attr3_icleaf_hdr	ichdr;
315 	struct xfs_mount		*mp = bp->b_mount;
316 	struct xfs_attr_leafblock	*leaf = bp->b_addr;
317 	struct xfs_attr_leaf_entry	*entries;
318 	struct xfs_attr_leaf_entry	*ent;
319 	char				*buf_end;
320 	uint32_t			end;	/* must be 32bit - see below */
321 	__u32				last_hashval = 0;
322 	int				i;
323 	xfs_failaddr_t			fa;
324 
325 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, leaf);
326 
327 	fa = xfs_da3_blkinfo_verify(bp, bp->b_addr);
328 	if (fa)
329 		return fa;
330 
331 	/*
332 	 * firstused is the block offset of the first name info structure.
333 	 * Make sure it doesn't go off the block or crash into the header.
334 	 */
335 	if (ichdr.firstused > mp->m_attr_geo->blksize)
336 		return __this_address;
337 	if (ichdr.firstused < xfs_attr3_leaf_hdr_size(leaf))
338 		return __this_address;
339 
340 	/* Make sure the entries array doesn't crash into the name info. */
341 	entries = xfs_attr3_leaf_entryp(bp->b_addr);
342 	if ((char *)&entries[ichdr.count] >
343 	    (char *)bp->b_addr + ichdr.firstused)
344 		return __this_address;
345 
346 	/*
347 	 * NOTE: This verifier historically failed empty leaf buffers because
348 	 * we expect the fork to be in another format. Empty attr fork format
349 	 * conversions are possible during xattr set, however, and format
350 	 * conversion is not atomic with the xattr set that triggers it. We
351 	 * cannot assume leaf blocks are non-empty until that is addressed.
352 	*/
353 	buf_end = (char *)bp->b_addr + mp->m_attr_geo->blksize;
354 	for (i = 0, ent = entries; i < ichdr.count; ent++, i++) {
355 		fa = xfs_attr3_leaf_verify_entry(mp, buf_end, leaf, &ichdr,
356 				ent, i, &last_hashval);
357 		if (fa)
358 			return fa;
359 	}
360 
361 	/*
362 	 * Quickly check the freemap information.  Attribute data has to be
363 	 * aligned to 4-byte boundaries, and likewise for the free space.
364 	 *
365 	 * Note that for 64k block size filesystems, the freemap entries cannot
366 	 * overflow as they are only be16 fields. However, when checking end
367 	 * pointer of the freemap, we have to be careful to detect overflows and
368 	 * so use uint32_t for those checks.
369 	 */
370 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
371 		if (ichdr.freemap[i].base > mp->m_attr_geo->blksize)
372 			return __this_address;
373 		if (ichdr.freemap[i].base & 0x3)
374 			return __this_address;
375 		if (ichdr.freemap[i].size > mp->m_attr_geo->blksize)
376 			return __this_address;
377 		if (ichdr.freemap[i].size & 0x3)
378 			return __this_address;
379 
380 		/* be care of 16 bit overflows here */
381 		end = (uint32_t)ichdr.freemap[i].base + ichdr.freemap[i].size;
382 		if (end < ichdr.freemap[i].base)
383 			return __this_address;
384 		if (end > mp->m_attr_geo->blksize)
385 			return __this_address;
386 	}
387 
388 	return NULL;
389 }
390 
391 xfs_failaddr_t
392 xfs_attr3_leaf_header_check(
393 	struct xfs_buf		*bp,
394 	xfs_ino_t		owner)
395 {
396 	struct xfs_mount	*mp = bp->b_mount;
397 
398 	if (xfs_has_crc(mp)) {
399 		struct xfs_attr3_leafblock *hdr3 = bp->b_addr;
400 
401 		if (hdr3->hdr.info.hdr.magic !=
402 				cpu_to_be16(XFS_ATTR3_LEAF_MAGIC))
403 			return __this_address;
404 
405 		if (be64_to_cpu(hdr3->hdr.info.owner) != owner)
406 			return __this_address;
407 	}
408 
409 	return NULL;
410 }
411 
412 static void
413 xfs_attr3_leaf_write_verify(
414 	struct xfs_buf	*bp)
415 {
416 	struct xfs_mount	*mp = bp->b_mount;
417 	struct xfs_buf_log_item	*bip = bp->b_log_item;
418 	struct xfs_attr3_leaf_hdr *hdr3 = bp->b_addr;
419 	xfs_failaddr_t		fa;
420 
421 	fa = xfs_attr3_leaf_verify(bp);
422 	if (fa) {
423 		xfs_verifier_error(bp, -EFSCORRUPTED, fa);
424 		return;
425 	}
426 
427 	if (!xfs_has_crc(mp))
428 		return;
429 
430 	if (bip)
431 		hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
432 
433 	xfs_buf_update_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF);
434 }
435 
436 /*
437  * leaf/node format detection on trees is sketchy, so a node read can be done on
438  * leaf level blocks when detection identifies the tree as a node format tree
439  * incorrectly. In this case, we need to swap the verifier to match the correct
440  * format of the block being read.
441  */
442 static void
443 xfs_attr3_leaf_read_verify(
444 	struct xfs_buf		*bp)
445 {
446 	struct xfs_mount	*mp = bp->b_mount;
447 	xfs_failaddr_t		fa;
448 
449 	if (xfs_has_crc(mp) &&
450 	     !xfs_buf_verify_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF))
451 		xfs_verifier_error(bp, -EFSBADCRC, __this_address);
452 	else {
453 		fa = xfs_attr3_leaf_verify(bp);
454 		if (fa)
455 			xfs_verifier_error(bp, -EFSCORRUPTED, fa);
456 	}
457 }
458 
459 const struct xfs_buf_ops xfs_attr3_leaf_buf_ops = {
460 	.name = "xfs_attr3_leaf",
461 	.magic16 = { cpu_to_be16(XFS_ATTR_LEAF_MAGIC),
462 		     cpu_to_be16(XFS_ATTR3_LEAF_MAGIC) },
463 	.verify_read = xfs_attr3_leaf_read_verify,
464 	.verify_write = xfs_attr3_leaf_write_verify,
465 	.verify_struct = xfs_attr3_leaf_verify,
466 };
467 
468 int
469 xfs_attr3_leaf_read(
470 	struct xfs_trans	*tp,
471 	struct xfs_inode	*dp,
472 	xfs_ino_t		owner,
473 	xfs_dablk_t		bno,
474 	struct xfs_buf		**bpp)
475 {
476 	xfs_failaddr_t		fa;
477 	int			err;
478 
479 	err = xfs_da_read_buf(tp, dp, bno, 0, bpp, XFS_ATTR_FORK,
480 			&xfs_attr3_leaf_buf_ops);
481 	if (err || !(*bpp))
482 		return err;
483 
484 	fa = xfs_attr3_leaf_header_check(*bpp, owner);
485 	if (fa) {
486 		__xfs_buf_mark_corrupt(*bpp, fa);
487 		xfs_trans_brelse(tp, *bpp);
488 		*bpp = NULL;
489 		xfs_dirattr_mark_sick(dp, XFS_ATTR_FORK);
490 		return -EFSCORRUPTED;
491 	}
492 
493 	if (tp)
494 		xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_ATTR_LEAF_BUF);
495 	return 0;
496 }
497 
498 /*========================================================================
499  * Namespace helper routines
500  *========================================================================*/
501 
502 /*
503  * If we are in log recovery, then we want the lookup to ignore the INCOMPLETE
504  * flag on disk - if there's an incomplete attr then recovery needs to tear it
505  * down. If there's no incomplete attr, then recovery needs to tear that attr
506  * down to replace it with the attr that has been logged. In this case, the
507  * INCOMPLETE flag will not be set in attr->attr_filter, but rather
508  * XFS_DA_OP_RECOVERY will be set in args->op_flags.
509  */
510 static inline unsigned int xfs_attr_match_mask(const struct xfs_da_args *args)
511 {
512 	if (args->op_flags & XFS_DA_OP_RECOVERY)
513 		return XFS_ATTR_NSP_ONDISK_MASK;
514 	return XFS_ATTR_NSP_ONDISK_MASK | XFS_ATTR_INCOMPLETE;
515 }
516 
517 static inline bool
518 xfs_attr_parent_match(
519 	const struct xfs_da_args	*args,
520 	const void			*value,
521 	unsigned int			valuelen)
522 {
523 	ASSERT(args->value != NULL);
524 
525 	/* Parent pointers do not use remote values */
526 	if (!value)
527 		return false;
528 
529 	/*
530 	 * The only value we support is a parent rec.  However, we'll accept
531 	 * any valuelen so that offline repair can delete ATTR_PARENT values
532 	 * that are not parent pointers.
533 	 */
534 	if (valuelen != args->valuelen)
535 		return false;
536 
537 	return memcmp(args->value, value, valuelen) == 0;
538 }
539 
540 static bool
541 xfs_attr_match(
542 	struct xfs_da_args	*args,
543 	unsigned int		attr_flags,
544 	const unsigned char	*name,
545 	unsigned int		namelen,
546 	const void		*value,
547 	unsigned int		valuelen)
548 {
549 	unsigned int		mask = xfs_attr_match_mask(args);
550 
551 	if (args->namelen != namelen)
552 		return false;
553 	if ((args->attr_filter & mask) != (attr_flags & mask))
554 		return false;
555 	if (memcmp(args->name, name, namelen) != 0)
556 		return false;
557 
558 	if (attr_flags & XFS_ATTR_PARENT)
559 		return xfs_attr_parent_match(args, value, valuelen);
560 
561 	return true;
562 }
563 
564 static int
565 xfs_attr_copy_value(
566 	struct xfs_da_args	*args,
567 	unsigned char		*value,
568 	int			valuelen)
569 {
570 	/*
571 	 * Parent pointer lookups require the caller to specify the name and
572 	 * value, so don't copy anything.
573 	 */
574 	if (args->attr_filter & XFS_ATTR_PARENT)
575 		return 0;
576 
577 	/*
578 	 * No copy if all we have to do is get the length
579 	 */
580 	if (!args->valuelen) {
581 		args->valuelen = valuelen;
582 		return 0;
583 	}
584 
585 	/*
586 	 * No copy if the length of the existing buffer is too small
587 	 */
588 	if (args->valuelen < valuelen) {
589 		args->valuelen = valuelen;
590 		return -ERANGE;
591 	}
592 
593 	if (!args->value) {
594 		args->value = kvmalloc(valuelen, GFP_KERNEL | __GFP_NOLOCKDEP);
595 		if (!args->value)
596 			return -ENOMEM;
597 	}
598 	args->valuelen = valuelen;
599 
600 	/* remote block xattr requires IO for copy-in */
601 	if (args->rmtblkno)
602 		return xfs_attr_rmtval_get(args);
603 
604 	/*
605 	 * This is to prevent a GCC warning because the remote xattr case
606 	 * doesn't have a value to pass in. In that case, we never reach here,
607 	 * but GCC can't work that out and so throws a "passing NULL to
608 	 * memcpy" warning.
609 	 */
610 	if (!value)
611 		return -EINVAL;
612 	memcpy(args->value, value, valuelen);
613 	return 0;
614 }
615 
616 /*========================================================================
617  * External routines when attribute fork size < XFS_LITINO(mp).
618  *========================================================================*/
619 
620 /*
621  * Query whether the total requested number of attr fork bytes of extended
622  * attribute space will be able to fit inline.
623  *
624  * Returns zero if not, else the i_forkoff fork offset to be used in the
625  * literal area for attribute data once the new bytes have been added.
626  *
627  * i_forkoff must be 8 byte aligned, hence is stored as a >>3 value;
628  * special case for dev/uuid inodes, they have fixed size data forks.
629  */
630 int
631 xfs_attr_shortform_bytesfit(
632 	struct xfs_inode	*dp,
633 	int			bytes)
634 {
635 	struct xfs_mount	*mp = dp->i_mount;
636 	int64_t			dsize;
637 	int			minforkoff;
638 	int			maxforkoff;
639 	int			offset;
640 
641 	/*
642 	 * Check if the new size could fit at all first:
643 	 */
644 	if (bytes > XFS_LITINO(mp))
645 		return 0;
646 
647 	/* rounded down */
648 	offset = (XFS_LITINO(mp) - bytes) >> 3;
649 
650 	if (dp->i_df.if_format == XFS_DINODE_FMT_DEV) {
651 		minforkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
652 		return (offset >= minforkoff) ? minforkoff : 0;
653 	}
654 
655 	/*
656 	 * If the requested numbers of bytes is smaller or equal to the
657 	 * current attribute fork size we can always proceed.
658 	 *
659 	 * Note that if_bytes in the data fork might actually be larger than
660 	 * the current data fork size is due to delalloc extents. In that
661 	 * case either the extent count will go down when they are converted
662 	 * to real extents, or the delalloc conversion will take care of the
663 	 * literal area rebalancing.
664 	 */
665 	if (bytes <= xfs_inode_attr_fork_size(dp))
666 		return dp->i_forkoff;
667 
668 	/*
669 	 * For attr2 we can try to move the forkoff if there is space in the
670 	 * literal area, but for the old format we are done if there is no
671 	 * space in the fixed attribute fork.
672 	 */
673 	if (!xfs_has_attr2(mp))
674 		return 0;
675 
676 	dsize = dp->i_df.if_bytes;
677 
678 	switch (dp->i_df.if_format) {
679 	case XFS_DINODE_FMT_EXTENTS:
680 		/*
681 		 * If there is no attr fork and the data fork is extents,
682 		 * determine if creating the default attr fork will result
683 		 * in the extents form migrating to btree. If so, the
684 		 * minimum offset only needs to be the space required for
685 		 * the btree root.
686 		 */
687 		if (!dp->i_forkoff && dp->i_df.if_bytes >
688 		    xfs_default_attroffset(dp))
689 			dsize = xfs_bmdr_space_calc(MINDBTPTRS);
690 		break;
691 	case XFS_DINODE_FMT_BTREE:
692 		/*
693 		 * If we have a data btree then keep forkoff if we have one,
694 		 * otherwise we are adding a new attr, so then we set
695 		 * minforkoff to where the btree root can finish so we have
696 		 * plenty of room for attrs
697 		 */
698 		if (dp->i_forkoff) {
699 			if (offset < dp->i_forkoff)
700 				return 0;
701 			return dp->i_forkoff;
702 		}
703 		dsize = xfs_bmap_bmdr_space(dp->i_df.if_broot);
704 		break;
705 	}
706 
707 	/*
708 	 * A data fork btree root must have space for at least
709 	 * MINDBTPTRS key/ptr pairs if the data fork is small or empty.
710 	 */
711 	minforkoff = max_t(int64_t, dsize, xfs_bmdr_space_calc(MINDBTPTRS));
712 	minforkoff = roundup(minforkoff, 8) >> 3;
713 
714 	/* attr fork btree root can have at least this many key/ptr pairs */
715 	maxforkoff = XFS_LITINO(mp) - xfs_bmdr_space_calc(MINABTPTRS);
716 	maxforkoff = maxforkoff >> 3;	/* rounded down */
717 
718 	if (offset >= maxforkoff)
719 		return maxforkoff;
720 	if (offset >= minforkoff)
721 		return offset;
722 	return 0;
723 }
724 
725 /*
726  * Switch on the ATTR2 superblock bit (implies also FEATURES2) unless:
727  * - noattr2 mount option is set,
728  * - on-disk version bit says it is already set, or
729  * - the attr2 mount option is not set to enable automatic upgrade from attr1.
730  */
731 STATIC void
732 xfs_sbversion_add_attr2(
733 	struct xfs_mount	*mp,
734 	struct xfs_trans	*tp)
735 {
736 	if (xfs_has_noattr2(mp))
737 		return;
738 	if (mp->m_sb.sb_features2 & XFS_SB_VERSION2_ATTR2BIT)
739 		return;
740 	if (!xfs_has_attr2(mp))
741 		return;
742 
743 	spin_lock(&mp->m_sb_lock);
744 	xfs_add_attr2(mp);
745 	spin_unlock(&mp->m_sb_lock);
746 	xfs_log_sb(tp);
747 }
748 
749 /*
750  * Create the initial contents of a shortform attribute list.
751  */
752 void
753 xfs_attr_shortform_create(
754 	struct xfs_da_args	*args)
755 {
756 	struct xfs_inode	*dp = args->dp;
757 	struct xfs_ifork	*ifp = &dp->i_af;
758 	struct xfs_attr_sf_hdr	*hdr;
759 
760 	trace_xfs_attr_sf_create(args);
761 
762 	ASSERT(ifp->if_bytes == 0);
763 	if (ifp->if_format == XFS_DINODE_FMT_EXTENTS)
764 		ifp->if_format = XFS_DINODE_FMT_LOCAL;
765 
766 	hdr = xfs_idata_realloc(dp, sizeof(*hdr), XFS_ATTR_FORK);
767 	memset(hdr, 0, sizeof(*hdr));
768 	hdr->totsize = cpu_to_be16(sizeof(*hdr));
769 	xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
770 }
771 
772 /*
773  * Return the entry if the attr in args is found, or NULL if not.
774  */
775 struct xfs_attr_sf_entry *
776 xfs_attr_sf_findname(
777 	struct xfs_da_args		*args)
778 {
779 	struct xfs_attr_sf_hdr		*sf = args->dp->i_af.if_data;
780 	struct xfs_attr_sf_entry	*sfe;
781 
782 	for (sfe = xfs_attr_sf_firstentry(sf);
783 	     sfe < xfs_attr_sf_endptr(sf);
784 	     sfe = xfs_attr_sf_nextentry(sfe)) {
785 		if (xfs_attr_match(args, sfe->flags, sfe->nameval,
786 				sfe->namelen, &sfe->nameval[sfe->namelen],
787 				sfe->valuelen))
788 			return sfe;
789 	}
790 
791 	return NULL;
792 }
793 
794 /*
795  * Add a name/value pair to the shortform attribute list.
796  * Overflow from the inode has already been checked for.
797  */
798 void
799 xfs_attr_shortform_add(
800 	struct xfs_da_args		*args,
801 	int				forkoff)
802 {
803 	struct xfs_inode		*dp = args->dp;
804 	struct xfs_mount		*mp = dp->i_mount;
805 	struct xfs_ifork		*ifp = &dp->i_af;
806 	struct xfs_attr_sf_hdr		*sf = ifp->if_data;
807 	struct xfs_attr_sf_entry	*sfe;
808 	int				size;
809 
810 	trace_xfs_attr_sf_add(args);
811 
812 	dp->i_forkoff = forkoff;
813 
814 	ASSERT(ifp->if_format == XFS_DINODE_FMT_LOCAL);
815 	ASSERT(!xfs_attr_sf_findname(args));
816 
817 	size = xfs_attr_sf_entsize_byname(args->namelen, args->valuelen);
818 	sf = xfs_idata_realloc(dp, size, XFS_ATTR_FORK);
819 
820 	sfe = xfs_attr_sf_endptr(sf);
821 	sfe->namelen = args->namelen;
822 	sfe->valuelen = args->valuelen;
823 	sfe->flags = args->attr_filter;
824 	memcpy(sfe->nameval, args->name, args->namelen);
825 	memcpy(&sfe->nameval[args->namelen], args->value, args->valuelen);
826 	sf->count++;
827 	be16_add_cpu(&sf->totsize, size);
828 	xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
829 
830 	xfs_sbversion_add_attr2(mp, args->trans);
831 }
832 
833 /*
834  * After the last attribute is removed revert to original inode format,
835  * making all literal area available to the data fork once more.
836  */
837 void
838 xfs_attr_fork_remove(
839 	struct xfs_inode	*ip,
840 	struct xfs_trans	*tp)
841 {
842 	ASSERT(ip->i_af.if_nextents == 0);
843 
844 	xfs_ifork_zap_attr(ip);
845 	ip->i_forkoff = 0;
846 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
847 }
848 
849 /*
850  * Remove an attribute from the shortform attribute list structure.
851  */
852 int
853 xfs_attr_sf_removename(
854 	struct xfs_da_args		*args)
855 {
856 	struct xfs_inode		*dp = args->dp;
857 	struct xfs_mount		*mp = dp->i_mount;
858 	struct xfs_attr_sf_hdr		*sf = dp->i_af.if_data;
859 	struct xfs_attr_sf_entry	*sfe;
860 	uint16_t			totsize = be16_to_cpu(sf->totsize);
861 	void				*next, *end;
862 	int				size = 0;
863 
864 	trace_xfs_attr_sf_remove(args);
865 
866 	sfe = xfs_attr_sf_findname(args);
867 	if (!sfe) {
868 		/*
869 		 * If we are recovering an operation, finding nothing to remove
870 		 * is not an error, it just means there was nothing to clean up.
871 		 */
872 		if (args->op_flags & XFS_DA_OP_RECOVERY)
873 			return 0;
874 		return -ENOATTR;
875 	}
876 
877 	/*
878 	 * Fix up the attribute fork data, covering the hole
879 	 */
880 	size = xfs_attr_sf_entsize(sfe);
881 	next = xfs_attr_sf_nextentry(sfe);
882 	end = xfs_attr_sf_endptr(sf);
883 	if (next < end)
884 		memmove(sfe, next, end - next);
885 	sf->count--;
886 	totsize -= size;
887 	sf->totsize = cpu_to_be16(totsize);
888 
889 	/*
890 	 * Fix up the start offset of the attribute fork
891 	 */
892 	if (totsize == sizeof(struct xfs_attr_sf_hdr) && xfs_has_attr2(mp) &&
893 	    (dp->i_df.if_format != XFS_DINODE_FMT_BTREE) &&
894 	    !(args->op_flags & (XFS_DA_OP_ADDNAME | XFS_DA_OP_REPLACE)) &&
895 	    !xfs_has_parent(mp)) {
896 		xfs_attr_fork_remove(dp, args->trans);
897 	} else {
898 		xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
899 		dp->i_forkoff = xfs_attr_shortform_bytesfit(dp, totsize);
900 		ASSERT(dp->i_forkoff);
901 		ASSERT(totsize > sizeof(struct xfs_attr_sf_hdr) ||
902 				(args->op_flags & XFS_DA_OP_ADDNAME) ||
903 				!xfs_has_attr2(mp) ||
904 				dp->i_df.if_format == XFS_DINODE_FMT_BTREE ||
905 				xfs_has_parent(mp));
906 		xfs_trans_log_inode(args->trans, dp,
907 					XFS_ILOG_CORE | XFS_ILOG_ADATA);
908 	}
909 
910 	xfs_sbversion_add_attr2(mp, args->trans);
911 
912 	return 0;
913 }
914 
915 /*
916  * Retrieve the attribute value and length.
917  *
918  * If args->valuelen is zero, only the length needs to be returned.  Unlike a
919  * lookup, we only return an error if the attribute does not exist or we can't
920  * retrieve the value.
921  */
922 int
923 xfs_attr_shortform_getvalue(
924 	struct xfs_da_args		*args)
925 {
926 	struct xfs_attr_sf_entry	*sfe;
927 
928 	ASSERT(args->dp->i_af.if_format == XFS_DINODE_FMT_LOCAL);
929 
930 	trace_xfs_attr_sf_lookup(args);
931 
932 	sfe = xfs_attr_sf_findname(args);
933 	if (!sfe)
934 		return -ENOATTR;
935 	return xfs_attr_copy_value(args, &sfe->nameval[args->namelen],
936 			sfe->valuelen);
937 }
938 
939 /* Convert from using the shortform to the leaf format. */
940 int
941 xfs_attr_shortform_to_leaf(
942 	struct xfs_da_args		*args)
943 {
944 	struct xfs_inode		*dp = args->dp;
945 	struct xfs_ifork		*ifp = &dp->i_af;
946 	struct xfs_attr_sf_hdr		*sf = ifp->if_data;
947 	struct xfs_attr_sf_entry	*sfe;
948 	int				size = be16_to_cpu(sf->totsize);
949 	struct xfs_da_args		nargs;
950 	char				*tmpbuffer;
951 	int				error, i;
952 	xfs_dablk_t			blkno;
953 	struct xfs_buf			*bp;
954 
955 	trace_xfs_attr_sf_to_leaf(args);
956 
957 	tmpbuffer = kmalloc(size, GFP_KERNEL | __GFP_NOFAIL);
958 	memcpy(tmpbuffer, ifp->if_data, size);
959 	sf = (struct xfs_attr_sf_hdr *)tmpbuffer;
960 
961 	xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
962 	xfs_bmap_local_to_extents_empty(args->trans, dp, XFS_ATTR_FORK);
963 
964 	bp = NULL;
965 	error = xfs_da_grow_inode(args, &blkno);
966 	if (error)
967 		goto out;
968 
969 	ASSERT(blkno == 0);
970 	error = xfs_attr3_leaf_create(args, blkno, &bp);
971 	if (error)
972 		goto out;
973 
974 	memset((char *)&nargs, 0, sizeof(nargs));
975 	nargs.dp = dp;
976 	nargs.geo = args->geo;
977 	nargs.total = args->total;
978 	nargs.whichfork = XFS_ATTR_FORK;
979 	nargs.trans = args->trans;
980 	nargs.op_flags = XFS_DA_OP_OKNOENT;
981 	nargs.owner = args->owner;
982 
983 	sfe = xfs_attr_sf_firstentry(sf);
984 	for (i = 0; i < sf->count; i++) {
985 		nargs.name = sfe->nameval;
986 		nargs.namelen = sfe->namelen;
987 		nargs.value = &sfe->nameval[nargs.namelen];
988 		nargs.valuelen = sfe->valuelen;
989 		nargs.attr_filter = sfe->flags & XFS_ATTR_NSP_ONDISK_MASK;
990 		if (!xfs_attr_check_namespace(sfe->flags)) {
991 			xfs_da_mark_sick(args);
992 			error = -EFSCORRUPTED;
993 			goto out;
994 		}
995 		xfs_attr_sethash(&nargs);
996 		error = xfs_attr3_leaf_lookup_int(bp, &nargs); /* set a->index */
997 		ASSERT(error == -ENOATTR);
998 		error = xfs_attr3_leaf_add(bp, &nargs);
999 		ASSERT(error != -ENOSPC);
1000 		if (error)
1001 			goto out;
1002 		sfe = xfs_attr_sf_nextentry(sfe);
1003 	}
1004 	error = 0;
1005 out:
1006 	kfree(tmpbuffer);
1007 	return error;
1008 }
1009 
1010 /*
1011  * Check a leaf attribute block to see if all the entries would fit into
1012  * a shortform attribute list.
1013  */
1014 int
1015 xfs_attr_shortform_allfit(
1016 	struct xfs_buf		*bp,
1017 	struct xfs_inode	*dp)
1018 {
1019 	struct xfs_attr_leafblock *leaf;
1020 	struct xfs_attr_leaf_entry *entry;
1021 	xfs_attr_leaf_name_local_t *name_loc;
1022 	struct xfs_attr3_icleaf_hdr leafhdr;
1023 	int			bytes;
1024 	int			i;
1025 	struct xfs_mount	*mp = bp->b_mount;
1026 
1027 	leaf = bp->b_addr;
1028 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf);
1029 	entry = xfs_attr3_leaf_entryp(leaf);
1030 
1031 	bytes = sizeof(struct xfs_attr_sf_hdr);
1032 	for (i = 0; i < leafhdr.count; entry++, i++) {
1033 		if (entry->flags & XFS_ATTR_INCOMPLETE)
1034 			continue;		/* don't copy partial entries */
1035 		if (!(entry->flags & XFS_ATTR_LOCAL))
1036 			return 0;
1037 		name_loc = xfs_attr3_leaf_name_local(leaf, i);
1038 		if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX)
1039 			return 0;
1040 		if (be16_to_cpu(name_loc->valuelen) >= XFS_ATTR_SF_ENTSIZE_MAX)
1041 			return 0;
1042 		bytes += xfs_attr_sf_entsize_byname(name_loc->namelen,
1043 					be16_to_cpu(name_loc->valuelen));
1044 	}
1045 	if (xfs_has_attr2(dp->i_mount) &&
1046 	    (dp->i_df.if_format != XFS_DINODE_FMT_BTREE) &&
1047 	    (bytes == sizeof(struct xfs_attr_sf_hdr)))
1048 		return -1;
1049 	return xfs_attr_shortform_bytesfit(dp, bytes);
1050 }
1051 
1052 /* Verify the consistency of a raw inline attribute fork. */
1053 xfs_failaddr_t
1054 xfs_attr_shortform_verify(
1055 	struct xfs_attr_sf_hdr		*sfp,
1056 	size_t				size)
1057 {
1058 	struct xfs_attr_sf_entry	*sfep = xfs_attr_sf_firstentry(sfp);
1059 	struct xfs_attr_sf_entry	*next_sfep;
1060 	char				*endp;
1061 	int				i;
1062 
1063 	/*
1064 	 * Give up if the attribute is way too short.
1065 	 */
1066 	if (size < sizeof(struct xfs_attr_sf_hdr))
1067 		return __this_address;
1068 
1069 	endp = (char *)sfp + size;
1070 
1071 	/* Check all reported entries */
1072 	for (i = 0; i < sfp->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 (!xfs_attr_check_namespace(sfep->flags))
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 = kvmalloc(args->geo->blksize, GFP_KERNEL | __GFP_NOFAIL);
1142 	memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1143 
1144 	leaf = (xfs_attr_leafblock_t *)tmpbuffer;
1145 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1146 	entry = xfs_attr3_leaf_entryp(leaf);
1147 
1148 	/* XXX (dgc): buffer is about to be marked stale - why zero it? */
1149 	memset(bp->b_addr, 0, args->geo->blksize);
1150 
1151 	/*
1152 	 * Clean out the prior contents of the attribute list.
1153 	 */
1154 	error = xfs_da_shrink_inode(args, 0, bp);
1155 	if (error)
1156 		goto out;
1157 
1158 	if (forkoff == -1) {
1159 		/*
1160 		 * Don't remove the attr fork if this operation is the first
1161 		 * part of a attr replace operations. We're going to add a new
1162 		 * attr immediately, so we need to keep the attr fork around in
1163 		 * this case.
1164 		 */
1165 		if (!(args->op_flags & XFS_DA_OP_REPLACE)) {
1166 			ASSERT(xfs_has_attr2(dp->i_mount));
1167 			ASSERT(dp->i_df.if_format != XFS_DINODE_FMT_BTREE);
1168 			xfs_attr_fork_remove(dp, args->trans);
1169 		}
1170 		goto out;
1171 	}
1172 
1173 	xfs_attr_shortform_create(args);
1174 
1175 	/*
1176 	 * Copy the attributes
1177 	 */
1178 	memset((char *)&nargs, 0, sizeof(nargs));
1179 	nargs.geo = args->geo;
1180 	nargs.dp = dp;
1181 	nargs.total = args->total;
1182 	nargs.whichfork = XFS_ATTR_FORK;
1183 	nargs.trans = args->trans;
1184 	nargs.op_flags = XFS_DA_OP_OKNOENT;
1185 	nargs.owner = args->owner;
1186 
1187 	for (i = 0; i < ichdr.count; entry++, i++) {
1188 		if (entry->flags & XFS_ATTR_INCOMPLETE)
1189 			continue;	/* don't copy partial entries */
1190 		if (!entry->nameidx)
1191 			continue;
1192 		ASSERT(entry->flags & XFS_ATTR_LOCAL);
1193 		name_loc = xfs_attr3_leaf_name_local(leaf, i);
1194 		nargs.name = name_loc->nameval;
1195 		nargs.namelen = name_loc->namelen;
1196 		nargs.value = &name_loc->nameval[nargs.namelen];
1197 		nargs.valuelen = be16_to_cpu(name_loc->valuelen);
1198 		nargs.hashval = be32_to_cpu(entry->hashval);
1199 		nargs.attr_filter = entry->flags & XFS_ATTR_NSP_ONDISK_MASK;
1200 		xfs_attr_shortform_add(&nargs, forkoff);
1201 	}
1202 	error = 0;
1203 
1204 out:
1205 	kvfree(tmpbuffer);
1206 	return error;
1207 }
1208 
1209 /*
1210  * Convert from using a single leaf to a root node and a leaf.
1211  */
1212 int
1213 xfs_attr3_leaf_to_node(
1214 	struct xfs_da_args	*args)
1215 {
1216 	struct xfs_attr_leafblock *leaf;
1217 	struct xfs_attr3_icleaf_hdr icleafhdr;
1218 	struct xfs_attr_leaf_entry *entries;
1219 	struct xfs_da3_icnode_hdr icnodehdr;
1220 	struct xfs_da_intnode	*node;
1221 	struct xfs_inode	*dp = args->dp;
1222 	struct xfs_mount	*mp = dp->i_mount;
1223 	struct xfs_buf		*bp1 = NULL;
1224 	struct xfs_buf		*bp2 = NULL;
1225 	xfs_dablk_t		blkno;
1226 	int			error;
1227 
1228 	trace_xfs_attr_leaf_to_node(args);
1229 
1230 	if (XFS_TEST_ERROR(false, mp, XFS_ERRTAG_ATTR_LEAF_TO_NODE)) {
1231 		error = -EIO;
1232 		goto out;
1233 	}
1234 
1235 	error = xfs_da_grow_inode(args, &blkno);
1236 	if (error)
1237 		goto out;
1238 	error = xfs_attr3_leaf_read(args->trans, dp, args->owner, 0, &bp1);
1239 	if (error)
1240 		goto out;
1241 
1242 	error = xfs_da_get_buf(args->trans, dp, blkno, &bp2, XFS_ATTR_FORK);
1243 	if (error)
1244 		goto out;
1245 
1246 	/*
1247 	 * Copy leaf to new buffer and log it.
1248 	 */
1249 	xfs_da_buf_copy(bp2, bp1, args->geo->blksize);
1250 	xfs_trans_log_buf(args->trans, bp2, 0, args->geo->blksize - 1);
1251 
1252 	/*
1253 	 * Set up the new root node.
1254 	 */
1255 	error = xfs_da3_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK);
1256 	if (error)
1257 		goto out;
1258 	node = bp1->b_addr;
1259 	xfs_da3_node_hdr_from_disk(mp, &icnodehdr, node);
1260 
1261 	leaf = bp2->b_addr;
1262 	xfs_attr3_leaf_hdr_from_disk(args->geo, &icleafhdr, leaf);
1263 	entries = xfs_attr3_leaf_entryp(leaf);
1264 
1265 	/* both on-disk, don't endian-flip twice */
1266 	icnodehdr.btree[0].hashval = entries[icleafhdr.count - 1].hashval;
1267 	icnodehdr.btree[0].before = cpu_to_be32(blkno);
1268 	icnodehdr.count = 1;
1269 	xfs_da3_node_hdr_to_disk(dp->i_mount, node, &icnodehdr);
1270 	xfs_trans_log_buf(args->trans, bp1, 0, args->geo->blksize - 1);
1271 	error = 0;
1272 out:
1273 	return error;
1274 }
1275 
1276 /*========================================================================
1277  * Routines used for growing the Btree.
1278  *========================================================================*/
1279 
1280 /*
1281  * Create the initial contents of a leaf attribute list
1282  * or a leaf in a node attribute list.
1283  */
1284 STATIC int
1285 xfs_attr3_leaf_create(
1286 	struct xfs_da_args	*args,
1287 	xfs_dablk_t		blkno,
1288 	struct xfs_buf		**bpp)
1289 {
1290 	struct xfs_attr_leafblock *leaf;
1291 	struct xfs_attr3_icleaf_hdr ichdr;
1292 	struct xfs_inode	*dp = args->dp;
1293 	struct xfs_mount	*mp = dp->i_mount;
1294 	struct xfs_buf		*bp;
1295 	int			error;
1296 
1297 	trace_xfs_attr_leaf_create(args);
1298 
1299 	error = xfs_da_get_buf(args->trans, args->dp, blkno, &bp,
1300 					    XFS_ATTR_FORK);
1301 	if (error)
1302 		return error;
1303 	bp->b_ops = &xfs_attr3_leaf_buf_ops;
1304 	xfs_trans_buf_set_type(args->trans, bp, XFS_BLFT_ATTR_LEAF_BUF);
1305 	leaf = bp->b_addr;
1306 	memset(leaf, 0, args->geo->blksize);
1307 
1308 	memset(&ichdr, 0, sizeof(ichdr));
1309 	ichdr.firstused = args->geo->blksize;
1310 
1311 	if (xfs_has_crc(mp)) {
1312 		struct xfs_da3_blkinfo *hdr3 = bp->b_addr;
1313 
1314 		ichdr.magic = XFS_ATTR3_LEAF_MAGIC;
1315 
1316 		hdr3->blkno = cpu_to_be64(xfs_buf_daddr(bp));
1317 		hdr3->owner = cpu_to_be64(args->owner);
1318 		uuid_copy(&hdr3->uuid, &mp->m_sb.sb_meta_uuid);
1319 
1320 		ichdr.freemap[0].base = sizeof(struct xfs_attr3_leaf_hdr);
1321 	} else {
1322 		ichdr.magic = XFS_ATTR_LEAF_MAGIC;
1323 		ichdr.freemap[0].base = sizeof(struct xfs_attr_leaf_hdr);
1324 	}
1325 	ichdr.freemap[0].size = ichdr.firstused - ichdr.freemap[0].base;
1326 
1327 	xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1328 	xfs_trans_log_buf(args->trans, bp, 0, args->geo->blksize - 1);
1329 
1330 	*bpp = bp;
1331 	return 0;
1332 }
1333 
1334 /*
1335  * Split the leaf node, rebalance, then add the new entry.
1336  */
1337 int
1338 xfs_attr3_leaf_split(
1339 	struct xfs_da_state	*state,
1340 	struct xfs_da_state_blk	*oldblk,
1341 	struct xfs_da_state_blk	*newblk)
1342 {
1343 	xfs_dablk_t blkno;
1344 	int error;
1345 
1346 	trace_xfs_attr_leaf_split(state->args);
1347 
1348 	/*
1349 	 * Allocate space for a new leaf node.
1350 	 */
1351 	ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
1352 	error = xfs_da_grow_inode(state->args, &blkno);
1353 	if (error)
1354 		return error;
1355 	error = xfs_attr3_leaf_create(state->args, blkno, &newblk->bp);
1356 	if (error)
1357 		return error;
1358 	newblk->blkno = blkno;
1359 	newblk->magic = XFS_ATTR_LEAF_MAGIC;
1360 
1361 	/*
1362 	 * Rebalance the entries across the two leaves.
1363 	 * NOTE: rebalance() currently depends on the 2nd block being empty.
1364 	 */
1365 	xfs_attr3_leaf_rebalance(state, oldblk, newblk);
1366 	error = xfs_da3_blk_link(state, oldblk, newblk);
1367 	if (error)
1368 		return error;
1369 
1370 	/*
1371 	 * Save info on "old" attribute for "atomic rename" ops, leaf_add()
1372 	 * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
1373 	 * "new" attrs info.  Will need the "old" info to remove it later.
1374 	 *
1375 	 * Insert the "new" entry in the correct block.
1376 	 */
1377 	if (state->inleaf) {
1378 		trace_xfs_attr_leaf_add_old(state->args);
1379 		error = xfs_attr3_leaf_add(oldblk->bp, state->args);
1380 	} else {
1381 		trace_xfs_attr_leaf_add_new(state->args);
1382 		error = xfs_attr3_leaf_add(newblk->bp, state->args);
1383 	}
1384 
1385 	/*
1386 	 * Update last hashval in each block since we added the name.
1387 	 */
1388 	oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
1389 	newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
1390 	return error;
1391 }
1392 
1393 /*
1394  * Add a name to the leaf attribute list structure.
1395  */
1396 int
1397 xfs_attr3_leaf_add(
1398 	struct xfs_buf		*bp,
1399 	struct xfs_da_args	*args)
1400 {
1401 	struct xfs_attr_leafblock *leaf;
1402 	struct xfs_attr3_icleaf_hdr ichdr;
1403 	int			tablesize;
1404 	int			entsize;
1405 	int			sum;
1406 	int			tmp;
1407 	int			i;
1408 
1409 	trace_xfs_attr_leaf_add(args);
1410 
1411 	leaf = bp->b_addr;
1412 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1413 	ASSERT(args->index >= 0 && args->index <= ichdr.count);
1414 	entsize = xfs_attr_leaf_newentsize(args, NULL);
1415 
1416 	/*
1417 	 * Search through freemap for first-fit on new name length.
1418 	 * (may need to figure in size of entry struct too)
1419 	 */
1420 	tablesize = (ichdr.count + 1) * sizeof(xfs_attr_leaf_entry_t)
1421 					+ xfs_attr3_leaf_hdr_size(leaf);
1422 	for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE - 1; i >= 0; i--) {
1423 		if (tablesize > ichdr.firstused) {
1424 			sum += ichdr.freemap[i].size;
1425 			continue;
1426 		}
1427 		if (!ichdr.freemap[i].size)
1428 			continue;	/* no space in this map */
1429 		tmp = entsize;
1430 		if (ichdr.freemap[i].base < ichdr.firstused)
1431 			tmp += sizeof(xfs_attr_leaf_entry_t);
1432 		if (ichdr.freemap[i].size >= tmp) {
1433 			tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, i);
1434 			goto out_log_hdr;
1435 		}
1436 		sum += ichdr.freemap[i].size;
1437 	}
1438 
1439 	/*
1440 	 * If there are no holes in the address space of the block,
1441 	 * and we don't have enough freespace, then compaction will do us
1442 	 * no good and we should just give up.
1443 	 */
1444 	if (!ichdr.holes && sum < entsize)
1445 		return -ENOSPC;
1446 
1447 	/*
1448 	 * Compact the entries to coalesce free space.
1449 	 * This may change the hdr->count via dropping INCOMPLETE entries.
1450 	 */
1451 	xfs_attr3_leaf_compact(args, &ichdr, bp);
1452 
1453 	/*
1454 	 * After compaction, the block is guaranteed to have only one
1455 	 * free region, in freemap[0].  If it is not big enough, give up.
1456 	 */
1457 	if (ichdr.freemap[0].size < (entsize + sizeof(xfs_attr_leaf_entry_t))) {
1458 		tmp = -ENOSPC;
1459 		goto out_log_hdr;
1460 	}
1461 
1462 	tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, 0);
1463 
1464 out_log_hdr:
1465 	xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1466 	xfs_trans_log_buf(args->trans, bp,
1467 		XFS_DA_LOGRANGE(leaf, &leaf->hdr,
1468 				xfs_attr3_leaf_hdr_size(leaf)));
1469 	return tmp;
1470 }
1471 
1472 /*
1473  * Add a name to a leaf attribute list structure.
1474  */
1475 STATIC int
1476 xfs_attr3_leaf_add_work(
1477 	struct xfs_buf		*bp,
1478 	struct xfs_attr3_icleaf_hdr *ichdr,
1479 	struct xfs_da_args	*args,
1480 	int			mapindex)
1481 {
1482 	struct xfs_attr_leafblock *leaf;
1483 	struct xfs_attr_leaf_entry *entry;
1484 	struct xfs_attr_leaf_name_local *name_loc;
1485 	struct xfs_attr_leaf_name_remote *name_rmt;
1486 	struct xfs_mount	*mp;
1487 	int			tmp;
1488 	int			i;
1489 
1490 	trace_xfs_attr_leaf_add_work(args);
1491 
1492 	leaf = bp->b_addr;
1493 	ASSERT(mapindex >= 0 && mapindex < XFS_ATTR_LEAF_MAPSIZE);
1494 	ASSERT(args->index >= 0 && args->index <= ichdr->count);
1495 
1496 	/*
1497 	 * Force open some space in the entry array and fill it in.
1498 	 */
1499 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1500 	if (args->index < ichdr->count) {
1501 		tmp  = ichdr->count - args->index;
1502 		tmp *= sizeof(xfs_attr_leaf_entry_t);
1503 		memmove(entry + 1, entry, tmp);
1504 		xfs_trans_log_buf(args->trans, bp,
1505 		    XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1506 	}
1507 	ichdr->count++;
1508 
1509 	/*
1510 	 * Allocate space for the new string (at the end of the run).
1511 	 */
1512 	mp = args->trans->t_mountp;
1513 	ASSERT(ichdr->freemap[mapindex].base < args->geo->blksize);
1514 	ASSERT((ichdr->freemap[mapindex].base & 0x3) == 0);
1515 	ASSERT(ichdr->freemap[mapindex].size >=
1516 		xfs_attr_leaf_newentsize(args, NULL));
1517 	ASSERT(ichdr->freemap[mapindex].size < args->geo->blksize);
1518 	ASSERT((ichdr->freemap[mapindex].size & 0x3) == 0);
1519 
1520 	ichdr->freemap[mapindex].size -= xfs_attr_leaf_newentsize(args, &tmp);
1521 
1522 	entry->nameidx = cpu_to_be16(ichdr->freemap[mapindex].base +
1523 				     ichdr->freemap[mapindex].size);
1524 	entry->hashval = cpu_to_be32(args->hashval);
1525 	entry->flags = args->attr_filter;
1526 	if (tmp)
1527 		entry->flags |= XFS_ATTR_LOCAL;
1528 	if (args->op_flags & XFS_DA_OP_REPLACE) {
1529 		if (!(args->op_flags & XFS_DA_OP_LOGGED))
1530 			entry->flags |= XFS_ATTR_INCOMPLETE;
1531 		if ((args->blkno2 == args->blkno) &&
1532 		    (args->index2 <= args->index)) {
1533 			args->index2++;
1534 		}
1535 	}
1536 	xfs_trans_log_buf(args->trans, bp,
1537 			  XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1538 	ASSERT((args->index == 0) ||
1539 	       (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1540 	ASSERT((args->index == ichdr->count - 1) ||
1541 	       (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1542 
1543 	/*
1544 	 * For "remote" attribute values, simply note that we need to
1545 	 * allocate space for the "remote" value.  We can't actually
1546 	 * allocate the extents in this transaction, and we can't decide
1547 	 * which blocks they should be as we might allocate more blocks
1548 	 * as part of this transaction (a split operation for example).
1549 	 */
1550 	if (entry->flags & XFS_ATTR_LOCAL) {
1551 		name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
1552 		name_loc->namelen = args->namelen;
1553 		name_loc->valuelen = cpu_to_be16(args->valuelen);
1554 		memcpy((char *)name_loc->nameval, args->name, args->namelen);
1555 		memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1556 				   be16_to_cpu(name_loc->valuelen));
1557 	} else {
1558 		name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
1559 		name_rmt->namelen = args->namelen;
1560 		memcpy((char *)name_rmt->name, args->name, args->namelen);
1561 		entry->flags |= XFS_ATTR_INCOMPLETE;
1562 		/* just in case */
1563 		name_rmt->valuelen = 0;
1564 		name_rmt->valueblk = 0;
1565 		args->rmtblkno = 1;
1566 		args->rmtblkcnt = xfs_attr3_rmt_blocks(mp, args->valuelen);
1567 		args->rmtvaluelen = args->valuelen;
1568 	}
1569 	xfs_trans_log_buf(args->trans, bp,
1570 	     XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1571 				   xfs_attr_leaf_entsize(leaf, args->index)));
1572 
1573 	/*
1574 	 * Update the control info for this leaf node
1575 	 */
1576 	if (be16_to_cpu(entry->nameidx) < ichdr->firstused)
1577 		ichdr->firstused = be16_to_cpu(entry->nameidx);
1578 
1579 	ASSERT(ichdr->firstused >= ichdr->count * sizeof(xfs_attr_leaf_entry_t)
1580 					+ xfs_attr3_leaf_hdr_size(leaf));
1581 	tmp = (ichdr->count - 1) * sizeof(xfs_attr_leaf_entry_t)
1582 					+ xfs_attr3_leaf_hdr_size(leaf);
1583 
1584 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1585 		if (ichdr->freemap[i].base == tmp) {
1586 			ichdr->freemap[i].base += sizeof(xfs_attr_leaf_entry_t);
1587 			ichdr->freemap[i].size -=
1588 				min_t(uint16_t, ichdr->freemap[i].size,
1589 						sizeof(xfs_attr_leaf_entry_t));
1590 		}
1591 	}
1592 	ichdr->usedbytes += xfs_attr_leaf_entsize(leaf, args->index);
1593 	return 0;
1594 }
1595 
1596 /*
1597  * Garbage collect a leaf attribute list block by copying it to a new buffer.
1598  */
1599 STATIC void
1600 xfs_attr3_leaf_compact(
1601 	struct xfs_da_args	*args,
1602 	struct xfs_attr3_icleaf_hdr *ichdr_dst,
1603 	struct xfs_buf		*bp)
1604 {
1605 	struct xfs_attr_leafblock *leaf_src;
1606 	struct xfs_attr_leafblock *leaf_dst;
1607 	struct xfs_attr3_icleaf_hdr ichdr_src;
1608 	struct xfs_trans	*trans = args->trans;
1609 	char			*tmpbuffer;
1610 
1611 	trace_xfs_attr_leaf_compact(args);
1612 
1613 	tmpbuffer = kvmalloc(args->geo->blksize, GFP_KERNEL | __GFP_NOFAIL);
1614 	memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1615 	memset(bp->b_addr, 0, args->geo->blksize);
1616 	leaf_src = (xfs_attr_leafblock_t *)tmpbuffer;
1617 	leaf_dst = bp->b_addr;
1618 
1619 	/*
1620 	 * Copy the on-disk header back into the destination buffer to ensure
1621 	 * all the information in the header that is not part of the incore
1622 	 * header structure is preserved.
1623 	 */
1624 	memcpy(bp->b_addr, tmpbuffer, xfs_attr3_leaf_hdr_size(leaf_src));
1625 
1626 	/* Initialise the incore headers */
1627 	ichdr_src = *ichdr_dst;	/* struct copy */
1628 	ichdr_dst->firstused = args->geo->blksize;
1629 	ichdr_dst->usedbytes = 0;
1630 	ichdr_dst->count = 0;
1631 	ichdr_dst->holes = 0;
1632 	ichdr_dst->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_src);
1633 	ichdr_dst->freemap[0].size = ichdr_dst->firstused -
1634 						ichdr_dst->freemap[0].base;
1635 
1636 	/* write the header back to initialise the underlying buffer */
1637 	xfs_attr3_leaf_hdr_to_disk(args->geo, leaf_dst, ichdr_dst);
1638 
1639 	/*
1640 	 * Copy all entry's in the same (sorted) order,
1641 	 * but allocate name/value pairs packed and in sequence.
1642 	 */
1643 	xfs_attr3_leaf_moveents(args, leaf_src, &ichdr_src, 0,
1644 				leaf_dst, ichdr_dst, 0, ichdr_src.count);
1645 	/*
1646 	 * this logs the entire buffer, but the caller must write the header
1647 	 * back to the buffer when it is finished modifying it.
1648 	 */
1649 	xfs_trans_log_buf(trans, bp, 0, args->geo->blksize - 1);
1650 
1651 	kvfree(tmpbuffer);
1652 }
1653 
1654 /*
1655  * Compare two leaf blocks "order".
1656  * Return 0 unless leaf2 should go before leaf1.
1657  */
1658 static int
1659 xfs_attr3_leaf_order(
1660 	struct xfs_buf	*leaf1_bp,
1661 	struct xfs_attr3_icleaf_hdr *leaf1hdr,
1662 	struct xfs_buf	*leaf2_bp,
1663 	struct xfs_attr3_icleaf_hdr *leaf2hdr)
1664 {
1665 	struct xfs_attr_leaf_entry *entries1;
1666 	struct xfs_attr_leaf_entry *entries2;
1667 
1668 	entries1 = xfs_attr3_leaf_entryp(leaf1_bp->b_addr);
1669 	entries2 = xfs_attr3_leaf_entryp(leaf2_bp->b_addr);
1670 	if (leaf1hdr->count > 0 && leaf2hdr->count > 0 &&
1671 	    ((be32_to_cpu(entries2[0].hashval) <
1672 	      be32_to_cpu(entries1[0].hashval)) ||
1673 	     (be32_to_cpu(entries2[leaf2hdr->count - 1].hashval) <
1674 	      be32_to_cpu(entries1[leaf1hdr->count - 1].hashval)))) {
1675 		return 1;
1676 	}
1677 	return 0;
1678 }
1679 
1680 int
1681 xfs_attr_leaf_order(
1682 	struct xfs_buf	*leaf1_bp,
1683 	struct xfs_buf	*leaf2_bp)
1684 {
1685 	struct xfs_attr3_icleaf_hdr ichdr1;
1686 	struct xfs_attr3_icleaf_hdr ichdr2;
1687 	struct xfs_mount *mp = leaf1_bp->b_mount;
1688 
1689 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr1, leaf1_bp->b_addr);
1690 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr2, leaf2_bp->b_addr);
1691 	return xfs_attr3_leaf_order(leaf1_bp, &ichdr1, leaf2_bp, &ichdr2);
1692 }
1693 
1694 /*
1695  * Redistribute the attribute list entries between two leaf nodes,
1696  * taking into account the size of the new entry.
1697  *
1698  * NOTE: if new block is empty, then it will get the upper half of the
1699  * old block.  At present, all (one) callers pass in an empty second block.
1700  *
1701  * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1702  * to match what it is doing in splitting the attribute leaf block.  Those
1703  * values are used in "atomic rename" operations on attributes.  Note that
1704  * the "new" and "old" values can end up in different blocks.
1705  */
1706 STATIC void
1707 xfs_attr3_leaf_rebalance(
1708 	struct xfs_da_state	*state,
1709 	struct xfs_da_state_blk	*blk1,
1710 	struct xfs_da_state_blk	*blk2)
1711 {
1712 	struct xfs_da_args	*args;
1713 	struct xfs_attr_leafblock *leaf1;
1714 	struct xfs_attr_leafblock *leaf2;
1715 	struct xfs_attr3_icleaf_hdr ichdr1;
1716 	struct xfs_attr3_icleaf_hdr ichdr2;
1717 	struct xfs_attr_leaf_entry *entries1;
1718 	struct xfs_attr_leaf_entry *entries2;
1719 	int			count;
1720 	int			totallen;
1721 	int			max;
1722 	int			space;
1723 	int			swap;
1724 
1725 	/*
1726 	 * Set up environment.
1727 	 */
1728 	ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1729 	ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1730 	leaf1 = blk1->bp->b_addr;
1731 	leaf2 = blk2->bp->b_addr;
1732 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr1, leaf1);
1733 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, leaf2);
1734 	ASSERT(ichdr2.count == 0);
1735 	args = state->args;
1736 
1737 	trace_xfs_attr_leaf_rebalance(args);
1738 
1739 	/*
1740 	 * Check ordering of blocks, reverse if it makes things simpler.
1741 	 *
1742 	 * NOTE: Given that all (current) callers pass in an empty
1743 	 * second block, this code should never set "swap".
1744 	 */
1745 	swap = 0;
1746 	if (xfs_attr3_leaf_order(blk1->bp, &ichdr1, blk2->bp, &ichdr2)) {
1747 		swap(blk1, blk2);
1748 
1749 		/* swap structures rather than reconverting them */
1750 		swap(ichdr1, ichdr2);
1751 
1752 		leaf1 = blk1->bp->b_addr;
1753 		leaf2 = blk2->bp->b_addr;
1754 		swap = 1;
1755 	}
1756 
1757 	/*
1758 	 * Examine entries until we reduce the absolute difference in
1759 	 * byte usage between the two blocks to a minimum.  Then get
1760 	 * the direction to copy and the number of elements to move.
1761 	 *
1762 	 * "inleaf" is true if the new entry should be inserted into blk1.
1763 	 * If "swap" is also true, then reverse the sense of "inleaf".
1764 	 */
1765 	state->inleaf = xfs_attr3_leaf_figure_balance(state, blk1, &ichdr1,
1766 						      blk2, &ichdr2,
1767 						      &count, &totallen);
1768 	if (swap)
1769 		state->inleaf = !state->inleaf;
1770 
1771 	/*
1772 	 * Move any entries required from leaf to leaf:
1773 	 */
1774 	if (count < ichdr1.count) {
1775 		/*
1776 		 * Figure the total bytes to be added to the destination leaf.
1777 		 */
1778 		/* number entries being moved */
1779 		count = ichdr1.count - count;
1780 		space  = ichdr1.usedbytes - totallen;
1781 		space += count * sizeof(xfs_attr_leaf_entry_t);
1782 
1783 		/*
1784 		 * leaf2 is the destination, compact it if it looks tight.
1785 		 */
1786 		max  = ichdr2.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1787 		max -= ichdr2.count * sizeof(xfs_attr_leaf_entry_t);
1788 		if (space > max)
1789 			xfs_attr3_leaf_compact(args, &ichdr2, blk2->bp);
1790 
1791 		/*
1792 		 * Move high entries from leaf1 to low end of leaf2.
1793 		 */
1794 		xfs_attr3_leaf_moveents(args, leaf1, &ichdr1,
1795 				ichdr1.count - count, leaf2, &ichdr2, 0, count);
1796 
1797 	} else if (count > ichdr1.count) {
1798 		/*
1799 		 * I assert that since all callers pass in an empty
1800 		 * second buffer, this code should never execute.
1801 		 */
1802 		ASSERT(0);
1803 
1804 		/*
1805 		 * Figure the total bytes to be added to the destination leaf.
1806 		 */
1807 		/* number entries being moved */
1808 		count -= ichdr1.count;
1809 		space  = totallen - ichdr1.usedbytes;
1810 		space += count * sizeof(xfs_attr_leaf_entry_t);
1811 
1812 		/*
1813 		 * leaf1 is the destination, compact it if it looks tight.
1814 		 */
1815 		max  = ichdr1.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1816 		max -= ichdr1.count * sizeof(xfs_attr_leaf_entry_t);
1817 		if (space > max)
1818 			xfs_attr3_leaf_compact(args, &ichdr1, blk1->bp);
1819 
1820 		/*
1821 		 * Move low entries from leaf2 to high end of leaf1.
1822 		 */
1823 		xfs_attr3_leaf_moveents(args, leaf2, &ichdr2, 0, leaf1, &ichdr1,
1824 					ichdr1.count, count);
1825 	}
1826 
1827 	xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf1, &ichdr1);
1828 	xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf2, &ichdr2);
1829 	xfs_trans_log_buf(args->trans, blk1->bp, 0, args->geo->blksize - 1);
1830 	xfs_trans_log_buf(args->trans, blk2->bp, 0, args->geo->blksize - 1);
1831 
1832 	/*
1833 	 * Copy out last hashval in each block for B-tree code.
1834 	 */
1835 	entries1 = xfs_attr3_leaf_entryp(leaf1);
1836 	entries2 = xfs_attr3_leaf_entryp(leaf2);
1837 	blk1->hashval = be32_to_cpu(entries1[ichdr1.count - 1].hashval);
1838 	blk2->hashval = be32_to_cpu(entries2[ichdr2.count - 1].hashval);
1839 
1840 	/*
1841 	 * Adjust the expected index for insertion.
1842 	 * NOTE: this code depends on the (current) situation that the
1843 	 * second block was originally empty.
1844 	 *
1845 	 * If the insertion point moved to the 2nd block, we must adjust
1846 	 * the index.  We must also track the entry just following the
1847 	 * new entry for use in an "atomic rename" operation, that entry
1848 	 * is always the "old" entry and the "new" entry is what we are
1849 	 * inserting.  The index/blkno fields refer to the "old" entry,
1850 	 * while the index2/blkno2 fields refer to the "new" entry.
1851 	 */
1852 	if (blk1->index > ichdr1.count) {
1853 		ASSERT(state->inleaf == 0);
1854 		blk2->index = blk1->index - ichdr1.count;
1855 		args->index = args->index2 = blk2->index;
1856 		args->blkno = args->blkno2 = blk2->blkno;
1857 	} else if (blk1->index == ichdr1.count) {
1858 		if (state->inleaf) {
1859 			args->index = blk1->index;
1860 			args->blkno = blk1->blkno;
1861 			args->index2 = 0;
1862 			args->blkno2 = blk2->blkno;
1863 		} else {
1864 			/*
1865 			 * On a double leaf split, the original attr location
1866 			 * is already stored in blkno2/index2, so don't
1867 			 * overwrite it overwise we corrupt the tree.
1868 			 */
1869 			blk2->index = blk1->index - ichdr1.count;
1870 			args->index = blk2->index;
1871 			args->blkno = blk2->blkno;
1872 			if (!state->extravalid) {
1873 				/*
1874 				 * set the new attr location to match the old
1875 				 * one and let the higher level split code
1876 				 * decide where in the leaf to place it.
1877 				 */
1878 				args->index2 = blk2->index;
1879 				args->blkno2 = blk2->blkno;
1880 			}
1881 		}
1882 	} else {
1883 		ASSERT(state->inleaf == 1);
1884 		args->index = args->index2 = blk1->index;
1885 		args->blkno = args->blkno2 = blk1->blkno;
1886 	}
1887 }
1888 
1889 /*
1890  * Examine entries until we reduce the absolute difference in
1891  * byte usage between the two blocks to a minimum.
1892  * GROT: Is this really necessary?  With other than a 512 byte blocksize,
1893  * GROT: there will always be enough room in either block for a new entry.
1894  * GROT: Do a double-split for this case?
1895  */
1896 STATIC int
1897 xfs_attr3_leaf_figure_balance(
1898 	struct xfs_da_state		*state,
1899 	struct xfs_da_state_blk		*blk1,
1900 	struct xfs_attr3_icleaf_hdr	*ichdr1,
1901 	struct xfs_da_state_blk		*blk2,
1902 	struct xfs_attr3_icleaf_hdr	*ichdr2,
1903 	int				*countarg,
1904 	int				*usedbytesarg)
1905 {
1906 	struct xfs_attr_leafblock	*leaf1 = blk1->bp->b_addr;
1907 	struct xfs_attr_leafblock	*leaf2 = blk2->bp->b_addr;
1908 	struct xfs_attr_leaf_entry	*entry;
1909 	int				count;
1910 	int				max;
1911 	int				index;
1912 	int				totallen = 0;
1913 	int				half;
1914 	int				lastdelta;
1915 	int				foundit = 0;
1916 	int				tmp;
1917 
1918 	/*
1919 	 * Examine entries until we reduce the absolute difference in
1920 	 * byte usage between the two blocks to a minimum.
1921 	 */
1922 	max = ichdr1->count + ichdr2->count;
1923 	half = (max + 1) * sizeof(*entry);
1924 	half += ichdr1->usedbytes + ichdr2->usedbytes +
1925 			xfs_attr_leaf_newentsize(state->args, NULL);
1926 	half /= 2;
1927 	lastdelta = state->args->geo->blksize;
1928 	entry = xfs_attr3_leaf_entryp(leaf1);
1929 	for (count = index = 0; count < max; entry++, index++, count++) {
1930 
1931 #define XFS_ATTR_ABS(A)	(((A) < 0) ? -(A) : (A))
1932 		/*
1933 		 * The new entry is in the first block, account for it.
1934 		 */
1935 		if (count == blk1->index) {
1936 			tmp = totallen + sizeof(*entry) +
1937 				xfs_attr_leaf_newentsize(state->args, NULL);
1938 			if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1939 				break;
1940 			lastdelta = XFS_ATTR_ABS(half - tmp);
1941 			totallen = tmp;
1942 			foundit = 1;
1943 		}
1944 
1945 		/*
1946 		 * Wrap around into the second block if necessary.
1947 		 */
1948 		if (count == ichdr1->count) {
1949 			leaf1 = leaf2;
1950 			entry = xfs_attr3_leaf_entryp(leaf1);
1951 			index = 0;
1952 		}
1953 
1954 		/*
1955 		 * Figure out if next leaf entry would be too much.
1956 		 */
1957 		tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1958 									index);
1959 		if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1960 			break;
1961 		lastdelta = XFS_ATTR_ABS(half - tmp);
1962 		totallen = tmp;
1963 #undef XFS_ATTR_ABS
1964 	}
1965 
1966 	/*
1967 	 * Calculate the number of usedbytes that will end up in lower block.
1968 	 * If new entry not in lower block, fix up the count.
1969 	 */
1970 	totallen -= count * sizeof(*entry);
1971 	if (foundit) {
1972 		totallen -= sizeof(*entry) +
1973 				xfs_attr_leaf_newentsize(state->args, NULL);
1974 	}
1975 
1976 	*countarg = count;
1977 	*usedbytesarg = totallen;
1978 	return foundit;
1979 }
1980 
1981 /*========================================================================
1982  * Routines used for shrinking the Btree.
1983  *========================================================================*/
1984 
1985 /*
1986  * Check a leaf block and its neighbors to see if the block should be
1987  * collapsed into one or the other neighbor.  Always keep the block
1988  * with the smaller block number.
1989  * If the current block is over 50% full, don't try to join it, return 0.
1990  * If the block is empty, fill in the state structure and return 2.
1991  * If it can be collapsed, fill in the state structure and return 1.
1992  * If nothing can be done, return 0.
1993  *
1994  * GROT: allow for INCOMPLETE entries in calculation.
1995  */
1996 int
1997 xfs_attr3_leaf_toosmall(
1998 	struct xfs_da_state	*state,
1999 	int			*action)
2000 {
2001 	struct xfs_attr_leafblock *leaf;
2002 	struct xfs_da_state_blk	*blk;
2003 	struct xfs_attr3_icleaf_hdr ichdr;
2004 	struct xfs_buf		*bp;
2005 	xfs_dablk_t		blkno;
2006 	int			bytes;
2007 	int			forward;
2008 	int			error;
2009 	int			retval;
2010 	int			i;
2011 
2012 	trace_xfs_attr_leaf_toosmall(state->args);
2013 
2014 	/*
2015 	 * Check for the degenerate case of the block being over 50% full.
2016 	 * If so, it's not worth even looking to see if we might be able
2017 	 * to coalesce with a sibling.
2018 	 */
2019 	blk = &state->path.blk[ state->path.active-1 ];
2020 	leaf = blk->bp->b_addr;
2021 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr, leaf);
2022 	bytes = xfs_attr3_leaf_hdr_size(leaf) +
2023 		ichdr.count * sizeof(xfs_attr_leaf_entry_t) +
2024 		ichdr.usedbytes;
2025 	if (bytes > (state->args->geo->blksize >> 1)) {
2026 		*action = 0;	/* blk over 50%, don't try to join */
2027 		return 0;
2028 	}
2029 
2030 	/*
2031 	 * Check for the degenerate case of the block being empty.
2032 	 * If the block is empty, we'll simply delete it, no need to
2033 	 * coalesce it with a sibling block.  We choose (arbitrarily)
2034 	 * to merge with the forward block unless it is NULL.
2035 	 */
2036 	if (ichdr.count == 0) {
2037 		/*
2038 		 * Make altpath point to the block we want to keep and
2039 		 * path point to the block we want to drop (this one).
2040 		 */
2041 		forward = (ichdr.forw != 0);
2042 		memcpy(&state->altpath, &state->path, sizeof(state->path));
2043 		error = xfs_da3_path_shift(state, &state->altpath, forward,
2044 						 0, &retval);
2045 		if (error)
2046 			return error;
2047 		if (retval) {
2048 			*action = 0;
2049 		} else {
2050 			*action = 2;
2051 		}
2052 		return 0;
2053 	}
2054 
2055 	/*
2056 	 * Examine each sibling block to see if we can coalesce with
2057 	 * at least 25% free space to spare.  We need to figure out
2058 	 * whether to merge with the forward or the backward block.
2059 	 * We prefer coalescing with the lower numbered sibling so as
2060 	 * to shrink an attribute list over time.
2061 	 */
2062 	/* start with smaller blk num */
2063 	forward = ichdr.forw < ichdr.back;
2064 	for (i = 0; i < 2; forward = !forward, i++) {
2065 		struct xfs_attr3_icleaf_hdr ichdr2;
2066 		if (forward)
2067 			blkno = ichdr.forw;
2068 		else
2069 			blkno = ichdr.back;
2070 		if (blkno == 0)
2071 			continue;
2072 		error = xfs_attr3_leaf_read(state->args->trans, state->args->dp,
2073 					state->args->owner, blkno, &bp);
2074 		if (error)
2075 			return error;
2076 
2077 		xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, bp->b_addr);
2078 
2079 		bytes = state->args->geo->blksize -
2080 			(state->args->geo->blksize >> 2) -
2081 			ichdr.usedbytes - ichdr2.usedbytes -
2082 			((ichdr.count + ichdr2.count) *
2083 					sizeof(xfs_attr_leaf_entry_t)) -
2084 			xfs_attr3_leaf_hdr_size(leaf);
2085 
2086 		xfs_trans_brelse(state->args->trans, bp);
2087 		if (bytes >= 0)
2088 			break;	/* fits with at least 25% to spare */
2089 	}
2090 	if (i >= 2) {
2091 		*action = 0;
2092 		return 0;
2093 	}
2094 
2095 	/*
2096 	 * Make altpath point to the block we want to keep (the lower
2097 	 * numbered block) and path point to the block we want to drop.
2098 	 */
2099 	memcpy(&state->altpath, &state->path, sizeof(state->path));
2100 	if (blkno < blk->blkno) {
2101 		error = xfs_da3_path_shift(state, &state->altpath, forward,
2102 						 0, &retval);
2103 	} else {
2104 		error = xfs_da3_path_shift(state, &state->path, forward,
2105 						 0, &retval);
2106 	}
2107 	if (error)
2108 		return error;
2109 	if (retval) {
2110 		*action = 0;
2111 	} else {
2112 		*action = 1;
2113 	}
2114 	return 0;
2115 }
2116 
2117 /*
2118  * Remove a name from the leaf attribute list structure.
2119  *
2120  * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
2121  * If two leaves are 37% full, when combined they will leave 25% free.
2122  */
2123 int
2124 xfs_attr3_leaf_remove(
2125 	struct xfs_buf		*bp,
2126 	struct xfs_da_args	*args)
2127 {
2128 	struct xfs_attr_leafblock *leaf;
2129 	struct xfs_attr3_icleaf_hdr ichdr;
2130 	struct xfs_attr_leaf_entry *entry;
2131 	int			before;
2132 	int			after;
2133 	int			smallest;
2134 	int			entsize;
2135 	int			tablesize;
2136 	int			tmp;
2137 	int			i;
2138 
2139 	trace_xfs_attr_leaf_remove(args);
2140 
2141 	leaf = bp->b_addr;
2142 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2143 
2144 	ASSERT(ichdr.count > 0 && ichdr.count < args->geo->blksize / 8);
2145 	ASSERT(args->index >= 0 && args->index < ichdr.count);
2146 	ASSERT(ichdr.firstused >= ichdr.count * sizeof(*entry) +
2147 					xfs_attr3_leaf_hdr_size(leaf));
2148 
2149 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2150 
2151 	ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2152 	ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
2153 
2154 	/*
2155 	 * Scan through free region table:
2156 	 *    check for adjacency of free'd entry with an existing one,
2157 	 *    find smallest free region in case we need to replace it,
2158 	 *    adjust any map that borders the entry table,
2159 	 */
2160 	tablesize = ichdr.count * sizeof(xfs_attr_leaf_entry_t)
2161 					+ xfs_attr3_leaf_hdr_size(leaf);
2162 	tmp = ichdr.freemap[0].size;
2163 	before = after = -1;
2164 	smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
2165 	entsize = xfs_attr_leaf_entsize(leaf, args->index);
2166 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
2167 		ASSERT(ichdr.freemap[i].base < args->geo->blksize);
2168 		ASSERT(ichdr.freemap[i].size < args->geo->blksize);
2169 		if (ichdr.freemap[i].base == tablesize) {
2170 			ichdr.freemap[i].base -= sizeof(xfs_attr_leaf_entry_t);
2171 			ichdr.freemap[i].size += sizeof(xfs_attr_leaf_entry_t);
2172 		}
2173 
2174 		if (ichdr.freemap[i].base + ichdr.freemap[i].size ==
2175 				be16_to_cpu(entry->nameidx)) {
2176 			before = i;
2177 		} else if (ichdr.freemap[i].base ==
2178 				(be16_to_cpu(entry->nameidx) + entsize)) {
2179 			after = i;
2180 		} else if (ichdr.freemap[i].size < tmp) {
2181 			tmp = ichdr.freemap[i].size;
2182 			smallest = i;
2183 		}
2184 	}
2185 
2186 	/*
2187 	 * Coalesce adjacent freemap regions,
2188 	 * or replace the smallest region.
2189 	 */
2190 	if ((before >= 0) || (after >= 0)) {
2191 		if ((before >= 0) && (after >= 0)) {
2192 			ichdr.freemap[before].size += entsize;
2193 			ichdr.freemap[before].size += ichdr.freemap[after].size;
2194 			ichdr.freemap[after].base = 0;
2195 			ichdr.freemap[after].size = 0;
2196 		} else if (before >= 0) {
2197 			ichdr.freemap[before].size += entsize;
2198 		} else {
2199 			ichdr.freemap[after].base = be16_to_cpu(entry->nameidx);
2200 			ichdr.freemap[after].size += entsize;
2201 		}
2202 	} else {
2203 		/*
2204 		 * Replace smallest region (if it is smaller than free'd entry)
2205 		 */
2206 		if (ichdr.freemap[smallest].size < entsize) {
2207 			ichdr.freemap[smallest].base = be16_to_cpu(entry->nameidx);
2208 			ichdr.freemap[smallest].size = entsize;
2209 		}
2210 	}
2211 
2212 	/*
2213 	 * Did we remove the first entry?
2214 	 */
2215 	if (be16_to_cpu(entry->nameidx) == ichdr.firstused)
2216 		smallest = 1;
2217 	else
2218 		smallest = 0;
2219 
2220 	/*
2221 	 * Compress the remaining entries and zero out the removed stuff.
2222 	 */
2223 	memset(xfs_attr3_leaf_name(leaf, args->index), 0, entsize);
2224 	ichdr.usedbytes -= entsize;
2225 	xfs_trans_log_buf(args->trans, bp,
2226 	     XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
2227 				   entsize));
2228 
2229 	tmp = (ichdr.count - args->index) * sizeof(xfs_attr_leaf_entry_t);
2230 	memmove(entry, entry + 1, tmp);
2231 	ichdr.count--;
2232 	xfs_trans_log_buf(args->trans, bp,
2233 	    XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(xfs_attr_leaf_entry_t)));
2234 
2235 	entry = &xfs_attr3_leaf_entryp(leaf)[ichdr.count];
2236 	memset(entry, 0, sizeof(xfs_attr_leaf_entry_t));
2237 
2238 	/*
2239 	 * If we removed the first entry, re-find the first used byte
2240 	 * in the name area.  Note that if the entry was the "firstused",
2241 	 * then we don't have a "hole" in our block resulting from
2242 	 * removing the name.
2243 	 */
2244 	if (smallest) {
2245 		tmp = args->geo->blksize;
2246 		entry = xfs_attr3_leaf_entryp(leaf);
2247 		for (i = ichdr.count - 1; i >= 0; entry++, i--) {
2248 			ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2249 			ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
2250 
2251 			if (be16_to_cpu(entry->nameidx) < tmp)
2252 				tmp = be16_to_cpu(entry->nameidx);
2253 		}
2254 		ichdr.firstused = tmp;
2255 		ASSERT(ichdr.firstused != 0);
2256 	} else {
2257 		ichdr.holes = 1;	/* mark as needing compaction */
2258 	}
2259 	xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
2260 	xfs_trans_log_buf(args->trans, bp,
2261 			  XFS_DA_LOGRANGE(leaf, &leaf->hdr,
2262 					  xfs_attr3_leaf_hdr_size(leaf)));
2263 
2264 	/*
2265 	 * Check if leaf is less than 50% full, caller may want to
2266 	 * "join" the leaf with a sibling if so.
2267 	 */
2268 	tmp = ichdr.usedbytes + xfs_attr3_leaf_hdr_size(leaf) +
2269 	      ichdr.count * sizeof(xfs_attr_leaf_entry_t);
2270 
2271 	return tmp < args->geo->magicpct; /* leaf is < 37% full */
2272 }
2273 
2274 /*
2275  * Move all the attribute list entries from drop_leaf into save_leaf.
2276  */
2277 void
2278 xfs_attr3_leaf_unbalance(
2279 	struct xfs_da_state	*state,
2280 	struct xfs_da_state_blk	*drop_blk,
2281 	struct xfs_da_state_blk	*save_blk)
2282 {
2283 	struct xfs_attr_leafblock *drop_leaf = drop_blk->bp->b_addr;
2284 	struct xfs_attr_leafblock *save_leaf = save_blk->bp->b_addr;
2285 	struct xfs_attr3_icleaf_hdr drophdr;
2286 	struct xfs_attr3_icleaf_hdr savehdr;
2287 	struct xfs_attr_leaf_entry *entry;
2288 
2289 	trace_xfs_attr_leaf_unbalance(state->args);
2290 
2291 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &drophdr, drop_leaf);
2292 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &savehdr, save_leaf);
2293 	entry = xfs_attr3_leaf_entryp(drop_leaf);
2294 
2295 	/*
2296 	 * Save last hashval from dying block for later Btree fixup.
2297 	 */
2298 	drop_blk->hashval = be32_to_cpu(entry[drophdr.count - 1].hashval);
2299 
2300 	/*
2301 	 * Check if we need a temp buffer, or can we do it in place.
2302 	 * Note that we don't check "leaf" for holes because we will
2303 	 * always be dropping it, toosmall() decided that for us already.
2304 	 */
2305 	if (savehdr.holes == 0) {
2306 		/*
2307 		 * dest leaf has no holes, so we add there.  May need
2308 		 * to make some room in the entry array.
2309 		 */
2310 		if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2311 					 drop_blk->bp, &drophdr)) {
2312 			xfs_attr3_leaf_moveents(state->args,
2313 						drop_leaf, &drophdr, 0,
2314 						save_leaf, &savehdr, 0,
2315 						drophdr.count);
2316 		} else {
2317 			xfs_attr3_leaf_moveents(state->args,
2318 						drop_leaf, &drophdr, 0,
2319 						save_leaf, &savehdr,
2320 						savehdr.count, drophdr.count);
2321 		}
2322 	} else {
2323 		/*
2324 		 * Destination has holes, so we make a temporary copy
2325 		 * of the leaf and add them both to that.
2326 		 */
2327 		struct xfs_attr_leafblock *tmp_leaf;
2328 		struct xfs_attr3_icleaf_hdr tmphdr;
2329 
2330 		tmp_leaf = kvzalloc(state->args->geo->blksize,
2331 				GFP_KERNEL | __GFP_NOFAIL);
2332 
2333 		/*
2334 		 * Copy the header into the temp leaf so that all the stuff
2335 		 * not in the incore header is present and gets copied back in
2336 		 * once we've moved all the entries.
2337 		 */
2338 		memcpy(tmp_leaf, save_leaf, xfs_attr3_leaf_hdr_size(save_leaf));
2339 
2340 		memset(&tmphdr, 0, sizeof(tmphdr));
2341 		tmphdr.magic = savehdr.magic;
2342 		tmphdr.forw = savehdr.forw;
2343 		tmphdr.back = savehdr.back;
2344 		tmphdr.firstused = state->args->geo->blksize;
2345 
2346 		/* write the header to the temp buffer to initialise it */
2347 		xfs_attr3_leaf_hdr_to_disk(state->args->geo, tmp_leaf, &tmphdr);
2348 
2349 		if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2350 					 drop_blk->bp, &drophdr)) {
2351 			xfs_attr3_leaf_moveents(state->args,
2352 						drop_leaf, &drophdr, 0,
2353 						tmp_leaf, &tmphdr, 0,
2354 						drophdr.count);
2355 			xfs_attr3_leaf_moveents(state->args,
2356 						save_leaf, &savehdr, 0,
2357 						tmp_leaf, &tmphdr, tmphdr.count,
2358 						savehdr.count);
2359 		} else {
2360 			xfs_attr3_leaf_moveents(state->args,
2361 						save_leaf, &savehdr, 0,
2362 						tmp_leaf, &tmphdr, 0,
2363 						savehdr.count);
2364 			xfs_attr3_leaf_moveents(state->args,
2365 						drop_leaf, &drophdr, 0,
2366 						tmp_leaf, &tmphdr, tmphdr.count,
2367 						drophdr.count);
2368 		}
2369 		memcpy(save_leaf, tmp_leaf, state->args->geo->blksize);
2370 		savehdr = tmphdr; /* struct copy */
2371 		kvfree(tmp_leaf);
2372 	}
2373 
2374 	xfs_attr3_leaf_hdr_to_disk(state->args->geo, save_leaf, &savehdr);
2375 	xfs_trans_log_buf(state->args->trans, save_blk->bp, 0,
2376 					   state->args->geo->blksize - 1);
2377 
2378 	/*
2379 	 * Copy out last hashval in each block for B-tree code.
2380 	 */
2381 	entry = xfs_attr3_leaf_entryp(save_leaf);
2382 	save_blk->hashval = be32_to_cpu(entry[savehdr.count - 1].hashval);
2383 }
2384 
2385 /*========================================================================
2386  * Routines used for finding things in the Btree.
2387  *========================================================================*/
2388 
2389 /*
2390  * Look up a name in a leaf attribute list structure.
2391  * This is the internal routine, it uses the caller's buffer.
2392  *
2393  * Note that duplicate keys are allowed, but only check within the
2394  * current leaf node.  The Btree code must check in adjacent leaf nodes.
2395  *
2396  * Return in args->index the index into the entry[] array of either
2397  * the found entry, or where the entry should have been (insert before
2398  * that entry).
2399  *
2400  * Don't change the args->value unless we find the attribute.
2401  */
2402 int
2403 xfs_attr3_leaf_lookup_int(
2404 	struct xfs_buf		*bp,
2405 	struct xfs_da_args	*args)
2406 {
2407 	struct xfs_attr_leafblock *leaf;
2408 	struct xfs_attr3_icleaf_hdr ichdr;
2409 	struct xfs_attr_leaf_entry *entry;
2410 	struct xfs_attr_leaf_entry *entries;
2411 	struct xfs_attr_leaf_name_local *name_loc;
2412 	struct xfs_attr_leaf_name_remote *name_rmt;
2413 	xfs_dahash_t		hashval;
2414 	int			probe;
2415 	int			span;
2416 
2417 	trace_xfs_attr_leaf_lookup(args);
2418 
2419 	leaf = bp->b_addr;
2420 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2421 	entries = xfs_attr3_leaf_entryp(leaf);
2422 	if (ichdr.count >= args->geo->blksize / 8) {
2423 		xfs_buf_mark_corrupt(bp);
2424 		xfs_da_mark_sick(args);
2425 		return -EFSCORRUPTED;
2426 	}
2427 
2428 	/*
2429 	 * Binary search.  (note: small blocks will skip this loop)
2430 	 */
2431 	hashval = args->hashval;
2432 	probe = span = ichdr.count / 2;
2433 	for (entry = &entries[probe]; span > 4; entry = &entries[probe]) {
2434 		span /= 2;
2435 		if (be32_to_cpu(entry->hashval) < hashval)
2436 			probe += span;
2437 		else if (be32_to_cpu(entry->hashval) > hashval)
2438 			probe -= span;
2439 		else
2440 			break;
2441 	}
2442 	if (!(probe >= 0 && (!ichdr.count || probe < ichdr.count))) {
2443 		xfs_buf_mark_corrupt(bp);
2444 		xfs_da_mark_sick(args);
2445 		return -EFSCORRUPTED;
2446 	}
2447 	if (!(span <= 4 || be32_to_cpu(entry->hashval) == hashval)) {
2448 		xfs_buf_mark_corrupt(bp);
2449 		xfs_da_mark_sick(args);
2450 		return -EFSCORRUPTED;
2451 	}
2452 
2453 	/*
2454 	 * Since we may have duplicate hashval's, find the first matching
2455 	 * hashval in the leaf.
2456 	 */
2457 	while (probe > 0 && be32_to_cpu(entry->hashval) >= hashval) {
2458 		entry--;
2459 		probe--;
2460 	}
2461 	while (probe < ichdr.count &&
2462 	       be32_to_cpu(entry->hashval) < hashval) {
2463 		entry++;
2464 		probe++;
2465 	}
2466 	if (probe == ichdr.count || be32_to_cpu(entry->hashval) != hashval) {
2467 		args->index = probe;
2468 		return -ENOATTR;
2469 	}
2470 
2471 	/*
2472 	 * Duplicate keys may be present, so search all of them for a match.
2473 	 */
2474 	for (; probe < ichdr.count && (be32_to_cpu(entry->hashval) == hashval);
2475 			entry++, probe++) {
2476 /*
2477  * GROT: Add code to remove incomplete entries.
2478  */
2479 		if (entry->flags & XFS_ATTR_LOCAL) {
2480 			name_loc = xfs_attr3_leaf_name_local(leaf, probe);
2481 			if (!xfs_attr_match(args, entry->flags,
2482 					name_loc->nameval, name_loc->namelen,
2483 					&name_loc->nameval[name_loc->namelen],
2484 					be16_to_cpu(name_loc->valuelen)))
2485 				continue;
2486 			args->index = probe;
2487 			return -EEXIST;
2488 		} else {
2489 			unsigned int	valuelen;
2490 
2491 			name_rmt = xfs_attr3_leaf_name_remote(leaf, probe);
2492 			valuelen = be32_to_cpu(name_rmt->valuelen);
2493 			if (!xfs_attr_match(args, entry->flags, name_rmt->name,
2494 					name_rmt->namelen, NULL, valuelen))
2495 				continue;
2496 			args->index = probe;
2497 			args->rmtvaluelen = valuelen;
2498 			args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2499 			args->rmtblkcnt = xfs_attr3_rmt_blocks(
2500 							args->dp->i_mount,
2501 							args->rmtvaluelen);
2502 			return -EEXIST;
2503 		}
2504 	}
2505 	args->index = probe;
2506 	return -ENOATTR;
2507 }
2508 
2509 /*
2510  * Get the value associated with an attribute name from a leaf attribute
2511  * list structure.
2512  *
2513  * If args->valuelen is zero, only the length needs to be returned.  Unlike a
2514  * lookup, we only return an error if the attribute does not exist or we can't
2515  * retrieve the value.
2516  */
2517 int
2518 xfs_attr3_leaf_getvalue(
2519 	struct xfs_buf		*bp,
2520 	struct xfs_da_args	*args)
2521 {
2522 	struct xfs_attr_leafblock *leaf;
2523 	struct xfs_attr3_icleaf_hdr ichdr;
2524 	struct xfs_attr_leaf_entry *entry;
2525 	struct xfs_attr_leaf_name_local *name_loc;
2526 	struct xfs_attr_leaf_name_remote *name_rmt;
2527 
2528 	leaf = bp->b_addr;
2529 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2530 	ASSERT(ichdr.count < args->geo->blksize / 8);
2531 	ASSERT(args->index < ichdr.count);
2532 
2533 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2534 	if (entry->flags & XFS_ATTR_LOCAL) {
2535 		name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2536 		ASSERT(name_loc->namelen == args->namelen);
2537 		ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2538 		return xfs_attr_copy_value(args,
2539 					&name_loc->nameval[args->namelen],
2540 					be16_to_cpu(name_loc->valuelen));
2541 	}
2542 
2543 	name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2544 	ASSERT(name_rmt->namelen == args->namelen);
2545 	ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2546 	args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2547 	args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2548 	args->rmtblkcnt = xfs_attr3_rmt_blocks(args->dp->i_mount,
2549 					       args->rmtvaluelen);
2550 	return xfs_attr_copy_value(args, NULL, args->rmtvaluelen);
2551 }
2552 
2553 /*========================================================================
2554  * Utility routines.
2555  *========================================================================*/
2556 
2557 /*
2558  * Move the indicated entries from one leaf to another.
2559  * NOTE: this routine modifies both source and destination leaves.
2560  */
2561 /*ARGSUSED*/
2562 STATIC void
2563 xfs_attr3_leaf_moveents(
2564 	struct xfs_da_args		*args,
2565 	struct xfs_attr_leafblock	*leaf_s,
2566 	struct xfs_attr3_icleaf_hdr	*ichdr_s,
2567 	int				start_s,
2568 	struct xfs_attr_leafblock	*leaf_d,
2569 	struct xfs_attr3_icleaf_hdr	*ichdr_d,
2570 	int				start_d,
2571 	int				count)
2572 {
2573 	struct xfs_attr_leaf_entry	*entry_s;
2574 	struct xfs_attr_leaf_entry	*entry_d;
2575 	int				desti;
2576 	int				tmp;
2577 	int				i;
2578 
2579 	/*
2580 	 * Check for nothing to do.
2581 	 */
2582 	if (count == 0)
2583 		return;
2584 
2585 	/*
2586 	 * Set up environment.
2587 	 */
2588 	ASSERT(ichdr_s->magic == XFS_ATTR_LEAF_MAGIC ||
2589 	       ichdr_s->magic == XFS_ATTR3_LEAF_MAGIC);
2590 	ASSERT(ichdr_s->magic == ichdr_d->magic);
2591 	ASSERT(ichdr_s->count > 0 && ichdr_s->count < args->geo->blksize / 8);
2592 	ASSERT(ichdr_s->firstused >= (ichdr_s->count * sizeof(*entry_s))
2593 					+ xfs_attr3_leaf_hdr_size(leaf_s));
2594 	ASSERT(ichdr_d->count < args->geo->blksize / 8);
2595 	ASSERT(ichdr_d->firstused >= (ichdr_d->count * sizeof(*entry_d))
2596 					+ xfs_attr3_leaf_hdr_size(leaf_d));
2597 
2598 	ASSERT(start_s < ichdr_s->count);
2599 	ASSERT(start_d <= ichdr_d->count);
2600 	ASSERT(count <= ichdr_s->count);
2601 
2602 
2603 	/*
2604 	 * Move the entries in the destination leaf up to make a hole?
2605 	 */
2606 	if (start_d < ichdr_d->count) {
2607 		tmp  = ichdr_d->count - start_d;
2608 		tmp *= sizeof(xfs_attr_leaf_entry_t);
2609 		entry_s = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2610 		entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d + count];
2611 		memmove(entry_d, entry_s, tmp);
2612 	}
2613 
2614 	/*
2615 	 * Copy all entry's in the same (sorted) order,
2616 	 * but allocate attribute info packed and in sequence.
2617 	 */
2618 	entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2619 	entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2620 	desti = start_d;
2621 	for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2622 		ASSERT(be16_to_cpu(entry_s->nameidx) >= ichdr_s->firstused);
2623 		tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2624 #ifdef GROT
2625 		/*
2626 		 * Code to drop INCOMPLETE entries.  Difficult to use as we
2627 		 * may also need to change the insertion index.  Code turned
2628 		 * off for 6.2, should be revisited later.
2629 		 */
2630 		if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2631 			memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2632 			ichdr_s->usedbytes -= tmp;
2633 			ichdr_s->count -= 1;
2634 			entry_d--;	/* to compensate for ++ in loop hdr */
2635 			desti--;
2636 			if ((start_s + i) < offset)
2637 				result++;	/* insertion index adjustment */
2638 		} else {
2639 #endif /* GROT */
2640 			ichdr_d->firstused -= tmp;
2641 			/* both on-disk, don't endian flip twice */
2642 			entry_d->hashval = entry_s->hashval;
2643 			entry_d->nameidx = cpu_to_be16(ichdr_d->firstused);
2644 			entry_d->flags = entry_s->flags;
2645 			ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2646 							<= args->geo->blksize);
2647 			memmove(xfs_attr3_leaf_name(leaf_d, desti),
2648 				xfs_attr3_leaf_name(leaf_s, start_s + i), tmp);
2649 			ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2650 							<= args->geo->blksize);
2651 			memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2652 			ichdr_s->usedbytes -= tmp;
2653 			ichdr_d->usedbytes += tmp;
2654 			ichdr_s->count -= 1;
2655 			ichdr_d->count += 1;
2656 			tmp = ichdr_d->count * sizeof(xfs_attr_leaf_entry_t)
2657 					+ xfs_attr3_leaf_hdr_size(leaf_d);
2658 			ASSERT(ichdr_d->firstused >= tmp);
2659 #ifdef GROT
2660 		}
2661 #endif /* GROT */
2662 	}
2663 
2664 	/*
2665 	 * Zero out the entries we just copied.
2666 	 */
2667 	if (start_s == ichdr_s->count) {
2668 		tmp = count * sizeof(xfs_attr_leaf_entry_t);
2669 		entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2670 		ASSERT(((char *)entry_s + tmp) <=
2671 		       ((char *)leaf_s + args->geo->blksize));
2672 		memset(entry_s, 0, tmp);
2673 	} else {
2674 		/*
2675 		 * Move the remaining entries down to fill the hole,
2676 		 * then zero the entries at the top.
2677 		 */
2678 		tmp  = (ichdr_s->count - count) * sizeof(xfs_attr_leaf_entry_t);
2679 		entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s + count];
2680 		entry_d = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2681 		memmove(entry_d, entry_s, tmp);
2682 
2683 		tmp = count * sizeof(xfs_attr_leaf_entry_t);
2684 		entry_s = &xfs_attr3_leaf_entryp(leaf_s)[ichdr_s->count];
2685 		ASSERT(((char *)entry_s + tmp) <=
2686 		       ((char *)leaf_s + args->geo->blksize));
2687 		memset(entry_s, 0, tmp);
2688 	}
2689 
2690 	/*
2691 	 * Fill in the freemap information
2692 	 */
2693 	ichdr_d->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_d);
2694 	ichdr_d->freemap[0].base += ichdr_d->count * sizeof(xfs_attr_leaf_entry_t);
2695 	ichdr_d->freemap[0].size = ichdr_d->firstused - ichdr_d->freemap[0].base;
2696 	ichdr_d->freemap[1].base = 0;
2697 	ichdr_d->freemap[2].base = 0;
2698 	ichdr_d->freemap[1].size = 0;
2699 	ichdr_d->freemap[2].size = 0;
2700 	ichdr_s->holes = 1;	/* leaf may not be compact */
2701 }
2702 
2703 /*
2704  * Pick up the last hashvalue from a leaf block.
2705  */
2706 xfs_dahash_t
2707 xfs_attr_leaf_lasthash(
2708 	struct xfs_buf	*bp,
2709 	int		*count)
2710 {
2711 	struct xfs_attr3_icleaf_hdr ichdr;
2712 	struct xfs_attr_leaf_entry *entries;
2713 	struct xfs_mount *mp = bp->b_mount;
2714 
2715 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, bp->b_addr);
2716 	entries = xfs_attr3_leaf_entryp(bp->b_addr);
2717 	if (count)
2718 		*count = ichdr.count;
2719 	if (!ichdr.count)
2720 		return 0;
2721 	return be32_to_cpu(entries[ichdr.count - 1].hashval);
2722 }
2723 
2724 /*
2725  * Calculate the number of bytes used to store the indicated attribute
2726  * (whether local or remote only calculate bytes in this block).
2727  */
2728 STATIC int
2729 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2730 {
2731 	struct xfs_attr_leaf_entry *entries;
2732 	xfs_attr_leaf_name_local_t *name_loc;
2733 	xfs_attr_leaf_name_remote_t *name_rmt;
2734 	int size;
2735 
2736 	entries = xfs_attr3_leaf_entryp(leaf);
2737 	if (entries[index].flags & XFS_ATTR_LOCAL) {
2738 		name_loc = xfs_attr3_leaf_name_local(leaf, index);
2739 		size = xfs_attr_leaf_entsize_local(name_loc->namelen,
2740 						   be16_to_cpu(name_loc->valuelen));
2741 	} else {
2742 		name_rmt = xfs_attr3_leaf_name_remote(leaf, index);
2743 		size = xfs_attr_leaf_entsize_remote(name_rmt->namelen);
2744 	}
2745 	return size;
2746 }
2747 
2748 /*
2749  * Calculate the number of bytes that would be required to store the new
2750  * attribute (whether local or remote only calculate bytes in this block).
2751  * This routine decides as a side effect whether the attribute will be
2752  * a "local" or a "remote" attribute.
2753  */
2754 int
2755 xfs_attr_leaf_newentsize(
2756 	struct xfs_da_args	*args,
2757 	int			*local)
2758 {
2759 	int			size;
2760 
2761 	size = xfs_attr_leaf_entsize_local(args->namelen, args->valuelen);
2762 	if (size < xfs_attr_leaf_entsize_local_max(args->geo->blksize)) {
2763 		if (local)
2764 			*local = 1;
2765 		return size;
2766 	}
2767 	if (local)
2768 		*local = 0;
2769 	return xfs_attr_leaf_entsize_remote(args->namelen);
2770 }
2771 
2772 
2773 /*========================================================================
2774  * Manage the INCOMPLETE flag in a leaf entry
2775  *========================================================================*/
2776 
2777 /*
2778  * Clear the INCOMPLETE flag on an entry in a leaf block.
2779  */
2780 int
2781 xfs_attr3_leaf_clearflag(
2782 	struct xfs_da_args	*args)
2783 {
2784 	struct xfs_attr_leafblock *leaf;
2785 	struct xfs_attr_leaf_entry *entry;
2786 	struct xfs_attr_leaf_name_remote *name_rmt;
2787 	struct xfs_buf		*bp;
2788 	int			error;
2789 #ifdef DEBUG
2790 	struct xfs_attr3_icleaf_hdr ichdr;
2791 	xfs_attr_leaf_name_local_t *name_loc;
2792 	int namelen;
2793 	char *name;
2794 #endif /* DEBUG */
2795 
2796 	trace_xfs_attr_leaf_clearflag(args);
2797 	/*
2798 	 * Set up the operation.
2799 	 */
2800 	error = xfs_attr3_leaf_read(args->trans, args->dp, args->owner,
2801 			args->blkno, &bp);
2802 	if (error)
2803 		return error;
2804 
2805 	leaf = bp->b_addr;
2806 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2807 	ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2808 
2809 #ifdef DEBUG
2810 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2811 	ASSERT(args->index < ichdr.count);
2812 	ASSERT(args->index >= 0);
2813 
2814 	if (entry->flags & XFS_ATTR_LOCAL) {
2815 		name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2816 		namelen = name_loc->namelen;
2817 		name = (char *)name_loc->nameval;
2818 	} else {
2819 		name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2820 		namelen = name_rmt->namelen;
2821 		name = (char *)name_rmt->name;
2822 	}
2823 	ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2824 	ASSERT(namelen == args->namelen);
2825 	ASSERT(memcmp(name, args->name, namelen) == 0);
2826 #endif /* DEBUG */
2827 
2828 	entry->flags &= ~XFS_ATTR_INCOMPLETE;
2829 	xfs_trans_log_buf(args->trans, bp,
2830 			 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2831 
2832 	if (args->rmtblkno) {
2833 		ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2834 		name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2835 		name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2836 		name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2837 		xfs_trans_log_buf(args->trans, bp,
2838 			 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2839 	}
2840 
2841 	return 0;
2842 }
2843 
2844 /*
2845  * Set the INCOMPLETE flag on an entry in a leaf block.
2846  */
2847 int
2848 xfs_attr3_leaf_setflag(
2849 	struct xfs_da_args	*args)
2850 {
2851 	struct xfs_attr_leafblock *leaf;
2852 	struct xfs_attr_leaf_entry *entry;
2853 	struct xfs_attr_leaf_name_remote *name_rmt;
2854 	struct xfs_buf		*bp;
2855 	int error;
2856 #ifdef DEBUG
2857 	struct xfs_attr3_icleaf_hdr ichdr;
2858 #endif
2859 
2860 	trace_xfs_attr_leaf_setflag(args);
2861 
2862 	/*
2863 	 * Set up the operation.
2864 	 */
2865 	error = xfs_attr3_leaf_read(args->trans, args->dp, args->owner,
2866 			args->blkno, &bp);
2867 	if (error)
2868 		return error;
2869 
2870 	leaf = bp->b_addr;
2871 #ifdef DEBUG
2872 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2873 	ASSERT(args->index < ichdr.count);
2874 	ASSERT(args->index >= 0);
2875 #endif
2876 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2877 
2878 	ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2879 	entry->flags |= XFS_ATTR_INCOMPLETE;
2880 	xfs_trans_log_buf(args->trans, bp,
2881 			XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2882 	if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2883 		name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2884 		name_rmt->valueblk = 0;
2885 		name_rmt->valuelen = 0;
2886 		xfs_trans_log_buf(args->trans, bp,
2887 			 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2888 	}
2889 
2890 	return 0;
2891 }
2892 
2893 /*
2894  * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2895  * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2896  * entry given by args->blkno2/index2.
2897  *
2898  * Note that they could be in different blocks, or in the same block.
2899  */
2900 int
2901 xfs_attr3_leaf_flipflags(
2902 	struct xfs_da_args	*args)
2903 {
2904 	struct xfs_attr_leafblock *leaf1;
2905 	struct xfs_attr_leafblock *leaf2;
2906 	struct xfs_attr_leaf_entry *entry1;
2907 	struct xfs_attr_leaf_entry *entry2;
2908 	struct xfs_attr_leaf_name_remote *name_rmt;
2909 	struct xfs_buf		*bp1;
2910 	struct xfs_buf		*bp2;
2911 	int error;
2912 #ifdef DEBUG
2913 	struct xfs_attr3_icleaf_hdr ichdr1;
2914 	struct xfs_attr3_icleaf_hdr ichdr2;
2915 	xfs_attr_leaf_name_local_t *name_loc;
2916 	int namelen1, namelen2;
2917 	char *name1, *name2;
2918 #endif /* DEBUG */
2919 
2920 	trace_xfs_attr_leaf_flipflags(args);
2921 
2922 	/*
2923 	 * Read the block containing the "old" attr
2924 	 */
2925 	error = xfs_attr3_leaf_read(args->trans, args->dp, args->owner,
2926 			args->blkno, &bp1);
2927 	if (error)
2928 		return error;
2929 
2930 	/*
2931 	 * Read the block containing the "new" attr, if it is different
2932 	 */
2933 	if (args->blkno2 != args->blkno) {
2934 		error = xfs_attr3_leaf_read(args->trans, args->dp, args->owner,
2935 				args->blkno2, &bp2);
2936 		if (error)
2937 			return error;
2938 	} else {
2939 		bp2 = bp1;
2940 	}
2941 
2942 	leaf1 = bp1->b_addr;
2943 	entry1 = &xfs_attr3_leaf_entryp(leaf1)[args->index];
2944 
2945 	leaf2 = bp2->b_addr;
2946 	entry2 = &xfs_attr3_leaf_entryp(leaf2)[args->index2];
2947 
2948 #ifdef DEBUG
2949 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr1, leaf1);
2950 	ASSERT(args->index < ichdr1.count);
2951 	ASSERT(args->index >= 0);
2952 
2953 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr2, leaf2);
2954 	ASSERT(args->index2 < ichdr2.count);
2955 	ASSERT(args->index2 >= 0);
2956 
2957 	if (entry1->flags & XFS_ATTR_LOCAL) {
2958 		name_loc = xfs_attr3_leaf_name_local(leaf1, args->index);
2959 		namelen1 = name_loc->namelen;
2960 		name1 = (char *)name_loc->nameval;
2961 	} else {
2962 		name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2963 		namelen1 = name_rmt->namelen;
2964 		name1 = (char *)name_rmt->name;
2965 	}
2966 	if (entry2->flags & XFS_ATTR_LOCAL) {
2967 		name_loc = xfs_attr3_leaf_name_local(leaf2, args->index2);
2968 		namelen2 = name_loc->namelen;
2969 		name2 = (char *)name_loc->nameval;
2970 	} else {
2971 		name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2972 		namelen2 = name_rmt->namelen;
2973 		name2 = (char *)name_rmt->name;
2974 	}
2975 	ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2976 	ASSERT(namelen1 == namelen2);
2977 	ASSERT(memcmp(name1, name2, namelen1) == 0);
2978 #endif /* DEBUG */
2979 
2980 	ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2981 	ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2982 
2983 	entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2984 	xfs_trans_log_buf(args->trans, bp1,
2985 			  XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2986 	if (args->rmtblkno) {
2987 		ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2988 		name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2989 		name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2990 		name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2991 		xfs_trans_log_buf(args->trans, bp1,
2992 			 XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2993 	}
2994 
2995 	entry2->flags |= XFS_ATTR_INCOMPLETE;
2996 	xfs_trans_log_buf(args->trans, bp2,
2997 			  XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2998 	if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
2999 		name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
3000 		name_rmt->valueblk = 0;
3001 		name_rmt->valuelen = 0;
3002 		xfs_trans_log_buf(args->trans, bp2,
3003 			 XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
3004 	}
3005 
3006 	return 0;
3007 }
3008