xref: /linux/fs/xfs/libxfs/xfs_attr_leaf.c (revision 266aa3b4812e97942a8ce5c7aafa7da059f7b5b8)
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_BROOT_SPACE(mp, 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 = kmalloc(args->geo->blksize, GFP_KERNEL | __GFP_NOFAIL);
1142 	if (!tmpbuffer)
1143 		return -ENOMEM;
1144 
1145 	memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1146 
1147 	leaf = (xfs_attr_leafblock_t *)tmpbuffer;
1148 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1149 	entry = xfs_attr3_leaf_entryp(leaf);
1150 
1151 	/* XXX (dgc): buffer is about to be marked stale - why zero it? */
1152 	memset(bp->b_addr, 0, args->geo->blksize);
1153 
1154 	/*
1155 	 * Clean out the prior contents of the attribute list.
1156 	 */
1157 	error = xfs_da_shrink_inode(args, 0, bp);
1158 	if (error)
1159 		goto out;
1160 
1161 	if (forkoff == -1) {
1162 		/*
1163 		 * Don't remove the attr fork if this operation is the first
1164 		 * part of a attr replace operations. We're going to add a new
1165 		 * attr immediately, so we need to keep the attr fork around in
1166 		 * this case.
1167 		 */
1168 		if (!(args->op_flags & XFS_DA_OP_REPLACE)) {
1169 			ASSERT(xfs_has_attr2(dp->i_mount));
1170 			ASSERT(dp->i_df.if_format != XFS_DINODE_FMT_BTREE);
1171 			xfs_attr_fork_remove(dp, args->trans);
1172 		}
1173 		goto out;
1174 	}
1175 
1176 	xfs_attr_shortform_create(args);
1177 
1178 	/*
1179 	 * Copy the attributes
1180 	 */
1181 	memset((char *)&nargs, 0, sizeof(nargs));
1182 	nargs.geo = args->geo;
1183 	nargs.dp = dp;
1184 	nargs.total = args->total;
1185 	nargs.whichfork = XFS_ATTR_FORK;
1186 	nargs.trans = args->trans;
1187 	nargs.op_flags = XFS_DA_OP_OKNOENT;
1188 	nargs.owner = args->owner;
1189 
1190 	for (i = 0; i < ichdr.count; entry++, i++) {
1191 		if (entry->flags & XFS_ATTR_INCOMPLETE)
1192 			continue;	/* don't copy partial entries */
1193 		if (!entry->nameidx)
1194 			continue;
1195 		ASSERT(entry->flags & XFS_ATTR_LOCAL);
1196 		name_loc = xfs_attr3_leaf_name_local(leaf, i);
1197 		nargs.name = name_loc->nameval;
1198 		nargs.namelen = name_loc->namelen;
1199 		nargs.value = &name_loc->nameval[nargs.namelen];
1200 		nargs.valuelen = be16_to_cpu(name_loc->valuelen);
1201 		nargs.hashval = be32_to_cpu(entry->hashval);
1202 		nargs.attr_filter = entry->flags & XFS_ATTR_NSP_ONDISK_MASK;
1203 		xfs_attr_shortform_add(&nargs, forkoff);
1204 	}
1205 	error = 0;
1206 
1207 out:
1208 	kfree(tmpbuffer);
1209 	return error;
1210 }
1211 
1212 /*
1213  * Convert from using a single leaf to a root node and a leaf.
1214  */
1215 int
1216 xfs_attr3_leaf_to_node(
1217 	struct xfs_da_args	*args)
1218 {
1219 	struct xfs_attr_leafblock *leaf;
1220 	struct xfs_attr3_icleaf_hdr icleafhdr;
1221 	struct xfs_attr_leaf_entry *entries;
1222 	struct xfs_da3_icnode_hdr icnodehdr;
1223 	struct xfs_da_intnode	*node;
1224 	struct xfs_inode	*dp = args->dp;
1225 	struct xfs_mount	*mp = dp->i_mount;
1226 	struct xfs_buf		*bp1 = NULL;
1227 	struct xfs_buf		*bp2 = NULL;
1228 	xfs_dablk_t		blkno;
1229 	int			error;
1230 
1231 	trace_xfs_attr_leaf_to_node(args);
1232 
1233 	if (XFS_TEST_ERROR(false, mp, XFS_ERRTAG_ATTR_LEAF_TO_NODE)) {
1234 		error = -EIO;
1235 		goto out;
1236 	}
1237 
1238 	error = xfs_da_grow_inode(args, &blkno);
1239 	if (error)
1240 		goto out;
1241 	error = xfs_attr3_leaf_read(args->trans, dp, args->owner, 0, &bp1);
1242 	if (error)
1243 		goto out;
1244 
1245 	error = xfs_da_get_buf(args->trans, dp, blkno, &bp2, XFS_ATTR_FORK);
1246 	if (error)
1247 		goto out;
1248 
1249 	/*
1250 	 * Copy leaf to new buffer and log it.
1251 	 */
1252 	xfs_da_buf_copy(bp2, bp1, args->geo->blksize);
1253 	xfs_trans_log_buf(args->trans, bp2, 0, args->geo->blksize - 1);
1254 
1255 	/*
1256 	 * Set up the new root node.
1257 	 */
1258 	error = xfs_da3_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK);
1259 	if (error)
1260 		goto out;
1261 	node = bp1->b_addr;
1262 	xfs_da3_node_hdr_from_disk(mp, &icnodehdr, node);
1263 
1264 	leaf = bp2->b_addr;
1265 	xfs_attr3_leaf_hdr_from_disk(args->geo, &icleafhdr, leaf);
1266 	entries = xfs_attr3_leaf_entryp(leaf);
1267 
1268 	/* both on-disk, don't endian-flip twice */
1269 	icnodehdr.btree[0].hashval = entries[icleafhdr.count - 1].hashval;
1270 	icnodehdr.btree[0].before = cpu_to_be32(blkno);
1271 	icnodehdr.count = 1;
1272 	xfs_da3_node_hdr_to_disk(dp->i_mount, node, &icnodehdr);
1273 	xfs_trans_log_buf(args->trans, bp1, 0, args->geo->blksize - 1);
1274 	error = 0;
1275 out:
1276 	return error;
1277 }
1278 
1279 /*========================================================================
1280  * Routines used for growing the Btree.
1281  *========================================================================*/
1282 
1283 /*
1284  * Create the initial contents of a leaf attribute list
1285  * or a leaf in a node attribute list.
1286  */
1287 STATIC int
1288 xfs_attr3_leaf_create(
1289 	struct xfs_da_args	*args,
1290 	xfs_dablk_t		blkno,
1291 	struct xfs_buf		**bpp)
1292 {
1293 	struct xfs_attr_leafblock *leaf;
1294 	struct xfs_attr3_icleaf_hdr ichdr;
1295 	struct xfs_inode	*dp = args->dp;
1296 	struct xfs_mount	*mp = dp->i_mount;
1297 	struct xfs_buf		*bp;
1298 	int			error;
1299 
1300 	trace_xfs_attr_leaf_create(args);
1301 
1302 	error = xfs_da_get_buf(args->trans, args->dp, blkno, &bp,
1303 					    XFS_ATTR_FORK);
1304 	if (error)
1305 		return error;
1306 	bp->b_ops = &xfs_attr3_leaf_buf_ops;
1307 	xfs_trans_buf_set_type(args->trans, bp, XFS_BLFT_ATTR_LEAF_BUF);
1308 	leaf = bp->b_addr;
1309 	memset(leaf, 0, args->geo->blksize);
1310 
1311 	memset(&ichdr, 0, sizeof(ichdr));
1312 	ichdr.firstused = args->geo->blksize;
1313 
1314 	if (xfs_has_crc(mp)) {
1315 		struct xfs_da3_blkinfo *hdr3 = bp->b_addr;
1316 
1317 		ichdr.magic = XFS_ATTR3_LEAF_MAGIC;
1318 
1319 		hdr3->blkno = cpu_to_be64(xfs_buf_daddr(bp));
1320 		hdr3->owner = cpu_to_be64(args->owner);
1321 		uuid_copy(&hdr3->uuid, &mp->m_sb.sb_meta_uuid);
1322 
1323 		ichdr.freemap[0].base = sizeof(struct xfs_attr3_leaf_hdr);
1324 	} else {
1325 		ichdr.magic = XFS_ATTR_LEAF_MAGIC;
1326 		ichdr.freemap[0].base = sizeof(struct xfs_attr_leaf_hdr);
1327 	}
1328 	ichdr.freemap[0].size = ichdr.firstused - ichdr.freemap[0].base;
1329 
1330 	xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1331 	xfs_trans_log_buf(args->trans, bp, 0, args->geo->blksize - 1);
1332 
1333 	*bpp = bp;
1334 	return 0;
1335 }
1336 
1337 /*
1338  * Split the leaf node, rebalance, then add the new entry.
1339  */
1340 int
1341 xfs_attr3_leaf_split(
1342 	struct xfs_da_state	*state,
1343 	struct xfs_da_state_blk	*oldblk,
1344 	struct xfs_da_state_blk	*newblk)
1345 {
1346 	xfs_dablk_t blkno;
1347 	int error;
1348 
1349 	trace_xfs_attr_leaf_split(state->args);
1350 
1351 	/*
1352 	 * Allocate space for a new leaf node.
1353 	 */
1354 	ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
1355 	error = xfs_da_grow_inode(state->args, &blkno);
1356 	if (error)
1357 		return error;
1358 	error = xfs_attr3_leaf_create(state->args, blkno, &newblk->bp);
1359 	if (error)
1360 		return error;
1361 	newblk->blkno = blkno;
1362 	newblk->magic = XFS_ATTR_LEAF_MAGIC;
1363 
1364 	/*
1365 	 * Rebalance the entries across the two leaves.
1366 	 * NOTE: rebalance() currently depends on the 2nd block being empty.
1367 	 */
1368 	xfs_attr3_leaf_rebalance(state, oldblk, newblk);
1369 	error = xfs_da3_blk_link(state, oldblk, newblk);
1370 	if (error)
1371 		return error;
1372 
1373 	/*
1374 	 * Save info on "old" attribute for "atomic rename" ops, leaf_add()
1375 	 * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
1376 	 * "new" attrs info.  Will need the "old" info to remove it later.
1377 	 *
1378 	 * Insert the "new" entry in the correct block.
1379 	 */
1380 	if (state->inleaf) {
1381 		trace_xfs_attr_leaf_add_old(state->args);
1382 		error = xfs_attr3_leaf_add(oldblk->bp, state->args);
1383 	} else {
1384 		trace_xfs_attr_leaf_add_new(state->args);
1385 		error = xfs_attr3_leaf_add(newblk->bp, state->args);
1386 	}
1387 
1388 	/*
1389 	 * Update last hashval in each block since we added the name.
1390 	 */
1391 	oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
1392 	newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
1393 	return error;
1394 }
1395 
1396 /*
1397  * Add a name to the leaf attribute list structure.
1398  */
1399 int
1400 xfs_attr3_leaf_add(
1401 	struct xfs_buf		*bp,
1402 	struct xfs_da_args	*args)
1403 {
1404 	struct xfs_attr_leafblock *leaf;
1405 	struct xfs_attr3_icleaf_hdr ichdr;
1406 	int			tablesize;
1407 	int			entsize;
1408 	int			sum;
1409 	int			tmp;
1410 	int			i;
1411 
1412 	trace_xfs_attr_leaf_add(args);
1413 
1414 	leaf = bp->b_addr;
1415 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1416 	ASSERT(args->index >= 0 && args->index <= ichdr.count);
1417 	entsize = xfs_attr_leaf_newentsize(args, NULL);
1418 
1419 	/*
1420 	 * Search through freemap for first-fit on new name length.
1421 	 * (may need to figure in size of entry struct too)
1422 	 */
1423 	tablesize = (ichdr.count + 1) * sizeof(xfs_attr_leaf_entry_t)
1424 					+ xfs_attr3_leaf_hdr_size(leaf);
1425 	for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE - 1; i >= 0; i--) {
1426 		if (tablesize > ichdr.firstused) {
1427 			sum += ichdr.freemap[i].size;
1428 			continue;
1429 		}
1430 		if (!ichdr.freemap[i].size)
1431 			continue;	/* no space in this map */
1432 		tmp = entsize;
1433 		if (ichdr.freemap[i].base < ichdr.firstused)
1434 			tmp += sizeof(xfs_attr_leaf_entry_t);
1435 		if (ichdr.freemap[i].size >= tmp) {
1436 			tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, i);
1437 			goto out_log_hdr;
1438 		}
1439 		sum += ichdr.freemap[i].size;
1440 	}
1441 
1442 	/*
1443 	 * If there are no holes in the address space of the block,
1444 	 * and we don't have enough freespace, then compaction will do us
1445 	 * no good and we should just give up.
1446 	 */
1447 	if (!ichdr.holes && sum < entsize)
1448 		return -ENOSPC;
1449 
1450 	/*
1451 	 * Compact the entries to coalesce free space.
1452 	 * This may change the hdr->count via dropping INCOMPLETE entries.
1453 	 */
1454 	xfs_attr3_leaf_compact(args, &ichdr, bp);
1455 
1456 	/*
1457 	 * After compaction, the block is guaranteed to have only one
1458 	 * free region, in freemap[0].  If it is not big enough, give up.
1459 	 */
1460 	if (ichdr.freemap[0].size < (entsize + sizeof(xfs_attr_leaf_entry_t))) {
1461 		tmp = -ENOSPC;
1462 		goto out_log_hdr;
1463 	}
1464 
1465 	tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, 0);
1466 
1467 out_log_hdr:
1468 	xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1469 	xfs_trans_log_buf(args->trans, bp,
1470 		XFS_DA_LOGRANGE(leaf, &leaf->hdr,
1471 				xfs_attr3_leaf_hdr_size(leaf)));
1472 	return tmp;
1473 }
1474 
1475 /*
1476  * Add a name to a leaf attribute list structure.
1477  */
1478 STATIC int
1479 xfs_attr3_leaf_add_work(
1480 	struct xfs_buf		*bp,
1481 	struct xfs_attr3_icleaf_hdr *ichdr,
1482 	struct xfs_da_args	*args,
1483 	int			mapindex)
1484 {
1485 	struct xfs_attr_leafblock *leaf;
1486 	struct xfs_attr_leaf_entry *entry;
1487 	struct xfs_attr_leaf_name_local *name_loc;
1488 	struct xfs_attr_leaf_name_remote *name_rmt;
1489 	struct xfs_mount	*mp;
1490 	int			tmp;
1491 	int			i;
1492 
1493 	trace_xfs_attr_leaf_add_work(args);
1494 
1495 	leaf = bp->b_addr;
1496 	ASSERT(mapindex >= 0 && mapindex < XFS_ATTR_LEAF_MAPSIZE);
1497 	ASSERT(args->index >= 0 && args->index <= ichdr->count);
1498 
1499 	/*
1500 	 * Force open some space in the entry array and fill it in.
1501 	 */
1502 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1503 	if (args->index < ichdr->count) {
1504 		tmp  = ichdr->count - args->index;
1505 		tmp *= sizeof(xfs_attr_leaf_entry_t);
1506 		memmove(entry + 1, entry, tmp);
1507 		xfs_trans_log_buf(args->trans, bp,
1508 		    XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1509 	}
1510 	ichdr->count++;
1511 
1512 	/*
1513 	 * Allocate space for the new string (at the end of the run).
1514 	 */
1515 	mp = args->trans->t_mountp;
1516 	ASSERT(ichdr->freemap[mapindex].base < args->geo->blksize);
1517 	ASSERT((ichdr->freemap[mapindex].base & 0x3) == 0);
1518 	ASSERT(ichdr->freemap[mapindex].size >=
1519 		xfs_attr_leaf_newentsize(args, NULL));
1520 	ASSERT(ichdr->freemap[mapindex].size < args->geo->blksize);
1521 	ASSERT((ichdr->freemap[mapindex].size & 0x3) == 0);
1522 
1523 	ichdr->freemap[mapindex].size -= xfs_attr_leaf_newentsize(args, &tmp);
1524 
1525 	entry->nameidx = cpu_to_be16(ichdr->freemap[mapindex].base +
1526 				     ichdr->freemap[mapindex].size);
1527 	entry->hashval = cpu_to_be32(args->hashval);
1528 	entry->flags = args->attr_filter;
1529 	if (tmp)
1530 		entry->flags |= XFS_ATTR_LOCAL;
1531 	if (args->op_flags & XFS_DA_OP_REPLACE) {
1532 		if (!(args->op_flags & XFS_DA_OP_LOGGED))
1533 			entry->flags |= XFS_ATTR_INCOMPLETE;
1534 		if ((args->blkno2 == args->blkno) &&
1535 		    (args->index2 <= args->index)) {
1536 			args->index2++;
1537 		}
1538 	}
1539 	xfs_trans_log_buf(args->trans, bp,
1540 			  XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1541 	ASSERT((args->index == 0) ||
1542 	       (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1543 	ASSERT((args->index == ichdr->count - 1) ||
1544 	       (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1545 
1546 	/*
1547 	 * For "remote" attribute values, simply note that we need to
1548 	 * allocate space for the "remote" value.  We can't actually
1549 	 * allocate the extents in this transaction, and we can't decide
1550 	 * which blocks they should be as we might allocate more blocks
1551 	 * as part of this transaction (a split operation for example).
1552 	 */
1553 	if (entry->flags & XFS_ATTR_LOCAL) {
1554 		name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
1555 		name_loc->namelen = args->namelen;
1556 		name_loc->valuelen = cpu_to_be16(args->valuelen);
1557 		memcpy((char *)name_loc->nameval, args->name, args->namelen);
1558 		memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1559 				   be16_to_cpu(name_loc->valuelen));
1560 	} else {
1561 		name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
1562 		name_rmt->namelen = args->namelen;
1563 		memcpy((char *)name_rmt->name, args->name, args->namelen);
1564 		entry->flags |= XFS_ATTR_INCOMPLETE;
1565 		/* just in case */
1566 		name_rmt->valuelen = 0;
1567 		name_rmt->valueblk = 0;
1568 		args->rmtblkno = 1;
1569 		args->rmtblkcnt = xfs_attr3_rmt_blocks(mp, args->valuelen);
1570 		args->rmtvaluelen = args->valuelen;
1571 	}
1572 	xfs_trans_log_buf(args->trans, bp,
1573 	     XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1574 				   xfs_attr_leaf_entsize(leaf, args->index)));
1575 
1576 	/*
1577 	 * Update the control info for this leaf node
1578 	 */
1579 	if (be16_to_cpu(entry->nameidx) < ichdr->firstused)
1580 		ichdr->firstused = be16_to_cpu(entry->nameidx);
1581 
1582 	ASSERT(ichdr->firstused >= ichdr->count * sizeof(xfs_attr_leaf_entry_t)
1583 					+ xfs_attr3_leaf_hdr_size(leaf));
1584 	tmp = (ichdr->count - 1) * sizeof(xfs_attr_leaf_entry_t)
1585 					+ xfs_attr3_leaf_hdr_size(leaf);
1586 
1587 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1588 		if (ichdr->freemap[i].base == tmp) {
1589 			ichdr->freemap[i].base += sizeof(xfs_attr_leaf_entry_t);
1590 			ichdr->freemap[i].size -=
1591 				min_t(uint16_t, ichdr->freemap[i].size,
1592 						sizeof(xfs_attr_leaf_entry_t));
1593 		}
1594 	}
1595 	ichdr->usedbytes += xfs_attr_leaf_entsize(leaf, args->index);
1596 	return 0;
1597 }
1598 
1599 /*
1600  * Garbage collect a leaf attribute list block by copying it to a new buffer.
1601  */
1602 STATIC void
1603 xfs_attr3_leaf_compact(
1604 	struct xfs_da_args	*args,
1605 	struct xfs_attr3_icleaf_hdr *ichdr_dst,
1606 	struct xfs_buf		*bp)
1607 {
1608 	struct xfs_attr_leafblock *leaf_src;
1609 	struct xfs_attr_leafblock *leaf_dst;
1610 	struct xfs_attr3_icleaf_hdr ichdr_src;
1611 	struct xfs_trans	*trans = args->trans;
1612 	char			*tmpbuffer;
1613 
1614 	trace_xfs_attr_leaf_compact(args);
1615 
1616 	tmpbuffer = kmalloc(args->geo->blksize, GFP_KERNEL | __GFP_NOFAIL);
1617 	memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1618 	memset(bp->b_addr, 0, args->geo->blksize);
1619 	leaf_src = (xfs_attr_leafblock_t *)tmpbuffer;
1620 	leaf_dst = bp->b_addr;
1621 
1622 	/*
1623 	 * Copy the on-disk header back into the destination buffer to ensure
1624 	 * all the information in the header that is not part of the incore
1625 	 * header structure is preserved.
1626 	 */
1627 	memcpy(bp->b_addr, tmpbuffer, xfs_attr3_leaf_hdr_size(leaf_src));
1628 
1629 	/* Initialise the incore headers */
1630 	ichdr_src = *ichdr_dst;	/* struct copy */
1631 	ichdr_dst->firstused = args->geo->blksize;
1632 	ichdr_dst->usedbytes = 0;
1633 	ichdr_dst->count = 0;
1634 	ichdr_dst->holes = 0;
1635 	ichdr_dst->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_src);
1636 	ichdr_dst->freemap[0].size = ichdr_dst->firstused -
1637 						ichdr_dst->freemap[0].base;
1638 
1639 	/* write the header back to initialise the underlying buffer */
1640 	xfs_attr3_leaf_hdr_to_disk(args->geo, leaf_dst, ichdr_dst);
1641 
1642 	/*
1643 	 * Copy all entry's in the same (sorted) order,
1644 	 * but allocate name/value pairs packed and in sequence.
1645 	 */
1646 	xfs_attr3_leaf_moveents(args, leaf_src, &ichdr_src, 0,
1647 				leaf_dst, ichdr_dst, 0, ichdr_src.count);
1648 	/*
1649 	 * this logs the entire buffer, but the caller must write the header
1650 	 * back to the buffer when it is finished modifying it.
1651 	 */
1652 	xfs_trans_log_buf(trans, bp, 0, args->geo->blksize - 1);
1653 
1654 	kfree(tmpbuffer);
1655 }
1656 
1657 /*
1658  * Compare two leaf blocks "order".
1659  * Return 0 unless leaf2 should go before leaf1.
1660  */
1661 static int
1662 xfs_attr3_leaf_order(
1663 	struct xfs_buf	*leaf1_bp,
1664 	struct xfs_attr3_icleaf_hdr *leaf1hdr,
1665 	struct xfs_buf	*leaf2_bp,
1666 	struct xfs_attr3_icleaf_hdr *leaf2hdr)
1667 {
1668 	struct xfs_attr_leaf_entry *entries1;
1669 	struct xfs_attr_leaf_entry *entries2;
1670 
1671 	entries1 = xfs_attr3_leaf_entryp(leaf1_bp->b_addr);
1672 	entries2 = xfs_attr3_leaf_entryp(leaf2_bp->b_addr);
1673 	if (leaf1hdr->count > 0 && leaf2hdr->count > 0 &&
1674 	    ((be32_to_cpu(entries2[0].hashval) <
1675 	      be32_to_cpu(entries1[0].hashval)) ||
1676 	     (be32_to_cpu(entries2[leaf2hdr->count - 1].hashval) <
1677 	      be32_to_cpu(entries1[leaf1hdr->count - 1].hashval)))) {
1678 		return 1;
1679 	}
1680 	return 0;
1681 }
1682 
1683 int
1684 xfs_attr_leaf_order(
1685 	struct xfs_buf	*leaf1_bp,
1686 	struct xfs_buf	*leaf2_bp)
1687 {
1688 	struct xfs_attr3_icleaf_hdr ichdr1;
1689 	struct xfs_attr3_icleaf_hdr ichdr2;
1690 	struct xfs_mount *mp = leaf1_bp->b_mount;
1691 
1692 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr1, leaf1_bp->b_addr);
1693 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr2, leaf2_bp->b_addr);
1694 	return xfs_attr3_leaf_order(leaf1_bp, &ichdr1, leaf2_bp, &ichdr2);
1695 }
1696 
1697 /*
1698  * Redistribute the attribute list entries between two leaf nodes,
1699  * taking into account the size of the new entry.
1700  *
1701  * NOTE: if new block is empty, then it will get the upper half of the
1702  * old block.  At present, all (one) callers pass in an empty second block.
1703  *
1704  * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1705  * to match what it is doing in splitting the attribute leaf block.  Those
1706  * values are used in "atomic rename" operations on attributes.  Note that
1707  * the "new" and "old" values can end up in different blocks.
1708  */
1709 STATIC void
1710 xfs_attr3_leaf_rebalance(
1711 	struct xfs_da_state	*state,
1712 	struct xfs_da_state_blk	*blk1,
1713 	struct xfs_da_state_blk	*blk2)
1714 {
1715 	struct xfs_da_args	*args;
1716 	struct xfs_attr_leafblock *leaf1;
1717 	struct xfs_attr_leafblock *leaf2;
1718 	struct xfs_attr3_icleaf_hdr ichdr1;
1719 	struct xfs_attr3_icleaf_hdr ichdr2;
1720 	struct xfs_attr_leaf_entry *entries1;
1721 	struct xfs_attr_leaf_entry *entries2;
1722 	int			count;
1723 	int			totallen;
1724 	int			max;
1725 	int			space;
1726 	int			swap;
1727 
1728 	/*
1729 	 * Set up environment.
1730 	 */
1731 	ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1732 	ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1733 	leaf1 = blk1->bp->b_addr;
1734 	leaf2 = blk2->bp->b_addr;
1735 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr1, leaf1);
1736 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, leaf2);
1737 	ASSERT(ichdr2.count == 0);
1738 	args = state->args;
1739 
1740 	trace_xfs_attr_leaf_rebalance(args);
1741 
1742 	/*
1743 	 * Check ordering of blocks, reverse if it makes things simpler.
1744 	 *
1745 	 * NOTE: Given that all (current) callers pass in an empty
1746 	 * second block, this code should never set "swap".
1747 	 */
1748 	swap = 0;
1749 	if (xfs_attr3_leaf_order(blk1->bp, &ichdr1, blk2->bp, &ichdr2)) {
1750 		swap(blk1, blk2);
1751 
1752 		/* swap structures rather than reconverting them */
1753 		swap(ichdr1, ichdr2);
1754 
1755 		leaf1 = blk1->bp->b_addr;
1756 		leaf2 = blk2->bp->b_addr;
1757 		swap = 1;
1758 	}
1759 
1760 	/*
1761 	 * Examine entries until we reduce the absolute difference in
1762 	 * byte usage between the two blocks to a minimum.  Then get
1763 	 * the direction to copy and the number of elements to move.
1764 	 *
1765 	 * "inleaf" is true if the new entry should be inserted into blk1.
1766 	 * If "swap" is also true, then reverse the sense of "inleaf".
1767 	 */
1768 	state->inleaf = xfs_attr3_leaf_figure_balance(state, blk1, &ichdr1,
1769 						      blk2, &ichdr2,
1770 						      &count, &totallen);
1771 	if (swap)
1772 		state->inleaf = !state->inleaf;
1773 
1774 	/*
1775 	 * Move any entries required from leaf to leaf:
1776 	 */
1777 	if (count < ichdr1.count) {
1778 		/*
1779 		 * Figure the total bytes to be added to the destination leaf.
1780 		 */
1781 		/* number entries being moved */
1782 		count = ichdr1.count - count;
1783 		space  = ichdr1.usedbytes - totallen;
1784 		space += count * sizeof(xfs_attr_leaf_entry_t);
1785 
1786 		/*
1787 		 * leaf2 is the destination, compact it if it looks tight.
1788 		 */
1789 		max  = ichdr2.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1790 		max -= ichdr2.count * sizeof(xfs_attr_leaf_entry_t);
1791 		if (space > max)
1792 			xfs_attr3_leaf_compact(args, &ichdr2, blk2->bp);
1793 
1794 		/*
1795 		 * Move high entries from leaf1 to low end of leaf2.
1796 		 */
1797 		xfs_attr3_leaf_moveents(args, leaf1, &ichdr1,
1798 				ichdr1.count - count, leaf2, &ichdr2, 0, count);
1799 
1800 	} else if (count > ichdr1.count) {
1801 		/*
1802 		 * I assert that since all callers pass in an empty
1803 		 * second buffer, this code should never execute.
1804 		 */
1805 		ASSERT(0);
1806 
1807 		/*
1808 		 * Figure the total bytes to be added to the destination leaf.
1809 		 */
1810 		/* number entries being moved */
1811 		count -= ichdr1.count;
1812 		space  = totallen - ichdr1.usedbytes;
1813 		space += count * sizeof(xfs_attr_leaf_entry_t);
1814 
1815 		/*
1816 		 * leaf1 is the destination, compact it if it looks tight.
1817 		 */
1818 		max  = ichdr1.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1819 		max -= ichdr1.count * sizeof(xfs_attr_leaf_entry_t);
1820 		if (space > max)
1821 			xfs_attr3_leaf_compact(args, &ichdr1, blk1->bp);
1822 
1823 		/*
1824 		 * Move low entries from leaf2 to high end of leaf1.
1825 		 */
1826 		xfs_attr3_leaf_moveents(args, leaf2, &ichdr2, 0, leaf1, &ichdr1,
1827 					ichdr1.count, count);
1828 	}
1829 
1830 	xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf1, &ichdr1);
1831 	xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf2, &ichdr2);
1832 	xfs_trans_log_buf(args->trans, blk1->bp, 0, args->geo->blksize - 1);
1833 	xfs_trans_log_buf(args->trans, blk2->bp, 0, args->geo->blksize - 1);
1834 
1835 	/*
1836 	 * Copy out last hashval in each block for B-tree code.
1837 	 */
1838 	entries1 = xfs_attr3_leaf_entryp(leaf1);
1839 	entries2 = xfs_attr3_leaf_entryp(leaf2);
1840 	blk1->hashval = be32_to_cpu(entries1[ichdr1.count - 1].hashval);
1841 	blk2->hashval = be32_to_cpu(entries2[ichdr2.count - 1].hashval);
1842 
1843 	/*
1844 	 * Adjust the expected index for insertion.
1845 	 * NOTE: this code depends on the (current) situation that the
1846 	 * second block was originally empty.
1847 	 *
1848 	 * If the insertion point moved to the 2nd block, we must adjust
1849 	 * the index.  We must also track the entry just following the
1850 	 * new entry for use in an "atomic rename" operation, that entry
1851 	 * is always the "old" entry and the "new" entry is what we are
1852 	 * inserting.  The index/blkno fields refer to the "old" entry,
1853 	 * while the index2/blkno2 fields refer to the "new" entry.
1854 	 */
1855 	if (blk1->index > ichdr1.count) {
1856 		ASSERT(state->inleaf == 0);
1857 		blk2->index = blk1->index - ichdr1.count;
1858 		args->index = args->index2 = blk2->index;
1859 		args->blkno = args->blkno2 = blk2->blkno;
1860 	} else if (blk1->index == ichdr1.count) {
1861 		if (state->inleaf) {
1862 			args->index = blk1->index;
1863 			args->blkno = blk1->blkno;
1864 			args->index2 = 0;
1865 			args->blkno2 = blk2->blkno;
1866 		} else {
1867 			/*
1868 			 * On a double leaf split, the original attr location
1869 			 * is already stored in blkno2/index2, so don't
1870 			 * overwrite it overwise we corrupt the tree.
1871 			 */
1872 			blk2->index = blk1->index - ichdr1.count;
1873 			args->index = blk2->index;
1874 			args->blkno = blk2->blkno;
1875 			if (!state->extravalid) {
1876 				/*
1877 				 * set the new attr location to match the old
1878 				 * one and let the higher level split code
1879 				 * decide where in the leaf to place it.
1880 				 */
1881 				args->index2 = blk2->index;
1882 				args->blkno2 = blk2->blkno;
1883 			}
1884 		}
1885 	} else {
1886 		ASSERT(state->inleaf == 1);
1887 		args->index = args->index2 = blk1->index;
1888 		args->blkno = args->blkno2 = blk1->blkno;
1889 	}
1890 }
1891 
1892 /*
1893  * Examine entries until we reduce the absolute difference in
1894  * byte usage between the two blocks to a minimum.
1895  * GROT: Is this really necessary?  With other than a 512 byte blocksize,
1896  * GROT: there will always be enough room in either block for a new entry.
1897  * GROT: Do a double-split for this case?
1898  */
1899 STATIC int
1900 xfs_attr3_leaf_figure_balance(
1901 	struct xfs_da_state		*state,
1902 	struct xfs_da_state_blk		*blk1,
1903 	struct xfs_attr3_icleaf_hdr	*ichdr1,
1904 	struct xfs_da_state_blk		*blk2,
1905 	struct xfs_attr3_icleaf_hdr	*ichdr2,
1906 	int				*countarg,
1907 	int				*usedbytesarg)
1908 {
1909 	struct xfs_attr_leafblock	*leaf1 = blk1->bp->b_addr;
1910 	struct xfs_attr_leafblock	*leaf2 = blk2->bp->b_addr;
1911 	struct xfs_attr_leaf_entry	*entry;
1912 	int				count;
1913 	int				max;
1914 	int				index;
1915 	int				totallen = 0;
1916 	int				half;
1917 	int				lastdelta;
1918 	int				foundit = 0;
1919 	int				tmp;
1920 
1921 	/*
1922 	 * Examine entries until we reduce the absolute difference in
1923 	 * byte usage between the two blocks to a minimum.
1924 	 */
1925 	max = ichdr1->count + ichdr2->count;
1926 	half = (max + 1) * sizeof(*entry);
1927 	half += ichdr1->usedbytes + ichdr2->usedbytes +
1928 			xfs_attr_leaf_newentsize(state->args, NULL);
1929 	half /= 2;
1930 	lastdelta = state->args->geo->blksize;
1931 	entry = xfs_attr3_leaf_entryp(leaf1);
1932 	for (count = index = 0; count < max; entry++, index++, count++) {
1933 
1934 #define XFS_ATTR_ABS(A)	(((A) < 0) ? -(A) : (A))
1935 		/*
1936 		 * The new entry is in the first block, account for it.
1937 		 */
1938 		if (count == blk1->index) {
1939 			tmp = totallen + sizeof(*entry) +
1940 				xfs_attr_leaf_newentsize(state->args, NULL);
1941 			if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1942 				break;
1943 			lastdelta = XFS_ATTR_ABS(half - tmp);
1944 			totallen = tmp;
1945 			foundit = 1;
1946 		}
1947 
1948 		/*
1949 		 * Wrap around into the second block if necessary.
1950 		 */
1951 		if (count == ichdr1->count) {
1952 			leaf1 = leaf2;
1953 			entry = xfs_attr3_leaf_entryp(leaf1);
1954 			index = 0;
1955 		}
1956 
1957 		/*
1958 		 * Figure out if next leaf entry would be too much.
1959 		 */
1960 		tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1961 									index);
1962 		if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1963 			break;
1964 		lastdelta = XFS_ATTR_ABS(half - tmp);
1965 		totallen = tmp;
1966 #undef XFS_ATTR_ABS
1967 	}
1968 
1969 	/*
1970 	 * Calculate the number of usedbytes that will end up in lower block.
1971 	 * If new entry not in lower block, fix up the count.
1972 	 */
1973 	totallen -= count * sizeof(*entry);
1974 	if (foundit) {
1975 		totallen -= sizeof(*entry) +
1976 				xfs_attr_leaf_newentsize(state->args, NULL);
1977 	}
1978 
1979 	*countarg = count;
1980 	*usedbytesarg = totallen;
1981 	return foundit;
1982 }
1983 
1984 /*========================================================================
1985  * Routines used for shrinking the Btree.
1986  *========================================================================*/
1987 
1988 /*
1989  * Check a leaf block and its neighbors to see if the block should be
1990  * collapsed into one or the other neighbor.  Always keep the block
1991  * with the smaller block number.
1992  * If the current block is over 50% full, don't try to join it, return 0.
1993  * If the block is empty, fill in the state structure and return 2.
1994  * If it can be collapsed, fill in the state structure and return 1.
1995  * If nothing can be done, return 0.
1996  *
1997  * GROT: allow for INCOMPLETE entries in calculation.
1998  */
1999 int
2000 xfs_attr3_leaf_toosmall(
2001 	struct xfs_da_state	*state,
2002 	int			*action)
2003 {
2004 	struct xfs_attr_leafblock *leaf;
2005 	struct xfs_da_state_blk	*blk;
2006 	struct xfs_attr3_icleaf_hdr ichdr;
2007 	struct xfs_buf		*bp;
2008 	xfs_dablk_t		blkno;
2009 	int			bytes;
2010 	int			forward;
2011 	int			error;
2012 	int			retval;
2013 	int			i;
2014 
2015 	trace_xfs_attr_leaf_toosmall(state->args);
2016 
2017 	/*
2018 	 * Check for the degenerate case of the block being over 50% full.
2019 	 * If so, it's not worth even looking to see if we might be able
2020 	 * to coalesce with a sibling.
2021 	 */
2022 	blk = &state->path.blk[ state->path.active-1 ];
2023 	leaf = blk->bp->b_addr;
2024 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr, leaf);
2025 	bytes = xfs_attr3_leaf_hdr_size(leaf) +
2026 		ichdr.count * sizeof(xfs_attr_leaf_entry_t) +
2027 		ichdr.usedbytes;
2028 	if (bytes > (state->args->geo->blksize >> 1)) {
2029 		*action = 0;	/* blk over 50%, don't try to join */
2030 		return 0;
2031 	}
2032 
2033 	/*
2034 	 * Check for the degenerate case of the block being empty.
2035 	 * If the block is empty, we'll simply delete it, no need to
2036 	 * coalesce it with a sibling block.  We choose (arbitrarily)
2037 	 * to merge with the forward block unless it is NULL.
2038 	 */
2039 	if (ichdr.count == 0) {
2040 		/*
2041 		 * Make altpath point to the block we want to keep and
2042 		 * path point to the block we want to drop (this one).
2043 		 */
2044 		forward = (ichdr.forw != 0);
2045 		memcpy(&state->altpath, &state->path, sizeof(state->path));
2046 		error = xfs_da3_path_shift(state, &state->altpath, forward,
2047 						 0, &retval);
2048 		if (error)
2049 			return error;
2050 		if (retval) {
2051 			*action = 0;
2052 		} else {
2053 			*action = 2;
2054 		}
2055 		return 0;
2056 	}
2057 
2058 	/*
2059 	 * Examine each sibling block to see if we can coalesce with
2060 	 * at least 25% free space to spare.  We need to figure out
2061 	 * whether to merge with the forward or the backward block.
2062 	 * We prefer coalescing with the lower numbered sibling so as
2063 	 * to shrink an attribute list over time.
2064 	 */
2065 	/* start with smaller blk num */
2066 	forward = ichdr.forw < ichdr.back;
2067 	for (i = 0; i < 2; forward = !forward, i++) {
2068 		struct xfs_attr3_icleaf_hdr ichdr2;
2069 		if (forward)
2070 			blkno = ichdr.forw;
2071 		else
2072 			blkno = ichdr.back;
2073 		if (blkno == 0)
2074 			continue;
2075 		error = xfs_attr3_leaf_read(state->args->trans, state->args->dp,
2076 					state->args->owner, blkno, &bp);
2077 		if (error)
2078 			return error;
2079 
2080 		xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, bp->b_addr);
2081 
2082 		bytes = state->args->geo->blksize -
2083 			(state->args->geo->blksize >> 2) -
2084 			ichdr.usedbytes - ichdr2.usedbytes -
2085 			((ichdr.count + ichdr2.count) *
2086 					sizeof(xfs_attr_leaf_entry_t)) -
2087 			xfs_attr3_leaf_hdr_size(leaf);
2088 
2089 		xfs_trans_brelse(state->args->trans, bp);
2090 		if (bytes >= 0)
2091 			break;	/* fits with at least 25% to spare */
2092 	}
2093 	if (i >= 2) {
2094 		*action = 0;
2095 		return 0;
2096 	}
2097 
2098 	/*
2099 	 * Make altpath point to the block we want to keep (the lower
2100 	 * numbered block) and path point to the block we want to drop.
2101 	 */
2102 	memcpy(&state->altpath, &state->path, sizeof(state->path));
2103 	if (blkno < blk->blkno) {
2104 		error = xfs_da3_path_shift(state, &state->altpath, forward,
2105 						 0, &retval);
2106 	} else {
2107 		error = xfs_da3_path_shift(state, &state->path, forward,
2108 						 0, &retval);
2109 	}
2110 	if (error)
2111 		return error;
2112 	if (retval) {
2113 		*action = 0;
2114 	} else {
2115 		*action = 1;
2116 	}
2117 	return 0;
2118 }
2119 
2120 /*
2121  * Remove a name from the leaf attribute list structure.
2122  *
2123  * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
2124  * If two leaves are 37% full, when combined they will leave 25% free.
2125  */
2126 int
2127 xfs_attr3_leaf_remove(
2128 	struct xfs_buf		*bp,
2129 	struct xfs_da_args	*args)
2130 {
2131 	struct xfs_attr_leafblock *leaf;
2132 	struct xfs_attr3_icleaf_hdr ichdr;
2133 	struct xfs_attr_leaf_entry *entry;
2134 	int			before;
2135 	int			after;
2136 	int			smallest;
2137 	int			entsize;
2138 	int			tablesize;
2139 	int			tmp;
2140 	int			i;
2141 
2142 	trace_xfs_attr_leaf_remove(args);
2143 
2144 	leaf = bp->b_addr;
2145 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2146 
2147 	ASSERT(ichdr.count > 0 && ichdr.count < args->geo->blksize / 8);
2148 	ASSERT(args->index >= 0 && args->index < ichdr.count);
2149 	ASSERT(ichdr.firstused >= ichdr.count * sizeof(*entry) +
2150 					xfs_attr3_leaf_hdr_size(leaf));
2151 
2152 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2153 
2154 	ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2155 	ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
2156 
2157 	/*
2158 	 * Scan through free region table:
2159 	 *    check for adjacency of free'd entry with an existing one,
2160 	 *    find smallest free region in case we need to replace it,
2161 	 *    adjust any map that borders the entry table,
2162 	 */
2163 	tablesize = ichdr.count * sizeof(xfs_attr_leaf_entry_t)
2164 					+ xfs_attr3_leaf_hdr_size(leaf);
2165 	tmp = ichdr.freemap[0].size;
2166 	before = after = -1;
2167 	smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
2168 	entsize = xfs_attr_leaf_entsize(leaf, args->index);
2169 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
2170 		ASSERT(ichdr.freemap[i].base < args->geo->blksize);
2171 		ASSERT(ichdr.freemap[i].size < args->geo->blksize);
2172 		if (ichdr.freemap[i].base == tablesize) {
2173 			ichdr.freemap[i].base -= sizeof(xfs_attr_leaf_entry_t);
2174 			ichdr.freemap[i].size += sizeof(xfs_attr_leaf_entry_t);
2175 		}
2176 
2177 		if (ichdr.freemap[i].base + ichdr.freemap[i].size ==
2178 				be16_to_cpu(entry->nameidx)) {
2179 			before = i;
2180 		} else if (ichdr.freemap[i].base ==
2181 				(be16_to_cpu(entry->nameidx) + entsize)) {
2182 			after = i;
2183 		} else if (ichdr.freemap[i].size < tmp) {
2184 			tmp = ichdr.freemap[i].size;
2185 			smallest = i;
2186 		}
2187 	}
2188 
2189 	/*
2190 	 * Coalesce adjacent freemap regions,
2191 	 * or replace the smallest region.
2192 	 */
2193 	if ((before >= 0) || (after >= 0)) {
2194 		if ((before >= 0) && (after >= 0)) {
2195 			ichdr.freemap[before].size += entsize;
2196 			ichdr.freemap[before].size += ichdr.freemap[after].size;
2197 			ichdr.freemap[after].base = 0;
2198 			ichdr.freemap[after].size = 0;
2199 		} else if (before >= 0) {
2200 			ichdr.freemap[before].size += entsize;
2201 		} else {
2202 			ichdr.freemap[after].base = be16_to_cpu(entry->nameidx);
2203 			ichdr.freemap[after].size += entsize;
2204 		}
2205 	} else {
2206 		/*
2207 		 * Replace smallest region (if it is smaller than free'd entry)
2208 		 */
2209 		if (ichdr.freemap[smallest].size < entsize) {
2210 			ichdr.freemap[smallest].base = be16_to_cpu(entry->nameidx);
2211 			ichdr.freemap[smallest].size = entsize;
2212 		}
2213 	}
2214 
2215 	/*
2216 	 * Did we remove the first entry?
2217 	 */
2218 	if (be16_to_cpu(entry->nameidx) == ichdr.firstused)
2219 		smallest = 1;
2220 	else
2221 		smallest = 0;
2222 
2223 	/*
2224 	 * Compress the remaining entries and zero out the removed stuff.
2225 	 */
2226 	memset(xfs_attr3_leaf_name(leaf, args->index), 0, entsize);
2227 	ichdr.usedbytes -= entsize;
2228 	xfs_trans_log_buf(args->trans, bp,
2229 	     XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
2230 				   entsize));
2231 
2232 	tmp = (ichdr.count - args->index) * sizeof(xfs_attr_leaf_entry_t);
2233 	memmove(entry, entry + 1, tmp);
2234 	ichdr.count--;
2235 	xfs_trans_log_buf(args->trans, bp,
2236 	    XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(xfs_attr_leaf_entry_t)));
2237 
2238 	entry = &xfs_attr3_leaf_entryp(leaf)[ichdr.count];
2239 	memset(entry, 0, sizeof(xfs_attr_leaf_entry_t));
2240 
2241 	/*
2242 	 * If we removed the first entry, re-find the first used byte
2243 	 * in the name area.  Note that if the entry was the "firstused",
2244 	 * then we don't have a "hole" in our block resulting from
2245 	 * removing the name.
2246 	 */
2247 	if (smallest) {
2248 		tmp = args->geo->blksize;
2249 		entry = xfs_attr3_leaf_entryp(leaf);
2250 		for (i = ichdr.count - 1; i >= 0; entry++, i--) {
2251 			ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2252 			ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
2253 
2254 			if (be16_to_cpu(entry->nameidx) < tmp)
2255 				tmp = be16_to_cpu(entry->nameidx);
2256 		}
2257 		ichdr.firstused = tmp;
2258 		ASSERT(ichdr.firstused != 0);
2259 	} else {
2260 		ichdr.holes = 1;	/* mark as needing compaction */
2261 	}
2262 	xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
2263 	xfs_trans_log_buf(args->trans, bp,
2264 			  XFS_DA_LOGRANGE(leaf, &leaf->hdr,
2265 					  xfs_attr3_leaf_hdr_size(leaf)));
2266 
2267 	/*
2268 	 * Check if leaf is less than 50% full, caller may want to
2269 	 * "join" the leaf with a sibling if so.
2270 	 */
2271 	tmp = ichdr.usedbytes + xfs_attr3_leaf_hdr_size(leaf) +
2272 	      ichdr.count * sizeof(xfs_attr_leaf_entry_t);
2273 
2274 	return tmp < args->geo->magicpct; /* leaf is < 37% full */
2275 }
2276 
2277 /*
2278  * Move all the attribute list entries from drop_leaf into save_leaf.
2279  */
2280 void
2281 xfs_attr3_leaf_unbalance(
2282 	struct xfs_da_state	*state,
2283 	struct xfs_da_state_blk	*drop_blk,
2284 	struct xfs_da_state_blk	*save_blk)
2285 {
2286 	struct xfs_attr_leafblock *drop_leaf = drop_blk->bp->b_addr;
2287 	struct xfs_attr_leafblock *save_leaf = save_blk->bp->b_addr;
2288 	struct xfs_attr3_icleaf_hdr drophdr;
2289 	struct xfs_attr3_icleaf_hdr savehdr;
2290 	struct xfs_attr_leaf_entry *entry;
2291 
2292 	trace_xfs_attr_leaf_unbalance(state->args);
2293 
2294 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &drophdr, drop_leaf);
2295 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &savehdr, save_leaf);
2296 	entry = xfs_attr3_leaf_entryp(drop_leaf);
2297 
2298 	/*
2299 	 * Save last hashval from dying block for later Btree fixup.
2300 	 */
2301 	drop_blk->hashval = be32_to_cpu(entry[drophdr.count - 1].hashval);
2302 
2303 	/*
2304 	 * Check if we need a temp buffer, or can we do it in place.
2305 	 * Note that we don't check "leaf" for holes because we will
2306 	 * always be dropping it, toosmall() decided that for us already.
2307 	 */
2308 	if (savehdr.holes == 0) {
2309 		/*
2310 		 * dest leaf has no holes, so we add there.  May need
2311 		 * to make some room in the entry array.
2312 		 */
2313 		if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2314 					 drop_blk->bp, &drophdr)) {
2315 			xfs_attr3_leaf_moveents(state->args,
2316 						drop_leaf, &drophdr, 0,
2317 						save_leaf, &savehdr, 0,
2318 						drophdr.count);
2319 		} else {
2320 			xfs_attr3_leaf_moveents(state->args,
2321 						drop_leaf, &drophdr, 0,
2322 						save_leaf, &savehdr,
2323 						savehdr.count, drophdr.count);
2324 		}
2325 	} else {
2326 		/*
2327 		 * Destination has holes, so we make a temporary copy
2328 		 * of the leaf and add them both to that.
2329 		 */
2330 		struct xfs_attr_leafblock *tmp_leaf;
2331 		struct xfs_attr3_icleaf_hdr tmphdr;
2332 
2333 		tmp_leaf = kzalloc(state->args->geo->blksize,
2334 				GFP_KERNEL | __GFP_NOFAIL);
2335 
2336 		/*
2337 		 * Copy the header into the temp leaf so that all the stuff
2338 		 * not in the incore header is present and gets copied back in
2339 		 * once we've moved all the entries.
2340 		 */
2341 		memcpy(tmp_leaf, save_leaf, xfs_attr3_leaf_hdr_size(save_leaf));
2342 
2343 		memset(&tmphdr, 0, sizeof(tmphdr));
2344 		tmphdr.magic = savehdr.magic;
2345 		tmphdr.forw = savehdr.forw;
2346 		tmphdr.back = savehdr.back;
2347 		tmphdr.firstused = state->args->geo->blksize;
2348 
2349 		/* write the header to the temp buffer to initialise it */
2350 		xfs_attr3_leaf_hdr_to_disk(state->args->geo, tmp_leaf, &tmphdr);
2351 
2352 		if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2353 					 drop_blk->bp, &drophdr)) {
2354 			xfs_attr3_leaf_moveents(state->args,
2355 						drop_leaf, &drophdr, 0,
2356 						tmp_leaf, &tmphdr, 0,
2357 						drophdr.count);
2358 			xfs_attr3_leaf_moveents(state->args,
2359 						save_leaf, &savehdr, 0,
2360 						tmp_leaf, &tmphdr, tmphdr.count,
2361 						savehdr.count);
2362 		} else {
2363 			xfs_attr3_leaf_moveents(state->args,
2364 						save_leaf, &savehdr, 0,
2365 						tmp_leaf, &tmphdr, 0,
2366 						savehdr.count);
2367 			xfs_attr3_leaf_moveents(state->args,
2368 						drop_leaf, &drophdr, 0,
2369 						tmp_leaf, &tmphdr, tmphdr.count,
2370 						drophdr.count);
2371 		}
2372 		memcpy(save_leaf, tmp_leaf, state->args->geo->blksize);
2373 		savehdr = tmphdr; /* struct copy */
2374 		kfree(tmp_leaf);
2375 	}
2376 
2377 	xfs_attr3_leaf_hdr_to_disk(state->args->geo, save_leaf, &savehdr);
2378 	xfs_trans_log_buf(state->args->trans, save_blk->bp, 0,
2379 					   state->args->geo->blksize - 1);
2380 
2381 	/*
2382 	 * Copy out last hashval in each block for B-tree code.
2383 	 */
2384 	entry = xfs_attr3_leaf_entryp(save_leaf);
2385 	save_blk->hashval = be32_to_cpu(entry[savehdr.count - 1].hashval);
2386 }
2387 
2388 /*========================================================================
2389  * Routines used for finding things in the Btree.
2390  *========================================================================*/
2391 
2392 /*
2393  * Look up a name in a leaf attribute list structure.
2394  * This is the internal routine, it uses the caller's buffer.
2395  *
2396  * Note that duplicate keys are allowed, but only check within the
2397  * current leaf node.  The Btree code must check in adjacent leaf nodes.
2398  *
2399  * Return in args->index the index into the entry[] array of either
2400  * the found entry, or where the entry should have been (insert before
2401  * that entry).
2402  *
2403  * Don't change the args->value unless we find the attribute.
2404  */
2405 int
2406 xfs_attr3_leaf_lookup_int(
2407 	struct xfs_buf		*bp,
2408 	struct xfs_da_args	*args)
2409 {
2410 	struct xfs_attr_leafblock *leaf;
2411 	struct xfs_attr3_icleaf_hdr ichdr;
2412 	struct xfs_attr_leaf_entry *entry;
2413 	struct xfs_attr_leaf_entry *entries;
2414 	struct xfs_attr_leaf_name_local *name_loc;
2415 	struct xfs_attr_leaf_name_remote *name_rmt;
2416 	xfs_dahash_t		hashval;
2417 	int			probe;
2418 	int			span;
2419 
2420 	trace_xfs_attr_leaf_lookup(args);
2421 
2422 	leaf = bp->b_addr;
2423 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2424 	entries = xfs_attr3_leaf_entryp(leaf);
2425 	if (ichdr.count >= args->geo->blksize / 8) {
2426 		xfs_buf_mark_corrupt(bp);
2427 		xfs_da_mark_sick(args);
2428 		return -EFSCORRUPTED;
2429 	}
2430 
2431 	/*
2432 	 * Binary search.  (note: small blocks will skip this loop)
2433 	 */
2434 	hashval = args->hashval;
2435 	probe = span = ichdr.count / 2;
2436 	for (entry = &entries[probe]; span > 4; entry = &entries[probe]) {
2437 		span /= 2;
2438 		if (be32_to_cpu(entry->hashval) < hashval)
2439 			probe += span;
2440 		else if (be32_to_cpu(entry->hashval) > hashval)
2441 			probe -= span;
2442 		else
2443 			break;
2444 	}
2445 	if (!(probe >= 0 && (!ichdr.count || probe < ichdr.count))) {
2446 		xfs_buf_mark_corrupt(bp);
2447 		xfs_da_mark_sick(args);
2448 		return -EFSCORRUPTED;
2449 	}
2450 	if (!(span <= 4 || be32_to_cpu(entry->hashval) == hashval)) {
2451 		xfs_buf_mark_corrupt(bp);
2452 		xfs_da_mark_sick(args);
2453 		return -EFSCORRUPTED;
2454 	}
2455 
2456 	/*
2457 	 * Since we may have duplicate hashval's, find the first matching
2458 	 * hashval in the leaf.
2459 	 */
2460 	while (probe > 0 && be32_to_cpu(entry->hashval) >= hashval) {
2461 		entry--;
2462 		probe--;
2463 	}
2464 	while (probe < ichdr.count &&
2465 	       be32_to_cpu(entry->hashval) < hashval) {
2466 		entry++;
2467 		probe++;
2468 	}
2469 	if (probe == ichdr.count || be32_to_cpu(entry->hashval) != hashval) {
2470 		args->index = probe;
2471 		return -ENOATTR;
2472 	}
2473 
2474 	/*
2475 	 * Duplicate keys may be present, so search all of them for a match.
2476 	 */
2477 	for (; probe < ichdr.count && (be32_to_cpu(entry->hashval) == hashval);
2478 			entry++, probe++) {
2479 /*
2480  * GROT: Add code to remove incomplete entries.
2481  */
2482 		if (entry->flags & XFS_ATTR_LOCAL) {
2483 			name_loc = xfs_attr3_leaf_name_local(leaf, probe);
2484 			if (!xfs_attr_match(args, entry->flags,
2485 					name_loc->nameval, name_loc->namelen,
2486 					&name_loc->nameval[name_loc->namelen],
2487 					be16_to_cpu(name_loc->valuelen)))
2488 				continue;
2489 			args->index = probe;
2490 			return -EEXIST;
2491 		} else {
2492 			unsigned int	valuelen;
2493 
2494 			name_rmt = xfs_attr3_leaf_name_remote(leaf, probe);
2495 			valuelen = be32_to_cpu(name_rmt->valuelen);
2496 			if (!xfs_attr_match(args, entry->flags, name_rmt->name,
2497 					name_rmt->namelen, NULL, valuelen))
2498 				continue;
2499 			args->index = probe;
2500 			args->rmtvaluelen = valuelen;
2501 			args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2502 			args->rmtblkcnt = xfs_attr3_rmt_blocks(
2503 							args->dp->i_mount,
2504 							args->rmtvaluelen);
2505 			return -EEXIST;
2506 		}
2507 	}
2508 	args->index = probe;
2509 	return -ENOATTR;
2510 }
2511 
2512 /*
2513  * Get the value associated with an attribute name from a leaf attribute
2514  * list structure.
2515  *
2516  * If args->valuelen is zero, only the length needs to be returned.  Unlike a
2517  * lookup, we only return an error if the attribute does not exist or we can't
2518  * retrieve the value.
2519  */
2520 int
2521 xfs_attr3_leaf_getvalue(
2522 	struct xfs_buf		*bp,
2523 	struct xfs_da_args	*args)
2524 {
2525 	struct xfs_attr_leafblock *leaf;
2526 	struct xfs_attr3_icleaf_hdr ichdr;
2527 	struct xfs_attr_leaf_entry *entry;
2528 	struct xfs_attr_leaf_name_local *name_loc;
2529 	struct xfs_attr_leaf_name_remote *name_rmt;
2530 
2531 	leaf = bp->b_addr;
2532 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2533 	ASSERT(ichdr.count < args->geo->blksize / 8);
2534 	ASSERT(args->index < ichdr.count);
2535 
2536 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2537 	if (entry->flags & XFS_ATTR_LOCAL) {
2538 		name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2539 		ASSERT(name_loc->namelen == args->namelen);
2540 		ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2541 		return xfs_attr_copy_value(args,
2542 					&name_loc->nameval[args->namelen],
2543 					be16_to_cpu(name_loc->valuelen));
2544 	}
2545 
2546 	name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2547 	ASSERT(name_rmt->namelen == args->namelen);
2548 	ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2549 	args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2550 	args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2551 	args->rmtblkcnt = xfs_attr3_rmt_blocks(args->dp->i_mount,
2552 					       args->rmtvaluelen);
2553 	return xfs_attr_copy_value(args, NULL, args->rmtvaluelen);
2554 }
2555 
2556 /*========================================================================
2557  * Utility routines.
2558  *========================================================================*/
2559 
2560 /*
2561  * Move the indicated entries from one leaf to another.
2562  * NOTE: this routine modifies both source and destination leaves.
2563  */
2564 /*ARGSUSED*/
2565 STATIC void
2566 xfs_attr3_leaf_moveents(
2567 	struct xfs_da_args		*args,
2568 	struct xfs_attr_leafblock	*leaf_s,
2569 	struct xfs_attr3_icleaf_hdr	*ichdr_s,
2570 	int				start_s,
2571 	struct xfs_attr_leafblock	*leaf_d,
2572 	struct xfs_attr3_icleaf_hdr	*ichdr_d,
2573 	int				start_d,
2574 	int				count)
2575 {
2576 	struct xfs_attr_leaf_entry	*entry_s;
2577 	struct xfs_attr_leaf_entry	*entry_d;
2578 	int				desti;
2579 	int				tmp;
2580 	int				i;
2581 
2582 	/*
2583 	 * Check for nothing to do.
2584 	 */
2585 	if (count == 0)
2586 		return;
2587 
2588 	/*
2589 	 * Set up environment.
2590 	 */
2591 	ASSERT(ichdr_s->magic == XFS_ATTR_LEAF_MAGIC ||
2592 	       ichdr_s->magic == XFS_ATTR3_LEAF_MAGIC);
2593 	ASSERT(ichdr_s->magic == ichdr_d->magic);
2594 	ASSERT(ichdr_s->count > 0 && ichdr_s->count < args->geo->blksize / 8);
2595 	ASSERT(ichdr_s->firstused >= (ichdr_s->count * sizeof(*entry_s))
2596 					+ xfs_attr3_leaf_hdr_size(leaf_s));
2597 	ASSERT(ichdr_d->count < args->geo->blksize / 8);
2598 	ASSERT(ichdr_d->firstused >= (ichdr_d->count * sizeof(*entry_d))
2599 					+ xfs_attr3_leaf_hdr_size(leaf_d));
2600 
2601 	ASSERT(start_s < ichdr_s->count);
2602 	ASSERT(start_d <= ichdr_d->count);
2603 	ASSERT(count <= ichdr_s->count);
2604 
2605 
2606 	/*
2607 	 * Move the entries in the destination leaf up to make a hole?
2608 	 */
2609 	if (start_d < ichdr_d->count) {
2610 		tmp  = ichdr_d->count - start_d;
2611 		tmp *= sizeof(xfs_attr_leaf_entry_t);
2612 		entry_s = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2613 		entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d + count];
2614 		memmove(entry_d, entry_s, tmp);
2615 	}
2616 
2617 	/*
2618 	 * Copy all entry's in the same (sorted) order,
2619 	 * but allocate attribute info packed and in sequence.
2620 	 */
2621 	entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2622 	entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2623 	desti = start_d;
2624 	for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2625 		ASSERT(be16_to_cpu(entry_s->nameidx) >= ichdr_s->firstused);
2626 		tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2627 #ifdef GROT
2628 		/*
2629 		 * Code to drop INCOMPLETE entries.  Difficult to use as we
2630 		 * may also need to change the insertion index.  Code turned
2631 		 * off for 6.2, should be revisited later.
2632 		 */
2633 		if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2634 			memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2635 			ichdr_s->usedbytes -= tmp;
2636 			ichdr_s->count -= 1;
2637 			entry_d--;	/* to compensate for ++ in loop hdr */
2638 			desti--;
2639 			if ((start_s + i) < offset)
2640 				result++;	/* insertion index adjustment */
2641 		} else {
2642 #endif /* GROT */
2643 			ichdr_d->firstused -= tmp;
2644 			/* both on-disk, don't endian flip twice */
2645 			entry_d->hashval = entry_s->hashval;
2646 			entry_d->nameidx = cpu_to_be16(ichdr_d->firstused);
2647 			entry_d->flags = entry_s->flags;
2648 			ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2649 							<= args->geo->blksize);
2650 			memmove(xfs_attr3_leaf_name(leaf_d, desti),
2651 				xfs_attr3_leaf_name(leaf_s, start_s + i), tmp);
2652 			ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2653 							<= args->geo->blksize);
2654 			memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2655 			ichdr_s->usedbytes -= tmp;
2656 			ichdr_d->usedbytes += tmp;
2657 			ichdr_s->count -= 1;
2658 			ichdr_d->count += 1;
2659 			tmp = ichdr_d->count * sizeof(xfs_attr_leaf_entry_t)
2660 					+ xfs_attr3_leaf_hdr_size(leaf_d);
2661 			ASSERT(ichdr_d->firstused >= tmp);
2662 #ifdef GROT
2663 		}
2664 #endif /* GROT */
2665 	}
2666 
2667 	/*
2668 	 * Zero out the entries we just copied.
2669 	 */
2670 	if (start_s == ichdr_s->count) {
2671 		tmp = count * sizeof(xfs_attr_leaf_entry_t);
2672 		entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2673 		ASSERT(((char *)entry_s + tmp) <=
2674 		       ((char *)leaf_s + args->geo->blksize));
2675 		memset(entry_s, 0, tmp);
2676 	} else {
2677 		/*
2678 		 * Move the remaining entries down to fill the hole,
2679 		 * then zero the entries at the top.
2680 		 */
2681 		tmp  = (ichdr_s->count - count) * sizeof(xfs_attr_leaf_entry_t);
2682 		entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s + count];
2683 		entry_d = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2684 		memmove(entry_d, entry_s, tmp);
2685 
2686 		tmp = count * sizeof(xfs_attr_leaf_entry_t);
2687 		entry_s = &xfs_attr3_leaf_entryp(leaf_s)[ichdr_s->count];
2688 		ASSERT(((char *)entry_s + tmp) <=
2689 		       ((char *)leaf_s + args->geo->blksize));
2690 		memset(entry_s, 0, tmp);
2691 	}
2692 
2693 	/*
2694 	 * Fill in the freemap information
2695 	 */
2696 	ichdr_d->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_d);
2697 	ichdr_d->freemap[0].base += ichdr_d->count * sizeof(xfs_attr_leaf_entry_t);
2698 	ichdr_d->freemap[0].size = ichdr_d->firstused - ichdr_d->freemap[0].base;
2699 	ichdr_d->freemap[1].base = 0;
2700 	ichdr_d->freemap[2].base = 0;
2701 	ichdr_d->freemap[1].size = 0;
2702 	ichdr_d->freemap[2].size = 0;
2703 	ichdr_s->holes = 1;	/* leaf may not be compact */
2704 }
2705 
2706 /*
2707  * Pick up the last hashvalue from a leaf block.
2708  */
2709 xfs_dahash_t
2710 xfs_attr_leaf_lasthash(
2711 	struct xfs_buf	*bp,
2712 	int		*count)
2713 {
2714 	struct xfs_attr3_icleaf_hdr ichdr;
2715 	struct xfs_attr_leaf_entry *entries;
2716 	struct xfs_mount *mp = bp->b_mount;
2717 
2718 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, bp->b_addr);
2719 	entries = xfs_attr3_leaf_entryp(bp->b_addr);
2720 	if (count)
2721 		*count = ichdr.count;
2722 	if (!ichdr.count)
2723 		return 0;
2724 	return be32_to_cpu(entries[ichdr.count - 1].hashval);
2725 }
2726 
2727 /*
2728  * Calculate the number of bytes used to store the indicated attribute
2729  * (whether local or remote only calculate bytes in this block).
2730  */
2731 STATIC int
2732 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2733 {
2734 	struct xfs_attr_leaf_entry *entries;
2735 	xfs_attr_leaf_name_local_t *name_loc;
2736 	xfs_attr_leaf_name_remote_t *name_rmt;
2737 	int size;
2738 
2739 	entries = xfs_attr3_leaf_entryp(leaf);
2740 	if (entries[index].flags & XFS_ATTR_LOCAL) {
2741 		name_loc = xfs_attr3_leaf_name_local(leaf, index);
2742 		size = xfs_attr_leaf_entsize_local(name_loc->namelen,
2743 						   be16_to_cpu(name_loc->valuelen));
2744 	} else {
2745 		name_rmt = xfs_attr3_leaf_name_remote(leaf, index);
2746 		size = xfs_attr_leaf_entsize_remote(name_rmt->namelen);
2747 	}
2748 	return size;
2749 }
2750 
2751 /*
2752  * Calculate the number of bytes that would be required to store the new
2753  * attribute (whether local or remote only calculate bytes in this block).
2754  * This routine decides as a side effect whether the attribute will be
2755  * a "local" or a "remote" attribute.
2756  */
2757 int
2758 xfs_attr_leaf_newentsize(
2759 	struct xfs_da_args	*args,
2760 	int			*local)
2761 {
2762 	int			size;
2763 
2764 	size = xfs_attr_leaf_entsize_local(args->namelen, args->valuelen);
2765 	if (size < xfs_attr_leaf_entsize_local_max(args->geo->blksize)) {
2766 		if (local)
2767 			*local = 1;
2768 		return size;
2769 	}
2770 	if (local)
2771 		*local = 0;
2772 	return xfs_attr_leaf_entsize_remote(args->namelen);
2773 }
2774 
2775 
2776 /*========================================================================
2777  * Manage the INCOMPLETE flag in a leaf entry
2778  *========================================================================*/
2779 
2780 /*
2781  * Clear the INCOMPLETE flag on an entry in a leaf block.
2782  */
2783 int
2784 xfs_attr3_leaf_clearflag(
2785 	struct xfs_da_args	*args)
2786 {
2787 	struct xfs_attr_leafblock *leaf;
2788 	struct xfs_attr_leaf_entry *entry;
2789 	struct xfs_attr_leaf_name_remote *name_rmt;
2790 	struct xfs_buf		*bp;
2791 	int			error;
2792 #ifdef DEBUG
2793 	struct xfs_attr3_icleaf_hdr ichdr;
2794 	xfs_attr_leaf_name_local_t *name_loc;
2795 	int namelen;
2796 	char *name;
2797 #endif /* DEBUG */
2798 
2799 	trace_xfs_attr_leaf_clearflag(args);
2800 	/*
2801 	 * Set up the operation.
2802 	 */
2803 	error = xfs_attr3_leaf_read(args->trans, args->dp, args->owner,
2804 			args->blkno, &bp);
2805 	if (error)
2806 		return error;
2807 
2808 	leaf = bp->b_addr;
2809 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2810 	ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2811 
2812 #ifdef DEBUG
2813 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2814 	ASSERT(args->index < ichdr.count);
2815 	ASSERT(args->index >= 0);
2816 
2817 	if (entry->flags & XFS_ATTR_LOCAL) {
2818 		name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2819 		namelen = name_loc->namelen;
2820 		name = (char *)name_loc->nameval;
2821 	} else {
2822 		name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2823 		namelen = name_rmt->namelen;
2824 		name = (char *)name_rmt->name;
2825 	}
2826 	ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2827 	ASSERT(namelen == args->namelen);
2828 	ASSERT(memcmp(name, args->name, namelen) == 0);
2829 #endif /* DEBUG */
2830 
2831 	entry->flags &= ~XFS_ATTR_INCOMPLETE;
2832 	xfs_trans_log_buf(args->trans, bp,
2833 			 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2834 
2835 	if (args->rmtblkno) {
2836 		ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2837 		name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2838 		name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2839 		name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2840 		xfs_trans_log_buf(args->trans, bp,
2841 			 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2842 	}
2843 
2844 	return 0;
2845 }
2846 
2847 /*
2848  * Set the INCOMPLETE flag on an entry in a leaf block.
2849  */
2850 int
2851 xfs_attr3_leaf_setflag(
2852 	struct xfs_da_args	*args)
2853 {
2854 	struct xfs_attr_leafblock *leaf;
2855 	struct xfs_attr_leaf_entry *entry;
2856 	struct xfs_attr_leaf_name_remote *name_rmt;
2857 	struct xfs_buf		*bp;
2858 	int error;
2859 #ifdef DEBUG
2860 	struct xfs_attr3_icleaf_hdr ichdr;
2861 #endif
2862 
2863 	trace_xfs_attr_leaf_setflag(args);
2864 
2865 	/*
2866 	 * Set up the operation.
2867 	 */
2868 	error = xfs_attr3_leaf_read(args->trans, args->dp, args->owner,
2869 			args->blkno, &bp);
2870 	if (error)
2871 		return error;
2872 
2873 	leaf = bp->b_addr;
2874 #ifdef DEBUG
2875 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2876 	ASSERT(args->index < ichdr.count);
2877 	ASSERT(args->index >= 0);
2878 #endif
2879 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2880 
2881 	ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2882 	entry->flags |= XFS_ATTR_INCOMPLETE;
2883 	xfs_trans_log_buf(args->trans, bp,
2884 			XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2885 	if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2886 		name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2887 		name_rmt->valueblk = 0;
2888 		name_rmt->valuelen = 0;
2889 		xfs_trans_log_buf(args->trans, bp,
2890 			 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2891 	}
2892 
2893 	return 0;
2894 }
2895 
2896 /*
2897  * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2898  * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2899  * entry given by args->blkno2/index2.
2900  *
2901  * Note that they could be in different blocks, or in the same block.
2902  */
2903 int
2904 xfs_attr3_leaf_flipflags(
2905 	struct xfs_da_args	*args)
2906 {
2907 	struct xfs_attr_leafblock *leaf1;
2908 	struct xfs_attr_leafblock *leaf2;
2909 	struct xfs_attr_leaf_entry *entry1;
2910 	struct xfs_attr_leaf_entry *entry2;
2911 	struct xfs_attr_leaf_name_remote *name_rmt;
2912 	struct xfs_buf		*bp1;
2913 	struct xfs_buf		*bp2;
2914 	int error;
2915 #ifdef DEBUG
2916 	struct xfs_attr3_icleaf_hdr ichdr1;
2917 	struct xfs_attr3_icleaf_hdr ichdr2;
2918 	xfs_attr_leaf_name_local_t *name_loc;
2919 	int namelen1, namelen2;
2920 	char *name1, *name2;
2921 #endif /* DEBUG */
2922 
2923 	trace_xfs_attr_leaf_flipflags(args);
2924 
2925 	/*
2926 	 * Read the block containing the "old" attr
2927 	 */
2928 	error = xfs_attr3_leaf_read(args->trans, args->dp, args->owner,
2929 			args->blkno, &bp1);
2930 	if (error)
2931 		return error;
2932 
2933 	/*
2934 	 * Read the block containing the "new" attr, if it is different
2935 	 */
2936 	if (args->blkno2 != args->blkno) {
2937 		error = xfs_attr3_leaf_read(args->trans, args->dp, args->owner,
2938 				args->blkno2, &bp2);
2939 		if (error)
2940 			return error;
2941 	} else {
2942 		bp2 = bp1;
2943 	}
2944 
2945 	leaf1 = bp1->b_addr;
2946 	entry1 = &xfs_attr3_leaf_entryp(leaf1)[args->index];
2947 
2948 	leaf2 = bp2->b_addr;
2949 	entry2 = &xfs_attr3_leaf_entryp(leaf2)[args->index2];
2950 
2951 #ifdef DEBUG
2952 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr1, leaf1);
2953 	ASSERT(args->index < ichdr1.count);
2954 	ASSERT(args->index >= 0);
2955 
2956 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr2, leaf2);
2957 	ASSERT(args->index2 < ichdr2.count);
2958 	ASSERT(args->index2 >= 0);
2959 
2960 	if (entry1->flags & XFS_ATTR_LOCAL) {
2961 		name_loc = xfs_attr3_leaf_name_local(leaf1, args->index);
2962 		namelen1 = name_loc->namelen;
2963 		name1 = (char *)name_loc->nameval;
2964 	} else {
2965 		name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2966 		namelen1 = name_rmt->namelen;
2967 		name1 = (char *)name_rmt->name;
2968 	}
2969 	if (entry2->flags & XFS_ATTR_LOCAL) {
2970 		name_loc = xfs_attr3_leaf_name_local(leaf2, args->index2);
2971 		namelen2 = name_loc->namelen;
2972 		name2 = (char *)name_loc->nameval;
2973 	} else {
2974 		name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2975 		namelen2 = name_rmt->namelen;
2976 		name2 = (char *)name_rmt->name;
2977 	}
2978 	ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2979 	ASSERT(namelen1 == namelen2);
2980 	ASSERT(memcmp(name1, name2, namelen1) == 0);
2981 #endif /* DEBUG */
2982 
2983 	ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2984 	ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2985 
2986 	entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2987 	xfs_trans_log_buf(args->trans, bp1,
2988 			  XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2989 	if (args->rmtblkno) {
2990 		ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2991 		name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2992 		name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2993 		name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2994 		xfs_trans_log_buf(args->trans, bp1,
2995 			 XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2996 	}
2997 
2998 	entry2->flags |= XFS_ATTR_INCOMPLETE;
2999 	xfs_trans_log_buf(args->trans, bp2,
3000 			  XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
3001 	if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
3002 		name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
3003 		name_rmt->valueblk = 0;
3004 		name_rmt->valuelen = 0;
3005 		xfs_trans_log_buf(args->trans, bp2,
3006 			 XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
3007 	}
3008 
3009 	return 0;
3010 }
3011