xref: /linux/fs/xfs/libxfs/xfs_attr_leaf.c (revision 24bce201d79807b668bf9d9e0aca801c5c0d5f78)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4  * Copyright (c) 2013 Red Hat, Inc.
5  * All Rights Reserved.
6  */
7 #include "xfs.h"
8 #include "xfs_fs.h"
9 #include "xfs_shared.h"
10 #include "xfs_format.h"
11 #include "xfs_log_format.h"
12 #include "xfs_trans_resv.h"
13 #include "xfs_sb.h"
14 #include "xfs_mount.h"
15 #include "xfs_da_format.h"
16 #include "xfs_da_btree.h"
17 #include "xfs_inode.h"
18 #include "xfs_trans.h"
19 #include "xfs_bmap_btree.h"
20 #include "xfs_bmap.h"
21 #include "xfs_attr_sf.h"
22 #include "xfs_attr.h"
23 #include "xfs_attr_remote.h"
24 #include "xfs_attr_leaf.h"
25 #include "xfs_error.h"
26 #include "xfs_trace.h"
27 #include "xfs_buf_item.h"
28 #include "xfs_dir2.h"
29 #include "xfs_log.h"
30 #include "xfs_ag.h"
31 #include "xfs_errortag.h"
32 
33 
34 /*
35  * xfs_attr_leaf.c
36  *
37  * Routines to implement leaf blocks of attributes as Btrees of hashed names.
38  */
39 
40 /*========================================================================
41  * Function prototypes for the kernel.
42  *========================================================================*/
43 
44 /*
45  * Routines used for growing the Btree.
46  */
47 STATIC int xfs_attr3_leaf_create(struct xfs_da_args *args,
48 				 xfs_dablk_t which_block, struct xfs_buf **bpp);
49 STATIC int xfs_attr3_leaf_add_work(struct xfs_buf *leaf_buffer,
50 				   struct xfs_attr3_icleaf_hdr *ichdr,
51 				   struct xfs_da_args *args, int freemap_index);
52 STATIC void xfs_attr3_leaf_compact(struct xfs_da_args *args,
53 				   struct xfs_attr3_icleaf_hdr *ichdr,
54 				   struct xfs_buf *leaf_buffer);
55 STATIC void xfs_attr3_leaf_rebalance(xfs_da_state_t *state,
56 						   xfs_da_state_blk_t *blk1,
57 						   xfs_da_state_blk_t *blk2);
58 STATIC int xfs_attr3_leaf_figure_balance(xfs_da_state_t *state,
59 			xfs_da_state_blk_t *leaf_blk_1,
60 			struct xfs_attr3_icleaf_hdr *ichdr1,
61 			xfs_da_state_blk_t *leaf_blk_2,
62 			struct xfs_attr3_icleaf_hdr *ichdr2,
63 			int *number_entries_in_blk1,
64 			int *number_usedbytes_in_blk1);
65 
66 /*
67  * Utility routines.
68  */
69 STATIC void xfs_attr3_leaf_moveents(struct xfs_da_args *args,
70 			struct xfs_attr_leafblock *src_leaf,
71 			struct xfs_attr3_icleaf_hdr *src_ichdr, int src_start,
72 			struct xfs_attr_leafblock *dst_leaf,
73 			struct xfs_attr3_icleaf_hdr *dst_ichdr, int dst_start,
74 			int move_count);
75 STATIC int xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index);
76 
77 /*
78  * attr3 block 'firstused' conversion helpers.
79  *
80  * firstused refers to the offset of the first used byte of the nameval region
81  * of an attr leaf block. The region starts at the tail of the block and expands
82  * backwards towards the middle. As such, firstused is initialized to the block
83  * size for an empty leaf block and is reduced from there.
84  *
85  * The attr3 block size is pegged to the fsb size and the maximum fsb is 64k.
86  * The in-core firstused field is 32-bit and thus supports the maximum fsb size.
87  * The on-disk field is only 16-bit, however, and overflows at 64k. Since this
88  * only occurs at exactly 64k, we use zero as a magic on-disk value to represent
89  * the attr block size. The following helpers manage the conversion between the
90  * in-core and on-disk formats.
91  */
92 
93 static void
94 xfs_attr3_leaf_firstused_from_disk(
95 	struct xfs_da_geometry		*geo,
96 	struct xfs_attr3_icleaf_hdr	*to,
97 	struct xfs_attr_leafblock	*from)
98 {
99 	struct xfs_attr3_leaf_hdr	*hdr3;
100 
101 	if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) {
102 		hdr3 = (struct xfs_attr3_leaf_hdr *) from;
103 		to->firstused = be16_to_cpu(hdr3->firstused);
104 	} else {
105 		to->firstused = be16_to_cpu(from->hdr.firstused);
106 	}
107 
108 	/*
109 	 * Convert from the magic fsb size value to actual blocksize. This
110 	 * should only occur for empty blocks when the block size overflows
111 	 * 16-bits.
112 	 */
113 	if (to->firstused == XFS_ATTR3_LEAF_NULLOFF) {
114 		ASSERT(!to->count && !to->usedbytes);
115 		ASSERT(geo->blksize > USHRT_MAX);
116 		to->firstused = geo->blksize;
117 	}
118 }
119 
120 static void
121 xfs_attr3_leaf_firstused_to_disk(
122 	struct xfs_da_geometry		*geo,
123 	struct xfs_attr_leafblock	*to,
124 	struct xfs_attr3_icleaf_hdr	*from)
125 {
126 	struct xfs_attr3_leaf_hdr	*hdr3;
127 	uint32_t			firstused;
128 
129 	/* magic value should only be seen on disk */
130 	ASSERT(from->firstused != XFS_ATTR3_LEAF_NULLOFF);
131 
132 	/*
133 	 * Scale down the 32-bit in-core firstused value to the 16-bit on-disk
134 	 * value. This only overflows at the max supported value of 64k. Use the
135 	 * magic on-disk value to represent block size in this case.
136 	 */
137 	firstused = from->firstused;
138 	if (firstused > USHRT_MAX) {
139 		ASSERT(from->firstused == geo->blksize);
140 		firstused = XFS_ATTR3_LEAF_NULLOFF;
141 	}
142 
143 	if (from->magic == XFS_ATTR3_LEAF_MAGIC) {
144 		hdr3 = (struct xfs_attr3_leaf_hdr *) to;
145 		hdr3->firstused = cpu_to_be16(firstused);
146 	} else {
147 		to->hdr.firstused = cpu_to_be16(firstused);
148 	}
149 }
150 
151 void
152 xfs_attr3_leaf_hdr_from_disk(
153 	struct xfs_da_geometry		*geo,
154 	struct xfs_attr3_icleaf_hdr	*to,
155 	struct xfs_attr_leafblock	*from)
156 {
157 	int	i;
158 
159 	ASSERT(from->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
160 	       from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
161 
162 	if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) {
163 		struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)from;
164 
165 		to->forw = be32_to_cpu(hdr3->info.hdr.forw);
166 		to->back = be32_to_cpu(hdr3->info.hdr.back);
167 		to->magic = be16_to_cpu(hdr3->info.hdr.magic);
168 		to->count = be16_to_cpu(hdr3->count);
169 		to->usedbytes = be16_to_cpu(hdr3->usedbytes);
170 		xfs_attr3_leaf_firstused_from_disk(geo, to, from);
171 		to->holes = hdr3->holes;
172 
173 		for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
174 			to->freemap[i].base = be16_to_cpu(hdr3->freemap[i].base);
175 			to->freemap[i].size = be16_to_cpu(hdr3->freemap[i].size);
176 		}
177 		return;
178 	}
179 	to->forw = be32_to_cpu(from->hdr.info.forw);
180 	to->back = be32_to_cpu(from->hdr.info.back);
181 	to->magic = be16_to_cpu(from->hdr.info.magic);
182 	to->count = be16_to_cpu(from->hdr.count);
183 	to->usedbytes = be16_to_cpu(from->hdr.usedbytes);
184 	xfs_attr3_leaf_firstused_from_disk(geo, to, from);
185 	to->holes = from->hdr.holes;
186 
187 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
188 		to->freemap[i].base = be16_to_cpu(from->hdr.freemap[i].base);
189 		to->freemap[i].size = be16_to_cpu(from->hdr.freemap[i].size);
190 	}
191 }
192 
193 void
194 xfs_attr3_leaf_hdr_to_disk(
195 	struct xfs_da_geometry		*geo,
196 	struct xfs_attr_leafblock	*to,
197 	struct xfs_attr3_icleaf_hdr	*from)
198 {
199 	int				i;
200 
201 	ASSERT(from->magic == XFS_ATTR_LEAF_MAGIC ||
202 	       from->magic == XFS_ATTR3_LEAF_MAGIC);
203 
204 	if (from->magic == XFS_ATTR3_LEAF_MAGIC) {
205 		struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)to;
206 
207 		hdr3->info.hdr.forw = cpu_to_be32(from->forw);
208 		hdr3->info.hdr.back = cpu_to_be32(from->back);
209 		hdr3->info.hdr.magic = cpu_to_be16(from->magic);
210 		hdr3->count = cpu_to_be16(from->count);
211 		hdr3->usedbytes = cpu_to_be16(from->usedbytes);
212 		xfs_attr3_leaf_firstused_to_disk(geo, to, from);
213 		hdr3->holes = from->holes;
214 		hdr3->pad1 = 0;
215 
216 		for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
217 			hdr3->freemap[i].base = cpu_to_be16(from->freemap[i].base);
218 			hdr3->freemap[i].size = cpu_to_be16(from->freemap[i].size);
219 		}
220 		return;
221 	}
222 	to->hdr.info.forw = cpu_to_be32(from->forw);
223 	to->hdr.info.back = cpu_to_be32(from->back);
224 	to->hdr.info.magic = cpu_to_be16(from->magic);
225 	to->hdr.count = cpu_to_be16(from->count);
226 	to->hdr.usedbytes = cpu_to_be16(from->usedbytes);
227 	xfs_attr3_leaf_firstused_to_disk(geo, to, from);
228 	to->hdr.holes = from->holes;
229 	to->hdr.pad1 = 0;
230 
231 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
232 		to->hdr.freemap[i].base = cpu_to_be16(from->freemap[i].base);
233 		to->hdr.freemap[i].size = cpu_to_be16(from->freemap[i].size);
234 	}
235 }
236 
237 static xfs_failaddr_t
238 xfs_attr3_leaf_verify_entry(
239 	struct xfs_mount			*mp,
240 	char					*buf_end,
241 	struct xfs_attr_leafblock		*leaf,
242 	struct xfs_attr3_icleaf_hdr		*leafhdr,
243 	struct xfs_attr_leaf_entry		*ent,
244 	int					idx,
245 	__u32					*last_hashval)
246 {
247 	struct xfs_attr_leaf_name_local		*lentry;
248 	struct xfs_attr_leaf_name_remote	*rentry;
249 	char					*name_end;
250 	unsigned int				nameidx;
251 	unsigned int				namesize;
252 	__u32					hashval;
253 
254 	/* hash order check */
255 	hashval = be32_to_cpu(ent->hashval);
256 	if (hashval < *last_hashval)
257 		return __this_address;
258 	*last_hashval = hashval;
259 
260 	nameidx = be16_to_cpu(ent->nameidx);
261 	if (nameidx < leafhdr->firstused || nameidx >= mp->m_attr_geo->blksize)
262 		return __this_address;
263 
264 	/*
265 	 * Check the name information.  The namelen fields are u8 so we can't
266 	 * possibly exceed the maximum name length of 255 bytes.
267 	 */
268 	if (ent->flags & XFS_ATTR_LOCAL) {
269 		lentry = xfs_attr3_leaf_name_local(leaf, idx);
270 		namesize = xfs_attr_leaf_entsize_local(lentry->namelen,
271 				be16_to_cpu(lentry->valuelen));
272 		name_end = (char *)lentry + namesize;
273 		if (lentry->namelen == 0)
274 			return __this_address;
275 	} else {
276 		rentry = xfs_attr3_leaf_name_remote(leaf, idx);
277 		namesize = xfs_attr_leaf_entsize_remote(rentry->namelen);
278 		name_end = (char *)rentry + namesize;
279 		if (rentry->namelen == 0)
280 			return __this_address;
281 		if (!(ent->flags & XFS_ATTR_INCOMPLETE) &&
282 		    rentry->valueblk == 0)
283 			return __this_address;
284 	}
285 
286 	if (name_end > buf_end)
287 		return __this_address;
288 
289 	return NULL;
290 }
291 
292 /*
293  * Validate an attribute leaf block.
294  *
295  * Empty leaf blocks can occur under the following circumstances:
296  *
297  * 1. setxattr adds a new extended attribute to a file;
298  * 2. The file has zero existing attributes;
299  * 3. The attribute is too large to fit in the attribute fork;
300  * 4. The attribute is small enough to fit in a leaf block;
301  * 5. A log flush occurs after committing the transaction that creates
302  *    the (empty) leaf block; and
303  * 6. The filesystem goes down after the log flush but before the new
304  *    attribute can be committed to the leaf block.
305  *
306  * Hence we need to ensure that we don't fail the validation purely
307  * because the leaf is empty.
308  */
309 static xfs_failaddr_t
310 xfs_attr3_leaf_verify(
311 	struct xfs_buf			*bp)
312 {
313 	struct xfs_attr3_icleaf_hdr	ichdr;
314 	struct xfs_mount		*mp = bp->b_mount;
315 	struct xfs_attr_leafblock	*leaf = bp->b_addr;
316 	struct xfs_attr_leaf_entry	*entries;
317 	struct xfs_attr_leaf_entry	*ent;
318 	char				*buf_end;
319 	uint32_t			end;	/* must be 32bit - see below */
320 	__u32				last_hashval = 0;
321 	int				i;
322 	xfs_failaddr_t			fa;
323 
324 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, leaf);
325 
326 	fa = xfs_da3_blkinfo_verify(bp, bp->b_addr);
327 	if (fa)
328 		return fa;
329 
330 	/*
331 	 * firstused is the block offset of the first name info structure.
332 	 * Make sure it doesn't go off the block or crash into the header.
333 	 */
334 	if (ichdr.firstused > mp->m_attr_geo->blksize)
335 		return __this_address;
336 	if (ichdr.firstused < xfs_attr3_leaf_hdr_size(leaf))
337 		return __this_address;
338 
339 	/* Make sure the entries array doesn't crash into the name info. */
340 	entries = xfs_attr3_leaf_entryp(bp->b_addr);
341 	if ((char *)&entries[ichdr.count] >
342 	    (char *)bp->b_addr + ichdr.firstused)
343 		return __this_address;
344 
345 	/*
346 	 * NOTE: This verifier historically failed empty leaf buffers because
347 	 * we expect the fork to be in another format. Empty attr fork format
348 	 * conversions are possible during xattr set, however, and format
349 	 * conversion is not atomic with the xattr set that triggers it. We
350 	 * cannot assume leaf blocks are non-empty until that is addressed.
351 	*/
352 	buf_end = (char *)bp->b_addr + mp->m_attr_geo->blksize;
353 	for (i = 0, ent = entries; i < ichdr.count; ent++, i++) {
354 		fa = xfs_attr3_leaf_verify_entry(mp, buf_end, leaf, &ichdr,
355 				ent, i, &last_hashval);
356 		if (fa)
357 			return fa;
358 	}
359 
360 	/*
361 	 * Quickly check the freemap information.  Attribute data has to be
362 	 * aligned to 4-byte boundaries, and likewise for the free space.
363 	 *
364 	 * Note that for 64k block size filesystems, the freemap entries cannot
365 	 * overflow as they are only be16 fields. However, when checking end
366 	 * pointer of the freemap, we have to be careful to detect overflows and
367 	 * so use uint32_t for those checks.
368 	 */
369 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
370 		if (ichdr.freemap[i].base > mp->m_attr_geo->blksize)
371 			return __this_address;
372 		if (ichdr.freemap[i].base & 0x3)
373 			return __this_address;
374 		if (ichdr.freemap[i].size > mp->m_attr_geo->blksize)
375 			return __this_address;
376 		if (ichdr.freemap[i].size & 0x3)
377 			return __this_address;
378 
379 		/* be care of 16 bit overflows here */
380 		end = (uint32_t)ichdr.freemap[i].base + ichdr.freemap[i].size;
381 		if (end < ichdr.freemap[i].base)
382 			return __this_address;
383 		if (end > mp->m_attr_geo->blksize)
384 			return __this_address;
385 	}
386 
387 	return NULL;
388 }
389 
390 static void
391 xfs_attr3_leaf_write_verify(
392 	struct xfs_buf	*bp)
393 {
394 	struct xfs_mount	*mp = bp->b_mount;
395 	struct xfs_buf_log_item	*bip = bp->b_log_item;
396 	struct xfs_attr3_leaf_hdr *hdr3 = bp->b_addr;
397 	xfs_failaddr_t		fa;
398 
399 	fa = xfs_attr3_leaf_verify(bp);
400 	if (fa) {
401 		xfs_verifier_error(bp, -EFSCORRUPTED, fa);
402 		return;
403 	}
404 
405 	if (!xfs_has_crc(mp))
406 		return;
407 
408 	if (bip)
409 		hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
410 
411 	xfs_buf_update_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF);
412 }
413 
414 /*
415  * leaf/node format detection on trees is sketchy, so a node read can be done on
416  * leaf level blocks when detection identifies the tree as a node format tree
417  * incorrectly. In this case, we need to swap the verifier to match the correct
418  * format of the block being read.
419  */
420 static void
421 xfs_attr3_leaf_read_verify(
422 	struct xfs_buf		*bp)
423 {
424 	struct xfs_mount	*mp = bp->b_mount;
425 	xfs_failaddr_t		fa;
426 
427 	if (xfs_has_crc(mp) &&
428 	     !xfs_buf_verify_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF))
429 		xfs_verifier_error(bp, -EFSBADCRC, __this_address);
430 	else {
431 		fa = xfs_attr3_leaf_verify(bp);
432 		if (fa)
433 			xfs_verifier_error(bp, -EFSCORRUPTED, fa);
434 	}
435 }
436 
437 const struct xfs_buf_ops xfs_attr3_leaf_buf_ops = {
438 	.name = "xfs_attr3_leaf",
439 	.magic16 = { cpu_to_be16(XFS_ATTR_LEAF_MAGIC),
440 		     cpu_to_be16(XFS_ATTR3_LEAF_MAGIC) },
441 	.verify_read = xfs_attr3_leaf_read_verify,
442 	.verify_write = xfs_attr3_leaf_write_verify,
443 	.verify_struct = xfs_attr3_leaf_verify,
444 };
445 
446 int
447 xfs_attr3_leaf_read(
448 	struct xfs_trans	*tp,
449 	struct xfs_inode	*dp,
450 	xfs_dablk_t		bno,
451 	struct xfs_buf		**bpp)
452 {
453 	int			err;
454 
455 	err = xfs_da_read_buf(tp, dp, bno, 0, bpp, XFS_ATTR_FORK,
456 			&xfs_attr3_leaf_buf_ops);
457 	if (!err && tp && *bpp)
458 		xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_ATTR_LEAF_BUF);
459 	return err;
460 }
461 
462 /*========================================================================
463  * Namespace helper routines
464  *========================================================================*/
465 
466 /*
467  * If we are in log recovery, then we want the lookup to ignore the INCOMPLETE
468  * flag on disk - if there's an incomplete attr then recovery needs to tear it
469  * down. If there's no incomplete attr, then recovery needs to tear that attr
470  * down to replace it with the attr that has been logged. In this case, the
471  * INCOMPLETE flag will not be set in attr->attr_filter, but rather
472  * XFS_DA_OP_RECOVERY will be set in args->op_flags.
473  */
474 static bool
475 xfs_attr_match(
476 	struct xfs_da_args	*args,
477 	uint8_t			namelen,
478 	unsigned char		*name,
479 	int			flags)
480 {
481 
482 	if (args->namelen != namelen)
483 		return false;
484 	if (memcmp(args->name, name, namelen) != 0)
485 		return false;
486 
487 	/* Recovery ignores the INCOMPLETE flag. */
488 	if ((args->op_flags & XFS_DA_OP_RECOVERY) &&
489 	    args->attr_filter == (flags & XFS_ATTR_NSP_ONDISK_MASK))
490 		return true;
491 
492 	/* All remaining matches need to be filtered by INCOMPLETE state. */
493 	if (args->attr_filter !=
494 	    (flags & (XFS_ATTR_NSP_ONDISK_MASK | XFS_ATTR_INCOMPLETE)))
495 		return false;
496 	return true;
497 }
498 
499 static int
500 xfs_attr_copy_value(
501 	struct xfs_da_args	*args,
502 	unsigned char		*value,
503 	int			valuelen)
504 {
505 	/*
506 	 * No copy if all we have to do is get the length
507 	 */
508 	if (!args->valuelen) {
509 		args->valuelen = valuelen;
510 		return 0;
511 	}
512 
513 	/*
514 	 * No copy if the length of the existing buffer is too small
515 	 */
516 	if (args->valuelen < valuelen) {
517 		args->valuelen = valuelen;
518 		return -ERANGE;
519 	}
520 
521 	if (!args->value) {
522 		args->value = kvmalloc(valuelen, GFP_KERNEL | __GFP_NOLOCKDEP);
523 		if (!args->value)
524 			return -ENOMEM;
525 	}
526 	args->valuelen = valuelen;
527 
528 	/* remote block xattr requires IO for copy-in */
529 	if (args->rmtblkno)
530 		return xfs_attr_rmtval_get(args);
531 
532 	/*
533 	 * This is to prevent a GCC warning because the remote xattr case
534 	 * doesn't have a value to pass in. In that case, we never reach here,
535 	 * but GCC can't work that out and so throws a "passing NULL to
536 	 * memcpy" warning.
537 	 */
538 	if (!value)
539 		return -EINVAL;
540 	memcpy(args->value, value, valuelen);
541 	return 0;
542 }
543 
544 /*========================================================================
545  * External routines when attribute fork size < XFS_LITINO(mp).
546  *========================================================================*/
547 
548 /*
549  * Query whether the total requested number of attr fork bytes of extended
550  * attribute space will be able to fit inline.
551  *
552  * Returns zero if not, else the i_forkoff fork offset to be used in the
553  * literal area for attribute data once the new bytes have been added.
554  *
555  * i_forkoff must be 8 byte aligned, hence is stored as a >>3 value;
556  * special case for dev/uuid inodes, they have fixed size data forks.
557  */
558 int
559 xfs_attr_shortform_bytesfit(
560 	struct xfs_inode	*dp,
561 	int			bytes)
562 {
563 	struct xfs_mount	*mp = dp->i_mount;
564 	int64_t			dsize;
565 	int			minforkoff;
566 	int			maxforkoff;
567 	int			offset;
568 
569 	/*
570 	 * Check if the new size could fit at all first:
571 	 */
572 	if (bytes > XFS_LITINO(mp))
573 		return 0;
574 
575 	/* rounded down */
576 	offset = (XFS_LITINO(mp) - bytes) >> 3;
577 
578 	if (dp->i_df.if_format == XFS_DINODE_FMT_DEV) {
579 		minforkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
580 		return (offset >= minforkoff) ? minforkoff : 0;
581 	}
582 
583 	/*
584 	 * If the requested numbers of bytes is smaller or equal to the
585 	 * current attribute fork size we can always proceed.
586 	 *
587 	 * Note that if_bytes in the data fork might actually be larger than
588 	 * the current data fork size is due to delalloc extents. In that
589 	 * case either the extent count will go down when they are converted
590 	 * to real extents, or the delalloc conversion will take care of the
591 	 * literal area rebalancing.
592 	 */
593 	if (bytes <= XFS_IFORK_ASIZE(dp))
594 		return dp->i_forkoff;
595 
596 	/*
597 	 * For attr2 we can try to move the forkoff if there is space in the
598 	 * literal area, but for the old format we are done if there is no
599 	 * space in the fixed attribute fork.
600 	 */
601 	if (!xfs_has_attr2(mp))
602 		return 0;
603 
604 	dsize = dp->i_df.if_bytes;
605 
606 	switch (dp->i_df.if_format) {
607 	case XFS_DINODE_FMT_EXTENTS:
608 		/*
609 		 * If there is no attr fork and the data fork is extents,
610 		 * determine if creating the default attr fork will result
611 		 * in the extents form migrating to btree. If so, the
612 		 * minimum offset only needs to be the space required for
613 		 * the btree root.
614 		 */
615 		if (!dp->i_forkoff && dp->i_df.if_bytes >
616 		    xfs_default_attroffset(dp))
617 			dsize = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
618 		break;
619 	case XFS_DINODE_FMT_BTREE:
620 		/*
621 		 * If we have a data btree then keep forkoff if we have one,
622 		 * otherwise we are adding a new attr, so then we set
623 		 * minforkoff to where the btree root can finish so we have
624 		 * plenty of room for attrs
625 		 */
626 		if (dp->i_forkoff) {
627 			if (offset < dp->i_forkoff)
628 				return 0;
629 			return dp->i_forkoff;
630 		}
631 		dsize = XFS_BMAP_BROOT_SPACE(mp, dp->i_df.if_broot);
632 		break;
633 	}
634 
635 	/*
636 	 * A data fork btree root must have space for at least
637 	 * MINDBTPTRS key/ptr pairs if the data fork is small or empty.
638 	 */
639 	minforkoff = max_t(int64_t, dsize, XFS_BMDR_SPACE_CALC(MINDBTPTRS));
640 	minforkoff = roundup(minforkoff, 8) >> 3;
641 
642 	/* attr fork btree root can have at least this many key/ptr pairs */
643 	maxforkoff = XFS_LITINO(mp) - XFS_BMDR_SPACE_CALC(MINABTPTRS);
644 	maxforkoff = maxforkoff >> 3;	/* rounded down */
645 
646 	if (offset >= maxforkoff)
647 		return maxforkoff;
648 	if (offset >= minforkoff)
649 		return offset;
650 	return 0;
651 }
652 
653 /*
654  * Switch on the ATTR2 superblock bit (implies also FEATURES2) unless:
655  * - noattr2 mount option is set,
656  * - on-disk version bit says it is already set, or
657  * - the attr2 mount option is not set to enable automatic upgrade from attr1.
658  */
659 STATIC void
660 xfs_sbversion_add_attr2(
661 	struct xfs_mount	*mp,
662 	struct xfs_trans	*tp)
663 {
664 	if (xfs_has_noattr2(mp))
665 		return;
666 	if (mp->m_sb.sb_features2 & XFS_SB_VERSION2_ATTR2BIT)
667 		return;
668 	if (!xfs_has_attr2(mp))
669 		return;
670 
671 	spin_lock(&mp->m_sb_lock);
672 	xfs_add_attr2(mp);
673 	spin_unlock(&mp->m_sb_lock);
674 	xfs_log_sb(tp);
675 }
676 
677 /*
678  * Create the initial contents of a shortform attribute list.
679  */
680 void
681 xfs_attr_shortform_create(
682 	struct xfs_da_args	*args)
683 {
684 	struct xfs_inode	*dp = args->dp;
685 	struct xfs_ifork	*ifp = dp->i_afp;
686 	struct xfs_attr_sf_hdr	*hdr;
687 
688 	trace_xfs_attr_sf_create(args);
689 
690 	ASSERT(ifp->if_bytes == 0);
691 	if (ifp->if_format == XFS_DINODE_FMT_EXTENTS)
692 		ifp->if_format = XFS_DINODE_FMT_LOCAL;
693 	xfs_idata_realloc(dp, sizeof(*hdr), XFS_ATTR_FORK);
694 	hdr = (struct xfs_attr_sf_hdr *)ifp->if_u1.if_data;
695 	memset(hdr, 0, sizeof(*hdr));
696 	hdr->totsize = cpu_to_be16(sizeof(*hdr));
697 	xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
698 }
699 
700 /*
701  * Return -EEXIST if attr is found, or -ENOATTR if not
702  * args:  args containing attribute name and namelen
703  * sfep:  If not null, pointer will be set to the last attr entry found on
704 	  -EEXIST.  On -ENOATTR pointer is left at the last entry in the list
705  * basep: If not null, pointer is set to the byte offset of the entry in the
706  *	  list on -EEXIST.  On -ENOATTR, pointer is left at the byte offset of
707  *	  the last entry in the list
708  */
709 int
710 xfs_attr_sf_findname(
711 	struct xfs_da_args	 *args,
712 	struct xfs_attr_sf_entry **sfep,
713 	unsigned int		 *basep)
714 {
715 	struct xfs_attr_shortform *sf;
716 	struct xfs_attr_sf_entry *sfe;
717 	unsigned int		base = sizeof(struct xfs_attr_sf_hdr);
718 	int			size = 0;
719 	int			end;
720 	int			i;
721 
722 	sf = (struct xfs_attr_shortform *)args->dp->i_afp->if_u1.if_data;
723 	sfe = &sf->list[0];
724 	end = sf->hdr.count;
725 	for (i = 0; i < end; sfe = xfs_attr_sf_nextentry(sfe),
726 			     base += size, i++) {
727 		size = xfs_attr_sf_entsize(sfe);
728 		if (!xfs_attr_match(args, sfe->namelen, sfe->nameval,
729 				    sfe->flags))
730 			continue;
731 		break;
732 	}
733 
734 	if (sfep != NULL)
735 		*sfep = sfe;
736 
737 	if (basep != NULL)
738 		*basep = base;
739 
740 	if (i == end)
741 		return -ENOATTR;
742 	return -EEXIST;
743 }
744 
745 /*
746  * Add a name/value pair to the shortform attribute list.
747  * Overflow from the inode has already been checked for.
748  */
749 void
750 xfs_attr_shortform_add(
751 	struct xfs_da_args		*args,
752 	int				forkoff)
753 {
754 	struct xfs_attr_shortform	*sf;
755 	struct xfs_attr_sf_entry	*sfe;
756 	int				offset, size;
757 	struct xfs_mount		*mp;
758 	struct xfs_inode		*dp;
759 	struct xfs_ifork		*ifp;
760 
761 	trace_xfs_attr_sf_add(args);
762 
763 	dp = args->dp;
764 	mp = dp->i_mount;
765 	dp->i_forkoff = forkoff;
766 
767 	ifp = dp->i_afp;
768 	ASSERT(ifp->if_format == XFS_DINODE_FMT_LOCAL);
769 	sf = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
770 	if (xfs_attr_sf_findname(args, &sfe, NULL) == -EEXIST)
771 		ASSERT(0);
772 
773 	offset = (char *)sfe - (char *)sf;
774 	size = xfs_attr_sf_entsize_byname(args->namelen, args->valuelen);
775 	xfs_idata_realloc(dp, size, XFS_ATTR_FORK);
776 	sf = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
777 	sfe = (struct xfs_attr_sf_entry *)((char *)sf + offset);
778 
779 	sfe->namelen = args->namelen;
780 	sfe->valuelen = args->valuelen;
781 	sfe->flags = args->attr_filter;
782 	memcpy(sfe->nameval, args->name, args->namelen);
783 	memcpy(&sfe->nameval[args->namelen], args->value, args->valuelen);
784 	sf->hdr.count++;
785 	be16_add_cpu(&sf->hdr.totsize, size);
786 	xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
787 
788 	xfs_sbversion_add_attr2(mp, args->trans);
789 }
790 
791 /*
792  * After the last attribute is removed revert to original inode format,
793  * making all literal area available to the data fork once more.
794  */
795 void
796 xfs_attr_fork_remove(
797 	struct xfs_inode	*ip,
798 	struct xfs_trans	*tp)
799 {
800 	ASSERT(ip->i_afp->if_nextents == 0);
801 
802 	xfs_idestroy_fork(ip->i_afp);
803 	kmem_cache_free(xfs_ifork_cache, ip->i_afp);
804 	ip->i_afp = NULL;
805 	ip->i_forkoff = 0;
806 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
807 }
808 
809 /*
810  * Remove an attribute from the shortform attribute list structure.
811  */
812 int
813 xfs_attr_sf_removename(
814 	struct xfs_da_args		*args)
815 {
816 	struct xfs_attr_shortform	*sf;
817 	struct xfs_attr_sf_entry	*sfe;
818 	int				size = 0, end, totsize;
819 	unsigned int			base;
820 	struct xfs_mount		*mp;
821 	struct xfs_inode		*dp;
822 	int				error;
823 
824 	trace_xfs_attr_sf_remove(args);
825 
826 	dp = args->dp;
827 	mp = dp->i_mount;
828 	sf = (struct xfs_attr_shortform *)dp->i_afp->if_u1.if_data;
829 
830 	error = xfs_attr_sf_findname(args, &sfe, &base);
831 
832 	/*
833 	 * If we are recovering an operation, finding nothing to
834 	 * remove is not an error - it just means there was nothing
835 	 * to clean up.
836 	 */
837 	if (error == -ENOATTR && (args->op_flags & XFS_DA_OP_RECOVERY))
838 		return 0;
839 	if (error != -EEXIST)
840 		return error;
841 	size = xfs_attr_sf_entsize(sfe);
842 
843 	/*
844 	 * Fix up the attribute fork data, covering the hole
845 	 */
846 	end = base + size;
847 	totsize = be16_to_cpu(sf->hdr.totsize);
848 	if (end != totsize)
849 		memmove(&((char *)sf)[base], &((char *)sf)[end], totsize - end);
850 	sf->hdr.count--;
851 	be16_add_cpu(&sf->hdr.totsize, -size);
852 
853 	/*
854 	 * Fix up the start offset of the attribute fork
855 	 */
856 	totsize -= size;
857 	if (totsize == sizeof(xfs_attr_sf_hdr_t) && xfs_has_attr2(mp) &&
858 	    (dp->i_df.if_format != XFS_DINODE_FMT_BTREE) &&
859 	    !(args->op_flags & (XFS_DA_OP_ADDNAME | XFS_DA_OP_REPLACE))) {
860 		xfs_attr_fork_remove(dp, args->trans);
861 	} else {
862 		xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
863 		dp->i_forkoff = xfs_attr_shortform_bytesfit(dp, totsize);
864 		ASSERT(dp->i_forkoff);
865 		ASSERT(totsize > sizeof(xfs_attr_sf_hdr_t) ||
866 				(args->op_flags & XFS_DA_OP_ADDNAME) ||
867 				!xfs_has_attr2(mp) ||
868 				dp->i_df.if_format == XFS_DINODE_FMT_BTREE);
869 		xfs_trans_log_inode(args->trans, dp,
870 					XFS_ILOG_CORE | XFS_ILOG_ADATA);
871 	}
872 
873 	xfs_sbversion_add_attr2(mp, args->trans);
874 
875 	return 0;
876 }
877 
878 /*
879  * Look up a name in a shortform attribute list structure.
880  */
881 /*ARGSUSED*/
882 int
883 xfs_attr_shortform_lookup(xfs_da_args_t *args)
884 {
885 	struct xfs_attr_shortform *sf;
886 	struct xfs_attr_sf_entry *sfe;
887 	int i;
888 	struct xfs_ifork *ifp;
889 
890 	trace_xfs_attr_sf_lookup(args);
891 
892 	ifp = args->dp->i_afp;
893 	ASSERT(ifp->if_format == XFS_DINODE_FMT_LOCAL);
894 	sf = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
895 	sfe = &sf->list[0];
896 	for (i = 0; i < sf->hdr.count;
897 				sfe = xfs_attr_sf_nextentry(sfe), i++) {
898 		if (xfs_attr_match(args, sfe->namelen, sfe->nameval,
899 				sfe->flags))
900 			return -EEXIST;
901 	}
902 	return -ENOATTR;
903 }
904 
905 /*
906  * Retrieve the attribute value and length.
907  *
908  * If args->valuelen is zero, only the length needs to be returned.  Unlike a
909  * lookup, we only return an error if the attribute does not exist or we can't
910  * retrieve the value.
911  */
912 int
913 xfs_attr_shortform_getvalue(
914 	struct xfs_da_args	*args)
915 {
916 	struct xfs_attr_shortform *sf;
917 	struct xfs_attr_sf_entry *sfe;
918 	int			i;
919 
920 	ASSERT(args->dp->i_afp->if_format == XFS_DINODE_FMT_LOCAL);
921 	sf = (struct xfs_attr_shortform *)args->dp->i_afp->if_u1.if_data;
922 	sfe = &sf->list[0];
923 	for (i = 0; i < sf->hdr.count;
924 				sfe = xfs_attr_sf_nextentry(sfe), i++) {
925 		if (xfs_attr_match(args, sfe->namelen, sfe->nameval,
926 				sfe->flags))
927 			return xfs_attr_copy_value(args,
928 				&sfe->nameval[args->namelen], sfe->valuelen);
929 	}
930 	return -ENOATTR;
931 }
932 
933 /* Convert from using the shortform to the leaf format. */
934 int
935 xfs_attr_shortform_to_leaf(
936 	struct xfs_da_args		*args)
937 {
938 	struct xfs_inode		*dp;
939 	struct xfs_attr_shortform	*sf;
940 	struct xfs_attr_sf_entry	*sfe;
941 	struct xfs_da_args		nargs;
942 	char				*tmpbuffer;
943 	int				error, i, size;
944 	xfs_dablk_t			blkno;
945 	struct xfs_buf			*bp;
946 	struct xfs_ifork		*ifp;
947 
948 	trace_xfs_attr_sf_to_leaf(args);
949 
950 	dp = args->dp;
951 	ifp = dp->i_afp;
952 	sf = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
953 	size = be16_to_cpu(sf->hdr.totsize);
954 	tmpbuffer = kmem_alloc(size, 0);
955 	ASSERT(tmpbuffer != NULL);
956 	memcpy(tmpbuffer, ifp->if_u1.if_data, size);
957 	sf = (struct xfs_attr_shortform *)tmpbuffer;
958 
959 	xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
960 	xfs_bmap_local_to_extents_empty(args->trans, dp, XFS_ATTR_FORK);
961 
962 	bp = NULL;
963 	error = xfs_da_grow_inode(args, &blkno);
964 	if (error)
965 		goto out;
966 
967 	ASSERT(blkno == 0);
968 	error = xfs_attr3_leaf_create(args, blkno, &bp);
969 	if (error)
970 		goto out;
971 
972 	memset((char *)&nargs, 0, sizeof(nargs));
973 	nargs.dp = dp;
974 	nargs.geo = args->geo;
975 	nargs.total = args->total;
976 	nargs.whichfork = XFS_ATTR_FORK;
977 	nargs.trans = args->trans;
978 	nargs.op_flags = XFS_DA_OP_OKNOENT;
979 
980 	sfe = &sf->list[0];
981 	for (i = 0; i < sf->hdr.count; i++) {
982 		nargs.name = sfe->nameval;
983 		nargs.namelen = sfe->namelen;
984 		nargs.value = &sfe->nameval[nargs.namelen];
985 		nargs.valuelen = sfe->valuelen;
986 		nargs.hashval = xfs_da_hashname(sfe->nameval,
987 						sfe->namelen);
988 		nargs.attr_filter = sfe->flags & XFS_ATTR_NSP_ONDISK_MASK;
989 		error = xfs_attr3_leaf_lookup_int(bp, &nargs); /* set a->index */
990 		ASSERT(error == -ENOATTR);
991 		error = xfs_attr3_leaf_add(bp, &nargs);
992 		ASSERT(error != -ENOSPC);
993 		if (error)
994 			goto out;
995 		sfe = xfs_attr_sf_nextentry(sfe);
996 	}
997 	error = 0;
998 out:
999 	kmem_free(tmpbuffer);
1000 	return error;
1001 }
1002 
1003 /*
1004  * Check a leaf attribute block to see if all the entries would fit into
1005  * a shortform attribute list.
1006  */
1007 int
1008 xfs_attr_shortform_allfit(
1009 	struct xfs_buf		*bp,
1010 	struct xfs_inode	*dp)
1011 {
1012 	struct xfs_attr_leafblock *leaf;
1013 	struct xfs_attr_leaf_entry *entry;
1014 	xfs_attr_leaf_name_local_t *name_loc;
1015 	struct xfs_attr3_icleaf_hdr leafhdr;
1016 	int			bytes;
1017 	int			i;
1018 	struct xfs_mount	*mp = bp->b_mount;
1019 
1020 	leaf = bp->b_addr;
1021 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf);
1022 	entry = xfs_attr3_leaf_entryp(leaf);
1023 
1024 	bytes = sizeof(struct xfs_attr_sf_hdr);
1025 	for (i = 0; i < leafhdr.count; entry++, i++) {
1026 		if (entry->flags & XFS_ATTR_INCOMPLETE)
1027 			continue;		/* don't copy partial entries */
1028 		if (!(entry->flags & XFS_ATTR_LOCAL))
1029 			return 0;
1030 		name_loc = xfs_attr3_leaf_name_local(leaf, i);
1031 		if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX)
1032 			return 0;
1033 		if (be16_to_cpu(name_loc->valuelen) >= XFS_ATTR_SF_ENTSIZE_MAX)
1034 			return 0;
1035 		bytes += xfs_attr_sf_entsize_byname(name_loc->namelen,
1036 					be16_to_cpu(name_loc->valuelen));
1037 	}
1038 	if (xfs_has_attr2(dp->i_mount) &&
1039 	    (dp->i_df.if_format != XFS_DINODE_FMT_BTREE) &&
1040 	    (bytes == sizeof(struct xfs_attr_sf_hdr)))
1041 		return -1;
1042 	return xfs_attr_shortform_bytesfit(dp, bytes);
1043 }
1044 
1045 /* Verify the consistency of an inline attribute fork. */
1046 xfs_failaddr_t
1047 xfs_attr_shortform_verify(
1048 	struct xfs_inode		*ip)
1049 {
1050 	struct xfs_attr_shortform	*sfp;
1051 	struct xfs_attr_sf_entry	*sfep;
1052 	struct xfs_attr_sf_entry	*next_sfep;
1053 	char				*endp;
1054 	struct xfs_ifork		*ifp;
1055 	int				i;
1056 	int64_t				size;
1057 
1058 	ASSERT(ip->i_afp->if_format == XFS_DINODE_FMT_LOCAL);
1059 	ifp = XFS_IFORK_PTR(ip, XFS_ATTR_FORK);
1060 	sfp = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
1061 	size = ifp->if_bytes;
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 	sfep = &sfp->list[0];
1073 	for (i = 0; i < sfp->hdr.count; i++) {
1074 		/*
1075 		 * struct xfs_attr_sf_entry has a variable length.
1076 		 * Check the fixed-offset parts of the structure are
1077 		 * within the data buffer.
1078 		 * xfs_attr_sf_entry is defined with a 1-byte variable
1079 		 * array at the end, so we must subtract that off.
1080 		 */
1081 		if (((char *)sfep + sizeof(*sfep)) >= endp)
1082 			return __this_address;
1083 
1084 		/* Don't allow names with known bad length. */
1085 		if (sfep->namelen == 0)
1086 			return __this_address;
1087 
1088 		/*
1089 		 * Check that the variable-length part of the structure is
1090 		 * within the data buffer.  The next entry starts after the
1091 		 * name component, so nextentry is an acceptable test.
1092 		 */
1093 		next_sfep = xfs_attr_sf_nextentry(sfep);
1094 		if ((char *)next_sfep > endp)
1095 			return __this_address;
1096 
1097 		/*
1098 		 * Check for unknown flags.  Short form doesn't support
1099 		 * the incomplete or local bits, so we can use the namespace
1100 		 * mask here.
1101 		 */
1102 		if (sfep->flags & ~XFS_ATTR_NSP_ONDISK_MASK)
1103 			return __this_address;
1104 
1105 		/*
1106 		 * Check for invalid namespace combinations.  We only allow
1107 		 * one namespace flag per xattr, so we can just count the
1108 		 * bits (i.e. hweight) here.
1109 		 */
1110 		if (hweight8(sfep->flags & XFS_ATTR_NSP_ONDISK_MASK) > 1)
1111 			return __this_address;
1112 
1113 		sfep = next_sfep;
1114 	}
1115 	if ((void *)sfep != (void *)endp)
1116 		return __this_address;
1117 
1118 	return NULL;
1119 }
1120 
1121 /*
1122  * Convert a leaf attribute list to shortform attribute list
1123  */
1124 int
1125 xfs_attr3_leaf_to_shortform(
1126 	struct xfs_buf		*bp,
1127 	struct xfs_da_args	*args,
1128 	int			forkoff)
1129 {
1130 	struct xfs_attr_leafblock *leaf;
1131 	struct xfs_attr3_icleaf_hdr ichdr;
1132 	struct xfs_attr_leaf_entry *entry;
1133 	struct xfs_attr_leaf_name_local *name_loc;
1134 	struct xfs_da_args	nargs;
1135 	struct xfs_inode	*dp = args->dp;
1136 	char			*tmpbuffer;
1137 	int			error;
1138 	int			i;
1139 
1140 	trace_xfs_attr_leaf_to_sf(args);
1141 
1142 	tmpbuffer = kmem_alloc(args->geo->blksize, 0);
1143 	if (!tmpbuffer)
1144 		return -ENOMEM;
1145 
1146 	memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1147 
1148 	leaf = (xfs_attr_leafblock_t *)tmpbuffer;
1149 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1150 	entry = xfs_attr3_leaf_entryp(leaf);
1151 
1152 	/* XXX (dgc): buffer is about to be marked stale - why zero it? */
1153 	memset(bp->b_addr, 0, args->geo->blksize);
1154 
1155 	/*
1156 	 * Clean out the prior contents of the attribute list.
1157 	 */
1158 	error = xfs_da_shrink_inode(args, 0, bp);
1159 	if (error)
1160 		goto out;
1161 
1162 	if (forkoff == -1) {
1163 		/*
1164 		 * Don't remove the attr fork if this operation is the first
1165 		 * part of a attr replace operations. We're going to add a new
1166 		 * attr immediately, so we need to keep the attr fork around in
1167 		 * this case.
1168 		 */
1169 		if (!(args->op_flags & XFS_DA_OP_REPLACE)) {
1170 			ASSERT(xfs_has_attr2(dp->i_mount));
1171 			ASSERT(dp->i_df.if_format != XFS_DINODE_FMT_BTREE);
1172 			xfs_attr_fork_remove(dp, args->trans);
1173 		}
1174 		goto out;
1175 	}
1176 
1177 	xfs_attr_shortform_create(args);
1178 
1179 	/*
1180 	 * Copy the attributes
1181 	 */
1182 	memset((char *)&nargs, 0, sizeof(nargs));
1183 	nargs.geo = args->geo;
1184 	nargs.dp = dp;
1185 	nargs.total = args->total;
1186 	nargs.whichfork = XFS_ATTR_FORK;
1187 	nargs.trans = args->trans;
1188 	nargs.op_flags = XFS_DA_OP_OKNOENT;
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 	kmem_free(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, 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 	/* copy leaf to new buffer, update identifiers */
1250 	xfs_trans_buf_set_type(args->trans, bp2, XFS_BLFT_ATTR_LEAF_BUF);
1251 	bp2->b_ops = bp1->b_ops;
1252 	memcpy(bp2->b_addr, bp1->b_addr, args->geo->blksize);
1253 	if (xfs_has_crc(mp)) {
1254 		struct xfs_da3_blkinfo *hdr3 = bp2->b_addr;
1255 		hdr3->blkno = cpu_to_be64(xfs_buf_daddr(bp2));
1256 	}
1257 	xfs_trans_log_buf(args->trans, bp2, 0, args->geo->blksize - 1);
1258 
1259 	/*
1260 	 * Set up the new root node.
1261 	 */
1262 	error = xfs_da3_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK);
1263 	if (error)
1264 		goto out;
1265 	node = bp1->b_addr;
1266 	xfs_da3_node_hdr_from_disk(mp, &icnodehdr, node);
1267 
1268 	leaf = bp2->b_addr;
1269 	xfs_attr3_leaf_hdr_from_disk(args->geo, &icleafhdr, leaf);
1270 	entries = xfs_attr3_leaf_entryp(leaf);
1271 
1272 	/* both on-disk, don't endian-flip twice */
1273 	icnodehdr.btree[0].hashval = entries[icleafhdr.count - 1].hashval;
1274 	icnodehdr.btree[0].before = cpu_to_be32(blkno);
1275 	icnodehdr.count = 1;
1276 	xfs_da3_node_hdr_to_disk(dp->i_mount, node, &icnodehdr);
1277 	xfs_trans_log_buf(args->trans, bp1, 0, args->geo->blksize - 1);
1278 	error = 0;
1279 out:
1280 	return error;
1281 }
1282 
1283 /*========================================================================
1284  * Routines used for growing the Btree.
1285  *========================================================================*/
1286 
1287 /*
1288  * Create the initial contents of a leaf attribute list
1289  * or a leaf in a node attribute list.
1290  */
1291 STATIC int
1292 xfs_attr3_leaf_create(
1293 	struct xfs_da_args	*args,
1294 	xfs_dablk_t		blkno,
1295 	struct xfs_buf		**bpp)
1296 {
1297 	struct xfs_attr_leafblock *leaf;
1298 	struct xfs_attr3_icleaf_hdr ichdr;
1299 	struct xfs_inode	*dp = args->dp;
1300 	struct xfs_mount	*mp = dp->i_mount;
1301 	struct xfs_buf		*bp;
1302 	int			error;
1303 
1304 	trace_xfs_attr_leaf_create(args);
1305 
1306 	error = xfs_da_get_buf(args->trans, args->dp, blkno, &bp,
1307 					    XFS_ATTR_FORK);
1308 	if (error)
1309 		return error;
1310 	bp->b_ops = &xfs_attr3_leaf_buf_ops;
1311 	xfs_trans_buf_set_type(args->trans, bp, XFS_BLFT_ATTR_LEAF_BUF);
1312 	leaf = bp->b_addr;
1313 	memset(leaf, 0, args->geo->blksize);
1314 
1315 	memset(&ichdr, 0, sizeof(ichdr));
1316 	ichdr.firstused = args->geo->blksize;
1317 
1318 	if (xfs_has_crc(mp)) {
1319 		struct xfs_da3_blkinfo *hdr3 = bp->b_addr;
1320 
1321 		ichdr.magic = XFS_ATTR3_LEAF_MAGIC;
1322 
1323 		hdr3->blkno = cpu_to_be64(xfs_buf_daddr(bp));
1324 		hdr3->owner = cpu_to_be64(dp->i_ino);
1325 		uuid_copy(&hdr3->uuid, &mp->m_sb.sb_meta_uuid);
1326 
1327 		ichdr.freemap[0].base = sizeof(struct xfs_attr3_leaf_hdr);
1328 	} else {
1329 		ichdr.magic = XFS_ATTR_LEAF_MAGIC;
1330 		ichdr.freemap[0].base = sizeof(struct xfs_attr_leaf_hdr);
1331 	}
1332 	ichdr.freemap[0].size = ichdr.firstused - ichdr.freemap[0].base;
1333 
1334 	xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1335 	xfs_trans_log_buf(args->trans, bp, 0, args->geo->blksize - 1);
1336 
1337 	*bpp = bp;
1338 	return 0;
1339 }
1340 
1341 /*
1342  * Split the leaf node, rebalance, then add the new entry.
1343  */
1344 int
1345 xfs_attr3_leaf_split(
1346 	struct xfs_da_state	*state,
1347 	struct xfs_da_state_blk	*oldblk,
1348 	struct xfs_da_state_blk	*newblk)
1349 {
1350 	xfs_dablk_t blkno;
1351 	int error;
1352 
1353 	trace_xfs_attr_leaf_split(state->args);
1354 
1355 	/*
1356 	 * Allocate space for a new leaf node.
1357 	 */
1358 	ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
1359 	error = xfs_da_grow_inode(state->args, &blkno);
1360 	if (error)
1361 		return error;
1362 	error = xfs_attr3_leaf_create(state->args, blkno, &newblk->bp);
1363 	if (error)
1364 		return error;
1365 	newblk->blkno = blkno;
1366 	newblk->magic = XFS_ATTR_LEAF_MAGIC;
1367 
1368 	/*
1369 	 * Rebalance the entries across the two leaves.
1370 	 * NOTE: rebalance() currently depends on the 2nd block being empty.
1371 	 */
1372 	xfs_attr3_leaf_rebalance(state, oldblk, newblk);
1373 	error = xfs_da3_blk_link(state, oldblk, newblk);
1374 	if (error)
1375 		return error;
1376 
1377 	/*
1378 	 * Save info on "old" attribute for "atomic rename" ops, leaf_add()
1379 	 * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
1380 	 * "new" attrs info.  Will need the "old" info to remove it later.
1381 	 *
1382 	 * Insert the "new" entry in the correct block.
1383 	 */
1384 	if (state->inleaf) {
1385 		trace_xfs_attr_leaf_add_old(state->args);
1386 		error = xfs_attr3_leaf_add(oldblk->bp, state->args);
1387 	} else {
1388 		trace_xfs_attr_leaf_add_new(state->args);
1389 		error = xfs_attr3_leaf_add(newblk->bp, state->args);
1390 	}
1391 
1392 	/*
1393 	 * Update last hashval in each block since we added the name.
1394 	 */
1395 	oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
1396 	newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
1397 	return error;
1398 }
1399 
1400 /*
1401  * Add a name to the leaf attribute list structure.
1402  */
1403 int
1404 xfs_attr3_leaf_add(
1405 	struct xfs_buf		*bp,
1406 	struct xfs_da_args	*args)
1407 {
1408 	struct xfs_attr_leafblock *leaf;
1409 	struct xfs_attr3_icleaf_hdr ichdr;
1410 	int			tablesize;
1411 	int			entsize;
1412 	int			sum;
1413 	int			tmp;
1414 	int			i;
1415 
1416 	trace_xfs_attr_leaf_add(args);
1417 
1418 	leaf = bp->b_addr;
1419 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1420 	ASSERT(args->index >= 0 && args->index <= ichdr.count);
1421 	entsize = xfs_attr_leaf_newentsize(args, NULL);
1422 
1423 	/*
1424 	 * Search through freemap for first-fit on new name length.
1425 	 * (may need to figure in size of entry struct too)
1426 	 */
1427 	tablesize = (ichdr.count + 1) * sizeof(xfs_attr_leaf_entry_t)
1428 					+ xfs_attr3_leaf_hdr_size(leaf);
1429 	for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE - 1; i >= 0; i--) {
1430 		if (tablesize > ichdr.firstused) {
1431 			sum += ichdr.freemap[i].size;
1432 			continue;
1433 		}
1434 		if (!ichdr.freemap[i].size)
1435 			continue;	/* no space in this map */
1436 		tmp = entsize;
1437 		if (ichdr.freemap[i].base < ichdr.firstused)
1438 			tmp += sizeof(xfs_attr_leaf_entry_t);
1439 		if (ichdr.freemap[i].size >= tmp) {
1440 			tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, i);
1441 			goto out_log_hdr;
1442 		}
1443 		sum += ichdr.freemap[i].size;
1444 	}
1445 
1446 	/*
1447 	 * If there are no holes in the address space of the block,
1448 	 * and we don't have enough freespace, then compaction will do us
1449 	 * no good and we should just give up.
1450 	 */
1451 	if (!ichdr.holes && sum < entsize)
1452 		return -ENOSPC;
1453 
1454 	/*
1455 	 * Compact the entries to coalesce free space.
1456 	 * This may change the hdr->count via dropping INCOMPLETE entries.
1457 	 */
1458 	xfs_attr3_leaf_compact(args, &ichdr, bp);
1459 
1460 	/*
1461 	 * After compaction, the block is guaranteed to have only one
1462 	 * free region, in freemap[0].  If it is not big enough, give up.
1463 	 */
1464 	if (ichdr.freemap[0].size < (entsize + sizeof(xfs_attr_leaf_entry_t))) {
1465 		tmp = -ENOSPC;
1466 		goto out_log_hdr;
1467 	}
1468 
1469 	tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, 0);
1470 
1471 out_log_hdr:
1472 	xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1473 	xfs_trans_log_buf(args->trans, bp,
1474 		XFS_DA_LOGRANGE(leaf, &leaf->hdr,
1475 				xfs_attr3_leaf_hdr_size(leaf)));
1476 	return tmp;
1477 }
1478 
1479 /*
1480  * Add a name to a leaf attribute list structure.
1481  */
1482 STATIC int
1483 xfs_attr3_leaf_add_work(
1484 	struct xfs_buf		*bp,
1485 	struct xfs_attr3_icleaf_hdr *ichdr,
1486 	struct xfs_da_args	*args,
1487 	int			mapindex)
1488 {
1489 	struct xfs_attr_leafblock *leaf;
1490 	struct xfs_attr_leaf_entry *entry;
1491 	struct xfs_attr_leaf_name_local *name_loc;
1492 	struct xfs_attr_leaf_name_remote *name_rmt;
1493 	struct xfs_mount	*mp;
1494 	int			tmp;
1495 	int			i;
1496 
1497 	trace_xfs_attr_leaf_add_work(args);
1498 
1499 	leaf = bp->b_addr;
1500 	ASSERT(mapindex >= 0 && mapindex < XFS_ATTR_LEAF_MAPSIZE);
1501 	ASSERT(args->index >= 0 && args->index <= ichdr->count);
1502 
1503 	/*
1504 	 * Force open some space in the entry array and fill it in.
1505 	 */
1506 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1507 	if (args->index < ichdr->count) {
1508 		tmp  = ichdr->count - args->index;
1509 		tmp *= sizeof(xfs_attr_leaf_entry_t);
1510 		memmove(entry + 1, entry, tmp);
1511 		xfs_trans_log_buf(args->trans, bp,
1512 		    XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1513 	}
1514 	ichdr->count++;
1515 
1516 	/*
1517 	 * Allocate space for the new string (at the end of the run).
1518 	 */
1519 	mp = args->trans->t_mountp;
1520 	ASSERT(ichdr->freemap[mapindex].base < args->geo->blksize);
1521 	ASSERT((ichdr->freemap[mapindex].base & 0x3) == 0);
1522 	ASSERT(ichdr->freemap[mapindex].size >=
1523 		xfs_attr_leaf_newentsize(args, NULL));
1524 	ASSERT(ichdr->freemap[mapindex].size < args->geo->blksize);
1525 	ASSERT((ichdr->freemap[mapindex].size & 0x3) == 0);
1526 
1527 	ichdr->freemap[mapindex].size -= xfs_attr_leaf_newentsize(args, &tmp);
1528 
1529 	entry->nameidx = cpu_to_be16(ichdr->freemap[mapindex].base +
1530 				     ichdr->freemap[mapindex].size);
1531 	entry->hashval = cpu_to_be32(args->hashval);
1532 	entry->flags = args->attr_filter;
1533 	if (tmp)
1534 		entry->flags |= XFS_ATTR_LOCAL;
1535 	if (args->op_flags & XFS_DA_OP_REPLACE) {
1536 		if (!(args->op_flags & XFS_DA_OP_LOGGED))
1537 			entry->flags |= XFS_ATTR_INCOMPLETE;
1538 		if ((args->blkno2 == args->blkno) &&
1539 		    (args->index2 <= args->index)) {
1540 			args->index2++;
1541 		}
1542 	}
1543 	xfs_trans_log_buf(args->trans, bp,
1544 			  XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1545 	ASSERT((args->index == 0) ||
1546 	       (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1547 	ASSERT((args->index == ichdr->count - 1) ||
1548 	       (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1549 
1550 	/*
1551 	 * For "remote" attribute values, simply note that we need to
1552 	 * allocate space for the "remote" value.  We can't actually
1553 	 * allocate the extents in this transaction, and we can't decide
1554 	 * which blocks they should be as we might allocate more blocks
1555 	 * as part of this transaction (a split operation for example).
1556 	 */
1557 	if (entry->flags & XFS_ATTR_LOCAL) {
1558 		name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
1559 		name_loc->namelen = args->namelen;
1560 		name_loc->valuelen = cpu_to_be16(args->valuelen);
1561 		memcpy((char *)name_loc->nameval, args->name, args->namelen);
1562 		memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1563 				   be16_to_cpu(name_loc->valuelen));
1564 	} else {
1565 		name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
1566 		name_rmt->namelen = args->namelen;
1567 		memcpy((char *)name_rmt->name, args->name, args->namelen);
1568 		entry->flags |= XFS_ATTR_INCOMPLETE;
1569 		/* just in case */
1570 		name_rmt->valuelen = 0;
1571 		name_rmt->valueblk = 0;
1572 		args->rmtblkno = 1;
1573 		args->rmtblkcnt = xfs_attr3_rmt_blocks(mp, args->valuelen);
1574 		args->rmtvaluelen = args->valuelen;
1575 	}
1576 	xfs_trans_log_buf(args->trans, bp,
1577 	     XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1578 				   xfs_attr_leaf_entsize(leaf, args->index)));
1579 
1580 	/*
1581 	 * Update the control info for this leaf node
1582 	 */
1583 	if (be16_to_cpu(entry->nameidx) < ichdr->firstused)
1584 		ichdr->firstused = be16_to_cpu(entry->nameidx);
1585 
1586 	ASSERT(ichdr->firstused >= ichdr->count * sizeof(xfs_attr_leaf_entry_t)
1587 					+ xfs_attr3_leaf_hdr_size(leaf));
1588 	tmp = (ichdr->count - 1) * sizeof(xfs_attr_leaf_entry_t)
1589 					+ xfs_attr3_leaf_hdr_size(leaf);
1590 
1591 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1592 		if (ichdr->freemap[i].base == tmp) {
1593 			ichdr->freemap[i].base += sizeof(xfs_attr_leaf_entry_t);
1594 			ichdr->freemap[i].size -=
1595 				min_t(uint16_t, ichdr->freemap[i].size,
1596 						sizeof(xfs_attr_leaf_entry_t));
1597 		}
1598 	}
1599 	ichdr->usedbytes += xfs_attr_leaf_entsize(leaf, args->index);
1600 	return 0;
1601 }
1602 
1603 /*
1604  * Garbage collect a leaf attribute list block by copying it to a new buffer.
1605  */
1606 STATIC void
1607 xfs_attr3_leaf_compact(
1608 	struct xfs_da_args	*args,
1609 	struct xfs_attr3_icleaf_hdr *ichdr_dst,
1610 	struct xfs_buf		*bp)
1611 {
1612 	struct xfs_attr_leafblock *leaf_src;
1613 	struct xfs_attr_leafblock *leaf_dst;
1614 	struct xfs_attr3_icleaf_hdr ichdr_src;
1615 	struct xfs_trans	*trans = args->trans;
1616 	char			*tmpbuffer;
1617 
1618 	trace_xfs_attr_leaf_compact(args);
1619 
1620 	tmpbuffer = kmem_alloc(args->geo->blksize, 0);
1621 	memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1622 	memset(bp->b_addr, 0, args->geo->blksize);
1623 	leaf_src = (xfs_attr_leafblock_t *)tmpbuffer;
1624 	leaf_dst = bp->b_addr;
1625 
1626 	/*
1627 	 * Copy the on-disk header back into the destination buffer to ensure
1628 	 * all the information in the header that is not part of the incore
1629 	 * header structure is preserved.
1630 	 */
1631 	memcpy(bp->b_addr, tmpbuffer, xfs_attr3_leaf_hdr_size(leaf_src));
1632 
1633 	/* Initialise the incore headers */
1634 	ichdr_src = *ichdr_dst;	/* struct copy */
1635 	ichdr_dst->firstused = args->geo->blksize;
1636 	ichdr_dst->usedbytes = 0;
1637 	ichdr_dst->count = 0;
1638 	ichdr_dst->holes = 0;
1639 	ichdr_dst->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_src);
1640 	ichdr_dst->freemap[0].size = ichdr_dst->firstused -
1641 						ichdr_dst->freemap[0].base;
1642 
1643 	/* write the header back to initialise the underlying buffer */
1644 	xfs_attr3_leaf_hdr_to_disk(args->geo, leaf_dst, ichdr_dst);
1645 
1646 	/*
1647 	 * Copy all entry's in the same (sorted) order,
1648 	 * but allocate name/value pairs packed and in sequence.
1649 	 */
1650 	xfs_attr3_leaf_moveents(args, leaf_src, &ichdr_src, 0,
1651 				leaf_dst, ichdr_dst, 0, ichdr_src.count);
1652 	/*
1653 	 * this logs the entire buffer, but the caller must write the header
1654 	 * back to the buffer when it is finished modifying it.
1655 	 */
1656 	xfs_trans_log_buf(trans, bp, 0, args->geo->blksize - 1);
1657 
1658 	kmem_free(tmpbuffer);
1659 }
1660 
1661 /*
1662  * Compare two leaf blocks "order".
1663  * Return 0 unless leaf2 should go before leaf1.
1664  */
1665 static int
1666 xfs_attr3_leaf_order(
1667 	struct xfs_buf	*leaf1_bp,
1668 	struct xfs_attr3_icleaf_hdr *leaf1hdr,
1669 	struct xfs_buf	*leaf2_bp,
1670 	struct xfs_attr3_icleaf_hdr *leaf2hdr)
1671 {
1672 	struct xfs_attr_leaf_entry *entries1;
1673 	struct xfs_attr_leaf_entry *entries2;
1674 
1675 	entries1 = xfs_attr3_leaf_entryp(leaf1_bp->b_addr);
1676 	entries2 = xfs_attr3_leaf_entryp(leaf2_bp->b_addr);
1677 	if (leaf1hdr->count > 0 && leaf2hdr->count > 0 &&
1678 	    ((be32_to_cpu(entries2[0].hashval) <
1679 	      be32_to_cpu(entries1[0].hashval)) ||
1680 	     (be32_to_cpu(entries2[leaf2hdr->count - 1].hashval) <
1681 	      be32_to_cpu(entries1[leaf1hdr->count - 1].hashval)))) {
1682 		return 1;
1683 	}
1684 	return 0;
1685 }
1686 
1687 int
1688 xfs_attr_leaf_order(
1689 	struct xfs_buf	*leaf1_bp,
1690 	struct xfs_buf	*leaf2_bp)
1691 {
1692 	struct xfs_attr3_icleaf_hdr ichdr1;
1693 	struct xfs_attr3_icleaf_hdr ichdr2;
1694 	struct xfs_mount *mp = leaf1_bp->b_mount;
1695 
1696 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr1, leaf1_bp->b_addr);
1697 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr2, leaf2_bp->b_addr);
1698 	return xfs_attr3_leaf_order(leaf1_bp, &ichdr1, leaf2_bp, &ichdr2);
1699 }
1700 
1701 /*
1702  * Redistribute the attribute list entries between two leaf nodes,
1703  * taking into account the size of the new entry.
1704  *
1705  * NOTE: if new block is empty, then it will get the upper half of the
1706  * old block.  At present, all (one) callers pass in an empty second block.
1707  *
1708  * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1709  * to match what it is doing in splitting the attribute leaf block.  Those
1710  * values are used in "atomic rename" operations on attributes.  Note that
1711  * the "new" and "old" values can end up in different blocks.
1712  */
1713 STATIC void
1714 xfs_attr3_leaf_rebalance(
1715 	struct xfs_da_state	*state,
1716 	struct xfs_da_state_blk	*blk1,
1717 	struct xfs_da_state_blk	*blk2)
1718 {
1719 	struct xfs_da_args	*args;
1720 	struct xfs_attr_leafblock *leaf1;
1721 	struct xfs_attr_leafblock *leaf2;
1722 	struct xfs_attr3_icleaf_hdr ichdr1;
1723 	struct xfs_attr3_icleaf_hdr ichdr2;
1724 	struct xfs_attr_leaf_entry *entries1;
1725 	struct xfs_attr_leaf_entry *entries2;
1726 	int			count;
1727 	int			totallen;
1728 	int			max;
1729 	int			space;
1730 	int			swap;
1731 
1732 	/*
1733 	 * Set up environment.
1734 	 */
1735 	ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1736 	ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1737 	leaf1 = blk1->bp->b_addr;
1738 	leaf2 = blk2->bp->b_addr;
1739 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr1, leaf1);
1740 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, leaf2);
1741 	ASSERT(ichdr2.count == 0);
1742 	args = state->args;
1743 
1744 	trace_xfs_attr_leaf_rebalance(args);
1745 
1746 	/*
1747 	 * Check ordering of blocks, reverse if it makes things simpler.
1748 	 *
1749 	 * NOTE: Given that all (current) callers pass in an empty
1750 	 * second block, this code should never set "swap".
1751 	 */
1752 	swap = 0;
1753 	if (xfs_attr3_leaf_order(blk1->bp, &ichdr1, blk2->bp, &ichdr2)) {
1754 		swap(blk1, blk2);
1755 
1756 		/* swap structures rather than reconverting them */
1757 		swap(ichdr1, ichdr2);
1758 
1759 		leaf1 = blk1->bp->b_addr;
1760 		leaf2 = blk2->bp->b_addr;
1761 		swap = 1;
1762 	}
1763 
1764 	/*
1765 	 * Examine entries until we reduce the absolute difference in
1766 	 * byte usage between the two blocks to a minimum.  Then get
1767 	 * the direction to copy and the number of elements to move.
1768 	 *
1769 	 * "inleaf" is true if the new entry should be inserted into blk1.
1770 	 * If "swap" is also true, then reverse the sense of "inleaf".
1771 	 */
1772 	state->inleaf = xfs_attr3_leaf_figure_balance(state, blk1, &ichdr1,
1773 						      blk2, &ichdr2,
1774 						      &count, &totallen);
1775 	if (swap)
1776 		state->inleaf = !state->inleaf;
1777 
1778 	/*
1779 	 * Move any entries required from leaf to leaf:
1780 	 */
1781 	if (count < ichdr1.count) {
1782 		/*
1783 		 * Figure the total bytes to be added to the destination leaf.
1784 		 */
1785 		/* number entries being moved */
1786 		count = ichdr1.count - count;
1787 		space  = ichdr1.usedbytes - totallen;
1788 		space += count * sizeof(xfs_attr_leaf_entry_t);
1789 
1790 		/*
1791 		 * leaf2 is the destination, compact it if it looks tight.
1792 		 */
1793 		max  = ichdr2.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1794 		max -= ichdr2.count * sizeof(xfs_attr_leaf_entry_t);
1795 		if (space > max)
1796 			xfs_attr3_leaf_compact(args, &ichdr2, blk2->bp);
1797 
1798 		/*
1799 		 * Move high entries from leaf1 to low end of leaf2.
1800 		 */
1801 		xfs_attr3_leaf_moveents(args, leaf1, &ichdr1,
1802 				ichdr1.count - count, leaf2, &ichdr2, 0, count);
1803 
1804 	} else if (count > ichdr1.count) {
1805 		/*
1806 		 * I assert that since all callers pass in an empty
1807 		 * second buffer, this code should never execute.
1808 		 */
1809 		ASSERT(0);
1810 
1811 		/*
1812 		 * Figure the total bytes to be added to the destination leaf.
1813 		 */
1814 		/* number entries being moved */
1815 		count -= ichdr1.count;
1816 		space  = totallen - ichdr1.usedbytes;
1817 		space += count * sizeof(xfs_attr_leaf_entry_t);
1818 
1819 		/*
1820 		 * leaf1 is the destination, compact it if it looks tight.
1821 		 */
1822 		max  = ichdr1.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1823 		max -= ichdr1.count * sizeof(xfs_attr_leaf_entry_t);
1824 		if (space > max)
1825 			xfs_attr3_leaf_compact(args, &ichdr1, blk1->bp);
1826 
1827 		/*
1828 		 * Move low entries from leaf2 to high end of leaf1.
1829 		 */
1830 		xfs_attr3_leaf_moveents(args, leaf2, &ichdr2, 0, leaf1, &ichdr1,
1831 					ichdr1.count, count);
1832 	}
1833 
1834 	xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf1, &ichdr1);
1835 	xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf2, &ichdr2);
1836 	xfs_trans_log_buf(args->trans, blk1->bp, 0, args->geo->blksize - 1);
1837 	xfs_trans_log_buf(args->trans, blk2->bp, 0, args->geo->blksize - 1);
1838 
1839 	/*
1840 	 * Copy out last hashval in each block for B-tree code.
1841 	 */
1842 	entries1 = xfs_attr3_leaf_entryp(leaf1);
1843 	entries2 = xfs_attr3_leaf_entryp(leaf2);
1844 	blk1->hashval = be32_to_cpu(entries1[ichdr1.count - 1].hashval);
1845 	blk2->hashval = be32_to_cpu(entries2[ichdr2.count - 1].hashval);
1846 
1847 	/*
1848 	 * Adjust the expected index for insertion.
1849 	 * NOTE: this code depends on the (current) situation that the
1850 	 * second block was originally empty.
1851 	 *
1852 	 * If the insertion point moved to the 2nd block, we must adjust
1853 	 * the index.  We must also track the entry just following the
1854 	 * new entry for use in an "atomic rename" operation, that entry
1855 	 * is always the "old" entry and the "new" entry is what we are
1856 	 * inserting.  The index/blkno fields refer to the "old" entry,
1857 	 * while the index2/blkno2 fields refer to the "new" entry.
1858 	 */
1859 	if (blk1->index > ichdr1.count) {
1860 		ASSERT(state->inleaf == 0);
1861 		blk2->index = blk1->index - ichdr1.count;
1862 		args->index = args->index2 = blk2->index;
1863 		args->blkno = args->blkno2 = blk2->blkno;
1864 	} else if (blk1->index == ichdr1.count) {
1865 		if (state->inleaf) {
1866 			args->index = blk1->index;
1867 			args->blkno = blk1->blkno;
1868 			args->index2 = 0;
1869 			args->blkno2 = blk2->blkno;
1870 		} else {
1871 			/*
1872 			 * On a double leaf split, the original attr location
1873 			 * is already stored in blkno2/index2, so don't
1874 			 * overwrite it overwise we corrupt the tree.
1875 			 */
1876 			blk2->index = blk1->index - ichdr1.count;
1877 			args->index = blk2->index;
1878 			args->blkno = blk2->blkno;
1879 			if (!state->extravalid) {
1880 				/*
1881 				 * set the new attr location to match the old
1882 				 * one and let the higher level split code
1883 				 * decide where in the leaf to place it.
1884 				 */
1885 				args->index2 = blk2->index;
1886 				args->blkno2 = blk2->blkno;
1887 			}
1888 		}
1889 	} else {
1890 		ASSERT(state->inleaf == 1);
1891 		args->index = args->index2 = blk1->index;
1892 		args->blkno = args->blkno2 = blk1->blkno;
1893 	}
1894 }
1895 
1896 /*
1897  * Examine entries until we reduce the absolute difference in
1898  * byte usage between the two blocks to a minimum.
1899  * GROT: Is this really necessary?  With other than a 512 byte blocksize,
1900  * GROT: there will always be enough room in either block for a new entry.
1901  * GROT: Do a double-split for this case?
1902  */
1903 STATIC int
1904 xfs_attr3_leaf_figure_balance(
1905 	struct xfs_da_state		*state,
1906 	struct xfs_da_state_blk		*blk1,
1907 	struct xfs_attr3_icleaf_hdr	*ichdr1,
1908 	struct xfs_da_state_blk		*blk2,
1909 	struct xfs_attr3_icleaf_hdr	*ichdr2,
1910 	int				*countarg,
1911 	int				*usedbytesarg)
1912 {
1913 	struct xfs_attr_leafblock	*leaf1 = blk1->bp->b_addr;
1914 	struct xfs_attr_leafblock	*leaf2 = blk2->bp->b_addr;
1915 	struct xfs_attr_leaf_entry	*entry;
1916 	int				count;
1917 	int				max;
1918 	int				index;
1919 	int				totallen = 0;
1920 	int				half;
1921 	int				lastdelta;
1922 	int				foundit = 0;
1923 	int				tmp;
1924 
1925 	/*
1926 	 * Examine entries until we reduce the absolute difference in
1927 	 * byte usage between the two blocks to a minimum.
1928 	 */
1929 	max = ichdr1->count + ichdr2->count;
1930 	half = (max + 1) * sizeof(*entry);
1931 	half += ichdr1->usedbytes + ichdr2->usedbytes +
1932 			xfs_attr_leaf_newentsize(state->args, NULL);
1933 	half /= 2;
1934 	lastdelta = state->args->geo->blksize;
1935 	entry = xfs_attr3_leaf_entryp(leaf1);
1936 	for (count = index = 0; count < max; entry++, index++, count++) {
1937 
1938 #define XFS_ATTR_ABS(A)	(((A) < 0) ? -(A) : (A))
1939 		/*
1940 		 * The new entry is in the first block, account for it.
1941 		 */
1942 		if (count == blk1->index) {
1943 			tmp = totallen + sizeof(*entry) +
1944 				xfs_attr_leaf_newentsize(state->args, NULL);
1945 			if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1946 				break;
1947 			lastdelta = XFS_ATTR_ABS(half - tmp);
1948 			totallen = tmp;
1949 			foundit = 1;
1950 		}
1951 
1952 		/*
1953 		 * Wrap around into the second block if necessary.
1954 		 */
1955 		if (count == ichdr1->count) {
1956 			leaf1 = leaf2;
1957 			entry = xfs_attr3_leaf_entryp(leaf1);
1958 			index = 0;
1959 		}
1960 
1961 		/*
1962 		 * Figure out if next leaf entry would be too much.
1963 		 */
1964 		tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1965 									index);
1966 		if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1967 			break;
1968 		lastdelta = XFS_ATTR_ABS(half - tmp);
1969 		totallen = tmp;
1970 #undef XFS_ATTR_ABS
1971 	}
1972 
1973 	/*
1974 	 * Calculate the number of usedbytes that will end up in lower block.
1975 	 * If new entry not in lower block, fix up the count.
1976 	 */
1977 	totallen -= count * sizeof(*entry);
1978 	if (foundit) {
1979 		totallen -= sizeof(*entry) +
1980 				xfs_attr_leaf_newentsize(state->args, NULL);
1981 	}
1982 
1983 	*countarg = count;
1984 	*usedbytesarg = totallen;
1985 	return foundit;
1986 }
1987 
1988 /*========================================================================
1989  * Routines used for shrinking the Btree.
1990  *========================================================================*/
1991 
1992 /*
1993  * Check a leaf block and its neighbors to see if the block should be
1994  * collapsed into one or the other neighbor.  Always keep the block
1995  * with the smaller block number.
1996  * If the current block is over 50% full, don't try to join it, return 0.
1997  * If the block is empty, fill in the state structure and return 2.
1998  * If it can be collapsed, fill in the state structure and return 1.
1999  * If nothing can be done, return 0.
2000  *
2001  * GROT: allow for INCOMPLETE entries in calculation.
2002  */
2003 int
2004 xfs_attr3_leaf_toosmall(
2005 	struct xfs_da_state	*state,
2006 	int			*action)
2007 {
2008 	struct xfs_attr_leafblock *leaf;
2009 	struct xfs_da_state_blk	*blk;
2010 	struct xfs_attr3_icleaf_hdr ichdr;
2011 	struct xfs_buf		*bp;
2012 	xfs_dablk_t		blkno;
2013 	int			bytes;
2014 	int			forward;
2015 	int			error;
2016 	int			retval;
2017 	int			i;
2018 
2019 	trace_xfs_attr_leaf_toosmall(state->args);
2020 
2021 	/*
2022 	 * Check for the degenerate case of the block being over 50% full.
2023 	 * If so, it's not worth even looking to see if we might be able
2024 	 * to coalesce with a sibling.
2025 	 */
2026 	blk = &state->path.blk[ state->path.active-1 ];
2027 	leaf = blk->bp->b_addr;
2028 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr, leaf);
2029 	bytes = xfs_attr3_leaf_hdr_size(leaf) +
2030 		ichdr.count * sizeof(xfs_attr_leaf_entry_t) +
2031 		ichdr.usedbytes;
2032 	if (bytes > (state->args->geo->blksize >> 1)) {
2033 		*action = 0;	/* blk over 50%, don't try to join */
2034 		return 0;
2035 	}
2036 
2037 	/*
2038 	 * Check for the degenerate case of the block being empty.
2039 	 * If the block is empty, we'll simply delete it, no need to
2040 	 * coalesce it with a sibling block.  We choose (arbitrarily)
2041 	 * to merge with the forward block unless it is NULL.
2042 	 */
2043 	if (ichdr.count == 0) {
2044 		/*
2045 		 * Make altpath point to the block we want to keep and
2046 		 * path point to the block we want to drop (this one).
2047 		 */
2048 		forward = (ichdr.forw != 0);
2049 		memcpy(&state->altpath, &state->path, sizeof(state->path));
2050 		error = xfs_da3_path_shift(state, &state->altpath, forward,
2051 						 0, &retval);
2052 		if (error)
2053 			return error;
2054 		if (retval) {
2055 			*action = 0;
2056 		} else {
2057 			*action = 2;
2058 		}
2059 		return 0;
2060 	}
2061 
2062 	/*
2063 	 * Examine each sibling block to see if we can coalesce with
2064 	 * at least 25% free space to spare.  We need to figure out
2065 	 * whether to merge with the forward or the backward block.
2066 	 * We prefer coalescing with the lower numbered sibling so as
2067 	 * to shrink an attribute list over time.
2068 	 */
2069 	/* start with smaller blk num */
2070 	forward = ichdr.forw < ichdr.back;
2071 	for (i = 0; i < 2; forward = !forward, i++) {
2072 		struct xfs_attr3_icleaf_hdr ichdr2;
2073 		if (forward)
2074 			blkno = ichdr.forw;
2075 		else
2076 			blkno = ichdr.back;
2077 		if (blkno == 0)
2078 			continue;
2079 		error = xfs_attr3_leaf_read(state->args->trans, state->args->dp,
2080 					blkno, &bp);
2081 		if (error)
2082 			return error;
2083 
2084 		xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, bp->b_addr);
2085 
2086 		bytes = state->args->geo->blksize -
2087 			(state->args->geo->blksize >> 2) -
2088 			ichdr.usedbytes - ichdr2.usedbytes -
2089 			((ichdr.count + ichdr2.count) *
2090 					sizeof(xfs_attr_leaf_entry_t)) -
2091 			xfs_attr3_leaf_hdr_size(leaf);
2092 
2093 		xfs_trans_brelse(state->args->trans, bp);
2094 		if (bytes >= 0)
2095 			break;	/* fits with at least 25% to spare */
2096 	}
2097 	if (i >= 2) {
2098 		*action = 0;
2099 		return 0;
2100 	}
2101 
2102 	/*
2103 	 * Make altpath point to the block we want to keep (the lower
2104 	 * numbered block) and path point to the block we want to drop.
2105 	 */
2106 	memcpy(&state->altpath, &state->path, sizeof(state->path));
2107 	if (blkno < blk->blkno) {
2108 		error = xfs_da3_path_shift(state, &state->altpath, forward,
2109 						 0, &retval);
2110 	} else {
2111 		error = xfs_da3_path_shift(state, &state->path, forward,
2112 						 0, &retval);
2113 	}
2114 	if (error)
2115 		return error;
2116 	if (retval) {
2117 		*action = 0;
2118 	} else {
2119 		*action = 1;
2120 	}
2121 	return 0;
2122 }
2123 
2124 /*
2125  * Remove a name from the leaf attribute list structure.
2126  *
2127  * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
2128  * If two leaves are 37% full, when combined they will leave 25% free.
2129  */
2130 int
2131 xfs_attr3_leaf_remove(
2132 	struct xfs_buf		*bp,
2133 	struct xfs_da_args	*args)
2134 {
2135 	struct xfs_attr_leafblock *leaf;
2136 	struct xfs_attr3_icleaf_hdr ichdr;
2137 	struct xfs_attr_leaf_entry *entry;
2138 	int			before;
2139 	int			after;
2140 	int			smallest;
2141 	int			entsize;
2142 	int			tablesize;
2143 	int			tmp;
2144 	int			i;
2145 
2146 	trace_xfs_attr_leaf_remove(args);
2147 
2148 	leaf = bp->b_addr;
2149 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2150 
2151 	ASSERT(ichdr.count > 0 && ichdr.count < args->geo->blksize / 8);
2152 	ASSERT(args->index >= 0 && args->index < ichdr.count);
2153 	ASSERT(ichdr.firstused >= ichdr.count * sizeof(*entry) +
2154 					xfs_attr3_leaf_hdr_size(leaf));
2155 
2156 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2157 
2158 	ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2159 	ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
2160 
2161 	/*
2162 	 * Scan through free region table:
2163 	 *    check for adjacency of free'd entry with an existing one,
2164 	 *    find smallest free region in case we need to replace it,
2165 	 *    adjust any map that borders the entry table,
2166 	 */
2167 	tablesize = ichdr.count * sizeof(xfs_attr_leaf_entry_t)
2168 					+ xfs_attr3_leaf_hdr_size(leaf);
2169 	tmp = ichdr.freemap[0].size;
2170 	before = after = -1;
2171 	smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
2172 	entsize = xfs_attr_leaf_entsize(leaf, args->index);
2173 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
2174 		ASSERT(ichdr.freemap[i].base < args->geo->blksize);
2175 		ASSERT(ichdr.freemap[i].size < args->geo->blksize);
2176 		if (ichdr.freemap[i].base == tablesize) {
2177 			ichdr.freemap[i].base -= sizeof(xfs_attr_leaf_entry_t);
2178 			ichdr.freemap[i].size += sizeof(xfs_attr_leaf_entry_t);
2179 		}
2180 
2181 		if (ichdr.freemap[i].base + ichdr.freemap[i].size ==
2182 				be16_to_cpu(entry->nameidx)) {
2183 			before = i;
2184 		} else if (ichdr.freemap[i].base ==
2185 				(be16_to_cpu(entry->nameidx) + entsize)) {
2186 			after = i;
2187 		} else if (ichdr.freemap[i].size < tmp) {
2188 			tmp = ichdr.freemap[i].size;
2189 			smallest = i;
2190 		}
2191 	}
2192 
2193 	/*
2194 	 * Coalesce adjacent freemap regions,
2195 	 * or replace the smallest region.
2196 	 */
2197 	if ((before >= 0) || (after >= 0)) {
2198 		if ((before >= 0) && (after >= 0)) {
2199 			ichdr.freemap[before].size += entsize;
2200 			ichdr.freemap[before].size += ichdr.freemap[after].size;
2201 			ichdr.freemap[after].base = 0;
2202 			ichdr.freemap[after].size = 0;
2203 		} else if (before >= 0) {
2204 			ichdr.freemap[before].size += entsize;
2205 		} else {
2206 			ichdr.freemap[after].base = be16_to_cpu(entry->nameidx);
2207 			ichdr.freemap[after].size += entsize;
2208 		}
2209 	} else {
2210 		/*
2211 		 * Replace smallest region (if it is smaller than free'd entry)
2212 		 */
2213 		if (ichdr.freemap[smallest].size < entsize) {
2214 			ichdr.freemap[smallest].base = be16_to_cpu(entry->nameidx);
2215 			ichdr.freemap[smallest].size = entsize;
2216 		}
2217 	}
2218 
2219 	/*
2220 	 * Did we remove the first entry?
2221 	 */
2222 	if (be16_to_cpu(entry->nameidx) == ichdr.firstused)
2223 		smallest = 1;
2224 	else
2225 		smallest = 0;
2226 
2227 	/*
2228 	 * Compress the remaining entries and zero out the removed stuff.
2229 	 */
2230 	memset(xfs_attr3_leaf_name(leaf, args->index), 0, entsize);
2231 	ichdr.usedbytes -= entsize;
2232 	xfs_trans_log_buf(args->trans, bp,
2233 	     XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
2234 				   entsize));
2235 
2236 	tmp = (ichdr.count - args->index) * sizeof(xfs_attr_leaf_entry_t);
2237 	memmove(entry, entry + 1, tmp);
2238 	ichdr.count--;
2239 	xfs_trans_log_buf(args->trans, bp,
2240 	    XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(xfs_attr_leaf_entry_t)));
2241 
2242 	entry = &xfs_attr3_leaf_entryp(leaf)[ichdr.count];
2243 	memset(entry, 0, sizeof(xfs_attr_leaf_entry_t));
2244 
2245 	/*
2246 	 * If we removed the first entry, re-find the first used byte
2247 	 * in the name area.  Note that if the entry was the "firstused",
2248 	 * then we don't have a "hole" in our block resulting from
2249 	 * removing the name.
2250 	 */
2251 	if (smallest) {
2252 		tmp = args->geo->blksize;
2253 		entry = xfs_attr3_leaf_entryp(leaf);
2254 		for (i = ichdr.count - 1; i >= 0; entry++, i--) {
2255 			ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2256 			ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
2257 
2258 			if (be16_to_cpu(entry->nameidx) < tmp)
2259 				tmp = be16_to_cpu(entry->nameidx);
2260 		}
2261 		ichdr.firstused = tmp;
2262 		ASSERT(ichdr.firstused != 0);
2263 	} else {
2264 		ichdr.holes = 1;	/* mark as needing compaction */
2265 	}
2266 	xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
2267 	xfs_trans_log_buf(args->trans, bp,
2268 			  XFS_DA_LOGRANGE(leaf, &leaf->hdr,
2269 					  xfs_attr3_leaf_hdr_size(leaf)));
2270 
2271 	/*
2272 	 * Check if leaf is less than 50% full, caller may want to
2273 	 * "join" the leaf with a sibling if so.
2274 	 */
2275 	tmp = ichdr.usedbytes + xfs_attr3_leaf_hdr_size(leaf) +
2276 	      ichdr.count * sizeof(xfs_attr_leaf_entry_t);
2277 
2278 	return tmp < args->geo->magicpct; /* leaf is < 37% full */
2279 }
2280 
2281 /*
2282  * Move all the attribute list entries from drop_leaf into save_leaf.
2283  */
2284 void
2285 xfs_attr3_leaf_unbalance(
2286 	struct xfs_da_state	*state,
2287 	struct xfs_da_state_blk	*drop_blk,
2288 	struct xfs_da_state_blk	*save_blk)
2289 {
2290 	struct xfs_attr_leafblock *drop_leaf = drop_blk->bp->b_addr;
2291 	struct xfs_attr_leafblock *save_leaf = save_blk->bp->b_addr;
2292 	struct xfs_attr3_icleaf_hdr drophdr;
2293 	struct xfs_attr3_icleaf_hdr savehdr;
2294 	struct xfs_attr_leaf_entry *entry;
2295 
2296 	trace_xfs_attr_leaf_unbalance(state->args);
2297 
2298 	drop_leaf = drop_blk->bp->b_addr;
2299 	save_leaf = save_blk->bp->b_addr;
2300 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &drophdr, drop_leaf);
2301 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &savehdr, save_leaf);
2302 	entry = xfs_attr3_leaf_entryp(drop_leaf);
2303 
2304 	/*
2305 	 * Save last hashval from dying block for later Btree fixup.
2306 	 */
2307 	drop_blk->hashval = be32_to_cpu(entry[drophdr.count - 1].hashval);
2308 
2309 	/*
2310 	 * Check if we need a temp buffer, or can we do it in place.
2311 	 * Note that we don't check "leaf" for holes because we will
2312 	 * always be dropping it, toosmall() decided that for us already.
2313 	 */
2314 	if (savehdr.holes == 0) {
2315 		/*
2316 		 * dest leaf has no holes, so we add there.  May need
2317 		 * to make some room in the entry array.
2318 		 */
2319 		if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2320 					 drop_blk->bp, &drophdr)) {
2321 			xfs_attr3_leaf_moveents(state->args,
2322 						drop_leaf, &drophdr, 0,
2323 						save_leaf, &savehdr, 0,
2324 						drophdr.count);
2325 		} else {
2326 			xfs_attr3_leaf_moveents(state->args,
2327 						drop_leaf, &drophdr, 0,
2328 						save_leaf, &savehdr,
2329 						savehdr.count, drophdr.count);
2330 		}
2331 	} else {
2332 		/*
2333 		 * Destination has holes, so we make a temporary copy
2334 		 * of the leaf and add them both to that.
2335 		 */
2336 		struct xfs_attr_leafblock *tmp_leaf;
2337 		struct xfs_attr3_icleaf_hdr tmphdr;
2338 
2339 		tmp_leaf = kmem_zalloc(state->args->geo->blksize, 0);
2340 
2341 		/*
2342 		 * Copy the header into the temp leaf so that all the stuff
2343 		 * not in the incore header is present and gets copied back in
2344 		 * once we've moved all the entries.
2345 		 */
2346 		memcpy(tmp_leaf, save_leaf, xfs_attr3_leaf_hdr_size(save_leaf));
2347 
2348 		memset(&tmphdr, 0, sizeof(tmphdr));
2349 		tmphdr.magic = savehdr.magic;
2350 		tmphdr.forw = savehdr.forw;
2351 		tmphdr.back = savehdr.back;
2352 		tmphdr.firstused = state->args->geo->blksize;
2353 
2354 		/* write the header to the temp buffer to initialise it */
2355 		xfs_attr3_leaf_hdr_to_disk(state->args->geo, tmp_leaf, &tmphdr);
2356 
2357 		if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2358 					 drop_blk->bp, &drophdr)) {
2359 			xfs_attr3_leaf_moveents(state->args,
2360 						drop_leaf, &drophdr, 0,
2361 						tmp_leaf, &tmphdr, 0,
2362 						drophdr.count);
2363 			xfs_attr3_leaf_moveents(state->args,
2364 						save_leaf, &savehdr, 0,
2365 						tmp_leaf, &tmphdr, tmphdr.count,
2366 						savehdr.count);
2367 		} else {
2368 			xfs_attr3_leaf_moveents(state->args,
2369 						save_leaf, &savehdr, 0,
2370 						tmp_leaf, &tmphdr, 0,
2371 						savehdr.count);
2372 			xfs_attr3_leaf_moveents(state->args,
2373 						drop_leaf, &drophdr, 0,
2374 						tmp_leaf, &tmphdr, tmphdr.count,
2375 						drophdr.count);
2376 		}
2377 		memcpy(save_leaf, tmp_leaf, state->args->geo->blksize);
2378 		savehdr = tmphdr; /* struct copy */
2379 		kmem_free(tmp_leaf);
2380 	}
2381 
2382 	xfs_attr3_leaf_hdr_to_disk(state->args->geo, save_leaf, &savehdr);
2383 	xfs_trans_log_buf(state->args->trans, save_blk->bp, 0,
2384 					   state->args->geo->blksize - 1);
2385 
2386 	/*
2387 	 * Copy out last hashval in each block for B-tree code.
2388 	 */
2389 	entry = xfs_attr3_leaf_entryp(save_leaf);
2390 	save_blk->hashval = be32_to_cpu(entry[savehdr.count - 1].hashval);
2391 }
2392 
2393 /*========================================================================
2394  * Routines used for finding things in the Btree.
2395  *========================================================================*/
2396 
2397 /*
2398  * Look up a name in a leaf attribute list structure.
2399  * This is the internal routine, it uses the caller's buffer.
2400  *
2401  * Note that duplicate keys are allowed, but only check within the
2402  * current leaf node.  The Btree code must check in adjacent leaf nodes.
2403  *
2404  * Return in args->index the index into the entry[] array of either
2405  * the found entry, or where the entry should have been (insert before
2406  * that entry).
2407  *
2408  * Don't change the args->value unless we find the attribute.
2409  */
2410 int
2411 xfs_attr3_leaf_lookup_int(
2412 	struct xfs_buf		*bp,
2413 	struct xfs_da_args	*args)
2414 {
2415 	struct xfs_attr_leafblock *leaf;
2416 	struct xfs_attr3_icleaf_hdr ichdr;
2417 	struct xfs_attr_leaf_entry *entry;
2418 	struct xfs_attr_leaf_entry *entries;
2419 	struct xfs_attr_leaf_name_local *name_loc;
2420 	struct xfs_attr_leaf_name_remote *name_rmt;
2421 	xfs_dahash_t		hashval;
2422 	int			probe;
2423 	int			span;
2424 
2425 	trace_xfs_attr_leaf_lookup(args);
2426 
2427 	leaf = bp->b_addr;
2428 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2429 	entries = xfs_attr3_leaf_entryp(leaf);
2430 	if (ichdr.count >= args->geo->blksize / 8) {
2431 		xfs_buf_mark_corrupt(bp);
2432 		return -EFSCORRUPTED;
2433 	}
2434 
2435 	/*
2436 	 * Binary search.  (note: small blocks will skip this loop)
2437 	 */
2438 	hashval = args->hashval;
2439 	probe = span = ichdr.count / 2;
2440 	for (entry = &entries[probe]; span > 4; entry = &entries[probe]) {
2441 		span /= 2;
2442 		if (be32_to_cpu(entry->hashval) < hashval)
2443 			probe += span;
2444 		else if (be32_to_cpu(entry->hashval) > hashval)
2445 			probe -= span;
2446 		else
2447 			break;
2448 	}
2449 	if (!(probe >= 0 && (!ichdr.count || probe < ichdr.count))) {
2450 		xfs_buf_mark_corrupt(bp);
2451 		return -EFSCORRUPTED;
2452 	}
2453 	if (!(span <= 4 || be32_to_cpu(entry->hashval) == hashval)) {
2454 		xfs_buf_mark_corrupt(bp);
2455 		return -EFSCORRUPTED;
2456 	}
2457 
2458 	/*
2459 	 * Since we may have duplicate hashval's, find the first matching
2460 	 * hashval in the leaf.
2461 	 */
2462 	while (probe > 0 && be32_to_cpu(entry->hashval) >= hashval) {
2463 		entry--;
2464 		probe--;
2465 	}
2466 	while (probe < ichdr.count &&
2467 	       be32_to_cpu(entry->hashval) < hashval) {
2468 		entry++;
2469 		probe++;
2470 	}
2471 	if (probe == ichdr.count || be32_to_cpu(entry->hashval) != hashval) {
2472 		args->index = probe;
2473 		return -ENOATTR;
2474 	}
2475 
2476 	/*
2477 	 * Duplicate keys may be present, so search all of them for a match.
2478 	 */
2479 	for (; probe < ichdr.count && (be32_to_cpu(entry->hashval) == hashval);
2480 			entry++, probe++) {
2481 /*
2482  * GROT: Add code to remove incomplete entries.
2483  */
2484 		if (entry->flags & XFS_ATTR_LOCAL) {
2485 			name_loc = xfs_attr3_leaf_name_local(leaf, probe);
2486 			if (!xfs_attr_match(args, name_loc->namelen,
2487 					name_loc->nameval, entry->flags))
2488 				continue;
2489 			args->index = probe;
2490 			return -EEXIST;
2491 		} else {
2492 			name_rmt = xfs_attr3_leaf_name_remote(leaf, probe);
2493 			if (!xfs_attr_match(args, name_rmt->namelen,
2494 					name_rmt->name, entry->flags))
2495 				continue;
2496 			args->index = probe;
2497 			args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2498 			args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2499 			args->rmtblkcnt = xfs_attr3_rmt_blocks(
2500 							args->dp->i_mount,
2501 							args->rmtvaluelen);
2502 			return -EEXIST;
2503 		}
2504 	}
2505 	args->index = probe;
2506 	return -ENOATTR;
2507 }
2508 
2509 /*
2510  * Get the value associated with an attribute name from a leaf attribute
2511  * list structure.
2512  *
2513  * If args->valuelen is zero, only the length needs to be returned.  Unlike a
2514  * lookup, we only return an error if the attribute does not exist or we can't
2515  * retrieve the value.
2516  */
2517 int
2518 xfs_attr3_leaf_getvalue(
2519 	struct xfs_buf		*bp,
2520 	struct xfs_da_args	*args)
2521 {
2522 	struct xfs_attr_leafblock *leaf;
2523 	struct xfs_attr3_icleaf_hdr ichdr;
2524 	struct xfs_attr_leaf_entry *entry;
2525 	struct xfs_attr_leaf_name_local *name_loc;
2526 	struct xfs_attr_leaf_name_remote *name_rmt;
2527 
2528 	leaf = bp->b_addr;
2529 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2530 	ASSERT(ichdr.count < args->geo->blksize / 8);
2531 	ASSERT(args->index < ichdr.count);
2532 
2533 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2534 	if (entry->flags & XFS_ATTR_LOCAL) {
2535 		name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2536 		ASSERT(name_loc->namelen == args->namelen);
2537 		ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2538 		return xfs_attr_copy_value(args,
2539 					&name_loc->nameval[args->namelen],
2540 					be16_to_cpu(name_loc->valuelen));
2541 	}
2542 
2543 	name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2544 	ASSERT(name_rmt->namelen == args->namelen);
2545 	ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2546 	args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2547 	args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2548 	args->rmtblkcnt = xfs_attr3_rmt_blocks(args->dp->i_mount,
2549 					       args->rmtvaluelen);
2550 	return xfs_attr_copy_value(args, NULL, args->rmtvaluelen);
2551 }
2552 
2553 /*========================================================================
2554  * Utility routines.
2555  *========================================================================*/
2556 
2557 /*
2558  * Move the indicated entries from one leaf to another.
2559  * NOTE: this routine modifies both source and destination leaves.
2560  */
2561 /*ARGSUSED*/
2562 STATIC void
2563 xfs_attr3_leaf_moveents(
2564 	struct xfs_da_args		*args,
2565 	struct xfs_attr_leafblock	*leaf_s,
2566 	struct xfs_attr3_icleaf_hdr	*ichdr_s,
2567 	int				start_s,
2568 	struct xfs_attr_leafblock	*leaf_d,
2569 	struct xfs_attr3_icleaf_hdr	*ichdr_d,
2570 	int				start_d,
2571 	int				count)
2572 {
2573 	struct xfs_attr_leaf_entry	*entry_s;
2574 	struct xfs_attr_leaf_entry	*entry_d;
2575 	int				desti;
2576 	int				tmp;
2577 	int				i;
2578 
2579 	/*
2580 	 * Check for nothing to do.
2581 	 */
2582 	if (count == 0)
2583 		return;
2584 
2585 	/*
2586 	 * Set up environment.
2587 	 */
2588 	ASSERT(ichdr_s->magic == XFS_ATTR_LEAF_MAGIC ||
2589 	       ichdr_s->magic == XFS_ATTR3_LEAF_MAGIC);
2590 	ASSERT(ichdr_s->magic == ichdr_d->magic);
2591 	ASSERT(ichdr_s->count > 0 && ichdr_s->count < args->geo->blksize / 8);
2592 	ASSERT(ichdr_s->firstused >= (ichdr_s->count * sizeof(*entry_s))
2593 					+ xfs_attr3_leaf_hdr_size(leaf_s));
2594 	ASSERT(ichdr_d->count < args->geo->blksize / 8);
2595 	ASSERT(ichdr_d->firstused >= (ichdr_d->count * sizeof(*entry_d))
2596 					+ xfs_attr3_leaf_hdr_size(leaf_d));
2597 
2598 	ASSERT(start_s < ichdr_s->count);
2599 	ASSERT(start_d <= ichdr_d->count);
2600 	ASSERT(count <= ichdr_s->count);
2601 
2602 
2603 	/*
2604 	 * Move the entries in the destination leaf up to make a hole?
2605 	 */
2606 	if (start_d < ichdr_d->count) {
2607 		tmp  = ichdr_d->count - start_d;
2608 		tmp *= sizeof(xfs_attr_leaf_entry_t);
2609 		entry_s = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2610 		entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d + count];
2611 		memmove(entry_d, entry_s, tmp);
2612 	}
2613 
2614 	/*
2615 	 * Copy all entry's in the same (sorted) order,
2616 	 * but allocate attribute info packed and in sequence.
2617 	 */
2618 	entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2619 	entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2620 	desti = start_d;
2621 	for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2622 		ASSERT(be16_to_cpu(entry_s->nameidx) >= ichdr_s->firstused);
2623 		tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2624 #ifdef GROT
2625 		/*
2626 		 * Code to drop INCOMPLETE entries.  Difficult to use as we
2627 		 * may also need to change the insertion index.  Code turned
2628 		 * off for 6.2, should be revisited later.
2629 		 */
2630 		if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2631 			memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2632 			ichdr_s->usedbytes -= tmp;
2633 			ichdr_s->count -= 1;
2634 			entry_d--;	/* to compensate for ++ in loop hdr */
2635 			desti--;
2636 			if ((start_s + i) < offset)
2637 				result++;	/* insertion index adjustment */
2638 		} else {
2639 #endif /* GROT */
2640 			ichdr_d->firstused -= tmp;
2641 			/* both on-disk, don't endian flip twice */
2642 			entry_d->hashval = entry_s->hashval;
2643 			entry_d->nameidx = cpu_to_be16(ichdr_d->firstused);
2644 			entry_d->flags = entry_s->flags;
2645 			ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2646 							<= args->geo->blksize);
2647 			memmove(xfs_attr3_leaf_name(leaf_d, desti),
2648 				xfs_attr3_leaf_name(leaf_s, start_s + i), tmp);
2649 			ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2650 							<= args->geo->blksize);
2651 			memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2652 			ichdr_s->usedbytes -= tmp;
2653 			ichdr_d->usedbytes += tmp;
2654 			ichdr_s->count -= 1;
2655 			ichdr_d->count += 1;
2656 			tmp = ichdr_d->count * sizeof(xfs_attr_leaf_entry_t)
2657 					+ xfs_attr3_leaf_hdr_size(leaf_d);
2658 			ASSERT(ichdr_d->firstused >= tmp);
2659 #ifdef GROT
2660 		}
2661 #endif /* GROT */
2662 	}
2663 
2664 	/*
2665 	 * Zero out the entries we just copied.
2666 	 */
2667 	if (start_s == ichdr_s->count) {
2668 		tmp = count * sizeof(xfs_attr_leaf_entry_t);
2669 		entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2670 		ASSERT(((char *)entry_s + tmp) <=
2671 		       ((char *)leaf_s + args->geo->blksize));
2672 		memset(entry_s, 0, tmp);
2673 	} else {
2674 		/*
2675 		 * Move the remaining entries down to fill the hole,
2676 		 * then zero the entries at the top.
2677 		 */
2678 		tmp  = (ichdr_s->count - count) * sizeof(xfs_attr_leaf_entry_t);
2679 		entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s + count];
2680 		entry_d = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2681 		memmove(entry_d, entry_s, tmp);
2682 
2683 		tmp = count * sizeof(xfs_attr_leaf_entry_t);
2684 		entry_s = &xfs_attr3_leaf_entryp(leaf_s)[ichdr_s->count];
2685 		ASSERT(((char *)entry_s + tmp) <=
2686 		       ((char *)leaf_s + args->geo->blksize));
2687 		memset(entry_s, 0, tmp);
2688 	}
2689 
2690 	/*
2691 	 * Fill in the freemap information
2692 	 */
2693 	ichdr_d->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_d);
2694 	ichdr_d->freemap[0].base += ichdr_d->count * sizeof(xfs_attr_leaf_entry_t);
2695 	ichdr_d->freemap[0].size = ichdr_d->firstused - ichdr_d->freemap[0].base;
2696 	ichdr_d->freemap[1].base = 0;
2697 	ichdr_d->freemap[2].base = 0;
2698 	ichdr_d->freemap[1].size = 0;
2699 	ichdr_d->freemap[2].size = 0;
2700 	ichdr_s->holes = 1;	/* leaf may not be compact */
2701 }
2702 
2703 /*
2704  * Pick up the last hashvalue from a leaf block.
2705  */
2706 xfs_dahash_t
2707 xfs_attr_leaf_lasthash(
2708 	struct xfs_buf	*bp,
2709 	int		*count)
2710 {
2711 	struct xfs_attr3_icleaf_hdr ichdr;
2712 	struct xfs_attr_leaf_entry *entries;
2713 	struct xfs_mount *mp = bp->b_mount;
2714 
2715 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, bp->b_addr);
2716 	entries = xfs_attr3_leaf_entryp(bp->b_addr);
2717 	if (count)
2718 		*count = ichdr.count;
2719 	if (!ichdr.count)
2720 		return 0;
2721 	return be32_to_cpu(entries[ichdr.count - 1].hashval);
2722 }
2723 
2724 /*
2725  * Calculate the number of bytes used to store the indicated attribute
2726  * (whether local or remote only calculate bytes in this block).
2727  */
2728 STATIC int
2729 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2730 {
2731 	struct xfs_attr_leaf_entry *entries;
2732 	xfs_attr_leaf_name_local_t *name_loc;
2733 	xfs_attr_leaf_name_remote_t *name_rmt;
2734 	int size;
2735 
2736 	entries = xfs_attr3_leaf_entryp(leaf);
2737 	if (entries[index].flags & XFS_ATTR_LOCAL) {
2738 		name_loc = xfs_attr3_leaf_name_local(leaf, index);
2739 		size = xfs_attr_leaf_entsize_local(name_loc->namelen,
2740 						   be16_to_cpu(name_loc->valuelen));
2741 	} else {
2742 		name_rmt = xfs_attr3_leaf_name_remote(leaf, index);
2743 		size = xfs_attr_leaf_entsize_remote(name_rmt->namelen);
2744 	}
2745 	return size;
2746 }
2747 
2748 /*
2749  * Calculate the number of bytes that would be required to store the new
2750  * attribute (whether local or remote only calculate bytes in this block).
2751  * This routine decides as a side effect whether the attribute will be
2752  * a "local" or a "remote" attribute.
2753  */
2754 int
2755 xfs_attr_leaf_newentsize(
2756 	struct xfs_da_args	*args,
2757 	int			*local)
2758 {
2759 	int			size;
2760 
2761 	size = xfs_attr_leaf_entsize_local(args->namelen, args->valuelen);
2762 	if (size < xfs_attr_leaf_entsize_local_max(args->geo->blksize)) {
2763 		if (local)
2764 			*local = 1;
2765 		return size;
2766 	}
2767 	if (local)
2768 		*local = 0;
2769 	return xfs_attr_leaf_entsize_remote(args->namelen);
2770 }
2771 
2772 
2773 /*========================================================================
2774  * Manage the INCOMPLETE flag in a leaf entry
2775  *========================================================================*/
2776 
2777 /*
2778  * Clear the INCOMPLETE flag on an entry in a leaf block.
2779  */
2780 int
2781 xfs_attr3_leaf_clearflag(
2782 	struct xfs_da_args	*args)
2783 {
2784 	struct xfs_attr_leafblock *leaf;
2785 	struct xfs_attr_leaf_entry *entry;
2786 	struct xfs_attr_leaf_name_remote *name_rmt;
2787 	struct xfs_buf		*bp;
2788 	int			error;
2789 #ifdef DEBUG
2790 	struct xfs_attr3_icleaf_hdr ichdr;
2791 	xfs_attr_leaf_name_local_t *name_loc;
2792 	int namelen;
2793 	char *name;
2794 #endif /* DEBUG */
2795 
2796 	trace_xfs_attr_leaf_clearflag(args);
2797 	/*
2798 	 * Set up the operation.
2799 	 */
2800 	error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp);
2801 	if (error)
2802 		return error;
2803 
2804 	leaf = bp->b_addr;
2805 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2806 	ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2807 
2808 #ifdef DEBUG
2809 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2810 	ASSERT(args->index < ichdr.count);
2811 	ASSERT(args->index >= 0);
2812 
2813 	if (entry->flags & XFS_ATTR_LOCAL) {
2814 		name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2815 		namelen = name_loc->namelen;
2816 		name = (char *)name_loc->nameval;
2817 	} else {
2818 		name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2819 		namelen = name_rmt->namelen;
2820 		name = (char *)name_rmt->name;
2821 	}
2822 	ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2823 	ASSERT(namelen == args->namelen);
2824 	ASSERT(memcmp(name, args->name, namelen) == 0);
2825 #endif /* DEBUG */
2826 
2827 	entry->flags &= ~XFS_ATTR_INCOMPLETE;
2828 	xfs_trans_log_buf(args->trans, bp,
2829 			 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2830 
2831 	if (args->rmtblkno) {
2832 		ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2833 		name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2834 		name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2835 		name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2836 		xfs_trans_log_buf(args->trans, bp,
2837 			 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2838 	}
2839 
2840 	return 0;
2841 }
2842 
2843 /*
2844  * Set the INCOMPLETE flag on an entry in a leaf block.
2845  */
2846 int
2847 xfs_attr3_leaf_setflag(
2848 	struct xfs_da_args	*args)
2849 {
2850 	struct xfs_attr_leafblock *leaf;
2851 	struct xfs_attr_leaf_entry *entry;
2852 	struct xfs_attr_leaf_name_remote *name_rmt;
2853 	struct xfs_buf		*bp;
2854 	int error;
2855 #ifdef DEBUG
2856 	struct xfs_attr3_icleaf_hdr ichdr;
2857 #endif
2858 
2859 	trace_xfs_attr_leaf_setflag(args);
2860 
2861 	/*
2862 	 * Set up the operation.
2863 	 */
2864 	error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp);
2865 	if (error)
2866 		return error;
2867 
2868 	leaf = bp->b_addr;
2869 #ifdef DEBUG
2870 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2871 	ASSERT(args->index < ichdr.count);
2872 	ASSERT(args->index >= 0);
2873 #endif
2874 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2875 
2876 	ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2877 	entry->flags |= XFS_ATTR_INCOMPLETE;
2878 	xfs_trans_log_buf(args->trans, bp,
2879 			XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2880 	if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2881 		name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2882 		name_rmt->valueblk = 0;
2883 		name_rmt->valuelen = 0;
2884 		xfs_trans_log_buf(args->trans, bp,
2885 			 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2886 	}
2887 
2888 	return 0;
2889 }
2890 
2891 /*
2892  * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2893  * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2894  * entry given by args->blkno2/index2.
2895  *
2896  * Note that they could be in different blocks, or in the same block.
2897  */
2898 int
2899 xfs_attr3_leaf_flipflags(
2900 	struct xfs_da_args	*args)
2901 {
2902 	struct xfs_attr_leafblock *leaf1;
2903 	struct xfs_attr_leafblock *leaf2;
2904 	struct xfs_attr_leaf_entry *entry1;
2905 	struct xfs_attr_leaf_entry *entry2;
2906 	struct xfs_attr_leaf_name_remote *name_rmt;
2907 	struct xfs_buf		*bp1;
2908 	struct xfs_buf		*bp2;
2909 	int error;
2910 #ifdef DEBUG
2911 	struct xfs_attr3_icleaf_hdr ichdr1;
2912 	struct xfs_attr3_icleaf_hdr ichdr2;
2913 	xfs_attr_leaf_name_local_t *name_loc;
2914 	int namelen1, namelen2;
2915 	char *name1, *name2;
2916 #endif /* DEBUG */
2917 
2918 	trace_xfs_attr_leaf_flipflags(args);
2919 
2920 	/*
2921 	 * Read the block containing the "old" attr
2922 	 */
2923 	error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp1);
2924 	if (error)
2925 		return error;
2926 
2927 	/*
2928 	 * Read the block containing the "new" attr, if it is different
2929 	 */
2930 	if (args->blkno2 != args->blkno) {
2931 		error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno2,
2932 					   &bp2);
2933 		if (error)
2934 			return error;
2935 	} else {
2936 		bp2 = bp1;
2937 	}
2938 
2939 	leaf1 = bp1->b_addr;
2940 	entry1 = &xfs_attr3_leaf_entryp(leaf1)[args->index];
2941 
2942 	leaf2 = bp2->b_addr;
2943 	entry2 = &xfs_attr3_leaf_entryp(leaf2)[args->index2];
2944 
2945 #ifdef DEBUG
2946 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr1, leaf1);
2947 	ASSERT(args->index < ichdr1.count);
2948 	ASSERT(args->index >= 0);
2949 
2950 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr2, leaf2);
2951 	ASSERT(args->index2 < ichdr2.count);
2952 	ASSERT(args->index2 >= 0);
2953 
2954 	if (entry1->flags & XFS_ATTR_LOCAL) {
2955 		name_loc = xfs_attr3_leaf_name_local(leaf1, args->index);
2956 		namelen1 = name_loc->namelen;
2957 		name1 = (char *)name_loc->nameval;
2958 	} else {
2959 		name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2960 		namelen1 = name_rmt->namelen;
2961 		name1 = (char *)name_rmt->name;
2962 	}
2963 	if (entry2->flags & XFS_ATTR_LOCAL) {
2964 		name_loc = xfs_attr3_leaf_name_local(leaf2, args->index2);
2965 		namelen2 = name_loc->namelen;
2966 		name2 = (char *)name_loc->nameval;
2967 	} else {
2968 		name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2969 		namelen2 = name_rmt->namelen;
2970 		name2 = (char *)name_rmt->name;
2971 	}
2972 	ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2973 	ASSERT(namelen1 == namelen2);
2974 	ASSERT(memcmp(name1, name2, namelen1) == 0);
2975 #endif /* DEBUG */
2976 
2977 	ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2978 	ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2979 
2980 	entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2981 	xfs_trans_log_buf(args->trans, bp1,
2982 			  XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2983 	if (args->rmtblkno) {
2984 		ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2985 		name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2986 		name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2987 		name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2988 		xfs_trans_log_buf(args->trans, bp1,
2989 			 XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2990 	}
2991 
2992 	entry2->flags |= XFS_ATTR_INCOMPLETE;
2993 	xfs_trans_log_buf(args->trans, bp2,
2994 			  XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2995 	if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
2996 		name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2997 		name_rmt->valueblk = 0;
2998 		name_rmt->valuelen = 0;
2999 		xfs_trans_log_buf(args->trans, bp2,
3000 			 XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
3001 	}
3002 
3003 	return 0;
3004 }
3005