1	// SPDX-License-Identifier: GPL-2.0
2	/*
3	* Copyright (c) 2000-2005 Silicon Graphics, Inc.
4	* Copyright (c) 2013 Red Hat, Inc.
5	* All Rights Reserved.
6	*/
7	#include "xfs.h"
8	#include "xfs_fs.h"
9	#include "xfs_shared.h"
10	#include "xfs_format.h"
11	#include "xfs_log_format.h"
12	#include "xfs_trans_resv.h"
13	#include "xfs_sb.h"
14	#include "xfs_mount.h"
15	#include "xfs_da_format.h"
16	#include "xfs_da_btree.h"
17	#include "xfs_inode.h"
18	#include "xfs_trans.h"
19	#include "xfs_bmap_btree.h"
20	#include "xfs_bmap.h"
21	#include "xfs_attr_sf.h"
22	#include "xfs_attr.h"
23	#include "xfs_attr_remote.h"
24	#include "xfs_attr_leaf.h"
25	#include "xfs_error.h"
26	#include "xfs_trace.h"
27	#include "xfs_buf_item.h"
28	#include "xfs_dir2.h"
29	#include "xfs_log.h"
30	#include "xfs_ag.h"
31	#include "xfs_errortag.h"
32	#include "xfs_health.h"
33
34
35	/*
36	* xfs_attr_leaf.c
37	*
38	* Routines to implement leaf blocks of attributes as Btrees of hashed names.
39	*/
40
41	/*========================================================================
42	* Function prototypes for the kernel.
43	*========================================================================*/
44
45	/*
46	* Routines used for growing the Btree.
47	*/
48	STATIC int xfs_attr3_leaf_create(struct xfs_da_args *args,
49	xfs_dablk_t which_block, struct xfs_buf **bpp);
50	STATIC void xfs_attr3_leaf_add_work(struct xfs_buf *leaf_buffer,
51	struct xfs_attr3_icleaf_hdr *ichdr,
52	struct xfs_da_args *args, int freemap_index);
53	STATIC void xfs_attr3_leaf_compact(struct xfs_da_args *args,
54	struct xfs_attr3_icleaf_hdr *ichdr,
55	struct xfs_buf *leaf_buffer);
56	STATIC void xfs_attr3_leaf_rebalance(xfs_da_state_t *state,
57	xfs_da_state_blk_t *blk1,
58	xfs_da_state_blk_t *blk2);
59	STATIC int xfs_attr3_leaf_figure_balance(xfs_da_state_t *state,
60	xfs_da_state_blk_t *leaf_blk_1,
61	struct xfs_attr3_icleaf_hdr *ichdr1,
62	xfs_da_state_blk_t *leaf_blk_2,
63	struct xfs_attr3_icleaf_hdr *ichdr2,
64	int *number_entries_in_blk1,
65	int *number_usedbytes_in_blk1);
66
67	/*
68	* Utility routines.
69	*/
70	STATIC void xfs_attr3_leaf_moveents(struct xfs_da_args *args,
71	struct xfs_attr_leafblock *src_leaf,
72	struct xfs_attr3_icleaf_hdr *src_ichdr, int src_start,
73	struct xfs_attr_leafblock *dst_leaf,
74	struct xfs_attr3_icleaf_hdr *dst_ichdr, int dst_start,
75	int move_count);
76	STATIC int xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index);
77
78	/*
79	* attr3 block 'firstused' conversion helpers.
80	*
81	* firstused refers to the offset of the first used byte of the nameval region
82	* of an attr leaf block. The region starts at the tail of the block and expands
83	* backwards towards the middle. As such, firstused is initialized to the block
84	* size for an empty leaf block and is reduced from there.
85	*
86	* The attr3 block size is pegged to the fsb size and the maximum fsb is 64k.
87	* The in-core firstused field is 32-bit and thus supports the maximum fsb size.
88	* The on-disk field is only 16-bit, however, and overflows at 64k. Since this
89	* only occurs at exactly 64k, we use zero as a magic on-disk value to represent
90	* the attr block size. The following helpers manage the conversion between the
91	* in-core and on-disk formats.
92	*/
93
94	static void
95	xfs_attr3_leaf_firstused_from_disk(
96	struct xfs_da_geometry *geo,
97	struct xfs_attr3_icleaf_hdr *to,
98	struct xfs_attr_leafblock *from)
99	{
100	struct xfs_attr3_leaf_hdr *hdr3;
101
102	if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) {
103	hdr3 = (struct xfs_attr3_leaf_hdr *) from;
104	to->firstused = be16_to_cpu(hdr3->firstused);
105	} else {
106	to->firstused = be16_to_cpu(from->hdr.firstused);
107	}
108
109	/*
110	* Convert from the magic fsb size value to actual blocksize. This
111	* should only occur for empty blocks when the block size overflows
112	* 16-bits.
113	*/
114	if (to->firstused == XFS_ATTR3_LEAF_NULLOFF) {
115	ASSERT(!to->count && !to->usedbytes);
116	ASSERT(geo->blksize > USHRT_MAX);
117	to->firstused = geo->blksize;
118	}
119	}
120
121	static void
122	xfs_attr3_leaf_firstused_to_disk(
123	struct xfs_da_geometry *geo,
124	struct xfs_attr_leafblock *to,
125	struct xfs_attr3_icleaf_hdr *from)
126	{
127	struct xfs_attr3_leaf_hdr *hdr3;
128	uint32_t firstused;
129
130	/* magic value should only be seen on disk */
131	ASSERT(from->firstused != XFS_ATTR3_LEAF_NULLOFF);
132
133	/*
134	* Scale down the 32-bit in-core firstused value to the 16-bit on-disk
135	* value. This only overflows at the max supported value of 64k. Use the
136	* magic on-disk value to represent block size in this case.
137	*/
138	firstused = from->firstused;
139	if (firstused > USHRT_MAX) {
140	ASSERT(from->firstused == geo->blksize);
141	firstused = XFS_ATTR3_LEAF_NULLOFF;
142	}
143
144	if (from->magic == XFS_ATTR3_LEAF_MAGIC) {
145	hdr3 = (struct xfs_attr3_leaf_hdr *) to;
146	hdr3->firstused = cpu_to_be16(firstused);
147	} else {
148	to->hdr.firstused = cpu_to_be16(firstused);
149	}
150	}
151
152	void
153	xfs_attr3_leaf_hdr_from_disk(
154	struct xfs_da_geometry *geo,
155	struct xfs_attr3_icleaf_hdr *to,
156	struct xfs_attr_leafblock *from)
157	{
158	int i;
159
160	ASSERT(from->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
161	from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
162
163	if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) {
164	struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)from;
165
166	to->forw = be32_to_cpu(hdr3->info.hdr.forw);
167	to->back = be32_to_cpu(hdr3->info.hdr.back);
168	to->magic = be16_to_cpu(hdr3->info.hdr.magic);
169	to->count = be16_to_cpu(hdr3->count);
170	to->usedbytes = be16_to_cpu(hdr3->usedbytes);
171	xfs_attr3_leaf_firstused_from_disk(geo, to, from);
172	to->holes = hdr3->holes;
173
174	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
175	to->freemap[i].base = be16_to_cpu(hdr3->freemap[i].base);
176	to->freemap[i].size = be16_to_cpu(hdr3->freemap[i].size);
177	}
178	return;
179	}
180	to->forw = be32_to_cpu(from->hdr.info.forw);
181	to->back = be32_to_cpu(from->hdr.info.back);
182	to->magic = be16_to_cpu(from->hdr.info.magic);
183	to->count = be16_to_cpu(from->hdr.count);
184	to->usedbytes = be16_to_cpu(from->hdr.usedbytes);
185	xfs_attr3_leaf_firstused_from_disk(geo, to, from);
186	to->holes = from->hdr.holes;
187
188	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
189	to->freemap[i].base = be16_to_cpu(from->hdr.freemap[i].base);
190	to->freemap[i].size = be16_to_cpu(from->hdr.freemap[i].size);
191	}
192	}
193
194	void
195	xfs_attr3_leaf_hdr_to_disk(
196	struct xfs_da_geometry *geo,
197	struct xfs_attr_leafblock *to,
198	struct xfs_attr3_icleaf_hdr *from)
199	{
200	int i;
201
202	ASSERT(from->magic == XFS_ATTR_LEAF_MAGIC ||
203	from->magic == XFS_ATTR3_LEAF_MAGIC);
204
205	if (from->magic == XFS_ATTR3_LEAF_MAGIC) {
206	struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)to;
207
208	hdr3->info.hdr.forw = cpu_to_be32(from->forw);
209	hdr3->info.hdr.back = cpu_to_be32(from->back);
210	hdr3->info.hdr.magic = cpu_to_be16(from->magic);
211	hdr3->count = cpu_to_be16(from->count);
212	hdr3->usedbytes = cpu_to_be16(from->usedbytes);
213	xfs_attr3_leaf_firstused_to_disk(geo, to, from);
214	hdr3->holes = from->holes;
215	hdr3->pad1 = 0;
216
217	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
218	hdr3->freemap[i].base = cpu_to_be16(from->freemap[i].base);
219	hdr3->freemap[i].size = cpu_to_be16(from->freemap[i].size);
220	}
221	return;
222	}
223	to->hdr.info.forw = cpu_to_be32(from->forw);
224	to->hdr.info.back = cpu_to_be32(from->back);
225	to->hdr.info.magic = cpu_to_be16(from->magic);
226	to->hdr.count = cpu_to_be16(from->count);
227	to->hdr.usedbytes = cpu_to_be16(from->usedbytes);
228	xfs_attr3_leaf_firstused_to_disk(geo, to, from);
229	to->hdr.holes = from->holes;
230	to->hdr.pad1 = 0;
231
232	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
233	to->hdr.freemap[i].base = cpu_to_be16(from->freemap[i].base);
234	to->hdr.freemap[i].size = cpu_to_be16(from->freemap[i].size);
235	}
236	}
237
238	static xfs_failaddr_t
239	xfs_attr3_leaf_verify_entry(
240	struct xfs_mount *mp,
241	char *buf_end,
242	struct xfs_attr_leafblock *leaf,
243	struct xfs_attr3_icleaf_hdr *leafhdr,
244	struct xfs_attr_leaf_entry *ent,
245	int idx,
246	__u32 *last_hashval)
247	{
248	struct xfs_attr_leaf_name_local *lentry;
249	struct xfs_attr_leaf_name_remote *rentry;
250	char *name_end;
251	unsigned int nameidx;
252	unsigned int namesize;
253	__u32 hashval;
254
255	/* hash order check */
256	hashval = be32_to_cpu(ent->hashval);
257	if (hashval < *last_hashval)
258	return __this_address;
259	*last_hashval = hashval;
260
261	nameidx = be16_to_cpu(ent->nameidx);
262	if (nameidx < leafhdr->firstused || nameidx >= mp->m_attr_geo->blksize)
263	return __this_address;
264
265	/*
266	* Check the name information. The namelen fields are u8 so we can't
267	* possibly exceed the maximum name length of 255 bytes.
268	*/
269	if (ent->flags & XFS_ATTR_LOCAL) {
270	lentry = xfs_attr3_leaf_name_local(leafp: leaf, idx);
271	namesize = xfs_attr_leaf_entsize_local(nlen: lentry->namelen,
272	vlen: be16_to_cpu(lentry->valuelen));
273	name_end = (char *)lentry + namesize;
274	if (lentry->namelen == 0)
275	return __this_address;
276	} else {
277	rentry = xfs_attr3_leaf_name_remote(leafp: leaf, idx);
278	namesize = xfs_attr_leaf_entsize_remote(nlen: rentry->namelen);
279	name_end = (char *)rentry + namesize;
280	if (rentry->namelen == 0)
281	return __this_address;
282	if (!(ent->flags & XFS_ATTR_INCOMPLETE) &&
283	rentry->valueblk == 0)
284	return __this_address;
285	}
286
287	if (name_end > buf_end)
288	return __this_address;
289
290	return NULL;
291	}
292
293	/*
294	* Validate an attribute leaf block.
295	*
296	* Empty leaf blocks can occur under the following circumstances:
297	*
298	* 1. setxattr adds a new extended attribute to a file;
299	* 2. The file has zero existing attributes;
300	* 3. The attribute is too large to fit in the attribute fork;
301	* 4. The attribute is small enough to fit in a leaf block;
302	* 5. A log flush occurs after committing the transaction that creates
303	* the (empty) leaf block; and
304	* 6. The filesystem goes down after the log flush but before the new
305	* attribute can be committed to the leaf block.
306	*
307	* Hence we need to ensure that we don't fail the validation purely
308	* because the leaf is empty.
309	*/
310	static xfs_failaddr_t
311	xfs_attr3_leaf_verify(
312	struct xfs_buf *bp)
313	{
314	struct xfs_attr3_icleaf_hdr ichdr;
315	struct xfs_mount *mp = bp->b_mount;
316	struct xfs_attr_leafblock *leaf = bp->b_addr;
317	struct xfs_attr_leaf_entry *entries;
318	struct xfs_attr_leaf_entry *ent;
319	char *buf_end;
320	uint32_t end; /* must be 32bit - see below */
321	__u32 last_hashval = 0;
322	int i;
323	xfs_failaddr_t fa;
324
325	xfs_attr3_leaf_hdr_from_disk(geo: mp->m_attr_geo, to: &ichdr, from: leaf);
326
327	fa = xfs_da3_blkinfo_verify(bp, bp->b_addr);
328	if (fa)
329	return fa;
330
331	/*
332	* firstused is the block offset of the first name info structure.
333	* Make sure it doesn't go off the block or crash into the header.
334	*/
335	if (ichdr.firstused > mp->m_attr_geo->blksize)
336	return __this_address;
337	if (ichdr.firstused < xfs_attr3_leaf_hdr_size(leaf))
338	return __this_address;
339
340	/* Make sure the entries array doesn't crash into the name info. */
341	entries = xfs_attr3_leaf_entryp(leafp: bp->b_addr);
342	if ((char *)&entries[ichdr.count] >
343	(char *)bp->b_addr + ichdr.firstused)
344	return __this_address;
345
346	/*
347	* NOTE: This verifier historically failed empty leaf buffers because
348	* we expect the fork to be in another format. Empty attr fork format
349	* conversions are possible during xattr set, however, and format
350	* conversion is not atomic with the xattr set that triggers it. We
351	* cannot assume leaf blocks are non-empty until that is addressed.
352	*/
353	buf_end = (char *)bp->b_addr + mp->m_attr_geo->blksize;
354	for (i = 0, ent = entries; i < ichdr.count; ent++, i++) {
355	fa = xfs_attr3_leaf_verify_entry(mp, buf_end, leaf, &ichdr,
356	ent, i, &last_hashval);
357	if (fa)
358	return fa;
359	}
360
361	/*
362	* Quickly check the freemap information. Attribute data has to be
363	* aligned to 4-byte boundaries, and likewise for the free space.
364	*
365	* Note that for 64k block size filesystems, the freemap entries cannot
366	* overflow as they are only be16 fields. However, when checking end
367	* pointer of the freemap, we have to be careful to detect overflows and
368	* so use uint32_t for those checks.
369	*/
370	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
371	if (ichdr.freemap[i].base > mp->m_attr_geo->blksize)
372	return __this_address;
373	if (ichdr.freemap[i].base & 0x3)
374	return __this_address;
375	if (ichdr.freemap[i].size > mp->m_attr_geo->blksize)
376	return __this_address;
377	if (ichdr.freemap[i].size & 0x3)
378	return __this_address;
379
380	/* be care of 16 bit overflows here */
381	end = (uint32_t)ichdr.freemap[i].base + ichdr.freemap[i].size;
382	if (end < ichdr.freemap[i].base)
383	return __this_address;
384	if (end > mp->m_attr_geo->blksize)
385	return __this_address;
386	}
387
388	return NULL;
389	}
390
391	xfs_failaddr_t
392	xfs_attr3_leaf_header_check(
393	struct xfs_buf *bp,
394	xfs_ino_t owner)
395	{
396	struct xfs_mount *mp = bp->b_mount;
397
398	if (xfs_has_crc(mp)) {
399	struct xfs_attr3_leafblock *hdr3 = bp->b_addr;
400
401	if (hdr3->hdr.info.hdr.magic !=
402	cpu_to_be16(XFS_ATTR3_LEAF_MAGIC))
403	return __this_address;
404
405	if (be64_to_cpu(hdr3->hdr.info.owner) != owner)
406	return __this_address;
407	}
408
409	return NULL;
410	}
411
412	static void
413	xfs_attr3_leaf_write_verify(
414	struct xfs_buf *bp)
415	{
416	struct xfs_mount *mp = bp->b_mount;
417	struct xfs_buf_log_item *bip = bp->b_log_item;
418	struct xfs_attr3_leaf_hdr *hdr3 = bp->b_addr;
419	xfs_failaddr_t fa;
420
421	fa = xfs_attr3_leaf_verify(bp);
422	if (fa) {
423	xfs_verifier_error(bp, -EFSCORRUPTED, fa);
424	return;
425	}
426
427	if (!xfs_has_crc(mp))
428	return;
429
430	if (bip)
431	hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
432
433	xfs_buf_update_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF);
434	}
435
436	/*
437	* leaf/node format detection on trees is sketchy, so a node read can be done on
438	* leaf level blocks when detection identifies the tree as a node format tree
439	* incorrectly. In this case, we need to swap the verifier to match the correct
440	* format of the block being read.
441	*/
442	static void
443	xfs_attr3_leaf_read_verify(
444	struct xfs_buf *bp)
445	{
446	struct xfs_mount *mp = bp->b_mount;
447	xfs_failaddr_t fa;
448
449	if (xfs_has_crc(mp) &&
450	!xfs_buf_verify_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF))
451	xfs_verifier_error(bp, -EFSBADCRC, __this_address);
452	else {
453	fa = xfs_attr3_leaf_verify(bp);
454	if (fa)
455	xfs_verifier_error(bp, -EFSCORRUPTED, fa);
456	}
457	}
458
459	const struct xfs_buf_ops xfs_attr3_leaf_buf_ops = {
460	.name = "xfs_attr3_leaf",
461	.magic16 = { cpu_to_be16(XFS_ATTR_LEAF_MAGIC),
462	cpu_to_be16(XFS_ATTR3_LEAF_MAGIC) },
463	.verify_read = xfs_attr3_leaf_read_verify,
464	.verify_write = xfs_attr3_leaf_write_verify,
465	.verify_struct = xfs_attr3_leaf_verify,
466	};
467
468	int
469	xfs_attr3_leaf_read(
470	struct xfs_trans *tp,
471	struct xfs_inode *dp,
472	xfs_ino_t owner,
473	xfs_dablk_t bno,
474	struct xfs_buf **bpp)
475	{
476	xfs_failaddr_t fa;
477	int err;
478
479	err = xfs_da_read_buf(tp, dp, bno, 0, bpp, XFS_ATTR_FORK,
480	&xfs_attr3_leaf_buf_ops);
481	if (err || !(*bpp))
482	return err;
483
484	fa = xfs_attr3_leaf_header_check(*bpp, owner);
485	if (fa) {
486	__xfs_buf_mark_corrupt(*bpp, fa);
487	xfs_trans_brelse(tp, *bpp);
488	*bpp = NULL;
489	xfs_dirattr_mark_sick(dp, XFS_ATTR_FORK);
490	return -EFSCORRUPTED;
491	}
492
493	if (tp)
494	xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_ATTR_LEAF_BUF);
495	return 0;
496	}
497
498	/*========================================================================
499	* Namespace helper routines
500	*========================================================================*/
501
502	/*
503	* If we are in log recovery, then we want the lookup to ignore the INCOMPLETE
504	* flag on disk - if there's an incomplete attr then recovery needs to tear it
505	* down. If there's no incomplete attr, then recovery needs to tear that attr
506	* down to replace it with the attr that has been logged. In this case, the
507	* INCOMPLETE flag will not be set in attr->attr_filter, but rather
508	* XFS_DA_OP_RECOVERY will be set in args->op_flags.
509	*/
510	static inline unsigned int xfs_attr_match_mask(const struct xfs_da_args *args)
511	{
512	if (args->op_flags & XFS_DA_OP_RECOVERY)
513	return XFS_ATTR_NSP_ONDISK_MASK;
514	return XFS_ATTR_NSP_ONDISK_MASK | XFS_ATTR_INCOMPLETE;
515	}
516
517	static inline bool
518	xfs_attr_parent_match(
519	const struct xfs_da_args *args,
520	const void *value,
521	unsigned int valuelen)
522	{
523	ASSERT(args->value != NULL);
524
525	/* Parent pointers do not use remote values */
526	if (!value)
527	return false;
528
529	/*
530	* The only value we support is a parent rec. However, we'll accept
531	* any valuelen so that offline repair can delete ATTR_PARENT values
532	* that are not parent pointers.
533	*/
534	if (valuelen != args->valuelen)
535	return false;
536
537	return memcmp(args->value, value, valuelen) == 0;
538	}
539
540	static bool
541	xfs_attr_match(
542	struct xfs_da_args *args,
543	unsigned int attr_flags,
544	const unsigned char *name,
545	unsigned int namelen,
546	const void *value,
547	unsigned int valuelen)
548	{
549	unsigned int mask = xfs_attr_match_mask(args);
550
551	if (args->namelen != namelen)
552	return false;
553	if ((args->attr_filter & mask) != (attr_flags & mask))
554	return false;
555	if (memcmp(args->name, name, namelen) != 0)
556	return false;
557
558	if (attr_flags & XFS_ATTR_PARENT)
559	return xfs_attr_parent_match(args, value, valuelen);
560
561	return true;
562	}
563
564	static int
565	xfs_attr_copy_value(
566	struct xfs_da_args *args,
567	unsigned char *value,
568	int valuelen)
569	{
570	/*
571	* Parent pointer lookups require the caller to specify the name and
572	* value, so don't copy anything.
573	*/
574	if (args->attr_filter & XFS_ATTR_PARENT)
575	return 0;
576
577	/*
578	* No copy if all we have to do is get the length
579	*/
580	if (!args->valuelen) {
581	args->valuelen = valuelen;
582	return 0;
583	}
584
585	/*
586	* No copy if the length of the existing buffer is too small
587	*/
588	if (args->valuelen < valuelen) {
589	args->valuelen = valuelen;
590	return -ERANGE;
591	}
592
593	if (!args->value) {
594	args->value = kvmalloc(valuelen, GFP_KERNEL | __GFP_NOLOCKDEP);
595	if (!args->value)
596	return -ENOMEM;
597	}
598	args->valuelen = valuelen;
599
600	/* remote block xattr requires IO for copy-in */
601	if (args->rmtblkno)
602	return xfs_attr_rmtval_get(args);
603
604	/*
605	* This is to prevent a GCC warning because the remote xattr case
606	* doesn't have a value to pass in. In that case, we never reach here,
607	* but GCC can't work that out and so throws a "passing NULL to
608	* memcpy" warning.
609	*/
610	if (!value)
611	return -EINVAL;
612	memcpy(args->value, value, valuelen);
613	return 0;
614	}
615
616	/*========================================================================
617	* External routines when attribute fork size < XFS_LITINO(mp).
618	*========================================================================*/
619
620	/*
621	* Query whether the total requested number of attr fork bytes of extended
622	* attribute space will be able to fit inline.
623	*
624	* Returns zero if not, else the i_forkoff fork offset to be used in the
625	* literal area for attribute data once the new bytes have been added.
626	*
627	* i_forkoff must be 8 byte aligned, hence is stored as a >>3 value;
628	* special case for dev/uuid inodes, they have fixed size data forks.
629	*/
630	int
631	xfs_attr_shortform_bytesfit(
632	struct xfs_inode *dp,
633	int bytes)
634	{
635	struct xfs_mount *mp = dp->i_mount;
636	int64_t dsize;
637	int minforkoff;
638	int maxforkoff;
639	int offset;
640
641	/*
642	* Check if the new size could fit at all first:
643	*/
644	if (bytes > XFS_LITINO(mp))
645	return 0;
646
647	/* rounded down */
648	offset = (XFS_LITINO(mp) - bytes) >> 3;
649
650	if (dp->i_df.if_format == XFS_DINODE_FMT_DEV) {
651	minforkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
652	return (offset >= minforkoff) ? minforkoff : 0;
653	}
654
655	/*
656	* If the requested numbers of bytes is smaller or equal to the
657	* current attribute fork size we can always proceed.
658	*
659	* Note that if_bytes in the data fork might actually be larger than
660	* the current data fork size is due to delalloc extents. In that
661	* case either the extent count will go down when they are converted
662	* to real extents, or the delalloc conversion will take care of the
663	* literal area rebalancing.
664	*/
665	if (bytes <= xfs_inode_attr_fork_size(dp))
666	return dp->i_forkoff;
667
668	/*
669	* For attr2 we can try to move the forkoff if there is space in the
670	* literal area
671	*/
672	dsize = dp->i_df.if_bytes;
673
674	switch (dp->i_df.if_format) {
675	case XFS_DINODE_FMT_EXTENTS:
676	/*
677	* If there is no attr fork and the data fork is extents,
678	* determine if creating the default attr fork will result
679	* in the extents form migrating to btree. If so, the
680	* minimum offset only needs to be the space required for
681	* the btree root.
682	*/
683	if (!dp->i_forkoff && dp->i_df.if_bytes >
684	xfs_default_attroffset(dp))
685	dsize = xfs_bmdr_space_calc(MINDBTPTRS);
686	break;
687	case XFS_DINODE_FMT_BTREE:
688	/*
689	* If we have a data btree then keep forkoff if we have one,
690	* otherwise we are adding a new attr, so then we set
691	* minforkoff to where the btree root can finish so we have
692	* plenty of room for attrs
693	*/
694	if (dp->i_forkoff) {
695	if (offset < dp->i_forkoff)
696	return 0;
697	return dp->i_forkoff;
698	}
699	dsize = xfs_bmap_bmdr_space(dp->i_df.if_broot);
700	break;
701	}
702
703	/*
704	* A data fork btree root must have space for at least
705	* MINDBTPTRS key/ptr pairs if the data fork is small or empty.
706	*/
707	minforkoff = max_t(int64_t, dsize, xfs_bmdr_space_calc(MINDBTPTRS));
708	minforkoff = roundup(minforkoff, 8) >> 3;
709
710	/* attr fork btree root can have at least this many key/ptr pairs */
711	maxforkoff = XFS_LITINO(mp) - xfs_bmdr_space_calc(MINABTPTRS);
712	maxforkoff = maxforkoff >> 3; /* rounded down */
713
714	if (offset >= maxforkoff)
715	return maxforkoff;
716	if (offset >= minforkoff)
717	return offset;
718	return 0;
719	}
720
721	/*
722	* Switch on the ATTR2 superblock bit (implies also FEATURES2) unless
723	* on-disk version bit says it is already set
724	*/
725	STATIC void
726	xfs_sbversion_add_attr2(
727	struct xfs_mount *mp,
728	struct xfs_trans *tp)
729	{
730	if (mp->m_sb.sb_features2 & XFS_SB_VERSION2_ATTR2BIT)
731	return;
732
733	spin_lock(&mp->m_sb_lock);
734	xfs_add_attr2(mp);
735	spin_unlock(&mp->m_sb_lock);
736	xfs_log_sb(tp);
737	}
738
739	/*
740	* Create the initial contents of a shortform attribute list.
741	*/
742	void
743	xfs_attr_shortform_create(
744	struct xfs_da_args *args)
745	{
746	struct xfs_inode *dp = args->dp;
747	struct xfs_ifork *ifp = &dp->i_af;
748	struct xfs_attr_sf_hdr *hdr;
749
750	trace_xfs_attr_sf_create(args);
751
752	ASSERT(ifp->if_bytes == 0);
753	if (ifp->if_format == XFS_DINODE_FMT_EXTENTS)
754	ifp->if_format = XFS_DINODE_FMT_LOCAL;
755
756	hdr = xfs_idata_realloc(dp, sizeof(*hdr), XFS_ATTR_FORK);
757	memset(hdr, 0, sizeof(*hdr));
758	hdr->totsize = cpu_to_be16(sizeof(*hdr));
759	xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
760	}
761
762	/*
763	* Return the entry if the attr in args is found, or NULL if not.
764	*/
765	struct xfs_attr_sf_entry *
766	xfs_attr_sf_findname(
767	struct xfs_da_args *args)
768	{
769	struct xfs_attr_sf_hdr *sf = args->dp->i_af.if_data;
770	struct xfs_attr_sf_entry *sfe;
771
772	for (sfe = xfs_attr_sf_firstentry(hdr: sf);
773	sfe < xfs_attr_sf_endptr(sf);
774	sfe = xfs_attr_sf_nextentry(sfep: sfe)) {
775	if (xfs_attr_match(args, sfe->flags, sfe->nameval,
776	sfe->namelen, &sfe->nameval[sfe->namelen],
777	sfe->valuelen))
778	return sfe;
779	}
780
781	return NULL;
782	}
783
784	/*
785	* Add a name/value pair to the shortform attribute list.
786	* Overflow from the inode has already been checked for.
787	*/
788	void
789	xfs_attr_shortform_add(
790	struct xfs_da_args *args,
791	int forkoff)
792	{
793	struct xfs_inode *dp = args->dp;
794	struct xfs_mount *mp = dp->i_mount;
795	struct xfs_ifork *ifp = &dp->i_af;
796	struct xfs_attr_sf_hdr *sf = ifp->if_data;
797	struct xfs_attr_sf_entry *sfe;
798	int size;
799
800	trace_xfs_attr_sf_add(args);
801
802	dp->i_forkoff = forkoff;
803
804	ASSERT(ifp->if_format == XFS_DINODE_FMT_LOCAL);
805	ASSERT(!xfs_attr_sf_findname(args));
806
807	size = xfs_attr_sf_entsize_byname(args->namelen, args->valuelen);
808	sf = xfs_idata_realloc(dp, size, XFS_ATTR_FORK);
809
810	sfe = xfs_attr_sf_endptr(sf);
811	sfe->namelen = args->namelen;
812	sfe->valuelen = args->valuelen;
813	sfe->flags = args->attr_filter;
814	memcpy(sfe->nameval, args->name, args->namelen);
815	memcpy(&sfe->nameval[args->namelen], args->value, args->valuelen);
816	sf->count++;
817	be16_add_cpu(&sf->totsize, size);
818	xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
819
820	xfs_sbversion_add_attr2(mp, args->trans);
821	}
822
823	/*
824	* After the last attribute is removed revert to original inode format,
825	* making all literal area available to the data fork once more.
826	*/
827	void
828	xfs_attr_fork_remove(
829	struct xfs_inode *ip,
830	struct xfs_trans *tp)
831	{
832	ASSERT(ip->i_af.if_nextents == 0);
833
834	xfs_ifork_zap_attr(ip);
835	ip->i_forkoff = 0;
836	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
837	}
838
839	/*
840	* Remove an attribute from the shortform attribute list structure.
841	*/
842	int
843	xfs_attr_sf_removename(
844	struct xfs_da_args *args)
845	{
846	struct xfs_inode *dp = args->dp;
847	struct xfs_mount *mp = dp->i_mount;
848	struct xfs_attr_sf_hdr *sf = dp->i_af.if_data;
849	struct xfs_attr_sf_entry *sfe;
850	uint16_t totsize = be16_to_cpu(sf->totsize);
851	void *next, *end;
852	int size = 0;
853
854	trace_xfs_attr_sf_remove(args);
855
856	sfe = xfs_attr_sf_findname(args);
857	if (!sfe) {
858	/*
859	* If we are recovering an operation, finding nothing to remove
860	* is not an error, it just means there was nothing to clean up.
861	*/
862	if (args->op_flags & XFS_DA_OP_RECOVERY)
863	return 0;
864	return -ENOATTR;
865	}
866
867	/*
868	* Fix up the attribute fork data, covering the hole
869	*/
870	size = xfs_attr_sf_entsize(sfep: sfe);
871	next = xfs_attr_sf_nextentry(sfep: sfe);
872	end = xfs_attr_sf_endptr(sf);
873	if (next < end)
874	memmove(sfe, next, end - next);
875	sf->count--;
876	totsize -= size;
877	sf->totsize = cpu_to_be16(totsize);
878
879	/*
880	* Fix up the start offset of the attribute fork
881	*/
882	if (totsize == sizeof(struct xfs_attr_sf_hdr) &&
883	(dp->i_df.if_format != XFS_DINODE_FMT_BTREE) &&
884	!(args->op_flags & (XFS_DA_OP_ADDNAME | XFS_DA_OP_REPLACE)) &&
885	!xfs_has_parent(mp)) {
886	xfs_attr_fork_remove(ip: dp, tp: args->trans);
887	} else {
888	xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
889	dp->i_forkoff = xfs_attr_shortform_bytesfit(dp, totsize);
890	ASSERT(dp->i_forkoff);
891	ASSERT(totsize > sizeof(struct xfs_attr_sf_hdr) ||
892	(args->op_flags & XFS_DA_OP_ADDNAME) ||
893	dp->i_df.if_format == XFS_DINODE_FMT_BTREE ||
894	xfs_has_parent(mp));
895	xfs_trans_log_inode(args->trans, dp,
896	XFS_ILOG_CORE | XFS_ILOG_ADATA);
897	}
898
899	xfs_sbversion_add_attr2(mp, args->trans);
900
901	return 0;
902	}
903
904	/*
905	* Retrieve the attribute value and length.
906	*
907	* If args->valuelen is zero, only the length needs to be returned. Unlike a
908	* lookup, we only return an error if the attribute does not exist or we can't
909	* retrieve the value.
910	*/
911	int
912	xfs_attr_shortform_getvalue(
913	struct xfs_da_args *args)
914	{
915	struct xfs_attr_sf_entry *sfe;
916
917	ASSERT(args->dp->i_af.if_format == XFS_DINODE_FMT_LOCAL);
918
919	trace_xfs_attr_sf_lookup(args);
920
921	sfe = xfs_attr_sf_findname(args);
922	if (!sfe)
923	return -ENOATTR;
924	return xfs_attr_copy_value(args, value: &sfe->nameval[args->namelen],
925	valuelen: sfe->valuelen);
926	}
927
928	/* Convert from using the shortform to the leaf format. */
929	int
930	xfs_attr_shortform_to_leaf(
931	struct xfs_da_args *args)
932	{
933	struct xfs_inode *dp = args->dp;
934	struct xfs_ifork *ifp = &dp->i_af;
935	struct xfs_attr_sf_hdr *sf = ifp->if_data;
936	struct xfs_attr_sf_entry *sfe;
937	int size = be16_to_cpu(sf->totsize);
938	struct xfs_da_args nargs;
939	char *tmpbuffer;
940	int error, i;
941	xfs_dablk_t blkno;
942	struct xfs_buf *bp;
943
944	trace_xfs_attr_sf_to_leaf(args);
945
946	tmpbuffer = kmalloc(size, GFP_KERNEL | __GFP_NOFAIL);
947	memcpy(tmpbuffer, ifp->if_data, size);
948	sf = (struct xfs_attr_sf_hdr *)tmpbuffer;
949
950	xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
951	xfs_bmap_local_to_extents_empty(args->trans, dp, XFS_ATTR_FORK);
952
953	bp = NULL;
954	error = xfs_da_grow_inode(args, &blkno);
955	if (error)
956	goto out;
957
958	ASSERT(blkno == 0);
959	error = xfs_attr3_leaf_create(args, blkno, &bp);
960	if (error)
961	goto out;
962
963	memset((char *)&nargs, 0, sizeof(nargs));
964	nargs.dp = dp;
965	nargs.geo = args->geo;
966	nargs.total = args->total;
967	nargs.whichfork = XFS_ATTR_FORK;
968	nargs.trans = args->trans;
969	nargs.op_flags = XFS_DA_OP_OKNOENT;
970	nargs.owner = args->owner;
971
972	sfe = xfs_attr_sf_firstentry(hdr: sf);
973	for (i = 0; i < sf->count; i++) {
974	nargs.name = sfe->nameval;
975	nargs.namelen = sfe->namelen;
976	nargs.value = &sfe->nameval[nargs.namelen];
977	nargs.valuelen = sfe->valuelen;
978	nargs.attr_filter = sfe->flags & XFS_ATTR_NSP_ONDISK_MASK;
979	if (!xfs_attr_check_namespace(sfe->flags)) {
980	xfs_da_mark_sick(args);
981	error = -EFSCORRUPTED;
982	goto out;
983	}
984	xfs_attr_sethash(args: &nargs);
985	error = xfs_attr3_leaf_lookup_int(leaf: bp, args: &nargs); /* set a->index */
986	ASSERT(error == -ENOATTR);
987	if (!xfs_attr3_leaf_add(bp, &nargs))
988	ASSERT(0);
989	sfe = xfs_attr_sf_nextentry(sfep: sfe);
990	}
991	error = 0;
992	out:
993	kfree(tmpbuffer);
994	return error;
995	}
996
997	/*
998	* Check a leaf attribute block to see if all the entries would fit into
999	* a shortform attribute list.
1000	*/
1001	int
1002	xfs_attr_shortform_allfit(
1003	struct xfs_buf *bp,
1004	struct xfs_inode *dp)
1005	{
1006	struct xfs_attr_leafblock *leaf;
1007	struct xfs_attr_leaf_entry *entry;
1008	xfs_attr_leaf_name_local_t *name_loc;
1009	struct xfs_attr3_icleaf_hdr leafhdr;
1010	int bytes;
1011	int i;
1012	struct xfs_mount *mp = bp->b_mount;
1013
1014	leaf = bp->b_addr;
1015	xfs_attr3_leaf_hdr_from_disk(geo: mp->m_attr_geo, to: &leafhdr, from: leaf);
1016	entry = xfs_attr3_leaf_entryp(leafp: leaf);
1017
1018	bytes = sizeof(struct xfs_attr_sf_hdr);
1019	for (i = 0; i < leafhdr.count; entry++, i++) {
1020	if (entry->flags & XFS_ATTR_INCOMPLETE)
1021	continue; /* don't copy partial entries */
1022	if (!(entry->flags & XFS_ATTR_LOCAL))
1023	return 0;
1024	name_loc = xfs_attr3_leaf_name_local(leafp: leaf, idx: i);
1025	if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX)
1026	return 0;
1027	if (be16_to_cpu(name_loc->valuelen) >= XFS_ATTR_SF_ENTSIZE_MAX)
1028	return 0;
1029	bytes += xfs_attr_sf_entsize_byname(name_loc->namelen,
1030	be16_to_cpu(name_loc->valuelen));
1031	}
1032	if ((dp->i_df.if_format != XFS_DINODE_FMT_BTREE) &&
1033	(bytes == sizeof(struct xfs_attr_sf_hdr)))
1034	return -1;
1035	return xfs_attr_shortform_bytesfit(dp, bytes);
1036	}
1037
1038	/* Verify the consistency of a raw inline attribute fork. */
1039	xfs_failaddr_t
1040	xfs_attr_shortform_verify(
1041	struct xfs_attr_sf_hdr *sfp,
1042	size_t size)
1043	{
1044	struct xfs_attr_sf_entry *sfep = xfs_attr_sf_firstentry(hdr: sfp);
1045	struct xfs_attr_sf_entry *next_sfep;
1046	char *endp;
1047	int i;
1048
1049	/*
1050	* Give up if the attribute is way too short.
1051	*/
1052	if (size < sizeof(struct xfs_attr_sf_hdr))
1053	return __this_address;
1054
1055	endp = (char *)sfp + size;
1056
1057	/* Check all reported entries */
1058	for (i = 0; i < sfp->count; i++) {
1059	/*
1060	* struct xfs_attr_sf_entry has a variable length.
1061	* Check the fixed-offset parts of the structure are
1062	* within the data buffer.
1063	* xfs_attr_sf_entry is defined with a 1-byte variable
1064	* array at the end, so we must subtract that off.
1065	*/
1066	if (((char *)sfep + sizeof(*sfep)) >= endp)
1067	return __this_address;
1068
1069	/* Don't allow names with known bad length. */
1070	if (sfep->namelen == 0)
1071	return __this_address;
1072
1073	/*
1074	* Check that the variable-length part of the structure is
1075	* within the data buffer. The next entry starts after the
1076	* name component, so nextentry is an acceptable test.
1077	*/
1078	next_sfep = xfs_attr_sf_nextentry(sfep);
1079	if ((char *)next_sfep > endp)
1080	return __this_address;
1081
1082	/*
1083	* Check for unknown flags. Short form doesn't support
1084	* the incomplete or local bits, so we can use the namespace
1085	* mask here.
1086	*/
1087	if (sfep->flags & ~XFS_ATTR_NSP_ONDISK_MASK)
1088	return __this_address;
1089
1090	/*
1091	* Check for invalid namespace combinations. We only allow
1092	* one namespace flag per xattr, so we can just count the
1093	* bits (i.e. hweight) here.
1094	*/
1095	if (!xfs_attr_check_namespace(sfep->flags))
1096	return __this_address;
1097
1098	sfep = next_sfep;
1099	}
1100	if ((void *)sfep != (void *)endp)
1101	return __this_address;
1102
1103	return NULL;
1104	}
1105
1106	/*
1107	* Convert a leaf attribute list to shortform attribute list
1108	*/
1109	int
1110	xfs_attr3_leaf_to_shortform(
1111	struct xfs_buf *bp,
1112	struct xfs_da_args *args,
1113	int forkoff)
1114	{
1115	struct xfs_attr_leafblock *leaf;
1116	struct xfs_attr3_icleaf_hdr ichdr;
1117	struct xfs_attr_leaf_entry *entry;
1118	struct xfs_attr_leaf_name_local *name_loc;
1119	struct xfs_da_args nargs;
1120	struct xfs_inode *dp = args->dp;
1121	char *tmpbuffer;
1122	int error;
1123	int i;
1124
1125	trace_xfs_attr_leaf_to_sf(args);
1126
1127	tmpbuffer = kvmalloc(args->geo->blksize, GFP_KERNEL | __GFP_NOFAIL);
1128	memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1129
1130	leaf = (xfs_attr_leafblock_t *)tmpbuffer;
1131	xfs_attr3_leaf_hdr_from_disk(geo: args->geo, to: &ichdr, from: leaf);
1132	entry = xfs_attr3_leaf_entryp(leafp: leaf);
1133
1134	/* XXX (dgc): buffer is about to be marked stale - why zero it? */
1135	memset(bp->b_addr, 0, args->geo->blksize);
1136
1137	/*
1138	* Clean out the prior contents of the attribute list.
1139	*/
1140	error = xfs_da_shrink_inode(args, 0, bp);
1141	if (error)
1142	goto out;
1143
1144	if (forkoff == -1) {
1145	/*
1146	* Don't remove the attr fork if this operation is the first
1147	* part of a attr replace operations. We're going to add a new
1148	* attr immediately, so we need to keep the attr fork around in
1149	* this case.
1150	*/
1151	if (!(args->op_flags & XFS_DA_OP_REPLACE)) {
1152	ASSERT(dp->i_df.if_format != XFS_DINODE_FMT_BTREE);
1153	xfs_attr_fork_remove(ip: dp, tp: args->trans);
1154	}
1155	goto out;
1156	}
1157
1158	xfs_attr_shortform_create(args);
1159
1160	/*
1161	* Copy the attributes
1162	*/
1163	memset((char *)&nargs, 0, sizeof(nargs));
1164	nargs.geo = args->geo;
1165	nargs.dp = dp;
1166	nargs.total = args->total;
1167	nargs.whichfork = XFS_ATTR_FORK;
1168	nargs.trans = args->trans;
1169	nargs.op_flags = XFS_DA_OP_OKNOENT;
1170	nargs.owner = args->owner;
1171
1172	for (i = 0; i < ichdr.count; entry++, i++) {
1173	if (entry->flags & XFS_ATTR_INCOMPLETE)
1174	continue; /* don't copy partial entries */
1175	if (!entry->nameidx)
1176	continue;
1177	ASSERT(entry->flags & XFS_ATTR_LOCAL);
1178	name_loc = xfs_attr3_leaf_name_local(leafp: leaf, idx: i);
1179	nargs.name = name_loc->nameval;
1180	nargs.namelen = name_loc->namelen;
1181	nargs.value = &name_loc->nameval[nargs.namelen];
1182	nargs.valuelen = be16_to_cpu(name_loc->valuelen);
1183	nargs.hashval = be32_to_cpu(entry->hashval);
1184	nargs.attr_filter = entry->flags & XFS_ATTR_NSP_ONDISK_MASK;
1185	xfs_attr_shortform_add(args: &nargs, forkoff);
1186	}
1187	error = 0;
1188
1189	out:
1190	kvfree(tmpbuffer);
1191	return error;
1192	}
1193
1194	/*
1195	* Convert from using a single leaf to a root node and a leaf.
1196	*/
1197	int
1198	xfs_attr3_leaf_to_node(
1199	struct xfs_da_args *args)
1200	{
1201	struct xfs_attr_leafblock *leaf;
1202	struct xfs_attr3_icleaf_hdr icleafhdr;
1203	struct xfs_attr_leaf_entry *entries;
1204	struct xfs_da3_icnode_hdr icnodehdr;
1205	struct xfs_da_intnode *node;
1206	struct xfs_inode *dp = args->dp;
1207	struct xfs_mount *mp = dp->i_mount;
1208	struct xfs_buf *bp1 = NULL;
1209	struct xfs_buf *bp2 = NULL;
1210	xfs_dablk_t blkno;
1211	int error;
1212
1213	trace_xfs_attr_leaf_to_node(args);
1214
1215	if (XFS_TEST_ERROR(mp, XFS_ERRTAG_ATTR_LEAF_TO_NODE)) {
1216	error = -EIO;
1217	goto out;
1218	}
1219
1220	error = xfs_da_grow_inode(args, &blkno);
1221	if (error)
1222	goto out;
1223	error = xfs_attr3_leaf_read(args->trans, dp, args->owner, 0, &bp1);
1224	if (error)
1225	goto out;
1226
1227	error = xfs_da_get_buf(args->trans, dp, blkno, &bp2, XFS_ATTR_FORK);
1228	if (error)
1229	goto out;
1230
1231	/*
1232	* Copy leaf to new buffer and log it.
1233	*/
1234	xfs_da_buf_copy(bp2, bp1, args->geo->blksize);
1235	xfs_trans_log_buf(args->trans, bp2, 0, args->geo->blksize - 1);
1236
1237	/*
1238	* Set up the new root node.
1239	*/
1240	error = xfs_da3_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK);
1241	if (error)
1242	goto out;
1243	node = bp1->b_addr;
1244	xfs_da3_node_hdr_from_disk(mp, to: &icnodehdr, from: node);
1245
1246	leaf = bp2->b_addr;
1247	xfs_attr3_leaf_hdr_from_disk(geo: args->geo, to: &icleafhdr, from: leaf);
1248	entries = xfs_attr3_leaf_entryp(leafp: leaf);
1249
1250	/* both on-disk, don't endian-flip twice */
1251	icnodehdr.btree[0].hashval = entries[icleafhdr.count - 1].hashval;
1252	icnodehdr.btree[0].before = cpu_to_be32(blkno);
1253	icnodehdr.count = 1;
1254	xfs_da3_node_hdr_to_disk(mp: dp->i_mount, to: node, from: &icnodehdr);
1255	xfs_trans_log_buf(args->trans, bp1, 0, args->geo->blksize - 1);
1256	error = 0;
1257	out:
1258	return error;
1259	}
1260
1261	/*========================================================================
1262	* Routines used for growing the Btree.
1263	*========================================================================*/
1264
1265	/*
1266	* Create the initial contents of a leaf attribute list
1267	* or a leaf in a node attribute list.
1268	*/
1269	STATIC int
1270	xfs_attr3_leaf_create(
1271	struct xfs_da_args *args,
1272	xfs_dablk_t blkno,
1273	struct xfs_buf **bpp)
1274	{
1275	struct xfs_attr_leafblock *leaf;
1276	struct xfs_attr3_icleaf_hdr ichdr;
1277	struct xfs_inode *dp = args->dp;
1278	struct xfs_mount *mp = dp->i_mount;
1279	struct xfs_buf *bp;
1280	int error;
1281
1282	trace_xfs_attr_leaf_create(args);
1283
1284	error = xfs_da_get_buf(args->trans, args->dp, blkno, &bp,
1285	XFS_ATTR_FORK);
1286	if (error)
1287	return error;
1288	bp->b_ops = &xfs_attr3_leaf_buf_ops;
1289	xfs_trans_buf_set_type(args->trans, bp, XFS_BLFT_ATTR_LEAF_BUF);
1290	leaf = bp->b_addr;
1291	memset(leaf, 0, args->geo->blksize);
1292
1293	memset(&ichdr, 0, sizeof(ichdr));
1294	ichdr.firstused = args->geo->blksize;
1295
1296	if (xfs_has_crc(mp)) {
1297	struct xfs_da3_blkinfo *hdr3 = bp->b_addr;
1298
1299	ichdr.magic = XFS_ATTR3_LEAF_MAGIC;
1300
1301	hdr3->blkno = cpu_to_be64(xfs_buf_daddr(bp));
1302	hdr3->owner = cpu_to_be64(args->owner);
1303	uuid_copy(&hdr3->uuid, &mp->m_sb.sb_meta_uuid);
1304
1305	ichdr.freemap[0].base = sizeof(struct xfs_attr3_leaf_hdr);
1306	} else {
1307	ichdr.magic = XFS_ATTR_LEAF_MAGIC;
1308	ichdr.freemap[0].base = sizeof(struct xfs_attr_leaf_hdr);
1309	}
1310	ichdr.freemap[0].size = ichdr.firstused - ichdr.freemap[0].base;
1311
1312	xfs_attr3_leaf_hdr_to_disk(geo: args->geo, to: leaf, from: &ichdr);
1313	xfs_trans_log_buf(args->trans, bp, 0, args->geo->blksize - 1);
1314
1315	*bpp = bp;
1316	return 0;
1317	}
1318
1319	/*
1320	* Split the leaf node, rebalance, then add the new entry.
1321	*
1322	* Returns 0 if the entry was added, 1 if a further split is needed or a
1323	* negative error number otherwise.
1324	*/
1325	int
1326	xfs_attr3_leaf_split(
1327	struct xfs_da_state *state,
1328	struct xfs_da_state_blk *oldblk,
1329	struct xfs_da_state_blk *newblk)
1330	{
1331	bool added;
1332	xfs_dablk_t blkno;
1333	int error;
1334
1335	trace_xfs_attr_leaf_split(state->args);
1336
1337	/*
1338	* Allocate space for a new leaf node.
1339	*/
1340	ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
1341	error = xfs_da_grow_inode(state->args, &blkno);
1342	if (error)
1343	return error;
1344	error = xfs_attr3_leaf_create(state->args, blkno, &newblk->bp);
1345	if (error)
1346	return error;
1347	newblk->blkno = blkno;
1348	newblk->magic = XFS_ATTR_LEAF_MAGIC;
1349
1350	/*
1351	* Rebalance the entries across the two leaves.
1352	* NOTE: rebalance() currently depends on the 2nd block being empty.
1353	*/
1354	xfs_attr3_leaf_rebalance(state, oldblk, newblk);
1355	error = xfs_da3_blk_link(state, old_blk: oldblk, new_blk: newblk);
1356	if (error)
1357	return error;
1358
1359	/*
1360	* Save info on "old" attribute for "atomic rename" ops, leaf_add()
1361	* modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
1362	* "new" attrs info. Will need the "old" info to remove it later.
1363	*
1364	* Insert the "new" entry in the correct block.
1365	*/
1366	if (state->inleaf) {
1367	trace_xfs_attr_leaf_add_old(state->args);
1368	added = xfs_attr3_leaf_add(oldblk->bp, state->args);
1369	} else {
1370	trace_xfs_attr_leaf_add_new(state->args);
1371	added = xfs_attr3_leaf_add(newblk->bp, state->args);
1372	}
1373
1374	/*
1375	* Update last hashval in each block since we added the name.
1376	*/
1377	oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
1378	newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
1379	if (!added)
1380	return 1;
1381	return 0;
1382	}
1383
1384	/*
1385	* Add a name to the leaf attribute list structure.
1386	*/
1387	bool
1388	xfs_attr3_leaf_add(
1389	struct xfs_buf *bp,
1390	struct xfs_da_args *args)
1391	{
1392	struct xfs_attr_leafblock *leaf;
1393	struct xfs_attr3_icleaf_hdr ichdr;
1394	int tablesize;
1395	int entsize;
1396	bool added = true;
1397	int sum;
1398	int tmp;
1399	int i;
1400
1401	trace_xfs_attr_leaf_add(args);
1402
1403	leaf = bp->b_addr;
1404	xfs_attr3_leaf_hdr_from_disk(geo: args->geo, to: &ichdr, from: leaf);
1405	ASSERT(args->index >= 0 && args->index <= ichdr.count);
1406	entsize = xfs_attr_leaf_newentsize(args, NULL);
1407
1408	/*
1409	* Search through freemap for first-fit on new name length.
1410	* (may need to figure in size of entry struct too)
1411	*/
1412	tablesize = (ichdr.count + 1) * sizeof(xfs_attr_leaf_entry_t)
1413	+ xfs_attr3_leaf_hdr_size(leafp: leaf);
1414	for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE - 1; i >= 0; i--) {
1415	if (tablesize > ichdr.firstused) {
1416	sum += ichdr.freemap[i].size;
1417	continue;
1418	}
1419	if (!ichdr.freemap[i].size)
1420	continue; /* no space in this map */
1421	tmp = entsize;
1422	if (ichdr.freemap[i].base < ichdr.firstused)
1423	tmp += sizeof(xfs_attr_leaf_entry_t);
1424	if (ichdr.freemap[i].size >= tmp) {
1425	xfs_attr3_leaf_add_work(bp, &ichdr, args, i);
1426	goto out_log_hdr;
1427	}
1428	sum += ichdr.freemap[i].size;
1429	}
1430
1431	/*
1432	* If there are no holes in the address space of the block,
1433	* and we don't have enough freespace, then compaction will do us
1434	* no good and we should just give up.
1435	*/
1436	if (!ichdr.holes && sum < entsize)
1437	return false;
1438
1439	/*
1440	* Compact the entries to coalesce free space.
1441	* This may change the hdr->count via dropping INCOMPLETE entries.
1442	*/
1443	xfs_attr3_leaf_compact(args, &ichdr, bp);
1444
1445	/*
1446	* After compaction, the block is guaranteed to have only one
1447	* free region, in freemap[0]. If it is not big enough, give up.
1448	*/
1449	if (ichdr.freemap[0].size < (entsize + sizeof(xfs_attr_leaf_entry_t))) {
1450	added = false;
1451	goto out_log_hdr;
1452	}
1453
1454	xfs_attr3_leaf_add_work(bp, &ichdr, args, 0);
1455
1456	out_log_hdr:
1457	xfs_attr3_leaf_hdr_to_disk(geo: args->geo, to: leaf, from: &ichdr);
1458	xfs_trans_log_buf(args->trans, bp,
1459	XFS_DA_LOGRANGE(leaf, &leaf->hdr,
1460	xfs_attr3_leaf_hdr_size(leaf)));
1461	return added;
1462	}
1463
1464	/*
1465	* Add a name to a leaf attribute list structure.
1466	*/
1467	STATIC void
1468	xfs_attr3_leaf_add_work(
1469	struct xfs_buf *bp,
1470	struct xfs_attr3_icleaf_hdr *ichdr,
1471	struct xfs_da_args *args,
1472	int mapindex)
1473	{
1474	struct xfs_attr_leafblock *leaf;
1475	struct xfs_attr_leaf_entry *entry;
1476	struct xfs_attr_leaf_name_local *name_loc;
1477	struct xfs_attr_leaf_name_remote *name_rmt;
1478	struct xfs_mount *mp;
1479	int tmp;
1480	int i;
1481
1482	trace_xfs_attr_leaf_add_work(args);
1483
1484	leaf = bp->b_addr;
1485	ASSERT(mapindex >= 0 && mapindex < XFS_ATTR_LEAF_MAPSIZE);
1486	ASSERT(args->index >= 0 && args->index <= ichdr->count);
1487
1488	/*
1489	* Force open some space in the entry array and fill it in.
1490	*/
1491	entry = &xfs_attr3_leaf_entryp(leafp: leaf)[args->index];
1492	if (args->index < ichdr->count) {
1493	tmp = ichdr->count - args->index;
1494	tmp *= sizeof(xfs_attr_leaf_entry_t);
1495	memmove(entry + 1, entry, tmp);
1496	xfs_trans_log_buf(args->trans, bp,
1497	XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1498	}
1499	ichdr->count++;
1500
1501	/*
1502	* Allocate space for the new string (at the end of the run).
1503	*/
1504	mp = args->trans->t_mountp;
1505	ASSERT(ichdr->freemap[mapindex].base < args->geo->blksize);
1506	ASSERT((ichdr->freemap[mapindex].base & 0x3) == 0);
1507	ASSERT(ichdr->freemap[mapindex].size >=
1508	xfs_attr_leaf_newentsize(args, NULL));
1509	ASSERT(ichdr->freemap[mapindex].size < args->geo->blksize);
1510	ASSERT((ichdr->freemap[mapindex].size & 0x3) == 0);
1511
1512	ichdr->freemap[mapindex].size -= xfs_attr_leaf_newentsize(args, local: &tmp);
1513
1514	entry->nameidx = cpu_to_be16(ichdr->freemap[mapindex].base +
1515	ichdr->freemap[mapindex].size);
1516	entry->hashval = cpu_to_be32(args->hashval);
1517	entry->flags = args->attr_filter;
1518	if (tmp)
1519	entry->flags |= XFS_ATTR_LOCAL;
1520	if (args->op_flags & XFS_DA_OP_REPLACE) {
1521	if (!(args->op_flags & XFS_DA_OP_LOGGED))
1522	entry->flags |= XFS_ATTR_INCOMPLETE;
1523	if ((args->blkno2 == args->blkno) &&
1524	(args->index2 <= args->index)) {
1525	args->index2++;
1526	}
1527	}
1528	xfs_trans_log_buf(args->trans, bp,
1529	XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1530	ASSERT((args->index == 0) ||
1531	(be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1532	ASSERT((args->index == ichdr->count - 1) ||
1533	(be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1534
1535	/*
1536	* For "remote" attribute values, simply note that we need to
1537	* allocate space for the "remote" value. We can't actually
1538	* allocate the extents in this transaction, and we can't decide
1539	* which blocks they should be as we might allocate more blocks
1540	* as part of this transaction (a split operation for example).
1541	*/
1542	if (entry->flags & XFS_ATTR_LOCAL) {
1543	name_loc = xfs_attr3_leaf_name_local(leafp: leaf, idx: args->index);
1544	name_loc->namelen = args->namelen;
1545	name_loc->valuelen = cpu_to_be16(args->valuelen);
1546	memcpy((char *)name_loc->nameval, args->name, args->namelen);
1547	memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1548	be16_to_cpu(name_loc->valuelen));
1549	} else {
1550	name_rmt = xfs_attr3_leaf_name_remote(leafp: leaf, idx: args->index);
1551	name_rmt->namelen = args->namelen;
1552	memcpy((char *)name_rmt->name, args->name, args->namelen);
1553	entry->flags |= XFS_ATTR_INCOMPLETE;
1554	/* just in case */
1555	name_rmt->valuelen = 0;
1556	name_rmt->valueblk = 0;
1557	args->rmtblkno = 1;
1558	args->rmtblkcnt = xfs_attr3_rmt_blocks(mp, attrlen: args->valuelen);
1559	args->rmtvaluelen = args->valuelen;
1560	}
1561	xfs_trans_log_buf(args->trans, bp,
1562	XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1563	xfs_attr_leaf_entsize(leaf, args->index)));
1564
1565	/*
1566	* Update the control info for this leaf node
1567	*/
1568	if (be16_to_cpu(entry->nameidx) < ichdr->firstused)
1569	ichdr->firstused = be16_to_cpu(entry->nameidx);
1570
1571	ASSERT(ichdr->firstused >= ichdr->count * sizeof(xfs_attr_leaf_entry_t)
1572	+ xfs_attr3_leaf_hdr_size(leafp: leaf));
1573	tmp = (ichdr->count - 1) * sizeof(xfs_attr_leaf_entry_t)
1574	+ xfs_attr3_leaf_hdr_size(leafp: leaf);
1575
1576	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1577	if (ichdr->freemap[i].base == tmp) {
1578	ichdr->freemap[i].base += sizeof(xfs_attr_leaf_entry_t);
1579	ichdr->freemap[i].size -=
1580	min_t(uint16_t, ichdr->freemap[i].size,
1581	sizeof(xfs_attr_leaf_entry_t));
1582	}
1583	}
1584	ichdr->usedbytes += xfs_attr_leaf_entsize(leaf, args->index);
1585	}
1586
1587	/*
1588	* Garbage collect a leaf attribute list block by copying it to a new buffer.
1589	*/
1590	STATIC void
1591	xfs_attr3_leaf_compact(
1592	struct xfs_da_args *args,
1593	struct xfs_attr3_icleaf_hdr *ichdr_dst,
1594	struct xfs_buf *bp)
1595	{
1596	struct xfs_attr_leafblock *leaf_src;
1597	struct xfs_attr_leafblock *leaf_dst;
1598	struct xfs_attr3_icleaf_hdr ichdr_src;
1599	struct xfs_trans *trans = args->trans;
1600	char *tmpbuffer;
1601
1602	trace_xfs_attr_leaf_compact(args);
1603
1604	tmpbuffer = kvmalloc(args->geo->blksize, GFP_KERNEL | __GFP_NOFAIL);
1605	memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1606	memset(bp->b_addr, 0, args->geo->blksize);
1607	leaf_src = (xfs_attr_leafblock_t *)tmpbuffer;
1608	leaf_dst = bp->b_addr;
1609
1610	/*
1611	* Copy the on-disk header back into the destination buffer to ensure
1612	* all the information in the header that is not part of the incore
1613	* header structure is preserved.
1614	*/
1615	memcpy(bp->b_addr, tmpbuffer, xfs_attr3_leaf_hdr_size(leafp: leaf_src));
1616
1617	/* Initialise the incore headers */
1618	ichdr_src = *ichdr_dst; /* struct copy */
1619	ichdr_dst->firstused = args->geo->blksize;
1620	ichdr_dst->usedbytes = 0;
1621	ichdr_dst->count = 0;
1622	ichdr_dst->holes = 0;
1623	ichdr_dst->freemap[0].base = xfs_attr3_leaf_hdr_size(leafp: leaf_src);
1624	ichdr_dst->freemap[0].size = ichdr_dst->firstused -
1625	ichdr_dst->freemap[0].base;
1626
1627	/* write the header back to initialise the underlying buffer */
1628	xfs_attr3_leaf_hdr_to_disk(geo: args->geo, to: leaf_dst, from: ichdr_dst);
1629
1630	/*
1631	* Copy all entry's in the same (sorted) order,
1632	* but allocate name/value pairs packed and in sequence.
1633	*/
1634	xfs_attr3_leaf_moveents(args, leaf_src, &ichdr_src, 0,
1635	leaf_dst, ichdr_dst, 0, ichdr_src.count);
1636	/*
1637	* this logs the entire buffer, but the caller must write the header
1638	* back to the buffer when it is finished modifying it.
1639	*/
1640	xfs_trans_log_buf(trans, bp, 0, args->geo->blksize - 1);
1641
1642	kvfree(tmpbuffer);
1643	}
1644
1645	/*
1646	* Compare two leaf blocks "order".
1647	* Return 0 unless leaf2 should go before leaf1.
1648	*/
1649	static int
1650	xfs_attr3_leaf_order(
1651	struct xfs_buf *leaf1_bp,
1652	struct xfs_attr3_icleaf_hdr *leaf1hdr,
1653	struct xfs_buf *leaf2_bp,
1654	struct xfs_attr3_icleaf_hdr *leaf2hdr)
1655	{
1656	struct xfs_attr_leaf_entry *entries1;
1657	struct xfs_attr_leaf_entry *entries2;
1658
1659	entries1 = xfs_attr3_leaf_entryp(leafp: leaf1_bp->b_addr);
1660	entries2 = xfs_attr3_leaf_entryp(leafp: leaf2_bp->b_addr);
1661	if (leaf1hdr->count > 0 && leaf2hdr->count > 0 &&
1662	((be32_to_cpu(entries2[0].hashval) <
1663	be32_to_cpu(entries1[0].hashval)) ||
1664	(be32_to_cpu(entries2[leaf2hdr->count - 1].hashval) <
1665	be32_to_cpu(entries1[leaf1hdr->count - 1].hashval)))) {
1666	return 1;
1667	}
1668	return 0;
1669	}
1670
1671	int
1672	xfs_attr_leaf_order(
1673	struct xfs_buf *leaf1_bp,
1674	struct xfs_buf *leaf2_bp)
1675	{
1676	struct xfs_attr3_icleaf_hdr ichdr1;
1677	struct xfs_attr3_icleaf_hdr ichdr2;
1678	struct xfs_mount *mp = leaf1_bp->b_mount;
1679
1680	xfs_attr3_leaf_hdr_from_disk(geo: mp->m_attr_geo, to: &ichdr1, from: leaf1_bp->b_addr);
1681	xfs_attr3_leaf_hdr_from_disk(geo: mp->m_attr_geo, to: &ichdr2, from: leaf2_bp->b_addr);
1682	return xfs_attr3_leaf_order(leaf1_bp, leaf1hdr: &ichdr1, leaf2_bp, leaf2hdr: &ichdr2);
1683	}
1684
1685	/*
1686	* Redistribute the attribute list entries between two leaf nodes,
1687	* taking into account the size of the new entry.
1688	*
1689	* NOTE: if new block is empty, then it will get the upper half of the
1690	* old block. At present, all (one) callers pass in an empty second block.
1691	*
1692	* This code adjusts the args->index/blkno and args->index2/blkno2 fields
1693	* to match what it is doing in splitting the attribute leaf block. Those
1694	* values are used in "atomic rename" operations on attributes. Note that
1695	* the "new" and "old" values can end up in different blocks.
1696	*/
1697	STATIC void
1698	xfs_attr3_leaf_rebalance(
1699	struct xfs_da_state *state,
1700	struct xfs_da_state_blk *blk1,
1701	struct xfs_da_state_blk *blk2)
1702	{
1703	struct xfs_da_args *args;
1704	struct xfs_attr_leafblock *leaf1;
1705	struct xfs_attr_leafblock *leaf2;
1706	struct xfs_attr3_icleaf_hdr ichdr1;
1707	struct xfs_attr3_icleaf_hdr ichdr2;
1708	struct xfs_attr_leaf_entry *entries1;
1709	struct xfs_attr_leaf_entry *entries2;
1710	int count;
1711	int totallen;
1712	int max;
1713	int space;
1714	int swap;
1715
1716	/*
1717	* Set up environment.
1718	*/
1719	ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1720	ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1721	leaf1 = blk1->bp->b_addr;
1722	leaf2 = blk2->bp->b_addr;
1723	xfs_attr3_leaf_hdr_from_disk(geo: state->args->geo, to: &ichdr1, from: leaf1);
1724	xfs_attr3_leaf_hdr_from_disk(geo: state->args->geo, to: &ichdr2, from: leaf2);
1725	ASSERT(ichdr2.count == 0);
1726	args = state->args;
1727
1728	trace_xfs_attr_leaf_rebalance(args);
1729
1730	/*
1731	* Check ordering of blocks, reverse if it makes things simpler.
1732	*
1733	* NOTE: Given that all (current) callers pass in an empty
1734	* second block, this code should never set "swap".
1735	*/
1736	swap = 0;
1737	if (xfs_attr3_leaf_order(leaf1_bp: blk1->bp, leaf1hdr: &ichdr1, leaf2_bp: blk2->bp, leaf2hdr: &ichdr2)) {
1738	swap(blk1, blk2);
1739
1740	/* swap structures rather than reconverting them */
1741	swap(ichdr1, ichdr2);
1742
1743	leaf1 = blk1->bp->b_addr;
1744	leaf2 = blk2->bp->b_addr;
1745	swap = 1;
1746	}
1747
1748	/*
1749	* Examine entries until we reduce the absolute difference in
1750	* byte usage between the two blocks to a minimum. Then get
1751	* the direction to copy and the number of elements to move.
1752	*
1753	* "inleaf" is true if the new entry should be inserted into blk1.
1754	* If "swap" is also true, then reverse the sense of "inleaf".
1755	*/
1756	state->inleaf = xfs_attr3_leaf_figure_balance(state, blk1, &ichdr1,
1757	blk2, &ichdr2,
1758	&count, &totallen);
1759	if (swap)
1760	state->inleaf = !state->inleaf;
1761
1762	/*
1763	* Move any entries required from leaf to leaf:
1764	*/
1765	if (count < ichdr1.count) {
1766	/*
1767	* Figure the total bytes to be added to the destination leaf.
1768	*/
1769	/* number entries being moved */
1770	count = ichdr1.count - count;
1771	space = ichdr1.usedbytes - totallen;
1772	space += count * sizeof(xfs_attr_leaf_entry_t);
1773
1774	/*
1775	* leaf2 is the destination, compact it if it looks tight.
1776	*/
1777	max = ichdr2.firstused - xfs_attr3_leaf_hdr_size(leafp: leaf1);
1778	max -= ichdr2.count * sizeof(xfs_attr_leaf_entry_t);
1779	if (space > max)
1780	xfs_attr3_leaf_compact(args, &ichdr2, blk2->bp);
1781
1782	/*
1783	* Move high entries from leaf1 to low end of leaf2.
1784	*/
1785	xfs_attr3_leaf_moveents(args, leaf1, &ichdr1,
1786	ichdr1.count - count, leaf2, &ichdr2, 0, count);
1787
1788	} else if (count > ichdr1.count) {
1789	/*
1790	* I assert that since all callers pass in an empty
1791	* second buffer, this code should never execute.
1792	*/
1793	ASSERT(0);
1794
1795	/*
1796	* Figure the total bytes to be added to the destination leaf.
1797	*/
1798	/* number entries being moved */
1799	count -= ichdr1.count;
1800	space = totallen - ichdr1.usedbytes;
1801	space += count * sizeof(xfs_attr_leaf_entry_t);
1802
1803	/*
1804	* leaf1 is the destination, compact it if it looks tight.
1805	*/
1806	max = ichdr1.firstused - xfs_attr3_leaf_hdr_size(leafp: leaf1);
1807	max -= ichdr1.count * sizeof(xfs_attr_leaf_entry_t);
1808	if (space > max)
1809	xfs_attr3_leaf_compact(args, &ichdr1, blk1->bp);
1810
1811	/*
1812	* Move low entries from leaf2 to high end of leaf1.
1813	*/
1814	xfs_attr3_leaf_moveents(args, leaf2, &ichdr2, 0, leaf1, &ichdr1,
1815	ichdr1.count, count);
1816	}
1817
1818	xfs_attr3_leaf_hdr_to_disk(geo: state->args->geo, to: leaf1, from: &ichdr1);
1819	xfs_attr3_leaf_hdr_to_disk(geo: state->args->geo, to: leaf2, from: &ichdr2);
1820	xfs_trans_log_buf(args->trans, blk1->bp, 0, args->geo->blksize - 1);
1821	xfs_trans_log_buf(args->trans, blk2->bp, 0, args->geo->blksize - 1);
1822
1823	/*
1824	* Copy out last hashval in each block for B-tree code.
1825	*/
1826	entries1 = xfs_attr3_leaf_entryp(leafp: leaf1);
1827	entries2 = xfs_attr3_leaf_entryp(leafp: leaf2);
1828	blk1->hashval = be32_to_cpu(entries1[ichdr1.count - 1].hashval);
1829	blk2->hashval = be32_to_cpu(entries2[ichdr2.count - 1].hashval);
1830
1831	/*
1832	* Adjust the expected index for insertion.
1833	* NOTE: this code depends on the (current) situation that the
1834	* second block was originally empty.
1835	*
1836	* If the insertion point moved to the 2nd block, we must adjust
1837	* the index. We must also track the entry just following the
1838	* new entry for use in an "atomic rename" operation, that entry
1839	* is always the "old" entry and the "new" entry is what we are
1840	* inserting. The index/blkno fields refer to the "old" entry,
1841	* while the index2/blkno2 fields refer to the "new" entry.
1842	*/
1843	if (blk1->index > ichdr1.count) {
1844	ASSERT(state->inleaf == 0);
1845	blk2->index = blk1->index - ichdr1.count;
1846	args->index = args->index2 = blk2->index;
1847	args->blkno = args->blkno2 = blk2->blkno;
1848	} else if (blk1->index == ichdr1.count) {
1849	if (state->inleaf) {
1850	args->index = blk1->index;
1851	args->blkno = blk1->blkno;
1852	args->index2 = 0;
1853	args->blkno2 = blk2->blkno;
1854	} else {
1855	/*
1856	* On a double leaf split, the original attr location
1857	* is already stored in blkno2/index2, so don't
1858	* overwrite it overwise we corrupt the tree.
1859	*/
1860	blk2->index = blk1->index - ichdr1.count;
1861	args->index = blk2->index;
1862	args->blkno = blk2->blkno;
1863	if (!state->extravalid) {
1864	/*
1865	* set the new attr location to match the old
1866	* one and let the higher level split code
1867	* decide where in the leaf to place it.
1868	*/
1869	args->index2 = blk2->index;
1870	args->blkno2 = blk2->blkno;
1871	}
1872	}
1873	} else {
1874	ASSERT(state->inleaf == 1);
1875	args->index = args->index2 = blk1->index;
1876	args->blkno = args->blkno2 = blk1->blkno;
1877	}
1878	}
1879
1880	/*
1881	* Examine entries until we reduce the absolute difference in
1882	* byte usage between the two blocks to a minimum.
1883	* GROT: Is this really necessary? With other than a 512 byte blocksize,
1884	* GROT: there will always be enough room in either block for a new entry.
1885	* GROT: Do a double-split for this case?
1886	*/
1887	STATIC int
1888	xfs_attr3_leaf_figure_balance(
1889	struct xfs_da_state *state,
1890	struct xfs_da_state_blk *blk1,
1891	struct xfs_attr3_icleaf_hdr *ichdr1,
1892	struct xfs_da_state_blk *blk2,
1893	struct xfs_attr3_icleaf_hdr *ichdr2,
1894	int *countarg,
1895	int *usedbytesarg)
1896	{
1897	struct xfs_attr_leafblock *leaf1 = blk1->bp->b_addr;
1898	struct xfs_attr_leafblock *leaf2 = blk2->bp->b_addr;
1899	struct xfs_attr_leaf_entry *entry;
1900	int count;
1901	int max;
1902	int index;
1903	int totallen = 0;
1904	int half;
1905	int lastdelta;
1906	int foundit = 0;
1907	int tmp;
1908
1909	/*
1910	* Examine entries until we reduce the absolute difference in
1911	* byte usage between the two blocks to a minimum.
1912	*/
1913	max = ichdr1->count + ichdr2->count;
1914	half = (max + 1) * sizeof(*entry);
1915	half += ichdr1->usedbytes + ichdr2->usedbytes +
1916	xfs_attr_leaf_newentsize(state->args, NULL);
1917	half /= 2;
1918	lastdelta = state->args->geo->blksize;
1919	entry = xfs_attr3_leaf_entryp(leafp: leaf1);
1920	for (count = index = 0; count < max; entry++, index++, count++) {
1921
1922	#define XFS_ATTR_ABS(A) (((A) < 0) ? -(A) : (A))
1923	/*
1924	* The new entry is in the first block, account for it.
1925	*/
1926	if (count == blk1->index) {
1927	tmp = totallen + sizeof(*entry) +
1928	xfs_attr_leaf_newentsize(state->args, NULL);
1929	if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1930	break;
1931	lastdelta = XFS_ATTR_ABS(half - tmp);
1932	totallen = tmp;
1933	foundit = 1;
1934	}
1935
1936	/*
1937	* Wrap around into the second block if necessary.
1938	*/
1939	if (count == ichdr1->count) {
1940	leaf1 = leaf2;
1941	entry = xfs_attr3_leaf_entryp(leafp: leaf1);
1942	index = 0;
1943	}
1944
1945	/*
1946	* Figure out if next leaf entry would be too much.
1947	*/
1948	tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1949	index);
1950	if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1951	break;
1952	lastdelta = XFS_ATTR_ABS(half - tmp);
1953	totallen = tmp;
1954	#undef XFS_ATTR_ABS
1955	}
1956
1957	/*
1958	* Calculate the number of usedbytes that will end up in lower block.
1959	* If new entry not in lower block, fix up the count.
1960	*/
1961	totallen -= count * sizeof(*entry);
1962	if (foundit) {
1963	totallen -= sizeof(*entry) +
1964	xfs_attr_leaf_newentsize(state->args, NULL);
1965	}
1966
1967	*countarg = count;
1968	*usedbytesarg = totallen;
1969	return foundit;
1970	}
1971
1972	/*========================================================================
1973	* Routines used for shrinking the Btree.
1974	*========================================================================*/
1975
1976	/*
1977	* Check a leaf block and its neighbors to see if the block should be
1978	* collapsed into one or the other neighbor. Always keep the block
1979	* with the smaller block number.
1980	* If the current block is over 50% full, don't try to join it, return 0.
1981	* If the block is empty, fill in the state structure and return 2.
1982	* If it can be collapsed, fill in the state structure and return 1.
1983	* If nothing can be done, return 0.
1984	*
1985	* GROT: allow for INCOMPLETE entries in calculation.
1986	*/
1987	int
1988	xfs_attr3_leaf_toosmall(
1989	struct xfs_da_state *state,
1990	int *action)
1991	{
1992	struct xfs_attr_leafblock *leaf;
1993	struct xfs_da_state_blk *blk;
1994	struct xfs_attr3_icleaf_hdr ichdr;
1995	struct xfs_buf *bp;
1996	xfs_dablk_t blkno;
1997	int bytes;
1998	int forward;
1999	int error;
2000	int retval;
2001	int i;
2002
2003	trace_xfs_attr_leaf_toosmall(state->args);
2004
2005	/*
2006	* Check for the degenerate case of the block being over 50% full.
2007	* If so, it's not worth even looking to see if we might be able
2008	* to coalesce with a sibling.
2009	*/
2010	blk = &state->path.blk[ state->path.active-1 ];
2011	leaf = blk->bp->b_addr;
2012	xfs_attr3_leaf_hdr_from_disk(geo: state->args->geo, to: &ichdr, from: leaf);
2013	bytes = xfs_attr3_leaf_hdr_size(leafp: leaf) +
2014	ichdr.count * sizeof(xfs_attr_leaf_entry_t) +
2015	ichdr.usedbytes;
2016	if (bytes > (state->args->geo->blksize >> 1)) {
2017	*action = 0; /* blk over 50%, don't try to join */
2018	return 0;
2019	}
2020
2021	/*
2022	* Check for the degenerate case of the block being empty.
2023	* If the block is empty, we'll simply delete it, no need to
2024	* coalesce it with a sibling block. We choose (arbitrarily)
2025	* to merge with the forward block unless it is NULL.
2026	*/
2027	if (ichdr.count == 0) {
2028	/*
2029	* Make altpath point to the block we want to keep and
2030	* path point to the block we want to drop (this one).
2031	*/
2032	forward = (ichdr.forw != 0);
2033	memcpy(&state->altpath, &state->path, sizeof(state->path));
2034	error = xfs_da3_path_shift(state, path: &state->altpath, forward,
2035	release: 0, result: &retval);
2036	if (error)
2037	return error;
2038	if (retval) {
2039	*action = 0;
2040	} else {
2041	*action = 2;
2042	}
2043	return 0;
2044	}
2045
2046	/*
2047	* Examine each sibling block to see if we can coalesce with
2048	* at least 25% free space to spare. We need to figure out
2049	* whether to merge with the forward or the backward block.
2050	* We prefer coalescing with the lower numbered sibling so as
2051	* to shrink an attribute list over time.
2052	*/
2053	/* start with smaller blk num */
2054	forward = ichdr.forw < ichdr.back;
2055	for (i = 0; i < 2; forward = !forward, i++) {
2056	struct xfs_attr3_icleaf_hdr ichdr2;
2057	if (forward)
2058	blkno = ichdr.forw;
2059	else
2060	blkno = ichdr.back;
2061	if (blkno == 0)
2062	continue;
2063	error = xfs_attr3_leaf_read(state->args->trans, state->args->dp,
2064	state->args->owner, blkno, &bp);
2065	if (error)
2066	return error;
2067
2068	xfs_attr3_leaf_hdr_from_disk(geo: state->args->geo, to: &ichdr2, from: bp->b_addr);
2069
2070	bytes = state->args->geo->blksize -
2071	(state->args->geo->blksize >> 2) -
2072	ichdr.usedbytes - ichdr2.usedbytes -
2073	((ichdr.count + ichdr2.count) *
2074	sizeof(xfs_attr_leaf_entry_t)) -
2075	xfs_attr3_leaf_hdr_size(leafp: leaf);
2076
2077	xfs_trans_brelse(state->args->trans, bp);
2078	if (bytes >= 0)
2079	break; /* fits with at least 25% to spare */
2080	}
2081	if (i >= 2) {
2082	*action = 0;
2083	return 0;
2084	}
2085
2086	/*
2087	* Make altpath point to the block we want to keep (the lower
2088	* numbered block) and path point to the block we want to drop.
2089	*/
2090	memcpy(&state->altpath, &state->path, sizeof(state->path));
2091	if (blkno < blk->blkno) {
2092	error = xfs_da3_path_shift(state, path: &state->altpath, forward,
2093	release: 0, result: &retval);
2094	} else {
2095	error = xfs_da3_path_shift(state, path: &state->path, forward,
2096	release: 0, result: &retval);
2097	}
2098	if (error)
2099	return error;
2100	if (retval) {
2101	*action = 0;
2102	} else {
2103	*action = 1;
2104	}
2105	return 0;
2106	}
2107
2108	/*
2109	* Remove a name from the leaf attribute list structure.
2110	*
2111	* Return 1 if leaf is less than 37% full, 0 if >= 37% full.
2112	* If two leaves are 37% full, when combined they will leave 25% free.
2113	*/
2114	int
2115	xfs_attr3_leaf_remove(
2116	struct xfs_buf *bp,
2117	struct xfs_da_args *args)
2118	{
2119	struct xfs_attr_leafblock *leaf;
2120	struct xfs_attr3_icleaf_hdr ichdr;
2121	struct xfs_attr_leaf_entry *entry;
2122	int before;
2123	int after;
2124	int smallest;
2125	int entsize;
2126	int tablesize;
2127	int tmp;
2128	int i;
2129
2130	trace_xfs_attr_leaf_remove(args);
2131
2132	leaf = bp->b_addr;
2133	xfs_attr3_leaf_hdr_from_disk(geo: args->geo, to: &ichdr, from: leaf);
2134
2135	ASSERT(ichdr.count > 0 && ichdr.count < args->geo->blksize / 8);
2136	ASSERT(args->index >= 0 && args->index < ichdr.count);
2137	ASSERT(ichdr.firstused >= ichdr.count * sizeof(*entry) +
2138	xfs_attr3_leaf_hdr_size(leafp: leaf));
2139
2140	entry = &xfs_attr3_leaf_entryp(leafp: leaf)[args->index];
2141
2142	ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2143	ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
2144
2145	/*
2146	* Scan through free region table:
2147	* check for adjacency of free'd entry with an existing one,
2148	* find smallest free region in case we need to replace it,
2149	* adjust any map that borders the entry table,
2150	*/
2151	tablesize = ichdr.count * sizeof(xfs_attr_leaf_entry_t)
2152	+ xfs_attr3_leaf_hdr_size(leafp: leaf);
2153	tmp = ichdr.freemap[0].size;
2154	before = after = -1;
2155	smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
2156	entsize = xfs_attr_leaf_entsize(leaf, args->index);
2157	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
2158	ASSERT(ichdr.freemap[i].base < args->geo->blksize);
2159	ASSERT(ichdr.freemap[i].size < args->geo->blksize);
2160	if (ichdr.freemap[i].base == tablesize) {
2161	ichdr.freemap[i].base -= sizeof(xfs_attr_leaf_entry_t);
2162	ichdr.freemap[i].size += sizeof(xfs_attr_leaf_entry_t);
2163	}
2164
2165	if (ichdr.freemap[i].base + ichdr.freemap[i].size ==
2166	be16_to_cpu(entry->nameidx)) {
2167	before = i;
2168	} else if (ichdr.freemap[i].base ==
2169	(be16_to_cpu(entry->nameidx) + entsize)) {
2170	after = i;
2171	} else if (ichdr.freemap[i].size < tmp) {
2172	tmp = ichdr.freemap[i].size;
2173	smallest = i;
2174	}
2175	}
2176
2177	/*
2178	* Coalesce adjacent freemap regions,
2179	* or replace the smallest region.
2180	*/
2181	if ((before >= 0) || (after >= 0)) {
2182	if ((before >= 0) && (after >= 0)) {
2183	ichdr.freemap[before].size += entsize;
2184	ichdr.freemap[before].size += ichdr.freemap[after].size;
2185	ichdr.freemap[after].base = 0;
2186	ichdr.freemap[after].size = 0;
2187	} else if (before >= 0) {
2188	ichdr.freemap[before].size += entsize;
2189	} else {
2190	ichdr.freemap[after].base = be16_to_cpu(entry->nameidx);
2191	ichdr.freemap[after].size += entsize;
2192	}
2193	} else {
2194	/*
2195	* Replace smallest region (if it is smaller than free'd entry)
2196	*/
2197	if (ichdr.freemap[smallest].size < entsize) {
2198	ichdr.freemap[smallest].base = be16_to_cpu(entry->nameidx);
2199	ichdr.freemap[smallest].size = entsize;
2200	}
2201	}
2202
2203	/*
2204	* Did we remove the first entry?
2205	*/
2206	if (be16_to_cpu(entry->nameidx) == ichdr.firstused)
2207	smallest = 1;
2208	else
2209	smallest = 0;
2210
2211	/*
2212	* Compress the remaining entries and zero out the removed stuff.
2213	*/
2214	memset(xfs_attr3_leaf_name(leafp: leaf, idx: args->index), 0, entsize);
2215	ichdr.usedbytes -= entsize;
2216	xfs_trans_log_buf(args->trans, bp,
2217	XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
2218	entsize));
2219
2220	tmp = (ichdr.count - args->index) * sizeof(xfs_attr_leaf_entry_t);
2221	memmove(entry, entry + 1, tmp);
2222	ichdr.count--;
2223	xfs_trans_log_buf(args->trans, bp,
2224	XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(xfs_attr_leaf_entry_t)));
2225
2226	entry = &xfs_attr3_leaf_entryp(leafp: leaf)[ichdr.count];
2227	memset(entry, 0, sizeof(xfs_attr_leaf_entry_t));
2228
2229	/*
2230	* If we removed the first entry, re-find the first used byte
2231	* in the name area. Note that if the entry was the "firstused",
2232	* then we don't have a "hole" in our block resulting from
2233	* removing the name.
2234	*/
2235	if (smallest) {
2236	tmp = args->geo->blksize;
2237	entry = xfs_attr3_leaf_entryp(leafp: leaf);
2238	for (i = ichdr.count - 1; i >= 0; entry++, i--) {
2239	ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2240	ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
2241
2242	if (be16_to_cpu(entry->nameidx) < tmp)
2243	tmp = be16_to_cpu(entry->nameidx);
2244	}
2245	ichdr.firstused = tmp;
2246	ASSERT(ichdr.firstused != 0);
2247	} else {
2248	ichdr.holes = 1; /* mark as needing compaction */
2249	}
2250	xfs_attr3_leaf_hdr_to_disk(geo: args->geo, to: leaf, from: &ichdr);
2251	xfs_trans_log_buf(args->trans, bp,
2252	XFS_DA_LOGRANGE(leaf, &leaf->hdr,
2253	xfs_attr3_leaf_hdr_size(leaf)));
2254
2255	/*
2256	* Check if leaf is less than 50% full, caller may want to
2257	* "join" the leaf with a sibling if so.
2258	*/
2259	tmp = ichdr.usedbytes + xfs_attr3_leaf_hdr_size(leafp: leaf) +
2260	ichdr.count * sizeof(xfs_attr_leaf_entry_t);
2261
2262	return tmp < args->geo->magicpct; /* leaf is < 37% full */
2263	}
2264
2265	/*
2266	* Move all the attribute list entries from drop_leaf into save_leaf.
2267	*/
2268	void
2269	xfs_attr3_leaf_unbalance(
2270	struct xfs_da_state *state,
2271	struct xfs_da_state_blk *drop_blk,
2272	struct xfs_da_state_blk *save_blk)
2273	{
2274	struct xfs_attr_leafblock *drop_leaf = drop_blk->bp->b_addr;
2275	struct xfs_attr_leafblock *save_leaf = save_blk->bp->b_addr;
2276	struct xfs_attr3_icleaf_hdr drophdr;
2277	struct xfs_attr3_icleaf_hdr savehdr;
2278	struct xfs_attr_leaf_entry *entry;
2279
2280	trace_xfs_attr_leaf_unbalance(state->args);
2281
2282	xfs_attr3_leaf_hdr_from_disk(geo: state->args->geo, to: &drophdr, from: drop_leaf);
2283	xfs_attr3_leaf_hdr_from_disk(geo: state->args->geo, to: &savehdr, from: save_leaf);
2284	entry = xfs_attr3_leaf_entryp(leafp: drop_leaf);
2285
2286	/*
2287	* Save last hashval from dying block for later Btree fixup.
2288	*/
2289	drop_blk->hashval = be32_to_cpu(entry[drophdr.count - 1].hashval);
2290
2291	/*
2292	* Check if we need a temp buffer, or can we do it in place.
2293	* Note that we don't check "leaf" for holes because we will
2294	* always be dropping it, toosmall() decided that for us already.
2295	*/
2296	if (savehdr.holes == 0) {
2297	/*
2298	* dest leaf has no holes, so we add there. May need
2299	* to make some room in the entry array.
2300	*/
2301	if (xfs_attr3_leaf_order(leaf1_bp: save_blk->bp, leaf1hdr: &savehdr,
2302	leaf2_bp: drop_blk->bp, leaf2hdr: &drophdr)) {
2303	xfs_attr3_leaf_moveents(state->args,
2304	drop_leaf, &drophdr, 0,
2305	save_leaf, &savehdr, 0,
2306	drophdr.count);
2307	} else {
2308	xfs_attr3_leaf_moveents(state->args,
2309	drop_leaf, &drophdr, 0,
2310	save_leaf, &savehdr,
2311	savehdr.count, drophdr.count);
2312	}
2313	} else {
2314	/*
2315	* Destination has holes, so we make a temporary copy
2316	* of the leaf and add them both to that.
2317	*/
2318	struct xfs_attr_leafblock *tmp_leaf;
2319	struct xfs_attr3_icleaf_hdr tmphdr;
2320
2321	tmp_leaf = kvzalloc(state->args->geo->blksize,
2322	GFP_KERNEL | __GFP_NOFAIL);
2323
2324	/*
2325	* Copy the header into the temp leaf so that all the stuff
2326	* not in the incore header is present and gets copied back in
2327	* once we've moved all the entries.
2328	*/
2329	memcpy(tmp_leaf, save_leaf, xfs_attr3_leaf_hdr_size(leafp: save_leaf));
2330
2331	memset(&tmphdr, 0, sizeof(tmphdr));
2332	tmphdr.magic = savehdr.magic;
2333	tmphdr.forw = savehdr.forw;
2334	tmphdr.back = savehdr.back;
2335	tmphdr.firstused = state->args->geo->blksize;
2336
2337	/* write the header to the temp buffer to initialise it */
2338	xfs_attr3_leaf_hdr_to_disk(geo: state->args->geo, to: tmp_leaf, from: &tmphdr);
2339
2340	if (xfs_attr3_leaf_order(leaf1_bp: save_blk->bp, leaf1hdr: &savehdr,
2341	leaf2_bp: drop_blk->bp, leaf2hdr: &drophdr)) {
2342	xfs_attr3_leaf_moveents(state->args,
2343	drop_leaf, &drophdr, 0,
2344	tmp_leaf, &tmphdr, 0,
2345	drophdr.count);
2346	xfs_attr3_leaf_moveents(state->args,
2347	save_leaf, &savehdr, 0,
2348	tmp_leaf, &tmphdr, tmphdr.count,
2349	savehdr.count);
2350	} else {
2351	xfs_attr3_leaf_moveents(state->args,
2352	save_leaf, &savehdr, 0,
2353	tmp_leaf, &tmphdr, 0,
2354	savehdr.count);
2355	xfs_attr3_leaf_moveents(state->args,
2356	drop_leaf, &drophdr, 0,
2357	tmp_leaf, &tmphdr, tmphdr.count,
2358	drophdr.count);
2359	}
2360	memcpy(save_leaf, tmp_leaf, state->args->geo->blksize);
2361	savehdr = tmphdr; /* struct copy */
2362	kvfree(tmp_leaf);
2363	}
2364
2365	xfs_attr3_leaf_hdr_to_disk(geo: state->args->geo, to: save_leaf, from: &savehdr);
2366	xfs_trans_log_buf(state->args->trans, save_blk->bp, 0,
2367	state->args->geo->blksize - 1);
2368
2369	/*
2370	* Copy out last hashval in each block for B-tree code.
2371	*/
2372	entry = xfs_attr3_leaf_entryp(leafp: save_leaf);
2373	save_blk->hashval = be32_to_cpu(entry[savehdr.count - 1].hashval);
2374	}
2375
2376	/*========================================================================
2377	* Routines used for finding things in the Btree.
2378	*========================================================================*/
2379
2380	/*
2381	* Look up a name in a leaf attribute list structure.
2382	* This is the internal routine, it uses the caller's buffer.
2383	*
2384	* Note that duplicate keys are allowed, but only check within the
2385	* current leaf node. The Btree code must check in adjacent leaf nodes.
2386	*
2387	* Return in args->index the index into the entry[] array of either
2388	* the found entry, or where the entry should have been (insert before
2389	* that entry).
2390	*
2391	* Don't change the args->value unless we find the attribute.
2392	*/
2393	int
2394	xfs_attr3_leaf_lookup_int(
2395	struct xfs_buf *bp,
2396	struct xfs_da_args *args)
2397	{
2398	struct xfs_attr_leafblock *leaf;
2399	struct xfs_attr3_icleaf_hdr ichdr;
2400	struct xfs_attr_leaf_entry *entry;
2401	struct xfs_attr_leaf_entry *entries;
2402	struct xfs_attr_leaf_name_local *name_loc;
2403	struct xfs_attr_leaf_name_remote *name_rmt;
2404	xfs_dahash_t hashval;
2405	int probe;
2406	int span;
2407
2408	trace_xfs_attr_leaf_lookup(args);
2409
2410	leaf = bp->b_addr;
2411	xfs_attr3_leaf_hdr_from_disk(geo: args->geo, to: &ichdr, from: leaf);
2412	entries = xfs_attr3_leaf_entryp(leafp: leaf);
2413	if (ichdr.count >= args->geo->blksize / 8) {
2414	xfs_buf_mark_corrupt(bp);
2415	xfs_da_mark_sick(args);
2416	return -EFSCORRUPTED;
2417	}
2418
2419	/*
2420	* Binary search. (note: small blocks will skip this loop)
2421	*/
2422	hashval = args->hashval;
2423	probe = span = ichdr.count / 2;
2424	for (entry = &entries[probe]; span > 4; entry = &entries[probe]) {
2425	span /= 2;
2426	if (be32_to_cpu(entry->hashval) < hashval)
2427	probe += span;
2428	else if (be32_to_cpu(entry->hashval) > hashval)
2429	probe -= span;
2430	else
2431	break;
2432	}
2433	if (!(probe >= 0 && (!ichdr.count || probe < ichdr.count))) {
2434	xfs_buf_mark_corrupt(bp);
2435	xfs_da_mark_sick(args);
2436	return -EFSCORRUPTED;
2437	}
2438	if (!(span <= 4 || be32_to_cpu(entry->hashval) == hashval)) {
2439	xfs_buf_mark_corrupt(bp);
2440	xfs_da_mark_sick(args);
2441	return -EFSCORRUPTED;
2442	}
2443
2444	/*
2445	* Since we may have duplicate hashval's, find the first matching
2446	* hashval in the leaf.
2447	*/
2448	while (probe > 0 && be32_to_cpu(entry->hashval) >= hashval) {
2449	entry--;
2450	probe--;
2451	}
2452	while (probe < ichdr.count &&
2453	be32_to_cpu(entry->hashval) < hashval) {
2454	entry++;
2455	probe++;
2456	}
2457	if (probe == ichdr.count || be32_to_cpu(entry->hashval) != hashval) {
2458	args->index = probe;
2459	return -ENOATTR;
2460	}
2461
2462	/*
2463	* Duplicate keys may be present, so search all of them for a match.
2464	*/
2465	for (; probe < ichdr.count && (be32_to_cpu(entry->hashval) == hashval);
2466	entry++, probe++) {
2467	/*
2468	* GROT: Add code to remove incomplete entries.
2469	*/
2470	if (entry->flags & XFS_ATTR_LOCAL) {
2471	name_loc = xfs_attr3_leaf_name_local(leaf, probe);
2472	if (!xfs_attr_match(args, entry->flags,
2473	name_loc->nameval, name_loc->namelen,
2474	&name_loc->nameval[name_loc->namelen],
2475	be16_to_cpu(name_loc->valuelen)))
2476	continue;
2477	args->index = probe;
2478	return -EEXIST;
2479	} else {
2480	unsigned int valuelen;
2481
2482	name_rmt = xfs_attr3_leaf_name_remote(leaf, probe);
2483	valuelen = be32_to_cpu(name_rmt->valuelen);
2484	if (!xfs_attr_match(args, entry->flags, name_rmt->name,
2485	name_rmt->namelen, NULL, valuelen))
2486	continue;
2487	args->index = probe;
2488	args->rmtvaluelen = valuelen;
2489	args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2490	args->rmtblkcnt = xfs_attr3_rmt_blocks(
2491	args->dp->i_mount,
2492	args->rmtvaluelen);
2493	return -EEXIST;
2494	}
2495	}
2496	args->index = probe;
2497	return -ENOATTR;
2498	}
2499
2500	/*
2501	* Get the value associated with an attribute name from a leaf attribute
2502	* list structure.
2503	*
2504	* If args->valuelen is zero, only the length needs to be returned. Unlike a
2505	* lookup, we only return an error if the attribute does not exist or we can't
2506	* retrieve the value.
2507	*/
2508	int
2509	xfs_attr3_leaf_getvalue(
2510	struct xfs_buf *bp,
2511	struct xfs_da_args *args)
2512	{
2513	struct xfs_attr_leafblock *leaf;
2514	struct xfs_attr3_icleaf_hdr ichdr;
2515	struct xfs_attr_leaf_entry *entry;
2516	struct xfs_attr_leaf_name_local *name_loc;
2517	struct xfs_attr_leaf_name_remote *name_rmt;
2518
2519	leaf = bp->b_addr;
2520	xfs_attr3_leaf_hdr_from_disk(geo: args->geo, to: &ichdr, from: leaf);
2521	ASSERT(ichdr.count < args->geo->blksize / 8);
2522	ASSERT(args->index < ichdr.count);
2523
2524	entry = &xfs_attr3_leaf_entryp(leafp: leaf)[args->index];
2525	if (entry->flags & XFS_ATTR_LOCAL) {
2526	name_loc = xfs_attr3_leaf_name_local(leafp: leaf, idx: args->index);
2527	ASSERT(name_loc->namelen == args->namelen);
2528	ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2529	return xfs_attr_copy_value(args,
2530	value: &name_loc->nameval[args->namelen],
2531	valuelen: be16_to_cpu(name_loc->valuelen));
2532	}
2533
2534	name_rmt = xfs_attr3_leaf_name_remote(leafp: leaf, idx: args->index);
2535	ASSERT(name_rmt->namelen == args->namelen);
2536	ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2537	args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2538	args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2539	args->rmtblkcnt = xfs_attr3_rmt_blocks(mp: args->dp->i_mount,
2540	attrlen: args->rmtvaluelen);
2541	return xfs_attr_copy_value(args, NULL, args->rmtvaluelen);
2542	}
2543
2544	/*========================================================================
2545	* Utility routines.
2546	*========================================================================*/
2547
2548	/*
2549	* Move the indicated entries from one leaf to another.
2550	* NOTE: this routine modifies both source and destination leaves.
2551	*/
2552	/*ARGSUSED*/
2553	STATIC void
2554	xfs_attr3_leaf_moveents(
2555	struct xfs_da_args *args,
2556	struct xfs_attr_leafblock *leaf_s,
2557	struct xfs_attr3_icleaf_hdr *ichdr_s,
2558	int start_s,
2559	struct xfs_attr_leafblock *leaf_d,
2560	struct xfs_attr3_icleaf_hdr *ichdr_d,
2561	int start_d,
2562	int count)
2563	{
2564	struct xfs_attr_leaf_entry *entry_s;
2565	struct xfs_attr_leaf_entry *entry_d;
2566	int desti;
2567	int tmp;
2568	int i;
2569
2570	/*
2571	* Check for nothing to do.
2572	*/
2573	if (count == 0)
2574	return;
2575
2576	/*
2577	* Set up environment.
2578	*/
2579	ASSERT(ichdr_s->magic == XFS_ATTR_LEAF_MAGIC ||
2580	ichdr_s->magic == XFS_ATTR3_LEAF_MAGIC);
2581	ASSERT(ichdr_s->magic == ichdr_d->magic);
2582	ASSERT(ichdr_s->count > 0 && ichdr_s->count < args->geo->blksize / 8);
2583	ASSERT(ichdr_s->firstused >= (ichdr_s->count * sizeof(*entry_s))
2584	+ xfs_attr3_leaf_hdr_size(leafp: leaf_s));
2585	ASSERT(ichdr_d->count < args->geo->blksize / 8);
2586	ASSERT(ichdr_d->firstused >= (ichdr_d->count * sizeof(*entry_d))
2587	+ xfs_attr3_leaf_hdr_size(leafp: leaf_d));
2588
2589	ASSERT(start_s < ichdr_s->count);
2590	ASSERT(start_d <= ichdr_d->count);
2591	ASSERT(count <= ichdr_s->count);
2592
2593
2594	/*
2595	* Move the entries in the destination leaf up to make a hole?
2596	*/
2597	if (start_d < ichdr_d->count) {
2598	tmp = ichdr_d->count - start_d;
2599	tmp *= sizeof(xfs_attr_leaf_entry_t);
2600	entry_s = &xfs_attr3_leaf_entryp(leafp: leaf_d)[start_d];
2601	entry_d = &xfs_attr3_leaf_entryp(leafp: leaf_d)[start_d + count];
2602	memmove(entry_d, entry_s, tmp);
2603	}
2604
2605	/*
2606	* Copy all entry's in the same (sorted) order,
2607	* but allocate attribute info packed and in sequence.
2608	*/
2609	entry_s = &xfs_attr3_leaf_entryp(leafp: leaf_s)[start_s];
2610	entry_d = &xfs_attr3_leaf_entryp(leafp: leaf_d)[start_d];
2611	desti = start_d;
2612	for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2613	ASSERT(be16_to_cpu(entry_s->nameidx) >= ichdr_s->firstused);
2614	tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2615	#ifdef GROT
2616	/*
2617	* Code to drop INCOMPLETE entries. Difficult to use as we
2618	* may also need to change the insertion index. Code turned
2619	* off for 6.2, should be revisited later.
2620	*/
2621	if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2622	memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2623	ichdr_s->usedbytes -= tmp;
2624	ichdr_s->count -= 1;
2625	entry_d--; /* to compensate for ++ in loop hdr */
2626	desti--;
2627	if ((start_s + i) < offset)
2628	result++; /* insertion index adjustment */
2629	} else {
2630	#endif /* GROT */
2631	ichdr_d->firstused -= tmp;
2632	/* both on-disk, don't endian flip twice */
2633	entry_d->hashval = entry_s->hashval;
2634	entry_d->nameidx = cpu_to_be16(ichdr_d->firstused);
2635	entry_d->flags = entry_s->flags;
2636	ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2637	<= args->geo->blksize);
2638	memmove(xfs_attr3_leaf_name(leafp: leaf_d, idx: desti),
2639	xfs_attr3_leaf_name(leafp: leaf_s, idx: start_s + i), tmp);
2640	ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2641	<= args->geo->blksize);
2642	memset(xfs_attr3_leaf_name(leafp: leaf_s, idx: start_s + i), 0, tmp);
2643	ichdr_s->usedbytes -= tmp;
2644	ichdr_d->usedbytes += tmp;
2645	ichdr_s->count -= 1;
2646	ichdr_d->count += 1;
2647	tmp = ichdr_d->count * sizeof(xfs_attr_leaf_entry_t)
2648	+ xfs_attr3_leaf_hdr_size(leafp: leaf_d);
2649	ASSERT(ichdr_d->firstused >= tmp);
2650	#ifdef GROT
2651	}
2652	#endif /* GROT */
2653	}
2654
2655	/*
2656	* Zero out the entries we just copied.
2657	*/
2658	if (start_s == ichdr_s->count) {
2659	tmp = count * sizeof(xfs_attr_leaf_entry_t);
2660	entry_s = &xfs_attr3_leaf_entryp(leafp: leaf_s)[start_s];
2661	ASSERT(((char *)entry_s + tmp) <=
2662	((char *)leaf_s + args->geo->blksize));
2663	memset(entry_s, 0, tmp);
2664	} else {
2665	/*
2666	* Move the remaining entries down to fill the hole,
2667	* then zero the entries at the top.
2668	*/
2669	tmp = (ichdr_s->count - count) * sizeof(xfs_attr_leaf_entry_t);
2670	entry_s = &xfs_attr3_leaf_entryp(leafp: leaf_s)[start_s + count];
2671	entry_d = &xfs_attr3_leaf_entryp(leafp: leaf_s)[start_s];
2672	memmove(entry_d, entry_s, tmp);
2673
2674	tmp = count * sizeof(xfs_attr_leaf_entry_t);
2675	entry_s = &xfs_attr3_leaf_entryp(leafp: leaf_s)[ichdr_s->count];
2676	ASSERT(((char *)entry_s + tmp) <=
2677	((char *)leaf_s + args->geo->blksize));
2678	memset(entry_s, 0, tmp);
2679	}
2680
2681	/*
2682	* Fill in the freemap information
2683	*/
2684	ichdr_d->freemap[0].base = xfs_attr3_leaf_hdr_size(leafp: leaf_d);
2685	ichdr_d->freemap[0].base += ichdr_d->count * sizeof(xfs_attr_leaf_entry_t);
2686	ichdr_d->freemap[0].size = ichdr_d->firstused - ichdr_d->freemap[0].base;
2687	ichdr_d->freemap[1].base = 0;
2688	ichdr_d->freemap[2].base = 0;
2689	ichdr_d->freemap[1].size = 0;
2690	ichdr_d->freemap[2].size = 0;
2691	ichdr_s->holes = 1; /* leaf may not be compact */
2692	}
2693
2694	/*
2695	* Pick up the last hashvalue from a leaf block.
2696	*/
2697	xfs_dahash_t
2698	xfs_attr_leaf_lasthash(
2699	struct xfs_buf *bp,
2700	int *count)
2701	{
2702	struct xfs_attr3_icleaf_hdr ichdr;
2703	struct xfs_attr_leaf_entry *entries;
2704	struct xfs_mount *mp = bp->b_mount;
2705
2706	xfs_attr3_leaf_hdr_from_disk(geo: mp->m_attr_geo, to: &ichdr, from: bp->b_addr);
2707	entries = xfs_attr3_leaf_entryp(leafp: bp->b_addr);
2708	if (count)
2709	*count = ichdr.count;
2710	if (!ichdr.count)
2711	return 0;
2712	return be32_to_cpu(entries[ichdr.count - 1].hashval);
2713	}
2714
2715	/*
2716	* Calculate the number of bytes used to store the indicated attribute
2717	* (whether local or remote only calculate bytes in this block).
2718	*/
2719	STATIC int
2720	xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2721	{
2722	struct xfs_attr_leaf_entry *entries;
2723	xfs_attr_leaf_name_local_t *name_loc;
2724	xfs_attr_leaf_name_remote_t *name_rmt;
2725	int size;
2726
2727	entries = xfs_attr3_leaf_entryp(leafp: leaf);
2728	if (entries[index].flags & XFS_ATTR_LOCAL) {
2729	name_loc = xfs_attr3_leaf_name_local(leafp: leaf, idx: index);
2730	size = xfs_attr_leaf_entsize_local(nlen: name_loc->namelen,
2731	vlen: be16_to_cpu(name_loc->valuelen));
2732	} else {
2733	name_rmt = xfs_attr3_leaf_name_remote(leafp: leaf, idx: index);
2734	size = xfs_attr_leaf_entsize_remote(nlen: name_rmt->namelen);
2735	}
2736	return size;
2737	}
2738
2739	/*
2740	* Calculate the number of bytes that would be required to store the new
2741	* attribute (whether local or remote only calculate bytes in this block).
2742	* This routine decides as a side effect whether the attribute will be
2743	* a "local" or a "remote" attribute.
2744	*/
2745	int
2746	xfs_attr_leaf_newentsize(
2747	struct xfs_da_args *args,
2748	int *local)
2749	{
2750	int size;
2751
2752	size = xfs_attr_leaf_entsize_local(nlen: args->namelen, vlen: args->valuelen);
2753	if (size < xfs_attr_leaf_entsize_local_max(bsize: args->geo->blksize)) {
2754	if (local)
2755	*local = 1;
2756	return size;
2757	}
2758	if (local)
2759	*local = 0;
2760	return xfs_attr_leaf_entsize_remote(nlen: args->namelen);
2761	}
2762
2763
2764	/*========================================================================
2765	* Manage the INCOMPLETE flag in a leaf entry
2766	*========================================================================*/
2767
2768	/*
2769	* Clear the INCOMPLETE flag on an entry in a leaf block.
2770	*/
2771	int
2772	xfs_attr3_leaf_clearflag(
2773	struct xfs_da_args *args)
2774	{
2775	struct xfs_attr_leafblock *leaf;
2776	struct xfs_attr_leaf_entry *entry;
2777	struct xfs_attr_leaf_name_remote *name_rmt;
2778	struct xfs_buf *bp;
2779	int error;
2780	#ifdef DEBUG
2781	struct xfs_attr3_icleaf_hdr ichdr;
2782	xfs_attr_leaf_name_local_t *name_loc;
2783	int namelen;
2784	char *name;
2785	#endif /* DEBUG */
2786
2787	trace_xfs_attr_leaf_clearflag(args);
2788	/*
2789	* Set up the operation.
2790	*/
2791	error = xfs_attr3_leaf_read(args->trans, args->dp, args->owner,
2792	args->blkno, &bp);
2793	if (error)
2794	return error;
2795
2796	leaf = bp->b_addr;
2797	entry = &xfs_attr3_leaf_entryp(leafp: leaf)[args->index];
2798	ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2799
2800	#ifdef DEBUG
2801	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2802	ASSERT(args->index < ichdr.count);
2803	ASSERT(args->index >= 0);
2804
2805	if (entry->flags & XFS_ATTR_LOCAL) {
2806	name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2807	namelen = name_loc->namelen;
2808	name = (char *)name_loc->nameval;
2809	} else {
2810	name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2811	namelen = name_rmt->namelen;
2812	name = (char *)name_rmt->name;
2813	}
2814	ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2815	ASSERT(namelen == args->namelen);
2816	ASSERT(memcmp(name, args->name, namelen) == 0);
2817	#endif /* DEBUG */
2818
2819	entry->flags &= ~XFS_ATTR_INCOMPLETE;
2820	xfs_trans_log_buf(args->trans, bp,
2821	XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2822
2823	if (args->rmtblkno) {
2824	ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2825	name_rmt = xfs_attr3_leaf_name_remote(leafp: leaf, idx: args->index);
2826	name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2827	name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2828	xfs_trans_log_buf(args->trans, bp,
2829	XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2830	}
2831
2832	return 0;
2833	}
2834
2835	/*
2836	* Set the INCOMPLETE flag on an entry in a leaf block.
2837	*/
2838	int
2839	xfs_attr3_leaf_setflag(
2840	struct xfs_da_args *args)
2841	{
2842	struct xfs_attr_leafblock *leaf;
2843	struct xfs_attr_leaf_entry *entry;
2844	struct xfs_attr_leaf_name_remote *name_rmt;
2845	struct xfs_buf *bp;
2846	int error;
2847	#ifdef DEBUG
2848	struct xfs_attr3_icleaf_hdr ichdr;
2849	#endif
2850
2851	trace_xfs_attr_leaf_setflag(args);
2852
2853	/*
2854	* Set up the operation.
2855	*/
2856	error = xfs_attr3_leaf_read(args->trans, args->dp, args->owner,
2857	args->blkno, &bp);
2858	if (error)
2859	return error;
2860
2861	leaf = bp->b_addr;
2862	#ifdef DEBUG
2863	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2864	ASSERT(args->index < ichdr.count);
2865	ASSERT(args->index >= 0);
2866	#endif
2867	entry = &xfs_attr3_leaf_entryp(leafp: leaf)[args->index];
2868
2869	ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2870	entry->flags |= XFS_ATTR_INCOMPLETE;
2871	xfs_trans_log_buf(args->trans, bp,
2872	XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2873	if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2874	name_rmt = xfs_attr3_leaf_name_remote(leafp: leaf, idx: args->index);
2875	name_rmt->valueblk = 0;
2876	name_rmt->valuelen = 0;
2877	xfs_trans_log_buf(args->trans, bp,
2878	XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2879	}
2880
2881	return 0;
2882	}
2883
2884	/*
2885	* In a single transaction, clear the INCOMPLETE flag on the leaf entry
2886	* given by args->blkno/index and set the INCOMPLETE flag on the leaf
2887	* entry given by args->blkno2/index2.
2888	*
2889	* Note that they could be in different blocks, or in the same block.
2890	*/
2891	int
2892	xfs_attr3_leaf_flipflags(
2893	struct xfs_da_args *args)
2894	{
2895	struct xfs_attr_leafblock *leaf1;
2896	struct xfs_attr_leafblock *leaf2;
2897	struct xfs_attr_leaf_entry *entry1;
2898	struct xfs_attr_leaf_entry *entry2;
2899	struct xfs_attr_leaf_name_remote *name_rmt;
2900	struct xfs_buf *bp1;
2901	struct xfs_buf *bp2;
2902	int error;
2903	#ifdef DEBUG
2904	struct xfs_attr3_icleaf_hdr ichdr1;
2905	struct xfs_attr3_icleaf_hdr ichdr2;
2906	xfs_attr_leaf_name_local_t *name_loc;
2907	int namelen1, namelen2;
2908	char *name1, *name2;
2909	#endif /* DEBUG */
2910
2911	trace_xfs_attr_leaf_flipflags(args);
2912
2913	/*
2914	* Read the block containing the "old" attr
2915	*/
2916	error = xfs_attr3_leaf_read(args->trans, args->dp, args->owner,
2917	args->blkno, &bp1);
2918	if (error)
2919	return error;
2920
2921	/*
2922	* Read the block containing the "new" attr, if it is different
2923	*/
2924	if (args->blkno2 != args->blkno) {
2925	error = xfs_attr3_leaf_read(args->trans, args->dp, args->owner,
2926	args->blkno2, &bp2);
2927	if (error)
2928	return error;
2929	} else {
2930	bp2 = bp1;
2931	}
2932
2933	leaf1 = bp1->b_addr;
2934	entry1 = &xfs_attr3_leaf_entryp(leafp: leaf1)[args->index];
2935
2936	leaf2 = bp2->b_addr;
2937	entry2 = &xfs_attr3_leaf_entryp(leafp: leaf2)[args->index2];
2938
2939	#ifdef DEBUG
2940	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr1, leaf1);
2941	ASSERT(args->index < ichdr1.count);
2942	ASSERT(args->index >= 0);
2943
2944	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr2, leaf2);
2945	ASSERT(args->index2 < ichdr2.count);
2946	ASSERT(args->index2 >= 0);
2947
2948	if (entry1->flags & XFS_ATTR_LOCAL) {
2949	name_loc = xfs_attr3_leaf_name_local(leaf1, args->index);
2950	namelen1 = name_loc->namelen;
2951	name1 = (char *)name_loc->nameval;
2952	} else {
2953	name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2954	namelen1 = name_rmt->namelen;
2955	name1 = (char *)name_rmt->name;
2956	}
2957	if (entry2->flags & XFS_ATTR_LOCAL) {
2958	name_loc = xfs_attr3_leaf_name_local(leaf2, args->index2);
2959	namelen2 = name_loc->namelen;
2960	name2 = (char *)name_loc->nameval;
2961	} else {
2962	name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2963	namelen2 = name_rmt->namelen;
2964	name2 = (char *)name_rmt->name;
2965	}
2966	ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2967	ASSERT(namelen1 == namelen2);
2968	ASSERT(memcmp(name1, name2, namelen1) == 0);
2969	#endif /* DEBUG */
2970
2971	ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2972	ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2973
2974	entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2975	xfs_trans_log_buf(args->trans, bp1,
2976	XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2977	if (args->rmtblkno) {
2978	ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2979	name_rmt = xfs_attr3_leaf_name_remote(leafp: leaf1, idx: args->index);
2980	name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2981	name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2982	xfs_trans_log_buf(args->trans, bp1,
2983	XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2984	}
2985
2986	entry2->flags |= XFS_ATTR_INCOMPLETE;
2987	xfs_trans_log_buf(args->trans, bp2,
2988	XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2989	if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
2990	name_rmt = xfs_attr3_leaf_name_remote(leafp: leaf2, idx: args->index2);
2991	name_rmt->valueblk = 0;
2992	name_rmt->valuelen = 0;
2993	xfs_trans_log_buf(args->trans, bp2,
2994	XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
2995	}
2996
2997	return 0;
2998	}
2999