1	// SPDX-License-Identifier: GPL-2.0
2	/*
3	* Copyright (c) 2000-2005 Silicon Graphics, Inc.
4	* Copyright (c) 2013 Red Hat, Inc.
5	* All Rights Reserved.
6	*/
7	#include "xfs.h"
8	#include "xfs_fs.h"
9	#include "xfs_shared.h"
10	#include "xfs_format.h"
11	#include "xfs_log_format.h"
12	#include "xfs_trans_resv.h"
13	#include "xfs_sb.h"
14	#include "xfs_mount.h"
15	#include "xfs_da_format.h"
16	#include "xfs_da_btree.h"
17	#include "xfs_inode.h"
18	#include "xfs_trans.h"
19	#include "xfs_bmap_btree.h"
20	#include "xfs_bmap.h"
21	#include "xfs_attr_sf.h"
22	#include "xfs_attr.h"
23	#include "xfs_attr_remote.h"
24	#include "xfs_attr_leaf.h"
25	#include "xfs_error.h"
26	#include "xfs_trace.h"
27	#include "xfs_buf_item.h"
28	#include "xfs_dir2.h"
29	#include "xfs_log.h"
30	#include "xfs_ag.h"
31	#include "xfs_errortag.h"
32	#include "xfs_health.h"
33
34
35	/*
36	* xfs_attr_leaf.c
37	*
38	* Routines to implement leaf blocks of attributes as Btrees of hashed names.
39	*/
40
41	/*========================================================================
42	* Function prototypes for the kernel.
43	*========================================================================*/
44
45	/*
46	* Routines used for growing the Btree.
47	*/
48	STATIC int xfs_attr3_leaf_create(struct xfs_da_args *args,
49	xfs_dablk_t which_block, struct xfs_buf **bpp);
50	STATIC int xfs_attr3_leaf_add_work(struct xfs_buf *leaf_buffer,
51	struct xfs_attr3_icleaf_hdr *ichdr,
52	struct xfs_da_args *args, int freemap_index);
53	STATIC void xfs_attr3_leaf_compact(struct xfs_da_args *args,
54	struct xfs_attr3_icleaf_hdr *ichdr,
55	struct xfs_buf *leaf_buffer);
56	STATIC void xfs_attr3_leaf_rebalance(xfs_da_state_t *state,
57	xfs_da_state_blk_t *blk1,
58	xfs_da_state_blk_t *blk2);
59	STATIC int xfs_attr3_leaf_figure_balance(xfs_da_state_t *state,
60	xfs_da_state_blk_t *leaf_blk_1,
61	struct xfs_attr3_icleaf_hdr *ichdr1,
62	xfs_da_state_blk_t *leaf_blk_2,
63	struct xfs_attr3_icleaf_hdr *ichdr2,
64	int *number_entries_in_blk1,
65	int *number_usedbytes_in_blk1);
66
67	/*
68	* Utility routines.
69	*/
70	STATIC void xfs_attr3_leaf_moveents(struct xfs_da_args *args,
71	struct xfs_attr_leafblock *src_leaf,
72	struct xfs_attr3_icleaf_hdr *src_ichdr, int src_start,
73	struct xfs_attr_leafblock *dst_leaf,
74	struct xfs_attr3_icleaf_hdr *dst_ichdr, int dst_start,
75	int move_count);
76	STATIC int xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index);
77
78	/*
79	* attr3 block 'firstused' conversion helpers.
80	*
81	* firstused refers to the offset of the first used byte of the nameval region
82	* of an attr leaf block. The region starts at the tail of the block and expands
83	* backwards towards the middle. As such, firstused is initialized to the block
84	* size for an empty leaf block and is reduced from there.
85	*
86	* The attr3 block size is pegged to the fsb size and the maximum fsb is 64k.
87	* The in-core firstused field is 32-bit and thus supports the maximum fsb size.
88	* The on-disk field is only 16-bit, however, and overflows at 64k. Since this
89	* only occurs at exactly 64k, we use zero as a magic on-disk value to represent
90	* the attr block size. The following helpers manage the conversion between the
91	* in-core and on-disk formats.
92	*/
93
94	static void
95	xfs_attr3_leaf_firstused_from_disk(
96	struct xfs_da_geometry *geo,
97	struct xfs_attr3_icleaf_hdr *to,
98	struct xfs_attr_leafblock *from)
99	{
100	struct xfs_attr3_leaf_hdr *hdr3;
101
102	if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) {
103	hdr3 = (struct xfs_attr3_leaf_hdr *) from;
104	to->firstused = be16_to_cpu(hdr3->firstused);
105	} else {
106	to->firstused = be16_to_cpu(from->hdr.firstused);
107	}
108
109	/*
110	* Convert from the magic fsb size value to actual blocksize. This
111	* should only occur for empty blocks when the block size overflows
112	* 16-bits.
113	*/
114	if (to->firstused == XFS_ATTR3_LEAF_NULLOFF) {
115	ASSERT(!to->count && !to->usedbytes);
116	ASSERT(geo->blksize > USHRT_MAX);
117	to->firstused = geo->blksize;
118	}
119	}
120
121	static void
122	xfs_attr3_leaf_firstused_to_disk(
123	struct xfs_da_geometry *geo,
124	struct xfs_attr_leafblock *to,
125	struct xfs_attr3_icleaf_hdr *from)
126	{
127	struct xfs_attr3_leaf_hdr *hdr3;
128	uint32_t firstused;
129
130	/* magic value should only be seen on disk */
131	ASSERT(from->firstused != XFS_ATTR3_LEAF_NULLOFF);
132
133	/*
134	* Scale down the 32-bit in-core firstused value to the 16-bit on-disk
135	* value. This only overflows at the max supported value of 64k. Use the
136	* magic on-disk value to represent block size in this case.
137	*/
138	firstused = from->firstused;
139	if (firstused > USHRT_MAX) {
140	ASSERT(from->firstused == geo->blksize);
141	firstused = XFS_ATTR3_LEAF_NULLOFF;
142	}
143
144	if (from->magic == XFS_ATTR3_LEAF_MAGIC) {
145	hdr3 = (struct xfs_attr3_leaf_hdr *) to;
146	hdr3->firstused = cpu_to_be16(firstused);
147	} else {
148	to->hdr.firstused = cpu_to_be16(firstused);
149	}
150	}
151
152	void
153	xfs_attr3_leaf_hdr_from_disk(
154	struct xfs_da_geometry *geo,
155	struct xfs_attr3_icleaf_hdr *to,
156	struct xfs_attr_leafblock *from)
157	{
158	int i;
159
160	ASSERT(from->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
161	from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
162
163	if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) {
164	struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)from;
165
166	to->forw = be32_to_cpu(hdr3->info.hdr.forw);
167	to->back = be32_to_cpu(hdr3->info.hdr.back);
168	to->magic = be16_to_cpu(hdr3->info.hdr.magic);
169	to->count = be16_to_cpu(hdr3->count);
170	to->usedbytes = be16_to_cpu(hdr3->usedbytes);
171	xfs_attr3_leaf_firstused_from_disk(geo, to, from);
172	to->holes = hdr3->holes;
173
174	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
175	to->freemap[i].base = be16_to_cpu(hdr3->freemap[i].base);
176	to->freemap[i].size = be16_to_cpu(hdr3->freemap[i].size);
177	}
178	return;
179	}
180	to->forw = be32_to_cpu(from->hdr.info.forw);
181	to->back = be32_to_cpu(from->hdr.info.back);
182	to->magic = be16_to_cpu(from->hdr.info.magic);
183	to->count = be16_to_cpu(from->hdr.count);
184	to->usedbytes = be16_to_cpu(from->hdr.usedbytes);
185	xfs_attr3_leaf_firstused_from_disk(geo, to, from);
186	to->holes = from->hdr.holes;
187
188	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
189	to->freemap[i].base = be16_to_cpu(from->hdr.freemap[i].base);
190	to->freemap[i].size = be16_to_cpu(from->hdr.freemap[i].size);
191	}
192	}
193
194	void
195	xfs_attr3_leaf_hdr_to_disk(
196	struct xfs_da_geometry *geo,
197	struct xfs_attr_leafblock *to,
198	struct xfs_attr3_icleaf_hdr *from)
199	{
200	int i;
201
202	ASSERT(from->magic == XFS_ATTR_LEAF_MAGIC ||
203	from->magic == XFS_ATTR3_LEAF_MAGIC);
204
205	if (from->magic == XFS_ATTR3_LEAF_MAGIC) {
206	struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)to;
207
208	hdr3->info.hdr.forw = cpu_to_be32(from->forw);
209	hdr3->info.hdr.back = cpu_to_be32(from->back);
210	hdr3->info.hdr.magic = cpu_to_be16(from->magic);
211	hdr3->count = cpu_to_be16(from->count);
212	hdr3->usedbytes = cpu_to_be16(from->usedbytes);
213	xfs_attr3_leaf_firstused_to_disk(geo, to, from);
214	hdr3->holes = from->holes;
215	hdr3->pad1 = 0;
216
217	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
218	hdr3->freemap[i].base = cpu_to_be16(from->freemap[i].base);
219	hdr3->freemap[i].size = cpu_to_be16(from->freemap[i].size);
220	}
221	return;
222	}
223	to->hdr.info.forw = cpu_to_be32(from->forw);
224	to->hdr.info.back = cpu_to_be32(from->back);
225	to->hdr.info.magic = cpu_to_be16(from->magic);
226	to->hdr.count = cpu_to_be16(from->count);
227	to->hdr.usedbytes = cpu_to_be16(from->usedbytes);
228	xfs_attr3_leaf_firstused_to_disk(geo, to, from);
229	to->hdr.holes = from->holes;
230	to->hdr.pad1 = 0;
231
232	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
233	to->hdr.freemap[i].base = cpu_to_be16(from->freemap[i].base);
234	to->hdr.freemap[i].size = cpu_to_be16(from->freemap[i].size);
235	}
236	}
237
238	static xfs_failaddr_t
239	xfs_attr3_leaf_verify_entry(
240	struct xfs_mount *mp,
241	char *buf_end,
242	struct xfs_attr_leafblock *leaf,
243	struct xfs_attr3_icleaf_hdr *leafhdr,
244	struct xfs_attr_leaf_entry *ent,
245	int idx,
246	__u32 *last_hashval)
247	{
248	struct xfs_attr_leaf_name_local *lentry;
249	struct xfs_attr_leaf_name_remote *rentry;
250	char *name_end;
251	unsigned int nameidx;
252	unsigned int namesize;
253	__u32 hashval;
254
255	/* hash order check */
256	hashval = be32_to_cpu(ent->hashval);
257	if (hashval < *last_hashval)
258	return __this_address;
259	*last_hashval = hashval;
260
261	nameidx = be16_to_cpu(ent->nameidx);
262	if (nameidx < leafhdr->firstused || nameidx >= mp->m_attr_geo->blksize)
263	return __this_address;
264
265	/*
266	* Check the name information. The namelen fields are u8 so we can't
267	* possibly exceed the maximum name length of 255 bytes.
268	*/
269	if (ent->flags & XFS_ATTR_LOCAL) {
270	lentry = xfs_attr3_leaf_name_local(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	static void
392	xfs_attr3_leaf_write_verify(
393	struct xfs_buf *bp)
394	{
395	struct xfs_mount *mp = bp->b_mount;
396	struct xfs_buf_log_item *bip = bp->b_log_item;
397	struct xfs_attr3_leaf_hdr *hdr3 = bp->b_addr;
398	xfs_failaddr_t fa;
399
400	fa = xfs_attr3_leaf_verify(bp);
401	if (fa) {
402	xfs_verifier_error(bp, -EFSCORRUPTED, fa);
403	return;
404	}
405
406	if (!xfs_has_crc(mp))
407	return;
408
409	if (bip)
410	hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
411
412	xfs_buf_update_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF);
413	}
414
415	/*
416	* leaf/node format detection on trees is sketchy, so a node read can be done on
417	* leaf level blocks when detection identifies the tree as a node format tree
418	* incorrectly. In this case, we need to swap the verifier to match the correct
419	* format of the block being read.
420	*/
421	static void
422	xfs_attr3_leaf_read_verify(
423	struct xfs_buf *bp)
424	{
425	struct xfs_mount *mp = bp->b_mount;
426	xfs_failaddr_t fa;
427
428	if (xfs_has_crc(mp) &&
429	!xfs_buf_verify_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF))
430	xfs_verifier_error(bp, -EFSBADCRC, __this_address);
431	else {
432	fa = xfs_attr3_leaf_verify(bp);
433	if (fa)
434	xfs_verifier_error(bp, -EFSCORRUPTED, fa);
435	}
436	}
437
438	const struct xfs_buf_ops xfs_attr3_leaf_buf_ops = {
439	.name = "xfs_attr3_leaf",
440	.magic16 = { cpu_to_be16(XFS_ATTR_LEAF_MAGIC),
441	cpu_to_be16(XFS_ATTR3_LEAF_MAGIC) },
442	.verify_read = xfs_attr3_leaf_read_verify,
443	.verify_write = xfs_attr3_leaf_write_verify,
444	.verify_struct = xfs_attr3_leaf_verify,
445	};
446
447	int
448	xfs_attr3_leaf_read(
449	struct xfs_trans *tp,
450	struct xfs_inode *dp,
451	xfs_dablk_t bno,
452	struct xfs_buf **bpp)
453	{
454	int err;
455
456	err = xfs_da_read_buf(tp, dp, bno, 0, bpp, XFS_ATTR_FORK,
457	&xfs_attr3_leaf_buf_ops);
458	if (!err && tp && *bpp)
459	xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_ATTR_LEAF_BUF);
460	return err;
461	}
462
463	/*========================================================================
464	* Namespace helper routines
465	*========================================================================*/
466
467	/*
468	* If we are in log recovery, then we want the lookup to ignore the INCOMPLETE
469	* flag on disk - if there's an incomplete attr then recovery needs to tear it
470	* down. If there's no incomplete attr, then recovery needs to tear that attr
471	* down to replace it with the attr that has been logged. In this case, the
472	* INCOMPLETE flag will not be set in attr->attr_filter, but rather
473	* XFS_DA_OP_RECOVERY will be set in args->op_flags.
474	*/
475	static bool
476	xfs_attr_match(
477	struct xfs_da_args *args,
478	uint8_t namelen,
479	unsigned char *name,
480	int flags)
481	{
482
483	if (args->namelen != namelen)
484	return false;
485	if (memcmp(args->name, name, namelen) != 0)
486	return false;
487
488	/* Recovery ignores the INCOMPLETE flag. */
489	if ((args->op_flags & XFS_DA_OP_RECOVERY) &&
490	args->attr_filter == (flags & XFS_ATTR_NSP_ONDISK_MASK))
491	return true;
492
493	/* All remaining matches need to be filtered by INCOMPLETE state. */
494	if (args->attr_filter !=
495	(flags & (XFS_ATTR_NSP_ONDISK_MASK | XFS_ATTR_INCOMPLETE)))
496	return false;
497	return true;
498	}
499
500	static int
501	xfs_attr_copy_value(
502	struct xfs_da_args *args,
503	unsigned char *value,
504	int valuelen)
505	{
506	/*
507	* No copy if all we have to do is get the length
508	*/
509	if (!args->valuelen) {
510	args->valuelen = valuelen;
511	return 0;
512	}
513
514	/*
515	* No copy if the length of the existing buffer is too small
516	*/
517	if (args->valuelen < valuelen) {
518	args->valuelen = valuelen;
519	return -ERANGE;
520	}
521
522	if (!args->value) {
523	args->value = kvmalloc(valuelen, GFP_KERNEL | __GFP_NOLOCKDEP);
524	if (!args->value)
525	return -ENOMEM;
526	}
527	args->valuelen = valuelen;
528
529	/* remote block xattr requires IO for copy-in */
530	if (args->rmtblkno)
531	return xfs_attr_rmtval_get(args);
532
533	/*
534	* This is to prevent a GCC warning because the remote xattr case
535	* doesn't have a value to pass in. In that case, we never reach here,
536	* but GCC can't work that out and so throws a "passing NULL to
537	* memcpy" warning.
538	*/
539	if (!value)
540	return -EINVAL;
541	memcpy(args->value, value, valuelen);
542	return 0;
543	}
544
545	/*========================================================================
546	* External routines when attribute fork size < XFS_LITINO(mp).
547	*========================================================================*/
548
549	/*
550	* Query whether the total requested number of attr fork bytes of extended
551	* attribute space will be able to fit inline.
552	*
553	* Returns zero if not, else the i_forkoff fork offset to be used in the
554	* literal area for attribute data once the new bytes have been added.
555	*
556	* i_forkoff must be 8 byte aligned, hence is stored as a >>3 value;
557	* special case for dev/uuid inodes, they have fixed size data forks.
558	*/
559	int
560	xfs_attr_shortform_bytesfit(
561	struct xfs_inode *dp,
562	int bytes)
563	{
564	struct xfs_mount *mp = dp->i_mount;
565	int64_t dsize;
566	int minforkoff;
567	int maxforkoff;
568	int offset;
569
570	/*
571	* Check if the new size could fit at all first:
572	*/
573	if (bytes > XFS_LITINO(mp))
574	return 0;
575
576	/* rounded down */
577	offset = (XFS_LITINO(mp) - bytes) >> 3;
578
579	if (dp->i_df.if_format == XFS_DINODE_FMT_DEV) {
580	minforkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
581	return (offset >= minforkoff) ? minforkoff : 0;
582	}
583
584	/*
585	* If the requested numbers of bytes is smaller or equal to the
586	* current attribute fork size we can always proceed.
587	*
588	* Note that if_bytes in the data fork might actually be larger than
589	* the current data fork size is due to delalloc extents. In that
590	* case either the extent count will go down when they are converted
591	* to real extents, or the delalloc conversion will take care of the
592	* literal area rebalancing.
593	*/
594	if (bytes <= xfs_inode_attr_fork_size(dp))
595	return dp->i_forkoff;
596
597	/*
598	* For attr2 we can try to move the forkoff if there is space in the
599	* literal area, but for the old format we are done if there is no
600	* space in the fixed attribute fork.
601	*/
602	if (!xfs_has_attr2(mp))
603	return 0;
604
605	dsize = dp->i_df.if_bytes;
606
607	switch (dp->i_df.if_format) {
608	case XFS_DINODE_FMT_EXTENTS:
609	/*
610	* If there is no attr fork and the data fork is extents,
611	* determine if creating the default attr fork will result
612	* in the extents form migrating to btree. If so, the
613	* minimum offset only needs to be the space required for
614	* the btree root.
615	*/
616	if (!dp->i_forkoff && dp->i_df.if_bytes >
617	xfs_default_attroffset(dp))
618	dsize = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
619	break;
620	case XFS_DINODE_FMT_BTREE:
621	/*
622	* If we have a data btree then keep forkoff if we have one,
623	* otherwise we are adding a new attr, so then we set
624	* minforkoff to where the btree root can finish so we have
625	* plenty of room for attrs
626	*/
627	if (dp->i_forkoff) {
628	if (offset < dp->i_forkoff)
629	return 0;
630	return dp->i_forkoff;
631	}
632	dsize = XFS_BMAP_BROOT_SPACE(mp, dp->i_df.if_broot);
633	break;
634	}
635
636	/*
637	* A data fork btree root must have space for at least
638	* MINDBTPTRS key/ptr pairs if the data fork is small or empty.
639	*/
640	minforkoff = max_t(int64_t, dsize, XFS_BMDR_SPACE_CALC(MINDBTPTRS));
641	minforkoff = roundup(minforkoff, 8) >> 3;
642
643	/* attr fork btree root can have at least this many key/ptr pairs */
644	maxforkoff = XFS_LITINO(mp) - XFS_BMDR_SPACE_CALC(MINABTPTRS);
645	maxforkoff = maxforkoff >> 3; /* rounded down */
646
647	if (offset >= maxforkoff)
648	return maxforkoff;
649	if (offset >= minforkoff)
650	return offset;
651	return 0;
652	}
653
654	/*
655	* Switch on the ATTR2 superblock bit (implies also FEATURES2) unless:
656	* - noattr2 mount option is set,
657	* - on-disk version bit says it is already set, or
658	* - the attr2 mount option is not set to enable automatic upgrade from attr1.
659	*/
660	STATIC void
661	xfs_sbversion_add_attr2(
662	struct xfs_mount *mp,
663	struct xfs_trans *tp)
664	{
665	if (xfs_has_noattr2(mp))
666	return;
667	if (mp->m_sb.sb_features2 & XFS_SB_VERSION2_ATTR2BIT)
668	return;
669	if (!xfs_has_attr2(mp))
670	return;
671
672	spin_lock(&mp->m_sb_lock);
673	xfs_add_attr2(mp);
674	spin_unlock(&mp->m_sb_lock);
675	xfs_log_sb(tp);
676	}
677
678	/*
679	* Create the initial contents of a shortform attribute list.
680	*/
681	void
682	xfs_attr_shortform_create(
683	struct xfs_da_args *args)
684	{
685	struct xfs_inode *dp = args->dp;
686	struct xfs_ifork *ifp = &dp->i_af;
687	struct xfs_attr_sf_hdr *hdr;
688
689	trace_xfs_attr_sf_create(args);
690
691	ASSERT(ifp->if_bytes == 0);
692	if (ifp->if_format == XFS_DINODE_FMT_EXTENTS)
693	ifp->if_format = XFS_DINODE_FMT_LOCAL;
694
695	hdr = xfs_idata_realloc(dp, sizeof(*hdr), XFS_ATTR_FORK);
696	memset(hdr, 0, sizeof(*hdr));
697	hdr->totsize = cpu_to_be16(sizeof(*hdr));
698	xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
699	}
700
701	/*
702	* Return the entry if the attr in args is found, or NULL if not.
703	*/
704	struct xfs_attr_sf_entry *
705	xfs_attr_sf_findname(
706	struct xfs_da_args *args)
707	{
708	struct xfs_attr_sf_hdr *sf = args->dp->i_af.if_data;
709	struct xfs_attr_sf_entry *sfe;
710
711	for (sfe = xfs_attr_sf_firstentry(hdr: sf);
712	sfe < xfs_attr_sf_endptr(sf);
713	sfe = xfs_attr_sf_nextentry(sfep: sfe)) {
714	if (xfs_attr_match(args, sfe->namelen, sfe->nameval,
715	sfe->flags))
716	return sfe;
717	}
718
719	return NULL;
720	}
721
722	/*
723	* Add a name/value pair to the shortform attribute list.
724	* Overflow from the inode has already been checked for.
725	*/
726	void
727	xfs_attr_shortform_add(
728	struct xfs_da_args *args,
729	int forkoff)
730	{
731	struct xfs_inode *dp = args->dp;
732	struct xfs_mount *mp = dp->i_mount;
733	struct xfs_ifork *ifp = &dp->i_af;
734	struct xfs_attr_sf_hdr *sf = ifp->if_data;
735	struct xfs_attr_sf_entry *sfe;
736	int size;
737
738	trace_xfs_attr_sf_add(args);
739
740	dp->i_forkoff = forkoff;
741
742	ASSERT(ifp->if_format == XFS_DINODE_FMT_LOCAL);
743	ASSERT(!xfs_attr_sf_findname(args));
744
745	size = xfs_attr_sf_entsize_byname(args->namelen, args->valuelen);
746	sf = xfs_idata_realloc(dp, size, XFS_ATTR_FORK);
747
748	sfe = xfs_attr_sf_endptr(sf);
749	sfe->namelen = args->namelen;
750	sfe->valuelen = args->valuelen;
751	sfe->flags = args->attr_filter;
752	memcpy(sfe->nameval, args->name, args->namelen);
753	memcpy(&sfe->nameval[args->namelen], args->value, args->valuelen);
754	sf->count++;
755	be16_add_cpu(&sf->totsize, size);
756	xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
757
758	xfs_sbversion_add_attr2(mp, args->trans);
759	}
760
761	/*
762	* After the last attribute is removed revert to original inode format,
763	* making all literal area available to the data fork once more.
764	*/
765	void
766	xfs_attr_fork_remove(
767	struct xfs_inode *ip,
768	struct xfs_trans *tp)
769	{
770	ASSERT(ip->i_af.if_nextents == 0);
771
772	xfs_ifork_zap_attr(ip);
773	ip->i_forkoff = 0;
774	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
775	}
776
777	/*
778	* Remove an attribute from the shortform attribute list structure.
779	*/
780	int
781	xfs_attr_sf_removename(
782	struct xfs_da_args *args)
783	{
784	struct xfs_inode *dp = args->dp;
785	struct xfs_mount *mp = dp->i_mount;
786	struct xfs_attr_sf_hdr *sf = dp->i_af.if_data;
787	struct xfs_attr_sf_entry *sfe;
788	uint16_t totsize = be16_to_cpu(sf->totsize);
789	void *next, *end;
790	int size = 0;
791
792	trace_xfs_attr_sf_remove(args);
793
794	sfe = xfs_attr_sf_findname(args);
795	if (!sfe) {
796	/*
797	* If we are recovering an operation, finding nothing to remove
798	* is not an error, it just means there was nothing to clean up.
799	*/
800	if (args->op_flags & XFS_DA_OP_RECOVERY)
801	return 0;
802	return -ENOATTR;
803	}
804
805	/*
806	* Fix up the attribute fork data, covering the hole
807	*/
808	size = xfs_attr_sf_entsize(sfep: sfe);
809	next = xfs_attr_sf_nextentry(sfep: sfe);
810	end = xfs_attr_sf_endptr(sf);
811	if (next < end)
812	memmove(sfe, next, end - next);
813	sf->count--;
814	totsize -= size;
815	sf->totsize = cpu_to_be16(totsize);
816
817	/*
818	* Fix up the start offset of the attribute fork
819	*/
820	if (totsize == sizeof(struct xfs_attr_sf_hdr) && xfs_has_attr2(mp) &&
821	(dp->i_df.if_format != XFS_DINODE_FMT_BTREE) &&
822	!(args->op_flags & (XFS_DA_OP_ADDNAME | XFS_DA_OP_REPLACE))) {
823	xfs_attr_fork_remove(ip: dp, tp: args->trans);
824	} else {
825	xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
826	dp->i_forkoff = xfs_attr_shortform_bytesfit(dp, totsize);
827	ASSERT(dp->i_forkoff);
828	ASSERT(totsize > sizeof(struct xfs_attr_sf_hdr) ||
829	(args->op_flags & XFS_DA_OP_ADDNAME) ||
830	!xfs_has_attr2(mp) ||
831	dp->i_df.if_format == XFS_DINODE_FMT_BTREE);
832	xfs_trans_log_inode(args->trans, dp,
833	XFS_ILOG_CORE | XFS_ILOG_ADATA);
834	}
835
836	xfs_sbversion_add_attr2(mp, args->trans);
837
838	return 0;
839	}
840
841	/*
842	* Retrieve the attribute value and length.
843	*
844	* If args->valuelen is zero, only the length needs to be returned. Unlike a
845	* lookup, we only return an error if the attribute does not exist or we can't
846	* retrieve the value.
847	*/
848	int
849	xfs_attr_shortform_getvalue(
850	struct xfs_da_args *args)
851	{
852	struct xfs_attr_sf_entry *sfe;
853
854	ASSERT(args->dp->i_af.if_format == XFS_DINODE_FMT_LOCAL);
855
856	trace_xfs_attr_sf_lookup(args);
857
858	sfe = xfs_attr_sf_findname(args);
859	if (!sfe)
860	return -ENOATTR;
861	return xfs_attr_copy_value(args, value: &sfe->nameval[args->namelen],
862	valuelen: sfe->valuelen);
863	}
864
865	/* Convert from using the shortform to the leaf format. */
866	int
867	xfs_attr_shortform_to_leaf(
868	struct xfs_da_args *args)
869	{
870	struct xfs_inode *dp = args->dp;
871	struct xfs_ifork *ifp = &dp->i_af;
872	struct xfs_attr_sf_hdr *sf = ifp->if_data;
873	struct xfs_attr_sf_entry *sfe;
874	int size = be16_to_cpu(sf->totsize);
875	struct xfs_da_args nargs;
876	char *tmpbuffer;
877	int error, i;
878	xfs_dablk_t blkno;
879	struct xfs_buf *bp;
880
881	trace_xfs_attr_sf_to_leaf(args);
882
883	tmpbuffer = kmalloc(size, GFP_KERNEL | __GFP_NOFAIL);
884	memcpy(tmpbuffer, ifp->if_data, size);
885	sf = (struct xfs_attr_sf_hdr *)tmpbuffer;
886
887	xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
888	xfs_bmap_local_to_extents_empty(args->trans, dp, XFS_ATTR_FORK);
889
890	bp = NULL;
891	error = xfs_da_grow_inode(args, &blkno);
892	if (error)
893	goto out;
894
895	ASSERT(blkno == 0);
896	error = xfs_attr3_leaf_create(args, blkno, &bp);
897	if (error)
898	goto out;
899
900	memset((char *)&nargs, 0, sizeof(nargs));
901	nargs.dp = dp;
902	nargs.geo = args->geo;
903	nargs.total = args->total;
904	nargs.whichfork = XFS_ATTR_FORK;
905	nargs.trans = args->trans;
906	nargs.op_flags = XFS_DA_OP_OKNOENT;
907
908	sfe = xfs_attr_sf_firstentry(hdr: sf);
909	for (i = 0; i < sf->count; i++) {
910	nargs.name = sfe->nameval;
911	nargs.namelen = sfe->namelen;
912	nargs.value = &sfe->nameval[nargs.namelen];
913	nargs.valuelen = sfe->valuelen;
914	nargs.hashval = xfs_da_hashname(sfe->nameval,
915	sfe->namelen);
916	nargs.attr_filter = sfe->flags & XFS_ATTR_NSP_ONDISK_MASK;
917	error = xfs_attr3_leaf_lookup_int(leaf: bp, args: &nargs); /* set a->index */
918	ASSERT(error == -ENOATTR);
919	error = xfs_attr3_leaf_add(leaf_buffer: bp, args: &nargs);
920	ASSERT(error != -ENOSPC);
921	if (error)
922	goto out;
923	sfe = xfs_attr_sf_nextentry(sfep: sfe);
924	}
925	error = 0;
926	out:
927	kfree(tmpbuffer);
928	return error;
929	}
930
931	/*
932	* Check a leaf attribute block to see if all the entries would fit into
933	* a shortform attribute list.
934	*/
935	int
936	xfs_attr_shortform_allfit(
937	struct xfs_buf *bp,
938	struct xfs_inode *dp)
939	{
940	struct xfs_attr_leafblock *leaf;
941	struct xfs_attr_leaf_entry *entry;
942	xfs_attr_leaf_name_local_t *name_loc;
943	struct xfs_attr3_icleaf_hdr leafhdr;
944	int bytes;
945	int i;
946	struct xfs_mount *mp = bp->b_mount;
947
948	leaf = bp->b_addr;
949	xfs_attr3_leaf_hdr_from_disk(geo: mp->m_attr_geo, to: &leafhdr, from: leaf);
950	entry = xfs_attr3_leaf_entryp(leafp: leaf);
951
952	bytes = sizeof(struct xfs_attr_sf_hdr);
953	for (i = 0; i < leafhdr.count; entry++, i++) {
954	if (entry->flags & XFS_ATTR_INCOMPLETE)
955	continue; /* don't copy partial entries */
956	if (!(entry->flags & XFS_ATTR_LOCAL))
957	return 0;
958	name_loc = xfs_attr3_leaf_name_local(leafp: leaf, idx: i);
959	if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX)
960	return 0;
961	if (be16_to_cpu(name_loc->valuelen) >= XFS_ATTR_SF_ENTSIZE_MAX)
962	return 0;
963	bytes += xfs_attr_sf_entsize_byname(name_loc->namelen,
964	be16_to_cpu(name_loc->valuelen));
965	}
966	if (xfs_has_attr2(dp->i_mount) &&
967	(dp->i_df.if_format != XFS_DINODE_FMT_BTREE) &&
968	(bytes == sizeof(struct xfs_attr_sf_hdr)))
969	return -1;
970	return xfs_attr_shortform_bytesfit(dp, bytes);
971	}
972
973	/* Verify the consistency of a raw inline attribute fork. */
974	xfs_failaddr_t
975	xfs_attr_shortform_verify(
976	struct xfs_attr_sf_hdr *sfp,
977	size_t size)
978	{
979	struct xfs_attr_sf_entry *sfep = xfs_attr_sf_firstentry(hdr: sfp);
980	struct xfs_attr_sf_entry *next_sfep;
981	char *endp;
982	int i;
983
984	/*
985	* Give up if the attribute is way too short.
986	*/
987	if (size < sizeof(struct xfs_attr_sf_hdr))
988	return __this_address;
989
990	endp = (char *)sfp + size;
991
992	/* Check all reported entries */
993	for (i = 0; i < sfp->count; i++) {
994	/*
995	* struct xfs_attr_sf_entry has a variable length.
996	* Check the fixed-offset parts of the structure are
997	* within the data buffer.
998	* xfs_attr_sf_entry is defined with a 1-byte variable
999	* array at the end, so we must subtract that off.
1000	*/
1001	if (((char *)sfep + sizeof(*sfep)) >= endp)
1002	return __this_address;
1003
1004	/* Don't allow names with known bad length. */
1005	if (sfep->namelen == 0)
1006	return __this_address;
1007
1008	/*
1009	* Check that the variable-length part of the structure is
1010	* within the data buffer. The next entry starts after the
1011	* name component, so nextentry is an acceptable test.
1012	*/
1013	next_sfep = xfs_attr_sf_nextentry(sfep);
1014	if ((char *)next_sfep > endp)
1015	return __this_address;
1016
1017	/*
1018	* Check for unknown flags. Short form doesn't support
1019	* the incomplete or local bits, so we can use the namespace
1020	* mask here.
1021	*/
1022	if (sfep->flags & ~XFS_ATTR_NSP_ONDISK_MASK)
1023	return __this_address;
1024
1025	/*
1026	* Check for invalid namespace combinations. We only allow
1027	* one namespace flag per xattr, so we can just count the
1028	* bits (i.e. hweight) here.
1029	*/
1030	if (hweight8(sfep->flags & XFS_ATTR_NSP_ONDISK_MASK) > 1)
1031	return __this_address;
1032
1033	sfep = next_sfep;
1034	}
1035	if ((void *)sfep != (void *)endp)
1036	return __this_address;
1037
1038	return NULL;
1039	}
1040
1041	/*
1042	* Convert a leaf attribute list to shortform attribute list
1043	*/
1044	int
1045	xfs_attr3_leaf_to_shortform(
1046	struct xfs_buf *bp,
1047	struct xfs_da_args *args,
1048	int forkoff)
1049	{
1050	struct xfs_attr_leafblock *leaf;
1051	struct xfs_attr3_icleaf_hdr ichdr;
1052	struct xfs_attr_leaf_entry *entry;
1053	struct xfs_attr_leaf_name_local *name_loc;
1054	struct xfs_da_args nargs;
1055	struct xfs_inode *dp = args->dp;
1056	char *tmpbuffer;
1057	int error;
1058	int i;
1059
1060	trace_xfs_attr_leaf_to_sf(args);
1061
1062	tmpbuffer = kmalloc(args->geo->blksize, GFP_KERNEL | __GFP_NOFAIL);
1063	if (!tmpbuffer)
1064	return -ENOMEM;
1065
1066	memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1067
1068	leaf = (xfs_attr_leafblock_t *)tmpbuffer;
1069	xfs_attr3_leaf_hdr_from_disk(geo: args->geo, to: &ichdr, from: leaf);
1070	entry = xfs_attr3_leaf_entryp(leafp: leaf);
1071
1072	/* XXX (dgc): buffer is about to be marked stale - why zero it? */
1073	memset(bp->b_addr, 0, args->geo->blksize);
1074
1075	/*
1076	* Clean out the prior contents of the attribute list.
1077	*/
1078	error = xfs_da_shrink_inode(args, 0, bp);
1079	if (error)
1080	goto out;
1081
1082	if (forkoff == -1) {
1083	/*
1084	* Don't remove the attr fork if this operation is the first
1085	* part of a attr replace operations. We're going to add a new
1086	* attr immediately, so we need to keep the attr fork around in
1087	* this case.
1088	*/
1089	if (!(args->op_flags & XFS_DA_OP_REPLACE)) {
1090	ASSERT(xfs_has_attr2(dp->i_mount));
1091	ASSERT(dp->i_df.if_format != XFS_DINODE_FMT_BTREE);
1092	xfs_attr_fork_remove(ip: dp, tp: args->trans);
1093	}
1094	goto out;
1095	}
1096
1097	xfs_attr_shortform_create(args);
1098
1099	/*
1100	* Copy the attributes
1101	*/
1102	memset((char *)&nargs, 0, sizeof(nargs));
1103	nargs.geo = args->geo;
1104	nargs.dp = dp;
1105	nargs.total = args->total;
1106	nargs.whichfork = XFS_ATTR_FORK;
1107	nargs.trans = args->trans;
1108	nargs.op_flags = XFS_DA_OP_OKNOENT;
1109
1110	for (i = 0; i < ichdr.count; entry++, i++) {
1111	if (entry->flags & XFS_ATTR_INCOMPLETE)
1112	continue; /* don't copy partial entries */
1113	if (!entry->nameidx)
1114	continue;
1115	ASSERT(entry->flags & XFS_ATTR_LOCAL);
1116	name_loc = xfs_attr3_leaf_name_local(leafp: leaf, idx: i);
1117	nargs.name = name_loc->nameval;
1118	nargs.namelen = name_loc->namelen;
1119	nargs.value = &name_loc->nameval[nargs.namelen];
1120	nargs.valuelen = be16_to_cpu(name_loc->valuelen);
1121	nargs.hashval = be32_to_cpu(entry->hashval);
1122	nargs.attr_filter = entry->flags & XFS_ATTR_NSP_ONDISK_MASK;
1123	xfs_attr_shortform_add(args: &nargs, forkoff);
1124	}
1125	error = 0;
1126
1127	out:
1128	kfree(tmpbuffer);
1129	return error;
1130	}
1131
1132	/*
1133	* Convert from using a single leaf to a root node and a leaf.
1134	*/
1135	int
1136	xfs_attr3_leaf_to_node(
1137	struct xfs_da_args *args)
1138	{
1139	struct xfs_attr_leafblock *leaf;
1140	struct xfs_attr3_icleaf_hdr icleafhdr;
1141	struct xfs_attr_leaf_entry *entries;
1142	struct xfs_da3_icnode_hdr icnodehdr;
1143	struct xfs_da_intnode *node;
1144	struct xfs_inode *dp = args->dp;
1145	struct xfs_mount *mp = dp->i_mount;
1146	struct xfs_buf *bp1 = NULL;
1147	struct xfs_buf *bp2 = NULL;
1148	xfs_dablk_t blkno;
1149	int error;
1150
1151	trace_xfs_attr_leaf_to_node(args);
1152
1153	if (XFS_TEST_ERROR(false, mp, XFS_ERRTAG_ATTR_LEAF_TO_NODE)) {
1154	error = -EIO;
1155	goto out;
1156	}
1157
1158	error = xfs_da_grow_inode(args, &blkno);
1159	if (error)
1160	goto out;
1161	error = xfs_attr3_leaf_read(args->trans, dp, 0, &bp1);
1162	if (error)
1163	goto out;
1164
1165	error = xfs_da_get_buf(args->trans, dp, blkno, &bp2, XFS_ATTR_FORK);
1166	if (error)
1167	goto out;
1168
1169	/*
1170	* Copy leaf to new buffer and log it.
1171	*/
1172	xfs_da_buf_copy(bp2, bp1, args->geo->blksize);
1173	xfs_trans_log_buf(args->trans, bp2, 0, args->geo->blksize - 1);
1174
1175	/*
1176	* Set up the new root node.
1177	*/
1178	error = xfs_da3_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK);
1179	if (error)
1180	goto out;
1181	node = bp1->b_addr;
1182	xfs_da3_node_hdr_from_disk(mp, to: &icnodehdr, from: node);
1183
1184	leaf = bp2->b_addr;
1185	xfs_attr3_leaf_hdr_from_disk(geo: args->geo, to: &icleafhdr, from: leaf);
1186	entries = xfs_attr3_leaf_entryp(leafp: leaf);
1187
1188	/* both on-disk, don't endian-flip twice */
1189	icnodehdr.btree[0].hashval = entries[icleafhdr.count - 1].hashval;
1190	icnodehdr.btree[0].before = cpu_to_be32(blkno);
1191	icnodehdr.count = 1;
1192	xfs_da3_node_hdr_to_disk(mp: dp->i_mount, to: node, from: &icnodehdr);
1193	xfs_trans_log_buf(args->trans, bp1, 0, args->geo->blksize - 1);
1194	error = 0;
1195	out:
1196	return error;
1197	}
1198
1199	/*========================================================================
1200	* Routines used for growing the Btree.
1201	*========================================================================*/
1202
1203	/*
1204	* Create the initial contents of a leaf attribute list
1205	* or a leaf in a node attribute list.
1206	*/
1207	STATIC int
1208	xfs_attr3_leaf_create(
1209	struct xfs_da_args *args,
1210	xfs_dablk_t blkno,
1211	struct xfs_buf **bpp)
1212	{
1213	struct xfs_attr_leafblock *leaf;
1214	struct xfs_attr3_icleaf_hdr ichdr;
1215	struct xfs_inode *dp = args->dp;
1216	struct xfs_mount *mp = dp->i_mount;
1217	struct xfs_buf *bp;
1218	int error;
1219
1220	trace_xfs_attr_leaf_create(args);
1221
1222	error = xfs_da_get_buf(args->trans, args->dp, blkno, &bp,
1223	XFS_ATTR_FORK);
1224	if (error)
1225	return error;
1226	bp->b_ops = &xfs_attr3_leaf_buf_ops;
1227	xfs_trans_buf_set_type(args->trans, bp, XFS_BLFT_ATTR_LEAF_BUF);
1228	leaf = bp->b_addr;
1229	memset(leaf, 0, args->geo->blksize);
1230
1231	memset(&ichdr, 0, sizeof(ichdr));
1232	ichdr.firstused = args->geo->blksize;
1233
1234	if (xfs_has_crc(mp)) {
1235	struct xfs_da3_blkinfo *hdr3 = bp->b_addr;
1236
1237	ichdr.magic = XFS_ATTR3_LEAF_MAGIC;
1238
1239	hdr3->blkno = cpu_to_be64(xfs_buf_daddr(bp));
1240	hdr3->owner = cpu_to_be64(dp->i_ino);
1241	uuid_copy(&hdr3->uuid, &mp->m_sb.sb_meta_uuid);
1242
1243	ichdr.freemap[0].base = sizeof(struct xfs_attr3_leaf_hdr);
1244	} else {
1245	ichdr.magic = XFS_ATTR_LEAF_MAGIC;
1246	ichdr.freemap[0].base = sizeof(struct xfs_attr_leaf_hdr);
1247	}
1248	ichdr.freemap[0].size = ichdr.firstused - ichdr.freemap[0].base;
1249
1250	xfs_attr3_leaf_hdr_to_disk(geo: args->geo, to: leaf, from: &ichdr);
1251	xfs_trans_log_buf(args->trans, bp, 0, args->geo->blksize - 1);
1252
1253	*bpp = bp;
1254	return 0;
1255	}
1256
1257	/*
1258	* Split the leaf node, rebalance, then add the new entry.
1259	*/
1260	int
1261	xfs_attr3_leaf_split(
1262	struct xfs_da_state *state,
1263	struct xfs_da_state_blk *oldblk,
1264	struct xfs_da_state_blk *newblk)
1265	{
1266	xfs_dablk_t blkno;
1267	int error;
1268
1269	trace_xfs_attr_leaf_split(state->args);
1270
1271	/*
1272	* Allocate space for a new leaf node.
1273	*/
1274	ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
1275	error = xfs_da_grow_inode(state->args, &blkno);
1276	if (error)
1277	return error;
1278	error = xfs_attr3_leaf_create(state->args, blkno, &newblk->bp);
1279	if (error)
1280	return error;
1281	newblk->blkno = blkno;
1282	newblk->magic = XFS_ATTR_LEAF_MAGIC;
1283
1284	/*
1285	* Rebalance the entries across the two leaves.
1286	* NOTE: rebalance() currently depends on the 2nd block being empty.
1287	*/
1288	xfs_attr3_leaf_rebalance(state, oldblk, newblk);
1289	error = xfs_da3_blk_link(state, old_blk: oldblk, new_blk: newblk);
1290	if (error)
1291	return error;
1292
1293	/*
1294	* Save info on "old" attribute for "atomic rename" ops, leaf_add()
1295	* modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
1296	* "new" attrs info. Will need the "old" info to remove it later.
1297	*
1298	* Insert the "new" entry in the correct block.
1299	*/
1300	if (state->inleaf) {
1301	trace_xfs_attr_leaf_add_old(state->args);
1302	error = xfs_attr3_leaf_add(leaf_buffer: oldblk->bp, args: state->args);
1303	} else {
1304	trace_xfs_attr_leaf_add_new(state->args);
1305	error = xfs_attr3_leaf_add(leaf_buffer: newblk->bp, args: state->args);
1306	}
1307
1308	/*
1309	* Update last hashval in each block since we added the name.
1310	*/
1311	oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
1312	newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
1313	return error;
1314	}
1315
1316	/*
1317	* Add a name to the leaf attribute list structure.
1318	*/
1319	int
1320	xfs_attr3_leaf_add(
1321	struct xfs_buf *bp,
1322	struct xfs_da_args *args)
1323	{
1324	struct xfs_attr_leafblock *leaf;
1325	struct xfs_attr3_icleaf_hdr ichdr;
1326	int tablesize;
1327	int entsize;
1328	int sum;
1329	int tmp;
1330	int i;
1331
1332	trace_xfs_attr_leaf_add(args);
1333
1334	leaf = bp->b_addr;
1335	xfs_attr3_leaf_hdr_from_disk(geo: args->geo, to: &ichdr, from: leaf);
1336	ASSERT(args->index >= 0 && args->index <= ichdr.count);
1337	entsize = xfs_attr_leaf_newentsize(args, NULL);
1338
1339	/*
1340	* Search through freemap for first-fit on new name length.
1341	* (may need to figure in size of entry struct too)
1342	*/
1343	tablesize = (ichdr.count + 1) * sizeof(xfs_attr_leaf_entry_t)
1344	+ xfs_attr3_leaf_hdr_size(leafp: leaf);
1345	for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE - 1; i >= 0; i--) {
1346	if (tablesize > ichdr.firstused) {
1347	sum += ichdr.freemap[i].size;
1348	continue;
1349	}
1350	if (!ichdr.freemap[i].size)
1351	continue; /* no space in this map */
1352	tmp = entsize;
1353	if (ichdr.freemap[i].base < ichdr.firstused)
1354	tmp += sizeof(xfs_attr_leaf_entry_t);
1355	if (ichdr.freemap[i].size >= tmp) {
1356	tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, i);
1357	goto out_log_hdr;
1358	}
1359	sum += ichdr.freemap[i].size;
1360	}
1361
1362	/*
1363	* If there are no holes in the address space of the block,
1364	* and we don't have enough freespace, then compaction will do us
1365	* no good and we should just give up.
1366	*/
1367	if (!ichdr.holes && sum < entsize)
1368	return -ENOSPC;
1369
1370	/*
1371	* Compact the entries to coalesce free space.
1372	* This may change the hdr->count via dropping INCOMPLETE entries.
1373	*/
1374	xfs_attr3_leaf_compact(args, &ichdr, bp);
1375
1376	/*
1377	* After compaction, the block is guaranteed to have only one
1378	* free region, in freemap[0]. If it is not big enough, give up.
1379	*/
1380	if (ichdr.freemap[0].size < (entsize + sizeof(xfs_attr_leaf_entry_t))) {
1381	tmp = -ENOSPC;
1382	goto out_log_hdr;
1383	}
1384
1385	tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, 0);
1386
1387	out_log_hdr:
1388	xfs_attr3_leaf_hdr_to_disk(geo: args->geo, to: leaf, from: &ichdr);
1389	xfs_trans_log_buf(args->trans, bp,
1390	XFS_DA_LOGRANGE(leaf, &leaf->hdr,
1391	xfs_attr3_leaf_hdr_size(leaf)));
1392	return tmp;
1393	}
1394
1395	/*
1396	* Add a name to a leaf attribute list structure.
1397	*/
1398	STATIC int
1399	xfs_attr3_leaf_add_work(
1400	struct xfs_buf *bp,
1401	struct xfs_attr3_icleaf_hdr *ichdr,
1402	struct xfs_da_args *args,
1403	int mapindex)
1404	{
1405	struct xfs_attr_leafblock *leaf;
1406	struct xfs_attr_leaf_entry *entry;
1407	struct xfs_attr_leaf_name_local *name_loc;
1408	struct xfs_attr_leaf_name_remote *name_rmt;
1409	struct xfs_mount *mp;
1410	int tmp;
1411	int i;
1412
1413	trace_xfs_attr_leaf_add_work(args);
1414
1415	leaf = bp->b_addr;
1416	ASSERT(mapindex >= 0 && mapindex < XFS_ATTR_LEAF_MAPSIZE);
1417	ASSERT(args->index >= 0 && args->index <= ichdr->count);
1418
1419	/*
1420	* Force open some space in the entry array and fill it in.
1421	*/
1422	entry = &xfs_attr3_leaf_entryp(leafp: leaf)[args->index];
1423	if (args->index < ichdr->count) {
1424	tmp = ichdr->count - args->index;
1425	tmp *= sizeof(xfs_attr_leaf_entry_t);
1426	memmove(entry + 1, entry, tmp);
1427	xfs_trans_log_buf(args->trans, bp,
1428	XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1429	}
1430	ichdr->count++;
1431
1432	/*
1433	* Allocate space for the new string (at the end of the run).
1434	*/
1435	mp = args->trans->t_mountp;
1436	ASSERT(ichdr->freemap[mapindex].base < args->geo->blksize);
1437	ASSERT((ichdr->freemap[mapindex].base & 0x3) == 0);
1438	ASSERT(ichdr->freemap[mapindex].size >=
1439	xfs_attr_leaf_newentsize(args, NULL));
1440	ASSERT(ichdr->freemap[mapindex].size < args->geo->blksize);
1441	ASSERT((ichdr->freemap[mapindex].size & 0x3) == 0);
1442
1443	ichdr->freemap[mapindex].size -= xfs_attr_leaf_newentsize(args, local: &tmp);
1444
1445	entry->nameidx = cpu_to_be16(ichdr->freemap[mapindex].base +
1446	ichdr->freemap[mapindex].size);
1447	entry->hashval = cpu_to_be32(args->hashval);
1448	entry->flags = args->attr_filter;
1449	if (tmp)
1450	entry->flags |= XFS_ATTR_LOCAL;
1451	if (args->op_flags & XFS_DA_OP_REPLACE) {
1452	if (!(args->op_flags & XFS_DA_OP_LOGGED))
1453	entry->flags |= XFS_ATTR_INCOMPLETE;
1454	if ((args->blkno2 == args->blkno) &&
1455	(args->index2 <= args->index)) {
1456	args->index2++;
1457	}
1458	}
1459	xfs_trans_log_buf(args->trans, bp,
1460	XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1461	ASSERT((args->index == 0) ||
1462	(be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1463	ASSERT((args->index == ichdr->count - 1) ||
1464	(be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1465
1466	/*
1467	* For "remote" attribute values, simply note that we need to
1468	* allocate space for the "remote" value. We can't actually
1469	* allocate the extents in this transaction, and we can't decide
1470	* which blocks they should be as we might allocate more blocks
1471	* as part of this transaction (a split operation for example).
1472	*/
1473	if (entry->flags & XFS_ATTR_LOCAL) {
1474	name_loc = xfs_attr3_leaf_name_local(leafp: leaf, idx: args->index);
1475	name_loc->namelen = args->namelen;
1476	name_loc->valuelen = cpu_to_be16(args->valuelen);
1477	memcpy((char *)name_loc->nameval, args->name, args->namelen);
1478	memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1479	be16_to_cpu(name_loc->valuelen));
1480	} else {
1481	name_rmt = xfs_attr3_leaf_name_remote(leafp: leaf, idx: args->index);
1482	name_rmt->namelen = args->namelen;
1483	memcpy((char *)name_rmt->name, args->name, args->namelen);
1484	entry->flags |= XFS_ATTR_INCOMPLETE;
1485	/* just in case */
1486	name_rmt->valuelen = 0;
1487	name_rmt->valueblk = 0;
1488	args->rmtblkno = 1;
1489	args->rmtblkcnt = xfs_attr3_rmt_blocks(mp, attrlen: args->valuelen);
1490	args->rmtvaluelen = args->valuelen;
1491	}
1492	xfs_trans_log_buf(args->trans, bp,
1493	XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1494	xfs_attr_leaf_entsize(leaf, args->index)));
1495
1496	/*
1497	* Update the control info for this leaf node
1498	*/
1499	if (be16_to_cpu(entry->nameidx) < ichdr->firstused)
1500	ichdr->firstused = be16_to_cpu(entry->nameidx);
1501
1502	ASSERT(ichdr->firstused >= ichdr->count * sizeof(xfs_attr_leaf_entry_t)
1503	+ xfs_attr3_leaf_hdr_size(leafp: leaf));
1504	tmp = (ichdr->count - 1) * sizeof(xfs_attr_leaf_entry_t)
1505	+ xfs_attr3_leaf_hdr_size(leafp: leaf);
1506
1507	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1508	if (ichdr->freemap[i].base == tmp) {
1509	ichdr->freemap[i].base += sizeof(xfs_attr_leaf_entry_t);
1510	ichdr->freemap[i].size -=
1511	min_t(uint16_t, ichdr->freemap[i].size,
1512	sizeof(xfs_attr_leaf_entry_t));
1513	}
1514	}
1515	ichdr->usedbytes += xfs_attr_leaf_entsize(leaf, args->index);
1516	return 0;
1517	}
1518
1519	/*
1520	* Garbage collect a leaf attribute list block by copying it to a new buffer.
1521	*/
1522	STATIC void
1523	xfs_attr3_leaf_compact(
1524	struct xfs_da_args *args,
1525	struct xfs_attr3_icleaf_hdr *ichdr_dst,
1526	struct xfs_buf *bp)
1527	{
1528	struct xfs_attr_leafblock *leaf_src;
1529	struct xfs_attr_leafblock *leaf_dst;
1530	struct xfs_attr3_icleaf_hdr ichdr_src;
1531	struct xfs_trans *trans = args->trans;
1532	char *tmpbuffer;
1533
1534	trace_xfs_attr_leaf_compact(args);
1535
1536	tmpbuffer = kmalloc(args->geo->blksize, GFP_KERNEL | __GFP_NOFAIL);
1537	memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1538	memset(bp->b_addr, 0, args->geo->blksize);
1539	leaf_src = (xfs_attr_leafblock_t *)tmpbuffer;
1540	leaf_dst = bp->b_addr;
1541
1542	/*
1543	* Copy the on-disk header back into the destination buffer to ensure
1544	* all the information in the header that is not part of the incore
1545	* header structure is preserved.
1546	*/
1547	memcpy(bp->b_addr, tmpbuffer, xfs_attr3_leaf_hdr_size(leafp: leaf_src));
1548
1549	/* Initialise the incore headers */
1550	ichdr_src = *ichdr_dst; /* struct copy */
1551	ichdr_dst->firstused = args->geo->blksize;
1552	ichdr_dst->usedbytes = 0;
1553	ichdr_dst->count = 0;
1554	ichdr_dst->holes = 0;
1555	ichdr_dst->freemap[0].base = xfs_attr3_leaf_hdr_size(leafp: leaf_src);
1556	ichdr_dst->freemap[0].size = ichdr_dst->firstused -
1557	ichdr_dst->freemap[0].base;
1558
1559	/* write the header back to initialise the underlying buffer */
1560	xfs_attr3_leaf_hdr_to_disk(geo: args->geo, to: leaf_dst, from: ichdr_dst);
1561
1562	/*
1563	* Copy all entry's in the same (sorted) order,
1564	* but allocate name/value pairs packed and in sequence.
1565	*/
1566	xfs_attr3_leaf_moveents(args, leaf_src, &ichdr_src, 0,
1567	leaf_dst, ichdr_dst, 0, ichdr_src.count);
1568	/*
1569	* this logs the entire buffer, but the caller must write the header
1570	* back to the buffer when it is finished modifying it.
1571	*/
1572	xfs_trans_log_buf(trans, bp, 0, args->geo->blksize - 1);
1573
1574	kfree(tmpbuffer);
1575	}
1576
1577	/*
1578	* Compare two leaf blocks "order".
1579	* Return 0 unless leaf2 should go before leaf1.
1580	*/
1581	static int
1582	xfs_attr3_leaf_order(
1583	struct xfs_buf *leaf1_bp,
1584	struct xfs_attr3_icleaf_hdr *leaf1hdr,
1585	struct xfs_buf *leaf2_bp,
1586	struct xfs_attr3_icleaf_hdr *leaf2hdr)
1587	{
1588	struct xfs_attr_leaf_entry *entries1;
1589	struct xfs_attr_leaf_entry *entries2;
1590
1591	entries1 = xfs_attr3_leaf_entryp(leafp: leaf1_bp->b_addr);
1592	entries2 = xfs_attr3_leaf_entryp(leafp: leaf2_bp->b_addr);
1593	if (leaf1hdr->count > 0 && leaf2hdr->count > 0 &&
1594	((be32_to_cpu(entries2[0].hashval) <
1595	be32_to_cpu(entries1[0].hashval)) ||
1596	(be32_to_cpu(entries2[leaf2hdr->count - 1].hashval) <
1597	be32_to_cpu(entries1[leaf1hdr->count - 1].hashval)))) {
1598	return 1;
1599	}
1600	return 0;
1601	}
1602
1603	int
1604	xfs_attr_leaf_order(
1605	struct xfs_buf *leaf1_bp,
1606	struct xfs_buf *leaf2_bp)
1607	{
1608	struct xfs_attr3_icleaf_hdr ichdr1;
1609	struct xfs_attr3_icleaf_hdr ichdr2;
1610	struct xfs_mount *mp = leaf1_bp->b_mount;
1611
1612	xfs_attr3_leaf_hdr_from_disk(geo: mp->m_attr_geo, to: &ichdr1, from: leaf1_bp->b_addr);
1613	xfs_attr3_leaf_hdr_from_disk(geo: mp->m_attr_geo, to: &ichdr2, from: leaf2_bp->b_addr);
1614	return xfs_attr3_leaf_order(leaf1_bp, leaf1hdr: &ichdr1, leaf2_bp, leaf2hdr: &ichdr2);
1615	}
1616
1617	/*
1618	* Redistribute the attribute list entries between two leaf nodes,
1619	* taking into account the size of the new entry.
1620	*
1621	* NOTE: if new block is empty, then it will get the upper half of the
1622	* old block. At present, all (one) callers pass in an empty second block.
1623	*
1624	* This code adjusts the args->index/blkno and args->index2/blkno2 fields
1625	* to match what it is doing in splitting the attribute leaf block. Those
1626	* values are used in "atomic rename" operations on attributes. Note that
1627	* the "new" and "old" values can end up in different blocks.
1628	*/
1629	STATIC void
1630	xfs_attr3_leaf_rebalance(
1631	struct xfs_da_state *state,
1632	struct xfs_da_state_blk *blk1,
1633	struct xfs_da_state_blk *blk2)
1634	{
1635	struct xfs_da_args *args;
1636	struct xfs_attr_leafblock *leaf1;
1637	struct xfs_attr_leafblock *leaf2;
1638	struct xfs_attr3_icleaf_hdr ichdr1;
1639	struct xfs_attr3_icleaf_hdr ichdr2;
1640	struct xfs_attr_leaf_entry *entries1;
1641	struct xfs_attr_leaf_entry *entries2;
1642	int count;
1643	int totallen;
1644	int max;
1645	int space;
1646	int swap;
1647
1648	/*
1649	* Set up environment.
1650	*/
1651	ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1652	ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1653	leaf1 = blk1->bp->b_addr;
1654	leaf2 = blk2->bp->b_addr;
1655	xfs_attr3_leaf_hdr_from_disk(geo: state->args->geo, to: &ichdr1, from: leaf1);
1656	xfs_attr3_leaf_hdr_from_disk(geo: state->args->geo, to: &ichdr2, from: leaf2);
1657	ASSERT(ichdr2.count == 0);
1658	args = state->args;
1659
1660	trace_xfs_attr_leaf_rebalance(args);
1661
1662	/*
1663	* Check ordering of blocks, reverse if it makes things simpler.
1664	*
1665	* NOTE: Given that all (current) callers pass in an empty
1666	* second block, this code should never set "swap".
1667	*/
1668	swap = 0;
1669	if (xfs_attr3_leaf_order(leaf1_bp: blk1->bp, leaf1hdr: &ichdr1, leaf2_bp: blk2->bp, leaf2hdr: &ichdr2)) {
1670	swap(blk1, blk2);
1671
1672	/* swap structures rather than reconverting them */
1673	swap(ichdr1, ichdr2);
1674
1675	leaf1 = blk1->bp->b_addr;
1676	leaf2 = blk2->bp->b_addr;
1677	swap = 1;
1678	}
1679
1680	/*
1681	* Examine entries until we reduce the absolute difference in
1682	* byte usage between the two blocks to a minimum. Then get
1683	* the direction to copy and the number of elements to move.
1684	*
1685	* "inleaf" is true if the new entry should be inserted into blk1.
1686	* If "swap" is also true, then reverse the sense of "inleaf".
1687	*/
1688	state->inleaf = xfs_attr3_leaf_figure_balance(state, blk1, &ichdr1,
1689	blk2, &ichdr2,
1690	&count, &totallen);
1691	if (swap)
1692	state->inleaf = !state->inleaf;
1693
1694	/*
1695	* Move any entries required from leaf to leaf:
1696	*/
1697	if (count < ichdr1.count) {
1698	/*
1699	* Figure the total bytes to be added to the destination leaf.
1700	*/
1701	/* number entries being moved */
1702	count = ichdr1.count - count;
1703	space = ichdr1.usedbytes - totallen;
1704	space += count * sizeof(xfs_attr_leaf_entry_t);
1705
1706	/*
1707	* leaf2 is the destination, compact it if it looks tight.
1708	*/
1709	max = ichdr2.firstused - xfs_attr3_leaf_hdr_size(leafp: leaf1);
1710	max -= ichdr2.count * sizeof(xfs_attr_leaf_entry_t);
1711	if (space > max)
1712	xfs_attr3_leaf_compact(args, &ichdr2, blk2->bp);
1713
1714	/*
1715	* Move high entries from leaf1 to low end of leaf2.
1716	*/
1717	xfs_attr3_leaf_moveents(args, leaf1, &ichdr1,
1718	ichdr1.count - count, leaf2, &ichdr2, 0, count);
1719
1720	} else if (count > ichdr1.count) {
1721	/*
1722	* I assert that since all callers pass in an empty
1723	* second buffer, this code should never execute.
1724	*/
1725	ASSERT(0);
1726
1727	/*
1728	* Figure the total bytes to be added to the destination leaf.
1729	*/
1730	/* number entries being moved */
1731	count -= ichdr1.count;
1732	space = totallen - ichdr1.usedbytes;
1733	space += count * sizeof(xfs_attr_leaf_entry_t);
1734
1735	/*
1736	* leaf1 is the destination, compact it if it looks tight.
1737	*/
1738	max = ichdr1.firstused - xfs_attr3_leaf_hdr_size(leafp: leaf1);
1739	max -= ichdr1.count * sizeof(xfs_attr_leaf_entry_t);
1740	if (space > max)
1741	xfs_attr3_leaf_compact(args, &ichdr1, blk1->bp);
1742
1743	/*
1744	* Move low entries from leaf2 to high end of leaf1.
1745	*/
1746	xfs_attr3_leaf_moveents(args, leaf2, &ichdr2, 0, leaf1, &ichdr1,
1747	ichdr1.count, count);
1748	}
1749
1750	xfs_attr3_leaf_hdr_to_disk(geo: state->args->geo, to: leaf1, from: &ichdr1);
1751	xfs_attr3_leaf_hdr_to_disk(geo: state->args->geo, to: leaf2, from: &ichdr2);
1752	xfs_trans_log_buf(args->trans, blk1->bp, 0, args->geo->blksize - 1);
1753	xfs_trans_log_buf(args->trans, blk2->bp, 0, args->geo->blksize - 1);
1754
1755	/*
1756	* Copy out last hashval in each block for B-tree code.
1757	*/
1758	entries1 = xfs_attr3_leaf_entryp(leafp: leaf1);
1759	entries2 = xfs_attr3_leaf_entryp(leafp: leaf2);
1760	blk1->hashval = be32_to_cpu(entries1[ichdr1.count - 1].hashval);
1761	blk2->hashval = be32_to_cpu(entries2[ichdr2.count - 1].hashval);
1762
1763	/*
1764	* Adjust the expected index for insertion.
1765	* NOTE: this code depends on the (current) situation that the
1766	* second block was originally empty.
1767	*
1768	* If the insertion point moved to the 2nd block, we must adjust
1769	* the index. We must also track the entry just following the
1770	* new entry for use in an "atomic rename" operation, that entry
1771	* is always the "old" entry and the "new" entry is what we are
1772	* inserting. The index/blkno fields refer to the "old" entry,
1773	* while the index2/blkno2 fields refer to the "new" entry.
1774	*/
1775	if (blk1->index > ichdr1.count) {
1776	ASSERT(state->inleaf == 0);
1777	blk2->index = blk1->index - ichdr1.count;
1778	args->index = args->index2 = blk2->index;
1779	args->blkno = args->blkno2 = blk2->blkno;
1780	} else if (blk1->index == ichdr1.count) {
1781	if (state->inleaf) {
1782	args->index = blk1->index;
1783	args->blkno = blk1->blkno;
1784	args->index2 = 0;
1785	args->blkno2 = blk2->blkno;
1786	} else {
1787	/*
1788	* On a double leaf split, the original attr location
1789	* is already stored in blkno2/index2, so don't
1790	* overwrite it overwise we corrupt the tree.
1791	*/
1792	blk2->index = blk1->index - ichdr1.count;
1793	args->index = blk2->index;
1794	args->blkno = blk2->blkno;
1795	if (!state->extravalid) {
1796	/*
1797	* set the new attr location to match the old
1798	* one and let the higher level split code
1799	* decide where in the leaf to place it.
1800	*/
1801	args->index2 = blk2->index;
1802	args->blkno2 = blk2->blkno;
1803	}
1804	}
1805	} else {
1806	ASSERT(state->inleaf == 1);
1807	args->index = args->index2 = blk1->index;
1808	args->blkno = args->blkno2 = blk1->blkno;
1809	}
1810	}
1811
1812	/*
1813	* Examine entries until we reduce the absolute difference in
1814	* byte usage between the two blocks to a minimum.
1815	* GROT: Is this really necessary? With other than a 512 byte blocksize,
1816	* GROT: there will always be enough room in either block for a new entry.
1817	* GROT: Do a double-split for this case?
1818	*/
1819	STATIC int
1820	xfs_attr3_leaf_figure_balance(
1821	struct xfs_da_state *state,
1822	struct xfs_da_state_blk *blk1,
1823	struct xfs_attr3_icleaf_hdr *ichdr1,
1824	struct xfs_da_state_blk *blk2,
1825	struct xfs_attr3_icleaf_hdr *ichdr2,
1826	int *countarg,
1827	int *usedbytesarg)
1828	{
1829	struct xfs_attr_leafblock *leaf1 = blk1->bp->b_addr;
1830	struct xfs_attr_leafblock *leaf2 = blk2->bp->b_addr;
1831	struct xfs_attr_leaf_entry *entry;
1832	int count;
1833	int max;
1834	int index;
1835	int totallen = 0;
1836	int half;
1837	int lastdelta;
1838	int foundit = 0;
1839	int tmp;
1840
1841	/*
1842	* Examine entries until we reduce the absolute difference in
1843	* byte usage between the two blocks to a minimum.
1844	*/
1845	max = ichdr1->count + ichdr2->count;
1846	half = (max + 1) * sizeof(*entry);
1847	half += ichdr1->usedbytes + ichdr2->usedbytes +
1848	xfs_attr_leaf_newentsize(state->args, NULL);
1849	half /= 2;
1850	lastdelta = state->args->geo->blksize;
1851	entry = xfs_attr3_leaf_entryp(leafp: leaf1);
1852	for (count = index = 0; count < max; entry++, index++, count++) {
1853
1854	#define XFS_ATTR_ABS(A) (((A) < 0) ? -(A) : (A))
1855	/*
1856	* The new entry is in the first block, account for it.
1857	*/
1858	if (count == blk1->index) {
1859	tmp = totallen + sizeof(*entry) +
1860	xfs_attr_leaf_newentsize(state->args, NULL);
1861	if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1862	break;
1863	lastdelta = XFS_ATTR_ABS(half - tmp);
1864	totallen = tmp;
1865	foundit = 1;
1866	}
1867
1868	/*
1869	* Wrap around into the second block if necessary.
1870	*/
1871	if (count == ichdr1->count) {
1872	leaf1 = leaf2;
1873	entry = xfs_attr3_leaf_entryp(leafp: leaf1);
1874	index = 0;
1875	}
1876
1877	/*
1878	* Figure out if next leaf entry would be too much.
1879	*/
1880	tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1881	index);
1882	if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1883	break;
1884	lastdelta = XFS_ATTR_ABS(half - tmp);
1885	totallen = tmp;
1886	#undef XFS_ATTR_ABS
1887	}
1888
1889	/*
1890	* Calculate the number of usedbytes that will end up in lower block.
1891	* If new entry not in lower block, fix up the count.
1892	*/
1893	totallen -= count * sizeof(*entry);
1894	if (foundit) {
1895	totallen -= sizeof(*entry) +
1896	xfs_attr_leaf_newentsize(state->args, NULL);
1897	}
1898
1899	*countarg = count;
1900	*usedbytesarg = totallen;
1901	return foundit;
1902	}
1903
1904	/*========================================================================
1905	* Routines used for shrinking the Btree.
1906	*========================================================================*/
1907
1908	/*
1909	* Check a leaf block and its neighbors to see if the block should be
1910	* collapsed into one or the other neighbor. Always keep the block
1911	* with the smaller block number.
1912	* If the current block is over 50% full, don't try to join it, return 0.
1913	* If the block is empty, fill in the state structure and return 2.
1914	* If it can be collapsed, fill in the state structure and return 1.
1915	* If nothing can be done, return 0.
1916	*
1917	* GROT: allow for INCOMPLETE entries in calculation.
1918	*/
1919	int
1920	xfs_attr3_leaf_toosmall(
1921	struct xfs_da_state *state,
1922	int *action)
1923	{
1924	struct xfs_attr_leafblock *leaf;
1925	struct xfs_da_state_blk *blk;
1926	struct xfs_attr3_icleaf_hdr ichdr;
1927	struct xfs_buf *bp;
1928	xfs_dablk_t blkno;
1929	int bytes;
1930	int forward;
1931	int error;
1932	int retval;
1933	int i;
1934
1935	trace_xfs_attr_leaf_toosmall(state->args);
1936
1937	/*
1938	* Check for the degenerate case of the block being over 50% full.
1939	* If so, it's not worth even looking to see if we might be able
1940	* to coalesce with a sibling.
1941	*/
1942	blk = &state->path.blk[ state->path.active-1 ];
1943	leaf = blk->bp->b_addr;
1944	xfs_attr3_leaf_hdr_from_disk(geo: state->args->geo, to: &ichdr, from: leaf);
1945	bytes = xfs_attr3_leaf_hdr_size(leafp: leaf) +
1946	ichdr.count * sizeof(xfs_attr_leaf_entry_t) +
1947	ichdr.usedbytes;
1948	if (bytes > (state->args->geo->blksize >> 1)) {
1949	*action = 0; /* blk over 50%, don't try to join */
1950	return 0;
1951	}
1952
1953	/*
1954	* Check for the degenerate case of the block being empty.
1955	* If the block is empty, we'll simply delete it, no need to
1956	* coalesce it with a sibling block. We choose (arbitrarily)
1957	* to merge with the forward block unless it is NULL.
1958	*/
1959	if (ichdr.count == 0) {
1960	/*
1961	* Make altpath point to the block we want to keep and
1962	* path point to the block we want to drop (this one).
1963	*/
1964	forward = (ichdr.forw != 0);
1965	memcpy(&state->altpath, &state->path, sizeof(state->path));
1966	error = xfs_da3_path_shift(state, path: &state->altpath, forward,
1967	release: 0, result: &retval);
1968	if (error)
1969	return error;
1970	if (retval) {
1971	*action = 0;
1972	} else {
1973	*action = 2;
1974	}
1975	return 0;
1976	}
1977
1978	/*
1979	* Examine each sibling block to see if we can coalesce with
1980	* at least 25% free space to spare. We need to figure out
1981	* whether to merge with the forward or the backward block.
1982	* We prefer coalescing with the lower numbered sibling so as
1983	* to shrink an attribute list over time.
1984	*/
1985	/* start with smaller blk num */
1986	forward = ichdr.forw < ichdr.back;
1987	for (i = 0; i < 2; forward = !forward, i++) {
1988	struct xfs_attr3_icleaf_hdr ichdr2;
1989	if (forward)
1990	blkno = ichdr.forw;
1991	else
1992	blkno = ichdr.back;
1993	if (blkno == 0)
1994	continue;
1995	error = xfs_attr3_leaf_read(state->args->trans, state->args->dp,
1996	blkno, &bp);
1997	if (error)
1998	return error;
1999
2000	xfs_attr3_leaf_hdr_from_disk(geo: state->args->geo, to: &ichdr2, from: bp->b_addr);
2001
2002	bytes = state->args->geo->blksize -
2003	(state->args->geo->blksize >> 2) -
2004	ichdr.usedbytes - ichdr2.usedbytes -
2005	((ichdr.count + ichdr2.count) *
2006	sizeof(xfs_attr_leaf_entry_t)) -
2007	xfs_attr3_leaf_hdr_size(leafp: leaf);
2008
2009	xfs_trans_brelse(state->args->trans, bp);
2010	if (bytes >= 0)
2011	break; /* fits with at least 25% to spare */
2012	}
2013	if (i >= 2) {
2014	*action = 0;
2015	return 0;
2016	}
2017
2018	/*
2019	* Make altpath point to the block we want to keep (the lower
2020	* numbered block) and path point to the block we want to drop.
2021	*/
2022	memcpy(&state->altpath, &state->path, sizeof(state->path));
2023	if (blkno < blk->blkno) {
2024	error = xfs_da3_path_shift(state, path: &state->altpath, forward,
2025	release: 0, result: &retval);
2026	} else {
2027	error = xfs_da3_path_shift(state, path: &state->path, forward,
2028	release: 0, result: &retval);
2029	}
2030	if (error)
2031	return error;
2032	if (retval) {
2033	*action = 0;
2034	} else {
2035	*action = 1;
2036	}
2037	return 0;
2038	}
2039
2040	/*
2041	* Remove a name from the leaf attribute list structure.
2042	*
2043	* Return 1 if leaf is less than 37% full, 0 if >= 37% full.
2044	* If two leaves are 37% full, when combined they will leave 25% free.
2045	*/
2046	int
2047	xfs_attr3_leaf_remove(
2048	struct xfs_buf *bp,
2049	struct xfs_da_args *args)
2050	{
2051	struct xfs_attr_leafblock *leaf;
2052	struct xfs_attr3_icleaf_hdr ichdr;
2053	struct xfs_attr_leaf_entry *entry;
2054	int before;
2055	int after;
2056	int smallest;
2057	int entsize;
2058	int tablesize;
2059	int tmp;
2060	int i;
2061
2062	trace_xfs_attr_leaf_remove(args);
2063
2064	leaf = bp->b_addr;
2065	xfs_attr3_leaf_hdr_from_disk(geo: args->geo, to: &ichdr, from: leaf);
2066
2067	ASSERT(ichdr.count > 0 && ichdr.count < args->geo->blksize / 8);
2068	ASSERT(args->index >= 0 && args->index < ichdr.count);
2069	ASSERT(ichdr.firstused >= ichdr.count * sizeof(*entry) +
2070	xfs_attr3_leaf_hdr_size(leafp: leaf));
2071
2072	entry = &xfs_attr3_leaf_entryp(leafp: leaf)[args->index];
2073
2074	ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2075	ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
2076
2077	/*
2078	* Scan through free region table:
2079	* check for adjacency of free'd entry with an existing one,
2080	* find smallest free region in case we need to replace it,
2081	* adjust any map that borders the entry table,
2082	*/
2083	tablesize = ichdr.count * sizeof(xfs_attr_leaf_entry_t)
2084	+ xfs_attr3_leaf_hdr_size(leafp: leaf);
2085	tmp = ichdr.freemap[0].size;
2086	before = after = -1;
2087	smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
2088	entsize = xfs_attr_leaf_entsize(leaf, args->index);
2089	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
2090	ASSERT(ichdr.freemap[i].base < args->geo->blksize);
2091	ASSERT(ichdr.freemap[i].size < args->geo->blksize);
2092	if (ichdr.freemap[i].base == tablesize) {
2093	ichdr.freemap[i].base -= sizeof(xfs_attr_leaf_entry_t);
2094	ichdr.freemap[i].size += sizeof(xfs_attr_leaf_entry_t);
2095	}
2096
2097	if (ichdr.freemap[i].base + ichdr.freemap[i].size ==
2098	be16_to_cpu(entry->nameidx)) {
2099	before = i;
2100	} else if (ichdr.freemap[i].base ==
2101	(be16_to_cpu(entry->nameidx) + entsize)) {
2102	after = i;
2103	} else if (ichdr.freemap[i].size < tmp) {
2104	tmp = ichdr.freemap[i].size;
2105	smallest = i;
2106	}
2107	}
2108
2109	/*
2110	* Coalesce adjacent freemap regions,
2111	* or replace the smallest region.
2112	*/
2113	if ((before >= 0) || (after >= 0)) {
2114	if ((before >= 0) && (after >= 0)) {
2115	ichdr.freemap[before].size += entsize;
2116	ichdr.freemap[before].size += ichdr.freemap[after].size;
2117	ichdr.freemap[after].base = 0;
2118	ichdr.freemap[after].size = 0;
2119	} else if (before >= 0) {
2120	ichdr.freemap[before].size += entsize;
2121	} else {
2122	ichdr.freemap[after].base = be16_to_cpu(entry->nameidx);
2123	ichdr.freemap[after].size += entsize;
2124	}
2125	} else {
2126	/*
2127	* Replace smallest region (if it is smaller than free'd entry)
2128	*/
2129	if (ichdr.freemap[smallest].size < entsize) {
2130	ichdr.freemap[smallest].base = be16_to_cpu(entry->nameidx);
2131	ichdr.freemap[smallest].size = entsize;
2132	}
2133	}
2134
2135	/*
2136	* Did we remove the first entry?
2137	*/
2138	if (be16_to_cpu(entry->nameidx) == ichdr.firstused)
2139	smallest = 1;
2140	else
2141	smallest = 0;
2142
2143	/*
2144	* Compress the remaining entries and zero out the removed stuff.
2145	*/
2146	memset(xfs_attr3_leaf_name(leafp: leaf, idx: args->index), 0, entsize);
2147	ichdr.usedbytes -= entsize;
2148	xfs_trans_log_buf(args->trans, bp,
2149	XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
2150	entsize));
2151
2152	tmp = (ichdr.count - args->index) * sizeof(xfs_attr_leaf_entry_t);
2153	memmove(entry, entry + 1, tmp);
2154	ichdr.count--;
2155	xfs_trans_log_buf(args->trans, bp,
2156	XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(xfs_attr_leaf_entry_t)));
2157
2158	entry = &xfs_attr3_leaf_entryp(leafp: leaf)[ichdr.count];
2159	memset(entry, 0, sizeof(xfs_attr_leaf_entry_t));
2160
2161	/*
2162	* If we removed the first entry, re-find the first used byte
2163	* in the name area. Note that if the entry was the "firstused",
2164	* then we don't have a "hole" in our block resulting from
2165	* removing the name.
2166	*/
2167	if (smallest) {
2168	tmp = args->geo->blksize;
2169	entry = xfs_attr3_leaf_entryp(leafp: leaf);
2170	for (i = ichdr.count - 1; i >= 0; entry++, i--) {
2171	ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2172	ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
2173
2174	if (be16_to_cpu(entry->nameidx) < tmp)
2175	tmp = be16_to_cpu(entry->nameidx);
2176	}
2177	ichdr.firstused = tmp;
2178	ASSERT(ichdr.firstused != 0);
2179	} else {
2180	ichdr.holes = 1; /* mark as needing compaction */
2181	}
2182	xfs_attr3_leaf_hdr_to_disk(geo: args->geo, to: leaf, from: &ichdr);
2183	xfs_trans_log_buf(args->trans, bp,
2184	XFS_DA_LOGRANGE(leaf, &leaf->hdr,
2185	xfs_attr3_leaf_hdr_size(leaf)));
2186
2187	/*
2188	* Check if leaf is less than 50% full, caller may want to
2189	* "join" the leaf with a sibling if so.
2190	*/
2191	tmp = ichdr.usedbytes + xfs_attr3_leaf_hdr_size(leafp: leaf) +
2192	ichdr.count * sizeof(xfs_attr_leaf_entry_t);
2193
2194	return tmp < args->geo->magicpct; /* leaf is < 37% full */
2195	}
2196
2197	/*
2198	* Move all the attribute list entries from drop_leaf into save_leaf.
2199	*/
2200	void
2201	xfs_attr3_leaf_unbalance(
2202	struct xfs_da_state *state,
2203	struct xfs_da_state_blk *drop_blk,
2204	struct xfs_da_state_blk *save_blk)
2205	{
2206	struct xfs_attr_leafblock *drop_leaf = drop_blk->bp->b_addr;
2207	struct xfs_attr_leafblock *save_leaf = save_blk->bp->b_addr;
2208	struct xfs_attr3_icleaf_hdr drophdr;
2209	struct xfs_attr3_icleaf_hdr savehdr;
2210	struct xfs_attr_leaf_entry *entry;
2211
2212	trace_xfs_attr_leaf_unbalance(state->args);
2213
2214	xfs_attr3_leaf_hdr_from_disk(geo: state->args->geo, to: &drophdr, from: drop_leaf);
2215	xfs_attr3_leaf_hdr_from_disk(geo: state->args->geo, to: &savehdr, from: save_leaf);
2216	entry = xfs_attr3_leaf_entryp(leafp: drop_leaf);
2217
2218	/*
2219	* Save last hashval from dying block for later Btree fixup.
2220	*/
2221	drop_blk->hashval = be32_to_cpu(entry[drophdr.count - 1].hashval);
2222
2223	/*
2224	* Check if we need a temp buffer, or can we do it in place.
2225	* Note that we don't check "leaf" for holes because we will
2226	* always be dropping it, toosmall() decided that for us already.
2227	*/
2228	if (savehdr.holes == 0) {
2229	/*
2230	* dest leaf has no holes, so we add there. May need
2231	* to make some room in the entry array.
2232	*/
2233	if (xfs_attr3_leaf_order(leaf1_bp: save_blk->bp, leaf1hdr: &savehdr,
2234	leaf2_bp: drop_blk->bp, leaf2hdr: &drophdr)) {
2235	xfs_attr3_leaf_moveents(state->args,
2236	drop_leaf, &drophdr, 0,
2237	save_leaf, &savehdr, 0,
2238	drophdr.count);
2239	} else {
2240	xfs_attr3_leaf_moveents(state->args,
2241	drop_leaf, &drophdr, 0,
2242	save_leaf, &savehdr,
2243	savehdr.count, drophdr.count);
2244	}
2245	} else {
2246	/*
2247	* Destination has holes, so we make a temporary copy
2248	* of the leaf and add them both to that.
2249	*/
2250	struct xfs_attr_leafblock *tmp_leaf;
2251	struct xfs_attr3_icleaf_hdr tmphdr;
2252
2253	tmp_leaf = kzalloc(state->args->geo->blksize,
2254	GFP_KERNEL | __GFP_NOFAIL);
2255
2256	/*
2257	* Copy the header into the temp leaf so that all the stuff
2258	* not in the incore header is present and gets copied back in
2259	* once we've moved all the entries.
2260	*/
2261	memcpy(tmp_leaf, save_leaf, xfs_attr3_leaf_hdr_size(leafp: save_leaf));
2262
2263	memset(&tmphdr, 0, sizeof(tmphdr));
2264	tmphdr.magic = savehdr.magic;
2265	tmphdr.forw = savehdr.forw;
2266	tmphdr.back = savehdr.back;
2267	tmphdr.firstused = state->args->geo->blksize;
2268
2269	/* write the header to the temp buffer to initialise it */
2270	xfs_attr3_leaf_hdr_to_disk(geo: state->args->geo, to: tmp_leaf, from: &tmphdr);
2271
2272	if (xfs_attr3_leaf_order(leaf1_bp: save_blk->bp, leaf1hdr: &savehdr,
2273	leaf2_bp: drop_blk->bp, leaf2hdr: &drophdr)) {
2274	xfs_attr3_leaf_moveents(state->args,
2275	drop_leaf, &drophdr, 0,
2276	tmp_leaf, &tmphdr, 0,
2277	drophdr.count);
2278	xfs_attr3_leaf_moveents(state->args,
2279	save_leaf, &savehdr, 0,
2280	tmp_leaf, &tmphdr, tmphdr.count,
2281	savehdr.count);
2282	} else {
2283	xfs_attr3_leaf_moveents(state->args,
2284	save_leaf, &savehdr, 0,
2285	tmp_leaf, &tmphdr, 0,
2286	savehdr.count);
2287	xfs_attr3_leaf_moveents(state->args,
2288	drop_leaf, &drophdr, 0,
2289	tmp_leaf, &tmphdr, tmphdr.count,
2290	drophdr.count);
2291	}
2292	memcpy(save_leaf, tmp_leaf, state->args->geo->blksize);
2293	savehdr = tmphdr; /* struct copy */
2294	kfree(tmp_leaf);
2295	}
2296
2297	xfs_attr3_leaf_hdr_to_disk(geo: state->args->geo, to: save_leaf, from: &savehdr);
2298	xfs_trans_log_buf(state->args->trans, save_blk->bp, 0,
2299	state->args->geo->blksize - 1);
2300
2301	/*
2302	* Copy out last hashval in each block for B-tree code.
2303	*/
2304	entry = xfs_attr3_leaf_entryp(leafp: save_leaf);
2305	save_blk->hashval = be32_to_cpu(entry[savehdr.count - 1].hashval);
2306	}
2307
2308	/*========================================================================
2309	* Routines used for finding things in the Btree.
2310	*========================================================================*/
2311
2312	/*
2313	* Look up a name in a leaf attribute list structure.
2314	* This is the internal routine, it uses the caller's buffer.
2315	*
2316	* Note that duplicate keys are allowed, but only check within the
2317	* current leaf node. The Btree code must check in adjacent leaf nodes.
2318	*
2319	* Return in args->index the index into the entry[] array of either
2320	* the found entry, or where the entry should have been (insert before
2321	* that entry).
2322	*
2323	* Don't change the args->value unless we find the attribute.
2324	*/
2325	int
2326	xfs_attr3_leaf_lookup_int(
2327	struct xfs_buf *bp,
2328	struct xfs_da_args *args)
2329	{
2330	struct xfs_attr_leafblock *leaf;
2331	struct xfs_attr3_icleaf_hdr ichdr;
2332	struct xfs_attr_leaf_entry *entry;
2333	struct xfs_attr_leaf_entry *entries;
2334	struct xfs_attr_leaf_name_local *name_loc;
2335	struct xfs_attr_leaf_name_remote *name_rmt;
2336	xfs_dahash_t hashval;
2337	int probe;
2338	int span;
2339
2340	trace_xfs_attr_leaf_lookup(args);
2341
2342	leaf = bp->b_addr;
2343	xfs_attr3_leaf_hdr_from_disk(geo: args->geo, to: &ichdr, from: leaf);
2344	entries = xfs_attr3_leaf_entryp(leafp: leaf);
2345	if (ichdr.count >= args->geo->blksize / 8) {
2346	xfs_buf_mark_corrupt(bp);
2347	xfs_da_mark_sick(args);
2348	return -EFSCORRUPTED;
2349	}
2350
2351	/*
2352	* Binary search. (note: small blocks will skip this loop)
2353	*/
2354	hashval = args->hashval;
2355	probe = span = ichdr.count / 2;
2356	for (entry = &entries[probe]; span > 4; entry = &entries[probe]) {
2357	span /= 2;
2358	if (be32_to_cpu(entry->hashval) < hashval)
2359	probe += span;
2360	else if (be32_to_cpu(entry->hashval) > hashval)
2361	probe -= span;
2362	else
2363	break;
2364	}
2365	if (!(probe >= 0 && (!ichdr.count || probe < ichdr.count))) {
2366	xfs_buf_mark_corrupt(bp);
2367	xfs_da_mark_sick(args);
2368	return -EFSCORRUPTED;
2369	}
2370	if (!(span <= 4 || be32_to_cpu(entry->hashval) == hashval)) {
2371	xfs_buf_mark_corrupt(bp);
2372	xfs_da_mark_sick(args);
2373	return -EFSCORRUPTED;
2374	}
2375
2376	/*
2377	* Since we may have duplicate hashval's, find the first matching
2378	* hashval in the leaf.
2379	*/
2380	while (probe > 0 && be32_to_cpu(entry->hashval) >= hashval) {
2381	entry--;
2382	probe--;
2383	}
2384	while (probe < ichdr.count &&
2385	be32_to_cpu(entry->hashval) < hashval) {
2386	entry++;
2387	probe++;
2388	}
2389	if (probe == ichdr.count || be32_to_cpu(entry->hashval) != hashval) {
2390	args->index = probe;
2391	return -ENOATTR;
2392	}
2393
2394	/*
2395	* Duplicate keys may be present, so search all of them for a match.
2396	*/
2397	for (; probe < ichdr.count && (be32_to_cpu(entry->hashval) == hashval);
2398	entry++, probe++) {
2399	/*
2400	* GROT: Add code to remove incomplete entries.
2401	*/
2402	if (entry->flags & XFS_ATTR_LOCAL) {
2403	name_loc = xfs_attr3_leaf_name_local(leaf, probe);
2404	if (!xfs_attr_match(args, name_loc->namelen,
2405	name_loc->nameval, entry->flags))
2406	continue;
2407	args->index = probe;
2408	return -EEXIST;
2409	} else {
2410	name_rmt = xfs_attr3_leaf_name_remote(leaf, probe);
2411	if (!xfs_attr_match(args, name_rmt->namelen,
2412	name_rmt->name, entry->flags))
2413	continue;
2414	args->index = probe;
2415	args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2416	args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2417	args->rmtblkcnt = xfs_attr3_rmt_blocks(
2418	args->dp->i_mount,
2419	args->rmtvaluelen);
2420	return -EEXIST;
2421	}
2422	}
2423	args->index = probe;
2424	return -ENOATTR;
2425	}
2426
2427	/*
2428	* Get the value associated with an attribute name from a leaf attribute
2429	* list structure.
2430	*
2431	* If args->valuelen is zero, only the length needs to be returned. Unlike a
2432	* lookup, we only return an error if the attribute does not exist or we can't
2433	* retrieve the value.
2434	*/
2435	int
2436	xfs_attr3_leaf_getvalue(
2437	struct xfs_buf *bp,
2438	struct xfs_da_args *args)
2439	{
2440	struct xfs_attr_leafblock *leaf;
2441	struct xfs_attr3_icleaf_hdr ichdr;
2442	struct xfs_attr_leaf_entry *entry;
2443	struct xfs_attr_leaf_name_local *name_loc;
2444	struct xfs_attr_leaf_name_remote *name_rmt;
2445
2446	leaf = bp->b_addr;
2447	xfs_attr3_leaf_hdr_from_disk(geo: args->geo, to: &ichdr, from: leaf);
2448	ASSERT(ichdr.count < args->geo->blksize / 8);
2449	ASSERT(args->index < ichdr.count);
2450
2451	entry = &xfs_attr3_leaf_entryp(leafp: leaf)[args->index];
2452	if (entry->flags & XFS_ATTR_LOCAL) {
2453	name_loc = xfs_attr3_leaf_name_local(leafp: leaf, idx: args->index);
2454	ASSERT(name_loc->namelen == args->namelen);
2455	ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2456	return xfs_attr_copy_value(args,
2457	value: &name_loc->nameval[args->namelen],
2458	valuelen: be16_to_cpu(name_loc->valuelen));
2459	}
2460
2461	name_rmt = xfs_attr3_leaf_name_remote(leafp: leaf, idx: args->index);
2462	ASSERT(name_rmt->namelen == args->namelen);
2463	ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2464	args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2465	args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2466	args->rmtblkcnt = xfs_attr3_rmt_blocks(mp: args->dp->i_mount,
2467	attrlen: args->rmtvaluelen);
2468	return xfs_attr_copy_value(args, NULL, args->rmtvaluelen);
2469	}
2470
2471	/*========================================================================
2472	* Utility routines.
2473	*========================================================================*/
2474
2475	/*
2476	* Move the indicated entries from one leaf to another.
2477	* NOTE: this routine modifies both source and destination leaves.
2478	*/
2479	/*ARGSUSED*/
2480	STATIC void
2481	xfs_attr3_leaf_moveents(
2482	struct xfs_da_args *args,
2483	struct xfs_attr_leafblock *leaf_s,
2484	struct xfs_attr3_icleaf_hdr *ichdr_s,
2485	int start_s,
2486	struct xfs_attr_leafblock *leaf_d,
2487	struct xfs_attr3_icleaf_hdr *ichdr_d,
2488	int start_d,
2489	int count)
2490	{
2491	struct xfs_attr_leaf_entry *entry_s;
2492	struct xfs_attr_leaf_entry *entry_d;
2493	int desti;
2494	int tmp;
2495	int i;
2496
2497	/*
2498	* Check for nothing to do.
2499	*/
2500	if (count == 0)
2501	return;
2502
2503	/*
2504	* Set up environment.
2505	*/
2506	ASSERT(ichdr_s->magic == XFS_ATTR_LEAF_MAGIC ||
2507	ichdr_s->magic == XFS_ATTR3_LEAF_MAGIC);
2508	ASSERT(ichdr_s->magic == ichdr_d->magic);
2509	ASSERT(ichdr_s->count > 0 && ichdr_s->count < args->geo->blksize / 8);
2510	ASSERT(ichdr_s->firstused >= (ichdr_s->count * sizeof(*entry_s))
2511	+ xfs_attr3_leaf_hdr_size(leafp: leaf_s));
2512	ASSERT(ichdr_d->count < args->geo->blksize / 8);
2513	ASSERT(ichdr_d->firstused >= (ichdr_d->count * sizeof(*entry_d))
2514	+ xfs_attr3_leaf_hdr_size(leafp: leaf_d));
2515
2516	ASSERT(start_s < ichdr_s->count);
2517	ASSERT(start_d <= ichdr_d->count);
2518	ASSERT(count <= ichdr_s->count);
2519
2520
2521	/*
2522	* Move the entries in the destination leaf up to make a hole?
2523	*/
2524	if (start_d < ichdr_d->count) {
2525	tmp = ichdr_d->count - start_d;
2526	tmp *= sizeof(xfs_attr_leaf_entry_t);
2527	entry_s = &xfs_attr3_leaf_entryp(leafp: leaf_d)[start_d];
2528	entry_d = &xfs_attr3_leaf_entryp(leafp: leaf_d)[start_d + count];
2529	memmove(entry_d, entry_s, tmp);
2530	}
2531
2532	/*
2533	* Copy all entry's in the same (sorted) order,
2534	* but allocate attribute info packed and in sequence.
2535	*/
2536	entry_s = &xfs_attr3_leaf_entryp(leafp: leaf_s)[start_s];
2537	entry_d = &xfs_attr3_leaf_entryp(leafp: leaf_d)[start_d];
2538	desti = start_d;
2539	for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2540	ASSERT(be16_to_cpu(entry_s->nameidx) >= ichdr_s->firstused);
2541	tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2542	#ifdef GROT
2543	/*
2544	* Code to drop INCOMPLETE entries. Difficult to use as we
2545	* may also need to change the insertion index. Code turned
2546	* off for 6.2, should be revisited later.
2547	*/
2548	if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2549	memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2550	ichdr_s->usedbytes -= tmp;
2551	ichdr_s->count -= 1;
2552	entry_d--; /* to compensate for ++ in loop hdr */
2553	desti--;
2554	if ((start_s + i) < offset)
2555	result++; /* insertion index adjustment */
2556	} else {
2557	#endif /* GROT */
2558	ichdr_d->firstused -= tmp;
2559	/* both on-disk, don't endian flip twice */
2560	entry_d->hashval = entry_s->hashval;
2561	entry_d->nameidx = cpu_to_be16(ichdr_d->firstused);
2562	entry_d->flags = entry_s->flags;
2563	ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2564	<= args->geo->blksize);
2565	memmove(xfs_attr3_leaf_name(leafp: leaf_d, idx: desti),
2566	xfs_attr3_leaf_name(leafp: leaf_s, idx: start_s + i), tmp);
2567	ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2568	<= args->geo->blksize);
2569	memset(xfs_attr3_leaf_name(leafp: leaf_s, idx: start_s + i), 0, tmp);
2570	ichdr_s->usedbytes -= tmp;
2571	ichdr_d->usedbytes += tmp;
2572	ichdr_s->count -= 1;
2573	ichdr_d->count += 1;
2574	tmp = ichdr_d->count * sizeof(xfs_attr_leaf_entry_t)
2575	+ xfs_attr3_leaf_hdr_size(leafp: leaf_d);
2576	ASSERT(ichdr_d->firstused >= tmp);
2577	#ifdef GROT
2578	}
2579	#endif /* GROT */
2580	}
2581
2582	/*
2583	* Zero out the entries we just copied.
2584	*/
2585	if (start_s == ichdr_s->count) {
2586	tmp = count * sizeof(xfs_attr_leaf_entry_t);
2587	entry_s = &xfs_attr3_leaf_entryp(leafp: leaf_s)[start_s];
2588	ASSERT(((char *)entry_s + tmp) <=
2589	((char *)leaf_s + args->geo->blksize));
2590	memset(entry_s, 0, tmp);
2591	} else {
2592	/*
2593	* Move the remaining entries down to fill the hole,
2594	* then zero the entries at the top.
2595	*/
2596	tmp = (ichdr_s->count - count) * sizeof(xfs_attr_leaf_entry_t);
2597	entry_s = &xfs_attr3_leaf_entryp(leafp: leaf_s)[start_s + count];
2598	entry_d = &xfs_attr3_leaf_entryp(leafp: leaf_s)[start_s];
2599	memmove(entry_d, entry_s, tmp);
2600
2601	tmp = count * sizeof(xfs_attr_leaf_entry_t);
2602	entry_s = &xfs_attr3_leaf_entryp(leafp: leaf_s)[ichdr_s->count];
2603	ASSERT(((char *)entry_s + tmp) <=
2604	((char *)leaf_s + args->geo->blksize));
2605	memset(entry_s, 0, tmp);
2606	}
2607
2608	/*
2609	* Fill in the freemap information
2610	*/
2611	ichdr_d->freemap[0].base = xfs_attr3_leaf_hdr_size(leafp: leaf_d);
2612	ichdr_d->freemap[0].base += ichdr_d->count * sizeof(xfs_attr_leaf_entry_t);
2613	ichdr_d->freemap[0].size = ichdr_d->firstused - ichdr_d->freemap[0].base;
2614	ichdr_d->freemap[1].base = 0;
2615	ichdr_d->freemap[2].base = 0;
2616	ichdr_d->freemap[1].size = 0;
2617	ichdr_d->freemap[2].size = 0;
2618	ichdr_s->holes = 1; /* leaf may not be compact */
2619	}
2620
2621	/*
2622	* Pick up the last hashvalue from a leaf block.
2623	*/
2624	xfs_dahash_t
2625	xfs_attr_leaf_lasthash(
2626	struct xfs_buf *bp,
2627	int *count)
2628	{
2629	struct xfs_attr3_icleaf_hdr ichdr;
2630	struct xfs_attr_leaf_entry *entries;
2631	struct xfs_mount *mp = bp->b_mount;
2632
2633	xfs_attr3_leaf_hdr_from_disk(geo: mp->m_attr_geo, to: &ichdr, from: bp->b_addr);
2634	entries = xfs_attr3_leaf_entryp(leafp: bp->b_addr);
2635	if (count)
2636	*count = ichdr.count;
2637	if (!ichdr.count)
2638	return 0;
2639	return be32_to_cpu(entries[ichdr.count - 1].hashval);
2640	}
2641
2642	/*
2643	* Calculate the number of bytes used to store the indicated attribute
2644	* (whether local or remote only calculate bytes in this block).
2645	*/
2646	STATIC int
2647	xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2648	{
2649	struct xfs_attr_leaf_entry *entries;
2650	xfs_attr_leaf_name_local_t *name_loc;
2651	xfs_attr_leaf_name_remote_t *name_rmt;
2652	int size;
2653
2654	entries = xfs_attr3_leaf_entryp(leafp: leaf);
2655	if (entries[index].flags & XFS_ATTR_LOCAL) {
2656	name_loc = xfs_attr3_leaf_name_local(leafp: leaf, idx: index);
2657	size = xfs_attr_leaf_entsize_local(nlen: name_loc->namelen,
2658	vlen: be16_to_cpu(name_loc->valuelen));
2659	} else {
2660	name_rmt = xfs_attr3_leaf_name_remote(leafp: leaf, idx: index);
2661	size = xfs_attr_leaf_entsize_remote(nlen: name_rmt->namelen);
2662	}
2663	return size;
2664	}
2665
2666	/*
2667	* Calculate the number of bytes that would be required to store the new
2668	* attribute (whether local or remote only calculate bytes in this block).
2669	* This routine decides as a side effect whether the attribute will be
2670	* a "local" or a "remote" attribute.
2671	*/
2672	int
2673	xfs_attr_leaf_newentsize(
2674	struct xfs_da_args *args,
2675	int *local)
2676	{
2677	int size;
2678
2679	size = xfs_attr_leaf_entsize_local(nlen: args->namelen, vlen: args->valuelen);
2680	if (size < xfs_attr_leaf_entsize_local_max(bsize: args->geo->blksize)) {
2681	if (local)
2682	*local = 1;
2683	return size;
2684	}
2685	if (local)
2686	*local = 0;
2687	return xfs_attr_leaf_entsize_remote(nlen: args->namelen);
2688	}
2689
2690
2691	/*========================================================================
2692	* Manage the INCOMPLETE flag in a leaf entry
2693	*========================================================================*/
2694
2695	/*
2696	* Clear the INCOMPLETE flag on an entry in a leaf block.
2697	*/
2698	int
2699	xfs_attr3_leaf_clearflag(
2700	struct xfs_da_args *args)
2701	{
2702	struct xfs_attr_leafblock *leaf;
2703	struct xfs_attr_leaf_entry *entry;
2704	struct xfs_attr_leaf_name_remote *name_rmt;
2705	struct xfs_buf *bp;
2706	int error;
2707	#ifdef DEBUG
2708	struct xfs_attr3_icleaf_hdr ichdr;
2709	xfs_attr_leaf_name_local_t *name_loc;
2710	int namelen;
2711	char *name;
2712	#endif /* DEBUG */
2713
2714	trace_xfs_attr_leaf_clearflag(args);
2715	/*
2716	* Set up the operation.
2717	*/
2718	error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp);
2719	if (error)
2720	return error;
2721
2722	leaf = bp->b_addr;
2723	entry = &xfs_attr3_leaf_entryp(leafp: leaf)[args->index];
2724	ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2725
2726	#ifdef DEBUG
2727	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2728	ASSERT(args->index < ichdr.count);
2729	ASSERT(args->index >= 0);
2730
2731	if (entry->flags & XFS_ATTR_LOCAL) {
2732	name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2733	namelen = name_loc->namelen;
2734	name = (char *)name_loc->nameval;
2735	} else {
2736	name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2737	namelen = name_rmt->namelen;
2738	name = (char *)name_rmt->name;
2739	}
2740	ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2741	ASSERT(namelen == args->namelen);
2742	ASSERT(memcmp(name, args->name, namelen) == 0);
2743	#endif /* DEBUG */
2744
2745	entry->flags &= ~XFS_ATTR_INCOMPLETE;
2746	xfs_trans_log_buf(args->trans, bp,
2747	XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2748
2749	if (args->rmtblkno) {
2750	ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2751	name_rmt = xfs_attr3_leaf_name_remote(leafp: leaf, idx: args->index);
2752	name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2753	name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2754	xfs_trans_log_buf(args->trans, bp,
2755	XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2756	}
2757
2758	return 0;
2759	}
2760
2761	/*
2762	* Set the INCOMPLETE flag on an entry in a leaf block.
2763	*/
2764	int
2765	xfs_attr3_leaf_setflag(
2766	struct xfs_da_args *args)
2767	{
2768	struct xfs_attr_leafblock *leaf;
2769	struct xfs_attr_leaf_entry *entry;
2770	struct xfs_attr_leaf_name_remote *name_rmt;
2771	struct xfs_buf *bp;
2772	int error;
2773	#ifdef DEBUG
2774	struct xfs_attr3_icleaf_hdr ichdr;
2775	#endif
2776
2777	trace_xfs_attr_leaf_setflag(args);
2778
2779	/*
2780	* Set up the operation.
2781	*/
2782	error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp);
2783	if (error)
2784	return error;
2785
2786	leaf = bp->b_addr;
2787	#ifdef DEBUG
2788	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2789	ASSERT(args->index < ichdr.count);
2790	ASSERT(args->index >= 0);
2791	#endif
2792	entry = &xfs_attr3_leaf_entryp(leafp: leaf)[args->index];
2793
2794	ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2795	entry->flags |= XFS_ATTR_INCOMPLETE;
2796	xfs_trans_log_buf(args->trans, bp,
2797	XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2798	if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2799	name_rmt = xfs_attr3_leaf_name_remote(leafp: leaf, idx: args->index);
2800	name_rmt->valueblk = 0;
2801	name_rmt->valuelen = 0;
2802	xfs_trans_log_buf(args->trans, bp,
2803	XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2804	}
2805
2806	return 0;
2807	}
2808
2809	/*
2810	* In a single transaction, clear the INCOMPLETE flag on the leaf entry
2811	* given by args->blkno/index and set the INCOMPLETE flag on the leaf
2812	* entry given by args->blkno2/index2.
2813	*
2814	* Note that they could be in different blocks, or in the same block.
2815	*/
2816	int
2817	xfs_attr3_leaf_flipflags(
2818	struct xfs_da_args *args)
2819	{
2820	struct xfs_attr_leafblock *leaf1;
2821	struct xfs_attr_leafblock *leaf2;
2822	struct xfs_attr_leaf_entry *entry1;
2823	struct xfs_attr_leaf_entry *entry2;
2824	struct xfs_attr_leaf_name_remote *name_rmt;
2825	struct xfs_buf *bp1;
2826	struct xfs_buf *bp2;
2827	int error;
2828	#ifdef DEBUG
2829	struct xfs_attr3_icleaf_hdr ichdr1;
2830	struct xfs_attr3_icleaf_hdr ichdr2;
2831	xfs_attr_leaf_name_local_t *name_loc;
2832	int namelen1, namelen2;
2833	char *name1, *name2;
2834	#endif /* DEBUG */
2835
2836	trace_xfs_attr_leaf_flipflags(args);
2837
2838	/*
2839	* Read the block containing the "old" attr
2840	*/
2841	error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp1);
2842	if (error)
2843	return error;
2844
2845	/*
2846	* Read the block containing the "new" attr, if it is different
2847	*/
2848	if (args->blkno2 != args->blkno) {
2849	error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno2,
2850	&bp2);
2851	if (error)
2852	return error;
2853	} else {
2854	bp2 = bp1;
2855	}
2856
2857	leaf1 = bp1->b_addr;
2858	entry1 = &xfs_attr3_leaf_entryp(leafp: leaf1)[args->index];
2859
2860	leaf2 = bp2->b_addr;
2861	entry2 = &xfs_attr3_leaf_entryp(leafp: leaf2)[args->index2];
2862
2863	#ifdef DEBUG
2864	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr1, leaf1);
2865	ASSERT(args->index < ichdr1.count);
2866	ASSERT(args->index >= 0);
2867
2868	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr2, leaf2);
2869	ASSERT(args->index2 < ichdr2.count);
2870	ASSERT(args->index2 >= 0);
2871
2872	if (entry1->flags & XFS_ATTR_LOCAL) {
2873	name_loc = xfs_attr3_leaf_name_local(leaf1, args->index);
2874	namelen1 = name_loc->namelen;
2875	name1 = (char *)name_loc->nameval;
2876	} else {
2877	name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2878	namelen1 = name_rmt->namelen;
2879	name1 = (char *)name_rmt->name;
2880	}
2881	if (entry2->flags & XFS_ATTR_LOCAL) {
2882	name_loc = xfs_attr3_leaf_name_local(leaf2, args->index2);
2883	namelen2 = name_loc->namelen;
2884	name2 = (char *)name_loc->nameval;
2885	} else {
2886	name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2887	namelen2 = name_rmt->namelen;
2888	name2 = (char *)name_rmt->name;
2889	}
2890	ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2891	ASSERT(namelen1 == namelen2);
2892	ASSERT(memcmp(name1, name2, namelen1) == 0);
2893	#endif /* DEBUG */
2894
2895	ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2896	ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2897
2898	entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2899	xfs_trans_log_buf(args->trans, bp1,
2900	XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2901	if (args->rmtblkno) {
2902	ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2903	name_rmt = xfs_attr3_leaf_name_remote(leafp: leaf1, idx: args->index);
2904	name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2905	name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2906	xfs_trans_log_buf(args->trans, bp1,
2907	XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2908	}
2909
2910	entry2->flags |= XFS_ATTR_INCOMPLETE;
2911	xfs_trans_log_buf(args->trans, bp2,
2912	XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2913	if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
2914	name_rmt = xfs_attr3_leaf_name_remote(leafp: leaf2, idx: args->index2);
2915	name_rmt->valueblk = 0;
2916	name_rmt->valuelen = 0;
2917	xfs_trans_log_buf(args->trans, bp2,
2918	XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
2919	}
2920
2921	return 0;
2922	}
2923