xref: /openbmc/linux/fs/xfs/libxfs/xfs_attr_leaf.c (revision 2a9eb57e)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4  * Copyright (c) 2013 Red Hat, Inc.
5  * All Rights Reserved.
6  */
7 #include "xfs.h"
8 #include "xfs_fs.h"
9 #include "xfs_shared.h"
10 #include "xfs_format.h"
11 #include "xfs_log_format.h"
12 #include "xfs_trans_resv.h"
13 #include "xfs_sb.h"
14 #include "xfs_mount.h"
15 #include "xfs_da_format.h"
16 #include "xfs_da_btree.h"
17 #include "xfs_inode.h"
18 #include "xfs_trans.h"
19 #include "xfs_bmap_btree.h"
20 #include "xfs_bmap.h"
21 #include "xfs_attr_sf.h"
22 #include "xfs_attr.h"
23 #include "xfs_attr_remote.h"
24 #include "xfs_attr_leaf.h"
25 #include "xfs_error.h"
26 #include "xfs_trace.h"
27 #include "xfs_buf_item.h"
28 #include "xfs_dir2.h"
29 #include "xfs_log.h"
30 #include "xfs_ag.h"
31 #include "xfs_errortag.h"
32 
33 
34 /*
35  * xfs_attr_leaf.c
36  *
37  * Routines to implement leaf blocks of attributes as Btrees of hashed names.
38  */
39 
40 /*========================================================================
41  * Function prototypes for the kernel.
42  *========================================================================*/
43 
44 /*
45  * Routines used for growing the Btree.
46  */
47 STATIC int xfs_attr3_leaf_create(struct xfs_da_args *args,
48 				 xfs_dablk_t which_block, struct xfs_buf **bpp);
49 STATIC int xfs_attr3_leaf_add_work(struct xfs_buf *leaf_buffer,
50 				   struct xfs_attr3_icleaf_hdr *ichdr,
51 				   struct xfs_da_args *args, int freemap_index);
52 STATIC void xfs_attr3_leaf_compact(struct xfs_da_args *args,
53 				   struct xfs_attr3_icleaf_hdr *ichdr,
54 				   struct xfs_buf *leaf_buffer);
55 STATIC void xfs_attr3_leaf_rebalance(xfs_da_state_t *state,
56 						   xfs_da_state_blk_t *blk1,
57 						   xfs_da_state_blk_t *blk2);
58 STATIC int xfs_attr3_leaf_figure_balance(xfs_da_state_t *state,
59 			xfs_da_state_blk_t *leaf_blk_1,
60 			struct xfs_attr3_icleaf_hdr *ichdr1,
61 			xfs_da_state_blk_t *leaf_blk_2,
62 			struct xfs_attr3_icleaf_hdr *ichdr2,
63 			int *number_entries_in_blk1,
64 			int *number_usedbytes_in_blk1);
65 
66 /*
67  * Utility routines.
68  */
69 STATIC void xfs_attr3_leaf_moveents(struct xfs_da_args *args,
70 			struct xfs_attr_leafblock *src_leaf,
71 			struct xfs_attr3_icleaf_hdr *src_ichdr, int src_start,
72 			struct xfs_attr_leafblock *dst_leaf,
73 			struct xfs_attr3_icleaf_hdr *dst_ichdr, int dst_start,
74 			int move_count);
75 STATIC int xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index);
76 
77 /*
78  * attr3 block 'firstused' conversion helpers.
79  *
80  * firstused refers to the offset of the first used byte of the nameval region
81  * of an attr leaf block. The region starts at the tail of the block and expands
82  * backwards towards the middle. As such, firstused is initialized to the block
83  * size for an empty leaf block and is reduced from there.
84  *
85  * The attr3 block size is pegged to the fsb size and the maximum fsb is 64k.
86  * The in-core firstused field is 32-bit and thus supports the maximum fsb size.
87  * The on-disk field is only 16-bit, however, and overflows at 64k. Since this
88  * only occurs at exactly 64k, we use zero as a magic on-disk value to represent
89  * the attr block size. The following helpers manage the conversion between the
90  * in-core and on-disk formats.
91  */
92 
93 static void
94 xfs_attr3_leaf_firstused_from_disk(
95 	struct xfs_da_geometry		*geo,
96 	struct xfs_attr3_icleaf_hdr	*to,
97 	struct xfs_attr_leafblock	*from)
98 {
99 	struct xfs_attr3_leaf_hdr	*hdr3;
100 
101 	if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) {
102 		hdr3 = (struct xfs_attr3_leaf_hdr *) from;
103 		to->firstused = be16_to_cpu(hdr3->firstused);
104 	} else {
105 		to->firstused = be16_to_cpu(from->hdr.firstused);
106 	}
107 
108 	/*
109 	 * Convert from the magic fsb size value to actual blocksize. This
110 	 * should only occur for empty blocks when the block size overflows
111 	 * 16-bits.
112 	 */
113 	if (to->firstused == XFS_ATTR3_LEAF_NULLOFF) {
114 		ASSERT(!to->count && !to->usedbytes);
115 		ASSERT(geo->blksize > USHRT_MAX);
116 		to->firstused = geo->blksize;
117 	}
118 }
119 
120 static void
121 xfs_attr3_leaf_firstused_to_disk(
122 	struct xfs_da_geometry		*geo,
123 	struct xfs_attr_leafblock	*to,
124 	struct xfs_attr3_icleaf_hdr	*from)
125 {
126 	struct xfs_attr3_leaf_hdr	*hdr3;
127 	uint32_t			firstused;
128 
129 	/* magic value should only be seen on disk */
130 	ASSERT(from->firstused != XFS_ATTR3_LEAF_NULLOFF);
131 
132 	/*
133 	 * Scale down the 32-bit in-core firstused value to the 16-bit on-disk
134 	 * value. This only overflows at the max supported value of 64k. Use the
135 	 * magic on-disk value to represent block size in this case.
136 	 */
137 	firstused = from->firstused;
138 	if (firstused > USHRT_MAX) {
139 		ASSERT(from->firstused == geo->blksize);
140 		firstused = XFS_ATTR3_LEAF_NULLOFF;
141 	}
142 
143 	if (from->magic == XFS_ATTR3_LEAF_MAGIC) {
144 		hdr3 = (struct xfs_attr3_leaf_hdr *) to;
145 		hdr3->firstused = cpu_to_be16(firstused);
146 	} else {
147 		to->hdr.firstused = cpu_to_be16(firstused);
148 	}
149 }
150 
151 void
152 xfs_attr3_leaf_hdr_from_disk(
153 	struct xfs_da_geometry		*geo,
154 	struct xfs_attr3_icleaf_hdr	*to,
155 	struct xfs_attr_leafblock	*from)
156 {
157 	int	i;
158 
159 	ASSERT(from->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
160 	       from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
161 
162 	if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) {
163 		struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)from;
164 
165 		to->forw = be32_to_cpu(hdr3->info.hdr.forw);
166 		to->back = be32_to_cpu(hdr3->info.hdr.back);
167 		to->magic = be16_to_cpu(hdr3->info.hdr.magic);
168 		to->count = be16_to_cpu(hdr3->count);
169 		to->usedbytes = be16_to_cpu(hdr3->usedbytes);
170 		xfs_attr3_leaf_firstused_from_disk(geo, to, from);
171 		to->holes = hdr3->holes;
172 
173 		for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
174 			to->freemap[i].base = be16_to_cpu(hdr3->freemap[i].base);
175 			to->freemap[i].size = be16_to_cpu(hdr3->freemap[i].size);
176 		}
177 		return;
178 	}
179 	to->forw = be32_to_cpu(from->hdr.info.forw);
180 	to->back = be32_to_cpu(from->hdr.info.back);
181 	to->magic = be16_to_cpu(from->hdr.info.magic);
182 	to->count = be16_to_cpu(from->hdr.count);
183 	to->usedbytes = be16_to_cpu(from->hdr.usedbytes);
184 	xfs_attr3_leaf_firstused_from_disk(geo, to, from);
185 	to->holes = from->hdr.holes;
186 
187 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
188 		to->freemap[i].base = be16_to_cpu(from->hdr.freemap[i].base);
189 		to->freemap[i].size = be16_to_cpu(from->hdr.freemap[i].size);
190 	}
191 }
192 
193 void
194 xfs_attr3_leaf_hdr_to_disk(
195 	struct xfs_da_geometry		*geo,
196 	struct xfs_attr_leafblock	*to,
197 	struct xfs_attr3_icleaf_hdr	*from)
198 {
199 	int				i;
200 
201 	ASSERT(from->magic == XFS_ATTR_LEAF_MAGIC ||
202 	       from->magic == XFS_ATTR3_LEAF_MAGIC);
203 
204 	if (from->magic == XFS_ATTR3_LEAF_MAGIC) {
205 		struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)to;
206 
207 		hdr3->info.hdr.forw = cpu_to_be32(from->forw);
208 		hdr3->info.hdr.back = cpu_to_be32(from->back);
209 		hdr3->info.hdr.magic = cpu_to_be16(from->magic);
210 		hdr3->count = cpu_to_be16(from->count);
211 		hdr3->usedbytes = cpu_to_be16(from->usedbytes);
212 		xfs_attr3_leaf_firstused_to_disk(geo, to, from);
213 		hdr3->holes = from->holes;
214 		hdr3->pad1 = 0;
215 
216 		for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
217 			hdr3->freemap[i].base = cpu_to_be16(from->freemap[i].base);
218 			hdr3->freemap[i].size = cpu_to_be16(from->freemap[i].size);
219 		}
220 		return;
221 	}
222 	to->hdr.info.forw = cpu_to_be32(from->forw);
223 	to->hdr.info.back = cpu_to_be32(from->back);
224 	to->hdr.info.magic = cpu_to_be16(from->magic);
225 	to->hdr.count = cpu_to_be16(from->count);
226 	to->hdr.usedbytes = cpu_to_be16(from->usedbytes);
227 	xfs_attr3_leaf_firstused_to_disk(geo, to, from);
228 	to->hdr.holes = from->holes;
229 	to->hdr.pad1 = 0;
230 
231 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
232 		to->hdr.freemap[i].base = cpu_to_be16(from->freemap[i].base);
233 		to->hdr.freemap[i].size = cpu_to_be16(from->freemap[i].size);
234 	}
235 }
236 
237 static xfs_failaddr_t
238 xfs_attr3_leaf_verify_entry(
239 	struct xfs_mount			*mp,
240 	char					*buf_end,
241 	struct xfs_attr_leafblock		*leaf,
242 	struct xfs_attr3_icleaf_hdr		*leafhdr,
243 	struct xfs_attr_leaf_entry		*ent,
244 	int					idx,
245 	__u32					*last_hashval)
246 {
247 	struct xfs_attr_leaf_name_local		*lentry;
248 	struct xfs_attr_leaf_name_remote	*rentry;
249 	char					*name_end;
250 	unsigned int				nameidx;
251 	unsigned int				namesize;
252 	__u32					hashval;
253 
254 	/* hash order check */
255 	hashval = be32_to_cpu(ent->hashval);
256 	if (hashval < *last_hashval)
257 		return __this_address;
258 	*last_hashval = hashval;
259 
260 	nameidx = be16_to_cpu(ent->nameidx);
261 	if (nameidx < leafhdr->firstused || nameidx >= mp->m_attr_geo->blksize)
262 		return __this_address;
263 
264 	/*
265 	 * Check the name information.  The namelen fields are u8 so we can't
266 	 * possibly exceed the maximum name length of 255 bytes.
267 	 */
268 	if (ent->flags & XFS_ATTR_LOCAL) {
269 		lentry = xfs_attr3_leaf_name_local(leaf, idx);
270 		namesize = xfs_attr_leaf_entsize_local(lentry->namelen,
271 				be16_to_cpu(lentry->valuelen));
272 		name_end = (char *)lentry + namesize;
273 		if (lentry->namelen == 0)
274 			return __this_address;
275 	} else {
276 		rentry = xfs_attr3_leaf_name_remote(leaf, idx);
277 		namesize = xfs_attr_leaf_entsize_remote(rentry->namelen);
278 		name_end = (char *)rentry + namesize;
279 		if (rentry->namelen == 0)
280 			return __this_address;
281 		if (!(ent->flags & XFS_ATTR_INCOMPLETE) &&
282 		    rentry->valueblk == 0)
283 			return __this_address;
284 	}
285 
286 	if (name_end > buf_end)
287 		return __this_address;
288 
289 	return NULL;
290 }
291 
292 /*
293  * Validate an attribute leaf block.
294  *
295  * Empty leaf blocks can occur under the following circumstances:
296  *
297  * 1. setxattr adds a new extended attribute to a file;
298  * 2. The file has zero existing attributes;
299  * 3. The attribute is too large to fit in the attribute fork;
300  * 4. The attribute is small enough to fit in a leaf block;
301  * 5. A log flush occurs after committing the transaction that creates
302  *    the (empty) leaf block; and
303  * 6. The filesystem goes down after the log flush but before the new
304  *    attribute can be committed to the leaf block.
305  *
306  * Hence we need to ensure that we don't fail the validation purely
307  * because the leaf is empty.
308  */
309 static xfs_failaddr_t
310 xfs_attr3_leaf_verify(
311 	struct xfs_buf			*bp)
312 {
313 	struct xfs_attr3_icleaf_hdr	ichdr;
314 	struct xfs_mount		*mp = bp->b_mount;
315 	struct xfs_attr_leafblock	*leaf = bp->b_addr;
316 	struct xfs_attr_leaf_entry	*entries;
317 	struct xfs_attr_leaf_entry	*ent;
318 	char				*buf_end;
319 	uint32_t			end;	/* must be 32bit - see below */
320 	__u32				last_hashval = 0;
321 	int				i;
322 	xfs_failaddr_t			fa;
323 
324 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, leaf);
325 
326 	fa = xfs_da3_blkinfo_verify(bp, bp->b_addr);
327 	if (fa)
328 		return fa;
329 
330 	/*
331 	 * firstused is the block offset of the first name info structure.
332 	 * Make sure it doesn't go off the block or crash into the header.
333 	 */
334 	if (ichdr.firstused > mp->m_attr_geo->blksize)
335 		return __this_address;
336 	if (ichdr.firstused < xfs_attr3_leaf_hdr_size(leaf))
337 		return __this_address;
338 
339 	/* Make sure the entries array doesn't crash into the name info. */
340 	entries = xfs_attr3_leaf_entryp(bp->b_addr);
341 	if ((char *)&entries[ichdr.count] >
342 	    (char *)bp->b_addr + ichdr.firstused)
343 		return __this_address;
344 
345 	/*
346 	 * NOTE: This verifier historically failed empty leaf buffers because
347 	 * we expect the fork to be in another format. Empty attr fork format
348 	 * conversions are possible during xattr set, however, and format
349 	 * conversion is not atomic with the xattr set that triggers it. We
350 	 * cannot assume leaf blocks are non-empty until that is addressed.
351 	*/
352 	buf_end = (char *)bp->b_addr + mp->m_attr_geo->blksize;
353 	for (i = 0, ent = entries; i < ichdr.count; ent++, i++) {
354 		fa = xfs_attr3_leaf_verify_entry(mp, buf_end, leaf, &ichdr,
355 				ent, i, &last_hashval);
356 		if (fa)
357 			return fa;
358 	}
359 
360 	/*
361 	 * Quickly check the freemap information.  Attribute data has to be
362 	 * aligned to 4-byte boundaries, and likewise for the free space.
363 	 *
364 	 * Note that for 64k block size filesystems, the freemap entries cannot
365 	 * overflow as they are only be16 fields. However, when checking end
366 	 * pointer of the freemap, we have to be careful to detect overflows and
367 	 * so use uint32_t for those checks.
368 	 */
369 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
370 		if (ichdr.freemap[i].base > mp->m_attr_geo->blksize)
371 			return __this_address;
372 		if (ichdr.freemap[i].base & 0x3)
373 			return __this_address;
374 		if (ichdr.freemap[i].size > mp->m_attr_geo->blksize)
375 			return __this_address;
376 		if (ichdr.freemap[i].size & 0x3)
377 			return __this_address;
378 
379 		/* be care of 16 bit overflows here */
380 		end = (uint32_t)ichdr.freemap[i].base + ichdr.freemap[i].size;
381 		if (end < ichdr.freemap[i].base)
382 			return __this_address;
383 		if (end > mp->m_attr_geo->blksize)
384 			return __this_address;
385 	}
386 
387 	return NULL;
388 }
389 
390 static void
391 xfs_attr3_leaf_write_verify(
392 	struct xfs_buf	*bp)
393 {
394 	struct xfs_mount	*mp = bp->b_mount;
395 	struct xfs_buf_log_item	*bip = bp->b_log_item;
396 	struct xfs_attr3_leaf_hdr *hdr3 = bp->b_addr;
397 	xfs_failaddr_t		fa;
398 
399 	fa = xfs_attr3_leaf_verify(bp);
400 	if (fa) {
401 		xfs_verifier_error(bp, -EFSCORRUPTED, fa);
402 		return;
403 	}
404 
405 	if (!xfs_has_crc(mp))
406 		return;
407 
408 	if (bip)
409 		hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
410 
411 	xfs_buf_update_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF);
412 }
413 
414 /*
415  * leaf/node format detection on trees is sketchy, so a node read can be done on
416  * leaf level blocks when detection identifies the tree as a node format tree
417  * incorrectly. In this case, we need to swap the verifier to match the correct
418  * format of the block being read.
419  */
420 static void
421 xfs_attr3_leaf_read_verify(
422 	struct xfs_buf		*bp)
423 {
424 	struct xfs_mount	*mp = bp->b_mount;
425 	xfs_failaddr_t		fa;
426 
427 	if (xfs_has_crc(mp) &&
428 	     !xfs_buf_verify_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF))
429 		xfs_verifier_error(bp, -EFSBADCRC, __this_address);
430 	else {
431 		fa = xfs_attr3_leaf_verify(bp);
432 		if (fa)
433 			xfs_verifier_error(bp, -EFSCORRUPTED, fa);
434 	}
435 }
436 
437 const struct xfs_buf_ops xfs_attr3_leaf_buf_ops = {
438 	.name = "xfs_attr3_leaf",
439 	.magic16 = { cpu_to_be16(XFS_ATTR_LEAF_MAGIC),
440 		     cpu_to_be16(XFS_ATTR3_LEAF_MAGIC) },
441 	.verify_read = xfs_attr3_leaf_read_verify,
442 	.verify_write = xfs_attr3_leaf_write_verify,
443 	.verify_struct = xfs_attr3_leaf_verify,
444 };
445 
446 int
447 xfs_attr3_leaf_read(
448 	struct xfs_trans	*tp,
449 	struct xfs_inode	*dp,
450 	xfs_dablk_t		bno,
451 	struct xfs_buf		**bpp)
452 {
453 	int			err;
454 
455 	err = xfs_da_read_buf(tp, dp, bno, 0, bpp, XFS_ATTR_FORK,
456 			&xfs_attr3_leaf_buf_ops);
457 	if (!err && tp && *bpp)
458 		xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_ATTR_LEAF_BUF);
459 	return err;
460 }
461 
462 /*========================================================================
463  * Namespace helper routines
464  *========================================================================*/
465 
466 /*
467  * If we are in log recovery, then we want the lookup to ignore the INCOMPLETE
468  * flag on disk - if there's an incomplete attr then recovery needs to tear it
469  * down. If there's no incomplete attr, then recovery needs to tear that attr
470  * down to replace it with the attr that has been logged. In this case, the
471  * INCOMPLETE flag will not be set in attr->attr_filter, but rather
472  * XFS_DA_OP_RECOVERY will be set in args->op_flags.
473  */
474 static bool
475 xfs_attr_match(
476 	struct xfs_da_args	*args,
477 	uint8_t			namelen,
478 	unsigned char		*name,
479 	int			flags)
480 {
481 
482 	if (args->namelen != namelen)
483 		return false;
484 	if (memcmp(args->name, name, namelen) != 0)
485 		return false;
486 
487 	/* Recovery ignores the INCOMPLETE flag. */
488 	if ((args->op_flags & XFS_DA_OP_RECOVERY) &&
489 	    args->attr_filter == (flags & XFS_ATTR_NSP_ONDISK_MASK))
490 		return true;
491 
492 	/* All remaining matches need to be filtered by INCOMPLETE state. */
493 	if (args->attr_filter !=
494 	    (flags & (XFS_ATTR_NSP_ONDISK_MASK | XFS_ATTR_INCOMPLETE)))
495 		return false;
496 	return true;
497 }
498 
499 static int
500 xfs_attr_copy_value(
501 	struct xfs_da_args	*args,
502 	unsigned char		*value,
503 	int			valuelen)
504 {
505 	/*
506 	 * No copy if all we have to do is get the length
507 	 */
508 	if (!args->valuelen) {
509 		args->valuelen = valuelen;
510 		return 0;
511 	}
512 
513 	/*
514 	 * No copy if the length of the existing buffer is too small
515 	 */
516 	if (args->valuelen < valuelen) {
517 		args->valuelen = valuelen;
518 		return -ERANGE;
519 	}
520 
521 	if (!args->value) {
522 		args->value = kvmalloc(valuelen, GFP_KERNEL | __GFP_NOLOCKDEP);
523 		if (!args->value)
524 			return -ENOMEM;
525 	}
526 	args->valuelen = valuelen;
527 
528 	/* remote block xattr requires IO for copy-in */
529 	if (args->rmtblkno)
530 		return xfs_attr_rmtval_get(args);
531 
532 	/*
533 	 * This is to prevent a GCC warning because the remote xattr case
534 	 * doesn't have a value to pass in. In that case, we never reach here,
535 	 * but GCC can't work that out and so throws a "passing NULL to
536 	 * memcpy" warning.
537 	 */
538 	if (!value)
539 		return -EINVAL;
540 	memcpy(args->value, value, valuelen);
541 	return 0;
542 }
543 
544 /*========================================================================
545  * External routines when attribute fork size < XFS_LITINO(mp).
546  *========================================================================*/
547 
548 /*
549  * Query whether the total requested number of attr fork bytes of extended
550  * attribute space will be able to fit inline.
551  *
552  * Returns zero if not, else the i_forkoff fork offset to be used in the
553  * literal area for attribute data once the new bytes have been added.
554  *
555  * i_forkoff must be 8 byte aligned, hence is stored as a >>3 value;
556  * special case for dev/uuid inodes, they have fixed size data forks.
557  */
558 int
559 xfs_attr_shortform_bytesfit(
560 	struct xfs_inode	*dp,
561 	int			bytes)
562 {
563 	struct xfs_mount	*mp = dp->i_mount;
564 	int64_t			dsize;
565 	int			minforkoff;
566 	int			maxforkoff;
567 	int			offset;
568 
569 	/*
570 	 * Check if the new size could fit at all first:
571 	 */
572 	if (bytes > XFS_LITINO(mp))
573 		return 0;
574 
575 	/* rounded down */
576 	offset = (XFS_LITINO(mp) - bytes) >> 3;
577 
578 	if (dp->i_df.if_format == XFS_DINODE_FMT_DEV) {
579 		minforkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
580 		return (offset >= minforkoff) ? minforkoff : 0;
581 	}
582 
583 	/*
584 	 * If the requested numbers of bytes is smaller or equal to the
585 	 * current attribute fork size we can always proceed.
586 	 *
587 	 * Note that if_bytes in the data fork might actually be larger than
588 	 * the current data fork size is due to delalloc extents. In that
589 	 * case either the extent count will go down when they are converted
590 	 * to real extents, or the delalloc conversion will take care of the
591 	 * literal area rebalancing.
592 	 */
593 	if (bytes <= xfs_inode_attr_fork_size(dp))
594 		return dp->i_forkoff;
595 
596 	/*
597 	 * For attr2 we can try to move the forkoff if there is space in the
598 	 * literal area, but for the old format we are done if there is no
599 	 * space in the fixed attribute fork.
600 	 */
601 	if (!xfs_has_attr2(mp))
602 		return 0;
603 
604 	dsize = dp->i_df.if_bytes;
605 
606 	switch (dp->i_df.if_format) {
607 	case XFS_DINODE_FMT_EXTENTS:
608 		/*
609 		 * If there is no attr fork and the data fork is extents,
610 		 * determine if creating the default attr fork will result
611 		 * in the extents form migrating to btree. If so, the
612 		 * minimum offset only needs to be the space required for
613 		 * the btree root.
614 		 */
615 		if (!dp->i_forkoff && dp->i_df.if_bytes >
616 		    xfs_default_attroffset(dp))
617 			dsize = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
618 		break;
619 	case XFS_DINODE_FMT_BTREE:
620 		/*
621 		 * If we have a data btree then keep forkoff if we have one,
622 		 * otherwise we are adding a new attr, so then we set
623 		 * minforkoff to where the btree root can finish so we have
624 		 * plenty of room for attrs
625 		 */
626 		if (dp->i_forkoff) {
627 			if (offset < dp->i_forkoff)
628 				return 0;
629 			return dp->i_forkoff;
630 		}
631 		dsize = XFS_BMAP_BROOT_SPACE(mp, dp->i_df.if_broot);
632 		break;
633 	}
634 
635 	/*
636 	 * A data fork btree root must have space for at least
637 	 * MINDBTPTRS key/ptr pairs if the data fork is small or empty.
638 	 */
639 	minforkoff = max_t(int64_t, dsize, XFS_BMDR_SPACE_CALC(MINDBTPTRS));
640 	minforkoff = roundup(minforkoff, 8) >> 3;
641 
642 	/* attr fork btree root can have at least this many key/ptr pairs */
643 	maxforkoff = XFS_LITINO(mp) - XFS_BMDR_SPACE_CALC(MINABTPTRS);
644 	maxforkoff = maxforkoff >> 3;	/* rounded down */
645 
646 	if (offset >= maxforkoff)
647 		return maxforkoff;
648 	if (offset >= minforkoff)
649 		return offset;
650 	return 0;
651 }
652 
653 /*
654  * Switch on the ATTR2 superblock bit (implies also FEATURES2) unless:
655  * - noattr2 mount option is set,
656  * - on-disk version bit says it is already set, or
657  * - the attr2 mount option is not set to enable automatic upgrade from attr1.
658  */
659 STATIC void
660 xfs_sbversion_add_attr2(
661 	struct xfs_mount	*mp,
662 	struct xfs_trans	*tp)
663 {
664 	if (xfs_has_noattr2(mp))
665 		return;
666 	if (mp->m_sb.sb_features2 & XFS_SB_VERSION2_ATTR2BIT)
667 		return;
668 	if (!xfs_has_attr2(mp))
669 		return;
670 
671 	spin_lock(&mp->m_sb_lock);
672 	xfs_add_attr2(mp);
673 	spin_unlock(&mp->m_sb_lock);
674 	xfs_log_sb(tp);
675 }
676 
677 /*
678  * Create the initial contents of a shortform attribute list.
679  */
680 void
681 xfs_attr_shortform_create(
682 	struct xfs_da_args	*args)
683 {
684 	struct xfs_inode	*dp = args->dp;
685 	struct xfs_ifork	*ifp = &dp->i_af;
686 	struct xfs_attr_sf_hdr	*hdr;
687 
688 	trace_xfs_attr_sf_create(args);
689 
690 	ASSERT(ifp->if_bytes == 0);
691 	if (ifp->if_format == XFS_DINODE_FMT_EXTENTS)
692 		ifp->if_format = XFS_DINODE_FMT_LOCAL;
693 	xfs_idata_realloc(dp, sizeof(*hdr), XFS_ATTR_FORK);
694 	hdr = (struct xfs_attr_sf_hdr *)ifp->if_u1.if_data;
695 	memset(hdr, 0, sizeof(*hdr));
696 	hdr->totsize = cpu_to_be16(sizeof(*hdr));
697 	xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
698 }
699 
700 /*
701  * Return -EEXIST if attr is found, or -ENOATTR if not
702  * args:  args containing attribute name and namelen
703  * sfep:  If not null, pointer will be set to the last attr entry found on
704 	  -EEXIST.  On -ENOATTR pointer is left at the last entry in the list
705  * basep: If not null, pointer is set to the byte offset of the entry in the
706  *	  list on -EEXIST.  On -ENOATTR, pointer is left at the byte offset of
707  *	  the last entry in the list
708  */
709 int
710 xfs_attr_sf_findname(
711 	struct xfs_da_args	 *args,
712 	struct xfs_attr_sf_entry **sfep,
713 	unsigned int		 *basep)
714 {
715 	struct xfs_attr_shortform *sf;
716 	struct xfs_attr_sf_entry *sfe;
717 	unsigned int		base = sizeof(struct xfs_attr_sf_hdr);
718 	int			size = 0;
719 	int			end;
720 	int			i;
721 
722 	sf = (struct xfs_attr_shortform *)args->dp->i_af.if_u1.if_data;
723 	sfe = &sf->list[0];
724 	end = sf->hdr.count;
725 	for (i = 0; i < end; sfe = xfs_attr_sf_nextentry(sfe),
726 			     base += size, i++) {
727 		size = xfs_attr_sf_entsize(sfe);
728 		if (!xfs_attr_match(args, sfe->namelen, sfe->nameval,
729 				    sfe->flags))
730 			continue;
731 		break;
732 	}
733 
734 	if (sfep != NULL)
735 		*sfep = sfe;
736 
737 	if (basep != NULL)
738 		*basep = base;
739 
740 	if (i == end)
741 		return -ENOATTR;
742 	return -EEXIST;
743 }
744 
745 /*
746  * Add a name/value pair to the shortform attribute list.
747  * Overflow from the inode has already been checked for.
748  */
749 void
750 xfs_attr_shortform_add(
751 	struct xfs_da_args		*args,
752 	int				forkoff)
753 {
754 	struct xfs_attr_shortform	*sf;
755 	struct xfs_attr_sf_entry	*sfe;
756 	int				offset, size;
757 	struct xfs_mount		*mp;
758 	struct xfs_inode		*dp;
759 	struct xfs_ifork		*ifp;
760 
761 	trace_xfs_attr_sf_add(args);
762 
763 	dp = args->dp;
764 	mp = dp->i_mount;
765 	dp->i_forkoff = forkoff;
766 
767 	ifp = &dp->i_af;
768 	ASSERT(ifp->if_format == XFS_DINODE_FMT_LOCAL);
769 	sf = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
770 	if (xfs_attr_sf_findname(args, &sfe, NULL) == -EEXIST)
771 		ASSERT(0);
772 
773 	offset = (char *)sfe - (char *)sf;
774 	size = xfs_attr_sf_entsize_byname(args->namelen, args->valuelen);
775 	xfs_idata_realloc(dp, size, XFS_ATTR_FORK);
776 	sf = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
777 	sfe = (struct xfs_attr_sf_entry *)((char *)sf + offset);
778 
779 	sfe->namelen = args->namelen;
780 	sfe->valuelen = args->valuelen;
781 	sfe->flags = args->attr_filter;
782 	memcpy(sfe->nameval, args->name, args->namelen);
783 	memcpy(&sfe->nameval[args->namelen], args->value, args->valuelen);
784 	sf->hdr.count++;
785 	be16_add_cpu(&sf->hdr.totsize, size);
786 	xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
787 
788 	xfs_sbversion_add_attr2(mp, args->trans);
789 }
790 
791 /*
792  * After the last attribute is removed revert to original inode format,
793  * making all literal area available to the data fork once more.
794  */
795 void
796 xfs_attr_fork_remove(
797 	struct xfs_inode	*ip,
798 	struct xfs_trans	*tp)
799 {
800 	ASSERT(ip->i_af.if_nextents == 0);
801 
802 	xfs_ifork_zap_attr(ip);
803 	ip->i_forkoff = 0;
804 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
805 }
806 
807 /*
808  * Remove an attribute from the shortform attribute list structure.
809  */
810 int
811 xfs_attr_sf_removename(
812 	struct xfs_da_args		*args)
813 {
814 	struct xfs_attr_shortform	*sf;
815 	struct xfs_attr_sf_entry	*sfe;
816 	int				size = 0, end, totsize;
817 	unsigned int			base;
818 	struct xfs_mount		*mp;
819 	struct xfs_inode		*dp;
820 	int				error;
821 
822 	trace_xfs_attr_sf_remove(args);
823 
824 	dp = args->dp;
825 	mp = dp->i_mount;
826 	sf = (struct xfs_attr_shortform *)dp->i_af.if_u1.if_data;
827 
828 	error = xfs_attr_sf_findname(args, &sfe, &base);
829 
830 	/*
831 	 * If we are recovering an operation, finding nothing to
832 	 * remove is not an error - it just means there was nothing
833 	 * to clean up.
834 	 */
835 	if (error == -ENOATTR && (args->op_flags & XFS_DA_OP_RECOVERY))
836 		return 0;
837 	if (error != -EEXIST)
838 		return error;
839 	size = xfs_attr_sf_entsize(sfe);
840 
841 	/*
842 	 * Fix up the attribute fork data, covering the hole
843 	 */
844 	end = base + size;
845 	totsize = be16_to_cpu(sf->hdr.totsize);
846 	if (end != totsize)
847 		memmove(&((char *)sf)[base], &((char *)sf)[end], totsize - end);
848 	sf->hdr.count--;
849 	be16_add_cpu(&sf->hdr.totsize, -size);
850 
851 	/*
852 	 * Fix up the start offset of the attribute fork
853 	 */
854 	totsize -= size;
855 	if (totsize == sizeof(xfs_attr_sf_hdr_t) && xfs_has_attr2(mp) &&
856 	    (dp->i_df.if_format != XFS_DINODE_FMT_BTREE) &&
857 	    !(args->op_flags & (XFS_DA_OP_ADDNAME | XFS_DA_OP_REPLACE))) {
858 		xfs_attr_fork_remove(dp, args->trans);
859 	} else {
860 		xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
861 		dp->i_forkoff = xfs_attr_shortform_bytesfit(dp, totsize);
862 		ASSERT(dp->i_forkoff);
863 		ASSERT(totsize > sizeof(xfs_attr_sf_hdr_t) ||
864 				(args->op_flags & XFS_DA_OP_ADDNAME) ||
865 				!xfs_has_attr2(mp) ||
866 				dp->i_df.if_format == XFS_DINODE_FMT_BTREE);
867 		xfs_trans_log_inode(args->trans, dp,
868 					XFS_ILOG_CORE | XFS_ILOG_ADATA);
869 	}
870 
871 	xfs_sbversion_add_attr2(mp, args->trans);
872 
873 	return 0;
874 }
875 
876 /*
877  * Look up a name in a shortform attribute list structure.
878  */
879 /*ARGSUSED*/
880 int
881 xfs_attr_shortform_lookup(xfs_da_args_t *args)
882 {
883 	struct xfs_attr_shortform *sf;
884 	struct xfs_attr_sf_entry *sfe;
885 	int i;
886 	struct xfs_ifork *ifp;
887 
888 	trace_xfs_attr_sf_lookup(args);
889 
890 	ifp = &args->dp->i_af;
891 	ASSERT(ifp->if_format == XFS_DINODE_FMT_LOCAL);
892 	sf = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
893 	sfe = &sf->list[0];
894 	for (i = 0; i < sf->hdr.count;
895 				sfe = xfs_attr_sf_nextentry(sfe), i++) {
896 		if (xfs_attr_match(args, sfe->namelen, sfe->nameval,
897 				sfe->flags))
898 			return -EEXIST;
899 	}
900 	return -ENOATTR;
901 }
902 
903 /*
904  * Retrieve the attribute value and length.
905  *
906  * If args->valuelen is zero, only the length needs to be returned.  Unlike a
907  * lookup, we only return an error if the attribute does not exist or we can't
908  * retrieve the value.
909  */
910 int
911 xfs_attr_shortform_getvalue(
912 	struct xfs_da_args	*args)
913 {
914 	struct xfs_attr_shortform *sf;
915 	struct xfs_attr_sf_entry *sfe;
916 	int			i;
917 
918 	ASSERT(args->dp->i_af.if_format == XFS_DINODE_FMT_LOCAL);
919 	sf = (struct xfs_attr_shortform *)args->dp->i_af.if_u1.if_data;
920 	sfe = &sf->list[0];
921 	for (i = 0; i < sf->hdr.count;
922 				sfe = xfs_attr_sf_nextentry(sfe), i++) {
923 		if (xfs_attr_match(args, sfe->namelen, sfe->nameval,
924 				sfe->flags))
925 			return xfs_attr_copy_value(args,
926 				&sfe->nameval[args->namelen], sfe->valuelen);
927 	}
928 	return -ENOATTR;
929 }
930 
931 /* Convert from using the shortform to the leaf format. */
932 int
933 xfs_attr_shortform_to_leaf(
934 	struct xfs_da_args		*args)
935 {
936 	struct xfs_inode		*dp;
937 	struct xfs_attr_shortform	*sf;
938 	struct xfs_attr_sf_entry	*sfe;
939 	struct xfs_da_args		nargs;
940 	char				*tmpbuffer;
941 	int				error, i, size;
942 	xfs_dablk_t			blkno;
943 	struct xfs_buf			*bp;
944 	struct xfs_ifork		*ifp;
945 
946 	trace_xfs_attr_sf_to_leaf(args);
947 
948 	dp = args->dp;
949 	ifp = &dp->i_af;
950 	sf = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
951 	size = be16_to_cpu(sf->hdr.totsize);
952 	tmpbuffer = kmem_alloc(size, 0);
953 	ASSERT(tmpbuffer != NULL);
954 	memcpy(tmpbuffer, ifp->if_u1.if_data, size);
955 	sf = (struct xfs_attr_shortform *)tmpbuffer;
956 
957 	xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
958 	xfs_bmap_local_to_extents_empty(args->trans, dp, XFS_ATTR_FORK);
959 
960 	bp = NULL;
961 	error = xfs_da_grow_inode(args, &blkno);
962 	if (error)
963 		goto out;
964 
965 	ASSERT(blkno == 0);
966 	error = xfs_attr3_leaf_create(args, blkno, &bp);
967 	if (error)
968 		goto out;
969 
970 	memset((char *)&nargs, 0, sizeof(nargs));
971 	nargs.dp = dp;
972 	nargs.geo = args->geo;
973 	nargs.total = args->total;
974 	nargs.whichfork = XFS_ATTR_FORK;
975 	nargs.trans = args->trans;
976 	nargs.op_flags = XFS_DA_OP_OKNOENT;
977 
978 	sfe = &sf->list[0];
979 	for (i = 0; i < sf->hdr.count; i++) {
980 		nargs.name = sfe->nameval;
981 		nargs.namelen = sfe->namelen;
982 		nargs.value = &sfe->nameval[nargs.namelen];
983 		nargs.valuelen = sfe->valuelen;
984 		nargs.hashval = xfs_da_hashname(sfe->nameval,
985 						sfe->namelen);
986 		nargs.attr_filter = sfe->flags & XFS_ATTR_NSP_ONDISK_MASK;
987 		error = xfs_attr3_leaf_lookup_int(bp, &nargs); /* set a->index */
988 		ASSERT(error == -ENOATTR);
989 		error = xfs_attr3_leaf_add(bp, &nargs);
990 		ASSERT(error != -ENOSPC);
991 		if (error)
992 			goto out;
993 		sfe = xfs_attr_sf_nextentry(sfe);
994 	}
995 	error = 0;
996 out:
997 	kmem_free(tmpbuffer);
998 	return error;
999 }
1000 
1001 /*
1002  * Check a leaf attribute block to see if all the entries would fit into
1003  * a shortform attribute list.
1004  */
1005 int
1006 xfs_attr_shortform_allfit(
1007 	struct xfs_buf		*bp,
1008 	struct xfs_inode	*dp)
1009 {
1010 	struct xfs_attr_leafblock *leaf;
1011 	struct xfs_attr_leaf_entry *entry;
1012 	xfs_attr_leaf_name_local_t *name_loc;
1013 	struct xfs_attr3_icleaf_hdr leafhdr;
1014 	int			bytes;
1015 	int			i;
1016 	struct xfs_mount	*mp = bp->b_mount;
1017 
1018 	leaf = bp->b_addr;
1019 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf);
1020 	entry = xfs_attr3_leaf_entryp(leaf);
1021 
1022 	bytes = sizeof(struct xfs_attr_sf_hdr);
1023 	for (i = 0; i < leafhdr.count; entry++, i++) {
1024 		if (entry->flags & XFS_ATTR_INCOMPLETE)
1025 			continue;		/* don't copy partial entries */
1026 		if (!(entry->flags & XFS_ATTR_LOCAL))
1027 			return 0;
1028 		name_loc = xfs_attr3_leaf_name_local(leaf, i);
1029 		if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX)
1030 			return 0;
1031 		if (be16_to_cpu(name_loc->valuelen) >= XFS_ATTR_SF_ENTSIZE_MAX)
1032 			return 0;
1033 		bytes += xfs_attr_sf_entsize_byname(name_loc->namelen,
1034 					be16_to_cpu(name_loc->valuelen));
1035 	}
1036 	if (xfs_has_attr2(dp->i_mount) &&
1037 	    (dp->i_df.if_format != XFS_DINODE_FMT_BTREE) &&
1038 	    (bytes == sizeof(struct xfs_attr_sf_hdr)))
1039 		return -1;
1040 	return xfs_attr_shortform_bytesfit(dp, bytes);
1041 }
1042 
1043 /* Verify the consistency of an inline attribute fork. */
1044 xfs_failaddr_t
1045 xfs_attr_shortform_verify(
1046 	struct xfs_inode		*ip)
1047 {
1048 	struct xfs_attr_shortform	*sfp;
1049 	struct xfs_attr_sf_entry	*sfep;
1050 	struct xfs_attr_sf_entry	*next_sfep;
1051 	char				*endp;
1052 	struct xfs_ifork		*ifp;
1053 	int				i;
1054 	int64_t				size;
1055 
1056 	ASSERT(ip->i_af.if_format == XFS_DINODE_FMT_LOCAL);
1057 	ifp = xfs_ifork_ptr(ip, XFS_ATTR_FORK);
1058 	sfp = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
1059 	size = ifp->if_bytes;
1060 
1061 	/*
1062 	 * Give up if the attribute is way too short.
1063 	 */
1064 	if (size < sizeof(struct xfs_attr_sf_hdr))
1065 		return __this_address;
1066 
1067 	endp = (char *)sfp + size;
1068 
1069 	/* Check all reported entries */
1070 	sfep = &sfp->list[0];
1071 	for (i = 0; i < sfp->hdr.count; i++) {
1072 		/*
1073 		 * struct xfs_attr_sf_entry has a variable length.
1074 		 * Check the fixed-offset parts of the structure are
1075 		 * within the data buffer.
1076 		 * xfs_attr_sf_entry is defined with a 1-byte variable
1077 		 * array at the end, so we must subtract that off.
1078 		 */
1079 		if (((char *)sfep + sizeof(*sfep)) >= endp)
1080 			return __this_address;
1081 
1082 		/* Don't allow names with known bad length. */
1083 		if (sfep->namelen == 0)
1084 			return __this_address;
1085 
1086 		/*
1087 		 * Check that the variable-length part of the structure is
1088 		 * within the data buffer.  The next entry starts after the
1089 		 * name component, so nextentry is an acceptable test.
1090 		 */
1091 		next_sfep = xfs_attr_sf_nextentry(sfep);
1092 		if ((char *)next_sfep > endp)
1093 			return __this_address;
1094 
1095 		/*
1096 		 * Check for unknown flags.  Short form doesn't support
1097 		 * the incomplete or local bits, so we can use the namespace
1098 		 * mask here.
1099 		 */
1100 		if (sfep->flags & ~XFS_ATTR_NSP_ONDISK_MASK)
1101 			return __this_address;
1102 
1103 		/*
1104 		 * Check for invalid namespace combinations.  We only allow
1105 		 * one namespace flag per xattr, so we can just count the
1106 		 * bits (i.e. hweight) here.
1107 		 */
1108 		if (hweight8(sfep->flags & XFS_ATTR_NSP_ONDISK_MASK) > 1)
1109 			return __this_address;
1110 
1111 		sfep = next_sfep;
1112 	}
1113 	if ((void *)sfep != (void *)endp)
1114 		return __this_address;
1115 
1116 	return NULL;
1117 }
1118 
1119 /*
1120  * Convert a leaf attribute list to shortform attribute list
1121  */
1122 int
1123 xfs_attr3_leaf_to_shortform(
1124 	struct xfs_buf		*bp,
1125 	struct xfs_da_args	*args,
1126 	int			forkoff)
1127 {
1128 	struct xfs_attr_leafblock *leaf;
1129 	struct xfs_attr3_icleaf_hdr ichdr;
1130 	struct xfs_attr_leaf_entry *entry;
1131 	struct xfs_attr_leaf_name_local *name_loc;
1132 	struct xfs_da_args	nargs;
1133 	struct xfs_inode	*dp = args->dp;
1134 	char			*tmpbuffer;
1135 	int			error;
1136 	int			i;
1137 
1138 	trace_xfs_attr_leaf_to_sf(args);
1139 
1140 	tmpbuffer = kmem_alloc(args->geo->blksize, 0);
1141 	if (!tmpbuffer)
1142 		return -ENOMEM;
1143 
1144 	memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1145 
1146 	leaf = (xfs_attr_leafblock_t *)tmpbuffer;
1147 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1148 	entry = xfs_attr3_leaf_entryp(leaf);
1149 
1150 	/* XXX (dgc): buffer is about to be marked stale - why zero it? */
1151 	memset(bp->b_addr, 0, args->geo->blksize);
1152 
1153 	/*
1154 	 * Clean out the prior contents of the attribute list.
1155 	 */
1156 	error = xfs_da_shrink_inode(args, 0, bp);
1157 	if (error)
1158 		goto out;
1159 
1160 	if (forkoff == -1) {
1161 		/*
1162 		 * Don't remove the attr fork if this operation is the first
1163 		 * part of a attr replace operations. We're going to add a new
1164 		 * attr immediately, so we need to keep the attr fork around in
1165 		 * this case.
1166 		 */
1167 		if (!(args->op_flags & XFS_DA_OP_REPLACE)) {
1168 			ASSERT(xfs_has_attr2(dp->i_mount));
1169 			ASSERT(dp->i_df.if_format != XFS_DINODE_FMT_BTREE);
1170 			xfs_attr_fork_remove(dp, args->trans);
1171 		}
1172 		goto out;
1173 	}
1174 
1175 	xfs_attr_shortform_create(args);
1176 
1177 	/*
1178 	 * Copy the attributes
1179 	 */
1180 	memset((char *)&nargs, 0, sizeof(nargs));
1181 	nargs.geo = args->geo;
1182 	nargs.dp = dp;
1183 	nargs.total = args->total;
1184 	nargs.whichfork = XFS_ATTR_FORK;
1185 	nargs.trans = args->trans;
1186 	nargs.op_flags = XFS_DA_OP_OKNOENT;
1187 
1188 	for (i = 0; i < ichdr.count; entry++, i++) {
1189 		if (entry->flags & XFS_ATTR_INCOMPLETE)
1190 			continue;	/* don't copy partial entries */
1191 		if (!entry->nameidx)
1192 			continue;
1193 		ASSERT(entry->flags & XFS_ATTR_LOCAL);
1194 		name_loc = xfs_attr3_leaf_name_local(leaf, i);
1195 		nargs.name = name_loc->nameval;
1196 		nargs.namelen = name_loc->namelen;
1197 		nargs.value = &name_loc->nameval[nargs.namelen];
1198 		nargs.valuelen = be16_to_cpu(name_loc->valuelen);
1199 		nargs.hashval = be32_to_cpu(entry->hashval);
1200 		nargs.attr_filter = entry->flags & XFS_ATTR_NSP_ONDISK_MASK;
1201 		xfs_attr_shortform_add(&nargs, forkoff);
1202 	}
1203 	error = 0;
1204 
1205 out:
1206 	kmem_free(tmpbuffer);
1207 	return error;
1208 }
1209 
1210 /*
1211  * Convert from using a single leaf to a root node and a leaf.
1212  */
1213 int
1214 xfs_attr3_leaf_to_node(
1215 	struct xfs_da_args	*args)
1216 {
1217 	struct xfs_attr_leafblock *leaf;
1218 	struct xfs_attr3_icleaf_hdr icleafhdr;
1219 	struct xfs_attr_leaf_entry *entries;
1220 	struct xfs_da3_icnode_hdr icnodehdr;
1221 	struct xfs_da_intnode	*node;
1222 	struct xfs_inode	*dp = args->dp;
1223 	struct xfs_mount	*mp = dp->i_mount;
1224 	struct xfs_buf		*bp1 = NULL;
1225 	struct xfs_buf		*bp2 = NULL;
1226 	xfs_dablk_t		blkno;
1227 	int			error;
1228 
1229 	trace_xfs_attr_leaf_to_node(args);
1230 
1231 	if (XFS_TEST_ERROR(false, mp, XFS_ERRTAG_ATTR_LEAF_TO_NODE)) {
1232 		error = -EIO;
1233 		goto out;
1234 	}
1235 
1236 	error = xfs_da_grow_inode(args, &blkno);
1237 	if (error)
1238 		goto out;
1239 	error = xfs_attr3_leaf_read(args->trans, dp, 0, &bp1);
1240 	if (error)
1241 		goto out;
1242 
1243 	error = xfs_da_get_buf(args->trans, dp, blkno, &bp2, XFS_ATTR_FORK);
1244 	if (error)
1245 		goto out;
1246 
1247 	/* copy leaf to new buffer, update identifiers */
1248 	xfs_trans_buf_set_type(args->trans, bp2, XFS_BLFT_ATTR_LEAF_BUF);
1249 	bp2->b_ops = bp1->b_ops;
1250 	memcpy(bp2->b_addr, bp1->b_addr, args->geo->blksize);
1251 	if (xfs_has_crc(mp)) {
1252 		struct xfs_da3_blkinfo *hdr3 = bp2->b_addr;
1253 		hdr3->blkno = cpu_to_be64(xfs_buf_daddr(bp2));
1254 	}
1255 	xfs_trans_log_buf(args->trans, bp2, 0, args->geo->blksize - 1);
1256 
1257 	/*
1258 	 * Set up the new root node.
1259 	 */
1260 	error = xfs_da3_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK);
1261 	if (error)
1262 		goto out;
1263 	node = bp1->b_addr;
1264 	xfs_da3_node_hdr_from_disk(mp, &icnodehdr, node);
1265 
1266 	leaf = bp2->b_addr;
1267 	xfs_attr3_leaf_hdr_from_disk(args->geo, &icleafhdr, leaf);
1268 	entries = xfs_attr3_leaf_entryp(leaf);
1269 
1270 	/* both on-disk, don't endian-flip twice */
1271 	icnodehdr.btree[0].hashval = entries[icleafhdr.count - 1].hashval;
1272 	icnodehdr.btree[0].before = cpu_to_be32(blkno);
1273 	icnodehdr.count = 1;
1274 	xfs_da3_node_hdr_to_disk(dp->i_mount, node, &icnodehdr);
1275 	xfs_trans_log_buf(args->trans, bp1, 0, args->geo->blksize - 1);
1276 	error = 0;
1277 out:
1278 	return error;
1279 }
1280 
1281 /*========================================================================
1282  * Routines used for growing the Btree.
1283  *========================================================================*/
1284 
1285 /*
1286  * Create the initial contents of a leaf attribute list
1287  * or a leaf in a node attribute list.
1288  */
1289 STATIC int
1290 xfs_attr3_leaf_create(
1291 	struct xfs_da_args	*args,
1292 	xfs_dablk_t		blkno,
1293 	struct xfs_buf		**bpp)
1294 {
1295 	struct xfs_attr_leafblock *leaf;
1296 	struct xfs_attr3_icleaf_hdr ichdr;
1297 	struct xfs_inode	*dp = args->dp;
1298 	struct xfs_mount	*mp = dp->i_mount;
1299 	struct xfs_buf		*bp;
1300 	int			error;
1301 
1302 	trace_xfs_attr_leaf_create(args);
1303 
1304 	error = xfs_da_get_buf(args->trans, args->dp, blkno, &bp,
1305 					    XFS_ATTR_FORK);
1306 	if (error)
1307 		return error;
1308 	bp->b_ops = &xfs_attr3_leaf_buf_ops;
1309 	xfs_trans_buf_set_type(args->trans, bp, XFS_BLFT_ATTR_LEAF_BUF);
1310 	leaf = bp->b_addr;
1311 	memset(leaf, 0, args->geo->blksize);
1312 
1313 	memset(&ichdr, 0, sizeof(ichdr));
1314 	ichdr.firstused = args->geo->blksize;
1315 
1316 	if (xfs_has_crc(mp)) {
1317 		struct xfs_da3_blkinfo *hdr3 = bp->b_addr;
1318 
1319 		ichdr.magic = XFS_ATTR3_LEAF_MAGIC;
1320 
1321 		hdr3->blkno = cpu_to_be64(xfs_buf_daddr(bp));
1322 		hdr3->owner = cpu_to_be64(dp->i_ino);
1323 		uuid_copy(&hdr3->uuid, &mp->m_sb.sb_meta_uuid);
1324 
1325 		ichdr.freemap[0].base = sizeof(struct xfs_attr3_leaf_hdr);
1326 	} else {
1327 		ichdr.magic = XFS_ATTR_LEAF_MAGIC;
1328 		ichdr.freemap[0].base = sizeof(struct xfs_attr_leaf_hdr);
1329 	}
1330 	ichdr.freemap[0].size = ichdr.firstused - ichdr.freemap[0].base;
1331 
1332 	xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1333 	xfs_trans_log_buf(args->trans, bp, 0, args->geo->blksize - 1);
1334 
1335 	*bpp = bp;
1336 	return 0;
1337 }
1338 
1339 /*
1340  * Split the leaf node, rebalance, then add the new entry.
1341  */
1342 int
1343 xfs_attr3_leaf_split(
1344 	struct xfs_da_state	*state,
1345 	struct xfs_da_state_blk	*oldblk,
1346 	struct xfs_da_state_blk	*newblk)
1347 {
1348 	xfs_dablk_t blkno;
1349 	int error;
1350 
1351 	trace_xfs_attr_leaf_split(state->args);
1352 
1353 	/*
1354 	 * Allocate space for a new leaf node.
1355 	 */
1356 	ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
1357 	error = xfs_da_grow_inode(state->args, &blkno);
1358 	if (error)
1359 		return error;
1360 	error = xfs_attr3_leaf_create(state->args, blkno, &newblk->bp);
1361 	if (error)
1362 		return error;
1363 	newblk->blkno = blkno;
1364 	newblk->magic = XFS_ATTR_LEAF_MAGIC;
1365 
1366 	/*
1367 	 * Rebalance the entries across the two leaves.
1368 	 * NOTE: rebalance() currently depends on the 2nd block being empty.
1369 	 */
1370 	xfs_attr3_leaf_rebalance(state, oldblk, newblk);
1371 	error = xfs_da3_blk_link(state, oldblk, newblk);
1372 	if (error)
1373 		return error;
1374 
1375 	/*
1376 	 * Save info on "old" attribute for "atomic rename" ops, leaf_add()
1377 	 * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
1378 	 * "new" attrs info.  Will need the "old" info to remove it later.
1379 	 *
1380 	 * Insert the "new" entry in the correct block.
1381 	 */
1382 	if (state->inleaf) {
1383 		trace_xfs_attr_leaf_add_old(state->args);
1384 		error = xfs_attr3_leaf_add(oldblk->bp, state->args);
1385 	} else {
1386 		trace_xfs_attr_leaf_add_new(state->args);
1387 		error = xfs_attr3_leaf_add(newblk->bp, state->args);
1388 	}
1389 
1390 	/*
1391 	 * Update last hashval in each block since we added the name.
1392 	 */
1393 	oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
1394 	newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
1395 	return error;
1396 }
1397 
1398 /*
1399  * Add a name to the leaf attribute list structure.
1400  */
1401 int
1402 xfs_attr3_leaf_add(
1403 	struct xfs_buf		*bp,
1404 	struct xfs_da_args	*args)
1405 {
1406 	struct xfs_attr_leafblock *leaf;
1407 	struct xfs_attr3_icleaf_hdr ichdr;
1408 	int			tablesize;
1409 	int			entsize;
1410 	int			sum;
1411 	int			tmp;
1412 	int			i;
1413 
1414 	trace_xfs_attr_leaf_add(args);
1415 
1416 	leaf = bp->b_addr;
1417 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1418 	ASSERT(args->index >= 0 && args->index <= ichdr.count);
1419 	entsize = xfs_attr_leaf_newentsize(args, NULL);
1420 
1421 	/*
1422 	 * Search through freemap for first-fit on new name length.
1423 	 * (may need to figure in size of entry struct too)
1424 	 */
1425 	tablesize = (ichdr.count + 1) * sizeof(xfs_attr_leaf_entry_t)
1426 					+ xfs_attr3_leaf_hdr_size(leaf);
1427 	for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE - 1; i >= 0; i--) {
1428 		if (tablesize > ichdr.firstused) {
1429 			sum += ichdr.freemap[i].size;
1430 			continue;
1431 		}
1432 		if (!ichdr.freemap[i].size)
1433 			continue;	/* no space in this map */
1434 		tmp = entsize;
1435 		if (ichdr.freemap[i].base < ichdr.firstused)
1436 			tmp += sizeof(xfs_attr_leaf_entry_t);
1437 		if (ichdr.freemap[i].size >= tmp) {
1438 			tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, i);
1439 			goto out_log_hdr;
1440 		}
1441 		sum += ichdr.freemap[i].size;
1442 	}
1443 
1444 	/*
1445 	 * If there are no holes in the address space of the block,
1446 	 * and we don't have enough freespace, then compaction will do us
1447 	 * no good and we should just give up.
1448 	 */
1449 	if (!ichdr.holes && sum < entsize)
1450 		return -ENOSPC;
1451 
1452 	/*
1453 	 * Compact the entries to coalesce free space.
1454 	 * This may change the hdr->count via dropping INCOMPLETE entries.
1455 	 */
1456 	xfs_attr3_leaf_compact(args, &ichdr, bp);
1457 
1458 	/*
1459 	 * After compaction, the block is guaranteed to have only one
1460 	 * free region, in freemap[0].  If it is not big enough, give up.
1461 	 */
1462 	if (ichdr.freemap[0].size < (entsize + sizeof(xfs_attr_leaf_entry_t))) {
1463 		tmp = -ENOSPC;
1464 		goto out_log_hdr;
1465 	}
1466 
1467 	tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, 0);
1468 
1469 out_log_hdr:
1470 	xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1471 	xfs_trans_log_buf(args->trans, bp,
1472 		XFS_DA_LOGRANGE(leaf, &leaf->hdr,
1473 				xfs_attr3_leaf_hdr_size(leaf)));
1474 	return tmp;
1475 }
1476 
1477 /*
1478  * Add a name to a leaf attribute list structure.
1479  */
1480 STATIC int
1481 xfs_attr3_leaf_add_work(
1482 	struct xfs_buf		*bp,
1483 	struct xfs_attr3_icleaf_hdr *ichdr,
1484 	struct xfs_da_args	*args,
1485 	int			mapindex)
1486 {
1487 	struct xfs_attr_leafblock *leaf;
1488 	struct xfs_attr_leaf_entry *entry;
1489 	struct xfs_attr_leaf_name_local *name_loc;
1490 	struct xfs_attr_leaf_name_remote *name_rmt;
1491 	struct xfs_mount	*mp;
1492 	int			tmp;
1493 	int			i;
1494 
1495 	trace_xfs_attr_leaf_add_work(args);
1496 
1497 	leaf = bp->b_addr;
1498 	ASSERT(mapindex >= 0 && mapindex < XFS_ATTR_LEAF_MAPSIZE);
1499 	ASSERT(args->index >= 0 && args->index <= ichdr->count);
1500 
1501 	/*
1502 	 * Force open some space in the entry array and fill it in.
1503 	 */
1504 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1505 	if (args->index < ichdr->count) {
1506 		tmp  = ichdr->count - args->index;
1507 		tmp *= sizeof(xfs_attr_leaf_entry_t);
1508 		memmove(entry + 1, entry, tmp);
1509 		xfs_trans_log_buf(args->trans, bp,
1510 		    XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1511 	}
1512 	ichdr->count++;
1513 
1514 	/*
1515 	 * Allocate space for the new string (at the end of the run).
1516 	 */
1517 	mp = args->trans->t_mountp;
1518 	ASSERT(ichdr->freemap[mapindex].base < args->geo->blksize);
1519 	ASSERT((ichdr->freemap[mapindex].base & 0x3) == 0);
1520 	ASSERT(ichdr->freemap[mapindex].size >=
1521 		xfs_attr_leaf_newentsize(args, NULL));
1522 	ASSERT(ichdr->freemap[mapindex].size < args->geo->blksize);
1523 	ASSERT((ichdr->freemap[mapindex].size & 0x3) == 0);
1524 
1525 	ichdr->freemap[mapindex].size -= xfs_attr_leaf_newentsize(args, &tmp);
1526 
1527 	entry->nameidx = cpu_to_be16(ichdr->freemap[mapindex].base +
1528 				     ichdr->freemap[mapindex].size);
1529 	entry->hashval = cpu_to_be32(args->hashval);
1530 	entry->flags = args->attr_filter;
1531 	if (tmp)
1532 		entry->flags |= XFS_ATTR_LOCAL;
1533 	if (args->op_flags & XFS_DA_OP_REPLACE) {
1534 		if (!(args->op_flags & XFS_DA_OP_LOGGED))
1535 			entry->flags |= XFS_ATTR_INCOMPLETE;
1536 		if ((args->blkno2 == args->blkno) &&
1537 		    (args->index2 <= args->index)) {
1538 			args->index2++;
1539 		}
1540 	}
1541 	xfs_trans_log_buf(args->trans, bp,
1542 			  XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1543 	ASSERT((args->index == 0) ||
1544 	       (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1545 	ASSERT((args->index == ichdr->count - 1) ||
1546 	       (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1547 
1548 	/*
1549 	 * For "remote" attribute values, simply note that we need to
1550 	 * allocate space for the "remote" value.  We can't actually
1551 	 * allocate the extents in this transaction, and we can't decide
1552 	 * which blocks they should be as we might allocate more blocks
1553 	 * as part of this transaction (a split operation for example).
1554 	 */
1555 	if (entry->flags & XFS_ATTR_LOCAL) {
1556 		name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
1557 		name_loc->namelen = args->namelen;
1558 		name_loc->valuelen = cpu_to_be16(args->valuelen);
1559 		memcpy((char *)name_loc->nameval, args->name, args->namelen);
1560 		memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1561 				   be16_to_cpu(name_loc->valuelen));
1562 	} else {
1563 		name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
1564 		name_rmt->namelen = args->namelen;
1565 		memcpy((char *)name_rmt->name, args->name, args->namelen);
1566 		entry->flags |= XFS_ATTR_INCOMPLETE;
1567 		/* just in case */
1568 		name_rmt->valuelen = 0;
1569 		name_rmt->valueblk = 0;
1570 		args->rmtblkno = 1;
1571 		args->rmtblkcnt = xfs_attr3_rmt_blocks(mp, args->valuelen);
1572 		args->rmtvaluelen = args->valuelen;
1573 	}
1574 	xfs_trans_log_buf(args->trans, bp,
1575 	     XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1576 				   xfs_attr_leaf_entsize(leaf, args->index)));
1577 
1578 	/*
1579 	 * Update the control info for this leaf node
1580 	 */
1581 	if (be16_to_cpu(entry->nameidx) < ichdr->firstused)
1582 		ichdr->firstused = be16_to_cpu(entry->nameidx);
1583 
1584 	ASSERT(ichdr->firstused >= ichdr->count * sizeof(xfs_attr_leaf_entry_t)
1585 					+ xfs_attr3_leaf_hdr_size(leaf));
1586 	tmp = (ichdr->count - 1) * sizeof(xfs_attr_leaf_entry_t)
1587 					+ xfs_attr3_leaf_hdr_size(leaf);
1588 
1589 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1590 		if (ichdr->freemap[i].base == tmp) {
1591 			ichdr->freemap[i].base += sizeof(xfs_attr_leaf_entry_t);
1592 			ichdr->freemap[i].size -=
1593 				min_t(uint16_t, ichdr->freemap[i].size,
1594 						sizeof(xfs_attr_leaf_entry_t));
1595 		}
1596 	}
1597 	ichdr->usedbytes += xfs_attr_leaf_entsize(leaf, args->index);
1598 	return 0;
1599 }
1600 
1601 /*
1602  * Garbage collect a leaf attribute list block by copying it to a new buffer.
1603  */
1604 STATIC void
1605 xfs_attr3_leaf_compact(
1606 	struct xfs_da_args	*args,
1607 	struct xfs_attr3_icleaf_hdr *ichdr_dst,
1608 	struct xfs_buf		*bp)
1609 {
1610 	struct xfs_attr_leafblock *leaf_src;
1611 	struct xfs_attr_leafblock *leaf_dst;
1612 	struct xfs_attr3_icleaf_hdr ichdr_src;
1613 	struct xfs_trans	*trans = args->trans;
1614 	char			*tmpbuffer;
1615 
1616 	trace_xfs_attr_leaf_compact(args);
1617 
1618 	tmpbuffer = kmem_alloc(args->geo->blksize, 0);
1619 	memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1620 	memset(bp->b_addr, 0, args->geo->blksize);
1621 	leaf_src = (xfs_attr_leafblock_t *)tmpbuffer;
1622 	leaf_dst = bp->b_addr;
1623 
1624 	/*
1625 	 * Copy the on-disk header back into the destination buffer to ensure
1626 	 * all the information in the header that is not part of the incore
1627 	 * header structure is preserved.
1628 	 */
1629 	memcpy(bp->b_addr, tmpbuffer, xfs_attr3_leaf_hdr_size(leaf_src));
1630 
1631 	/* Initialise the incore headers */
1632 	ichdr_src = *ichdr_dst;	/* struct copy */
1633 	ichdr_dst->firstused = args->geo->blksize;
1634 	ichdr_dst->usedbytes = 0;
1635 	ichdr_dst->count = 0;
1636 	ichdr_dst->holes = 0;
1637 	ichdr_dst->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_src);
1638 	ichdr_dst->freemap[0].size = ichdr_dst->firstused -
1639 						ichdr_dst->freemap[0].base;
1640 
1641 	/* write the header back to initialise the underlying buffer */
1642 	xfs_attr3_leaf_hdr_to_disk(args->geo, leaf_dst, ichdr_dst);
1643 
1644 	/*
1645 	 * Copy all entry's in the same (sorted) order,
1646 	 * but allocate name/value pairs packed and in sequence.
1647 	 */
1648 	xfs_attr3_leaf_moveents(args, leaf_src, &ichdr_src, 0,
1649 				leaf_dst, ichdr_dst, 0, ichdr_src.count);
1650 	/*
1651 	 * this logs the entire buffer, but the caller must write the header
1652 	 * back to the buffer when it is finished modifying it.
1653 	 */
1654 	xfs_trans_log_buf(trans, bp, 0, args->geo->blksize - 1);
1655 
1656 	kmem_free(tmpbuffer);
1657 }
1658 
1659 /*
1660  * Compare two leaf blocks "order".
1661  * Return 0 unless leaf2 should go before leaf1.
1662  */
1663 static int
1664 xfs_attr3_leaf_order(
1665 	struct xfs_buf	*leaf1_bp,
1666 	struct xfs_attr3_icleaf_hdr *leaf1hdr,
1667 	struct xfs_buf	*leaf2_bp,
1668 	struct xfs_attr3_icleaf_hdr *leaf2hdr)
1669 {
1670 	struct xfs_attr_leaf_entry *entries1;
1671 	struct xfs_attr_leaf_entry *entries2;
1672 
1673 	entries1 = xfs_attr3_leaf_entryp(leaf1_bp->b_addr);
1674 	entries2 = xfs_attr3_leaf_entryp(leaf2_bp->b_addr);
1675 	if (leaf1hdr->count > 0 && leaf2hdr->count > 0 &&
1676 	    ((be32_to_cpu(entries2[0].hashval) <
1677 	      be32_to_cpu(entries1[0].hashval)) ||
1678 	     (be32_to_cpu(entries2[leaf2hdr->count - 1].hashval) <
1679 	      be32_to_cpu(entries1[leaf1hdr->count - 1].hashval)))) {
1680 		return 1;
1681 	}
1682 	return 0;
1683 }
1684 
1685 int
1686 xfs_attr_leaf_order(
1687 	struct xfs_buf	*leaf1_bp,
1688 	struct xfs_buf	*leaf2_bp)
1689 {
1690 	struct xfs_attr3_icleaf_hdr ichdr1;
1691 	struct xfs_attr3_icleaf_hdr ichdr2;
1692 	struct xfs_mount *mp = leaf1_bp->b_mount;
1693 
1694 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr1, leaf1_bp->b_addr);
1695 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr2, leaf2_bp->b_addr);
1696 	return xfs_attr3_leaf_order(leaf1_bp, &ichdr1, leaf2_bp, &ichdr2);
1697 }
1698 
1699 /*
1700  * Redistribute the attribute list entries between two leaf nodes,
1701  * taking into account the size of the new entry.
1702  *
1703  * NOTE: if new block is empty, then it will get the upper half of the
1704  * old block.  At present, all (one) callers pass in an empty second block.
1705  *
1706  * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1707  * to match what it is doing in splitting the attribute leaf block.  Those
1708  * values are used in "atomic rename" operations on attributes.  Note that
1709  * the "new" and "old" values can end up in different blocks.
1710  */
1711 STATIC void
1712 xfs_attr3_leaf_rebalance(
1713 	struct xfs_da_state	*state,
1714 	struct xfs_da_state_blk	*blk1,
1715 	struct xfs_da_state_blk	*blk2)
1716 {
1717 	struct xfs_da_args	*args;
1718 	struct xfs_attr_leafblock *leaf1;
1719 	struct xfs_attr_leafblock *leaf2;
1720 	struct xfs_attr3_icleaf_hdr ichdr1;
1721 	struct xfs_attr3_icleaf_hdr ichdr2;
1722 	struct xfs_attr_leaf_entry *entries1;
1723 	struct xfs_attr_leaf_entry *entries2;
1724 	int			count;
1725 	int			totallen;
1726 	int			max;
1727 	int			space;
1728 	int			swap;
1729 
1730 	/*
1731 	 * Set up environment.
1732 	 */
1733 	ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1734 	ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1735 	leaf1 = blk1->bp->b_addr;
1736 	leaf2 = blk2->bp->b_addr;
1737 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr1, leaf1);
1738 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, leaf2);
1739 	ASSERT(ichdr2.count == 0);
1740 	args = state->args;
1741 
1742 	trace_xfs_attr_leaf_rebalance(args);
1743 
1744 	/*
1745 	 * Check ordering of blocks, reverse if it makes things simpler.
1746 	 *
1747 	 * NOTE: Given that all (current) callers pass in an empty
1748 	 * second block, this code should never set "swap".
1749 	 */
1750 	swap = 0;
1751 	if (xfs_attr3_leaf_order(blk1->bp, &ichdr1, blk2->bp, &ichdr2)) {
1752 		swap(blk1, blk2);
1753 
1754 		/* swap structures rather than reconverting them */
1755 		swap(ichdr1, ichdr2);
1756 
1757 		leaf1 = blk1->bp->b_addr;
1758 		leaf2 = blk2->bp->b_addr;
1759 		swap = 1;
1760 	}
1761 
1762 	/*
1763 	 * Examine entries until we reduce the absolute difference in
1764 	 * byte usage between the two blocks to a minimum.  Then get
1765 	 * the direction to copy and the number of elements to move.
1766 	 *
1767 	 * "inleaf" is true if the new entry should be inserted into blk1.
1768 	 * If "swap" is also true, then reverse the sense of "inleaf".
1769 	 */
1770 	state->inleaf = xfs_attr3_leaf_figure_balance(state, blk1, &ichdr1,
1771 						      blk2, &ichdr2,
1772 						      &count, &totallen);
1773 	if (swap)
1774 		state->inleaf = !state->inleaf;
1775 
1776 	/*
1777 	 * Move any entries required from leaf to leaf:
1778 	 */
1779 	if (count < ichdr1.count) {
1780 		/*
1781 		 * Figure the total bytes to be added to the destination leaf.
1782 		 */
1783 		/* number entries being moved */
1784 		count = ichdr1.count - count;
1785 		space  = ichdr1.usedbytes - totallen;
1786 		space += count * sizeof(xfs_attr_leaf_entry_t);
1787 
1788 		/*
1789 		 * leaf2 is the destination, compact it if it looks tight.
1790 		 */
1791 		max  = ichdr2.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1792 		max -= ichdr2.count * sizeof(xfs_attr_leaf_entry_t);
1793 		if (space > max)
1794 			xfs_attr3_leaf_compact(args, &ichdr2, blk2->bp);
1795 
1796 		/*
1797 		 * Move high entries from leaf1 to low end of leaf2.
1798 		 */
1799 		xfs_attr3_leaf_moveents(args, leaf1, &ichdr1,
1800 				ichdr1.count - count, leaf2, &ichdr2, 0, count);
1801 
1802 	} else if (count > ichdr1.count) {
1803 		/*
1804 		 * I assert that since all callers pass in an empty
1805 		 * second buffer, this code should never execute.
1806 		 */
1807 		ASSERT(0);
1808 
1809 		/*
1810 		 * Figure the total bytes to be added to the destination leaf.
1811 		 */
1812 		/* number entries being moved */
1813 		count -= ichdr1.count;
1814 		space  = totallen - ichdr1.usedbytes;
1815 		space += count * sizeof(xfs_attr_leaf_entry_t);
1816 
1817 		/*
1818 		 * leaf1 is the destination, compact it if it looks tight.
1819 		 */
1820 		max  = ichdr1.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1821 		max -= ichdr1.count * sizeof(xfs_attr_leaf_entry_t);
1822 		if (space > max)
1823 			xfs_attr3_leaf_compact(args, &ichdr1, blk1->bp);
1824 
1825 		/*
1826 		 * Move low entries from leaf2 to high end of leaf1.
1827 		 */
1828 		xfs_attr3_leaf_moveents(args, leaf2, &ichdr2, 0, leaf1, &ichdr1,
1829 					ichdr1.count, count);
1830 	}
1831 
1832 	xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf1, &ichdr1);
1833 	xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf2, &ichdr2);
1834 	xfs_trans_log_buf(args->trans, blk1->bp, 0, args->geo->blksize - 1);
1835 	xfs_trans_log_buf(args->trans, blk2->bp, 0, args->geo->blksize - 1);
1836 
1837 	/*
1838 	 * Copy out last hashval in each block for B-tree code.
1839 	 */
1840 	entries1 = xfs_attr3_leaf_entryp(leaf1);
1841 	entries2 = xfs_attr3_leaf_entryp(leaf2);
1842 	blk1->hashval = be32_to_cpu(entries1[ichdr1.count - 1].hashval);
1843 	blk2->hashval = be32_to_cpu(entries2[ichdr2.count - 1].hashval);
1844 
1845 	/*
1846 	 * Adjust the expected index for insertion.
1847 	 * NOTE: this code depends on the (current) situation that the
1848 	 * second block was originally empty.
1849 	 *
1850 	 * If the insertion point moved to the 2nd block, we must adjust
1851 	 * the index.  We must also track the entry just following the
1852 	 * new entry for use in an "atomic rename" operation, that entry
1853 	 * is always the "old" entry and the "new" entry is what we are
1854 	 * inserting.  The index/blkno fields refer to the "old" entry,
1855 	 * while the index2/blkno2 fields refer to the "new" entry.
1856 	 */
1857 	if (blk1->index > ichdr1.count) {
1858 		ASSERT(state->inleaf == 0);
1859 		blk2->index = blk1->index - ichdr1.count;
1860 		args->index = args->index2 = blk2->index;
1861 		args->blkno = args->blkno2 = blk2->blkno;
1862 	} else if (blk1->index == ichdr1.count) {
1863 		if (state->inleaf) {
1864 			args->index = blk1->index;
1865 			args->blkno = blk1->blkno;
1866 			args->index2 = 0;
1867 			args->blkno2 = blk2->blkno;
1868 		} else {
1869 			/*
1870 			 * On a double leaf split, the original attr location
1871 			 * is already stored in blkno2/index2, so don't
1872 			 * overwrite it overwise we corrupt the tree.
1873 			 */
1874 			blk2->index = blk1->index - ichdr1.count;
1875 			args->index = blk2->index;
1876 			args->blkno = blk2->blkno;
1877 			if (!state->extravalid) {
1878 				/*
1879 				 * set the new attr location to match the old
1880 				 * one and let the higher level split code
1881 				 * decide where in the leaf to place it.
1882 				 */
1883 				args->index2 = blk2->index;
1884 				args->blkno2 = blk2->blkno;
1885 			}
1886 		}
1887 	} else {
1888 		ASSERT(state->inleaf == 1);
1889 		args->index = args->index2 = blk1->index;
1890 		args->blkno = args->blkno2 = blk1->blkno;
1891 	}
1892 }
1893 
1894 /*
1895  * Examine entries until we reduce the absolute difference in
1896  * byte usage between the two blocks to a minimum.
1897  * GROT: Is this really necessary?  With other than a 512 byte blocksize,
1898  * GROT: there will always be enough room in either block for a new entry.
1899  * GROT: Do a double-split for this case?
1900  */
1901 STATIC int
1902 xfs_attr3_leaf_figure_balance(
1903 	struct xfs_da_state		*state,
1904 	struct xfs_da_state_blk		*blk1,
1905 	struct xfs_attr3_icleaf_hdr	*ichdr1,
1906 	struct xfs_da_state_blk		*blk2,
1907 	struct xfs_attr3_icleaf_hdr	*ichdr2,
1908 	int				*countarg,
1909 	int				*usedbytesarg)
1910 {
1911 	struct xfs_attr_leafblock	*leaf1 = blk1->bp->b_addr;
1912 	struct xfs_attr_leafblock	*leaf2 = blk2->bp->b_addr;
1913 	struct xfs_attr_leaf_entry	*entry;
1914 	int				count;
1915 	int				max;
1916 	int				index;
1917 	int				totallen = 0;
1918 	int				half;
1919 	int				lastdelta;
1920 	int				foundit = 0;
1921 	int				tmp;
1922 
1923 	/*
1924 	 * Examine entries until we reduce the absolute difference in
1925 	 * byte usage between the two blocks to a minimum.
1926 	 */
1927 	max = ichdr1->count + ichdr2->count;
1928 	half = (max + 1) * sizeof(*entry);
1929 	half += ichdr1->usedbytes + ichdr2->usedbytes +
1930 			xfs_attr_leaf_newentsize(state->args, NULL);
1931 	half /= 2;
1932 	lastdelta = state->args->geo->blksize;
1933 	entry = xfs_attr3_leaf_entryp(leaf1);
1934 	for (count = index = 0; count < max; entry++, index++, count++) {
1935 
1936 #define XFS_ATTR_ABS(A)	(((A) < 0) ? -(A) : (A))
1937 		/*
1938 		 * The new entry is in the first block, account for it.
1939 		 */
1940 		if (count == blk1->index) {
1941 			tmp = totallen + sizeof(*entry) +
1942 				xfs_attr_leaf_newentsize(state->args, NULL);
1943 			if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1944 				break;
1945 			lastdelta = XFS_ATTR_ABS(half - tmp);
1946 			totallen = tmp;
1947 			foundit = 1;
1948 		}
1949 
1950 		/*
1951 		 * Wrap around into the second block if necessary.
1952 		 */
1953 		if (count == ichdr1->count) {
1954 			leaf1 = leaf2;
1955 			entry = xfs_attr3_leaf_entryp(leaf1);
1956 			index = 0;
1957 		}
1958 
1959 		/*
1960 		 * Figure out if next leaf entry would be too much.
1961 		 */
1962 		tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1963 									index);
1964 		if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1965 			break;
1966 		lastdelta = XFS_ATTR_ABS(half - tmp);
1967 		totallen = tmp;
1968 #undef XFS_ATTR_ABS
1969 	}
1970 
1971 	/*
1972 	 * Calculate the number of usedbytes that will end up in lower block.
1973 	 * If new entry not in lower block, fix up the count.
1974 	 */
1975 	totallen -= count * sizeof(*entry);
1976 	if (foundit) {
1977 		totallen -= sizeof(*entry) +
1978 				xfs_attr_leaf_newentsize(state->args, NULL);
1979 	}
1980 
1981 	*countarg = count;
1982 	*usedbytesarg = totallen;
1983 	return foundit;
1984 }
1985 
1986 /*========================================================================
1987  * Routines used for shrinking the Btree.
1988  *========================================================================*/
1989 
1990 /*
1991  * Check a leaf block and its neighbors to see if the block should be
1992  * collapsed into one or the other neighbor.  Always keep the block
1993  * with the smaller block number.
1994  * If the current block is over 50% full, don't try to join it, return 0.
1995  * If the block is empty, fill in the state structure and return 2.
1996  * If it can be collapsed, fill in the state structure and return 1.
1997  * If nothing can be done, return 0.
1998  *
1999  * GROT: allow for INCOMPLETE entries in calculation.
2000  */
2001 int
2002 xfs_attr3_leaf_toosmall(
2003 	struct xfs_da_state	*state,
2004 	int			*action)
2005 {
2006 	struct xfs_attr_leafblock *leaf;
2007 	struct xfs_da_state_blk	*blk;
2008 	struct xfs_attr3_icleaf_hdr ichdr;
2009 	struct xfs_buf		*bp;
2010 	xfs_dablk_t		blkno;
2011 	int			bytes;
2012 	int			forward;
2013 	int			error;
2014 	int			retval;
2015 	int			i;
2016 
2017 	trace_xfs_attr_leaf_toosmall(state->args);
2018 
2019 	/*
2020 	 * Check for the degenerate case of the block being over 50% full.
2021 	 * If so, it's not worth even looking to see if we might be able
2022 	 * to coalesce with a sibling.
2023 	 */
2024 	blk = &state->path.blk[ state->path.active-1 ];
2025 	leaf = blk->bp->b_addr;
2026 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr, leaf);
2027 	bytes = xfs_attr3_leaf_hdr_size(leaf) +
2028 		ichdr.count * sizeof(xfs_attr_leaf_entry_t) +
2029 		ichdr.usedbytes;
2030 	if (bytes > (state->args->geo->blksize >> 1)) {
2031 		*action = 0;	/* blk over 50%, don't try to join */
2032 		return 0;
2033 	}
2034 
2035 	/*
2036 	 * Check for the degenerate case of the block being empty.
2037 	 * If the block is empty, we'll simply delete it, no need to
2038 	 * coalesce it with a sibling block.  We choose (arbitrarily)
2039 	 * to merge with the forward block unless it is NULL.
2040 	 */
2041 	if (ichdr.count == 0) {
2042 		/*
2043 		 * Make altpath point to the block we want to keep and
2044 		 * path point to the block we want to drop (this one).
2045 		 */
2046 		forward = (ichdr.forw != 0);
2047 		memcpy(&state->altpath, &state->path, sizeof(state->path));
2048 		error = xfs_da3_path_shift(state, &state->altpath, forward,
2049 						 0, &retval);
2050 		if (error)
2051 			return error;
2052 		if (retval) {
2053 			*action = 0;
2054 		} else {
2055 			*action = 2;
2056 		}
2057 		return 0;
2058 	}
2059 
2060 	/*
2061 	 * Examine each sibling block to see if we can coalesce with
2062 	 * at least 25% free space to spare.  We need to figure out
2063 	 * whether to merge with the forward or the backward block.
2064 	 * We prefer coalescing with the lower numbered sibling so as
2065 	 * to shrink an attribute list over time.
2066 	 */
2067 	/* start with smaller blk num */
2068 	forward = ichdr.forw < ichdr.back;
2069 	for (i = 0; i < 2; forward = !forward, i++) {
2070 		struct xfs_attr3_icleaf_hdr ichdr2;
2071 		if (forward)
2072 			blkno = ichdr.forw;
2073 		else
2074 			blkno = ichdr.back;
2075 		if (blkno == 0)
2076 			continue;
2077 		error = xfs_attr3_leaf_read(state->args->trans, state->args->dp,
2078 					blkno, &bp);
2079 		if (error)
2080 			return error;
2081 
2082 		xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, bp->b_addr);
2083 
2084 		bytes = state->args->geo->blksize -
2085 			(state->args->geo->blksize >> 2) -
2086 			ichdr.usedbytes - ichdr2.usedbytes -
2087 			((ichdr.count + ichdr2.count) *
2088 					sizeof(xfs_attr_leaf_entry_t)) -
2089 			xfs_attr3_leaf_hdr_size(leaf);
2090 
2091 		xfs_trans_brelse(state->args->trans, bp);
2092 		if (bytes >= 0)
2093 			break;	/* fits with at least 25% to spare */
2094 	}
2095 	if (i >= 2) {
2096 		*action = 0;
2097 		return 0;
2098 	}
2099 
2100 	/*
2101 	 * Make altpath point to the block we want to keep (the lower
2102 	 * numbered block) and path point to the block we want to drop.
2103 	 */
2104 	memcpy(&state->altpath, &state->path, sizeof(state->path));
2105 	if (blkno < blk->blkno) {
2106 		error = xfs_da3_path_shift(state, &state->altpath, forward,
2107 						 0, &retval);
2108 	} else {
2109 		error = xfs_da3_path_shift(state, &state->path, forward,
2110 						 0, &retval);
2111 	}
2112 	if (error)
2113 		return error;
2114 	if (retval) {
2115 		*action = 0;
2116 	} else {
2117 		*action = 1;
2118 	}
2119 	return 0;
2120 }
2121 
2122 /*
2123  * Remove a name from the leaf attribute list structure.
2124  *
2125  * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
2126  * If two leaves are 37% full, when combined they will leave 25% free.
2127  */
2128 int
2129 xfs_attr3_leaf_remove(
2130 	struct xfs_buf		*bp,
2131 	struct xfs_da_args	*args)
2132 {
2133 	struct xfs_attr_leafblock *leaf;
2134 	struct xfs_attr3_icleaf_hdr ichdr;
2135 	struct xfs_attr_leaf_entry *entry;
2136 	int			before;
2137 	int			after;
2138 	int			smallest;
2139 	int			entsize;
2140 	int			tablesize;
2141 	int			tmp;
2142 	int			i;
2143 
2144 	trace_xfs_attr_leaf_remove(args);
2145 
2146 	leaf = bp->b_addr;
2147 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2148 
2149 	ASSERT(ichdr.count > 0 && ichdr.count < args->geo->blksize / 8);
2150 	ASSERT(args->index >= 0 && args->index < ichdr.count);
2151 	ASSERT(ichdr.firstused >= ichdr.count * sizeof(*entry) +
2152 					xfs_attr3_leaf_hdr_size(leaf));
2153 
2154 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2155 
2156 	ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2157 	ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
2158 
2159 	/*
2160 	 * Scan through free region table:
2161 	 *    check for adjacency of free'd entry with an existing one,
2162 	 *    find smallest free region in case we need to replace it,
2163 	 *    adjust any map that borders the entry table,
2164 	 */
2165 	tablesize = ichdr.count * sizeof(xfs_attr_leaf_entry_t)
2166 					+ xfs_attr3_leaf_hdr_size(leaf);
2167 	tmp = ichdr.freemap[0].size;
2168 	before = after = -1;
2169 	smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
2170 	entsize = xfs_attr_leaf_entsize(leaf, args->index);
2171 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
2172 		ASSERT(ichdr.freemap[i].base < args->geo->blksize);
2173 		ASSERT(ichdr.freemap[i].size < args->geo->blksize);
2174 		if (ichdr.freemap[i].base == tablesize) {
2175 			ichdr.freemap[i].base -= sizeof(xfs_attr_leaf_entry_t);
2176 			ichdr.freemap[i].size += sizeof(xfs_attr_leaf_entry_t);
2177 		}
2178 
2179 		if (ichdr.freemap[i].base + ichdr.freemap[i].size ==
2180 				be16_to_cpu(entry->nameidx)) {
2181 			before = i;
2182 		} else if (ichdr.freemap[i].base ==
2183 				(be16_to_cpu(entry->nameidx) + entsize)) {
2184 			after = i;
2185 		} else if (ichdr.freemap[i].size < tmp) {
2186 			tmp = ichdr.freemap[i].size;
2187 			smallest = i;
2188 		}
2189 	}
2190 
2191 	/*
2192 	 * Coalesce adjacent freemap regions,
2193 	 * or replace the smallest region.
2194 	 */
2195 	if ((before >= 0) || (after >= 0)) {
2196 		if ((before >= 0) && (after >= 0)) {
2197 			ichdr.freemap[before].size += entsize;
2198 			ichdr.freemap[before].size += ichdr.freemap[after].size;
2199 			ichdr.freemap[after].base = 0;
2200 			ichdr.freemap[after].size = 0;
2201 		} else if (before >= 0) {
2202 			ichdr.freemap[before].size += entsize;
2203 		} else {
2204 			ichdr.freemap[after].base = be16_to_cpu(entry->nameidx);
2205 			ichdr.freemap[after].size += entsize;
2206 		}
2207 	} else {
2208 		/*
2209 		 * Replace smallest region (if it is smaller than free'd entry)
2210 		 */
2211 		if (ichdr.freemap[smallest].size < entsize) {
2212 			ichdr.freemap[smallest].base = be16_to_cpu(entry->nameidx);
2213 			ichdr.freemap[smallest].size = entsize;
2214 		}
2215 	}
2216 
2217 	/*
2218 	 * Did we remove the first entry?
2219 	 */
2220 	if (be16_to_cpu(entry->nameidx) == ichdr.firstused)
2221 		smallest = 1;
2222 	else
2223 		smallest = 0;
2224 
2225 	/*
2226 	 * Compress the remaining entries and zero out the removed stuff.
2227 	 */
2228 	memset(xfs_attr3_leaf_name(leaf, args->index), 0, entsize);
2229 	ichdr.usedbytes -= entsize;
2230 	xfs_trans_log_buf(args->trans, bp,
2231 	     XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
2232 				   entsize));
2233 
2234 	tmp = (ichdr.count - args->index) * sizeof(xfs_attr_leaf_entry_t);
2235 	memmove(entry, entry + 1, tmp);
2236 	ichdr.count--;
2237 	xfs_trans_log_buf(args->trans, bp,
2238 	    XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(xfs_attr_leaf_entry_t)));
2239 
2240 	entry = &xfs_attr3_leaf_entryp(leaf)[ichdr.count];
2241 	memset(entry, 0, sizeof(xfs_attr_leaf_entry_t));
2242 
2243 	/*
2244 	 * If we removed the first entry, re-find the first used byte
2245 	 * in the name area.  Note that if the entry was the "firstused",
2246 	 * then we don't have a "hole" in our block resulting from
2247 	 * removing the name.
2248 	 */
2249 	if (smallest) {
2250 		tmp = args->geo->blksize;
2251 		entry = xfs_attr3_leaf_entryp(leaf);
2252 		for (i = ichdr.count - 1; i >= 0; entry++, i--) {
2253 			ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2254 			ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
2255 
2256 			if (be16_to_cpu(entry->nameidx) < tmp)
2257 				tmp = be16_to_cpu(entry->nameidx);
2258 		}
2259 		ichdr.firstused = tmp;
2260 		ASSERT(ichdr.firstused != 0);
2261 	} else {
2262 		ichdr.holes = 1;	/* mark as needing compaction */
2263 	}
2264 	xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
2265 	xfs_trans_log_buf(args->trans, bp,
2266 			  XFS_DA_LOGRANGE(leaf, &leaf->hdr,
2267 					  xfs_attr3_leaf_hdr_size(leaf)));
2268 
2269 	/*
2270 	 * Check if leaf is less than 50% full, caller may want to
2271 	 * "join" the leaf with a sibling if so.
2272 	 */
2273 	tmp = ichdr.usedbytes + xfs_attr3_leaf_hdr_size(leaf) +
2274 	      ichdr.count * sizeof(xfs_attr_leaf_entry_t);
2275 
2276 	return tmp < args->geo->magicpct; /* leaf is < 37% full */
2277 }
2278 
2279 /*
2280  * Move all the attribute list entries from drop_leaf into save_leaf.
2281  */
2282 void
2283 xfs_attr3_leaf_unbalance(
2284 	struct xfs_da_state	*state,
2285 	struct xfs_da_state_blk	*drop_blk,
2286 	struct xfs_da_state_blk	*save_blk)
2287 {
2288 	struct xfs_attr_leafblock *drop_leaf = drop_blk->bp->b_addr;
2289 	struct xfs_attr_leafblock *save_leaf = save_blk->bp->b_addr;
2290 	struct xfs_attr3_icleaf_hdr drophdr;
2291 	struct xfs_attr3_icleaf_hdr savehdr;
2292 	struct xfs_attr_leaf_entry *entry;
2293 
2294 	trace_xfs_attr_leaf_unbalance(state->args);
2295 
2296 	drop_leaf = drop_blk->bp->b_addr;
2297 	save_leaf = save_blk->bp->b_addr;
2298 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &drophdr, drop_leaf);
2299 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &savehdr, save_leaf);
2300 	entry = xfs_attr3_leaf_entryp(drop_leaf);
2301 
2302 	/*
2303 	 * Save last hashval from dying block for later Btree fixup.
2304 	 */
2305 	drop_blk->hashval = be32_to_cpu(entry[drophdr.count - 1].hashval);
2306 
2307 	/*
2308 	 * Check if we need a temp buffer, or can we do it in place.
2309 	 * Note that we don't check "leaf" for holes because we will
2310 	 * always be dropping it, toosmall() decided that for us already.
2311 	 */
2312 	if (savehdr.holes == 0) {
2313 		/*
2314 		 * dest leaf has no holes, so we add there.  May need
2315 		 * to make some room in the entry array.
2316 		 */
2317 		if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2318 					 drop_blk->bp, &drophdr)) {
2319 			xfs_attr3_leaf_moveents(state->args,
2320 						drop_leaf, &drophdr, 0,
2321 						save_leaf, &savehdr, 0,
2322 						drophdr.count);
2323 		} else {
2324 			xfs_attr3_leaf_moveents(state->args,
2325 						drop_leaf, &drophdr, 0,
2326 						save_leaf, &savehdr,
2327 						savehdr.count, drophdr.count);
2328 		}
2329 	} else {
2330 		/*
2331 		 * Destination has holes, so we make a temporary copy
2332 		 * of the leaf and add them both to that.
2333 		 */
2334 		struct xfs_attr_leafblock *tmp_leaf;
2335 		struct xfs_attr3_icleaf_hdr tmphdr;
2336 
2337 		tmp_leaf = kmem_zalloc(state->args->geo->blksize, 0);
2338 
2339 		/*
2340 		 * Copy the header into the temp leaf so that all the stuff
2341 		 * not in the incore header is present and gets copied back in
2342 		 * once we've moved all the entries.
2343 		 */
2344 		memcpy(tmp_leaf, save_leaf, xfs_attr3_leaf_hdr_size(save_leaf));
2345 
2346 		memset(&tmphdr, 0, sizeof(tmphdr));
2347 		tmphdr.magic = savehdr.magic;
2348 		tmphdr.forw = savehdr.forw;
2349 		tmphdr.back = savehdr.back;
2350 		tmphdr.firstused = state->args->geo->blksize;
2351 
2352 		/* write the header to the temp buffer to initialise it */
2353 		xfs_attr3_leaf_hdr_to_disk(state->args->geo, tmp_leaf, &tmphdr);
2354 
2355 		if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2356 					 drop_blk->bp, &drophdr)) {
2357 			xfs_attr3_leaf_moveents(state->args,
2358 						drop_leaf, &drophdr, 0,
2359 						tmp_leaf, &tmphdr, 0,
2360 						drophdr.count);
2361 			xfs_attr3_leaf_moveents(state->args,
2362 						save_leaf, &savehdr, 0,
2363 						tmp_leaf, &tmphdr, tmphdr.count,
2364 						savehdr.count);
2365 		} else {
2366 			xfs_attr3_leaf_moveents(state->args,
2367 						save_leaf, &savehdr, 0,
2368 						tmp_leaf, &tmphdr, 0,
2369 						savehdr.count);
2370 			xfs_attr3_leaf_moveents(state->args,
2371 						drop_leaf, &drophdr, 0,
2372 						tmp_leaf, &tmphdr, tmphdr.count,
2373 						drophdr.count);
2374 		}
2375 		memcpy(save_leaf, tmp_leaf, state->args->geo->blksize);
2376 		savehdr = tmphdr; /* struct copy */
2377 		kmem_free(tmp_leaf);
2378 	}
2379 
2380 	xfs_attr3_leaf_hdr_to_disk(state->args->geo, save_leaf, &savehdr);
2381 	xfs_trans_log_buf(state->args->trans, save_blk->bp, 0,
2382 					   state->args->geo->blksize - 1);
2383 
2384 	/*
2385 	 * Copy out last hashval in each block for B-tree code.
2386 	 */
2387 	entry = xfs_attr3_leaf_entryp(save_leaf);
2388 	save_blk->hashval = be32_to_cpu(entry[savehdr.count - 1].hashval);
2389 }
2390 
2391 /*========================================================================
2392  * Routines used for finding things in the Btree.
2393  *========================================================================*/
2394 
2395 /*
2396  * Look up a name in a leaf attribute list structure.
2397  * This is the internal routine, it uses the caller's buffer.
2398  *
2399  * Note that duplicate keys are allowed, but only check within the
2400  * current leaf node.  The Btree code must check in adjacent leaf nodes.
2401  *
2402  * Return in args->index the index into the entry[] array of either
2403  * the found entry, or where the entry should have been (insert before
2404  * that entry).
2405  *
2406  * Don't change the args->value unless we find the attribute.
2407  */
2408 int
2409 xfs_attr3_leaf_lookup_int(
2410 	struct xfs_buf		*bp,
2411 	struct xfs_da_args	*args)
2412 {
2413 	struct xfs_attr_leafblock *leaf;
2414 	struct xfs_attr3_icleaf_hdr ichdr;
2415 	struct xfs_attr_leaf_entry *entry;
2416 	struct xfs_attr_leaf_entry *entries;
2417 	struct xfs_attr_leaf_name_local *name_loc;
2418 	struct xfs_attr_leaf_name_remote *name_rmt;
2419 	xfs_dahash_t		hashval;
2420 	int			probe;
2421 	int			span;
2422 
2423 	trace_xfs_attr_leaf_lookup(args);
2424 
2425 	leaf = bp->b_addr;
2426 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2427 	entries = xfs_attr3_leaf_entryp(leaf);
2428 	if (ichdr.count >= args->geo->blksize / 8) {
2429 		xfs_buf_mark_corrupt(bp);
2430 		return -EFSCORRUPTED;
2431 	}
2432 
2433 	/*
2434 	 * Binary search.  (note: small blocks will skip this loop)
2435 	 */
2436 	hashval = args->hashval;
2437 	probe = span = ichdr.count / 2;
2438 	for (entry = &entries[probe]; span > 4; entry = &entries[probe]) {
2439 		span /= 2;
2440 		if (be32_to_cpu(entry->hashval) < hashval)
2441 			probe += span;
2442 		else if (be32_to_cpu(entry->hashval) > hashval)
2443 			probe -= span;
2444 		else
2445 			break;
2446 	}
2447 	if (!(probe >= 0 && (!ichdr.count || probe < ichdr.count))) {
2448 		xfs_buf_mark_corrupt(bp);
2449 		return -EFSCORRUPTED;
2450 	}
2451 	if (!(span <= 4 || be32_to_cpu(entry->hashval) == hashval)) {
2452 		xfs_buf_mark_corrupt(bp);
2453 		return -EFSCORRUPTED;
2454 	}
2455 
2456 	/*
2457 	 * Since we may have duplicate hashval's, find the first matching
2458 	 * hashval in the leaf.
2459 	 */
2460 	while (probe > 0 && be32_to_cpu(entry->hashval) >= hashval) {
2461 		entry--;
2462 		probe--;
2463 	}
2464 	while (probe < ichdr.count &&
2465 	       be32_to_cpu(entry->hashval) < hashval) {
2466 		entry++;
2467 		probe++;
2468 	}
2469 	if (probe == ichdr.count || be32_to_cpu(entry->hashval) != hashval) {
2470 		args->index = probe;
2471 		return -ENOATTR;
2472 	}
2473 
2474 	/*
2475 	 * Duplicate keys may be present, so search all of them for a match.
2476 	 */
2477 	for (; probe < ichdr.count && (be32_to_cpu(entry->hashval) == hashval);
2478 			entry++, probe++) {
2479 /*
2480  * GROT: Add code to remove incomplete entries.
2481  */
2482 		if (entry->flags & XFS_ATTR_LOCAL) {
2483 			name_loc = xfs_attr3_leaf_name_local(leaf, probe);
2484 			if (!xfs_attr_match(args, name_loc->namelen,
2485 					name_loc->nameval, entry->flags))
2486 				continue;
2487 			args->index = probe;
2488 			return -EEXIST;
2489 		} else {
2490 			name_rmt = xfs_attr3_leaf_name_remote(leaf, probe);
2491 			if (!xfs_attr_match(args, name_rmt->namelen,
2492 					name_rmt->name, entry->flags))
2493 				continue;
2494 			args->index = probe;
2495 			args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2496 			args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2497 			args->rmtblkcnt = xfs_attr3_rmt_blocks(
2498 							args->dp->i_mount,
2499 							args->rmtvaluelen);
2500 			return -EEXIST;
2501 		}
2502 	}
2503 	args->index = probe;
2504 	return -ENOATTR;
2505 }
2506 
2507 /*
2508  * Get the value associated with an attribute name from a leaf attribute
2509  * list structure.
2510  *
2511  * If args->valuelen is zero, only the length needs to be returned.  Unlike a
2512  * lookup, we only return an error if the attribute does not exist or we can't
2513  * retrieve the value.
2514  */
2515 int
2516 xfs_attr3_leaf_getvalue(
2517 	struct xfs_buf		*bp,
2518 	struct xfs_da_args	*args)
2519 {
2520 	struct xfs_attr_leafblock *leaf;
2521 	struct xfs_attr3_icleaf_hdr ichdr;
2522 	struct xfs_attr_leaf_entry *entry;
2523 	struct xfs_attr_leaf_name_local *name_loc;
2524 	struct xfs_attr_leaf_name_remote *name_rmt;
2525 
2526 	leaf = bp->b_addr;
2527 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2528 	ASSERT(ichdr.count < args->geo->blksize / 8);
2529 	ASSERT(args->index < ichdr.count);
2530 
2531 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2532 	if (entry->flags & XFS_ATTR_LOCAL) {
2533 		name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2534 		ASSERT(name_loc->namelen == args->namelen);
2535 		ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2536 		return xfs_attr_copy_value(args,
2537 					&name_loc->nameval[args->namelen],
2538 					be16_to_cpu(name_loc->valuelen));
2539 	}
2540 
2541 	name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2542 	ASSERT(name_rmt->namelen == args->namelen);
2543 	ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2544 	args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2545 	args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2546 	args->rmtblkcnt = xfs_attr3_rmt_blocks(args->dp->i_mount,
2547 					       args->rmtvaluelen);
2548 	return xfs_attr_copy_value(args, NULL, args->rmtvaluelen);
2549 }
2550 
2551 /*========================================================================
2552  * Utility routines.
2553  *========================================================================*/
2554 
2555 /*
2556  * Move the indicated entries from one leaf to another.
2557  * NOTE: this routine modifies both source and destination leaves.
2558  */
2559 /*ARGSUSED*/
2560 STATIC void
2561 xfs_attr3_leaf_moveents(
2562 	struct xfs_da_args		*args,
2563 	struct xfs_attr_leafblock	*leaf_s,
2564 	struct xfs_attr3_icleaf_hdr	*ichdr_s,
2565 	int				start_s,
2566 	struct xfs_attr_leafblock	*leaf_d,
2567 	struct xfs_attr3_icleaf_hdr	*ichdr_d,
2568 	int				start_d,
2569 	int				count)
2570 {
2571 	struct xfs_attr_leaf_entry	*entry_s;
2572 	struct xfs_attr_leaf_entry	*entry_d;
2573 	int				desti;
2574 	int				tmp;
2575 	int				i;
2576 
2577 	/*
2578 	 * Check for nothing to do.
2579 	 */
2580 	if (count == 0)
2581 		return;
2582 
2583 	/*
2584 	 * Set up environment.
2585 	 */
2586 	ASSERT(ichdr_s->magic == XFS_ATTR_LEAF_MAGIC ||
2587 	       ichdr_s->magic == XFS_ATTR3_LEAF_MAGIC);
2588 	ASSERT(ichdr_s->magic == ichdr_d->magic);
2589 	ASSERT(ichdr_s->count > 0 && ichdr_s->count < args->geo->blksize / 8);
2590 	ASSERT(ichdr_s->firstused >= (ichdr_s->count * sizeof(*entry_s))
2591 					+ xfs_attr3_leaf_hdr_size(leaf_s));
2592 	ASSERT(ichdr_d->count < args->geo->blksize / 8);
2593 	ASSERT(ichdr_d->firstused >= (ichdr_d->count * sizeof(*entry_d))
2594 					+ xfs_attr3_leaf_hdr_size(leaf_d));
2595 
2596 	ASSERT(start_s < ichdr_s->count);
2597 	ASSERT(start_d <= ichdr_d->count);
2598 	ASSERT(count <= ichdr_s->count);
2599 
2600 
2601 	/*
2602 	 * Move the entries in the destination leaf up to make a hole?
2603 	 */
2604 	if (start_d < ichdr_d->count) {
2605 		tmp  = ichdr_d->count - start_d;
2606 		tmp *= sizeof(xfs_attr_leaf_entry_t);
2607 		entry_s = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2608 		entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d + count];
2609 		memmove(entry_d, entry_s, tmp);
2610 	}
2611 
2612 	/*
2613 	 * Copy all entry's in the same (sorted) order,
2614 	 * but allocate attribute info packed and in sequence.
2615 	 */
2616 	entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2617 	entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2618 	desti = start_d;
2619 	for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2620 		ASSERT(be16_to_cpu(entry_s->nameidx) >= ichdr_s->firstused);
2621 		tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2622 #ifdef GROT
2623 		/*
2624 		 * Code to drop INCOMPLETE entries.  Difficult to use as we
2625 		 * may also need to change the insertion index.  Code turned
2626 		 * off for 6.2, should be revisited later.
2627 		 */
2628 		if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2629 			memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2630 			ichdr_s->usedbytes -= tmp;
2631 			ichdr_s->count -= 1;
2632 			entry_d--;	/* to compensate for ++ in loop hdr */
2633 			desti--;
2634 			if ((start_s + i) < offset)
2635 				result++;	/* insertion index adjustment */
2636 		} else {
2637 #endif /* GROT */
2638 			ichdr_d->firstused -= tmp;
2639 			/* both on-disk, don't endian flip twice */
2640 			entry_d->hashval = entry_s->hashval;
2641 			entry_d->nameidx = cpu_to_be16(ichdr_d->firstused);
2642 			entry_d->flags = entry_s->flags;
2643 			ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2644 							<= args->geo->blksize);
2645 			memmove(xfs_attr3_leaf_name(leaf_d, desti),
2646 				xfs_attr3_leaf_name(leaf_s, start_s + i), tmp);
2647 			ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2648 							<= args->geo->blksize);
2649 			memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2650 			ichdr_s->usedbytes -= tmp;
2651 			ichdr_d->usedbytes += tmp;
2652 			ichdr_s->count -= 1;
2653 			ichdr_d->count += 1;
2654 			tmp = ichdr_d->count * sizeof(xfs_attr_leaf_entry_t)
2655 					+ xfs_attr3_leaf_hdr_size(leaf_d);
2656 			ASSERT(ichdr_d->firstused >= tmp);
2657 #ifdef GROT
2658 		}
2659 #endif /* GROT */
2660 	}
2661 
2662 	/*
2663 	 * Zero out the entries we just copied.
2664 	 */
2665 	if (start_s == ichdr_s->count) {
2666 		tmp = count * sizeof(xfs_attr_leaf_entry_t);
2667 		entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2668 		ASSERT(((char *)entry_s + tmp) <=
2669 		       ((char *)leaf_s + args->geo->blksize));
2670 		memset(entry_s, 0, tmp);
2671 	} else {
2672 		/*
2673 		 * Move the remaining entries down to fill the hole,
2674 		 * then zero the entries at the top.
2675 		 */
2676 		tmp  = (ichdr_s->count - count) * sizeof(xfs_attr_leaf_entry_t);
2677 		entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s + count];
2678 		entry_d = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2679 		memmove(entry_d, entry_s, tmp);
2680 
2681 		tmp = count * sizeof(xfs_attr_leaf_entry_t);
2682 		entry_s = &xfs_attr3_leaf_entryp(leaf_s)[ichdr_s->count];
2683 		ASSERT(((char *)entry_s + tmp) <=
2684 		       ((char *)leaf_s + args->geo->blksize));
2685 		memset(entry_s, 0, tmp);
2686 	}
2687 
2688 	/*
2689 	 * Fill in the freemap information
2690 	 */
2691 	ichdr_d->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_d);
2692 	ichdr_d->freemap[0].base += ichdr_d->count * sizeof(xfs_attr_leaf_entry_t);
2693 	ichdr_d->freemap[0].size = ichdr_d->firstused - ichdr_d->freemap[0].base;
2694 	ichdr_d->freemap[1].base = 0;
2695 	ichdr_d->freemap[2].base = 0;
2696 	ichdr_d->freemap[1].size = 0;
2697 	ichdr_d->freemap[2].size = 0;
2698 	ichdr_s->holes = 1;	/* leaf may not be compact */
2699 }
2700 
2701 /*
2702  * Pick up the last hashvalue from a leaf block.
2703  */
2704 xfs_dahash_t
2705 xfs_attr_leaf_lasthash(
2706 	struct xfs_buf	*bp,
2707 	int		*count)
2708 {
2709 	struct xfs_attr3_icleaf_hdr ichdr;
2710 	struct xfs_attr_leaf_entry *entries;
2711 	struct xfs_mount *mp = bp->b_mount;
2712 
2713 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, bp->b_addr);
2714 	entries = xfs_attr3_leaf_entryp(bp->b_addr);
2715 	if (count)
2716 		*count = ichdr.count;
2717 	if (!ichdr.count)
2718 		return 0;
2719 	return be32_to_cpu(entries[ichdr.count - 1].hashval);
2720 }
2721 
2722 /*
2723  * Calculate the number of bytes used to store the indicated attribute
2724  * (whether local or remote only calculate bytes in this block).
2725  */
2726 STATIC int
2727 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2728 {
2729 	struct xfs_attr_leaf_entry *entries;
2730 	xfs_attr_leaf_name_local_t *name_loc;
2731 	xfs_attr_leaf_name_remote_t *name_rmt;
2732 	int size;
2733 
2734 	entries = xfs_attr3_leaf_entryp(leaf);
2735 	if (entries[index].flags & XFS_ATTR_LOCAL) {
2736 		name_loc = xfs_attr3_leaf_name_local(leaf, index);
2737 		size = xfs_attr_leaf_entsize_local(name_loc->namelen,
2738 						   be16_to_cpu(name_loc->valuelen));
2739 	} else {
2740 		name_rmt = xfs_attr3_leaf_name_remote(leaf, index);
2741 		size = xfs_attr_leaf_entsize_remote(name_rmt->namelen);
2742 	}
2743 	return size;
2744 }
2745 
2746 /*
2747  * Calculate the number of bytes that would be required to store the new
2748  * attribute (whether local or remote only calculate bytes in this block).
2749  * This routine decides as a side effect whether the attribute will be
2750  * a "local" or a "remote" attribute.
2751  */
2752 int
2753 xfs_attr_leaf_newentsize(
2754 	struct xfs_da_args	*args,
2755 	int			*local)
2756 {
2757 	int			size;
2758 
2759 	size = xfs_attr_leaf_entsize_local(args->namelen, args->valuelen);
2760 	if (size < xfs_attr_leaf_entsize_local_max(args->geo->blksize)) {
2761 		if (local)
2762 			*local = 1;
2763 		return size;
2764 	}
2765 	if (local)
2766 		*local = 0;
2767 	return xfs_attr_leaf_entsize_remote(args->namelen);
2768 }
2769 
2770 
2771 /*========================================================================
2772  * Manage the INCOMPLETE flag in a leaf entry
2773  *========================================================================*/
2774 
2775 /*
2776  * Clear the INCOMPLETE flag on an entry in a leaf block.
2777  */
2778 int
2779 xfs_attr3_leaf_clearflag(
2780 	struct xfs_da_args	*args)
2781 {
2782 	struct xfs_attr_leafblock *leaf;
2783 	struct xfs_attr_leaf_entry *entry;
2784 	struct xfs_attr_leaf_name_remote *name_rmt;
2785 	struct xfs_buf		*bp;
2786 	int			error;
2787 #ifdef DEBUG
2788 	struct xfs_attr3_icleaf_hdr ichdr;
2789 	xfs_attr_leaf_name_local_t *name_loc;
2790 	int namelen;
2791 	char *name;
2792 #endif /* DEBUG */
2793 
2794 	trace_xfs_attr_leaf_clearflag(args);
2795 	/*
2796 	 * Set up the operation.
2797 	 */
2798 	error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp);
2799 	if (error)
2800 		return error;
2801 
2802 	leaf = bp->b_addr;
2803 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2804 	ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2805 
2806 #ifdef DEBUG
2807 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2808 	ASSERT(args->index < ichdr.count);
2809 	ASSERT(args->index >= 0);
2810 
2811 	if (entry->flags & XFS_ATTR_LOCAL) {
2812 		name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2813 		namelen = name_loc->namelen;
2814 		name = (char *)name_loc->nameval;
2815 	} else {
2816 		name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2817 		namelen = name_rmt->namelen;
2818 		name = (char *)name_rmt->name;
2819 	}
2820 	ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2821 	ASSERT(namelen == args->namelen);
2822 	ASSERT(memcmp(name, args->name, namelen) == 0);
2823 #endif /* DEBUG */
2824 
2825 	entry->flags &= ~XFS_ATTR_INCOMPLETE;
2826 	xfs_trans_log_buf(args->trans, bp,
2827 			 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2828 
2829 	if (args->rmtblkno) {
2830 		ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2831 		name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2832 		name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2833 		name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2834 		xfs_trans_log_buf(args->trans, bp,
2835 			 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2836 	}
2837 
2838 	return 0;
2839 }
2840 
2841 /*
2842  * Set the INCOMPLETE flag on an entry in a leaf block.
2843  */
2844 int
2845 xfs_attr3_leaf_setflag(
2846 	struct xfs_da_args	*args)
2847 {
2848 	struct xfs_attr_leafblock *leaf;
2849 	struct xfs_attr_leaf_entry *entry;
2850 	struct xfs_attr_leaf_name_remote *name_rmt;
2851 	struct xfs_buf		*bp;
2852 	int error;
2853 #ifdef DEBUG
2854 	struct xfs_attr3_icleaf_hdr ichdr;
2855 #endif
2856 
2857 	trace_xfs_attr_leaf_setflag(args);
2858 
2859 	/*
2860 	 * Set up the operation.
2861 	 */
2862 	error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp);
2863 	if (error)
2864 		return error;
2865 
2866 	leaf = bp->b_addr;
2867 #ifdef DEBUG
2868 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2869 	ASSERT(args->index < ichdr.count);
2870 	ASSERT(args->index >= 0);
2871 #endif
2872 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2873 
2874 	ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2875 	entry->flags |= XFS_ATTR_INCOMPLETE;
2876 	xfs_trans_log_buf(args->trans, bp,
2877 			XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2878 	if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2879 		name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2880 		name_rmt->valueblk = 0;
2881 		name_rmt->valuelen = 0;
2882 		xfs_trans_log_buf(args->trans, bp,
2883 			 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2884 	}
2885 
2886 	return 0;
2887 }
2888 
2889 /*
2890  * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2891  * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2892  * entry given by args->blkno2/index2.
2893  *
2894  * Note that they could be in different blocks, or in the same block.
2895  */
2896 int
2897 xfs_attr3_leaf_flipflags(
2898 	struct xfs_da_args	*args)
2899 {
2900 	struct xfs_attr_leafblock *leaf1;
2901 	struct xfs_attr_leafblock *leaf2;
2902 	struct xfs_attr_leaf_entry *entry1;
2903 	struct xfs_attr_leaf_entry *entry2;
2904 	struct xfs_attr_leaf_name_remote *name_rmt;
2905 	struct xfs_buf		*bp1;
2906 	struct xfs_buf		*bp2;
2907 	int error;
2908 #ifdef DEBUG
2909 	struct xfs_attr3_icleaf_hdr ichdr1;
2910 	struct xfs_attr3_icleaf_hdr ichdr2;
2911 	xfs_attr_leaf_name_local_t *name_loc;
2912 	int namelen1, namelen2;
2913 	char *name1, *name2;
2914 #endif /* DEBUG */
2915 
2916 	trace_xfs_attr_leaf_flipflags(args);
2917 
2918 	/*
2919 	 * Read the block containing the "old" attr
2920 	 */
2921 	error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp1);
2922 	if (error)
2923 		return error;
2924 
2925 	/*
2926 	 * Read the block containing the "new" attr, if it is different
2927 	 */
2928 	if (args->blkno2 != args->blkno) {
2929 		error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno2,
2930 					   &bp2);
2931 		if (error)
2932 			return error;
2933 	} else {
2934 		bp2 = bp1;
2935 	}
2936 
2937 	leaf1 = bp1->b_addr;
2938 	entry1 = &xfs_attr3_leaf_entryp(leaf1)[args->index];
2939 
2940 	leaf2 = bp2->b_addr;
2941 	entry2 = &xfs_attr3_leaf_entryp(leaf2)[args->index2];
2942 
2943 #ifdef DEBUG
2944 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr1, leaf1);
2945 	ASSERT(args->index < ichdr1.count);
2946 	ASSERT(args->index >= 0);
2947 
2948 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr2, leaf2);
2949 	ASSERT(args->index2 < ichdr2.count);
2950 	ASSERT(args->index2 >= 0);
2951 
2952 	if (entry1->flags & XFS_ATTR_LOCAL) {
2953 		name_loc = xfs_attr3_leaf_name_local(leaf1, args->index);
2954 		namelen1 = name_loc->namelen;
2955 		name1 = (char *)name_loc->nameval;
2956 	} else {
2957 		name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2958 		namelen1 = name_rmt->namelen;
2959 		name1 = (char *)name_rmt->name;
2960 	}
2961 	if (entry2->flags & XFS_ATTR_LOCAL) {
2962 		name_loc = xfs_attr3_leaf_name_local(leaf2, args->index2);
2963 		namelen2 = name_loc->namelen;
2964 		name2 = (char *)name_loc->nameval;
2965 	} else {
2966 		name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2967 		namelen2 = name_rmt->namelen;
2968 		name2 = (char *)name_rmt->name;
2969 	}
2970 	ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2971 	ASSERT(namelen1 == namelen2);
2972 	ASSERT(memcmp(name1, name2, namelen1) == 0);
2973 #endif /* DEBUG */
2974 
2975 	ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2976 	ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2977 
2978 	entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2979 	xfs_trans_log_buf(args->trans, bp1,
2980 			  XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2981 	if (args->rmtblkno) {
2982 		ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2983 		name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2984 		name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2985 		name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2986 		xfs_trans_log_buf(args->trans, bp1,
2987 			 XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2988 	}
2989 
2990 	entry2->flags |= XFS_ATTR_INCOMPLETE;
2991 	xfs_trans_log_buf(args->trans, bp2,
2992 			  XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2993 	if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
2994 		name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2995 		name_rmt->valueblk = 0;
2996 		name_rmt->valuelen = 0;
2997 		xfs_trans_log_buf(args->trans, bp2,
2998 			 XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
2999 	}
3000 
3001 	return 0;
3002 }
3003