xref: /openbmc/linux/fs/xfs/libxfs/xfs_dir2_leaf.c (revision eb3fcf00)
1 /*
2  * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
3  * Copyright (c) 2013 Red Hat, Inc.
4  * All Rights Reserved.
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public License as
8  * published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it would be useful,
11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13  * GNU General Public License for more details.
14  *
15  * You should have received a copy of the GNU General Public License
16  * along with this program; if not, write the Free Software Foundation,
17  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
18  */
19 #include "xfs.h"
20 #include "xfs_fs.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
24 #include "xfs_mount.h"
25 #include "xfs_da_format.h"
26 #include "xfs_da_btree.h"
27 #include "xfs_inode.h"
28 #include "xfs_bmap.h"
29 #include "xfs_dir2.h"
30 #include "xfs_dir2_priv.h"
31 #include "xfs_error.h"
32 #include "xfs_trace.h"
33 #include "xfs_trans.h"
34 #include "xfs_buf_item.h"
35 #include "xfs_cksum.h"
36 
37 /*
38  * Local function declarations.
39  */
40 static int xfs_dir2_leaf_lookup_int(xfs_da_args_t *args, struct xfs_buf **lbpp,
41 				    int *indexp, struct xfs_buf **dbpp);
42 static void xfs_dir3_leaf_log_bests(struct xfs_da_args *args,
43 				    struct xfs_buf *bp, int first, int last);
44 static void xfs_dir3_leaf_log_tail(struct xfs_da_args *args,
45 				   struct xfs_buf *bp);
46 
47 /*
48  * Check the internal consistency of a leaf1 block.
49  * Pop an assert if something is wrong.
50  */
51 #ifdef DEBUG
52 #define	xfs_dir3_leaf_check(dp, bp) \
53 do { \
54 	if (!xfs_dir3_leaf1_check((dp), (bp))) \
55 		ASSERT(0); \
56 } while (0);
57 
58 STATIC bool
59 xfs_dir3_leaf1_check(
60 	struct xfs_inode	*dp,
61 	struct xfs_buf		*bp)
62 {
63 	struct xfs_dir2_leaf	*leaf = bp->b_addr;
64 	struct xfs_dir3_icleaf_hdr leafhdr;
65 
66 	dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
67 
68 	if (leafhdr.magic == XFS_DIR3_LEAF1_MAGIC) {
69 		struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
70 		if (be64_to_cpu(leaf3->info.blkno) != bp->b_bn)
71 			return false;
72 	} else if (leafhdr.magic != XFS_DIR2_LEAF1_MAGIC)
73 		return false;
74 
75 	return xfs_dir3_leaf_check_int(dp->i_mount, dp, &leafhdr, leaf);
76 }
77 #else
78 #define	xfs_dir3_leaf_check(dp, bp)
79 #endif
80 
81 bool
82 xfs_dir3_leaf_check_int(
83 	struct xfs_mount	*mp,
84 	struct xfs_inode	*dp,
85 	struct xfs_dir3_icleaf_hdr *hdr,
86 	struct xfs_dir2_leaf	*leaf)
87 {
88 	struct xfs_dir2_leaf_entry *ents;
89 	xfs_dir2_leaf_tail_t	*ltp;
90 	int			stale;
91 	int			i;
92 	const struct xfs_dir_ops *ops;
93 	struct xfs_dir3_icleaf_hdr leafhdr;
94 	struct xfs_da_geometry	*geo = mp->m_dir_geo;
95 
96 	/*
97 	 * we can be passed a null dp here from a verifier, so we need to go the
98 	 * hard way to get them.
99 	 */
100 	ops = xfs_dir_get_ops(mp, dp);
101 
102 	if (!hdr) {
103 		ops->leaf_hdr_from_disk(&leafhdr, leaf);
104 		hdr = &leafhdr;
105 	}
106 
107 	ents = ops->leaf_ents_p(leaf);
108 	ltp = xfs_dir2_leaf_tail_p(geo, leaf);
109 
110 	/*
111 	 * XXX (dgc): This value is not restrictive enough.
112 	 * Should factor in the size of the bests table as well.
113 	 * We can deduce a value for that from di_size.
114 	 */
115 	if (hdr->count > ops->leaf_max_ents(geo))
116 		return false;
117 
118 	/* Leaves and bests don't overlap in leaf format. */
119 	if ((hdr->magic == XFS_DIR2_LEAF1_MAGIC ||
120 	     hdr->magic == XFS_DIR3_LEAF1_MAGIC) &&
121 	    (char *)&ents[hdr->count] > (char *)xfs_dir2_leaf_bests_p(ltp))
122 		return false;
123 
124 	/* Check hash value order, count stale entries.  */
125 	for (i = stale = 0; i < hdr->count; i++) {
126 		if (i + 1 < hdr->count) {
127 			if (be32_to_cpu(ents[i].hashval) >
128 					be32_to_cpu(ents[i + 1].hashval))
129 				return false;
130 		}
131 		if (ents[i].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
132 			stale++;
133 	}
134 	if (hdr->stale != stale)
135 		return false;
136 	return true;
137 }
138 
139 /*
140  * We verify the magic numbers before decoding the leaf header so that on debug
141  * kernels we don't get assertion failures in xfs_dir3_leaf_hdr_from_disk() due
142  * to incorrect magic numbers.
143  */
144 static bool
145 xfs_dir3_leaf_verify(
146 	struct xfs_buf		*bp,
147 	__uint16_t		magic)
148 {
149 	struct xfs_mount	*mp = bp->b_target->bt_mount;
150 	struct xfs_dir2_leaf	*leaf = bp->b_addr;
151 
152 	ASSERT(magic == XFS_DIR2_LEAF1_MAGIC || magic == XFS_DIR2_LEAFN_MAGIC);
153 
154 	if (xfs_sb_version_hascrc(&mp->m_sb)) {
155 		struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
156 		__uint16_t		magic3;
157 
158 		magic3 = (magic == XFS_DIR2_LEAF1_MAGIC) ? XFS_DIR3_LEAF1_MAGIC
159 							 : XFS_DIR3_LEAFN_MAGIC;
160 
161 		if (leaf3->info.hdr.magic != cpu_to_be16(magic3))
162 			return false;
163 		if (!uuid_equal(&leaf3->info.uuid, &mp->m_sb.sb_meta_uuid))
164 			return false;
165 		if (be64_to_cpu(leaf3->info.blkno) != bp->b_bn)
166 			return false;
167 	} else {
168 		if (leaf->hdr.info.magic != cpu_to_be16(magic))
169 			return false;
170 	}
171 
172 	return xfs_dir3_leaf_check_int(mp, NULL, NULL, leaf);
173 }
174 
175 static void
176 __read_verify(
177 	struct xfs_buf  *bp,
178 	__uint16_t	magic)
179 {
180 	struct xfs_mount	*mp = bp->b_target->bt_mount;
181 
182 	if (xfs_sb_version_hascrc(&mp->m_sb) &&
183 	     !xfs_buf_verify_cksum(bp, XFS_DIR3_LEAF_CRC_OFF))
184 		xfs_buf_ioerror(bp, -EFSBADCRC);
185 	else if (!xfs_dir3_leaf_verify(bp, magic))
186 		xfs_buf_ioerror(bp, -EFSCORRUPTED);
187 
188 	if (bp->b_error)
189 		xfs_verifier_error(bp);
190 }
191 
192 static void
193 __write_verify(
194 	struct xfs_buf  *bp,
195 	__uint16_t	magic)
196 {
197 	struct xfs_mount	*mp = bp->b_target->bt_mount;
198 	struct xfs_buf_log_item	*bip = bp->b_fspriv;
199 	struct xfs_dir3_leaf_hdr *hdr3 = bp->b_addr;
200 
201 	if (!xfs_dir3_leaf_verify(bp, magic)) {
202 		xfs_buf_ioerror(bp, -EFSCORRUPTED);
203 		xfs_verifier_error(bp);
204 		return;
205 	}
206 
207 	if (!xfs_sb_version_hascrc(&mp->m_sb))
208 		return;
209 
210 	if (bip)
211 		hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
212 
213 	xfs_buf_update_cksum(bp, XFS_DIR3_LEAF_CRC_OFF);
214 }
215 
216 static void
217 xfs_dir3_leaf1_read_verify(
218 	struct xfs_buf	*bp)
219 {
220 	__read_verify(bp, XFS_DIR2_LEAF1_MAGIC);
221 }
222 
223 static void
224 xfs_dir3_leaf1_write_verify(
225 	struct xfs_buf	*bp)
226 {
227 	__write_verify(bp, XFS_DIR2_LEAF1_MAGIC);
228 }
229 
230 static void
231 xfs_dir3_leafn_read_verify(
232 	struct xfs_buf	*bp)
233 {
234 	__read_verify(bp, XFS_DIR2_LEAFN_MAGIC);
235 }
236 
237 static void
238 xfs_dir3_leafn_write_verify(
239 	struct xfs_buf	*bp)
240 {
241 	__write_verify(bp, XFS_DIR2_LEAFN_MAGIC);
242 }
243 
244 const struct xfs_buf_ops xfs_dir3_leaf1_buf_ops = {
245 	.verify_read = xfs_dir3_leaf1_read_verify,
246 	.verify_write = xfs_dir3_leaf1_write_verify,
247 };
248 
249 const struct xfs_buf_ops xfs_dir3_leafn_buf_ops = {
250 	.verify_read = xfs_dir3_leafn_read_verify,
251 	.verify_write = xfs_dir3_leafn_write_verify,
252 };
253 
254 static int
255 xfs_dir3_leaf_read(
256 	struct xfs_trans	*tp,
257 	struct xfs_inode	*dp,
258 	xfs_dablk_t		fbno,
259 	xfs_daddr_t		mappedbno,
260 	struct xfs_buf		**bpp)
261 {
262 	int			err;
263 
264 	err = xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
265 				XFS_DATA_FORK, &xfs_dir3_leaf1_buf_ops);
266 	if (!err && tp)
267 		xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_LEAF1_BUF);
268 	return err;
269 }
270 
271 int
272 xfs_dir3_leafn_read(
273 	struct xfs_trans	*tp,
274 	struct xfs_inode	*dp,
275 	xfs_dablk_t		fbno,
276 	xfs_daddr_t		mappedbno,
277 	struct xfs_buf		**bpp)
278 {
279 	int			err;
280 
281 	err = xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
282 				XFS_DATA_FORK, &xfs_dir3_leafn_buf_ops);
283 	if (!err && tp)
284 		xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_LEAFN_BUF);
285 	return err;
286 }
287 
288 /*
289  * Initialize a new leaf block, leaf1 or leafn magic accepted.
290  */
291 static void
292 xfs_dir3_leaf_init(
293 	struct xfs_mount	*mp,
294 	struct xfs_trans	*tp,
295 	struct xfs_buf		*bp,
296 	xfs_ino_t		owner,
297 	__uint16_t		type)
298 {
299 	struct xfs_dir2_leaf	*leaf = bp->b_addr;
300 
301 	ASSERT(type == XFS_DIR2_LEAF1_MAGIC || type == XFS_DIR2_LEAFN_MAGIC);
302 
303 	if (xfs_sb_version_hascrc(&mp->m_sb)) {
304 		struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
305 
306 		memset(leaf3, 0, sizeof(*leaf3));
307 
308 		leaf3->info.hdr.magic = (type == XFS_DIR2_LEAF1_MAGIC)
309 					 ? cpu_to_be16(XFS_DIR3_LEAF1_MAGIC)
310 					 : cpu_to_be16(XFS_DIR3_LEAFN_MAGIC);
311 		leaf3->info.blkno = cpu_to_be64(bp->b_bn);
312 		leaf3->info.owner = cpu_to_be64(owner);
313 		uuid_copy(&leaf3->info.uuid, &mp->m_sb.sb_meta_uuid);
314 	} else {
315 		memset(leaf, 0, sizeof(*leaf));
316 		leaf->hdr.info.magic = cpu_to_be16(type);
317 	}
318 
319 	/*
320 	 * If it's a leaf-format directory initialize the tail.
321 	 * Caller is responsible for initialising the bests table.
322 	 */
323 	if (type == XFS_DIR2_LEAF1_MAGIC) {
324 		struct xfs_dir2_leaf_tail *ltp;
325 
326 		ltp = xfs_dir2_leaf_tail_p(mp->m_dir_geo, leaf);
327 		ltp->bestcount = 0;
328 		bp->b_ops = &xfs_dir3_leaf1_buf_ops;
329 		xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAF1_BUF);
330 	} else {
331 		bp->b_ops = &xfs_dir3_leafn_buf_ops;
332 		xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAFN_BUF);
333 	}
334 }
335 
336 int
337 xfs_dir3_leaf_get_buf(
338 	xfs_da_args_t		*args,
339 	xfs_dir2_db_t		bno,
340 	struct xfs_buf		**bpp,
341 	__uint16_t		magic)
342 {
343 	struct xfs_inode	*dp = args->dp;
344 	struct xfs_trans	*tp = args->trans;
345 	struct xfs_mount	*mp = dp->i_mount;
346 	struct xfs_buf		*bp;
347 	int			error;
348 
349 	ASSERT(magic == XFS_DIR2_LEAF1_MAGIC || magic == XFS_DIR2_LEAFN_MAGIC);
350 	ASSERT(bno >= xfs_dir2_byte_to_db(args->geo, XFS_DIR2_LEAF_OFFSET) &&
351 	       bno < xfs_dir2_byte_to_db(args->geo, XFS_DIR2_FREE_OFFSET));
352 
353 	error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(args->geo, bno),
354 			       -1, &bp, XFS_DATA_FORK);
355 	if (error)
356 		return error;
357 
358 	xfs_dir3_leaf_init(mp, tp, bp, dp->i_ino, magic);
359 	xfs_dir3_leaf_log_header(args, bp);
360 	if (magic == XFS_DIR2_LEAF1_MAGIC)
361 		xfs_dir3_leaf_log_tail(args, bp);
362 	*bpp = bp;
363 	return 0;
364 }
365 
366 /*
367  * Convert a block form directory to a leaf form directory.
368  */
369 int						/* error */
370 xfs_dir2_block_to_leaf(
371 	xfs_da_args_t		*args,		/* operation arguments */
372 	struct xfs_buf		*dbp)		/* input block's buffer */
373 {
374 	__be16			*bestsp;	/* leaf's bestsp entries */
375 	xfs_dablk_t		blkno;		/* leaf block's bno */
376 	xfs_dir2_data_hdr_t	*hdr;		/* block header */
377 	xfs_dir2_leaf_entry_t	*blp;		/* block's leaf entries */
378 	xfs_dir2_block_tail_t	*btp;		/* block's tail */
379 	xfs_inode_t		*dp;		/* incore directory inode */
380 	int			error;		/* error return code */
381 	struct xfs_buf		*lbp;		/* leaf block's buffer */
382 	xfs_dir2_db_t		ldb;		/* leaf block's bno */
383 	xfs_dir2_leaf_t		*leaf;		/* leaf structure */
384 	xfs_dir2_leaf_tail_t	*ltp;		/* leaf's tail */
385 	int			needlog;	/* need to log block header */
386 	int			needscan;	/* need to rescan bestfree */
387 	xfs_trans_t		*tp;		/* transaction pointer */
388 	struct xfs_dir2_data_free *bf;
389 	struct xfs_dir2_leaf_entry *ents;
390 	struct xfs_dir3_icleaf_hdr leafhdr;
391 
392 	trace_xfs_dir2_block_to_leaf(args);
393 
394 	dp = args->dp;
395 	tp = args->trans;
396 	/*
397 	 * Add the leaf block to the inode.
398 	 * This interface will only put blocks in the leaf/node range.
399 	 * Since that's empty now, we'll get the root (block 0 in range).
400 	 */
401 	if ((error = xfs_da_grow_inode(args, &blkno))) {
402 		return error;
403 	}
404 	ldb = xfs_dir2_da_to_db(args->geo, blkno);
405 	ASSERT(ldb == xfs_dir2_byte_to_db(args->geo, XFS_DIR2_LEAF_OFFSET));
406 	/*
407 	 * Initialize the leaf block, get a buffer for it.
408 	 */
409 	error = xfs_dir3_leaf_get_buf(args, ldb, &lbp, XFS_DIR2_LEAF1_MAGIC);
410 	if (error)
411 		return error;
412 
413 	leaf = lbp->b_addr;
414 	hdr = dbp->b_addr;
415 	xfs_dir3_data_check(dp, dbp);
416 	btp = xfs_dir2_block_tail_p(args->geo, hdr);
417 	blp = xfs_dir2_block_leaf_p(btp);
418 	bf = dp->d_ops->data_bestfree_p(hdr);
419 	ents = dp->d_ops->leaf_ents_p(leaf);
420 
421 	/*
422 	 * Set the counts in the leaf header.
423 	 */
424 	dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
425 	leafhdr.count = be32_to_cpu(btp->count);
426 	leafhdr.stale = be32_to_cpu(btp->stale);
427 	dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
428 	xfs_dir3_leaf_log_header(args, lbp);
429 
430 	/*
431 	 * Could compact these but I think we always do the conversion
432 	 * after squeezing out stale entries.
433 	 */
434 	memcpy(ents, blp, be32_to_cpu(btp->count) * sizeof(xfs_dir2_leaf_entry_t));
435 	xfs_dir3_leaf_log_ents(args, lbp, 0, leafhdr.count - 1);
436 	needscan = 0;
437 	needlog = 1;
438 	/*
439 	 * Make the space formerly occupied by the leaf entries and block
440 	 * tail be free.
441 	 */
442 	xfs_dir2_data_make_free(args, dbp,
443 		(xfs_dir2_data_aoff_t)((char *)blp - (char *)hdr),
444 		(xfs_dir2_data_aoff_t)((char *)hdr + args->geo->blksize -
445 				       (char *)blp),
446 		&needlog, &needscan);
447 	/*
448 	 * Fix up the block header, make it a data block.
449 	 */
450 	dbp->b_ops = &xfs_dir3_data_buf_ops;
451 	xfs_trans_buf_set_type(tp, dbp, XFS_BLFT_DIR_DATA_BUF);
452 	if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC))
453 		hdr->magic = cpu_to_be32(XFS_DIR2_DATA_MAGIC);
454 	else
455 		hdr->magic = cpu_to_be32(XFS_DIR3_DATA_MAGIC);
456 
457 	if (needscan)
458 		xfs_dir2_data_freescan(dp, hdr, &needlog);
459 	/*
460 	 * Set up leaf tail and bests table.
461 	 */
462 	ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
463 	ltp->bestcount = cpu_to_be32(1);
464 	bestsp = xfs_dir2_leaf_bests_p(ltp);
465 	bestsp[0] =  bf[0].length;
466 	/*
467 	 * Log the data header and leaf bests table.
468 	 */
469 	if (needlog)
470 		xfs_dir2_data_log_header(args, dbp);
471 	xfs_dir3_leaf_check(dp, lbp);
472 	xfs_dir3_data_check(dp, dbp);
473 	xfs_dir3_leaf_log_bests(args, lbp, 0, 0);
474 	return 0;
475 }
476 
477 STATIC void
478 xfs_dir3_leaf_find_stale(
479 	struct xfs_dir3_icleaf_hdr *leafhdr,
480 	struct xfs_dir2_leaf_entry *ents,
481 	int			index,
482 	int			*lowstale,
483 	int			*highstale)
484 {
485 	/*
486 	 * Find the first stale entry before our index, if any.
487 	 */
488 	for (*lowstale = index - 1; *lowstale >= 0; --*lowstale) {
489 		if (ents[*lowstale].address ==
490 		    cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
491 			break;
492 	}
493 
494 	/*
495 	 * Find the first stale entry at or after our index, if any.
496 	 * Stop if the result would require moving more entries than using
497 	 * lowstale.
498 	 */
499 	for (*highstale = index; *highstale < leafhdr->count; ++*highstale) {
500 		if (ents[*highstale].address ==
501 		    cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
502 			break;
503 		if (*lowstale >= 0 && index - *lowstale <= *highstale - index)
504 			break;
505 	}
506 }
507 
508 struct xfs_dir2_leaf_entry *
509 xfs_dir3_leaf_find_entry(
510 	struct xfs_dir3_icleaf_hdr *leafhdr,
511 	struct xfs_dir2_leaf_entry *ents,
512 	int			index,		/* leaf table position */
513 	int			compact,	/* need to compact leaves */
514 	int			lowstale,	/* index of prev stale leaf */
515 	int			highstale,	/* index of next stale leaf */
516 	int			*lfloglow,	/* low leaf logging index */
517 	int			*lfloghigh)	/* high leaf logging index */
518 {
519 	if (!leafhdr->stale) {
520 		xfs_dir2_leaf_entry_t	*lep;	/* leaf entry table pointer */
521 
522 		/*
523 		 * Now we need to make room to insert the leaf entry.
524 		 *
525 		 * If there are no stale entries, just insert a hole at index.
526 		 */
527 		lep = &ents[index];
528 		if (index < leafhdr->count)
529 			memmove(lep + 1, lep,
530 				(leafhdr->count - index) * sizeof(*lep));
531 
532 		/*
533 		 * Record low and high logging indices for the leaf.
534 		 */
535 		*lfloglow = index;
536 		*lfloghigh = leafhdr->count++;
537 		return lep;
538 	}
539 
540 	/*
541 	 * There are stale entries.
542 	 *
543 	 * We will use one of them for the new entry.  It's probably not at
544 	 * the right location, so we'll have to shift some up or down first.
545 	 *
546 	 * If we didn't compact before, we need to find the nearest stale
547 	 * entries before and after our insertion point.
548 	 */
549 	if (compact == 0)
550 		xfs_dir3_leaf_find_stale(leafhdr, ents, index,
551 					 &lowstale, &highstale);
552 
553 	/*
554 	 * If the low one is better, use it.
555 	 */
556 	if (lowstale >= 0 &&
557 	    (highstale == leafhdr->count ||
558 	     index - lowstale - 1 < highstale - index)) {
559 		ASSERT(index - lowstale - 1 >= 0);
560 		ASSERT(ents[lowstale].address ==
561 		       cpu_to_be32(XFS_DIR2_NULL_DATAPTR));
562 
563 		/*
564 		 * Copy entries up to cover the stale entry and make room
565 		 * for the new entry.
566 		 */
567 		if (index - lowstale - 1 > 0) {
568 			memmove(&ents[lowstale], &ents[lowstale + 1],
569 				(index - lowstale - 1) *
570 					sizeof(xfs_dir2_leaf_entry_t));
571 		}
572 		*lfloglow = MIN(lowstale, *lfloglow);
573 		*lfloghigh = MAX(index - 1, *lfloghigh);
574 		leafhdr->stale--;
575 		return &ents[index - 1];
576 	}
577 
578 	/*
579 	 * The high one is better, so use that one.
580 	 */
581 	ASSERT(highstale - index >= 0);
582 	ASSERT(ents[highstale].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR));
583 
584 	/*
585 	 * Copy entries down to cover the stale entry and make room for the
586 	 * new entry.
587 	 */
588 	if (highstale - index > 0) {
589 		memmove(&ents[index + 1], &ents[index],
590 			(highstale - index) * sizeof(xfs_dir2_leaf_entry_t));
591 	}
592 	*lfloglow = MIN(index, *lfloglow);
593 	*lfloghigh = MAX(highstale, *lfloghigh);
594 	leafhdr->stale--;
595 	return &ents[index];
596 }
597 
598 /*
599  * Add an entry to a leaf form directory.
600  */
601 int						/* error */
602 xfs_dir2_leaf_addname(
603 	xfs_da_args_t		*args)		/* operation arguments */
604 {
605 	__be16			*bestsp;	/* freespace table in leaf */
606 	int			compact;	/* need to compact leaves */
607 	xfs_dir2_data_hdr_t	*hdr;		/* data block header */
608 	struct xfs_buf		*dbp;		/* data block buffer */
609 	xfs_dir2_data_entry_t	*dep;		/* data block entry */
610 	xfs_inode_t		*dp;		/* incore directory inode */
611 	xfs_dir2_data_unused_t	*dup;		/* data unused entry */
612 	int			error;		/* error return value */
613 	int			grown;		/* allocated new data block */
614 	int			highstale;	/* index of next stale leaf */
615 	int			i;		/* temporary, index */
616 	int			index;		/* leaf table position */
617 	struct xfs_buf		*lbp;		/* leaf's buffer */
618 	xfs_dir2_leaf_t		*leaf;		/* leaf structure */
619 	int			length;		/* length of new entry */
620 	xfs_dir2_leaf_entry_t	*lep;		/* leaf entry table pointer */
621 	int			lfloglow;	/* low leaf logging index */
622 	int			lfloghigh;	/* high leaf logging index */
623 	int			lowstale;	/* index of prev stale leaf */
624 	xfs_dir2_leaf_tail_t	*ltp;		/* leaf tail pointer */
625 	int			needbytes;	/* leaf block bytes needed */
626 	int			needlog;	/* need to log data header */
627 	int			needscan;	/* need to rescan data free */
628 	__be16			*tagp;		/* end of data entry */
629 	xfs_trans_t		*tp;		/* transaction pointer */
630 	xfs_dir2_db_t		use_block;	/* data block number */
631 	struct xfs_dir2_data_free *bf;		/* bestfree table */
632 	struct xfs_dir2_leaf_entry *ents;
633 	struct xfs_dir3_icleaf_hdr leafhdr;
634 
635 	trace_xfs_dir2_leaf_addname(args);
636 
637 	dp = args->dp;
638 	tp = args->trans;
639 
640 	error = xfs_dir3_leaf_read(tp, dp, args->geo->leafblk, -1, &lbp);
641 	if (error)
642 		return error;
643 
644 	/*
645 	 * Look up the entry by hash value and name.
646 	 * We know it's not there, our caller has already done a lookup.
647 	 * So the index is of the entry to insert in front of.
648 	 * But if there are dup hash values the index is of the first of those.
649 	 */
650 	index = xfs_dir2_leaf_search_hash(args, lbp);
651 	leaf = lbp->b_addr;
652 	ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
653 	ents = dp->d_ops->leaf_ents_p(leaf);
654 	dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
655 	bestsp = xfs_dir2_leaf_bests_p(ltp);
656 	length = dp->d_ops->data_entsize(args->namelen);
657 
658 	/*
659 	 * See if there are any entries with the same hash value
660 	 * and space in their block for the new entry.
661 	 * This is good because it puts multiple same-hash value entries
662 	 * in a data block, improving the lookup of those entries.
663 	 */
664 	for (use_block = -1, lep = &ents[index];
665 	     index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
666 	     index++, lep++) {
667 		if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
668 			continue;
669 		i = xfs_dir2_dataptr_to_db(args->geo, be32_to_cpu(lep->address));
670 		ASSERT(i < be32_to_cpu(ltp->bestcount));
671 		ASSERT(bestsp[i] != cpu_to_be16(NULLDATAOFF));
672 		if (be16_to_cpu(bestsp[i]) >= length) {
673 			use_block = i;
674 			break;
675 		}
676 	}
677 	/*
678 	 * Didn't find a block yet, linear search all the data blocks.
679 	 */
680 	if (use_block == -1) {
681 		for (i = 0; i < be32_to_cpu(ltp->bestcount); i++) {
682 			/*
683 			 * Remember a block we see that's missing.
684 			 */
685 			if (bestsp[i] == cpu_to_be16(NULLDATAOFF) &&
686 			    use_block == -1)
687 				use_block = i;
688 			else if (be16_to_cpu(bestsp[i]) >= length) {
689 				use_block = i;
690 				break;
691 			}
692 		}
693 	}
694 	/*
695 	 * How many bytes do we need in the leaf block?
696 	 */
697 	needbytes = 0;
698 	if (!leafhdr.stale)
699 		needbytes += sizeof(xfs_dir2_leaf_entry_t);
700 	if (use_block == -1)
701 		needbytes += sizeof(xfs_dir2_data_off_t);
702 
703 	/*
704 	 * Now kill use_block if it refers to a missing block, so we
705 	 * can use it as an indication of allocation needed.
706 	 */
707 	if (use_block != -1 && bestsp[use_block] == cpu_to_be16(NULLDATAOFF))
708 		use_block = -1;
709 	/*
710 	 * If we don't have enough free bytes but we can make enough
711 	 * by compacting out stale entries, we'll do that.
712 	 */
713 	if ((char *)bestsp - (char *)&ents[leafhdr.count] < needbytes &&
714 	    leafhdr.stale > 1)
715 		compact = 1;
716 
717 	/*
718 	 * Otherwise if we don't have enough free bytes we need to
719 	 * convert to node form.
720 	 */
721 	else if ((char *)bestsp - (char *)&ents[leafhdr.count] < needbytes) {
722 		/*
723 		 * Just checking or no space reservation, give up.
724 		 */
725 		if ((args->op_flags & XFS_DA_OP_JUSTCHECK) ||
726 							args->total == 0) {
727 			xfs_trans_brelse(tp, lbp);
728 			return -ENOSPC;
729 		}
730 		/*
731 		 * Convert to node form.
732 		 */
733 		error = xfs_dir2_leaf_to_node(args, lbp);
734 		if (error)
735 			return error;
736 		/*
737 		 * Then add the new entry.
738 		 */
739 		return xfs_dir2_node_addname(args);
740 	}
741 	/*
742 	 * Otherwise it will fit without compaction.
743 	 */
744 	else
745 		compact = 0;
746 	/*
747 	 * If just checking, then it will fit unless we needed to allocate
748 	 * a new data block.
749 	 */
750 	if (args->op_flags & XFS_DA_OP_JUSTCHECK) {
751 		xfs_trans_brelse(tp, lbp);
752 		return use_block == -1 ? -ENOSPC : 0;
753 	}
754 	/*
755 	 * If no allocations are allowed, return now before we've
756 	 * changed anything.
757 	 */
758 	if (args->total == 0 && use_block == -1) {
759 		xfs_trans_brelse(tp, lbp);
760 		return -ENOSPC;
761 	}
762 	/*
763 	 * Need to compact the leaf entries, removing stale ones.
764 	 * Leave one stale entry behind - the one closest to our
765 	 * insertion index - and we'll shift that one to our insertion
766 	 * point later.
767 	 */
768 	if (compact) {
769 		xfs_dir3_leaf_compact_x1(&leafhdr, ents, &index, &lowstale,
770 			&highstale, &lfloglow, &lfloghigh);
771 	}
772 	/*
773 	 * There are stale entries, so we'll need log-low and log-high
774 	 * impossibly bad values later.
775 	 */
776 	else if (leafhdr.stale) {
777 		lfloglow = leafhdr.count;
778 		lfloghigh = -1;
779 	}
780 	/*
781 	 * If there was no data block space found, we need to allocate
782 	 * a new one.
783 	 */
784 	if (use_block == -1) {
785 		/*
786 		 * Add the new data block.
787 		 */
788 		if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE,
789 				&use_block))) {
790 			xfs_trans_brelse(tp, lbp);
791 			return error;
792 		}
793 		/*
794 		 * Initialize the block.
795 		 */
796 		if ((error = xfs_dir3_data_init(args, use_block, &dbp))) {
797 			xfs_trans_brelse(tp, lbp);
798 			return error;
799 		}
800 		/*
801 		 * If we're adding a new data block on the end we need to
802 		 * extend the bests table.  Copy it up one entry.
803 		 */
804 		if (use_block >= be32_to_cpu(ltp->bestcount)) {
805 			bestsp--;
806 			memmove(&bestsp[0], &bestsp[1],
807 				be32_to_cpu(ltp->bestcount) * sizeof(bestsp[0]));
808 			be32_add_cpu(&ltp->bestcount, 1);
809 			xfs_dir3_leaf_log_tail(args, lbp);
810 			xfs_dir3_leaf_log_bests(args, lbp, 0,
811 						be32_to_cpu(ltp->bestcount) - 1);
812 		}
813 		/*
814 		 * If we're filling in a previously empty block just log it.
815 		 */
816 		else
817 			xfs_dir3_leaf_log_bests(args, lbp, use_block, use_block);
818 		hdr = dbp->b_addr;
819 		bf = dp->d_ops->data_bestfree_p(hdr);
820 		bestsp[use_block] = bf[0].length;
821 		grown = 1;
822 	} else {
823 		/*
824 		 * Already had space in some data block.
825 		 * Just read that one in.
826 		 */
827 		error = xfs_dir3_data_read(tp, dp,
828 				   xfs_dir2_db_to_da(args->geo, use_block),
829 				   -1, &dbp);
830 		if (error) {
831 			xfs_trans_brelse(tp, lbp);
832 			return error;
833 		}
834 		hdr = dbp->b_addr;
835 		bf = dp->d_ops->data_bestfree_p(hdr);
836 		grown = 0;
837 	}
838 	/*
839 	 * Point to the biggest freespace in our data block.
840 	 */
841 	dup = (xfs_dir2_data_unused_t *)
842 	      ((char *)hdr + be16_to_cpu(bf[0].offset));
843 	ASSERT(be16_to_cpu(dup->length) >= length);
844 	needscan = needlog = 0;
845 	/*
846 	 * Mark the initial part of our freespace in use for the new entry.
847 	 */
848 	xfs_dir2_data_use_free(args, dbp, dup,
849 		(xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr), length,
850 		&needlog, &needscan);
851 	/*
852 	 * Initialize our new entry (at last).
853 	 */
854 	dep = (xfs_dir2_data_entry_t *)dup;
855 	dep->inumber = cpu_to_be64(args->inumber);
856 	dep->namelen = args->namelen;
857 	memcpy(dep->name, args->name, dep->namelen);
858 	dp->d_ops->data_put_ftype(dep, args->filetype);
859 	tagp = dp->d_ops->data_entry_tag_p(dep);
860 	*tagp = cpu_to_be16((char *)dep - (char *)hdr);
861 	/*
862 	 * Need to scan fix up the bestfree table.
863 	 */
864 	if (needscan)
865 		xfs_dir2_data_freescan(dp, hdr, &needlog);
866 	/*
867 	 * Need to log the data block's header.
868 	 */
869 	if (needlog)
870 		xfs_dir2_data_log_header(args, dbp);
871 	xfs_dir2_data_log_entry(args, dbp, dep);
872 	/*
873 	 * If the bests table needs to be changed, do it.
874 	 * Log the change unless we've already done that.
875 	 */
876 	if (be16_to_cpu(bestsp[use_block]) != be16_to_cpu(bf[0].length)) {
877 		bestsp[use_block] = bf[0].length;
878 		if (!grown)
879 			xfs_dir3_leaf_log_bests(args, lbp, use_block, use_block);
880 	}
881 
882 	lep = xfs_dir3_leaf_find_entry(&leafhdr, ents, index, compact, lowstale,
883 				       highstale, &lfloglow, &lfloghigh);
884 
885 	/*
886 	 * Fill in the new leaf entry.
887 	 */
888 	lep->hashval = cpu_to_be32(args->hashval);
889 	lep->address = cpu_to_be32(
890 				xfs_dir2_db_off_to_dataptr(args->geo, use_block,
891 				be16_to_cpu(*tagp)));
892 	/*
893 	 * Log the leaf fields and give up the buffers.
894 	 */
895 	dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
896 	xfs_dir3_leaf_log_header(args, lbp);
897 	xfs_dir3_leaf_log_ents(args, lbp, lfloglow, lfloghigh);
898 	xfs_dir3_leaf_check(dp, lbp);
899 	xfs_dir3_data_check(dp, dbp);
900 	return 0;
901 }
902 
903 /*
904  * Compact out any stale entries in the leaf.
905  * Log the header and changed leaf entries, if any.
906  */
907 void
908 xfs_dir3_leaf_compact(
909 	xfs_da_args_t	*args,		/* operation arguments */
910 	struct xfs_dir3_icleaf_hdr *leafhdr,
911 	struct xfs_buf	*bp)		/* leaf buffer */
912 {
913 	int		from;		/* source leaf index */
914 	xfs_dir2_leaf_t	*leaf;		/* leaf structure */
915 	int		loglow;		/* first leaf entry to log */
916 	int		to;		/* target leaf index */
917 	struct xfs_dir2_leaf_entry *ents;
918 	struct xfs_inode *dp = args->dp;
919 
920 	leaf = bp->b_addr;
921 	if (!leafhdr->stale)
922 		return;
923 
924 	/*
925 	 * Compress out the stale entries in place.
926 	 */
927 	ents = dp->d_ops->leaf_ents_p(leaf);
928 	for (from = to = 0, loglow = -1; from < leafhdr->count; from++) {
929 		if (ents[from].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
930 			continue;
931 		/*
932 		 * Only actually copy the entries that are different.
933 		 */
934 		if (from > to) {
935 			if (loglow == -1)
936 				loglow = to;
937 			ents[to] = ents[from];
938 		}
939 		to++;
940 	}
941 	/*
942 	 * Update and log the header, log the leaf entries.
943 	 */
944 	ASSERT(leafhdr->stale == from - to);
945 	leafhdr->count -= leafhdr->stale;
946 	leafhdr->stale = 0;
947 
948 	dp->d_ops->leaf_hdr_to_disk(leaf, leafhdr);
949 	xfs_dir3_leaf_log_header(args, bp);
950 	if (loglow != -1)
951 		xfs_dir3_leaf_log_ents(args, bp, loglow, to - 1);
952 }
953 
954 /*
955  * Compact the leaf entries, removing stale ones.
956  * Leave one stale entry behind - the one closest to our
957  * insertion index - and the caller will shift that one to our insertion
958  * point later.
959  * Return new insertion index, where the remaining stale entry is,
960  * and leaf logging indices.
961  */
962 void
963 xfs_dir3_leaf_compact_x1(
964 	struct xfs_dir3_icleaf_hdr *leafhdr,
965 	struct xfs_dir2_leaf_entry *ents,
966 	int		*indexp,	/* insertion index */
967 	int		*lowstalep,	/* out: stale entry before us */
968 	int		*highstalep,	/* out: stale entry after us */
969 	int		*lowlogp,	/* out: low log index */
970 	int		*highlogp)	/* out: high log index */
971 {
972 	int		from;		/* source copy index */
973 	int		highstale;	/* stale entry at/after index */
974 	int		index;		/* insertion index */
975 	int		keepstale;	/* source index of kept stale */
976 	int		lowstale;	/* stale entry before index */
977 	int		newindex=0;	/* new insertion index */
978 	int		to;		/* destination copy index */
979 
980 	ASSERT(leafhdr->stale > 1);
981 	index = *indexp;
982 
983 	xfs_dir3_leaf_find_stale(leafhdr, ents, index, &lowstale, &highstale);
984 
985 	/*
986 	 * Pick the better of lowstale and highstale.
987 	 */
988 	if (lowstale >= 0 &&
989 	    (highstale == leafhdr->count ||
990 	     index - lowstale <= highstale - index))
991 		keepstale = lowstale;
992 	else
993 		keepstale = highstale;
994 	/*
995 	 * Copy the entries in place, removing all the stale entries
996 	 * except keepstale.
997 	 */
998 	for (from = to = 0; from < leafhdr->count; from++) {
999 		/*
1000 		 * Notice the new value of index.
1001 		 */
1002 		if (index == from)
1003 			newindex = to;
1004 		if (from != keepstale &&
1005 		    ents[from].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR)) {
1006 			if (from == to)
1007 				*lowlogp = to;
1008 			continue;
1009 		}
1010 		/*
1011 		 * Record the new keepstale value for the insertion.
1012 		 */
1013 		if (from == keepstale)
1014 			lowstale = highstale = to;
1015 		/*
1016 		 * Copy only the entries that have moved.
1017 		 */
1018 		if (from > to)
1019 			ents[to] = ents[from];
1020 		to++;
1021 	}
1022 	ASSERT(from > to);
1023 	/*
1024 	 * If the insertion point was past the last entry,
1025 	 * set the new insertion point accordingly.
1026 	 */
1027 	if (index == from)
1028 		newindex = to;
1029 	*indexp = newindex;
1030 	/*
1031 	 * Adjust the leaf header values.
1032 	 */
1033 	leafhdr->count -= from - to;
1034 	leafhdr->stale = 1;
1035 	/*
1036 	 * Remember the low/high stale value only in the "right"
1037 	 * direction.
1038 	 */
1039 	if (lowstale >= newindex)
1040 		lowstale = -1;
1041 	else
1042 		highstale = leafhdr->count;
1043 	*highlogp = leafhdr->count - 1;
1044 	*lowstalep = lowstale;
1045 	*highstalep = highstale;
1046 }
1047 
1048 /*
1049  * Log the bests entries indicated from a leaf1 block.
1050  */
1051 static void
1052 xfs_dir3_leaf_log_bests(
1053 	struct xfs_da_args	*args,
1054 	struct xfs_buf		*bp,		/* leaf buffer */
1055 	int			first,		/* first entry to log */
1056 	int			last)		/* last entry to log */
1057 {
1058 	__be16			*firstb;	/* pointer to first entry */
1059 	__be16			*lastb;		/* pointer to last entry */
1060 	struct xfs_dir2_leaf	*leaf = bp->b_addr;
1061 	xfs_dir2_leaf_tail_t	*ltp;		/* leaf tail structure */
1062 
1063 	ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1064 	       leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC));
1065 
1066 	ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1067 	firstb = xfs_dir2_leaf_bests_p(ltp) + first;
1068 	lastb = xfs_dir2_leaf_bests_p(ltp) + last;
1069 	xfs_trans_log_buf(args->trans, bp,
1070 		(uint)((char *)firstb - (char *)leaf),
1071 		(uint)((char *)lastb - (char *)leaf + sizeof(*lastb) - 1));
1072 }
1073 
1074 /*
1075  * Log the leaf entries indicated from a leaf1 or leafn block.
1076  */
1077 void
1078 xfs_dir3_leaf_log_ents(
1079 	struct xfs_da_args	*args,
1080 	struct xfs_buf		*bp,
1081 	int			first,
1082 	int			last)
1083 {
1084 	xfs_dir2_leaf_entry_t	*firstlep;	/* pointer to first entry */
1085 	xfs_dir2_leaf_entry_t	*lastlep;	/* pointer to last entry */
1086 	struct xfs_dir2_leaf	*leaf = bp->b_addr;
1087 	struct xfs_dir2_leaf_entry *ents;
1088 
1089 	ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1090 	       leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1091 	       leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1092 	       leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1093 
1094 	ents = args->dp->d_ops->leaf_ents_p(leaf);
1095 	firstlep = &ents[first];
1096 	lastlep = &ents[last];
1097 	xfs_trans_log_buf(args->trans, bp,
1098 		(uint)((char *)firstlep - (char *)leaf),
1099 		(uint)((char *)lastlep - (char *)leaf + sizeof(*lastlep) - 1));
1100 }
1101 
1102 /*
1103  * Log the header of the leaf1 or leafn block.
1104  */
1105 void
1106 xfs_dir3_leaf_log_header(
1107 	struct xfs_da_args	*args,
1108 	struct xfs_buf		*bp)
1109 {
1110 	struct xfs_dir2_leaf	*leaf = bp->b_addr;
1111 
1112 	ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1113 	       leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1114 	       leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1115 	       leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1116 
1117 	xfs_trans_log_buf(args->trans, bp,
1118 			  (uint)((char *)&leaf->hdr - (char *)leaf),
1119 			  args->dp->d_ops->leaf_hdr_size - 1);
1120 }
1121 
1122 /*
1123  * Log the tail of the leaf1 block.
1124  */
1125 STATIC void
1126 xfs_dir3_leaf_log_tail(
1127 	struct xfs_da_args	*args,
1128 	struct xfs_buf		*bp)
1129 {
1130 	struct xfs_dir2_leaf	*leaf = bp->b_addr;
1131 	xfs_dir2_leaf_tail_t	*ltp;		/* leaf tail structure */
1132 
1133 	ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1134 	       leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1135 	       leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1136 	       leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1137 
1138 	ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1139 	xfs_trans_log_buf(args->trans, bp, (uint)((char *)ltp - (char *)leaf),
1140 		(uint)(args->geo->blksize - 1));
1141 }
1142 
1143 /*
1144  * Look up the entry referred to by args in the leaf format directory.
1145  * Most of the work is done by the xfs_dir2_leaf_lookup_int routine which
1146  * is also used by the node-format code.
1147  */
1148 int
1149 xfs_dir2_leaf_lookup(
1150 	xfs_da_args_t		*args)		/* operation arguments */
1151 {
1152 	struct xfs_buf		*dbp;		/* data block buffer */
1153 	xfs_dir2_data_entry_t	*dep;		/* data block entry */
1154 	xfs_inode_t		*dp;		/* incore directory inode */
1155 	int			error;		/* error return code */
1156 	int			index;		/* found entry index */
1157 	struct xfs_buf		*lbp;		/* leaf buffer */
1158 	xfs_dir2_leaf_t		*leaf;		/* leaf structure */
1159 	xfs_dir2_leaf_entry_t	*lep;		/* leaf entry */
1160 	xfs_trans_t		*tp;		/* transaction pointer */
1161 	struct xfs_dir2_leaf_entry *ents;
1162 
1163 	trace_xfs_dir2_leaf_lookup(args);
1164 
1165 	/*
1166 	 * Look up name in the leaf block, returning both buffers and index.
1167 	 */
1168 	if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1169 		return error;
1170 	}
1171 	tp = args->trans;
1172 	dp = args->dp;
1173 	xfs_dir3_leaf_check(dp, lbp);
1174 	leaf = lbp->b_addr;
1175 	ents = dp->d_ops->leaf_ents_p(leaf);
1176 	/*
1177 	 * Get to the leaf entry and contained data entry address.
1178 	 */
1179 	lep = &ents[index];
1180 
1181 	/*
1182 	 * Point to the data entry.
1183 	 */
1184 	dep = (xfs_dir2_data_entry_t *)
1185 	      ((char *)dbp->b_addr +
1186 	       xfs_dir2_dataptr_to_off(args->geo, be32_to_cpu(lep->address)));
1187 	/*
1188 	 * Return the found inode number & CI name if appropriate
1189 	 */
1190 	args->inumber = be64_to_cpu(dep->inumber);
1191 	args->filetype = dp->d_ops->data_get_ftype(dep);
1192 	error = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
1193 	xfs_trans_brelse(tp, dbp);
1194 	xfs_trans_brelse(tp, lbp);
1195 	return error;
1196 }
1197 
1198 /*
1199  * Look up name/hash in the leaf block.
1200  * Fill in indexp with the found index, and dbpp with the data buffer.
1201  * If not found dbpp will be NULL, and ENOENT comes back.
1202  * lbpp will always be filled in with the leaf buffer unless there's an error.
1203  */
1204 static int					/* error */
1205 xfs_dir2_leaf_lookup_int(
1206 	xfs_da_args_t		*args,		/* operation arguments */
1207 	struct xfs_buf		**lbpp,		/* out: leaf buffer */
1208 	int			*indexp,	/* out: index in leaf block */
1209 	struct xfs_buf		**dbpp)		/* out: data buffer */
1210 {
1211 	xfs_dir2_db_t		curdb = -1;	/* current data block number */
1212 	struct xfs_buf		*dbp = NULL;	/* data buffer */
1213 	xfs_dir2_data_entry_t	*dep;		/* data entry */
1214 	xfs_inode_t		*dp;		/* incore directory inode */
1215 	int			error;		/* error return code */
1216 	int			index;		/* index in leaf block */
1217 	struct xfs_buf		*lbp;		/* leaf buffer */
1218 	xfs_dir2_leaf_entry_t	*lep;		/* leaf entry */
1219 	xfs_dir2_leaf_t		*leaf;		/* leaf structure */
1220 	xfs_mount_t		*mp;		/* filesystem mount point */
1221 	xfs_dir2_db_t		newdb;		/* new data block number */
1222 	xfs_trans_t		*tp;		/* transaction pointer */
1223 	xfs_dir2_db_t		cidb = -1;	/* case match data block no. */
1224 	enum xfs_dacmp		cmp;		/* name compare result */
1225 	struct xfs_dir2_leaf_entry *ents;
1226 	struct xfs_dir3_icleaf_hdr leafhdr;
1227 
1228 	dp = args->dp;
1229 	tp = args->trans;
1230 	mp = dp->i_mount;
1231 
1232 	error = xfs_dir3_leaf_read(tp, dp, args->geo->leafblk, -1, &lbp);
1233 	if (error)
1234 		return error;
1235 
1236 	*lbpp = lbp;
1237 	leaf = lbp->b_addr;
1238 	xfs_dir3_leaf_check(dp, lbp);
1239 	ents = dp->d_ops->leaf_ents_p(leaf);
1240 	dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1241 
1242 	/*
1243 	 * Look for the first leaf entry with our hash value.
1244 	 */
1245 	index = xfs_dir2_leaf_search_hash(args, lbp);
1246 	/*
1247 	 * Loop over all the entries with the right hash value
1248 	 * looking to match the name.
1249 	 */
1250 	for (lep = &ents[index];
1251 	     index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
1252 	     lep++, index++) {
1253 		/*
1254 		 * Skip over stale leaf entries.
1255 		 */
1256 		if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
1257 			continue;
1258 		/*
1259 		 * Get the new data block number.
1260 		 */
1261 		newdb = xfs_dir2_dataptr_to_db(args->geo,
1262 					       be32_to_cpu(lep->address));
1263 		/*
1264 		 * If it's not the same as the old data block number,
1265 		 * need to pitch the old one and read the new one.
1266 		 */
1267 		if (newdb != curdb) {
1268 			if (dbp)
1269 				xfs_trans_brelse(tp, dbp);
1270 			error = xfs_dir3_data_read(tp, dp,
1271 					   xfs_dir2_db_to_da(args->geo, newdb),
1272 					   -1, &dbp);
1273 			if (error) {
1274 				xfs_trans_brelse(tp, lbp);
1275 				return error;
1276 			}
1277 			curdb = newdb;
1278 		}
1279 		/*
1280 		 * Point to the data entry.
1281 		 */
1282 		dep = (xfs_dir2_data_entry_t *)((char *)dbp->b_addr +
1283 			xfs_dir2_dataptr_to_off(args->geo,
1284 						be32_to_cpu(lep->address)));
1285 		/*
1286 		 * Compare name and if it's an exact match, return the index
1287 		 * and buffer. If it's the first case-insensitive match, store
1288 		 * the index and buffer and continue looking for an exact match.
1289 		 */
1290 		cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
1291 		if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
1292 			args->cmpresult = cmp;
1293 			*indexp = index;
1294 			/* case exact match: return the current buffer. */
1295 			if (cmp == XFS_CMP_EXACT) {
1296 				*dbpp = dbp;
1297 				return 0;
1298 			}
1299 			cidb = curdb;
1300 		}
1301 	}
1302 	ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
1303 	/*
1304 	 * Here, we can only be doing a lookup (not a rename or remove).
1305 	 * If a case-insensitive match was found earlier, re-read the
1306 	 * appropriate data block if required and return it.
1307 	 */
1308 	if (args->cmpresult == XFS_CMP_CASE) {
1309 		ASSERT(cidb != -1);
1310 		if (cidb != curdb) {
1311 			xfs_trans_brelse(tp, dbp);
1312 			error = xfs_dir3_data_read(tp, dp,
1313 					   xfs_dir2_db_to_da(args->geo, cidb),
1314 					   -1, &dbp);
1315 			if (error) {
1316 				xfs_trans_brelse(tp, lbp);
1317 				return error;
1318 			}
1319 		}
1320 		*dbpp = dbp;
1321 		return 0;
1322 	}
1323 	/*
1324 	 * No match found, return -ENOENT.
1325 	 */
1326 	ASSERT(cidb == -1);
1327 	if (dbp)
1328 		xfs_trans_brelse(tp, dbp);
1329 	xfs_trans_brelse(tp, lbp);
1330 	return -ENOENT;
1331 }
1332 
1333 /*
1334  * Remove an entry from a leaf format directory.
1335  */
1336 int						/* error */
1337 xfs_dir2_leaf_removename(
1338 	xfs_da_args_t		*args)		/* operation arguments */
1339 {
1340 	__be16			*bestsp;	/* leaf block best freespace */
1341 	xfs_dir2_data_hdr_t	*hdr;		/* data block header */
1342 	xfs_dir2_db_t		db;		/* data block number */
1343 	struct xfs_buf		*dbp;		/* data block buffer */
1344 	xfs_dir2_data_entry_t	*dep;		/* data entry structure */
1345 	xfs_inode_t		*dp;		/* incore directory inode */
1346 	int			error;		/* error return code */
1347 	xfs_dir2_db_t		i;		/* temporary data block # */
1348 	int			index;		/* index into leaf entries */
1349 	struct xfs_buf		*lbp;		/* leaf buffer */
1350 	xfs_dir2_leaf_t		*leaf;		/* leaf structure */
1351 	xfs_dir2_leaf_entry_t	*lep;		/* leaf entry */
1352 	xfs_dir2_leaf_tail_t	*ltp;		/* leaf tail structure */
1353 	int			needlog;	/* need to log data header */
1354 	int			needscan;	/* need to rescan data frees */
1355 	xfs_dir2_data_off_t	oldbest;	/* old value of best free */
1356 	struct xfs_dir2_data_free *bf;		/* bestfree table */
1357 	struct xfs_dir2_leaf_entry *ents;
1358 	struct xfs_dir3_icleaf_hdr leafhdr;
1359 
1360 	trace_xfs_dir2_leaf_removename(args);
1361 
1362 	/*
1363 	 * Lookup the leaf entry, get the leaf and data blocks read in.
1364 	 */
1365 	if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1366 		return error;
1367 	}
1368 	dp = args->dp;
1369 	leaf = lbp->b_addr;
1370 	hdr = dbp->b_addr;
1371 	xfs_dir3_data_check(dp, dbp);
1372 	bf = dp->d_ops->data_bestfree_p(hdr);
1373 	dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1374 	ents = dp->d_ops->leaf_ents_p(leaf);
1375 	/*
1376 	 * Point to the leaf entry, use that to point to the data entry.
1377 	 */
1378 	lep = &ents[index];
1379 	db = xfs_dir2_dataptr_to_db(args->geo, be32_to_cpu(lep->address));
1380 	dep = (xfs_dir2_data_entry_t *)((char *)hdr +
1381 		xfs_dir2_dataptr_to_off(args->geo, be32_to_cpu(lep->address)));
1382 	needscan = needlog = 0;
1383 	oldbest = be16_to_cpu(bf[0].length);
1384 	ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1385 	bestsp = xfs_dir2_leaf_bests_p(ltp);
1386 	ASSERT(be16_to_cpu(bestsp[db]) == oldbest);
1387 	/*
1388 	 * Mark the former data entry unused.
1389 	 */
1390 	xfs_dir2_data_make_free(args, dbp,
1391 		(xfs_dir2_data_aoff_t)((char *)dep - (char *)hdr),
1392 		dp->d_ops->data_entsize(dep->namelen), &needlog, &needscan);
1393 	/*
1394 	 * We just mark the leaf entry stale by putting a null in it.
1395 	 */
1396 	leafhdr.stale++;
1397 	dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
1398 	xfs_dir3_leaf_log_header(args, lbp);
1399 
1400 	lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1401 	xfs_dir3_leaf_log_ents(args, lbp, index, index);
1402 
1403 	/*
1404 	 * Scan the freespace in the data block again if necessary,
1405 	 * log the data block header if necessary.
1406 	 */
1407 	if (needscan)
1408 		xfs_dir2_data_freescan(dp, hdr, &needlog);
1409 	if (needlog)
1410 		xfs_dir2_data_log_header(args, dbp);
1411 	/*
1412 	 * If the longest freespace in the data block has changed,
1413 	 * put the new value in the bests table and log that.
1414 	 */
1415 	if (be16_to_cpu(bf[0].length) != oldbest) {
1416 		bestsp[db] = bf[0].length;
1417 		xfs_dir3_leaf_log_bests(args, lbp, db, db);
1418 	}
1419 	xfs_dir3_data_check(dp, dbp);
1420 	/*
1421 	 * If the data block is now empty then get rid of the data block.
1422 	 */
1423 	if (be16_to_cpu(bf[0].length) ==
1424 			args->geo->blksize - dp->d_ops->data_entry_offset) {
1425 		ASSERT(db != args->geo->datablk);
1426 		if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1427 			/*
1428 			 * Nope, can't get rid of it because it caused
1429 			 * allocation of a bmap btree block to do so.
1430 			 * Just go on, returning success, leaving the
1431 			 * empty block in place.
1432 			 */
1433 			if (error == -ENOSPC && args->total == 0)
1434 				error = 0;
1435 			xfs_dir3_leaf_check(dp, lbp);
1436 			return error;
1437 		}
1438 		dbp = NULL;
1439 		/*
1440 		 * If this is the last data block then compact the
1441 		 * bests table by getting rid of entries.
1442 		 */
1443 		if (db == be32_to_cpu(ltp->bestcount) - 1) {
1444 			/*
1445 			 * Look for the last active entry (i).
1446 			 */
1447 			for (i = db - 1; i > 0; i--) {
1448 				if (bestsp[i] != cpu_to_be16(NULLDATAOFF))
1449 					break;
1450 			}
1451 			/*
1452 			 * Copy the table down so inactive entries at the
1453 			 * end are removed.
1454 			 */
1455 			memmove(&bestsp[db - i], bestsp,
1456 				(be32_to_cpu(ltp->bestcount) - (db - i)) * sizeof(*bestsp));
1457 			be32_add_cpu(&ltp->bestcount, -(db - i));
1458 			xfs_dir3_leaf_log_tail(args, lbp);
1459 			xfs_dir3_leaf_log_bests(args, lbp, 0,
1460 						be32_to_cpu(ltp->bestcount) - 1);
1461 		} else
1462 			bestsp[db] = cpu_to_be16(NULLDATAOFF);
1463 	}
1464 	/*
1465 	 * If the data block was not the first one, drop it.
1466 	 */
1467 	else if (db != args->geo->datablk)
1468 		dbp = NULL;
1469 
1470 	xfs_dir3_leaf_check(dp, lbp);
1471 	/*
1472 	 * See if we can convert to block form.
1473 	 */
1474 	return xfs_dir2_leaf_to_block(args, lbp, dbp);
1475 }
1476 
1477 /*
1478  * Replace the inode number in a leaf format directory entry.
1479  */
1480 int						/* error */
1481 xfs_dir2_leaf_replace(
1482 	xfs_da_args_t		*args)		/* operation arguments */
1483 {
1484 	struct xfs_buf		*dbp;		/* data block buffer */
1485 	xfs_dir2_data_entry_t	*dep;		/* data block entry */
1486 	xfs_inode_t		*dp;		/* incore directory inode */
1487 	int			error;		/* error return code */
1488 	int			index;		/* index of leaf entry */
1489 	struct xfs_buf		*lbp;		/* leaf buffer */
1490 	xfs_dir2_leaf_t		*leaf;		/* leaf structure */
1491 	xfs_dir2_leaf_entry_t	*lep;		/* leaf entry */
1492 	xfs_trans_t		*tp;		/* transaction pointer */
1493 	struct xfs_dir2_leaf_entry *ents;
1494 
1495 	trace_xfs_dir2_leaf_replace(args);
1496 
1497 	/*
1498 	 * Look up the entry.
1499 	 */
1500 	if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1501 		return error;
1502 	}
1503 	dp = args->dp;
1504 	leaf = lbp->b_addr;
1505 	ents = dp->d_ops->leaf_ents_p(leaf);
1506 	/*
1507 	 * Point to the leaf entry, get data address from it.
1508 	 */
1509 	lep = &ents[index];
1510 	/*
1511 	 * Point to the data entry.
1512 	 */
1513 	dep = (xfs_dir2_data_entry_t *)
1514 	      ((char *)dbp->b_addr +
1515 	       xfs_dir2_dataptr_to_off(args->geo, be32_to_cpu(lep->address)));
1516 	ASSERT(args->inumber != be64_to_cpu(dep->inumber));
1517 	/*
1518 	 * Put the new inode number in, log it.
1519 	 */
1520 	dep->inumber = cpu_to_be64(args->inumber);
1521 	dp->d_ops->data_put_ftype(dep, args->filetype);
1522 	tp = args->trans;
1523 	xfs_dir2_data_log_entry(args, dbp, dep);
1524 	xfs_dir3_leaf_check(dp, lbp);
1525 	xfs_trans_brelse(tp, lbp);
1526 	return 0;
1527 }
1528 
1529 /*
1530  * Return index in the leaf block (lbp) which is either the first
1531  * one with this hash value, or if there are none, the insert point
1532  * for that hash value.
1533  */
1534 int						/* index value */
1535 xfs_dir2_leaf_search_hash(
1536 	xfs_da_args_t		*args,		/* operation arguments */
1537 	struct xfs_buf		*lbp)		/* leaf buffer */
1538 {
1539 	xfs_dahash_t		hash=0;		/* hash from this entry */
1540 	xfs_dahash_t		hashwant;	/* hash value looking for */
1541 	int			high;		/* high leaf index */
1542 	int			low;		/* low leaf index */
1543 	xfs_dir2_leaf_t		*leaf;		/* leaf structure */
1544 	xfs_dir2_leaf_entry_t	*lep;		/* leaf entry */
1545 	int			mid=0;		/* current leaf index */
1546 	struct xfs_dir2_leaf_entry *ents;
1547 	struct xfs_dir3_icleaf_hdr leafhdr;
1548 
1549 	leaf = lbp->b_addr;
1550 	ents = args->dp->d_ops->leaf_ents_p(leaf);
1551 	args->dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1552 
1553 	/*
1554 	 * Note, the table cannot be empty, so we have to go through the loop.
1555 	 * Binary search the leaf entries looking for our hash value.
1556 	 */
1557 	for (lep = ents, low = 0, high = leafhdr.count - 1,
1558 		hashwant = args->hashval;
1559 	     low <= high; ) {
1560 		mid = (low + high) >> 1;
1561 		if ((hash = be32_to_cpu(lep[mid].hashval)) == hashwant)
1562 			break;
1563 		if (hash < hashwant)
1564 			low = mid + 1;
1565 		else
1566 			high = mid - 1;
1567 	}
1568 	/*
1569 	 * Found one, back up through all the equal hash values.
1570 	 */
1571 	if (hash == hashwant) {
1572 		while (mid > 0 && be32_to_cpu(lep[mid - 1].hashval) == hashwant) {
1573 			mid--;
1574 		}
1575 	}
1576 	/*
1577 	 * Need to point to an entry higher than ours.
1578 	 */
1579 	else if (hash < hashwant)
1580 		mid++;
1581 	return mid;
1582 }
1583 
1584 /*
1585  * Trim off a trailing data block.  We know it's empty since the leaf
1586  * freespace table says so.
1587  */
1588 int						/* error */
1589 xfs_dir2_leaf_trim_data(
1590 	xfs_da_args_t		*args,		/* operation arguments */
1591 	struct xfs_buf		*lbp,		/* leaf buffer */
1592 	xfs_dir2_db_t		db)		/* data block number */
1593 {
1594 	__be16			*bestsp;	/* leaf bests table */
1595 	struct xfs_buf		*dbp;		/* data block buffer */
1596 	xfs_inode_t		*dp;		/* incore directory inode */
1597 	int			error;		/* error return value */
1598 	xfs_dir2_leaf_t		*leaf;		/* leaf structure */
1599 	xfs_dir2_leaf_tail_t	*ltp;		/* leaf tail structure */
1600 	xfs_trans_t		*tp;		/* transaction pointer */
1601 
1602 	dp = args->dp;
1603 	tp = args->trans;
1604 	/*
1605 	 * Read the offending data block.  We need its buffer.
1606 	 */
1607 	error = xfs_dir3_data_read(tp, dp, xfs_dir2_db_to_da(args->geo, db),
1608 				   -1, &dbp);
1609 	if (error)
1610 		return error;
1611 
1612 	leaf = lbp->b_addr;
1613 	ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1614 
1615 #ifdef DEBUG
1616 {
1617 	struct xfs_dir2_data_hdr *hdr = dbp->b_addr;
1618 	struct xfs_dir2_data_free *bf = dp->d_ops->data_bestfree_p(hdr);
1619 
1620 	ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
1621 	       hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC));
1622 	ASSERT(be16_to_cpu(bf[0].length) ==
1623 	       args->geo->blksize - dp->d_ops->data_entry_offset);
1624 	ASSERT(db == be32_to_cpu(ltp->bestcount) - 1);
1625 }
1626 #endif
1627 
1628 	/*
1629 	 * Get rid of the data block.
1630 	 */
1631 	if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1632 		ASSERT(error != -ENOSPC);
1633 		xfs_trans_brelse(tp, dbp);
1634 		return error;
1635 	}
1636 	/*
1637 	 * Eliminate the last bests entry from the table.
1638 	 */
1639 	bestsp = xfs_dir2_leaf_bests_p(ltp);
1640 	be32_add_cpu(&ltp->bestcount, -1);
1641 	memmove(&bestsp[1], &bestsp[0], be32_to_cpu(ltp->bestcount) * sizeof(*bestsp));
1642 	xfs_dir3_leaf_log_tail(args, lbp);
1643 	xfs_dir3_leaf_log_bests(args, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1644 	return 0;
1645 }
1646 
1647 static inline size_t
1648 xfs_dir3_leaf_size(
1649 	struct xfs_dir3_icleaf_hdr	*hdr,
1650 	int				counts)
1651 {
1652 	int	entries;
1653 	int	hdrsize;
1654 
1655 	entries = hdr->count - hdr->stale;
1656 	if (hdr->magic == XFS_DIR2_LEAF1_MAGIC ||
1657 	    hdr->magic == XFS_DIR2_LEAFN_MAGIC)
1658 		hdrsize = sizeof(struct xfs_dir2_leaf_hdr);
1659 	else
1660 		hdrsize = sizeof(struct xfs_dir3_leaf_hdr);
1661 
1662 	return hdrsize + entries * sizeof(xfs_dir2_leaf_entry_t)
1663 	               + counts * sizeof(xfs_dir2_data_off_t)
1664 		       + sizeof(xfs_dir2_leaf_tail_t);
1665 }
1666 
1667 /*
1668  * Convert node form directory to leaf form directory.
1669  * The root of the node form dir needs to already be a LEAFN block.
1670  * Just return if we can't do anything.
1671  */
1672 int						/* error */
1673 xfs_dir2_node_to_leaf(
1674 	xfs_da_state_t		*state)		/* directory operation state */
1675 {
1676 	xfs_da_args_t		*args;		/* operation arguments */
1677 	xfs_inode_t		*dp;		/* incore directory inode */
1678 	int			error;		/* error return code */
1679 	struct xfs_buf		*fbp;		/* buffer for freespace block */
1680 	xfs_fileoff_t		fo;		/* freespace file offset */
1681 	xfs_dir2_free_t		*free;		/* freespace structure */
1682 	struct xfs_buf		*lbp;		/* buffer for leaf block */
1683 	xfs_dir2_leaf_tail_t	*ltp;		/* tail of leaf structure */
1684 	xfs_dir2_leaf_t		*leaf;		/* leaf structure */
1685 	xfs_mount_t		*mp;		/* filesystem mount point */
1686 	int			rval;		/* successful free trim? */
1687 	xfs_trans_t		*tp;		/* transaction pointer */
1688 	struct xfs_dir3_icleaf_hdr leafhdr;
1689 	struct xfs_dir3_icfree_hdr freehdr;
1690 
1691 	/*
1692 	 * There's more than a leaf level in the btree, so there must
1693 	 * be multiple leafn blocks.  Give up.
1694 	 */
1695 	if (state->path.active > 1)
1696 		return 0;
1697 	args = state->args;
1698 
1699 	trace_xfs_dir2_node_to_leaf(args);
1700 
1701 	mp = state->mp;
1702 	dp = args->dp;
1703 	tp = args->trans;
1704 	/*
1705 	 * Get the last offset in the file.
1706 	 */
1707 	if ((error = xfs_bmap_last_offset(dp, &fo, XFS_DATA_FORK))) {
1708 		return error;
1709 	}
1710 	fo -= args->geo->fsbcount;
1711 	/*
1712 	 * If there are freespace blocks other than the first one,
1713 	 * take this opportunity to remove trailing empty freespace blocks
1714 	 * that may have been left behind during no-space-reservation
1715 	 * operations.
1716 	 */
1717 	while (fo > args->geo->freeblk) {
1718 		if ((error = xfs_dir2_node_trim_free(args, fo, &rval))) {
1719 			return error;
1720 		}
1721 		if (rval)
1722 			fo -= args->geo->fsbcount;
1723 		else
1724 			return 0;
1725 	}
1726 	/*
1727 	 * Now find the block just before the freespace block.
1728 	 */
1729 	if ((error = xfs_bmap_last_before(tp, dp, &fo, XFS_DATA_FORK))) {
1730 		return error;
1731 	}
1732 	/*
1733 	 * If it's not the single leaf block, give up.
1734 	 */
1735 	if (XFS_FSB_TO_B(mp, fo) > XFS_DIR2_LEAF_OFFSET + args->geo->blksize)
1736 		return 0;
1737 	lbp = state->path.blk[0].bp;
1738 	leaf = lbp->b_addr;
1739 	dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1740 
1741 	ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC ||
1742 	       leafhdr.magic == XFS_DIR3_LEAFN_MAGIC);
1743 
1744 	/*
1745 	 * Read the freespace block.
1746 	 */
1747 	error = xfs_dir2_free_read(tp, dp,  args->geo->freeblk, &fbp);
1748 	if (error)
1749 		return error;
1750 	free = fbp->b_addr;
1751 	dp->d_ops->free_hdr_from_disk(&freehdr, free);
1752 
1753 	ASSERT(!freehdr.firstdb);
1754 
1755 	/*
1756 	 * Now see if the leafn and free data will fit in a leaf1.
1757 	 * If not, release the buffer and give up.
1758 	 */
1759 	if (xfs_dir3_leaf_size(&leafhdr, freehdr.nvalid) > args->geo->blksize) {
1760 		xfs_trans_brelse(tp, fbp);
1761 		return 0;
1762 	}
1763 
1764 	/*
1765 	 * If the leaf has any stale entries in it, compress them out.
1766 	 */
1767 	if (leafhdr.stale)
1768 		xfs_dir3_leaf_compact(args, &leafhdr, lbp);
1769 
1770 	lbp->b_ops = &xfs_dir3_leaf1_buf_ops;
1771 	xfs_trans_buf_set_type(tp, lbp, XFS_BLFT_DIR_LEAF1_BUF);
1772 	leafhdr.magic = (leafhdr.magic == XFS_DIR2_LEAFN_MAGIC)
1773 					? XFS_DIR2_LEAF1_MAGIC
1774 					: XFS_DIR3_LEAF1_MAGIC;
1775 
1776 	/*
1777 	 * Set up the leaf tail from the freespace block.
1778 	 */
1779 	ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1780 	ltp->bestcount = cpu_to_be32(freehdr.nvalid);
1781 
1782 	/*
1783 	 * Set up the leaf bests table.
1784 	 */
1785 	memcpy(xfs_dir2_leaf_bests_p(ltp), dp->d_ops->free_bests_p(free),
1786 		freehdr.nvalid * sizeof(xfs_dir2_data_off_t));
1787 
1788 	dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
1789 	xfs_dir3_leaf_log_header(args, lbp);
1790 	xfs_dir3_leaf_log_bests(args, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1791 	xfs_dir3_leaf_log_tail(args, lbp);
1792 	xfs_dir3_leaf_check(dp, lbp);
1793 
1794 	/*
1795 	 * Get rid of the freespace block.
1796 	 */
1797 	error = xfs_dir2_shrink_inode(args,
1798 			xfs_dir2_byte_to_db(args->geo, XFS_DIR2_FREE_OFFSET),
1799 			fbp);
1800 	if (error) {
1801 		/*
1802 		 * This can't fail here because it can only happen when
1803 		 * punching out the middle of an extent, and this is an
1804 		 * isolated block.
1805 		 */
1806 		ASSERT(error != -ENOSPC);
1807 		return error;
1808 	}
1809 	fbp = NULL;
1810 	/*
1811 	 * Now see if we can convert the single-leaf directory
1812 	 * down to a block form directory.
1813 	 * This routine always kills the dabuf for the leaf, so
1814 	 * eliminate it from the path.
1815 	 */
1816 	error = xfs_dir2_leaf_to_block(args, lbp, NULL);
1817 	state->path.blk[0].bp = NULL;
1818 	return error;
1819 }
1820