xref: /openbmc/linux/fs/xfs/libxfs/xfs_dir2_leaf.c (revision 9a8f3203)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2000-2003,2005 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_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
12 #include "xfs_mount.h"
13 #include "xfs_da_format.h"
14 #include "xfs_da_btree.h"
15 #include "xfs_inode.h"
16 #include "xfs_bmap.h"
17 #include "xfs_dir2.h"
18 #include "xfs_dir2_priv.h"
19 #include "xfs_error.h"
20 #include "xfs_trace.h"
21 #include "xfs_trans.h"
22 #include "xfs_buf_item.h"
23 #include "xfs_cksum.h"
24 #include "xfs_log.h"
25 
26 /*
27  * Local function declarations.
28  */
29 static int xfs_dir2_leaf_lookup_int(xfs_da_args_t *args, struct xfs_buf **lbpp,
30 				    int *indexp, struct xfs_buf **dbpp);
31 static void xfs_dir3_leaf_log_bests(struct xfs_da_args *args,
32 				    struct xfs_buf *bp, int first, int last);
33 static void xfs_dir3_leaf_log_tail(struct xfs_da_args *args,
34 				   struct xfs_buf *bp);
35 
36 /*
37  * Check the internal consistency of a leaf1 block.
38  * Pop an assert if something is wrong.
39  */
40 #ifdef DEBUG
41 static xfs_failaddr_t
42 xfs_dir3_leaf1_check(
43 	struct xfs_inode	*dp,
44 	struct xfs_buf		*bp)
45 {
46 	struct xfs_dir2_leaf	*leaf = bp->b_addr;
47 	struct xfs_dir3_icleaf_hdr leafhdr;
48 
49 	dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
50 
51 	if (leafhdr.magic == XFS_DIR3_LEAF1_MAGIC) {
52 		struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
53 		if (be64_to_cpu(leaf3->info.blkno) != bp->b_bn)
54 			return __this_address;
55 	} else if (leafhdr.magic != XFS_DIR2_LEAF1_MAGIC)
56 		return __this_address;
57 
58 	return xfs_dir3_leaf_check_int(dp->i_mount, dp, &leafhdr, leaf);
59 }
60 
61 static inline void
62 xfs_dir3_leaf_check(
63 	struct xfs_inode	*dp,
64 	struct xfs_buf		*bp)
65 {
66 	xfs_failaddr_t		fa;
67 
68 	fa = xfs_dir3_leaf1_check(dp, bp);
69 	if (!fa)
70 		return;
71 	xfs_corruption_error(__func__, XFS_ERRLEVEL_LOW, dp->i_mount,
72 			bp->b_addr, BBTOB(bp->b_length), __FILE__, __LINE__,
73 			fa);
74 	ASSERT(0);
75 }
76 #else
77 #define	xfs_dir3_leaf_check(dp, bp)
78 #endif
79 
80 xfs_failaddr_t
81 xfs_dir3_leaf_check_int(
82 	struct xfs_mount	*mp,
83 	struct xfs_inode	*dp,
84 	struct xfs_dir3_icleaf_hdr *hdr,
85 	struct xfs_dir2_leaf	*leaf)
86 {
87 	struct xfs_dir2_leaf_entry *ents;
88 	xfs_dir2_leaf_tail_t	*ltp;
89 	int			stale;
90 	int			i;
91 	const struct xfs_dir_ops *ops;
92 	struct xfs_dir3_icleaf_hdr leafhdr;
93 	struct xfs_da_geometry	*geo = mp->m_dir_geo;
94 
95 	/*
96 	 * we can be passed a null dp here from a verifier, so we need to go the
97 	 * hard way to get them.
98 	 */
99 	ops = xfs_dir_get_ops(mp, dp);
100 
101 	if (!hdr) {
102 		ops->leaf_hdr_from_disk(&leafhdr, leaf);
103 		hdr = &leafhdr;
104 	}
105 
106 	ents = ops->leaf_ents_p(leaf);
107 	ltp = xfs_dir2_leaf_tail_p(geo, leaf);
108 
109 	/*
110 	 * XXX (dgc): This value is not restrictive enough.
111 	 * Should factor in the size of the bests table as well.
112 	 * We can deduce a value for that from di_size.
113 	 */
114 	if (hdr->count > ops->leaf_max_ents(geo))
115 		return __this_address;
116 
117 	/* Leaves and bests don't overlap in leaf format. */
118 	if ((hdr->magic == XFS_DIR2_LEAF1_MAGIC ||
119 	     hdr->magic == XFS_DIR3_LEAF1_MAGIC) &&
120 	    (char *)&ents[hdr->count] > (char *)xfs_dir2_leaf_bests_p(ltp))
121 		return __this_address;
122 
123 	/* Check hash value order, count stale entries.  */
124 	for (i = stale = 0; i < hdr->count; i++) {
125 		if (i + 1 < hdr->count) {
126 			if (be32_to_cpu(ents[i].hashval) >
127 					be32_to_cpu(ents[i + 1].hashval))
128 				return __this_address;
129 		}
130 		if (ents[i].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
131 			stale++;
132 	}
133 	if (hdr->stale != stale)
134 		return __this_address;
135 	return NULL;
136 }
137 
138 /*
139  * We verify the magic numbers before decoding the leaf header so that on debug
140  * kernels we don't get assertion failures in xfs_dir3_leaf_hdr_from_disk() due
141  * to incorrect magic numbers.
142  */
143 static xfs_failaddr_t
144 xfs_dir3_leaf_verify(
145 	struct xfs_buf		*bp)
146 {
147 	struct xfs_mount	*mp = bp->b_target->bt_mount;
148 	struct xfs_dir2_leaf	*leaf = bp->b_addr;
149 	xfs_failaddr_t		fa;
150 
151 	fa = xfs_da3_blkinfo_verify(bp, bp->b_addr);
152 	if (fa)
153 		return fa;
154 
155 	return xfs_dir3_leaf_check_int(mp, NULL, NULL, leaf);
156 }
157 
158 static void
159 xfs_dir3_leaf_read_verify(
160 	struct xfs_buf  *bp)
161 {
162 	struct xfs_mount	*mp = bp->b_target->bt_mount;
163 	xfs_failaddr_t		fa;
164 
165 	if (xfs_sb_version_hascrc(&mp->m_sb) &&
166 	     !xfs_buf_verify_cksum(bp, XFS_DIR3_LEAF_CRC_OFF))
167 		xfs_verifier_error(bp, -EFSBADCRC, __this_address);
168 	else {
169 		fa = xfs_dir3_leaf_verify(bp);
170 		if (fa)
171 			xfs_verifier_error(bp, -EFSCORRUPTED, fa);
172 	}
173 }
174 
175 static void
176 xfs_dir3_leaf_write_verify(
177 	struct xfs_buf  *bp)
178 {
179 	struct xfs_mount	*mp = bp->b_target->bt_mount;
180 	struct xfs_buf_log_item	*bip = bp->b_log_item;
181 	struct xfs_dir3_leaf_hdr *hdr3 = bp->b_addr;
182 	xfs_failaddr_t		fa;
183 
184 	fa = xfs_dir3_leaf_verify(bp);
185 	if (fa) {
186 		xfs_verifier_error(bp, -EFSCORRUPTED, fa);
187 		return;
188 	}
189 
190 	if (!xfs_sb_version_hascrc(&mp->m_sb))
191 		return;
192 
193 	if (bip)
194 		hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
195 
196 	xfs_buf_update_cksum(bp, XFS_DIR3_LEAF_CRC_OFF);
197 }
198 
199 const struct xfs_buf_ops xfs_dir3_leaf1_buf_ops = {
200 	.name = "xfs_dir3_leaf1",
201 	.magic16 = { cpu_to_be16(XFS_DIR2_LEAF1_MAGIC),
202 		     cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) },
203 	.verify_read = xfs_dir3_leaf_read_verify,
204 	.verify_write = xfs_dir3_leaf_write_verify,
205 	.verify_struct = xfs_dir3_leaf_verify,
206 };
207 
208 const struct xfs_buf_ops xfs_dir3_leafn_buf_ops = {
209 	.name = "xfs_dir3_leafn",
210 	.magic16 = { cpu_to_be16(XFS_DIR2_LEAFN_MAGIC),
211 		     cpu_to_be16(XFS_DIR3_LEAFN_MAGIC) },
212 	.verify_read = xfs_dir3_leaf_read_verify,
213 	.verify_write = xfs_dir3_leaf_write_verify,
214 	.verify_struct = xfs_dir3_leaf_verify,
215 };
216 
217 int
218 xfs_dir3_leaf_read(
219 	struct xfs_trans	*tp,
220 	struct xfs_inode	*dp,
221 	xfs_dablk_t		fbno,
222 	xfs_daddr_t		mappedbno,
223 	struct xfs_buf		**bpp)
224 {
225 	int			err;
226 
227 	err = xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
228 				XFS_DATA_FORK, &xfs_dir3_leaf1_buf_ops);
229 	if (!err && tp && *bpp)
230 		xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_LEAF1_BUF);
231 	return err;
232 }
233 
234 int
235 xfs_dir3_leafn_read(
236 	struct xfs_trans	*tp,
237 	struct xfs_inode	*dp,
238 	xfs_dablk_t		fbno,
239 	xfs_daddr_t		mappedbno,
240 	struct xfs_buf		**bpp)
241 {
242 	int			err;
243 
244 	err = xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
245 				XFS_DATA_FORK, &xfs_dir3_leafn_buf_ops);
246 	if (!err && tp && *bpp)
247 		xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_LEAFN_BUF);
248 	return err;
249 }
250 
251 /*
252  * Initialize a new leaf block, leaf1 or leafn magic accepted.
253  */
254 static void
255 xfs_dir3_leaf_init(
256 	struct xfs_mount	*mp,
257 	struct xfs_trans	*tp,
258 	struct xfs_buf		*bp,
259 	xfs_ino_t		owner,
260 	uint16_t		type)
261 {
262 	struct xfs_dir2_leaf	*leaf = bp->b_addr;
263 
264 	ASSERT(type == XFS_DIR2_LEAF1_MAGIC || type == XFS_DIR2_LEAFN_MAGIC);
265 
266 	if (xfs_sb_version_hascrc(&mp->m_sb)) {
267 		struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
268 
269 		memset(leaf3, 0, sizeof(*leaf3));
270 
271 		leaf3->info.hdr.magic = (type == XFS_DIR2_LEAF1_MAGIC)
272 					 ? cpu_to_be16(XFS_DIR3_LEAF1_MAGIC)
273 					 : cpu_to_be16(XFS_DIR3_LEAFN_MAGIC);
274 		leaf3->info.blkno = cpu_to_be64(bp->b_bn);
275 		leaf3->info.owner = cpu_to_be64(owner);
276 		uuid_copy(&leaf3->info.uuid, &mp->m_sb.sb_meta_uuid);
277 	} else {
278 		memset(leaf, 0, sizeof(*leaf));
279 		leaf->hdr.info.magic = cpu_to_be16(type);
280 	}
281 
282 	/*
283 	 * If it's a leaf-format directory initialize the tail.
284 	 * Caller is responsible for initialising the bests table.
285 	 */
286 	if (type == XFS_DIR2_LEAF1_MAGIC) {
287 		struct xfs_dir2_leaf_tail *ltp;
288 
289 		ltp = xfs_dir2_leaf_tail_p(mp->m_dir_geo, leaf);
290 		ltp->bestcount = 0;
291 		bp->b_ops = &xfs_dir3_leaf1_buf_ops;
292 		xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAF1_BUF);
293 	} else {
294 		bp->b_ops = &xfs_dir3_leafn_buf_ops;
295 		xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAFN_BUF);
296 	}
297 }
298 
299 int
300 xfs_dir3_leaf_get_buf(
301 	xfs_da_args_t		*args,
302 	xfs_dir2_db_t		bno,
303 	struct xfs_buf		**bpp,
304 	uint16_t		magic)
305 {
306 	struct xfs_inode	*dp = args->dp;
307 	struct xfs_trans	*tp = args->trans;
308 	struct xfs_mount	*mp = dp->i_mount;
309 	struct xfs_buf		*bp;
310 	int			error;
311 
312 	ASSERT(magic == XFS_DIR2_LEAF1_MAGIC || magic == XFS_DIR2_LEAFN_MAGIC);
313 	ASSERT(bno >= xfs_dir2_byte_to_db(args->geo, XFS_DIR2_LEAF_OFFSET) &&
314 	       bno < xfs_dir2_byte_to_db(args->geo, XFS_DIR2_FREE_OFFSET));
315 
316 	error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(args->geo, bno),
317 			       -1, &bp, XFS_DATA_FORK);
318 	if (error)
319 		return error;
320 
321 	xfs_dir3_leaf_init(mp, tp, bp, dp->i_ino, magic);
322 	xfs_dir3_leaf_log_header(args, bp);
323 	if (magic == XFS_DIR2_LEAF1_MAGIC)
324 		xfs_dir3_leaf_log_tail(args, bp);
325 	*bpp = bp;
326 	return 0;
327 }
328 
329 /*
330  * Convert a block form directory to a leaf form directory.
331  */
332 int						/* error */
333 xfs_dir2_block_to_leaf(
334 	xfs_da_args_t		*args,		/* operation arguments */
335 	struct xfs_buf		*dbp)		/* input block's buffer */
336 {
337 	__be16			*bestsp;	/* leaf's bestsp entries */
338 	xfs_dablk_t		blkno;		/* leaf block's bno */
339 	xfs_dir2_data_hdr_t	*hdr;		/* block header */
340 	xfs_dir2_leaf_entry_t	*blp;		/* block's leaf entries */
341 	xfs_dir2_block_tail_t	*btp;		/* block's tail */
342 	xfs_inode_t		*dp;		/* incore directory inode */
343 	int			error;		/* error return code */
344 	struct xfs_buf		*lbp;		/* leaf block's buffer */
345 	xfs_dir2_db_t		ldb;		/* leaf block's bno */
346 	xfs_dir2_leaf_t		*leaf;		/* leaf structure */
347 	xfs_dir2_leaf_tail_t	*ltp;		/* leaf's tail */
348 	int			needlog;	/* need to log block header */
349 	int			needscan;	/* need to rescan bestfree */
350 	xfs_trans_t		*tp;		/* transaction pointer */
351 	struct xfs_dir2_data_free *bf;
352 	struct xfs_dir2_leaf_entry *ents;
353 	struct xfs_dir3_icleaf_hdr leafhdr;
354 
355 	trace_xfs_dir2_block_to_leaf(args);
356 
357 	dp = args->dp;
358 	tp = args->trans;
359 	/*
360 	 * Add the leaf block to the inode.
361 	 * This interface will only put blocks in the leaf/node range.
362 	 * Since that's empty now, we'll get the root (block 0 in range).
363 	 */
364 	if ((error = xfs_da_grow_inode(args, &blkno))) {
365 		return error;
366 	}
367 	ldb = xfs_dir2_da_to_db(args->geo, blkno);
368 	ASSERT(ldb == xfs_dir2_byte_to_db(args->geo, XFS_DIR2_LEAF_OFFSET));
369 	/*
370 	 * Initialize the leaf block, get a buffer for it.
371 	 */
372 	error = xfs_dir3_leaf_get_buf(args, ldb, &lbp, XFS_DIR2_LEAF1_MAGIC);
373 	if (error)
374 		return error;
375 
376 	leaf = lbp->b_addr;
377 	hdr = dbp->b_addr;
378 	xfs_dir3_data_check(dp, dbp);
379 	btp = xfs_dir2_block_tail_p(args->geo, hdr);
380 	blp = xfs_dir2_block_leaf_p(btp);
381 	bf = dp->d_ops->data_bestfree_p(hdr);
382 	ents = dp->d_ops->leaf_ents_p(leaf);
383 
384 	/*
385 	 * Set the counts in the leaf header.
386 	 */
387 	dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
388 	leafhdr.count = be32_to_cpu(btp->count);
389 	leafhdr.stale = be32_to_cpu(btp->stale);
390 	dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
391 	xfs_dir3_leaf_log_header(args, lbp);
392 
393 	/*
394 	 * Could compact these but I think we always do the conversion
395 	 * after squeezing out stale entries.
396 	 */
397 	memcpy(ents, blp, be32_to_cpu(btp->count) * sizeof(xfs_dir2_leaf_entry_t));
398 	xfs_dir3_leaf_log_ents(args, lbp, 0, leafhdr.count - 1);
399 	needscan = 0;
400 	needlog = 1;
401 	/*
402 	 * Make the space formerly occupied by the leaf entries and block
403 	 * tail be free.
404 	 */
405 	xfs_dir2_data_make_free(args, dbp,
406 		(xfs_dir2_data_aoff_t)((char *)blp - (char *)hdr),
407 		(xfs_dir2_data_aoff_t)((char *)hdr + args->geo->blksize -
408 				       (char *)blp),
409 		&needlog, &needscan);
410 	/*
411 	 * Fix up the block header, make it a data block.
412 	 */
413 	dbp->b_ops = &xfs_dir3_data_buf_ops;
414 	xfs_trans_buf_set_type(tp, dbp, XFS_BLFT_DIR_DATA_BUF);
415 	if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC))
416 		hdr->magic = cpu_to_be32(XFS_DIR2_DATA_MAGIC);
417 	else
418 		hdr->magic = cpu_to_be32(XFS_DIR3_DATA_MAGIC);
419 
420 	if (needscan)
421 		xfs_dir2_data_freescan(dp, hdr, &needlog);
422 	/*
423 	 * Set up leaf tail and bests table.
424 	 */
425 	ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
426 	ltp->bestcount = cpu_to_be32(1);
427 	bestsp = xfs_dir2_leaf_bests_p(ltp);
428 	bestsp[0] =  bf[0].length;
429 	/*
430 	 * Log the data header and leaf bests table.
431 	 */
432 	if (needlog)
433 		xfs_dir2_data_log_header(args, dbp);
434 	xfs_dir3_leaf_check(dp, lbp);
435 	xfs_dir3_data_check(dp, dbp);
436 	xfs_dir3_leaf_log_bests(args, lbp, 0, 0);
437 	return 0;
438 }
439 
440 STATIC void
441 xfs_dir3_leaf_find_stale(
442 	struct xfs_dir3_icleaf_hdr *leafhdr,
443 	struct xfs_dir2_leaf_entry *ents,
444 	int			index,
445 	int			*lowstale,
446 	int			*highstale)
447 {
448 	/*
449 	 * Find the first stale entry before our index, if any.
450 	 */
451 	for (*lowstale = index - 1; *lowstale >= 0; --*lowstale) {
452 		if (ents[*lowstale].address ==
453 		    cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
454 			break;
455 	}
456 
457 	/*
458 	 * Find the first stale entry at or after our index, if any.
459 	 * Stop if the result would require moving more entries than using
460 	 * lowstale.
461 	 */
462 	for (*highstale = index; *highstale < leafhdr->count; ++*highstale) {
463 		if (ents[*highstale].address ==
464 		    cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
465 			break;
466 		if (*lowstale >= 0 && index - *lowstale <= *highstale - index)
467 			break;
468 	}
469 }
470 
471 struct xfs_dir2_leaf_entry *
472 xfs_dir3_leaf_find_entry(
473 	struct xfs_dir3_icleaf_hdr *leafhdr,
474 	struct xfs_dir2_leaf_entry *ents,
475 	int			index,		/* leaf table position */
476 	int			compact,	/* need to compact leaves */
477 	int			lowstale,	/* index of prev stale leaf */
478 	int			highstale,	/* index of next stale leaf */
479 	int			*lfloglow,	/* low leaf logging index */
480 	int			*lfloghigh)	/* high leaf logging index */
481 {
482 	if (!leafhdr->stale) {
483 		xfs_dir2_leaf_entry_t	*lep;	/* leaf entry table pointer */
484 
485 		/*
486 		 * Now we need to make room to insert the leaf entry.
487 		 *
488 		 * If there are no stale entries, just insert a hole at index.
489 		 */
490 		lep = &ents[index];
491 		if (index < leafhdr->count)
492 			memmove(lep + 1, lep,
493 				(leafhdr->count - index) * sizeof(*lep));
494 
495 		/*
496 		 * Record low and high logging indices for the leaf.
497 		 */
498 		*lfloglow = index;
499 		*lfloghigh = leafhdr->count++;
500 		return lep;
501 	}
502 
503 	/*
504 	 * There are stale entries.
505 	 *
506 	 * We will use one of them for the new entry.  It's probably not at
507 	 * the right location, so we'll have to shift some up or down first.
508 	 *
509 	 * If we didn't compact before, we need to find the nearest stale
510 	 * entries before and after our insertion point.
511 	 */
512 	if (compact == 0)
513 		xfs_dir3_leaf_find_stale(leafhdr, ents, index,
514 					 &lowstale, &highstale);
515 
516 	/*
517 	 * If the low one is better, use it.
518 	 */
519 	if (lowstale >= 0 &&
520 	    (highstale == leafhdr->count ||
521 	     index - lowstale - 1 < highstale - index)) {
522 		ASSERT(index - lowstale - 1 >= 0);
523 		ASSERT(ents[lowstale].address ==
524 		       cpu_to_be32(XFS_DIR2_NULL_DATAPTR));
525 
526 		/*
527 		 * Copy entries up to cover the stale entry and make room
528 		 * for the new entry.
529 		 */
530 		if (index - lowstale - 1 > 0) {
531 			memmove(&ents[lowstale], &ents[lowstale + 1],
532 				(index - lowstale - 1) *
533 					sizeof(xfs_dir2_leaf_entry_t));
534 		}
535 		*lfloglow = min(lowstale, *lfloglow);
536 		*lfloghigh = max(index - 1, *lfloghigh);
537 		leafhdr->stale--;
538 		return &ents[index - 1];
539 	}
540 
541 	/*
542 	 * The high one is better, so use that one.
543 	 */
544 	ASSERT(highstale - index >= 0);
545 	ASSERT(ents[highstale].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR));
546 
547 	/*
548 	 * Copy entries down to cover the stale entry and make room for the
549 	 * new entry.
550 	 */
551 	if (highstale - index > 0) {
552 		memmove(&ents[index + 1], &ents[index],
553 			(highstale - index) * sizeof(xfs_dir2_leaf_entry_t));
554 	}
555 	*lfloglow = min(index, *lfloglow);
556 	*lfloghigh = max(highstale, *lfloghigh);
557 	leafhdr->stale--;
558 	return &ents[index];
559 }
560 
561 /*
562  * Add an entry to a leaf form directory.
563  */
564 int						/* error */
565 xfs_dir2_leaf_addname(
566 	struct xfs_da_args	*args)		/* operation arguments */
567 {
568 	struct xfs_dir3_icleaf_hdr leafhdr;
569 	struct xfs_trans	*tp = args->trans;
570 	__be16			*bestsp;	/* freespace table in leaf */
571 	__be16			*tagp;		/* end of data entry */
572 	struct xfs_buf		*dbp;		/* data block buffer */
573 	struct xfs_buf		*lbp;		/* leaf's buffer */
574 	struct xfs_dir2_leaf	*leaf;		/* leaf structure */
575 	struct xfs_inode	*dp = args->dp;	/* incore directory inode */
576 	struct xfs_dir2_data_hdr *hdr;		/* data block header */
577 	struct xfs_dir2_data_entry *dep;	/* data block entry */
578 	struct xfs_dir2_leaf_entry *lep;	/* leaf entry table pointer */
579 	struct xfs_dir2_leaf_entry *ents;
580 	struct xfs_dir2_data_unused *dup;	/* data unused entry */
581 	struct xfs_dir2_leaf_tail *ltp;		/* leaf tail pointer */
582 	struct xfs_dir2_data_free *bf;		/* bestfree table */
583 	int			compact;	/* need to compact leaves */
584 	int			error;		/* error return value */
585 	int			grown;		/* allocated new data block */
586 	int			highstale = 0;	/* index of next stale leaf */
587 	int			i;		/* temporary, index */
588 	int			index;		/* leaf table position */
589 	int			length;		/* length of new entry */
590 	int			lfloglow;	/* low leaf logging index */
591 	int			lfloghigh;	/* high leaf logging index */
592 	int			lowstale = 0;	/* index of prev stale leaf */
593 	int			needbytes;	/* leaf block bytes needed */
594 	int			needlog;	/* need to log data header */
595 	int			needscan;	/* need to rescan data free */
596 	xfs_dir2_db_t		use_block;	/* data block number */
597 
598 	trace_xfs_dir2_leaf_addname(args);
599 
600 	error = xfs_dir3_leaf_read(tp, dp, args->geo->leafblk, -1, &lbp);
601 	if (error)
602 		return error;
603 
604 	/*
605 	 * Look up the entry by hash value and name.
606 	 * We know it's not there, our caller has already done a lookup.
607 	 * So the index is of the entry to insert in front of.
608 	 * But if there are dup hash values the index is of the first of those.
609 	 */
610 	index = xfs_dir2_leaf_search_hash(args, lbp);
611 	leaf = lbp->b_addr;
612 	ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
613 	ents = dp->d_ops->leaf_ents_p(leaf);
614 	dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
615 	bestsp = xfs_dir2_leaf_bests_p(ltp);
616 	length = dp->d_ops->data_entsize(args->namelen);
617 
618 	/*
619 	 * See if there are any entries with the same hash value
620 	 * and space in their block for the new entry.
621 	 * This is good because it puts multiple same-hash value entries
622 	 * in a data block, improving the lookup of those entries.
623 	 */
624 	for (use_block = -1, lep = &ents[index];
625 	     index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
626 	     index++, lep++) {
627 		if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
628 			continue;
629 		i = xfs_dir2_dataptr_to_db(args->geo, be32_to_cpu(lep->address));
630 		ASSERT(i < be32_to_cpu(ltp->bestcount));
631 		ASSERT(bestsp[i] != cpu_to_be16(NULLDATAOFF));
632 		if (be16_to_cpu(bestsp[i]) >= length) {
633 			use_block = i;
634 			break;
635 		}
636 	}
637 	/*
638 	 * Didn't find a block yet, linear search all the data blocks.
639 	 */
640 	if (use_block == -1) {
641 		for (i = 0; i < be32_to_cpu(ltp->bestcount); i++) {
642 			/*
643 			 * Remember a block we see that's missing.
644 			 */
645 			if (bestsp[i] == cpu_to_be16(NULLDATAOFF) &&
646 			    use_block == -1)
647 				use_block = i;
648 			else if (be16_to_cpu(bestsp[i]) >= length) {
649 				use_block = i;
650 				break;
651 			}
652 		}
653 	}
654 	/*
655 	 * How many bytes do we need in the leaf block?
656 	 */
657 	needbytes = 0;
658 	if (!leafhdr.stale)
659 		needbytes += sizeof(xfs_dir2_leaf_entry_t);
660 	if (use_block == -1)
661 		needbytes += sizeof(xfs_dir2_data_off_t);
662 
663 	/*
664 	 * Now kill use_block if it refers to a missing block, so we
665 	 * can use it as an indication of allocation needed.
666 	 */
667 	if (use_block != -1 && bestsp[use_block] == cpu_to_be16(NULLDATAOFF))
668 		use_block = -1;
669 	/*
670 	 * If we don't have enough free bytes but we can make enough
671 	 * by compacting out stale entries, we'll do that.
672 	 */
673 	if ((char *)bestsp - (char *)&ents[leafhdr.count] < needbytes &&
674 	    leafhdr.stale > 1)
675 		compact = 1;
676 
677 	/*
678 	 * Otherwise if we don't have enough free bytes we need to
679 	 * convert to node form.
680 	 */
681 	else if ((char *)bestsp - (char *)&ents[leafhdr.count] < needbytes) {
682 		/*
683 		 * Just checking or no space reservation, give up.
684 		 */
685 		if ((args->op_flags & XFS_DA_OP_JUSTCHECK) ||
686 							args->total == 0) {
687 			xfs_trans_brelse(tp, lbp);
688 			return -ENOSPC;
689 		}
690 		/*
691 		 * Convert to node form.
692 		 */
693 		error = xfs_dir2_leaf_to_node(args, lbp);
694 		if (error)
695 			return error;
696 		/*
697 		 * Then add the new entry.
698 		 */
699 		return xfs_dir2_node_addname(args);
700 	}
701 	/*
702 	 * Otherwise it will fit without compaction.
703 	 */
704 	else
705 		compact = 0;
706 	/*
707 	 * If just checking, then it will fit unless we needed to allocate
708 	 * a new data block.
709 	 */
710 	if (args->op_flags & XFS_DA_OP_JUSTCHECK) {
711 		xfs_trans_brelse(tp, lbp);
712 		return use_block == -1 ? -ENOSPC : 0;
713 	}
714 	/*
715 	 * If no allocations are allowed, return now before we've
716 	 * changed anything.
717 	 */
718 	if (args->total == 0 && use_block == -1) {
719 		xfs_trans_brelse(tp, lbp);
720 		return -ENOSPC;
721 	}
722 	/*
723 	 * Need to compact the leaf entries, removing stale ones.
724 	 * Leave one stale entry behind - the one closest to our
725 	 * insertion index - and we'll shift that one to our insertion
726 	 * point later.
727 	 */
728 	if (compact) {
729 		xfs_dir3_leaf_compact_x1(&leafhdr, ents, &index, &lowstale,
730 			&highstale, &lfloglow, &lfloghigh);
731 	}
732 	/*
733 	 * There are stale entries, so we'll need log-low and log-high
734 	 * impossibly bad values later.
735 	 */
736 	else if (leafhdr.stale) {
737 		lfloglow = leafhdr.count;
738 		lfloghigh = -1;
739 	}
740 	/*
741 	 * If there was no data block space found, we need to allocate
742 	 * a new one.
743 	 */
744 	if (use_block == -1) {
745 		/*
746 		 * Add the new data block.
747 		 */
748 		if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE,
749 				&use_block))) {
750 			xfs_trans_brelse(tp, lbp);
751 			return error;
752 		}
753 		/*
754 		 * Initialize the block.
755 		 */
756 		if ((error = xfs_dir3_data_init(args, use_block, &dbp))) {
757 			xfs_trans_brelse(tp, lbp);
758 			return error;
759 		}
760 		/*
761 		 * If we're adding a new data block on the end we need to
762 		 * extend the bests table.  Copy it up one entry.
763 		 */
764 		if (use_block >= be32_to_cpu(ltp->bestcount)) {
765 			bestsp--;
766 			memmove(&bestsp[0], &bestsp[1],
767 				be32_to_cpu(ltp->bestcount) * sizeof(bestsp[0]));
768 			be32_add_cpu(&ltp->bestcount, 1);
769 			xfs_dir3_leaf_log_tail(args, lbp);
770 			xfs_dir3_leaf_log_bests(args, lbp, 0,
771 						be32_to_cpu(ltp->bestcount) - 1);
772 		}
773 		/*
774 		 * If we're filling in a previously empty block just log it.
775 		 */
776 		else
777 			xfs_dir3_leaf_log_bests(args, lbp, use_block, use_block);
778 		hdr = dbp->b_addr;
779 		bf = dp->d_ops->data_bestfree_p(hdr);
780 		bestsp[use_block] = bf[0].length;
781 		grown = 1;
782 	} else {
783 		/*
784 		 * Already had space in some data block.
785 		 * Just read that one in.
786 		 */
787 		error = xfs_dir3_data_read(tp, dp,
788 				   xfs_dir2_db_to_da(args->geo, use_block),
789 				   -1, &dbp);
790 		if (error) {
791 			xfs_trans_brelse(tp, lbp);
792 			return error;
793 		}
794 		hdr = dbp->b_addr;
795 		bf = dp->d_ops->data_bestfree_p(hdr);
796 		grown = 0;
797 	}
798 	/*
799 	 * Point to the biggest freespace in our data block.
800 	 */
801 	dup = (xfs_dir2_data_unused_t *)
802 	      ((char *)hdr + be16_to_cpu(bf[0].offset));
803 	needscan = needlog = 0;
804 	/*
805 	 * Mark the initial part of our freespace in use for the new entry.
806 	 */
807 	error = xfs_dir2_data_use_free(args, dbp, dup,
808 			(xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr),
809 			length, &needlog, &needscan);
810 	if (error) {
811 		xfs_trans_brelse(tp, lbp);
812 		return error;
813 	}
814 	/*
815 	 * Initialize our new entry (at last).
816 	 */
817 	dep = (xfs_dir2_data_entry_t *)dup;
818 	dep->inumber = cpu_to_be64(args->inumber);
819 	dep->namelen = args->namelen;
820 	memcpy(dep->name, args->name, dep->namelen);
821 	dp->d_ops->data_put_ftype(dep, args->filetype);
822 	tagp = dp->d_ops->data_entry_tag_p(dep);
823 	*tagp = cpu_to_be16((char *)dep - (char *)hdr);
824 	/*
825 	 * Need to scan fix up the bestfree table.
826 	 */
827 	if (needscan)
828 		xfs_dir2_data_freescan(dp, hdr, &needlog);
829 	/*
830 	 * Need to log the data block's header.
831 	 */
832 	if (needlog)
833 		xfs_dir2_data_log_header(args, dbp);
834 	xfs_dir2_data_log_entry(args, dbp, dep);
835 	/*
836 	 * If the bests table needs to be changed, do it.
837 	 * Log the change unless we've already done that.
838 	 */
839 	if (be16_to_cpu(bestsp[use_block]) != be16_to_cpu(bf[0].length)) {
840 		bestsp[use_block] = bf[0].length;
841 		if (!grown)
842 			xfs_dir3_leaf_log_bests(args, lbp, use_block, use_block);
843 	}
844 
845 	lep = xfs_dir3_leaf_find_entry(&leafhdr, ents, index, compact, lowstale,
846 				       highstale, &lfloglow, &lfloghigh);
847 
848 	/*
849 	 * Fill in the new leaf entry.
850 	 */
851 	lep->hashval = cpu_to_be32(args->hashval);
852 	lep->address = cpu_to_be32(
853 				xfs_dir2_db_off_to_dataptr(args->geo, use_block,
854 				be16_to_cpu(*tagp)));
855 	/*
856 	 * Log the leaf fields and give up the buffers.
857 	 */
858 	dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
859 	xfs_dir3_leaf_log_header(args, lbp);
860 	xfs_dir3_leaf_log_ents(args, lbp, lfloglow, lfloghigh);
861 	xfs_dir3_leaf_check(dp, lbp);
862 	xfs_dir3_data_check(dp, dbp);
863 	return 0;
864 }
865 
866 /*
867  * Compact out any stale entries in the leaf.
868  * Log the header and changed leaf entries, if any.
869  */
870 void
871 xfs_dir3_leaf_compact(
872 	xfs_da_args_t	*args,		/* operation arguments */
873 	struct xfs_dir3_icleaf_hdr *leafhdr,
874 	struct xfs_buf	*bp)		/* leaf buffer */
875 {
876 	int		from;		/* source leaf index */
877 	xfs_dir2_leaf_t	*leaf;		/* leaf structure */
878 	int		loglow;		/* first leaf entry to log */
879 	int		to;		/* target leaf index */
880 	struct xfs_dir2_leaf_entry *ents;
881 	struct xfs_inode *dp = args->dp;
882 
883 	leaf = bp->b_addr;
884 	if (!leafhdr->stale)
885 		return;
886 
887 	/*
888 	 * Compress out the stale entries in place.
889 	 */
890 	ents = dp->d_ops->leaf_ents_p(leaf);
891 	for (from = to = 0, loglow = -1; from < leafhdr->count; from++) {
892 		if (ents[from].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
893 			continue;
894 		/*
895 		 * Only actually copy the entries that are different.
896 		 */
897 		if (from > to) {
898 			if (loglow == -1)
899 				loglow = to;
900 			ents[to] = ents[from];
901 		}
902 		to++;
903 	}
904 	/*
905 	 * Update and log the header, log the leaf entries.
906 	 */
907 	ASSERT(leafhdr->stale == from - to);
908 	leafhdr->count -= leafhdr->stale;
909 	leafhdr->stale = 0;
910 
911 	dp->d_ops->leaf_hdr_to_disk(leaf, leafhdr);
912 	xfs_dir3_leaf_log_header(args, bp);
913 	if (loglow != -1)
914 		xfs_dir3_leaf_log_ents(args, bp, loglow, to - 1);
915 }
916 
917 /*
918  * Compact the leaf entries, removing stale ones.
919  * Leave one stale entry behind - the one closest to our
920  * insertion index - and the caller will shift that one to our insertion
921  * point later.
922  * Return new insertion index, where the remaining stale entry is,
923  * and leaf logging indices.
924  */
925 void
926 xfs_dir3_leaf_compact_x1(
927 	struct xfs_dir3_icleaf_hdr *leafhdr,
928 	struct xfs_dir2_leaf_entry *ents,
929 	int		*indexp,	/* insertion index */
930 	int		*lowstalep,	/* out: stale entry before us */
931 	int		*highstalep,	/* out: stale entry after us */
932 	int		*lowlogp,	/* out: low log index */
933 	int		*highlogp)	/* out: high log index */
934 {
935 	int		from;		/* source copy index */
936 	int		highstale;	/* stale entry at/after index */
937 	int		index;		/* insertion index */
938 	int		keepstale;	/* source index of kept stale */
939 	int		lowstale;	/* stale entry before index */
940 	int		newindex=0;	/* new insertion index */
941 	int		to;		/* destination copy index */
942 
943 	ASSERT(leafhdr->stale > 1);
944 	index = *indexp;
945 
946 	xfs_dir3_leaf_find_stale(leafhdr, ents, index, &lowstale, &highstale);
947 
948 	/*
949 	 * Pick the better of lowstale and highstale.
950 	 */
951 	if (lowstale >= 0 &&
952 	    (highstale == leafhdr->count ||
953 	     index - lowstale <= highstale - index))
954 		keepstale = lowstale;
955 	else
956 		keepstale = highstale;
957 	/*
958 	 * Copy the entries in place, removing all the stale entries
959 	 * except keepstale.
960 	 */
961 	for (from = to = 0; from < leafhdr->count; from++) {
962 		/*
963 		 * Notice the new value of index.
964 		 */
965 		if (index == from)
966 			newindex = to;
967 		if (from != keepstale &&
968 		    ents[from].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR)) {
969 			if (from == to)
970 				*lowlogp = to;
971 			continue;
972 		}
973 		/*
974 		 * Record the new keepstale value for the insertion.
975 		 */
976 		if (from == keepstale)
977 			lowstale = highstale = to;
978 		/*
979 		 * Copy only the entries that have moved.
980 		 */
981 		if (from > to)
982 			ents[to] = ents[from];
983 		to++;
984 	}
985 	ASSERT(from > to);
986 	/*
987 	 * If the insertion point was past the last entry,
988 	 * set the new insertion point accordingly.
989 	 */
990 	if (index == from)
991 		newindex = to;
992 	*indexp = newindex;
993 	/*
994 	 * Adjust the leaf header values.
995 	 */
996 	leafhdr->count -= from - to;
997 	leafhdr->stale = 1;
998 	/*
999 	 * Remember the low/high stale value only in the "right"
1000 	 * direction.
1001 	 */
1002 	if (lowstale >= newindex)
1003 		lowstale = -1;
1004 	else
1005 		highstale = leafhdr->count;
1006 	*highlogp = leafhdr->count - 1;
1007 	*lowstalep = lowstale;
1008 	*highstalep = highstale;
1009 }
1010 
1011 /*
1012  * Log the bests entries indicated from a leaf1 block.
1013  */
1014 static void
1015 xfs_dir3_leaf_log_bests(
1016 	struct xfs_da_args	*args,
1017 	struct xfs_buf		*bp,		/* leaf buffer */
1018 	int			first,		/* first entry to log */
1019 	int			last)		/* last entry to log */
1020 {
1021 	__be16			*firstb;	/* pointer to first entry */
1022 	__be16			*lastb;		/* pointer to last entry */
1023 	struct xfs_dir2_leaf	*leaf = bp->b_addr;
1024 	xfs_dir2_leaf_tail_t	*ltp;		/* leaf tail structure */
1025 
1026 	ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1027 	       leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC));
1028 
1029 	ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1030 	firstb = xfs_dir2_leaf_bests_p(ltp) + first;
1031 	lastb = xfs_dir2_leaf_bests_p(ltp) + last;
1032 	xfs_trans_log_buf(args->trans, bp,
1033 		(uint)((char *)firstb - (char *)leaf),
1034 		(uint)((char *)lastb - (char *)leaf + sizeof(*lastb) - 1));
1035 }
1036 
1037 /*
1038  * Log the leaf entries indicated from a leaf1 or leafn block.
1039  */
1040 void
1041 xfs_dir3_leaf_log_ents(
1042 	struct xfs_da_args	*args,
1043 	struct xfs_buf		*bp,
1044 	int			first,
1045 	int			last)
1046 {
1047 	xfs_dir2_leaf_entry_t	*firstlep;	/* pointer to first entry */
1048 	xfs_dir2_leaf_entry_t	*lastlep;	/* pointer to last entry */
1049 	struct xfs_dir2_leaf	*leaf = bp->b_addr;
1050 	struct xfs_dir2_leaf_entry *ents;
1051 
1052 	ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1053 	       leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1054 	       leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1055 	       leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1056 
1057 	ents = args->dp->d_ops->leaf_ents_p(leaf);
1058 	firstlep = &ents[first];
1059 	lastlep = &ents[last];
1060 	xfs_trans_log_buf(args->trans, bp,
1061 		(uint)((char *)firstlep - (char *)leaf),
1062 		(uint)((char *)lastlep - (char *)leaf + sizeof(*lastlep) - 1));
1063 }
1064 
1065 /*
1066  * Log the header of the leaf1 or leafn block.
1067  */
1068 void
1069 xfs_dir3_leaf_log_header(
1070 	struct xfs_da_args	*args,
1071 	struct xfs_buf		*bp)
1072 {
1073 	struct xfs_dir2_leaf	*leaf = bp->b_addr;
1074 
1075 	ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1076 	       leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1077 	       leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1078 	       leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1079 
1080 	xfs_trans_log_buf(args->trans, bp,
1081 			  (uint)((char *)&leaf->hdr - (char *)leaf),
1082 			  args->dp->d_ops->leaf_hdr_size - 1);
1083 }
1084 
1085 /*
1086  * Log the tail of the leaf1 block.
1087  */
1088 STATIC void
1089 xfs_dir3_leaf_log_tail(
1090 	struct xfs_da_args	*args,
1091 	struct xfs_buf		*bp)
1092 {
1093 	struct xfs_dir2_leaf	*leaf = bp->b_addr;
1094 	xfs_dir2_leaf_tail_t	*ltp;		/* leaf tail structure */
1095 
1096 	ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1097 	       leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1098 	       leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1099 	       leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1100 
1101 	ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1102 	xfs_trans_log_buf(args->trans, bp, (uint)((char *)ltp - (char *)leaf),
1103 		(uint)(args->geo->blksize - 1));
1104 }
1105 
1106 /*
1107  * Look up the entry referred to by args in the leaf format directory.
1108  * Most of the work is done by the xfs_dir2_leaf_lookup_int routine which
1109  * is also used by the node-format code.
1110  */
1111 int
1112 xfs_dir2_leaf_lookup(
1113 	xfs_da_args_t		*args)		/* operation arguments */
1114 {
1115 	struct xfs_buf		*dbp;		/* data block buffer */
1116 	xfs_dir2_data_entry_t	*dep;		/* data block entry */
1117 	xfs_inode_t		*dp;		/* incore directory inode */
1118 	int			error;		/* error return code */
1119 	int			index;		/* found entry index */
1120 	struct xfs_buf		*lbp;		/* leaf buffer */
1121 	xfs_dir2_leaf_t		*leaf;		/* leaf structure */
1122 	xfs_dir2_leaf_entry_t	*lep;		/* leaf entry */
1123 	xfs_trans_t		*tp;		/* transaction pointer */
1124 	struct xfs_dir2_leaf_entry *ents;
1125 
1126 	trace_xfs_dir2_leaf_lookup(args);
1127 
1128 	/*
1129 	 * Look up name in the leaf block, returning both buffers and index.
1130 	 */
1131 	if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1132 		return error;
1133 	}
1134 	tp = args->trans;
1135 	dp = args->dp;
1136 	xfs_dir3_leaf_check(dp, lbp);
1137 	leaf = lbp->b_addr;
1138 	ents = dp->d_ops->leaf_ents_p(leaf);
1139 	/*
1140 	 * Get to the leaf entry and contained data entry address.
1141 	 */
1142 	lep = &ents[index];
1143 
1144 	/*
1145 	 * Point to the data entry.
1146 	 */
1147 	dep = (xfs_dir2_data_entry_t *)
1148 	      ((char *)dbp->b_addr +
1149 	       xfs_dir2_dataptr_to_off(args->geo, be32_to_cpu(lep->address)));
1150 	/*
1151 	 * Return the found inode number & CI name if appropriate
1152 	 */
1153 	args->inumber = be64_to_cpu(dep->inumber);
1154 	args->filetype = dp->d_ops->data_get_ftype(dep);
1155 	error = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
1156 	xfs_trans_brelse(tp, dbp);
1157 	xfs_trans_brelse(tp, lbp);
1158 	return error;
1159 }
1160 
1161 /*
1162  * Look up name/hash in the leaf block.
1163  * Fill in indexp with the found index, and dbpp with the data buffer.
1164  * If not found dbpp will be NULL, and ENOENT comes back.
1165  * lbpp will always be filled in with the leaf buffer unless there's an error.
1166  */
1167 static int					/* error */
1168 xfs_dir2_leaf_lookup_int(
1169 	xfs_da_args_t		*args,		/* operation arguments */
1170 	struct xfs_buf		**lbpp,		/* out: leaf buffer */
1171 	int			*indexp,	/* out: index in leaf block */
1172 	struct xfs_buf		**dbpp)		/* out: data buffer */
1173 {
1174 	xfs_dir2_db_t		curdb = -1;	/* current data block number */
1175 	struct xfs_buf		*dbp = NULL;	/* data buffer */
1176 	xfs_dir2_data_entry_t	*dep;		/* data entry */
1177 	xfs_inode_t		*dp;		/* incore directory inode */
1178 	int			error;		/* error return code */
1179 	int			index;		/* index in leaf block */
1180 	struct xfs_buf		*lbp;		/* leaf buffer */
1181 	xfs_dir2_leaf_entry_t	*lep;		/* leaf entry */
1182 	xfs_dir2_leaf_t		*leaf;		/* leaf structure */
1183 	xfs_mount_t		*mp;		/* filesystem mount point */
1184 	xfs_dir2_db_t		newdb;		/* new data block number */
1185 	xfs_trans_t		*tp;		/* transaction pointer */
1186 	xfs_dir2_db_t		cidb = -1;	/* case match data block no. */
1187 	enum xfs_dacmp		cmp;		/* name compare result */
1188 	struct xfs_dir2_leaf_entry *ents;
1189 	struct xfs_dir3_icleaf_hdr leafhdr;
1190 
1191 	dp = args->dp;
1192 	tp = args->trans;
1193 	mp = dp->i_mount;
1194 
1195 	error = xfs_dir3_leaf_read(tp, dp, args->geo->leafblk, -1, &lbp);
1196 	if (error)
1197 		return error;
1198 
1199 	*lbpp = lbp;
1200 	leaf = lbp->b_addr;
1201 	xfs_dir3_leaf_check(dp, lbp);
1202 	ents = dp->d_ops->leaf_ents_p(leaf);
1203 	dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1204 
1205 	/*
1206 	 * Look for the first leaf entry with our hash value.
1207 	 */
1208 	index = xfs_dir2_leaf_search_hash(args, lbp);
1209 	/*
1210 	 * Loop over all the entries with the right hash value
1211 	 * looking to match the name.
1212 	 */
1213 	for (lep = &ents[index];
1214 	     index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
1215 	     lep++, index++) {
1216 		/*
1217 		 * Skip over stale leaf entries.
1218 		 */
1219 		if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
1220 			continue;
1221 		/*
1222 		 * Get the new data block number.
1223 		 */
1224 		newdb = xfs_dir2_dataptr_to_db(args->geo,
1225 					       be32_to_cpu(lep->address));
1226 		/*
1227 		 * If it's not the same as the old data block number,
1228 		 * need to pitch the old one and read the new one.
1229 		 */
1230 		if (newdb != curdb) {
1231 			if (dbp)
1232 				xfs_trans_brelse(tp, dbp);
1233 			error = xfs_dir3_data_read(tp, dp,
1234 					   xfs_dir2_db_to_da(args->geo, newdb),
1235 					   -1, &dbp);
1236 			if (error) {
1237 				xfs_trans_brelse(tp, lbp);
1238 				return error;
1239 			}
1240 			curdb = newdb;
1241 		}
1242 		/*
1243 		 * Point to the data entry.
1244 		 */
1245 		dep = (xfs_dir2_data_entry_t *)((char *)dbp->b_addr +
1246 			xfs_dir2_dataptr_to_off(args->geo,
1247 						be32_to_cpu(lep->address)));
1248 		/*
1249 		 * Compare name and if it's an exact match, return the index
1250 		 * and buffer. If it's the first case-insensitive match, store
1251 		 * the index and buffer and continue looking for an exact match.
1252 		 */
1253 		cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
1254 		if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
1255 			args->cmpresult = cmp;
1256 			*indexp = index;
1257 			/* case exact match: return the current buffer. */
1258 			if (cmp == XFS_CMP_EXACT) {
1259 				*dbpp = dbp;
1260 				return 0;
1261 			}
1262 			cidb = curdb;
1263 		}
1264 	}
1265 	ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
1266 	/*
1267 	 * Here, we can only be doing a lookup (not a rename or remove).
1268 	 * If a case-insensitive match was found earlier, re-read the
1269 	 * appropriate data block if required and return it.
1270 	 */
1271 	if (args->cmpresult == XFS_CMP_CASE) {
1272 		ASSERT(cidb != -1);
1273 		if (cidb != curdb) {
1274 			xfs_trans_brelse(tp, dbp);
1275 			error = xfs_dir3_data_read(tp, dp,
1276 					   xfs_dir2_db_to_da(args->geo, cidb),
1277 					   -1, &dbp);
1278 			if (error) {
1279 				xfs_trans_brelse(tp, lbp);
1280 				return error;
1281 			}
1282 		}
1283 		*dbpp = dbp;
1284 		return 0;
1285 	}
1286 	/*
1287 	 * No match found, return -ENOENT.
1288 	 */
1289 	ASSERT(cidb == -1);
1290 	if (dbp)
1291 		xfs_trans_brelse(tp, dbp);
1292 	xfs_trans_brelse(tp, lbp);
1293 	return -ENOENT;
1294 }
1295 
1296 /*
1297  * Remove an entry from a leaf format directory.
1298  */
1299 int						/* error */
1300 xfs_dir2_leaf_removename(
1301 	xfs_da_args_t		*args)		/* operation arguments */
1302 {
1303 	__be16			*bestsp;	/* leaf block best freespace */
1304 	xfs_dir2_data_hdr_t	*hdr;		/* data block header */
1305 	xfs_dir2_db_t		db;		/* data block number */
1306 	struct xfs_buf		*dbp;		/* data block buffer */
1307 	xfs_dir2_data_entry_t	*dep;		/* data entry structure */
1308 	xfs_inode_t		*dp;		/* incore directory inode */
1309 	int			error;		/* error return code */
1310 	xfs_dir2_db_t		i;		/* temporary data block # */
1311 	int			index;		/* index into leaf entries */
1312 	struct xfs_buf		*lbp;		/* leaf buffer */
1313 	xfs_dir2_leaf_t		*leaf;		/* leaf structure */
1314 	xfs_dir2_leaf_entry_t	*lep;		/* leaf entry */
1315 	xfs_dir2_leaf_tail_t	*ltp;		/* leaf tail structure */
1316 	int			needlog;	/* need to log data header */
1317 	int			needscan;	/* need to rescan data frees */
1318 	xfs_dir2_data_off_t	oldbest;	/* old value of best free */
1319 	struct xfs_dir2_data_free *bf;		/* bestfree table */
1320 	struct xfs_dir2_leaf_entry *ents;
1321 	struct xfs_dir3_icleaf_hdr leafhdr;
1322 
1323 	trace_xfs_dir2_leaf_removename(args);
1324 
1325 	/*
1326 	 * Lookup the leaf entry, get the leaf and data blocks read in.
1327 	 */
1328 	if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1329 		return error;
1330 	}
1331 	dp = args->dp;
1332 	leaf = lbp->b_addr;
1333 	hdr = dbp->b_addr;
1334 	xfs_dir3_data_check(dp, dbp);
1335 	bf = dp->d_ops->data_bestfree_p(hdr);
1336 	dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1337 	ents = dp->d_ops->leaf_ents_p(leaf);
1338 	/*
1339 	 * Point to the leaf entry, use that to point to the data entry.
1340 	 */
1341 	lep = &ents[index];
1342 	db = xfs_dir2_dataptr_to_db(args->geo, be32_to_cpu(lep->address));
1343 	dep = (xfs_dir2_data_entry_t *)((char *)hdr +
1344 		xfs_dir2_dataptr_to_off(args->geo, be32_to_cpu(lep->address)));
1345 	needscan = needlog = 0;
1346 	oldbest = be16_to_cpu(bf[0].length);
1347 	ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1348 	bestsp = xfs_dir2_leaf_bests_p(ltp);
1349 	if (be16_to_cpu(bestsp[db]) != oldbest)
1350 		return -EFSCORRUPTED;
1351 	/*
1352 	 * Mark the former data entry unused.
1353 	 */
1354 	xfs_dir2_data_make_free(args, dbp,
1355 		(xfs_dir2_data_aoff_t)((char *)dep - (char *)hdr),
1356 		dp->d_ops->data_entsize(dep->namelen), &needlog, &needscan);
1357 	/*
1358 	 * We just mark the leaf entry stale by putting a null in it.
1359 	 */
1360 	leafhdr.stale++;
1361 	dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
1362 	xfs_dir3_leaf_log_header(args, lbp);
1363 
1364 	lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1365 	xfs_dir3_leaf_log_ents(args, lbp, index, index);
1366 
1367 	/*
1368 	 * Scan the freespace in the data block again if necessary,
1369 	 * log the data block header if necessary.
1370 	 */
1371 	if (needscan)
1372 		xfs_dir2_data_freescan(dp, hdr, &needlog);
1373 	if (needlog)
1374 		xfs_dir2_data_log_header(args, dbp);
1375 	/*
1376 	 * If the longest freespace in the data block has changed,
1377 	 * put the new value in the bests table and log that.
1378 	 */
1379 	if (be16_to_cpu(bf[0].length) != oldbest) {
1380 		bestsp[db] = bf[0].length;
1381 		xfs_dir3_leaf_log_bests(args, lbp, db, db);
1382 	}
1383 	xfs_dir3_data_check(dp, dbp);
1384 	/*
1385 	 * If the data block is now empty then get rid of the data block.
1386 	 */
1387 	if (be16_to_cpu(bf[0].length) ==
1388 			args->geo->blksize - dp->d_ops->data_entry_offset) {
1389 		ASSERT(db != args->geo->datablk);
1390 		if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1391 			/*
1392 			 * Nope, can't get rid of it because it caused
1393 			 * allocation of a bmap btree block to do so.
1394 			 * Just go on, returning success, leaving the
1395 			 * empty block in place.
1396 			 */
1397 			if (error == -ENOSPC && args->total == 0)
1398 				error = 0;
1399 			xfs_dir3_leaf_check(dp, lbp);
1400 			return error;
1401 		}
1402 		dbp = NULL;
1403 		/*
1404 		 * If this is the last data block then compact the
1405 		 * bests table by getting rid of entries.
1406 		 */
1407 		if (db == be32_to_cpu(ltp->bestcount) - 1) {
1408 			/*
1409 			 * Look for the last active entry (i).
1410 			 */
1411 			for (i = db - 1; i > 0; i--) {
1412 				if (bestsp[i] != cpu_to_be16(NULLDATAOFF))
1413 					break;
1414 			}
1415 			/*
1416 			 * Copy the table down so inactive entries at the
1417 			 * end are removed.
1418 			 */
1419 			memmove(&bestsp[db - i], bestsp,
1420 				(be32_to_cpu(ltp->bestcount) - (db - i)) * sizeof(*bestsp));
1421 			be32_add_cpu(&ltp->bestcount, -(db - i));
1422 			xfs_dir3_leaf_log_tail(args, lbp);
1423 			xfs_dir3_leaf_log_bests(args, lbp, 0,
1424 						be32_to_cpu(ltp->bestcount) - 1);
1425 		} else
1426 			bestsp[db] = cpu_to_be16(NULLDATAOFF);
1427 	}
1428 	/*
1429 	 * If the data block was not the first one, drop it.
1430 	 */
1431 	else if (db != args->geo->datablk)
1432 		dbp = NULL;
1433 
1434 	xfs_dir3_leaf_check(dp, lbp);
1435 	/*
1436 	 * See if we can convert to block form.
1437 	 */
1438 	return xfs_dir2_leaf_to_block(args, lbp, dbp);
1439 }
1440 
1441 /*
1442  * Replace the inode number in a leaf format directory entry.
1443  */
1444 int						/* error */
1445 xfs_dir2_leaf_replace(
1446 	xfs_da_args_t		*args)		/* operation arguments */
1447 {
1448 	struct xfs_buf		*dbp;		/* data block buffer */
1449 	xfs_dir2_data_entry_t	*dep;		/* data block entry */
1450 	xfs_inode_t		*dp;		/* incore directory inode */
1451 	int			error;		/* error return code */
1452 	int			index;		/* index of leaf entry */
1453 	struct xfs_buf		*lbp;		/* leaf buffer */
1454 	xfs_dir2_leaf_t		*leaf;		/* leaf structure */
1455 	xfs_dir2_leaf_entry_t	*lep;		/* leaf entry */
1456 	xfs_trans_t		*tp;		/* transaction pointer */
1457 	struct xfs_dir2_leaf_entry *ents;
1458 
1459 	trace_xfs_dir2_leaf_replace(args);
1460 
1461 	/*
1462 	 * Look up the entry.
1463 	 */
1464 	if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1465 		return error;
1466 	}
1467 	dp = args->dp;
1468 	leaf = lbp->b_addr;
1469 	ents = dp->d_ops->leaf_ents_p(leaf);
1470 	/*
1471 	 * Point to the leaf entry, get data address from it.
1472 	 */
1473 	lep = &ents[index];
1474 	/*
1475 	 * Point to the data entry.
1476 	 */
1477 	dep = (xfs_dir2_data_entry_t *)
1478 	      ((char *)dbp->b_addr +
1479 	       xfs_dir2_dataptr_to_off(args->geo, be32_to_cpu(lep->address)));
1480 	ASSERT(args->inumber != be64_to_cpu(dep->inumber));
1481 	/*
1482 	 * Put the new inode number in, log it.
1483 	 */
1484 	dep->inumber = cpu_to_be64(args->inumber);
1485 	dp->d_ops->data_put_ftype(dep, args->filetype);
1486 	tp = args->trans;
1487 	xfs_dir2_data_log_entry(args, dbp, dep);
1488 	xfs_dir3_leaf_check(dp, lbp);
1489 	xfs_trans_brelse(tp, lbp);
1490 	return 0;
1491 }
1492 
1493 /*
1494  * Return index in the leaf block (lbp) which is either the first
1495  * one with this hash value, or if there are none, the insert point
1496  * for that hash value.
1497  */
1498 int						/* index value */
1499 xfs_dir2_leaf_search_hash(
1500 	xfs_da_args_t		*args,		/* operation arguments */
1501 	struct xfs_buf		*lbp)		/* leaf buffer */
1502 {
1503 	xfs_dahash_t		hash=0;		/* hash from this entry */
1504 	xfs_dahash_t		hashwant;	/* hash value looking for */
1505 	int			high;		/* high leaf index */
1506 	int			low;		/* low leaf index */
1507 	xfs_dir2_leaf_t		*leaf;		/* leaf structure */
1508 	xfs_dir2_leaf_entry_t	*lep;		/* leaf entry */
1509 	int			mid=0;		/* current leaf index */
1510 	struct xfs_dir2_leaf_entry *ents;
1511 	struct xfs_dir3_icleaf_hdr leafhdr;
1512 
1513 	leaf = lbp->b_addr;
1514 	ents = args->dp->d_ops->leaf_ents_p(leaf);
1515 	args->dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1516 
1517 	/*
1518 	 * Note, the table cannot be empty, so we have to go through the loop.
1519 	 * Binary search the leaf entries looking for our hash value.
1520 	 */
1521 	for (lep = ents, low = 0, high = leafhdr.count - 1,
1522 		hashwant = args->hashval;
1523 	     low <= high; ) {
1524 		mid = (low + high) >> 1;
1525 		if ((hash = be32_to_cpu(lep[mid].hashval)) == hashwant)
1526 			break;
1527 		if (hash < hashwant)
1528 			low = mid + 1;
1529 		else
1530 			high = mid - 1;
1531 	}
1532 	/*
1533 	 * Found one, back up through all the equal hash values.
1534 	 */
1535 	if (hash == hashwant) {
1536 		while (mid > 0 && be32_to_cpu(lep[mid - 1].hashval) == hashwant) {
1537 			mid--;
1538 		}
1539 	}
1540 	/*
1541 	 * Need to point to an entry higher than ours.
1542 	 */
1543 	else if (hash < hashwant)
1544 		mid++;
1545 	return mid;
1546 }
1547 
1548 /*
1549  * Trim off a trailing data block.  We know it's empty since the leaf
1550  * freespace table says so.
1551  */
1552 int						/* error */
1553 xfs_dir2_leaf_trim_data(
1554 	xfs_da_args_t		*args,		/* operation arguments */
1555 	struct xfs_buf		*lbp,		/* leaf buffer */
1556 	xfs_dir2_db_t		db)		/* data block number */
1557 {
1558 	__be16			*bestsp;	/* leaf bests table */
1559 	struct xfs_buf		*dbp;		/* data block buffer */
1560 	xfs_inode_t		*dp;		/* incore directory inode */
1561 	int			error;		/* error return value */
1562 	xfs_dir2_leaf_t		*leaf;		/* leaf structure */
1563 	xfs_dir2_leaf_tail_t	*ltp;		/* leaf tail structure */
1564 	xfs_trans_t		*tp;		/* transaction pointer */
1565 
1566 	dp = args->dp;
1567 	tp = args->trans;
1568 	/*
1569 	 * Read the offending data block.  We need its buffer.
1570 	 */
1571 	error = xfs_dir3_data_read(tp, dp, xfs_dir2_db_to_da(args->geo, db),
1572 				   -1, &dbp);
1573 	if (error)
1574 		return error;
1575 
1576 	leaf = lbp->b_addr;
1577 	ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1578 
1579 #ifdef DEBUG
1580 {
1581 	struct xfs_dir2_data_hdr *hdr = dbp->b_addr;
1582 	struct xfs_dir2_data_free *bf = dp->d_ops->data_bestfree_p(hdr);
1583 
1584 	ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
1585 	       hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC));
1586 	ASSERT(be16_to_cpu(bf[0].length) ==
1587 	       args->geo->blksize - dp->d_ops->data_entry_offset);
1588 	ASSERT(db == be32_to_cpu(ltp->bestcount) - 1);
1589 }
1590 #endif
1591 
1592 	/*
1593 	 * Get rid of the data block.
1594 	 */
1595 	if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1596 		ASSERT(error != -ENOSPC);
1597 		xfs_trans_brelse(tp, dbp);
1598 		return error;
1599 	}
1600 	/*
1601 	 * Eliminate the last bests entry from the table.
1602 	 */
1603 	bestsp = xfs_dir2_leaf_bests_p(ltp);
1604 	be32_add_cpu(&ltp->bestcount, -1);
1605 	memmove(&bestsp[1], &bestsp[0], be32_to_cpu(ltp->bestcount) * sizeof(*bestsp));
1606 	xfs_dir3_leaf_log_tail(args, lbp);
1607 	xfs_dir3_leaf_log_bests(args, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1608 	return 0;
1609 }
1610 
1611 static inline size_t
1612 xfs_dir3_leaf_size(
1613 	struct xfs_dir3_icleaf_hdr	*hdr,
1614 	int				counts)
1615 {
1616 	int	entries;
1617 	int	hdrsize;
1618 
1619 	entries = hdr->count - hdr->stale;
1620 	if (hdr->magic == XFS_DIR2_LEAF1_MAGIC ||
1621 	    hdr->magic == XFS_DIR2_LEAFN_MAGIC)
1622 		hdrsize = sizeof(struct xfs_dir2_leaf_hdr);
1623 	else
1624 		hdrsize = sizeof(struct xfs_dir3_leaf_hdr);
1625 
1626 	return hdrsize + entries * sizeof(xfs_dir2_leaf_entry_t)
1627 	               + counts * sizeof(xfs_dir2_data_off_t)
1628 		       + sizeof(xfs_dir2_leaf_tail_t);
1629 }
1630 
1631 /*
1632  * Convert node form directory to leaf form directory.
1633  * The root of the node form dir needs to already be a LEAFN block.
1634  * Just return if we can't do anything.
1635  */
1636 int						/* error */
1637 xfs_dir2_node_to_leaf(
1638 	xfs_da_state_t		*state)		/* directory operation state */
1639 {
1640 	xfs_da_args_t		*args;		/* operation arguments */
1641 	xfs_inode_t		*dp;		/* incore directory inode */
1642 	int			error;		/* error return code */
1643 	struct xfs_buf		*fbp;		/* buffer for freespace block */
1644 	xfs_fileoff_t		fo;		/* freespace file offset */
1645 	xfs_dir2_free_t		*free;		/* freespace structure */
1646 	struct xfs_buf		*lbp;		/* buffer for leaf block */
1647 	xfs_dir2_leaf_tail_t	*ltp;		/* tail of leaf structure */
1648 	xfs_dir2_leaf_t		*leaf;		/* leaf structure */
1649 	xfs_mount_t		*mp;		/* filesystem mount point */
1650 	int			rval;		/* successful free trim? */
1651 	xfs_trans_t		*tp;		/* transaction pointer */
1652 	struct xfs_dir3_icleaf_hdr leafhdr;
1653 	struct xfs_dir3_icfree_hdr freehdr;
1654 
1655 	/*
1656 	 * There's more than a leaf level in the btree, so there must
1657 	 * be multiple leafn blocks.  Give up.
1658 	 */
1659 	if (state->path.active > 1)
1660 		return 0;
1661 	args = state->args;
1662 
1663 	trace_xfs_dir2_node_to_leaf(args);
1664 
1665 	mp = state->mp;
1666 	dp = args->dp;
1667 	tp = args->trans;
1668 	/*
1669 	 * Get the last offset in the file.
1670 	 */
1671 	if ((error = xfs_bmap_last_offset(dp, &fo, XFS_DATA_FORK))) {
1672 		return error;
1673 	}
1674 	fo -= args->geo->fsbcount;
1675 	/*
1676 	 * If there are freespace blocks other than the first one,
1677 	 * take this opportunity to remove trailing empty freespace blocks
1678 	 * that may have been left behind during no-space-reservation
1679 	 * operations.
1680 	 */
1681 	while (fo > args->geo->freeblk) {
1682 		if ((error = xfs_dir2_node_trim_free(args, fo, &rval))) {
1683 			return error;
1684 		}
1685 		if (rval)
1686 			fo -= args->geo->fsbcount;
1687 		else
1688 			return 0;
1689 	}
1690 	/*
1691 	 * Now find the block just before the freespace block.
1692 	 */
1693 	if ((error = xfs_bmap_last_before(tp, dp, &fo, XFS_DATA_FORK))) {
1694 		return error;
1695 	}
1696 	/*
1697 	 * If it's not the single leaf block, give up.
1698 	 */
1699 	if (XFS_FSB_TO_B(mp, fo) > XFS_DIR2_LEAF_OFFSET + args->geo->blksize)
1700 		return 0;
1701 	lbp = state->path.blk[0].bp;
1702 	leaf = lbp->b_addr;
1703 	dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1704 
1705 	ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC ||
1706 	       leafhdr.magic == XFS_DIR3_LEAFN_MAGIC);
1707 
1708 	/*
1709 	 * Read the freespace block.
1710 	 */
1711 	error = xfs_dir2_free_read(tp, dp,  args->geo->freeblk, &fbp);
1712 	if (error)
1713 		return error;
1714 	free = fbp->b_addr;
1715 	dp->d_ops->free_hdr_from_disk(&freehdr, free);
1716 
1717 	ASSERT(!freehdr.firstdb);
1718 
1719 	/*
1720 	 * Now see if the leafn and free data will fit in a leaf1.
1721 	 * If not, release the buffer and give up.
1722 	 */
1723 	if (xfs_dir3_leaf_size(&leafhdr, freehdr.nvalid) > args->geo->blksize) {
1724 		xfs_trans_brelse(tp, fbp);
1725 		return 0;
1726 	}
1727 
1728 	/*
1729 	 * If the leaf has any stale entries in it, compress them out.
1730 	 */
1731 	if (leafhdr.stale)
1732 		xfs_dir3_leaf_compact(args, &leafhdr, lbp);
1733 
1734 	lbp->b_ops = &xfs_dir3_leaf1_buf_ops;
1735 	xfs_trans_buf_set_type(tp, lbp, XFS_BLFT_DIR_LEAF1_BUF);
1736 	leafhdr.magic = (leafhdr.magic == XFS_DIR2_LEAFN_MAGIC)
1737 					? XFS_DIR2_LEAF1_MAGIC
1738 					: XFS_DIR3_LEAF1_MAGIC;
1739 
1740 	/*
1741 	 * Set up the leaf tail from the freespace block.
1742 	 */
1743 	ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1744 	ltp->bestcount = cpu_to_be32(freehdr.nvalid);
1745 
1746 	/*
1747 	 * Set up the leaf bests table.
1748 	 */
1749 	memcpy(xfs_dir2_leaf_bests_p(ltp), dp->d_ops->free_bests_p(free),
1750 		freehdr.nvalid * sizeof(xfs_dir2_data_off_t));
1751 
1752 	dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
1753 	xfs_dir3_leaf_log_header(args, lbp);
1754 	xfs_dir3_leaf_log_bests(args, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1755 	xfs_dir3_leaf_log_tail(args, lbp);
1756 	xfs_dir3_leaf_check(dp, lbp);
1757 
1758 	/*
1759 	 * Get rid of the freespace block.
1760 	 */
1761 	error = xfs_dir2_shrink_inode(args,
1762 			xfs_dir2_byte_to_db(args->geo, XFS_DIR2_FREE_OFFSET),
1763 			fbp);
1764 	if (error) {
1765 		/*
1766 		 * This can't fail here because it can only happen when
1767 		 * punching out the middle of an extent, and this is an
1768 		 * isolated block.
1769 		 */
1770 		ASSERT(error != -ENOSPC);
1771 		return error;
1772 	}
1773 	fbp = NULL;
1774 	/*
1775 	 * Now see if we can convert the single-leaf directory
1776 	 * down to a block form directory.
1777 	 * This routine always kills the dabuf for the leaf, so
1778 	 * eliminate it from the path.
1779 	 */
1780 	error = xfs_dir2_leaf_to_block(args, lbp, NULL);
1781 	state->path.blk[0].bp = NULL;
1782 	return error;
1783 }
1784