xref: /openbmc/linux/fs/xfs/libxfs/xfs_dir2_leaf.c (revision 349f631d)
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_shared.h"
10 #include "xfs_format.h"
11 #include "xfs_log_format.h"
12 #include "xfs_trans_resv.h"
13 #include "xfs_mount.h"
14 #include "xfs_inode.h"
15 #include "xfs_bmap.h"
16 #include "xfs_dir2.h"
17 #include "xfs_dir2_priv.h"
18 #include "xfs_error.h"
19 #include "xfs_trace.h"
20 #include "xfs_trans.h"
21 #include "xfs_buf_item.h"
22 
23 /*
24  * Local function declarations.
25  */
26 static int xfs_dir2_leaf_lookup_int(xfs_da_args_t *args, struct xfs_buf **lbpp,
27 				    int *indexp, struct xfs_buf **dbpp,
28 				    struct xfs_dir3_icleaf_hdr *leafhdr);
29 static void xfs_dir3_leaf_log_bests(struct xfs_da_args *args,
30 				    struct xfs_buf *bp, int first, int last);
31 static void xfs_dir3_leaf_log_tail(struct xfs_da_args *args,
32 				   struct xfs_buf *bp);
33 
34 void
35 xfs_dir2_leaf_hdr_from_disk(
36 	struct xfs_mount		*mp,
37 	struct xfs_dir3_icleaf_hdr	*to,
38 	struct xfs_dir2_leaf		*from)
39 {
40 	if (xfs_has_crc(mp)) {
41 		struct xfs_dir3_leaf *from3 = (struct xfs_dir3_leaf *)from;
42 
43 		to->forw = be32_to_cpu(from3->hdr.info.hdr.forw);
44 		to->back = be32_to_cpu(from3->hdr.info.hdr.back);
45 		to->magic = be16_to_cpu(from3->hdr.info.hdr.magic);
46 		to->count = be16_to_cpu(from3->hdr.count);
47 		to->stale = be16_to_cpu(from3->hdr.stale);
48 		to->ents = from3->__ents;
49 
50 		ASSERT(to->magic == XFS_DIR3_LEAF1_MAGIC ||
51 		       to->magic == XFS_DIR3_LEAFN_MAGIC);
52 	} else {
53 		to->forw = be32_to_cpu(from->hdr.info.forw);
54 		to->back = be32_to_cpu(from->hdr.info.back);
55 		to->magic = be16_to_cpu(from->hdr.info.magic);
56 		to->count = be16_to_cpu(from->hdr.count);
57 		to->stale = be16_to_cpu(from->hdr.stale);
58 		to->ents = from->__ents;
59 
60 		ASSERT(to->magic == XFS_DIR2_LEAF1_MAGIC ||
61 		       to->magic == XFS_DIR2_LEAFN_MAGIC);
62 	}
63 }
64 
65 void
66 xfs_dir2_leaf_hdr_to_disk(
67 	struct xfs_mount		*mp,
68 	struct xfs_dir2_leaf		*to,
69 	struct xfs_dir3_icleaf_hdr	*from)
70 {
71 	if (xfs_has_crc(mp)) {
72 		struct xfs_dir3_leaf *to3 = (struct xfs_dir3_leaf *)to;
73 
74 		ASSERT(from->magic == XFS_DIR3_LEAF1_MAGIC ||
75 		       from->magic == XFS_DIR3_LEAFN_MAGIC);
76 
77 		to3->hdr.info.hdr.forw = cpu_to_be32(from->forw);
78 		to3->hdr.info.hdr.back = cpu_to_be32(from->back);
79 		to3->hdr.info.hdr.magic = cpu_to_be16(from->magic);
80 		to3->hdr.count = cpu_to_be16(from->count);
81 		to3->hdr.stale = cpu_to_be16(from->stale);
82 	} else {
83 		ASSERT(from->magic == XFS_DIR2_LEAF1_MAGIC ||
84 		       from->magic == XFS_DIR2_LEAFN_MAGIC);
85 
86 		to->hdr.info.forw = cpu_to_be32(from->forw);
87 		to->hdr.info.back = cpu_to_be32(from->back);
88 		to->hdr.info.magic = cpu_to_be16(from->magic);
89 		to->hdr.count = cpu_to_be16(from->count);
90 		to->hdr.stale = cpu_to_be16(from->stale);
91 	}
92 }
93 
94 /*
95  * Check the internal consistency of a leaf1 block.
96  * Pop an assert if something is wrong.
97  */
98 #ifdef DEBUG
99 static xfs_failaddr_t
100 xfs_dir3_leaf1_check(
101 	struct xfs_inode	*dp,
102 	struct xfs_buf		*bp)
103 {
104 	struct xfs_dir2_leaf	*leaf = bp->b_addr;
105 	struct xfs_dir3_icleaf_hdr leafhdr;
106 
107 	xfs_dir2_leaf_hdr_from_disk(dp->i_mount, &leafhdr, leaf);
108 
109 	if (leafhdr.magic == XFS_DIR3_LEAF1_MAGIC) {
110 		struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
111 		if (be64_to_cpu(leaf3->info.blkno) != xfs_buf_daddr(bp))
112 			return __this_address;
113 	} else if (leafhdr.magic != XFS_DIR2_LEAF1_MAGIC)
114 		return __this_address;
115 
116 	return xfs_dir3_leaf_check_int(dp->i_mount, &leafhdr, leaf, false);
117 }
118 
119 static inline void
120 xfs_dir3_leaf_check(
121 	struct xfs_inode	*dp,
122 	struct xfs_buf		*bp)
123 {
124 	xfs_failaddr_t		fa;
125 
126 	fa = xfs_dir3_leaf1_check(dp, bp);
127 	if (!fa)
128 		return;
129 	xfs_corruption_error(__func__, XFS_ERRLEVEL_LOW, dp->i_mount,
130 			bp->b_addr, BBTOB(bp->b_length), __FILE__, __LINE__,
131 			fa);
132 	ASSERT(0);
133 }
134 #else
135 #define	xfs_dir3_leaf_check(dp, bp)
136 #endif
137 
138 xfs_failaddr_t
139 xfs_dir3_leaf_check_int(
140 	struct xfs_mount		*mp,
141 	struct xfs_dir3_icleaf_hdr	*hdr,
142 	struct xfs_dir2_leaf		*leaf,
143 	bool				expensive_checking)
144 {
145 	struct xfs_da_geometry		*geo = mp->m_dir_geo;
146 	xfs_dir2_leaf_tail_t		*ltp;
147 	int				stale;
148 	int				i;
149 
150 	ltp = xfs_dir2_leaf_tail_p(geo, leaf);
151 
152 	/*
153 	 * XXX (dgc): This value is not restrictive enough.
154 	 * Should factor in the size of the bests table as well.
155 	 * We can deduce a value for that from i_disk_size.
156 	 */
157 	if (hdr->count > geo->leaf_max_ents)
158 		return __this_address;
159 
160 	/* Leaves and bests don't overlap in leaf format. */
161 	if ((hdr->magic == XFS_DIR2_LEAF1_MAGIC ||
162 	     hdr->magic == XFS_DIR3_LEAF1_MAGIC) &&
163 	    (char *)&hdr->ents[hdr->count] > (char *)xfs_dir2_leaf_bests_p(ltp))
164 		return __this_address;
165 
166 	if (!expensive_checking)
167 		return NULL;
168 
169 	/* Check hash value order, count stale entries.  */
170 	for (i = stale = 0; i < hdr->count; i++) {
171 		if (i + 1 < hdr->count) {
172 			if (be32_to_cpu(hdr->ents[i].hashval) >
173 					be32_to_cpu(hdr->ents[i + 1].hashval))
174 				return __this_address;
175 		}
176 		if (hdr->ents[i].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
177 			stale++;
178 	}
179 	if (hdr->stale != stale)
180 		return __this_address;
181 	return NULL;
182 }
183 
184 /*
185  * We verify the magic numbers before decoding the leaf header so that on debug
186  * kernels we don't get assertion failures in xfs_dir3_leaf_hdr_from_disk() due
187  * to incorrect magic numbers.
188  */
189 static xfs_failaddr_t
190 xfs_dir3_leaf_verify(
191 	struct xfs_buf			*bp)
192 {
193 	struct xfs_mount		*mp = bp->b_mount;
194 	struct xfs_dir3_icleaf_hdr	leafhdr;
195 	xfs_failaddr_t			fa;
196 
197 	fa = xfs_da3_blkinfo_verify(bp, bp->b_addr);
198 	if (fa)
199 		return fa;
200 
201 	xfs_dir2_leaf_hdr_from_disk(mp, &leafhdr, bp->b_addr);
202 	return xfs_dir3_leaf_check_int(mp, &leafhdr, bp->b_addr, true);
203 }
204 
205 static void
206 xfs_dir3_leaf_read_verify(
207 	struct xfs_buf  *bp)
208 {
209 	struct xfs_mount	*mp = bp->b_mount;
210 	xfs_failaddr_t		fa;
211 
212 	if (xfs_has_crc(mp) &&
213 	     !xfs_buf_verify_cksum(bp, XFS_DIR3_LEAF_CRC_OFF))
214 		xfs_verifier_error(bp, -EFSBADCRC, __this_address);
215 	else {
216 		fa = xfs_dir3_leaf_verify(bp);
217 		if (fa)
218 			xfs_verifier_error(bp, -EFSCORRUPTED, fa);
219 	}
220 }
221 
222 static void
223 xfs_dir3_leaf_write_verify(
224 	struct xfs_buf  *bp)
225 {
226 	struct xfs_mount	*mp = bp->b_mount;
227 	struct xfs_buf_log_item	*bip = bp->b_log_item;
228 	struct xfs_dir3_leaf_hdr *hdr3 = bp->b_addr;
229 	xfs_failaddr_t		fa;
230 
231 	fa = xfs_dir3_leaf_verify(bp);
232 	if (fa) {
233 		xfs_verifier_error(bp, -EFSCORRUPTED, fa);
234 		return;
235 	}
236 
237 	if (!xfs_has_crc(mp))
238 		return;
239 
240 	if (bip)
241 		hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
242 
243 	xfs_buf_update_cksum(bp, XFS_DIR3_LEAF_CRC_OFF);
244 }
245 
246 const struct xfs_buf_ops xfs_dir3_leaf1_buf_ops = {
247 	.name = "xfs_dir3_leaf1",
248 	.magic16 = { cpu_to_be16(XFS_DIR2_LEAF1_MAGIC),
249 		     cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) },
250 	.verify_read = xfs_dir3_leaf_read_verify,
251 	.verify_write = xfs_dir3_leaf_write_verify,
252 	.verify_struct = xfs_dir3_leaf_verify,
253 };
254 
255 const struct xfs_buf_ops xfs_dir3_leafn_buf_ops = {
256 	.name = "xfs_dir3_leafn",
257 	.magic16 = { cpu_to_be16(XFS_DIR2_LEAFN_MAGIC),
258 		     cpu_to_be16(XFS_DIR3_LEAFN_MAGIC) },
259 	.verify_read = xfs_dir3_leaf_read_verify,
260 	.verify_write = xfs_dir3_leaf_write_verify,
261 	.verify_struct = xfs_dir3_leaf_verify,
262 };
263 
264 int
265 xfs_dir3_leaf_read(
266 	struct xfs_trans	*tp,
267 	struct xfs_inode	*dp,
268 	xfs_dablk_t		fbno,
269 	struct xfs_buf		**bpp)
270 {
271 	int			err;
272 
273 	err = xfs_da_read_buf(tp, dp, fbno, 0, bpp, XFS_DATA_FORK,
274 			&xfs_dir3_leaf1_buf_ops);
275 	if (!err && tp && *bpp)
276 		xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_LEAF1_BUF);
277 	return err;
278 }
279 
280 int
281 xfs_dir3_leafn_read(
282 	struct xfs_trans	*tp,
283 	struct xfs_inode	*dp,
284 	xfs_dablk_t		fbno,
285 	struct xfs_buf		**bpp)
286 {
287 	int			err;
288 
289 	err = xfs_da_read_buf(tp, dp, fbno, 0, bpp, XFS_DATA_FORK,
290 			&xfs_dir3_leafn_buf_ops);
291 	if (!err && tp && *bpp)
292 		xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_LEAFN_BUF);
293 	return err;
294 }
295 
296 /*
297  * Initialize a new leaf block, leaf1 or leafn magic accepted.
298  */
299 static void
300 xfs_dir3_leaf_init(
301 	struct xfs_mount	*mp,
302 	struct xfs_trans	*tp,
303 	struct xfs_buf		*bp,
304 	xfs_ino_t		owner,
305 	uint16_t		type)
306 {
307 	struct xfs_dir2_leaf	*leaf = bp->b_addr;
308 
309 	ASSERT(type == XFS_DIR2_LEAF1_MAGIC || type == XFS_DIR2_LEAFN_MAGIC);
310 
311 	if (xfs_has_crc(mp)) {
312 		struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
313 
314 		memset(leaf3, 0, sizeof(*leaf3));
315 
316 		leaf3->info.hdr.magic = (type == XFS_DIR2_LEAF1_MAGIC)
317 					 ? cpu_to_be16(XFS_DIR3_LEAF1_MAGIC)
318 					 : cpu_to_be16(XFS_DIR3_LEAFN_MAGIC);
319 		leaf3->info.blkno = cpu_to_be64(xfs_buf_daddr(bp));
320 		leaf3->info.owner = cpu_to_be64(owner);
321 		uuid_copy(&leaf3->info.uuid, &mp->m_sb.sb_meta_uuid);
322 	} else {
323 		memset(leaf, 0, sizeof(*leaf));
324 		leaf->hdr.info.magic = cpu_to_be16(type);
325 	}
326 
327 	/*
328 	 * If it's a leaf-format directory initialize the tail.
329 	 * Caller is responsible for initialising the bests table.
330 	 */
331 	if (type == XFS_DIR2_LEAF1_MAGIC) {
332 		struct xfs_dir2_leaf_tail *ltp;
333 
334 		ltp = xfs_dir2_leaf_tail_p(mp->m_dir_geo, leaf);
335 		ltp->bestcount = 0;
336 		bp->b_ops = &xfs_dir3_leaf1_buf_ops;
337 		xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAF1_BUF);
338 	} else {
339 		bp->b_ops = &xfs_dir3_leafn_buf_ops;
340 		xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAFN_BUF);
341 	}
342 }
343 
344 int
345 xfs_dir3_leaf_get_buf(
346 	xfs_da_args_t		*args,
347 	xfs_dir2_db_t		bno,
348 	struct xfs_buf		**bpp,
349 	uint16_t		magic)
350 {
351 	struct xfs_inode	*dp = args->dp;
352 	struct xfs_trans	*tp = args->trans;
353 	struct xfs_mount	*mp = dp->i_mount;
354 	struct xfs_buf		*bp;
355 	int			error;
356 
357 	ASSERT(magic == XFS_DIR2_LEAF1_MAGIC || magic == XFS_DIR2_LEAFN_MAGIC);
358 	ASSERT(bno >= xfs_dir2_byte_to_db(args->geo, XFS_DIR2_LEAF_OFFSET) &&
359 	       bno < xfs_dir2_byte_to_db(args->geo, XFS_DIR2_FREE_OFFSET));
360 
361 	error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(args->geo, bno),
362 			       &bp, XFS_DATA_FORK);
363 	if (error)
364 		return error;
365 
366 	xfs_dir3_leaf_init(mp, tp, bp, dp->i_ino, magic);
367 	xfs_dir3_leaf_log_header(args, bp);
368 	if (magic == XFS_DIR2_LEAF1_MAGIC)
369 		xfs_dir3_leaf_log_tail(args, bp);
370 	*bpp = bp;
371 	return 0;
372 }
373 
374 /*
375  * Convert a block form directory to a leaf form directory.
376  */
377 int						/* error */
378 xfs_dir2_block_to_leaf(
379 	xfs_da_args_t		*args,		/* operation arguments */
380 	struct xfs_buf		*dbp)		/* input block's buffer */
381 {
382 	__be16			*bestsp;	/* leaf's bestsp entries */
383 	xfs_dablk_t		blkno;		/* leaf block's bno */
384 	xfs_dir2_data_hdr_t	*hdr;		/* block header */
385 	xfs_dir2_leaf_entry_t	*blp;		/* block's leaf entries */
386 	xfs_dir2_block_tail_t	*btp;		/* block's tail */
387 	xfs_inode_t		*dp;		/* incore directory inode */
388 	int			error;		/* error return code */
389 	struct xfs_buf		*lbp;		/* leaf block's buffer */
390 	xfs_dir2_db_t		ldb;		/* leaf block's bno */
391 	xfs_dir2_leaf_t		*leaf;		/* leaf structure */
392 	xfs_dir2_leaf_tail_t	*ltp;		/* leaf's tail */
393 	int			needlog;	/* need to log block header */
394 	int			needscan;	/* need to rescan bestfree */
395 	xfs_trans_t		*tp;		/* transaction pointer */
396 	struct xfs_dir2_data_free *bf;
397 	struct xfs_dir3_icleaf_hdr leafhdr;
398 
399 	trace_xfs_dir2_block_to_leaf(args);
400 
401 	dp = args->dp;
402 	tp = args->trans;
403 	/*
404 	 * Add the leaf block to the inode.
405 	 * This interface will only put blocks in the leaf/node range.
406 	 * Since that's empty now, we'll get the root (block 0 in range).
407 	 */
408 	if ((error = xfs_da_grow_inode(args, &blkno))) {
409 		return error;
410 	}
411 	ldb = xfs_dir2_da_to_db(args->geo, blkno);
412 	ASSERT(ldb == xfs_dir2_byte_to_db(args->geo, XFS_DIR2_LEAF_OFFSET));
413 	/*
414 	 * Initialize the leaf block, get a buffer for it.
415 	 */
416 	error = xfs_dir3_leaf_get_buf(args, ldb, &lbp, XFS_DIR2_LEAF1_MAGIC);
417 	if (error)
418 		return error;
419 
420 	leaf = lbp->b_addr;
421 	hdr = dbp->b_addr;
422 	xfs_dir3_data_check(dp, dbp);
423 	btp = xfs_dir2_block_tail_p(args->geo, hdr);
424 	blp = xfs_dir2_block_leaf_p(btp);
425 	bf = xfs_dir2_data_bestfree_p(dp->i_mount, hdr);
426 
427 	/*
428 	 * Set the counts in the leaf header.
429 	 */
430 	xfs_dir2_leaf_hdr_from_disk(dp->i_mount, &leafhdr, leaf);
431 	leafhdr.count = be32_to_cpu(btp->count);
432 	leafhdr.stale = be32_to_cpu(btp->stale);
433 	xfs_dir2_leaf_hdr_to_disk(dp->i_mount, leaf, &leafhdr);
434 	xfs_dir3_leaf_log_header(args, lbp);
435 
436 	/*
437 	 * Could compact these but I think we always do the conversion
438 	 * after squeezing out stale entries.
439 	 */
440 	memcpy(leafhdr.ents, blp,
441 		be32_to_cpu(btp->count) * sizeof(struct xfs_dir2_leaf_entry));
442 	xfs_dir3_leaf_log_ents(args, &leafhdr, lbp, 0, leafhdr.count - 1);
443 	needscan = 0;
444 	needlog = 1;
445 	/*
446 	 * Make the space formerly occupied by the leaf entries and block
447 	 * tail be free.
448 	 */
449 	xfs_dir2_data_make_free(args, dbp,
450 		(xfs_dir2_data_aoff_t)((char *)blp - (char *)hdr),
451 		(xfs_dir2_data_aoff_t)((char *)hdr + args->geo->blksize -
452 				       (char *)blp),
453 		&needlog, &needscan);
454 	/*
455 	 * Fix up the block header, make it a data block.
456 	 */
457 	dbp->b_ops = &xfs_dir3_data_buf_ops;
458 	xfs_trans_buf_set_type(tp, dbp, XFS_BLFT_DIR_DATA_BUF);
459 	if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC))
460 		hdr->magic = cpu_to_be32(XFS_DIR2_DATA_MAGIC);
461 	else
462 		hdr->magic = cpu_to_be32(XFS_DIR3_DATA_MAGIC);
463 
464 	if (needscan)
465 		xfs_dir2_data_freescan(dp->i_mount, hdr, &needlog);
466 	/*
467 	 * Set up leaf tail and bests table.
468 	 */
469 	ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
470 	ltp->bestcount = cpu_to_be32(1);
471 	bestsp = xfs_dir2_leaf_bests_p(ltp);
472 	bestsp[0] =  bf[0].length;
473 	/*
474 	 * Log the data header and leaf bests table.
475 	 */
476 	if (needlog)
477 		xfs_dir2_data_log_header(args, dbp);
478 	xfs_dir3_leaf_check(dp, lbp);
479 	xfs_dir3_data_check(dp, dbp);
480 	xfs_dir3_leaf_log_bests(args, lbp, 0, 0);
481 	return 0;
482 }
483 
484 STATIC void
485 xfs_dir3_leaf_find_stale(
486 	struct xfs_dir3_icleaf_hdr *leafhdr,
487 	struct xfs_dir2_leaf_entry *ents,
488 	int			index,
489 	int			*lowstale,
490 	int			*highstale)
491 {
492 	/*
493 	 * Find the first stale entry before our index, if any.
494 	 */
495 	for (*lowstale = index - 1; *lowstale >= 0; --*lowstale) {
496 		if (ents[*lowstale].address ==
497 		    cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
498 			break;
499 	}
500 
501 	/*
502 	 * Find the first stale entry at or after our index, if any.
503 	 * Stop if the result would require moving more entries than using
504 	 * lowstale.
505 	 */
506 	for (*highstale = index; *highstale < leafhdr->count; ++*highstale) {
507 		if (ents[*highstale].address ==
508 		    cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
509 			break;
510 		if (*lowstale >= 0 && index - *lowstale <= *highstale - index)
511 			break;
512 	}
513 }
514 
515 struct xfs_dir2_leaf_entry *
516 xfs_dir3_leaf_find_entry(
517 	struct xfs_dir3_icleaf_hdr *leafhdr,
518 	struct xfs_dir2_leaf_entry *ents,
519 	int			index,		/* leaf table position */
520 	int			compact,	/* need to compact leaves */
521 	int			lowstale,	/* index of prev stale leaf */
522 	int			highstale,	/* index of next stale leaf */
523 	int			*lfloglow,	/* low leaf logging index */
524 	int			*lfloghigh)	/* high leaf logging index */
525 {
526 	if (!leafhdr->stale) {
527 		xfs_dir2_leaf_entry_t	*lep;	/* leaf entry table pointer */
528 
529 		/*
530 		 * Now we need to make room to insert the leaf entry.
531 		 *
532 		 * If there are no stale entries, just insert a hole at index.
533 		 */
534 		lep = &ents[index];
535 		if (index < leafhdr->count)
536 			memmove(lep + 1, lep,
537 				(leafhdr->count - index) * sizeof(*lep));
538 
539 		/*
540 		 * Record low and high logging indices for the leaf.
541 		 */
542 		*lfloglow = index;
543 		*lfloghigh = leafhdr->count++;
544 		return lep;
545 	}
546 
547 	/*
548 	 * There are stale entries.
549 	 *
550 	 * We will use one of them for the new entry.  It's probably not at
551 	 * the right location, so we'll have to shift some up or down first.
552 	 *
553 	 * If we didn't compact before, we need to find the nearest stale
554 	 * entries before and after our insertion point.
555 	 */
556 	if (compact == 0)
557 		xfs_dir3_leaf_find_stale(leafhdr, ents, index,
558 					 &lowstale, &highstale);
559 
560 	/*
561 	 * If the low one is better, use it.
562 	 */
563 	if (lowstale >= 0 &&
564 	    (highstale == leafhdr->count ||
565 	     index - lowstale - 1 < highstale - index)) {
566 		ASSERT(index - lowstale - 1 >= 0);
567 		ASSERT(ents[lowstale].address ==
568 		       cpu_to_be32(XFS_DIR2_NULL_DATAPTR));
569 
570 		/*
571 		 * Copy entries up to cover the stale entry and make room
572 		 * for the new entry.
573 		 */
574 		if (index - lowstale - 1 > 0) {
575 			memmove(&ents[lowstale], &ents[lowstale + 1],
576 				(index - lowstale - 1) *
577 					sizeof(xfs_dir2_leaf_entry_t));
578 		}
579 		*lfloglow = min(lowstale, *lfloglow);
580 		*lfloghigh = max(index - 1, *lfloghigh);
581 		leafhdr->stale--;
582 		return &ents[index - 1];
583 	}
584 
585 	/*
586 	 * The high one is better, so use that one.
587 	 */
588 	ASSERT(highstale - index >= 0);
589 	ASSERT(ents[highstale].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR));
590 
591 	/*
592 	 * Copy entries down to cover the stale entry and make room for the
593 	 * new entry.
594 	 */
595 	if (highstale - index > 0) {
596 		memmove(&ents[index + 1], &ents[index],
597 			(highstale - index) * sizeof(xfs_dir2_leaf_entry_t));
598 	}
599 	*lfloglow = min(index, *lfloglow);
600 	*lfloghigh = max(highstale, *lfloghigh);
601 	leafhdr->stale--;
602 	return &ents[index];
603 }
604 
605 /*
606  * Add an entry to a leaf form directory.
607  */
608 int						/* error */
609 xfs_dir2_leaf_addname(
610 	struct xfs_da_args	*args)		/* operation arguments */
611 {
612 	struct xfs_dir3_icleaf_hdr leafhdr;
613 	struct xfs_trans	*tp = args->trans;
614 	__be16			*bestsp;	/* freespace table in leaf */
615 	__be16			*tagp;		/* end of data entry */
616 	struct xfs_buf		*dbp;		/* data block buffer */
617 	struct xfs_buf		*lbp;		/* leaf's buffer */
618 	struct xfs_dir2_leaf	*leaf;		/* leaf structure */
619 	struct xfs_inode	*dp = args->dp;	/* incore directory inode */
620 	struct xfs_dir2_data_hdr *hdr;		/* data block header */
621 	struct xfs_dir2_data_entry *dep;	/* data block entry */
622 	struct xfs_dir2_leaf_entry *lep;	/* leaf entry table pointer */
623 	struct xfs_dir2_leaf_entry *ents;
624 	struct xfs_dir2_data_unused *dup;	/* data unused entry */
625 	struct xfs_dir2_leaf_tail *ltp;		/* leaf tail pointer */
626 	struct xfs_dir2_data_free *bf;		/* bestfree table */
627 	int			compact;	/* need to compact leaves */
628 	int			error;		/* error return value */
629 	int			grown;		/* allocated new data block */
630 	int			highstale = 0;	/* index of next stale leaf */
631 	int			i;		/* temporary, index */
632 	int			index;		/* leaf table position */
633 	int			length;		/* length of new entry */
634 	int			lfloglow;	/* low leaf logging index */
635 	int			lfloghigh;	/* high leaf logging index */
636 	int			lowstale = 0;	/* index of prev stale leaf */
637 	int			needbytes;	/* leaf block bytes needed */
638 	int			needlog;	/* need to log data header */
639 	int			needscan;	/* need to rescan data free */
640 	xfs_dir2_db_t		use_block;	/* data block number */
641 
642 	trace_xfs_dir2_leaf_addname(args);
643 
644 	error = xfs_dir3_leaf_read(tp, dp, args->geo->leafblk, &lbp);
645 	if (error)
646 		return error;
647 
648 	/*
649 	 * Look up the entry by hash value and name.
650 	 * We know it's not there, our caller has already done a lookup.
651 	 * So the index is of the entry to insert in front of.
652 	 * But if there are dup hash values the index is of the first of those.
653 	 */
654 	index = xfs_dir2_leaf_search_hash(args, lbp);
655 	leaf = lbp->b_addr;
656 	ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
657 	xfs_dir2_leaf_hdr_from_disk(dp->i_mount, &leafhdr, leaf);
658 	ents = leafhdr.ents;
659 	bestsp = xfs_dir2_leaf_bests_p(ltp);
660 	length = xfs_dir2_data_entsize(dp->i_mount, args->namelen);
661 
662 	/*
663 	 * See if there are any entries with the same hash value
664 	 * and space in their block for the new entry.
665 	 * This is good because it puts multiple same-hash value entries
666 	 * in a data block, improving the lookup of those entries.
667 	 */
668 	for (use_block = -1, lep = &ents[index];
669 	     index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
670 	     index++, lep++) {
671 		if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
672 			continue;
673 		i = xfs_dir2_dataptr_to_db(args->geo, be32_to_cpu(lep->address));
674 		ASSERT(i < be32_to_cpu(ltp->bestcount));
675 		ASSERT(bestsp[i] != cpu_to_be16(NULLDATAOFF));
676 		if (be16_to_cpu(bestsp[i]) >= length) {
677 			use_block = i;
678 			break;
679 		}
680 	}
681 	/*
682 	 * Didn't find a block yet, linear search all the data blocks.
683 	 */
684 	if (use_block == -1) {
685 		for (i = 0; i < be32_to_cpu(ltp->bestcount); i++) {
686 			/*
687 			 * Remember a block we see that's missing.
688 			 */
689 			if (bestsp[i] == cpu_to_be16(NULLDATAOFF) &&
690 			    use_block == -1)
691 				use_block = i;
692 			else if (be16_to_cpu(bestsp[i]) >= length) {
693 				use_block = i;
694 				break;
695 			}
696 		}
697 	}
698 	/*
699 	 * How many bytes do we need in the leaf block?
700 	 */
701 	needbytes = 0;
702 	if (!leafhdr.stale)
703 		needbytes += sizeof(xfs_dir2_leaf_entry_t);
704 	if (use_block == -1)
705 		needbytes += sizeof(xfs_dir2_data_off_t);
706 
707 	/*
708 	 * Now kill use_block if it refers to a missing block, so we
709 	 * can use it as an indication of allocation needed.
710 	 */
711 	if (use_block != -1 && bestsp[use_block] == cpu_to_be16(NULLDATAOFF))
712 		use_block = -1;
713 	/*
714 	 * If we don't have enough free bytes but we can make enough
715 	 * by compacting out stale entries, we'll do that.
716 	 */
717 	if ((char *)bestsp - (char *)&ents[leafhdr.count] < needbytes &&
718 	    leafhdr.stale > 1)
719 		compact = 1;
720 
721 	/*
722 	 * Otherwise if we don't have enough free bytes we need to
723 	 * convert to node form.
724 	 */
725 	else if ((char *)bestsp - (char *)&ents[leafhdr.count] < needbytes) {
726 		/*
727 		 * Just checking or no space reservation, give up.
728 		 */
729 		if ((args->op_flags & XFS_DA_OP_JUSTCHECK) ||
730 							args->total == 0) {
731 			xfs_trans_brelse(tp, lbp);
732 			return -ENOSPC;
733 		}
734 		/*
735 		 * Convert to node form.
736 		 */
737 		error = xfs_dir2_leaf_to_node(args, lbp);
738 		if (error)
739 			return error;
740 		/*
741 		 * Then add the new entry.
742 		 */
743 		return xfs_dir2_node_addname(args);
744 	}
745 	/*
746 	 * Otherwise it will fit without compaction.
747 	 */
748 	else
749 		compact = 0;
750 	/*
751 	 * If just checking, then it will fit unless we needed to allocate
752 	 * a new data block.
753 	 */
754 	if (args->op_flags & XFS_DA_OP_JUSTCHECK) {
755 		xfs_trans_brelse(tp, lbp);
756 		return use_block == -1 ? -ENOSPC : 0;
757 	}
758 	/*
759 	 * If no allocations are allowed, return now before we've
760 	 * changed anything.
761 	 */
762 	if (args->total == 0 && use_block == -1) {
763 		xfs_trans_brelse(tp, lbp);
764 		return -ENOSPC;
765 	}
766 	/*
767 	 * Need to compact the leaf entries, removing stale ones.
768 	 * Leave one stale entry behind - the one closest to our
769 	 * insertion index - and we'll shift that one to our insertion
770 	 * point later.
771 	 */
772 	if (compact) {
773 		xfs_dir3_leaf_compact_x1(&leafhdr, ents, &index, &lowstale,
774 			&highstale, &lfloglow, &lfloghigh);
775 	}
776 	/*
777 	 * There are stale entries, so we'll need log-low and log-high
778 	 * impossibly bad values later.
779 	 */
780 	else if (leafhdr.stale) {
781 		lfloglow = leafhdr.count;
782 		lfloghigh = -1;
783 	}
784 	/*
785 	 * If there was no data block space found, we need to allocate
786 	 * a new one.
787 	 */
788 	if (use_block == -1) {
789 		/*
790 		 * Add the new data block.
791 		 */
792 		if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE,
793 				&use_block))) {
794 			xfs_trans_brelse(tp, lbp);
795 			return error;
796 		}
797 		/*
798 		 * Initialize the block.
799 		 */
800 		if ((error = xfs_dir3_data_init(args, use_block, &dbp))) {
801 			xfs_trans_brelse(tp, lbp);
802 			return error;
803 		}
804 		/*
805 		 * If we're adding a new data block on the end we need to
806 		 * extend the bests table.  Copy it up one entry.
807 		 */
808 		if (use_block >= be32_to_cpu(ltp->bestcount)) {
809 			bestsp--;
810 			memmove(&bestsp[0], &bestsp[1],
811 				be32_to_cpu(ltp->bestcount) * sizeof(bestsp[0]));
812 			be32_add_cpu(&ltp->bestcount, 1);
813 			xfs_dir3_leaf_log_tail(args, lbp);
814 			xfs_dir3_leaf_log_bests(args, lbp, 0,
815 						be32_to_cpu(ltp->bestcount) - 1);
816 		}
817 		/*
818 		 * If we're filling in a previously empty block just log it.
819 		 */
820 		else
821 			xfs_dir3_leaf_log_bests(args, lbp, use_block, use_block);
822 		hdr = dbp->b_addr;
823 		bf = xfs_dir2_data_bestfree_p(dp->i_mount, hdr);
824 		bestsp[use_block] = bf[0].length;
825 		grown = 1;
826 	} else {
827 		/*
828 		 * Already had space in some data block.
829 		 * Just read that one in.
830 		 */
831 		error = xfs_dir3_data_read(tp, dp,
832 				   xfs_dir2_db_to_da(args->geo, use_block),
833 				   0, &dbp);
834 		if (error) {
835 			xfs_trans_brelse(tp, lbp);
836 			return error;
837 		}
838 		hdr = dbp->b_addr;
839 		bf = xfs_dir2_data_bestfree_p(dp->i_mount, hdr);
840 		grown = 0;
841 	}
842 	/*
843 	 * Point to the biggest freespace in our data block.
844 	 */
845 	dup = (xfs_dir2_data_unused_t *)
846 	      ((char *)hdr + be16_to_cpu(bf[0].offset));
847 	needscan = needlog = 0;
848 	/*
849 	 * Mark the initial part of our freespace in use for the new entry.
850 	 */
851 	error = xfs_dir2_data_use_free(args, dbp, dup,
852 			(xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr),
853 			length, &needlog, &needscan);
854 	if (error) {
855 		xfs_trans_brelse(tp, lbp);
856 		return error;
857 	}
858 	/*
859 	 * Initialize our new entry (at last).
860 	 */
861 	dep = (xfs_dir2_data_entry_t *)dup;
862 	dep->inumber = cpu_to_be64(args->inumber);
863 	dep->namelen = args->namelen;
864 	memcpy(dep->name, args->name, dep->namelen);
865 	xfs_dir2_data_put_ftype(dp->i_mount, dep, args->filetype);
866 	tagp = xfs_dir2_data_entry_tag_p(dp->i_mount, dep);
867 	*tagp = cpu_to_be16((char *)dep - (char *)hdr);
868 	/*
869 	 * Need to scan fix up the bestfree table.
870 	 */
871 	if (needscan)
872 		xfs_dir2_data_freescan(dp->i_mount, hdr, &needlog);
873 	/*
874 	 * Need to log the data block's header.
875 	 */
876 	if (needlog)
877 		xfs_dir2_data_log_header(args, dbp);
878 	xfs_dir2_data_log_entry(args, dbp, dep);
879 	/*
880 	 * If the bests table needs to be changed, do it.
881 	 * Log the change unless we've already done that.
882 	 */
883 	if (be16_to_cpu(bestsp[use_block]) != be16_to_cpu(bf[0].length)) {
884 		bestsp[use_block] = bf[0].length;
885 		if (!grown)
886 			xfs_dir3_leaf_log_bests(args, lbp, use_block, use_block);
887 	}
888 
889 	lep = xfs_dir3_leaf_find_entry(&leafhdr, ents, index, compact, lowstale,
890 				       highstale, &lfloglow, &lfloghigh);
891 
892 	/*
893 	 * Fill in the new leaf entry.
894 	 */
895 	lep->hashval = cpu_to_be32(args->hashval);
896 	lep->address = cpu_to_be32(
897 				xfs_dir2_db_off_to_dataptr(args->geo, use_block,
898 				be16_to_cpu(*tagp)));
899 	/*
900 	 * Log the leaf fields and give up the buffers.
901 	 */
902 	xfs_dir2_leaf_hdr_to_disk(dp->i_mount, leaf, &leafhdr);
903 	xfs_dir3_leaf_log_header(args, lbp);
904 	xfs_dir3_leaf_log_ents(args, &leafhdr, lbp, lfloglow, lfloghigh);
905 	xfs_dir3_leaf_check(dp, lbp);
906 	xfs_dir3_data_check(dp, dbp);
907 	return 0;
908 }
909 
910 /*
911  * Compact out any stale entries in the leaf.
912  * Log the header and changed leaf entries, if any.
913  */
914 void
915 xfs_dir3_leaf_compact(
916 	xfs_da_args_t	*args,		/* operation arguments */
917 	struct xfs_dir3_icleaf_hdr *leafhdr,
918 	struct xfs_buf	*bp)		/* leaf buffer */
919 {
920 	int		from;		/* source leaf index */
921 	xfs_dir2_leaf_t	*leaf;		/* leaf structure */
922 	int		loglow;		/* first leaf entry to log */
923 	int		to;		/* target leaf index */
924 	struct xfs_inode *dp = args->dp;
925 
926 	leaf = bp->b_addr;
927 	if (!leafhdr->stale)
928 		return;
929 
930 	/*
931 	 * Compress out the stale entries in place.
932 	 */
933 	for (from = to = 0, loglow = -1; from < leafhdr->count; from++) {
934 		if (leafhdr->ents[from].address ==
935 		    cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
936 			continue;
937 		/*
938 		 * Only actually copy the entries that are different.
939 		 */
940 		if (from > to) {
941 			if (loglow == -1)
942 				loglow = to;
943 			leafhdr->ents[to] = leafhdr->ents[from];
944 		}
945 		to++;
946 	}
947 	/*
948 	 * Update and log the header, log the leaf entries.
949 	 */
950 	ASSERT(leafhdr->stale == from - to);
951 	leafhdr->count -= leafhdr->stale;
952 	leafhdr->stale = 0;
953 
954 	xfs_dir2_leaf_hdr_to_disk(dp->i_mount, leaf, leafhdr);
955 	xfs_dir3_leaf_log_header(args, bp);
956 	if (loglow != -1)
957 		xfs_dir3_leaf_log_ents(args, leafhdr, bp, loglow, to - 1);
958 }
959 
960 /*
961  * Compact the leaf entries, removing stale ones.
962  * Leave one stale entry behind - the one closest to our
963  * insertion index - and the caller will shift that one to our insertion
964  * point later.
965  * Return new insertion index, where the remaining stale entry is,
966  * and leaf logging indices.
967  */
968 void
969 xfs_dir3_leaf_compact_x1(
970 	struct xfs_dir3_icleaf_hdr *leafhdr,
971 	struct xfs_dir2_leaf_entry *ents,
972 	int		*indexp,	/* insertion index */
973 	int		*lowstalep,	/* out: stale entry before us */
974 	int		*highstalep,	/* out: stale entry after us */
975 	int		*lowlogp,	/* out: low log index */
976 	int		*highlogp)	/* out: high log index */
977 {
978 	int		from;		/* source copy index */
979 	int		highstale;	/* stale entry at/after index */
980 	int		index;		/* insertion index */
981 	int		keepstale;	/* source index of kept stale */
982 	int		lowstale;	/* stale entry before index */
983 	int		newindex=0;	/* new insertion index */
984 	int		to;		/* destination copy index */
985 
986 	ASSERT(leafhdr->stale > 1);
987 	index = *indexp;
988 
989 	xfs_dir3_leaf_find_stale(leafhdr, ents, index, &lowstale, &highstale);
990 
991 	/*
992 	 * Pick the better of lowstale and highstale.
993 	 */
994 	if (lowstale >= 0 &&
995 	    (highstale == leafhdr->count ||
996 	     index - lowstale <= highstale - index))
997 		keepstale = lowstale;
998 	else
999 		keepstale = highstale;
1000 	/*
1001 	 * Copy the entries in place, removing all the stale entries
1002 	 * except keepstale.
1003 	 */
1004 	for (from = to = 0; from < leafhdr->count; from++) {
1005 		/*
1006 		 * Notice the new value of index.
1007 		 */
1008 		if (index == from)
1009 			newindex = to;
1010 		if (from != keepstale &&
1011 		    ents[from].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR)) {
1012 			if (from == to)
1013 				*lowlogp = to;
1014 			continue;
1015 		}
1016 		/*
1017 		 * Record the new keepstale value for the insertion.
1018 		 */
1019 		if (from == keepstale)
1020 			lowstale = highstale = to;
1021 		/*
1022 		 * Copy only the entries that have moved.
1023 		 */
1024 		if (from > to)
1025 			ents[to] = ents[from];
1026 		to++;
1027 	}
1028 	ASSERT(from > to);
1029 	/*
1030 	 * If the insertion point was past the last entry,
1031 	 * set the new insertion point accordingly.
1032 	 */
1033 	if (index == from)
1034 		newindex = to;
1035 	*indexp = newindex;
1036 	/*
1037 	 * Adjust the leaf header values.
1038 	 */
1039 	leafhdr->count -= from - to;
1040 	leafhdr->stale = 1;
1041 	/*
1042 	 * Remember the low/high stale value only in the "right"
1043 	 * direction.
1044 	 */
1045 	if (lowstale >= newindex)
1046 		lowstale = -1;
1047 	else
1048 		highstale = leafhdr->count;
1049 	*highlogp = leafhdr->count - 1;
1050 	*lowstalep = lowstale;
1051 	*highstalep = highstale;
1052 }
1053 
1054 /*
1055  * Log the bests entries indicated from a leaf1 block.
1056  */
1057 static void
1058 xfs_dir3_leaf_log_bests(
1059 	struct xfs_da_args	*args,
1060 	struct xfs_buf		*bp,		/* leaf buffer */
1061 	int			first,		/* first entry to log */
1062 	int			last)		/* last entry to log */
1063 {
1064 	__be16			*firstb;	/* pointer to first entry */
1065 	__be16			*lastb;		/* pointer to last entry */
1066 	struct xfs_dir2_leaf	*leaf = bp->b_addr;
1067 	xfs_dir2_leaf_tail_t	*ltp;		/* leaf tail structure */
1068 
1069 	ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1070 	       leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC));
1071 
1072 	ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1073 	firstb = xfs_dir2_leaf_bests_p(ltp) + first;
1074 	lastb = xfs_dir2_leaf_bests_p(ltp) + last;
1075 	xfs_trans_log_buf(args->trans, bp,
1076 		(uint)((char *)firstb - (char *)leaf),
1077 		(uint)((char *)lastb - (char *)leaf + sizeof(*lastb) - 1));
1078 }
1079 
1080 /*
1081  * Log the leaf entries indicated from a leaf1 or leafn block.
1082  */
1083 void
1084 xfs_dir3_leaf_log_ents(
1085 	struct xfs_da_args	*args,
1086 	struct xfs_dir3_icleaf_hdr *hdr,
1087 	struct xfs_buf		*bp,
1088 	int			first,
1089 	int			last)
1090 {
1091 	xfs_dir2_leaf_entry_t	*firstlep;	/* pointer to first entry */
1092 	xfs_dir2_leaf_entry_t	*lastlep;	/* pointer to last entry */
1093 	struct xfs_dir2_leaf	*leaf = bp->b_addr;
1094 
1095 	ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1096 	       leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1097 	       leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1098 	       leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1099 
1100 	firstlep = &hdr->ents[first];
1101 	lastlep = &hdr->ents[last];
1102 	xfs_trans_log_buf(args->trans, bp,
1103 		(uint)((char *)firstlep - (char *)leaf),
1104 		(uint)((char *)lastlep - (char *)leaf + sizeof(*lastlep) - 1));
1105 }
1106 
1107 /*
1108  * Log the header of the leaf1 or leafn block.
1109  */
1110 void
1111 xfs_dir3_leaf_log_header(
1112 	struct xfs_da_args	*args,
1113 	struct xfs_buf		*bp)
1114 {
1115 	struct xfs_dir2_leaf	*leaf = bp->b_addr;
1116 
1117 	ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1118 	       leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1119 	       leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1120 	       leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1121 
1122 	xfs_trans_log_buf(args->trans, bp,
1123 			  (uint)((char *)&leaf->hdr - (char *)leaf),
1124 			  args->geo->leaf_hdr_size - 1);
1125 }
1126 
1127 /*
1128  * Log the tail of the leaf1 block.
1129  */
1130 STATIC void
1131 xfs_dir3_leaf_log_tail(
1132 	struct xfs_da_args	*args,
1133 	struct xfs_buf		*bp)
1134 {
1135 	struct xfs_dir2_leaf	*leaf = bp->b_addr;
1136 	xfs_dir2_leaf_tail_t	*ltp;		/* leaf tail structure */
1137 
1138 	ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1139 	       leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1140 	       leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1141 	       leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1142 
1143 	ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1144 	xfs_trans_log_buf(args->trans, bp, (uint)((char *)ltp - (char *)leaf),
1145 		(uint)(args->geo->blksize - 1));
1146 }
1147 
1148 /*
1149  * Look up the entry referred to by args in the leaf format directory.
1150  * Most of the work is done by the xfs_dir2_leaf_lookup_int routine which
1151  * is also used by the node-format code.
1152  */
1153 int
1154 xfs_dir2_leaf_lookup(
1155 	xfs_da_args_t		*args)		/* operation arguments */
1156 {
1157 	struct xfs_buf		*dbp;		/* data block buffer */
1158 	xfs_dir2_data_entry_t	*dep;		/* data block entry */
1159 	xfs_inode_t		*dp;		/* incore directory inode */
1160 	int			error;		/* error return code */
1161 	int			index;		/* found entry index */
1162 	struct xfs_buf		*lbp;		/* leaf buffer */
1163 	xfs_dir2_leaf_entry_t	*lep;		/* leaf entry */
1164 	xfs_trans_t		*tp;		/* transaction pointer */
1165 	struct xfs_dir3_icleaf_hdr leafhdr;
1166 
1167 	trace_xfs_dir2_leaf_lookup(args);
1168 
1169 	/*
1170 	 * Look up name in the leaf block, returning both buffers and index.
1171 	 */
1172 	error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp, &leafhdr);
1173 	if (error)
1174 		return error;
1175 
1176 	tp = args->trans;
1177 	dp = args->dp;
1178 	xfs_dir3_leaf_check(dp, lbp);
1179 
1180 	/*
1181 	 * Get to the leaf entry and contained data entry address.
1182 	 */
1183 	lep = &leafhdr.ents[index];
1184 
1185 	/*
1186 	 * Point to the data entry.
1187 	 */
1188 	dep = (xfs_dir2_data_entry_t *)
1189 	      ((char *)dbp->b_addr +
1190 	       xfs_dir2_dataptr_to_off(args->geo, be32_to_cpu(lep->address)));
1191 	/*
1192 	 * Return the found inode number & CI name if appropriate
1193 	 */
1194 	args->inumber = be64_to_cpu(dep->inumber);
1195 	args->filetype = xfs_dir2_data_get_ftype(dp->i_mount, dep);
1196 	error = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
1197 	xfs_trans_brelse(tp, dbp);
1198 	xfs_trans_brelse(tp, lbp);
1199 	return error;
1200 }
1201 
1202 /*
1203  * Look up name/hash in the leaf block.
1204  * Fill in indexp with the found index, and dbpp with the data buffer.
1205  * If not found dbpp will be NULL, and ENOENT comes back.
1206  * lbpp will always be filled in with the leaf buffer unless there's an error.
1207  */
1208 static int					/* error */
1209 xfs_dir2_leaf_lookup_int(
1210 	xfs_da_args_t		*args,		/* operation arguments */
1211 	struct xfs_buf		**lbpp,		/* out: leaf buffer */
1212 	int			*indexp,	/* out: index in leaf block */
1213 	struct xfs_buf		**dbpp,		/* out: data buffer */
1214 	struct xfs_dir3_icleaf_hdr *leafhdr)
1215 {
1216 	xfs_dir2_db_t		curdb = -1;	/* current data block number */
1217 	struct xfs_buf		*dbp = NULL;	/* data buffer */
1218 	xfs_dir2_data_entry_t	*dep;		/* data entry */
1219 	xfs_inode_t		*dp;		/* incore directory inode */
1220 	int			error;		/* error return code */
1221 	int			index;		/* index in leaf block */
1222 	struct xfs_buf		*lbp;		/* leaf buffer */
1223 	xfs_dir2_leaf_entry_t	*lep;		/* leaf entry */
1224 	xfs_dir2_leaf_t		*leaf;		/* leaf structure */
1225 	xfs_mount_t		*mp;		/* filesystem mount point */
1226 	xfs_dir2_db_t		newdb;		/* new data block number */
1227 	xfs_trans_t		*tp;		/* transaction pointer */
1228 	xfs_dir2_db_t		cidb = -1;	/* case match data block no. */
1229 	enum xfs_dacmp		cmp;		/* name compare result */
1230 
1231 	dp = args->dp;
1232 	tp = args->trans;
1233 	mp = dp->i_mount;
1234 
1235 	error = xfs_dir3_leaf_read(tp, dp, args->geo->leafblk, &lbp);
1236 	if (error)
1237 		return error;
1238 
1239 	*lbpp = lbp;
1240 	leaf = lbp->b_addr;
1241 	xfs_dir3_leaf_check(dp, lbp);
1242 	xfs_dir2_leaf_hdr_from_disk(mp, leafhdr, leaf);
1243 
1244 	/*
1245 	 * Look for the first leaf entry with our hash value.
1246 	 */
1247 	index = xfs_dir2_leaf_search_hash(args, lbp);
1248 	/*
1249 	 * Loop over all the entries with the right hash value
1250 	 * looking to match the name.
1251 	 */
1252 	for (lep = &leafhdr->ents[index];
1253 	     index < leafhdr->count &&
1254 			be32_to_cpu(lep->hashval) == args->hashval;
1255 	     lep++, index++) {
1256 		/*
1257 		 * Skip over stale leaf entries.
1258 		 */
1259 		if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
1260 			continue;
1261 		/*
1262 		 * Get the new data block number.
1263 		 */
1264 		newdb = xfs_dir2_dataptr_to_db(args->geo,
1265 					       be32_to_cpu(lep->address));
1266 		/*
1267 		 * If it's not the same as the old data block number,
1268 		 * need to pitch the old one and read the new one.
1269 		 */
1270 		if (newdb != curdb) {
1271 			if (dbp)
1272 				xfs_trans_brelse(tp, dbp);
1273 			error = xfs_dir3_data_read(tp, dp,
1274 					   xfs_dir2_db_to_da(args->geo, newdb),
1275 					   0, &dbp);
1276 			if (error) {
1277 				xfs_trans_brelse(tp, lbp);
1278 				return error;
1279 			}
1280 			curdb = newdb;
1281 		}
1282 		/*
1283 		 * Point to the data entry.
1284 		 */
1285 		dep = (xfs_dir2_data_entry_t *)((char *)dbp->b_addr +
1286 			xfs_dir2_dataptr_to_off(args->geo,
1287 						be32_to_cpu(lep->address)));
1288 		/*
1289 		 * Compare name and if it's an exact match, return the index
1290 		 * and buffer. If it's the first case-insensitive match, store
1291 		 * the index and buffer and continue looking for an exact match.
1292 		 */
1293 		cmp = xfs_dir2_compname(args, dep->name, dep->namelen);
1294 		if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
1295 			args->cmpresult = cmp;
1296 			*indexp = index;
1297 			/* case exact match: return the current buffer. */
1298 			if (cmp == XFS_CMP_EXACT) {
1299 				*dbpp = dbp;
1300 				return 0;
1301 			}
1302 			cidb = curdb;
1303 		}
1304 	}
1305 	ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
1306 	/*
1307 	 * Here, we can only be doing a lookup (not a rename or remove).
1308 	 * If a case-insensitive match was found earlier, re-read the
1309 	 * appropriate data block if required and return it.
1310 	 */
1311 	if (args->cmpresult == XFS_CMP_CASE) {
1312 		ASSERT(cidb != -1);
1313 		if (cidb != curdb) {
1314 			xfs_trans_brelse(tp, dbp);
1315 			error = xfs_dir3_data_read(tp, dp,
1316 					   xfs_dir2_db_to_da(args->geo, cidb),
1317 					   0, &dbp);
1318 			if (error) {
1319 				xfs_trans_brelse(tp, lbp);
1320 				return error;
1321 			}
1322 		}
1323 		*dbpp = dbp;
1324 		return 0;
1325 	}
1326 	/*
1327 	 * No match found, return -ENOENT.
1328 	 */
1329 	ASSERT(cidb == -1);
1330 	if (dbp)
1331 		xfs_trans_brelse(tp, dbp);
1332 	xfs_trans_brelse(tp, lbp);
1333 	return -ENOENT;
1334 }
1335 
1336 /*
1337  * Remove an entry from a leaf format directory.
1338  */
1339 int						/* error */
1340 xfs_dir2_leaf_removename(
1341 	xfs_da_args_t		*args)		/* operation arguments */
1342 {
1343 	struct xfs_da_geometry	*geo = args->geo;
1344 	__be16			*bestsp;	/* leaf block best freespace */
1345 	xfs_dir2_data_hdr_t	*hdr;		/* data block header */
1346 	xfs_dir2_db_t		db;		/* data block number */
1347 	struct xfs_buf		*dbp;		/* data block buffer */
1348 	xfs_dir2_data_entry_t	*dep;		/* data entry structure */
1349 	xfs_inode_t		*dp;		/* incore directory inode */
1350 	int			error;		/* error return code */
1351 	xfs_dir2_db_t		i;		/* temporary data block # */
1352 	int			index;		/* index into leaf entries */
1353 	struct xfs_buf		*lbp;		/* leaf buffer */
1354 	xfs_dir2_leaf_t		*leaf;		/* leaf structure */
1355 	xfs_dir2_leaf_entry_t	*lep;		/* leaf entry */
1356 	xfs_dir2_leaf_tail_t	*ltp;		/* leaf tail structure */
1357 	int			needlog;	/* need to log data header */
1358 	int			needscan;	/* need to rescan data frees */
1359 	xfs_dir2_data_off_t	oldbest;	/* old value of best free */
1360 	struct xfs_dir2_data_free *bf;		/* bestfree table */
1361 	struct xfs_dir3_icleaf_hdr leafhdr;
1362 
1363 	trace_xfs_dir2_leaf_removename(args);
1364 
1365 	/*
1366 	 * Lookup the leaf entry, get the leaf and data blocks read in.
1367 	 */
1368 	error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp, &leafhdr);
1369 	if (error)
1370 		return error;
1371 
1372 	dp = args->dp;
1373 	leaf = lbp->b_addr;
1374 	hdr = dbp->b_addr;
1375 	xfs_dir3_data_check(dp, dbp);
1376 	bf = xfs_dir2_data_bestfree_p(dp->i_mount, hdr);
1377 
1378 	/*
1379 	 * Point to the leaf entry, use that to point to the data entry.
1380 	 */
1381 	lep = &leafhdr.ents[index];
1382 	db = xfs_dir2_dataptr_to_db(geo, be32_to_cpu(lep->address));
1383 	dep = (xfs_dir2_data_entry_t *)((char *)hdr +
1384 		xfs_dir2_dataptr_to_off(geo, be32_to_cpu(lep->address)));
1385 	needscan = needlog = 0;
1386 	oldbest = be16_to_cpu(bf[0].length);
1387 	ltp = xfs_dir2_leaf_tail_p(geo, leaf);
1388 	bestsp = xfs_dir2_leaf_bests_p(ltp);
1389 	if (be16_to_cpu(bestsp[db]) != oldbest) {
1390 		xfs_buf_mark_corrupt(lbp);
1391 		return -EFSCORRUPTED;
1392 	}
1393 	/*
1394 	 * Mark the former data entry unused.
1395 	 */
1396 	xfs_dir2_data_make_free(args, dbp,
1397 		(xfs_dir2_data_aoff_t)((char *)dep - (char *)hdr),
1398 		xfs_dir2_data_entsize(dp->i_mount, dep->namelen), &needlog,
1399 		&needscan);
1400 	/*
1401 	 * We just mark the leaf entry stale by putting a null in it.
1402 	 */
1403 	leafhdr.stale++;
1404 	xfs_dir2_leaf_hdr_to_disk(dp->i_mount, leaf, &leafhdr);
1405 	xfs_dir3_leaf_log_header(args, lbp);
1406 
1407 	lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1408 	xfs_dir3_leaf_log_ents(args, &leafhdr, lbp, index, index);
1409 
1410 	/*
1411 	 * Scan the freespace in the data block again if necessary,
1412 	 * log the data block header if necessary.
1413 	 */
1414 	if (needscan)
1415 		xfs_dir2_data_freescan(dp->i_mount, hdr, &needlog);
1416 	if (needlog)
1417 		xfs_dir2_data_log_header(args, dbp);
1418 	/*
1419 	 * If the longest freespace in the data block has changed,
1420 	 * put the new value in the bests table and log that.
1421 	 */
1422 	if (be16_to_cpu(bf[0].length) != oldbest) {
1423 		bestsp[db] = bf[0].length;
1424 		xfs_dir3_leaf_log_bests(args, lbp, db, db);
1425 	}
1426 	xfs_dir3_data_check(dp, dbp);
1427 	/*
1428 	 * If the data block is now empty then get rid of the data block.
1429 	 */
1430 	if (be16_to_cpu(bf[0].length) ==
1431 	    geo->blksize - geo->data_entry_offset) {
1432 		ASSERT(db != geo->datablk);
1433 		if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1434 			/*
1435 			 * Nope, can't get rid of it because it caused
1436 			 * allocation of a bmap btree block to do so.
1437 			 * Just go on, returning success, leaving the
1438 			 * empty block in place.
1439 			 */
1440 			if (error == -ENOSPC && args->total == 0)
1441 				error = 0;
1442 			xfs_dir3_leaf_check(dp, lbp);
1443 			return error;
1444 		}
1445 		dbp = NULL;
1446 		/*
1447 		 * If this is the last data block then compact the
1448 		 * bests table by getting rid of entries.
1449 		 */
1450 		if (db == be32_to_cpu(ltp->bestcount) - 1) {
1451 			/*
1452 			 * Look for the last active entry (i).
1453 			 */
1454 			for (i = db - 1; i > 0; i--) {
1455 				if (bestsp[i] != cpu_to_be16(NULLDATAOFF))
1456 					break;
1457 			}
1458 			/*
1459 			 * Copy the table down so inactive entries at the
1460 			 * end are removed.
1461 			 */
1462 			memmove(&bestsp[db - i], bestsp,
1463 				(be32_to_cpu(ltp->bestcount) - (db - i)) * sizeof(*bestsp));
1464 			be32_add_cpu(&ltp->bestcount, -(db - i));
1465 			xfs_dir3_leaf_log_tail(args, lbp);
1466 			xfs_dir3_leaf_log_bests(args, lbp, 0,
1467 						be32_to_cpu(ltp->bestcount) - 1);
1468 		} else
1469 			bestsp[db] = cpu_to_be16(NULLDATAOFF);
1470 	}
1471 	/*
1472 	 * If the data block was not the first one, drop it.
1473 	 */
1474 	else if (db != geo->datablk)
1475 		dbp = NULL;
1476 
1477 	xfs_dir3_leaf_check(dp, lbp);
1478 	/*
1479 	 * See if we can convert to block form.
1480 	 */
1481 	return xfs_dir2_leaf_to_block(args, lbp, dbp);
1482 }
1483 
1484 /*
1485  * Replace the inode number in a leaf format directory entry.
1486  */
1487 int						/* error */
1488 xfs_dir2_leaf_replace(
1489 	xfs_da_args_t		*args)		/* operation arguments */
1490 {
1491 	struct xfs_buf		*dbp;		/* data block buffer */
1492 	xfs_dir2_data_entry_t	*dep;		/* data block entry */
1493 	xfs_inode_t		*dp;		/* incore directory inode */
1494 	int			error;		/* error return code */
1495 	int			index;		/* index of leaf entry */
1496 	struct xfs_buf		*lbp;		/* leaf buffer */
1497 	xfs_dir2_leaf_entry_t	*lep;		/* leaf entry */
1498 	xfs_trans_t		*tp;		/* transaction pointer */
1499 	struct xfs_dir3_icleaf_hdr leafhdr;
1500 
1501 	trace_xfs_dir2_leaf_replace(args);
1502 
1503 	/*
1504 	 * Look up the entry.
1505 	 */
1506 	error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp, &leafhdr);
1507 	if (error)
1508 		return error;
1509 
1510 	dp = args->dp;
1511 	/*
1512 	 * Point to the leaf entry, get data address from it.
1513 	 */
1514 	lep = &leafhdr.ents[index];
1515 	/*
1516 	 * Point to the data entry.
1517 	 */
1518 	dep = (xfs_dir2_data_entry_t *)
1519 	      ((char *)dbp->b_addr +
1520 	       xfs_dir2_dataptr_to_off(args->geo, be32_to_cpu(lep->address)));
1521 	ASSERT(args->inumber != be64_to_cpu(dep->inumber));
1522 	/*
1523 	 * Put the new inode number in, log it.
1524 	 */
1525 	dep->inumber = cpu_to_be64(args->inumber);
1526 	xfs_dir2_data_put_ftype(dp->i_mount, dep, args->filetype);
1527 	tp = args->trans;
1528 	xfs_dir2_data_log_entry(args, dbp, dep);
1529 	xfs_dir3_leaf_check(dp, lbp);
1530 	xfs_trans_brelse(tp, lbp);
1531 	return 0;
1532 }
1533 
1534 /*
1535  * Return index in the leaf block (lbp) which is either the first
1536  * one with this hash value, or if there are none, the insert point
1537  * for that hash value.
1538  */
1539 int						/* index value */
1540 xfs_dir2_leaf_search_hash(
1541 	xfs_da_args_t		*args,		/* operation arguments */
1542 	struct xfs_buf		*lbp)		/* leaf buffer */
1543 {
1544 	xfs_dahash_t		hash=0;		/* hash from this entry */
1545 	xfs_dahash_t		hashwant;	/* hash value looking for */
1546 	int			high;		/* high leaf index */
1547 	int			low;		/* low leaf index */
1548 	xfs_dir2_leaf_entry_t	*lep;		/* leaf entry */
1549 	int			mid=0;		/* current leaf index */
1550 	struct xfs_dir3_icleaf_hdr leafhdr;
1551 
1552 	xfs_dir2_leaf_hdr_from_disk(args->dp->i_mount, &leafhdr, lbp->b_addr);
1553 
1554 	/*
1555 	 * Note, the table cannot be empty, so we have to go through the loop.
1556 	 * Binary search the leaf entries looking for our hash value.
1557 	 */
1558 	for (lep = leafhdr.ents, low = 0, high = leafhdr.count - 1,
1559 		hashwant = args->hashval;
1560 	     low <= high; ) {
1561 		mid = (low + high) >> 1;
1562 		if ((hash = be32_to_cpu(lep[mid].hashval)) == hashwant)
1563 			break;
1564 		if (hash < hashwant)
1565 			low = mid + 1;
1566 		else
1567 			high = mid - 1;
1568 	}
1569 	/*
1570 	 * Found one, back up through all the equal hash values.
1571 	 */
1572 	if (hash == hashwant) {
1573 		while (mid > 0 && be32_to_cpu(lep[mid - 1].hashval) == hashwant) {
1574 			mid--;
1575 		}
1576 	}
1577 	/*
1578 	 * Need to point to an entry higher than ours.
1579 	 */
1580 	else if (hash < hashwant)
1581 		mid++;
1582 	return mid;
1583 }
1584 
1585 /*
1586  * Trim off a trailing data block.  We know it's empty since the leaf
1587  * freespace table says so.
1588  */
1589 int						/* error */
1590 xfs_dir2_leaf_trim_data(
1591 	xfs_da_args_t		*args,		/* operation arguments */
1592 	struct xfs_buf		*lbp,		/* leaf buffer */
1593 	xfs_dir2_db_t		db)		/* data block number */
1594 {
1595 	struct xfs_da_geometry	*geo = args->geo;
1596 	__be16			*bestsp;	/* leaf bests table */
1597 	struct xfs_buf		*dbp;		/* data block buffer */
1598 	xfs_inode_t		*dp;		/* incore directory inode */
1599 	int			error;		/* error return value */
1600 	xfs_dir2_leaf_t		*leaf;		/* leaf structure */
1601 	xfs_dir2_leaf_tail_t	*ltp;		/* leaf tail structure */
1602 	xfs_trans_t		*tp;		/* transaction pointer */
1603 
1604 	dp = args->dp;
1605 	tp = args->trans;
1606 	/*
1607 	 * Read the offending data block.  We need its buffer.
1608 	 */
1609 	error = xfs_dir3_data_read(tp, dp, xfs_dir2_db_to_da(geo, db), 0, &dbp);
1610 	if (error)
1611 		return error;
1612 
1613 	leaf = lbp->b_addr;
1614 	ltp = xfs_dir2_leaf_tail_p(geo, leaf);
1615 
1616 #ifdef DEBUG
1617 {
1618 	struct xfs_dir2_data_hdr *hdr = dbp->b_addr;
1619 	struct xfs_dir2_data_free *bf =
1620 		xfs_dir2_data_bestfree_p(dp->i_mount, hdr);
1621 
1622 	ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
1623 	       hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC));
1624 	ASSERT(be16_to_cpu(bf[0].length) ==
1625 	       geo->blksize - geo->data_entry_offset);
1626 	ASSERT(db == be32_to_cpu(ltp->bestcount) - 1);
1627 }
1628 #endif
1629 
1630 	/*
1631 	 * Get rid of the data block.
1632 	 */
1633 	if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1634 		ASSERT(error != -ENOSPC);
1635 		xfs_trans_brelse(tp, dbp);
1636 		return error;
1637 	}
1638 	/*
1639 	 * Eliminate the last bests entry from the table.
1640 	 */
1641 	bestsp = xfs_dir2_leaf_bests_p(ltp);
1642 	be32_add_cpu(&ltp->bestcount, -1);
1643 	memmove(&bestsp[1], &bestsp[0], be32_to_cpu(ltp->bestcount) * sizeof(*bestsp));
1644 	xfs_dir3_leaf_log_tail(args, lbp);
1645 	xfs_dir3_leaf_log_bests(args, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1646 	return 0;
1647 }
1648 
1649 static inline size_t
1650 xfs_dir3_leaf_size(
1651 	struct xfs_dir3_icleaf_hdr	*hdr,
1652 	int				counts)
1653 {
1654 	int	entries;
1655 	int	hdrsize;
1656 
1657 	entries = hdr->count - hdr->stale;
1658 	if (hdr->magic == XFS_DIR2_LEAF1_MAGIC ||
1659 	    hdr->magic == XFS_DIR2_LEAFN_MAGIC)
1660 		hdrsize = sizeof(struct xfs_dir2_leaf_hdr);
1661 	else
1662 		hdrsize = sizeof(struct xfs_dir3_leaf_hdr);
1663 
1664 	return hdrsize + entries * sizeof(xfs_dir2_leaf_entry_t)
1665 	               + counts * sizeof(xfs_dir2_data_off_t)
1666 		       + sizeof(xfs_dir2_leaf_tail_t);
1667 }
1668 
1669 /*
1670  * Convert node form directory to leaf form directory.
1671  * The root of the node form dir needs to already be a LEAFN block.
1672  * Just return if we can't do anything.
1673  */
1674 int						/* error */
1675 xfs_dir2_node_to_leaf(
1676 	xfs_da_state_t		*state)		/* directory operation state */
1677 {
1678 	xfs_da_args_t		*args;		/* operation arguments */
1679 	xfs_inode_t		*dp;		/* incore directory inode */
1680 	int			error;		/* error return code */
1681 	struct xfs_buf		*fbp;		/* buffer for freespace block */
1682 	xfs_fileoff_t		fo;		/* freespace file offset */
1683 	struct xfs_buf		*lbp;		/* buffer for leaf block */
1684 	xfs_dir2_leaf_tail_t	*ltp;		/* tail of leaf structure */
1685 	xfs_dir2_leaf_t		*leaf;		/* leaf structure */
1686 	xfs_mount_t		*mp;		/* filesystem mount point */
1687 	int			rval;		/* successful free trim? */
1688 	xfs_trans_t		*tp;		/* transaction pointer */
1689 	struct xfs_dir3_icleaf_hdr leafhdr;
1690 	struct xfs_dir3_icfree_hdr freehdr;
1691 
1692 	/*
1693 	 * There's more than a leaf level in the btree, so there must
1694 	 * be multiple leafn blocks.  Give up.
1695 	 */
1696 	if (state->path.active > 1)
1697 		return 0;
1698 	args = state->args;
1699 
1700 	trace_xfs_dir2_node_to_leaf(args);
1701 
1702 	mp = state->mp;
1703 	dp = args->dp;
1704 	tp = args->trans;
1705 	/*
1706 	 * Get the last offset in the file.
1707 	 */
1708 	if ((error = xfs_bmap_last_offset(dp, &fo, XFS_DATA_FORK))) {
1709 		return error;
1710 	}
1711 	fo -= args->geo->fsbcount;
1712 	/*
1713 	 * If there are freespace blocks other than the first one,
1714 	 * take this opportunity to remove trailing empty freespace blocks
1715 	 * that may have been left behind during no-space-reservation
1716 	 * operations.
1717 	 */
1718 	while (fo > args->geo->freeblk) {
1719 		if ((error = xfs_dir2_node_trim_free(args, fo, &rval))) {
1720 			return error;
1721 		}
1722 		if (rval)
1723 			fo -= args->geo->fsbcount;
1724 		else
1725 			return 0;
1726 	}
1727 	/*
1728 	 * Now find the block just before the freespace block.
1729 	 */
1730 	if ((error = xfs_bmap_last_before(tp, dp, &fo, XFS_DATA_FORK))) {
1731 		return error;
1732 	}
1733 	/*
1734 	 * If it's not the single leaf block, give up.
1735 	 */
1736 	if (XFS_FSB_TO_B(mp, fo) > XFS_DIR2_LEAF_OFFSET + args->geo->blksize)
1737 		return 0;
1738 	lbp = state->path.blk[0].bp;
1739 	leaf = lbp->b_addr;
1740 	xfs_dir2_leaf_hdr_from_disk(mp, &leafhdr, leaf);
1741 
1742 	ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC ||
1743 	       leafhdr.magic == XFS_DIR3_LEAFN_MAGIC);
1744 
1745 	/*
1746 	 * Read the freespace block.
1747 	 */
1748 	error = xfs_dir2_free_read(tp, dp,  args->geo->freeblk, &fbp);
1749 	if (error)
1750 		return error;
1751 	xfs_dir2_free_hdr_from_disk(mp, &freehdr, fbp->b_addr);
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), freehdr.bests,
1786 		freehdr.nvalid * sizeof(xfs_dir2_data_off_t));
1787 
1788 	xfs_dir2_leaf_hdr_to_disk(mp, 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