• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
3  * Copyright (c) 2013 Red Hat, Inc.
4  * All Rights Reserved.
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public License as
8  * published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it would be useful,
11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13  * GNU General Public License for more details.
14  *
15  * You should have received a copy of the GNU General Public License
16  * along with this program; if not, write the Free Software Foundation,
17  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
18  */
19 #include "xfs.h"
20 #include "xfs_fs.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
24 #include "xfs_sb.h"
25 #include "xfs_ag.h"
26 #include "xfs_mount.h"
27 #include "xfs_da_format.h"
28 #include "xfs_da_btree.h"
29 #include "xfs_inode.h"
30 #include "xfs_trans.h"
31 #include "xfs_inode_item.h"
32 #include "xfs_bmap.h"
33 #include "xfs_buf_item.h"
34 #include "xfs_dir2.h"
35 #include "xfs_dir2_priv.h"
36 #include "xfs_error.h"
37 #include "xfs_trace.h"
38 #include "xfs_cksum.h"
39 #include "xfs_dinode.h"
40 
41 /*
42  * Local function prototypes.
43  */
44 static void xfs_dir2_block_log_leaf(xfs_trans_t *tp, struct xfs_buf *bp,
45 				    int first, int last);
46 static void xfs_dir2_block_log_tail(xfs_trans_t *tp, struct xfs_buf *bp);
47 static int xfs_dir2_block_lookup_int(xfs_da_args_t *args, struct xfs_buf **bpp,
48 				     int *entno);
49 static int xfs_dir2_block_sort(const void *a, const void *b);
50 
51 static xfs_dahash_t xfs_dir_hash_dot, xfs_dir_hash_dotdot;
52 
53 /*
54  * One-time startup routine called from xfs_init().
55  */
56 void
xfs_dir_startup(void)57 xfs_dir_startup(void)
58 {
59 	xfs_dir_hash_dot = xfs_da_hashname((unsigned char *)".", 1);
60 	xfs_dir_hash_dotdot = xfs_da_hashname((unsigned char *)"..", 2);
61 }
62 
63 static bool
xfs_dir3_block_verify(struct xfs_buf * bp)64 xfs_dir3_block_verify(
65 	struct xfs_buf		*bp)
66 {
67 	struct xfs_mount	*mp = bp->b_target->bt_mount;
68 	struct xfs_dir3_blk_hdr	*hdr3 = bp->b_addr;
69 
70 	if (xfs_sb_version_hascrc(&mp->m_sb)) {
71 		if (hdr3->magic != cpu_to_be32(XFS_DIR3_BLOCK_MAGIC))
72 			return false;
73 		if (!uuid_equal(&hdr3->uuid, &mp->m_sb.sb_uuid))
74 			return false;
75 		if (be64_to_cpu(hdr3->blkno) != bp->b_bn)
76 			return false;
77 	} else {
78 		if (hdr3->magic != cpu_to_be32(XFS_DIR2_BLOCK_MAGIC))
79 			return false;
80 	}
81 	if (__xfs_dir3_data_check(NULL, bp))
82 		return false;
83 	return true;
84 }
85 
86 static void
xfs_dir3_block_read_verify(struct xfs_buf * bp)87 xfs_dir3_block_read_verify(
88 	struct xfs_buf	*bp)
89 {
90 	struct xfs_mount	*mp = bp->b_target->bt_mount;
91 
92 	if (xfs_sb_version_hascrc(&mp->m_sb) &&
93 	     !xfs_buf_verify_cksum(bp, XFS_DIR3_DATA_CRC_OFF))
94 		xfs_buf_ioerror(bp, -EFSBADCRC);
95 	else if (!xfs_dir3_block_verify(bp))
96 		xfs_buf_ioerror(bp, -EFSCORRUPTED);
97 
98 	if (bp->b_error)
99 		xfs_verifier_error(bp);
100 }
101 
102 static void
xfs_dir3_block_write_verify(struct xfs_buf * bp)103 xfs_dir3_block_write_verify(
104 	struct xfs_buf	*bp)
105 {
106 	struct xfs_mount	*mp = bp->b_target->bt_mount;
107 	struct xfs_buf_log_item	*bip = bp->b_fspriv;
108 	struct xfs_dir3_blk_hdr	*hdr3 = bp->b_addr;
109 
110 	if (!xfs_dir3_block_verify(bp)) {
111 		xfs_buf_ioerror(bp, -EFSCORRUPTED);
112 		xfs_verifier_error(bp);
113 		return;
114 	}
115 
116 	if (!xfs_sb_version_hascrc(&mp->m_sb))
117 		return;
118 
119 	if (bip)
120 		hdr3->lsn = cpu_to_be64(bip->bli_item.li_lsn);
121 
122 	xfs_buf_update_cksum(bp, XFS_DIR3_DATA_CRC_OFF);
123 }
124 
125 const struct xfs_buf_ops xfs_dir3_block_buf_ops = {
126 	.name = "xfs_dir3_block",
127 	.verify_read = xfs_dir3_block_read_verify,
128 	.verify_write = xfs_dir3_block_write_verify,
129 };
130 
131 int
xfs_dir3_block_read(struct xfs_trans * tp,struct xfs_inode * dp,struct xfs_buf ** bpp)132 xfs_dir3_block_read(
133 	struct xfs_trans	*tp,
134 	struct xfs_inode	*dp,
135 	struct xfs_buf		**bpp)
136 {
137 	struct xfs_mount	*mp = dp->i_mount;
138 	int			err;
139 
140 	err = xfs_da_read_buf(tp, dp, mp->m_dir_geo->datablk, -1, bpp,
141 				XFS_DATA_FORK, &xfs_dir3_block_buf_ops);
142 	if (!err && tp)
143 		xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_BLOCK_BUF);
144 	return err;
145 }
146 
147 static void
xfs_dir3_block_init(struct xfs_mount * mp,struct xfs_trans * tp,struct xfs_buf * bp,struct xfs_inode * dp)148 xfs_dir3_block_init(
149 	struct xfs_mount	*mp,
150 	struct xfs_trans	*tp,
151 	struct xfs_buf		*bp,
152 	struct xfs_inode	*dp)
153 {
154 	struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
155 
156 	bp->b_ops = &xfs_dir3_block_buf_ops;
157 	xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_BLOCK_BUF);
158 
159 	if (xfs_sb_version_hascrc(&mp->m_sb)) {
160 		memset(hdr3, 0, sizeof(*hdr3));
161 		hdr3->magic = cpu_to_be32(XFS_DIR3_BLOCK_MAGIC);
162 		hdr3->blkno = cpu_to_be64(bp->b_bn);
163 		hdr3->owner = cpu_to_be64(dp->i_ino);
164 		uuid_copy(&hdr3->uuid, &mp->m_sb.sb_uuid);
165 		return;
166 
167 	}
168 	hdr3->magic = cpu_to_be32(XFS_DIR2_BLOCK_MAGIC);
169 }
170 
171 static void
xfs_dir2_block_need_space(struct xfs_inode * dp,struct xfs_dir2_data_hdr * hdr,struct xfs_dir2_block_tail * btp,struct xfs_dir2_leaf_entry * blp,__be16 ** tagpp,struct xfs_dir2_data_unused ** dupp,struct xfs_dir2_data_unused ** enddupp,int * compact,int len)172 xfs_dir2_block_need_space(
173 	struct xfs_inode		*dp,
174 	struct xfs_dir2_data_hdr	*hdr,
175 	struct xfs_dir2_block_tail	*btp,
176 	struct xfs_dir2_leaf_entry	*blp,
177 	__be16				**tagpp,
178 	struct xfs_dir2_data_unused	**dupp,
179 	struct xfs_dir2_data_unused	**enddupp,
180 	int				*compact,
181 	int				len)
182 {
183 	struct xfs_dir2_data_free	*bf;
184 	__be16				*tagp = NULL;
185 	struct xfs_dir2_data_unused	*dup = NULL;
186 	struct xfs_dir2_data_unused	*enddup = NULL;
187 
188 	*compact = 0;
189 	bf = dp->d_ops->data_bestfree_p(hdr);
190 
191 	/*
192 	 * If there are stale entries we'll use one for the leaf.
193 	 */
194 	if (btp->stale) {
195 		if (be16_to_cpu(bf[0].length) >= len) {
196 			/*
197 			 * The biggest entry enough to avoid compaction.
198 			 */
199 			dup = (xfs_dir2_data_unused_t *)
200 			      ((char *)hdr + be16_to_cpu(bf[0].offset));
201 			goto out;
202 		}
203 
204 		/*
205 		 * Will need to compact to make this work.
206 		 * Tag just before the first leaf entry.
207 		 */
208 		*compact = 1;
209 		tagp = (__be16 *)blp - 1;
210 
211 		/* Data object just before the first leaf entry.  */
212 		dup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
213 
214 		/*
215 		 * If it's not free then the data will go where the
216 		 * leaf data starts now, if it works at all.
217 		 */
218 		if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
219 			if (be16_to_cpu(dup->length) + (be32_to_cpu(btp->stale) - 1) *
220 			    (uint)sizeof(*blp) < len)
221 				dup = NULL;
222 		} else if ((be32_to_cpu(btp->stale) - 1) * (uint)sizeof(*blp) < len)
223 			dup = NULL;
224 		else
225 			dup = (xfs_dir2_data_unused_t *)blp;
226 		goto out;
227 	}
228 
229 	/*
230 	 * no stale entries, so just use free space.
231 	 * Tag just before the first leaf entry.
232 	 */
233 	tagp = (__be16 *)blp - 1;
234 
235 	/* Data object just before the first leaf entry.  */
236 	enddup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
237 
238 	/*
239 	 * If it's not free then can't do this add without cleaning up:
240 	 * the space before the first leaf entry needs to be free so it
241 	 * can be expanded to hold the pointer to the new entry.
242 	 */
243 	if (be16_to_cpu(enddup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
244 		/*
245 		 * Check out the biggest freespace and see if it's the same one.
246 		 */
247 		dup = (xfs_dir2_data_unused_t *)
248 		      ((char *)hdr + be16_to_cpu(bf[0].offset));
249 		if (dup != enddup) {
250 			/*
251 			 * Not the same free entry, just check its length.
252 			 */
253 			if (be16_to_cpu(dup->length) < len)
254 				dup = NULL;
255 			goto out;
256 		}
257 
258 		/*
259 		 * It is the biggest freespace, can it hold the leaf too?
260 		 */
261 		if (be16_to_cpu(dup->length) < len + (uint)sizeof(*blp)) {
262 			/*
263 			 * Yes, use the second-largest entry instead if it works.
264 			 */
265 			if (be16_to_cpu(bf[1].length) >= len)
266 				dup = (xfs_dir2_data_unused_t *)
267 				      ((char *)hdr + be16_to_cpu(bf[1].offset));
268 			else
269 				dup = NULL;
270 		}
271 	}
272 out:
273 	*tagpp = tagp;
274 	*dupp = dup;
275 	*enddupp = enddup;
276 }
277 
278 /*
279  * compact the leaf entries.
280  * Leave the highest-numbered stale entry stale.
281  * XXX should be the one closest to mid but mid is not yet computed.
282  */
283 static void
xfs_dir2_block_compact(struct xfs_da_args * args,struct xfs_buf * bp,struct xfs_dir2_data_hdr * hdr,struct xfs_dir2_block_tail * btp,struct xfs_dir2_leaf_entry * blp,int * needlog,int * lfloghigh,int * lfloglow)284 xfs_dir2_block_compact(
285 	struct xfs_da_args		*args,
286 	struct xfs_buf			*bp,
287 	struct xfs_dir2_data_hdr	*hdr,
288 	struct xfs_dir2_block_tail	*btp,
289 	struct xfs_dir2_leaf_entry	*blp,
290 	int				*needlog,
291 	int				*lfloghigh,
292 	int				*lfloglow)
293 {
294 	int			fromidx;	/* source leaf index */
295 	int			toidx;		/* target leaf index */
296 	int			needscan = 0;
297 	int			highstale;	/* high stale index */
298 
299 	fromidx = toidx = be32_to_cpu(btp->count) - 1;
300 	highstale = *lfloghigh = -1;
301 	for (; fromidx >= 0; fromidx--) {
302 		if (blp[fromidx].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR)) {
303 			if (highstale == -1)
304 				highstale = toidx;
305 			else {
306 				if (*lfloghigh == -1)
307 					*lfloghigh = toidx;
308 				continue;
309 			}
310 		}
311 		if (fromidx < toidx)
312 			blp[toidx] = blp[fromidx];
313 		toidx--;
314 	}
315 	*lfloglow = toidx + 1 - (be32_to_cpu(btp->stale) - 1);
316 	*lfloghigh -= be32_to_cpu(btp->stale) - 1;
317 	be32_add_cpu(&btp->count, -(be32_to_cpu(btp->stale) - 1));
318 	xfs_dir2_data_make_free(args, bp,
319 		(xfs_dir2_data_aoff_t)((char *)blp - (char *)hdr),
320 		(xfs_dir2_data_aoff_t)((be32_to_cpu(btp->stale) - 1) * sizeof(*blp)),
321 		needlog, &needscan);
322 	btp->stale = cpu_to_be32(1);
323 	/*
324 	 * If we now need to rebuild the bestfree map, do so.
325 	 * This needs to happen before the next call to use_free.
326 	 */
327 	if (needscan)
328 		xfs_dir2_data_freescan(args->dp, hdr, needlog);
329 }
330 
331 /*
332  * Add an entry to a block directory.
333  */
334 int						/* error */
xfs_dir2_block_addname(xfs_da_args_t * args)335 xfs_dir2_block_addname(
336 	xfs_da_args_t		*args)		/* directory op arguments */
337 {
338 	xfs_dir2_data_hdr_t	*hdr;		/* block header */
339 	xfs_dir2_leaf_entry_t	*blp;		/* block leaf entries */
340 	struct xfs_buf		*bp;		/* buffer for block */
341 	xfs_dir2_block_tail_t	*btp;		/* block tail */
342 	int			compact;	/* need to compact leaf ents */
343 	xfs_dir2_data_entry_t	*dep;		/* block data entry */
344 	xfs_inode_t		*dp;		/* directory inode */
345 	xfs_dir2_data_unused_t	*dup;		/* block unused entry */
346 	int			error;		/* error return value */
347 	xfs_dir2_data_unused_t	*enddup=NULL;	/* unused at end of data */
348 	xfs_dahash_t		hash;		/* hash value of found entry */
349 	int			high;		/* high index for binary srch */
350 	int			highstale;	/* high stale index */
351 	int			lfloghigh=0;	/* last final leaf to log */
352 	int			lfloglow=0;	/* first final leaf to log */
353 	int			len;		/* length of the new entry */
354 	int			low;		/* low index for binary srch */
355 	int			lowstale;	/* low stale index */
356 	int			mid=0;		/* midpoint for binary srch */
357 	xfs_mount_t		*mp;		/* filesystem mount point */
358 	int			needlog;	/* need to log header */
359 	int			needscan;	/* need to rescan freespace */
360 	__be16			*tagp;		/* pointer to tag value */
361 	xfs_trans_t		*tp;		/* transaction structure */
362 
363 	trace_xfs_dir2_block_addname(args);
364 
365 	dp = args->dp;
366 	tp = args->trans;
367 	mp = dp->i_mount;
368 
369 	/* Read the (one and only) directory block into bp. */
370 	error = xfs_dir3_block_read(tp, dp, &bp);
371 	if (error)
372 		return error;
373 
374 	len = dp->d_ops->data_entsize(args->namelen);
375 
376 	/*
377 	 * Set up pointers to parts of the block.
378 	 */
379 	hdr = bp->b_addr;
380 	btp = xfs_dir2_block_tail_p(args->geo, hdr);
381 	blp = xfs_dir2_block_leaf_p(btp);
382 
383 	/*
384 	 * Find out if we can reuse stale entries or whether we need extra
385 	 * space for entry and new leaf.
386 	 */
387 	xfs_dir2_block_need_space(dp, hdr, btp, blp, &tagp, &dup,
388 				  &enddup, &compact, len);
389 
390 	/*
391 	 * Done everything we need for a space check now.
392 	 */
393 	if (args->op_flags & XFS_DA_OP_JUSTCHECK) {
394 		xfs_trans_brelse(tp, bp);
395 		if (!dup)
396 			return -ENOSPC;
397 		return 0;
398 	}
399 
400 	/*
401 	 * If we don't have space for the new entry & leaf ...
402 	 */
403 	if (!dup) {
404 		/* Don't have a space reservation: return no-space.  */
405 		if (args->total == 0)
406 			return -ENOSPC;
407 		/*
408 		 * Convert to the next larger format.
409 		 * Then add the new entry in that format.
410 		 */
411 		error = xfs_dir2_block_to_leaf(args, bp);
412 		if (error)
413 			return error;
414 		return xfs_dir2_leaf_addname(args);
415 	}
416 
417 	needlog = needscan = 0;
418 
419 	/*
420 	 * If need to compact the leaf entries, do it now.
421 	 */
422 	if (compact) {
423 		xfs_dir2_block_compact(args, bp, hdr, btp, blp, &needlog,
424 				      &lfloghigh, &lfloglow);
425 		/* recalculate blp post-compaction */
426 		blp = xfs_dir2_block_leaf_p(btp);
427 	} else if (btp->stale) {
428 		/*
429 		 * Set leaf logging boundaries to impossible state.
430 		 * For the no-stale case they're set explicitly.
431 		 */
432 		lfloglow = be32_to_cpu(btp->count);
433 		lfloghigh = -1;
434 	}
435 
436 	/*
437 	 * Find the slot that's first lower than our hash value, -1 if none.
438 	 */
439 	for (low = 0, high = be32_to_cpu(btp->count) - 1; low <= high; ) {
440 		mid = (low + high) >> 1;
441 		if ((hash = be32_to_cpu(blp[mid].hashval)) == args->hashval)
442 			break;
443 		if (hash < args->hashval)
444 			low = mid + 1;
445 		else
446 			high = mid - 1;
447 	}
448 	while (mid >= 0 && be32_to_cpu(blp[mid].hashval) >= args->hashval) {
449 		mid--;
450 	}
451 	/*
452 	 * No stale entries, will use enddup space to hold new leaf.
453 	 */
454 	if (!btp->stale) {
455 		/*
456 		 * Mark the space needed for the new leaf entry, now in use.
457 		 */
458 		xfs_dir2_data_use_free(args, bp, enddup,
459 			(xfs_dir2_data_aoff_t)
460 			((char *)enddup - (char *)hdr + be16_to_cpu(enddup->length) -
461 			 sizeof(*blp)),
462 			(xfs_dir2_data_aoff_t)sizeof(*blp),
463 			&needlog, &needscan);
464 		/*
465 		 * Update the tail (entry count).
466 		 */
467 		be32_add_cpu(&btp->count, 1);
468 		/*
469 		 * If we now need to rebuild the bestfree map, do so.
470 		 * This needs to happen before the next call to use_free.
471 		 */
472 		if (needscan) {
473 			xfs_dir2_data_freescan(dp, hdr, &needlog);
474 			needscan = 0;
475 		}
476 		/*
477 		 * Adjust pointer to the first leaf entry, we're about to move
478 		 * the table up one to open up space for the new leaf entry.
479 		 * Then adjust our index to match.
480 		 */
481 		blp--;
482 		mid++;
483 		if (mid)
484 			memmove(blp, &blp[1], mid * sizeof(*blp));
485 		lfloglow = 0;
486 		lfloghigh = mid;
487 	}
488 	/*
489 	 * Use a stale leaf for our new entry.
490 	 */
491 	else {
492 		for (lowstale = mid;
493 		     lowstale >= 0 &&
494 			blp[lowstale].address !=
495 			cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
496 		     lowstale--)
497 			continue;
498 		for (highstale = mid + 1;
499 		     highstale < be32_to_cpu(btp->count) &&
500 			blp[highstale].address !=
501 			cpu_to_be32(XFS_DIR2_NULL_DATAPTR) &&
502 			(lowstale < 0 || mid - lowstale > highstale - mid);
503 		     highstale++)
504 			continue;
505 		/*
506 		 * Move entries toward the low-numbered stale entry.
507 		 */
508 		if (lowstale >= 0 &&
509 		    (highstale == be32_to_cpu(btp->count) ||
510 		     mid - lowstale <= highstale - mid)) {
511 			if (mid - lowstale)
512 				memmove(&blp[lowstale], &blp[lowstale + 1],
513 					(mid - lowstale) * sizeof(*blp));
514 			lfloglow = MIN(lowstale, lfloglow);
515 			lfloghigh = MAX(mid, lfloghigh);
516 		}
517 		/*
518 		 * Move entries toward the high-numbered stale entry.
519 		 */
520 		else {
521 			ASSERT(highstale < be32_to_cpu(btp->count));
522 			mid++;
523 			if (highstale - mid)
524 				memmove(&blp[mid + 1], &blp[mid],
525 					(highstale - mid) * sizeof(*blp));
526 			lfloglow = MIN(mid, lfloglow);
527 			lfloghigh = MAX(highstale, lfloghigh);
528 		}
529 		be32_add_cpu(&btp->stale, -1);
530 	}
531 	/*
532 	 * Point to the new data entry.
533 	 */
534 	dep = (xfs_dir2_data_entry_t *)dup;
535 	/*
536 	 * Fill in the leaf entry.
537 	 */
538 	blp[mid].hashval = cpu_to_be32(args->hashval);
539 	blp[mid].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(
540 				(char *)dep - (char *)hdr));
541 	xfs_dir2_block_log_leaf(tp, bp, lfloglow, lfloghigh);
542 	/*
543 	 * Mark space for the data entry used.
544 	 */
545 	xfs_dir2_data_use_free(args, bp, dup,
546 		(xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr),
547 		(xfs_dir2_data_aoff_t)len, &needlog, &needscan);
548 	/*
549 	 * Create the new data entry.
550 	 */
551 	dep->inumber = cpu_to_be64(args->inumber);
552 	dep->namelen = args->namelen;
553 	memcpy(dep->name, args->name, args->namelen);
554 	dp->d_ops->data_put_ftype(dep, args->filetype);
555 	tagp = dp->d_ops->data_entry_tag_p(dep);
556 	*tagp = cpu_to_be16((char *)dep - (char *)hdr);
557 	/*
558 	 * Clean up the bestfree array and log the header, tail, and entry.
559 	 */
560 	if (needscan)
561 		xfs_dir2_data_freescan(dp, hdr, &needlog);
562 	if (needlog)
563 		xfs_dir2_data_log_header(args, bp);
564 	xfs_dir2_block_log_tail(tp, bp);
565 	xfs_dir2_data_log_entry(args, bp, dep);
566 	xfs_dir3_data_check(dp, bp);
567 	return 0;
568 }
569 
570 /*
571  * Log leaf entries from the block.
572  */
573 static void
xfs_dir2_block_log_leaf(xfs_trans_t * tp,struct xfs_buf * bp,int first,int last)574 xfs_dir2_block_log_leaf(
575 	xfs_trans_t		*tp,		/* transaction structure */
576 	struct xfs_buf		*bp,		/* block buffer */
577 	int			first,		/* index of first logged leaf */
578 	int			last)		/* index of last logged leaf */
579 {
580 	xfs_dir2_data_hdr_t	*hdr = bp->b_addr;
581 	xfs_dir2_leaf_entry_t	*blp;
582 	xfs_dir2_block_tail_t	*btp;
583 
584 	btp = xfs_dir2_block_tail_p(tp->t_mountp->m_dir_geo, hdr);
585 	blp = xfs_dir2_block_leaf_p(btp);
586 	xfs_trans_log_buf(tp, bp, (uint)((char *)&blp[first] - (char *)hdr),
587 		(uint)((char *)&blp[last + 1] - (char *)hdr - 1));
588 }
589 
590 /*
591  * Log the block tail.
592  */
593 static void
xfs_dir2_block_log_tail(xfs_trans_t * tp,struct xfs_buf * bp)594 xfs_dir2_block_log_tail(
595 	xfs_trans_t		*tp,		/* transaction structure */
596 	struct xfs_buf		*bp)		/* block buffer */
597 {
598 	xfs_dir2_data_hdr_t	*hdr = bp->b_addr;
599 	xfs_dir2_block_tail_t	*btp;
600 
601 	btp = xfs_dir2_block_tail_p(tp->t_mountp->m_dir_geo, hdr);
602 	xfs_trans_log_buf(tp, bp, (uint)((char *)btp - (char *)hdr),
603 		(uint)((char *)(btp + 1) - (char *)hdr - 1));
604 }
605 
606 /*
607  * Look up an entry in the block.  This is the external routine,
608  * xfs_dir2_block_lookup_int does the real work.
609  */
610 int						/* error */
xfs_dir2_block_lookup(xfs_da_args_t * args)611 xfs_dir2_block_lookup(
612 	xfs_da_args_t		*args)		/* dir lookup arguments */
613 {
614 	xfs_dir2_data_hdr_t	*hdr;		/* block header */
615 	xfs_dir2_leaf_entry_t	*blp;		/* block leaf entries */
616 	struct xfs_buf		*bp;		/* block buffer */
617 	xfs_dir2_block_tail_t	*btp;		/* block tail */
618 	xfs_dir2_data_entry_t	*dep;		/* block data entry */
619 	xfs_inode_t		*dp;		/* incore inode */
620 	int			ent;		/* entry index */
621 	int			error;		/* error return value */
622 	xfs_mount_t		*mp;		/* filesystem mount point */
623 
624 	trace_xfs_dir2_block_lookup(args);
625 
626 	/*
627 	 * Get the buffer, look up the entry.
628 	 * If not found (ENOENT) then return, have no buffer.
629 	 */
630 	if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent)))
631 		return error;
632 	dp = args->dp;
633 	mp = dp->i_mount;
634 	hdr = bp->b_addr;
635 	xfs_dir3_data_check(dp, bp);
636 	btp = xfs_dir2_block_tail_p(args->geo, hdr);
637 	blp = xfs_dir2_block_leaf_p(btp);
638 	/*
639 	 * Get the offset from the leaf entry, to point to the data.
640 	 */
641 	dep = (xfs_dir2_data_entry_t *)((char *)hdr +
642 			xfs_dir2_dataptr_to_off(args->geo,
643 						be32_to_cpu(blp[ent].address)));
644 	/*
645 	 * Fill in inode number, CI name if appropriate, release the block.
646 	 */
647 	args->inumber = be64_to_cpu(dep->inumber);
648 	args->filetype = dp->d_ops->data_get_ftype(dep);
649 	error = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
650 	xfs_trans_brelse(args->trans, bp);
651 	return error;
652 }
653 
654 /*
655  * Internal block lookup routine.
656  */
657 static int					/* error */
xfs_dir2_block_lookup_int(xfs_da_args_t * args,struct xfs_buf ** bpp,int * entno)658 xfs_dir2_block_lookup_int(
659 	xfs_da_args_t		*args,		/* dir lookup arguments */
660 	struct xfs_buf		**bpp,		/* returned block buffer */
661 	int			*entno)		/* returned entry number */
662 {
663 	xfs_dir2_dataptr_t	addr;		/* data entry address */
664 	xfs_dir2_data_hdr_t	*hdr;		/* block header */
665 	xfs_dir2_leaf_entry_t	*blp;		/* block leaf entries */
666 	struct xfs_buf		*bp;		/* block buffer */
667 	xfs_dir2_block_tail_t	*btp;		/* block tail */
668 	xfs_dir2_data_entry_t	*dep;		/* block data entry */
669 	xfs_inode_t		*dp;		/* incore inode */
670 	int			error;		/* error return value */
671 	xfs_dahash_t		hash;		/* found hash value */
672 	int			high;		/* binary search high index */
673 	int			low;		/* binary search low index */
674 	int			mid;		/* binary search current idx */
675 	xfs_mount_t		*mp;		/* filesystem mount point */
676 	xfs_trans_t		*tp;		/* transaction pointer */
677 	enum xfs_dacmp		cmp;		/* comparison result */
678 
679 	dp = args->dp;
680 	tp = args->trans;
681 	mp = dp->i_mount;
682 
683 	error = xfs_dir3_block_read(tp, dp, &bp);
684 	if (error)
685 		return error;
686 
687 	hdr = bp->b_addr;
688 	xfs_dir3_data_check(dp, bp);
689 	btp = xfs_dir2_block_tail_p(args->geo, hdr);
690 	blp = xfs_dir2_block_leaf_p(btp);
691 	/*
692 	 * Loop doing a binary search for our hash value.
693 	 * Find our entry, ENOENT if it's not there.
694 	 */
695 	for (low = 0, high = be32_to_cpu(btp->count) - 1; ; ) {
696 		ASSERT(low <= high);
697 		mid = (low + high) >> 1;
698 		if ((hash = be32_to_cpu(blp[mid].hashval)) == args->hashval)
699 			break;
700 		if (hash < args->hashval)
701 			low = mid + 1;
702 		else
703 			high = mid - 1;
704 		if (low > high) {
705 			ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
706 			xfs_trans_brelse(tp, bp);
707 			return -ENOENT;
708 		}
709 	}
710 	/*
711 	 * Back up to the first one with the right hash value.
712 	 */
713 	while (mid > 0 && be32_to_cpu(blp[mid - 1].hashval) == args->hashval) {
714 		mid--;
715 	}
716 	/*
717 	 * Now loop forward through all the entries with the
718 	 * right hash value looking for our name.
719 	 */
720 	do {
721 		if ((addr = be32_to_cpu(blp[mid].address)) == XFS_DIR2_NULL_DATAPTR)
722 			continue;
723 		/*
724 		 * Get pointer to the entry from the leaf.
725 		 */
726 		dep = (xfs_dir2_data_entry_t *)
727 			((char *)hdr + xfs_dir2_dataptr_to_off(args->geo, addr));
728 		/*
729 		 * Compare name and if it's an exact match, return the index
730 		 * and buffer. If it's the first case-insensitive match, store
731 		 * the index and buffer and continue looking for an exact match.
732 		 */
733 		cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
734 		if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
735 			args->cmpresult = cmp;
736 			*bpp = bp;
737 			*entno = mid;
738 			if (cmp == XFS_CMP_EXACT)
739 				return 0;
740 		}
741 	} while (++mid < be32_to_cpu(btp->count) &&
742 			be32_to_cpu(blp[mid].hashval) == hash);
743 
744 	ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
745 	/*
746 	 * Here, we can only be doing a lookup (not a rename or replace).
747 	 * If a case-insensitive match was found earlier, return success.
748 	 */
749 	if (args->cmpresult == XFS_CMP_CASE)
750 		return 0;
751 	/*
752 	 * No match, release the buffer and return ENOENT.
753 	 */
754 	xfs_trans_brelse(tp, bp);
755 	return -ENOENT;
756 }
757 
758 /*
759  * Remove an entry from a block format directory.
760  * If that makes the block small enough to fit in shortform, transform it.
761  */
762 int						/* error */
xfs_dir2_block_removename(xfs_da_args_t * args)763 xfs_dir2_block_removename(
764 	xfs_da_args_t		*args)		/* directory operation args */
765 {
766 	xfs_dir2_data_hdr_t	*hdr;		/* block header */
767 	xfs_dir2_leaf_entry_t	*blp;		/* block leaf pointer */
768 	struct xfs_buf		*bp;		/* block buffer */
769 	xfs_dir2_block_tail_t	*btp;		/* block tail */
770 	xfs_dir2_data_entry_t	*dep;		/* block data entry */
771 	xfs_inode_t		*dp;		/* incore inode */
772 	int			ent;		/* block leaf entry index */
773 	int			error;		/* error return value */
774 	xfs_mount_t		*mp;		/* filesystem mount point */
775 	int			needlog;	/* need to log block header */
776 	int			needscan;	/* need to fixup bestfree */
777 	xfs_dir2_sf_hdr_t	sfh;		/* shortform header */
778 	int			size;		/* shortform size */
779 	xfs_trans_t		*tp;		/* transaction pointer */
780 
781 	trace_xfs_dir2_block_removename(args);
782 
783 	/*
784 	 * Look up the entry in the block.  Gets the buffer and entry index.
785 	 * It will always be there, the vnodeops level does a lookup first.
786 	 */
787 	if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent))) {
788 		return error;
789 	}
790 	dp = args->dp;
791 	tp = args->trans;
792 	mp = dp->i_mount;
793 	hdr = bp->b_addr;
794 	btp = xfs_dir2_block_tail_p(args->geo, hdr);
795 	blp = xfs_dir2_block_leaf_p(btp);
796 	/*
797 	 * Point to the data entry using the leaf entry.
798 	 */
799 	dep = (xfs_dir2_data_entry_t *)((char *)hdr +
800 			xfs_dir2_dataptr_to_off(args->geo,
801 						be32_to_cpu(blp[ent].address)));
802 	/*
803 	 * Mark the data entry's space free.
804 	 */
805 	needlog = needscan = 0;
806 	xfs_dir2_data_make_free(args, bp,
807 		(xfs_dir2_data_aoff_t)((char *)dep - (char *)hdr),
808 		dp->d_ops->data_entsize(dep->namelen), &needlog, &needscan);
809 	/*
810 	 * Fix up the block tail.
811 	 */
812 	be32_add_cpu(&btp->stale, 1);
813 	xfs_dir2_block_log_tail(tp, bp);
814 	/*
815 	 * Remove the leaf entry by marking it stale.
816 	 */
817 	blp[ent].address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
818 	xfs_dir2_block_log_leaf(tp, bp, ent, ent);
819 	/*
820 	 * Fix up bestfree, log the header if necessary.
821 	 */
822 	if (needscan)
823 		xfs_dir2_data_freescan(dp, hdr, &needlog);
824 	if (needlog)
825 		xfs_dir2_data_log_header(args, bp);
826 	xfs_dir3_data_check(dp, bp);
827 	/*
828 	 * See if the size as a shortform is good enough.
829 	 */
830 	size = xfs_dir2_block_sfsize(dp, hdr, &sfh);
831 	if (size > XFS_IFORK_DSIZE(dp))
832 		return 0;
833 
834 	/*
835 	 * If it works, do the conversion.
836 	 */
837 	return xfs_dir2_block_to_sf(args, bp, size, &sfh);
838 }
839 
840 /*
841  * Replace an entry in a V2 block directory.
842  * Change the inode number to the new value.
843  */
844 int						/* error */
xfs_dir2_block_replace(xfs_da_args_t * args)845 xfs_dir2_block_replace(
846 	xfs_da_args_t		*args)		/* directory operation args */
847 {
848 	xfs_dir2_data_hdr_t	*hdr;		/* block header */
849 	xfs_dir2_leaf_entry_t	*blp;		/* block leaf entries */
850 	struct xfs_buf		*bp;		/* block buffer */
851 	xfs_dir2_block_tail_t	*btp;		/* block tail */
852 	xfs_dir2_data_entry_t	*dep;		/* block data entry */
853 	xfs_inode_t		*dp;		/* incore inode */
854 	int			ent;		/* leaf entry index */
855 	int			error;		/* error return value */
856 	xfs_mount_t		*mp;		/* filesystem mount point */
857 
858 	trace_xfs_dir2_block_replace(args);
859 
860 	/*
861 	 * Lookup the entry in the directory.  Get buffer and entry index.
862 	 * This will always succeed since the caller has already done a lookup.
863 	 */
864 	if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent))) {
865 		return error;
866 	}
867 	dp = args->dp;
868 	mp = dp->i_mount;
869 	hdr = bp->b_addr;
870 	btp = xfs_dir2_block_tail_p(args->geo, hdr);
871 	blp = xfs_dir2_block_leaf_p(btp);
872 	/*
873 	 * Point to the data entry we need to change.
874 	 */
875 	dep = (xfs_dir2_data_entry_t *)((char *)hdr +
876 			xfs_dir2_dataptr_to_off(args->geo,
877 						be32_to_cpu(blp[ent].address)));
878 	ASSERT(be64_to_cpu(dep->inumber) != args->inumber);
879 	/*
880 	 * Change the inode number to the new value.
881 	 */
882 	dep->inumber = cpu_to_be64(args->inumber);
883 	dp->d_ops->data_put_ftype(dep, args->filetype);
884 	xfs_dir2_data_log_entry(args, bp, dep);
885 	xfs_dir3_data_check(dp, bp);
886 	return 0;
887 }
888 
889 /*
890  * Qsort comparison routine for the block leaf entries.
891  */
892 static int					/* sort order */
xfs_dir2_block_sort(const void * a,const void * b)893 xfs_dir2_block_sort(
894 	const void			*a,	/* first leaf entry */
895 	const void			*b)	/* second leaf entry */
896 {
897 	const xfs_dir2_leaf_entry_t	*la;	/* first leaf entry */
898 	const xfs_dir2_leaf_entry_t	*lb;	/* second leaf entry */
899 
900 	la = a;
901 	lb = b;
902 	return be32_to_cpu(la->hashval) < be32_to_cpu(lb->hashval) ? -1 :
903 		(be32_to_cpu(la->hashval) > be32_to_cpu(lb->hashval) ? 1 : 0);
904 }
905 
906 /*
907  * Convert a V2 leaf directory to a V2 block directory if possible.
908  */
909 int						/* error */
xfs_dir2_leaf_to_block(xfs_da_args_t * args,struct xfs_buf * lbp,struct xfs_buf * dbp)910 xfs_dir2_leaf_to_block(
911 	xfs_da_args_t		*args,		/* operation arguments */
912 	struct xfs_buf		*lbp,		/* leaf buffer */
913 	struct xfs_buf		*dbp)		/* data buffer */
914 {
915 	__be16			*bestsp;	/* leaf bests table */
916 	xfs_dir2_data_hdr_t	*hdr;		/* block header */
917 	xfs_dir2_block_tail_t	*btp;		/* block tail */
918 	xfs_inode_t		*dp;		/* incore directory inode */
919 	xfs_dir2_data_unused_t	*dup;		/* unused data entry */
920 	int			error;		/* error return value */
921 	int			from;		/* leaf from index */
922 	xfs_dir2_leaf_t		*leaf;		/* leaf structure */
923 	xfs_dir2_leaf_entry_t	*lep;		/* leaf entry */
924 	xfs_dir2_leaf_tail_t	*ltp;		/* leaf tail structure */
925 	xfs_mount_t		*mp;		/* file system mount point */
926 	int			needlog;	/* need to log data header */
927 	int			needscan;	/* need to scan for bestfree */
928 	xfs_dir2_sf_hdr_t	sfh;		/* shortform header */
929 	int			size;		/* bytes used */
930 	__be16			*tagp;		/* end of entry (tag) */
931 	int			to;		/* block/leaf to index */
932 	xfs_trans_t		*tp;		/* transaction pointer */
933 	struct xfs_dir2_leaf_entry *ents;
934 	struct xfs_dir3_icleaf_hdr leafhdr;
935 
936 	trace_xfs_dir2_leaf_to_block(args);
937 
938 	dp = args->dp;
939 	tp = args->trans;
940 	mp = dp->i_mount;
941 	leaf = lbp->b_addr;
942 	dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
943 	ents = dp->d_ops->leaf_ents_p(leaf);
944 	ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
945 
946 	ASSERT(leafhdr.magic == XFS_DIR2_LEAF1_MAGIC ||
947 	       leafhdr.magic == XFS_DIR3_LEAF1_MAGIC);
948 	/*
949 	 * If there are data blocks other than the first one, take this
950 	 * opportunity to remove trailing empty data blocks that may have
951 	 * been left behind during no-space-reservation operations.
952 	 * These will show up in the leaf bests table.
953 	 */
954 	while (dp->i_d.di_size > args->geo->blksize) {
955 		int hdrsz;
956 
957 		hdrsz = dp->d_ops->data_entry_offset;
958 		bestsp = xfs_dir2_leaf_bests_p(ltp);
959 		if (be16_to_cpu(bestsp[be32_to_cpu(ltp->bestcount) - 1]) ==
960 					    args->geo->blksize - hdrsz) {
961 			if ((error =
962 			    xfs_dir2_leaf_trim_data(args, lbp,
963 				    (xfs_dir2_db_t)(be32_to_cpu(ltp->bestcount) - 1))))
964 				return error;
965 		} else
966 			return 0;
967 	}
968 	/*
969 	 * Read the data block if we don't already have it, give up if it fails.
970 	 */
971 	if (!dbp) {
972 		error = xfs_dir3_data_read(tp, dp, args->geo->datablk, -1, &dbp);
973 		if (error)
974 			return error;
975 	}
976 	hdr = dbp->b_addr;
977 	ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
978 	       hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC));
979 
980 	/*
981 	 * Size of the "leaf" area in the block.
982 	 */
983 	size = (uint)sizeof(xfs_dir2_block_tail_t) +
984 	       (uint)sizeof(*lep) * (leafhdr.count - leafhdr.stale);
985 	/*
986 	 * Look at the last data entry.
987 	 */
988 	tagp = (__be16 *)((char *)hdr + args->geo->blksize) - 1;
989 	dup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
990 	/*
991 	 * If it's not free or is too short we can't do it.
992 	 */
993 	if (be16_to_cpu(dup->freetag) != XFS_DIR2_DATA_FREE_TAG ||
994 	    be16_to_cpu(dup->length) < size)
995 		return 0;
996 
997 	/*
998 	 * Start converting it to block form.
999 	 */
1000 	xfs_dir3_block_init(mp, tp, dbp, dp);
1001 
1002 	needlog = 1;
1003 	needscan = 0;
1004 	/*
1005 	 * Use up the space at the end of the block (blp/btp).
1006 	 */
1007 	xfs_dir2_data_use_free(args, dbp, dup, args->geo->blksize - size, size,
1008 		&needlog, &needscan);
1009 	/*
1010 	 * Initialize the block tail.
1011 	 */
1012 	btp = xfs_dir2_block_tail_p(args->geo, hdr);
1013 	btp->count = cpu_to_be32(leafhdr.count - leafhdr.stale);
1014 	btp->stale = 0;
1015 	xfs_dir2_block_log_tail(tp, dbp);
1016 	/*
1017 	 * Initialize the block leaf area.  We compact out stale entries.
1018 	 */
1019 	lep = xfs_dir2_block_leaf_p(btp);
1020 	for (from = to = 0; from < leafhdr.count; from++) {
1021 		if (ents[from].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
1022 			continue;
1023 		lep[to++] = ents[from];
1024 	}
1025 	ASSERT(to == be32_to_cpu(btp->count));
1026 	xfs_dir2_block_log_leaf(tp, dbp, 0, be32_to_cpu(btp->count) - 1);
1027 	/*
1028 	 * Scan the bestfree if we need it and log the data block header.
1029 	 */
1030 	if (needscan)
1031 		xfs_dir2_data_freescan(dp, hdr, &needlog);
1032 	if (needlog)
1033 		xfs_dir2_data_log_header(args, dbp);
1034 	/*
1035 	 * Pitch the old leaf block.
1036 	 */
1037 	error = xfs_da_shrink_inode(args, args->geo->leafblk, lbp);
1038 	if (error)
1039 		return error;
1040 
1041 	/*
1042 	 * Now see if the resulting block can be shrunken to shortform.
1043 	 */
1044 	size = xfs_dir2_block_sfsize(dp, hdr, &sfh);
1045 	if (size > XFS_IFORK_DSIZE(dp))
1046 		return 0;
1047 
1048 	return xfs_dir2_block_to_sf(args, dbp, size, &sfh);
1049 }
1050 
1051 /*
1052  * Convert the shortform directory to block form.
1053  */
1054 int						/* error */
xfs_dir2_sf_to_block(xfs_da_args_t * args)1055 xfs_dir2_sf_to_block(
1056 	xfs_da_args_t		*args)		/* operation arguments */
1057 {
1058 	xfs_dir2_db_t		blkno;		/* dir-relative block # (0) */
1059 	xfs_dir2_data_hdr_t	*hdr;		/* block header */
1060 	xfs_dir2_leaf_entry_t	*blp;		/* block leaf entries */
1061 	struct xfs_buf		*bp;		/* block buffer */
1062 	xfs_dir2_block_tail_t	*btp;		/* block tail pointer */
1063 	xfs_dir2_data_entry_t	*dep;		/* data entry pointer */
1064 	xfs_inode_t		*dp;		/* incore directory inode */
1065 	int			dummy;		/* trash */
1066 	xfs_dir2_data_unused_t	*dup;		/* unused entry pointer */
1067 	int			endoffset;	/* end of data objects */
1068 	int			error;		/* error return value */
1069 	int			i;		/* index */
1070 	xfs_mount_t		*mp;		/* filesystem mount point */
1071 	int			needlog;	/* need to log block header */
1072 	int			needscan;	/* need to scan block freespc */
1073 	int			newoffset;	/* offset from current entry */
1074 	int			offset;		/* target block offset */
1075 	xfs_dir2_sf_entry_t	*sfep;		/* sf entry pointer */
1076 	xfs_dir2_sf_hdr_t	*oldsfp;	/* old shortform header  */
1077 	xfs_dir2_sf_hdr_t	*sfp;		/* shortform header  */
1078 	__be16			*tagp;		/* end of data entry */
1079 	xfs_trans_t		*tp;		/* transaction pointer */
1080 	struct xfs_name		name;
1081 	struct xfs_ifork	*ifp;
1082 
1083 	trace_xfs_dir2_sf_to_block(args);
1084 
1085 	dp = args->dp;
1086 	tp = args->trans;
1087 	mp = dp->i_mount;
1088 	ifp = XFS_IFORK_PTR(dp, XFS_DATA_FORK);
1089 	ASSERT(ifp->if_flags & XFS_IFINLINE);
1090 	/*
1091 	 * Bomb out if the shortform directory is way too short.
1092 	 */
1093 	if (dp->i_d.di_size < offsetof(xfs_dir2_sf_hdr_t, parent)) {
1094 		ASSERT(XFS_FORCED_SHUTDOWN(mp));
1095 		return -EIO;
1096 	}
1097 
1098 	oldsfp = (xfs_dir2_sf_hdr_t *)ifp->if_u1.if_data;
1099 
1100 	ASSERT(ifp->if_bytes == dp->i_d.di_size);
1101 	ASSERT(ifp->if_u1.if_data != NULL);
1102 	ASSERT(dp->i_d.di_size >= xfs_dir2_sf_hdr_size(oldsfp->i8count));
1103 	ASSERT(dp->i_d.di_nextents == 0);
1104 
1105 	/*
1106 	 * Copy the directory into a temporary buffer.
1107 	 * Then pitch the incore inode data so we can make extents.
1108 	 */
1109 	sfp = kmem_alloc(ifp->if_bytes, KM_SLEEP);
1110 	memcpy(sfp, oldsfp, ifp->if_bytes);
1111 
1112 	xfs_idata_realloc(dp, -ifp->if_bytes, XFS_DATA_FORK);
1113 	xfs_bmap_local_to_extents_empty(dp, XFS_DATA_FORK);
1114 	dp->i_d.di_size = 0;
1115 
1116 	/*
1117 	 * Add block 0 to the inode.
1118 	 */
1119 	error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE, &blkno);
1120 	if (error) {
1121 		kmem_free(sfp);
1122 		return error;
1123 	}
1124 	/*
1125 	 * Initialize the data block, then convert it to block format.
1126 	 */
1127 	error = xfs_dir3_data_init(args, blkno, &bp);
1128 	if (error) {
1129 		kmem_free(sfp);
1130 		return error;
1131 	}
1132 	xfs_dir3_block_init(mp, tp, bp, dp);
1133 	hdr = bp->b_addr;
1134 
1135 	/*
1136 	 * Compute size of block "tail" area.
1137 	 */
1138 	i = (uint)sizeof(*btp) +
1139 	    (sfp->count + 2) * (uint)sizeof(xfs_dir2_leaf_entry_t);
1140 	/*
1141 	 * The whole thing is initialized to free by the init routine.
1142 	 * Say we're using the leaf and tail area.
1143 	 */
1144 	dup = dp->d_ops->data_unused_p(hdr);
1145 	needlog = needscan = 0;
1146 	xfs_dir2_data_use_free(args, bp, dup, args->geo->blksize - i,
1147 			       i, &needlog, &needscan);
1148 	ASSERT(needscan == 0);
1149 	/*
1150 	 * Fill in the tail.
1151 	 */
1152 	btp = xfs_dir2_block_tail_p(args->geo, hdr);
1153 	btp->count = cpu_to_be32(sfp->count + 2);	/* ., .. */
1154 	btp->stale = 0;
1155 	blp = xfs_dir2_block_leaf_p(btp);
1156 	endoffset = (uint)((char *)blp - (char *)hdr);
1157 	/*
1158 	 * Remove the freespace, we'll manage it.
1159 	 */
1160 	xfs_dir2_data_use_free(args, bp, dup,
1161 		(xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr),
1162 		be16_to_cpu(dup->length), &needlog, &needscan);
1163 	/*
1164 	 * Create entry for .
1165 	 */
1166 	dep = dp->d_ops->data_dot_entry_p(hdr);
1167 	dep->inumber = cpu_to_be64(dp->i_ino);
1168 	dep->namelen = 1;
1169 	dep->name[0] = '.';
1170 	dp->d_ops->data_put_ftype(dep, XFS_DIR3_FT_DIR);
1171 	tagp = dp->d_ops->data_entry_tag_p(dep);
1172 	*tagp = cpu_to_be16((char *)dep - (char *)hdr);
1173 	xfs_dir2_data_log_entry(args, bp, dep);
1174 	blp[0].hashval = cpu_to_be32(xfs_dir_hash_dot);
1175 	blp[0].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(
1176 				(char *)dep - (char *)hdr));
1177 	/*
1178 	 * Create entry for ..
1179 	 */
1180 	dep = dp->d_ops->data_dotdot_entry_p(hdr);
1181 	dep->inumber = cpu_to_be64(dp->d_ops->sf_get_parent_ino(sfp));
1182 	dep->namelen = 2;
1183 	dep->name[0] = dep->name[1] = '.';
1184 	dp->d_ops->data_put_ftype(dep, XFS_DIR3_FT_DIR);
1185 	tagp = dp->d_ops->data_entry_tag_p(dep);
1186 	*tagp = cpu_to_be16((char *)dep - (char *)hdr);
1187 	xfs_dir2_data_log_entry(args, bp, dep);
1188 	blp[1].hashval = cpu_to_be32(xfs_dir_hash_dotdot);
1189 	blp[1].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(
1190 				(char *)dep - (char *)hdr));
1191 	offset = dp->d_ops->data_first_offset;
1192 	/*
1193 	 * Loop over existing entries, stuff them in.
1194 	 */
1195 	i = 0;
1196 	if (!sfp->count)
1197 		sfep = NULL;
1198 	else
1199 		sfep = xfs_dir2_sf_firstentry(sfp);
1200 	/*
1201 	 * Need to preserve the existing offset values in the sf directory.
1202 	 * Insert holes (unused entries) where necessary.
1203 	 */
1204 	while (offset < endoffset) {
1205 		/*
1206 		 * sfep is null when we reach the end of the list.
1207 		 */
1208 		if (sfep == NULL)
1209 			newoffset = endoffset;
1210 		else
1211 			newoffset = xfs_dir2_sf_get_offset(sfep);
1212 		/*
1213 		 * There should be a hole here, make one.
1214 		 */
1215 		if (offset < newoffset) {
1216 			dup = (xfs_dir2_data_unused_t *)((char *)hdr + offset);
1217 			dup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
1218 			dup->length = cpu_to_be16(newoffset - offset);
1219 			*xfs_dir2_data_unused_tag_p(dup) = cpu_to_be16(
1220 				((char *)dup - (char *)hdr));
1221 			xfs_dir2_data_log_unused(args, bp, dup);
1222 			xfs_dir2_data_freeinsert(hdr,
1223 						 dp->d_ops->data_bestfree_p(hdr),
1224 						 dup, &dummy);
1225 			offset += be16_to_cpu(dup->length);
1226 			continue;
1227 		}
1228 		/*
1229 		 * Copy a real entry.
1230 		 */
1231 		dep = (xfs_dir2_data_entry_t *)((char *)hdr + newoffset);
1232 		dep->inumber = cpu_to_be64(dp->d_ops->sf_get_ino(sfp, sfep));
1233 		dep->namelen = sfep->namelen;
1234 		dp->d_ops->data_put_ftype(dep, dp->d_ops->sf_get_ftype(sfep));
1235 		memcpy(dep->name, sfep->name, dep->namelen);
1236 		tagp = dp->d_ops->data_entry_tag_p(dep);
1237 		*tagp = cpu_to_be16((char *)dep - (char *)hdr);
1238 		xfs_dir2_data_log_entry(args, bp, dep);
1239 		name.name = sfep->name;
1240 		name.len = sfep->namelen;
1241 		blp[2 + i].hashval = cpu_to_be32(mp->m_dirnameops->
1242 							hashname(&name));
1243 		blp[2 + i].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(
1244 						 (char *)dep - (char *)hdr));
1245 		offset = (int)((char *)(tagp + 1) - (char *)hdr);
1246 		if (++i == sfp->count)
1247 			sfep = NULL;
1248 		else
1249 			sfep = dp->d_ops->sf_nextentry(sfp, sfep);
1250 	}
1251 	/* Done with the temporary buffer */
1252 	kmem_free(sfp);
1253 	/*
1254 	 * Sort the leaf entries by hash value.
1255 	 */
1256 	xfs_sort(blp, be32_to_cpu(btp->count), sizeof(*blp), xfs_dir2_block_sort);
1257 	/*
1258 	 * Log the leaf entry area and tail.
1259 	 * Already logged the header in data_init, ignore needlog.
1260 	 */
1261 	ASSERT(needscan == 0);
1262 	xfs_dir2_block_log_leaf(tp, bp, 0, be32_to_cpu(btp->count) - 1);
1263 	xfs_dir2_block_log_tail(tp, bp);
1264 	xfs_dir3_data_check(dp, bp);
1265 	return 0;
1266 }
1267