xref: /openbmc/linux/fs/xfs/libxfs/xfs_dir2_data.c (revision d3964221)
1 /*
2  * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
3  * Copyright (c) 2013 Red Hat, Inc.
4  * All Rights Reserved.
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public License as
8  * published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it would be useful,
11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13  * GNU General Public License for more details.
14  *
15  * You should have received a copy of the GNU General Public License
16  * along with this program; if not, write the Free Software Foundation,
17  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
18  */
19 #include "xfs.h"
20 #include "xfs_fs.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
24 #include "xfs_mount.h"
25 #include "xfs_da_format.h"
26 #include "xfs_da_btree.h"
27 #include "xfs_inode.h"
28 #include "xfs_dir2.h"
29 #include "xfs_dir2_priv.h"
30 #include "xfs_error.h"
31 #include "xfs_trans.h"
32 #include "xfs_buf_item.h"
33 #include "xfs_cksum.h"
34 #include "xfs_log.h"
35 
36 /*
37  * Check the consistency of the data block.
38  * The input can also be a block-format directory.
39  * Return 0 is the buffer is good, otherwise an error.
40  */
41 int
42 __xfs_dir3_data_check(
43 	struct xfs_inode	*dp,		/* incore inode pointer */
44 	struct xfs_buf		*bp)		/* data block's buffer */
45 {
46 	xfs_dir2_dataptr_t	addr;		/* addr for leaf lookup */
47 	xfs_dir2_data_free_t	*bf;		/* bestfree table */
48 	xfs_dir2_block_tail_t	*btp=NULL;	/* block tail */
49 	int			count;		/* count of entries found */
50 	xfs_dir2_data_hdr_t	*hdr;		/* data block header */
51 	xfs_dir2_data_entry_t	*dep;		/* data entry */
52 	xfs_dir2_data_free_t	*dfp;		/* bestfree entry */
53 	xfs_dir2_data_unused_t	*dup;		/* unused entry */
54 	char			*endp;		/* end of useful data */
55 	int			freeseen;	/* mask of bestfrees seen */
56 	xfs_dahash_t		hash;		/* hash of current name */
57 	int			i;		/* leaf index */
58 	int			lastfree;	/* last entry was unused */
59 	xfs_dir2_leaf_entry_t	*lep=NULL;	/* block leaf entries */
60 	xfs_mount_t		*mp;		/* filesystem mount point */
61 	char			*p;		/* current data position */
62 	int			stale;		/* count of stale leaves */
63 	struct xfs_name		name;
64 	const struct xfs_dir_ops *ops;
65 	struct xfs_da_geometry	*geo;
66 
67 	mp = bp->b_target->bt_mount;
68 	geo = mp->m_dir_geo;
69 
70 	/*
71 	 * We can be passed a null dp here from a verifier, so we need to go the
72 	 * hard way to get them.
73 	 */
74 	ops = xfs_dir_get_ops(mp, dp);
75 
76 	hdr = bp->b_addr;
77 	p = (char *)ops->data_entry_p(hdr);
78 
79 	switch (hdr->magic) {
80 	case cpu_to_be32(XFS_DIR3_BLOCK_MAGIC):
81 	case cpu_to_be32(XFS_DIR2_BLOCK_MAGIC):
82 		btp = xfs_dir2_block_tail_p(geo, hdr);
83 		lep = xfs_dir2_block_leaf_p(btp);
84 		endp = (char *)lep;
85 
86 		/*
87 		 * The number of leaf entries is limited by the size of the
88 		 * block and the amount of space used by the data entries.
89 		 * We don't know how much space is used by the data entries yet,
90 		 * so just ensure that the count falls somewhere inside the
91 		 * block right now.
92 		 */
93 		XFS_WANT_CORRUPTED_RETURN(mp, be32_to_cpu(btp->count) <
94 			((char *)btp - p) / sizeof(struct xfs_dir2_leaf_entry));
95 		break;
96 	case cpu_to_be32(XFS_DIR3_DATA_MAGIC):
97 	case cpu_to_be32(XFS_DIR2_DATA_MAGIC):
98 		endp = (char *)hdr + geo->blksize;
99 		break;
100 	default:
101 		XFS_ERROR_REPORT("Bad Magic", XFS_ERRLEVEL_LOW, mp);
102 		return -EFSCORRUPTED;
103 	}
104 
105 	/*
106 	 * Account for zero bestfree entries.
107 	 */
108 	bf = ops->data_bestfree_p(hdr);
109 	count = lastfree = freeseen = 0;
110 	if (!bf[0].length) {
111 		XFS_WANT_CORRUPTED_RETURN(mp, !bf[0].offset);
112 		freeseen |= 1 << 0;
113 	}
114 	if (!bf[1].length) {
115 		XFS_WANT_CORRUPTED_RETURN(mp, !bf[1].offset);
116 		freeseen |= 1 << 1;
117 	}
118 	if (!bf[2].length) {
119 		XFS_WANT_CORRUPTED_RETURN(mp, !bf[2].offset);
120 		freeseen |= 1 << 2;
121 	}
122 
123 	XFS_WANT_CORRUPTED_RETURN(mp, be16_to_cpu(bf[0].length) >=
124 						be16_to_cpu(bf[1].length));
125 	XFS_WANT_CORRUPTED_RETURN(mp, be16_to_cpu(bf[1].length) >=
126 						be16_to_cpu(bf[2].length));
127 	/*
128 	 * Loop over the data/unused entries.
129 	 */
130 	while (p < endp) {
131 		dup = (xfs_dir2_data_unused_t *)p;
132 		/*
133 		 * If it's unused, look for the space in the bestfree table.
134 		 * If we find it, account for that, else make sure it
135 		 * doesn't need to be there.
136 		 */
137 		if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
138 			XFS_WANT_CORRUPTED_RETURN(mp, lastfree == 0);
139 			XFS_WANT_CORRUPTED_RETURN(mp, endp >=
140 					p + be16_to_cpu(dup->length));
141 			XFS_WANT_CORRUPTED_RETURN(mp,
142 				be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup)) ==
143 					       (char *)dup - (char *)hdr);
144 			dfp = xfs_dir2_data_freefind(hdr, bf, dup);
145 			if (dfp) {
146 				i = (int)(dfp - bf);
147 				XFS_WANT_CORRUPTED_RETURN(mp,
148 					(freeseen & (1 << i)) == 0);
149 				freeseen |= 1 << i;
150 			} else {
151 				XFS_WANT_CORRUPTED_RETURN(mp,
152 					be16_to_cpu(dup->length) <=
153 						be16_to_cpu(bf[2].length));
154 			}
155 			p += be16_to_cpu(dup->length);
156 			lastfree = 1;
157 			continue;
158 		}
159 		/*
160 		 * It's a real entry.  Validate the fields.
161 		 * If this is a block directory then make sure it's
162 		 * in the leaf section of the block.
163 		 * The linear search is crude but this is DEBUG code.
164 		 */
165 		dep = (xfs_dir2_data_entry_t *)p;
166 		XFS_WANT_CORRUPTED_RETURN(mp, dep->namelen != 0);
167 		XFS_WANT_CORRUPTED_RETURN(mp,
168 			!xfs_dir_ino_validate(mp, be64_to_cpu(dep->inumber)));
169 		XFS_WANT_CORRUPTED_RETURN(mp, endp >=
170 				p + ops->data_entsize(dep->namelen));
171 		XFS_WANT_CORRUPTED_RETURN(mp,
172 			be16_to_cpu(*ops->data_entry_tag_p(dep)) ==
173 					       (char *)dep - (char *)hdr);
174 		XFS_WANT_CORRUPTED_RETURN(mp,
175 				ops->data_get_ftype(dep) < XFS_DIR3_FT_MAX);
176 		count++;
177 		lastfree = 0;
178 		if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
179 		    hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC)) {
180 			addr = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
181 						(xfs_dir2_data_aoff_t)
182 						((char *)dep - (char *)hdr));
183 			name.name = dep->name;
184 			name.len = dep->namelen;
185 			hash = mp->m_dirnameops->hashname(&name);
186 			for (i = 0; i < be32_to_cpu(btp->count); i++) {
187 				if (be32_to_cpu(lep[i].address) == addr &&
188 				    be32_to_cpu(lep[i].hashval) == hash)
189 					break;
190 			}
191 			XFS_WANT_CORRUPTED_RETURN(mp,
192 						  i < be32_to_cpu(btp->count));
193 		}
194 		p += ops->data_entsize(dep->namelen);
195 	}
196 	/*
197 	 * Need to have seen all the entries and all the bestfree slots.
198 	 */
199 	XFS_WANT_CORRUPTED_RETURN(mp, freeseen == 7);
200 	if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
201 	    hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC)) {
202 		for (i = stale = 0; i < be32_to_cpu(btp->count); i++) {
203 			if (lep[i].address ==
204 			    cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
205 				stale++;
206 			if (i > 0)
207 				XFS_WANT_CORRUPTED_RETURN(mp,
208 					be32_to_cpu(lep[i].hashval) >=
209 						be32_to_cpu(lep[i - 1].hashval));
210 		}
211 		XFS_WANT_CORRUPTED_RETURN(mp, count ==
212 			be32_to_cpu(btp->count) - be32_to_cpu(btp->stale));
213 		XFS_WANT_CORRUPTED_RETURN(mp, stale == be32_to_cpu(btp->stale));
214 	}
215 	return 0;
216 }
217 
218 static bool
219 xfs_dir3_data_verify(
220 	struct xfs_buf		*bp)
221 {
222 	struct xfs_mount	*mp = bp->b_target->bt_mount;
223 	struct xfs_dir3_blk_hdr	*hdr3 = bp->b_addr;
224 
225 	if (xfs_sb_version_hascrc(&mp->m_sb)) {
226 		if (hdr3->magic != cpu_to_be32(XFS_DIR3_DATA_MAGIC))
227 			return false;
228 		if (!uuid_equal(&hdr3->uuid, &mp->m_sb.sb_meta_uuid))
229 			return false;
230 		if (be64_to_cpu(hdr3->blkno) != bp->b_bn)
231 			return false;
232 		if (!xfs_log_check_lsn(mp, be64_to_cpu(hdr3->lsn)))
233 			return false;
234 	} else {
235 		if (hdr3->magic != cpu_to_be32(XFS_DIR2_DATA_MAGIC))
236 			return false;
237 	}
238 	if (__xfs_dir3_data_check(NULL, bp))
239 		return false;
240 	return true;
241 }
242 
243 /*
244  * Readahead of the first block of the directory when it is opened is completely
245  * oblivious to the format of the directory. Hence we can either get a block
246  * format buffer or a data format buffer on readahead.
247  */
248 static void
249 xfs_dir3_data_reada_verify(
250 	struct xfs_buf		*bp)
251 {
252 	struct xfs_dir2_data_hdr *hdr = bp->b_addr;
253 
254 	switch (hdr->magic) {
255 	case cpu_to_be32(XFS_DIR2_BLOCK_MAGIC):
256 	case cpu_to_be32(XFS_DIR3_BLOCK_MAGIC):
257 		bp->b_ops = &xfs_dir3_block_buf_ops;
258 		bp->b_ops->verify_read(bp);
259 		return;
260 	case cpu_to_be32(XFS_DIR2_DATA_MAGIC):
261 	case cpu_to_be32(XFS_DIR3_DATA_MAGIC):
262 		bp->b_ops = &xfs_dir3_data_buf_ops;
263 		bp->b_ops->verify_read(bp);
264 		return;
265 	default:
266 		xfs_buf_ioerror(bp, -EFSCORRUPTED);
267 		xfs_verifier_error(bp);
268 		break;
269 	}
270 }
271 
272 static void
273 xfs_dir3_data_read_verify(
274 	struct xfs_buf	*bp)
275 {
276 	struct xfs_mount	*mp = bp->b_target->bt_mount;
277 
278 	if (xfs_sb_version_hascrc(&mp->m_sb) &&
279 	     !xfs_buf_verify_cksum(bp, XFS_DIR3_DATA_CRC_OFF))
280 		 xfs_buf_ioerror(bp, -EFSBADCRC);
281 	else if (!xfs_dir3_data_verify(bp))
282 		xfs_buf_ioerror(bp, -EFSCORRUPTED);
283 
284 	if (bp->b_error)
285 		xfs_verifier_error(bp);
286 }
287 
288 static void
289 xfs_dir3_data_write_verify(
290 	struct xfs_buf	*bp)
291 {
292 	struct xfs_mount	*mp = bp->b_target->bt_mount;
293 	struct xfs_buf_log_item	*bip = bp->b_fspriv;
294 	struct xfs_dir3_blk_hdr	*hdr3 = bp->b_addr;
295 
296 	if (!xfs_dir3_data_verify(bp)) {
297 		xfs_buf_ioerror(bp, -EFSCORRUPTED);
298 		xfs_verifier_error(bp);
299 		return;
300 	}
301 
302 	if (!xfs_sb_version_hascrc(&mp->m_sb))
303 		return;
304 
305 	if (bip)
306 		hdr3->lsn = cpu_to_be64(bip->bli_item.li_lsn);
307 
308 	xfs_buf_update_cksum(bp, XFS_DIR3_DATA_CRC_OFF);
309 }
310 
311 const struct xfs_buf_ops xfs_dir3_data_buf_ops = {
312 	.name = "xfs_dir3_data",
313 	.verify_read = xfs_dir3_data_read_verify,
314 	.verify_write = xfs_dir3_data_write_verify,
315 };
316 
317 static const struct xfs_buf_ops xfs_dir3_data_reada_buf_ops = {
318 	.name = "xfs_dir3_data_reada",
319 	.verify_read = xfs_dir3_data_reada_verify,
320 	.verify_write = xfs_dir3_data_write_verify,
321 };
322 
323 
324 int
325 xfs_dir3_data_read(
326 	struct xfs_trans	*tp,
327 	struct xfs_inode	*dp,
328 	xfs_dablk_t		bno,
329 	xfs_daddr_t		mapped_bno,
330 	struct xfs_buf		**bpp)
331 {
332 	int			err;
333 
334 	err = xfs_da_read_buf(tp, dp, bno, mapped_bno, bpp,
335 				XFS_DATA_FORK, &xfs_dir3_data_buf_ops);
336 	if (!err && tp && *bpp)
337 		xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_DATA_BUF);
338 	return err;
339 }
340 
341 int
342 xfs_dir3_data_readahead(
343 	struct xfs_inode	*dp,
344 	xfs_dablk_t		bno,
345 	xfs_daddr_t		mapped_bno)
346 {
347 	return xfs_da_reada_buf(dp, bno, mapped_bno,
348 				XFS_DATA_FORK, &xfs_dir3_data_reada_buf_ops);
349 }
350 
351 /*
352  * Given a data block and an unused entry from that block,
353  * return the bestfree entry if any that corresponds to it.
354  */
355 xfs_dir2_data_free_t *
356 xfs_dir2_data_freefind(
357 	struct xfs_dir2_data_hdr *hdr,		/* data block header */
358 	struct xfs_dir2_data_free *bf,		/* bestfree table pointer */
359 	struct xfs_dir2_data_unused *dup)	/* unused space */
360 {
361 	xfs_dir2_data_free_t	*dfp;		/* bestfree entry */
362 	xfs_dir2_data_aoff_t	off;		/* offset value needed */
363 #ifdef DEBUG
364 	int			matched;	/* matched the value */
365 	int			seenzero;	/* saw a 0 bestfree entry */
366 #endif
367 
368 	off = (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr);
369 
370 #ifdef DEBUG
371 	/*
372 	 * Validate some consistency in the bestfree table.
373 	 * Check order, non-overlapping entries, and if we find the
374 	 * one we're looking for it has to be exact.
375 	 */
376 	ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
377 	       hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
378 	       hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
379 	       hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
380 	for (dfp = &bf[0], seenzero = matched = 0;
381 	     dfp < &bf[XFS_DIR2_DATA_FD_COUNT];
382 	     dfp++) {
383 		if (!dfp->offset) {
384 			ASSERT(!dfp->length);
385 			seenzero = 1;
386 			continue;
387 		}
388 		ASSERT(seenzero == 0);
389 		if (be16_to_cpu(dfp->offset) == off) {
390 			matched = 1;
391 			ASSERT(dfp->length == dup->length);
392 		} else if (off < be16_to_cpu(dfp->offset))
393 			ASSERT(off + be16_to_cpu(dup->length) <= be16_to_cpu(dfp->offset));
394 		else
395 			ASSERT(be16_to_cpu(dfp->offset) + be16_to_cpu(dfp->length) <= off);
396 		ASSERT(matched || be16_to_cpu(dfp->length) >= be16_to_cpu(dup->length));
397 		if (dfp > &bf[0])
398 			ASSERT(be16_to_cpu(dfp[-1].length) >= be16_to_cpu(dfp[0].length));
399 	}
400 #endif
401 	/*
402 	 * If this is smaller than the smallest bestfree entry,
403 	 * it can't be there since they're sorted.
404 	 */
405 	if (be16_to_cpu(dup->length) <
406 	    be16_to_cpu(bf[XFS_DIR2_DATA_FD_COUNT - 1].length))
407 		return NULL;
408 	/*
409 	 * Look at the three bestfree entries for our guy.
410 	 */
411 	for (dfp = &bf[0]; dfp < &bf[XFS_DIR2_DATA_FD_COUNT]; dfp++) {
412 		if (!dfp->offset)
413 			return NULL;
414 		if (be16_to_cpu(dfp->offset) == off)
415 			return dfp;
416 	}
417 	/*
418 	 * Didn't find it.  This only happens if there are duplicate lengths.
419 	 */
420 	return NULL;
421 }
422 
423 /*
424  * Insert an unused-space entry into the bestfree table.
425  */
426 xfs_dir2_data_free_t *				/* entry inserted */
427 xfs_dir2_data_freeinsert(
428 	struct xfs_dir2_data_hdr *hdr,		/* data block pointer */
429 	struct xfs_dir2_data_free *dfp,		/* bestfree table pointer */
430 	struct xfs_dir2_data_unused *dup,	/* unused space */
431 	int			*loghead)	/* log the data header (out) */
432 {
433 	xfs_dir2_data_free_t	new;		/* new bestfree entry */
434 
435 	ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
436 	       hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
437 	       hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
438 	       hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
439 
440 	new.length = dup->length;
441 	new.offset = cpu_to_be16((char *)dup - (char *)hdr);
442 
443 	/*
444 	 * Insert at position 0, 1, or 2; or not at all.
445 	 */
446 	if (be16_to_cpu(new.length) > be16_to_cpu(dfp[0].length)) {
447 		dfp[2] = dfp[1];
448 		dfp[1] = dfp[0];
449 		dfp[0] = new;
450 		*loghead = 1;
451 		return &dfp[0];
452 	}
453 	if (be16_to_cpu(new.length) > be16_to_cpu(dfp[1].length)) {
454 		dfp[2] = dfp[1];
455 		dfp[1] = new;
456 		*loghead = 1;
457 		return &dfp[1];
458 	}
459 	if (be16_to_cpu(new.length) > be16_to_cpu(dfp[2].length)) {
460 		dfp[2] = new;
461 		*loghead = 1;
462 		return &dfp[2];
463 	}
464 	return NULL;
465 }
466 
467 /*
468  * Remove a bestfree entry from the table.
469  */
470 STATIC void
471 xfs_dir2_data_freeremove(
472 	struct xfs_dir2_data_hdr *hdr,		/* data block header */
473 	struct xfs_dir2_data_free *bf,		/* bestfree table pointer */
474 	struct xfs_dir2_data_free *dfp,		/* bestfree entry pointer */
475 	int			*loghead)	/* out: log data header */
476 {
477 
478 	ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
479 	       hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
480 	       hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
481 	       hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
482 
483 	/*
484 	 * It's the first entry, slide the next 2 up.
485 	 */
486 	if (dfp == &bf[0]) {
487 		bf[0] = bf[1];
488 		bf[1] = bf[2];
489 	}
490 	/*
491 	 * It's the second entry, slide the 3rd entry up.
492 	 */
493 	else if (dfp == &bf[1])
494 		bf[1] = bf[2];
495 	/*
496 	 * Must be the last entry.
497 	 */
498 	else
499 		ASSERT(dfp == &bf[2]);
500 	/*
501 	 * Clear the 3rd entry, must be zero now.
502 	 */
503 	bf[2].length = 0;
504 	bf[2].offset = 0;
505 	*loghead = 1;
506 }
507 
508 /*
509  * Given a data block, reconstruct its bestfree map.
510  */
511 void
512 xfs_dir2_data_freescan_int(
513 	struct xfs_da_geometry	*geo,
514 	const struct xfs_dir_ops *ops,
515 	struct xfs_dir2_data_hdr *hdr,
516 	int			*loghead)
517 {
518 	xfs_dir2_block_tail_t	*btp;		/* block tail */
519 	xfs_dir2_data_entry_t	*dep;		/* active data entry */
520 	xfs_dir2_data_unused_t	*dup;		/* unused data entry */
521 	struct xfs_dir2_data_free *bf;
522 	char			*endp;		/* end of block's data */
523 	char			*p;		/* current entry pointer */
524 
525 	ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
526 	       hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
527 	       hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
528 	       hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
529 
530 	/*
531 	 * Start by clearing the table.
532 	 */
533 	bf = ops->data_bestfree_p(hdr);
534 	memset(bf, 0, sizeof(*bf) * XFS_DIR2_DATA_FD_COUNT);
535 	*loghead = 1;
536 	/*
537 	 * Set up pointers.
538 	 */
539 	p = (char *)ops->data_entry_p(hdr);
540 	if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
541 	    hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC)) {
542 		btp = xfs_dir2_block_tail_p(geo, hdr);
543 		endp = (char *)xfs_dir2_block_leaf_p(btp);
544 	} else
545 		endp = (char *)hdr + geo->blksize;
546 	/*
547 	 * Loop over the block's entries.
548 	 */
549 	while (p < endp) {
550 		dup = (xfs_dir2_data_unused_t *)p;
551 		/*
552 		 * If it's a free entry, insert it.
553 		 */
554 		if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
555 			ASSERT((char *)dup - (char *)hdr ==
556 			       be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup)));
557 			xfs_dir2_data_freeinsert(hdr, bf, dup, loghead);
558 			p += be16_to_cpu(dup->length);
559 		}
560 		/*
561 		 * For active entries, check their tags and skip them.
562 		 */
563 		else {
564 			dep = (xfs_dir2_data_entry_t *)p;
565 			ASSERT((char *)dep - (char *)hdr ==
566 			       be16_to_cpu(*ops->data_entry_tag_p(dep)));
567 			p += ops->data_entsize(dep->namelen);
568 		}
569 	}
570 }
571 
572 void
573 xfs_dir2_data_freescan(
574 	struct xfs_inode	*dp,
575 	struct xfs_dir2_data_hdr *hdr,
576 	int			*loghead)
577 {
578 	return xfs_dir2_data_freescan_int(dp->i_mount->m_dir_geo, dp->d_ops,
579 			hdr, loghead);
580 }
581 
582 /*
583  * Initialize a data block at the given block number in the directory.
584  * Give back the buffer for the created block.
585  */
586 int						/* error */
587 xfs_dir3_data_init(
588 	xfs_da_args_t		*args,		/* directory operation args */
589 	xfs_dir2_db_t		blkno,		/* logical dir block number */
590 	struct xfs_buf		**bpp)		/* output block buffer */
591 {
592 	struct xfs_buf		*bp;		/* block buffer */
593 	xfs_dir2_data_hdr_t	*hdr;		/* data block header */
594 	xfs_inode_t		*dp;		/* incore directory inode */
595 	xfs_dir2_data_unused_t	*dup;		/* unused entry pointer */
596 	struct xfs_dir2_data_free *bf;
597 	int			error;		/* error return value */
598 	int			i;		/* bestfree index */
599 	xfs_mount_t		*mp;		/* filesystem mount point */
600 	xfs_trans_t		*tp;		/* transaction pointer */
601 	int                     t;              /* temp */
602 
603 	dp = args->dp;
604 	mp = dp->i_mount;
605 	tp = args->trans;
606 	/*
607 	 * Get the buffer set up for the block.
608 	 */
609 	error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(args->geo, blkno),
610 			       -1, &bp, XFS_DATA_FORK);
611 	if (error)
612 		return error;
613 	bp->b_ops = &xfs_dir3_data_buf_ops;
614 	xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_DATA_BUF);
615 
616 	/*
617 	 * Initialize the header.
618 	 */
619 	hdr = bp->b_addr;
620 	if (xfs_sb_version_hascrc(&mp->m_sb)) {
621 		struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
622 
623 		memset(hdr3, 0, sizeof(*hdr3));
624 		hdr3->magic = cpu_to_be32(XFS_DIR3_DATA_MAGIC);
625 		hdr3->blkno = cpu_to_be64(bp->b_bn);
626 		hdr3->owner = cpu_to_be64(dp->i_ino);
627 		uuid_copy(&hdr3->uuid, &mp->m_sb.sb_meta_uuid);
628 
629 	} else
630 		hdr->magic = cpu_to_be32(XFS_DIR2_DATA_MAGIC);
631 
632 	bf = dp->d_ops->data_bestfree_p(hdr);
633 	bf[0].offset = cpu_to_be16(dp->d_ops->data_entry_offset);
634 	for (i = 1; i < XFS_DIR2_DATA_FD_COUNT; i++) {
635 		bf[i].length = 0;
636 		bf[i].offset = 0;
637 	}
638 
639 	/*
640 	 * Set up an unused entry for the block's body.
641 	 */
642 	dup = dp->d_ops->data_unused_p(hdr);
643 	dup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
644 
645 	t = args->geo->blksize - (uint)dp->d_ops->data_entry_offset;
646 	bf[0].length = cpu_to_be16(t);
647 	dup->length = cpu_to_be16(t);
648 	*xfs_dir2_data_unused_tag_p(dup) = cpu_to_be16((char *)dup - (char *)hdr);
649 	/*
650 	 * Log it and return it.
651 	 */
652 	xfs_dir2_data_log_header(args, bp);
653 	xfs_dir2_data_log_unused(args, bp, dup);
654 	*bpp = bp;
655 	return 0;
656 }
657 
658 /*
659  * Log an active data entry from the block.
660  */
661 void
662 xfs_dir2_data_log_entry(
663 	struct xfs_da_args	*args,
664 	struct xfs_buf		*bp,
665 	xfs_dir2_data_entry_t	*dep)		/* data entry pointer */
666 {
667 	struct xfs_dir2_data_hdr *hdr = bp->b_addr;
668 
669 	ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
670 	       hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
671 	       hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
672 	       hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
673 
674 	xfs_trans_log_buf(args->trans, bp, (uint)((char *)dep - (char *)hdr),
675 		(uint)((char *)(args->dp->d_ops->data_entry_tag_p(dep) + 1) -
676 		       (char *)hdr - 1));
677 }
678 
679 /*
680  * Log a data block header.
681  */
682 void
683 xfs_dir2_data_log_header(
684 	struct xfs_da_args	*args,
685 	struct xfs_buf		*bp)
686 {
687 #ifdef DEBUG
688 	struct xfs_dir2_data_hdr *hdr = bp->b_addr;
689 
690 	ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
691 	       hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
692 	       hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
693 	       hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
694 #endif
695 
696 	xfs_trans_log_buf(args->trans, bp, 0,
697 			  args->dp->d_ops->data_entry_offset - 1);
698 }
699 
700 /*
701  * Log a data unused entry.
702  */
703 void
704 xfs_dir2_data_log_unused(
705 	struct xfs_da_args	*args,
706 	struct xfs_buf		*bp,
707 	xfs_dir2_data_unused_t	*dup)		/* data unused pointer */
708 {
709 	xfs_dir2_data_hdr_t	*hdr = bp->b_addr;
710 
711 	ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
712 	       hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
713 	       hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
714 	       hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
715 
716 	/*
717 	 * Log the first part of the unused entry.
718 	 */
719 	xfs_trans_log_buf(args->trans, bp, (uint)((char *)dup - (char *)hdr),
720 		(uint)((char *)&dup->length + sizeof(dup->length) -
721 		       1 - (char *)hdr));
722 	/*
723 	 * Log the end (tag) of the unused entry.
724 	 */
725 	xfs_trans_log_buf(args->trans, bp,
726 		(uint)((char *)xfs_dir2_data_unused_tag_p(dup) - (char *)hdr),
727 		(uint)((char *)xfs_dir2_data_unused_tag_p(dup) - (char *)hdr +
728 		       sizeof(xfs_dir2_data_off_t) - 1));
729 }
730 
731 /*
732  * Make a byte range in the data block unused.
733  * Its current contents are unimportant.
734  */
735 void
736 xfs_dir2_data_make_free(
737 	struct xfs_da_args	*args,
738 	struct xfs_buf		*bp,
739 	xfs_dir2_data_aoff_t	offset,		/* starting byte offset */
740 	xfs_dir2_data_aoff_t	len,		/* length in bytes */
741 	int			*needlogp,	/* out: log header */
742 	int			*needscanp)	/* out: regen bestfree */
743 {
744 	xfs_dir2_data_hdr_t	*hdr;		/* data block pointer */
745 	xfs_dir2_data_free_t	*dfp;		/* bestfree pointer */
746 	char			*endptr;	/* end of data area */
747 	int			needscan;	/* need to regen bestfree */
748 	xfs_dir2_data_unused_t	*newdup;	/* new unused entry */
749 	xfs_dir2_data_unused_t	*postdup;	/* unused entry after us */
750 	xfs_dir2_data_unused_t	*prevdup;	/* unused entry before us */
751 	struct xfs_dir2_data_free *bf;
752 
753 	hdr = bp->b_addr;
754 
755 	/*
756 	 * Figure out where the end of the data area is.
757 	 */
758 	if (hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
759 	    hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC))
760 		endptr = (char *)hdr + args->geo->blksize;
761 	else {
762 		xfs_dir2_block_tail_t	*btp;	/* block tail */
763 
764 		ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
765 			hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
766 		btp = xfs_dir2_block_tail_p(args->geo, hdr);
767 		endptr = (char *)xfs_dir2_block_leaf_p(btp);
768 	}
769 	/*
770 	 * If this isn't the start of the block, then back up to
771 	 * the previous entry and see if it's free.
772 	 */
773 	if (offset > args->dp->d_ops->data_entry_offset) {
774 		__be16			*tagp;	/* tag just before us */
775 
776 		tagp = (__be16 *)((char *)hdr + offset) - 1;
777 		prevdup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
778 		if (be16_to_cpu(prevdup->freetag) != XFS_DIR2_DATA_FREE_TAG)
779 			prevdup = NULL;
780 	} else
781 		prevdup = NULL;
782 	/*
783 	 * If this isn't the end of the block, see if the entry after
784 	 * us is free.
785 	 */
786 	if ((char *)hdr + offset + len < endptr) {
787 		postdup =
788 			(xfs_dir2_data_unused_t *)((char *)hdr + offset + len);
789 		if (be16_to_cpu(postdup->freetag) != XFS_DIR2_DATA_FREE_TAG)
790 			postdup = NULL;
791 	} else
792 		postdup = NULL;
793 	ASSERT(*needscanp == 0);
794 	needscan = 0;
795 	/*
796 	 * Previous and following entries are both free,
797 	 * merge everything into a single free entry.
798 	 */
799 	bf = args->dp->d_ops->data_bestfree_p(hdr);
800 	if (prevdup && postdup) {
801 		xfs_dir2_data_free_t	*dfp2;	/* another bestfree pointer */
802 
803 		/*
804 		 * See if prevdup and/or postdup are in bestfree table.
805 		 */
806 		dfp = xfs_dir2_data_freefind(hdr, bf, prevdup);
807 		dfp2 = xfs_dir2_data_freefind(hdr, bf, postdup);
808 		/*
809 		 * We need a rescan unless there are exactly 2 free entries
810 		 * namely our two.  Then we know what's happening, otherwise
811 		 * since the third bestfree is there, there might be more
812 		 * entries.
813 		 */
814 		needscan = (bf[2].length != 0);
815 		/*
816 		 * Fix up the new big freespace.
817 		 */
818 		be16_add_cpu(&prevdup->length, len + be16_to_cpu(postdup->length));
819 		*xfs_dir2_data_unused_tag_p(prevdup) =
820 			cpu_to_be16((char *)prevdup - (char *)hdr);
821 		xfs_dir2_data_log_unused(args, bp, prevdup);
822 		if (!needscan) {
823 			/*
824 			 * Has to be the case that entries 0 and 1 are
825 			 * dfp and dfp2 (don't know which is which), and
826 			 * entry 2 is empty.
827 			 * Remove entry 1 first then entry 0.
828 			 */
829 			ASSERT(dfp && dfp2);
830 			if (dfp == &bf[1]) {
831 				dfp = &bf[0];
832 				ASSERT(dfp2 == dfp);
833 				dfp2 = &bf[1];
834 			}
835 			xfs_dir2_data_freeremove(hdr, bf, dfp2, needlogp);
836 			xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
837 			/*
838 			 * Now insert the new entry.
839 			 */
840 			dfp = xfs_dir2_data_freeinsert(hdr, bf, prevdup,
841 						       needlogp);
842 			ASSERT(dfp == &bf[0]);
843 			ASSERT(dfp->length == prevdup->length);
844 			ASSERT(!dfp[1].length);
845 			ASSERT(!dfp[2].length);
846 		}
847 	}
848 	/*
849 	 * The entry before us is free, merge with it.
850 	 */
851 	else if (prevdup) {
852 		dfp = xfs_dir2_data_freefind(hdr, bf, prevdup);
853 		be16_add_cpu(&prevdup->length, len);
854 		*xfs_dir2_data_unused_tag_p(prevdup) =
855 			cpu_to_be16((char *)prevdup - (char *)hdr);
856 		xfs_dir2_data_log_unused(args, bp, prevdup);
857 		/*
858 		 * If the previous entry was in the table, the new entry
859 		 * is longer, so it will be in the table too.  Remove
860 		 * the old one and add the new one.
861 		 */
862 		if (dfp) {
863 			xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
864 			xfs_dir2_data_freeinsert(hdr, bf, prevdup, needlogp);
865 		}
866 		/*
867 		 * Otherwise we need a scan if the new entry is big enough.
868 		 */
869 		else {
870 			needscan = be16_to_cpu(prevdup->length) >
871 				   be16_to_cpu(bf[2].length);
872 		}
873 	}
874 	/*
875 	 * The following entry is free, merge with it.
876 	 */
877 	else if (postdup) {
878 		dfp = xfs_dir2_data_freefind(hdr, bf, postdup);
879 		newdup = (xfs_dir2_data_unused_t *)((char *)hdr + offset);
880 		newdup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
881 		newdup->length = cpu_to_be16(len + be16_to_cpu(postdup->length));
882 		*xfs_dir2_data_unused_tag_p(newdup) =
883 			cpu_to_be16((char *)newdup - (char *)hdr);
884 		xfs_dir2_data_log_unused(args, bp, newdup);
885 		/*
886 		 * If the following entry was in the table, the new entry
887 		 * is longer, so it will be in the table too.  Remove
888 		 * the old one and add the new one.
889 		 */
890 		if (dfp) {
891 			xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
892 			xfs_dir2_data_freeinsert(hdr, bf, newdup, needlogp);
893 		}
894 		/*
895 		 * Otherwise we need a scan if the new entry is big enough.
896 		 */
897 		else {
898 			needscan = be16_to_cpu(newdup->length) >
899 				   be16_to_cpu(bf[2].length);
900 		}
901 	}
902 	/*
903 	 * Neither neighbor is free.  Make a new entry.
904 	 */
905 	else {
906 		newdup = (xfs_dir2_data_unused_t *)((char *)hdr + offset);
907 		newdup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
908 		newdup->length = cpu_to_be16(len);
909 		*xfs_dir2_data_unused_tag_p(newdup) =
910 			cpu_to_be16((char *)newdup - (char *)hdr);
911 		xfs_dir2_data_log_unused(args, bp, newdup);
912 		xfs_dir2_data_freeinsert(hdr, bf, newdup, needlogp);
913 	}
914 	*needscanp = needscan;
915 }
916 
917 /*
918  * Take a byte range out of an existing unused space and make it un-free.
919  */
920 void
921 xfs_dir2_data_use_free(
922 	struct xfs_da_args	*args,
923 	struct xfs_buf		*bp,
924 	xfs_dir2_data_unused_t	*dup,		/* unused entry */
925 	xfs_dir2_data_aoff_t	offset,		/* starting offset to use */
926 	xfs_dir2_data_aoff_t	len,		/* length to use */
927 	int			*needlogp,	/* out: need to log header */
928 	int			*needscanp)	/* out: need regen bestfree */
929 {
930 	xfs_dir2_data_hdr_t	*hdr;		/* data block header */
931 	xfs_dir2_data_free_t	*dfp;		/* bestfree pointer */
932 	int			matchback;	/* matches end of freespace */
933 	int			matchfront;	/* matches start of freespace */
934 	int			needscan;	/* need to regen bestfree */
935 	xfs_dir2_data_unused_t	*newdup;	/* new unused entry */
936 	xfs_dir2_data_unused_t	*newdup2;	/* another new unused entry */
937 	int			oldlen;		/* old unused entry's length */
938 	struct xfs_dir2_data_free *bf;
939 
940 	hdr = bp->b_addr;
941 	ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
942 	       hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
943 	       hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
944 	       hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
945 	ASSERT(be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG);
946 	ASSERT(offset >= (char *)dup - (char *)hdr);
947 	ASSERT(offset + len <= (char *)dup + be16_to_cpu(dup->length) - (char *)hdr);
948 	ASSERT((char *)dup - (char *)hdr == be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup)));
949 	/*
950 	 * Look up the entry in the bestfree table.
951 	 */
952 	oldlen = be16_to_cpu(dup->length);
953 	bf = args->dp->d_ops->data_bestfree_p(hdr);
954 	dfp = xfs_dir2_data_freefind(hdr, bf, dup);
955 	ASSERT(dfp || oldlen <= be16_to_cpu(bf[2].length));
956 	/*
957 	 * Check for alignment with front and back of the entry.
958 	 */
959 	matchfront = (char *)dup - (char *)hdr == offset;
960 	matchback = (char *)dup + oldlen - (char *)hdr == offset + len;
961 	ASSERT(*needscanp == 0);
962 	needscan = 0;
963 	/*
964 	 * If we matched it exactly we just need to get rid of it from
965 	 * the bestfree table.
966 	 */
967 	if (matchfront && matchback) {
968 		if (dfp) {
969 			needscan = (bf[2].offset != 0);
970 			if (!needscan)
971 				xfs_dir2_data_freeremove(hdr, bf, dfp,
972 							 needlogp);
973 		}
974 	}
975 	/*
976 	 * We match the first part of the entry.
977 	 * Make a new entry with the remaining freespace.
978 	 */
979 	else if (matchfront) {
980 		newdup = (xfs_dir2_data_unused_t *)((char *)hdr + offset + len);
981 		newdup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
982 		newdup->length = cpu_to_be16(oldlen - len);
983 		*xfs_dir2_data_unused_tag_p(newdup) =
984 			cpu_to_be16((char *)newdup - (char *)hdr);
985 		xfs_dir2_data_log_unused(args, bp, newdup);
986 		/*
987 		 * If it was in the table, remove it and add the new one.
988 		 */
989 		if (dfp) {
990 			xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
991 			dfp = xfs_dir2_data_freeinsert(hdr, bf, newdup,
992 						       needlogp);
993 			ASSERT(dfp != NULL);
994 			ASSERT(dfp->length == newdup->length);
995 			ASSERT(be16_to_cpu(dfp->offset) == (char *)newdup - (char *)hdr);
996 			/*
997 			 * If we got inserted at the last slot,
998 			 * that means we don't know if there was a better
999 			 * choice for the last slot, or not.  Rescan.
1000 			 */
1001 			needscan = dfp == &bf[2];
1002 		}
1003 	}
1004 	/*
1005 	 * We match the last part of the entry.
1006 	 * Trim the allocated space off the tail of the entry.
1007 	 */
1008 	else if (matchback) {
1009 		newdup = dup;
1010 		newdup->length = cpu_to_be16(((char *)hdr + offset) - (char *)newdup);
1011 		*xfs_dir2_data_unused_tag_p(newdup) =
1012 			cpu_to_be16((char *)newdup - (char *)hdr);
1013 		xfs_dir2_data_log_unused(args, bp, newdup);
1014 		/*
1015 		 * If it was in the table, remove it and add the new one.
1016 		 */
1017 		if (dfp) {
1018 			xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
1019 			dfp = xfs_dir2_data_freeinsert(hdr, bf, newdup,
1020 						       needlogp);
1021 			ASSERT(dfp != NULL);
1022 			ASSERT(dfp->length == newdup->length);
1023 			ASSERT(be16_to_cpu(dfp->offset) == (char *)newdup - (char *)hdr);
1024 			/*
1025 			 * If we got inserted at the last slot,
1026 			 * that means we don't know if there was a better
1027 			 * choice for the last slot, or not.  Rescan.
1028 			 */
1029 			needscan = dfp == &bf[2];
1030 		}
1031 	}
1032 	/*
1033 	 * Poking out the middle of an entry.
1034 	 * Make two new entries.
1035 	 */
1036 	else {
1037 		newdup = dup;
1038 		newdup->length = cpu_to_be16(((char *)hdr + offset) - (char *)newdup);
1039 		*xfs_dir2_data_unused_tag_p(newdup) =
1040 			cpu_to_be16((char *)newdup - (char *)hdr);
1041 		xfs_dir2_data_log_unused(args, bp, newdup);
1042 		newdup2 = (xfs_dir2_data_unused_t *)((char *)hdr + offset + len);
1043 		newdup2->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
1044 		newdup2->length = cpu_to_be16(oldlen - len - be16_to_cpu(newdup->length));
1045 		*xfs_dir2_data_unused_tag_p(newdup2) =
1046 			cpu_to_be16((char *)newdup2 - (char *)hdr);
1047 		xfs_dir2_data_log_unused(args, bp, newdup2);
1048 		/*
1049 		 * If the old entry was in the table, we need to scan
1050 		 * if the 3rd entry was valid, since these entries
1051 		 * are smaller than the old one.
1052 		 * If we don't need to scan that means there were 1 or 2
1053 		 * entries in the table, and removing the old and adding
1054 		 * the 2 new will work.
1055 		 */
1056 		if (dfp) {
1057 			needscan = (bf[2].length != 0);
1058 			if (!needscan) {
1059 				xfs_dir2_data_freeremove(hdr, bf, dfp,
1060 							 needlogp);
1061 				xfs_dir2_data_freeinsert(hdr, bf, newdup,
1062 							 needlogp);
1063 				xfs_dir2_data_freeinsert(hdr, bf, newdup2,
1064 							 needlogp);
1065 			}
1066 		}
1067 	}
1068 	*needscanp = needscan;
1069 }
1070