xref: /openbmc/linux/fs/xfs/libxfs/xfs_dir2_data.c (revision a06c488d)
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,
140 				be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup)) ==
141 					       (char *)dup - (char *)hdr);
142 			dfp = xfs_dir2_data_freefind(hdr, bf, dup);
143 			if (dfp) {
144 				i = (int)(dfp - bf);
145 				XFS_WANT_CORRUPTED_RETURN(mp,
146 					(freeseen & (1 << i)) == 0);
147 				freeseen |= 1 << i;
148 			} else {
149 				XFS_WANT_CORRUPTED_RETURN(mp,
150 					be16_to_cpu(dup->length) <=
151 						be16_to_cpu(bf[2].length));
152 			}
153 			p += be16_to_cpu(dup->length);
154 			lastfree = 1;
155 			continue;
156 		}
157 		/*
158 		 * It's a real entry.  Validate the fields.
159 		 * If this is a block directory then make sure it's
160 		 * in the leaf section of the block.
161 		 * The linear search is crude but this is DEBUG code.
162 		 */
163 		dep = (xfs_dir2_data_entry_t *)p;
164 		XFS_WANT_CORRUPTED_RETURN(mp, dep->namelen != 0);
165 		XFS_WANT_CORRUPTED_RETURN(mp,
166 			!xfs_dir_ino_validate(mp, be64_to_cpu(dep->inumber)));
167 		XFS_WANT_CORRUPTED_RETURN(mp,
168 			be16_to_cpu(*ops->data_entry_tag_p(dep)) ==
169 					       (char *)dep - (char *)hdr);
170 		XFS_WANT_CORRUPTED_RETURN(mp,
171 				ops->data_get_ftype(dep) < XFS_DIR3_FT_MAX);
172 		count++;
173 		lastfree = 0;
174 		if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
175 		    hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC)) {
176 			addr = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
177 						(xfs_dir2_data_aoff_t)
178 						((char *)dep - (char *)hdr));
179 			name.name = dep->name;
180 			name.len = dep->namelen;
181 			hash = mp->m_dirnameops->hashname(&name);
182 			for (i = 0; i < be32_to_cpu(btp->count); i++) {
183 				if (be32_to_cpu(lep[i].address) == addr &&
184 				    be32_to_cpu(lep[i].hashval) == hash)
185 					break;
186 			}
187 			XFS_WANT_CORRUPTED_RETURN(mp,
188 						  i < be32_to_cpu(btp->count));
189 		}
190 		p += ops->data_entsize(dep->namelen);
191 	}
192 	/*
193 	 * Need to have seen all the entries and all the bestfree slots.
194 	 */
195 	XFS_WANT_CORRUPTED_RETURN(mp, freeseen == 7);
196 	if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
197 	    hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC)) {
198 		for (i = stale = 0; i < be32_to_cpu(btp->count); i++) {
199 			if (lep[i].address ==
200 			    cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
201 				stale++;
202 			if (i > 0)
203 				XFS_WANT_CORRUPTED_RETURN(mp,
204 					be32_to_cpu(lep[i].hashval) >=
205 						be32_to_cpu(lep[i - 1].hashval));
206 		}
207 		XFS_WANT_CORRUPTED_RETURN(mp, count ==
208 			be32_to_cpu(btp->count) - be32_to_cpu(btp->stale));
209 		XFS_WANT_CORRUPTED_RETURN(mp, stale == be32_to_cpu(btp->stale));
210 	}
211 	return 0;
212 }
213 
214 static bool
215 xfs_dir3_data_verify(
216 	struct xfs_buf		*bp)
217 {
218 	struct xfs_mount	*mp = bp->b_target->bt_mount;
219 	struct xfs_dir3_blk_hdr	*hdr3 = bp->b_addr;
220 
221 	if (xfs_sb_version_hascrc(&mp->m_sb)) {
222 		if (hdr3->magic != cpu_to_be32(XFS_DIR3_DATA_MAGIC))
223 			return false;
224 		if (!uuid_equal(&hdr3->uuid, &mp->m_sb.sb_meta_uuid))
225 			return false;
226 		if (be64_to_cpu(hdr3->blkno) != bp->b_bn)
227 			return false;
228 		if (!xfs_log_check_lsn(mp, be64_to_cpu(hdr3->lsn)))
229 			return false;
230 	} else {
231 		if (hdr3->magic != cpu_to_be32(XFS_DIR2_DATA_MAGIC))
232 			return false;
233 	}
234 	if (__xfs_dir3_data_check(NULL, bp))
235 		return false;
236 	return true;
237 }
238 
239 /*
240  * Readahead of the first block of the directory when it is opened is completely
241  * oblivious to the format of the directory. Hence we can either get a block
242  * format buffer or a data format buffer on readahead.
243  */
244 static void
245 xfs_dir3_data_reada_verify(
246 	struct xfs_buf		*bp)
247 {
248 	struct xfs_dir2_data_hdr *hdr = bp->b_addr;
249 
250 	switch (hdr->magic) {
251 	case cpu_to_be32(XFS_DIR2_BLOCK_MAGIC):
252 	case cpu_to_be32(XFS_DIR3_BLOCK_MAGIC):
253 		bp->b_ops = &xfs_dir3_block_buf_ops;
254 		bp->b_ops->verify_read(bp);
255 		return;
256 	case cpu_to_be32(XFS_DIR2_DATA_MAGIC):
257 	case cpu_to_be32(XFS_DIR3_DATA_MAGIC):
258 		bp->b_ops = &xfs_dir3_data_buf_ops;
259 		bp->b_ops->verify_read(bp);
260 		return;
261 	default:
262 		xfs_buf_ioerror(bp, -EFSCORRUPTED);
263 		xfs_verifier_error(bp);
264 		break;
265 	}
266 }
267 
268 static void
269 xfs_dir3_data_read_verify(
270 	struct xfs_buf	*bp)
271 {
272 	struct xfs_mount	*mp = bp->b_target->bt_mount;
273 
274 	if (xfs_sb_version_hascrc(&mp->m_sb) &&
275 	     !xfs_buf_verify_cksum(bp, XFS_DIR3_DATA_CRC_OFF))
276 		 xfs_buf_ioerror(bp, -EFSBADCRC);
277 	else if (!xfs_dir3_data_verify(bp))
278 		xfs_buf_ioerror(bp, -EFSCORRUPTED);
279 
280 	if (bp->b_error)
281 		xfs_verifier_error(bp);
282 }
283 
284 static void
285 xfs_dir3_data_write_verify(
286 	struct xfs_buf	*bp)
287 {
288 	struct xfs_mount	*mp = bp->b_target->bt_mount;
289 	struct xfs_buf_log_item	*bip = bp->b_fspriv;
290 	struct xfs_dir3_blk_hdr	*hdr3 = bp->b_addr;
291 
292 	if (!xfs_dir3_data_verify(bp)) {
293 		xfs_buf_ioerror(bp, -EFSCORRUPTED);
294 		xfs_verifier_error(bp);
295 		return;
296 	}
297 
298 	if (!xfs_sb_version_hascrc(&mp->m_sb))
299 		return;
300 
301 	if (bip)
302 		hdr3->lsn = cpu_to_be64(bip->bli_item.li_lsn);
303 
304 	xfs_buf_update_cksum(bp, XFS_DIR3_DATA_CRC_OFF);
305 }
306 
307 const struct xfs_buf_ops xfs_dir3_data_buf_ops = {
308 	.name = "xfs_dir3_data",
309 	.verify_read = xfs_dir3_data_read_verify,
310 	.verify_write = xfs_dir3_data_write_verify,
311 };
312 
313 static const struct xfs_buf_ops xfs_dir3_data_reada_buf_ops = {
314 	.name = "xfs_dir3_data_reada",
315 	.verify_read = xfs_dir3_data_reada_verify,
316 	.verify_write = xfs_dir3_data_write_verify,
317 };
318 
319 
320 int
321 xfs_dir3_data_read(
322 	struct xfs_trans	*tp,
323 	struct xfs_inode	*dp,
324 	xfs_dablk_t		bno,
325 	xfs_daddr_t		mapped_bno,
326 	struct xfs_buf		**bpp)
327 {
328 	int			err;
329 
330 	err = xfs_da_read_buf(tp, dp, bno, mapped_bno, bpp,
331 				XFS_DATA_FORK, &xfs_dir3_data_buf_ops);
332 	if (!err && tp)
333 		xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_DATA_BUF);
334 	return err;
335 }
336 
337 int
338 xfs_dir3_data_readahead(
339 	struct xfs_inode	*dp,
340 	xfs_dablk_t		bno,
341 	xfs_daddr_t		mapped_bno)
342 {
343 	return xfs_da_reada_buf(dp, bno, mapped_bno,
344 				XFS_DATA_FORK, &xfs_dir3_data_reada_buf_ops);
345 }
346 
347 /*
348  * Given a data block and an unused entry from that block,
349  * return the bestfree entry if any that corresponds to it.
350  */
351 xfs_dir2_data_free_t *
352 xfs_dir2_data_freefind(
353 	struct xfs_dir2_data_hdr *hdr,		/* data block header */
354 	struct xfs_dir2_data_free *bf,		/* bestfree table pointer */
355 	struct xfs_dir2_data_unused *dup)	/* unused space */
356 {
357 	xfs_dir2_data_free_t	*dfp;		/* bestfree entry */
358 	xfs_dir2_data_aoff_t	off;		/* offset value needed */
359 #ifdef DEBUG
360 	int			matched;	/* matched the value */
361 	int			seenzero;	/* saw a 0 bestfree entry */
362 #endif
363 
364 	off = (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr);
365 
366 #ifdef DEBUG
367 	/*
368 	 * Validate some consistency in the bestfree table.
369 	 * Check order, non-overlapping entries, and if we find the
370 	 * one we're looking for it has to be exact.
371 	 */
372 	ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
373 	       hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
374 	       hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
375 	       hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
376 	for (dfp = &bf[0], seenzero = matched = 0;
377 	     dfp < &bf[XFS_DIR2_DATA_FD_COUNT];
378 	     dfp++) {
379 		if (!dfp->offset) {
380 			ASSERT(!dfp->length);
381 			seenzero = 1;
382 			continue;
383 		}
384 		ASSERT(seenzero == 0);
385 		if (be16_to_cpu(dfp->offset) == off) {
386 			matched = 1;
387 			ASSERT(dfp->length == dup->length);
388 		} else if (off < be16_to_cpu(dfp->offset))
389 			ASSERT(off + be16_to_cpu(dup->length) <= be16_to_cpu(dfp->offset));
390 		else
391 			ASSERT(be16_to_cpu(dfp->offset) + be16_to_cpu(dfp->length) <= off);
392 		ASSERT(matched || be16_to_cpu(dfp->length) >= be16_to_cpu(dup->length));
393 		if (dfp > &bf[0])
394 			ASSERT(be16_to_cpu(dfp[-1].length) >= be16_to_cpu(dfp[0].length));
395 	}
396 #endif
397 	/*
398 	 * If this is smaller than the smallest bestfree entry,
399 	 * it can't be there since they're sorted.
400 	 */
401 	if (be16_to_cpu(dup->length) <
402 	    be16_to_cpu(bf[XFS_DIR2_DATA_FD_COUNT - 1].length))
403 		return NULL;
404 	/*
405 	 * Look at the three bestfree entries for our guy.
406 	 */
407 	for (dfp = &bf[0]; dfp < &bf[XFS_DIR2_DATA_FD_COUNT]; dfp++) {
408 		if (!dfp->offset)
409 			return NULL;
410 		if (be16_to_cpu(dfp->offset) == off)
411 			return dfp;
412 	}
413 	/*
414 	 * Didn't find it.  This only happens if there are duplicate lengths.
415 	 */
416 	return NULL;
417 }
418 
419 /*
420  * Insert an unused-space entry into the bestfree table.
421  */
422 xfs_dir2_data_free_t *				/* entry inserted */
423 xfs_dir2_data_freeinsert(
424 	struct xfs_dir2_data_hdr *hdr,		/* data block pointer */
425 	struct xfs_dir2_data_free *dfp,		/* bestfree table pointer */
426 	struct xfs_dir2_data_unused *dup,	/* unused space */
427 	int			*loghead)	/* log the data header (out) */
428 {
429 	xfs_dir2_data_free_t	new;		/* new bestfree entry */
430 
431 	ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
432 	       hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
433 	       hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
434 	       hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
435 
436 	new.length = dup->length;
437 	new.offset = cpu_to_be16((char *)dup - (char *)hdr);
438 
439 	/*
440 	 * Insert at position 0, 1, or 2; or not at all.
441 	 */
442 	if (be16_to_cpu(new.length) > be16_to_cpu(dfp[0].length)) {
443 		dfp[2] = dfp[1];
444 		dfp[1] = dfp[0];
445 		dfp[0] = new;
446 		*loghead = 1;
447 		return &dfp[0];
448 	}
449 	if (be16_to_cpu(new.length) > be16_to_cpu(dfp[1].length)) {
450 		dfp[2] = dfp[1];
451 		dfp[1] = new;
452 		*loghead = 1;
453 		return &dfp[1];
454 	}
455 	if (be16_to_cpu(new.length) > be16_to_cpu(dfp[2].length)) {
456 		dfp[2] = new;
457 		*loghead = 1;
458 		return &dfp[2];
459 	}
460 	return NULL;
461 }
462 
463 /*
464  * Remove a bestfree entry from the table.
465  */
466 STATIC void
467 xfs_dir2_data_freeremove(
468 	struct xfs_dir2_data_hdr *hdr,		/* data block header */
469 	struct xfs_dir2_data_free *bf,		/* bestfree table pointer */
470 	struct xfs_dir2_data_free *dfp,		/* bestfree entry pointer */
471 	int			*loghead)	/* out: log data header */
472 {
473 
474 	ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
475 	       hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
476 	       hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
477 	       hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
478 
479 	/*
480 	 * It's the first entry, slide the next 2 up.
481 	 */
482 	if (dfp == &bf[0]) {
483 		bf[0] = bf[1];
484 		bf[1] = bf[2];
485 	}
486 	/*
487 	 * It's the second entry, slide the 3rd entry up.
488 	 */
489 	else if (dfp == &bf[1])
490 		bf[1] = bf[2];
491 	/*
492 	 * Must be the last entry.
493 	 */
494 	else
495 		ASSERT(dfp == &bf[2]);
496 	/*
497 	 * Clear the 3rd entry, must be zero now.
498 	 */
499 	bf[2].length = 0;
500 	bf[2].offset = 0;
501 	*loghead = 1;
502 }
503 
504 /*
505  * Given a data block, reconstruct its bestfree map.
506  */
507 void
508 xfs_dir2_data_freescan(
509 	struct xfs_inode	*dp,
510 	struct xfs_dir2_data_hdr *hdr,
511 	int			*loghead)
512 {
513 	xfs_dir2_block_tail_t	*btp;		/* block tail */
514 	xfs_dir2_data_entry_t	*dep;		/* active data entry */
515 	xfs_dir2_data_unused_t	*dup;		/* unused data entry */
516 	struct xfs_dir2_data_free *bf;
517 	char			*endp;		/* end of block's data */
518 	char			*p;		/* current entry pointer */
519 	struct xfs_da_geometry	*geo = dp->i_mount->m_dir_geo;
520 
521 	ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
522 	       hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
523 	       hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
524 	       hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
525 
526 	/*
527 	 * Start by clearing the table.
528 	 */
529 	bf = dp->d_ops->data_bestfree_p(hdr);
530 	memset(bf, 0, sizeof(*bf) * XFS_DIR2_DATA_FD_COUNT);
531 	*loghead = 1;
532 	/*
533 	 * Set up pointers.
534 	 */
535 	p = (char *)dp->d_ops->data_entry_p(hdr);
536 	if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
537 	    hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC)) {
538 		btp = xfs_dir2_block_tail_p(geo, hdr);
539 		endp = (char *)xfs_dir2_block_leaf_p(btp);
540 	} else
541 		endp = (char *)hdr + geo->blksize;
542 	/*
543 	 * Loop over the block's entries.
544 	 */
545 	while (p < endp) {
546 		dup = (xfs_dir2_data_unused_t *)p;
547 		/*
548 		 * If it's a free entry, insert it.
549 		 */
550 		if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
551 			ASSERT((char *)dup - (char *)hdr ==
552 			       be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup)));
553 			xfs_dir2_data_freeinsert(hdr, bf, dup, loghead);
554 			p += be16_to_cpu(dup->length);
555 		}
556 		/*
557 		 * For active entries, check their tags and skip them.
558 		 */
559 		else {
560 			dep = (xfs_dir2_data_entry_t *)p;
561 			ASSERT((char *)dep - (char *)hdr ==
562 			       be16_to_cpu(*dp->d_ops->data_entry_tag_p(dep)));
563 			p += dp->d_ops->data_entsize(dep->namelen);
564 		}
565 	}
566 }
567 
568 /*
569  * Initialize a data block at the given block number in the directory.
570  * Give back the buffer for the created block.
571  */
572 int						/* error */
573 xfs_dir3_data_init(
574 	xfs_da_args_t		*args,		/* directory operation args */
575 	xfs_dir2_db_t		blkno,		/* logical dir block number */
576 	struct xfs_buf		**bpp)		/* output block buffer */
577 {
578 	struct xfs_buf		*bp;		/* block buffer */
579 	xfs_dir2_data_hdr_t	*hdr;		/* data block header */
580 	xfs_inode_t		*dp;		/* incore directory inode */
581 	xfs_dir2_data_unused_t	*dup;		/* unused entry pointer */
582 	struct xfs_dir2_data_free *bf;
583 	int			error;		/* error return value */
584 	int			i;		/* bestfree index */
585 	xfs_mount_t		*mp;		/* filesystem mount point */
586 	xfs_trans_t		*tp;		/* transaction pointer */
587 	int                     t;              /* temp */
588 
589 	dp = args->dp;
590 	mp = dp->i_mount;
591 	tp = args->trans;
592 	/*
593 	 * Get the buffer set up for the block.
594 	 */
595 	error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(args->geo, blkno),
596 			       -1, &bp, XFS_DATA_FORK);
597 	if (error)
598 		return error;
599 	bp->b_ops = &xfs_dir3_data_buf_ops;
600 	xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_DATA_BUF);
601 
602 	/*
603 	 * Initialize the header.
604 	 */
605 	hdr = bp->b_addr;
606 	if (xfs_sb_version_hascrc(&mp->m_sb)) {
607 		struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
608 
609 		memset(hdr3, 0, sizeof(*hdr3));
610 		hdr3->magic = cpu_to_be32(XFS_DIR3_DATA_MAGIC);
611 		hdr3->blkno = cpu_to_be64(bp->b_bn);
612 		hdr3->owner = cpu_to_be64(dp->i_ino);
613 		uuid_copy(&hdr3->uuid, &mp->m_sb.sb_meta_uuid);
614 
615 	} else
616 		hdr->magic = cpu_to_be32(XFS_DIR2_DATA_MAGIC);
617 
618 	bf = dp->d_ops->data_bestfree_p(hdr);
619 	bf[0].offset = cpu_to_be16(dp->d_ops->data_entry_offset);
620 	for (i = 1; i < XFS_DIR2_DATA_FD_COUNT; i++) {
621 		bf[i].length = 0;
622 		bf[i].offset = 0;
623 	}
624 
625 	/*
626 	 * Set up an unused entry for the block's body.
627 	 */
628 	dup = dp->d_ops->data_unused_p(hdr);
629 	dup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
630 
631 	t = args->geo->blksize - (uint)dp->d_ops->data_entry_offset;
632 	bf[0].length = cpu_to_be16(t);
633 	dup->length = cpu_to_be16(t);
634 	*xfs_dir2_data_unused_tag_p(dup) = cpu_to_be16((char *)dup - (char *)hdr);
635 	/*
636 	 * Log it and return it.
637 	 */
638 	xfs_dir2_data_log_header(args, bp);
639 	xfs_dir2_data_log_unused(args, bp, dup);
640 	*bpp = bp;
641 	return 0;
642 }
643 
644 /*
645  * Log an active data entry from the block.
646  */
647 void
648 xfs_dir2_data_log_entry(
649 	struct xfs_da_args	*args,
650 	struct xfs_buf		*bp,
651 	xfs_dir2_data_entry_t	*dep)		/* data entry pointer */
652 {
653 	struct xfs_dir2_data_hdr *hdr = bp->b_addr;
654 
655 	ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
656 	       hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
657 	       hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
658 	       hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
659 
660 	xfs_trans_log_buf(args->trans, bp, (uint)((char *)dep - (char *)hdr),
661 		(uint)((char *)(args->dp->d_ops->data_entry_tag_p(dep) + 1) -
662 		       (char *)hdr - 1));
663 }
664 
665 /*
666  * Log a data block header.
667  */
668 void
669 xfs_dir2_data_log_header(
670 	struct xfs_da_args	*args,
671 	struct xfs_buf		*bp)
672 {
673 #ifdef DEBUG
674 	struct xfs_dir2_data_hdr *hdr = bp->b_addr;
675 
676 	ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
677 	       hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
678 	       hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
679 	       hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
680 #endif
681 
682 	xfs_trans_log_buf(args->trans, bp, 0,
683 			  args->dp->d_ops->data_entry_offset - 1);
684 }
685 
686 /*
687  * Log a data unused entry.
688  */
689 void
690 xfs_dir2_data_log_unused(
691 	struct xfs_da_args	*args,
692 	struct xfs_buf		*bp,
693 	xfs_dir2_data_unused_t	*dup)		/* data unused pointer */
694 {
695 	xfs_dir2_data_hdr_t	*hdr = bp->b_addr;
696 
697 	ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
698 	       hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
699 	       hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
700 	       hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
701 
702 	/*
703 	 * Log the first part of the unused entry.
704 	 */
705 	xfs_trans_log_buf(args->trans, bp, (uint)((char *)dup - (char *)hdr),
706 		(uint)((char *)&dup->length + sizeof(dup->length) -
707 		       1 - (char *)hdr));
708 	/*
709 	 * Log the end (tag) of the unused entry.
710 	 */
711 	xfs_trans_log_buf(args->trans, bp,
712 		(uint)((char *)xfs_dir2_data_unused_tag_p(dup) - (char *)hdr),
713 		(uint)((char *)xfs_dir2_data_unused_tag_p(dup) - (char *)hdr +
714 		       sizeof(xfs_dir2_data_off_t) - 1));
715 }
716 
717 /*
718  * Make a byte range in the data block unused.
719  * Its current contents are unimportant.
720  */
721 void
722 xfs_dir2_data_make_free(
723 	struct xfs_da_args	*args,
724 	struct xfs_buf		*bp,
725 	xfs_dir2_data_aoff_t	offset,		/* starting byte offset */
726 	xfs_dir2_data_aoff_t	len,		/* length in bytes */
727 	int			*needlogp,	/* out: log header */
728 	int			*needscanp)	/* out: regen bestfree */
729 {
730 	xfs_dir2_data_hdr_t	*hdr;		/* data block pointer */
731 	xfs_dir2_data_free_t	*dfp;		/* bestfree pointer */
732 	char			*endptr;	/* end of data area */
733 	int			needscan;	/* need to regen bestfree */
734 	xfs_dir2_data_unused_t	*newdup;	/* new unused entry */
735 	xfs_dir2_data_unused_t	*postdup;	/* unused entry after us */
736 	xfs_dir2_data_unused_t	*prevdup;	/* unused entry before us */
737 	struct xfs_dir2_data_free *bf;
738 
739 	hdr = bp->b_addr;
740 
741 	/*
742 	 * Figure out where the end of the data area is.
743 	 */
744 	if (hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
745 	    hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC))
746 		endptr = (char *)hdr + args->geo->blksize;
747 	else {
748 		xfs_dir2_block_tail_t	*btp;	/* block tail */
749 
750 		ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
751 			hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
752 		btp = xfs_dir2_block_tail_p(args->geo, hdr);
753 		endptr = (char *)xfs_dir2_block_leaf_p(btp);
754 	}
755 	/*
756 	 * If this isn't the start of the block, then back up to
757 	 * the previous entry and see if it's free.
758 	 */
759 	if (offset > args->dp->d_ops->data_entry_offset) {
760 		__be16			*tagp;	/* tag just before us */
761 
762 		tagp = (__be16 *)((char *)hdr + offset) - 1;
763 		prevdup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
764 		if (be16_to_cpu(prevdup->freetag) != XFS_DIR2_DATA_FREE_TAG)
765 			prevdup = NULL;
766 	} else
767 		prevdup = NULL;
768 	/*
769 	 * If this isn't the end of the block, see if the entry after
770 	 * us is free.
771 	 */
772 	if ((char *)hdr + offset + len < endptr) {
773 		postdup =
774 			(xfs_dir2_data_unused_t *)((char *)hdr + offset + len);
775 		if (be16_to_cpu(postdup->freetag) != XFS_DIR2_DATA_FREE_TAG)
776 			postdup = NULL;
777 	} else
778 		postdup = NULL;
779 	ASSERT(*needscanp == 0);
780 	needscan = 0;
781 	/*
782 	 * Previous and following entries are both free,
783 	 * merge everything into a single free entry.
784 	 */
785 	bf = args->dp->d_ops->data_bestfree_p(hdr);
786 	if (prevdup && postdup) {
787 		xfs_dir2_data_free_t	*dfp2;	/* another bestfree pointer */
788 
789 		/*
790 		 * See if prevdup and/or postdup are in bestfree table.
791 		 */
792 		dfp = xfs_dir2_data_freefind(hdr, bf, prevdup);
793 		dfp2 = xfs_dir2_data_freefind(hdr, bf, postdup);
794 		/*
795 		 * We need a rescan unless there are exactly 2 free entries
796 		 * namely our two.  Then we know what's happening, otherwise
797 		 * since the third bestfree is there, there might be more
798 		 * entries.
799 		 */
800 		needscan = (bf[2].length != 0);
801 		/*
802 		 * Fix up the new big freespace.
803 		 */
804 		be16_add_cpu(&prevdup->length, len + be16_to_cpu(postdup->length));
805 		*xfs_dir2_data_unused_tag_p(prevdup) =
806 			cpu_to_be16((char *)prevdup - (char *)hdr);
807 		xfs_dir2_data_log_unused(args, bp, prevdup);
808 		if (!needscan) {
809 			/*
810 			 * Has to be the case that entries 0 and 1 are
811 			 * dfp and dfp2 (don't know which is which), and
812 			 * entry 2 is empty.
813 			 * Remove entry 1 first then entry 0.
814 			 */
815 			ASSERT(dfp && dfp2);
816 			if (dfp == &bf[1]) {
817 				dfp = &bf[0];
818 				ASSERT(dfp2 == dfp);
819 				dfp2 = &bf[1];
820 			}
821 			xfs_dir2_data_freeremove(hdr, bf, dfp2, needlogp);
822 			xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
823 			/*
824 			 * Now insert the new entry.
825 			 */
826 			dfp = xfs_dir2_data_freeinsert(hdr, bf, prevdup,
827 						       needlogp);
828 			ASSERT(dfp == &bf[0]);
829 			ASSERT(dfp->length == prevdup->length);
830 			ASSERT(!dfp[1].length);
831 			ASSERT(!dfp[2].length);
832 		}
833 	}
834 	/*
835 	 * The entry before us is free, merge with it.
836 	 */
837 	else if (prevdup) {
838 		dfp = xfs_dir2_data_freefind(hdr, bf, prevdup);
839 		be16_add_cpu(&prevdup->length, len);
840 		*xfs_dir2_data_unused_tag_p(prevdup) =
841 			cpu_to_be16((char *)prevdup - (char *)hdr);
842 		xfs_dir2_data_log_unused(args, bp, prevdup);
843 		/*
844 		 * If the previous entry was in the table, the new entry
845 		 * is longer, so it will be in the table too.  Remove
846 		 * the old one and add the new one.
847 		 */
848 		if (dfp) {
849 			xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
850 			xfs_dir2_data_freeinsert(hdr, bf, prevdup, needlogp);
851 		}
852 		/*
853 		 * Otherwise we need a scan if the new entry is big enough.
854 		 */
855 		else {
856 			needscan = be16_to_cpu(prevdup->length) >
857 				   be16_to_cpu(bf[2].length);
858 		}
859 	}
860 	/*
861 	 * The following entry is free, merge with it.
862 	 */
863 	else if (postdup) {
864 		dfp = xfs_dir2_data_freefind(hdr, bf, postdup);
865 		newdup = (xfs_dir2_data_unused_t *)((char *)hdr + offset);
866 		newdup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
867 		newdup->length = cpu_to_be16(len + be16_to_cpu(postdup->length));
868 		*xfs_dir2_data_unused_tag_p(newdup) =
869 			cpu_to_be16((char *)newdup - (char *)hdr);
870 		xfs_dir2_data_log_unused(args, bp, newdup);
871 		/*
872 		 * If the following entry was in the table, the new entry
873 		 * is longer, so it will be in the table too.  Remove
874 		 * the old one and add the new one.
875 		 */
876 		if (dfp) {
877 			xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
878 			xfs_dir2_data_freeinsert(hdr, bf, newdup, needlogp);
879 		}
880 		/*
881 		 * Otherwise we need a scan if the new entry is big enough.
882 		 */
883 		else {
884 			needscan = be16_to_cpu(newdup->length) >
885 				   be16_to_cpu(bf[2].length);
886 		}
887 	}
888 	/*
889 	 * Neither neighbor is free.  Make a new entry.
890 	 */
891 	else {
892 		newdup = (xfs_dir2_data_unused_t *)((char *)hdr + offset);
893 		newdup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
894 		newdup->length = cpu_to_be16(len);
895 		*xfs_dir2_data_unused_tag_p(newdup) =
896 			cpu_to_be16((char *)newdup - (char *)hdr);
897 		xfs_dir2_data_log_unused(args, bp, newdup);
898 		xfs_dir2_data_freeinsert(hdr, bf, newdup, needlogp);
899 	}
900 	*needscanp = needscan;
901 }
902 
903 /*
904  * Take a byte range out of an existing unused space and make it un-free.
905  */
906 void
907 xfs_dir2_data_use_free(
908 	struct xfs_da_args	*args,
909 	struct xfs_buf		*bp,
910 	xfs_dir2_data_unused_t	*dup,		/* unused entry */
911 	xfs_dir2_data_aoff_t	offset,		/* starting offset to use */
912 	xfs_dir2_data_aoff_t	len,		/* length to use */
913 	int			*needlogp,	/* out: need to log header */
914 	int			*needscanp)	/* out: need regen bestfree */
915 {
916 	xfs_dir2_data_hdr_t	*hdr;		/* data block header */
917 	xfs_dir2_data_free_t	*dfp;		/* bestfree pointer */
918 	int			matchback;	/* matches end of freespace */
919 	int			matchfront;	/* matches start of freespace */
920 	int			needscan;	/* need to regen bestfree */
921 	xfs_dir2_data_unused_t	*newdup;	/* new unused entry */
922 	xfs_dir2_data_unused_t	*newdup2;	/* another new unused entry */
923 	int			oldlen;		/* old unused entry's length */
924 	struct xfs_dir2_data_free *bf;
925 
926 	hdr = bp->b_addr;
927 	ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
928 	       hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
929 	       hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
930 	       hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
931 	ASSERT(be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG);
932 	ASSERT(offset >= (char *)dup - (char *)hdr);
933 	ASSERT(offset + len <= (char *)dup + be16_to_cpu(dup->length) - (char *)hdr);
934 	ASSERT((char *)dup - (char *)hdr == be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup)));
935 	/*
936 	 * Look up the entry in the bestfree table.
937 	 */
938 	oldlen = be16_to_cpu(dup->length);
939 	bf = args->dp->d_ops->data_bestfree_p(hdr);
940 	dfp = xfs_dir2_data_freefind(hdr, bf, dup);
941 	ASSERT(dfp || oldlen <= be16_to_cpu(bf[2].length));
942 	/*
943 	 * Check for alignment with front and back of the entry.
944 	 */
945 	matchfront = (char *)dup - (char *)hdr == offset;
946 	matchback = (char *)dup + oldlen - (char *)hdr == offset + len;
947 	ASSERT(*needscanp == 0);
948 	needscan = 0;
949 	/*
950 	 * If we matched it exactly we just need to get rid of it from
951 	 * the bestfree table.
952 	 */
953 	if (matchfront && matchback) {
954 		if (dfp) {
955 			needscan = (bf[2].offset != 0);
956 			if (!needscan)
957 				xfs_dir2_data_freeremove(hdr, bf, dfp,
958 							 needlogp);
959 		}
960 	}
961 	/*
962 	 * We match the first part of the entry.
963 	 * Make a new entry with the remaining freespace.
964 	 */
965 	else if (matchfront) {
966 		newdup = (xfs_dir2_data_unused_t *)((char *)hdr + offset + len);
967 		newdup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
968 		newdup->length = cpu_to_be16(oldlen - len);
969 		*xfs_dir2_data_unused_tag_p(newdup) =
970 			cpu_to_be16((char *)newdup - (char *)hdr);
971 		xfs_dir2_data_log_unused(args, bp, newdup);
972 		/*
973 		 * If it was in the table, remove it and add the new one.
974 		 */
975 		if (dfp) {
976 			xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
977 			dfp = xfs_dir2_data_freeinsert(hdr, bf, newdup,
978 						       needlogp);
979 			ASSERT(dfp != NULL);
980 			ASSERT(dfp->length == newdup->length);
981 			ASSERT(be16_to_cpu(dfp->offset) == (char *)newdup - (char *)hdr);
982 			/*
983 			 * If we got inserted at the last slot,
984 			 * that means we don't know if there was a better
985 			 * choice for the last slot, or not.  Rescan.
986 			 */
987 			needscan = dfp == &bf[2];
988 		}
989 	}
990 	/*
991 	 * We match the last part of the entry.
992 	 * Trim the allocated space off the tail of the entry.
993 	 */
994 	else if (matchback) {
995 		newdup = dup;
996 		newdup->length = cpu_to_be16(((char *)hdr + offset) - (char *)newdup);
997 		*xfs_dir2_data_unused_tag_p(newdup) =
998 			cpu_to_be16((char *)newdup - (char *)hdr);
999 		xfs_dir2_data_log_unused(args, bp, newdup);
1000 		/*
1001 		 * If it was in the table, remove it and add the new one.
1002 		 */
1003 		if (dfp) {
1004 			xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
1005 			dfp = xfs_dir2_data_freeinsert(hdr, bf, newdup,
1006 						       needlogp);
1007 			ASSERT(dfp != NULL);
1008 			ASSERT(dfp->length == newdup->length);
1009 			ASSERT(be16_to_cpu(dfp->offset) == (char *)newdup - (char *)hdr);
1010 			/*
1011 			 * If we got inserted at the last slot,
1012 			 * that means we don't know if there was a better
1013 			 * choice for the last slot, or not.  Rescan.
1014 			 */
1015 			needscan = dfp == &bf[2];
1016 		}
1017 	}
1018 	/*
1019 	 * Poking out the middle of an entry.
1020 	 * Make two new entries.
1021 	 */
1022 	else {
1023 		newdup = dup;
1024 		newdup->length = cpu_to_be16(((char *)hdr + offset) - (char *)newdup);
1025 		*xfs_dir2_data_unused_tag_p(newdup) =
1026 			cpu_to_be16((char *)newdup - (char *)hdr);
1027 		xfs_dir2_data_log_unused(args, bp, newdup);
1028 		newdup2 = (xfs_dir2_data_unused_t *)((char *)hdr + offset + len);
1029 		newdup2->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
1030 		newdup2->length = cpu_to_be16(oldlen - len - be16_to_cpu(newdup->length));
1031 		*xfs_dir2_data_unused_tag_p(newdup2) =
1032 			cpu_to_be16((char *)newdup2 - (char *)hdr);
1033 		xfs_dir2_data_log_unused(args, bp, newdup2);
1034 		/*
1035 		 * If the old entry was in the table, we need to scan
1036 		 * if the 3rd entry was valid, since these entries
1037 		 * are smaller than the old one.
1038 		 * If we don't need to scan that means there were 1 or 2
1039 		 * entries in the table, and removing the old and adding
1040 		 * the 2 new will work.
1041 		 */
1042 		if (dfp) {
1043 			needscan = (bf[2].length != 0);
1044 			if (!needscan) {
1045 				xfs_dir2_data_freeremove(hdr, bf, dfp,
1046 							 needlogp);
1047 				xfs_dir2_data_freeinsert(hdr, bf, newdup,
1048 							 needlogp);
1049 				xfs_dir2_data_freeinsert(hdr, bf, newdup2,
1050 							 needlogp);
1051 			}
1052 		}
1053 	}
1054 	*needscanp = needscan;
1055 }
1056