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_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_dir2.h"
31 #include "xfs_dir2_priv.h"
32 #include "xfs_error.h"
33 #include "xfs_trans.h"
34 #include "xfs_buf_item.h"
35 #include "xfs_cksum.h"
36
37 /*
38 * Check the consistency of the data block.
39 * The input can also be a block-format directory.
40 * Return 0 is the buffer is good, otherwise an error.
41 */
42 int
__xfs_dir3_data_check(struct xfs_inode * dp,struct xfs_buf * bp)43 __xfs_dir3_data_check(
44 struct xfs_inode *dp, /* incore inode pointer */
45 struct xfs_buf *bp) /* data block's buffer */
46 {
47 xfs_dir2_dataptr_t addr; /* addr for leaf lookup */
48 xfs_dir2_data_free_t *bf; /* bestfree table */
49 xfs_dir2_block_tail_t *btp=NULL; /* block tail */
50 int count; /* count of entries found */
51 xfs_dir2_data_hdr_t *hdr; /* data block header */
52 xfs_dir2_data_entry_t *dep; /* data entry */
53 xfs_dir2_data_free_t *dfp; /* bestfree entry */
54 xfs_dir2_data_unused_t *dup; /* unused entry */
55 char *endp; /* end of useful data */
56 int freeseen; /* mask of bestfrees seen */
57 xfs_dahash_t hash; /* hash of current name */
58 int i; /* leaf index */
59 int lastfree; /* last entry was unused */
60 xfs_dir2_leaf_entry_t *lep=NULL; /* block leaf entries */
61 xfs_mount_t *mp; /* filesystem mount point */
62 char *p; /* current data position */
63 int stale; /* count of stale leaves */
64 struct xfs_name name;
65 const struct xfs_dir_ops *ops;
66 struct xfs_da_geometry *geo;
67
68 mp = bp->b_target->bt_mount;
69 geo = mp->m_dir_geo;
70
71 /*
72 * We can be passed a null dp here from a verifier, so we need to go the
73 * hard way to get them.
74 */
75 ops = xfs_dir_get_ops(mp, dp);
76
77 hdr = bp->b_addr;
78 p = (char *)ops->data_entry_p(hdr);
79
80 switch (hdr->magic) {
81 case cpu_to_be32(XFS_DIR3_BLOCK_MAGIC):
82 case cpu_to_be32(XFS_DIR2_BLOCK_MAGIC):
83 btp = xfs_dir2_block_tail_p(geo, hdr);
84 lep = xfs_dir2_block_leaf_p(btp);
85 endp = (char *)lep;
86
87 /*
88 * The number of leaf entries is limited by the size of the
89 * block and the amount of space used by the data entries.
90 * We don't know how much space is used by the data entries yet,
91 * so just ensure that the count falls somewhere inside the
92 * block right now.
93 */
94 XFS_WANT_CORRUPTED_RETURN(be32_to_cpu(btp->count) <
95 ((char *)btp - p) / sizeof(struct xfs_dir2_leaf_entry));
96 break;
97 case cpu_to_be32(XFS_DIR3_DATA_MAGIC):
98 case cpu_to_be32(XFS_DIR2_DATA_MAGIC):
99 endp = (char *)hdr + geo->blksize;
100 break;
101 default:
102 XFS_ERROR_REPORT("Bad Magic", XFS_ERRLEVEL_LOW, mp);
103 return -EFSCORRUPTED;
104 }
105
106 /*
107 * Account for zero bestfree entries.
108 */
109 bf = ops->data_bestfree_p(hdr);
110 count = lastfree = freeseen = 0;
111 if (!bf[0].length) {
112 XFS_WANT_CORRUPTED_RETURN(!bf[0].offset);
113 freeseen |= 1 << 0;
114 }
115 if (!bf[1].length) {
116 XFS_WANT_CORRUPTED_RETURN(!bf[1].offset);
117 freeseen |= 1 << 1;
118 }
119 if (!bf[2].length) {
120 XFS_WANT_CORRUPTED_RETURN(!bf[2].offset);
121 freeseen |= 1 << 2;
122 }
123
124 XFS_WANT_CORRUPTED_RETURN(be16_to_cpu(bf[0].length) >=
125 be16_to_cpu(bf[1].length));
126 XFS_WANT_CORRUPTED_RETURN(be16_to_cpu(bf[1].length) >=
127 be16_to_cpu(bf[2].length));
128 /*
129 * Loop over the data/unused entries.
130 */
131 while (p < endp) {
132 dup = (xfs_dir2_data_unused_t *)p;
133 /*
134 * If it's unused, look for the space in the bestfree table.
135 * If we find it, account for that, else make sure it
136 * doesn't need to be there.
137 */
138 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
139 XFS_WANT_CORRUPTED_RETURN(lastfree == 0);
140 XFS_WANT_CORRUPTED_RETURN(
141 be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup)) ==
142 (char *)dup - (char *)hdr);
143 dfp = xfs_dir2_data_freefind(hdr, bf, dup);
144 if (dfp) {
145 i = (int)(dfp - bf);
146 XFS_WANT_CORRUPTED_RETURN(
147 (freeseen & (1 << i)) == 0);
148 freeseen |= 1 << i;
149 } else {
150 XFS_WANT_CORRUPTED_RETURN(
151 be16_to_cpu(dup->length) <=
152 be16_to_cpu(bf[2].length));
153 }
154 p += be16_to_cpu(dup->length);
155 lastfree = 1;
156 continue;
157 }
158 /*
159 * It's a real entry. Validate the fields.
160 * If this is a block directory then make sure it's
161 * in the leaf section of the block.
162 * The linear search is crude but this is DEBUG code.
163 */
164 dep = (xfs_dir2_data_entry_t *)p;
165 XFS_WANT_CORRUPTED_RETURN(dep->namelen != 0);
166 XFS_WANT_CORRUPTED_RETURN(
167 !xfs_dir_ino_validate(mp, be64_to_cpu(dep->inumber)));
168 XFS_WANT_CORRUPTED_RETURN(
169 be16_to_cpu(*ops->data_entry_tag_p(dep)) ==
170 (char *)dep - (char *)hdr);
171 XFS_WANT_CORRUPTED_RETURN(
172 ops->data_get_ftype(dep) < XFS_DIR3_FT_MAX);
173 count++;
174 lastfree = 0;
175 if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
176 hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC)) {
177 addr = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
178 (xfs_dir2_data_aoff_t)
179 ((char *)dep - (char *)hdr));
180 name.name = dep->name;
181 name.len = dep->namelen;
182 hash = mp->m_dirnameops->hashname(&name);
183 for (i = 0; i < be32_to_cpu(btp->count); i++) {
184 if (be32_to_cpu(lep[i].address) == addr &&
185 be32_to_cpu(lep[i].hashval) == hash)
186 break;
187 }
188 XFS_WANT_CORRUPTED_RETURN(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(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(
204 be32_to_cpu(lep[i].hashval) >=
205 be32_to_cpu(lep[i - 1].hashval));
206 }
207 XFS_WANT_CORRUPTED_RETURN(count ==
208 be32_to_cpu(btp->count) - be32_to_cpu(btp->stale));
209 XFS_WANT_CORRUPTED_RETURN(stale == be32_to_cpu(btp->stale));
210 }
211 return 0;
212 }
213
214 static bool
xfs_dir3_data_verify(struct xfs_buf * bp)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_uuid))
225 return false;
226 if (be64_to_cpu(hdr3->blkno) != bp->b_bn)
227 return false;
228 } else {
229 if (hdr3->magic != cpu_to_be32(XFS_DIR2_DATA_MAGIC))
230 return false;
231 }
232 if (__xfs_dir3_data_check(NULL, bp))
233 return false;
234 return true;
235 }
236
237 /*
238 * Readahead of the first block of the directory when it is opened is completely
239 * oblivious to the format of the directory. Hence we can either get a block
240 * format buffer or a data format buffer on readahead.
241 */
242 static void
xfs_dir3_data_reada_verify(struct xfs_buf * bp)243 xfs_dir3_data_reada_verify(
244 struct xfs_buf *bp)
245 {
246 struct xfs_dir2_data_hdr *hdr = bp->b_addr;
247
248 switch (hdr->magic) {
249 case cpu_to_be32(XFS_DIR2_BLOCK_MAGIC):
250 case cpu_to_be32(XFS_DIR3_BLOCK_MAGIC):
251 bp->b_ops = &xfs_dir3_block_buf_ops;
252 bp->b_ops->verify_read(bp);
253 return;
254 case cpu_to_be32(XFS_DIR2_DATA_MAGIC):
255 case cpu_to_be32(XFS_DIR3_DATA_MAGIC):
256 xfs_dir3_data_verify(bp);
257 return;
258 default:
259 xfs_buf_ioerror(bp, -EFSCORRUPTED);
260 xfs_verifier_error(bp);
261 break;
262 }
263 }
264
265 static void
xfs_dir3_data_read_verify(struct xfs_buf * bp)266 xfs_dir3_data_read_verify(
267 struct xfs_buf *bp)
268 {
269 struct xfs_mount *mp = bp->b_target->bt_mount;
270
271 if (xfs_sb_version_hascrc(&mp->m_sb) &&
272 !xfs_buf_verify_cksum(bp, XFS_DIR3_DATA_CRC_OFF))
273 xfs_buf_ioerror(bp, -EFSBADCRC);
274 else if (!xfs_dir3_data_verify(bp))
275 xfs_buf_ioerror(bp, -EFSCORRUPTED);
276
277 if (bp->b_error)
278 xfs_verifier_error(bp);
279 }
280
281 static void
xfs_dir3_data_write_verify(struct xfs_buf * bp)282 xfs_dir3_data_write_verify(
283 struct xfs_buf *bp)
284 {
285 struct xfs_mount *mp = bp->b_target->bt_mount;
286 struct xfs_buf_log_item *bip = bp->b_fspriv;
287 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
288
289 if (!xfs_dir3_data_verify(bp)) {
290 xfs_buf_ioerror(bp, -EFSCORRUPTED);
291 xfs_verifier_error(bp);
292 return;
293 }
294
295 if (!xfs_sb_version_hascrc(&mp->m_sb))
296 return;
297
298 if (bip)
299 hdr3->lsn = cpu_to_be64(bip->bli_item.li_lsn);
300
301 xfs_buf_update_cksum(bp, XFS_DIR3_DATA_CRC_OFF);
302 }
303
304 const struct xfs_buf_ops xfs_dir3_data_buf_ops = {
305 .name = "xfs_dir3_data",
306 .verify_read = xfs_dir3_data_read_verify,
307 .verify_write = xfs_dir3_data_write_verify,
308 };
309
310 static const struct xfs_buf_ops xfs_dir3_data_reada_buf_ops = {
311 .name = "xfs_dir3_data_reada",
312 .verify_read = xfs_dir3_data_reada_verify,
313 .verify_write = xfs_dir3_data_write_verify,
314 };
315
316
317 int
xfs_dir3_data_read(struct xfs_trans * tp,struct xfs_inode * dp,xfs_dablk_t bno,xfs_daddr_t mapped_bno,struct xfs_buf ** bpp)318 xfs_dir3_data_read(
319 struct xfs_trans *tp,
320 struct xfs_inode *dp,
321 xfs_dablk_t bno,
322 xfs_daddr_t mapped_bno,
323 struct xfs_buf **bpp)
324 {
325 int err;
326
327 err = xfs_da_read_buf(tp, dp, bno, mapped_bno, bpp,
328 XFS_DATA_FORK, &xfs_dir3_data_buf_ops);
329 if (!err && tp)
330 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_DATA_BUF);
331 return err;
332 }
333
334 int
xfs_dir3_data_readahead(struct xfs_inode * dp,xfs_dablk_t bno,xfs_daddr_t mapped_bno)335 xfs_dir3_data_readahead(
336 struct xfs_inode *dp,
337 xfs_dablk_t bno,
338 xfs_daddr_t mapped_bno)
339 {
340 return xfs_da_reada_buf(dp, bno, mapped_bno,
341 XFS_DATA_FORK, &xfs_dir3_data_reada_buf_ops);
342 }
343
344 /*
345 * Given a data block and an unused entry from that block,
346 * return the bestfree entry if any that corresponds to it.
347 */
348 xfs_dir2_data_free_t *
xfs_dir2_data_freefind(struct xfs_dir2_data_hdr * hdr,struct xfs_dir2_data_free * bf,struct xfs_dir2_data_unused * dup)349 xfs_dir2_data_freefind(
350 struct xfs_dir2_data_hdr *hdr, /* data block header */
351 struct xfs_dir2_data_free *bf, /* bestfree table pointer */
352 struct xfs_dir2_data_unused *dup) /* unused space */
353 {
354 xfs_dir2_data_free_t *dfp; /* bestfree entry */
355 xfs_dir2_data_aoff_t off; /* offset value needed */
356 #ifdef DEBUG
357 int matched; /* matched the value */
358 int seenzero; /* saw a 0 bestfree entry */
359 #endif
360
361 off = (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr);
362
363 #ifdef DEBUG
364 /*
365 * Validate some consistency in the bestfree table.
366 * Check order, non-overlapping entries, and if we find the
367 * one we're looking for it has to be exact.
368 */
369 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
370 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
371 hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
372 hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
373 for (dfp = &bf[0], seenzero = matched = 0;
374 dfp < &bf[XFS_DIR2_DATA_FD_COUNT];
375 dfp++) {
376 if (!dfp->offset) {
377 ASSERT(!dfp->length);
378 seenzero = 1;
379 continue;
380 }
381 ASSERT(seenzero == 0);
382 if (be16_to_cpu(dfp->offset) == off) {
383 matched = 1;
384 ASSERT(dfp->length == dup->length);
385 } else if (off < be16_to_cpu(dfp->offset))
386 ASSERT(off + be16_to_cpu(dup->length) <= be16_to_cpu(dfp->offset));
387 else
388 ASSERT(be16_to_cpu(dfp->offset) + be16_to_cpu(dfp->length) <= off);
389 ASSERT(matched || be16_to_cpu(dfp->length) >= be16_to_cpu(dup->length));
390 if (dfp > &bf[0])
391 ASSERT(be16_to_cpu(dfp[-1].length) >= be16_to_cpu(dfp[0].length));
392 }
393 #endif
394 /*
395 * If this is smaller than the smallest bestfree entry,
396 * it can't be there since they're sorted.
397 */
398 if (be16_to_cpu(dup->length) <
399 be16_to_cpu(bf[XFS_DIR2_DATA_FD_COUNT - 1].length))
400 return NULL;
401 /*
402 * Look at the three bestfree entries for our guy.
403 */
404 for (dfp = &bf[0]; dfp < &bf[XFS_DIR2_DATA_FD_COUNT]; dfp++) {
405 if (!dfp->offset)
406 return NULL;
407 if (be16_to_cpu(dfp->offset) == off)
408 return dfp;
409 }
410 /*
411 * Didn't find it. This only happens if there are duplicate lengths.
412 */
413 return NULL;
414 }
415
416 /*
417 * Insert an unused-space entry into the bestfree table.
418 */
419 xfs_dir2_data_free_t * /* entry inserted */
xfs_dir2_data_freeinsert(struct xfs_dir2_data_hdr * hdr,struct xfs_dir2_data_free * dfp,struct xfs_dir2_data_unused * dup,int * loghead)420 xfs_dir2_data_freeinsert(
421 struct xfs_dir2_data_hdr *hdr, /* data block pointer */
422 struct xfs_dir2_data_free *dfp, /* bestfree table pointer */
423 struct xfs_dir2_data_unused *dup, /* unused space */
424 int *loghead) /* log the data header (out) */
425 {
426 xfs_dir2_data_free_t new; /* new bestfree entry */
427
428 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
429 hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
430 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
431 hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
432
433 new.length = dup->length;
434 new.offset = cpu_to_be16((char *)dup - (char *)hdr);
435
436 /*
437 * Insert at position 0, 1, or 2; or not at all.
438 */
439 if (be16_to_cpu(new.length) > be16_to_cpu(dfp[0].length)) {
440 dfp[2] = dfp[1];
441 dfp[1] = dfp[0];
442 dfp[0] = new;
443 *loghead = 1;
444 return &dfp[0];
445 }
446 if (be16_to_cpu(new.length) > be16_to_cpu(dfp[1].length)) {
447 dfp[2] = dfp[1];
448 dfp[1] = new;
449 *loghead = 1;
450 return &dfp[1];
451 }
452 if (be16_to_cpu(new.length) > be16_to_cpu(dfp[2].length)) {
453 dfp[2] = new;
454 *loghead = 1;
455 return &dfp[2];
456 }
457 return NULL;
458 }
459
460 /*
461 * Remove a bestfree entry from the table.
462 */
463 STATIC void
xfs_dir2_data_freeremove(struct xfs_dir2_data_hdr * hdr,struct xfs_dir2_data_free * bf,struct xfs_dir2_data_free * dfp,int * loghead)464 xfs_dir2_data_freeremove(
465 struct xfs_dir2_data_hdr *hdr, /* data block header */
466 struct xfs_dir2_data_free *bf, /* bestfree table pointer */
467 struct xfs_dir2_data_free *dfp, /* bestfree entry pointer */
468 int *loghead) /* out: log data header */
469 {
470
471 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
472 hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
473 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
474 hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
475
476 /*
477 * It's the first entry, slide the next 2 up.
478 */
479 if (dfp == &bf[0]) {
480 bf[0] = bf[1];
481 bf[1] = bf[2];
482 }
483 /*
484 * It's the second entry, slide the 3rd entry up.
485 */
486 else if (dfp == &bf[1])
487 bf[1] = bf[2];
488 /*
489 * Must be the last entry.
490 */
491 else
492 ASSERT(dfp == &bf[2]);
493 /*
494 * Clear the 3rd entry, must be zero now.
495 */
496 bf[2].length = 0;
497 bf[2].offset = 0;
498 *loghead = 1;
499 }
500
501 /*
502 * Given a data block, reconstruct its bestfree map.
503 */
504 void
xfs_dir2_data_freescan(struct xfs_inode * dp,struct xfs_dir2_data_hdr * hdr,int * loghead)505 xfs_dir2_data_freescan(
506 struct xfs_inode *dp,
507 struct xfs_dir2_data_hdr *hdr,
508 int *loghead)
509 {
510 xfs_dir2_block_tail_t *btp; /* block tail */
511 xfs_dir2_data_entry_t *dep; /* active data entry */
512 xfs_dir2_data_unused_t *dup; /* unused data entry */
513 struct xfs_dir2_data_free *bf;
514 char *endp; /* end of block's data */
515 char *p; /* current entry pointer */
516 struct xfs_da_geometry *geo = dp->i_mount->m_dir_geo;
517
518 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
519 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
520 hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
521 hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
522
523 /*
524 * Start by clearing the table.
525 */
526 bf = dp->d_ops->data_bestfree_p(hdr);
527 memset(bf, 0, sizeof(*bf) * XFS_DIR2_DATA_FD_COUNT);
528 *loghead = 1;
529 /*
530 * Set up pointers.
531 */
532 p = (char *)dp->d_ops->data_entry_p(hdr);
533 if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
534 hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC)) {
535 btp = xfs_dir2_block_tail_p(geo, hdr);
536 endp = (char *)xfs_dir2_block_leaf_p(btp);
537 } else
538 endp = (char *)hdr + geo->blksize;
539 /*
540 * Loop over the block's entries.
541 */
542 while (p < endp) {
543 dup = (xfs_dir2_data_unused_t *)p;
544 /*
545 * If it's a free entry, insert it.
546 */
547 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
548 ASSERT((char *)dup - (char *)hdr ==
549 be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup)));
550 xfs_dir2_data_freeinsert(hdr, bf, dup, loghead);
551 p += be16_to_cpu(dup->length);
552 }
553 /*
554 * For active entries, check their tags and skip them.
555 */
556 else {
557 dep = (xfs_dir2_data_entry_t *)p;
558 ASSERT((char *)dep - (char *)hdr ==
559 be16_to_cpu(*dp->d_ops->data_entry_tag_p(dep)));
560 p += dp->d_ops->data_entsize(dep->namelen);
561 }
562 }
563 }
564
565 /*
566 * Initialize a data block at the given block number in the directory.
567 * Give back the buffer for the created block.
568 */
569 int /* error */
xfs_dir3_data_init(xfs_da_args_t * args,xfs_dir2_db_t blkno,struct xfs_buf ** bpp)570 xfs_dir3_data_init(
571 xfs_da_args_t *args, /* directory operation args */
572 xfs_dir2_db_t blkno, /* logical dir block number */
573 struct xfs_buf **bpp) /* output block buffer */
574 {
575 struct xfs_buf *bp; /* block buffer */
576 xfs_dir2_data_hdr_t *hdr; /* data block header */
577 xfs_inode_t *dp; /* incore directory inode */
578 xfs_dir2_data_unused_t *dup; /* unused entry pointer */
579 struct xfs_dir2_data_free *bf;
580 int error; /* error return value */
581 int i; /* bestfree index */
582 xfs_mount_t *mp; /* filesystem mount point */
583 xfs_trans_t *tp; /* transaction pointer */
584 int t; /* temp */
585
586 dp = args->dp;
587 mp = dp->i_mount;
588 tp = args->trans;
589 /*
590 * Get the buffer set up for the block.
591 */
592 error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(args->geo, blkno),
593 -1, &bp, XFS_DATA_FORK);
594 if (error)
595 return error;
596 bp->b_ops = &xfs_dir3_data_buf_ops;
597 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_DATA_BUF);
598
599 /*
600 * Initialize the header.
601 */
602 hdr = bp->b_addr;
603 if (xfs_sb_version_hascrc(&mp->m_sb)) {
604 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
605
606 memset(hdr3, 0, sizeof(*hdr3));
607 hdr3->magic = cpu_to_be32(XFS_DIR3_DATA_MAGIC);
608 hdr3->blkno = cpu_to_be64(bp->b_bn);
609 hdr3->owner = cpu_to_be64(dp->i_ino);
610 uuid_copy(&hdr3->uuid, &mp->m_sb.sb_uuid);
611
612 } else
613 hdr->magic = cpu_to_be32(XFS_DIR2_DATA_MAGIC);
614
615 bf = dp->d_ops->data_bestfree_p(hdr);
616 bf[0].offset = cpu_to_be16(dp->d_ops->data_entry_offset);
617 for (i = 1; i < XFS_DIR2_DATA_FD_COUNT; i++) {
618 bf[i].length = 0;
619 bf[i].offset = 0;
620 }
621
622 /*
623 * Set up an unused entry for the block's body.
624 */
625 dup = dp->d_ops->data_unused_p(hdr);
626 dup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
627
628 t = args->geo->blksize - (uint)dp->d_ops->data_entry_offset;
629 bf[0].length = cpu_to_be16(t);
630 dup->length = cpu_to_be16(t);
631 *xfs_dir2_data_unused_tag_p(dup) = cpu_to_be16((char *)dup - (char *)hdr);
632 /*
633 * Log it and return it.
634 */
635 xfs_dir2_data_log_header(args, bp);
636 xfs_dir2_data_log_unused(args, bp, dup);
637 *bpp = bp;
638 return 0;
639 }
640
641 /*
642 * Log an active data entry from the block.
643 */
644 void
xfs_dir2_data_log_entry(struct xfs_da_args * args,struct xfs_buf * bp,xfs_dir2_data_entry_t * dep)645 xfs_dir2_data_log_entry(
646 struct xfs_da_args *args,
647 struct xfs_buf *bp,
648 xfs_dir2_data_entry_t *dep) /* data entry pointer */
649 {
650 struct xfs_dir2_data_hdr *hdr = bp->b_addr;
651
652 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
653 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
654 hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
655 hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
656
657 xfs_trans_log_buf(args->trans, bp, (uint)((char *)dep - (char *)hdr),
658 (uint)((char *)(args->dp->d_ops->data_entry_tag_p(dep) + 1) -
659 (char *)hdr - 1));
660 }
661
662 /*
663 * Log a data block header.
664 */
665 void
xfs_dir2_data_log_header(struct xfs_da_args * args,struct xfs_buf * bp)666 xfs_dir2_data_log_header(
667 struct xfs_da_args *args,
668 struct xfs_buf *bp)
669 {
670 #ifdef DEBUG
671 struct xfs_dir2_data_hdr *hdr = bp->b_addr;
672
673 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
674 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
675 hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
676 hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
677 #endif
678
679 xfs_trans_log_buf(args->trans, bp, 0,
680 args->dp->d_ops->data_entry_offset - 1);
681 }
682
683 /*
684 * Log a data unused entry.
685 */
686 void
xfs_dir2_data_log_unused(struct xfs_da_args * args,struct xfs_buf * bp,xfs_dir2_data_unused_t * dup)687 xfs_dir2_data_log_unused(
688 struct xfs_da_args *args,
689 struct xfs_buf *bp,
690 xfs_dir2_data_unused_t *dup) /* data unused pointer */
691 {
692 xfs_dir2_data_hdr_t *hdr = bp->b_addr;
693
694 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
695 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
696 hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
697 hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
698
699 /*
700 * Log the first part of the unused entry.
701 */
702 xfs_trans_log_buf(args->trans, bp, (uint)((char *)dup - (char *)hdr),
703 (uint)((char *)&dup->length + sizeof(dup->length) -
704 1 - (char *)hdr));
705 /*
706 * Log the end (tag) of the unused entry.
707 */
708 xfs_trans_log_buf(args->trans, bp,
709 (uint)((char *)xfs_dir2_data_unused_tag_p(dup) - (char *)hdr),
710 (uint)((char *)xfs_dir2_data_unused_tag_p(dup) - (char *)hdr +
711 sizeof(xfs_dir2_data_off_t) - 1));
712 }
713
714 /*
715 * Make a byte range in the data block unused.
716 * Its current contents are unimportant.
717 */
718 void
xfs_dir2_data_make_free(struct xfs_da_args * args,struct xfs_buf * bp,xfs_dir2_data_aoff_t offset,xfs_dir2_data_aoff_t len,int * needlogp,int * needscanp)719 xfs_dir2_data_make_free(
720 struct xfs_da_args *args,
721 struct xfs_buf *bp,
722 xfs_dir2_data_aoff_t offset, /* starting byte offset */
723 xfs_dir2_data_aoff_t len, /* length in bytes */
724 int *needlogp, /* out: log header */
725 int *needscanp) /* out: regen bestfree */
726 {
727 xfs_dir2_data_hdr_t *hdr; /* data block pointer */
728 xfs_dir2_data_free_t *dfp; /* bestfree pointer */
729 char *endptr; /* end of data area */
730 int needscan; /* need to regen bestfree */
731 xfs_dir2_data_unused_t *newdup; /* new unused entry */
732 xfs_dir2_data_unused_t *postdup; /* unused entry after us */
733 xfs_dir2_data_unused_t *prevdup; /* unused entry before us */
734 struct xfs_dir2_data_free *bf;
735
736 hdr = bp->b_addr;
737
738 /*
739 * Figure out where the end of the data area is.
740 */
741 if (hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
742 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC))
743 endptr = (char *)hdr + args->geo->blksize;
744 else {
745 xfs_dir2_block_tail_t *btp; /* block tail */
746
747 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
748 hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
749 btp = xfs_dir2_block_tail_p(args->geo, hdr);
750 endptr = (char *)xfs_dir2_block_leaf_p(btp);
751 }
752 /*
753 * If this isn't the start of the block, then back up to
754 * the previous entry and see if it's free.
755 */
756 if (offset > args->dp->d_ops->data_entry_offset) {
757 __be16 *tagp; /* tag just before us */
758
759 tagp = (__be16 *)((char *)hdr + offset) - 1;
760 prevdup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
761 if (be16_to_cpu(prevdup->freetag) != XFS_DIR2_DATA_FREE_TAG)
762 prevdup = NULL;
763 } else
764 prevdup = NULL;
765 /*
766 * If this isn't the end of the block, see if the entry after
767 * us is free.
768 */
769 if ((char *)hdr + offset + len < endptr) {
770 postdup =
771 (xfs_dir2_data_unused_t *)((char *)hdr + offset + len);
772 if (be16_to_cpu(postdup->freetag) != XFS_DIR2_DATA_FREE_TAG)
773 postdup = NULL;
774 } else
775 postdup = NULL;
776 ASSERT(*needscanp == 0);
777 needscan = 0;
778 /*
779 * Previous and following entries are both free,
780 * merge everything into a single free entry.
781 */
782 bf = args->dp->d_ops->data_bestfree_p(hdr);
783 if (prevdup && postdup) {
784 xfs_dir2_data_free_t *dfp2; /* another bestfree pointer */
785
786 /*
787 * See if prevdup and/or postdup are in bestfree table.
788 */
789 dfp = xfs_dir2_data_freefind(hdr, bf, prevdup);
790 dfp2 = xfs_dir2_data_freefind(hdr, bf, postdup);
791 /*
792 * We need a rescan unless there are exactly 2 free entries
793 * namely our two. Then we know what's happening, otherwise
794 * since the third bestfree is there, there might be more
795 * entries.
796 */
797 needscan = (bf[2].length != 0);
798 /*
799 * Fix up the new big freespace.
800 */
801 be16_add_cpu(&prevdup->length, len + be16_to_cpu(postdup->length));
802 *xfs_dir2_data_unused_tag_p(prevdup) =
803 cpu_to_be16((char *)prevdup - (char *)hdr);
804 xfs_dir2_data_log_unused(args, bp, prevdup);
805 if (!needscan) {
806 /*
807 * Has to be the case that entries 0 and 1 are
808 * dfp and dfp2 (don't know which is which), and
809 * entry 2 is empty.
810 * Remove entry 1 first then entry 0.
811 */
812 ASSERT(dfp && dfp2);
813 if (dfp == &bf[1]) {
814 dfp = &bf[0];
815 ASSERT(dfp2 == dfp);
816 dfp2 = &bf[1];
817 }
818 xfs_dir2_data_freeremove(hdr, bf, dfp2, needlogp);
819 xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
820 /*
821 * Now insert the new entry.
822 */
823 dfp = xfs_dir2_data_freeinsert(hdr, bf, prevdup,
824 needlogp);
825 ASSERT(dfp == &bf[0]);
826 ASSERT(dfp->length == prevdup->length);
827 ASSERT(!dfp[1].length);
828 ASSERT(!dfp[2].length);
829 }
830 }
831 /*
832 * The entry before us is free, merge with it.
833 */
834 else if (prevdup) {
835 dfp = xfs_dir2_data_freefind(hdr, bf, prevdup);
836 be16_add_cpu(&prevdup->length, len);
837 *xfs_dir2_data_unused_tag_p(prevdup) =
838 cpu_to_be16((char *)prevdup - (char *)hdr);
839 xfs_dir2_data_log_unused(args, bp, prevdup);
840 /*
841 * If the previous entry was in the table, the new entry
842 * is longer, so it will be in the table too. Remove
843 * the old one and add the new one.
844 */
845 if (dfp) {
846 xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
847 xfs_dir2_data_freeinsert(hdr, bf, prevdup, needlogp);
848 }
849 /*
850 * Otherwise we need a scan if the new entry is big enough.
851 */
852 else {
853 needscan = be16_to_cpu(prevdup->length) >
854 be16_to_cpu(bf[2].length);
855 }
856 }
857 /*
858 * The following entry is free, merge with it.
859 */
860 else if (postdup) {
861 dfp = xfs_dir2_data_freefind(hdr, bf, postdup);
862 newdup = (xfs_dir2_data_unused_t *)((char *)hdr + offset);
863 newdup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
864 newdup->length = cpu_to_be16(len + be16_to_cpu(postdup->length));
865 *xfs_dir2_data_unused_tag_p(newdup) =
866 cpu_to_be16((char *)newdup - (char *)hdr);
867 xfs_dir2_data_log_unused(args, bp, newdup);
868 /*
869 * If the following entry was in the table, the new entry
870 * is longer, so it will be in the table too. Remove
871 * the old one and add the new one.
872 */
873 if (dfp) {
874 xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
875 xfs_dir2_data_freeinsert(hdr, bf, newdup, needlogp);
876 }
877 /*
878 * Otherwise we need a scan if the new entry is big enough.
879 */
880 else {
881 needscan = be16_to_cpu(newdup->length) >
882 be16_to_cpu(bf[2].length);
883 }
884 }
885 /*
886 * Neither neighbor is free. Make a new entry.
887 */
888 else {
889 newdup = (xfs_dir2_data_unused_t *)((char *)hdr + offset);
890 newdup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
891 newdup->length = cpu_to_be16(len);
892 *xfs_dir2_data_unused_tag_p(newdup) =
893 cpu_to_be16((char *)newdup - (char *)hdr);
894 xfs_dir2_data_log_unused(args, bp, newdup);
895 xfs_dir2_data_freeinsert(hdr, bf, newdup, needlogp);
896 }
897 *needscanp = needscan;
898 }
899
900 /*
901 * Take a byte range out of an existing unused space and make it un-free.
902 */
903 void
xfs_dir2_data_use_free(struct xfs_da_args * args,struct xfs_buf * bp,xfs_dir2_data_unused_t * dup,xfs_dir2_data_aoff_t offset,xfs_dir2_data_aoff_t len,int * needlogp,int * needscanp)904 xfs_dir2_data_use_free(
905 struct xfs_da_args *args,
906 struct xfs_buf *bp,
907 xfs_dir2_data_unused_t *dup, /* unused entry */
908 xfs_dir2_data_aoff_t offset, /* starting offset to use */
909 xfs_dir2_data_aoff_t len, /* length to use */
910 int *needlogp, /* out: need to log header */
911 int *needscanp) /* out: need regen bestfree */
912 {
913 xfs_dir2_data_hdr_t *hdr; /* data block header */
914 xfs_dir2_data_free_t *dfp; /* bestfree pointer */
915 int matchback; /* matches end of freespace */
916 int matchfront; /* matches start of freespace */
917 int needscan; /* need to regen bestfree */
918 xfs_dir2_data_unused_t *newdup; /* new unused entry */
919 xfs_dir2_data_unused_t *newdup2; /* another new unused entry */
920 int oldlen; /* old unused entry's length */
921 struct xfs_dir2_data_free *bf;
922
923 hdr = bp->b_addr;
924 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
925 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
926 hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
927 hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
928 ASSERT(be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG);
929 ASSERT(offset >= (char *)dup - (char *)hdr);
930 ASSERT(offset + len <= (char *)dup + be16_to_cpu(dup->length) - (char *)hdr);
931 ASSERT((char *)dup - (char *)hdr == be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup)));
932 /*
933 * Look up the entry in the bestfree table.
934 */
935 oldlen = be16_to_cpu(dup->length);
936 bf = args->dp->d_ops->data_bestfree_p(hdr);
937 dfp = xfs_dir2_data_freefind(hdr, bf, dup);
938 ASSERT(dfp || oldlen <= be16_to_cpu(bf[2].length));
939 /*
940 * Check for alignment with front and back of the entry.
941 */
942 matchfront = (char *)dup - (char *)hdr == offset;
943 matchback = (char *)dup + oldlen - (char *)hdr == offset + len;
944 ASSERT(*needscanp == 0);
945 needscan = 0;
946 /*
947 * If we matched it exactly we just need to get rid of it from
948 * the bestfree table.
949 */
950 if (matchfront && matchback) {
951 if (dfp) {
952 needscan = (bf[2].offset != 0);
953 if (!needscan)
954 xfs_dir2_data_freeremove(hdr, bf, dfp,
955 needlogp);
956 }
957 }
958 /*
959 * We match the first part of the entry.
960 * Make a new entry with the remaining freespace.
961 */
962 else if (matchfront) {
963 newdup = (xfs_dir2_data_unused_t *)((char *)hdr + offset + len);
964 newdup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
965 newdup->length = cpu_to_be16(oldlen - len);
966 *xfs_dir2_data_unused_tag_p(newdup) =
967 cpu_to_be16((char *)newdup - (char *)hdr);
968 xfs_dir2_data_log_unused(args, bp, newdup);
969 /*
970 * If it was in the table, remove it and add the new one.
971 */
972 if (dfp) {
973 xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
974 dfp = xfs_dir2_data_freeinsert(hdr, bf, newdup,
975 needlogp);
976 ASSERT(dfp != NULL);
977 ASSERT(dfp->length == newdup->length);
978 ASSERT(be16_to_cpu(dfp->offset) == (char *)newdup - (char *)hdr);
979 /*
980 * If we got inserted at the last slot,
981 * that means we don't know if there was a better
982 * choice for the last slot, or not. Rescan.
983 */
984 needscan = dfp == &bf[2];
985 }
986 }
987 /*
988 * We match the last part of the entry.
989 * Trim the allocated space off the tail of the entry.
990 */
991 else if (matchback) {
992 newdup = dup;
993 newdup->length = cpu_to_be16(((char *)hdr + offset) - (char *)newdup);
994 *xfs_dir2_data_unused_tag_p(newdup) =
995 cpu_to_be16((char *)newdup - (char *)hdr);
996 xfs_dir2_data_log_unused(args, bp, newdup);
997 /*
998 * If it was in the table, remove it and add the new one.
999 */
1000 if (dfp) {
1001 xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
1002 dfp = xfs_dir2_data_freeinsert(hdr, bf, newdup,
1003 needlogp);
1004 ASSERT(dfp != NULL);
1005 ASSERT(dfp->length == newdup->length);
1006 ASSERT(be16_to_cpu(dfp->offset) == (char *)newdup - (char *)hdr);
1007 /*
1008 * If we got inserted at the last slot,
1009 * that means we don't know if there was a better
1010 * choice for the last slot, or not. Rescan.
1011 */
1012 needscan = dfp == &bf[2];
1013 }
1014 }
1015 /*
1016 * Poking out the middle of an entry.
1017 * Make two new entries.
1018 */
1019 else {
1020 newdup = dup;
1021 newdup->length = cpu_to_be16(((char *)hdr + offset) - (char *)newdup);
1022 *xfs_dir2_data_unused_tag_p(newdup) =
1023 cpu_to_be16((char *)newdup - (char *)hdr);
1024 xfs_dir2_data_log_unused(args, bp, newdup);
1025 newdup2 = (xfs_dir2_data_unused_t *)((char *)hdr + offset + len);
1026 newdup2->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
1027 newdup2->length = cpu_to_be16(oldlen - len - be16_to_cpu(newdup->length));
1028 *xfs_dir2_data_unused_tag_p(newdup2) =
1029 cpu_to_be16((char *)newdup2 - (char *)hdr);
1030 xfs_dir2_data_log_unused(args, bp, newdup2);
1031 /*
1032 * If the old entry was in the table, we need to scan
1033 * if the 3rd entry was valid, since these entries
1034 * are smaller than the old one.
1035 * If we don't need to scan that means there were 1 or 2
1036 * entries in the table, and removing the old and adding
1037 * the 2 new will work.
1038 */
1039 if (dfp) {
1040 needscan = (bf[2].length != 0);
1041 if (!needscan) {
1042 xfs_dir2_data_freeremove(hdr, bf, dfp,
1043 needlogp);
1044 xfs_dir2_data_freeinsert(hdr, bf, newdup,
1045 needlogp);
1046 xfs_dir2_data_freeinsert(hdr, bf, newdup2,
1047 needlogp);
1048 }
1049 }
1050 }
1051 *needscanp = needscan;
1052 }
1053