mm: fix exec activate_mm vs TLB shootdown and lazy tlb switching race
[linux/fpc-iii.git] / fs / xfs / libxfs / xfs_dir2_block.c
blob43c902f7a68d4da0c6eb9853aeb14b5fbbe52340
1 /*
2 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
3 * Copyright (c) 2013 Red Hat, Inc.
4 * All Rights Reserved.
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.
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.
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
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_trans.h"
29 #include "xfs_inode_item.h"
30 #include "xfs_bmap.h"
31 #include "xfs_buf_item.h"
32 #include "xfs_dir2.h"
33 #include "xfs_dir2_priv.h"
34 #include "xfs_error.h"
35 #include "xfs_trace.h"
36 #include "xfs_cksum.h"
37 #include "xfs_log.h"
40 * Local function prototypes.
42 static void xfs_dir2_block_log_leaf(xfs_trans_t *tp, struct xfs_buf *bp,
43 int first, int last);
44 static void xfs_dir2_block_log_tail(xfs_trans_t *tp, struct xfs_buf *bp);
45 static int xfs_dir2_block_lookup_int(xfs_da_args_t *args, struct xfs_buf **bpp,
46 int *entno);
47 static int xfs_dir2_block_sort(const void *a, const void *b);
49 static xfs_dahash_t xfs_dir_hash_dot, xfs_dir_hash_dotdot;
52 * One-time startup routine called from xfs_init().
54 void
55 xfs_dir_startup(void)
57 xfs_dir_hash_dot = xfs_da_hashname((unsigned char *)".", 1);
58 xfs_dir_hash_dotdot = xfs_da_hashname((unsigned char *)"..", 2);
61 static bool
62 xfs_dir3_block_verify(
63 struct xfs_buf *bp)
65 struct xfs_mount *mp = bp->b_target->bt_mount;
66 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
68 if (xfs_sb_version_hascrc(&mp->m_sb)) {
69 if (hdr3->magic != cpu_to_be32(XFS_DIR3_BLOCK_MAGIC))
70 return false;
71 if (!uuid_equal(&hdr3->uuid, &mp->m_sb.sb_meta_uuid))
72 return false;
73 if (be64_to_cpu(hdr3->blkno) != bp->b_bn)
74 return false;
75 if (!xfs_log_check_lsn(mp, be64_to_cpu(hdr3->lsn)))
76 return false;
77 } else {
78 if (hdr3->magic != cpu_to_be32(XFS_DIR2_BLOCK_MAGIC))
79 return false;
81 if (__xfs_dir3_data_check(NULL, bp))
82 return false;
83 return true;
86 static void
87 xfs_dir3_block_read_verify(
88 struct xfs_buf *bp)
90 struct xfs_mount *mp = bp->b_target->bt_mount;
92 if (xfs_sb_version_hascrc(&mp->m_sb) &&
93 !xfs_buf_verify_cksum(bp, XFS_DIR3_DATA_CRC_OFF))
94 xfs_buf_ioerror(bp, -EFSBADCRC);
95 else if (!xfs_dir3_block_verify(bp))
96 xfs_buf_ioerror(bp, -EFSCORRUPTED);
98 if (bp->b_error)
99 xfs_verifier_error(bp);
102 static void
103 xfs_dir3_block_write_verify(
104 struct xfs_buf *bp)
106 struct xfs_mount *mp = bp->b_target->bt_mount;
107 struct xfs_buf_log_item *bip = bp->b_fspriv;
108 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
110 if (!xfs_dir3_block_verify(bp)) {
111 xfs_buf_ioerror(bp, -EFSCORRUPTED);
112 xfs_verifier_error(bp);
113 return;
116 if (!xfs_sb_version_hascrc(&mp->m_sb))
117 return;
119 if (bip)
120 hdr3->lsn = cpu_to_be64(bip->bli_item.li_lsn);
122 xfs_buf_update_cksum(bp, XFS_DIR3_DATA_CRC_OFF);
125 const struct xfs_buf_ops xfs_dir3_block_buf_ops = {
126 .name = "xfs_dir3_block",
127 .verify_read = xfs_dir3_block_read_verify,
128 .verify_write = xfs_dir3_block_write_verify,
132 xfs_dir3_block_read(
133 struct xfs_trans *tp,
134 struct xfs_inode *dp,
135 struct xfs_buf **bpp)
137 struct xfs_mount *mp = dp->i_mount;
138 int err;
140 err = xfs_da_read_buf(tp, dp, mp->m_dir_geo->datablk, -1, bpp,
141 XFS_DATA_FORK, &xfs_dir3_block_buf_ops);
142 if (!err && tp && *bpp)
143 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_BLOCK_BUF);
144 return err;
147 static void
148 xfs_dir3_block_init(
149 struct xfs_mount *mp,
150 struct xfs_trans *tp,
151 struct xfs_buf *bp,
152 struct xfs_inode *dp)
154 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
156 bp->b_ops = &xfs_dir3_block_buf_ops;
157 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_BLOCK_BUF);
159 if (xfs_sb_version_hascrc(&mp->m_sb)) {
160 memset(hdr3, 0, sizeof(*hdr3));
161 hdr3->magic = cpu_to_be32(XFS_DIR3_BLOCK_MAGIC);
162 hdr3->blkno = cpu_to_be64(bp->b_bn);
163 hdr3->owner = cpu_to_be64(dp->i_ino);
164 uuid_copy(&hdr3->uuid, &mp->m_sb.sb_meta_uuid);
165 return;
168 hdr3->magic = cpu_to_be32(XFS_DIR2_BLOCK_MAGIC);
171 static void
172 xfs_dir2_block_need_space(
173 struct xfs_inode *dp,
174 struct xfs_dir2_data_hdr *hdr,
175 struct xfs_dir2_block_tail *btp,
176 struct xfs_dir2_leaf_entry *blp,
177 __be16 **tagpp,
178 struct xfs_dir2_data_unused **dupp,
179 struct xfs_dir2_data_unused **enddupp,
180 int *compact,
181 int len)
183 struct xfs_dir2_data_free *bf;
184 __be16 *tagp = NULL;
185 struct xfs_dir2_data_unused *dup = NULL;
186 struct xfs_dir2_data_unused *enddup = NULL;
188 *compact = 0;
189 bf = dp->d_ops->data_bestfree_p(hdr);
192 * If there are stale entries we'll use one for the leaf.
194 if (btp->stale) {
195 if (be16_to_cpu(bf[0].length) >= len) {
197 * The biggest entry enough to avoid compaction.
199 dup = (xfs_dir2_data_unused_t *)
200 ((char *)hdr + be16_to_cpu(bf[0].offset));
201 goto out;
205 * Will need to compact to make this work.
206 * Tag just before the first leaf entry.
208 *compact = 1;
209 tagp = (__be16 *)blp - 1;
211 /* Data object just before the first leaf entry. */
212 dup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
215 * If it's not free then the data will go where the
216 * leaf data starts now, if it works at all.
218 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
219 if (be16_to_cpu(dup->length) + (be32_to_cpu(btp->stale) - 1) *
220 (uint)sizeof(*blp) < len)
221 dup = NULL;
222 } else if ((be32_to_cpu(btp->stale) - 1) * (uint)sizeof(*blp) < len)
223 dup = NULL;
224 else
225 dup = (xfs_dir2_data_unused_t *)blp;
226 goto out;
230 * no stale entries, so just use free space.
231 * Tag just before the first leaf entry.
233 tagp = (__be16 *)blp - 1;
235 /* Data object just before the first leaf entry. */
236 enddup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
239 * If it's not free then can't do this add without cleaning up:
240 * the space before the first leaf entry needs to be free so it
241 * can be expanded to hold the pointer to the new entry.
243 if (be16_to_cpu(enddup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
245 * Check out the biggest freespace and see if it's the same one.
247 dup = (xfs_dir2_data_unused_t *)
248 ((char *)hdr + be16_to_cpu(bf[0].offset));
249 if (dup != enddup) {
251 * Not the same free entry, just check its length.
253 if (be16_to_cpu(dup->length) < len)
254 dup = NULL;
255 goto out;
259 * It is the biggest freespace, can it hold the leaf too?
261 if (be16_to_cpu(dup->length) < len + (uint)sizeof(*blp)) {
263 * Yes, use the second-largest entry instead if it works.
265 if (be16_to_cpu(bf[1].length) >= len)
266 dup = (xfs_dir2_data_unused_t *)
267 ((char *)hdr + be16_to_cpu(bf[1].offset));
268 else
269 dup = NULL;
272 out:
273 *tagpp = tagp;
274 *dupp = dup;
275 *enddupp = enddup;
279 * compact the leaf entries.
280 * Leave the highest-numbered stale entry stale.
281 * XXX should be the one closest to mid but mid is not yet computed.
283 static void
284 xfs_dir2_block_compact(
285 struct xfs_da_args *args,
286 struct xfs_buf *bp,
287 struct xfs_dir2_data_hdr *hdr,
288 struct xfs_dir2_block_tail *btp,
289 struct xfs_dir2_leaf_entry *blp,
290 int *needlog,
291 int *lfloghigh,
292 int *lfloglow)
294 int fromidx; /* source leaf index */
295 int toidx; /* target leaf index */
296 int needscan = 0;
297 int highstale; /* high stale index */
299 fromidx = toidx = be32_to_cpu(btp->count) - 1;
300 highstale = *lfloghigh = -1;
301 for (; fromidx >= 0; fromidx--) {
302 if (blp[fromidx].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR)) {
303 if (highstale == -1)
304 highstale = toidx;
305 else {
306 if (*lfloghigh == -1)
307 *lfloghigh = toidx;
308 continue;
311 if (fromidx < toidx)
312 blp[toidx] = blp[fromidx];
313 toidx--;
315 *lfloglow = toidx + 1 - (be32_to_cpu(btp->stale) - 1);
316 *lfloghigh -= be32_to_cpu(btp->stale) - 1;
317 be32_add_cpu(&btp->count, -(be32_to_cpu(btp->stale) - 1));
318 xfs_dir2_data_make_free(args, bp,
319 (xfs_dir2_data_aoff_t)((char *)blp - (char *)hdr),
320 (xfs_dir2_data_aoff_t)((be32_to_cpu(btp->stale) - 1) * sizeof(*blp)),
321 needlog, &needscan);
322 btp->stale = cpu_to_be32(1);
324 * If we now need to rebuild the bestfree map, do so.
325 * This needs to happen before the next call to use_free.
327 if (needscan)
328 xfs_dir2_data_freescan(args->dp, hdr, needlog);
332 * Add an entry to a block directory.
334 int /* error */
335 xfs_dir2_block_addname(
336 xfs_da_args_t *args) /* directory op arguments */
338 xfs_dir2_data_hdr_t *hdr; /* block header */
339 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
340 struct xfs_buf *bp; /* buffer for block */
341 xfs_dir2_block_tail_t *btp; /* block tail */
342 int compact; /* need to compact leaf ents */
343 xfs_dir2_data_entry_t *dep; /* block data entry */
344 xfs_inode_t *dp; /* directory inode */
345 xfs_dir2_data_unused_t *dup; /* block unused entry */
346 int error; /* error return value */
347 xfs_dir2_data_unused_t *enddup=NULL; /* unused at end of data */
348 xfs_dahash_t hash; /* hash value of found entry */
349 int high; /* high index for binary srch */
350 int highstale; /* high stale index */
351 int lfloghigh=0; /* last final leaf to log */
352 int lfloglow=0; /* first final leaf to log */
353 int len; /* length of the new entry */
354 int low; /* low index for binary srch */
355 int lowstale; /* low stale index */
356 int mid=0; /* midpoint for binary srch */
357 int needlog; /* need to log header */
358 int needscan; /* need to rescan freespace */
359 __be16 *tagp; /* pointer to tag value */
360 xfs_trans_t *tp; /* transaction structure */
362 trace_xfs_dir2_block_addname(args);
364 dp = args->dp;
365 tp = args->trans;
367 /* Read the (one and only) directory block into bp. */
368 error = xfs_dir3_block_read(tp, dp, &bp);
369 if (error)
370 return error;
372 len = dp->d_ops->data_entsize(args->namelen);
375 * Set up pointers to parts of the block.
377 hdr = bp->b_addr;
378 btp = xfs_dir2_block_tail_p(args->geo, hdr);
379 blp = xfs_dir2_block_leaf_p(btp);
382 * Find out if we can reuse stale entries or whether we need extra
383 * space for entry and new leaf.
385 xfs_dir2_block_need_space(dp, hdr, btp, blp, &tagp, &dup,
386 &enddup, &compact, len);
389 * Done everything we need for a space check now.
391 if (args->op_flags & XFS_DA_OP_JUSTCHECK) {
392 xfs_trans_brelse(tp, bp);
393 if (!dup)
394 return -ENOSPC;
395 return 0;
399 * If we don't have space for the new entry & leaf ...
401 if (!dup) {
402 /* Don't have a space reservation: return no-space. */
403 if (args->total == 0)
404 return -ENOSPC;
406 * Convert to the next larger format.
407 * Then add the new entry in that format.
409 error = xfs_dir2_block_to_leaf(args, bp);
410 if (error)
411 return error;
412 return xfs_dir2_leaf_addname(args);
415 needlog = needscan = 0;
418 * If need to compact the leaf entries, do it now.
420 if (compact) {
421 xfs_dir2_block_compact(args, bp, hdr, btp, blp, &needlog,
422 &lfloghigh, &lfloglow);
423 /* recalculate blp post-compaction */
424 blp = xfs_dir2_block_leaf_p(btp);
425 } else if (btp->stale) {
427 * Set leaf logging boundaries to impossible state.
428 * For the no-stale case they're set explicitly.
430 lfloglow = be32_to_cpu(btp->count);
431 lfloghigh = -1;
435 * Find the slot that's first lower than our hash value, -1 if none.
437 for (low = 0, high = be32_to_cpu(btp->count) - 1; low <= high; ) {
438 mid = (low + high) >> 1;
439 if ((hash = be32_to_cpu(blp[mid].hashval)) == args->hashval)
440 break;
441 if (hash < args->hashval)
442 low = mid + 1;
443 else
444 high = mid - 1;
446 while (mid >= 0 && be32_to_cpu(blp[mid].hashval) >= args->hashval) {
447 mid--;
450 * No stale entries, will use enddup space to hold new leaf.
452 if (!btp->stale) {
454 * Mark the space needed for the new leaf entry, now in use.
456 xfs_dir2_data_use_free(args, bp, enddup,
457 (xfs_dir2_data_aoff_t)
458 ((char *)enddup - (char *)hdr + be16_to_cpu(enddup->length) -
459 sizeof(*blp)),
460 (xfs_dir2_data_aoff_t)sizeof(*blp),
461 &needlog, &needscan);
463 * Update the tail (entry count).
465 be32_add_cpu(&btp->count, 1);
467 * If we now need to rebuild the bestfree map, do so.
468 * This needs to happen before the next call to use_free.
470 if (needscan) {
471 xfs_dir2_data_freescan(dp, hdr, &needlog);
472 needscan = 0;
475 * Adjust pointer to the first leaf entry, we're about to move
476 * the table up one to open up space for the new leaf entry.
477 * Then adjust our index to match.
479 blp--;
480 mid++;
481 if (mid)
482 memmove(blp, &blp[1], mid * sizeof(*blp));
483 lfloglow = 0;
484 lfloghigh = mid;
487 * Use a stale leaf for our new entry.
489 else {
490 for (lowstale = mid;
491 lowstale >= 0 &&
492 blp[lowstale].address !=
493 cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
494 lowstale--)
495 continue;
496 for (highstale = mid + 1;
497 highstale < be32_to_cpu(btp->count) &&
498 blp[highstale].address !=
499 cpu_to_be32(XFS_DIR2_NULL_DATAPTR) &&
500 (lowstale < 0 || mid - lowstale > highstale - mid);
501 highstale++)
502 continue;
504 * Move entries toward the low-numbered stale entry.
506 if (lowstale >= 0 &&
507 (highstale == be32_to_cpu(btp->count) ||
508 mid - lowstale <= highstale - mid)) {
509 if (mid - lowstale)
510 memmove(&blp[lowstale], &blp[lowstale + 1],
511 (mid - lowstale) * sizeof(*blp));
512 lfloglow = MIN(lowstale, lfloglow);
513 lfloghigh = MAX(mid, lfloghigh);
516 * Move entries toward the high-numbered stale entry.
518 else {
519 ASSERT(highstale < be32_to_cpu(btp->count));
520 mid++;
521 if (highstale - mid)
522 memmove(&blp[mid + 1], &blp[mid],
523 (highstale - mid) * sizeof(*blp));
524 lfloglow = MIN(mid, lfloglow);
525 lfloghigh = MAX(highstale, lfloghigh);
527 be32_add_cpu(&btp->stale, -1);
530 * Point to the new data entry.
532 dep = (xfs_dir2_data_entry_t *)dup;
534 * Fill in the leaf entry.
536 blp[mid].hashval = cpu_to_be32(args->hashval);
537 blp[mid].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(
538 (char *)dep - (char *)hdr));
539 xfs_dir2_block_log_leaf(tp, bp, lfloglow, lfloghigh);
541 * Mark space for the data entry used.
543 xfs_dir2_data_use_free(args, bp, dup,
544 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr),
545 (xfs_dir2_data_aoff_t)len, &needlog, &needscan);
547 * Create the new data entry.
549 dep->inumber = cpu_to_be64(args->inumber);
550 dep->namelen = args->namelen;
551 memcpy(dep->name, args->name, args->namelen);
552 dp->d_ops->data_put_ftype(dep, args->filetype);
553 tagp = dp->d_ops->data_entry_tag_p(dep);
554 *tagp = cpu_to_be16((char *)dep - (char *)hdr);
556 * Clean up the bestfree array and log the header, tail, and entry.
558 if (needscan)
559 xfs_dir2_data_freescan(dp, hdr, &needlog);
560 if (needlog)
561 xfs_dir2_data_log_header(args, bp);
562 xfs_dir2_block_log_tail(tp, bp);
563 xfs_dir2_data_log_entry(args, bp, dep);
564 xfs_dir3_data_check(dp, bp);
565 return 0;
569 * Log leaf entries from the block.
571 static void
572 xfs_dir2_block_log_leaf(
573 xfs_trans_t *tp, /* transaction structure */
574 struct xfs_buf *bp, /* block buffer */
575 int first, /* index of first logged leaf */
576 int last) /* index of last logged leaf */
578 xfs_dir2_data_hdr_t *hdr = bp->b_addr;
579 xfs_dir2_leaf_entry_t *blp;
580 xfs_dir2_block_tail_t *btp;
582 btp = xfs_dir2_block_tail_p(tp->t_mountp->m_dir_geo, hdr);
583 blp = xfs_dir2_block_leaf_p(btp);
584 xfs_trans_log_buf(tp, bp, (uint)((char *)&blp[first] - (char *)hdr),
585 (uint)((char *)&blp[last + 1] - (char *)hdr - 1));
589 * Log the block tail.
591 static void
592 xfs_dir2_block_log_tail(
593 xfs_trans_t *tp, /* transaction structure */
594 struct xfs_buf *bp) /* block buffer */
596 xfs_dir2_data_hdr_t *hdr = bp->b_addr;
597 xfs_dir2_block_tail_t *btp;
599 btp = xfs_dir2_block_tail_p(tp->t_mountp->m_dir_geo, hdr);
600 xfs_trans_log_buf(tp, bp, (uint)((char *)btp - (char *)hdr),
601 (uint)((char *)(btp + 1) - (char *)hdr - 1));
605 * Look up an entry in the block. This is the external routine,
606 * xfs_dir2_block_lookup_int does the real work.
608 int /* error */
609 xfs_dir2_block_lookup(
610 xfs_da_args_t *args) /* dir lookup arguments */
612 xfs_dir2_data_hdr_t *hdr; /* block header */
613 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
614 struct xfs_buf *bp; /* block buffer */
615 xfs_dir2_block_tail_t *btp; /* block tail */
616 xfs_dir2_data_entry_t *dep; /* block data entry */
617 xfs_inode_t *dp; /* incore inode */
618 int ent; /* entry index */
619 int error; /* error return value */
621 trace_xfs_dir2_block_lookup(args);
624 * Get the buffer, look up the entry.
625 * If not found (ENOENT) then return, have no buffer.
627 if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent)))
628 return error;
629 dp = args->dp;
630 hdr = bp->b_addr;
631 xfs_dir3_data_check(dp, bp);
632 btp = xfs_dir2_block_tail_p(args->geo, hdr);
633 blp = xfs_dir2_block_leaf_p(btp);
635 * Get the offset from the leaf entry, to point to the data.
637 dep = (xfs_dir2_data_entry_t *)((char *)hdr +
638 xfs_dir2_dataptr_to_off(args->geo,
639 be32_to_cpu(blp[ent].address)));
641 * Fill in inode number, CI name if appropriate, release the block.
643 args->inumber = be64_to_cpu(dep->inumber);
644 args->filetype = dp->d_ops->data_get_ftype(dep);
645 error = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
646 xfs_trans_brelse(args->trans, bp);
647 return error;
651 * Internal block lookup routine.
653 static int /* error */
654 xfs_dir2_block_lookup_int(
655 xfs_da_args_t *args, /* dir lookup arguments */
656 struct xfs_buf **bpp, /* returned block buffer */
657 int *entno) /* returned entry number */
659 xfs_dir2_dataptr_t addr; /* data entry address */
660 xfs_dir2_data_hdr_t *hdr; /* block header */
661 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
662 struct xfs_buf *bp; /* block buffer */
663 xfs_dir2_block_tail_t *btp; /* block tail */
664 xfs_dir2_data_entry_t *dep; /* block data entry */
665 xfs_inode_t *dp; /* incore inode */
666 int error; /* error return value */
667 xfs_dahash_t hash; /* found hash value */
668 int high; /* binary search high index */
669 int low; /* binary search low index */
670 int mid; /* binary search current idx */
671 xfs_mount_t *mp; /* filesystem mount point */
672 xfs_trans_t *tp; /* transaction pointer */
673 enum xfs_dacmp cmp; /* comparison result */
675 dp = args->dp;
676 tp = args->trans;
677 mp = dp->i_mount;
679 error = xfs_dir3_block_read(tp, dp, &bp);
680 if (error)
681 return error;
683 hdr = bp->b_addr;
684 xfs_dir3_data_check(dp, bp);
685 btp = xfs_dir2_block_tail_p(args->geo, hdr);
686 blp = xfs_dir2_block_leaf_p(btp);
688 * Loop doing a binary search for our hash value.
689 * Find our entry, ENOENT if it's not there.
691 for (low = 0, high = be32_to_cpu(btp->count) - 1; ; ) {
692 ASSERT(low <= high);
693 mid = (low + high) >> 1;
694 if ((hash = be32_to_cpu(blp[mid].hashval)) == args->hashval)
695 break;
696 if (hash < args->hashval)
697 low = mid + 1;
698 else
699 high = mid - 1;
700 if (low > high) {
701 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
702 xfs_trans_brelse(tp, bp);
703 return -ENOENT;
707 * Back up to the first one with the right hash value.
709 while (mid > 0 && be32_to_cpu(blp[mid - 1].hashval) == args->hashval) {
710 mid--;
713 * Now loop forward through all the entries with the
714 * right hash value looking for our name.
716 do {
717 if ((addr = be32_to_cpu(blp[mid].address)) == XFS_DIR2_NULL_DATAPTR)
718 continue;
720 * Get pointer to the entry from the leaf.
722 dep = (xfs_dir2_data_entry_t *)
723 ((char *)hdr + xfs_dir2_dataptr_to_off(args->geo, addr));
725 * Compare name and if it's an exact match, return the index
726 * and buffer. If it's the first case-insensitive match, store
727 * the index and buffer and continue looking for an exact match.
729 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
730 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
731 args->cmpresult = cmp;
732 *bpp = bp;
733 *entno = mid;
734 if (cmp == XFS_CMP_EXACT)
735 return 0;
737 } while (++mid < be32_to_cpu(btp->count) &&
738 be32_to_cpu(blp[mid].hashval) == hash);
740 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
742 * Here, we can only be doing a lookup (not a rename or replace).
743 * If a case-insensitive match was found earlier, return success.
745 if (args->cmpresult == XFS_CMP_CASE)
746 return 0;
748 * No match, release the buffer and return ENOENT.
750 xfs_trans_brelse(tp, bp);
751 return -ENOENT;
755 * Remove an entry from a block format directory.
756 * If that makes the block small enough to fit in shortform, transform it.
758 int /* error */
759 xfs_dir2_block_removename(
760 xfs_da_args_t *args) /* directory operation args */
762 xfs_dir2_data_hdr_t *hdr; /* block header */
763 xfs_dir2_leaf_entry_t *blp; /* block leaf pointer */
764 struct xfs_buf *bp; /* block buffer */
765 xfs_dir2_block_tail_t *btp; /* block tail */
766 xfs_dir2_data_entry_t *dep; /* block data entry */
767 xfs_inode_t *dp; /* incore inode */
768 int ent; /* block leaf entry index */
769 int error; /* error return value */
770 int needlog; /* need to log block header */
771 int needscan; /* need to fixup bestfree */
772 xfs_dir2_sf_hdr_t sfh; /* shortform header */
773 int size; /* shortform size */
774 xfs_trans_t *tp; /* transaction pointer */
776 trace_xfs_dir2_block_removename(args);
779 * Look up the entry in the block. Gets the buffer and entry index.
780 * It will always be there, the vnodeops level does a lookup first.
782 if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent))) {
783 return error;
785 dp = args->dp;
786 tp = args->trans;
787 hdr = bp->b_addr;
788 btp = xfs_dir2_block_tail_p(args->geo, hdr);
789 blp = xfs_dir2_block_leaf_p(btp);
791 * Point to the data entry using the leaf entry.
793 dep = (xfs_dir2_data_entry_t *)((char *)hdr +
794 xfs_dir2_dataptr_to_off(args->geo,
795 be32_to_cpu(blp[ent].address)));
797 * Mark the data entry's space free.
799 needlog = needscan = 0;
800 xfs_dir2_data_make_free(args, bp,
801 (xfs_dir2_data_aoff_t)((char *)dep - (char *)hdr),
802 dp->d_ops->data_entsize(dep->namelen), &needlog, &needscan);
804 * Fix up the block tail.
806 be32_add_cpu(&btp->stale, 1);
807 xfs_dir2_block_log_tail(tp, bp);
809 * Remove the leaf entry by marking it stale.
811 blp[ent].address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
812 xfs_dir2_block_log_leaf(tp, bp, ent, ent);
814 * Fix up bestfree, log the header if necessary.
816 if (needscan)
817 xfs_dir2_data_freescan(dp, hdr, &needlog);
818 if (needlog)
819 xfs_dir2_data_log_header(args, bp);
820 xfs_dir3_data_check(dp, bp);
822 * See if the size as a shortform is good enough.
824 size = xfs_dir2_block_sfsize(dp, hdr, &sfh);
825 if (size > XFS_IFORK_DSIZE(dp))
826 return 0;
829 * If it works, do the conversion.
831 return xfs_dir2_block_to_sf(args, bp, size, &sfh);
835 * Replace an entry in a V2 block directory.
836 * Change the inode number to the new value.
838 int /* error */
839 xfs_dir2_block_replace(
840 xfs_da_args_t *args) /* directory operation args */
842 xfs_dir2_data_hdr_t *hdr; /* block header */
843 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
844 struct xfs_buf *bp; /* block buffer */
845 xfs_dir2_block_tail_t *btp; /* block tail */
846 xfs_dir2_data_entry_t *dep; /* block data entry */
847 xfs_inode_t *dp; /* incore inode */
848 int ent; /* leaf entry index */
849 int error; /* error return value */
851 trace_xfs_dir2_block_replace(args);
854 * Lookup the entry in the directory. Get buffer and entry index.
855 * This will always succeed since the caller has already done a lookup.
857 if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent))) {
858 return error;
860 dp = args->dp;
861 hdr = bp->b_addr;
862 btp = xfs_dir2_block_tail_p(args->geo, hdr);
863 blp = xfs_dir2_block_leaf_p(btp);
865 * Point to the data entry we need to change.
867 dep = (xfs_dir2_data_entry_t *)((char *)hdr +
868 xfs_dir2_dataptr_to_off(args->geo,
869 be32_to_cpu(blp[ent].address)));
870 ASSERT(be64_to_cpu(dep->inumber) != args->inumber);
872 * Change the inode number to the new value.
874 dep->inumber = cpu_to_be64(args->inumber);
875 dp->d_ops->data_put_ftype(dep, args->filetype);
876 xfs_dir2_data_log_entry(args, bp, dep);
877 xfs_dir3_data_check(dp, bp);
878 return 0;
882 * Qsort comparison routine for the block leaf entries.
884 static int /* sort order */
885 xfs_dir2_block_sort(
886 const void *a, /* first leaf entry */
887 const void *b) /* second leaf entry */
889 const xfs_dir2_leaf_entry_t *la; /* first leaf entry */
890 const xfs_dir2_leaf_entry_t *lb; /* second leaf entry */
892 la = a;
893 lb = b;
894 return be32_to_cpu(la->hashval) < be32_to_cpu(lb->hashval) ? -1 :
895 (be32_to_cpu(la->hashval) > be32_to_cpu(lb->hashval) ? 1 : 0);
899 * Convert a V2 leaf directory to a V2 block directory if possible.
901 int /* error */
902 xfs_dir2_leaf_to_block(
903 xfs_da_args_t *args, /* operation arguments */
904 struct xfs_buf *lbp, /* leaf buffer */
905 struct xfs_buf *dbp) /* data buffer */
907 __be16 *bestsp; /* leaf bests table */
908 xfs_dir2_data_hdr_t *hdr; /* block header */
909 xfs_dir2_block_tail_t *btp; /* block tail */
910 xfs_inode_t *dp; /* incore directory inode */
911 xfs_dir2_data_unused_t *dup; /* unused data entry */
912 int error; /* error return value */
913 int from; /* leaf from index */
914 xfs_dir2_leaf_t *leaf; /* leaf structure */
915 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
916 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
917 xfs_mount_t *mp; /* file system mount point */
918 int needlog; /* need to log data header */
919 int needscan; /* need to scan for bestfree */
920 xfs_dir2_sf_hdr_t sfh; /* shortform header */
921 int size; /* bytes used */
922 __be16 *tagp; /* end of entry (tag) */
923 int to; /* block/leaf to index */
924 xfs_trans_t *tp; /* transaction pointer */
925 struct xfs_dir2_leaf_entry *ents;
926 struct xfs_dir3_icleaf_hdr leafhdr;
928 trace_xfs_dir2_leaf_to_block(args);
930 dp = args->dp;
931 tp = args->trans;
932 mp = dp->i_mount;
933 leaf = lbp->b_addr;
934 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
935 ents = dp->d_ops->leaf_ents_p(leaf);
936 ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
938 ASSERT(leafhdr.magic == XFS_DIR2_LEAF1_MAGIC ||
939 leafhdr.magic == XFS_DIR3_LEAF1_MAGIC);
941 * If there are data blocks other than the first one, take this
942 * opportunity to remove trailing empty data blocks that may have
943 * been left behind during no-space-reservation operations.
944 * These will show up in the leaf bests table.
946 while (dp->i_d.di_size > args->geo->blksize) {
947 int hdrsz;
949 hdrsz = dp->d_ops->data_entry_offset;
950 bestsp = xfs_dir2_leaf_bests_p(ltp);
951 if (be16_to_cpu(bestsp[be32_to_cpu(ltp->bestcount) - 1]) ==
952 args->geo->blksize - hdrsz) {
953 if ((error =
954 xfs_dir2_leaf_trim_data(args, lbp,
955 (xfs_dir2_db_t)(be32_to_cpu(ltp->bestcount) - 1))))
956 return error;
957 } else
958 return 0;
961 * Read the data block if we don't already have it, give up if it fails.
963 if (!dbp) {
964 error = xfs_dir3_data_read(tp, dp, args->geo->datablk, -1, &dbp);
965 if (error)
966 return error;
968 hdr = dbp->b_addr;
969 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
970 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC));
973 * Size of the "leaf" area in the block.
975 size = (uint)sizeof(xfs_dir2_block_tail_t) +
976 (uint)sizeof(*lep) * (leafhdr.count - leafhdr.stale);
978 * Look at the last data entry.
980 tagp = (__be16 *)((char *)hdr + args->geo->blksize) - 1;
981 dup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
983 * If it's not free or is too short we can't do it.
985 if (be16_to_cpu(dup->freetag) != XFS_DIR2_DATA_FREE_TAG ||
986 be16_to_cpu(dup->length) < size)
987 return 0;
990 * Start converting it to block form.
992 xfs_dir3_block_init(mp, tp, dbp, dp);
994 needlog = 1;
995 needscan = 0;
997 * Use up the space at the end of the block (blp/btp).
999 xfs_dir2_data_use_free(args, dbp, dup, args->geo->blksize - size, size,
1000 &needlog, &needscan);
1002 * Initialize the block tail.
1004 btp = xfs_dir2_block_tail_p(args->geo, hdr);
1005 btp->count = cpu_to_be32(leafhdr.count - leafhdr.stale);
1006 btp->stale = 0;
1007 xfs_dir2_block_log_tail(tp, dbp);
1009 * Initialize the block leaf area. We compact out stale entries.
1011 lep = xfs_dir2_block_leaf_p(btp);
1012 for (from = to = 0; from < leafhdr.count; from++) {
1013 if (ents[from].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
1014 continue;
1015 lep[to++] = ents[from];
1017 ASSERT(to == be32_to_cpu(btp->count));
1018 xfs_dir2_block_log_leaf(tp, dbp, 0, be32_to_cpu(btp->count) - 1);
1020 * Scan the bestfree if we need it and log the data block header.
1022 if (needscan)
1023 xfs_dir2_data_freescan(dp, hdr, &needlog);
1024 if (needlog)
1025 xfs_dir2_data_log_header(args, dbp);
1027 * Pitch the old leaf block.
1029 error = xfs_da_shrink_inode(args, args->geo->leafblk, lbp);
1030 if (error)
1031 return error;
1034 * Now see if the resulting block can be shrunken to shortform.
1036 size = xfs_dir2_block_sfsize(dp, hdr, &sfh);
1037 if (size > XFS_IFORK_DSIZE(dp))
1038 return 0;
1040 return xfs_dir2_block_to_sf(args, dbp, size, &sfh);
1044 * Convert the shortform directory to block form.
1046 int /* error */
1047 xfs_dir2_sf_to_block(
1048 xfs_da_args_t *args) /* operation arguments */
1050 xfs_dir2_db_t blkno; /* dir-relative block # (0) */
1051 xfs_dir2_data_hdr_t *hdr; /* block header */
1052 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
1053 struct xfs_buf *bp; /* block buffer */
1054 xfs_dir2_block_tail_t *btp; /* block tail pointer */
1055 xfs_dir2_data_entry_t *dep; /* data entry pointer */
1056 xfs_inode_t *dp; /* incore directory inode */
1057 int dummy; /* trash */
1058 xfs_dir2_data_unused_t *dup; /* unused entry pointer */
1059 int endoffset; /* end of data objects */
1060 int error; /* error return value */
1061 int i; /* index */
1062 xfs_mount_t *mp; /* filesystem mount point */
1063 int needlog; /* need to log block header */
1064 int needscan; /* need to scan block freespc */
1065 int newoffset; /* offset from current entry */
1066 int offset; /* target block offset */
1067 xfs_dir2_sf_entry_t *sfep; /* sf entry pointer */
1068 xfs_dir2_sf_hdr_t *oldsfp; /* old shortform header */
1069 xfs_dir2_sf_hdr_t *sfp; /* shortform header */
1070 __be16 *tagp; /* end of data entry */
1071 xfs_trans_t *tp; /* transaction pointer */
1072 struct xfs_name name;
1073 struct xfs_ifork *ifp;
1075 trace_xfs_dir2_sf_to_block(args);
1077 dp = args->dp;
1078 tp = args->trans;
1079 mp = dp->i_mount;
1080 ifp = XFS_IFORK_PTR(dp, XFS_DATA_FORK);
1081 ASSERT(ifp->if_flags & XFS_IFINLINE);
1083 * Bomb out if the shortform directory is way too short.
1085 if (dp->i_d.di_size < offsetof(xfs_dir2_sf_hdr_t, parent)) {
1086 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1087 return -EIO;
1090 oldsfp = (xfs_dir2_sf_hdr_t *)ifp->if_u1.if_data;
1092 ASSERT(ifp->if_bytes == dp->i_d.di_size);
1093 ASSERT(ifp->if_u1.if_data != NULL);
1094 ASSERT(dp->i_d.di_size >= xfs_dir2_sf_hdr_size(oldsfp->i8count));
1095 ASSERT(dp->i_d.di_nextents == 0);
1098 * Copy the directory into a temporary buffer.
1099 * Then pitch the incore inode data so we can make extents.
1101 sfp = kmem_alloc(ifp->if_bytes, KM_SLEEP);
1102 memcpy(sfp, oldsfp, ifp->if_bytes);
1104 xfs_idata_realloc(dp, -ifp->if_bytes, XFS_DATA_FORK);
1105 xfs_bmap_local_to_extents_empty(dp, XFS_DATA_FORK);
1106 dp->i_d.di_size = 0;
1109 * Add block 0 to the inode.
1111 error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE, &blkno);
1112 if (error) {
1113 kmem_free(sfp);
1114 return error;
1117 * Initialize the data block, then convert it to block format.
1119 error = xfs_dir3_data_init(args, blkno, &bp);
1120 if (error) {
1121 kmem_free(sfp);
1122 return error;
1124 xfs_dir3_block_init(mp, tp, bp, dp);
1125 hdr = bp->b_addr;
1128 * Compute size of block "tail" area.
1130 i = (uint)sizeof(*btp) +
1131 (sfp->count + 2) * (uint)sizeof(xfs_dir2_leaf_entry_t);
1133 * The whole thing is initialized to free by the init routine.
1134 * Say we're using the leaf and tail area.
1136 dup = dp->d_ops->data_unused_p(hdr);
1137 needlog = needscan = 0;
1138 xfs_dir2_data_use_free(args, bp, dup, args->geo->blksize - i,
1139 i, &needlog, &needscan);
1140 ASSERT(needscan == 0);
1142 * Fill in the tail.
1144 btp = xfs_dir2_block_tail_p(args->geo, hdr);
1145 btp->count = cpu_to_be32(sfp->count + 2); /* ., .. */
1146 btp->stale = 0;
1147 blp = xfs_dir2_block_leaf_p(btp);
1148 endoffset = (uint)((char *)blp - (char *)hdr);
1150 * Remove the freespace, we'll manage it.
1152 xfs_dir2_data_use_free(args, bp, dup,
1153 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr),
1154 be16_to_cpu(dup->length), &needlog, &needscan);
1156 * Create entry for .
1158 dep = dp->d_ops->data_dot_entry_p(hdr);
1159 dep->inumber = cpu_to_be64(dp->i_ino);
1160 dep->namelen = 1;
1161 dep->name[0] = '.';
1162 dp->d_ops->data_put_ftype(dep, XFS_DIR3_FT_DIR);
1163 tagp = dp->d_ops->data_entry_tag_p(dep);
1164 *tagp = cpu_to_be16((char *)dep - (char *)hdr);
1165 xfs_dir2_data_log_entry(args, bp, dep);
1166 blp[0].hashval = cpu_to_be32(xfs_dir_hash_dot);
1167 blp[0].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(
1168 (char *)dep - (char *)hdr));
1170 * Create entry for ..
1172 dep = dp->d_ops->data_dotdot_entry_p(hdr);
1173 dep->inumber = cpu_to_be64(dp->d_ops->sf_get_parent_ino(sfp));
1174 dep->namelen = 2;
1175 dep->name[0] = dep->name[1] = '.';
1176 dp->d_ops->data_put_ftype(dep, XFS_DIR3_FT_DIR);
1177 tagp = dp->d_ops->data_entry_tag_p(dep);
1178 *tagp = cpu_to_be16((char *)dep - (char *)hdr);
1179 xfs_dir2_data_log_entry(args, bp, dep);
1180 blp[1].hashval = cpu_to_be32(xfs_dir_hash_dotdot);
1181 blp[1].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(
1182 (char *)dep - (char *)hdr));
1183 offset = dp->d_ops->data_first_offset;
1185 * Loop over existing entries, stuff them in.
1187 i = 0;
1188 if (!sfp->count)
1189 sfep = NULL;
1190 else
1191 sfep = xfs_dir2_sf_firstentry(sfp);
1193 * Need to preserve the existing offset values in the sf directory.
1194 * Insert holes (unused entries) where necessary.
1196 while (offset < endoffset) {
1198 * sfep is null when we reach the end of the list.
1200 if (sfep == NULL)
1201 newoffset = endoffset;
1202 else
1203 newoffset = xfs_dir2_sf_get_offset(sfep);
1205 * There should be a hole here, make one.
1207 if (offset < newoffset) {
1208 dup = (xfs_dir2_data_unused_t *)((char *)hdr + offset);
1209 dup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
1210 dup->length = cpu_to_be16(newoffset - offset);
1211 *xfs_dir2_data_unused_tag_p(dup) = cpu_to_be16(
1212 ((char *)dup - (char *)hdr));
1213 xfs_dir2_data_log_unused(args, bp, dup);
1214 xfs_dir2_data_freeinsert(hdr,
1215 dp->d_ops->data_bestfree_p(hdr),
1216 dup, &dummy);
1217 offset += be16_to_cpu(dup->length);
1218 continue;
1221 * Copy a real entry.
1223 dep = (xfs_dir2_data_entry_t *)((char *)hdr + newoffset);
1224 dep->inumber = cpu_to_be64(dp->d_ops->sf_get_ino(sfp, sfep));
1225 dep->namelen = sfep->namelen;
1226 dp->d_ops->data_put_ftype(dep, dp->d_ops->sf_get_ftype(sfep));
1227 memcpy(dep->name, sfep->name, dep->namelen);
1228 tagp = dp->d_ops->data_entry_tag_p(dep);
1229 *tagp = cpu_to_be16((char *)dep - (char *)hdr);
1230 xfs_dir2_data_log_entry(args, bp, dep);
1231 name.name = sfep->name;
1232 name.len = sfep->namelen;
1233 blp[2 + i].hashval = cpu_to_be32(mp->m_dirnameops->
1234 hashname(&name));
1235 blp[2 + i].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(
1236 (char *)dep - (char *)hdr));
1237 offset = (int)((char *)(tagp + 1) - (char *)hdr);
1238 if (++i == sfp->count)
1239 sfep = NULL;
1240 else
1241 sfep = dp->d_ops->sf_nextentry(sfp, sfep);
1243 /* Done with the temporary buffer */
1244 kmem_free(sfp);
1246 * Sort the leaf entries by hash value.
1248 xfs_sort(blp, be32_to_cpu(btp->count), sizeof(*blp), xfs_dir2_block_sort);
1250 * Log the leaf entry area and tail.
1251 * Already logged the header in data_init, ignore needlog.
1253 ASSERT(needscan == 0);
1254 xfs_dir2_block_log_leaf(tp, bp, 0, be32_to_cpu(btp->count) - 1);
1255 xfs_dir2_block_log_tail(tp, bp);
1256 xfs_dir3_data_check(dp, bp);
1257 return 0;