aufs: possible bugfix, allow the nested aufsd between several aufs mounts
[zen-stable.git] / fs / xfs / xfs_dir2_node.c
blob0179a41d9e5a651543252bbf23c4981640f88d39
1 /*
2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_types.h"
21 #include "xfs_log.h"
22 #include "xfs_inum.h"
23 #include "xfs_trans.h"
24 #include "xfs_sb.h"
25 #include "xfs_ag.h"
26 #include "xfs_mount.h"
27 #include "xfs_da_btree.h"
28 #include "xfs_bmap_btree.h"
29 #include "xfs_dinode.h"
30 #include "xfs_inode.h"
31 #include "xfs_bmap.h"
32 #include "xfs_dir2_format.h"
33 #include "xfs_dir2_priv.h"
34 #include "xfs_error.h"
35 #include "xfs_trace.h"
38 * Function declarations.
40 static void xfs_dir2_free_log_header(xfs_trans_t *tp, xfs_dabuf_t *bp);
41 static int xfs_dir2_leafn_add(xfs_dabuf_t *bp, xfs_da_args_t *args, int index);
42 #ifdef DEBUG
43 static void xfs_dir2_leafn_check(xfs_inode_t *dp, xfs_dabuf_t *bp);
44 #else
45 #define xfs_dir2_leafn_check(dp, bp)
46 #endif
47 static void xfs_dir2_leafn_moveents(xfs_da_args_t *args, xfs_dabuf_t *bp_s,
48 int start_s, xfs_dabuf_t *bp_d, int start_d,
49 int count);
50 static void xfs_dir2_leafn_rebalance(xfs_da_state_t *state,
51 xfs_da_state_blk_t *blk1,
52 xfs_da_state_blk_t *blk2);
53 static int xfs_dir2_leafn_remove(xfs_da_args_t *args, xfs_dabuf_t *bp,
54 int index, xfs_da_state_blk_t *dblk,
55 int *rval);
56 static int xfs_dir2_node_addname_int(xfs_da_args_t *args,
57 xfs_da_state_blk_t *fblk);
60 * Log entries from a freespace block.
62 STATIC void
63 xfs_dir2_free_log_bests(
64 xfs_trans_t *tp, /* transaction pointer */
65 xfs_dabuf_t *bp, /* freespace buffer */
66 int first, /* first entry to log */
67 int last) /* last entry to log */
69 xfs_dir2_free_t *free; /* freespace structure */
71 free = bp->data;
72 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC));
73 xfs_da_log_buf(tp, bp,
74 (uint)((char *)&free->bests[first] - (char *)free),
75 (uint)((char *)&free->bests[last] - (char *)free +
76 sizeof(free->bests[0]) - 1));
80 * Log header from a freespace block.
82 static void
83 xfs_dir2_free_log_header(
84 xfs_trans_t *tp, /* transaction pointer */
85 xfs_dabuf_t *bp) /* freespace buffer */
87 xfs_dir2_free_t *free; /* freespace structure */
89 free = bp->data;
90 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC));
91 xfs_da_log_buf(tp, bp, (uint)((char *)&free->hdr - (char *)free),
92 (uint)(sizeof(xfs_dir2_free_hdr_t) - 1));
96 * Convert a leaf-format directory to a node-format directory.
97 * We need to change the magic number of the leaf block, and copy
98 * the freespace table out of the leaf block into its own block.
100 int /* error */
101 xfs_dir2_leaf_to_node(
102 xfs_da_args_t *args, /* operation arguments */
103 xfs_dabuf_t *lbp) /* leaf buffer */
105 xfs_inode_t *dp; /* incore directory inode */
106 int error; /* error return value */
107 xfs_dabuf_t *fbp; /* freespace buffer */
108 xfs_dir2_db_t fdb; /* freespace block number */
109 xfs_dir2_free_t *free; /* freespace structure */
110 __be16 *from; /* pointer to freespace entry */
111 int i; /* leaf freespace index */
112 xfs_dir2_leaf_t *leaf; /* leaf structure */
113 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
114 xfs_mount_t *mp; /* filesystem mount point */
115 int n; /* count of live freespc ents */
116 xfs_dir2_data_off_t off; /* freespace entry value */
117 __be16 *to; /* pointer to freespace entry */
118 xfs_trans_t *tp; /* transaction pointer */
120 trace_xfs_dir2_leaf_to_node(args);
122 dp = args->dp;
123 mp = dp->i_mount;
124 tp = args->trans;
126 * Add a freespace block to the directory.
128 if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE, &fdb))) {
129 return error;
131 ASSERT(fdb == XFS_DIR2_FREE_FIRSTDB(mp));
133 * Get the buffer for the new freespace block.
135 if ((error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(mp, fdb), -1, &fbp,
136 XFS_DATA_FORK))) {
137 return error;
139 ASSERT(fbp != NULL);
140 free = fbp->data;
141 leaf = lbp->data;
142 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
144 * Initialize the freespace block header.
146 free->hdr.magic = cpu_to_be32(XFS_DIR2_FREE_MAGIC);
147 free->hdr.firstdb = 0;
148 ASSERT(be32_to_cpu(ltp->bestcount) <= (uint)dp->i_d.di_size / mp->m_dirblksize);
149 free->hdr.nvalid = ltp->bestcount;
151 * Copy freespace entries from the leaf block to the new block.
152 * Count active entries.
154 for (i = n = 0, from = xfs_dir2_leaf_bests_p(ltp), to = free->bests;
155 i < be32_to_cpu(ltp->bestcount); i++, from++, to++) {
156 if ((off = be16_to_cpu(*from)) != NULLDATAOFF)
157 n++;
158 *to = cpu_to_be16(off);
160 free->hdr.nused = cpu_to_be32(n);
161 leaf->hdr.info.magic = cpu_to_be16(XFS_DIR2_LEAFN_MAGIC);
163 * Log everything.
165 xfs_dir2_leaf_log_header(tp, lbp);
166 xfs_dir2_free_log_header(tp, fbp);
167 xfs_dir2_free_log_bests(tp, fbp, 0, be32_to_cpu(free->hdr.nvalid) - 1);
168 xfs_da_buf_done(fbp);
169 xfs_dir2_leafn_check(dp, lbp);
170 return 0;
174 * Add a leaf entry to a leaf block in a node-form directory.
175 * The other work necessary is done from the caller.
177 static int /* error */
178 xfs_dir2_leafn_add(
179 xfs_dabuf_t *bp, /* leaf buffer */
180 xfs_da_args_t *args, /* operation arguments */
181 int index) /* insertion pt for new entry */
183 int compact; /* compacting stale leaves */
184 xfs_inode_t *dp; /* incore directory inode */
185 int highstale; /* next stale entry */
186 xfs_dir2_leaf_t *leaf; /* leaf structure */
187 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
188 int lfloghigh; /* high leaf entry logging */
189 int lfloglow; /* low leaf entry logging */
190 int lowstale; /* previous stale entry */
191 xfs_mount_t *mp; /* filesystem mount point */
192 xfs_trans_t *tp; /* transaction pointer */
194 trace_xfs_dir2_leafn_add(args, index);
196 dp = args->dp;
197 mp = dp->i_mount;
198 tp = args->trans;
199 leaf = bp->data;
202 * Quick check just to make sure we are not going to index
203 * into other peoples memory
205 if (index < 0)
206 return XFS_ERROR(EFSCORRUPTED);
209 * If there are already the maximum number of leaf entries in
210 * the block, if there are no stale entries it won't fit.
211 * Caller will do a split. If there are stale entries we'll do
212 * a compact.
215 if (be16_to_cpu(leaf->hdr.count) == xfs_dir2_max_leaf_ents(mp)) {
216 if (!leaf->hdr.stale)
217 return XFS_ERROR(ENOSPC);
218 compact = be16_to_cpu(leaf->hdr.stale) > 1;
219 } else
220 compact = 0;
221 ASSERT(index == 0 || be32_to_cpu(leaf->ents[index - 1].hashval) <= args->hashval);
222 ASSERT(index == be16_to_cpu(leaf->hdr.count) ||
223 be32_to_cpu(leaf->ents[index].hashval) >= args->hashval);
225 if (args->op_flags & XFS_DA_OP_JUSTCHECK)
226 return 0;
229 * Compact out all but one stale leaf entry. Leaves behind
230 * the entry closest to index.
232 if (compact) {
233 xfs_dir2_leaf_compact_x1(bp, &index, &lowstale, &highstale,
234 &lfloglow, &lfloghigh);
237 * Set impossible logging indices for this case.
239 else if (leaf->hdr.stale) {
240 lfloglow = be16_to_cpu(leaf->hdr.count);
241 lfloghigh = -1;
245 * Insert the new entry, log everything.
247 lep = xfs_dir2_leaf_find_entry(leaf, index, compact, lowstale,
248 highstale, &lfloglow, &lfloghigh);
250 lep->hashval = cpu_to_be32(args->hashval);
251 lep->address = cpu_to_be32(xfs_dir2_db_off_to_dataptr(mp,
252 args->blkno, args->index));
253 xfs_dir2_leaf_log_header(tp, bp);
254 xfs_dir2_leaf_log_ents(tp, bp, lfloglow, lfloghigh);
255 xfs_dir2_leafn_check(dp, bp);
256 return 0;
259 #ifdef DEBUG
261 * Check internal consistency of a leafn block.
263 void
264 xfs_dir2_leafn_check(
265 xfs_inode_t *dp, /* incore directory inode */
266 xfs_dabuf_t *bp) /* leaf buffer */
268 int i; /* leaf index */
269 xfs_dir2_leaf_t *leaf; /* leaf structure */
270 xfs_mount_t *mp; /* filesystem mount point */
271 int stale; /* count of stale leaves */
273 leaf = bp->data;
274 mp = dp->i_mount;
275 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
276 ASSERT(be16_to_cpu(leaf->hdr.count) <= xfs_dir2_max_leaf_ents(mp));
277 for (i = stale = 0; i < be16_to_cpu(leaf->hdr.count); i++) {
278 if (i + 1 < be16_to_cpu(leaf->hdr.count)) {
279 ASSERT(be32_to_cpu(leaf->ents[i].hashval) <=
280 be32_to_cpu(leaf->ents[i + 1].hashval));
282 if (leaf->ents[i].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
283 stale++;
285 ASSERT(be16_to_cpu(leaf->hdr.stale) == stale);
287 #endif /* DEBUG */
290 * Return the last hash value in the leaf.
291 * Stale entries are ok.
293 xfs_dahash_t /* hash value */
294 xfs_dir2_leafn_lasthash(
295 xfs_dabuf_t *bp, /* leaf buffer */
296 int *count) /* count of entries in leaf */
298 xfs_dir2_leaf_t *leaf; /* leaf structure */
300 leaf = bp->data;
301 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
302 if (count)
303 *count = be16_to_cpu(leaf->hdr.count);
304 if (!leaf->hdr.count)
305 return 0;
306 return be32_to_cpu(leaf->ents[be16_to_cpu(leaf->hdr.count) - 1].hashval);
310 * Look up a leaf entry for space to add a name in a node-format leaf block.
311 * The extrablk in state is a freespace block.
313 STATIC int
314 xfs_dir2_leafn_lookup_for_addname(
315 xfs_dabuf_t *bp, /* leaf buffer */
316 xfs_da_args_t *args, /* operation arguments */
317 int *indexp, /* out: leaf entry index */
318 xfs_da_state_t *state) /* state to fill in */
320 xfs_dabuf_t *curbp = NULL; /* current data/free buffer */
321 xfs_dir2_db_t curdb = -1; /* current data block number */
322 xfs_dir2_db_t curfdb = -1; /* current free block number */
323 xfs_inode_t *dp; /* incore directory inode */
324 int error; /* error return value */
325 int fi; /* free entry index */
326 xfs_dir2_free_t *free = NULL; /* free block structure */
327 int index; /* leaf entry index */
328 xfs_dir2_leaf_t *leaf; /* leaf structure */
329 int length; /* length of new data entry */
330 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
331 xfs_mount_t *mp; /* filesystem mount point */
332 xfs_dir2_db_t newdb; /* new data block number */
333 xfs_dir2_db_t newfdb; /* new free block number */
334 xfs_trans_t *tp; /* transaction pointer */
336 dp = args->dp;
337 tp = args->trans;
338 mp = dp->i_mount;
339 leaf = bp->data;
340 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
341 #ifdef __KERNEL__
342 ASSERT(be16_to_cpu(leaf->hdr.count) > 0);
343 #endif
344 xfs_dir2_leafn_check(dp, bp);
346 * Look up the hash value in the leaf entries.
348 index = xfs_dir2_leaf_search_hash(args, bp);
350 * Do we have a buffer coming in?
352 if (state->extravalid) {
353 /* If so, it's a free block buffer, get the block number. */
354 curbp = state->extrablk.bp;
355 curfdb = state->extrablk.blkno;
356 free = curbp->data;
357 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC));
359 length = xfs_dir2_data_entsize(args->namelen);
361 * Loop over leaf entries with the right hash value.
363 for (lep = &leaf->ents[index]; index < be16_to_cpu(leaf->hdr.count) &&
364 be32_to_cpu(lep->hashval) == args->hashval;
365 lep++, index++) {
367 * Skip stale leaf entries.
369 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
370 continue;
372 * Pull the data block number from the entry.
374 newdb = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
376 * For addname, we're looking for a place to put the new entry.
377 * We want to use a data block with an entry of equal
378 * hash value to ours if there is one with room.
380 * If this block isn't the data block we already have
381 * in hand, take a look at it.
383 if (newdb != curdb) {
384 curdb = newdb;
386 * Convert the data block to the free block
387 * holding its freespace information.
389 newfdb = xfs_dir2_db_to_fdb(mp, newdb);
391 * If it's not the one we have in hand, read it in.
393 if (newfdb != curfdb) {
395 * If we had one before, drop it.
397 if (curbp)
398 xfs_da_brelse(tp, curbp);
400 * Read the free block.
402 error = xfs_da_read_buf(tp, dp,
403 xfs_dir2_db_to_da(mp, newfdb),
404 -1, &curbp, XFS_DATA_FORK);
405 if (error)
406 return error;
407 free = curbp->data;
408 ASSERT(be32_to_cpu(free->hdr.magic) ==
409 XFS_DIR2_FREE_MAGIC);
410 ASSERT((be32_to_cpu(free->hdr.firstdb) %
411 xfs_dir2_free_max_bests(mp)) == 0);
412 ASSERT(be32_to_cpu(free->hdr.firstdb) <= curdb);
413 ASSERT(curdb < be32_to_cpu(free->hdr.firstdb) +
414 be32_to_cpu(free->hdr.nvalid));
417 * Get the index for our entry.
419 fi = xfs_dir2_db_to_fdindex(mp, curdb);
421 * If it has room, return it.
423 if (unlikely(free->bests[fi] ==
424 cpu_to_be16(NULLDATAOFF))) {
425 XFS_ERROR_REPORT("xfs_dir2_leafn_lookup_int",
426 XFS_ERRLEVEL_LOW, mp);
427 if (curfdb != newfdb)
428 xfs_da_brelse(tp, curbp);
429 return XFS_ERROR(EFSCORRUPTED);
431 curfdb = newfdb;
432 if (be16_to_cpu(free->bests[fi]) >= length)
433 goto out;
436 /* Didn't find any space */
437 fi = -1;
438 out:
439 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
440 if (curbp) {
441 /* Giving back a free block. */
442 state->extravalid = 1;
443 state->extrablk.bp = curbp;
444 state->extrablk.index = fi;
445 state->extrablk.blkno = curfdb;
446 state->extrablk.magic = XFS_DIR2_FREE_MAGIC;
447 } else {
448 state->extravalid = 0;
451 * Return the index, that will be the insertion point.
453 *indexp = index;
454 return XFS_ERROR(ENOENT);
458 * Look up a leaf entry in a node-format leaf block.
459 * The extrablk in state a data block.
461 STATIC int
462 xfs_dir2_leafn_lookup_for_entry(
463 xfs_dabuf_t *bp, /* leaf buffer */
464 xfs_da_args_t *args, /* operation arguments */
465 int *indexp, /* out: leaf entry index */
466 xfs_da_state_t *state) /* state to fill in */
468 xfs_dabuf_t *curbp = NULL; /* current data/free buffer */
469 xfs_dir2_db_t curdb = -1; /* current data block number */
470 xfs_dir2_data_entry_t *dep; /* data block entry */
471 xfs_inode_t *dp; /* incore directory inode */
472 int error; /* error return value */
473 int index; /* leaf entry index */
474 xfs_dir2_leaf_t *leaf; /* leaf structure */
475 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
476 xfs_mount_t *mp; /* filesystem mount point */
477 xfs_dir2_db_t newdb; /* new data block number */
478 xfs_trans_t *tp; /* transaction pointer */
479 enum xfs_dacmp cmp; /* comparison result */
481 dp = args->dp;
482 tp = args->trans;
483 mp = dp->i_mount;
484 leaf = bp->data;
485 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
486 #ifdef __KERNEL__
487 ASSERT(be16_to_cpu(leaf->hdr.count) > 0);
488 #endif
489 xfs_dir2_leafn_check(dp, bp);
491 * Look up the hash value in the leaf entries.
493 index = xfs_dir2_leaf_search_hash(args, bp);
495 * Do we have a buffer coming in?
497 if (state->extravalid) {
498 curbp = state->extrablk.bp;
499 curdb = state->extrablk.blkno;
502 * Loop over leaf entries with the right hash value.
504 for (lep = &leaf->ents[index]; index < be16_to_cpu(leaf->hdr.count) &&
505 be32_to_cpu(lep->hashval) == args->hashval;
506 lep++, index++) {
508 * Skip stale leaf entries.
510 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
511 continue;
513 * Pull the data block number from the entry.
515 newdb = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
517 * Not adding a new entry, so we really want to find
518 * the name given to us.
520 * If it's a different data block, go get it.
522 if (newdb != curdb) {
524 * If we had a block before that we aren't saving
525 * for a CI name, drop it
527 if (curbp && (args->cmpresult == XFS_CMP_DIFFERENT ||
528 curdb != state->extrablk.blkno))
529 xfs_da_brelse(tp, curbp);
531 * If needing the block that is saved with a CI match,
532 * use it otherwise read in the new data block.
534 if (args->cmpresult != XFS_CMP_DIFFERENT &&
535 newdb == state->extrablk.blkno) {
536 ASSERT(state->extravalid);
537 curbp = state->extrablk.bp;
538 } else {
539 error = xfs_da_read_buf(tp, dp,
540 xfs_dir2_db_to_da(mp, newdb),
541 -1, &curbp, XFS_DATA_FORK);
542 if (error)
543 return error;
545 xfs_dir2_data_check(dp, curbp);
546 curdb = newdb;
549 * Point to the data entry.
551 dep = (xfs_dir2_data_entry_t *)((char *)curbp->data +
552 xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address)));
554 * Compare the entry and if it's an exact match, return
555 * EEXIST immediately. If it's the first case-insensitive
556 * match, store the block & inode number and continue looking.
558 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
559 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
560 /* If there is a CI match block, drop it */
561 if (args->cmpresult != XFS_CMP_DIFFERENT &&
562 curdb != state->extrablk.blkno)
563 xfs_da_brelse(tp, state->extrablk.bp);
564 args->cmpresult = cmp;
565 args->inumber = be64_to_cpu(dep->inumber);
566 *indexp = index;
567 state->extravalid = 1;
568 state->extrablk.bp = curbp;
569 state->extrablk.blkno = curdb;
570 state->extrablk.index = (int)((char *)dep -
571 (char *)curbp->data);
572 state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
573 if (cmp == XFS_CMP_EXACT)
574 return XFS_ERROR(EEXIST);
577 ASSERT(index == be16_to_cpu(leaf->hdr.count) ||
578 (args->op_flags & XFS_DA_OP_OKNOENT));
579 if (curbp) {
580 if (args->cmpresult == XFS_CMP_DIFFERENT) {
581 /* Giving back last used data block. */
582 state->extravalid = 1;
583 state->extrablk.bp = curbp;
584 state->extrablk.index = -1;
585 state->extrablk.blkno = curdb;
586 state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
587 } else {
588 /* If the curbp is not the CI match block, drop it */
589 if (state->extrablk.bp != curbp)
590 xfs_da_brelse(tp, curbp);
592 } else {
593 state->extravalid = 0;
595 *indexp = index;
596 return XFS_ERROR(ENOENT);
600 * Look up a leaf entry in a node-format leaf block.
601 * If this is an addname then the extrablk in state is a freespace block,
602 * otherwise it's a data block.
605 xfs_dir2_leafn_lookup_int(
606 xfs_dabuf_t *bp, /* leaf buffer */
607 xfs_da_args_t *args, /* operation arguments */
608 int *indexp, /* out: leaf entry index */
609 xfs_da_state_t *state) /* state to fill in */
611 if (args->op_flags & XFS_DA_OP_ADDNAME)
612 return xfs_dir2_leafn_lookup_for_addname(bp, args, indexp,
613 state);
614 return xfs_dir2_leafn_lookup_for_entry(bp, args, indexp, state);
618 * Move count leaf entries from source to destination leaf.
619 * Log entries and headers. Stale entries are preserved.
621 static void
622 xfs_dir2_leafn_moveents(
623 xfs_da_args_t *args, /* operation arguments */
624 xfs_dabuf_t *bp_s, /* source leaf buffer */
625 int start_s, /* source leaf index */
626 xfs_dabuf_t *bp_d, /* destination leaf buffer */
627 int start_d, /* destination leaf index */
628 int count) /* count of leaves to copy */
630 xfs_dir2_leaf_t *leaf_d; /* destination leaf structure */
631 xfs_dir2_leaf_t *leaf_s; /* source leaf structure */
632 int stale; /* count stale leaves copied */
633 xfs_trans_t *tp; /* transaction pointer */
635 trace_xfs_dir2_leafn_moveents(args, start_s, start_d, count);
638 * Silently return if nothing to do.
640 if (count == 0) {
641 return;
643 tp = args->trans;
644 leaf_s = bp_s->data;
645 leaf_d = bp_d->data;
647 * If the destination index is not the end of the current
648 * destination leaf entries, open up a hole in the destination
649 * to hold the new entries.
651 if (start_d < be16_to_cpu(leaf_d->hdr.count)) {
652 memmove(&leaf_d->ents[start_d + count], &leaf_d->ents[start_d],
653 (be16_to_cpu(leaf_d->hdr.count) - start_d) *
654 sizeof(xfs_dir2_leaf_entry_t));
655 xfs_dir2_leaf_log_ents(tp, bp_d, start_d + count,
656 count + be16_to_cpu(leaf_d->hdr.count) - 1);
659 * If the source has stale leaves, count the ones in the copy range
660 * so we can update the header correctly.
662 if (leaf_s->hdr.stale) {
663 int i; /* temp leaf index */
665 for (i = start_s, stale = 0; i < start_s + count; i++) {
666 if (leaf_s->ents[i].address ==
667 cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
668 stale++;
670 } else
671 stale = 0;
673 * Copy the leaf entries from source to destination.
675 memcpy(&leaf_d->ents[start_d], &leaf_s->ents[start_s],
676 count * sizeof(xfs_dir2_leaf_entry_t));
677 xfs_dir2_leaf_log_ents(tp, bp_d, start_d, start_d + count - 1);
679 * If there are source entries after the ones we copied,
680 * delete the ones we copied by sliding the next ones down.
682 if (start_s + count < be16_to_cpu(leaf_s->hdr.count)) {
683 memmove(&leaf_s->ents[start_s], &leaf_s->ents[start_s + count],
684 count * sizeof(xfs_dir2_leaf_entry_t));
685 xfs_dir2_leaf_log_ents(tp, bp_s, start_s, start_s + count - 1);
688 * Update the headers and log them.
690 be16_add_cpu(&leaf_s->hdr.count, -(count));
691 be16_add_cpu(&leaf_s->hdr.stale, -(stale));
692 be16_add_cpu(&leaf_d->hdr.count, count);
693 be16_add_cpu(&leaf_d->hdr.stale, stale);
694 xfs_dir2_leaf_log_header(tp, bp_s);
695 xfs_dir2_leaf_log_header(tp, bp_d);
696 xfs_dir2_leafn_check(args->dp, bp_s);
697 xfs_dir2_leafn_check(args->dp, bp_d);
701 * Determine the sort order of two leaf blocks.
702 * Returns 1 if both are valid and leaf2 should be before leaf1, else 0.
704 int /* sort order */
705 xfs_dir2_leafn_order(
706 xfs_dabuf_t *leaf1_bp, /* leaf1 buffer */
707 xfs_dabuf_t *leaf2_bp) /* leaf2 buffer */
709 xfs_dir2_leaf_t *leaf1; /* leaf1 structure */
710 xfs_dir2_leaf_t *leaf2; /* leaf2 structure */
712 leaf1 = leaf1_bp->data;
713 leaf2 = leaf2_bp->data;
714 ASSERT(leaf1->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
715 ASSERT(leaf2->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
716 if (be16_to_cpu(leaf1->hdr.count) > 0 &&
717 be16_to_cpu(leaf2->hdr.count) > 0 &&
718 (be32_to_cpu(leaf2->ents[0].hashval) < be32_to_cpu(leaf1->ents[0].hashval) ||
719 be32_to_cpu(leaf2->ents[be16_to_cpu(leaf2->hdr.count) - 1].hashval) <
720 be32_to_cpu(leaf1->ents[be16_to_cpu(leaf1->hdr.count) - 1].hashval)))
721 return 1;
722 return 0;
726 * Rebalance leaf entries between two leaf blocks.
727 * This is actually only called when the second block is new,
728 * though the code deals with the general case.
729 * A new entry will be inserted in one of the blocks, and that
730 * entry is taken into account when balancing.
732 static void
733 xfs_dir2_leafn_rebalance(
734 xfs_da_state_t *state, /* btree cursor */
735 xfs_da_state_blk_t *blk1, /* first btree block */
736 xfs_da_state_blk_t *blk2) /* second btree block */
738 xfs_da_args_t *args; /* operation arguments */
739 int count; /* count (& direction) leaves */
740 int isleft; /* new goes in left leaf */
741 xfs_dir2_leaf_t *leaf1; /* first leaf structure */
742 xfs_dir2_leaf_t *leaf2; /* second leaf structure */
743 int mid; /* midpoint leaf index */
744 #ifdef DEBUG
745 int oldstale; /* old count of stale leaves */
746 #endif
747 int oldsum; /* old total leaf count */
748 int swap; /* swapped leaf blocks */
750 args = state->args;
752 * If the block order is wrong, swap the arguments.
754 if ((swap = xfs_dir2_leafn_order(blk1->bp, blk2->bp))) {
755 xfs_da_state_blk_t *tmp; /* temp for block swap */
757 tmp = blk1;
758 blk1 = blk2;
759 blk2 = tmp;
761 leaf1 = blk1->bp->data;
762 leaf2 = blk2->bp->data;
763 oldsum = be16_to_cpu(leaf1->hdr.count) + be16_to_cpu(leaf2->hdr.count);
764 #ifdef DEBUG
765 oldstale = be16_to_cpu(leaf1->hdr.stale) + be16_to_cpu(leaf2->hdr.stale);
766 #endif
767 mid = oldsum >> 1;
769 * If the old leaf count was odd then the new one will be even,
770 * so we need to divide the new count evenly.
772 if (oldsum & 1) {
773 xfs_dahash_t midhash; /* middle entry hash value */
775 if (mid >= be16_to_cpu(leaf1->hdr.count))
776 midhash = be32_to_cpu(leaf2->ents[mid - be16_to_cpu(leaf1->hdr.count)].hashval);
777 else
778 midhash = be32_to_cpu(leaf1->ents[mid].hashval);
779 isleft = args->hashval <= midhash;
782 * If the old count is even then the new count is odd, so there's
783 * no preferred side for the new entry.
784 * Pick the left one.
786 else
787 isleft = 1;
789 * Calculate moved entry count. Positive means left-to-right,
790 * negative means right-to-left. Then move the entries.
792 count = be16_to_cpu(leaf1->hdr.count) - mid + (isleft == 0);
793 if (count > 0)
794 xfs_dir2_leafn_moveents(args, blk1->bp,
795 be16_to_cpu(leaf1->hdr.count) - count, blk2->bp, 0, count);
796 else if (count < 0)
797 xfs_dir2_leafn_moveents(args, blk2->bp, 0, blk1->bp,
798 be16_to_cpu(leaf1->hdr.count), count);
799 ASSERT(be16_to_cpu(leaf1->hdr.count) + be16_to_cpu(leaf2->hdr.count) == oldsum);
800 ASSERT(be16_to_cpu(leaf1->hdr.stale) + be16_to_cpu(leaf2->hdr.stale) == oldstale);
802 * Mark whether we're inserting into the old or new leaf.
804 if (be16_to_cpu(leaf1->hdr.count) < be16_to_cpu(leaf2->hdr.count))
805 state->inleaf = swap;
806 else if (be16_to_cpu(leaf1->hdr.count) > be16_to_cpu(leaf2->hdr.count))
807 state->inleaf = !swap;
808 else
809 state->inleaf =
810 swap ^ (blk1->index <= be16_to_cpu(leaf1->hdr.count));
812 * Adjust the expected index for insertion.
814 if (!state->inleaf)
815 blk2->index = blk1->index - be16_to_cpu(leaf1->hdr.count);
818 * Finally sanity check just to make sure we are not returning a
819 * negative index
821 if(blk2->index < 0) {
822 state->inleaf = 1;
823 blk2->index = 0;
824 xfs_alert(args->dp->i_mount,
825 "%s: picked the wrong leaf? reverting original leaf: blk1->index %d\n",
826 __func__, blk1->index);
831 * Remove an entry from a node directory.
832 * This removes the leaf entry and the data entry,
833 * and updates the free block if necessary.
835 static int /* error */
836 xfs_dir2_leafn_remove(
837 xfs_da_args_t *args, /* operation arguments */
838 xfs_dabuf_t *bp, /* leaf buffer */
839 int index, /* leaf entry index */
840 xfs_da_state_blk_t *dblk, /* data block */
841 int *rval) /* resulting block needs join */
843 xfs_dir2_data_hdr_t *hdr; /* data block header */
844 xfs_dir2_db_t db; /* data block number */
845 xfs_dabuf_t *dbp; /* data block buffer */
846 xfs_dir2_data_entry_t *dep; /* data block entry */
847 xfs_inode_t *dp; /* incore directory inode */
848 xfs_dir2_leaf_t *leaf; /* leaf structure */
849 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
850 int longest; /* longest data free entry */
851 int off; /* data block entry offset */
852 xfs_mount_t *mp; /* filesystem mount point */
853 int needlog; /* need to log data header */
854 int needscan; /* need to rescan data frees */
855 xfs_trans_t *tp; /* transaction pointer */
857 trace_xfs_dir2_leafn_remove(args, index);
859 dp = args->dp;
860 tp = args->trans;
861 mp = dp->i_mount;
862 leaf = bp->data;
863 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
865 * Point to the entry we're removing.
867 lep = &leaf->ents[index];
869 * Extract the data block and offset from the entry.
871 db = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
872 ASSERT(dblk->blkno == db);
873 off = xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address));
874 ASSERT(dblk->index == off);
876 * Kill the leaf entry by marking it stale.
877 * Log the leaf block changes.
879 be16_add_cpu(&leaf->hdr.stale, 1);
880 xfs_dir2_leaf_log_header(tp, bp);
881 lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
882 xfs_dir2_leaf_log_ents(tp, bp, index, index);
884 * Make the data entry free. Keep track of the longest freespace
885 * in the data block in case it changes.
887 dbp = dblk->bp;
888 hdr = dbp->data;
889 dep = (xfs_dir2_data_entry_t *)((char *)hdr + off);
890 longest = be16_to_cpu(hdr->bestfree[0].length);
891 needlog = needscan = 0;
892 xfs_dir2_data_make_free(tp, dbp, off,
893 xfs_dir2_data_entsize(dep->namelen), &needlog, &needscan);
895 * Rescan the data block freespaces for bestfree.
896 * Log the data block header if needed.
898 if (needscan)
899 xfs_dir2_data_freescan(mp, hdr, &needlog);
900 if (needlog)
901 xfs_dir2_data_log_header(tp, dbp);
902 xfs_dir2_data_check(dp, dbp);
904 * If the longest data block freespace changes, need to update
905 * the corresponding freeblock entry.
907 if (longest < be16_to_cpu(hdr->bestfree[0].length)) {
908 int error; /* error return value */
909 xfs_dabuf_t *fbp; /* freeblock buffer */
910 xfs_dir2_db_t fdb; /* freeblock block number */
911 int findex; /* index in freeblock entries */
912 xfs_dir2_free_t *free; /* freeblock structure */
913 int logfree; /* need to log free entry */
916 * Convert the data block number to a free block,
917 * read in the free block.
919 fdb = xfs_dir2_db_to_fdb(mp, db);
920 if ((error = xfs_da_read_buf(tp, dp, xfs_dir2_db_to_da(mp, fdb),
921 -1, &fbp, XFS_DATA_FORK))) {
922 return error;
924 free = fbp->data;
925 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC));
926 ASSERT(be32_to_cpu(free->hdr.firstdb) ==
927 xfs_dir2_free_max_bests(mp) *
928 (fdb - XFS_DIR2_FREE_FIRSTDB(mp)));
930 * Calculate which entry we need to fix.
932 findex = xfs_dir2_db_to_fdindex(mp, db);
933 longest = be16_to_cpu(hdr->bestfree[0].length);
935 * If the data block is now empty we can get rid of it
936 * (usually).
938 if (longest == mp->m_dirblksize - (uint)sizeof(*hdr)) {
940 * Try to punch out the data block.
942 error = xfs_dir2_shrink_inode(args, db, dbp);
943 if (error == 0) {
944 dblk->bp = NULL;
945 hdr = NULL;
948 * We can get ENOSPC if there's no space reservation.
949 * In this case just drop the buffer and some one else
950 * will eventually get rid of the empty block.
952 else if (error == ENOSPC && args->total == 0)
953 xfs_da_buf_done(dbp);
954 else
955 return error;
958 * If we got rid of the data block, we can eliminate that entry
959 * in the free block.
961 if (hdr == NULL) {
963 * One less used entry in the free table.
965 be32_add_cpu(&free->hdr.nused, -1);
966 xfs_dir2_free_log_header(tp, fbp);
968 * If this was the last entry in the table, we can
969 * trim the table size back. There might be other
970 * entries at the end referring to non-existent
971 * data blocks, get those too.
973 if (findex == be32_to_cpu(free->hdr.nvalid) - 1) {
974 int i; /* free entry index */
976 for (i = findex - 1;
977 i >= 0 &&
978 free->bests[i] == cpu_to_be16(NULLDATAOFF);
979 i--)
980 continue;
981 free->hdr.nvalid = cpu_to_be32(i + 1);
982 logfree = 0;
985 * Not the last entry, just punch it out.
987 else {
988 free->bests[findex] = cpu_to_be16(NULLDATAOFF);
989 logfree = 1;
992 * If there are no useful entries left in the block,
993 * get rid of the block if we can.
995 if (!free->hdr.nused) {
996 error = xfs_dir2_shrink_inode(args, fdb, fbp);
997 if (error == 0) {
998 fbp = NULL;
999 logfree = 0;
1000 } else if (error != ENOSPC || args->total != 0)
1001 return error;
1003 * It's possible to get ENOSPC if there is no
1004 * space reservation. In this case some one
1005 * else will eventually get rid of this block.
1010 * Data block is not empty, just set the free entry to
1011 * the new value.
1013 else {
1014 free->bests[findex] = cpu_to_be16(longest);
1015 logfree = 1;
1018 * Log the free entry that changed, unless we got rid of it.
1020 if (logfree)
1021 xfs_dir2_free_log_bests(tp, fbp, findex, findex);
1023 * Drop the buffer if we still have it.
1025 if (fbp)
1026 xfs_da_buf_done(fbp);
1028 xfs_dir2_leafn_check(dp, bp);
1030 * Return indication of whether this leaf block is empty enough
1031 * to justify trying to join it with a neighbor.
1033 *rval =
1034 ((uint)sizeof(leaf->hdr) +
1035 (uint)sizeof(leaf->ents[0]) *
1036 (be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale))) <
1037 mp->m_dir_magicpct;
1038 return 0;
1042 * Split the leaf entries in the old block into old and new blocks.
1044 int /* error */
1045 xfs_dir2_leafn_split(
1046 xfs_da_state_t *state, /* btree cursor */
1047 xfs_da_state_blk_t *oldblk, /* original block */
1048 xfs_da_state_blk_t *newblk) /* newly created block */
1050 xfs_da_args_t *args; /* operation arguments */
1051 xfs_dablk_t blkno; /* new leaf block number */
1052 int error; /* error return value */
1053 xfs_mount_t *mp; /* filesystem mount point */
1056 * Allocate space for a new leaf node.
1058 args = state->args;
1059 mp = args->dp->i_mount;
1060 ASSERT(args != NULL);
1061 ASSERT(oldblk->magic == XFS_DIR2_LEAFN_MAGIC);
1062 error = xfs_da_grow_inode(args, &blkno);
1063 if (error) {
1064 return error;
1067 * Initialize the new leaf block.
1069 error = xfs_dir2_leaf_init(args, xfs_dir2_da_to_db(mp, blkno),
1070 &newblk->bp, XFS_DIR2_LEAFN_MAGIC);
1071 if (error) {
1072 return error;
1074 newblk->blkno = blkno;
1075 newblk->magic = XFS_DIR2_LEAFN_MAGIC;
1077 * Rebalance the entries across the two leaves, link the new
1078 * block into the leaves.
1080 xfs_dir2_leafn_rebalance(state, oldblk, newblk);
1081 error = xfs_da_blk_link(state, oldblk, newblk);
1082 if (error) {
1083 return error;
1086 * Insert the new entry in the correct block.
1088 if (state->inleaf)
1089 error = xfs_dir2_leafn_add(oldblk->bp, args, oldblk->index);
1090 else
1091 error = xfs_dir2_leafn_add(newblk->bp, args, newblk->index);
1093 * Update last hashval in each block since we added the name.
1095 oldblk->hashval = xfs_dir2_leafn_lasthash(oldblk->bp, NULL);
1096 newblk->hashval = xfs_dir2_leafn_lasthash(newblk->bp, NULL);
1097 xfs_dir2_leafn_check(args->dp, oldblk->bp);
1098 xfs_dir2_leafn_check(args->dp, newblk->bp);
1099 return error;
1103 * Check a leaf block and its neighbors to see if the block should be
1104 * collapsed into one or the other neighbor. Always keep the block
1105 * with the smaller block number.
1106 * If the current block is over 50% full, don't try to join it, return 0.
1107 * If the block is empty, fill in the state structure and return 2.
1108 * If it can be collapsed, fill in the state structure and return 1.
1109 * If nothing can be done, return 0.
1111 int /* error */
1112 xfs_dir2_leafn_toosmall(
1113 xfs_da_state_t *state, /* btree cursor */
1114 int *action) /* resulting action to take */
1116 xfs_da_state_blk_t *blk; /* leaf block */
1117 xfs_dablk_t blkno; /* leaf block number */
1118 xfs_dabuf_t *bp; /* leaf buffer */
1119 int bytes; /* bytes in use */
1120 int count; /* leaf live entry count */
1121 int error; /* error return value */
1122 int forward; /* sibling block direction */
1123 int i; /* sibling counter */
1124 xfs_da_blkinfo_t *info; /* leaf block header */
1125 xfs_dir2_leaf_t *leaf; /* leaf structure */
1126 int rval; /* result from path_shift */
1129 * Check for the degenerate case of the block being over 50% full.
1130 * If so, it's not worth even looking to see if we might be able
1131 * to coalesce with a sibling.
1133 blk = &state->path.blk[state->path.active - 1];
1134 info = blk->bp->data;
1135 ASSERT(info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
1136 leaf = (xfs_dir2_leaf_t *)info;
1137 count = be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale);
1138 bytes = (uint)sizeof(leaf->hdr) + count * (uint)sizeof(leaf->ents[0]);
1139 if (bytes > (state->blocksize >> 1)) {
1141 * Blk over 50%, don't try to join.
1143 *action = 0;
1144 return 0;
1147 * Check for the degenerate case of the block being empty.
1148 * If the block is empty, we'll simply delete it, no need to
1149 * coalesce it with a sibling block. We choose (arbitrarily)
1150 * to merge with the forward block unless it is NULL.
1152 if (count == 0) {
1154 * Make altpath point to the block we want to keep and
1155 * path point to the block we want to drop (this one).
1157 forward = (info->forw != 0);
1158 memcpy(&state->altpath, &state->path, sizeof(state->path));
1159 error = xfs_da_path_shift(state, &state->altpath, forward, 0,
1160 &rval);
1161 if (error)
1162 return error;
1163 *action = rval ? 2 : 0;
1164 return 0;
1167 * Examine each sibling block to see if we can coalesce with
1168 * at least 25% free space to spare. We need to figure out
1169 * whether to merge with the forward or the backward block.
1170 * We prefer coalescing with the lower numbered sibling so as
1171 * to shrink a directory over time.
1173 forward = be32_to_cpu(info->forw) < be32_to_cpu(info->back);
1174 for (i = 0, bp = NULL; i < 2; forward = !forward, i++) {
1175 blkno = forward ? be32_to_cpu(info->forw) : be32_to_cpu(info->back);
1176 if (blkno == 0)
1177 continue;
1179 * Read the sibling leaf block.
1181 if ((error =
1182 xfs_da_read_buf(state->args->trans, state->args->dp, blkno,
1183 -1, &bp, XFS_DATA_FORK))) {
1184 return error;
1186 ASSERT(bp != NULL);
1188 * Count bytes in the two blocks combined.
1190 leaf = (xfs_dir2_leaf_t *)info;
1191 count = be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale);
1192 bytes = state->blocksize - (state->blocksize >> 2);
1193 leaf = bp->data;
1194 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
1195 count += be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale);
1196 bytes -= count * (uint)sizeof(leaf->ents[0]);
1198 * Fits with at least 25% to spare.
1200 if (bytes >= 0)
1201 break;
1202 xfs_da_brelse(state->args->trans, bp);
1205 * Didn't like either block, give up.
1207 if (i >= 2) {
1208 *action = 0;
1209 return 0;
1212 * Done with the sibling leaf block here, drop the dabuf
1213 * so path_shift can get it.
1215 xfs_da_buf_done(bp);
1217 * Make altpath point to the block we want to keep (the lower
1218 * numbered block) and path point to the block we want to drop.
1220 memcpy(&state->altpath, &state->path, sizeof(state->path));
1221 if (blkno < blk->blkno)
1222 error = xfs_da_path_shift(state, &state->altpath, forward, 0,
1223 &rval);
1224 else
1225 error = xfs_da_path_shift(state, &state->path, forward, 0,
1226 &rval);
1227 if (error) {
1228 return error;
1230 *action = rval ? 0 : 1;
1231 return 0;
1235 * Move all the leaf entries from drop_blk to save_blk.
1236 * This is done as part of a join operation.
1238 void
1239 xfs_dir2_leafn_unbalance(
1240 xfs_da_state_t *state, /* cursor */
1241 xfs_da_state_blk_t *drop_blk, /* dead block */
1242 xfs_da_state_blk_t *save_blk) /* surviving block */
1244 xfs_da_args_t *args; /* operation arguments */
1245 xfs_dir2_leaf_t *drop_leaf; /* dead leaf structure */
1246 xfs_dir2_leaf_t *save_leaf; /* surviving leaf structure */
1248 args = state->args;
1249 ASSERT(drop_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1250 ASSERT(save_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1251 drop_leaf = drop_blk->bp->data;
1252 save_leaf = save_blk->bp->data;
1253 ASSERT(drop_leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
1254 ASSERT(save_leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
1256 * If there are any stale leaf entries, take this opportunity
1257 * to purge them.
1259 if (drop_leaf->hdr.stale)
1260 xfs_dir2_leaf_compact(args, drop_blk->bp);
1261 if (save_leaf->hdr.stale)
1262 xfs_dir2_leaf_compact(args, save_blk->bp);
1264 * Move the entries from drop to the appropriate end of save.
1266 drop_blk->hashval = be32_to_cpu(drop_leaf->ents[be16_to_cpu(drop_leaf->hdr.count) - 1].hashval);
1267 if (xfs_dir2_leafn_order(save_blk->bp, drop_blk->bp))
1268 xfs_dir2_leafn_moveents(args, drop_blk->bp, 0, save_blk->bp, 0,
1269 be16_to_cpu(drop_leaf->hdr.count));
1270 else
1271 xfs_dir2_leafn_moveents(args, drop_blk->bp, 0, save_blk->bp,
1272 be16_to_cpu(save_leaf->hdr.count), be16_to_cpu(drop_leaf->hdr.count));
1273 save_blk->hashval = be32_to_cpu(save_leaf->ents[be16_to_cpu(save_leaf->hdr.count) - 1].hashval);
1274 xfs_dir2_leafn_check(args->dp, save_blk->bp);
1278 * Top-level node form directory addname routine.
1280 int /* error */
1281 xfs_dir2_node_addname(
1282 xfs_da_args_t *args) /* operation arguments */
1284 xfs_da_state_blk_t *blk; /* leaf block for insert */
1285 int error; /* error return value */
1286 int rval; /* sub-return value */
1287 xfs_da_state_t *state; /* btree cursor */
1289 trace_xfs_dir2_node_addname(args);
1292 * Allocate and initialize the state (btree cursor).
1294 state = xfs_da_state_alloc();
1295 state->args = args;
1296 state->mp = args->dp->i_mount;
1297 state->blocksize = state->mp->m_dirblksize;
1298 state->node_ents = state->mp->m_dir_node_ents;
1300 * Look up the name. We're not supposed to find it, but
1301 * this gives us the insertion point.
1303 error = xfs_da_node_lookup_int(state, &rval);
1304 if (error)
1305 rval = error;
1306 if (rval != ENOENT) {
1307 goto done;
1310 * Add the data entry to a data block.
1311 * Extravalid is set to a freeblock found by lookup.
1313 rval = xfs_dir2_node_addname_int(args,
1314 state->extravalid ? &state->extrablk : NULL);
1315 if (rval) {
1316 goto done;
1318 blk = &state->path.blk[state->path.active - 1];
1319 ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
1321 * Add the new leaf entry.
1323 rval = xfs_dir2_leafn_add(blk->bp, args, blk->index);
1324 if (rval == 0) {
1326 * It worked, fix the hash values up the btree.
1328 if (!(args->op_flags & XFS_DA_OP_JUSTCHECK))
1329 xfs_da_fixhashpath(state, &state->path);
1330 } else {
1332 * It didn't work, we need to split the leaf block.
1334 if (args->total == 0) {
1335 ASSERT(rval == ENOSPC);
1336 goto done;
1339 * Split the leaf block and insert the new entry.
1341 rval = xfs_da_split(state);
1343 done:
1344 xfs_da_state_free(state);
1345 return rval;
1349 * Add the data entry for a node-format directory name addition.
1350 * The leaf entry is added in xfs_dir2_leafn_add.
1351 * We may enter with a freespace block that the lookup found.
1353 static int /* error */
1354 xfs_dir2_node_addname_int(
1355 xfs_da_args_t *args, /* operation arguments */
1356 xfs_da_state_blk_t *fblk) /* optional freespace block */
1358 xfs_dir2_data_hdr_t *hdr; /* data block header */
1359 xfs_dir2_db_t dbno; /* data block number */
1360 xfs_dabuf_t *dbp; /* data block buffer */
1361 xfs_dir2_data_entry_t *dep; /* data entry pointer */
1362 xfs_inode_t *dp; /* incore directory inode */
1363 xfs_dir2_data_unused_t *dup; /* data unused entry pointer */
1364 int error; /* error return value */
1365 xfs_dir2_db_t fbno; /* freespace block number */
1366 xfs_dabuf_t *fbp; /* freespace buffer */
1367 int findex; /* freespace entry index */
1368 xfs_dir2_free_t *free=NULL; /* freespace block structure */
1369 xfs_dir2_db_t ifbno; /* initial freespace block no */
1370 xfs_dir2_db_t lastfbno=0; /* highest freespace block no */
1371 int length; /* length of the new entry */
1372 int logfree; /* need to log free entry */
1373 xfs_mount_t *mp; /* filesystem mount point */
1374 int needlog; /* need to log data header */
1375 int needscan; /* need to rescan data frees */
1376 __be16 *tagp; /* data entry tag pointer */
1377 xfs_trans_t *tp; /* transaction pointer */
1379 dp = args->dp;
1380 mp = dp->i_mount;
1381 tp = args->trans;
1382 length = xfs_dir2_data_entsize(args->namelen);
1384 * If we came in with a freespace block that means that lookup
1385 * found an entry with our hash value. This is the freespace
1386 * block for that data entry.
1388 if (fblk) {
1389 fbp = fblk->bp;
1391 * Remember initial freespace block number.
1393 ifbno = fblk->blkno;
1394 free = fbp->data;
1395 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC));
1396 findex = fblk->index;
1398 * This means the free entry showed that the data block had
1399 * space for our entry, so we remembered it.
1400 * Use that data block.
1402 if (findex >= 0) {
1403 ASSERT(findex < be32_to_cpu(free->hdr.nvalid));
1404 ASSERT(be16_to_cpu(free->bests[findex]) != NULLDATAOFF);
1405 ASSERT(be16_to_cpu(free->bests[findex]) >= length);
1406 dbno = be32_to_cpu(free->hdr.firstdb) + findex;
1409 * The data block looked at didn't have enough room.
1410 * We'll start at the beginning of the freespace entries.
1412 else {
1413 dbno = -1;
1414 findex = 0;
1418 * Didn't come in with a freespace block, so don't have a data block.
1420 else {
1421 ifbno = dbno = -1;
1422 fbp = NULL;
1423 findex = 0;
1426 * If we don't have a data block yet, we're going to scan the
1427 * freespace blocks looking for one. Figure out what the
1428 * highest freespace block number is.
1430 if (dbno == -1) {
1431 xfs_fileoff_t fo; /* freespace block number */
1433 if ((error = xfs_bmap_last_offset(tp, dp, &fo, XFS_DATA_FORK)))
1434 return error;
1435 lastfbno = xfs_dir2_da_to_db(mp, (xfs_dablk_t)fo);
1436 fbno = ifbno;
1439 * While we haven't identified a data block, search the freeblock
1440 * data for a good data block. If we find a null freeblock entry,
1441 * indicating a hole in the data blocks, remember that.
1443 while (dbno == -1) {
1445 * If we don't have a freeblock in hand, get the next one.
1447 if (fbp == NULL) {
1449 * Happens the first time through unless lookup gave
1450 * us a freespace block to start with.
1452 if (++fbno == 0)
1453 fbno = XFS_DIR2_FREE_FIRSTDB(mp);
1455 * If it's ifbno we already looked at it.
1457 if (fbno == ifbno)
1458 fbno++;
1460 * If it's off the end we're done.
1462 if (fbno >= lastfbno)
1463 break;
1465 * Read the block. There can be holes in the
1466 * freespace blocks, so this might not succeed.
1467 * This should be really rare, so there's no reason
1468 * to avoid it.
1470 if ((error = xfs_da_read_buf(tp, dp,
1471 xfs_dir2_db_to_da(mp, fbno), -2, &fbp,
1472 XFS_DATA_FORK))) {
1473 return error;
1475 if (unlikely(fbp == NULL)) {
1476 continue;
1478 free = fbp->data;
1479 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC));
1480 findex = 0;
1483 * Look at the current free entry. Is it good enough?
1485 if (be16_to_cpu(free->bests[findex]) != NULLDATAOFF &&
1486 be16_to_cpu(free->bests[findex]) >= length)
1487 dbno = be32_to_cpu(free->hdr.firstdb) + findex;
1488 else {
1490 * Are we done with the freeblock?
1492 if (++findex == be32_to_cpu(free->hdr.nvalid)) {
1494 * Drop the block.
1496 xfs_da_brelse(tp, fbp);
1497 fbp = NULL;
1498 if (fblk && fblk->bp)
1499 fblk->bp = NULL;
1504 * If we don't have a data block, we need to allocate one and make
1505 * the freespace entries refer to it.
1507 if (unlikely(dbno == -1)) {
1509 * Not allowed to allocate, return failure.
1511 if ((args->op_flags & XFS_DA_OP_JUSTCHECK) ||
1512 args->total == 0) {
1514 * Drop the freespace buffer unless it came from our
1515 * caller.
1517 if ((fblk == NULL || fblk->bp == NULL) && fbp != NULL)
1518 xfs_da_buf_done(fbp);
1519 return XFS_ERROR(ENOSPC);
1522 * Allocate and initialize the new data block.
1524 if (unlikely((error = xfs_dir2_grow_inode(args,
1525 XFS_DIR2_DATA_SPACE,
1526 &dbno)) ||
1527 (error = xfs_dir2_data_init(args, dbno, &dbp)))) {
1529 * Drop the freespace buffer unless it came from our
1530 * caller.
1532 if ((fblk == NULL || fblk->bp == NULL) && fbp != NULL)
1533 xfs_da_buf_done(fbp);
1534 return error;
1537 * If (somehow) we have a freespace block, get rid of it.
1539 if (fbp)
1540 xfs_da_brelse(tp, fbp);
1541 if (fblk && fblk->bp)
1542 fblk->bp = NULL;
1545 * Get the freespace block corresponding to the data block
1546 * that was just allocated.
1548 fbno = xfs_dir2_db_to_fdb(mp, dbno);
1549 if (unlikely(error = xfs_da_read_buf(tp, dp,
1550 xfs_dir2_db_to_da(mp, fbno), -2, &fbp,
1551 XFS_DATA_FORK))) {
1552 xfs_da_buf_done(dbp);
1553 return error;
1556 * If there wasn't a freespace block, the read will
1557 * return a NULL fbp. Allocate and initialize a new one.
1559 if( fbp == NULL ) {
1560 if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE,
1561 &fbno))) {
1562 return error;
1565 if (unlikely(xfs_dir2_db_to_fdb(mp, dbno) != fbno)) {
1566 xfs_alert(mp,
1567 "%s: dir ino %llu needed freesp block %lld for\n"
1568 " data block %lld, got %lld ifbno %llu lastfbno %d",
1569 __func__, (unsigned long long)dp->i_ino,
1570 (long long)xfs_dir2_db_to_fdb(mp, dbno),
1571 (long long)dbno, (long long)fbno,
1572 (unsigned long long)ifbno, lastfbno);
1573 if (fblk) {
1574 xfs_alert(mp,
1575 " fblk 0x%p blkno %llu index %d magic 0x%x",
1576 fblk,
1577 (unsigned long long)fblk->blkno,
1578 fblk->index,
1579 fblk->magic);
1580 } else {
1581 xfs_alert(mp, " ... fblk is NULL");
1583 XFS_ERROR_REPORT("xfs_dir2_node_addname_int",
1584 XFS_ERRLEVEL_LOW, mp);
1585 return XFS_ERROR(EFSCORRUPTED);
1589 * Get a buffer for the new block.
1591 if ((error = xfs_da_get_buf(tp, dp,
1592 xfs_dir2_db_to_da(mp, fbno),
1593 -1, &fbp, XFS_DATA_FORK))) {
1594 return error;
1596 ASSERT(fbp != NULL);
1599 * Initialize the new block to be empty, and remember
1600 * its first slot as our empty slot.
1602 free = fbp->data;
1603 free->hdr.magic = cpu_to_be32(XFS_DIR2_FREE_MAGIC);
1604 free->hdr.firstdb = cpu_to_be32(
1605 (fbno - XFS_DIR2_FREE_FIRSTDB(mp)) *
1606 xfs_dir2_free_max_bests(mp));
1607 free->hdr.nvalid = 0;
1608 free->hdr.nused = 0;
1609 } else {
1610 free = fbp->data;
1611 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC));
1615 * Set the freespace block index from the data block number.
1617 findex = xfs_dir2_db_to_fdindex(mp, dbno);
1619 * If it's after the end of the current entries in the
1620 * freespace block, extend that table.
1622 if (findex >= be32_to_cpu(free->hdr.nvalid)) {
1623 ASSERT(findex < xfs_dir2_free_max_bests(mp));
1624 free->hdr.nvalid = cpu_to_be32(findex + 1);
1626 * Tag new entry so nused will go up.
1628 free->bests[findex] = cpu_to_be16(NULLDATAOFF);
1631 * If this entry was for an empty data block
1632 * (this should always be true) then update the header.
1634 if (free->bests[findex] == cpu_to_be16(NULLDATAOFF)) {
1635 be32_add_cpu(&free->hdr.nused, 1);
1636 xfs_dir2_free_log_header(tp, fbp);
1639 * Update the real value in the table.
1640 * We haven't allocated the data entry yet so this will
1641 * change again.
1643 hdr = dbp->data;
1644 free->bests[findex] = hdr->bestfree[0].length;
1645 logfree = 1;
1648 * We had a data block so we don't have to make a new one.
1650 else {
1652 * If just checking, we succeeded.
1654 if (args->op_flags & XFS_DA_OP_JUSTCHECK) {
1655 if ((fblk == NULL || fblk->bp == NULL) && fbp != NULL)
1656 xfs_da_buf_done(fbp);
1657 return 0;
1660 * Read the data block in.
1662 if (unlikely(
1663 error = xfs_da_read_buf(tp, dp, xfs_dir2_db_to_da(mp, dbno),
1664 -1, &dbp, XFS_DATA_FORK))) {
1665 if ((fblk == NULL || fblk->bp == NULL) && fbp != NULL)
1666 xfs_da_buf_done(fbp);
1667 return error;
1669 hdr = dbp->data;
1670 logfree = 0;
1672 ASSERT(be16_to_cpu(hdr->bestfree[0].length) >= length);
1674 * Point to the existing unused space.
1676 dup = (xfs_dir2_data_unused_t *)
1677 ((char *)hdr + be16_to_cpu(hdr->bestfree[0].offset));
1678 needscan = needlog = 0;
1680 * Mark the first part of the unused space, inuse for us.
1682 xfs_dir2_data_use_free(tp, dbp, dup,
1683 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr), length,
1684 &needlog, &needscan);
1686 * Fill in the new entry and log it.
1688 dep = (xfs_dir2_data_entry_t *)dup;
1689 dep->inumber = cpu_to_be64(args->inumber);
1690 dep->namelen = args->namelen;
1691 memcpy(dep->name, args->name, dep->namelen);
1692 tagp = xfs_dir2_data_entry_tag_p(dep);
1693 *tagp = cpu_to_be16((char *)dep - (char *)hdr);
1694 xfs_dir2_data_log_entry(tp, dbp, dep);
1696 * Rescan the block for bestfree if needed.
1698 if (needscan)
1699 xfs_dir2_data_freescan(mp, hdr, &needlog);
1701 * Log the data block header if needed.
1703 if (needlog)
1704 xfs_dir2_data_log_header(tp, dbp);
1706 * If the freespace entry is now wrong, update it.
1708 if (be16_to_cpu(free->bests[findex]) != be16_to_cpu(hdr->bestfree[0].length)) {
1709 free->bests[findex] = hdr->bestfree[0].length;
1710 logfree = 1;
1713 * Log the freespace entry if needed.
1715 if (logfree)
1716 xfs_dir2_free_log_bests(tp, fbp, findex, findex);
1718 * If the caller didn't hand us the freespace block, drop it.
1720 if ((fblk == NULL || fblk->bp == NULL) && fbp != NULL)
1721 xfs_da_buf_done(fbp);
1723 * Return the data block and offset in args, then drop the data block.
1725 args->blkno = (xfs_dablk_t)dbno;
1726 args->index = be16_to_cpu(*tagp);
1727 xfs_da_buf_done(dbp);
1728 return 0;
1732 * Lookup an entry in a node-format directory.
1733 * All the real work happens in xfs_da_node_lookup_int.
1734 * The only real output is the inode number of the entry.
1736 int /* error */
1737 xfs_dir2_node_lookup(
1738 xfs_da_args_t *args) /* operation arguments */
1740 int error; /* error return value */
1741 int i; /* btree level */
1742 int rval; /* operation return value */
1743 xfs_da_state_t *state; /* btree cursor */
1745 trace_xfs_dir2_node_lookup(args);
1748 * Allocate and initialize the btree cursor.
1750 state = xfs_da_state_alloc();
1751 state->args = args;
1752 state->mp = args->dp->i_mount;
1753 state->blocksize = state->mp->m_dirblksize;
1754 state->node_ents = state->mp->m_dir_node_ents;
1756 * Fill in the path to the entry in the cursor.
1758 error = xfs_da_node_lookup_int(state, &rval);
1759 if (error)
1760 rval = error;
1761 else if (rval == ENOENT && args->cmpresult == XFS_CMP_CASE) {
1762 /* If a CI match, dup the actual name and return EEXIST */
1763 xfs_dir2_data_entry_t *dep;
1765 dep = (xfs_dir2_data_entry_t *)((char *)state->extrablk.bp->
1766 data + state->extrablk.index);
1767 rval = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
1770 * Release the btree blocks and leaf block.
1772 for (i = 0; i < state->path.active; i++) {
1773 xfs_da_brelse(args->trans, state->path.blk[i].bp);
1774 state->path.blk[i].bp = NULL;
1777 * Release the data block if we have it.
1779 if (state->extravalid && state->extrablk.bp) {
1780 xfs_da_brelse(args->trans, state->extrablk.bp);
1781 state->extrablk.bp = NULL;
1783 xfs_da_state_free(state);
1784 return rval;
1788 * Remove an entry from a node-format directory.
1790 int /* error */
1791 xfs_dir2_node_removename(
1792 xfs_da_args_t *args) /* operation arguments */
1794 xfs_da_state_blk_t *blk; /* leaf block */
1795 int error; /* error return value */
1796 int rval; /* operation return value */
1797 xfs_da_state_t *state; /* btree cursor */
1799 trace_xfs_dir2_node_removename(args);
1802 * Allocate and initialize the btree cursor.
1804 state = xfs_da_state_alloc();
1805 state->args = args;
1806 state->mp = args->dp->i_mount;
1807 state->blocksize = state->mp->m_dirblksize;
1808 state->node_ents = state->mp->m_dir_node_ents;
1810 * Look up the entry we're deleting, set up the cursor.
1812 error = xfs_da_node_lookup_int(state, &rval);
1813 if (error)
1814 rval = error;
1816 * Didn't find it, upper layer screwed up.
1818 if (rval != EEXIST) {
1819 xfs_da_state_free(state);
1820 return rval;
1822 blk = &state->path.blk[state->path.active - 1];
1823 ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
1824 ASSERT(state->extravalid);
1826 * Remove the leaf and data entries.
1827 * Extrablk refers to the data block.
1829 error = xfs_dir2_leafn_remove(args, blk->bp, blk->index,
1830 &state->extrablk, &rval);
1831 if (error)
1832 return error;
1834 * Fix the hash values up the btree.
1836 xfs_da_fixhashpath(state, &state->path);
1838 * If we need to join leaf blocks, do it.
1840 if (rval && state->path.active > 1)
1841 error = xfs_da_join(state);
1843 * If no errors so far, try conversion to leaf format.
1845 if (!error)
1846 error = xfs_dir2_node_to_leaf(state);
1847 xfs_da_state_free(state);
1848 return error;
1852 * Replace an entry's inode number in a node-format directory.
1854 int /* error */
1855 xfs_dir2_node_replace(
1856 xfs_da_args_t *args) /* operation arguments */
1858 xfs_da_state_blk_t *blk; /* leaf block */
1859 xfs_dir2_data_hdr_t *hdr; /* data block header */
1860 xfs_dir2_data_entry_t *dep; /* data entry changed */
1861 int error; /* error return value */
1862 int i; /* btree level */
1863 xfs_ino_t inum; /* new inode number */
1864 xfs_dir2_leaf_t *leaf; /* leaf structure */
1865 xfs_dir2_leaf_entry_t *lep; /* leaf entry being changed */
1866 int rval; /* internal return value */
1867 xfs_da_state_t *state; /* btree cursor */
1869 trace_xfs_dir2_node_replace(args);
1872 * Allocate and initialize the btree cursor.
1874 state = xfs_da_state_alloc();
1875 state->args = args;
1876 state->mp = args->dp->i_mount;
1877 state->blocksize = state->mp->m_dirblksize;
1878 state->node_ents = state->mp->m_dir_node_ents;
1879 inum = args->inumber;
1881 * Lookup the entry to change in the btree.
1883 error = xfs_da_node_lookup_int(state, &rval);
1884 if (error) {
1885 rval = error;
1888 * It should be found, since the vnodeops layer has looked it up
1889 * and locked it. But paranoia is good.
1891 if (rval == EEXIST) {
1893 * Find the leaf entry.
1895 blk = &state->path.blk[state->path.active - 1];
1896 ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
1897 leaf = blk->bp->data;
1898 lep = &leaf->ents[blk->index];
1899 ASSERT(state->extravalid);
1901 * Point to the data entry.
1903 hdr = state->extrablk.bp->data;
1904 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC));
1905 dep = (xfs_dir2_data_entry_t *)
1906 ((char *)hdr +
1907 xfs_dir2_dataptr_to_off(state->mp, be32_to_cpu(lep->address)));
1908 ASSERT(inum != be64_to_cpu(dep->inumber));
1910 * Fill in the new inode number and log the entry.
1912 dep->inumber = cpu_to_be64(inum);
1913 xfs_dir2_data_log_entry(args->trans, state->extrablk.bp, dep);
1914 rval = 0;
1917 * Didn't find it, and we're holding a data block. Drop it.
1919 else if (state->extravalid) {
1920 xfs_da_brelse(args->trans, state->extrablk.bp);
1921 state->extrablk.bp = NULL;
1924 * Release all the buffers in the cursor.
1926 for (i = 0; i < state->path.active; i++) {
1927 xfs_da_brelse(args->trans, state->path.blk[i].bp);
1928 state->path.blk[i].bp = NULL;
1930 xfs_da_state_free(state);
1931 return rval;
1935 * Trim off a trailing empty freespace block.
1936 * Return (in rvalp) 1 if we did it, 0 if not.
1938 int /* error */
1939 xfs_dir2_node_trim_free(
1940 xfs_da_args_t *args, /* operation arguments */
1941 xfs_fileoff_t fo, /* free block number */
1942 int *rvalp) /* out: did something */
1944 xfs_dabuf_t *bp; /* freespace buffer */
1945 xfs_inode_t *dp; /* incore directory inode */
1946 int error; /* error return code */
1947 xfs_dir2_free_t *free; /* freespace structure */
1948 xfs_mount_t *mp; /* filesystem mount point */
1949 xfs_trans_t *tp; /* transaction pointer */
1951 dp = args->dp;
1952 mp = dp->i_mount;
1953 tp = args->trans;
1955 * Read the freespace block.
1957 if (unlikely(error = xfs_da_read_buf(tp, dp, (xfs_dablk_t)fo, -2, &bp,
1958 XFS_DATA_FORK))) {
1959 return error;
1963 * There can be holes in freespace. If fo is a hole, there's
1964 * nothing to do.
1966 if (bp == NULL) {
1967 return 0;
1969 free = bp->data;
1970 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC));
1972 * If there are used entries, there's nothing to do.
1974 if (be32_to_cpu(free->hdr.nused) > 0) {
1975 xfs_da_brelse(tp, bp);
1976 *rvalp = 0;
1977 return 0;
1980 * Blow the block away.
1982 if ((error =
1983 xfs_dir2_shrink_inode(args, xfs_dir2_da_to_db(mp, (xfs_dablk_t)fo),
1984 bp))) {
1986 * Can't fail with ENOSPC since that only happens with no
1987 * space reservation, when breaking up an extent into two
1988 * pieces. This is the last block of an extent.
1990 ASSERT(error != ENOSPC);
1991 xfs_da_brelse(tp, bp);
1992 return error;
1995 * Return that we succeeded.
1997 *rvalp = 1;
1998 return 0;