]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blob - fs/xfs/libxfs/xfs_dir2_node.c
projects / mirror_ubuntu-zesty-kernel.git / blob
? search:


bbd1238852b3c4f66b76b159847d33e827ced077
[mirror_ubuntu-zesty-kernel.git] / fs / xfs / libxfs / xfs_dir2_node.c
1 /*
2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * Copyright (c) 2013 Red Hat, Inc.
4 * All Rights Reserved.
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License as
8 * published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it would be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
18 */
19 #include "xfs.h"
20 #include "xfs_fs.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
24 #include "xfs_mount.h"
25 #include "xfs_da_format.h"
26 #include "xfs_da_btree.h"
27 #include "xfs_inode.h"
28 #include "xfs_bmap.h"
29 #include "xfs_dir2.h"
30 #include "xfs_dir2_priv.h"
31 #include "xfs_error.h"
32 #include "xfs_trace.h"
33 #include "xfs_trans.h"
34 #include "xfs_buf_item.h"
35 #include "xfs_cksum.h"
36 #include "xfs_log.h"
37
38 /*
39 * Function declarations.
40 */
41 static int xfs_dir2_leafn_add(struct xfs_buf *bp, xfs_da_args_t *args,
42 int index);
43 static void xfs_dir2_leafn_rebalance(xfs_da_state_t *state,
44 xfs_da_state_blk_t *blk1,
45 xfs_da_state_blk_t *blk2);
46 static int xfs_dir2_leafn_remove(xfs_da_args_t *args, struct xfs_buf *bp,
47 int index, xfs_da_state_blk_t *dblk,
48 int *rval);
49 static int xfs_dir2_node_addname_int(xfs_da_args_t *args,
50 xfs_da_state_blk_t *fblk);
51
52 /*
53 * Check internal consistency of a leafn block.
54 */
55 #ifdef DEBUG
56 #define xfs_dir3_leaf_check(dp, bp) \
57 do { \
58 if (!xfs_dir3_leafn_check((dp), (bp))) \
59 ASSERT(0); \
60 } while (0);
61
62 static bool
63 xfs_dir3_leafn_check(
64 struct xfs_inode *dp,
65 struct xfs_buf *bp)
66 {
67 struct xfs_dir2_leaf *leaf = bp->b_addr;
68 struct xfs_dir3_icleaf_hdr leafhdr;
69
70 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
71
72 if (leafhdr.magic == XFS_DIR3_LEAFN_MAGIC) {
73 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
74 if (be64_to_cpu(leaf3->info.blkno) != bp->b_bn)
75 return false;
76 } else if (leafhdr.magic != XFS_DIR2_LEAFN_MAGIC)
77 return false;
78
79 return xfs_dir3_leaf_check_int(dp->i_mount, dp, &leafhdr, leaf);
80 }
81 #else
82 #define xfs_dir3_leaf_check(dp, bp)
83 #endif
84
85 static bool
86 xfs_dir3_free_verify(
87 struct xfs_buf *bp)
88 {
89 struct xfs_mount *mp = bp->b_target->bt_mount;
90 struct xfs_dir2_free_hdr *hdr = bp->b_addr;
91
92 if (xfs_sb_version_hascrc(&mp->m_sb)) {
93 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
94
95 if (hdr3->magic != cpu_to_be32(XFS_DIR3_FREE_MAGIC))
96 return false;
97 if (!uuid_equal(&hdr3->uuid, &mp->m_sb.sb_meta_uuid))
98 return false;
99 if (be64_to_cpu(hdr3->blkno) != bp->b_bn)
100 return false;
101 if (!xfs_log_check_lsn(mp, be64_to_cpu(hdr3->lsn)))
102 return false;
103 } else {
104 if (hdr->magic != cpu_to_be32(XFS_DIR2_FREE_MAGIC))
105 return false;
106 }
107
108 /* XXX: should bounds check the xfs_dir3_icfree_hdr here */
109
110 return true;
111 }
112
113 static void
114 xfs_dir3_free_read_verify(
115 struct xfs_buf *bp)
116 {
117 struct xfs_mount *mp = bp->b_target->bt_mount;
118
119 if (xfs_sb_version_hascrc(&mp->m_sb) &&
120 !xfs_buf_verify_cksum(bp, XFS_DIR3_FREE_CRC_OFF))
121 xfs_buf_ioerror(bp, -EFSBADCRC);
122 else if (!xfs_dir3_free_verify(bp))
123 xfs_buf_ioerror(bp, -EFSCORRUPTED);
124
125 if (bp->b_error)
126 xfs_verifier_error(bp);
127 }
128
129 static void
130 xfs_dir3_free_write_verify(
131 struct xfs_buf *bp)
132 {
133 struct xfs_mount *mp = bp->b_target->bt_mount;
134 struct xfs_buf_log_item *bip = bp->b_fspriv;
135 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
136
137 if (!xfs_dir3_free_verify(bp)) {
138 xfs_buf_ioerror(bp, -EFSCORRUPTED);
139 xfs_verifier_error(bp);
140 return;
141 }
142
143 if (!xfs_sb_version_hascrc(&mp->m_sb))
144 return;
145
146 if (bip)
147 hdr3->lsn = cpu_to_be64(bip->bli_item.li_lsn);
148
149 xfs_buf_update_cksum(bp, XFS_DIR3_FREE_CRC_OFF);
150 }
151
152 const struct xfs_buf_ops xfs_dir3_free_buf_ops = {
153 .name = "xfs_dir3_free",
154 .verify_read = xfs_dir3_free_read_verify,
155 .verify_write = xfs_dir3_free_write_verify,
156 };
157
158 /* Everything ok in the free block header? */
159 static bool
160 xfs_dir3_free_header_check(
161 struct xfs_inode *dp,
162 xfs_dablk_t fbno,
163 struct xfs_buf *bp)
164 {
165 struct xfs_mount *mp = dp->i_mount;
166 unsigned int firstdb;
167 int maxbests;
168
169 maxbests = dp->d_ops->free_max_bests(mp->m_dir_geo);
170 firstdb = (xfs_dir2_da_to_db(mp->m_dir_geo, fbno) -
171 xfs_dir2_byte_to_db(mp->m_dir_geo, XFS_DIR2_FREE_OFFSET)) *
172 maxbests;
173 if (xfs_sb_version_hascrc(&mp->m_sb)) {
174 struct xfs_dir3_free_hdr *hdr3 = bp->b_addr;
175
176 if (be32_to_cpu(hdr3->firstdb) != firstdb)
177 return false;
178 if (be32_to_cpu(hdr3->nvalid) > maxbests)
179 return false;
180 if (be32_to_cpu(hdr3->nvalid) < be32_to_cpu(hdr3->nused))
181 return false;
182 } else {
183 struct xfs_dir2_free_hdr *hdr = bp->b_addr;
184
185 if (be32_to_cpu(hdr->firstdb) != firstdb)
186 return false;
187 if (be32_to_cpu(hdr->nvalid) > maxbests)
188 return false;
189 if (be32_to_cpu(hdr->nvalid) < be32_to_cpu(hdr->nused))
190 return false;
191 }
192 return true;
193 }
194
195 static int
196 __xfs_dir3_free_read(
197 struct xfs_trans *tp,
198 struct xfs_inode *dp,
199 xfs_dablk_t fbno,
200 xfs_daddr_t mappedbno,
201 struct xfs_buf **bpp)
202 {
203 int err;
204
205 err = xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
206 XFS_DATA_FORK, &xfs_dir3_free_buf_ops);
207 if (err || !*bpp)
208 return err;
209
210 /* Check things that we can't do in the verifier. */
211 if (!xfs_dir3_free_header_check(dp, fbno, *bpp)) {
212 xfs_buf_ioerror(*bpp, -EFSCORRUPTED);
213 xfs_verifier_error(*bpp);
214 xfs_trans_brelse(tp, *bpp);
215 return -EFSCORRUPTED;
216 }
217
218 /* try read returns without an error or *bpp if it lands in a hole */
219 if (tp)
220 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_FREE_BUF);
221
222 return 0;
223 }
224
225 int
226 xfs_dir2_free_read(
227 struct xfs_trans *tp,
228 struct xfs_inode *dp,
229 xfs_dablk_t fbno,
230 struct xfs_buf **bpp)
231 {
232 return __xfs_dir3_free_read(tp, dp, fbno, -1, bpp);
233 }
234
235 static int
236 xfs_dir2_free_try_read(
237 struct xfs_trans *tp,
238 struct xfs_inode *dp,
239 xfs_dablk_t fbno,
240 struct xfs_buf **bpp)
241 {
242 return __xfs_dir3_free_read(tp, dp, fbno, -2, bpp);
243 }
244
245 static int
246 xfs_dir3_free_get_buf(
247 xfs_da_args_t *args,
248 xfs_dir2_db_t fbno,
249 struct xfs_buf **bpp)
250 {
251 struct xfs_trans *tp = args->trans;
252 struct xfs_inode *dp = args->dp;
253 struct xfs_mount *mp = dp->i_mount;
254 struct xfs_buf *bp;
255 int error;
256 struct xfs_dir3_icfree_hdr hdr;
257
258 error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(args->geo, fbno),
259 -1, &bp, XFS_DATA_FORK);
260 if (error)
261 return error;
262
263 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_FREE_BUF);
264 bp->b_ops = &xfs_dir3_free_buf_ops;
265
266 /*
267 * Initialize the new block to be empty, and remember
268 * its first slot as our empty slot.
269 */
270 memset(bp->b_addr, 0, sizeof(struct xfs_dir3_free_hdr));
271 memset(&hdr, 0, sizeof(hdr));
272
273 if (xfs_sb_version_hascrc(&mp->m_sb)) {
274 struct xfs_dir3_free_hdr *hdr3 = bp->b_addr;
275
276 hdr.magic = XFS_DIR3_FREE_MAGIC;
277
278 hdr3->hdr.blkno = cpu_to_be64(bp->b_bn);
279 hdr3->hdr.owner = cpu_to_be64(dp->i_ino);
280 uuid_copy(&hdr3->hdr.uuid, &mp->m_sb.sb_meta_uuid);
281 } else
282 hdr.magic = XFS_DIR2_FREE_MAGIC;
283 dp->d_ops->free_hdr_to_disk(bp->b_addr, &hdr);
284 *bpp = bp;
285 return 0;
286 }
287
288 /*
289 * Log entries from a freespace block.
290 */
291 STATIC void
292 xfs_dir2_free_log_bests(
293 struct xfs_da_args *args,
294 struct xfs_buf *bp,
295 int first, /* first entry to log */
296 int last) /* last entry to log */
297 {
298 xfs_dir2_free_t *free; /* freespace structure */
299 __be16 *bests;
300
301 free = bp->b_addr;
302 bests = args->dp->d_ops->free_bests_p(free);
303 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC) ||
304 free->hdr.magic == cpu_to_be32(XFS_DIR3_FREE_MAGIC));
305 xfs_trans_log_buf(args->trans, bp,
306 (uint)((char *)&bests[first] - (char *)free),
307 (uint)((char *)&bests[last] - (char *)free +
308 sizeof(bests[0]) - 1));
309 }
310
311 /*
312 * Log header from a freespace block.
313 */
314 static void
315 xfs_dir2_free_log_header(
316 struct xfs_da_args *args,
317 struct xfs_buf *bp)
318 {
319 #ifdef DEBUG
320 xfs_dir2_free_t *free; /* freespace structure */
321
322 free = bp->b_addr;
323 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC) ||
324 free->hdr.magic == cpu_to_be32(XFS_DIR3_FREE_MAGIC));
325 #endif
326 xfs_trans_log_buf(args->trans, bp, 0,
327 args->dp->d_ops->free_hdr_size - 1);
328 }
329
330 /*
331 * Convert a leaf-format directory to a node-format directory.
332 * We need to change the magic number of the leaf block, and copy
333 * the freespace table out of the leaf block into its own block.
334 */
335 int /* error */
336 xfs_dir2_leaf_to_node(
337 xfs_da_args_t *args, /* operation arguments */
338 struct xfs_buf *lbp) /* leaf buffer */
339 {
340 xfs_inode_t *dp; /* incore directory inode */
341 int error; /* error return value */
342 struct xfs_buf *fbp; /* freespace buffer */
343 xfs_dir2_db_t fdb; /* freespace block number */
344 xfs_dir2_free_t *free; /* freespace structure */
345 __be16 *from; /* pointer to freespace entry */
346 int i; /* leaf freespace index */
347 xfs_dir2_leaf_t *leaf; /* leaf structure */
348 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
349 int n; /* count of live freespc ents */
350 xfs_dir2_data_off_t off; /* freespace entry value */
351 __be16 *to; /* pointer to freespace entry */
352 xfs_trans_t *tp; /* transaction pointer */
353 struct xfs_dir3_icfree_hdr freehdr;
354
355 trace_xfs_dir2_leaf_to_node(args);
356
357 dp = args->dp;
358 tp = args->trans;
359 /*
360 * Add a freespace block to the directory.
361 */
362 if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE, &fdb))) {
363 return error;
364 }
365 ASSERT(fdb == xfs_dir2_byte_to_db(args->geo, XFS_DIR2_FREE_OFFSET));
366 /*
367 * Get the buffer for the new freespace block.
368 */
369 error = xfs_dir3_free_get_buf(args, fdb, &fbp);
370 if (error)
371 return error;
372
373 free = fbp->b_addr;
374 dp->d_ops->free_hdr_from_disk(&freehdr, free);
375 leaf = lbp->b_addr;
376 ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
377 ASSERT(be32_to_cpu(ltp->bestcount) <=
378 (uint)dp->i_d.di_size / args->geo->blksize);
379
380 /*
381 * Copy freespace entries from the leaf block to the new block.
382 * Count active entries.
383 */
384 from = xfs_dir2_leaf_bests_p(ltp);
385 to = dp->d_ops->free_bests_p(free);
386 for (i = n = 0; i < be32_to_cpu(ltp->bestcount); i++, from++, to++) {
387 if ((off = be16_to_cpu(*from)) != NULLDATAOFF)
388 n++;
389 *to = cpu_to_be16(off);
390 }
391
392 /*
393 * Now initialize the freespace block header.
394 */
395 freehdr.nused = n;
396 freehdr.nvalid = be32_to_cpu(ltp->bestcount);
397
398 dp->d_ops->free_hdr_to_disk(fbp->b_addr, &freehdr);
399 xfs_dir2_free_log_bests(args, fbp, 0, freehdr.nvalid - 1);
400 xfs_dir2_free_log_header(args, fbp);
401
402 /*
403 * Converting the leaf to a leafnode is just a matter of changing the
404 * magic number and the ops. Do the change directly to the buffer as
405 * it's less work (and less code) than decoding the header to host
406 * format and back again.
407 */
408 if (leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC))
409 leaf->hdr.info.magic = cpu_to_be16(XFS_DIR2_LEAFN_MAGIC);
410 else
411 leaf->hdr.info.magic = cpu_to_be16(XFS_DIR3_LEAFN_MAGIC);
412 lbp->b_ops = &xfs_dir3_leafn_buf_ops;
413 xfs_trans_buf_set_type(tp, lbp, XFS_BLFT_DIR_LEAFN_BUF);
414 xfs_dir3_leaf_log_header(args, lbp);
415 xfs_dir3_leaf_check(dp, lbp);
416 return 0;
417 }
418
419 /*
420 * Add a leaf entry to a leaf block in a node-form directory.
421 * The other work necessary is done from the caller.
422 */
423 static int /* error */
424 xfs_dir2_leafn_add(
425 struct xfs_buf *bp, /* leaf buffer */
426 xfs_da_args_t *args, /* operation arguments */
427 int index) /* insertion pt for new entry */
428 {
429 int compact; /* compacting stale leaves */
430 xfs_inode_t *dp; /* incore directory inode */
431 int highstale; /* next stale entry */
432 xfs_dir2_leaf_t *leaf; /* leaf structure */
433 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
434 int lfloghigh; /* high leaf entry logging */
435 int lfloglow; /* low leaf entry logging */
436 int lowstale; /* previous stale entry */
437 struct xfs_dir3_icleaf_hdr leafhdr;
438 struct xfs_dir2_leaf_entry *ents;
439
440 trace_xfs_dir2_leafn_add(args, index);
441
442 dp = args->dp;
443 leaf = bp->b_addr;
444 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
445 ents = dp->d_ops->leaf_ents_p(leaf);
446
447 /*
448 * Quick check just to make sure we are not going to index
449 * into other peoples memory
450 */
451 if (index < 0)
452 return -EFSCORRUPTED;
453
454 /*
455 * If there are already the maximum number of leaf entries in
456 * the block, if there are no stale entries it won't fit.
457 * Caller will do a split. If there are stale entries we'll do
458 * a compact.
459 */
460
461 if (leafhdr.count == dp->d_ops->leaf_max_ents(args->geo)) {
462 if (!leafhdr.stale)
463 return -ENOSPC;
464 compact = leafhdr.stale > 1;
465 } else
466 compact = 0;
467 ASSERT(index == 0 || be32_to_cpu(ents[index - 1].hashval) <= args->hashval);
468 ASSERT(index == leafhdr.count ||
469 be32_to_cpu(ents[index].hashval) >= args->hashval);
470
471 if (args->op_flags & XFS_DA_OP_JUSTCHECK)
472 return 0;
473
474 /*
475 * Compact out all but one stale leaf entry. Leaves behind
476 * the entry closest to index.
477 */
478 if (compact)
479 xfs_dir3_leaf_compact_x1(&leafhdr, ents, &index, &lowstale,
480 &highstale, &lfloglow, &lfloghigh);
481 else if (leafhdr.stale) {
482 /*
483 * Set impossible logging indices for this case.
484 */
485 lfloglow = leafhdr.count;
486 lfloghigh = -1;
487 }
488
489 /*
490 * Insert the new entry, log everything.
491 */
492 lep = xfs_dir3_leaf_find_entry(&leafhdr, ents, index, compact, lowstale,
493 highstale, &lfloglow, &lfloghigh);
494
495 lep->hashval = cpu_to_be32(args->hashval);
496 lep->address = cpu_to_be32(xfs_dir2_db_off_to_dataptr(args->geo,
497 args->blkno, args->index));
498
499 dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
500 xfs_dir3_leaf_log_header(args, bp);
501 xfs_dir3_leaf_log_ents(args, bp, lfloglow, lfloghigh);
502 xfs_dir3_leaf_check(dp, bp);
503 return 0;
504 }
505
506 #ifdef DEBUG
507 static void
508 xfs_dir2_free_hdr_check(
509 struct xfs_inode *dp,
510 struct xfs_buf *bp,
511 xfs_dir2_db_t db)
512 {
513 struct xfs_dir3_icfree_hdr hdr;
514
515 dp->d_ops->free_hdr_from_disk(&hdr, bp->b_addr);
516
517 ASSERT((hdr.firstdb %
518 dp->d_ops->free_max_bests(dp->i_mount->m_dir_geo)) == 0);
519 ASSERT(hdr.firstdb <= db);
520 ASSERT(db < hdr.firstdb + hdr.nvalid);
521 }
522 #else
523 #define xfs_dir2_free_hdr_check(dp, bp, db)
524 #endif /* DEBUG */
525
526 /*
527 * Return the last hash value in the leaf.
528 * Stale entries are ok.
529 */
530 xfs_dahash_t /* hash value */
531 xfs_dir2_leafn_lasthash(
532 struct xfs_inode *dp,
533 struct xfs_buf *bp, /* leaf buffer */
534 int *count) /* count of entries in leaf */
535 {
536 struct xfs_dir2_leaf *leaf = bp->b_addr;
537 struct xfs_dir2_leaf_entry *ents;
538 struct xfs_dir3_icleaf_hdr leafhdr;
539
540 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
541
542 ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC ||
543 leafhdr.magic == XFS_DIR3_LEAFN_MAGIC);
544
545 if (count)
546 *count = leafhdr.count;
547 if (!leafhdr.count)
548 return 0;
549
550 ents = dp->d_ops->leaf_ents_p(leaf);
551 return be32_to_cpu(ents[leafhdr.count - 1].hashval);
552 }
553
554 /*
555 * Look up a leaf entry for space to add a name in a node-format leaf block.
556 * The extrablk in state is a freespace block.
557 */
558 STATIC int
559 xfs_dir2_leafn_lookup_for_addname(
560 struct xfs_buf *bp, /* leaf buffer */
561 xfs_da_args_t *args, /* operation arguments */
562 int *indexp, /* out: leaf entry index */
563 xfs_da_state_t *state) /* state to fill in */
564 {
565 struct xfs_buf *curbp = NULL; /* current data/free buffer */
566 xfs_dir2_db_t curdb = -1; /* current data block number */
567 xfs_dir2_db_t curfdb = -1; /* current free block number */
568 xfs_inode_t *dp; /* incore directory inode */
569 int error; /* error return value */
570 int fi; /* free entry index */
571 xfs_dir2_free_t *free = NULL; /* free block structure */
572 int index; /* leaf entry index */
573 xfs_dir2_leaf_t *leaf; /* leaf structure */
574 int length; /* length of new data entry */
575 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
576 xfs_mount_t *mp; /* filesystem mount point */
577 xfs_dir2_db_t newdb; /* new data block number */
578 xfs_dir2_db_t newfdb; /* new free block number */
579 xfs_trans_t *tp; /* transaction pointer */
580 struct xfs_dir2_leaf_entry *ents;
581 struct xfs_dir3_icleaf_hdr leafhdr;
582
583 dp = args->dp;
584 tp = args->trans;
585 mp = dp->i_mount;
586 leaf = bp->b_addr;
587 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
588 ents = dp->d_ops->leaf_ents_p(leaf);
589
590 xfs_dir3_leaf_check(dp, bp);
591 ASSERT(leafhdr.count > 0);
592
593 /*
594 * Look up the hash value in the leaf entries.
595 */
596 index = xfs_dir2_leaf_search_hash(args, bp);
597 /*
598 * Do we have a buffer coming in?
599 */
600 if (state->extravalid) {
601 /* If so, it's a free block buffer, get the block number. */
602 curbp = state->extrablk.bp;
603 curfdb = state->extrablk.blkno;
604 free = curbp->b_addr;
605 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC) ||
606 free->hdr.magic == cpu_to_be32(XFS_DIR3_FREE_MAGIC));
607 }
608 length = dp->d_ops->data_entsize(args->namelen);
609 /*
610 * Loop over leaf entries with the right hash value.
611 */
612 for (lep = &ents[index];
613 index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
614 lep++, index++) {
615 /*
616 * Skip stale leaf entries.
617 */
618 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
619 continue;
620 /*
621 * Pull the data block number from the entry.
622 */
623 newdb = xfs_dir2_dataptr_to_db(args->geo,
624 be32_to_cpu(lep->address));
625 /*
626 * For addname, we're looking for a place to put the new entry.
627 * We want to use a data block with an entry of equal
628 * hash value to ours if there is one with room.
629 *
630 * If this block isn't the data block we already have
631 * in hand, take a look at it.
632 */
633 if (newdb != curdb) {
634 __be16 *bests;
635
636 curdb = newdb;
637 /*
638 * Convert the data block to the free block
639 * holding its freespace information.
640 */
641 newfdb = dp->d_ops->db_to_fdb(args->geo, newdb);
642 /*
643 * If it's not the one we have in hand, read it in.
644 */
645 if (newfdb != curfdb) {
646 /*
647 * If we had one before, drop it.
648 */
649 if (curbp)
650 xfs_trans_brelse(tp, curbp);
651
652 error = xfs_dir2_free_read(tp, dp,
653 xfs_dir2_db_to_da(args->geo,
654 newfdb),
655 &curbp);
656 if (error)
657 return error;
658 free = curbp->b_addr;
659
660 xfs_dir2_free_hdr_check(dp, curbp, curdb);
661 }
662 /*
663 * Get the index for our entry.
664 */
665 fi = dp->d_ops->db_to_fdindex(args->geo, curdb);
666 /*
667 * If it has room, return it.
668 */
669 bests = dp->d_ops->free_bests_p(free);
670 if (unlikely(bests[fi] == cpu_to_be16(NULLDATAOFF))) {
671 XFS_ERROR_REPORT("xfs_dir2_leafn_lookup_int",
672 XFS_ERRLEVEL_LOW, mp);
673 if (curfdb != newfdb)
674 xfs_trans_brelse(tp, curbp);
675 return -EFSCORRUPTED;
676 }
677 curfdb = newfdb;
678 if (be16_to_cpu(bests[fi]) >= length)
679 goto out;
680 }
681 }
682 /* Didn't find any space */
683 fi = -1;
684 out:
685 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
686 if (curbp) {
687 /* Giving back a free block. */
688 state->extravalid = 1;
689 state->extrablk.bp = curbp;
690 state->extrablk.index = fi;
691 state->extrablk.blkno = curfdb;
692
693 /*
694 * Important: this magic number is not in the buffer - it's for
695 * buffer type information and therefore only the free/data type
696 * matters here, not whether CRCs are enabled or not.
697 */
698 state->extrablk.magic = XFS_DIR2_FREE_MAGIC;
699 } else {
700 state->extravalid = 0;
701 }
702 /*
703 * Return the index, that will be the insertion point.
704 */
705 *indexp = index;
706 return -ENOENT;
707 }
708
709 /*
710 * Look up a leaf entry in a node-format leaf block.
711 * The extrablk in state a data block.
712 */
713 STATIC int
714 xfs_dir2_leafn_lookup_for_entry(
715 struct xfs_buf *bp, /* leaf buffer */
716 xfs_da_args_t *args, /* operation arguments */
717 int *indexp, /* out: leaf entry index */
718 xfs_da_state_t *state) /* state to fill in */
719 {
720 struct xfs_buf *curbp = NULL; /* current data/free buffer */
721 xfs_dir2_db_t curdb = -1; /* current data block number */
722 xfs_dir2_data_entry_t *dep; /* data block entry */
723 xfs_inode_t *dp; /* incore directory inode */
724 int error; /* error return value */
725 int index; /* leaf entry index */
726 xfs_dir2_leaf_t *leaf; /* leaf structure */
727 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
728 xfs_mount_t *mp; /* filesystem mount point */
729 xfs_dir2_db_t newdb; /* new data block number */
730 xfs_trans_t *tp; /* transaction pointer */
731 enum xfs_dacmp cmp; /* comparison result */
732 struct xfs_dir2_leaf_entry *ents;
733 struct xfs_dir3_icleaf_hdr leafhdr;
734
735 dp = args->dp;
736 tp = args->trans;
737 mp = dp->i_mount;
738 leaf = bp->b_addr;
739 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
740 ents = dp->d_ops->leaf_ents_p(leaf);
741
742 xfs_dir3_leaf_check(dp, bp);
743 ASSERT(leafhdr.count > 0);
744
745 /*
746 * Look up the hash value in the leaf entries.
747 */
748 index = xfs_dir2_leaf_search_hash(args, bp);
749 /*
750 * Do we have a buffer coming in?
751 */
752 if (state->extravalid) {
753 curbp = state->extrablk.bp;
754 curdb = state->extrablk.blkno;
755 }
756 /*
757 * Loop over leaf entries with the right hash value.
758 */
759 for (lep = &ents[index];
760 index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
761 lep++, index++) {
762 /*
763 * Skip stale leaf entries.
764 */
765 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
766 continue;
767 /*
768 * Pull the data block number from the entry.
769 */
770 newdb = xfs_dir2_dataptr_to_db(args->geo,
771 be32_to_cpu(lep->address));
772 /*
773 * Not adding a new entry, so we really want to find
774 * the name given to us.
775 *
776 * If it's a different data block, go get it.
777 */
778 if (newdb != curdb) {
779 /*
780 * If we had a block before that we aren't saving
781 * for a CI name, drop it
782 */
783 if (curbp && (args->cmpresult == XFS_CMP_DIFFERENT ||
784 curdb != state->extrablk.blkno))
785 xfs_trans_brelse(tp, curbp);
786 /*
787 * If needing the block that is saved with a CI match,
788 * use it otherwise read in the new data block.
789 */
790 if (args->cmpresult != XFS_CMP_DIFFERENT &&
791 newdb == state->extrablk.blkno) {
792 ASSERT(state->extravalid);
793 curbp = state->extrablk.bp;
794 } else {
795 error = xfs_dir3_data_read(tp, dp,
796 xfs_dir2_db_to_da(args->geo,
797 newdb),
798 -1, &curbp);
799 if (error)
800 return error;
801 }
802 xfs_dir3_data_check(dp, curbp);
803 curdb = newdb;
804 }
805 /*
806 * Point to the data entry.
807 */
808 dep = (xfs_dir2_data_entry_t *)((char *)curbp->b_addr +
809 xfs_dir2_dataptr_to_off(args->geo,
810 be32_to_cpu(lep->address)));
811 /*
812 * Compare the entry and if it's an exact match, return
813 * EEXIST immediately. If it's the first case-insensitive
814 * match, store the block & inode number and continue looking.
815 */
816 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
817 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
818 /* If there is a CI match block, drop it */
819 if (args->cmpresult != XFS_CMP_DIFFERENT &&
820 curdb != state->extrablk.blkno)
821 xfs_trans_brelse(tp, state->extrablk.bp);
822 args->cmpresult = cmp;
823 args->inumber = be64_to_cpu(dep->inumber);
824 args->filetype = dp->d_ops->data_get_ftype(dep);
825 *indexp = index;
826 state->extravalid = 1;
827 state->extrablk.bp = curbp;
828 state->extrablk.blkno = curdb;
829 state->extrablk.index = (int)((char *)dep -
830 (char *)curbp->b_addr);
831 state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
832 curbp->b_ops = &xfs_dir3_data_buf_ops;
833 xfs_trans_buf_set_type(tp, curbp, XFS_BLFT_DIR_DATA_BUF);
834 if (cmp == XFS_CMP_EXACT)
835 return -EEXIST;
836 }
837 }
838 ASSERT(index == leafhdr.count || (args->op_flags & XFS_DA_OP_OKNOENT));
839 if (curbp) {
840 if (args->cmpresult == XFS_CMP_DIFFERENT) {
841 /* Giving back last used data block. */
842 state->extravalid = 1;
843 state->extrablk.bp = curbp;
844 state->extrablk.index = -1;
845 state->extrablk.blkno = curdb;
846 state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
847 curbp->b_ops = &xfs_dir3_data_buf_ops;
848 xfs_trans_buf_set_type(tp, curbp, XFS_BLFT_DIR_DATA_BUF);
849 } else {
850 /* If the curbp is not the CI match block, drop it */
851 if (state->extrablk.bp != curbp)
852 xfs_trans_brelse(tp, curbp);
853 }
854 } else {
855 state->extravalid = 0;
856 }
857 *indexp = index;
858 return -ENOENT;
859 }
860
861 /*
862 * Look up a leaf entry in a node-format leaf block.
863 * If this is an addname then the extrablk in state is a freespace block,
864 * otherwise it's a data block.
865 */
866 int
867 xfs_dir2_leafn_lookup_int(
868 struct xfs_buf *bp, /* leaf buffer */
869 xfs_da_args_t *args, /* operation arguments */
870 int *indexp, /* out: leaf entry index */
871 xfs_da_state_t *state) /* state to fill in */
872 {
873 if (args->op_flags & XFS_DA_OP_ADDNAME)
874 return xfs_dir2_leafn_lookup_for_addname(bp, args, indexp,
875 state);
876 return xfs_dir2_leafn_lookup_for_entry(bp, args, indexp, state);
877 }
878
879 /*
880 * Move count leaf entries from source to destination leaf.
881 * Log entries and headers. Stale entries are preserved.
882 */
883 static void
884 xfs_dir3_leafn_moveents(
885 xfs_da_args_t *args, /* operation arguments */
886 struct xfs_buf *bp_s, /* source */
887 struct xfs_dir3_icleaf_hdr *shdr,
888 struct xfs_dir2_leaf_entry *sents,
889 int start_s,/* source leaf index */
890 struct xfs_buf *bp_d, /* destination */
891 struct xfs_dir3_icleaf_hdr *dhdr,
892 struct xfs_dir2_leaf_entry *dents,
893 int start_d,/* destination leaf index */
894 int count) /* count of leaves to copy */
895 {
896 int stale; /* count stale leaves copied */
897
898 trace_xfs_dir2_leafn_moveents(args, start_s, start_d, count);
899
900 /*
901 * Silently return if nothing to do.
902 */
903 if (count == 0)
904 return;
905
906 /*
907 * If the destination index is not the end of the current
908 * destination leaf entries, open up a hole in the destination
909 * to hold the new entries.
910 */
911 if (start_d < dhdr->count) {
912 memmove(&dents[start_d + count], &dents[start_d],
913 (dhdr->count - start_d) * sizeof(xfs_dir2_leaf_entry_t));
914 xfs_dir3_leaf_log_ents(args, bp_d, start_d + count,
915 count + dhdr->count - 1);
916 }
917 /*
918 * If the source has stale leaves, count the ones in the copy range
919 * so we can update the header correctly.
920 */
921 if (shdr->stale) {
922 int i; /* temp leaf index */
923
924 for (i = start_s, stale = 0; i < start_s + count; i++) {
925 if (sents[i].address ==
926 cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
927 stale++;
928 }
929 } else
930 stale = 0;
931 /*
932 * Copy the leaf entries from source to destination.
933 */
934 memcpy(&dents[start_d], &sents[start_s],
935 count * sizeof(xfs_dir2_leaf_entry_t));
936 xfs_dir3_leaf_log_ents(args, bp_d, start_d, start_d + count - 1);
937
938 /*
939 * If there are source entries after the ones we copied,
940 * delete the ones we copied by sliding the next ones down.
941 */
942 if (start_s + count < shdr->count) {
943 memmove(&sents[start_s], &sents[start_s + count],
944 count * sizeof(xfs_dir2_leaf_entry_t));
945 xfs_dir3_leaf_log_ents(args, bp_s, start_s, start_s + count - 1);
946 }
947
948 /*
949 * Update the headers and log them.
950 */
951 shdr->count -= count;
952 shdr->stale -= stale;
953 dhdr->count += count;
954 dhdr->stale += stale;
955 }
956
957 /*
958 * Determine the sort order of two leaf blocks.
959 * Returns 1 if both are valid and leaf2 should be before leaf1, else 0.
960 */
961 int /* sort order */
962 xfs_dir2_leafn_order(
963 struct xfs_inode *dp,
964 struct xfs_buf *leaf1_bp, /* leaf1 buffer */
965 struct xfs_buf *leaf2_bp) /* leaf2 buffer */
966 {
967 struct xfs_dir2_leaf *leaf1 = leaf1_bp->b_addr;
968 struct xfs_dir2_leaf *leaf2 = leaf2_bp->b_addr;
969 struct xfs_dir2_leaf_entry *ents1;
970 struct xfs_dir2_leaf_entry *ents2;
971 struct xfs_dir3_icleaf_hdr hdr1;
972 struct xfs_dir3_icleaf_hdr hdr2;
973
974 dp->d_ops->leaf_hdr_from_disk(&hdr1, leaf1);
975 dp->d_ops->leaf_hdr_from_disk(&hdr2, leaf2);
976 ents1 = dp->d_ops->leaf_ents_p(leaf1);
977 ents2 = dp->d_ops->leaf_ents_p(leaf2);
978
979 if (hdr1.count > 0 && hdr2.count > 0 &&
980 (be32_to_cpu(ents2[0].hashval) < be32_to_cpu(ents1[0].hashval) ||
981 be32_to_cpu(ents2[hdr2.count - 1].hashval) <
982 be32_to_cpu(ents1[hdr1.count - 1].hashval)))
983 return 1;
984 return 0;
985 }
986
987 /*
988 * Rebalance leaf entries between two leaf blocks.
989 * This is actually only called when the second block is new,
990 * though the code deals with the general case.
991 * A new entry will be inserted in one of the blocks, and that
992 * entry is taken into account when balancing.
993 */
994 static void
995 xfs_dir2_leafn_rebalance(
996 xfs_da_state_t *state, /* btree cursor */
997 xfs_da_state_blk_t *blk1, /* first btree block */
998 xfs_da_state_blk_t *blk2) /* second btree block */
999 {
1000 xfs_da_args_t *args; /* operation arguments */
1001 int count; /* count (& direction) leaves */
1002 int isleft; /* new goes in left leaf */
1003 xfs_dir2_leaf_t *leaf1; /* first leaf structure */
1004 xfs_dir2_leaf_t *leaf2; /* second leaf structure */
1005 int mid; /* midpoint leaf index */
1006 #if defined(DEBUG) || defined(XFS_WARN)
1007 int oldstale; /* old count of stale leaves */
1008 #endif
1009 int oldsum; /* old total leaf count */
1010 int swap; /* swapped leaf blocks */
1011 struct xfs_dir2_leaf_entry *ents1;
1012 struct xfs_dir2_leaf_entry *ents2;
1013 struct xfs_dir3_icleaf_hdr hdr1;
1014 struct xfs_dir3_icleaf_hdr hdr2;
1015 struct xfs_inode *dp = state->args->dp;
1016
1017 args = state->args;
1018 /*
1019 * If the block order is wrong, swap the arguments.
1020 */
1021 if ((swap = xfs_dir2_leafn_order(dp, blk1->bp, blk2->bp))) {
1022 xfs_da_state_blk_t *tmp; /* temp for block swap */
1023
1024 tmp = blk1;
1025 blk1 = blk2;
1026 blk2 = tmp;
1027 }
1028 leaf1 = blk1->bp->b_addr;
1029 leaf2 = blk2->bp->b_addr;
1030 dp->d_ops->leaf_hdr_from_disk(&hdr1, leaf1);
1031 dp->d_ops->leaf_hdr_from_disk(&hdr2, leaf2);
1032 ents1 = dp->d_ops->leaf_ents_p(leaf1);
1033 ents2 = dp->d_ops->leaf_ents_p(leaf2);
1034
1035 oldsum = hdr1.count + hdr2.count;
1036 #if defined(DEBUG) || defined(XFS_WARN)
1037 oldstale = hdr1.stale + hdr2.stale;
1038 #endif
1039 mid = oldsum >> 1;
1040
1041 /*
1042 * If the old leaf count was odd then the new one will be even,
1043 * so we need to divide the new count evenly.
1044 */
1045 if (oldsum & 1) {
1046 xfs_dahash_t midhash; /* middle entry hash value */
1047
1048 if (mid >= hdr1.count)
1049 midhash = be32_to_cpu(ents2[mid - hdr1.count].hashval);
1050 else
1051 midhash = be32_to_cpu(ents1[mid].hashval);
1052 isleft = args->hashval <= midhash;
1053 }
1054 /*
1055 * If the old count is even then the new count is odd, so there's
1056 * no preferred side for the new entry.
1057 * Pick the left one.
1058 */
1059 else
1060 isleft = 1;
1061 /*
1062 * Calculate moved entry count. Positive means left-to-right,
1063 * negative means right-to-left. Then move the entries.
1064 */
1065 count = hdr1.count - mid + (isleft == 0);
1066 if (count > 0)
1067 xfs_dir3_leafn_moveents(args, blk1->bp, &hdr1, ents1,
1068 hdr1.count - count, blk2->bp,
1069 &hdr2, ents2, 0, count);
1070 else if (count < 0)
1071 xfs_dir3_leafn_moveents(args, blk2->bp, &hdr2, ents2, 0,
1072 blk1->bp, &hdr1, ents1,
1073 hdr1.count, count);
1074
1075 ASSERT(hdr1.count + hdr2.count == oldsum);
1076 ASSERT(hdr1.stale + hdr2.stale == oldstale);
1077
1078 /* log the changes made when moving the entries */
1079 dp->d_ops->leaf_hdr_to_disk(leaf1, &hdr1);
1080 dp->d_ops->leaf_hdr_to_disk(leaf2, &hdr2);
1081 xfs_dir3_leaf_log_header(args, blk1->bp);
1082 xfs_dir3_leaf_log_header(args, blk2->bp);
1083
1084 xfs_dir3_leaf_check(dp, blk1->bp);
1085 xfs_dir3_leaf_check(dp, blk2->bp);
1086
1087 /*
1088 * Mark whether we're inserting into the old or new leaf.
1089 */
1090 if (hdr1.count < hdr2.count)
1091 state->inleaf = swap;
1092 else if (hdr1.count > hdr2.count)
1093 state->inleaf = !swap;
1094 else
1095 state->inleaf = swap ^ (blk1->index <= hdr1.count);
1096 /*
1097 * Adjust the expected index for insertion.
1098 */
1099 if (!state->inleaf)
1100 blk2->index = blk1->index - hdr1.count;
1101
1102 /*
1103 * Finally sanity check just to make sure we are not returning a
1104 * negative index
1105 */
1106 if (blk2->index < 0) {
1107 state->inleaf = 1;
1108 blk2->index = 0;
1109 xfs_alert(dp->i_mount,
1110 "%s: picked the wrong leaf? reverting original leaf: blk1->index %d",
1111 __func__, blk1->index);
1112 }
1113 }
1114
1115 static int
1116 xfs_dir3_data_block_free(
1117 xfs_da_args_t *args,
1118 struct xfs_dir2_data_hdr *hdr,
1119 struct xfs_dir2_free *free,
1120 xfs_dir2_db_t fdb,
1121 int findex,
1122 struct xfs_buf *fbp,
1123 int longest)
1124 {
1125 int logfree = 0;
1126 __be16 *bests;
1127 struct xfs_dir3_icfree_hdr freehdr;
1128 struct xfs_inode *dp = args->dp;
1129
1130 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1131 bests = dp->d_ops->free_bests_p(free);
1132 if (hdr) {
1133 /*
1134 * Data block is not empty, just set the free entry to the new
1135 * value.
1136 */
1137 bests[findex] = cpu_to_be16(longest);
1138 xfs_dir2_free_log_bests(args, fbp, findex, findex);
1139 return 0;
1140 }
1141
1142 /* One less used entry in the free table. */
1143 freehdr.nused--;
1144
1145 /*
1146 * If this was the last entry in the table, we can trim the table size
1147 * back. There might be other entries at the end referring to
1148 * non-existent data blocks, get those too.
1149 */
1150 if (findex == freehdr.nvalid - 1) {
1151 int i; /* free entry index */
1152
1153 for (i = findex - 1; i >= 0; i--) {
1154 if (bests[i] != cpu_to_be16(NULLDATAOFF))
1155 break;
1156 }
1157 freehdr.nvalid = i + 1;
1158 logfree = 0;
1159 } else {
1160 /* Not the last entry, just punch it out. */
1161 bests[findex] = cpu_to_be16(NULLDATAOFF);
1162 logfree = 1;
1163 }
1164
1165 dp->d_ops->free_hdr_to_disk(free, &freehdr);
1166 xfs_dir2_free_log_header(args, fbp);
1167
1168 /*
1169 * If there are no useful entries left in the block, get rid of the
1170 * block if we can.
1171 */
1172 if (!freehdr.nused) {
1173 int error;
1174
1175 error = xfs_dir2_shrink_inode(args, fdb, fbp);
1176 if (error == 0) {
1177 fbp = NULL;
1178 logfree = 0;
1179 } else if (error != -ENOSPC || args->total != 0)
1180 return error;
1181 /*
1182 * It's possible to get ENOSPC if there is no
1183 * space reservation. In this case some one
1184 * else will eventually get rid of this block.
1185 */
1186 }
1187
1188 /* Log the free entry that changed, unless we got rid of it. */
1189 if (logfree)
1190 xfs_dir2_free_log_bests(args, fbp, findex, findex);
1191 return 0;
1192 }
1193
1194 /*
1195 * Remove an entry from a node directory.
1196 * This removes the leaf entry and the data entry,
1197 * and updates the free block if necessary.
1198 */
1199 static int /* error */
1200 xfs_dir2_leafn_remove(
1201 xfs_da_args_t *args, /* operation arguments */
1202 struct xfs_buf *bp, /* leaf buffer */
1203 int index, /* leaf entry index */
1204 xfs_da_state_blk_t *dblk, /* data block */
1205 int *rval) /* resulting block needs join */
1206 {
1207 xfs_dir2_data_hdr_t *hdr; /* data block header */
1208 xfs_dir2_db_t db; /* data block number */
1209 struct xfs_buf *dbp; /* data block buffer */
1210 xfs_dir2_data_entry_t *dep; /* data block entry */
1211 xfs_inode_t *dp; /* incore directory inode */
1212 xfs_dir2_leaf_t *leaf; /* leaf structure */
1213 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1214 int longest; /* longest data free entry */
1215 int off; /* data block entry offset */
1216 int needlog; /* need to log data header */
1217 int needscan; /* need to rescan data frees */
1218 xfs_trans_t *tp; /* transaction pointer */
1219 struct xfs_dir2_data_free *bf; /* bestfree table */
1220 struct xfs_dir3_icleaf_hdr leafhdr;
1221 struct xfs_dir2_leaf_entry *ents;
1222
1223 trace_xfs_dir2_leafn_remove(args, index);
1224
1225 dp = args->dp;
1226 tp = args->trans;
1227 leaf = bp->b_addr;
1228 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1229 ents = dp->d_ops->leaf_ents_p(leaf);
1230
1231 /*
1232 * Point to the entry we're removing.
1233 */
1234 lep = &ents[index];
1235
1236 /*
1237 * Extract the data block and offset from the entry.
1238 */
1239 db = xfs_dir2_dataptr_to_db(args->geo, be32_to_cpu(lep->address));
1240 ASSERT(dblk->blkno == db);
1241 off = xfs_dir2_dataptr_to_off(args->geo, be32_to_cpu(lep->address));
1242 ASSERT(dblk->index == off);
1243
1244 /*
1245 * Kill the leaf entry by marking it stale.
1246 * Log the leaf block changes.
1247 */
1248 leafhdr.stale++;
1249 dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
1250 xfs_dir3_leaf_log_header(args, bp);
1251
1252 lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1253 xfs_dir3_leaf_log_ents(args, bp, index, index);
1254
1255 /*
1256 * Make the data entry free. Keep track of the longest freespace
1257 * in the data block in case it changes.
1258 */
1259 dbp = dblk->bp;
1260 hdr = dbp->b_addr;
1261 dep = (xfs_dir2_data_entry_t *)((char *)hdr + off);
1262 bf = dp->d_ops->data_bestfree_p(hdr);
1263 longest = be16_to_cpu(bf[0].length);
1264 needlog = needscan = 0;
1265 xfs_dir2_data_make_free(args, dbp, off,
1266 dp->d_ops->data_entsize(dep->namelen), &needlog, &needscan);
1267 /*
1268 * Rescan the data block freespaces for bestfree.
1269 * Log the data block header if needed.
1270 */
1271 if (needscan)
1272 xfs_dir2_data_freescan(dp, hdr, &needlog);
1273 if (needlog)
1274 xfs_dir2_data_log_header(args, dbp);
1275 xfs_dir3_data_check(dp, dbp);
1276 /*
1277 * If the longest data block freespace changes, need to update
1278 * the corresponding freeblock entry.
1279 */
1280 if (longest < be16_to_cpu(bf[0].length)) {
1281 int error; /* error return value */
1282 struct xfs_buf *fbp; /* freeblock buffer */
1283 xfs_dir2_db_t fdb; /* freeblock block number */
1284 int findex; /* index in freeblock entries */
1285 xfs_dir2_free_t *free; /* freeblock structure */
1286
1287 /*
1288 * Convert the data block number to a free block,
1289 * read in the free block.
1290 */
1291 fdb = dp->d_ops->db_to_fdb(args->geo, db);
1292 error = xfs_dir2_free_read(tp, dp,
1293 xfs_dir2_db_to_da(args->geo, fdb),
1294 &fbp);
1295 if (error)
1296 return error;
1297 free = fbp->b_addr;
1298 #ifdef DEBUG
1299 {
1300 struct xfs_dir3_icfree_hdr freehdr;
1301 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1302 ASSERT(freehdr.firstdb == dp->d_ops->free_max_bests(args->geo) *
1303 (fdb - xfs_dir2_byte_to_db(args->geo,
1304 XFS_DIR2_FREE_OFFSET)));
1305 }
1306 #endif
1307 /*
1308 * Calculate which entry we need to fix.
1309 */
1310 findex = dp->d_ops->db_to_fdindex(args->geo, db);
1311 longest = be16_to_cpu(bf[0].length);
1312 /*
1313 * If the data block is now empty we can get rid of it
1314 * (usually).
1315 */
1316 if (longest == args->geo->blksize -
1317 dp->d_ops->data_entry_offset) {
1318 /*
1319 * Try to punch out the data block.
1320 */
1321 error = xfs_dir2_shrink_inode(args, db, dbp);
1322 if (error == 0) {
1323 dblk->bp = NULL;
1324 hdr = NULL;
1325 }
1326 /*
1327 * We can get ENOSPC if there's no space reservation.
1328 * In this case just drop the buffer and some one else
1329 * will eventually get rid of the empty block.
1330 */
1331 else if (!(error == -ENOSPC && args->total == 0))
1332 return error;
1333 }
1334 /*
1335 * If we got rid of the data block, we can eliminate that entry
1336 * in the free block.
1337 */
1338 error = xfs_dir3_data_block_free(args, hdr, free,
1339 fdb, findex, fbp, longest);
1340 if (error)
1341 return error;
1342 }
1343
1344 xfs_dir3_leaf_check(dp, bp);
1345 /*
1346 * Return indication of whether this leaf block is empty enough
1347 * to justify trying to join it with a neighbor.
1348 */
1349 *rval = (dp->d_ops->leaf_hdr_size +
1350 (uint)sizeof(ents[0]) * (leafhdr.count - leafhdr.stale)) <
1351 args->geo->magicpct;
1352 return 0;
1353 }
1354
1355 /*
1356 * Split the leaf entries in the old block into old and new blocks.
1357 */
1358 int /* error */
1359 xfs_dir2_leafn_split(
1360 xfs_da_state_t *state, /* btree cursor */
1361 xfs_da_state_blk_t *oldblk, /* original block */
1362 xfs_da_state_blk_t *newblk) /* newly created block */
1363 {
1364 xfs_da_args_t *args; /* operation arguments */
1365 xfs_dablk_t blkno; /* new leaf block number */
1366 int error; /* error return value */
1367 struct xfs_inode *dp;
1368
1369 /*
1370 * Allocate space for a new leaf node.
1371 */
1372 args = state->args;
1373 dp = args->dp;
1374 ASSERT(oldblk->magic == XFS_DIR2_LEAFN_MAGIC);
1375 error = xfs_da_grow_inode(args, &blkno);
1376 if (error) {
1377 return error;
1378 }
1379 /*
1380 * Initialize the new leaf block.
1381 */
1382 error = xfs_dir3_leaf_get_buf(args, xfs_dir2_da_to_db(args->geo, blkno),
1383 &newblk->bp, XFS_DIR2_LEAFN_MAGIC);
1384 if (error)
1385 return error;
1386
1387 newblk->blkno = blkno;
1388 newblk->magic = XFS_DIR2_LEAFN_MAGIC;
1389 /*
1390 * Rebalance the entries across the two leaves, link the new
1391 * block into the leaves.
1392 */
1393 xfs_dir2_leafn_rebalance(state, oldblk, newblk);
1394 error = xfs_da3_blk_link(state, oldblk, newblk);
1395 if (error) {
1396 return error;
1397 }
1398 /*
1399 * Insert the new entry in the correct block.
1400 */
1401 if (state->inleaf)
1402 error = xfs_dir2_leafn_add(oldblk->bp, args, oldblk->index);
1403 else
1404 error = xfs_dir2_leafn_add(newblk->bp, args, newblk->index);
1405 /*
1406 * Update last hashval in each block since we added the name.
1407 */
1408 oldblk->hashval = xfs_dir2_leafn_lasthash(dp, oldblk->bp, NULL);
1409 newblk->hashval = xfs_dir2_leafn_lasthash(dp, newblk->bp, NULL);
1410 xfs_dir3_leaf_check(dp, oldblk->bp);
1411 xfs_dir3_leaf_check(dp, newblk->bp);
1412 return error;
1413 }
1414
1415 /*
1416 * Check a leaf block and its neighbors to see if the block should be
1417 * collapsed into one or the other neighbor. Always keep the block
1418 * with the smaller block number.
1419 * If the current block is over 50% full, don't try to join it, return 0.
1420 * If the block is empty, fill in the state structure and return 2.
1421 * If it can be collapsed, fill in the state structure and return 1.
1422 * If nothing can be done, return 0.
1423 */
1424 int /* error */
1425 xfs_dir2_leafn_toosmall(
1426 xfs_da_state_t *state, /* btree cursor */
1427 int *action) /* resulting action to take */
1428 {
1429 xfs_da_state_blk_t *blk; /* leaf block */
1430 xfs_dablk_t blkno; /* leaf block number */
1431 struct xfs_buf *bp; /* leaf buffer */
1432 int bytes; /* bytes in use */
1433 int count; /* leaf live entry count */
1434 int error; /* error return value */
1435 int forward; /* sibling block direction */
1436 int i; /* sibling counter */
1437 xfs_dir2_leaf_t *leaf; /* leaf structure */
1438 int rval; /* result from path_shift */
1439 struct xfs_dir3_icleaf_hdr leafhdr;
1440 struct xfs_dir2_leaf_entry *ents;
1441 struct xfs_inode *dp = state->args->dp;
1442
1443 /*
1444 * Check for the degenerate case of the block being over 50% full.
1445 * If so, it's not worth even looking to see if we might be able
1446 * to coalesce with a sibling.
1447 */
1448 blk = &state->path.blk[state->path.active - 1];
1449 leaf = blk->bp->b_addr;
1450 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1451 ents = dp->d_ops->leaf_ents_p(leaf);
1452 xfs_dir3_leaf_check(dp, blk->bp);
1453
1454 count = leafhdr.count - leafhdr.stale;
1455 bytes = dp->d_ops->leaf_hdr_size + count * sizeof(ents[0]);
1456 if (bytes > (state->args->geo->blksize >> 1)) {
1457 /*
1458 * Blk over 50%, don't try to join.
1459 */
1460 *action = 0;
1461 return 0;
1462 }
1463 /*
1464 * Check for the degenerate case of the block being empty.
1465 * If the block is empty, we'll simply delete it, no need to
1466 * coalesce it with a sibling block. We choose (arbitrarily)
1467 * to merge with the forward block unless it is NULL.
1468 */
1469 if (count == 0) {
1470 /*
1471 * Make altpath point to the block we want to keep and
1472 * path point to the block we want to drop (this one).
1473 */
1474 forward = (leafhdr.forw != 0);
1475 memcpy(&state->altpath, &state->path, sizeof(state->path));
1476 error = xfs_da3_path_shift(state, &state->altpath, forward, 0,
1477 &rval);
1478 if (error)
1479 return error;
1480 *action = rval ? 2 : 0;
1481 return 0;
1482 }
1483 /*
1484 * Examine each sibling block to see if we can coalesce with
1485 * at least 25% free space to spare. We need to figure out
1486 * whether to merge with the forward or the backward block.
1487 * We prefer coalescing with the lower numbered sibling so as
1488 * to shrink a directory over time.
1489 */
1490 forward = leafhdr.forw < leafhdr.back;
1491 for (i = 0, bp = NULL; i < 2; forward = !forward, i++) {
1492 struct xfs_dir3_icleaf_hdr hdr2;
1493
1494 blkno = forward ? leafhdr.forw : leafhdr.back;
1495 if (blkno == 0)
1496 continue;
1497 /*
1498 * Read the sibling leaf block.
1499 */
1500 error = xfs_dir3_leafn_read(state->args->trans, dp,
1501 blkno, -1, &bp);
1502 if (error)
1503 return error;
1504
1505 /*
1506 * Count bytes in the two blocks combined.
1507 */
1508 count = leafhdr.count - leafhdr.stale;
1509 bytes = state->args->geo->blksize -
1510 (state->args->geo->blksize >> 2);
1511
1512 leaf = bp->b_addr;
1513 dp->d_ops->leaf_hdr_from_disk(&hdr2, leaf);
1514 ents = dp->d_ops->leaf_ents_p(leaf);
1515 count += hdr2.count - hdr2.stale;
1516 bytes -= count * sizeof(ents[0]);
1517
1518 /*
1519 * Fits with at least 25% to spare.
1520 */
1521 if (bytes >= 0)
1522 break;
1523 xfs_trans_brelse(state->args->trans, bp);
1524 }
1525 /*
1526 * Didn't like either block, give up.
1527 */
1528 if (i >= 2) {
1529 *action = 0;
1530 return 0;
1531 }
1532
1533 /*
1534 * Make altpath point to the block we want to keep (the lower
1535 * numbered block) and path point to the block we want to drop.
1536 */
1537 memcpy(&state->altpath, &state->path, sizeof(state->path));
1538 if (blkno < blk->blkno)
1539 error = xfs_da3_path_shift(state, &state->altpath, forward, 0,
1540 &rval);
1541 else
1542 error = xfs_da3_path_shift(state, &state->path, forward, 0,
1543 &rval);
1544 if (error) {
1545 return error;
1546 }
1547 *action = rval ? 0 : 1;
1548 return 0;
1549 }
1550
1551 /*
1552 * Move all the leaf entries from drop_blk to save_blk.
1553 * This is done as part of a join operation.
1554 */
1555 void
1556 xfs_dir2_leafn_unbalance(
1557 xfs_da_state_t *state, /* cursor */
1558 xfs_da_state_blk_t *drop_blk, /* dead block */
1559 xfs_da_state_blk_t *save_blk) /* surviving block */
1560 {
1561 xfs_da_args_t *args; /* operation arguments */
1562 xfs_dir2_leaf_t *drop_leaf; /* dead leaf structure */
1563 xfs_dir2_leaf_t *save_leaf; /* surviving leaf structure */
1564 struct xfs_dir3_icleaf_hdr savehdr;
1565 struct xfs_dir3_icleaf_hdr drophdr;
1566 struct xfs_dir2_leaf_entry *sents;
1567 struct xfs_dir2_leaf_entry *dents;
1568 struct xfs_inode *dp = state->args->dp;
1569
1570 args = state->args;
1571 ASSERT(drop_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1572 ASSERT(save_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1573 drop_leaf = drop_blk->bp->b_addr;
1574 save_leaf = save_blk->bp->b_addr;
1575
1576 dp->d_ops->leaf_hdr_from_disk(&savehdr, save_leaf);
1577 dp->d_ops->leaf_hdr_from_disk(&drophdr, drop_leaf);
1578 sents = dp->d_ops->leaf_ents_p(save_leaf);
1579 dents = dp->d_ops->leaf_ents_p(drop_leaf);
1580
1581 /*
1582 * If there are any stale leaf entries, take this opportunity
1583 * to purge them.
1584 */
1585 if (drophdr.stale)
1586 xfs_dir3_leaf_compact(args, &drophdr, drop_blk->bp);
1587 if (savehdr.stale)
1588 xfs_dir3_leaf_compact(args, &savehdr, save_blk->bp);
1589
1590 /*
1591 * Move the entries from drop to the appropriate end of save.
1592 */
1593 drop_blk->hashval = be32_to_cpu(dents[drophdr.count - 1].hashval);
1594 if (xfs_dir2_leafn_order(dp, save_blk->bp, drop_blk->bp))
1595 xfs_dir3_leafn_moveents(args, drop_blk->bp, &drophdr, dents, 0,
1596 save_blk->bp, &savehdr, sents, 0,
1597 drophdr.count);
1598 else
1599 xfs_dir3_leafn_moveents(args, drop_blk->bp, &drophdr, dents, 0,
1600 save_blk->bp, &savehdr, sents,
1601 savehdr.count, drophdr.count);
1602 save_blk->hashval = be32_to_cpu(sents[savehdr.count - 1].hashval);
1603
1604 /* log the changes made when moving the entries */
1605 dp->d_ops->leaf_hdr_to_disk(save_leaf, &savehdr);
1606 dp->d_ops->leaf_hdr_to_disk(drop_leaf, &drophdr);
1607 xfs_dir3_leaf_log_header(args, save_blk->bp);
1608 xfs_dir3_leaf_log_header(args, drop_blk->bp);
1609
1610 xfs_dir3_leaf_check(dp, save_blk->bp);
1611 xfs_dir3_leaf_check(dp, drop_blk->bp);
1612 }
1613
1614 /*
1615 * Top-level node form directory addname routine.
1616 */
1617 int /* error */
1618 xfs_dir2_node_addname(
1619 xfs_da_args_t *args) /* operation arguments */
1620 {
1621 xfs_da_state_blk_t *blk; /* leaf block for insert */
1622 int error; /* error return value */
1623 int rval; /* sub-return value */
1624 xfs_da_state_t *state; /* btree cursor */
1625
1626 trace_xfs_dir2_node_addname(args);
1627
1628 /*
1629 * Allocate and initialize the state (btree cursor).
1630 */
1631 state = xfs_da_state_alloc();
1632 state->args = args;
1633 state->mp = args->dp->i_mount;
1634 /*
1635 * Look up the name. We're not supposed to find it, but
1636 * this gives us the insertion point.
1637 */
1638 error = xfs_da3_node_lookup_int(state, &rval);
1639 if (error)
1640 rval = error;
1641 if (rval != -ENOENT) {
1642 goto done;
1643 }
1644 /*
1645 * Add the data entry to a data block.
1646 * Extravalid is set to a freeblock found by lookup.
1647 */
1648 rval = xfs_dir2_node_addname_int(args,
1649 state->extravalid ? &state->extrablk : NULL);
1650 if (rval) {
1651 goto done;
1652 }
1653 blk = &state->path.blk[state->path.active - 1];
1654 ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
1655 /*
1656 * Add the new leaf entry.
1657 */
1658 rval = xfs_dir2_leafn_add(blk->bp, args, blk->index);
1659 if (rval == 0) {
1660 /*
1661 * It worked, fix the hash values up the btree.
1662 */
1663 if (!(args->op_flags & XFS_DA_OP_JUSTCHECK))
1664 xfs_da3_fixhashpath(state, &state->path);
1665 } else {
1666 /*
1667 * It didn't work, we need to split the leaf block.
1668 */
1669 if (args->total == 0) {
1670 ASSERT(rval == -ENOSPC);
1671 goto done;
1672 }
1673 /*
1674 * Split the leaf block and insert the new entry.
1675 */
1676 rval = xfs_da3_split(state);
1677 }
1678 done:
1679 xfs_da_state_free(state);
1680 return rval;
1681 }
1682
1683 /*
1684 * Add the data entry for a node-format directory name addition.
1685 * The leaf entry is added in xfs_dir2_leafn_add.
1686 * We may enter with a freespace block that the lookup found.
1687 */
1688 static int /* error */
1689 xfs_dir2_node_addname_int(
1690 xfs_da_args_t *args, /* operation arguments */
1691 xfs_da_state_blk_t *fblk) /* optional freespace block */
1692 {
1693 xfs_dir2_data_hdr_t *hdr; /* data block header */
1694 xfs_dir2_db_t dbno; /* data block number */
1695 struct xfs_buf *dbp; /* data block buffer */
1696 xfs_dir2_data_entry_t *dep; /* data entry pointer */
1697 xfs_inode_t *dp; /* incore directory inode */
1698 xfs_dir2_data_unused_t *dup; /* data unused entry pointer */
1699 int error; /* error return value */
1700 xfs_dir2_db_t fbno; /* freespace block number */
1701 struct xfs_buf *fbp; /* freespace buffer */
1702 int findex; /* freespace entry index */
1703 xfs_dir2_free_t *free=NULL; /* freespace block structure */
1704 xfs_dir2_db_t ifbno; /* initial freespace block no */
1705 xfs_dir2_db_t lastfbno=0; /* highest freespace block no */
1706 int length; /* length of the new entry */
1707 int logfree; /* need to log free entry */
1708 xfs_mount_t *mp; /* filesystem mount point */
1709 int needlog; /* need to log data header */
1710 int needscan; /* need to rescan data frees */
1711 __be16 *tagp; /* data entry tag pointer */
1712 xfs_trans_t *tp; /* transaction pointer */
1713 __be16 *bests;
1714 struct xfs_dir3_icfree_hdr freehdr;
1715 struct xfs_dir2_data_free *bf;
1716
1717 dp = args->dp;
1718 mp = dp->i_mount;
1719 tp = args->trans;
1720 length = dp->d_ops->data_entsize(args->namelen);
1721 /*
1722 * If we came in with a freespace block that means that lookup
1723 * found an entry with our hash value. This is the freespace
1724 * block for that data entry.
1725 */
1726 if (fblk) {
1727 fbp = fblk->bp;
1728 /*
1729 * Remember initial freespace block number.
1730 */
1731 ifbno = fblk->blkno;
1732 free = fbp->b_addr;
1733 findex = fblk->index;
1734 bests = dp->d_ops->free_bests_p(free);
1735 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1736
1737 /*
1738 * This means the free entry showed that the data block had
1739 * space for our entry, so we remembered it.
1740 * Use that data block.
1741 */
1742 if (findex >= 0) {
1743 ASSERT(findex < freehdr.nvalid);
1744 ASSERT(be16_to_cpu(bests[findex]) != NULLDATAOFF);
1745 ASSERT(be16_to_cpu(bests[findex]) >= length);
1746 dbno = freehdr.firstdb + findex;
1747 } else {
1748 /*
1749 * The data block looked at didn't have enough room.
1750 * We'll start at the beginning of the freespace entries.
1751 */
1752 dbno = -1;
1753 findex = 0;
1754 }
1755 } else {
1756 /*
1757 * Didn't come in with a freespace block, so no data block.
1758 */
1759 ifbno = dbno = -1;
1760 fbp = NULL;
1761 findex = 0;
1762 }
1763
1764 /*
1765 * If we don't have a data block yet, we're going to scan the
1766 * freespace blocks looking for one. Figure out what the
1767 * highest freespace block number is.
1768 */
1769 if (dbno == -1) {
1770 xfs_fileoff_t fo; /* freespace block number */
1771
1772 if ((error = xfs_bmap_last_offset(dp, &fo, XFS_DATA_FORK)))
1773 return error;
1774 lastfbno = xfs_dir2_da_to_db(args->geo, (xfs_dablk_t)fo);
1775 fbno = ifbno;
1776 }
1777 /*
1778 * While we haven't identified a data block, search the freeblock
1779 * data for a good data block. If we find a null freeblock entry,
1780 * indicating a hole in the data blocks, remember that.
1781 */
1782 while (dbno == -1) {
1783 /*
1784 * If we don't have a freeblock in hand, get the next one.
1785 */
1786 if (fbp == NULL) {
1787 /*
1788 * Happens the first time through unless lookup gave
1789 * us a freespace block to start with.
1790 */
1791 if (++fbno == 0)
1792 fbno = xfs_dir2_byte_to_db(args->geo,
1793 XFS_DIR2_FREE_OFFSET);
1794 /*
1795 * If it's ifbno we already looked at it.
1796 */
1797 if (fbno == ifbno)
1798 fbno++;
1799 /*
1800 * If it's off the end we're done.
1801 */
1802 if (fbno >= lastfbno)
1803 break;
1804 /*
1805 * Read the block. There can be holes in the
1806 * freespace blocks, so this might not succeed.
1807 * This should be really rare, so there's no reason
1808 * to avoid it.
1809 */
1810 error = xfs_dir2_free_try_read(tp, dp,
1811 xfs_dir2_db_to_da(args->geo, fbno),
1812 &fbp);
1813 if (error)
1814 return error;
1815 if (!fbp)
1816 continue;
1817 free = fbp->b_addr;
1818 findex = 0;
1819 }
1820 /*
1821 * Look at the current free entry. Is it good enough?
1822 *
1823 * The bests initialisation should be where the bufer is read in
1824 * the above branch. But gcc is too stupid to realise that bests
1825 * and the freehdr are actually initialised if they are placed
1826 * there, so we have to do it here to avoid warnings. Blech.
1827 */
1828 bests = dp->d_ops->free_bests_p(free);
1829 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1830 if (be16_to_cpu(bests[findex]) != NULLDATAOFF &&
1831 be16_to_cpu(bests[findex]) >= length)
1832 dbno = freehdr.firstdb + findex;
1833 else {
1834 /*
1835 * Are we done with the freeblock?
1836 */
1837 if (++findex == freehdr.nvalid) {
1838 /*
1839 * Drop the block.
1840 */
1841 xfs_trans_brelse(tp, fbp);
1842 fbp = NULL;
1843 if (fblk && fblk->bp)
1844 fblk->bp = NULL;
1845 }
1846 }
1847 }
1848 /*
1849 * If we don't have a data block, we need to allocate one and make
1850 * the freespace entries refer to it.
1851 */
1852 if (unlikely(dbno == -1)) {
1853 /*
1854 * Not allowed to allocate, return failure.
1855 */
1856 if ((args->op_flags & XFS_DA_OP_JUSTCHECK) || args->total == 0)
1857 return -ENOSPC;
1858
1859 /*
1860 * Allocate and initialize the new data block.
1861 */
1862 if (unlikely((error = xfs_dir2_grow_inode(args,
1863 XFS_DIR2_DATA_SPACE,
1864 &dbno)) ||
1865 (error = xfs_dir3_data_init(args, dbno, &dbp))))
1866 return error;
1867
1868 /*
1869 * If (somehow) we have a freespace block, get rid of it.
1870 */
1871 if (fbp)
1872 xfs_trans_brelse(tp, fbp);
1873 if (fblk && fblk->bp)
1874 fblk->bp = NULL;
1875
1876 /*
1877 * Get the freespace block corresponding to the data block
1878 * that was just allocated.
1879 */
1880 fbno = dp->d_ops->db_to_fdb(args->geo, dbno);
1881 error = xfs_dir2_free_try_read(tp, dp,
1882 xfs_dir2_db_to_da(args->geo, fbno),
1883 &fbp);
1884 if (error)
1885 return error;
1886
1887 /*
1888 * If there wasn't a freespace block, the read will
1889 * return a NULL fbp. Allocate and initialize a new one.
1890 */
1891 if (!fbp) {
1892 error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE,
1893 &fbno);
1894 if (error)
1895 return error;
1896
1897 if (dp->d_ops->db_to_fdb(args->geo, dbno) != fbno) {
1898 xfs_alert(mp,
1899 "%s: dir ino %llu needed freesp block %lld for data block %lld, got %lld ifbno %llu lastfbno %d",
1900 __func__, (unsigned long long)dp->i_ino,
1901 (long long)dp->d_ops->db_to_fdb(
1902 args->geo, dbno),
1903 (long long)dbno, (long long)fbno,
1904 (unsigned long long)ifbno, lastfbno);
1905 if (fblk) {
1906 xfs_alert(mp,
1907 " fblk 0x%p blkno %llu index %d magic 0x%x",
1908 fblk,
1909 (unsigned long long)fblk->blkno,
1910 fblk->index,
1911 fblk->magic);
1912 } else {
1913 xfs_alert(mp, " ... fblk is NULL");
1914 }
1915 XFS_ERROR_REPORT("xfs_dir2_node_addname_int",
1916 XFS_ERRLEVEL_LOW, mp);
1917 return -EFSCORRUPTED;
1918 }
1919
1920 /*
1921 * Get a buffer for the new block.
1922 */
1923 error = xfs_dir3_free_get_buf(args, fbno, &fbp);
1924 if (error)
1925 return error;
1926 free = fbp->b_addr;
1927 bests = dp->d_ops->free_bests_p(free);
1928 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1929
1930 /*
1931 * Remember the first slot as our empty slot.
1932 */
1933 freehdr.firstdb =
1934 (fbno - xfs_dir2_byte_to_db(args->geo,
1935 XFS_DIR2_FREE_OFFSET)) *
1936 dp->d_ops->free_max_bests(args->geo);
1937 } else {
1938 free = fbp->b_addr;
1939 bests = dp->d_ops->free_bests_p(free);
1940 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1941 }
1942
1943 /*
1944 * Set the freespace block index from the data block number.
1945 */
1946 findex = dp->d_ops->db_to_fdindex(args->geo, dbno);
1947 /*
1948 * If it's after the end of the current entries in the
1949 * freespace block, extend that table.
1950 */
1951 if (findex >= freehdr.nvalid) {
1952 ASSERT(findex < dp->d_ops->free_max_bests(args->geo));
1953 freehdr.nvalid = findex + 1;
1954 /*
1955 * Tag new entry so nused will go up.
1956 */
1957 bests[findex] = cpu_to_be16(NULLDATAOFF);
1958 }
1959 /*
1960 * If this entry was for an empty data block
1961 * (this should always be true) then update the header.
1962 */
1963 if (bests[findex] == cpu_to_be16(NULLDATAOFF)) {
1964 freehdr.nused++;
1965 dp->d_ops->free_hdr_to_disk(fbp->b_addr, &freehdr);
1966 xfs_dir2_free_log_header(args, fbp);
1967 }
1968 /*
1969 * Update the real value in the table.
1970 * We haven't allocated the data entry yet so this will
1971 * change again.
1972 */
1973 hdr = dbp->b_addr;
1974 bf = dp->d_ops->data_bestfree_p(hdr);
1975 bests[findex] = bf[0].length;
1976 logfree = 1;
1977 }
1978 /*
1979 * We had a data block so we don't have to make a new one.
1980 */
1981 else {
1982 /*
1983 * If just checking, we succeeded.
1984 */
1985 if (args->op_flags & XFS_DA_OP_JUSTCHECK)
1986 return 0;
1987
1988 /*
1989 * Read the data block in.
1990 */
1991 error = xfs_dir3_data_read(tp, dp,
1992 xfs_dir2_db_to_da(args->geo, dbno),
1993 -1, &dbp);
1994 if (error)
1995 return error;
1996 hdr = dbp->b_addr;
1997 bf = dp->d_ops->data_bestfree_p(hdr);
1998 logfree = 0;
1999 }
2000 ASSERT(be16_to_cpu(bf[0].length) >= length);
2001 /*
2002 * Point to the existing unused space.
2003 */
2004 dup = (xfs_dir2_data_unused_t *)
2005 ((char *)hdr + be16_to_cpu(bf[0].offset));
2006 needscan = needlog = 0;
2007 /*
2008 * Mark the first part of the unused space, inuse for us.
2009 */
2010 xfs_dir2_data_use_free(args, dbp, dup,
2011 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr), length,
2012 &needlog, &needscan);
2013 /*
2014 * Fill in the new entry and log it.
2015 */
2016 dep = (xfs_dir2_data_entry_t *)dup;
2017 dep->inumber = cpu_to_be64(args->inumber);
2018 dep->namelen = args->namelen;
2019 memcpy(dep->name, args->name, dep->namelen);
2020 dp->d_ops->data_put_ftype(dep, args->filetype);
2021 tagp = dp->d_ops->data_entry_tag_p(dep);
2022 *tagp = cpu_to_be16((char *)dep - (char *)hdr);
2023 xfs_dir2_data_log_entry(args, dbp, dep);
2024 /*
2025 * Rescan the block for bestfree if needed.
2026 */
2027 if (needscan)
2028 xfs_dir2_data_freescan(dp, hdr, &needlog);
2029 /*
2030 * Log the data block header if needed.
2031 */
2032 if (needlog)
2033 xfs_dir2_data_log_header(args, dbp);
2034 /*
2035 * If the freespace entry is now wrong, update it.
2036 */
2037 bests = dp->d_ops->free_bests_p(free); /* gcc is so stupid */
2038 if (be16_to_cpu(bests[findex]) != be16_to_cpu(bf[0].length)) {
2039 bests[findex] = bf[0].length;
2040 logfree = 1;
2041 }
2042 /*
2043 * Log the freespace entry if needed.
2044 */
2045 if (logfree)
2046 xfs_dir2_free_log_bests(args, fbp, findex, findex);
2047 /*
2048 * Return the data block and offset in args, then drop the data block.
2049 */
2050 args->blkno = (xfs_dablk_t)dbno;
2051 args->index = be16_to_cpu(*tagp);
2052 return 0;
2053 }
2054
2055 /*
2056 * Lookup an entry in a node-format directory.
2057 * All the real work happens in xfs_da3_node_lookup_int.
2058 * The only real output is the inode number of the entry.
2059 */
2060 int /* error */
2061 xfs_dir2_node_lookup(
2062 xfs_da_args_t *args) /* operation arguments */
2063 {
2064 int error; /* error return value */
2065 int i; /* btree level */
2066 int rval; /* operation return value */
2067 xfs_da_state_t *state; /* btree cursor */
2068
2069 trace_xfs_dir2_node_lookup(args);
2070
2071 /*
2072 * Allocate and initialize the btree cursor.
2073 */
2074 state = xfs_da_state_alloc();
2075 state->args = args;
2076 state->mp = args->dp->i_mount;
2077 /*
2078 * Fill in the path to the entry in the cursor.
2079 */
2080 error = xfs_da3_node_lookup_int(state, &rval);
2081 if (error)
2082 rval = error;
2083 else if (rval == -ENOENT && args->cmpresult == XFS_CMP_CASE) {
2084 /* If a CI match, dup the actual name and return -EEXIST */
2085 xfs_dir2_data_entry_t *dep;
2086
2087 dep = (xfs_dir2_data_entry_t *)
2088 ((char *)state->extrablk.bp->b_addr +
2089 state->extrablk.index);
2090 rval = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
2091 }
2092 /*
2093 * Release the btree blocks and leaf block.
2094 */
2095 for (i = 0; i < state->path.active; i++) {
2096 xfs_trans_brelse(args->trans, state->path.blk[i].bp);
2097 state->path.blk[i].bp = NULL;
2098 }
2099 /*
2100 * Release the data block if we have it.
2101 */
2102 if (state->extravalid && state->extrablk.bp) {
2103 xfs_trans_brelse(args->trans, state->extrablk.bp);
2104 state->extrablk.bp = NULL;
2105 }
2106 xfs_da_state_free(state);
2107 return rval;
2108 }
2109
2110 /*
2111 * Remove an entry from a node-format directory.
2112 */
2113 int /* error */
2114 xfs_dir2_node_removename(
2115 struct xfs_da_args *args) /* operation arguments */
2116 {
2117 struct xfs_da_state_blk *blk; /* leaf block */
2118 int error; /* error return value */
2119 int rval; /* operation return value */
2120 struct xfs_da_state *state; /* btree cursor */
2121
2122 trace_xfs_dir2_node_removename(args);
2123
2124 /*
2125 * Allocate and initialize the btree cursor.
2126 */
2127 state = xfs_da_state_alloc();
2128 state->args = args;
2129 state->mp = args->dp->i_mount;
2130
2131 /* Look up the entry we're deleting, set up the cursor. */
2132 error = xfs_da3_node_lookup_int(state, &rval);
2133 if (error)
2134 goto out_free;
2135
2136 /* Didn't find it, upper layer screwed up. */
2137 if (rval != -EEXIST) {
2138 error = rval;
2139 goto out_free;
2140 }
2141
2142 blk = &state->path.blk[state->path.active - 1];
2143 ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
2144 ASSERT(state->extravalid);
2145 /*
2146 * Remove the leaf and data entries.
2147 * Extrablk refers to the data block.
2148 */
2149 error = xfs_dir2_leafn_remove(args, blk->bp, blk->index,
2150 &state->extrablk, &rval);
2151 if (error)
2152 goto out_free;
2153 /*
2154 * Fix the hash values up the btree.
2155 */
2156 xfs_da3_fixhashpath(state, &state->path);
2157 /*
2158 * If we need to join leaf blocks, do it.
2159 */
2160 if (rval && state->path.active > 1)
2161 error = xfs_da3_join(state);
2162 /*
2163 * If no errors so far, try conversion to leaf format.
2164 */
2165 if (!error)
2166 error = xfs_dir2_node_to_leaf(state);
2167 out_free:
2168 xfs_da_state_free(state);
2169 return error;
2170 }
2171
2172 /*
2173 * Replace an entry's inode number in a node-format directory.
2174 */
2175 int /* error */
2176 xfs_dir2_node_replace(
2177 xfs_da_args_t *args) /* operation arguments */
2178 {
2179 xfs_da_state_blk_t *blk; /* leaf block */
2180 xfs_dir2_data_hdr_t *hdr; /* data block header */
2181 xfs_dir2_data_entry_t *dep; /* data entry changed */
2182 int error; /* error return value */
2183 int i; /* btree level */
2184 xfs_ino_t inum; /* new inode number */
2185 int ftype; /* new file type */
2186 xfs_dir2_leaf_t *leaf; /* leaf structure */
2187 xfs_dir2_leaf_entry_t *lep; /* leaf entry being changed */
2188 int rval; /* internal return value */
2189 xfs_da_state_t *state; /* btree cursor */
2190
2191 trace_xfs_dir2_node_replace(args);
2192
2193 /*
2194 * Allocate and initialize the btree cursor.
2195 */
2196 state = xfs_da_state_alloc();
2197 state->args = args;
2198 state->mp = args->dp->i_mount;
2199
2200 /*
2201 * We have to save new inode number and ftype since
2202 * xfs_da3_node_lookup_int() is going to overwrite them
2203 */
2204 inum = args->inumber;
2205 ftype = args->filetype;
2206
2207 /*
2208 * Lookup the entry to change in the btree.
2209 */
2210 error = xfs_da3_node_lookup_int(state, &rval);
2211 if (error) {
2212 rval = error;
2213 }
2214 /*
2215 * It should be found, since the vnodeops layer has looked it up
2216 * and locked it. But paranoia is good.
2217 */
2218 if (rval == -EEXIST) {
2219 struct xfs_dir2_leaf_entry *ents;
2220 /*
2221 * Find the leaf entry.
2222 */
2223 blk = &state->path.blk[state->path.active - 1];
2224 ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
2225 leaf = blk->bp->b_addr;
2226 ents = args->dp->d_ops->leaf_ents_p(leaf);
2227 lep = &ents[blk->index];
2228 ASSERT(state->extravalid);
2229 /*
2230 * Point to the data entry.
2231 */
2232 hdr = state->extrablk.bp->b_addr;
2233 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
2234 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC));
2235 dep = (xfs_dir2_data_entry_t *)
2236 ((char *)hdr +
2237 xfs_dir2_dataptr_to_off(args->geo,
2238 be32_to_cpu(lep->address)));
2239 ASSERT(inum != be64_to_cpu(dep->inumber));
2240 /*
2241 * Fill in the new inode number and log the entry.
2242 */
2243 dep->inumber = cpu_to_be64(inum);
2244 args->dp->d_ops->data_put_ftype(dep, ftype);
2245 xfs_dir2_data_log_entry(args, state->extrablk.bp, dep);
2246 rval = 0;
2247 }
2248 /*
2249 * Didn't find it, and we're holding a data block. Drop it.
2250 */
2251 else if (state->extravalid) {
2252 xfs_trans_brelse(args->trans, state->extrablk.bp);
2253 state->extrablk.bp = NULL;
2254 }
2255 /*
2256 * Release all the buffers in the cursor.
2257 */
2258 for (i = 0; i < state->path.active; i++) {
2259 xfs_trans_brelse(args->trans, state->path.blk[i].bp);
2260 state->path.blk[i].bp = NULL;
2261 }
2262 xfs_da_state_free(state);
2263 return rval;
2264 }
2265
2266 /*
2267 * Trim off a trailing empty freespace block.
2268 * Return (in rvalp) 1 if we did it, 0 if not.
2269 */
2270 int /* error */
2271 xfs_dir2_node_trim_free(
2272 xfs_da_args_t *args, /* operation arguments */
2273 xfs_fileoff_t fo, /* free block number */
2274 int *rvalp) /* out: did something */
2275 {
2276 struct xfs_buf *bp; /* freespace buffer */
2277 xfs_inode_t *dp; /* incore directory inode */
2278 int error; /* error return code */
2279 xfs_dir2_free_t *free; /* freespace structure */
2280 xfs_trans_t *tp; /* transaction pointer */
2281 struct xfs_dir3_icfree_hdr freehdr;
2282
2283 dp = args->dp;
2284 tp = args->trans;
2285
2286 *rvalp = 0;
2287
2288 /*
2289 * Read the freespace block.
2290 */
2291 error = xfs_dir2_free_try_read(tp, dp, fo, &bp);
2292 if (error)
2293 return error;
2294 /*
2295 * There can be holes in freespace. If fo is a hole, there's
2296 * nothing to do.
2297 */
2298 if (!bp)
2299 return 0;
2300 free = bp->b_addr;
2301 dp->d_ops->free_hdr_from_disk(&freehdr, free);
2302
2303 /*
2304 * If there are used entries, there's nothing to do.
2305 */
2306 if (freehdr.nused > 0) {
2307 xfs_trans_brelse(tp, bp);
2308 return 0;
2309 }
2310 /*
2311 * Blow the block away.
2312 */
2313 error = xfs_dir2_shrink_inode(args,
2314 xfs_dir2_da_to_db(args->geo, (xfs_dablk_t)fo), bp);
2315 if (error) {
2316 /*
2317 * Can't fail with ENOSPC since that only happens with no
2318 * space reservation, when breaking up an extent into two
2319 * pieces. This is the last block of an extent.
2320 */
2321 ASSERT(error != -ENOSPC);
2322 xfs_trans_brelse(tp, bp);
2323 return error;
2324 }
2325 /*
2326 * Return that we succeeded.
2327 */
2328 *rvalp = 1;
2329 return 0;
2330 }
ubuntu-zesty-kernel mirror