Merge remote-tracking branch 'pfdo/drm-fixes' into drm-next
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / fs / xfs / xfs_btree.c
1 /*
2 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
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.
8 *
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.
13 *
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
17 */
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_types.h"
21 #include "xfs_bit.h"
22 #include "xfs_log.h"
23 #include "xfs_trans.h"
24 #include "xfs_sb.h"
25 #include "xfs_ag.h"
26 #include "xfs_mount.h"
27 #include "xfs_bmap_btree.h"
28 #include "xfs_alloc_btree.h"
29 #include "xfs_ialloc_btree.h"
30 #include "xfs_dinode.h"
31 #include "xfs_inode.h"
32 #include "xfs_inode_item.h"
33 #include "xfs_buf_item.h"
34 #include "xfs_btree.h"
35 #include "xfs_error.h"
36 #include "xfs_trace.h"
37 #include "xfs_cksum.h"
38
39 /*
40 * Cursor allocation zone.
41 */
42 kmem_zone_t *xfs_btree_cur_zone;
43
44 /*
45 * Btree magic numbers.
46 */
47 static const __uint32_t xfs_magics[2][XFS_BTNUM_MAX] = {
48 { XFS_ABTB_MAGIC, XFS_ABTC_MAGIC, XFS_BMAP_MAGIC, XFS_IBT_MAGIC },
49 { XFS_ABTB_CRC_MAGIC, XFS_ABTC_CRC_MAGIC,
50 XFS_BMAP_CRC_MAGIC, XFS_IBT_CRC_MAGIC }
51 };
52 #define xfs_btree_magic(cur) \
53 xfs_magics[!!((cur)->bc_flags & XFS_BTREE_CRC_BLOCKS)][cur->bc_btnum]
54
55
56 STATIC int /* error (0 or EFSCORRUPTED) */
57 xfs_btree_check_lblock(
58 struct xfs_btree_cur *cur, /* btree cursor */
59 struct xfs_btree_block *block, /* btree long form block pointer */
60 int level, /* level of the btree block */
61 struct xfs_buf *bp) /* buffer for block, if any */
62 {
63 int lblock_ok = 1; /* block passes checks */
64 struct xfs_mount *mp; /* file system mount point */
65
66 mp = cur->bc_mp;
67
68 if (xfs_sb_version_hascrc(&mp->m_sb)) {
69 lblock_ok = lblock_ok &&
70 uuid_equal(&block->bb_u.l.bb_uuid, &mp->m_sb.sb_uuid) &&
71 block->bb_u.l.bb_blkno == cpu_to_be64(
72 bp ? bp->b_bn : XFS_BUF_DADDR_NULL);
73 }
74
75 lblock_ok = lblock_ok &&
76 be32_to_cpu(block->bb_magic) == xfs_btree_magic(cur) &&
77 be16_to_cpu(block->bb_level) == level &&
78 be16_to_cpu(block->bb_numrecs) <=
79 cur->bc_ops->get_maxrecs(cur, level) &&
80 block->bb_u.l.bb_leftsib &&
81 (block->bb_u.l.bb_leftsib == cpu_to_be64(NULLDFSBNO) ||
82 XFS_FSB_SANITY_CHECK(mp,
83 be64_to_cpu(block->bb_u.l.bb_leftsib))) &&
84 block->bb_u.l.bb_rightsib &&
85 (block->bb_u.l.bb_rightsib == cpu_to_be64(NULLDFSBNO) ||
86 XFS_FSB_SANITY_CHECK(mp,
87 be64_to_cpu(block->bb_u.l.bb_rightsib)));
88
89 if (unlikely(XFS_TEST_ERROR(!lblock_ok, mp,
90 XFS_ERRTAG_BTREE_CHECK_LBLOCK,
91 XFS_RANDOM_BTREE_CHECK_LBLOCK))) {
92 if (bp)
93 trace_xfs_btree_corrupt(bp, _RET_IP_);
94 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp);
95 return XFS_ERROR(EFSCORRUPTED);
96 }
97 return 0;
98 }
99
100 STATIC int /* error (0 or EFSCORRUPTED) */
101 xfs_btree_check_sblock(
102 struct xfs_btree_cur *cur, /* btree cursor */
103 struct xfs_btree_block *block, /* btree short form block pointer */
104 int level, /* level of the btree block */
105 struct xfs_buf *bp) /* buffer containing block */
106 {
107 struct xfs_mount *mp; /* file system mount point */
108 struct xfs_buf *agbp; /* buffer for ag. freespace struct */
109 struct xfs_agf *agf; /* ag. freespace structure */
110 xfs_agblock_t agflen; /* native ag. freespace length */
111 int sblock_ok = 1; /* block passes checks */
112
113 mp = cur->bc_mp;
114 agbp = cur->bc_private.a.agbp;
115 agf = XFS_BUF_TO_AGF(agbp);
116 agflen = be32_to_cpu(agf->agf_length);
117
118 if (xfs_sb_version_hascrc(&mp->m_sb)) {
119 sblock_ok = sblock_ok &&
120 uuid_equal(&block->bb_u.s.bb_uuid, &mp->m_sb.sb_uuid) &&
121 block->bb_u.s.bb_blkno == cpu_to_be64(
122 bp ? bp->b_bn : XFS_BUF_DADDR_NULL);
123 }
124
125 sblock_ok = sblock_ok &&
126 be32_to_cpu(block->bb_magic) == xfs_btree_magic(cur) &&
127 be16_to_cpu(block->bb_level) == level &&
128 be16_to_cpu(block->bb_numrecs) <=
129 cur->bc_ops->get_maxrecs(cur, level) &&
130 (block->bb_u.s.bb_leftsib == cpu_to_be32(NULLAGBLOCK) ||
131 be32_to_cpu(block->bb_u.s.bb_leftsib) < agflen) &&
132 block->bb_u.s.bb_leftsib &&
133 (block->bb_u.s.bb_rightsib == cpu_to_be32(NULLAGBLOCK) ||
134 be32_to_cpu(block->bb_u.s.bb_rightsib) < agflen) &&
135 block->bb_u.s.bb_rightsib;
136
137 if (unlikely(XFS_TEST_ERROR(!sblock_ok, mp,
138 XFS_ERRTAG_BTREE_CHECK_SBLOCK,
139 XFS_RANDOM_BTREE_CHECK_SBLOCK))) {
140 if (bp)
141 trace_xfs_btree_corrupt(bp, _RET_IP_);
142 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp);
143 return XFS_ERROR(EFSCORRUPTED);
144 }
145 return 0;
146 }
147
148 /*
149 * Debug routine: check that block header is ok.
150 */
151 int
152 xfs_btree_check_block(
153 struct xfs_btree_cur *cur, /* btree cursor */
154 struct xfs_btree_block *block, /* generic btree block pointer */
155 int level, /* level of the btree block */
156 struct xfs_buf *bp) /* buffer containing block, if any */
157 {
158 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
159 return xfs_btree_check_lblock(cur, block, level, bp);
160 else
161 return xfs_btree_check_sblock(cur, block, level, bp);
162 }
163
164 /*
165 * Check that (long) pointer is ok.
166 */
167 int /* error (0 or EFSCORRUPTED) */
168 xfs_btree_check_lptr(
169 struct xfs_btree_cur *cur, /* btree cursor */
170 xfs_dfsbno_t bno, /* btree block disk address */
171 int level) /* btree block level */
172 {
173 XFS_WANT_CORRUPTED_RETURN(
174 level > 0 &&
175 bno != NULLDFSBNO &&
176 XFS_FSB_SANITY_CHECK(cur->bc_mp, bno));
177 return 0;
178 }
179
180 #ifdef DEBUG
181 /*
182 * Check that (short) pointer is ok.
183 */
184 STATIC int /* error (0 or EFSCORRUPTED) */
185 xfs_btree_check_sptr(
186 struct xfs_btree_cur *cur, /* btree cursor */
187 xfs_agblock_t bno, /* btree block disk address */
188 int level) /* btree block level */
189 {
190 xfs_agblock_t agblocks = cur->bc_mp->m_sb.sb_agblocks;
191
192 XFS_WANT_CORRUPTED_RETURN(
193 level > 0 &&
194 bno != NULLAGBLOCK &&
195 bno != 0 &&
196 bno < agblocks);
197 return 0;
198 }
199
200 /*
201 * Check that block ptr is ok.
202 */
203 STATIC int /* error (0 or EFSCORRUPTED) */
204 xfs_btree_check_ptr(
205 struct xfs_btree_cur *cur, /* btree cursor */
206 union xfs_btree_ptr *ptr, /* btree block disk address */
207 int index, /* offset from ptr to check */
208 int level) /* btree block level */
209 {
210 if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
211 return xfs_btree_check_lptr(cur,
212 be64_to_cpu((&ptr->l)[index]), level);
213 } else {
214 return xfs_btree_check_sptr(cur,
215 be32_to_cpu((&ptr->s)[index]), level);
216 }
217 }
218 #endif
219
220 /*
221 * Calculate CRC on the whole btree block and stuff it into the
222 * long-form btree header.
223 *
224 * Prior to calculting the CRC, pull the LSN out of the buffer log item and put
225 * it into the buffer so recovery knows what the last modifcation was that made
226 * it to disk.
227 */
228 void
229 xfs_btree_lblock_calc_crc(
230 struct xfs_buf *bp)
231 {
232 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
233 struct xfs_buf_log_item *bip = bp->b_fspriv;
234
235 if (!xfs_sb_version_hascrc(&bp->b_target->bt_mount->m_sb))
236 return;
237 if (bip)
238 block->bb_u.l.bb_lsn = cpu_to_be64(bip->bli_item.li_lsn);
239 xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length),
240 XFS_BTREE_LBLOCK_CRC_OFF);
241 }
242
243 bool
244 xfs_btree_lblock_verify_crc(
245 struct xfs_buf *bp)
246 {
247 if (xfs_sb_version_hascrc(&bp->b_target->bt_mount->m_sb))
248 return xfs_verify_cksum(bp->b_addr, BBTOB(bp->b_length),
249 XFS_BTREE_LBLOCK_CRC_OFF);
250 return true;
251 }
252
253 /*
254 * Calculate CRC on the whole btree block and stuff it into the
255 * short-form btree header.
256 *
257 * Prior to calculting the CRC, pull the LSN out of the buffer log item and put
258 * it into the buffer so recovery knows what the last modifcation was that made
259 * it to disk.
260 */
261 void
262 xfs_btree_sblock_calc_crc(
263 struct xfs_buf *bp)
264 {
265 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
266 struct xfs_buf_log_item *bip = bp->b_fspriv;
267
268 if (!xfs_sb_version_hascrc(&bp->b_target->bt_mount->m_sb))
269 return;
270 if (bip)
271 block->bb_u.s.bb_lsn = cpu_to_be64(bip->bli_item.li_lsn);
272 xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length),
273 XFS_BTREE_SBLOCK_CRC_OFF);
274 }
275
276 bool
277 xfs_btree_sblock_verify_crc(
278 struct xfs_buf *bp)
279 {
280 if (xfs_sb_version_hascrc(&bp->b_target->bt_mount->m_sb))
281 return xfs_verify_cksum(bp->b_addr, BBTOB(bp->b_length),
282 XFS_BTREE_SBLOCK_CRC_OFF);
283 return true;
284 }
285
286 /*
287 * Delete the btree cursor.
288 */
289 void
290 xfs_btree_del_cursor(
291 xfs_btree_cur_t *cur, /* btree cursor */
292 int error) /* del because of error */
293 {
294 int i; /* btree level */
295
296 /*
297 * Clear the buffer pointers, and release the buffers.
298 * If we're doing this in the face of an error, we
299 * need to make sure to inspect all of the entries
300 * in the bc_bufs array for buffers to be unlocked.
301 * This is because some of the btree code works from
302 * level n down to 0, and if we get an error along
303 * the way we won't have initialized all the entries
304 * down to 0.
305 */
306 for (i = 0; i < cur->bc_nlevels; i++) {
307 if (cur->bc_bufs[i])
308 xfs_trans_brelse(cur->bc_tp, cur->bc_bufs[i]);
309 else if (!error)
310 break;
311 }
312 /*
313 * Can't free a bmap cursor without having dealt with the
314 * allocated indirect blocks' accounting.
315 */
316 ASSERT(cur->bc_btnum != XFS_BTNUM_BMAP ||
317 cur->bc_private.b.allocated == 0);
318 /*
319 * Free the cursor.
320 */
321 kmem_zone_free(xfs_btree_cur_zone, cur);
322 }
323
324 /*
325 * Duplicate the btree cursor.
326 * Allocate a new one, copy the record, re-get the buffers.
327 */
328 int /* error */
329 xfs_btree_dup_cursor(
330 xfs_btree_cur_t *cur, /* input cursor */
331 xfs_btree_cur_t **ncur) /* output cursor */
332 {
333 xfs_buf_t *bp; /* btree block's buffer pointer */
334 int error; /* error return value */
335 int i; /* level number of btree block */
336 xfs_mount_t *mp; /* mount structure for filesystem */
337 xfs_btree_cur_t *new; /* new cursor value */
338 xfs_trans_t *tp; /* transaction pointer, can be NULL */
339
340 tp = cur->bc_tp;
341 mp = cur->bc_mp;
342
343 /*
344 * Allocate a new cursor like the old one.
345 */
346 new = cur->bc_ops->dup_cursor(cur);
347
348 /*
349 * Copy the record currently in the cursor.
350 */
351 new->bc_rec = cur->bc_rec;
352
353 /*
354 * For each level current, re-get the buffer and copy the ptr value.
355 */
356 for (i = 0; i < new->bc_nlevels; i++) {
357 new->bc_ptrs[i] = cur->bc_ptrs[i];
358 new->bc_ra[i] = cur->bc_ra[i];
359 bp = cur->bc_bufs[i];
360 if (bp) {
361 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
362 XFS_BUF_ADDR(bp), mp->m_bsize,
363 0, &bp,
364 cur->bc_ops->buf_ops);
365 if (error) {
366 xfs_btree_del_cursor(new, error);
367 *ncur = NULL;
368 return error;
369 }
370 }
371 new->bc_bufs[i] = bp;
372 }
373 *ncur = new;
374 return 0;
375 }
376
377 /*
378 * XFS btree block layout and addressing:
379 *
380 * There are two types of blocks in the btree: leaf and non-leaf blocks.
381 *
382 * The leaf record start with a header then followed by records containing
383 * the values. A non-leaf block also starts with the same header, and
384 * then first contains lookup keys followed by an equal number of pointers
385 * to the btree blocks at the previous level.
386 *
387 * +--------+-------+-------+-------+-------+-------+-------+
388 * Leaf: | header | rec 1 | rec 2 | rec 3 | rec 4 | rec 5 | rec N |
389 * +--------+-------+-------+-------+-------+-------+-------+
390 *
391 * +--------+-------+-------+-------+-------+-------+-------+
392 * Non-Leaf: | header | key 1 | key 2 | key N | ptr 1 | ptr 2 | ptr N |
393 * +--------+-------+-------+-------+-------+-------+-------+
394 *
395 * The header is called struct xfs_btree_block for reasons better left unknown
396 * and comes in different versions for short (32bit) and long (64bit) block
397 * pointers. The record and key structures are defined by the btree instances
398 * and opaque to the btree core. The block pointers are simple disk endian
399 * integers, available in a short (32bit) and long (64bit) variant.
400 *
401 * The helpers below calculate the offset of a given record, key or pointer
402 * into a btree block (xfs_btree_*_offset) or return a pointer to the given
403 * record, key or pointer (xfs_btree_*_addr). Note that all addressing
404 * inside the btree block is done using indices starting at one, not zero!
405 */
406
407 /*
408 * Return size of the btree block header for this btree instance.
409 */
410 static inline size_t xfs_btree_block_len(struct xfs_btree_cur *cur)
411 {
412 if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
413 if (cur->bc_flags & XFS_BTREE_CRC_BLOCKS)
414 return XFS_BTREE_LBLOCK_CRC_LEN;
415 return XFS_BTREE_LBLOCK_LEN;
416 }
417 if (cur->bc_flags & XFS_BTREE_CRC_BLOCKS)
418 return XFS_BTREE_SBLOCK_CRC_LEN;
419 return XFS_BTREE_SBLOCK_LEN;
420 }
421
422 /*
423 * Return size of btree block pointers for this btree instance.
424 */
425 static inline size_t xfs_btree_ptr_len(struct xfs_btree_cur *cur)
426 {
427 return (cur->bc_flags & XFS_BTREE_LONG_PTRS) ?
428 sizeof(__be64) : sizeof(__be32);
429 }
430
431 /*
432 * Calculate offset of the n-th record in a btree block.
433 */
434 STATIC size_t
435 xfs_btree_rec_offset(
436 struct xfs_btree_cur *cur,
437 int n)
438 {
439 return xfs_btree_block_len(cur) +
440 (n - 1) * cur->bc_ops->rec_len;
441 }
442
443 /*
444 * Calculate offset of the n-th key in a btree block.
445 */
446 STATIC size_t
447 xfs_btree_key_offset(
448 struct xfs_btree_cur *cur,
449 int n)
450 {
451 return xfs_btree_block_len(cur) +
452 (n - 1) * cur->bc_ops->key_len;
453 }
454
455 /*
456 * Calculate offset of the n-th block pointer in a btree block.
457 */
458 STATIC size_t
459 xfs_btree_ptr_offset(
460 struct xfs_btree_cur *cur,
461 int n,
462 int level)
463 {
464 return xfs_btree_block_len(cur) +
465 cur->bc_ops->get_maxrecs(cur, level) * cur->bc_ops->key_len +
466 (n - 1) * xfs_btree_ptr_len(cur);
467 }
468
469 /*
470 * Return a pointer to the n-th record in the btree block.
471 */
472 STATIC union xfs_btree_rec *
473 xfs_btree_rec_addr(
474 struct xfs_btree_cur *cur,
475 int n,
476 struct xfs_btree_block *block)
477 {
478 return (union xfs_btree_rec *)
479 ((char *)block + xfs_btree_rec_offset(cur, n));
480 }
481
482 /*
483 * Return a pointer to the n-th key in the btree block.
484 */
485 STATIC union xfs_btree_key *
486 xfs_btree_key_addr(
487 struct xfs_btree_cur *cur,
488 int n,
489 struct xfs_btree_block *block)
490 {
491 return (union xfs_btree_key *)
492 ((char *)block + xfs_btree_key_offset(cur, n));
493 }
494
495 /*
496 * Return a pointer to the n-th block pointer in the btree block.
497 */
498 STATIC union xfs_btree_ptr *
499 xfs_btree_ptr_addr(
500 struct xfs_btree_cur *cur,
501 int n,
502 struct xfs_btree_block *block)
503 {
504 int level = xfs_btree_get_level(block);
505
506 ASSERT(block->bb_level != 0);
507
508 return (union xfs_btree_ptr *)
509 ((char *)block + xfs_btree_ptr_offset(cur, n, level));
510 }
511
512 /*
513 * Get a the root block which is stored in the inode.
514 *
515 * For now this btree implementation assumes the btree root is always
516 * stored in the if_broot field of an inode fork.
517 */
518 STATIC struct xfs_btree_block *
519 xfs_btree_get_iroot(
520 struct xfs_btree_cur *cur)
521 {
522 struct xfs_ifork *ifp;
523
524 ifp = XFS_IFORK_PTR(cur->bc_private.b.ip, cur->bc_private.b.whichfork);
525 return (struct xfs_btree_block *)ifp->if_broot;
526 }
527
528 /*
529 * Retrieve the block pointer from the cursor at the given level.
530 * This may be an inode btree root or from a buffer.
531 */
532 STATIC struct xfs_btree_block * /* generic btree block pointer */
533 xfs_btree_get_block(
534 struct xfs_btree_cur *cur, /* btree cursor */
535 int level, /* level in btree */
536 struct xfs_buf **bpp) /* buffer containing the block */
537 {
538 if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
539 (level == cur->bc_nlevels - 1)) {
540 *bpp = NULL;
541 return xfs_btree_get_iroot(cur);
542 }
543
544 *bpp = cur->bc_bufs[level];
545 return XFS_BUF_TO_BLOCK(*bpp);
546 }
547
548 /*
549 * Get a buffer for the block, return it with no data read.
550 * Long-form addressing.
551 */
552 xfs_buf_t * /* buffer for fsbno */
553 xfs_btree_get_bufl(
554 xfs_mount_t *mp, /* file system mount point */
555 xfs_trans_t *tp, /* transaction pointer */
556 xfs_fsblock_t fsbno, /* file system block number */
557 uint lock) /* lock flags for get_buf */
558 {
559 xfs_buf_t *bp; /* buffer pointer (return value) */
560 xfs_daddr_t d; /* real disk block address */
561
562 ASSERT(fsbno != NULLFSBLOCK);
563 d = XFS_FSB_TO_DADDR(mp, fsbno);
564 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d, mp->m_bsize, lock);
565 ASSERT(!xfs_buf_geterror(bp));
566 return bp;
567 }
568
569 /*
570 * Get a buffer for the block, return it with no data read.
571 * Short-form addressing.
572 */
573 xfs_buf_t * /* buffer for agno/agbno */
574 xfs_btree_get_bufs(
575 xfs_mount_t *mp, /* file system mount point */
576 xfs_trans_t *tp, /* transaction pointer */
577 xfs_agnumber_t agno, /* allocation group number */
578 xfs_agblock_t agbno, /* allocation group block number */
579 uint lock) /* lock flags for get_buf */
580 {
581 xfs_buf_t *bp; /* buffer pointer (return value) */
582 xfs_daddr_t d; /* real disk block address */
583
584 ASSERT(agno != NULLAGNUMBER);
585 ASSERT(agbno != NULLAGBLOCK);
586 d = XFS_AGB_TO_DADDR(mp, agno, agbno);
587 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d, mp->m_bsize, lock);
588 ASSERT(!xfs_buf_geterror(bp));
589 return bp;
590 }
591
592 /*
593 * Check for the cursor referring to the last block at the given level.
594 */
595 int /* 1=is last block, 0=not last block */
596 xfs_btree_islastblock(
597 xfs_btree_cur_t *cur, /* btree cursor */
598 int level) /* level to check */
599 {
600 struct xfs_btree_block *block; /* generic btree block pointer */
601 xfs_buf_t *bp; /* buffer containing block */
602
603 block = xfs_btree_get_block(cur, level, &bp);
604 xfs_btree_check_block(cur, block, level, bp);
605 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
606 return block->bb_u.l.bb_rightsib == cpu_to_be64(NULLDFSBNO);
607 else
608 return block->bb_u.s.bb_rightsib == cpu_to_be32(NULLAGBLOCK);
609 }
610
611 /*
612 * Change the cursor to point to the first record at the given level.
613 * Other levels are unaffected.
614 */
615 STATIC int /* success=1, failure=0 */
616 xfs_btree_firstrec(
617 xfs_btree_cur_t *cur, /* btree cursor */
618 int level) /* level to change */
619 {
620 struct xfs_btree_block *block; /* generic btree block pointer */
621 xfs_buf_t *bp; /* buffer containing block */
622
623 /*
624 * Get the block pointer for this level.
625 */
626 block = xfs_btree_get_block(cur, level, &bp);
627 xfs_btree_check_block(cur, block, level, bp);
628 /*
629 * It's empty, there is no such record.
630 */
631 if (!block->bb_numrecs)
632 return 0;
633 /*
634 * Set the ptr value to 1, that's the first record/key.
635 */
636 cur->bc_ptrs[level] = 1;
637 return 1;
638 }
639
640 /*
641 * Change the cursor to point to the last record in the current block
642 * at the given level. Other levels are unaffected.
643 */
644 STATIC int /* success=1, failure=0 */
645 xfs_btree_lastrec(
646 xfs_btree_cur_t *cur, /* btree cursor */
647 int level) /* level to change */
648 {
649 struct xfs_btree_block *block; /* generic btree block pointer */
650 xfs_buf_t *bp; /* buffer containing block */
651
652 /*
653 * Get the block pointer for this level.
654 */
655 block = xfs_btree_get_block(cur, level, &bp);
656 xfs_btree_check_block(cur, block, level, bp);
657 /*
658 * It's empty, there is no such record.
659 */
660 if (!block->bb_numrecs)
661 return 0;
662 /*
663 * Set the ptr value to numrecs, that's the last record/key.
664 */
665 cur->bc_ptrs[level] = be16_to_cpu(block->bb_numrecs);
666 return 1;
667 }
668
669 /*
670 * Compute first and last byte offsets for the fields given.
671 * Interprets the offsets table, which contains struct field offsets.
672 */
673 void
674 xfs_btree_offsets(
675 __int64_t fields, /* bitmask of fields */
676 const short *offsets, /* table of field offsets */
677 int nbits, /* number of bits to inspect */
678 int *first, /* output: first byte offset */
679 int *last) /* output: last byte offset */
680 {
681 int i; /* current bit number */
682 __int64_t imask; /* mask for current bit number */
683
684 ASSERT(fields != 0);
685 /*
686 * Find the lowest bit, so the first byte offset.
687 */
688 for (i = 0, imask = 1LL; ; i++, imask <<= 1) {
689 if (imask & fields) {
690 *first = offsets[i];
691 break;
692 }
693 }
694 /*
695 * Find the highest bit, so the last byte offset.
696 */
697 for (i = nbits - 1, imask = 1LL << i; ; i--, imask >>= 1) {
698 if (imask & fields) {
699 *last = offsets[i + 1] - 1;
700 break;
701 }
702 }
703 }
704
705 /*
706 * Get a buffer for the block, return it read in.
707 * Long-form addressing.
708 */
709 int
710 xfs_btree_read_bufl(
711 struct xfs_mount *mp, /* file system mount point */
712 struct xfs_trans *tp, /* transaction pointer */
713 xfs_fsblock_t fsbno, /* file system block number */
714 uint lock, /* lock flags for read_buf */
715 struct xfs_buf **bpp, /* buffer for fsbno */
716 int refval, /* ref count value for buffer */
717 const struct xfs_buf_ops *ops)
718 {
719 struct xfs_buf *bp; /* return value */
720 xfs_daddr_t d; /* real disk block address */
721 int error;
722
723 ASSERT(fsbno != NULLFSBLOCK);
724 d = XFS_FSB_TO_DADDR(mp, fsbno);
725 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, d,
726 mp->m_bsize, lock, &bp, ops);
727 if (error)
728 return error;
729 ASSERT(!xfs_buf_geterror(bp));
730 if (bp)
731 xfs_buf_set_ref(bp, refval);
732 *bpp = bp;
733 return 0;
734 }
735
736 /*
737 * Read-ahead the block, don't wait for it, don't return a buffer.
738 * Long-form addressing.
739 */
740 /* ARGSUSED */
741 void
742 xfs_btree_reada_bufl(
743 struct xfs_mount *mp, /* file system mount point */
744 xfs_fsblock_t fsbno, /* file system block number */
745 xfs_extlen_t count, /* count of filesystem blocks */
746 const struct xfs_buf_ops *ops)
747 {
748 xfs_daddr_t d;
749
750 ASSERT(fsbno != NULLFSBLOCK);
751 d = XFS_FSB_TO_DADDR(mp, fsbno);
752 xfs_buf_readahead(mp->m_ddev_targp, d, mp->m_bsize * count, ops);
753 }
754
755 /*
756 * Read-ahead the block, don't wait for it, don't return a buffer.
757 * Short-form addressing.
758 */
759 /* ARGSUSED */
760 void
761 xfs_btree_reada_bufs(
762 struct xfs_mount *mp, /* file system mount point */
763 xfs_agnumber_t agno, /* allocation group number */
764 xfs_agblock_t agbno, /* allocation group block number */
765 xfs_extlen_t count, /* count of filesystem blocks */
766 const struct xfs_buf_ops *ops)
767 {
768 xfs_daddr_t d;
769
770 ASSERT(agno != NULLAGNUMBER);
771 ASSERT(agbno != NULLAGBLOCK);
772 d = XFS_AGB_TO_DADDR(mp, agno, agbno);
773 xfs_buf_readahead(mp->m_ddev_targp, d, mp->m_bsize * count, ops);
774 }
775
776 STATIC int
777 xfs_btree_readahead_lblock(
778 struct xfs_btree_cur *cur,
779 int lr,
780 struct xfs_btree_block *block)
781 {
782 int rval = 0;
783 xfs_dfsbno_t left = be64_to_cpu(block->bb_u.l.bb_leftsib);
784 xfs_dfsbno_t right = be64_to_cpu(block->bb_u.l.bb_rightsib);
785
786 if ((lr & XFS_BTCUR_LEFTRA) && left != NULLDFSBNO) {
787 xfs_btree_reada_bufl(cur->bc_mp, left, 1,
788 cur->bc_ops->buf_ops);
789 rval++;
790 }
791
792 if ((lr & XFS_BTCUR_RIGHTRA) && right != NULLDFSBNO) {
793 xfs_btree_reada_bufl(cur->bc_mp, right, 1,
794 cur->bc_ops->buf_ops);
795 rval++;
796 }
797
798 return rval;
799 }
800
801 STATIC int
802 xfs_btree_readahead_sblock(
803 struct xfs_btree_cur *cur,
804 int lr,
805 struct xfs_btree_block *block)
806 {
807 int rval = 0;
808 xfs_agblock_t left = be32_to_cpu(block->bb_u.s.bb_leftsib);
809 xfs_agblock_t right = be32_to_cpu(block->bb_u.s.bb_rightsib);
810
811
812 if ((lr & XFS_BTCUR_LEFTRA) && left != NULLAGBLOCK) {
813 xfs_btree_reada_bufs(cur->bc_mp, cur->bc_private.a.agno,
814 left, 1, cur->bc_ops->buf_ops);
815 rval++;
816 }
817
818 if ((lr & XFS_BTCUR_RIGHTRA) && right != NULLAGBLOCK) {
819 xfs_btree_reada_bufs(cur->bc_mp, cur->bc_private.a.agno,
820 right, 1, cur->bc_ops->buf_ops);
821 rval++;
822 }
823
824 return rval;
825 }
826
827 /*
828 * Read-ahead btree blocks, at the given level.
829 * Bits in lr are set from XFS_BTCUR_{LEFT,RIGHT}RA.
830 */
831 STATIC int
832 xfs_btree_readahead(
833 struct xfs_btree_cur *cur, /* btree cursor */
834 int lev, /* level in btree */
835 int lr) /* left/right bits */
836 {
837 struct xfs_btree_block *block;
838
839 /*
840 * No readahead needed if we are at the root level and the
841 * btree root is stored in the inode.
842 */
843 if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
844 (lev == cur->bc_nlevels - 1))
845 return 0;
846
847 if ((cur->bc_ra[lev] | lr) == cur->bc_ra[lev])
848 return 0;
849
850 cur->bc_ra[lev] |= lr;
851 block = XFS_BUF_TO_BLOCK(cur->bc_bufs[lev]);
852
853 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
854 return xfs_btree_readahead_lblock(cur, lr, block);
855 return xfs_btree_readahead_sblock(cur, lr, block);
856 }
857
858 /*
859 * Set the buffer for level "lev" in the cursor to bp, releasing
860 * any previous buffer.
861 */
862 STATIC void
863 xfs_btree_setbuf(
864 xfs_btree_cur_t *cur, /* btree cursor */
865 int lev, /* level in btree */
866 xfs_buf_t *bp) /* new buffer to set */
867 {
868 struct xfs_btree_block *b; /* btree block */
869
870 if (cur->bc_bufs[lev])
871 xfs_trans_brelse(cur->bc_tp, cur->bc_bufs[lev]);
872 cur->bc_bufs[lev] = bp;
873 cur->bc_ra[lev] = 0;
874
875 b = XFS_BUF_TO_BLOCK(bp);
876 if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
877 if (b->bb_u.l.bb_leftsib == cpu_to_be64(NULLDFSBNO))
878 cur->bc_ra[lev] |= XFS_BTCUR_LEFTRA;
879 if (b->bb_u.l.bb_rightsib == cpu_to_be64(NULLDFSBNO))
880 cur->bc_ra[lev] |= XFS_BTCUR_RIGHTRA;
881 } else {
882 if (b->bb_u.s.bb_leftsib == cpu_to_be32(NULLAGBLOCK))
883 cur->bc_ra[lev] |= XFS_BTCUR_LEFTRA;
884 if (b->bb_u.s.bb_rightsib == cpu_to_be32(NULLAGBLOCK))
885 cur->bc_ra[lev] |= XFS_BTCUR_RIGHTRA;
886 }
887 }
888
889 STATIC int
890 xfs_btree_ptr_is_null(
891 struct xfs_btree_cur *cur,
892 union xfs_btree_ptr *ptr)
893 {
894 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
895 return ptr->l == cpu_to_be64(NULLDFSBNO);
896 else
897 return ptr->s == cpu_to_be32(NULLAGBLOCK);
898 }
899
900 STATIC void
901 xfs_btree_set_ptr_null(
902 struct xfs_btree_cur *cur,
903 union xfs_btree_ptr *ptr)
904 {
905 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
906 ptr->l = cpu_to_be64(NULLDFSBNO);
907 else
908 ptr->s = cpu_to_be32(NULLAGBLOCK);
909 }
910
911 /*
912 * Get/set/init sibling pointers
913 */
914 STATIC void
915 xfs_btree_get_sibling(
916 struct xfs_btree_cur *cur,
917 struct xfs_btree_block *block,
918 union xfs_btree_ptr *ptr,
919 int lr)
920 {
921 ASSERT(lr == XFS_BB_LEFTSIB || lr == XFS_BB_RIGHTSIB);
922
923 if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
924 if (lr == XFS_BB_RIGHTSIB)
925 ptr->l = block->bb_u.l.bb_rightsib;
926 else
927 ptr->l = block->bb_u.l.bb_leftsib;
928 } else {
929 if (lr == XFS_BB_RIGHTSIB)
930 ptr->s = block->bb_u.s.bb_rightsib;
931 else
932 ptr->s = block->bb_u.s.bb_leftsib;
933 }
934 }
935
936 STATIC void
937 xfs_btree_set_sibling(
938 struct xfs_btree_cur *cur,
939 struct xfs_btree_block *block,
940 union xfs_btree_ptr *ptr,
941 int lr)
942 {
943 ASSERT(lr == XFS_BB_LEFTSIB || lr == XFS_BB_RIGHTSIB);
944
945 if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
946 if (lr == XFS_BB_RIGHTSIB)
947 block->bb_u.l.bb_rightsib = ptr->l;
948 else
949 block->bb_u.l.bb_leftsib = ptr->l;
950 } else {
951 if (lr == XFS_BB_RIGHTSIB)
952 block->bb_u.s.bb_rightsib = ptr->s;
953 else
954 block->bb_u.s.bb_leftsib = ptr->s;
955 }
956 }
957
958 void
959 xfs_btree_init_block_int(
960 struct xfs_mount *mp,
961 struct xfs_btree_block *buf,
962 xfs_daddr_t blkno,
963 __u32 magic,
964 __u16 level,
965 __u16 numrecs,
966 __u64 owner,
967 unsigned int flags)
968 {
969 buf->bb_magic = cpu_to_be32(magic);
970 buf->bb_level = cpu_to_be16(level);
971 buf->bb_numrecs = cpu_to_be16(numrecs);
972
973 if (flags & XFS_BTREE_LONG_PTRS) {
974 buf->bb_u.l.bb_leftsib = cpu_to_be64(NULLDFSBNO);
975 buf->bb_u.l.bb_rightsib = cpu_to_be64(NULLDFSBNO);
976 if (flags & XFS_BTREE_CRC_BLOCKS) {
977 buf->bb_u.l.bb_blkno = cpu_to_be64(blkno);
978 buf->bb_u.l.bb_owner = cpu_to_be64(owner);
979 uuid_copy(&buf->bb_u.l.bb_uuid, &mp->m_sb.sb_uuid);
980 buf->bb_u.l.bb_pad = 0;
981 }
982 } else {
983 /* owner is a 32 bit value on short blocks */
984 __u32 __owner = (__u32)owner;
985
986 buf->bb_u.s.bb_leftsib = cpu_to_be32(NULLAGBLOCK);
987 buf->bb_u.s.bb_rightsib = cpu_to_be32(NULLAGBLOCK);
988 if (flags & XFS_BTREE_CRC_BLOCKS) {
989 buf->bb_u.s.bb_blkno = cpu_to_be64(blkno);
990 buf->bb_u.s.bb_owner = cpu_to_be32(__owner);
991 uuid_copy(&buf->bb_u.s.bb_uuid, &mp->m_sb.sb_uuid);
992 }
993 }
994 }
995
996 void
997 xfs_btree_init_block(
998 struct xfs_mount *mp,
999 struct xfs_buf *bp,
1000 __u32 magic,
1001 __u16 level,
1002 __u16 numrecs,
1003 __u64 owner,
1004 unsigned int flags)
1005 {
1006 xfs_btree_init_block_int(mp, XFS_BUF_TO_BLOCK(bp), bp->b_bn,
1007 magic, level, numrecs, owner, flags);
1008 }
1009
1010 STATIC void
1011 xfs_btree_init_block_cur(
1012 struct xfs_btree_cur *cur,
1013 struct xfs_buf *bp,
1014 int level,
1015 int numrecs)
1016 {
1017 __u64 owner;
1018
1019 /*
1020 * we can pull the owner from the cursor right now as the different
1021 * owners align directly with the pointer size of the btree. This may
1022 * change in future, but is safe for current users of the generic btree
1023 * code.
1024 */
1025 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
1026 owner = cur->bc_private.b.ip->i_ino;
1027 else
1028 owner = cur->bc_private.a.agno;
1029
1030 xfs_btree_init_block_int(cur->bc_mp, XFS_BUF_TO_BLOCK(bp), bp->b_bn,
1031 xfs_btree_magic(cur), level, numrecs,
1032 owner, cur->bc_flags);
1033 }
1034
1035 /*
1036 * Return true if ptr is the last record in the btree and
1037 * we need to track updates to this record. The decision
1038 * will be further refined in the update_lastrec method.
1039 */
1040 STATIC int
1041 xfs_btree_is_lastrec(
1042 struct xfs_btree_cur *cur,
1043 struct xfs_btree_block *block,
1044 int level)
1045 {
1046 union xfs_btree_ptr ptr;
1047
1048 if (level > 0)
1049 return 0;
1050 if (!(cur->bc_flags & XFS_BTREE_LASTREC_UPDATE))
1051 return 0;
1052
1053 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
1054 if (!xfs_btree_ptr_is_null(cur, &ptr))
1055 return 0;
1056 return 1;
1057 }
1058
1059 STATIC void
1060 xfs_btree_buf_to_ptr(
1061 struct xfs_btree_cur *cur,
1062 struct xfs_buf *bp,
1063 union xfs_btree_ptr *ptr)
1064 {
1065 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
1066 ptr->l = cpu_to_be64(XFS_DADDR_TO_FSB(cur->bc_mp,
1067 XFS_BUF_ADDR(bp)));
1068 else {
1069 ptr->s = cpu_to_be32(xfs_daddr_to_agbno(cur->bc_mp,
1070 XFS_BUF_ADDR(bp)));
1071 }
1072 }
1073
1074 STATIC xfs_daddr_t
1075 xfs_btree_ptr_to_daddr(
1076 struct xfs_btree_cur *cur,
1077 union xfs_btree_ptr *ptr)
1078 {
1079 if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
1080 ASSERT(ptr->l != cpu_to_be64(NULLDFSBNO));
1081
1082 return XFS_FSB_TO_DADDR(cur->bc_mp, be64_to_cpu(ptr->l));
1083 } else {
1084 ASSERT(cur->bc_private.a.agno != NULLAGNUMBER);
1085 ASSERT(ptr->s != cpu_to_be32(NULLAGBLOCK));
1086
1087 return XFS_AGB_TO_DADDR(cur->bc_mp, cur->bc_private.a.agno,
1088 be32_to_cpu(ptr->s));
1089 }
1090 }
1091
1092 STATIC void
1093 xfs_btree_set_refs(
1094 struct xfs_btree_cur *cur,
1095 struct xfs_buf *bp)
1096 {
1097 switch (cur->bc_btnum) {
1098 case XFS_BTNUM_BNO:
1099 case XFS_BTNUM_CNT:
1100 xfs_buf_set_ref(bp, XFS_ALLOC_BTREE_REF);
1101 break;
1102 case XFS_BTNUM_INO:
1103 xfs_buf_set_ref(bp, XFS_INO_BTREE_REF);
1104 break;
1105 case XFS_BTNUM_BMAP:
1106 xfs_buf_set_ref(bp, XFS_BMAP_BTREE_REF);
1107 break;
1108 default:
1109 ASSERT(0);
1110 }
1111 }
1112
1113 STATIC int
1114 xfs_btree_get_buf_block(
1115 struct xfs_btree_cur *cur,
1116 union xfs_btree_ptr *ptr,
1117 int flags,
1118 struct xfs_btree_block **block,
1119 struct xfs_buf **bpp)
1120 {
1121 struct xfs_mount *mp = cur->bc_mp;
1122 xfs_daddr_t d;
1123
1124 /* need to sort out how callers deal with failures first */
1125 ASSERT(!(flags & XBF_TRYLOCK));
1126
1127 d = xfs_btree_ptr_to_daddr(cur, ptr);
1128 *bpp = xfs_trans_get_buf(cur->bc_tp, mp->m_ddev_targp, d,
1129 mp->m_bsize, flags);
1130
1131 if (!*bpp)
1132 return ENOMEM;
1133
1134 (*bpp)->b_ops = cur->bc_ops->buf_ops;
1135 *block = XFS_BUF_TO_BLOCK(*bpp);
1136 return 0;
1137 }
1138
1139 /*
1140 * Read in the buffer at the given ptr and return the buffer and
1141 * the block pointer within the buffer.
1142 */
1143 STATIC int
1144 xfs_btree_read_buf_block(
1145 struct xfs_btree_cur *cur,
1146 union xfs_btree_ptr *ptr,
1147 int level,
1148 int flags,
1149 struct xfs_btree_block **block,
1150 struct xfs_buf **bpp)
1151 {
1152 struct xfs_mount *mp = cur->bc_mp;
1153 xfs_daddr_t d;
1154 int error;
1155
1156 /* need to sort out how callers deal with failures first */
1157 ASSERT(!(flags & XBF_TRYLOCK));
1158
1159 d = xfs_btree_ptr_to_daddr(cur, ptr);
1160 error = xfs_trans_read_buf(mp, cur->bc_tp, mp->m_ddev_targp, d,
1161 mp->m_bsize, flags, bpp,
1162 cur->bc_ops->buf_ops);
1163 if (error)
1164 return error;
1165
1166 ASSERT(!xfs_buf_geterror(*bpp));
1167 xfs_btree_set_refs(cur, *bpp);
1168 *block = XFS_BUF_TO_BLOCK(*bpp);
1169 return 0;
1170 }
1171
1172 /*
1173 * Copy keys from one btree block to another.
1174 */
1175 STATIC void
1176 xfs_btree_copy_keys(
1177 struct xfs_btree_cur *cur,
1178 union xfs_btree_key *dst_key,
1179 union xfs_btree_key *src_key,
1180 int numkeys)
1181 {
1182 ASSERT(numkeys >= 0);
1183 memcpy(dst_key, src_key, numkeys * cur->bc_ops->key_len);
1184 }
1185
1186 /*
1187 * Copy records from one btree block to another.
1188 */
1189 STATIC void
1190 xfs_btree_copy_recs(
1191 struct xfs_btree_cur *cur,
1192 union xfs_btree_rec *dst_rec,
1193 union xfs_btree_rec *src_rec,
1194 int numrecs)
1195 {
1196 ASSERT(numrecs >= 0);
1197 memcpy(dst_rec, src_rec, numrecs * cur->bc_ops->rec_len);
1198 }
1199
1200 /*
1201 * Copy block pointers from one btree block to another.
1202 */
1203 STATIC void
1204 xfs_btree_copy_ptrs(
1205 struct xfs_btree_cur *cur,
1206 union xfs_btree_ptr *dst_ptr,
1207 union xfs_btree_ptr *src_ptr,
1208 int numptrs)
1209 {
1210 ASSERT(numptrs >= 0);
1211 memcpy(dst_ptr, src_ptr, numptrs * xfs_btree_ptr_len(cur));
1212 }
1213
1214 /*
1215 * Shift keys one index left/right inside a single btree block.
1216 */
1217 STATIC void
1218 xfs_btree_shift_keys(
1219 struct xfs_btree_cur *cur,
1220 union xfs_btree_key *key,
1221 int dir,
1222 int numkeys)
1223 {
1224 char *dst_key;
1225
1226 ASSERT(numkeys >= 0);
1227 ASSERT(dir == 1 || dir == -1);
1228
1229 dst_key = (char *)key + (dir * cur->bc_ops->key_len);
1230 memmove(dst_key, key, numkeys * cur->bc_ops->key_len);
1231 }
1232
1233 /*
1234 * Shift records one index left/right inside a single btree block.
1235 */
1236 STATIC void
1237 xfs_btree_shift_recs(
1238 struct xfs_btree_cur *cur,
1239 union xfs_btree_rec *rec,
1240 int dir,
1241 int numrecs)
1242 {
1243 char *dst_rec;
1244
1245 ASSERT(numrecs >= 0);
1246 ASSERT(dir == 1 || dir == -1);
1247
1248 dst_rec = (char *)rec + (dir * cur->bc_ops->rec_len);
1249 memmove(dst_rec, rec, numrecs * cur->bc_ops->rec_len);
1250 }
1251
1252 /*
1253 * Shift block pointers one index left/right inside a single btree block.
1254 */
1255 STATIC void
1256 xfs_btree_shift_ptrs(
1257 struct xfs_btree_cur *cur,
1258 union xfs_btree_ptr *ptr,
1259 int dir,
1260 int numptrs)
1261 {
1262 char *dst_ptr;
1263
1264 ASSERT(numptrs >= 0);
1265 ASSERT(dir == 1 || dir == -1);
1266
1267 dst_ptr = (char *)ptr + (dir * xfs_btree_ptr_len(cur));
1268 memmove(dst_ptr, ptr, numptrs * xfs_btree_ptr_len(cur));
1269 }
1270
1271 /*
1272 * Log key values from the btree block.
1273 */
1274 STATIC void
1275 xfs_btree_log_keys(
1276 struct xfs_btree_cur *cur,
1277 struct xfs_buf *bp,
1278 int first,
1279 int last)
1280 {
1281 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1282 XFS_BTREE_TRACE_ARGBII(cur, bp, first, last);
1283
1284 if (bp) {
1285 xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
1286 xfs_trans_log_buf(cur->bc_tp, bp,
1287 xfs_btree_key_offset(cur, first),
1288 xfs_btree_key_offset(cur, last + 1) - 1);
1289 } else {
1290 xfs_trans_log_inode(cur->bc_tp, cur->bc_private.b.ip,
1291 xfs_ilog_fbroot(cur->bc_private.b.whichfork));
1292 }
1293
1294 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1295 }
1296
1297 /*
1298 * Log record values from the btree block.
1299 */
1300 void
1301 xfs_btree_log_recs(
1302 struct xfs_btree_cur *cur,
1303 struct xfs_buf *bp,
1304 int first,
1305 int last)
1306 {
1307 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1308 XFS_BTREE_TRACE_ARGBII(cur, bp, first, last);
1309
1310 xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
1311 xfs_trans_log_buf(cur->bc_tp, bp,
1312 xfs_btree_rec_offset(cur, first),
1313 xfs_btree_rec_offset(cur, last + 1) - 1);
1314
1315 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1316 }
1317
1318 /*
1319 * Log block pointer fields from a btree block (nonleaf).
1320 */
1321 STATIC void
1322 xfs_btree_log_ptrs(
1323 struct xfs_btree_cur *cur, /* btree cursor */
1324 struct xfs_buf *bp, /* buffer containing btree block */
1325 int first, /* index of first pointer to log */
1326 int last) /* index of last pointer to log */
1327 {
1328 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1329 XFS_BTREE_TRACE_ARGBII(cur, bp, first, last);
1330
1331 if (bp) {
1332 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
1333 int level = xfs_btree_get_level(block);
1334
1335 xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
1336 xfs_trans_log_buf(cur->bc_tp, bp,
1337 xfs_btree_ptr_offset(cur, first, level),
1338 xfs_btree_ptr_offset(cur, last + 1, level) - 1);
1339 } else {
1340 xfs_trans_log_inode(cur->bc_tp, cur->bc_private.b.ip,
1341 xfs_ilog_fbroot(cur->bc_private.b.whichfork));
1342 }
1343
1344 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1345 }
1346
1347 /*
1348 * Log fields from a btree block header.
1349 */
1350 void
1351 xfs_btree_log_block(
1352 struct xfs_btree_cur *cur, /* btree cursor */
1353 struct xfs_buf *bp, /* buffer containing btree block */
1354 int fields) /* mask of fields: XFS_BB_... */
1355 {
1356 int first; /* first byte offset logged */
1357 int last; /* last byte offset logged */
1358 static const short soffsets[] = { /* table of offsets (short) */
1359 offsetof(struct xfs_btree_block, bb_magic),
1360 offsetof(struct xfs_btree_block, bb_level),
1361 offsetof(struct xfs_btree_block, bb_numrecs),
1362 offsetof(struct xfs_btree_block, bb_u.s.bb_leftsib),
1363 offsetof(struct xfs_btree_block, bb_u.s.bb_rightsib),
1364 offsetof(struct xfs_btree_block, bb_u.s.bb_blkno),
1365 offsetof(struct xfs_btree_block, bb_u.s.bb_lsn),
1366 offsetof(struct xfs_btree_block, bb_u.s.bb_uuid),
1367 offsetof(struct xfs_btree_block, bb_u.s.bb_owner),
1368 offsetof(struct xfs_btree_block, bb_u.s.bb_crc),
1369 XFS_BTREE_SBLOCK_CRC_LEN
1370 };
1371 static const short loffsets[] = { /* table of offsets (long) */
1372 offsetof(struct xfs_btree_block, bb_magic),
1373 offsetof(struct xfs_btree_block, bb_level),
1374 offsetof(struct xfs_btree_block, bb_numrecs),
1375 offsetof(struct xfs_btree_block, bb_u.l.bb_leftsib),
1376 offsetof(struct xfs_btree_block, bb_u.l.bb_rightsib),
1377 offsetof(struct xfs_btree_block, bb_u.l.bb_blkno),
1378 offsetof(struct xfs_btree_block, bb_u.l.bb_lsn),
1379 offsetof(struct xfs_btree_block, bb_u.l.bb_uuid),
1380 offsetof(struct xfs_btree_block, bb_u.l.bb_owner),
1381 offsetof(struct xfs_btree_block, bb_u.l.bb_crc),
1382 offsetof(struct xfs_btree_block, bb_u.l.bb_pad),
1383 XFS_BTREE_LBLOCK_CRC_LEN
1384 };
1385
1386 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1387 XFS_BTREE_TRACE_ARGBI(cur, bp, fields);
1388
1389 if (bp) {
1390 int nbits;
1391
1392 if (cur->bc_flags & XFS_BTREE_CRC_BLOCKS) {
1393 /*
1394 * We don't log the CRC when updating a btree
1395 * block but instead recreate it during log
1396 * recovery. As the log buffers have checksums
1397 * of their own this is safe and avoids logging a crc
1398 * update in a lot of places.
1399 */
1400 if (fields == XFS_BB_ALL_BITS)
1401 fields = XFS_BB_ALL_BITS_CRC;
1402 nbits = XFS_BB_NUM_BITS_CRC;
1403 } else {
1404 nbits = XFS_BB_NUM_BITS;
1405 }
1406 xfs_btree_offsets(fields,
1407 (cur->bc_flags & XFS_BTREE_LONG_PTRS) ?
1408 loffsets : soffsets,
1409 nbits, &first, &last);
1410 xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
1411 xfs_trans_log_buf(cur->bc_tp, bp, first, last);
1412 } else {
1413 xfs_trans_log_inode(cur->bc_tp, cur->bc_private.b.ip,
1414 xfs_ilog_fbroot(cur->bc_private.b.whichfork));
1415 }
1416
1417 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1418 }
1419
1420 /*
1421 * Increment cursor by one record at the level.
1422 * For nonzero levels the leaf-ward information is untouched.
1423 */
1424 int /* error */
1425 xfs_btree_increment(
1426 struct xfs_btree_cur *cur,
1427 int level,
1428 int *stat) /* success/failure */
1429 {
1430 struct xfs_btree_block *block;
1431 union xfs_btree_ptr ptr;
1432 struct xfs_buf *bp;
1433 int error; /* error return value */
1434 int lev;
1435
1436 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1437 XFS_BTREE_TRACE_ARGI(cur, level);
1438
1439 ASSERT(level < cur->bc_nlevels);
1440
1441 /* Read-ahead to the right at this level. */
1442 xfs_btree_readahead(cur, level, XFS_BTCUR_RIGHTRA);
1443
1444 /* Get a pointer to the btree block. */
1445 block = xfs_btree_get_block(cur, level, &bp);
1446
1447 #ifdef DEBUG
1448 error = xfs_btree_check_block(cur, block, level, bp);
1449 if (error)
1450 goto error0;
1451 #endif
1452
1453 /* We're done if we remain in the block after the increment. */
1454 if (++cur->bc_ptrs[level] <= xfs_btree_get_numrecs(block))
1455 goto out1;
1456
1457 /* Fail if we just went off the right edge of the tree. */
1458 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
1459 if (xfs_btree_ptr_is_null(cur, &ptr))
1460 goto out0;
1461
1462 XFS_BTREE_STATS_INC(cur, increment);
1463
1464 /*
1465 * March up the tree incrementing pointers.
1466 * Stop when we don't go off the right edge of a block.
1467 */
1468 for (lev = level + 1; lev < cur->bc_nlevels; lev++) {
1469 block = xfs_btree_get_block(cur, lev, &bp);
1470
1471 #ifdef DEBUG
1472 error = xfs_btree_check_block(cur, block, lev, bp);
1473 if (error)
1474 goto error0;
1475 #endif
1476
1477 if (++cur->bc_ptrs[lev] <= xfs_btree_get_numrecs(block))
1478 break;
1479
1480 /* Read-ahead the right block for the next loop. */
1481 xfs_btree_readahead(cur, lev, XFS_BTCUR_RIGHTRA);
1482 }
1483
1484 /*
1485 * If we went off the root then we are either seriously
1486 * confused or have the tree root in an inode.
1487 */
1488 if (lev == cur->bc_nlevels) {
1489 if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE)
1490 goto out0;
1491 ASSERT(0);
1492 error = EFSCORRUPTED;
1493 goto error0;
1494 }
1495 ASSERT(lev < cur->bc_nlevels);
1496
1497 /*
1498 * Now walk back down the tree, fixing up the cursor's buffer
1499 * pointers and key numbers.
1500 */
1501 for (block = xfs_btree_get_block(cur, lev, &bp); lev > level; ) {
1502 union xfs_btree_ptr *ptrp;
1503
1504 ptrp = xfs_btree_ptr_addr(cur, cur->bc_ptrs[lev], block);
1505 error = xfs_btree_read_buf_block(cur, ptrp, --lev,
1506 0, &block, &bp);
1507 if (error)
1508 goto error0;
1509
1510 xfs_btree_setbuf(cur, lev, bp);
1511 cur->bc_ptrs[lev] = 1;
1512 }
1513 out1:
1514 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1515 *stat = 1;
1516 return 0;
1517
1518 out0:
1519 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1520 *stat = 0;
1521 return 0;
1522
1523 error0:
1524 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
1525 return error;
1526 }
1527
1528 /*
1529 * Decrement cursor by one record at the level.
1530 * For nonzero levels the leaf-ward information is untouched.
1531 */
1532 int /* error */
1533 xfs_btree_decrement(
1534 struct xfs_btree_cur *cur,
1535 int level,
1536 int *stat) /* success/failure */
1537 {
1538 struct xfs_btree_block *block;
1539 xfs_buf_t *bp;
1540 int error; /* error return value */
1541 int lev;
1542 union xfs_btree_ptr ptr;
1543
1544 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1545 XFS_BTREE_TRACE_ARGI(cur, level);
1546
1547 ASSERT(level < cur->bc_nlevels);
1548
1549 /* Read-ahead to the left at this level. */
1550 xfs_btree_readahead(cur, level, XFS_BTCUR_LEFTRA);
1551
1552 /* We're done if we remain in the block after the decrement. */
1553 if (--cur->bc_ptrs[level] > 0)
1554 goto out1;
1555
1556 /* Get a pointer to the btree block. */
1557 block = xfs_btree_get_block(cur, level, &bp);
1558
1559 #ifdef DEBUG
1560 error = xfs_btree_check_block(cur, block, level, bp);
1561 if (error)
1562 goto error0;
1563 #endif
1564
1565 /* Fail if we just went off the left edge of the tree. */
1566 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_LEFTSIB);
1567 if (xfs_btree_ptr_is_null(cur, &ptr))
1568 goto out0;
1569
1570 XFS_BTREE_STATS_INC(cur, decrement);
1571
1572 /*
1573 * March up the tree decrementing pointers.
1574 * Stop when we don't go off the left edge of a block.
1575 */
1576 for (lev = level + 1; lev < cur->bc_nlevels; lev++) {
1577 if (--cur->bc_ptrs[lev] > 0)
1578 break;
1579 /* Read-ahead the left block for the next loop. */
1580 xfs_btree_readahead(cur, lev, XFS_BTCUR_LEFTRA);
1581 }
1582
1583 /*
1584 * If we went off the root then we are seriously confused.
1585 * or the root of the tree is in an inode.
1586 */
1587 if (lev == cur->bc_nlevels) {
1588 if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE)
1589 goto out0;
1590 ASSERT(0);
1591 error = EFSCORRUPTED;
1592 goto error0;
1593 }
1594 ASSERT(lev < cur->bc_nlevels);
1595
1596 /*
1597 * Now walk back down the tree, fixing up the cursor's buffer
1598 * pointers and key numbers.
1599 */
1600 for (block = xfs_btree_get_block(cur, lev, &bp); lev > level; ) {
1601 union xfs_btree_ptr *ptrp;
1602
1603 ptrp = xfs_btree_ptr_addr(cur, cur->bc_ptrs[lev], block);
1604 error = xfs_btree_read_buf_block(cur, ptrp, --lev,
1605 0, &block, &bp);
1606 if (error)
1607 goto error0;
1608 xfs_btree_setbuf(cur, lev, bp);
1609 cur->bc_ptrs[lev] = xfs_btree_get_numrecs(block);
1610 }
1611 out1:
1612 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1613 *stat = 1;
1614 return 0;
1615
1616 out0:
1617 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1618 *stat = 0;
1619 return 0;
1620
1621 error0:
1622 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
1623 return error;
1624 }
1625
1626 STATIC int
1627 xfs_btree_lookup_get_block(
1628 struct xfs_btree_cur *cur, /* btree cursor */
1629 int level, /* level in the btree */
1630 union xfs_btree_ptr *pp, /* ptr to btree block */
1631 struct xfs_btree_block **blkp) /* return btree block */
1632 {
1633 struct xfs_buf *bp; /* buffer pointer for btree block */
1634 int error = 0;
1635
1636 /* special case the root block if in an inode */
1637 if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
1638 (level == cur->bc_nlevels - 1)) {
1639 *blkp = xfs_btree_get_iroot(cur);
1640 return 0;
1641 }
1642
1643 /*
1644 * If the old buffer at this level for the disk address we are
1645 * looking for re-use it.
1646 *
1647 * Otherwise throw it away and get a new one.
1648 */
1649 bp = cur->bc_bufs[level];
1650 if (bp && XFS_BUF_ADDR(bp) == xfs_btree_ptr_to_daddr(cur, pp)) {
1651 *blkp = XFS_BUF_TO_BLOCK(bp);
1652 return 0;
1653 }
1654
1655 error = xfs_btree_read_buf_block(cur, pp, level, 0, blkp, &bp);
1656 if (error)
1657 return error;
1658
1659 xfs_btree_setbuf(cur, level, bp);
1660 return 0;
1661 }
1662
1663 /*
1664 * Get current search key. For level 0 we don't actually have a key
1665 * structure so we make one up from the record. For all other levels
1666 * we just return the right key.
1667 */
1668 STATIC union xfs_btree_key *
1669 xfs_lookup_get_search_key(
1670 struct xfs_btree_cur *cur,
1671 int level,
1672 int keyno,
1673 struct xfs_btree_block *block,
1674 union xfs_btree_key *kp)
1675 {
1676 if (level == 0) {
1677 cur->bc_ops->init_key_from_rec(kp,
1678 xfs_btree_rec_addr(cur, keyno, block));
1679 return kp;
1680 }
1681
1682 return xfs_btree_key_addr(cur, keyno, block);
1683 }
1684
1685 /*
1686 * Lookup the record. The cursor is made to point to it, based on dir.
1687 * Return 0 if can't find any such record, 1 for success.
1688 */
1689 int /* error */
1690 xfs_btree_lookup(
1691 struct xfs_btree_cur *cur, /* btree cursor */
1692 xfs_lookup_t dir, /* <=, ==, or >= */
1693 int *stat) /* success/failure */
1694 {
1695 struct xfs_btree_block *block; /* current btree block */
1696 __int64_t diff; /* difference for the current key */
1697 int error; /* error return value */
1698 int keyno; /* current key number */
1699 int level; /* level in the btree */
1700 union xfs_btree_ptr *pp; /* ptr to btree block */
1701 union xfs_btree_ptr ptr; /* ptr to btree block */
1702
1703 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1704 XFS_BTREE_TRACE_ARGI(cur, dir);
1705
1706 XFS_BTREE_STATS_INC(cur, lookup);
1707
1708 block = NULL;
1709 keyno = 0;
1710
1711 /* initialise start pointer from cursor */
1712 cur->bc_ops->init_ptr_from_cur(cur, &ptr);
1713 pp = &ptr;
1714
1715 /*
1716 * Iterate over each level in the btree, starting at the root.
1717 * For each level above the leaves, find the key we need, based
1718 * on the lookup record, then follow the corresponding block
1719 * pointer down to the next level.
1720 */
1721 for (level = cur->bc_nlevels - 1, diff = 1; level >= 0; level--) {
1722 /* Get the block we need to do the lookup on. */
1723 error = xfs_btree_lookup_get_block(cur, level, pp, &block);
1724 if (error)
1725 goto error0;
1726
1727 if (diff == 0) {
1728 /*
1729 * If we already had a key match at a higher level, we
1730 * know we need to use the first entry in this block.
1731 */
1732 keyno = 1;
1733 } else {
1734 /* Otherwise search this block. Do a binary search. */
1735
1736 int high; /* high entry number */
1737 int low; /* low entry number */
1738
1739 /* Set low and high entry numbers, 1-based. */
1740 low = 1;
1741 high = xfs_btree_get_numrecs(block);
1742 if (!high) {
1743 /* Block is empty, must be an empty leaf. */
1744 ASSERT(level == 0 && cur->bc_nlevels == 1);
1745
1746 cur->bc_ptrs[0] = dir != XFS_LOOKUP_LE;
1747 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1748 *stat = 0;
1749 return 0;
1750 }
1751
1752 /* Binary search the block. */
1753 while (low <= high) {
1754 union xfs_btree_key key;
1755 union xfs_btree_key *kp;
1756
1757 XFS_BTREE_STATS_INC(cur, compare);
1758
1759 /* keyno is average of low and high. */
1760 keyno = (low + high) >> 1;
1761
1762 /* Get current search key */
1763 kp = xfs_lookup_get_search_key(cur, level,
1764 keyno, block, &key);
1765
1766 /*
1767 * Compute difference to get next direction:
1768 * - less than, move right
1769 * - greater than, move left
1770 * - equal, we're done
1771 */
1772 diff = cur->bc_ops->key_diff(cur, kp);
1773 if (diff < 0)
1774 low = keyno + 1;
1775 else if (diff > 0)
1776 high = keyno - 1;
1777 else
1778 break;
1779 }
1780 }
1781
1782 /*
1783 * If there are more levels, set up for the next level
1784 * by getting the block number and filling in the cursor.
1785 */
1786 if (level > 0) {
1787 /*
1788 * If we moved left, need the previous key number,
1789 * unless there isn't one.
1790 */
1791 if (diff > 0 && --keyno < 1)
1792 keyno = 1;
1793 pp = xfs_btree_ptr_addr(cur, keyno, block);
1794
1795 #ifdef DEBUG
1796 error = xfs_btree_check_ptr(cur, pp, 0, level);
1797 if (error)
1798 goto error0;
1799 #endif
1800 cur->bc_ptrs[level] = keyno;
1801 }
1802 }
1803
1804 /* Done with the search. See if we need to adjust the results. */
1805 if (dir != XFS_LOOKUP_LE && diff < 0) {
1806 keyno++;
1807 /*
1808 * If ge search and we went off the end of the block, but it's
1809 * not the last block, we're in the wrong block.
1810 */
1811 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
1812 if (dir == XFS_LOOKUP_GE &&
1813 keyno > xfs_btree_get_numrecs(block) &&
1814 !xfs_btree_ptr_is_null(cur, &ptr)) {
1815 int i;
1816
1817 cur->bc_ptrs[0] = keyno;
1818 error = xfs_btree_increment(cur, 0, &i);
1819 if (error)
1820 goto error0;
1821 XFS_WANT_CORRUPTED_RETURN(i == 1);
1822 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1823 *stat = 1;
1824 return 0;
1825 }
1826 } else if (dir == XFS_LOOKUP_LE && diff > 0)
1827 keyno--;
1828 cur->bc_ptrs[0] = keyno;
1829
1830 /* Return if we succeeded or not. */
1831 if (keyno == 0 || keyno > xfs_btree_get_numrecs(block))
1832 *stat = 0;
1833 else if (dir != XFS_LOOKUP_EQ || diff == 0)
1834 *stat = 1;
1835 else
1836 *stat = 0;
1837 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1838 return 0;
1839
1840 error0:
1841 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
1842 return error;
1843 }
1844
1845 /*
1846 * Update keys at all levels from here to the root along the cursor's path.
1847 */
1848 STATIC int
1849 xfs_btree_updkey(
1850 struct xfs_btree_cur *cur,
1851 union xfs_btree_key *keyp,
1852 int level)
1853 {
1854 struct xfs_btree_block *block;
1855 struct xfs_buf *bp;
1856 union xfs_btree_key *kp;
1857 int ptr;
1858
1859 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1860 XFS_BTREE_TRACE_ARGIK(cur, level, keyp);
1861
1862 ASSERT(!(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) || level >= 1);
1863
1864 /*
1865 * Go up the tree from this level toward the root.
1866 * At each level, update the key value to the value input.
1867 * Stop when we reach a level where the cursor isn't pointing
1868 * at the first entry in the block.
1869 */
1870 for (ptr = 1; ptr == 1 && level < cur->bc_nlevels; level++) {
1871 #ifdef DEBUG
1872 int error;
1873 #endif
1874 block = xfs_btree_get_block(cur, level, &bp);
1875 #ifdef DEBUG
1876 error = xfs_btree_check_block(cur, block, level, bp);
1877 if (error) {
1878 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
1879 return error;
1880 }
1881 #endif
1882 ptr = cur->bc_ptrs[level];
1883 kp = xfs_btree_key_addr(cur, ptr, block);
1884 xfs_btree_copy_keys(cur, kp, keyp, 1);
1885 xfs_btree_log_keys(cur, bp, ptr, ptr);
1886 }
1887
1888 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1889 return 0;
1890 }
1891
1892 /*
1893 * Update the record referred to by cur to the value in the
1894 * given record. This either works (return 0) or gets an
1895 * EFSCORRUPTED error.
1896 */
1897 int
1898 xfs_btree_update(
1899 struct xfs_btree_cur *cur,
1900 union xfs_btree_rec *rec)
1901 {
1902 struct xfs_btree_block *block;
1903 struct xfs_buf *bp;
1904 int error;
1905 int ptr;
1906 union xfs_btree_rec *rp;
1907
1908 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1909 XFS_BTREE_TRACE_ARGR(cur, rec);
1910
1911 /* Pick up the current block. */
1912 block = xfs_btree_get_block(cur, 0, &bp);
1913
1914 #ifdef DEBUG
1915 error = xfs_btree_check_block(cur, block, 0, bp);
1916 if (error)
1917 goto error0;
1918 #endif
1919 /* Get the address of the rec to be updated. */
1920 ptr = cur->bc_ptrs[0];
1921 rp = xfs_btree_rec_addr(cur, ptr, block);
1922
1923 /* Fill in the new contents and log them. */
1924 xfs_btree_copy_recs(cur, rp, rec, 1);
1925 xfs_btree_log_recs(cur, bp, ptr, ptr);
1926
1927 /*
1928 * If we are tracking the last record in the tree and
1929 * we are at the far right edge of the tree, update it.
1930 */
1931 if (xfs_btree_is_lastrec(cur, block, 0)) {
1932 cur->bc_ops->update_lastrec(cur, block, rec,
1933 ptr, LASTREC_UPDATE);
1934 }
1935
1936 /* Updating first rec in leaf. Pass new key value up to our parent. */
1937 if (ptr == 1) {
1938 union xfs_btree_key key;
1939
1940 cur->bc_ops->init_key_from_rec(&key, rec);
1941 error = xfs_btree_updkey(cur, &key, 1);
1942 if (error)
1943 goto error0;
1944 }
1945
1946 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1947 return 0;
1948
1949 error0:
1950 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
1951 return error;
1952 }
1953
1954 /*
1955 * Move 1 record left from cur/level if possible.
1956 * Update cur to reflect the new path.
1957 */
1958 STATIC int /* error */
1959 xfs_btree_lshift(
1960 struct xfs_btree_cur *cur,
1961 int level,
1962 int *stat) /* success/failure */
1963 {
1964 union xfs_btree_key key; /* btree key */
1965 struct xfs_buf *lbp; /* left buffer pointer */
1966 struct xfs_btree_block *left; /* left btree block */
1967 int lrecs; /* left record count */
1968 struct xfs_buf *rbp; /* right buffer pointer */
1969 struct xfs_btree_block *right; /* right btree block */
1970 int rrecs; /* right record count */
1971 union xfs_btree_ptr lptr; /* left btree pointer */
1972 union xfs_btree_key *rkp = NULL; /* right btree key */
1973 union xfs_btree_ptr *rpp = NULL; /* right address pointer */
1974 union xfs_btree_rec *rrp = NULL; /* right record pointer */
1975 int error; /* error return value */
1976
1977 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1978 XFS_BTREE_TRACE_ARGI(cur, level);
1979
1980 if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
1981 level == cur->bc_nlevels - 1)
1982 goto out0;
1983
1984 /* Set up variables for this block as "right". */
1985 right = xfs_btree_get_block(cur, level, &rbp);
1986
1987 #ifdef DEBUG
1988 error = xfs_btree_check_block(cur, right, level, rbp);
1989 if (error)
1990 goto error0;
1991 #endif
1992
1993 /* If we've got no left sibling then we can't shift an entry left. */
1994 xfs_btree_get_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
1995 if (xfs_btree_ptr_is_null(cur, &lptr))
1996 goto out0;
1997
1998 /*
1999 * If the cursor entry is the one that would be moved, don't
2000 * do it... it's too complicated.
2001 */
2002 if (cur->bc_ptrs[level] <= 1)
2003 goto out0;
2004
2005 /* Set up the left neighbor as "left". */
2006 error = xfs_btree_read_buf_block(cur, &lptr, level, 0, &left, &lbp);
2007 if (error)
2008 goto error0;
2009
2010 /* If it's full, it can't take another entry. */
2011 lrecs = xfs_btree_get_numrecs(left);
2012 if (lrecs == cur->bc_ops->get_maxrecs(cur, level))
2013 goto out0;
2014
2015 rrecs = xfs_btree_get_numrecs(right);
2016
2017 /*
2018 * We add one entry to the left side and remove one for the right side.
2019 * Account for it here, the changes will be updated on disk and logged
2020 * later.
2021 */
2022 lrecs++;
2023 rrecs--;
2024
2025 XFS_BTREE_STATS_INC(cur, lshift);
2026 XFS_BTREE_STATS_ADD(cur, moves, 1);
2027
2028 /*
2029 * If non-leaf, copy a key and a ptr to the left block.
2030 * Log the changes to the left block.
2031 */
2032 if (level > 0) {
2033 /* It's a non-leaf. Move keys and pointers. */
2034 union xfs_btree_key *lkp; /* left btree key */
2035 union xfs_btree_ptr *lpp; /* left address pointer */
2036
2037 lkp = xfs_btree_key_addr(cur, lrecs, left);
2038 rkp = xfs_btree_key_addr(cur, 1, right);
2039
2040 lpp = xfs_btree_ptr_addr(cur, lrecs, left);
2041 rpp = xfs_btree_ptr_addr(cur, 1, right);
2042 #ifdef DEBUG
2043 error = xfs_btree_check_ptr(cur, rpp, 0, level);
2044 if (error)
2045 goto error0;
2046 #endif
2047 xfs_btree_copy_keys(cur, lkp, rkp, 1);
2048 xfs_btree_copy_ptrs(cur, lpp, rpp, 1);
2049
2050 xfs_btree_log_keys(cur, lbp, lrecs, lrecs);
2051 xfs_btree_log_ptrs(cur, lbp, lrecs, lrecs);
2052
2053 ASSERT(cur->bc_ops->keys_inorder(cur,
2054 xfs_btree_key_addr(cur, lrecs - 1, left), lkp));
2055 } else {
2056 /* It's a leaf. Move records. */
2057 union xfs_btree_rec *lrp; /* left record pointer */
2058
2059 lrp = xfs_btree_rec_addr(cur, lrecs, left);
2060 rrp = xfs_btree_rec_addr(cur, 1, right);
2061
2062 xfs_btree_copy_recs(cur, lrp, rrp, 1);
2063 xfs_btree_log_recs(cur, lbp, lrecs, lrecs);
2064
2065 ASSERT(cur->bc_ops->recs_inorder(cur,
2066 xfs_btree_rec_addr(cur, lrecs - 1, left), lrp));
2067 }
2068
2069 xfs_btree_set_numrecs(left, lrecs);
2070 xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS);
2071
2072 xfs_btree_set_numrecs(right, rrecs);
2073 xfs_btree_log_block(cur, rbp, XFS_BB_NUMRECS);
2074
2075 /*
2076 * Slide the contents of right down one entry.
2077 */
2078 XFS_BTREE_STATS_ADD(cur, moves, rrecs - 1);
2079 if (level > 0) {
2080 /* It's a nonleaf. operate on keys and ptrs */
2081 #ifdef DEBUG
2082 int i; /* loop index */
2083
2084 for (i = 0; i < rrecs; i++) {
2085 error = xfs_btree_check_ptr(cur, rpp, i + 1, level);
2086 if (error)
2087 goto error0;
2088 }
2089 #endif
2090 xfs_btree_shift_keys(cur,
2091 xfs_btree_key_addr(cur, 2, right),
2092 -1, rrecs);
2093 xfs_btree_shift_ptrs(cur,
2094 xfs_btree_ptr_addr(cur, 2, right),
2095 -1, rrecs);
2096
2097 xfs_btree_log_keys(cur, rbp, 1, rrecs);
2098 xfs_btree_log_ptrs(cur, rbp, 1, rrecs);
2099 } else {
2100 /* It's a leaf. operate on records */
2101 xfs_btree_shift_recs(cur,
2102 xfs_btree_rec_addr(cur, 2, right),
2103 -1, rrecs);
2104 xfs_btree_log_recs(cur, rbp, 1, rrecs);
2105
2106 /*
2107 * If it's the first record in the block, we'll need a key
2108 * structure to pass up to the next level (updkey).
2109 */
2110 cur->bc_ops->init_key_from_rec(&key,
2111 xfs_btree_rec_addr(cur, 1, right));
2112 rkp = &key;
2113 }
2114
2115 /* Update the parent key values of right. */
2116 error = xfs_btree_updkey(cur, rkp, level + 1);
2117 if (error)
2118 goto error0;
2119
2120 /* Slide the cursor value left one. */
2121 cur->bc_ptrs[level]--;
2122
2123 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2124 *stat = 1;
2125 return 0;
2126
2127 out0:
2128 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2129 *stat = 0;
2130 return 0;
2131
2132 error0:
2133 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2134 return error;
2135 }
2136
2137 /*
2138 * Move 1 record right from cur/level if possible.
2139 * Update cur to reflect the new path.
2140 */
2141 STATIC int /* error */
2142 xfs_btree_rshift(
2143 struct xfs_btree_cur *cur,
2144 int level,
2145 int *stat) /* success/failure */
2146 {
2147 union xfs_btree_key key; /* btree key */
2148 struct xfs_buf *lbp; /* left buffer pointer */
2149 struct xfs_btree_block *left; /* left btree block */
2150 struct xfs_buf *rbp; /* right buffer pointer */
2151 struct xfs_btree_block *right; /* right btree block */
2152 struct xfs_btree_cur *tcur; /* temporary btree cursor */
2153 union xfs_btree_ptr rptr; /* right block pointer */
2154 union xfs_btree_key *rkp; /* right btree key */
2155 int rrecs; /* right record count */
2156 int lrecs; /* left record count */
2157 int error; /* error return value */
2158 int i; /* loop counter */
2159
2160 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2161 XFS_BTREE_TRACE_ARGI(cur, level);
2162
2163 if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
2164 (level == cur->bc_nlevels - 1))
2165 goto out0;
2166
2167 /* Set up variables for this block as "left". */
2168 left = xfs_btree_get_block(cur, level, &lbp);
2169
2170 #ifdef DEBUG
2171 error = xfs_btree_check_block(cur, left, level, lbp);
2172 if (error)
2173 goto error0;
2174 #endif
2175
2176 /* If we've got no right sibling then we can't shift an entry right. */
2177 xfs_btree_get_sibling(cur, left, &rptr, XFS_BB_RIGHTSIB);
2178 if (xfs_btree_ptr_is_null(cur, &rptr))
2179 goto out0;
2180
2181 /*
2182 * If the cursor entry is the one that would be moved, don't
2183 * do it... it's too complicated.
2184 */
2185 lrecs = xfs_btree_get_numrecs(left);
2186 if (cur->bc_ptrs[level] >= lrecs)
2187 goto out0;
2188
2189 /* Set up the right neighbor as "right". */
2190 error = xfs_btree_read_buf_block(cur, &rptr, level, 0, &right, &rbp);
2191 if (error)
2192 goto error0;
2193
2194 /* If it's full, it can't take another entry. */
2195 rrecs = xfs_btree_get_numrecs(right);
2196 if (rrecs == cur->bc_ops->get_maxrecs(cur, level))
2197 goto out0;
2198
2199 XFS_BTREE_STATS_INC(cur, rshift);
2200 XFS_BTREE_STATS_ADD(cur, moves, rrecs);
2201
2202 /*
2203 * Make a hole at the start of the right neighbor block, then
2204 * copy the last left block entry to the hole.
2205 */
2206 if (level > 0) {
2207 /* It's a nonleaf. make a hole in the keys and ptrs */
2208 union xfs_btree_key *lkp;
2209 union xfs_btree_ptr *lpp;
2210 union xfs_btree_ptr *rpp;
2211
2212 lkp = xfs_btree_key_addr(cur, lrecs, left);
2213 lpp = xfs_btree_ptr_addr(cur, lrecs, left);
2214 rkp = xfs_btree_key_addr(cur, 1, right);
2215 rpp = xfs_btree_ptr_addr(cur, 1, right);
2216
2217 #ifdef DEBUG
2218 for (i = rrecs - 1; i >= 0; i--) {
2219 error = xfs_btree_check_ptr(cur, rpp, i, level);
2220 if (error)
2221 goto error0;
2222 }
2223 #endif
2224
2225 xfs_btree_shift_keys(cur, rkp, 1, rrecs);
2226 xfs_btree_shift_ptrs(cur, rpp, 1, rrecs);
2227
2228 #ifdef DEBUG
2229 error = xfs_btree_check_ptr(cur, lpp, 0, level);
2230 if (error)
2231 goto error0;
2232 #endif
2233
2234 /* Now put the new data in, and log it. */
2235 xfs_btree_copy_keys(cur, rkp, lkp, 1);
2236 xfs_btree_copy_ptrs(cur, rpp, lpp, 1);
2237
2238 xfs_btree_log_keys(cur, rbp, 1, rrecs + 1);
2239 xfs_btree_log_ptrs(cur, rbp, 1, rrecs + 1);
2240
2241 ASSERT(cur->bc_ops->keys_inorder(cur, rkp,
2242 xfs_btree_key_addr(cur, 2, right)));
2243 } else {
2244 /* It's a leaf. make a hole in the records */
2245 union xfs_btree_rec *lrp;
2246 union xfs_btree_rec *rrp;
2247
2248 lrp = xfs_btree_rec_addr(cur, lrecs, left);
2249 rrp = xfs_btree_rec_addr(cur, 1, right);
2250
2251 xfs_btree_shift_recs(cur, rrp, 1, rrecs);
2252
2253 /* Now put the new data in, and log it. */
2254 xfs_btree_copy_recs(cur, rrp, lrp, 1);
2255 xfs_btree_log_recs(cur, rbp, 1, rrecs + 1);
2256
2257 cur->bc_ops->init_key_from_rec(&key, rrp);
2258 rkp = &key;
2259
2260 ASSERT(cur->bc_ops->recs_inorder(cur, rrp,
2261 xfs_btree_rec_addr(cur, 2, right)));
2262 }
2263
2264 /*
2265 * Decrement and log left's numrecs, bump and log right's numrecs.
2266 */
2267 xfs_btree_set_numrecs(left, --lrecs);
2268 xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS);
2269
2270 xfs_btree_set_numrecs(right, ++rrecs);
2271 xfs_btree_log_block(cur, rbp, XFS_BB_NUMRECS);
2272
2273 /*
2274 * Using a temporary cursor, update the parent key values of the
2275 * block on the right.
2276 */
2277 error = xfs_btree_dup_cursor(cur, &tcur);
2278 if (error)
2279 goto error0;
2280 i = xfs_btree_lastrec(tcur, level);
2281 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
2282
2283 error = xfs_btree_increment(tcur, level, &i);
2284 if (error)
2285 goto error1;
2286
2287 error = xfs_btree_updkey(tcur, rkp, level + 1);
2288 if (error)
2289 goto error1;
2290
2291 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
2292
2293 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2294 *stat = 1;
2295 return 0;
2296
2297 out0:
2298 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2299 *stat = 0;
2300 return 0;
2301
2302 error0:
2303 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2304 return error;
2305
2306 error1:
2307 XFS_BTREE_TRACE_CURSOR(tcur, XBT_ERROR);
2308 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
2309 return error;
2310 }
2311
2312 /*
2313 * Split cur/level block in half.
2314 * Return new block number and the key to its first
2315 * record (to be inserted into parent).
2316 */
2317 STATIC int /* error */
2318 xfs_btree_split(
2319 struct xfs_btree_cur *cur,
2320 int level,
2321 union xfs_btree_ptr *ptrp,
2322 union xfs_btree_key *key,
2323 struct xfs_btree_cur **curp,
2324 int *stat) /* success/failure */
2325 {
2326 union xfs_btree_ptr lptr; /* left sibling block ptr */
2327 struct xfs_buf *lbp; /* left buffer pointer */
2328 struct xfs_btree_block *left; /* left btree block */
2329 union xfs_btree_ptr rptr; /* right sibling block ptr */
2330 struct xfs_buf *rbp; /* right buffer pointer */
2331 struct xfs_btree_block *right; /* right btree block */
2332 union xfs_btree_ptr rrptr; /* right-right sibling ptr */
2333 struct xfs_buf *rrbp; /* right-right buffer pointer */
2334 struct xfs_btree_block *rrblock; /* right-right btree block */
2335 int lrecs;
2336 int rrecs;
2337 int src_index;
2338 int error; /* error return value */
2339 #ifdef DEBUG
2340 int i;
2341 #endif
2342
2343 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2344 XFS_BTREE_TRACE_ARGIPK(cur, level, *ptrp, key);
2345
2346 XFS_BTREE_STATS_INC(cur, split);
2347
2348 /* Set up left block (current one). */
2349 left = xfs_btree_get_block(cur, level, &lbp);
2350
2351 #ifdef DEBUG
2352 error = xfs_btree_check_block(cur, left, level, lbp);
2353 if (error)
2354 goto error0;
2355 #endif
2356
2357 xfs_btree_buf_to_ptr(cur, lbp, &lptr);
2358
2359 /* Allocate the new block. If we can't do it, we're toast. Give up. */
2360 error = cur->bc_ops->alloc_block(cur, &lptr, &rptr, 1, stat);
2361 if (error)
2362 goto error0;
2363 if (*stat == 0)
2364 goto out0;
2365 XFS_BTREE_STATS_INC(cur, alloc);
2366
2367 /* Set up the new block as "right". */
2368 error = xfs_btree_get_buf_block(cur, &rptr, 0, &right, &rbp);
2369 if (error)
2370 goto error0;
2371
2372 /* Fill in the btree header for the new right block. */
2373 xfs_btree_init_block_cur(cur, rbp, xfs_btree_get_level(left), 0);
2374
2375 /*
2376 * Split the entries between the old and the new block evenly.
2377 * Make sure that if there's an odd number of entries now, that
2378 * each new block will have the same number of entries.
2379 */
2380 lrecs = xfs_btree_get_numrecs(left);
2381 rrecs = lrecs / 2;
2382 if ((lrecs & 1) && cur->bc_ptrs[level] <= rrecs + 1)
2383 rrecs++;
2384 src_index = (lrecs - rrecs + 1);
2385
2386 XFS_BTREE_STATS_ADD(cur, moves, rrecs);
2387
2388 /*
2389 * Copy btree block entries from the left block over to the
2390 * new block, the right. Update the right block and log the
2391 * changes.
2392 */
2393 if (level > 0) {
2394 /* It's a non-leaf. Move keys and pointers. */
2395 union xfs_btree_key *lkp; /* left btree key */
2396 union xfs_btree_ptr *lpp; /* left address pointer */
2397 union xfs_btree_key *rkp; /* right btree key */
2398 union xfs_btree_ptr *rpp; /* right address pointer */
2399
2400 lkp = xfs_btree_key_addr(cur, src_index, left);
2401 lpp = xfs_btree_ptr_addr(cur, src_index, left);
2402 rkp = xfs_btree_key_addr(cur, 1, right);
2403 rpp = xfs_btree_ptr_addr(cur, 1, right);
2404
2405 #ifdef DEBUG
2406 for (i = src_index; i < rrecs; i++) {
2407 error = xfs_btree_check_ptr(cur, lpp, i, level);
2408 if (error)
2409 goto error0;
2410 }
2411 #endif
2412
2413 xfs_btree_copy_keys(cur, rkp, lkp, rrecs);
2414 xfs_btree_copy_ptrs(cur, rpp, lpp, rrecs);
2415
2416 xfs_btree_log_keys(cur, rbp, 1, rrecs);
2417 xfs_btree_log_ptrs(cur, rbp, 1, rrecs);
2418
2419 /* Grab the keys to the entries moved to the right block */
2420 xfs_btree_copy_keys(cur, key, rkp, 1);
2421 } else {
2422 /* It's a leaf. Move records. */
2423 union xfs_btree_rec *lrp; /* left record pointer */
2424 union xfs_btree_rec *rrp; /* right record pointer */
2425
2426 lrp = xfs_btree_rec_addr(cur, src_index, left);
2427 rrp = xfs_btree_rec_addr(cur, 1, right);
2428
2429 xfs_btree_copy_recs(cur, rrp, lrp, rrecs);
2430 xfs_btree_log_recs(cur, rbp, 1, rrecs);
2431
2432 cur->bc_ops->init_key_from_rec(key,
2433 xfs_btree_rec_addr(cur, 1, right));
2434 }
2435
2436
2437 /*
2438 * Find the left block number by looking in the buffer.
2439 * Adjust numrecs, sibling pointers.
2440 */
2441 xfs_btree_get_sibling(cur, left, &rrptr, XFS_BB_RIGHTSIB);
2442 xfs_btree_set_sibling(cur, right, &rrptr, XFS_BB_RIGHTSIB);
2443 xfs_btree_set_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
2444 xfs_btree_set_sibling(cur, left, &rptr, XFS_BB_RIGHTSIB);
2445
2446 lrecs -= rrecs;
2447 xfs_btree_set_numrecs(left, lrecs);
2448 xfs_btree_set_numrecs(right, xfs_btree_get_numrecs(right) + rrecs);
2449
2450 xfs_btree_log_block(cur, rbp, XFS_BB_ALL_BITS);
2451 xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS | XFS_BB_RIGHTSIB);
2452
2453 /*
2454 * If there's a block to the new block's right, make that block
2455 * point back to right instead of to left.
2456 */
2457 if (!xfs_btree_ptr_is_null(cur, &rrptr)) {
2458 error = xfs_btree_read_buf_block(cur, &rrptr, level,
2459 0, &rrblock, &rrbp);
2460 if (error)
2461 goto error0;
2462 xfs_btree_set_sibling(cur, rrblock, &rptr, XFS_BB_LEFTSIB);
2463 xfs_btree_log_block(cur, rrbp, XFS_BB_LEFTSIB);
2464 }
2465 /*
2466 * If the cursor is really in the right block, move it there.
2467 * If it's just pointing past the last entry in left, then we'll
2468 * insert there, so don't change anything in that case.
2469 */
2470 if (cur->bc_ptrs[level] > lrecs + 1) {
2471 xfs_btree_setbuf(cur, level, rbp);
2472 cur->bc_ptrs[level] -= lrecs;
2473 }
2474 /*
2475 * If there are more levels, we'll need another cursor which refers
2476 * the right block, no matter where this cursor was.
2477 */
2478 if (level + 1 < cur->bc_nlevels) {
2479 error = xfs_btree_dup_cursor(cur, curp);
2480 if (error)
2481 goto error0;
2482 (*curp)->bc_ptrs[level + 1]++;
2483 }
2484 *ptrp = rptr;
2485 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2486 *stat = 1;
2487 return 0;
2488 out0:
2489 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2490 *stat = 0;
2491 return 0;
2492
2493 error0:
2494 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2495 return error;
2496 }
2497
2498 /*
2499 * Copy the old inode root contents into a real block and make the
2500 * broot point to it.
2501 */
2502 int /* error */
2503 xfs_btree_new_iroot(
2504 struct xfs_btree_cur *cur, /* btree cursor */
2505 int *logflags, /* logging flags for inode */
2506 int *stat) /* return status - 0 fail */
2507 {
2508 struct xfs_buf *cbp; /* buffer for cblock */
2509 struct xfs_btree_block *block; /* btree block */
2510 struct xfs_btree_block *cblock; /* child btree block */
2511 union xfs_btree_key *ckp; /* child key pointer */
2512 union xfs_btree_ptr *cpp; /* child ptr pointer */
2513 union xfs_btree_key *kp; /* pointer to btree key */
2514 union xfs_btree_ptr *pp; /* pointer to block addr */
2515 union xfs_btree_ptr nptr; /* new block addr */
2516 int level; /* btree level */
2517 int error; /* error return code */
2518 #ifdef DEBUG
2519 int i; /* loop counter */
2520 #endif
2521
2522 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2523 XFS_BTREE_STATS_INC(cur, newroot);
2524
2525 ASSERT(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE);
2526
2527 level = cur->bc_nlevels - 1;
2528
2529 block = xfs_btree_get_iroot(cur);
2530 pp = xfs_btree_ptr_addr(cur, 1, block);
2531
2532 /* Allocate the new block. If we can't do it, we're toast. Give up. */
2533 error = cur->bc_ops->alloc_block(cur, pp, &nptr, 1, stat);
2534 if (error)
2535 goto error0;
2536 if (*stat == 0) {
2537 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2538 return 0;
2539 }
2540 XFS_BTREE_STATS_INC(cur, alloc);
2541
2542 /* Copy the root into a real block. */
2543 error = xfs_btree_get_buf_block(cur, &nptr, 0, &cblock, &cbp);
2544 if (error)
2545 goto error0;
2546
2547 memcpy(cblock, block, xfs_btree_block_len(cur));
2548
2549 be16_add_cpu(&block->bb_level, 1);
2550 xfs_btree_set_numrecs(block, 1);
2551 cur->bc_nlevels++;
2552 cur->bc_ptrs[level + 1] = 1;
2553
2554 kp = xfs_btree_key_addr(cur, 1, block);
2555 ckp = xfs_btree_key_addr(cur, 1, cblock);
2556 xfs_btree_copy_keys(cur, ckp, kp, xfs_btree_get_numrecs(cblock));
2557
2558 cpp = xfs_btree_ptr_addr(cur, 1, cblock);
2559 #ifdef DEBUG
2560 for (i = 0; i < be16_to_cpu(cblock->bb_numrecs); i++) {
2561 error = xfs_btree_check_ptr(cur, pp, i, level);
2562 if (error)
2563 goto error0;
2564 }
2565 #endif
2566 xfs_btree_copy_ptrs(cur, cpp, pp, xfs_btree_get_numrecs(cblock));
2567
2568 #ifdef DEBUG
2569 error = xfs_btree_check_ptr(cur, &nptr, 0, level);
2570 if (error)
2571 goto error0;
2572 #endif
2573 xfs_btree_copy_ptrs(cur, pp, &nptr, 1);
2574
2575 xfs_iroot_realloc(cur->bc_private.b.ip,
2576 1 - xfs_btree_get_numrecs(cblock),
2577 cur->bc_private.b.whichfork);
2578
2579 xfs_btree_setbuf(cur, level, cbp);
2580
2581 /*
2582 * Do all this logging at the end so that
2583 * the root is at the right level.
2584 */
2585 xfs_btree_log_block(cur, cbp, XFS_BB_ALL_BITS);
2586 xfs_btree_log_keys(cur, cbp, 1, be16_to_cpu(cblock->bb_numrecs));
2587 xfs_btree_log_ptrs(cur, cbp, 1, be16_to_cpu(cblock->bb_numrecs));
2588
2589 *logflags |=
2590 XFS_ILOG_CORE | xfs_ilog_fbroot(cur->bc_private.b.whichfork);
2591 *stat = 1;
2592 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2593 return 0;
2594 error0:
2595 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2596 return error;
2597 }
2598
2599 /*
2600 * Allocate a new root block, fill it in.
2601 */
2602 STATIC int /* error */
2603 xfs_btree_new_root(
2604 struct xfs_btree_cur *cur, /* btree cursor */
2605 int *stat) /* success/failure */
2606 {
2607 struct xfs_btree_block *block; /* one half of the old root block */
2608 struct xfs_buf *bp; /* buffer containing block */
2609 int error; /* error return value */
2610 struct xfs_buf *lbp; /* left buffer pointer */
2611 struct xfs_btree_block *left; /* left btree block */
2612 struct xfs_buf *nbp; /* new (root) buffer */
2613 struct xfs_btree_block *new; /* new (root) btree block */
2614 int nptr; /* new value for key index, 1 or 2 */
2615 struct xfs_buf *rbp; /* right buffer pointer */
2616 struct xfs_btree_block *right; /* right btree block */
2617 union xfs_btree_ptr rptr;
2618 union xfs_btree_ptr lptr;
2619
2620 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2621 XFS_BTREE_STATS_INC(cur, newroot);
2622
2623 /* initialise our start point from the cursor */
2624 cur->bc_ops->init_ptr_from_cur(cur, &rptr);
2625
2626 /* Allocate the new block. If we can't do it, we're toast. Give up. */
2627 error = cur->bc_ops->alloc_block(cur, &rptr, &lptr, 1, stat);
2628 if (error)
2629 goto error0;
2630 if (*stat == 0)
2631 goto out0;
2632 XFS_BTREE_STATS_INC(cur, alloc);
2633
2634 /* Set up the new block. */
2635 error = xfs_btree_get_buf_block(cur, &lptr, 0, &new, &nbp);
2636 if (error)
2637 goto error0;
2638
2639 /* Set the root in the holding structure increasing the level by 1. */
2640 cur->bc_ops->set_root(cur, &lptr, 1);
2641
2642 /*
2643 * At the previous root level there are now two blocks: the old root,
2644 * and the new block generated when it was split. We don't know which
2645 * one the cursor is pointing at, so we set up variables "left" and
2646 * "right" for each case.
2647 */
2648 block = xfs_btree_get_block(cur, cur->bc_nlevels - 1, &bp);
2649
2650 #ifdef DEBUG
2651 error = xfs_btree_check_block(cur, block, cur->bc_nlevels - 1, bp);
2652 if (error)
2653 goto error0;
2654 #endif
2655
2656 xfs_btree_get_sibling(cur, block, &rptr, XFS_BB_RIGHTSIB);
2657 if (!xfs_btree_ptr_is_null(cur, &rptr)) {
2658 /* Our block is left, pick up the right block. */
2659 lbp = bp;
2660 xfs_btree_buf_to_ptr(cur, lbp, &lptr);
2661 left = block;
2662 error = xfs_btree_read_buf_block(cur, &rptr,
2663 cur->bc_nlevels - 1, 0, &right, &rbp);
2664 if (error)
2665 goto error0;
2666 bp = rbp;
2667 nptr = 1;
2668 } else {
2669 /* Our block is right, pick up the left block. */
2670 rbp = bp;
2671 xfs_btree_buf_to_ptr(cur, rbp, &rptr);
2672 right = block;
2673 xfs_btree_get_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
2674 error = xfs_btree_read_buf_block(cur, &lptr,
2675 cur->bc_nlevels - 1, 0, &left, &lbp);
2676 if (error)
2677 goto error0;
2678 bp = lbp;
2679 nptr = 2;
2680 }
2681 /* Fill in the new block's btree header and log it. */
2682 xfs_btree_init_block_cur(cur, nbp, cur->bc_nlevels, 2);
2683 xfs_btree_log_block(cur, nbp, XFS_BB_ALL_BITS);
2684 ASSERT(!xfs_btree_ptr_is_null(cur, &lptr) &&
2685 !xfs_btree_ptr_is_null(cur, &rptr));
2686
2687 /* Fill in the key data in the new root. */
2688 if (xfs_btree_get_level(left) > 0) {
2689 xfs_btree_copy_keys(cur,
2690 xfs_btree_key_addr(cur, 1, new),
2691 xfs_btree_key_addr(cur, 1, left), 1);
2692 xfs_btree_copy_keys(cur,
2693 xfs_btree_key_addr(cur, 2, new),
2694 xfs_btree_key_addr(cur, 1, right), 1);
2695 } else {
2696 cur->bc_ops->init_key_from_rec(
2697 xfs_btree_key_addr(cur, 1, new),
2698 xfs_btree_rec_addr(cur, 1, left));
2699 cur->bc_ops->init_key_from_rec(
2700 xfs_btree_key_addr(cur, 2, new),
2701 xfs_btree_rec_addr(cur, 1, right));
2702 }
2703 xfs_btree_log_keys(cur, nbp, 1, 2);
2704
2705 /* Fill in the pointer data in the new root. */
2706 xfs_btree_copy_ptrs(cur,
2707 xfs_btree_ptr_addr(cur, 1, new), &lptr, 1);
2708 xfs_btree_copy_ptrs(cur,
2709 xfs_btree_ptr_addr(cur, 2, new), &rptr, 1);
2710 xfs_btree_log_ptrs(cur, nbp, 1, 2);
2711
2712 /* Fix up the cursor. */
2713 xfs_btree_setbuf(cur, cur->bc_nlevels, nbp);
2714 cur->bc_ptrs[cur->bc_nlevels] = nptr;
2715 cur->bc_nlevels++;
2716 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2717 *stat = 1;
2718 return 0;
2719 error0:
2720 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2721 return error;
2722 out0:
2723 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2724 *stat = 0;
2725 return 0;
2726 }
2727
2728 STATIC int
2729 xfs_btree_make_block_unfull(
2730 struct xfs_btree_cur *cur, /* btree cursor */
2731 int level, /* btree level */
2732 int numrecs,/* # of recs in block */
2733 int *oindex,/* old tree index */
2734 int *index, /* new tree index */
2735 union xfs_btree_ptr *nptr, /* new btree ptr */
2736 struct xfs_btree_cur **ncur, /* new btree cursor */
2737 union xfs_btree_rec *nrec, /* new record */
2738 int *stat)
2739 {
2740 union xfs_btree_key key; /* new btree key value */
2741 int error = 0;
2742
2743 if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
2744 level == cur->bc_nlevels - 1) {
2745 struct xfs_inode *ip = cur->bc_private.b.ip;
2746
2747 if (numrecs < cur->bc_ops->get_dmaxrecs(cur, level)) {
2748 /* A root block that can be made bigger. */
2749
2750 xfs_iroot_realloc(ip, 1, cur->bc_private.b.whichfork);
2751 } else {
2752 /* A root block that needs replacing */
2753 int logflags = 0;
2754
2755 error = xfs_btree_new_iroot(cur, &logflags, stat);
2756 if (error || *stat == 0)
2757 return error;
2758
2759 xfs_trans_log_inode(cur->bc_tp, ip, logflags);
2760 }
2761
2762 return 0;
2763 }
2764
2765 /* First, try shifting an entry to the right neighbor. */
2766 error = xfs_btree_rshift(cur, level, stat);
2767 if (error || *stat)
2768 return error;
2769
2770 /* Next, try shifting an entry to the left neighbor. */
2771 error = xfs_btree_lshift(cur, level, stat);
2772 if (error)
2773 return error;
2774
2775 if (*stat) {
2776 *oindex = *index = cur->bc_ptrs[level];
2777 return 0;
2778 }
2779
2780 /*
2781 * Next, try splitting the current block in half.
2782 *
2783 * If this works we have to re-set our variables because we
2784 * could be in a different block now.
2785 */
2786 error = xfs_btree_split(cur, level, nptr, &key, ncur, stat);
2787 if (error || *stat == 0)
2788 return error;
2789
2790
2791 *index = cur->bc_ptrs[level];
2792 cur->bc_ops->init_rec_from_key(&key, nrec);
2793 return 0;
2794 }
2795
2796 /*
2797 * Insert one record/level. Return information to the caller
2798 * allowing the next level up to proceed if necessary.
2799 */
2800 STATIC int
2801 xfs_btree_insrec(
2802 struct xfs_btree_cur *cur, /* btree cursor */
2803 int level, /* level to insert record at */
2804 union xfs_btree_ptr *ptrp, /* i/o: block number inserted */
2805 union xfs_btree_rec *recp, /* i/o: record data inserted */
2806 struct xfs_btree_cur **curp, /* output: new cursor replacing cur */
2807 int *stat) /* success/failure */
2808 {
2809 struct xfs_btree_block *block; /* btree block */
2810 struct xfs_buf *bp; /* buffer for block */
2811 union xfs_btree_key key; /* btree key */
2812 union xfs_btree_ptr nptr; /* new block ptr */
2813 struct xfs_btree_cur *ncur; /* new btree cursor */
2814 union xfs_btree_rec nrec; /* new record count */
2815 int optr; /* old key/record index */
2816 int ptr; /* key/record index */
2817 int numrecs;/* number of records */
2818 int error; /* error return value */
2819 #ifdef DEBUG
2820 int i;
2821 #endif
2822
2823 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2824 XFS_BTREE_TRACE_ARGIPR(cur, level, *ptrp, recp);
2825
2826 ncur = NULL;
2827
2828 /*
2829 * If we have an external root pointer, and we've made it to the
2830 * root level, allocate a new root block and we're done.
2831 */
2832 if (!(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
2833 (level >= cur->bc_nlevels)) {
2834 error = xfs_btree_new_root(cur, stat);
2835 xfs_btree_set_ptr_null(cur, ptrp);
2836
2837 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2838 return error;
2839 }
2840
2841 /* If we're off the left edge, return failure. */
2842 ptr = cur->bc_ptrs[level];
2843 if (ptr == 0) {
2844 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2845 *stat = 0;
2846 return 0;
2847 }
2848
2849 /* Make a key out of the record data to be inserted, and save it. */
2850 cur->bc_ops->init_key_from_rec(&key, recp);
2851
2852 optr = ptr;
2853
2854 XFS_BTREE_STATS_INC(cur, insrec);
2855
2856 /* Get pointers to the btree buffer and block. */
2857 block = xfs_btree_get_block(cur, level, &bp);
2858 numrecs = xfs_btree_get_numrecs(block);
2859
2860 #ifdef DEBUG
2861 error = xfs_btree_check_block(cur, block, level, bp);
2862 if (error)
2863 goto error0;
2864
2865 /* Check that the new entry is being inserted in the right place. */
2866 if (ptr <= numrecs) {
2867 if (level == 0) {
2868 ASSERT(cur->bc_ops->recs_inorder(cur, recp,
2869 xfs_btree_rec_addr(cur, ptr, block)));
2870 } else {
2871 ASSERT(cur->bc_ops->keys_inorder(cur, &key,
2872 xfs_btree_key_addr(cur, ptr, block)));
2873 }
2874 }
2875 #endif
2876
2877 /*
2878 * If the block is full, we can't insert the new entry until we
2879 * make the block un-full.
2880 */
2881 xfs_btree_set_ptr_null(cur, &nptr);
2882 if (numrecs == cur->bc_ops->get_maxrecs(cur, level)) {
2883 error = xfs_btree_make_block_unfull(cur, level, numrecs,
2884 &optr, &ptr, &nptr, &ncur, &nrec, stat);
2885 if (error || *stat == 0)
2886 goto error0;
2887 }
2888
2889 /*
2890 * The current block may have changed if the block was
2891 * previously full and we have just made space in it.
2892 */
2893 block = xfs_btree_get_block(cur, level, &bp);
2894 numrecs = xfs_btree_get_numrecs(block);
2895
2896 #ifdef DEBUG
2897 error = xfs_btree_check_block(cur, block, level, bp);
2898 if (error)
2899 return error;
2900 #endif
2901
2902 /*
2903 * At this point we know there's room for our new entry in the block
2904 * we're pointing at.
2905 */
2906 XFS_BTREE_STATS_ADD(cur, moves, numrecs - ptr + 1);
2907
2908 if (level > 0) {
2909 /* It's a nonleaf. make a hole in the keys and ptrs */
2910 union xfs_btree_key *kp;
2911 union xfs_btree_ptr *pp;
2912
2913 kp = xfs_btree_key_addr(cur, ptr, block);
2914 pp = xfs_btree_ptr_addr(cur, ptr, block);
2915
2916 #ifdef DEBUG
2917 for (i = numrecs - ptr; i >= 0; i--) {
2918 error = xfs_btree_check_ptr(cur, pp, i, level);
2919 if (error)
2920 return error;
2921 }
2922 #endif
2923
2924 xfs_btree_shift_keys(cur, kp, 1, numrecs - ptr + 1);
2925 xfs_btree_shift_ptrs(cur, pp, 1, numrecs - ptr + 1);
2926
2927 #ifdef DEBUG
2928 error = xfs_btree_check_ptr(cur, ptrp, 0, level);
2929 if (error)
2930 goto error0;
2931 #endif
2932
2933 /* Now put the new data in, bump numrecs and log it. */
2934 xfs_btree_copy_keys(cur, kp, &key, 1);
2935 xfs_btree_copy_ptrs(cur, pp, ptrp, 1);
2936 numrecs++;
2937 xfs_btree_set_numrecs(block, numrecs);
2938 xfs_btree_log_ptrs(cur, bp, ptr, numrecs);
2939 xfs_btree_log_keys(cur, bp, ptr, numrecs);
2940 #ifdef DEBUG
2941 if (ptr < numrecs) {
2942 ASSERT(cur->bc_ops->keys_inorder(cur, kp,
2943 xfs_btree_key_addr(cur, ptr + 1, block)));
2944 }
2945 #endif
2946 } else {
2947 /* It's a leaf. make a hole in the records */
2948 union xfs_btree_rec *rp;
2949
2950 rp = xfs_btree_rec_addr(cur, ptr, block);
2951
2952 xfs_btree_shift_recs(cur, rp, 1, numrecs - ptr + 1);
2953
2954 /* Now put the new data in, bump numrecs and log it. */
2955 xfs_btree_copy_recs(cur, rp, recp, 1);
2956 xfs_btree_set_numrecs(block, ++numrecs);
2957 xfs_btree_log_recs(cur, bp, ptr, numrecs);
2958 #ifdef DEBUG
2959 if (ptr < numrecs) {
2960 ASSERT(cur->bc_ops->recs_inorder(cur, rp,
2961 xfs_btree_rec_addr(cur, ptr + 1, block)));
2962 }
2963 #endif
2964 }
2965
2966 /* Log the new number of records in the btree header. */
2967 xfs_btree_log_block(cur, bp, XFS_BB_NUMRECS);
2968
2969 /* If we inserted at the start of a block, update the parents' keys. */
2970 if (optr == 1) {
2971 error = xfs_btree_updkey(cur, &key, level + 1);
2972 if (error)
2973 goto error0;
2974 }
2975
2976 /*
2977 * If we are tracking the last record in the tree and
2978 * we are at the far right edge of the tree, update it.
2979 */
2980 if (xfs_btree_is_lastrec(cur, block, level)) {
2981 cur->bc_ops->update_lastrec(cur, block, recp,
2982 ptr, LASTREC_INSREC);
2983 }
2984
2985 /*
2986 * Return the new block number, if any.
2987 * If there is one, give back a record value and a cursor too.
2988 */
2989 *ptrp = nptr;
2990 if (!xfs_btree_ptr_is_null(cur, &nptr)) {
2991 *recp = nrec;
2992 *curp = ncur;
2993 }
2994
2995 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2996 *stat = 1;
2997 return 0;
2998
2999 error0:
3000 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3001 return error;
3002 }
3003
3004 /*
3005 * Insert the record at the point referenced by cur.
3006 *
3007 * A multi-level split of the tree on insert will invalidate the original
3008 * cursor. All callers of this function should assume that the cursor is
3009 * no longer valid and revalidate it.
3010 */
3011 int
3012 xfs_btree_insert(
3013 struct xfs_btree_cur *cur,
3014 int *stat)
3015 {
3016 int error; /* error return value */
3017 int i; /* result value, 0 for failure */
3018 int level; /* current level number in btree */
3019 union xfs_btree_ptr nptr; /* new block number (split result) */
3020 struct xfs_btree_cur *ncur; /* new cursor (split result) */
3021 struct xfs_btree_cur *pcur; /* previous level's cursor */
3022 union xfs_btree_rec rec; /* record to insert */
3023
3024 level = 0;
3025 ncur = NULL;
3026 pcur = cur;
3027
3028 xfs_btree_set_ptr_null(cur, &nptr);
3029 cur->bc_ops->init_rec_from_cur(cur, &rec);
3030
3031 /*
3032 * Loop going up the tree, starting at the leaf level.
3033 * Stop when we don't get a split block, that must mean that
3034 * the insert is finished with this level.
3035 */
3036 do {
3037 /*
3038 * Insert nrec/nptr into this level of the tree.
3039 * Note if we fail, nptr will be null.
3040 */
3041 error = xfs_btree_insrec(pcur, level, &nptr, &rec, &ncur, &i);
3042 if (error) {
3043 if (pcur != cur)
3044 xfs_btree_del_cursor(pcur, XFS_BTREE_ERROR);
3045 goto error0;
3046 }
3047
3048 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
3049 level++;
3050
3051 /*
3052 * See if the cursor we just used is trash.
3053 * Can't trash the caller's cursor, but otherwise we should
3054 * if ncur is a new cursor or we're about to be done.
3055 */
3056 if (pcur != cur &&
3057 (ncur || xfs_btree_ptr_is_null(cur, &nptr))) {
3058 /* Save the state from the cursor before we trash it */
3059 if (cur->bc_ops->update_cursor)
3060 cur->bc_ops->update_cursor(pcur, cur);
3061 cur->bc_nlevels = pcur->bc_nlevels;
3062 xfs_btree_del_cursor(pcur, XFS_BTREE_NOERROR);
3063 }
3064 /* If we got a new cursor, switch to it. */
3065 if (ncur) {
3066 pcur = ncur;
3067 ncur = NULL;
3068 }
3069 } while (!xfs_btree_ptr_is_null(cur, &nptr));
3070
3071 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3072 *stat = i;
3073 return 0;
3074 error0:
3075 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3076 return error;
3077 }
3078
3079 /*
3080 * Try to merge a non-leaf block back into the inode root.
3081 *
3082 * Note: the killroot names comes from the fact that we're effectively
3083 * killing the old root block. But because we can't just delete the
3084 * inode we have to copy the single block it was pointing to into the
3085 * inode.
3086 */
3087 STATIC int
3088 xfs_btree_kill_iroot(
3089 struct xfs_btree_cur *cur)
3090 {
3091 int whichfork = cur->bc_private.b.whichfork;
3092 struct xfs_inode *ip = cur->bc_private.b.ip;
3093 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
3094 struct xfs_btree_block *block;
3095 struct xfs_btree_block *cblock;
3096 union xfs_btree_key *kp;
3097 union xfs_btree_key *ckp;
3098 union xfs_btree_ptr *pp;
3099 union xfs_btree_ptr *cpp;
3100 struct xfs_buf *cbp;
3101 int level;
3102 int index;
3103 int numrecs;
3104 #ifdef DEBUG
3105 union xfs_btree_ptr ptr;
3106 int i;
3107 #endif
3108
3109 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
3110
3111 ASSERT(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE);
3112 ASSERT(cur->bc_nlevels > 1);
3113
3114 /*
3115 * Don't deal with the root block needs to be a leaf case.
3116 * We're just going to turn the thing back into extents anyway.
3117 */
3118 level = cur->bc_nlevels - 1;
3119 if (level == 1)
3120 goto out0;
3121
3122 /*
3123 * Give up if the root has multiple children.
3124 */
3125 block = xfs_btree_get_iroot(cur);
3126 if (xfs_btree_get_numrecs(block) != 1)
3127 goto out0;
3128
3129 cblock = xfs_btree_get_block(cur, level - 1, &cbp);
3130 numrecs = xfs_btree_get_numrecs(cblock);
3131
3132 /*
3133 * Only do this if the next level will fit.
3134 * Then the data must be copied up to the inode,
3135 * instead of freeing the root you free the next level.
3136 */
3137 if (numrecs > cur->bc_ops->get_dmaxrecs(cur, level))
3138 goto out0;
3139
3140 XFS_BTREE_STATS_INC(cur, killroot);
3141
3142 #ifdef DEBUG
3143 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_LEFTSIB);
3144 ASSERT(xfs_btree_ptr_is_null(cur, &ptr));
3145 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
3146 ASSERT(xfs_btree_ptr_is_null(cur, &ptr));
3147 #endif
3148
3149 index = numrecs - cur->bc_ops->get_maxrecs(cur, level);
3150 if (index) {
3151 xfs_iroot_realloc(cur->bc_private.b.ip, index,
3152 cur->bc_private.b.whichfork);
3153 block = ifp->if_broot;
3154 }
3155
3156 be16_add_cpu(&block->bb_numrecs, index);
3157 ASSERT(block->bb_numrecs == cblock->bb_numrecs);
3158
3159 kp = xfs_btree_key_addr(cur, 1, block);
3160 ckp = xfs_btree_key_addr(cur, 1, cblock);
3161 xfs_btree_copy_keys(cur, kp, ckp, numrecs);
3162
3163 pp = xfs_btree_ptr_addr(cur, 1, block);
3164 cpp = xfs_btree_ptr_addr(cur, 1, cblock);
3165 #ifdef DEBUG
3166 for (i = 0; i < numrecs; i++) {
3167 int error;
3168
3169 error = xfs_btree_check_ptr(cur, cpp, i, level - 1);
3170 if (error) {
3171 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3172 return error;
3173 }
3174 }
3175 #endif
3176 xfs_btree_copy_ptrs(cur, pp, cpp, numrecs);
3177
3178 cur->bc_ops->free_block(cur, cbp);
3179 XFS_BTREE_STATS_INC(cur, free);
3180
3181 cur->bc_bufs[level - 1] = NULL;
3182 be16_add_cpu(&block->bb_level, -1);
3183 xfs_trans_log_inode(cur->bc_tp, ip,
3184 XFS_ILOG_CORE | xfs_ilog_fbroot(cur->bc_private.b.whichfork));
3185 cur->bc_nlevels--;
3186 out0:
3187 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3188 return 0;
3189 }
3190
3191 /*
3192 * Kill the current root node, and replace it with it's only child node.
3193 */
3194 STATIC int
3195 xfs_btree_kill_root(
3196 struct xfs_btree_cur *cur,
3197 struct xfs_buf *bp,
3198 int level,
3199 union xfs_btree_ptr *newroot)
3200 {
3201 int error;
3202
3203 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
3204 XFS_BTREE_STATS_INC(cur, killroot);
3205
3206 /*
3207 * Update the root pointer, decreasing the level by 1 and then
3208 * free the old root.
3209 */
3210 cur->bc_ops->set_root(cur, newroot, -1);
3211
3212 error = cur->bc_ops->free_block(cur, bp);
3213 if (error) {
3214 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3215 return error;
3216 }
3217
3218 XFS_BTREE_STATS_INC(cur, free);
3219
3220 cur->bc_bufs[level] = NULL;
3221 cur->bc_ra[level] = 0;
3222 cur->bc_nlevels--;
3223
3224 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3225 return 0;
3226 }
3227
3228 STATIC int
3229 xfs_btree_dec_cursor(
3230 struct xfs_btree_cur *cur,
3231 int level,
3232 int *stat)
3233 {
3234 int error;
3235 int i;
3236
3237 if (level > 0) {
3238 error = xfs_btree_decrement(cur, level, &i);
3239 if (error)
3240 return error;
3241 }
3242
3243 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3244 *stat = 1;
3245 return 0;
3246 }
3247
3248 /*
3249 * Single level of the btree record deletion routine.
3250 * Delete record pointed to by cur/level.
3251 * Remove the record from its block then rebalance the tree.
3252 * Return 0 for error, 1 for done, 2 to go on to the next level.
3253 */
3254 STATIC int /* error */
3255 xfs_btree_delrec(
3256 struct xfs_btree_cur *cur, /* btree cursor */
3257 int level, /* level removing record from */
3258 int *stat) /* fail/done/go-on */
3259 {
3260 struct xfs_btree_block *block; /* btree block */
3261 union xfs_btree_ptr cptr; /* current block ptr */
3262 struct xfs_buf *bp; /* buffer for block */
3263 int error; /* error return value */
3264 int i; /* loop counter */
3265 union xfs_btree_key key; /* storage for keyp */
3266 union xfs_btree_key *keyp = &key; /* passed to the next level */
3267 union xfs_btree_ptr lptr; /* left sibling block ptr */
3268 struct xfs_buf *lbp; /* left buffer pointer */
3269 struct xfs_btree_block *left; /* left btree block */
3270 int lrecs = 0; /* left record count */
3271 int ptr; /* key/record index */
3272 union xfs_btree_ptr rptr; /* right sibling block ptr */
3273 struct xfs_buf *rbp; /* right buffer pointer */
3274 struct xfs_btree_block *right; /* right btree block */
3275 struct xfs_btree_block *rrblock; /* right-right btree block */
3276 struct xfs_buf *rrbp; /* right-right buffer pointer */
3277 int rrecs = 0; /* right record count */
3278 struct xfs_btree_cur *tcur; /* temporary btree cursor */
3279 int numrecs; /* temporary numrec count */
3280
3281 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
3282 XFS_BTREE_TRACE_ARGI(cur, level);
3283
3284 tcur = NULL;
3285
3286 /* Get the index of the entry being deleted, check for nothing there. */
3287 ptr = cur->bc_ptrs[level];
3288 if (ptr == 0) {
3289 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3290 *stat = 0;
3291 return 0;
3292 }
3293
3294 /* Get the buffer & block containing the record or key/ptr. */
3295 block = xfs_btree_get_block(cur, level, &bp);
3296 numrecs = xfs_btree_get_numrecs(block);
3297
3298 #ifdef DEBUG
3299 error = xfs_btree_check_block(cur, block, level, bp);
3300 if (error)
3301 goto error0;
3302 #endif
3303
3304 /* Fail if we're off the end of the block. */
3305 if (ptr > numrecs) {
3306 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3307 *stat = 0;
3308 return 0;
3309 }
3310
3311 XFS_BTREE_STATS_INC(cur, delrec);
3312 XFS_BTREE_STATS_ADD(cur, moves, numrecs - ptr);
3313
3314 /* Excise the entries being deleted. */
3315 if (level > 0) {
3316 /* It's a nonleaf. operate on keys and ptrs */
3317 union xfs_btree_key *lkp;
3318 union xfs_btree_ptr *lpp;
3319
3320 lkp = xfs_btree_key_addr(cur, ptr + 1, block);
3321 lpp = xfs_btree_ptr_addr(cur, ptr + 1, block);
3322
3323 #ifdef DEBUG
3324 for (i = 0; i < numrecs - ptr; i++) {
3325 error = xfs_btree_check_ptr(cur, lpp, i, level);
3326 if (error)
3327 goto error0;
3328 }
3329 #endif
3330
3331 if (ptr < numrecs) {
3332 xfs_btree_shift_keys(cur, lkp, -1, numrecs - ptr);
3333 xfs_btree_shift_ptrs(cur, lpp, -1, numrecs - ptr);
3334 xfs_btree_log_keys(cur, bp, ptr, numrecs - 1);
3335 xfs_btree_log_ptrs(cur, bp, ptr, numrecs - 1);
3336 }
3337
3338 /*
3339 * If it's the first record in the block, we'll need to pass a
3340 * key up to the next level (updkey).
3341 */
3342 if (ptr == 1)
3343 keyp = xfs_btree_key_addr(cur, 1, block);
3344 } else {
3345 /* It's a leaf. operate on records */
3346 if (ptr < numrecs) {
3347 xfs_btree_shift_recs(cur,
3348 xfs_btree_rec_addr(cur, ptr + 1, block),
3349 -1, numrecs - ptr);
3350 xfs_btree_log_recs(cur, bp, ptr, numrecs - 1);
3351 }
3352
3353 /*
3354 * If it's the first record in the block, we'll need a key
3355 * structure to pass up to the next level (updkey).
3356 */
3357 if (ptr == 1) {
3358 cur->bc_ops->init_key_from_rec(&key,
3359 xfs_btree_rec_addr(cur, 1, block));
3360 keyp = &key;
3361 }
3362 }
3363
3364 /*
3365 * Decrement and log the number of entries in the block.
3366 */
3367 xfs_btree_set_numrecs(block, --numrecs);
3368 xfs_btree_log_block(cur, bp, XFS_BB_NUMRECS);
3369
3370 /*
3371 * If we are tracking the last record in the tree and
3372 * we are at the far right edge of the tree, update it.
3373 */
3374 if (xfs_btree_is_lastrec(cur, block, level)) {
3375 cur->bc_ops->update_lastrec(cur, block, NULL,
3376 ptr, LASTREC_DELREC);
3377 }
3378
3379 /*
3380 * We're at the root level. First, shrink the root block in-memory.
3381 * Try to get rid of the next level down. If we can't then there's
3382 * nothing left to do.
3383 */
3384 if (level == cur->bc_nlevels - 1) {
3385 if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) {
3386 xfs_iroot_realloc(cur->bc_private.b.ip, -1,
3387 cur->bc_private.b.whichfork);
3388
3389 error = xfs_btree_kill_iroot(cur);
3390 if (error)
3391 goto error0;
3392
3393 error = xfs_btree_dec_cursor(cur, level, stat);
3394 if (error)
3395 goto error0;
3396 *stat = 1;
3397 return 0;
3398 }
3399
3400 /*
3401 * If this is the root level, and there's only one entry left,
3402 * and it's NOT the leaf level, then we can get rid of this
3403 * level.
3404 */
3405 if (numrecs == 1 && level > 0) {
3406 union xfs_btree_ptr *pp;
3407 /*
3408 * pp is still set to the first pointer in the block.
3409 * Make it the new root of the btree.
3410 */
3411 pp = xfs_btree_ptr_addr(cur, 1, block);
3412 error = xfs_btree_kill_root(cur, bp, level, pp);
3413 if (error)
3414 goto error0;
3415 } else if (level > 0) {
3416 error = xfs_btree_dec_cursor(cur, level, stat);
3417 if (error)
3418 goto error0;
3419 }
3420 *stat = 1;
3421 return 0;
3422 }
3423
3424 /*
3425 * If we deleted the leftmost entry in the block, update the
3426 * key values above us in the tree.
3427 */
3428 if (ptr == 1) {
3429 error = xfs_btree_updkey(cur, keyp, level + 1);
3430 if (error)
3431 goto error0;
3432 }
3433
3434 /*
3435 * If the number of records remaining in the block is at least
3436 * the minimum, we're done.
3437 */
3438 if (numrecs >= cur->bc_ops->get_minrecs(cur, level)) {
3439 error = xfs_btree_dec_cursor(cur, level, stat);
3440 if (error)
3441 goto error0;
3442 return 0;
3443 }
3444
3445 /*
3446 * Otherwise, we have to move some records around to keep the
3447 * tree balanced. Look at the left and right sibling blocks to
3448 * see if we can re-balance by moving only one record.
3449 */
3450 xfs_btree_get_sibling(cur, block, &rptr, XFS_BB_RIGHTSIB);
3451 xfs_btree_get_sibling(cur, block, &lptr, XFS_BB_LEFTSIB);
3452
3453 if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) {
3454 /*
3455 * One child of root, need to get a chance to copy its contents
3456 * into the root and delete it. Can't go up to next level,
3457 * there's nothing to delete there.
3458 */
3459 if (xfs_btree_ptr_is_null(cur, &rptr) &&
3460 xfs_btree_ptr_is_null(cur, &lptr) &&
3461 level == cur->bc_nlevels - 2) {
3462 error = xfs_btree_kill_iroot(cur);
3463 if (!error)
3464 error = xfs_btree_dec_cursor(cur, level, stat);
3465 if (error)
3466 goto error0;
3467 return 0;
3468 }
3469 }
3470
3471 ASSERT(!xfs_btree_ptr_is_null(cur, &rptr) ||
3472 !xfs_btree_ptr_is_null(cur, &lptr));
3473
3474 /*
3475 * Duplicate the cursor so our btree manipulations here won't
3476 * disrupt the next level up.
3477 */
3478 error = xfs_btree_dup_cursor(cur, &tcur);
3479 if (error)
3480 goto error0;
3481
3482 /*
3483 * If there's a right sibling, see if it's ok to shift an entry
3484 * out of it.
3485 */
3486 if (!xfs_btree_ptr_is_null(cur, &rptr)) {
3487 /*
3488 * Move the temp cursor to the last entry in the next block.
3489 * Actually any entry but the first would suffice.
3490 */
3491 i = xfs_btree_lastrec(tcur, level);
3492 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
3493
3494 error = xfs_btree_increment(tcur, level, &i);
3495 if (error)
3496 goto error0;
3497 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
3498
3499 i = xfs_btree_lastrec(tcur, level);
3500 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
3501
3502 /* Grab a pointer to the block. */
3503 right = xfs_btree_get_block(tcur, level, &rbp);
3504 #ifdef DEBUG
3505 error = xfs_btree_check_block(tcur, right, level, rbp);
3506 if (error)
3507 goto error0;
3508 #endif
3509 /* Grab the current block number, for future use. */
3510 xfs_btree_get_sibling(tcur, right, &cptr, XFS_BB_LEFTSIB);
3511
3512 /*
3513 * If right block is full enough so that removing one entry
3514 * won't make it too empty, and left-shifting an entry out
3515 * of right to us works, we're done.
3516 */
3517 if (xfs_btree_get_numrecs(right) - 1 >=
3518 cur->bc_ops->get_minrecs(tcur, level)) {
3519 error = xfs_btree_lshift(tcur, level, &i);
3520 if (error)
3521 goto error0;
3522 if (i) {
3523 ASSERT(xfs_btree_get_numrecs(block) >=
3524 cur->bc_ops->get_minrecs(tcur, level));
3525
3526 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
3527 tcur = NULL;
3528
3529 error = xfs_btree_dec_cursor(cur, level, stat);
3530 if (error)
3531 goto error0;
3532 return 0;
3533 }
3534 }
3535
3536 /*
3537 * Otherwise, grab the number of records in right for
3538 * future reference, and fix up the temp cursor to point
3539 * to our block again (last record).
3540 */
3541 rrecs = xfs_btree_get_numrecs(right);
3542 if (!xfs_btree_ptr_is_null(cur, &lptr)) {
3543 i = xfs_btree_firstrec(tcur, level);
3544 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
3545
3546 error = xfs_btree_decrement(tcur, level, &i);
3547 if (error)
3548 goto error0;
3549 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
3550 }
3551 }
3552
3553 /*
3554 * If there's a left sibling, see if it's ok to shift an entry
3555 * out of it.
3556 */
3557 if (!xfs_btree_ptr_is_null(cur, &lptr)) {
3558 /*
3559 * Move the temp cursor to the first entry in the
3560 * previous block.
3561 */
3562 i = xfs_btree_firstrec(tcur, level);
3563 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
3564
3565 error = xfs_btree_decrement(tcur, level, &i);
3566 if (error)
3567 goto error0;
3568 i = xfs_btree_firstrec(tcur, level);
3569 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
3570
3571 /* Grab a pointer to the block. */
3572 left = xfs_btree_get_block(tcur, level, &lbp);
3573 #ifdef DEBUG
3574 error = xfs_btree_check_block(cur, left, level, lbp);
3575 if (error)
3576 goto error0;
3577 #endif
3578 /* Grab the current block number, for future use. */
3579 xfs_btree_get_sibling(tcur, left, &cptr, XFS_BB_RIGHTSIB);
3580
3581 /*
3582 * If left block is full enough so that removing one entry
3583 * won't make it too empty, and right-shifting an entry out
3584 * of left to us works, we're done.
3585 */
3586 if (xfs_btree_get_numrecs(left) - 1 >=
3587 cur->bc_ops->get_minrecs(tcur, level)) {
3588 error = xfs_btree_rshift(tcur, level, &i);
3589 if (error)
3590 goto error0;
3591 if (i) {
3592 ASSERT(xfs_btree_get_numrecs(block) >=
3593 cur->bc_ops->get_minrecs(tcur, level));
3594 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
3595 tcur = NULL;
3596 if (level == 0)
3597 cur->bc_ptrs[0]++;
3598 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3599 *stat = 1;
3600 return 0;
3601 }
3602 }
3603
3604 /*
3605 * Otherwise, grab the number of records in right for
3606 * future reference.
3607 */
3608 lrecs = xfs_btree_get_numrecs(left);
3609 }
3610
3611 /* Delete the temp cursor, we're done with it. */
3612 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
3613 tcur = NULL;
3614
3615 /* If here, we need to do a join to keep the tree balanced. */
3616 ASSERT(!xfs_btree_ptr_is_null(cur, &cptr));
3617
3618 if (!xfs_btree_ptr_is_null(cur, &lptr) &&
3619 lrecs + xfs_btree_get_numrecs(block) <=
3620 cur->bc_ops->get_maxrecs(cur, level)) {
3621 /*
3622 * Set "right" to be the starting block,
3623 * "left" to be the left neighbor.
3624 */
3625 rptr = cptr;
3626 right = block;
3627 rbp = bp;
3628 error = xfs_btree_read_buf_block(cur, &lptr, level,
3629 0, &left, &lbp);
3630 if (error)
3631 goto error0;
3632
3633 /*
3634 * If that won't work, see if we can join with the right neighbor block.
3635 */
3636 } else if (!xfs_btree_ptr_is_null(cur, &rptr) &&
3637 rrecs + xfs_btree_get_numrecs(block) <=
3638 cur->bc_ops->get_maxrecs(cur, level)) {
3639 /*
3640 * Set "left" to be the starting block,
3641 * "right" to be the right neighbor.
3642 */
3643 lptr = cptr;
3644 left = block;
3645 lbp = bp;
3646 error = xfs_btree_read_buf_block(cur, &rptr, level,
3647 0, &right, &rbp);
3648 if (error)
3649 goto error0;
3650
3651 /*
3652 * Otherwise, we can't fix the imbalance.
3653 * Just return. This is probably a logic error, but it's not fatal.
3654 */
3655 } else {
3656 error = xfs_btree_dec_cursor(cur, level, stat);
3657 if (error)
3658 goto error0;
3659 return 0;
3660 }
3661
3662 rrecs = xfs_btree_get_numrecs(right);
3663 lrecs = xfs_btree_get_numrecs(left);
3664
3665 /*
3666 * We're now going to join "left" and "right" by moving all the stuff
3667 * in "right" to "left" and deleting "right".
3668 */
3669 XFS_BTREE_STATS_ADD(cur, moves, rrecs);
3670 if (level > 0) {
3671 /* It's a non-leaf. Move keys and pointers. */
3672 union xfs_btree_key *lkp; /* left btree key */
3673 union xfs_btree_ptr *lpp; /* left address pointer */
3674 union xfs_btree_key *rkp; /* right btree key */
3675 union xfs_btree_ptr *rpp; /* right address pointer */
3676
3677 lkp = xfs_btree_key_addr(cur, lrecs + 1, left);
3678 lpp = xfs_btree_ptr_addr(cur, lrecs + 1, left);
3679 rkp = xfs_btree_key_addr(cur, 1, right);
3680 rpp = xfs_btree_ptr_addr(cur, 1, right);
3681 #ifdef DEBUG
3682 for (i = 1; i < rrecs; i++) {
3683 error = xfs_btree_check_ptr(cur, rpp, i, level);
3684 if (error)
3685 goto error0;
3686 }
3687 #endif
3688 xfs_btree_copy_keys(cur, lkp, rkp, rrecs);
3689 xfs_btree_copy_ptrs(cur, lpp, rpp, rrecs);
3690
3691 xfs_btree_log_keys(cur, lbp, lrecs + 1, lrecs + rrecs);
3692 xfs_btree_log_ptrs(cur, lbp, lrecs + 1, lrecs + rrecs);
3693 } else {
3694 /* It's a leaf. Move records. */
3695 union xfs_btree_rec *lrp; /* left record pointer */
3696 union xfs_btree_rec *rrp; /* right record pointer */
3697
3698 lrp = xfs_btree_rec_addr(cur, lrecs + 1, left);
3699 rrp = xfs_btree_rec_addr(cur, 1, right);
3700
3701 xfs_btree_copy_recs(cur, lrp, rrp, rrecs);
3702 xfs_btree_log_recs(cur, lbp, lrecs + 1, lrecs + rrecs);
3703 }
3704
3705 XFS_BTREE_STATS_INC(cur, join);
3706
3707 /*
3708 * Fix up the number of records and right block pointer in the
3709 * surviving block, and log it.
3710 */
3711 xfs_btree_set_numrecs(left, lrecs + rrecs);
3712 xfs_btree_get_sibling(cur, right, &cptr, XFS_BB_RIGHTSIB),
3713 xfs_btree_set_sibling(cur, left, &cptr, XFS_BB_RIGHTSIB);
3714 xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS | XFS_BB_RIGHTSIB);
3715
3716 /* If there is a right sibling, point it to the remaining block. */
3717 xfs_btree_get_sibling(cur, left, &cptr, XFS_BB_RIGHTSIB);
3718 if (!xfs_btree_ptr_is_null(cur, &cptr)) {
3719 error = xfs_btree_read_buf_block(cur, &cptr, level,
3720 0, &rrblock, &rrbp);
3721 if (error)
3722 goto error0;
3723 xfs_btree_set_sibling(cur, rrblock, &lptr, XFS_BB_LEFTSIB);
3724 xfs_btree_log_block(cur, rrbp, XFS_BB_LEFTSIB);
3725 }
3726
3727 /* Free the deleted block. */
3728 error = cur->bc_ops->free_block(cur, rbp);
3729 if (error)
3730 goto error0;
3731 XFS_BTREE_STATS_INC(cur, free);
3732
3733 /*
3734 * If we joined with the left neighbor, set the buffer in the
3735 * cursor to the left block, and fix up the index.
3736 */
3737 if (bp != lbp) {
3738 cur->bc_bufs[level] = lbp;
3739 cur->bc_ptrs[level] += lrecs;
3740 cur->bc_ra[level] = 0;
3741 }
3742 /*
3743 * If we joined with the right neighbor and there's a level above
3744 * us, increment the cursor at that level.
3745 */
3746 else if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) ||
3747 (level + 1 < cur->bc_nlevels)) {
3748 error = xfs_btree_increment(cur, level + 1, &i);
3749 if (error)
3750 goto error0;
3751 }
3752
3753 /*
3754 * Readjust the ptr at this level if it's not a leaf, since it's
3755 * still pointing at the deletion point, which makes the cursor
3756 * inconsistent. If this makes the ptr 0, the caller fixes it up.
3757 * We can't use decrement because it would change the next level up.
3758 */
3759 if (level > 0)
3760 cur->bc_ptrs[level]--;
3761
3762 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3763 /* Return value means the next level up has something to do. */
3764 *stat = 2;
3765 return 0;
3766
3767 error0:
3768 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3769 if (tcur)
3770 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
3771 return error;
3772 }
3773
3774 /*
3775 * Delete the record pointed to by cur.
3776 * The cursor refers to the place where the record was (could be inserted)
3777 * when the operation returns.
3778 */
3779 int /* error */
3780 xfs_btree_delete(
3781 struct xfs_btree_cur *cur,
3782 int *stat) /* success/failure */
3783 {
3784 int error; /* error return value */
3785 int level;
3786 int i;
3787
3788 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
3789
3790 /*
3791 * Go up the tree, starting at leaf level.
3792 *
3793 * If 2 is returned then a join was done; go to the next level.
3794 * Otherwise we are done.
3795 */
3796 for (level = 0, i = 2; i == 2; level++) {
3797 error = xfs_btree_delrec(cur, level, &i);
3798 if (error)
3799 goto error0;
3800 }
3801
3802 if (i == 0) {
3803 for (level = 1; level < cur->bc_nlevels; level++) {
3804 if (cur->bc_ptrs[level] == 0) {
3805 error = xfs_btree_decrement(cur, level, &i);
3806 if (error)
3807 goto error0;
3808 break;
3809 }
3810 }
3811 }
3812
3813 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3814 *stat = i;
3815 return 0;
3816 error0:
3817 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3818 return error;
3819 }
3820
3821 /*
3822 * Get the data from the pointed-to record.
3823 */
3824 int /* error */
3825 xfs_btree_get_rec(
3826 struct xfs_btree_cur *cur, /* btree cursor */
3827 union xfs_btree_rec **recp, /* output: btree record */
3828 int *stat) /* output: success/failure */
3829 {
3830 struct xfs_btree_block *block; /* btree block */
3831 struct xfs_buf *bp; /* buffer pointer */
3832 int ptr; /* record number */
3833 #ifdef DEBUG
3834 int error; /* error return value */
3835 #endif
3836
3837 ptr = cur->bc_ptrs[0];
3838 block = xfs_btree_get_block(cur, 0, &bp);
3839
3840 #ifdef DEBUG
3841 error = xfs_btree_check_block(cur, block, 0, bp);
3842 if (error)
3843 return error;
3844 #endif
3845
3846 /*
3847 * Off the right end or left end, return failure.
3848 */
3849 if (ptr > xfs_btree_get_numrecs(block) || ptr <= 0) {
3850 *stat = 0;
3851 return 0;
3852 }
3853
3854 /*
3855 * Point to the record and extract its data.
3856 */
3857 *recp = xfs_btree_rec_addr(cur, ptr, block);
3858 *stat = 1;
3859 return 0;
3860 }