Merge tag 'sound-fix-4.4-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai...
[deliverable/linux.git] / fs / xfs / libxfs / 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_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
24 #include "xfs_bit.h"
25 #include "xfs_mount.h"
26 #include "xfs_inode.h"
27 #include "xfs_trans.h"
28 #include "xfs_inode_item.h"
29 #include "xfs_buf_item.h"
30 #include "xfs_btree.h"
31 #include "xfs_error.h"
32 #include "xfs_trace.h"
33 #include "xfs_cksum.h"
34 #include "xfs_alloc.h"
35 #include "xfs_log.h"
36
37 /*
38 * Cursor allocation zone.
39 */
40 kmem_zone_t *xfs_btree_cur_zone;
41
42 /*
43 * Btree magic numbers.
44 */
45 static const __uint32_t xfs_magics[2][XFS_BTNUM_MAX] = {
46 { XFS_ABTB_MAGIC, XFS_ABTC_MAGIC, XFS_BMAP_MAGIC, XFS_IBT_MAGIC,
47 XFS_FIBT_MAGIC },
48 { XFS_ABTB_CRC_MAGIC, XFS_ABTC_CRC_MAGIC,
49 XFS_BMAP_CRC_MAGIC, XFS_IBT_CRC_MAGIC, XFS_FIBT_CRC_MAGIC }
50 };
51 #define xfs_btree_magic(cur) \
52 xfs_magics[!!((cur)->bc_flags & XFS_BTREE_CRC_BLOCKS)][cur->bc_btnum]
53
54
55 STATIC int /* error (0 or EFSCORRUPTED) */
56 xfs_btree_check_lblock(
57 struct xfs_btree_cur *cur, /* btree cursor */
58 struct xfs_btree_block *block, /* btree long form block pointer */
59 int level, /* level of the btree block */
60 struct xfs_buf *bp) /* buffer for block, if any */
61 {
62 int lblock_ok = 1; /* block passes checks */
63 struct xfs_mount *mp; /* file system mount point */
64
65 mp = cur->bc_mp;
66
67 if (xfs_sb_version_hascrc(&mp->m_sb)) {
68 lblock_ok = lblock_ok &&
69 uuid_equal(&block->bb_u.l.bb_uuid,
70 &mp->m_sb.sb_meta_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(NULLFSBLOCK) ||
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(NULLFSBLOCK) ||
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 -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,
121 &mp->m_sb.sb_meta_uuid) &&
122 block->bb_u.s.bb_blkno == cpu_to_be64(
123 bp ? bp->b_bn : XFS_BUF_DADDR_NULL);
124 }
125
126 sblock_ok = sblock_ok &&
127 be32_to_cpu(block->bb_magic) == xfs_btree_magic(cur) &&
128 be16_to_cpu(block->bb_level) == level &&
129 be16_to_cpu(block->bb_numrecs) <=
130 cur->bc_ops->get_maxrecs(cur, level) &&
131 (block->bb_u.s.bb_leftsib == cpu_to_be32(NULLAGBLOCK) ||
132 be32_to_cpu(block->bb_u.s.bb_leftsib) < agflen) &&
133 block->bb_u.s.bb_leftsib &&
134 (block->bb_u.s.bb_rightsib == cpu_to_be32(NULLAGBLOCK) ||
135 be32_to_cpu(block->bb_u.s.bb_rightsib) < agflen) &&
136 block->bb_u.s.bb_rightsib;
137
138 if (unlikely(XFS_TEST_ERROR(!sblock_ok, mp,
139 XFS_ERRTAG_BTREE_CHECK_SBLOCK,
140 XFS_RANDOM_BTREE_CHECK_SBLOCK))) {
141 if (bp)
142 trace_xfs_btree_corrupt(bp, _RET_IP_);
143 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp);
144 return -EFSCORRUPTED;
145 }
146 return 0;
147 }
148
149 /*
150 * Debug routine: check that block header is ok.
151 */
152 int
153 xfs_btree_check_block(
154 struct xfs_btree_cur *cur, /* btree cursor */
155 struct xfs_btree_block *block, /* generic btree block pointer */
156 int level, /* level of the btree block */
157 struct xfs_buf *bp) /* buffer containing block, if any */
158 {
159 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
160 return xfs_btree_check_lblock(cur, block, level, bp);
161 else
162 return xfs_btree_check_sblock(cur, block, level, bp);
163 }
164
165 /*
166 * Check that (long) pointer is ok.
167 */
168 int /* error (0 or EFSCORRUPTED) */
169 xfs_btree_check_lptr(
170 struct xfs_btree_cur *cur, /* btree cursor */
171 xfs_fsblock_t bno, /* btree block disk address */
172 int level) /* btree block level */
173 {
174 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp,
175 level > 0 &&
176 bno != NULLFSBLOCK &&
177 XFS_FSB_SANITY_CHECK(cur->bc_mp, bno));
178 return 0;
179 }
180
181 #ifdef DEBUG
182 /*
183 * Check that (short) pointer is ok.
184 */
185 STATIC int /* error (0 or EFSCORRUPTED) */
186 xfs_btree_check_sptr(
187 struct xfs_btree_cur *cur, /* btree cursor */
188 xfs_agblock_t bno, /* btree block disk address */
189 int level) /* btree block level */
190 {
191 xfs_agblock_t agblocks = cur->bc_mp->m_sb.sb_agblocks;
192
193 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp,
194 level > 0 &&
195 bno != NULLAGBLOCK &&
196 bno != 0 &&
197 bno < agblocks);
198 return 0;
199 }
200
201 /*
202 * Check that block ptr is ok.
203 */
204 STATIC int /* error (0 or EFSCORRUPTED) */
205 xfs_btree_check_ptr(
206 struct xfs_btree_cur *cur, /* btree cursor */
207 union xfs_btree_ptr *ptr, /* btree block disk address */
208 int index, /* offset from ptr to check */
209 int level) /* btree block level */
210 {
211 if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
212 return xfs_btree_check_lptr(cur,
213 be64_to_cpu((&ptr->l)[index]), level);
214 } else {
215 return xfs_btree_check_sptr(cur,
216 be32_to_cpu((&ptr->s)[index]), level);
217 }
218 }
219 #endif
220
221 /*
222 * Calculate CRC on the whole btree block and stuff it into the
223 * long-form btree header.
224 *
225 * Prior to calculting the CRC, pull the LSN out of the buffer log item and put
226 * it into the buffer so recovery knows what the last modification was that made
227 * it to disk.
228 */
229 void
230 xfs_btree_lblock_calc_crc(
231 struct xfs_buf *bp)
232 {
233 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
234 struct xfs_buf_log_item *bip = bp->b_fspriv;
235
236 if (!xfs_sb_version_hascrc(&bp->b_target->bt_mount->m_sb))
237 return;
238 if (bip)
239 block->bb_u.l.bb_lsn = cpu_to_be64(bip->bli_item.li_lsn);
240 xfs_buf_update_cksum(bp, XFS_BTREE_LBLOCK_CRC_OFF);
241 }
242
243 bool
244 xfs_btree_lblock_verify_crc(
245 struct xfs_buf *bp)
246 {
247 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
248 struct xfs_mount *mp = bp->b_target->bt_mount;
249
250 if (xfs_sb_version_hascrc(&mp->m_sb)) {
251 if (!xfs_log_check_lsn(mp, be64_to_cpu(block->bb_u.l.bb_lsn)))
252 return false;
253 return xfs_buf_verify_cksum(bp, XFS_BTREE_LBLOCK_CRC_OFF);
254 }
255
256 return true;
257 }
258
259 /*
260 * Calculate CRC on the whole btree block and stuff it into the
261 * short-form btree header.
262 *
263 * Prior to calculting the CRC, pull the LSN out of the buffer log item and put
264 * it into the buffer so recovery knows what the last modification was that made
265 * it to disk.
266 */
267 void
268 xfs_btree_sblock_calc_crc(
269 struct xfs_buf *bp)
270 {
271 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
272 struct xfs_buf_log_item *bip = bp->b_fspriv;
273
274 if (!xfs_sb_version_hascrc(&bp->b_target->bt_mount->m_sb))
275 return;
276 if (bip)
277 block->bb_u.s.bb_lsn = cpu_to_be64(bip->bli_item.li_lsn);
278 xfs_buf_update_cksum(bp, XFS_BTREE_SBLOCK_CRC_OFF);
279 }
280
281 bool
282 xfs_btree_sblock_verify_crc(
283 struct xfs_buf *bp)
284 {
285 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
286 struct xfs_mount *mp = bp->b_target->bt_mount;
287
288 if (xfs_sb_version_hascrc(&mp->m_sb)) {
289 if (!xfs_log_check_lsn(mp, be64_to_cpu(block->bb_u.s.bb_lsn)))
290 return false;
291 return xfs_buf_verify_cksum(bp, XFS_BTREE_SBLOCK_CRC_OFF);
292 }
293
294 return true;
295 }
296
297 /*
298 * Delete the btree cursor.
299 */
300 void
301 xfs_btree_del_cursor(
302 xfs_btree_cur_t *cur, /* btree cursor */
303 int error) /* del because of error */
304 {
305 int i; /* btree level */
306
307 /*
308 * Clear the buffer pointers, and release the buffers.
309 * If we're doing this in the face of an error, we
310 * need to make sure to inspect all of the entries
311 * in the bc_bufs array for buffers to be unlocked.
312 * This is because some of the btree code works from
313 * level n down to 0, and if we get an error along
314 * the way we won't have initialized all the entries
315 * down to 0.
316 */
317 for (i = 0; i < cur->bc_nlevels; i++) {
318 if (cur->bc_bufs[i])
319 xfs_trans_brelse(cur->bc_tp, cur->bc_bufs[i]);
320 else if (!error)
321 break;
322 }
323 /*
324 * Can't free a bmap cursor without having dealt with the
325 * allocated indirect blocks' accounting.
326 */
327 ASSERT(cur->bc_btnum != XFS_BTNUM_BMAP ||
328 cur->bc_private.b.allocated == 0);
329 /*
330 * Free the cursor.
331 */
332 kmem_zone_free(xfs_btree_cur_zone, cur);
333 }
334
335 /*
336 * Duplicate the btree cursor.
337 * Allocate a new one, copy the record, re-get the buffers.
338 */
339 int /* error */
340 xfs_btree_dup_cursor(
341 xfs_btree_cur_t *cur, /* input cursor */
342 xfs_btree_cur_t **ncur) /* output cursor */
343 {
344 xfs_buf_t *bp; /* btree block's buffer pointer */
345 int error; /* error return value */
346 int i; /* level number of btree block */
347 xfs_mount_t *mp; /* mount structure for filesystem */
348 xfs_btree_cur_t *new; /* new cursor value */
349 xfs_trans_t *tp; /* transaction pointer, can be NULL */
350
351 tp = cur->bc_tp;
352 mp = cur->bc_mp;
353
354 /*
355 * Allocate a new cursor like the old one.
356 */
357 new = cur->bc_ops->dup_cursor(cur);
358
359 /*
360 * Copy the record currently in the cursor.
361 */
362 new->bc_rec = cur->bc_rec;
363
364 /*
365 * For each level current, re-get the buffer and copy the ptr value.
366 */
367 for (i = 0; i < new->bc_nlevels; i++) {
368 new->bc_ptrs[i] = cur->bc_ptrs[i];
369 new->bc_ra[i] = cur->bc_ra[i];
370 bp = cur->bc_bufs[i];
371 if (bp) {
372 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
373 XFS_BUF_ADDR(bp), mp->m_bsize,
374 0, &bp,
375 cur->bc_ops->buf_ops);
376 if (error) {
377 xfs_btree_del_cursor(new, error);
378 *ncur = NULL;
379 return error;
380 }
381 }
382 new->bc_bufs[i] = bp;
383 }
384 *ncur = new;
385 return 0;
386 }
387
388 /*
389 * XFS btree block layout and addressing:
390 *
391 * There are two types of blocks in the btree: leaf and non-leaf blocks.
392 *
393 * The leaf record start with a header then followed by records containing
394 * the values. A non-leaf block also starts with the same header, and
395 * then first contains lookup keys followed by an equal number of pointers
396 * to the btree blocks at the previous level.
397 *
398 * +--------+-------+-------+-------+-------+-------+-------+
399 * Leaf: | header | rec 1 | rec 2 | rec 3 | rec 4 | rec 5 | rec N |
400 * +--------+-------+-------+-------+-------+-------+-------+
401 *
402 * +--------+-------+-------+-------+-------+-------+-------+
403 * Non-Leaf: | header | key 1 | key 2 | key N | ptr 1 | ptr 2 | ptr N |
404 * +--------+-------+-------+-------+-------+-------+-------+
405 *
406 * The header is called struct xfs_btree_block for reasons better left unknown
407 * and comes in different versions for short (32bit) and long (64bit) block
408 * pointers. The record and key structures are defined by the btree instances
409 * and opaque to the btree core. The block pointers are simple disk endian
410 * integers, available in a short (32bit) and long (64bit) variant.
411 *
412 * The helpers below calculate the offset of a given record, key or pointer
413 * into a btree block (xfs_btree_*_offset) or return a pointer to the given
414 * record, key or pointer (xfs_btree_*_addr). Note that all addressing
415 * inside the btree block is done using indices starting at one, not zero!
416 */
417
418 /*
419 * Return size of the btree block header for this btree instance.
420 */
421 static inline size_t xfs_btree_block_len(struct xfs_btree_cur *cur)
422 {
423 if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
424 if (cur->bc_flags & XFS_BTREE_CRC_BLOCKS)
425 return XFS_BTREE_LBLOCK_CRC_LEN;
426 return XFS_BTREE_LBLOCK_LEN;
427 }
428 if (cur->bc_flags & XFS_BTREE_CRC_BLOCKS)
429 return XFS_BTREE_SBLOCK_CRC_LEN;
430 return XFS_BTREE_SBLOCK_LEN;
431 }
432
433 /*
434 * Return size of btree block pointers for this btree instance.
435 */
436 static inline size_t xfs_btree_ptr_len(struct xfs_btree_cur *cur)
437 {
438 return (cur->bc_flags & XFS_BTREE_LONG_PTRS) ?
439 sizeof(__be64) : sizeof(__be32);
440 }
441
442 /*
443 * Calculate offset of the n-th record in a btree block.
444 */
445 STATIC size_t
446 xfs_btree_rec_offset(
447 struct xfs_btree_cur *cur,
448 int n)
449 {
450 return xfs_btree_block_len(cur) +
451 (n - 1) * cur->bc_ops->rec_len;
452 }
453
454 /*
455 * Calculate offset of the n-th key in a btree block.
456 */
457 STATIC size_t
458 xfs_btree_key_offset(
459 struct xfs_btree_cur *cur,
460 int n)
461 {
462 return xfs_btree_block_len(cur) +
463 (n - 1) * cur->bc_ops->key_len;
464 }
465
466 /*
467 * Calculate offset of the n-th block pointer in a btree block.
468 */
469 STATIC size_t
470 xfs_btree_ptr_offset(
471 struct xfs_btree_cur *cur,
472 int n,
473 int level)
474 {
475 return xfs_btree_block_len(cur) +
476 cur->bc_ops->get_maxrecs(cur, level) * cur->bc_ops->key_len +
477 (n - 1) * xfs_btree_ptr_len(cur);
478 }
479
480 /*
481 * Return a pointer to the n-th record in the btree block.
482 */
483 STATIC union xfs_btree_rec *
484 xfs_btree_rec_addr(
485 struct xfs_btree_cur *cur,
486 int n,
487 struct xfs_btree_block *block)
488 {
489 return (union xfs_btree_rec *)
490 ((char *)block + xfs_btree_rec_offset(cur, n));
491 }
492
493 /*
494 * Return a pointer to the n-th key in the btree block.
495 */
496 STATIC union xfs_btree_key *
497 xfs_btree_key_addr(
498 struct xfs_btree_cur *cur,
499 int n,
500 struct xfs_btree_block *block)
501 {
502 return (union xfs_btree_key *)
503 ((char *)block + xfs_btree_key_offset(cur, n));
504 }
505
506 /*
507 * Return a pointer to the n-th block pointer in the btree block.
508 */
509 STATIC union xfs_btree_ptr *
510 xfs_btree_ptr_addr(
511 struct xfs_btree_cur *cur,
512 int n,
513 struct xfs_btree_block *block)
514 {
515 int level = xfs_btree_get_level(block);
516
517 ASSERT(block->bb_level != 0);
518
519 return (union xfs_btree_ptr *)
520 ((char *)block + xfs_btree_ptr_offset(cur, n, level));
521 }
522
523 /*
524 * Get the root block which is stored in the inode.
525 *
526 * For now this btree implementation assumes the btree root is always
527 * stored in the if_broot field of an inode fork.
528 */
529 STATIC struct xfs_btree_block *
530 xfs_btree_get_iroot(
531 struct xfs_btree_cur *cur)
532 {
533 struct xfs_ifork *ifp;
534
535 ifp = XFS_IFORK_PTR(cur->bc_private.b.ip, cur->bc_private.b.whichfork);
536 return (struct xfs_btree_block *)ifp->if_broot;
537 }
538
539 /*
540 * Retrieve the block pointer from the cursor at the given level.
541 * This may be an inode btree root or from a buffer.
542 */
543 STATIC struct xfs_btree_block * /* generic btree block pointer */
544 xfs_btree_get_block(
545 struct xfs_btree_cur *cur, /* btree cursor */
546 int level, /* level in btree */
547 struct xfs_buf **bpp) /* buffer containing the block */
548 {
549 if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
550 (level == cur->bc_nlevels - 1)) {
551 *bpp = NULL;
552 return xfs_btree_get_iroot(cur);
553 }
554
555 *bpp = cur->bc_bufs[level];
556 return XFS_BUF_TO_BLOCK(*bpp);
557 }
558
559 /*
560 * Get a buffer for the block, return it with no data read.
561 * Long-form addressing.
562 */
563 xfs_buf_t * /* buffer for fsbno */
564 xfs_btree_get_bufl(
565 xfs_mount_t *mp, /* file system mount point */
566 xfs_trans_t *tp, /* transaction pointer */
567 xfs_fsblock_t fsbno, /* file system block number */
568 uint lock) /* lock flags for get_buf */
569 {
570 xfs_daddr_t d; /* real disk block address */
571
572 ASSERT(fsbno != NULLFSBLOCK);
573 d = XFS_FSB_TO_DADDR(mp, fsbno);
574 return xfs_trans_get_buf(tp, mp->m_ddev_targp, d, mp->m_bsize, lock);
575 }
576
577 /*
578 * Get a buffer for the block, return it with no data read.
579 * Short-form addressing.
580 */
581 xfs_buf_t * /* buffer for agno/agbno */
582 xfs_btree_get_bufs(
583 xfs_mount_t *mp, /* file system mount point */
584 xfs_trans_t *tp, /* transaction pointer */
585 xfs_agnumber_t agno, /* allocation group number */
586 xfs_agblock_t agbno, /* allocation group block number */
587 uint lock) /* lock flags for get_buf */
588 {
589 xfs_daddr_t d; /* real disk block address */
590
591 ASSERT(agno != NULLAGNUMBER);
592 ASSERT(agbno != NULLAGBLOCK);
593 d = XFS_AGB_TO_DADDR(mp, agno, agbno);
594 return xfs_trans_get_buf(tp, mp->m_ddev_targp, d, mp->m_bsize, lock);
595 }
596
597 /*
598 * Check for the cursor referring to the last block at the given level.
599 */
600 int /* 1=is last block, 0=not last block */
601 xfs_btree_islastblock(
602 xfs_btree_cur_t *cur, /* btree cursor */
603 int level) /* level to check */
604 {
605 struct xfs_btree_block *block; /* generic btree block pointer */
606 xfs_buf_t *bp; /* buffer containing block */
607
608 block = xfs_btree_get_block(cur, level, &bp);
609 xfs_btree_check_block(cur, block, level, bp);
610 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
611 return block->bb_u.l.bb_rightsib == cpu_to_be64(NULLFSBLOCK);
612 else
613 return block->bb_u.s.bb_rightsib == cpu_to_be32(NULLAGBLOCK);
614 }
615
616 /*
617 * Change the cursor to point to the first record at the given level.
618 * Other levels are unaffected.
619 */
620 STATIC int /* success=1, failure=0 */
621 xfs_btree_firstrec(
622 xfs_btree_cur_t *cur, /* btree cursor */
623 int level) /* level to change */
624 {
625 struct xfs_btree_block *block; /* generic btree block pointer */
626 xfs_buf_t *bp; /* buffer containing block */
627
628 /*
629 * Get the block pointer for this level.
630 */
631 block = xfs_btree_get_block(cur, level, &bp);
632 xfs_btree_check_block(cur, block, level, bp);
633 /*
634 * It's empty, there is no such record.
635 */
636 if (!block->bb_numrecs)
637 return 0;
638 /*
639 * Set the ptr value to 1, that's the first record/key.
640 */
641 cur->bc_ptrs[level] = 1;
642 return 1;
643 }
644
645 /*
646 * Change the cursor to point to the last record in the current block
647 * at the given level. Other levels are unaffected.
648 */
649 STATIC int /* success=1, failure=0 */
650 xfs_btree_lastrec(
651 xfs_btree_cur_t *cur, /* btree cursor */
652 int level) /* level to change */
653 {
654 struct xfs_btree_block *block; /* generic btree block pointer */
655 xfs_buf_t *bp; /* buffer containing block */
656
657 /*
658 * Get the block pointer for this level.
659 */
660 block = xfs_btree_get_block(cur, level, &bp);
661 xfs_btree_check_block(cur, block, level, bp);
662 /*
663 * It's empty, there is no such record.
664 */
665 if (!block->bb_numrecs)
666 return 0;
667 /*
668 * Set the ptr value to numrecs, that's the last record/key.
669 */
670 cur->bc_ptrs[level] = be16_to_cpu(block->bb_numrecs);
671 return 1;
672 }
673
674 /*
675 * Compute first and last byte offsets for the fields given.
676 * Interprets the offsets table, which contains struct field offsets.
677 */
678 void
679 xfs_btree_offsets(
680 __int64_t fields, /* bitmask of fields */
681 const short *offsets, /* table of field offsets */
682 int nbits, /* number of bits to inspect */
683 int *first, /* output: first byte offset */
684 int *last) /* output: last byte offset */
685 {
686 int i; /* current bit number */
687 __int64_t imask; /* mask for current bit number */
688
689 ASSERT(fields != 0);
690 /*
691 * Find the lowest bit, so the first byte offset.
692 */
693 for (i = 0, imask = 1LL; ; i++, imask <<= 1) {
694 if (imask & fields) {
695 *first = offsets[i];
696 break;
697 }
698 }
699 /*
700 * Find the highest bit, so the last byte offset.
701 */
702 for (i = nbits - 1, imask = 1LL << i; ; i--, imask >>= 1) {
703 if (imask & fields) {
704 *last = offsets[i + 1] - 1;
705 break;
706 }
707 }
708 }
709
710 /*
711 * Get a buffer for the block, return it read in.
712 * Long-form addressing.
713 */
714 int
715 xfs_btree_read_bufl(
716 struct xfs_mount *mp, /* file system mount point */
717 struct xfs_trans *tp, /* transaction pointer */
718 xfs_fsblock_t fsbno, /* file system block number */
719 uint lock, /* lock flags for read_buf */
720 struct xfs_buf **bpp, /* buffer for fsbno */
721 int refval, /* ref count value for buffer */
722 const struct xfs_buf_ops *ops)
723 {
724 struct xfs_buf *bp; /* return value */
725 xfs_daddr_t d; /* real disk block address */
726 int error;
727
728 ASSERT(fsbno != NULLFSBLOCK);
729 d = XFS_FSB_TO_DADDR(mp, fsbno);
730 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, d,
731 mp->m_bsize, lock, &bp, ops);
732 if (error)
733 return error;
734 if (bp)
735 xfs_buf_set_ref(bp, refval);
736 *bpp = bp;
737 return 0;
738 }
739
740 /*
741 * Read-ahead the block, don't wait for it, don't return a buffer.
742 * Long-form addressing.
743 */
744 /* ARGSUSED */
745 void
746 xfs_btree_reada_bufl(
747 struct xfs_mount *mp, /* file system mount point */
748 xfs_fsblock_t fsbno, /* file system block number */
749 xfs_extlen_t count, /* count of filesystem blocks */
750 const struct xfs_buf_ops *ops)
751 {
752 xfs_daddr_t d;
753
754 ASSERT(fsbno != NULLFSBLOCK);
755 d = XFS_FSB_TO_DADDR(mp, fsbno);
756 xfs_buf_readahead(mp->m_ddev_targp, d, mp->m_bsize * count, ops);
757 }
758
759 /*
760 * Read-ahead the block, don't wait for it, don't return a buffer.
761 * Short-form addressing.
762 */
763 /* ARGSUSED */
764 void
765 xfs_btree_reada_bufs(
766 struct xfs_mount *mp, /* file system mount point */
767 xfs_agnumber_t agno, /* allocation group number */
768 xfs_agblock_t agbno, /* allocation group block number */
769 xfs_extlen_t count, /* count of filesystem blocks */
770 const struct xfs_buf_ops *ops)
771 {
772 xfs_daddr_t d;
773
774 ASSERT(agno != NULLAGNUMBER);
775 ASSERT(agbno != NULLAGBLOCK);
776 d = XFS_AGB_TO_DADDR(mp, agno, agbno);
777 xfs_buf_readahead(mp->m_ddev_targp, d, mp->m_bsize * count, ops);
778 }
779
780 STATIC int
781 xfs_btree_readahead_lblock(
782 struct xfs_btree_cur *cur,
783 int lr,
784 struct xfs_btree_block *block)
785 {
786 int rval = 0;
787 xfs_fsblock_t left = be64_to_cpu(block->bb_u.l.bb_leftsib);
788 xfs_fsblock_t right = be64_to_cpu(block->bb_u.l.bb_rightsib);
789
790 if ((lr & XFS_BTCUR_LEFTRA) && left != NULLFSBLOCK) {
791 xfs_btree_reada_bufl(cur->bc_mp, left, 1,
792 cur->bc_ops->buf_ops);
793 rval++;
794 }
795
796 if ((lr & XFS_BTCUR_RIGHTRA) && right != NULLFSBLOCK) {
797 xfs_btree_reada_bufl(cur->bc_mp, right, 1,
798 cur->bc_ops->buf_ops);
799 rval++;
800 }
801
802 return rval;
803 }
804
805 STATIC int
806 xfs_btree_readahead_sblock(
807 struct xfs_btree_cur *cur,
808 int lr,
809 struct xfs_btree_block *block)
810 {
811 int rval = 0;
812 xfs_agblock_t left = be32_to_cpu(block->bb_u.s.bb_leftsib);
813 xfs_agblock_t right = be32_to_cpu(block->bb_u.s.bb_rightsib);
814
815
816 if ((lr & XFS_BTCUR_LEFTRA) && left != NULLAGBLOCK) {
817 xfs_btree_reada_bufs(cur->bc_mp, cur->bc_private.a.agno,
818 left, 1, cur->bc_ops->buf_ops);
819 rval++;
820 }
821
822 if ((lr & XFS_BTCUR_RIGHTRA) && right != NULLAGBLOCK) {
823 xfs_btree_reada_bufs(cur->bc_mp, cur->bc_private.a.agno,
824 right, 1, cur->bc_ops->buf_ops);
825 rval++;
826 }
827
828 return rval;
829 }
830
831 /*
832 * Read-ahead btree blocks, at the given level.
833 * Bits in lr are set from XFS_BTCUR_{LEFT,RIGHT}RA.
834 */
835 STATIC int
836 xfs_btree_readahead(
837 struct xfs_btree_cur *cur, /* btree cursor */
838 int lev, /* level in btree */
839 int lr) /* left/right bits */
840 {
841 struct xfs_btree_block *block;
842
843 /*
844 * No readahead needed if we are at the root level and the
845 * btree root is stored in the inode.
846 */
847 if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
848 (lev == cur->bc_nlevels - 1))
849 return 0;
850
851 if ((cur->bc_ra[lev] | lr) == cur->bc_ra[lev])
852 return 0;
853
854 cur->bc_ra[lev] |= lr;
855 block = XFS_BUF_TO_BLOCK(cur->bc_bufs[lev]);
856
857 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
858 return xfs_btree_readahead_lblock(cur, lr, block);
859 return xfs_btree_readahead_sblock(cur, lr, block);
860 }
861
862 STATIC xfs_daddr_t
863 xfs_btree_ptr_to_daddr(
864 struct xfs_btree_cur *cur,
865 union xfs_btree_ptr *ptr)
866 {
867 if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
868 ASSERT(ptr->l != cpu_to_be64(NULLFSBLOCK));
869
870 return XFS_FSB_TO_DADDR(cur->bc_mp, be64_to_cpu(ptr->l));
871 } else {
872 ASSERT(cur->bc_private.a.agno != NULLAGNUMBER);
873 ASSERT(ptr->s != cpu_to_be32(NULLAGBLOCK));
874
875 return XFS_AGB_TO_DADDR(cur->bc_mp, cur->bc_private.a.agno,
876 be32_to_cpu(ptr->s));
877 }
878 }
879
880 /*
881 * Readahead @count btree blocks at the given @ptr location.
882 *
883 * We don't need to care about long or short form btrees here as we have a
884 * method of converting the ptr directly to a daddr available to us.
885 */
886 STATIC void
887 xfs_btree_readahead_ptr(
888 struct xfs_btree_cur *cur,
889 union xfs_btree_ptr *ptr,
890 xfs_extlen_t count)
891 {
892 xfs_buf_readahead(cur->bc_mp->m_ddev_targp,
893 xfs_btree_ptr_to_daddr(cur, ptr),
894 cur->bc_mp->m_bsize * count, cur->bc_ops->buf_ops);
895 }
896
897 /*
898 * Set the buffer for level "lev" in the cursor to bp, releasing
899 * any previous buffer.
900 */
901 STATIC void
902 xfs_btree_setbuf(
903 xfs_btree_cur_t *cur, /* btree cursor */
904 int lev, /* level in btree */
905 xfs_buf_t *bp) /* new buffer to set */
906 {
907 struct xfs_btree_block *b; /* btree block */
908
909 if (cur->bc_bufs[lev])
910 xfs_trans_brelse(cur->bc_tp, cur->bc_bufs[lev]);
911 cur->bc_bufs[lev] = bp;
912 cur->bc_ra[lev] = 0;
913
914 b = XFS_BUF_TO_BLOCK(bp);
915 if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
916 if (b->bb_u.l.bb_leftsib == cpu_to_be64(NULLFSBLOCK))
917 cur->bc_ra[lev] |= XFS_BTCUR_LEFTRA;
918 if (b->bb_u.l.bb_rightsib == cpu_to_be64(NULLFSBLOCK))
919 cur->bc_ra[lev] |= XFS_BTCUR_RIGHTRA;
920 } else {
921 if (b->bb_u.s.bb_leftsib == cpu_to_be32(NULLAGBLOCK))
922 cur->bc_ra[lev] |= XFS_BTCUR_LEFTRA;
923 if (b->bb_u.s.bb_rightsib == cpu_to_be32(NULLAGBLOCK))
924 cur->bc_ra[lev] |= XFS_BTCUR_RIGHTRA;
925 }
926 }
927
928 STATIC int
929 xfs_btree_ptr_is_null(
930 struct xfs_btree_cur *cur,
931 union xfs_btree_ptr *ptr)
932 {
933 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
934 return ptr->l == cpu_to_be64(NULLFSBLOCK);
935 else
936 return ptr->s == cpu_to_be32(NULLAGBLOCK);
937 }
938
939 STATIC void
940 xfs_btree_set_ptr_null(
941 struct xfs_btree_cur *cur,
942 union xfs_btree_ptr *ptr)
943 {
944 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
945 ptr->l = cpu_to_be64(NULLFSBLOCK);
946 else
947 ptr->s = cpu_to_be32(NULLAGBLOCK);
948 }
949
950 /*
951 * Get/set/init sibling pointers
952 */
953 STATIC void
954 xfs_btree_get_sibling(
955 struct xfs_btree_cur *cur,
956 struct xfs_btree_block *block,
957 union xfs_btree_ptr *ptr,
958 int lr)
959 {
960 ASSERT(lr == XFS_BB_LEFTSIB || lr == XFS_BB_RIGHTSIB);
961
962 if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
963 if (lr == XFS_BB_RIGHTSIB)
964 ptr->l = block->bb_u.l.bb_rightsib;
965 else
966 ptr->l = block->bb_u.l.bb_leftsib;
967 } else {
968 if (lr == XFS_BB_RIGHTSIB)
969 ptr->s = block->bb_u.s.bb_rightsib;
970 else
971 ptr->s = block->bb_u.s.bb_leftsib;
972 }
973 }
974
975 STATIC void
976 xfs_btree_set_sibling(
977 struct xfs_btree_cur *cur,
978 struct xfs_btree_block *block,
979 union xfs_btree_ptr *ptr,
980 int lr)
981 {
982 ASSERT(lr == XFS_BB_LEFTSIB || lr == XFS_BB_RIGHTSIB);
983
984 if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
985 if (lr == XFS_BB_RIGHTSIB)
986 block->bb_u.l.bb_rightsib = ptr->l;
987 else
988 block->bb_u.l.bb_leftsib = ptr->l;
989 } else {
990 if (lr == XFS_BB_RIGHTSIB)
991 block->bb_u.s.bb_rightsib = ptr->s;
992 else
993 block->bb_u.s.bb_leftsib = ptr->s;
994 }
995 }
996
997 void
998 xfs_btree_init_block_int(
999 struct xfs_mount *mp,
1000 struct xfs_btree_block *buf,
1001 xfs_daddr_t blkno,
1002 __u32 magic,
1003 __u16 level,
1004 __u16 numrecs,
1005 __u64 owner,
1006 unsigned int flags)
1007 {
1008 buf->bb_magic = cpu_to_be32(magic);
1009 buf->bb_level = cpu_to_be16(level);
1010 buf->bb_numrecs = cpu_to_be16(numrecs);
1011
1012 if (flags & XFS_BTREE_LONG_PTRS) {
1013 buf->bb_u.l.bb_leftsib = cpu_to_be64(NULLFSBLOCK);
1014 buf->bb_u.l.bb_rightsib = cpu_to_be64(NULLFSBLOCK);
1015 if (flags & XFS_BTREE_CRC_BLOCKS) {
1016 buf->bb_u.l.bb_blkno = cpu_to_be64(blkno);
1017 buf->bb_u.l.bb_owner = cpu_to_be64(owner);
1018 uuid_copy(&buf->bb_u.l.bb_uuid, &mp->m_sb.sb_meta_uuid);
1019 buf->bb_u.l.bb_pad = 0;
1020 buf->bb_u.l.bb_lsn = 0;
1021 }
1022 } else {
1023 /* owner is a 32 bit value on short blocks */
1024 __u32 __owner = (__u32)owner;
1025
1026 buf->bb_u.s.bb_leftsib = cpu_to_be32(NULLAGBLOCK);
1027 buf->bb_u.s.bb_rightsib = cpu_to_be32(NULLAGBLOCK);
1028 if (flags & XFS_BTREE_CRC_BLOCKS) {
1029 buf->bb_u.s.bb_blkno = cpu_to_be64(blkno);
1030 buf->bb_u.s.bb_owner = cpu_to_be32(__owner);
1031 uuid_copy(&buf->bb_u.s.bb_uuid, &mp->m_sb.sb_meta_uuid);
1032 buf->bb_u.s.bb_lsn = 0;
1033 }
1034 }
1035 }
1036
1037 void
1038 xfs_btree_init_block(
1039 struct xfs_mount *mp,
1040 struct xfs_buf *bp,
1041 __u32 magic,
1042 __u16 level,
1043 __u16 numrecs,
1044 __u64 owner,
1045 unsigned int flags)
1046 {
1047 xfs_btree_init_block_int(mp, XFS_BUF_TO_BLOCK(bp), bp->b_bn,
1048 magic, level, numrecs, owner, flags);
1049 }
1050
1051 STATIC void
1052 xfs_btree_init_block_cur(
1053 struct xfs_btree_cur *cur,
1054 struct xfs_buf *bp,
1055 int level,
1056 int numrecs)
1057 {
1058 __u64 owner;
1059
1060 /*
1061 * we can pull the owner from the cursor right now as the different
1062 * owners align directly with the pointer size of the btree. This may
1063 * change in future, but is safe for current users of the generic btree
1064 * code.
1065 */
1066 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
1067 owner = cur->bc_private.b.ip->i_ino;
1068 else
1069 owner = cur->bc_private.a.agno;
1070
1071 xfs_btree_init_block_int(cur->bc_mp, XFS_BUF_TO_BLOCK(bp), bp->b_bn,
1072 xfs_btree_magic(cur), level, numrecs,
1073 owner, cur->bc_flags);
1074 }
1075
1076 /*
1077 * Return true if ptr is the last record in the btree and
1078 * we need to track updates to this record. The decision
1079 * will be further refined in the update_lastrec method.
1080 */
1081 STATIC int
1082 xfs_btree_is_lastrec(
1083 struct xfs_btree_cur *cur,
1084 struct xfs_btree_block *block,
1085 int level)
1086 {
1087 union xfs_btree_ptr ptr;
1088
1089 if (level > 0)
1090 return 0;
1091 if (!(cur->bc_flags & XFS_BTREE_LASTREC_UPDATE))
1092 return 0;
1093
1094 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
1095 if (!xfs_btree_ptr_is_null(cur, &ptr))
1096 return 0;
1097 return 1;
1098 }
1099
1100 STATIC void
1101 xfs_btree_buf_to_ptr(
1102 struct xfs_btree_cur *cur,
1103 struct xfs_buf *bp,
1104 union xfs_btree_ptr *ptr)
1105 {
1106 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
1107 ptr->l = cpu_to_be64(XFS_DADDR_TO_FSB(cur->bc_mp,
1108 XFS_BUF_ADDR(bp)));
1109 else {
1110 ptr->s = cpu_to_be32(xfs_daddr_to_agbno(cur->bc_mp,
1111 XFS_BUF_ADDR(bp)));
1112 }
1113 }
1114
1115 STATIC void
1116 xfs_btree_set_refs(
1117 struct xfs_btree_cur *cur,
1118 struct xfs_buf *bp)
1119 {
1120 switch (cur->bc_btnum) {
1121 case XFS_BTNUM_BNO:
1122 case XFS_BTNUM_CNT:
1123 xfs_buf_set_ref(bp, XFS_ALLOC_BTREE_REF);
1124 break;
1125 case XFS_BTNUM_INO:
1126 case XFS_BTNUM_FINO:
1127 xfs_buf_set_ref(bp, XFS_INO_BTREE_REF);
1128 break;
1129 case XFS_BTNUM_BMAP:
1130 xfs_buf_set_ref(bp, XFS_BMAP_BTREE_REF);
1131 break;
1132 default:
1133 ASSERT(0);
1134 }
1135 }
1136
1137 STATIC int
1138 xfs_btree_get_buf_block(
1139 struct xfs_btree_cur *cur,
1140 union xfs_btree_ptr *ptr,
1141 int flags,
1142 struct xfs_btree_block **block,
1143 struct xfs_buf **bpp)
1144 {
1145 struct xfs_mount *mp = cur->bc_mp;
1146 xfs_daddr_t d;
1147
1148 /* need to sort out how callers deal with failures first */
1149 ASSERT(!(flags & XBF_TRYLOCK));
1150
1151 d = xfs_btree_ptr_to_daddr(cur, ptr);
1152 *bpp = xfs_trans_get_buf(cur->bc_tp, mp->m_ddev_targp, d,
1153 mp->m_bsize, flags);
1154
1155 if (!*bpp)
1156 return -ENOMEM;
1157
1158 (*bpp)->b_ops = cur->bc_ops->buf_ops;
1159 *block = XFS_BUF_TO_BLOCK(*bpp);
1160 return 0;
1161 }
1162
1163 /*
1164 * Read in the buffer at the given ptr and return the buffer and
1165 * the block pointer within the buffer.
1166 */
1167 STATIC int
1168 xfs_btree_read_buf_block(
1169 struct xfs_btree_cur *cur,
1170 union xfs_btree_ptr *ptr,
1171 int flags,
1172 struct xfs_btree_block **block,
1173 struct xfs_buf **bpp)
1174 {
1175 struct xfs_mount *mp = cur->bc_mp;
1176 xfs_daddr_t d;
1177 int error;
1178
1179 /* need to sort out how callers deal with failures first */
1180 ASSERT(!(flags & XBF_TRYLOCK));
1181
1182 d = xfs_btree_ptr_to_daddr(cur, ptr);
1183 error = xfs_trans_read_buf(mp, cur->bc_tp, mp->m_ddev_targp, d,
1184 mp->m_bsize, flags, bpp,
1185 cur->bc_ops->buf_ops);
1186 if (error)
1187 return error;
1188
1189 xfs_btree_set_refs(cur, *bpp);
1190 *block = XFS_BUF_TO_BLOCK(*bpp);
1191 return 0;
1192 }
1193
1194 /*
1195 * Copy keys from one btree block to another.
1196 */
1197 STATIC void
1198 xfs_btree_copy_keys(
1199 struct xfs_btree_cur *cur,
1200 union xfs_btree_key *dst_key,
1201 union xfs_btree_key *src_key,
1202 int numkeys)
1203 {
1204 ASSERT(numkeys >= 0);
1205 memcpy(dst_key, src_key, numkeys * cur->bc_ops->key_len);
1206 }
1207
1208 /*
1209 * Copy records from one btree block to another.
1210 */
1211 STATIC void
1212 xfs_btree_copy_recs(
1213 struct xfs_btree_cur *cur,
1214 union xfs_btree_rec *dst_rec,
1215 union xfs_btree_rec *src_rec,
1216 int numrecs)
1217 {
1218 ASSERT(numrecs >= 0);
1219 memcpy(dst_rec, src_rec, numrecs * cur->bc_ops->rec_len);
1220 }
1221
1222 /*
1223 * Copy block pointers from one btree block to another.
1224 */
1225 STATIC void
1226 xfs_btree_copy_ptrs(
1227 struct xfs_btree_cur *cur,
1228 union xfs_btree_ptr *dst_ptr,
1229 union xfs_btree_ptr *src_ptr,
1230 int numptrs)
1231 {
1232 ASSERT(numptrs >= 0);
1233 memcpy(dst_ptr, src_ptr, numptrs * xfs_btree_ptr_len(cur));
1234 }
1235
1236 /*
1237 * Shift keys one index left/right inside a single btree block.
1238 */
1239 STATIC void
1240 xfs_btree_shift_keys(
1241 struct xfs_btree_cur *cur,
1242 union xfs_btree_key *key,
1243 int dir,
1244 int numkeys)
1245 {
1246 char *dst_key;
1247
1248 ASSERT(numkeys >= 0);
1249 ASSERT(dir == 1 || dir == -1);
1250
1251 dst_key = (char *)key + (dir * cur->bc_ops->key_len);
1252 memmove(dst_key, key, numkeys * cur->bc_ops->key_len);
1253 }
1254
1255 /*
1256 * Shift records one index left/right inside a single btree block.
1257 */
1258 STATIC void
1259 xfs_btree_shift_recs(
1260 struct xfs_btree_cur *cur,
1261 union xfs_btree_rec *rec,
1262 int dir,
1263 int numrecs)
1264 {
1265 char *dst_rec;
1266
1267 ASSERT(numrecs >= 0);
1268 ASSERT(dir == 1 || dir == -1);
1269
1270 dst_rec = (char *)rec + (dir * cur->bc_ops->rec_len);
1271 memmove(dst_rec, rec, numrecs * cur->bc_ops->rec_len);
1272 }
1273
1274 /*
1275 * Shift block pointers one index left/right inside a single btree block.
1276 */
1277 STATIC void
1278 xfs_btree_shift_ptrs(
1279 struct xfs_btree_cur *cur,
1280 union xfs_btree_ptr *ptr,
1281 int dir,
1282 int numptrs)
1283 {
1284 char *dst_ptr;
1285
1286 ASSERT(numptrs >= 0);
1287 ASSERT(dir == 1 || dir == -1);
1288
1289 dst_ptr = (char *)ptr + (dir * xfs_btree_ptr_len(cur));
1290 memmove(dst_ptr, ptr, numptrs * xfs_btree_ptr_len(cur));
1291 }
1292
1293 /*
1294 * Log key values from the btree block.
1295 */
1296 STATIC void
1297 xfs_btree_log_keys(
1298 struct xfs_btree_cur *cur,
1299 struct xfs_buf *bp,
1300 int first,
1301 int last)
1302 {
1303 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1304 XFS_BTREE_TRACE_ARGBII(cur, bp, first, last);
1305
1306 if (bp) {
1307 xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
1308 xfs_trans_log_buf(cur->bc_tp, bp,
1309 xfs_btree_key_offset(cur, first),
1310 xfs_btree_key_offset(cur, last + 1) - 1);
1311 } else {
1312 xfs_trans_log_inode(cur->bc_tp, cur->bc_private.b.ip,
1313 xfs_ilog_fbroot(cur->bc_private.b.whichfork));
1314 }
1315
1316 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1317 }
1318
1319 /*
1320 * Log record values from the btree block.
1321 */
1322 void
1323 xfs_btree_log_recs(
1324 struct xfs_btree_cur *cur,
1325 struct xfs_buf *bp,
1326 int first,
1327 int last)
1328 {
1329 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1330 XFS_BTREE_TRACE_ARGBII(cur, bp, first, last);
1331
1332 xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
1333 xfs_trans_log_buf(cur->bc_tp, bp,
1334 xfs_btree_rec_offset(cur, first),
1335 xfs_btree_rec_offset(cur, last + 1) - 1);
1336
1337 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1338 }
1339
1340 /*
1341 * Log block pointer fields from a btree block (nonleaf).
1342 */
1343 STATIC void
1344 xfs_btree_log_ptrs(
1345 struct xfs_btree_cur *cur, /* btree cursor */
1346 struct xfs_buf *bp, /* buffer containing btree block */
1347 int first, /* index of first pointer to log */
1348 int last) /* index of last pointer to log */
1349 {
1350 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1351 XFS_BTREE_TRACE_ARGBII(cur, bp, first, last);
1352
1353 if (bp) {
1354 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
1355 int level = xfs_btree_get_level(block);
1356
1357 xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
1358 xfs_trans_log_buf(cur->bc_tp, bp,
1359 xfs_btree_ptr_offset(cur, first, level),
1360 xfs_btree_ptr_offset(cur, last + 1, level) - 1);
1361 } else {
1362 xfs_trans_log_inode(cur->bc_tp, cur->bc_private.b.ip,
1363 xfs_ilog_fbroot(cur->bc_private.b.whichfork));
1364 }
1365
1366 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1367 }
1368
1369 /*
1370 * Log fields from a btree block header.
1371 */
1372 void
1373 xfs_btree_log_block(
1374 struct xfs_btree_cur *cur, /* btree cursor */
1375 struct xfs_buf *bp, /* buffer containing btree block */
1376 int fields) /* mask of fields: XFS_BB_... */
1377 {
1378 int first; /* first byte offset logged */
1379 int last; /* last byte offset logged */
1380 static const short soffsets[] = { /* table of offsets (short) */
1381 offsetof(struct xfs_btree_block, bb_magic),
1382 offsetof(struct xfs_btree_block, bb_level),
1383 offsetof(struct xfs_btree_block, bb_numrecs),
1384 offsetof(struct xfs_btree_block, bb_u.s.bb_leftsib),
1385 offsetof(struct xfs_btree_block, bb_u.s.bb_rightsib),
1386 offsetof(struct xfs_btree_block, bb_u.s.bb_blkno),
1387 offsetof(struct xfs_btree_block, bb_u.s.bb_lsn),
1388 offsetof(struct xfs_btree_block, bb_u.s.bb_uuid),
1389 offsetof(struct xfs_btree_block, bb_u.s.bb_owner),
1390 offsetof(struct xfs_btree_block, bb_u.s.bb_crc),
1391 XFS_BTREE_SBLOCK_CRC_LEN
1392 };
1393 static const short loffsets[] = { /* table of offsets (long) */
1394 offsetof(struct xfs_btree_block, bb_magic),
1395 offsetof(struct xfs_btree_block, bb_level),
1396 offsetof(struct xfs_btree_block, bb_numrecs),
1397 offsetof(struct xfs_btree_block, bb_u.l.bb_leftsib),
1398 offsetof(struct xfs_btree_block, bb_u.l.bb_rightsib),
1399 offsetof(struct xfs_btree_block, bb_u.l.bb_blkno),
1400 offsetof(struct xfs_btree_block, bb_u.l.bb_lsn),
1401 offsetof(struct xfs_btree_block, bb_u.l.bb_uuid),
1402 offsetof(struct xfs_btree_block, bb_u.l.bb_owner),
1403 offsetof(struct xfs_btree_block, bb_u.l.bb_crc),
1404 offsetof(struct xfs_btree_block, bb_u.l.bb_pad),
1405 XFS_BTREE_LBLOCK_CRC_LEN
1406 };
1407
1408 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1409 XFS_BTREE_TRACE_ARGBI(cur, bp, fields);
1410
1411 if (bp) {
1412 int nbits;
1413
1414 if (cur->bc_flags & XFS_BTREE_CRC_BLOCKS) {
1415 /*
1416 * We don't log the CRC when updating a btree
1417 * block but instead recreate it during log
1418 * recovery. As the log buffers have checksums
1419 * of their own this is safe and avoids logging a crc
1420 * update in a lot of places.
1421 */
1422 if (fields == XFS_BB_ALL_BITS)
1423 fields = XFS_BB_ALL_BITS_CRC;
1424 nbits = XFS_BB_NUM_BITS_CRC;
1425 } else {
1426 nbits = XFS_BB_NUM_BITS;
1427 }
1428 xfs_btree_offsets(fields,
1429 (cur->bc_flags & XFS_BTREE_LONG_PTRS) ?
1430 loffsets : soffsets,
1431 nbits, &first, &last);
1432 xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
1433 xfs_trans_log_buf(cur->bc_tp, bp, first, last);
1434 } else {
1435 xfs_trans_log_inode(cur->bc_tp, cur->bc_private.b.ip,
1436 xfs_ilog_fbroot(cur->bc_private.b.whichfork));
1437 }
1438
1439 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1440 }
1441
1442 /*
1443 * Increment cursor by one record at the level.
1444 * For nonzero levels the leaf-ward information is untouched.
1445 */
1446 int /* error */
1447 xfs_btree_increment(
1448 struct xfs_btree_cur *cur,
1449 int level,
1450 int *stat) /* success/failure */
1451 {
1452 struct xfs_btree_block *block;
1453 union xfs_btree_ptr ptr;
1454 struct xfs_buf *bp;
1455 int error; /* error return value */
1456 int lev;
1457
1458 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1459 XFS_BTREE_TRACE_ARGI(cur, level);
1460
1461 ASSERT(level < cur->bc_nlevels);
1462
1463 /* Read-ahead to the right at this level. */
1464 xfs_btree_readahead(cur, level, XFS_BTCUR_RIGHTRA);
1465
1466 /* Get a pointer to the btree block. */
1467 block = xfs_btree_get_block(cur, level, &bp);
1468
1469 #ifdef DEBUG
1470 error = xfs_btree_check_block(cur, block, level, bp);
1471 if (error)
1472 goto error0;
1473 #endif
1474
1475 /* We're done if we remain in the block after the increment. */
1476 if (++cur->bc_ptrs[level] <= xfs_btree_get_numrecs(block))
1477 goto out1;
1478
1479 /* Fail if we just went off the right edge of the tree. */
1480 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
1481 if (xfs_btree_ptr_is_null(cur, &ptr))
1482 goto out0;
1483
1484 XFS_BTREE_STATS_INC(cur, increment);
1485
1486 /*
1487 * March up the tree incrementing pointers.
1488 * Stop when we don't go off the right edge of a block.
1489 */
1490 for (lev = level + 1; lev < cur->bc_nlevels; lev++) {
1491 block = xfs_btree_get_block(cur, lev, &bp);
1492
1493 #ifdef DEBUG
1494 error = xfs_btree_check_block(cur, block, lev, bp);
1495 if (error)
1496 goto error0;
1497 #endif
1498
1499 if (++cur->bc_ptrs[lev] <= xfs_btree_get_numrecs(block))
1500 break;
1501
1502 /* Read-ahead the right block for the next loop. */
1503 xfs_btree_readahead(cur, lev, XFS_BTCUR_RIGHTRA);
1504 }
1505
1506 /*
1507 * If we went off the root then we are either seriously
1508 * confused or have the tree root in an inode.
1509 */
1510 if (lev == cur->bc_nlevels) {
1511 if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE)
1512 goto out0;
1513 ASSERT(0);
1514 error = -EFSCORRUPTED;
1515 goto error0;
1516 }
1517 ASSERT(lev < cur->bc_nlevels);
1518
1519 /*
1520 * Now walk back down the tree, fixing up the cursor's buffer
1521 * pointers and key numbers.
1522 */
1523 for (block = xfs_btree_get_block(cur, lev, &bp); lev > level; ) {
1524 union xfs_btree_ptr *ptrp;
1525
1526 ptrp = xfs_btree_ptr_addr(cur, cur->bc_ptrs[lev], block);
1527 --lev;
1528 error = xfs_btree_read_buf_block(cur, ptrp, 0, &block, &bp);
1529 if (error)
1530 goto error0;
1531
1532 xfs_btree_setbuf(cur, lev, bp);
1533 cur->bc_ptrs[lev] = 1;
1534 }
1535 out1:
1536 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1537 *stat = 1;
1538 return 0;
1539
1540 out0:
1541 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1542 *stat = 0;
1543 return 0;
1544
1545 error0:
1546 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
1547 return error;
1548 }
1549
1550 /*
1551 * Decrement cursor by one record at the level.
1552 * For nonzero levels the leaf-ward information is untouched.
1553 */
1554 int /* error */
1555 xfs_btree_decrement(
1556 struct xfs_btree_cur *cur,
1557 int level,
1558 int *stat) /* success/failure */
1559 {
1560 struct xfs_btree_block *block;
1561 xfs_buf_t *bp;
1562 int error; /* error return value */
1563 int lev;
1564 union xfs_btree_ptr ptr;
1565
1566 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1567 XFS_BTREE_TRACE_ARGI(cur, level);
1568
1569 ASSERT(level < cur->bc_nlevels);
1570
1571 /* Read-ahead to the left at this level. */
1572 xfs_btree_readahead(cur, level, XFS_BTCUR_LEFTRA);
1573
1574 /* We're done if we remain in the block after the decrement. */
1575 if (--cur->bc_ptrs[level] > 0)
1576 goto out1;
1577
1578 /* Get a pointer to the btree block. */
1579 block = xfs_btree_get_block(cur, level, &bp);
1580
1581 #ifdef DEBUG
1582 error = xfs_btree_check_block(cur, block, level, bp);
1583 if (error)
1584 goto error0;
1585 #endif
1586
1587 /* Fail if we just went off the left edge of the tree. */
1588 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_LEFTSIB);
1589 if (xfs_btree_ptr_is_null(cur, &ptr))
1590 goto out0;
1591
1592 XFS_BTREE_STATS_INC(cur, decrement);
1593
1594 /*
1595 * March up the tree decrementing pointers.
1596 * Stop when we don't go off the left edge of a block.
1597 */
1598 for (lev = level + 1; lev < cur->bc_nlevels; lev++) {
1599 if (--cur->bc_ptrs[lev] > 0)
1600 break;
1601 /* Read-ahead the left block for the next loop. */
1602 xfs_btree_readahead(cur, lev, XFS_BTCUR_LEFTRA);
1603 }
1604
1605 /*
1606 * If we went off the root then we are seriously confused.
1607 * or the root of the tree is in an inode.
1608 */
1609 if (lev == cur->bc_nlevels) {
1610 if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE)
1611 goto out0;
1612 ASSERT(0);
1613 error = -EFSCORRUPTED;
1614 goto error0;
1615 }
1616 ASSERT(lev < cur->bc_nlevels);
1617
1618 /*
1619 * Now walk back down the tree, fixing up the cursor's buffer
1620 * pointers and key numbers.
1621 */
1622 for (block = xfs_btree_get_block(cur, lev, &bp); lev > level; ) {
1623 union xfs_btree_ptr *ptrp;
1624
1625 ptrp = xfs_btree_ptr_addr(cur, cur->bc_ptrs[lev], block);
1626 --lev;
1627 error = xfs_btree_read_buf_block(cur, ptrp, 0, &block, &bp);
1628 if (error)
1629 goto error0;
1630 xfs_btree_setbuf(cur, lev, bp);
1631 cur->bc_ptrs[lev] = xfs_btree_get_numrecs(block);
1632 }
1633 out1:
1634 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1635 *stat = 1;
1636 return 0;
1637
1638 out0:
1639 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1640 *stat = 0;
1641 return 0;
1642
1643 error0:
1644 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
1645 return error;
1646 }
1647
1648 STATIC int
1649 xfs_btree_lookup_get_block(
1650 struct xfs_btree_cur *cur, /* btree cursor */
1651 int level, /* level in the btree */
1652 union xfs_btree_ptr *pp, /* ptr to btree block */
1653 struct xfs_btree_block **blkp) /* return btree block */
1654 {
1655 struct xfs_buf *bp; /* buffer pointer for btree block */
1656 int error = 0;
1657
1658 /* special case the root block if in an inode */
1659 if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
1660 (level == cur->bc_nlevels - 1)) {
1661 *blkp = xfs_btree_get_iroot(cur);
1662 return 0;
1663 }
1664
1665 /*
1666 * If the old buffer at this level for the disk address we are
1667 * looking for re-use it.
1668 *
1669 * Otherwise throw it away and get a new one.
1670 */
1671 bp = cur->bc_bufs[level];
1672 if (bp && XFS_BUF_ADDR(bp) == xfs_btree_ptr_to_daddr(cur, pp)) {
1673 *blkp = XFS_BUF_TO_BLOCK(bp);
1674 return 0;
1675 }
1676
1677 error = xfs_btree_read_buf_block(cur, pp, 0, blkp, &bp);
1678 if (error)
1679 return error;
1680
1681 xfs_btree_setbuf(cur, level, bp);
1682 return 0;
1683 }
1684
1685 /*
1686 * Get current search key. For level 0 we don't actually have a key
1687 * structure so we make one up from the record. For all other levels
1688 * we just return the right key.
1689 */
1690 STATIC union xfs_btree_key *
1691 xfs_lookup_get_search_key(
1692 struct xfs_btree_cur *cur,
1693 int level,
1694 int keyno,
1695 struct xfs_btree_block *block,
1696 union xfs_btree_key *kp)
1697 {
1698 if (level == 0) {
1699 cur->bc_ops->init_key_from_rec(kp,
1700 xfs_btree_rec_addr(cur, keyno, block));
1701 return kp;
1702 }
1703
1704 return xfs_btree_key_addr(cur, keyno, block);
1705 }
1706
1707 /*
1708 * Lookup the record. The cursor is made to point to it, based on dir.
1709 * stat is set to 0 if can't find any such record, 1 for success.
1710 */
1711 int /* error */
1712 xfs_btree_lookup(
1713 struct xfs_btree_cur *cur, /* btree cursor */
1714 xfs_lookup_t dir, /* <=, ==, or >= */
1715 int *stat) /* success/failure */
1716 {
1717 struct xfs_btree_block *block; /* current btree block */
1718 __int64_t diff; /* difference for the current key */
1719 int error; /* error return value */
1720 int keyno; /* current key number */
1721 int level; /* level in the btree */
1722 union xfs_btree_ptr *pp; /* ptr to btree block */
1723 union xfs_btree_ptr ptr; /* ptr to btree block */
1724
1725 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1726 XFS_BTREE_TRACE_ARGI(cur, dir);
1727
1728 XFS_BTREE_STATS_INC(cur, lookup);
1729
1730 block = NULL;
1731 keyno = 0;
1732
1733 /* initialise start pointer from cursor */
1734 cur->bc_ops->init_ptr_from_cur(cur, &ptr);
1735 pp = &ptr;
1736
1737 /*
1738 * Iterate over each level in the btree, starting at the root.
1739 * For each level above the leaves, find the key we need, based
1740 * on the lookup record, then follow the corresponding block
1741 * pointer down to the next level.
1742 */
1743 for (level = cur->bc_nlevels - 1, diff = 1; level >= 0; level--) {
1744 /* Get the block we need to do the lookup on. */
1745 error = xfs_btree_lookup_get_block(cur, level, pp, &block);
1746 if (error)
1747 goto error0;
1748
1749 if (diff == 0) {
1750 /*
1751 * If we already had a key match at a higher level, we
1752 * know we need to use the first entry in this block.
1753 */
1754 keyno = 1;
1755 } else {
1756 /* Otherwise search this block. Do a binary search. */
1757
1758 int high; /* high entry number */
1759 int low; /* low entry number */
1760
1761 /* Set low and high entry numbers, 1-based. */
1762 low = 1;
1763 high = xfs_btree_get_numrecs(block);
1764 if (!high) {
1765 /* Block is empty, must be an empty leaf. */
1766 ASSERT(level == 0 && cur->bc_nlevels == 1);
1767
1768 cur->bc_ptrs[0] = dir != XFS_LOOKUP_LE;
1769 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1770 *stat = 0;
1771 return 0;
1772 }
1773
1774 /* Binary search the block. */
1775 while (low <= high) {
1776 union xfs_btree_key key;
1777 union xfs_btree_key *kp;
1778
1779 XFS_BTREE_STATS_INC(cur, compare);
1780
1781 /* keyno is average of low and high. */
1782 keyno = (low + high) >> 1;
1783
1784 /* Get current search key */
1785 kp = xfs_lookup_get_search_key(cur, level,
1786 keyno, block, &key);
1787
1788 /*
1789 * Compute difference to get next direction:
1790 * - less than, move right
1791 * - greater than, move left
1792 * - equal, we're done
1793 */
1794 diff = cur->bc_ops->key_diff(cur, kp);
1795 if (diff < 0)
1796 low = keyno + 1;
1797 else if (diff > 0)
1798 high = keyno - 1;
1799 else
1800 break;
1801 }
1802 }
1803
1804 /*
1805 * If there are more levels, set up for the next level
1806 * by getting the block number and filling in the cursor.
1807 */
1808 if (level > 0) {
1809 /*
1810 * If we moved left, need the previous key number,
1811 * unless there isn't one.
1812 */
1813 if (diff > 0 && --keyno < 1)
1814 keyno = 1;
1815 pp = xfs_btree_ptr_addr(cur, keyno, block);
1816
1817 #ifdef DEBUG
1818 error = xfs_btree_check_ptr(cur, pp, 0, level);
1819 if (error)
1820 goto error0;
1821 #endif
1822 cur->bc_ptrs[level] = keyno;
1823 }
1824 }
1825
1826 /* Done with the search. See if we need to adjust the results. */
1827 if (dir != XFS_LOOKUP_LE && diff < 0) {
1828 keyno++;
1829 /*
1830 * If ge search and we went off the end of the block, but it's
1831 * not the last block, we're in the wrong block.
1832 */
1833 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
1834 if (dir == XFS_LOOKUP_GE &&
1835 keyno > xfs_btree_get_numrecs(block) &&
1836 !xfs_btree_ptr_is_null(cur, &ptr)) {
1837 int i;
1838
1839 cur->bc_ptrs[0] = keyno;
1840 error = xfs_btree_increment(cur, 0, &i);
1841 if (error)
1842 goto error0;
1843 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
1844 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1845 *stat = 1;
1846 return 0;
1847 }
1848 } else if (dir == XFS_LOOKUP_LE && diff > 0)
1849 keyno--;
1850 cur->bc_ptrs[0] = keyno;
1851
1852 /* Return if we succeeded or not. */
1853 if (keyno == 0 || keyno > xfs_btree_get_numrecs(block))
1854 *stat = 0;
1855 else if (dir != XFS_LOOKUP_EQ || diff == 0)
1856 *stat = 1;
1857 else
1858 *stat = 0;
1859 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1860 return 0;
1861
1862 error0:
1863 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
1864 return error;
1865 }
1866
1867 /*
1868 * Update keys at all levels from here to the root along the cursor's path.
1869 */
1870 STATIC int
1871 xfs_btree_updkey(
1872 struct xfs_btree_cur *cur,
1873 union xfs_btree_key *keyp,
1874 int level)
1875 {
1876 struct xfs_btree_block *block;
1877 struct xfs_buf *bp;
1878 union xfs_btree_key *kp;
1879 int ptr;
1880
1881 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1882 XFS_BTREE_TRACE_ARGIK(cur, level, keyp);
1883
1884 ASSERT(!(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) || level >= 1);
1885
1886 /*
1887 * Go up the tree from this level toward the root.
1888 * At each level, update the key value to the value input.
1889 * Stop when we reach a level where the cursor isn't pointing
1890 * at the first entry in the block.
1891 */
1892 for (ptr = 1; ptr == 1 && level < cur->bc_nlevels; level++) {
1893 #ifdef DEBUG
1894 int error;
1895 #endif
1896 block = xfs_btree_get_block(cur, level, &bp);
1897 #ifdef DEBUG
1898 error = xfs_btree_check_block(cur, block, level, bp);
1899 if (error) {
1900 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
1901 return error;
1902 }
1903 #endif
1904 ptr = cur->bc_ptrs[level];
1905 kp = xfs_btree_key_addr(cur, ptr, block);
1906 xfs_btree_copy_keys(cur, kp, keyp, 1);
1907 xfs_btree_log_keys(cur, bp, ptr, ptr);
1908 }
1909
1910 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1911 return 0;
1912 }
1913
1914 /*
1915 * Update the record referred to by cur to the value in the
1916 * given record. This either works (return 0) or gets an
1917 * EFSCORRUPTED error.
1918 */
1919 int
1920 xfs_btree_update(
1921 struct xfs_btree_cur *cur,
1922 union xfs_btree_rec *rec)
1923 {
1924 struct xfs_btree_block *block;
1925 struct xfs_buf *bp;
1926 int error;
1927 int ptr;
1928 union xfs_btree_rec *rp;
1929
1930 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1931 XFS_BTREE_TRACE_ARGR(cur, rec);
1932
1933 /* Pick up the current block. */
1934 block = xfs_btree_get_block(cur, 0, &bp);
1935
1936 #ifdef DEBUG
1937 error = xfs_btree_check_block(cur, block, 0, bp);
1938 if (error)
1939 goto error0;
1940 #endif
1941 /* Get the address of the rec to be updated. */
1942 ptr = cur->bc_ptrs[0];
1943 rp = xfs_btree_rec_addr(cur, ptr, block);
1944
1945 /* Fill in the new contents and log them. */
1946 xfs_btree_copy_recs(cur, rp, rec, 1);
1947 xfs_btree_log_recs(cur, bp, ptr, ptr);
1948
1949 /*
1950 * If we are tracking the last record in the tree and
1951 * we are at the far right edge of the tree, update it.
1952 */
1953 if (xfs_btree_is_lastrec(cur, block, 0)) {
1954 cur->bc_ops->update_lastrec(cur, block, rec,
1955 ptr, LASTREC_UPDATE);
1956 }
1957
1958 /* Updating first rec in leaf. Pass new key value up to our parent. */
1959 if (ptr == 1) {
1960 union xfs_btree_key key;
1961
1962 cur->bc_ops->init_key_from_rec(&key, rec);
1963 error = xfs_btree_updkey(cur, &key, 1);
1964 if (error)
1965 goto error0;
1966 }
1967
1968 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1969 return 0;
1970
1971 error0:
1972 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
1973 return error;
1974 }
1975
1976 /*
1977 * Move 1 record left from cur/level if possible.
1978 * Update cur to reflect the new path.
1979 */
1980 STATIC int /* error */
1981 xfs_btree_lshift(
1982 struct xfs_btree_cur *cur,
1983 int level,
1984 int *stat) /* success/failure */
1985 {
1986 union xfs_btree_key key; /* btree key */
1987 struct xfs_buf *lbp; /* left buffer pointer */
1988 struct xfs_btree_block *left; /* left btree block */
1989 int lrecs; /* left record count */
1990 struct xfs_buf *rbp; /* right buffer pointer */
1991 struct xfs_btree_block *right; /* right btree block */
1992 int rrecs; /* right record count */
1993 union xfs_btree_ptr lptr; /* left btree pointer */
1994 union xfs_btree_key *rkp = NULL; /* right btree key */
1995 union xfs_btree_ptr *rpp = NULL; /* right address pointer */
1996 union xfs_btree_rec *rrp = NULL; /* right record pointer */
1997 int error; /* error return value */
1998
1999 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2000 XFS_BTREE_TRACE_ARGI(cur, level);
2001
2002 if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
2003 level == cur->bc_nlevels - 1)
2004 goto out0;
2005
2006 /* Set up variables for this block as "right". */
2007 right = xfs_btree_get_block(cur, level, &rbp);
2008
2009 #ifdef DEBUG
2010 error = xfs_btree_check_block(cur, right, level, rbp);
2011 if (error)
2012 goto error0;
2013 #endif
2014
2015 /* If we've got no left sibling then we can't shift an entry left. */
2016 xfs_btree_get_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
2017 if (xfs_btree_ptr_is_null(cur, &lptr))
2018 goto out0;
2019
2020 /*
2021 * If the cursor entry is the one that would be moved, don't
2022 * do it... it's too complicated.
2023 */
2024 if (cur->bc_ptrs[level] <= 1)
2025 goto out0;
2026
2027 /* Set up the left neighbor as "left". */
2028 error = xfs_btree_read_buf_block(cur, &lptr, 0, &left, &lbp);
2029 if (error)
2030 goto error0;
2031
2032 /* If it's full, it can't take another entry. */
2033 lrecs = xfs_btree_get_numrecs(left);
2034 if (lrecs == cur->bc_ops->get_maxrecs(cur, level))
2035 goto out0;
2036
2037 rrecs = xfs_btree_get_numrecs(right);
2038
2039 /*
2040 * We add one entry to the left side and remove one for the right side.
2041 * Account for it here, the changes will be updated on disk and logged
2042 * later.
2043 */
2044 lrecs++;
2045 rrecs--;
2046
2047 XFS_BTREE_STATS_INC(cur, lshift);
2048 XFS_BTREE_STATS_ADD(cur, moves, 1);
2049
2050 /*
2051 * If non-leaf, copy a key and a ptr to the left block.
2052 * Log the changes to the left block.
2053 */
2054 if (level > 0) {
2055 /* It's a non-leaf. Move keys and pointers. */
2056 union xfs_btree_key *lkp; /* left btree key */
2057 union xfs_btree_ptr *lpp; /* left address pointer */
2058
2059 lkp = xfs_btree_key_addr(cur, lrecs, left);
2060 rkp = xfs_btree_key_addr(cur, 1, right);
2061
2062 lpp = xfs_btree_ptr_addr(cur, lrecs, left);
2063 rpp = xfs_btree_ptr_addr(cur, 1, right);
2064 #ifdef DEBUG
2065 error = xfs_btree_check_ptr(cur, rpp, 0, level);
2066 if (error)
2067 goto error0;
2068 #endif
2069 xfs_btree_copy_keys(cur, lkp, rkp, 1);
2070 xfs_btree_copy_ptrs(cur, lpp, rpp, 1);
2071
2072 xfs_btree_log_keys(cur, lbp, lrecs, lrecs);
2073 xfs_btree_log_ptrs(cur, lbp, lrecs, lrecs);
2074
2075 ASSERT(cur->bc_ops->keys_inorder(cur,
2076 xfs_btree_key_addr(cur, lrecs - 1, left), lkp));
2077 } else {
2078 /* It's a leaf. Move records. */
2079 union xfs_btree_rec *lrp; /* left record pointer */
2080
2081 lrp = xfs_btree_rec_addr(cur, lrecs, left);
2082 rrp = xfs_btree_rec_addr(cur, 1, right);
2083
2084 xfs_btree_copy_recs(cur, lrp, rrp, 1);
2085 xfs_btree_log_recs(cur, lbp, lrecs, lrecs);
2086
2087 ASSERT(cur->bc_ops->recs_inorder(cur,
2088 xfs_btree_rec_addr(cur, lrecs - 1, left), lrp));
2089 }
2090
2091 xfs_btree_set_numrecs(left, lrecs);
2092 xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS);
2093
2094 xfs_btree_set_numrecs(right, rrecs);
2095 xfs_btree_log_block(cur, rbp, XFS_BB_NUMRECS);
2096
2097 /*
2098 * Slide the contents of right down one entry.
2099 */
2100 XFS_BTREE_STATS_ADD(cur, moves, rrecs - 1);
2101 if (level > 0) {
2102 /* It's a nonleaf. operate on keys and ptrs */
2103 #ifdef DEBUG
2104 int i; /* loop index */
2105
2106 for (i = 0; i < rrecs; i++) {
2107 error = xfs_btree_check_ptr(cur, rpp, i + 1, level);
2108 if (error)
2109 goto error0;
2110 }
2111 #endif
2112 xfs_btree_shift_keys(cur,
2113 xfs_btree_key_addr(cur, 2, right),
2114 -1, rrecs);
2115 xfs_btree_shift_ptrs(cur,
2116 xfs_btree_ptr_addr(cur, 2, right),
2117 -1, rrecs);
2118
2119 xfs_btree_log_keys(cur, rbp, 1, rrecs);
2120 xfs_btree_log_ptrs(cur, rbp, 1, rrecs);
2121 } else {
2122 /* It's a leaf. operate on records */
2123 xfs_btree_shift_recs(cur,
2124 xfs_btree_rec_addr(cur, 2, right),
2125 -1, rrecs);
2126 xfs_btree_log_recs(cur, rbp, 1, rrecs);
2127
2128 /*
2129 * If it's the first record in the block, we'll need a key
2130 * structure to pass up to the next level (updkey).
2131 */
2132 cur->bc_ops->init_key_from_rec(&key,
2133 xfs_btree_rec_addr(cur, 1, right));
2134 rkp = &key;
2135 }
2136
2137 /* Update the parent key values of right. */
2138 error = xfs_btree_updkey(cur, rkp, level + 1);
2139 if (error)
2140 goto error0;
2141
2142 /* Slide the cursor value left one. */
2143 cur->bc_ptrs[level]--;
2144
2145 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2146 *stat = 1;
2147 return 0;
2148
2149 out0:
2150 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2151 *stat = 0;
2152 return 0;
2153
2154 error0:
2155 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2156 return error;
2157 }
2158
2159 /*
2160 * Move 1 record right from cur/level if possible.
2161 * Update cur to reflect the new path.
2162 */
2163 STATIC int /* error */
2164 xfs_btree_rshift(
2165 struct xfs_btree_cur *cur,
2166 int level,
2167 int *stat) /* success/failure */
2168 {
2169 union xfs_btree_key key; /* btree key */
2170 struct xfs_buf *lbp; /* left buffer pointer */
2171 struct xfs_btree_block *left; /* left btree block */
2172 struct xfs_buf *rbp; /* right buffer pointer */
2173 struct xfs_btree_block *right; /* right btree block */
2174 struct xfs_btree_cur *tcur; /* temporary btree cursor */
2175 union xfs_btree_ptr rptr; /* right block pointer */
2176 union xfs_btree_key *rkp; /* right btree key */
2177 int rrecs; /* right record count */
2178 int lrecs; /* left record count */
2179 int error; /* error return value */
2180 int i; /* loop counter */
2181
2182 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2183 XFS_BTREE_TRACE_ARGI(cur, level);
2184
2185 if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
2186 (level == cur->bc_nlevels - 1))
2187 goto out0;
2188
2189 /* Set up variables for this block as "left". */
2190 left = xfs_btree_get_block(cur, level, &lbp);
2191
2192 #ifdef DEBUG
2193 error = xfs_btree_check_block(cur, left, level, lbp);
2194 if (error)
2195 goto error0;
2196 #endif
2197
2198 /* If we've got no right sibling then we can't shift an entry right. */
2199 xfs_btree_get_sibling(cur, left, &rptr, XFS_BB_RIGHTSIB);
2200 if (xfs_btree_ptr_is_null(cur, &rptr))
2201 goto out0;
2202
2203 /*
2204 * If the cursor entry is the one that would be moved, don't
2205 * do it... it's too complicated.
2206 */
2207 lrecs = xfs_btree_get_numrecs(left);
2208 if (cur->bc_ptrs[level] >= lrecs)
2209 goto out0;
2210
2211 /* Set up the right neighbor as "right". */
2212 error = xfs_btree_read_buf_block(cur, &rptr, 0, &right, &rbp);
2213 if (error)
2214 goto error0;
2215
2216 /* If it's full, it can't take another entry. */
2217 rrecs = xfs_btree_get_numrecs(right);
2218 if (rrecs == cur->bc_ops->get_maxrecs(cur, level))
2219 goto out0;
2220
2221 XFS_BTREE_STATS_INC(cur, rshift);
2222 XFS_BTREE_STATS_ADD(cur, moves, rrecs);
2223
2224 /*
2225 * Make a hole at the start of the right neighbor block, then
2226 * copy the last left block entry to the hole.
2227 */
2228 if (level > 0) {
2229 /* It's a nonleaf. make a hole in the keys and ptrs */
2230 union xfs_btree_key *lkp;
2231 union xfs_btree_ptr *lpp;
2232 union xfs_btree_ptr *rpp;
2233
2234 lkp = xfs_btree_key_addr(cur, lrecs, left);
2235 lpp = xfs_btree_ptr_addr(cur, lrecs, left);
2236 rkp = xfs_btree_key_addr(cur, 1, right);
2237 rpp = xfs_btree_ptr_addr(cur, 1, right);
2238
2239 #ifdef DEBUG
2240 for (i = rrecs - 1; i >= 0; i--) {
2241 error = xfs_btree_check_ptr(cur, rpp, i, level);
2242 if (error)
2243 goto error0;
2244 }
2245 #endif
2246
2247 xfs_btree_shift_keys(cur, rkp, 1, rrecs);
2248 xfs_btree_shift_ptrs(cur, rpp, 1, rrecs);
2249
2250 #ifdef DEBUG
2251 error = xfs_btree_check_ptr(cur, lpp, 0, level);
2252 if (error)
2253 goto error0;
2254 #endif
2255
2256 /* Now put the new data in, and log it. */
2257 xfs_btree_copy_keys(cur, rkp, lkp, 1);
2258 xfs_btree_copy_ptrs(cur, rpp, lpp, 1);
2259
2260 xfs_btree_log_keys(cur, rbp, 1, rrecs + 1);
2261 xfs_btree_log_ptrs(cur, rbp, 1, rrecs + 1);
2262
2263 ASSERT(cur->bc_ops->keys_inorder(cur, rkp,
2264 xfs_btree_key_addr(cur, 2, right)));
2265 } else {
2266 /* It's a leaf. make a hole in the records */
2267 union xfs_btree_rec *lrp;
2268 union xfs_btree_rec *rrp;
2269
2270 lrp = xfs_btree_rec_addr(cur, lrecs, left);
2271 rrp = xfs_btree_rec_addr(cur, 1, right);
2272
2273 xfs_btree_shift_recs(cur, rrp, 1, rrecs);
2274
2275 /* Now put the new data in, and log it. */
2276 xfs_btree_copy_recs(cur, rrp, lrp, 1);
2277 xfs_btree_log_recs(cur, rbp, 1, rrecs + 1);
2278
2279 cur->bc_ops->init_key_from_rec(&key, rrp);
2280 rkp = &key;
2281
2282 ASSERT(cur->bc_ops->recs_inorder(cur, rrp,
2283 xfs_btree_rec_addr(cur, 2, right)));
2284 }
2285
2286 /*
2287 * Decrement and log left's numrecs, bump and log right's numrecs.
2288 */
2289 xfs_btree_set_numrecs(left, --lrecs);
2290 xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS);
2291
2292 xfs_btree_set_numrecs(right, ++rrecs);
2293 xfs_btree_log_block(cur, rbp, XFS_BB_NUMRECS);
2294
2295 /*
2296 * Using a temporary cursor, update the parent key values of the
2297 * block on the right.
2298 */
2299 error = xfs_btree_dup_cursor(cur, &tcur);
2300 if (error)
2301 goto error0;
2302 i = xfs_btree_lastrec(tcur, level);
2303 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0);
2304
2305 error = xfs_btree_increment(tcur, level, &i);
2306 if (error)
2307 goto error1;
2308
2309 error = xfs_btree_updkey(tcur, rkp, level + 1);
2310 if (error)
2311 goto error1;
2312
2313 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
2314
2315 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2316 *stat = 1;
2317 return 0;
2318
2319 out0:
2320 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2321 *stat = 0;
2322 return 0;
2323
2324 error0:
2325 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2326 return error;
2327
2328 error1:
2329 XFS_BTREE_TRACE_CURSOR(tcur, XBT_ERROR);
2330 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
2331 return error;
2332 }
2333
2334 /*
2335 * Split cur/level block in half.
2336 * Return new block number and the key to its first
2337 * record (to be inserted into parent).
2338 */
2339 STATIC int /* error */
2340 __xfs_btree_split(
2341 struct xfs_btree_cur *cur,
2342 int level,
2343 union xfs_btree_ptr *ptrp,
2344 union xfs_btree_key *key,
2345 struct xfs_btree_cur **curp,
2346 int *stat) /* success/failure */
2347 {
2348 union xfs_btree_ptr lptr; /* left sibling block ptr */
2349 struct xfs_buf *lbp; /* left buffer pointer */
2350 struct xfs_btree_block *left; /* left btree block */
2351 union xfs_btree_ptr rptr; /* right sibling block ptr */
2352 struct xfs_buf *rbp; /* right buffer pointer */
2353 struct xfs_btree_block *right; /* right btree block */
2354 union xfs_btree_ptr rrptr; /* right-right sibling ptr */
2355 struct xfs_buf *rrbp; /* right-right buffer pointer */
2356 struct xfs_btree_block *rrblock; /* right-right btree block */
2357 int lrecs;
2358 int rrecs;
2359 int src_index;
2360 int error; /* error return value */
2361 #ifdef DEBUG
2362 int i;
2363 #endif
2364
2365 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2366 XFS_BTREE_TRACE_ARGIPK(cur, level, *ptrp, key);
2367
2368 XFS_BTREE_STATS_INC(cur, split);
2369
2370 /* Set up left block (current one). */
2371 left = xfs_btree_get_block(cur, level, &lbp);
2372
2373 #ifdef DEBUG
2374 error = xfs_btree_check_block(cur, left, level, lbp);
2375 if (error)
2376 goto error0;
2377 #endif
2378
2379 xfs_btree_buf_to_ptr(cur, lbp, &lptr);
2380
2381 /* Allocate the new block. If we can't do it, we're toast. Give up. */
2382 error = cur->bc_ops->alloc_block(cur, &lptr, &rptr, stat);
2383 if (error)
2384 goto error0;
2385 if (*stat == 0)
2386 goto out0;
2387 XFS_BTREE_STATS_INC(cur, alloc);
2388
2389 /* Set up the new block as "right". */
2390 error = xfs_btree_get_buf_block(cur, &rptr, 0, &right, &rbp);
2391 if (error)
2392 goto error0;
2393
2394 /* Fill in the btree header for the new right block. */
2395 xfs_btree_init_block_cur(cur, rbp, xfs_btree_get_level(left), 0);
2396
2397 /*
2398 * Split the entries between the old and the new block evenly.
2399 * Make sure that if there's an odd number of entries now, that
2400 * each new block will have the same number of entries.
2401 */
2402 lrecs = xfs_btree_get_numrecs(left);
2403 rrecs = lrecs / 2;
2404 if ((lrecs & 1) && cur->bc_ptrs[level] <= rrecs + 1)
2405 rrecs++;
2406 src_index = (lrecs - rrecs + 1);
2407
2408 XFS_BTREE_STATS_ADD(cur, moves, rrecs);
2409
2410 /*
2411 * Copy btree block entries from the left block over to the
2412 * new block, the right. Update the right block and log the
2413 * changes.
2414 */
2415 if (level > 0) {
2416 /* It's a non-leaf. Move keys and pointers. */
2417 union xfs_btree_key *lkp; /* left btree key */
2418 union xfs_btree_ptr *lpp; /* left address pointer */
2419 union xfs_btree_key *rkp; /* right btree key */
2420 union xfs_btree_ptr *rpp; /* right address pointer */
2421
2422 lkp = xfs_btree_key_addr(cur, src_index, left);
2423 lpp = xfs_btree_ptr_addr(cur, src_index, left);
2424 rkp = xfs_btree_key_addr(cur, 1, right);
2425 rpp = xfs_btree_ptr_addr(cur, 1, right);
2426
2427 #ifdef DEBUG
2428 for (i = src_index; i < rrecs; i++) {
2429 error = xfs_btree_check_ptr(cur, lpp, i, level);
2430 if (error)
2431 goto error0;
2432 }
2433 #endif
2434
2435 xfs_btree_copy_keys(cur, rkp, lkp, rrecs);
2436 xfs_btree_copy_ptrs(cur, rpp, lpp, rrecs);
2437
2438 xfs_btree_log_keys(cur, rbp, 1, rrecs);
2439 xfs_btree_log_ptrs(cur, rbp, 1, rrecs);
2440
2441 /* Grab the keys to the entries moved to the right block */
2442 xfs_btree_copy_keys(cur, key, rkp, 1);
2443 } else {
2444 /* It's a leaf. Move records. */
2445 union xfs_btree_rec *lrp; /* left record pointer */
2446 union xfs_btree_rec *rrp; /* right record pointer */
2447
2448 lrp = xfs_btree_rec_addr(cur, src_index, left);
2449 rrp = xfs_btree_rec_addr(cur, 1, right);
2450
2451 xfs_btree_copy_recs(cur, rrp, lrp, rrecs);
2452 xfs_btree_log_recs(cur, rbp, 1, rrecs);
2453
2454 cur->bc_ops->init_key_from_rec(key,
2455 xfs_btree_rec_addr(cur, 1, right));
2456 }
2457
2458
2459 /*
2460 * Find the left block number by looking in the buffer.
2461 * Adjust numrecs, sibling pointers.
2462 */
2463 xfs_btree_get_sibling(cur, left, &rrptr, XFS_BB_RIGHTSIB);
2464 xfs_btree_set_sibling(cur, right, &rrptr, XFS_BB_RIGHTSIB);
2465 xfs_btree_set_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
2466 xfs_btree_set_sibling(cur, left, &rptr, XFS_BB_RIGHTSIB);
2467
2468 lrecs -= rrecs;
2469 xfs_btree_set_numrecs(left, lrecs);
2470 xfs_btree_set_numrecs(right, xfs_btree_get_numrecs(right) + rrecs);
2471
2472 xfs_btree_log_block(cur, rbp, XFS_BB_ALL_BITS);
2473 xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS | XFS_BB_RIGHTSIB);
2474
2475 /*
2476 * If there's a block to the new block's right, make that block
2477 * point back to right instead of to left.
2478 */
2479 if (!xfs_btree_ptr_is_null(cur, &rrptr)) {
2480 error = xfs_btree_read_buf_block(cur, &rrptr,
2481 0, &rrblock, &rrbp);
2482 if (error)
2483 goto error0;
2484 xfs_btree_set_sibling(cur, rrblock, &rptr, XFS_BB_LEFTSIB);
2485 xfs_btree_log_block(cur, rrbp, XFS_BB_LEFTSIB);
2486 }
2487 /*
2488 * If the cursor is really in the right block, move it there.
2489 * If it's just pointing past the last entry in left, then we'll
2490 * insert there, so don't change anything in that case.
2491 */
2492 if (cur->bc_ptrs[level] > lrecs + 1) {
2493 xfs_btree_setbuf(cur, level, rbp);
2494 cur->bc_ptrs[level] -= lrecs;
2495 }
2496 /*
2497 * If there are more levels, we'll need another cursor which refers
2498 * the right block, no matter where this cursor was.
2499 */
2500 if (level + 1 < cur->bc_nlevels) {
2501 error = xfs_btree_dup_cursor(cur, curp);
2502 if (error)
2503 goto error0;
2504 (*curp)->bc_ptrs[level + 1]++;
2505 }
2506 *ptrp = rptr;
2507 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2508 *stat = 1;
2509 return 0;
2510 out0:
2511 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2512 *stat = 0;
2513 return 0;
2514
2515 error0:
2516 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2517 return error;
2518 }
2519
2520 struct xfs_btree_split_args {
2521 struct xfs_btree_cur *cur;
2522 int level;
2523 union xfs_btree_ptr *ptrp;
2524 union xfs_btree_key *key;
2525 struct xfs_btree_cur **curp;
2526 int *stat; /* success/failure */
2527 int result;
2528 bool kswapd; /* allocation in kswapd context */
2529 struct completion *done;
2530 struct work_struct work;
2531 };
2532
2533 /*
2534 * Stack switching interfaces for allocation
2535 */
2536 static void
2537 xfs_btree_split_worker(
2538 struct work_struct *work)
2539 {
2540 struct xfs_btree_split_args *args = container_of(work,
2541 struct xfs_btree_split_args, work);
2542 unsigned long pflags;
2543 unsigned long new_pflags = PF_FSTRANS;
2544
2545 /*
2546 * we are in a transaction context here, but may also be doing work
2547 * in kswapd context, and hence we may need to inherit that state
2548 * temporarily to ensure that we don't block waiting for memory reclaim
2549 * in any way.
2550 */
2551 if (args->kswapd)
2552 new_pflags |= PF_MEMALLOC | PF_SWAPWRITE | PF_KSWAPD;
2553
2554 current_set_flags_nested(&pflags, new_pflags);
2555
2556 args->result = __xfs_btree_split(args->cur, args->level, args->ptrp,
2557 args->key, args->curp, args->stat);
2558 complete(args->done);
2559
2560 current_restore_flags_nested(&pflags, new_pflags);
2561 }
2562
2563 /*
2564 * BMBT split requests often come in with little stack to work on. Push
2565 * them off to a worker thread so there is lots of stack to use. For the other
2566 * btree types, just call directly to avoid the context switch overhead here.
2567 */
2568 STATIC int /* error */
2569 xfs_btree_split(
2570 struct xfs_btree_cur *cur,
2571 int level,
2572 union xfs_btree_ptr *ptrp,
2573 union xfs_btree_key *key,
2574 struct xfs_btree_cur **curp,
2575 int *stat) /* success/failure */
2576 {
2577 struct xfs_btree_split_args args;
2578 DECLARE_COMPLETION_ONSTACK(done);
2579
2580 if (cur->bc_btnum != XFS_BTNUM_BMAP)
2581 return __xfs_btree_split(cur, level, ptrp, key, curp, stat);
2582
2583 args.cur = cur;
2584 args.level = level;
2585 args.ptrp = ptrp;
2586 args.key = key;
2587 args.curp = curp;
2588 args.stat = stat;
2589 args.done = &done;
2590 args.kswapd = current_is_kswapd();
2591 INIT_WORK_ONSTACK(&args.work, xfs_btree_split_worker);
2592 queue_work(xfs_alloc_wq, &args.work);
2593 wait_for_completion(&done);
2594 destroy_work_on_stack(&args.work);
2595 return args.result;
2596 }
2597
2598
2599 /*
2600 * Copy the old inode root contents into a real block and make the
2601 * broot point to it.
2602 */
2603 int /* error */
2604 xfs_btree_new_iroot(
2605 struct xfs_btree_cur *cur, /* btree cursor */
2606 int *logflags, /* logging flags for inode */
2607 int *stat) /* return status - 0 fail */
2608 {
2609 struct xfs_buf *cbp; /* buffer for cblock */
2610 struct xfs_btree_block *block; /* btree block */
2611 struct xfs_btree_block *cblock; /* child btree block */
2612 union xfs_btree_key *ckp; /* child key pointer */
2613 union xfs_btree_ptr *cpp; /* child ptr pointer */
2614 union xfs_btree_key *kp; /* pointer to btree key */
2615 union xfs_btree_ptr *pp; /* pointer to block addr */
2616 union xfs_btree_ptr nptr; /* new block addr */
2617 int level; /* btree level */
2618 int error; /* error return code */
2619 #ifdef DEBUG
2620 int i; /* loop counter */
2621 #endif
2622
2623 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2624 XFS_BTREE_STATS_INC(cur, newroot);
2625
2626 ASSERT(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE);
2627
2628 level = cur->bc_nlevels - 1;
2629
2630 block = xfs_btree_get_iroot(cur);
2631 pp = xfs_btree_ptr_addr(cur, 1, block);
2632
2633 /* Allocate the new block. If we can't do it, we're toast. Give up. */
2634 error = cur->bc_ops->alloc_block(cur, pp, &nptr, stat);
2635 if (error)
2636 goto error0;
2637 if (*stat == 0) {
2638 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2639 return 0;
2640 }
2641 XFS_BTREE_STATS_INC(cur, alloc);
2642
2643 /* Copy the root into a real block. */
2644 error = xfs_btree_get_buf_block(cur, &nptr, 0, &cblock, &cbp);
2645 if (error)
2646 goto error0;
2647
2648 /*
2649 * we can't just memcpy() the root in for CRC enabled btree blocks.
2650 * In that case have to also ensure the blkno remains correct
2651 */
2652 memcpy(cblock, block, xfs_btree_block_len(cur));
2653 if (cur->bc_flags & XFS_BTREE_CRC_BLOCKS) {
2654 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
2655 cblock->bb_u.l.bb_blkno = cpu_to_be64(cbp->b_bn);
2656 else
2657 cblock->bb_u.s.bb_blkno = cpu_to_be64(cbp->b_bn);
2658 }
2659
2660 be16_add_cpu(&block->bb_level, 1);
2661 xfs_btree_set_numrecs(block, 1);
2662 cur->bc_nlevels++;
2663 cur->bc_ptrs[level + 1] = 1;
2664
2665 kp = xfs_btree_key_addr(cur, 1, block);
2666 ckp = xfs_btree_key_addr(cur, 1, cblock);
2667 xfs_btree_copy_keys(cur, ckp, kp, xfs_btree_get_numrecs(cblock));
2668
2669 cpp = xfs_btree_ptr_addr(cur, 1, cblock);
2670 #ifdef DEBUG
2671 for (i = 0; i < be16_to_cpu(cblock->bb_numrecs); i++) {
2672 error = xfs_btree_check_ptr(cur, pp, i, level);
2673 if (error)
2674 goto error0;
2675 }
2676 #endif
2677 xfs_btree_copy_ptrs(cur, cpp, pp, xfs_btree_get_numrecs(cblock));
2678
2679 #ifdef DEBUG
2680 error = xfs_btree_check_ptr(cur, &nptr, 0, level);
2681 if (error)
2682 goto error0;
2683 #endif
2684 xfs_btree_copy_ptrs(cur, pp, &nptr, 1);
2685
2686 xfs_iroot_realloc(cur->bc_private.b.ip,
2687 1 - xfs_btree_get_numrecs(cblock),
2688 cur->bc_private.b.whichfork);
2689
2690 xfs_btree_setbuf(cur, level, cbp);
2691
2692 /*
2693 * Do all this logging at the end so that
2694 * the root is at the right level.
2695 */
2696 xfs_btree_log_block(cur, cbp, XFS_BB_ALL_BITS);
2697 xfs_btree_log_keys(cur, cbp, 1, be16_to_cpu(cblock->bb_numrecs));
2698 xfs_btree_log_ptrs(cur, cbp, 1, be16_to_cpu(cblock->bb_numrecs));
2699
2700 *logflags |=
2701 XFS_ILOG_CORE | xfs_ilog_fbroot(cur->bc_private.b.whichfork);
2702 *stat = 1;
2703 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2704 return 0;
2705 error0:
2706 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2707 return error;
2708 }
2709
2710 /*
2711 * Allocate a new root block, fill it in.
2712 */
2713 STATIC int /* error */
2714 xfs_btree_new_root(
2715 struct xfs_btree_cur *cur, /* btree cursor */
2716 int *stat) /* success/failure */
2717 {
2718 struct xfs_btree_block *block; /* one half of the old root block */
2719 struct xfs_buf *bp; /* buffer containing block */
2720 int error; /* error return value */
2721 struct xfs_buf *lbp; /* left buffer pointer */
2722 struct xfs_btree_block *left; /* left btree block */
2723 struct xfs_buf *nbp; /* new (root) buffer */
2724 struct xfs_btree_block *new; /* new (root) btree block */
2725 int nptr; /* new value for key index, 1 or 2 */
2726 struct xfs_buf *rbp; /* right buffer pointer */
2727 struct xfs_btree_block *right; /* right btree block */
2728 union xfs_btree_ptr rptr;
2729 union xfs_btree_ptr lptr;
2730
2731 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2732 XFS_BTREE_STATS_INC(cur, newroot);
2733
2734 /* initialise our start point from the cursor */
2735 cur->bc_ops->init_ptr_from_cur(cur, &rptr);
2736
2737 /* Allocate the new block. If we can't do it, we're toast. Give up. */
2738 error = cur->bc_ops->alloc_block(cur, &rptr, &lptr, stat);
2739 if (error)
2740 goto error0;
2741 if (*stat == 0)
2742 goto out0;
2743 XFS_BTREE_STATS_INC(cur, alloc);
2744
2745 /* Set up the new block. */
2746 error = xfs_btree_get_buf_block(cur, &lptr, 0, &new, &nbp);
2747 if (error)
2748 goto error0;
2749
2750 /* Set the root in the holding structure increasing the level by 1. */
2751 cur->bc_ops->set_root(cur, &lptr, 1);
2752
2753 /*
2754 * At the previous root level there are now two blocks: the old root,
2755 * and the new block generated when it was split. We don't know which
2756 * one the cursor is pointing at, so we set up variables "left" and
2757 * "right" for each case.
2758 */
2759 block = xfs_btree_get_block(cur, cur->bc_nlevels - 1, &bp);
2760
2761 #ifdef DEBUG
2762 error = xfs_btree_check_block(cur, block, cur->bc_nlevels - 1, bp);
2763 if (error)
2764 goto error0;
2765 #endif
2766
2767 xfs_btree_get_sibling(cur, block, &rptr, XFS_BB_RIGHTSIB);
2768 if (!xfs_btree_ptr_is_null(cur, &rptr)) {
2769 /* Our block is left, pick up the right block. */
2770 lbp = bp;
2771 xfs_btree_buf_to_ptr(cur, lbp, &lptr);
2772 left = block;
2773 error = xfs_btree_read_buf_block(cur, &rptr, 0, &right, &rbp);
2774 if (error)
2775 goto error0;
2776 bp = rbp;
2777 nptr = 1;
2778 } else {
2779 /* Our block is right, pick up the left block. */
2780 rbp = bp;
2781 xfs_btree_buf_to_ptr(cur, rbp, &rptr);
2782 right = block;
2783 xfs_btree_get_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
2784 error = xfs_btree_read_buf_block(cur, &lptr, 0, &left, &lbp);
2785 if (error)
2786 goto error0;
2787 bp = lbp;
2788 nptr = 2;
2789 }
2790 /* Fill in the new block's btree header and log it. */
2791 xfs_btree_init_block_cur(cur, nbp, cur->bc_nlevels, 2);
2792 xfs_btree_log_block(cur, nbp, XFS_BB_ALL_BITS);
2793 ASSERT(!xfs_btree_ptr_is_null(cur, &lptr) &&
2794 !xfs_btree_ptr_is_null(cur, &rptr));
2795
2796 /* Fill in the key data in the new root. */
2797 if (xfs_btree_get_level(left) > 0) {
2798 xfs_btree_copy_keys(cur,
2799 xfs_btree_key_addr(cur, 1, new),
2800 xfs_btree_key_addr(cur, 1, left), 1);
2801 xfs_btree_copy_keys(cur,
2802 xfs_btree_key_addr(cur, 2, new),
2803 xfs_btree_key_addr(cur, 1, right), 1);
2804 } else {
2805 cur->bc_ops->init_key_from_rec(
2806 xfs_btree_key_addr(cur, 1, new),
2807 xfs_btree_rec_addr(cur, 1, left));
2808 cur->bc_ops->init_key_from_rec(
2809 xfs_btree_key_addr(cur, 2, new),
2810 xfs_btree_rec_addr(cur, 1, right));
2811 }
2812 xfs_btree_log_keys(cur, nbp, 1, 2);
2813
2814 /* Fill in the pointer data in the new root. */
2815 xfs_btree_copy_ptrs(cur,
2816 xfs_btree_ptr_addr(cur, 1, new), &lptr, 1);
2817 xfs_btree_copy_ptrs(cur,
2818 xfs_btree_ptr_addr(cur, 2, new), &rptr, 1);
2819 xfs_btree_log_ptrs(cur, nbp, 1, 2);
2820
2821 /* Fix up the cursor. */
2822 xfs_btree_setbuf(cur, cur->bc_nlevels, nbp);
2823 cur->bc_ptrs[cur->bc_nlevels] = nptr;
2824 cur->bc_nlevels++;
2825 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2826 *stat = 1;
2827 return 0;
2828 error0:
2829 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2830 return error;
2831 out0:
2832 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2833 *stat = 0;
2834 return 0;
2835 }
2836
2837 STATIC int
2838 xfs_btree_make_block_unfull(
2839 struct xfs_btree_cur *cur, /* btree cursor */
2840 int level, /* btree level */
2841 int numrecs,/* # of recs in block */
2842 int *oindex,/* old tree index */
2843 int *index, /* new tree index */
2844 union xfs_btree_ptr *nptr, /* new btree ptr */
2845 struct xfs_btree_cur **ncur, /* new btree cursor */
2846 union xfs_btree_rec *nrec, /* new record */
2847 int *stat)
2848 {
2849 union xfs_btree_key key; /* new btree key value */
2850 int error = 0;
2851
2852 if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
2853 level == cur->bc_nlevels - 1) {
2854 struct xfs_inode *ip = cur->bc_private.b.ip;
2855
2856 if (numrecs < cur->bc_ops->get_dmaxrecs(cur, level)) {
2857 /* A root block that can be made bigger. */
2858 xfs_iroot_realloc(ip, 1, cur->bc_private.b.whichfork);
2859 } else {
2860 /* A root block that needs replacing */
2861 int logflags = 0;
2862
2863 error = xfs_btree_new_iroot(cur, &logflags, stat);
2864 if (error || *stat == 0)
2865 return error;
2866
2867 xfs_trans_log_inode(cur->bc_tp, ip, logflags);
2868 }
2869
2870 return 0;
2871 }
2872
2873 /* First, try shifting an entry to the right neighbor. */
2874 error = xfs_btree_rshift(cur, level, stat);
2875 if (error || *stat)
2876 return error;
2877
2878 /* Next, try shifting an entry to the left neighbor. */
2879 error = xfs_btree_lshift(cur, level, stat);
2880 if (error)
2881 return error;
2882
2883 if (*stat) {
2884 *oindex = *index = cur->bc_ptrs[level];
2885 return 0;
2886 }
2887
2888 /*
2889 * Next, try splitting the current block in half.
2890 *
2891 * If this works we have to re-set our variables because we
2892 * could be in a different block now.
2893 */
2894 error = xfs_btree_split(cur, level, nptr, &key, ncur, stat);
2895 if (error || *stat == 0)
2896 return error;
2897
2898
2899 *index = cur->bc_ptrs[level];
2900 cur->bc_ops->init_rec_from_key(&key, nrec);
2901 return 0;
2902 }
2903
2904 /*
2905 * Insert one record/level. Return information to the caller
2906 * allowing the next level up to proceed if necessary.
2907 */
2908 STATIC int
2909 xfs_btree_insrec(
2910 struct xfs_btree_cur *cur, /* btree cursor */
2911 int level, /* level to insert record at */
2912 union xfs_btree_ptr *ptrp, /* i/o: block number inserted */
2913 union xfs_btree_rec *recp, /* i/o: record data inserted */
2914 struct xfs_btree_cur **curp, /* output: new cursor replacing cur */
2915 int *stat) /* success/failure */
2916 {
2917 struct xfs_btree_block *block; /* btree block */
2918 struct xfs_buf *bp; /* buffer for block */
2919 union xfs_btree_key key; /* btree key */
2920 union xfs_btree_ptr nptr; /* new block ptr */
2921 struct xfs_btree_cur *ncur; /* new btree cursor */
2922 union xfs_btree_rec nrec; /* new record count */
2923 int optr; /* old key/record index */
2924 int ptr; /* key/record index */
2925 int numrecs;/* number of records */
2926 int error; /* error return value */
2927 #ifdef DEBUG
2928 int i;
2929 #endif
2930
2931 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2932 XFS_BTREE_TRACE_ARGIPR(cur, level, *ptrp, recp);
2933
2934 ncur = NULL;
2935
2936 /*
2937 * If we have an external root pointer, and we've made it to the
2938 * root level, allocate a new root block and we're done.
2939 */
2940 if (!(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
2941 (level >= cur->bc_nlevels)) {
2942 error = xfs_btree_new_root(cur, stat);
2943 xfs_btree_set_ptr_null(cur, ptrp);
2944
2945 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2946 return error;
2947 }
2948
2949 /* If we're off the left edge, return failure. */
2950 ptr = cur->bc_ptrs[level];
2951 if (ptr == 0) {
2952 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2953 *stat = 0;
2954 return 0;
2955 }
2956
2957 /* Make a key out of the record data to be inserted, and save it. */
2958 cur->bc_ops->init_key_from_rec(&key, recp);
2959
2960 optr = ptr;
2961
2962 XFS_BTREE_STATS_INC(cur, insrec);
2963
2964 /* Get pointers to the btree buffer and block. */
2965 block = xfs_btree_get_block(cur, level, &bp);
2966 numrecs = xfs_btree_get_numrecs(block);
2967
2968 #ifdef DEBUG
2969 error = xfs_btree_check_block(cur, block, level, bp);
2970 if (error)
2971 goto error0;
2972
2973 /* Check that the new entry is being inserted in the right place. */
2974 if (ptr <= numrecs) {
2975 if (level == 0) {
2976 ASSERT(cur->bc_ops->recs_inorder(cur, recp,
2977 xfs_btree_rec_addr(cur, ptr, block)));
2978 } else {
2979 ASSERT(cur->bc_ops->keys_inorder(cur, &key,
2980 xfs_btree_key_addr(cur, ptr, block)));
2981 }
2982 }
2983 #endif
2984
2985 /*
2986 * If the block is full, we can't insert the new entry until we
2987 * make the block un-full.
2988 */
2989 xfs_btree_set_ptr_null(cur, &nptr);
2990 if (numrecs == cur->bc_ops->get_maxrecs(cur, level)) {
2991 error = xfs_btree_make_block_unfull(cur, level, numrecs,
2992 &optr, &ptr, &nptr, &ncur, &nrec, stat);
2993 if (error || *stat == 0)
2994 goto error0;
2995 }
2996
2997 /*
2998 * The current block may have changed if the block was
2999 * previously full and we have just made space in it.
3000 */
3001 block = xfs_btree_get_block(cur, level, &bp);
3002 numrecs = xfs_btree_get_numrecs(block);
3003
3004 #ifdef DEBUG
3005 error = xfs_btree_check_block(cur, block, level, bp);
3006 if (error)
3007 return error;
3008 #endif
3009
3010 /*
3011 * At this point we know there's room for our new entry in the block
3012 * we're pointing at.
3013 */
3014 XFS_BTREE_STATS_ADD(cur, moves, numrecs - ptr + 1);
3015
3016 if (level > 0) {
3017 /* It's a nonleaf. make a hole in the keys and ptrs */
3018 union xfs_btree_key *kp;
3019 union xfs_btree_ptr *pp;
3020
3021 kp = xfs_btree_key_addr(cur, ptr, block);
3022 pp = xfs_btree_ptr_addr(cur, ptr, block);
3023
3024 #ifdef DEBUG
3025 for (i = numrecs - ptr; i >= 0; i--) {
3026 error = xfs_btree_check_ptr(cur, pp, i, level);
3027 if (error)
3028 return error;
3029 }
3030 #endif
3031
3032 xfs_btree_shift_keys(cur, kp, 1, numrecs - ptr + 1);
3033 xfs_btree_shift_ptrs(cur, pp, 1, numrecs - ptr + 1);
3034
3035 #ifdef DEBUG
3036 error = xfs_btree_check_ptr(cur, ptrp, 0, level);
3037 if (error)
3038 goto error0;
3039 #endif
3040
3041 /* Now put the new data in, bump numrecs and log it. */
3042 xfs_btree_copy_keys(cur, kp, &key, 1);
3043 xfs_btree_copy_ptrs(cur, pp, ptrp, 1);
3044 numrecs++;
3045 xfs_btree_set_numrecs(block, numrecs);
3046 xfs_btree_log_ptrs(cur, bp, ptr, numrecs);
3047 xfs_btree_log_keys(cur, bp, ptr, numrecs);
3048 #ifdef DEBUG
3049 if (ptr < numrecs) {
3050 ASSERT(cur->bc_ops->keys_inorder(cur, kp,
3051 xfs_btree_key_addr(cur, ptr + 1, block)));
3052 }
3053 #endif
3054 } else {
3055 /* It's a leaf. make a hole in the records */
3056 union xfs_btree_rec *rp;
3057
3058 rp = xfs_btree_rec_addr(cur, ptr, block);
3059
3060 xfs_btree_shift_recs(cur, rp, 1, numrecs - ptr + 1);
3061
3062 /* Now put the new data in, bump numrecs and log it. */
3063 xfs_btree_copy_recs(cur, rp, recp, 1);
3064 xfs_btree_set_numrecs(block, ++numrecs);
3065 xfs_btree_log_recs(cur, bp, ptr, numrecs);
3066 #ifdef DEBUG
3067 if (ptr < numrecs) {
3068 ASSERT(cur->bc_ops->recs_inorder(cur, rp,
3069 xfs_btree_rec_addr(cur, ptr + 1, block)));
3070 }
3071 #endif
3072 }
3073
3074 /* Log the new number of records in the btree header. */
3075 xfs_btree_log_block(cur, bp, XFS_BB_NUMRECS);
3076
3077 /* If we inserted at the start of a block, update the parents' keys. */
3078 if (optr == 1) {
3079 error = xfs_btree_updkey(cur, &key, level + 1);
3080 if (error)
3081 goto error0;
3082 }
3083
3084 /*
3085 * If we are tracking the last record in the tree and
3086 * we are at the far right edge of the tree, update it.
3087 */
3088 if (xfs_btree_is_lastrec(cur, block, level)) {
3089 cur->bc_ops->update_lastrec(cur, block, recp,
3090 ptr, LASTREC_INSREC);
3091 }
3092
3093 /*
3094 * Return the new block number, if any.
3095 * If there is one, give back a record value and a cursor too.
3096 */
3097 *ptrp = nptr;
3098 if (!xfs_btree_ptr_is_null(cur, &nptr)) {
3099 *recp = nrec;
3100 *curp = ncur;
3101 }
3102
3103 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3104 *stat = 1;
3105 return 0;
3106
3107 error0:
3108 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3109 return error;
3110 }
3111
3112 /*
3113 * Insert the record at the point referenced by cur.
3114 *
3115 * A multi-level split of the tree on insert will invalidate the original
3116 * cursor. All callers of this function should assume that the cursor is
3117 * no longer valid and revalidate it.
3118 */
3119 int
3120 xfs_btree_insert(
3121 struct xfs_btree_cur *cur,
3122 int *stat)
3123 {
3124 int error; /* error return value */
3125 int i; /* result value, 0 for failure */
3126 int level; /* current level number in btree */
3127 union xfs_btree_ptr nptr; /* new block number (split result) */
3128 struct xfs_btree_cur *ncur; /* new cursor (split result) */
3129 struct xfs_btree_cur *pcur; /* previous level's cursor */
3130 union xfs_btree_rec rec; /* record to insert */
3131
3132 level = 0;
3133 ncur = NULL;
3134 pcur = cur;
3135
3136 xfs_btree_set_ptr_null(cur, &nptr);
3137 cur->bc_ops->init_rec_from_cur(cur, &rec);
3138
3139 /*
3140 * Loop going up the tree, starting at the leaf level.
3141 * Stop when we don't get a split block, that must mean that
3142 * the insert is finished with this level.
3143 */
3144 do {
3145 /*
3146 * Insert nrec/nptr into this level of the tree.
3147 * Note if we fail, nptr will be null.
3148 */
3149 error = xfs_btree_insrec(pcur, level, &nptr, &rec, &ncur, &i);
3150 if (error) {
3151 if (pcur != cur)
3152 xfs_btree_del_cursor(pcur, XFS_BTREE_ERROR);
3153 goto error0;
3154 }
3155
3156 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0);
3157 level++;
3158
3159 /*
3160 * See if the cursor we just used is trash.
3161 * Can't trash the caller's cursor, but otherwise we should
3162 * if ncur is a new cursor or we're about to be done.
3163 */
3164 if (pcur != cur &&
3165 (ncur || xfs_btree_ptr_is_null(cur, &nptr))) {
3166 /* Save the state from the cursor before we trash it */
3167 if (cur->bc_ops->update_cursor)
3168 cur->bc_ops->update_cursor(pcur, cur);
3169 cur->bc_nlevels = pcur->bc_nlevels;
3170 xfs_btree_del_cursor(pcur, XFS_BTREE_NOERROR);
3171 }
3172 /* If we got a new cursor, switch to it. */
3173 if (ncur) {
3174 pcur = ncur;
3175 ncur = NULL;
3176 }
3177 } while (!xfs_btree_ptr_is_null(cur, &nptr));
3178
3179 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3180 *stat = i;
3181 return 0;
3182 error0:
3183 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3184 return error;
3185 }
3186
3187 /*
3188 * Try to merge a non-leaf block back into the inode root.
3189 *
3190 * Note: the killroot names comes from the fact that we're effectively
3191 * killing the old root block. But because we can't just delete the
3192 * inode we have to copy the single block it was pointing to into the
3193 * inode.
3194 */
3195 STATIC int
3196 xfs_btree_kill_iroot(
3197 struct xfs_btree_cur *cur)
3198 {
3199 int whichfork = cur->bc_private.b.whichfork;
3200 struct xfs_inode *ip = cur->bc_private.b.ip;
3201 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
3202 struct xfs_btree_block *block;
3203 struct xfs_btree_block *cblock;
3204 union xfs_btree_key *kp;
3205 union xfs_btree_key *ckp;
3206 union xfs_btree_ptr *pp;
3207 union xfs_btree_ptr *cpp;
3208 struct xfs_buf *cbp;
3209 int level;
3210 int index;
3211 int numrecs;
3212 #ifdef DEBUG
3213 union xfs_btree_ptr ptr;
3214 int i;
3215 #endif
3216
3217 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
3218
3219 ASSERT(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE);
3220 ASSERT(cur->bc_nlevels > 1);
3221
3222 /*
3223 * Don't deal with the root block needs to be a leaf case.
3224 * We're just going to turn the thing back into extents anyway.
3225 */
3226 level = cur->bc_nlevels - 1;
3227 if (level == 1)
3228 goto out0;
3229
3230 /*
3231 * Give up if the root has multiple children.
3232 */
3233 block = xfs_btree_get_iroot(cur);
3234 if (xfs_btree_get_numrecs(block) != 1)
3235 goto out0;
3236
3237 cblock = xfs_btree_get_block(cur, level - 1, &cbp);
3238 numrecs = xfs_btree_get_numrecs(cblock);
3239
3240 /*
3241 * Only do this if the next level will fit.
3242 * Then the data must be copied up to the inode,
3243 * instead of freeing the root you free the next level.
3244 */
3245 if (numrecs > cur->bc_ops->get_dmaxrecs(cur, level))
3246 goto out0;
3247
3248 XFS_BTREE_STATS_INC(cur, killroot);
3249
3250 #ifdef DEBUG
3251 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_LEFTSIB);
3252 ASSERT(xfs_btree_ptr_is_null(cur, &ptr));
3253 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
3254 ASSERT(xfs_btree_ptr_is_null(cur, &ptr));
3255 #endif
3256
3257 index = numrecs - cur->bc_ops->get_maxrecs(cur, level);
3258 if (index) {
3259 xfs_iroot_realloc(cur->bc_private.b.ip, index,
3260 cur->bc_private.b.whichfork);
3261 block = ifp->if_broot;
3262 }
3263
3264 be16_add_cpu(&block->bb_numrecs, index);
3265 ASSERT(block->bb_numrecs == cblock->bb_numrecs);
3266
3267 kp = xfs_btree_key_addr(cur, 1, block);
3268 ckp = xfs_btree_key_addr(cur, 1, cblock);
3269 xfs_btree_copy_keys(cur, kp, ckp, numrecs);
3270
3271 pp = xfs_btree_ptr_addr(cur, 1, block);
3272 cpp = xfs_btree_ptr_addr(cur, 1, cblock);
3273 #ifdef DEBUG
3274 for (i = 0; i < numrecs; i++) {
3275 int error;
3276
3277 error = xfs_btree_check_ptr(cur, cpp, i, level - 1);
3278 if (error) {
3279 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3280 return error;
3281 }
3282 }
3283 #endif
3284 xfs_btree_copy_ptrs(cur, pp, cpp, numrecs);
3285
3286 cur->bc_ops->free_block(cur, cbp);
3287 XFS_BTREE_STATS_INC(cur, free);
3288
3289 cur->bc_bufs[level - 1] = NULL;
3290 be16_add_cpu(&block->bb_level, -1);
3291 xfs_trans_log_inode(cur->bc_tp, ip,
3292 XFS_ILOG_CORE | xfs_ilog_fbroot(cur->bc_private.b.whichfork));
3293 cur->bc_nlevels--;
3294 out0:
3295 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3296 return 0;
3297 }
3298
3299 /*
3300 * Kill the current root node, and replace it with it's only child node.
3301 */
3302 STATIC int
3303 xfs_btree_kill_root(
3304 struct xfs_btree_cur *cur,
3305 struct xfs_buf *bp,
3306 int level,
3307 union xfs_btree_ptr *newroot)
3308 {
3309 int error;
3310
3311 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
3312 XFS_BTREE_STATS_INC(cur, killroot);
3313
3314 /*
3315 * Update the root pointer, decreasing the level by 1 and then
3316 * free the old root.
3317 */
3318 cur->bc_ops->set_root(cur, newroot, -1);
3319
3320 error = cur->bc_ops->free_block(cur, bp);
3321 if (error) {
3322 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3323 return error;
3324 }
3325
3326 XFS_BTREE_STATS_INC(cur, free);
3327
3328 cur->bc_bufs[level] = NULL;
3329 cur->bc_ra[level] = 0;
3330 cur->bc_nlevels--;
3331
3332 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3333 return 0;
3334 }
3335
3336 STATIC int
3337 xfs_btree_dec_cursor(
3338 struct xfs_btree_cur *cur,
3339 int level,
3340 int *stat)
3341 {
3342 int error;
3343 int i;
3344
3345 if (level > 0) {
3346 error = xfs_btree_decrement(cur, level, &i);
3347 if (error)
3348 return error;
3349 }
3350
3351 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3352 *stat = 1;
3353 return 0;
3354 }
3355
3356 /*
3357 * Single level of the btree record deletion routine.
3358 * Delete record pointed to by cur/level.
3359 * Remove the record from its block then rebalance the tree.
3360 * Return 0 for error, 1 for done, 2 to go on to the next level.
3361 */
3362 STATIC int /* error */
3363 xfs_btree_delrec(
3364 struct xfs_btree_cur *cur, /* btree cursor */
3365 int level, /* level removing record from */
3366 int *stat) /* fail/done/go-on */
3367 {
3368 struct xfs_btree_block *block; /* btree block */
3369 union xfs_btree_ptr cptr; /* current block ptr */
3370 struct xfs_buf *bp; /* buffer for block */
3371 int error; /* error return value */
3372 int i; /* loop counter */
3373 union xfs_btree_key key; /* storage for keyp */
3374 union xfs_btree_key *keyp = &key; /* passed to the next level */
3375 union xfs_btree_ptr lptr; /* left sibling block ptr */
3376 struct xfs_buf *lbp; /* left buffer pointer */
3377 struct xfs_btree_block *left; /* left btree block */
3378 int lrecs = 0; /* left record count */
3379 int ptr; /* key/record index */
3380 union xfs_btree_ptr rptr; /* right sibling block ptr */
3381 struct xfs_buf *rbp; /* right buffer pointer */
3382 struct xfs_btree_block *right; /* right btree block */
3383 struct xfs_btree_block *rrblock; /* right-right btree block */
3384 struct xfs_buf *rrbp; /* right-right buffer pointer */
3385 int rrecs = 0; /* right record count */
3386 struct xfs_btree_cur *tcur; /* temporary btree cursor */
3387 int numrecs; /* temporary numrec count */
3388
3389 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
3390 XFS_BTREE_TRACE_ARGI(cur, level);
3391
3392 tcur = NULL;
3393
3394 /* Get the index of the entry being deleted, check for nothing there. */
3395 ptr = cur->bc_ptrs[level];
3396 if (ptr == 0) {
3397 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3398 *stat = 0;
3399 return 0;
3400 }
3401
3402 /* Get the buffer & block containing the record or key/ptr. */
3403 block = xfs_btree_get_block(cur, level, &bp);
3404 numrecs = xfs_btree_get_numrecs(block);
3405
3406 #ifdef DEBUG
3407 error = xfs_btree_check_block(cur, block, level, bp);
3408 if (error)
3409 goto error0;
3410 #endif
3411
3412 /* Fail if we're off the end of the block. */
3413 if (ptr > numrecs) {
3414 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3415 *stat = 0;
3416 return 0;
3417 }
3418
3419 XFS_BTREE_STATS_INC(cur, delrec);
3420 XFS_BTREE_STATS_ADD(cur, moves, numrecs - ptr);
3421
3422 /* Excise the entries being deleted. */
3423 if (level > 0) {
3424 /* It's a nonleaf. operate on keys and ptrs */
3425 union xfs_btree_key *lkp;
3426 union xfs_btree_ptr *lpp;
3427
3428 lkp = xfs_btree_key_addr(cur, ptr + 1, block);
3429 lpp = xfs_btree_ptr_addr(cur, ptr + 1, block);
3430
3431 #ifdef DEBUG
3432 for (i = 0; i < numrecs - ptr; i++) {
3433 error = xfs_btree_check_ptr(cur, lpp, i, level);
3434 if (error)
3435 goto error0;
3436 }
3437 #endif
3438
3439 if (ptr < numrecs) {
3440 xfs_btree_shift_keys(cur, lkp, -1, numrecs - ptr);
3441 xfs_btree_shift_ptrs(cur, lpp, -1, numrecs - ptr);
3442 xfs_btree_log_keys(cur, bp, ptr, numrecs - 1);
3443 xfs_btree_log_ptrs(cur, bp, ptr, numrecs - 1);
3444 }
3445
3446 /*
3447 * If it's the first record in the block, we'll need to pass a
3448 * key up to the next level (updkey).
3449 */
3450 if (ptr == 1)
3451 keyp = xfs_btree_key_addr(cur, 1, block);
3452 } else {
3453 /* It's a leaf. operate on records */
3454 if (ptr < numrecs) {
3455 xfs_btree_shift_recs(cur,
3456 xfs_btree_rec_addr(cur, ptr + 1, block),
3457 -1, numrecs - ptr);
3458 xfs_btree_log_recs(cur, bp, ptr, numrecs - 1);
3459 }
3460
3461 /*
3462 * If it's the first record in the block, we'll need a key
3463 * structure to pass up to the next level (updkey).
3464 */
3465 if (ptr == 1) {
3466 cur->bc_ops->init_key_from_rec(&key,
3467 xfs_btree_rec_addr(cur, 1, block));
3468 keyp = &key;
3469 }
3470 }
3471
3472 /*
3473 * Decrement and log the number of entries in the block.
3474 */
3475 xfs_btree_set_numrecs(block, --numrecs);
3476 xfs_btree_log_block(cur, bp, XFS_BB_NUMRECS);
3477
3478 /*
3479 * If we are tracking the last record in the tree and
3480 * we are at the far right edge of the tree, update it.
3481 */
3482 if (xfs_btree_is_lastrec(cur, block, level)) {
3483 cur->bc_ops->update_lastrec(cur, block, NULL,
3484 ptr, LASTREC_DELREC);
3485 }
3486
3487 /*
3488 * We're at the root level. First, shrink the root block in-memory.
3489 * Try to get rid of the next level down. If we can't then there's
3490 * nothing left to do.
3491 */
3492 if (level == cur->bc_nlevels - 1) {
3493 if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) {
3494 xfs_iroot_realloc(cur->bc_private.b.ip, -1,
3495 cur->bc_private.b.whichfork);
3496
3497 error = xfs_btree_kill_iroot(cur);
3498 if (error)
3499 goto error0;
3500
3501 error = xfs_btree_dec_cursor(cur, level, stat);
3502 if (error)
3503 goto error0;
3504 *stat = 1;
3505 return 0;
3506 }
3507
3508 /*
3509 * If this is the root level, and there's only one entry left,
3510 * and it's NOT the leaf level, then we can get rid of this
3511 * level.
3512 */
3513 if (numrecs == 1 && level > 0) {
3514 union xfs_btree_ptr *pp;
3515 /*
3516 * pp is still set to the first pointer in the block.
3517 * Make it the new root of the btree.
3518 */
3519 pp = xfs_btree_ptr_addr(cur, 1, block);
3520 error = xfs_btree_kill_root(cur, bp, level, pp);
3521 if (error)
3522 goto error0;
3523 } else if (level > 0) {
3524 error = xfs_btree_dec_cursor(cur, level, stat);
3525 if (error)
3526 goto error0;
3527 }
3528 *stat = 1;
3529 return 0;
3530 }
3531
3532 /*
3533 * If we deleted the leftmost entry in the block, update the
3534 * key values above us in the tree.
3535 */
3536 if (ptr == 1) {
3537 error = xfs_btree_updkey(cur, keyp, level + 1);
3538 if (error)
3539 goto error0;
3540 }
3541
3542 /*
3543 * If the number of records remaining in the block is at least
3544 * the minimum, we're done.
3545 */
3546 if (numrecs >= cur->bc_ops->get_minrecs(cur, level)) {
3547 error = xfs_btree_dec_cursor(cur, level, stat);
3548 if (error)
3549 goto error0;
3550 return 0;
3551 }
3552
3553 /*
3554 * Otherwise, we have to move some records around to keep the
3555 * tree balanced. Look at the left and right sibling blocks to
3556 * see if we can re-balance by moving only one record.
3557 */
3558 xfs_btree_get_sibling(cur, block, &rptr, XFS_BB_RIGHTSIB);
3559 xfs_btree_get_sibling(cur, block, &lptr, XFS_BB_LEFTSIB);
3560
3561 if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) {
3562 /*
3563 * One child of root, need to get a chance to copy its contents
3564 * into the root and delete it. Can't go up to next level,
3565 * there's nothing to delete there.
3566 */
3567 if (xfs_btree_ptr_is_null(cur, &rptr) &&
3568 xfs_btree_ptr_is_null(cur, &lptr) &&
3569 level == cur->bc_nlevels - 2) {
3570 error = xfs_btree_kill_iroot(cur);
3571 if (!error)
3572 error = xfs_btree_dec_cursor(cur, level, stat);
3573 if (error)
3574 goto error0;
3575 return 0;
3576 }
3577 }
3578
3579 ASSERT(!xfs_btree_ptr_is_null(cur, &rptr) ||
3580 !xfs_btree_ptr_is_null(cur, &lptr));
3581
3582 /*
3583 * Duplicate the cursor so our btree manipulations here won't
3584 * disrupt the next level up.
3585 */
3586 error = xfs_btree_dup_cursor(cur, &tcur);
3587 if (error)
3588 goto error0;
3589
3590 /*
3591 * If there's a right sibling, see if it's ok to shift an entry
3592 * out of it.
3593 */
3594 if (!xfs_btree_ptr_is_null(cur, &rptr)) {
3595 /*
3596 * Move the temp cursor to the last entry in the next block.
3597 * Actually any entry but the first would suffice.
3598 */
3599 i = xfs_btree_lastrec(tcur, level);
3600 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0);
3601
3602 error = xfs_btree_increment(tcur, level, &i);
3603 if (error)
3604 goto error0;
3605 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0);
3606
3607 i = xfs_btree_lastrec(tcur, level);
3608 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0);
3609
3610 /* Grab a pointer to the block. */
3611 right = xfs_btree_get_block(tcur, level, &rbp);
3612 #ifdef DEBUG
3613 error = xfs_btree_check_block(tcur, right, level, rbp);
3614 if (error)
3615 goto error0;
3616 #endif
3617 /* Grab the current block number, for future use. */
3618 xfs_btree_get_sibling(tcur, right, &cptr, XFS_BB_LEFTSIB);
3619
3620 /*
3621 * If right block is full enough so that removing one entry
3622 * won't make it too empty, and left-shifting an entry out
3623 * of right to us works, we're done.
3624 */
3625 if (xfs_btree_get_numrecs(right) - 1 >=
3626 cur->bc_ops->get_minrecs(tcur, level)) {
3627 error = xfs_btree_lshift(tcur, level, &i);
3628 if (error)
3629 goto error0;
3630 if (i) {
3631 ASSERT(xfs_btree_get_numrecs(block) >=
3632 cur->bc_ops->get_minrecs(tcur, level));
3633
3634 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
3635 tcur = NULL;
3636
3637 error = xfs_btree_dec_cursor(cur, level, stat);
3638 if (error)
3639 goto error0;
3640 return 0;
3641 }
3642 }
3643
3644 /*
3645 * Otherwise, grab the number of records in right for
3646 * future reference, and fix up the temp cursor to point
3647 * to our block again (last record).
3648 */
3649 rrecs = xfs_btree_get_numrecs(right);
3650 if (!xfs_btree_ptr_is_null(cur, &lptr)) {
3651 i = xfs_btree_firstrec(tcur, level);
3652 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0);
3653
3654 error = xfs_btree_decrement(tcur, level, &i);
3655 if (error)
3656 goto error0;
3657 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0);
3658 }
3659 }
3660
3661 /*
3662 * If there's a left sibling, see if it's ok to shift an entry
3663 * out of it.
3664 */
3665 if (!xfs_btree_ptr_is_null(cur, &lptr)) {
3666 /*
3667 * Move the temp cursor to the first entry in the
3668 * previous block.
3669 */
3670 i = xfs_btree_firstrec(tcur, level);
3671 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0);
3672
3673 error = xfs_btree_decrement(tcur, level, &i);
3674 if (error)
3675 goto error0;
3676 i = xfs_btree_firstrec(tcur, level);
3677 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0);
3678
3679 /* Grab a pointer to the block. */
3680 left = xfs_btree_get_block(tcur, level, &lbp);
3681 #ifdef DEBUG
3682 error = xfs_btree_check_block(cur, left, level, lbp);
3683 if (error)
3684 goto error0;
3685 #endif
3686 /* Grab the current block number, for future use. */
3687 xfs_btree_get_sibling(tcur, left, &cptr, XFS_BB_RIGHTSIB);
3688
3689 /*
3690 * If left block is full enough so that removing one entry
3691 * won't make it too empty, and right-shifting an entry out
3692 * of left to us works, we're done.
3693 */
3694 if (xfs_btree_get_numrecs(left) - 1 >=
3695 cur->bc_ops->get_minrecs(tcur, level)) {
3696 error = xfs_btree_rshift(tcur, level, &i);
3697 if (error)
3698 goto error0;
3699 if (i) {
3700 ASSERT(xfs_btree_get_numrecs(block) >=
3701 cur->bc_ops->get_minrecs(tcur, level));
3702 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
3703 tcur = NULL;
3704 if (level == 0)
3705 cur->bc_ptrs[0]++;
3706 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3707 *stat = 1;
3708 return 0;
3709 }
3710 }
3711
3712 /*
3713 * Otherwise, grab the number of records in right for
3714 * future reference.
3715 */
3716 lrecs = xfs_btree_get_numrecs(left);
3717 }
3718
3719 /* Delete the temp cursor, we're done with it. */
3720 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
3721 tcur = NULL;
3722
3723 /* If here, we need to do a join to keep the tree balanced. */
3724 ASSERT(!xfs_btree_ptr_is_null(cur, &cptr));
3725
3726 if (!xfs_btree_ptr_is_null(cur, &lptr) &&
3727 lrecs + xfs_btree_get_numrecs(block) <=
3728 cur->bc_ops->get_maxrecs(cur, level)) {
3729 /*
3730 * Set "right" to be the starting block,
3731 * "left" to be the left neighbor.
3732 */
3733 rptr = cptr;
3734 right = block;
3735 rbp = bp;
3736 error = xfs_btree_read_buf_block(cur, &lptr, 0, &left, &lbp);
3737 if (error)
3738 goto error0;
3739
3740 /*
3741 * If that won't work, see if we can join with the right neighbor block.
3742 */
3743 } else if (!xfs_btree_ptr_is_null(cur, &rptr) &&
3744 rrecs + xfs_btree_get_numrecs(block) <=
3745 cur->bc_ops->get_maxrecs(cur, level)) {
3746 /*
3747 * Set "left" to be the starting block,
3748 * "right" to be the right neighbor.
3749 */
3750 lptr = cptr;
3751 left = block;
3752 lbp = bp;
3753 error = xfs_btree_read_buf_block(cur, &rptr, 0, &right, &rbp);
3754 if (error)
3755 goto error0;
3756
3757 /*
3758 * Otherwise, we can't fix the imbalance.
3759 * Just return. This is probably a logic error, but it's not fatal.
3760 */
3761 } else {
3762 error = xfs_btree_dec_cursor(cur, level, stat);
3763 if (error)
3764 goto error0;
3765 return 0;
3766 }
3767
3768 rrecs = xfs_btree_get_numrecs(right);
3769 lrecs = xfs_btree_get_numrecs(left);
3770
3771 /*
3772 * We're now going to join "left" and "right" by moving all the stuff
3773 * in "right" to "left" and deleting "right".
3774 */
3775 XFS_BTREE_STATS_ADD(cur, moves, rrecs);
3776 if (level > 0) {
3777 /* It's a non-leaf. Move keys and pointers. */
3778 union xfs_btree_key *lkp; /* left btree key */
3779 union xfs_btree_ptr *lpp; /* left address pointer */
3780 union xfs_btree_key *rkp; /* right btree key */
3781 union xfs_btree_ptr *rpp; /* right address pointer */
3782
3783 lkp = xfs_btree_key_addr(cur, lrecs + 1, left);
3784 lpp = xfs_btree_ptr_addr(cur, lrecs + 1, left);
3785 rkp = xfs_btree_key_addr(cur, 1, right);
3786 rpp = xfs_btree_ptr_addr(cur, 1, right);
3787 #ifdef DEBUG
3788 for (i = 1; i < rrecs; i++) {
3789 error = xfs_btree_check_ptr(cur, rpp, i, level);
3790 if (error)
3791 goto error0;
3792 }
3793 #endif
3794 xfs_btree_copy_keys(cur, lkp, rkp, rrecs);
3795 xfs_btree_copy_ptrs(cur, lpp, rpp, rrecs);
3796
3797 xfs_btree_log_keys(cur, lbp, lrecs + 1, lrecs + rrecs);
3798 xfs_btree_log_ptrs(cur, lbp, lrecs + 1, lrecs + rrecs);
3799 } else {
3800 /* It's a leaf. Move records. */
3801 union xfs_btree_rec *lrp; /* left record pointer */
3802 union xfs_btree_rec *rrp; /* right record pointer */
3803
3804 lrp = xfs_btree_rec_addr(cur, lrecs + 1, left);
3805 rrp = xfs_btree_rec_addr(cur, 1, right);
3806
3807 xfs_btree_copy_recs(cur, lrp, rrp, rrecs);
3808 xfs_btree_log_recs(cur, lbp, lrecs + 1, lrecs + rrecs);
3809 }
3810
3811 XFS_BTREE_STATS_INC(cur, join);
3812
3813 /*
3814 * Fix up the number of records and right block pointer in the
3815 * surviving block, and log it.
3816 */
3817 xfs_btree_set_numrecs(left, lrecs + rrecs);
3818 xfs_btree_get_sibling(cur, right, &cptr, XFS_BB_RIGHTSIB),
3819 xfs_btree_set_sibling(cur, left, &cptr, XFS_BB_RIGHTSIB);
3820 xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS | XFS_BB_RIGHTSIB);
3821
3822 /* If there is a right sibling, point it to the remaining block. */
3823 xfs_btree_get_sibling(cur, left, &cptr, XFS_BB_RIGHTSIB);
3824 if (!xfs_btree_ptr_is_null(cur, &cptr)) {
3825 error = xfs_btree_read_buf_block(cur, &cptr, 0, &rrblock, &rrbp);
3826 if (error)
3827 goto error0;
3828 xfs_btree_set_sibling(cur, rrblock, &lptr, XFS_BB_LEFTSIB);
3829 xfs_btree_log_block(cur, rrbp, XFS_BB_LEFTSIB);
3830 }
3831
3832 /* Free the deleted block. */
3833 error = cur->bc_ops->free_block(cur, rbp);
3834 if (error)
3835 goto error0;
3836 XFS_BTREE_STATS_INC(cur, free);
3837
3838 /*
3839 * If we joined with the left neighbor, set the buffer in the
3840 * cursor to the left block, and fix up the index.
3841 */
3842 if (bp != lbp) {
3843 cur->bc_bufs[level] = lbp;
3844 cur->bc_ptrs[level] += lrecs;
3845 cur->bc_ra[level] = 0;
3846 }
3847 /*
3848 * If we joined with the right neighbor and there's a level above
3849 * us, increment the cursor at that level.
3850 */
3851 else if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) ||
3852 (level + 1 < cur->bc_nlevels)) {
3853 error = xfs_btree_increment(cur, level + 1, &i);
3854 if (error)
3855 goto error0;
3856 }
3857
3858 /*
3859 * Readjust the ptr at this level if it's not a leaf, since it's
3860 * still pointing at the deletion point, which makes the cursor
3861 * inconsistent. If this makes the ptr 0, the caller fixes it up.
3862 * We can't use decrement because it would change the next level up.
3863 */
3864 if (level > 0)
3865 cur->bc_ptrs[level]--;
3866
3867 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3868 /* Return value means the next level up has something to do. */
3869 *stat = 2;
3870 return 0;
3871
3872 error0:
3873 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3874 if (tcur)
3875 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
3876 return error;
3877 }
3878
3879 /*
3880 * Delete the record pointed to by cur.
3881 * The cursor refers to the place where the record was (could be inserted)
3882 * when the operation returns.
3883 */
3884 int /* error */
3885 xfs_btree_delete(
3886 struct xfs_btree_cur *cur,
3887 int *stat) /* success/failure */
3888 {
3889 int error; /* error return value */
3890 int level;
3891 int i;
3892
3893 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
3894
3895 /*
3896 * Go up the tree, starting at leaf level.
3897 *
3898 * If 2 is returned then a join was done; go to the next level.
3899 * Otherwise we are done.
3900 */
3901 for (level = 0, i = 2; i == 2; level++) {
3902 error = xfs_btree_delrec(cur, level, &i);
3903 if (error)
3904 goto error0;
3905 }
3906
3907 if (i == 0) {
3908 for (level = 1; level < cur->bc_nlevels; level++) {
3909 if (cur->bc_ptrs[level] == 0) {
3910 error = xfs_btree_decrement(cur, level, &i);
3911 if (error)
3912 goto error0;
3913 break;
3914 }
3915 }
3916 }
3917
3918 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3919 *stat = i;
3920 return 0;
3921 error0:
3922 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3923 return error;
3924 }
3925
3926 /*
3927 * Get the data from the pointed-to record.
3928 */
3929 int /* error */
3930 xfs_btree_get_rec(
3931 struct xfs_btree_cur *cur, /* btree cursor */
3932 union xfs_btree_rec **recp, /* output: btree record */
3933 int *stat) /* output: success/failure */
3934 {
3935 struct xfs_btree_block *block; /* btree block */
3936 struct xfs_buf *bp; /* buffer pointer */
3937 int ptr; /* record number */
3938 #ifdef DEBUG
3939 int error; /* error return value */
3940 #endif
3941
3942 ptr = cur->bc_ptrs[0];
3943 block = xfs_btree_get_block(cur, 0, &bp);
3944
3945 #ifdef DEBUG
3946 error = xfs_btree_check_block(cur, block, 0, bp);
3947 if (error)
3948 return error;
3949 #endif
3950
3951 /*
3952 * Off the right end or left end, return failure.
3953 */
3954 if (ptr > xfs_btree_get_numrecs(block) || ptr <= 0) {
3955 *stat = 0;
3956 return 0;
3957 }
3958
3959 /*
3960 * Point to the record and extract its data.
3961 */
3962 *recp = xfs_btree_rec_addr(cur, ptr, block);
3963 *stat = 1;
3964 return 0;
3965 }
3966
3967 /*
3968 * Change the owner of a btree.
3969 *
3970 * The mechanism we use here is ordered buffer logging. Because we don't know
3971 * how many buffers were are going to need to modify, we don't really want to
3972 * have to make transaction reservations for the worst case of every buffer in a
3973 * full size btree as that may be more space that we can fit in the log....
3974 *
3975 * We do the btree walk in the most optimal manner possible - we have sibling
3976 * pointers so we can just walk all the blocks on each level from left to right
3977 * in a single pass, and then move to the next level and do the same. We can
3978 * also do readahead on the sibling pointers to get IO moving more quickly,
3979 * though for slow disks this is unlikely to make much difference to performance
3980 * as the amount of CPU work we have to do before moving to the next block is
3981 * relatively small.
3982 *
3983 * For each btree block that we load, modify the owner appropriately, set the
3984 * buffer as an ordered buffer and log it appropriately. We need to ensure that
3985 * we mark the region we change dirty so that if the buffer is relogged in
3986 * a subsequent transaction the changes we make here as an ordered buffer are
3987 * correctly relogged in that transaction. If we are in recovery context, then
3988 * just queue the modified buffer as delayed write buffer so the transaction
3989 * recovery completion writes the changes to disk.
3990 */
3991 static int
3992 xfs_btree_block_change_owner(
3993 struct xfs_btree_cur *cur,
3994 int level,
3995 __uint64_t new_owner,
3996 struct list_head *buffer_list)
3997 {
3998 struct xfs_btree_block *block;
3999 struct xfs_buf *bp;
4000 union xfs_btree_ptr rptr;
4001
4002 /* do right sibling readahead */
4003 xfs_btree_readahead(cur, level, XFS_BTCUR_RIGHTRA);
4004
4005 /* modify the owner */
4006 block = xfs_btree_get_block(cur, level, &bp);
4007 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
4008 block->bb_u.l.bb_owner = cpu_to_be64(new_owner);
4009 else
4010 block->bb_u.s.bb_owner = cpu_to_be32(new_owner);
4011
4012 /*
4013 * If the block is a root block hosted in an inode, we might not have a
4014 * buffer pointer here and we shouldn't attempt to log the change as the
4015 * information is already held in the inode and discarded when the root
4016 * block is formatted into the on-disk inode fork. We still change it,
4017 * though, so everything is consistent in memory.
4018 */
4019 if (bp) {
4020 if (cur->bc_tp) {
4021 xfs_trans_ordered_buf(cur->bc_tp, bp);
4022 xfs_btree_log_block(cur, bp, XFS_BB_OWNER);
4023 } else {
4024 xfs_buf_delwri_queue(bp, buffer_list);
4025 }
4026 } else {
4027 ASSERT(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE);
4028 ASSERT(level == cur->bc_nlevels - 1);
4029 }
4030
4031 /* now read rh sibling block for next iteration */
4032 xfs_btree_get_sibling(cur, block, &rptr, XFS_BB_RIGHTSIB);
4033 if (xfs_btree_ptr_is_null(cur, &rptr))
4034 return -ENOENT;
4035
4036 return xfs_btree_lookup_get_block(cur, level, &rptr, &block);
4037 }
4038
4039 int
4040 xfs_btree_change_owner(
4041 struct xfs_btree_cur *cur,
4042 __uint64_t new_owner,
4043 struct list_head *buffer_list)
4044 {
4045 union xfs_btree_ptr lptr;
4046 int level;
4047 struct xfs_btree_block *block = NULL;
4048 int error = 0;
4049
4050 cur->bc_ops->init_ptr_from_cur(cur, &lptr);
4051
4052 /* for each level */
4053 for (level = cur->bc_nlevels - 1; level >= 0; level--) {
4054 /* grab the left hand block */
4055 error = xfs_btree_lookup_get_block(cur, level, &lptr, &block);
4056 if (error)
4057 return error;
4058
4059 /* readahead the left most block for the next level down */
4060 if (level > 0) {
4061 union xfs_btree_ptr *ptr;
4062
4063 ptr = xfs_btree_ptr_addr(cur, 1, block);
4064 xfs_btree_readahead_ptr(cur, ptr, 1);
4065
4066 /* save for the next iteration of the loop */
4067 lptr = *ptr;
4068 }
4069
4070 /* for each buffer in the level */
4071 do {
4072 error = xfs_btree_block_change_owner(cur, level,
4073 new_owner,
4074 buffer_list);
4075 } while (!error);
4076
4077 if (error != -ENOENT)
4078 return error;
4079 }
4080
4081 return 0;
4082 }
This page took 0.119115 seconds and 5 git commands to generate.