]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - fs/xfs/xfs_btree.c
projects / mirror_ubuntu-bionic-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
xfs: decouple log and transaction headers
[mirror_ubuntu-bionic-kernel.git] / fs / xfs / xfs_btree.c
CommitLineData
1da177e4 1/*
7b718769
NS		 2 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
1da177e4 4 *
7b718769
NS		 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
1da177e4
LT		 7 * published by the Free Software Foundation.
8 *
7b718769
NS		 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.
1da177e4 13 *
7b718769
NS		 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
1da177e4 17 */
1da177e4 18#include "xfs.h"
a844f451 19#include "xfs_fs.h"
70a9883c 20#include "xfs_shared.h"
239880ef
DC		 21#include "xfs_log_format.h"
22#include "xfs_trans_resv.h"
a844f451 23#include "xfs_bit.h"
1da177e4
LT		 24#include "xfs_sb.h"
25#include "xfs_ag.h"
1da177e4 26#include "xfs_mount.h"
1da177e4 27#include "xfs_bmap_btree.h"
a844f451 28#include "xfs_alloc_btree.h"
1da177e4 29#include "xfs_ialloc_btree.h"
1da177e4
LT		 30#include "xfs_dinode.h"
31#include "xfs_inode.h"
239880ef 32#include "xfs_trans.h"
38bb7423 33#include "xfs_inode_item.h"
ee1a47ab 34#include "xfs_buf_item.h"
a844f451 35#include "xfs_btree.h"
1da177e4 36#include "xfs_error.h"
0b1b213f 37#include "xfs_trace.h"
ee1a47ab 38#include "xfs_cksum.h"
1da177e4
LT		 39
40/*
41 * Cursor allocation zone.
42 */
43kmem_zone_t *xfs_btree_cur_zone;
44
45/*
46 * Btree magic numbers.
47 */
ee1a47ab
CH		 48static const __uint32_t xfs_magics[2][XFS_BTNUM_MAX] = {
49 { XFS_ABTB_MAGIC, XFS_ABTC_MAGIC, XFS_BMAP_MAGIC, XFS_IBT_MAGIC },
50 { XFS_ABTB_CRC_MAGIC, XFS_ABTC_CRC_MAGIC,
51 XFS_BMAP_CRC_MAGIC, XFS_IBT_CRC_MAGIC }
1da177e4 52};
ee1a47ab
CH		 53#define xfs_btree_magic(cur) \
54 xfs_magics[!!((cur)->bc_flags & XFS_BTREE_CRC_BLOCKS)][cur->bc_btnum]
1da177e4 55
1da177e4 56
7cc95a82 57STATIC int /* error (0 or EFSCORRUPTED) */
a23f6ef8
CH		 58xfs_btree_check_lblock(
59 struct xfs_btree_cur *cur, /* btree cursor */
7cc95a82 60 struct xfs_btree_block *block, /* btree long form block pointer */
a23f6ef8
CH		 61 int level, /* level of the btree block */
62 struct xfs_buf *bp) /* buffer for block, if any */
63{
ee1a47ab 64 int lblock_ok = 1; /* block passes checks */
a23f6ef8
CH		 65 struct xfs_mount *mp; /* file system mount point */
66
67 mp = cur->bc_mp;
ee1a47ab
CH		 68
69 if (xfs_sb_version_hascrc(&mp->m_sb)) {
70 lblock_ok = lblock_ok &&
71 uuid_equal(&block->bb_u.l.bb_uuid, &mp->m_sb.sb_uuid) &&
72 block->bb_u.l.bb_blkno == cpu_to_be64(
73 bp ? bp->b_bn : XFS_BUF_DADDR_NULL);
74 }
75
76 lblock_ok = lblock_ok &&
77 be32_to_cpu(block->bb_magic) == xfs_btree_magic(cur) &&
a23f6ef8
CH		 78 be16_to_cpu(block->bb_level) == level &&
79 be16_to_cpu(block->bb_numrecs) <=
ce5e42db 80 cur->bc_ops->get_maxrecs(cur, level) &&
7cc95a82 81 block->bb_u.l.bb_leftsib &&
69ef921b 82 (block->bb_u.l.bb_leftsib == cpu_to_be64(NULLDFSBNO) ||
7cc95a82 83 XFS_FSB_SANITY_CHECK(mp,
ee1a47ab 84 be64_to_cpu(block->bb_u.l.bb_leftsib))) &&
7cc95a82 85 block->bb_u.l.bb_rightsib &&
69ef921b 86 (block->bb_u.l.bb_rightsib == cpu_to_be64(NULLDFSBNO) ||
7cc95a82 87 XFS_FSB_SANITY_CHECK(mp,
ee1a47ab
CH		 88 be64_to_cpu(block->bb_u.l.bb_rightsib)));
89
a23f6ef8
CH		 90 if (unlikely(XFS_TEST_ERROR(!lblock_ok, mp,
91 XFS_ERRTAG_BTREE_CHECK_LBLOCK,
92 XFS_RANDOM_BTREE_CHECK_LBLOCK))) {
93 if (bp)
0b1b213f 94 trace_xfs_btree_corrupt(bp, _RET_IP_);
ee1a47ab 95 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp);
a23f6ef8
CH		 96 return XFS_ERROR(EFSCORRUPTED);
97 }
98 return 0;
99}
100
3cc7524c 101STATIC int /* error (0 or EFSCORRUPTED) */
1da177e4 102xfs_btree_check_sblock(
a23f6ef8 103 struct xfs_btree_cur *cur, /* btree cursor */
7cc95a82 104 struct xfs_btree_block *block, /* btree short form block pointer */
1da177e4 105 int level, /* level of the btree block */
a23f6ef8 106 struct xfs_buf *bp) /* buffer containing block */
1da177e4 107{
ee1a47ab 108 struct xfs_mount *mp; /* file system mount point */
a23f6ef8
CH		 109 struct xfs_buf *agbp; /* buffer for ag. freespace struct */
110 struct xfs_agf *agf; /* ag. freespace structure */
1da177e4 111 xfs_agblock_t agflen; /* native ag. freespace length */
ee1a47ab 112 int sblock_ok = 1; /* block passes checks */
1da177e4 113
ee1a47ab 114 mp = cur->bc_mp;
1da177e4
LT		 115 agbp = cur->bc_private.a.agbp;
116 agf = XFS_BUF_TO_AGF(agbp);
16259e7d 117 agflen = be32_to_cpu(agf->agf_length);
ee1a47ab
CH		 118
119 if (xfs_sb_version_hascrc(&mp->m_sb)) {
120 sblock_ok = sblock_ok &&
121 uuid_equal(&block->bb_u.s.bb_uuid, &mp->m_sb.sb_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) &&
16259e7d
CH		 128 be16_to_cpu(block->bb_level) == level &&
129 be16_to_cpu(block->bb_numrecs) <=
ce5e42db 130 cur->bc_ops->get_maxrecs(cur, level) &&
69ef921b 131 (block->bb_u.s.bb_leftsib == cpu_to_be32(NULLAGBLOCK) ||
7cc95a82
CH		 132 be32_to_cpu(block->bb_u.s.bb_leftsib) < agflen) &&
133 block->bb_u.s.bb_leftsib &&
69ef921b 134 (block->bb_u.s.bb_rightsib == cpu_to_be32(NULLAGBLOCK) ||
7cc95a82
CH		 135 be32_to_cpu(block->bb_u.s.bb_rightsib) < agflen) &&
136 block->bb_u.s.bb_rightsib;
ee1a47ab
CH		 137
138 if (unlikely(XFS_TEST_ERROR(!sblock_ok, mp,
1da177e4
LT		 139 XFS_ERRTAG_BTREE_CHECK_SBLOCK,
140 XFS_RANDOM_BTREE_CHECK_SBLOCK))) {
141 if (bp)
0b1b213f 142 trace_xfs_btree_corrupt(bp, _RET_IP_);
ee1a47ab 143 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp);
1da177e4
LT		 144 return XFS_ERROR(EFSCORRUPTED);
145 }
146 return 0;
147}
148
149/*
a23f6ef8
CH		 150 * Debug routine: check that block header is ok.
151 */
152int
153xfs_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{
7cc95a82
CH		 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);
a23f6ef8
CH		 163}
164
165/*
166 * Check that (long) pointer is ok.
167 */
168int /* error (0 or EFSCORRUPTED) */
169xfs_btree_check_lptr(
170 struct xfs_btree_cur *cur, /* btree cursor */
171 xfs_dfsbno_t bno, /* btree block disk address */
172 int level) /* btree block level */
173{
174 XFS_WANT_CORRUPTED_RETURN(
175 level > 0 &&
176 bno != NULLDFSBNO &&
177 XFS_FSB_SANITY_CHECK(cur->bc_mp, bno));
178 return 0;
179}
180
24ee0e49 181#ifdef DEBUG
a23f6ef8
CH		 182/*
183 * Check that (short) pointer is ok.
1da177e4 184 */
3cc7524c 185STATIC int /* error (0 or EFSCORRUPTED) */
1da177e4 186xfs_btree_check_sptr(
a23f6ef8
CH		 187 struct xfs_btree_cur *cur, /* btree cursor */
188 xfs_agblock_t bno, /* btree block disk address */
189 int level) /* btree block level */
1da177e4 190{
a23f6ef8 191 xfs_agblock_t agblocks = cur->bc_mp->m_sb.sb_agblocks;
1da177e4 192
1da177e4
LT		 193 XFS_WANT_CORRUPTED_RETURN(
194 level > 0 &&
a23f6ef8
CH		 195 bno != NULLAGBLOCK &&
196 bno != 0 &&
197 bno < agblocks);
1da177e4
LT		 198 return 0;
199}
200
a23f6ef8
CH		 201/*
202 * Check that block ptr is ok.
203 */
3cc7524c 204STATIC int /* error (0 or EFSCORRUPTED) */
a23f6ef8
CH		 205xfs_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}
24ee0e49 219#endif
a23f6ef8 220
ee1a47ab
CH		 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 modifcation was that made
227 * it to disk.
228 */
229void
230xfs_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_update_cksum(bp->b_addr, BBTOB(bp->b_length),
241 XFS_BTREE_LBLOCK_CRC_OFF);
242}
243
244bool
245xfs_btree_lblock_verify_crc(
246 struct xfs_buf *bp)
247{
248 if (xfs_sb_version_hascrc(&bp->b_target->bt_mount->m_sb))
249 return xfs_verify_cksum(bp->b_addr, BBTOB(bp->b_length),
250 XFS_BTREE_LBLOCK_CRC_OFF);
251 return true;
252}
253
254/*
255 * Calculate CRC on the whole btree block and stuff it into the
256 * short-form btree header.
257 *
258 * Prior to calculting the CRC, pull the LSN out of the buffer log item and put
259 * it into the buffer so recovery knows what the last modifcation was that made
260 * it to disk.
261 */
262void
263xfs_btree_sblock_calc_crc(
264 struct xfs_buf *bp)
265{
266 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
267 struct xfs_buf_log_item *bip = bp->b_fspriv;
268
269 if (!xfs_sb_version_hascrc(&bp->b_target->bt_mount->m_sb))
270 return;
271 if (bip)
272 block->bb_u.s.bb_lsn = cpu_to_be64(bip->bli_item.li_lsn);
273 xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length),
274 XFS_BTREE_SBLOCK_CRC_OFF);
275}
276
277bool
278xfs_btree_sblock_verify_crc(
279 struct xfs_buf *bp)
280{
281 if (xfs_sb_version_hascrc(&bp->b_target->bt_mount->m_sb))
282 return xfs_verify_cksum(bp->b_addr, BBTOB(bp->b_length),
283 XFS_BTREE_SBLOCK_CRC_OFF);
284 return true;
285}
286
1da177e4
LT		 287/*
288 * Delete the btree cursor.
289 */
290void
291xfs_btree_del_cursor(
292 xfs_btree_cur_t *cur, /* btree cursor */
293 int error) /* del because of error */
294{
295 int i; /* btree level */
296
297 /*
298 * Clear the buffer pointers, and release the buffers.
299 * If we're doing this in the face of an error, we
300 * need to make sure to inspect all of the entries
301 * in the bc_bufs array for buffers to be unlocked.
302 * This is because some of the btree code works from
303 * level n down to 0, and if we get an error along
304 * the way we won't have initialized all the entries
305 * down to 0.
306 */
307 for (i = 0; i < cur->bc_nlevels; i++) {
308 if (cur->bc_bufs[i])
c0e59e1a 309 xfs_trans_brelse(cur->bc_tp, cur->bc_bufs[i]);
1da177e4
LT		 310 else if (!error)
311 break;
312 }
313 /*
314 * Can't free a bmap cursor without having dealt with the
315 * allocated indirect blocks' accounting.
316 */
317 ASSERT(cur->bc_btnum != XFS_BTNUM_BMAP ||
318 cur->bc_private.b.allocated == 0);
319 /*
320 * Free the cursor.
321 */
322 kmem_zone_free(xfs_btree_cur_zone, cur);
323}
324
325/*
326 * Duplicate the btree cursor.
327 * Allocate a new one, copy the record, re-get the buffers.
328 */
329int /* error */
330xfs_btree_dup_cursor(
331 xfs_btree_cur_t *cur, /* input cursor */
332 xfs_btree_cur_t **ncur) /* output cursor */
333{
334 xfs_buf_t *bp; /* btree block's buffer pointer */
335 int error; /* error return value */
336 int i; /* level number of btree block */
337 xfs_mount_t *mp; /* mount structure for filesystem */
338 xfs_btree_cur_t *new; /* new cursor value */
339 xfs_trans_t *tp; /* transaction pointer, can be NULL */
340
341 tp = cur->bc_tp;
342 mp = cur->bc_mp;
561f7d17 343
1da177e4
LT		 344 /*
345 * Allocate a new cursor like the old one.
346 */
561f7d17
CH		 347 new = cur->bc_ops->dup_cursor(cur);
348
1da177e4
LT		 349 /*
350 * Copy the record currently in the cursor.
351 */
352 new->bc_rec = cur->bc_rec;
561f7d17 353
1da177e4
LT		 354 /*
355 * For each level current, re-get the buffer and copy the ptr value.
356 */
357 for (i = 0; i < new->bc_nlevels; i++) {
358 new->bc_ptrs[i] = cur->bc_ptrs[i];
359 new->bc_ra[i] = cur->bc_ra[i];
c3f8fc73
DC		 360 bp = cur->bc_bufs[i];
361 if (bp) {
362 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
363 XFS_BUF_ADDR(bp), mp->m_bsize,
3d3e6f64 364 0, &bp,
1813dd64 365 cur->bc_ops->buf_ops);
c3f8fc73 366 if (error) {
1da177e4
LT		 367 xfs_btree_del_cursor(new, error);
368 *ncur = NULL;
369 return error;
370 }
ee1a47ab
CH		 371 }
372 new->bc_bufs[i] = bp;
1da177e4 373 }
1da177e4
LT		 374 *ncur = new;
375 return 0;
376}
377
65f1eaea
CH		 378/*
379 * XFS btree block layout and addressing:
380 *
381 * There are two types of blocks in the btree: leaf and non-leaf blocks.
382 *
383 * The leaf record start with a header then followed by records containing
384 * the values. A non-leaf block also starts with the same header, and
385 * then first contains lookup keys followed by an equal number of pointers
386 * to the btree blocks at the previous level.
387 *
388 * +--------+-------+-------+-------+-------+-------+-------+
389 * Leaf: | header | rec 1 | rec 2 | rec 3 | rec 4 | rec 5 | rec N |
390 * +--------+-------+-------+-------+-------+-------+-------+
391 *
392 * +--------+-------+-------+-------+-------+-------+-------+
393 * Non-Leaf: | header | key 1 | key 2 | key N | ptr 1 | ptr 2 | ptr N |
394 * +--------+-------+-------+-------+-------+-------+-------+
395 *
396 * The header is called struct xfs_btree_block for reasons better left unknown
397 * and comes in different versions for short (32bit) and long (64bit) block
398 * pointers. The record and key structures are defined by the btree instances
399 * and opaque to the btree core. The block pointers are simple disk endian
400 * integers, available in a short (32bit) and long (64bit) variant.
401 *
402 * The helpers below calculate the offset of a given record, key or pointer
403 * into a btree block (xfs_btree_*_offset) or return a pointer to the given
404 * record, key or pointer (xfs_btree_*_addr). Note that all addressing
405 * inside the btree block is done using indices starting at one, not zero!
406 */
407
408/*
409 * Return size of the btree block header for this btree instance.
410 */
411static inline size_t xfs_btree_block_len(struct xfs_btree_cur *cur)
412{
ee1a47ab
CH		 413 if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
414 if (cur->bc_flags & XFS_BTREE_CRC_BLOCKS)
415 return XFS_BTREE_LBLOCK_CRC_LEN;
416 return XFS_BTREE_LBLOCK_LEN;
417 }
418 if (cur->bc_flags & XFS_BTREE_CRC_BLOCKS)
419 return XFS_BTREE_SBLOCK_CRC_LEN;
420 return XFS_BTREE_SBLOCK_LEN;
65f1eaea
CH		 421}
422
423/*
424 * Return size of btree block pointers for this btree instance.
425 */
426static inline size_t xfs_btree_ptr_len(struct xfs_btree_cur *cur)
427{
428 return (cur->bc_flags & XFS_BTREE_LONG_PTRS) ?
429 sizeof(__be64) : sizeof(__be32);
430}
431
432/*
433 * Calculate offset of the n-th record in a btree block.
434 */
435STATIC size_t
436xfs_btree_rec_offset(
437 struct xfs_btree_cur *cur,
438 int n)
439{
440 return xfs_btree_block_len(cur) +
441 (n - 1) * cur->bc_ops->rec_len;
442}
443
444/*
445 * Calculate offset of the n-th key in a btree block.
446 */
447STATIC size_t
448xfs_btree_key_offset(
449 struct xfs_btree_cur *cur,
450 int n)
451{
452 return xfs_btree_block_len(cur) +
453 (n - 1) * cur->bc_ops->key_len;
454}
455
456/*
457 * Calculate offset of the n-th block pointer in a btree block.
458 */
459STATIC size_t
460xfs_btree_ptr_offset(
461 struct xfs_btree_cur *cur,
462 int n,
463 int level)
464{
465 return xfs_btree_block_len(cur) +
466 cur->bc_ops->get_maxrecs(cur, level) * cur->bc_ops->key_len +
467 (n - 1) * xfs_btree_ptr_len(cur);
468}
469
470/*
471 * Return a pointer to the n-th record in the btree block.
472 */
473STATIC union xfs_btree_rec *
474xfs_btree_rec_addr(
475 struct xfs_btree_cur *cur,
476 int n,
477 struct xfs_btree_block *block)
478{
479 return (union xfs_btree_rec *)
480 ((char *)block + xfs_btree_rec_offset(cur, n));
481}
482
483/*
484 * Return a pointer to the n-th key in the btree block.
485 */
486STATIC union xfs_btree_key *
487xfs_btree_key_addr(
488 struct xfs_btree_cur *cur,
489 int n,
490 struct xfs_btree_block *block)
491{
492 return (union xfs_btree_key *)
493 ((char *)block + xfs_btree_key_offset(cur, n));
494}
495
496/*
497 * Return a pointer to the n-th block pointer in the btree block.
498 */
499STATIC union xfs_btree_ptr *
500xfs_btree_ptr_addr(
501 struct xfs_btree_cur *cur,
502 int n,
503 struct xfs_btree_block *block)
504{
505 int level = xfs_btree_get_level(block);
506
507 ASSERT(block->bb_level != 0);
508
509 return (union xfs_btree_ptr *)
510 ((char *)block + xfs_btree_ptr_offset(cur, n, level));
511}
512
8186e517 513/*
1cb93863 514 * Get the root block which is stored in the inode.
8186e517
CH		 515 *
516 * For now this btree implementation assumes the btree root is always
517 * stored in the if_broot field of an inode fork.
518 */
519STATIC struct xfs_btree_block *
520xfs_btree_get_iroot(
521 struct xfs_btree_cur *cur)
522{
523 struct xfs_ifork *ifp;
524
525 ifp = XFS_IFORK_PTR(cur->bc_private.b.ip, cur->bc_private.b.whichfork);
526 return (struct xfs_btree_block *)ifp->if_broot;
527}
528
1da177e4
LT		 529/*
530 * Retrieve the block pointer from the cursor at the given level.
8186e517 531 * This may be an inode btree root or from a buffer.
1da177e4 532 */
8186e517 533STATIC struct xfs_btree_block * /* generic btree block pointer */
1da177e4 534xfs_btree_get_block(
8186e517 535 struct xfs_btree_cur *cur, /* btree cursor */
1da177e4 536 int level, /* level in btree */
8186e517 537 struct xfs_buf **bpp) /* buffer containing the block */
1da177e4 538{
8186e517
CH		 539 if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
540 (level == cur->bc_nlevels - 1)) {
541 *bpp = NULL;
542 return xfs_btree_get_iroot(cur);
1da177e4 543 }
8186e517
CH		 544
545 *bpp = cur->bc_bufs[level];
546 return XFS_BUF_TO_BLOCK(*bpp);
1da177e4
LT		 547}
548
549/*
550 * Get a buffer for the block, return it with no data read.
551 * Long-form addressing.
552 */
553xfs_buf_t * /* buffer for fsbno */
554xfs_btree_get_bufl(
555 xfs_mount_t *mp, /* file system mount point */
556 xfs_trans_t *tp, /* transaction pointer */
557 xfs_fsblock_t fsbno, /* file system block number */
558 uint lock) /* lock flags for get_buf */
559{
560 xfs_buf_t *bp; /* buffer pointer (return value) */
561 xfs_daddr_t d; /* real disk block address */
562
563 ASSERT(fsbno != NULLFSBLOCK);
564 d = XFS_FSB_TO_DADDR(mp, fsbno);
565 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d, mp->m_bsize, lock);
5a52c2a5 566 ASSERT(!xfs_buf_geterror(bp));
1da177e4
LT		 567 return bp;
568}
569
570/*
571 * Get a buffer for the block, return it with no data read.
572 * Short-form addressing.
573 */
574xfs_buf_t * /* buffer for agno/agbno */
575xfs_btree_get_bufs(
576 xfs_mount_t *mp, /* file system mount point */
577 xfs_trans_t *tp, /* transaction pointer */
578 xfs_agnumber_t agno, /* allocation group number */
579 xfs_agblock_t agbno, /* allocation group block number */
580 uint lock) /* lock flags for get_buf */
581{
582 xfs_buf_t *bp; /* buffer pointer (return value) */
583 xfs_daddr_t d; /* real disk block address */
584
585 ASSERT(agno != NULLAGNUMBER);
586 ASSERT(agbno != NULLAGBLOCK);
587 d = XFS_AGB_TO_DADDR(mp, agno, agbno);
588 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d, mp->m_bsize, lock);
5a52c2a5 589 ASSERT(!xfs_buf_geterror(bp));
1da177e4
LT		 590 return bp;
591}
592
1da177e4
LT		 593/*
594 * Check for the cursor referring to the last block at the given level.
595 */
596int /* 1=is last block, 0=not last block */
597xfs_btree_islastblock(
598 xfs_btree_cur_t *cur, /* btree cursor */
599 int level) /* level to check */
600{
7cc95a82 601 struct xfs_btree_block *block; /* generic btree block pointer */
1da177e4
LT		 602 xfs_buf_t *bp; /* buffer containing block */
603
604 block = xfs_btree_get_block(cur, level, &bp);
605 xfs_btree_check_block(cur, block, level, bp);
e99ab90d 606 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
69ef921b 607 return block->bb_u.l.bb_rightsib == cpu_to_be64(NULLDFSBNO);
1da177e4 608 else
69ef921b 609 return block->bb_u.s.bb_rightsib == cpu_to_be32(NULLAGBLOCK);
1da177e4
LT		 610}
611
cdcf4333
CH		 612/*
613 * Change the cursor to point to the first record at the given level.
614 * Other levels are unaffected.
615 */
3cc7524c 616STATIC int /* success=1, failure=0 */
cdcf4333
CH		 617xfs_btree_firstrec(
618 xfs_btree_cur_t *cur, /* btree cursor */
619 int level) /* level to change */
620{
7cc95a82 621 struct xfs_btree_block *block; /* generic btree block pointer */
cdcf4333
CH		 622 xfs_buf_t *bp; /* buffer containing block */
623
624 /*
625 * Get the block pointer for this level.
626 */
627 block = xfs_btree_get_block(cur, level, &bp);
628 xfs_btree_check_block(cur, block, level, bp);
629 /*
630 * It's empty, there is no such record.
631 */
f2277f06 632 if (!block->bb_numrecs)
cdcf4333
CH		 633 return 0;
634 /*
635 * Set the ptr value to 1, that's the first record/key.
636 */
637 cur->bc_ptrs[level] = 1;
638 return 1;
639}
640
1da177e4
LT		 641/*
642 * Change the cursor to point to the last record in the current block
643 * at the given level. Other levels are unaffected.
644 */
3cc7524c 645STATIC int /* success=1, failure=0 */
1da177e4
LT		 646xfs_btree_lastrec(
647 xfs_btree_cur_t *cur, /* btree cursor */
648 int level) /* level to change */
649{
7cc95a82 650 struct xfs_btree_block *block; /* generic btree block pointer */
1da177e4
LT		 651 xfs_buf_t *bp; /* buffer containing block */
652
653 /*
654 * Get the block pointer for this level.
655 */
656 block = xfs_btree_get_block(cur, level, &bp);
657 xfs_btree_check_block(cur, block, level, bp);
658 /*
659 * It's empty, there is no such record.
660 */
f2277f06 661 if (!block->bb_numrecs)
1da177e4
LT		 662 return 0;
663 /*
664 * Set the ptr value to numrecs, that's the last record/key.
665 */
f2277f06 666 cur->bc_ptrs[level] = be16_to_cpu(block->bb_numrecs);
1da177e4
LT		 667 return 1;
668}
669
670/*
671 * Compute first and last byte offsets for the fields given.
672 * Interprets the offsets table, which contains struct field offsets.
673 */
674void
675xfs_btree_offsets(
676 __int64_t fields, /* bitmask of fields */
677 const short *offsets, /* table of field offsets */
678 int nbits, /* number of bits to inspect */
679 int *first, /* output: first byte offset */
680 int *last) /* output: last byte offset */
681{
682 int i; /* current bit number */
683 __int64_t imask; /* mask for current bit number */
684
685 ASSERT(fields != 0);
686 /*
687 * Find the lowest bit, so the first byte offset.
688 */
689 for (i = 0, imask = 1LL; ; i++, imask <<= 1) {
690 if (imask & fields) {
691 *first = offsets[i];
692 break;
693 }
694 }
695 /*
696 * Find the highest bit, so the last byte offset.
697 */
698 for (i = nbits - 1, imask = 1LL << i; ; i--, imask >>= 1) {
699 if (imask & fields) {
700 *last = offsets[i + 1] - 1;
701 break;
702 }
703 }
704}
705
706/*
707 * Get a buffer for the block, return it read in.
708 * Long-form addressing.
709 */
3d3e6f64 710int
1da177e4 711xfs_btree_read_bufl(
3d3e6f64
DC		 712 struct xfs_mount *mp, /* file system mount point */
713 struct xfs_trans *tp, /* transaction pointer */
714 xfs_fsblock_t fsbno, /* file system block number */
715 uint lock, /* lock flags for read_buf */
716 struct xfs_buf **bpp, /* buffer for fsbno */
717 int refval, /* ref count value for buffer */
1813dd64 718 const struct xfs_buf_ops *ops)
1da177e4 719{
3d3e6f64 720 struct xfs_buf *bp; /* return value */
1da177e4 721 xfs_daddr_t d; /* real disk block address */
3d3e6f64 722 int error;
1da177e4
LT		 723
724 ASSERT(fsbno != NULLFSBLOCK);
725 d = XFS_FSB_TO_DADDR(mp, fsbno);
c3f8fc73 726 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, d,
1813dd64 727 mp->m_bsize, lock, &bp, ops);
c3f8fc73 728 if (error)
1da177e4 729 return error;
5a52c2a5 730 ASSERT(!xfs_buf_geterror(bp));
821eb21d 731 if (bp)
38f23232 732 xfs_buf_set_ref(bp, refval);
1da177e4
LT		 733 *bpp = bp;
734 return 0;
735}
736
1da177e4
LT		 737/*
738 * Read-ahead the block, don't wait for it, don't return a buffer.
739 * Long-form addressing.
740 */
741/* ARGSUSED */
742void
743xfs_btree_reada_bufl(
3d3e6f64
DC		 744 struct xfs_mount *mp, /* file system mount point */
745 xfs_fsblock_t fsbno, /* file system block number */
746 xfs_extlen_t count, /* count of filesystem blocks */
1813dd64 747 const struct xfs_buf_ops *ops)
1da177e4
LT		 748{
749 xfs_daddr_t d;
750
751 ASSERT(fsbno != NULLFSBLOCK);
752 d = XFS_FSB_TO_DADDR(mp, fsbno);
1813dd64 753 xfs_buf_readahead(mp->m_ddev_targp, d, mp->m_bsize * count, ops);
1da177e4
LT		 754}
755
756/*
757 * Read-ahead the block, don't wait for it, don't return a buffer.
758 * Short-form addressing.
759 */
760/* ARGSUSED */
761void
762xfs_btree_reada_bufs(
3d3e6f64
DC		 763 struct xfs_mount *mp, /* file system mount point */
764 xfs_agnumber_t agno, /* allocation group number */
765 xfs_agblock_t agbno, /* allocation group block number */
766 xfs_extlen_t count, /* count of filesystem blocks */
1813dd64 767 const struct xfs_buf_ops *ops)
1da177e4
LT		 768{
769 xfs_daddr_t d;
770
771 ASSERT(agno != NULLAGNUMBER);
772 ASSERT(agbno != NULLAGBLOCK);
773 d = XFS_AGB_TO_DADDR(mp, agno, agbno);
1813dd64 774 xfs_buf_readahead(mp->m_ddev_targp, d, mp->m_bsize * count, ops);
1da177e4
LT		 775}
776
b524bfee
CH		 777STATIC int
778xfs_btree_readahead_lblock(
779 struct xfs_btree_cur *cur,
780 int lr,
781 struct xfs_btree_block *block)
782{
783 int rval = 0;
e6edbd1c
CH		 784 xfs_dfsbno_t left = be64_to_cpu(block->bb_u.l.bb_leftsib);
785 xfs_dfsbno_t right = be64_to_cpu(block->bb_u.l.bb_rightsib);
b524bfee
CH		 786
787 if ((lr & XFS_BTCUR_LEFTRA) && left != NULLDFSBNO) {
3d3e6f64 788 xfs_btree_reada_bufl(cur->bc_mp, left, 1,
1813dd64 789 cur->bc_ops->buf_ops);
b524bfee
CH		 790 rval++;
791 }
792
793 if ((lr & XFS_BTCUR_RIGHTRA) && right != NULLDFSBNO) {
3d3e6f64 794 xfs_btree_reada_bufl(cur->bc_mp, right, 1,
1813dd64 795 cur->bc_ops->buf_ops);
b524bfee
CH		 796 rval++;
797 }
798
799 return rval;
800}
801
802STATIC int
803xfs_btree_readahead_sblock(
804 struct xfs_btree_cur *cur,
805 int lr,
806 struct xfs_btree_block *block)
807{
808 int rval = 0;
809 xfs_agblock_t left = be32_to_cpu(block->bb_u.s.bb_leftsib);
810 xfs_agblock_t right = be32_to_cpu(block->bb_u.s.bb_rightsib);
811
812
813 if ((lr & XFS_BTCUR_LEFTRA) && left != NULLAGBLOCK) {
814 xfs_btree_reada_bufs(cur->bc_mp, cur->bc_private.a.agno,
1813dd64 815 left, 1, cur->bc_ops->buf_ops);
b524bfee
CH		 816 rval++;
817 }
818
819 if ((lr & XFS_BTCUR_RIGHTRA) && right != NULLAGBLOCK) {
820 xfs_btree_reada_bufs(cur->bc_mp, cur->bc_private.a.agno,
1813dd64 821 right, 1, cur->bc_ops->buf_ops);
b524bfee
CH		 822 rval++;
823 }
824
825 return rval;
826}
827
1da177e4
LT		 828/*
829 * Read-ahead btree blocks, at the given level.
830 * Bits in lr are set from XFS_BTCUR_{LEFT,RIGHT}RA.
831 */
3cc7524c 832STATIC int
b524bfee
CH		 833xfs_btree_readahead(
834 struct xfs_btree_cur *cur, /* btree cursor */
1da177e4
LT		 835 int lev, /* level in btree */
836 int lr) /* left/right bits */
837{
b524bfee
CH		 838 struct xfs_btree_block *block;
839
840 /*
841 * No readahead needed if we are at the root level and the
842 * btree root is stored in the inode.
843 */
844 if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
845 (lev == cur->bc_nlevels - 1))
846 return 0;
847
848 if ((cur->bc_ra[lev] | lr) == cur->bc_ra[lev])
849 return 0;
1da177e4 850
1da177e4 851 cur->bc_ra[lev] |= lr;
b524bfee
CH		 852 block = XFS_BUF_TO_BLOCK(cur->bc_bufs[lev]);
853
854 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
855 return xfs_btree_readahead_lblock(cur, lr, block);
856 return xfs_btree_readahead_sblock(cur, lr, block);
1da177e4
LT		 857}
858
21b5c978
DC		 859STATIC xfs_daddr_t
860xfs_btree_ptr_to_daddr(
861 struct xfs_btree_cur *cur,
862 union xfs_btree_ptr *ptr)
863{
864 if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
865 ASSERT(ptr->l != cpu_to_be64(NULLDFSBNO));
866
867 return XFS_FSB_TO_DADDR(cur->bc_mp, be64_to_cpu(ptr->l));
868 } else {
869 ASSERT(cur->bc_private.a.agno != NULLAGNUMBER);
870 ASSERT(ptr->s != cpu_to_be32(NULLAGBLOCK));
871
872 return XFS_AGB_TO_DADDR(cur->bc_mp, cur->bc_private.a.agno,
873 be32_to_cpu(ptr->s));
874 }
875}
876
877/*
878 * Readahead @count btree blocks at the given @ptr location.
879 *
880 * We don't need to care about long or short form btrees here as we have a
881 * method of converting the ptr directly to a daddr available to us.
882 */
883STATIC void
884xfs_btree_readahead_ptr(
885 struct xfs_btree_cur *cur,
886 union xfs_btree_ptr *ptr,
887 xfs_extlen_t count)
888{
889 xfs_buf_readahead(cur->bc_mp->m_ddev_targp,
890 xfs_btree_ptr_to_daddr(cur, ptr),
891 cur->bc_mp->m_bsize * count, cur->bc_ops->buf_ops);
892}
893
1da177e4
LT		 894/*
895 * Set the buffer for level "lev" in the cursor to bp, releasing
896 * any previous buffer.
897 */
c0e59e1a 898STATIC void
1da177e4
LT		 899xfs_btree_setbuf(
900 xfs_btree_cur_t *cur, /* btree cursor */
901 int lev, /* level in btree */
902 xfs_buf_t *bp) /* new buffer to set */
903{
7cc95a82 904 struct xfs_btree_block *b; /* btree block */
1da177e4 905
c0e59e1a
CH		 906 if (cur->bc_bufs[lev])
907 xfs_trans_brelse(cur->bc_tp, cur->bc_bufs[lev]);
1da177e4
LT		 908 cur->bc_bufs[lev] = bp;
909 cur->bc_ra[lev] = 0;
c0e59e1a 910
1da177e4 911 b = XFS_BUF_TO_BLOCK(bp);
e99ab90d 912 if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
69ef921b 913 if (b->bb_u.l.bb_leftsib == cpu_to_be64(NULLDFSBNO))
1da177e4 914 cur->bc_ra[lev] |= XFS_BTCUR_LEFTRA;
69ef921b 915 if (b->bb_u.l.bb_rightsib == cpu_to_be64(NULLDFSBNO))
1da177e4
LT		 916 cur->bc_ra[lev] |= XFS_BTCUR_RIGHTRA;
917 } else {
69ef921b 918 if (b->bb_u.s.bb_leftsib == cpu_to_be32(NULLAGBLOCK))
1da177e4 919 cur->bc_ra[lev] |= XFS_BTCUR_LEFTRA;
69ef921b 920 if (b->bb_u.s.bb_rightsib == cpu_to_be32(NULLAGBLOCK))
1da177e4
LT		 921 cur->bc_ra[lev] |= XFS_BTCUR_RIGHTRA;
922 }
923}
637aa50f
CH		 924
925STATIC int
926xfs_btree_ptr_is_null(
927 struct xfs_btree_cur *cur,
928 union xfs_btree_ptr *ptr)
929{
930 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
69ef921b 931 return ptr->l == cpu_to_be64(NULLDFSBNO);
637aa50f 932 else
69ef921b 933 return ptr->s == cpu_to_be32(NULLAGBLOCK);
637aa50f
CH		 934}
935
4b22a571
CH		 936STATIC void
937xfs_btree_set_ptr_null(
938 struct xfs_btree_cur *cur,
939 union xfs_btree_ptr *ptr)
940{
941 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
33ad965d 942 ptr->l = cpu_to_be64(NULLDFSBNO);
4b22a571
CH		 943 else
944 ptr->s = cpu_to_be32(NULLAGBLOCK);
945}
946
637aa50f
CH		 947/*
948 * Get/set/init sibling pointers
949 */
950STATIC void
951xfs_btree_get_sibling(
952 struct xfs_btree_cur *cur,
953 struct xfs_btree_block *block,
954 union xfs_btree_ptr *ptr,
955 int lr)
956{
957 ASSERT(lr == XFS_BB_LEFTSIB || lr == XFS_BB_RIGHTSIB);
958
959 if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
960 if (lr == XFS_BB_RIGHTSIB)
961 ptr->l = block->bb_u.l.bb_rightsib;
962 else
963 ptr->l = block->bb_u.l.bb_leftsib;
964 } else {
965 if (lr == XFS_BB_RIGHTSIB)
966 ptr->s = block->bb_u.s.bb_rightsib;
967 else
968 ptr->s = block->bb_u.s.bb_leftsib;
969 }
970}
971
f5eb8e7c
CH		 972STATIC void
973xfs_btree_set_sibling(
974 struct xfs_btree_cur *cur,
975 struct xfs_btree_block *block,
976 union xfs_btree_ptr *ptr,
977 int lr)
978{
979 ASSERT(lr == XFS_BB_LEFTSIB || lr == XFS_BB_RIGHTSIB);
980
981 if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
982 if (lr == XFS_BB_RIGHTSIB)
983 block->bb_u.l.bb_rightsib = ptr->l;
984 else
985 block->bb_u.l.bb_leftsib = ptr->l;
986 } else {
987 if (lr == XFS_BB_RIGHTSIB)
988 block->bb_u.s.bb_rightsib = ptr->s;
989 else
990 block->bb_u.s.bb_leftsib = ptr->s;
991 }
992}
993
ee1a47ab
CH		 994void
995xfs_btree_init_block_int(
996 struct xfs_mount *mp,
997 struct xfs_btree_block *buf,
998 xfs_daddr_t blkno,
999 __u32 magic,
1000 __u16 level,
1001 __u16 numrecs,
1002 __u64 owner,
1003 unsigned int flags)
1004{
1005 buf->bb_magic = cpu_to_be32(magic);
1006 buf->bb_level = cpu_to_be16(level);
1007 buf->bb_numrecs = cpu_to_be16(numrecs);
1008
1009 if (flags & XFS_BTREE_LONG_PTRS) {
1010 buf->bb_u.l.bb_leftsib = cpu_to_be64(NULLDFSBNO);
1011 buf->bb_u.l.bb_rightsib = cpu_to_be64(NULLDFSBNO);
1012 if (flags & XFS_BTREE_CRC_BLOCKS) {
1013 buf->bb_u.l.bb_blkno = cpu_to_be64(blkno);
1014 buf->bb_u.l.bb_owner = cpu_to_be64(owner);
1015 uuid_copy(&buf->bb_u.l.bb_uuid, &mp->m_sb.sb_uuid);
1016 buf->bb_u.l.bb_pad = 0;
b58fa554 1017 buf->bb_u.l.bb_lsn = 0;
ee1a47ab
CH		 1018 }
1019 } else {
1020 /* owner is a 32 bit value on short blocks */
1021 __u32 __owner = (__u32)owner;
1022
1023 buf->bb_u.s.bb_leftsib = cpu_to_be32(NULLAGBLOCK);
1024 buf->bb_u.s.bb_rightsib = cpu_to_be32(NULLAGBLOCK);
1025 if (flags & XFS_BTREE_CRC_BLOCKS) {
1026 buf->bb_u.s.bb_blkno = cpu_to_be64(blkno);
1027 buf->bb_u.s.bb_owner = cpu_to_be32(__owner);
1028 uuid_copy(&buf->bb_u.s.bb_uuid, &mp->m_sb.sb_uuid);
b58fa554 1029 buf->bb_u.s.bb_lsn = 0;
ee1a47ab
CH		 1030 }
1031 }
1032}
1033
b64f3a39 1034void
f5eb8e7c 1035xfs_btree_init_block(
b64f3a39
DC		 1036 struct xfs_mount *mp,
1037 struct xfs_buf *bp,
1038 __u32 magic,
1039 __u16 level,
1040 __u16 numrecs,
ee1a47ab 1041 __u64 owner,
b64f3a39 1042 unsigned int flags)
f5eb8e7c 1043{
ee1a47ab
CH		 1044 xfs_btree_init_block_int(mp, XFS_BUF_TO_BLOCK(bp), bp->b_bn,
1045 magic, level, numrecs, owner, flags);
f5eb8e7c
CH		 1046}
1047
b64f3a39
DC		 1048STATIC void
1049xfs_btree_init_block_cur(
1050 struct xfs_btree_cur *cur,
ee1a47ab 1051 struct xfs_buf *bp,
b64f3a39 1052 int level,
ee1a47ab 1053 int numrecs)
b64f3a39 1054{
ee1a47ab
CH		 1055 __u64 owner;
1056
1057 /*
1058 * we can pull the owner from the cursor right now as the different
1059 * owners align directly with the pointer size of the btree. This may
1060 * change in future, but is safe for current users of the generic btree
1061 * code.
1062 */
1063 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
1064 owner = cur->bc_private.b.ip->i_ino;
1065 else
1066 owner = cur->bc_private.a.agno;
1067
1068 xfs_btree_init_block_int(cur->bc_mp, XFS_BUF_TO_BLOCK(bp), bp->b_bn,
1069 xfs_btree_magic(cur), level, numrecs,
1070 owner, cur->bc_flags);
b64f3a39
DC		 1071}
1072
278d0ca1
CH		 1073/*
1074 * Return true if ptr is the last record in the btree and
ee1a47ab 1075 * we need to track updates to this record. The decision
278d0ca1
CH		 1076 * will be further refined in the update_lastrec method.
1077 */
1078STATIC int
1079xfs_btree_is_lastrec(
1080 struct xfs_btree_cur *cur,
1081 struct xfs_btree_block *block,
1082 int level)
1083{
1084 union xfs_btree_ptr ptr;
1085
1086 if (level > 0)
1087 return 0;
1088 if (!(cur->bc_flags & XFS_BTREE_LASTREC_UPDATE))
1089 return 0;
1090
1091 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
1092 if (!xfs_btree_ptr_is_null(cur, &ptr))
1093 return 0;
1094 return 1;
1095}
1096
f5eb8e7c
CH		 1097STATIC void
1098xfs_btree_buf_to_ptr(
1099 struct xfs_btree_cur *cur,
1100 struct xfs_buf *bp,
1101 union xfs_btree_ptr *ptr)
1102{
1103 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
1104 ptr->l = cpu_to_be64(XFS_DADDR_TO_FSB(cur->bc_mp,
1105 XFS_BUF_ADDR(bp)));
1106 else {
9d87c319 1107 ptr->s = cpu_to_be32(xfs_daddr_to_agbno(cur->bc_mp,
f5eb8e7c
CH		 1108 XFS_BUF_ADDR(bp)));
1109 }
1110}
1111
637aa50f
CH		 1112STATIC void
1113xfs_btree_set_refs(
1114 struct xfs_btree_cur *cur,
1115 struct xfs_buf *bp)
1116{
1117 switch (cur->bc_btnum) {
1118 case XFS_BTNUM_BNO:
1119 case XFS_BTNUM_CNT:
38f23232 1120 xfs_buf_set_ref(bp, XFS_ALLOC_BTREE_REF);
637aa50f
CH		 1121 break;
1122 case XFS_BTNUM_INO:
38f23232 1123 xfs_buf_set_ref(bp, XFS_INO_BTREE_REF);
637aa50f
CH		 1124 break;
1125 case XFS_BTNUM_BMAP:
38f23232 1126 xfs_buf_set_ref(bp, XFS_BMAP_BTREE_REF);
637aa50f
CH		 1127 break;
1128 default:
1129 ASSERT(0);
1130 }
1131}
1132
f5eb8e7c
CH		 1133STATIC int
1134xfs_btree_get_buf_block(
1135 struct xfs_btree_cur *cur,
1136 union xfs_btree_ptr *ptr,
1137 int flags,
1138 struct xfs_btree_block **block,
1139 struct xfs_buf **bpp)
1140{
1141 struct xfs_mount *mp = cur->bc_mp;
1142 xfs_daddr_t d;
1143
1144 /* need to sort out how callers deal with failures first */
0cadda1c 1145 ASSERT(!(flags & XBF_TRYLOCK));
f5eb8e7c
CH		 1146
1147 d = xfs_btree_ptr_to_daddr(cur, ptr);
1148 *bpp = xfs_trans_get_buf(cur->bc_tp, mp->m_ddev_targp, d,
1149 mp->m_bsize, flags);
1150
2a30f36d
CS		 1151 if (!*bpp)
1152 return ENOMEM;
f5eb8e7c 1153
1813dd64 1154 (*bpp)->b_ops = cur->bc_ops->buf_ops;
f5eb8e7c
CH		 1155 *block = XFS_BUF_TO_BLOCK(*bpp);
1156 return 0;
1157}
1158
637aa50f
CH		 1159/*
1160 * Read in the buffer at the given ptr and return the buffer and
1161 * the block pointer within the buffer.
1162 */
1163STATIC int
1164xfs_btree_read_buf_block(
1165 struct xfs_btree_cur *cur,
1166 union xfs_btree_ptr *ptr,
1167 int level,
1168 int flags,
1169 struct xfs_btree_block **block,
1170 struct xfs_buf **bpp)
1171{
1172 struct xfs_mount *mp = cur->bc_mp;
1173 xfs_daddr_t d;
1174 int error;
1175
1176 /* need to sort out how callers deal with failures first */
0cadda1c 1177 ASSERT(!(flags & XBF_TRYLOCK));
637aa50f
CH		 1178
1179 d = xfs_btree_ptr_to_daddr(cur, ptr);
1180 error = xfs_trans_read_buf(mp, cur->bc_tp, mp->m_ddev_targp, d,
3d3e6f64 1181 mp->m_bsize, flags, bpp,
1813dd64 1182 cur->bc_ops->buf_ops);
637aa50f
CH		 1183 if (error)
1184 return error;
1185
5a52c2a5 1186 ASSERT(!xfs_buf_geterror(*bpp));
637aa50f
CH		 1187 xfs_btree_set_refs(cur, *bpp);
1188 *block = XFS_BUF_TO_BLOCK(*bpp);
3d3e6f64 1189 return 0;
637aa50f
CH		 1190}
1191
38bb7423
CH		 1192/*
1193 * Copy keys from one btree block to another.
1194 */
1195STATIC void
1196xfs_btree_copy_keys(
1197 struct xfs_btree_cur *cur,
1198 union xfs_btree_key *dst_key,
1199 union xfs_btree_key *src_key,
1200 int numkeys)
1201{
1202 ASSERT(numkeys >= 0);
1203 memcpy(dst_key, src_key, numkeys * cur->bc_ops->key_len);
1204}
1205
278d0ca1
CH		 1206/*
1207 * Copy records from one btree block to another.
1208 */
1209STATIC void
1210xfs_btree_copy_recs(
1211 struct xfs_btree_cur *cur,
1212 union xfs_btree_rec *dst_rec,
1213 union xfs_btree_rec *src_rec,
1214 int numrecs)
1215{
1216 ASSERT(numrecs >= 0);
1217 memcpy(dst_rec, src_rec, numrecs * cur->bc_ops->rec_len);
1218}
1219
9eaead51
CH		 1220/*
1221 * Copy block pointers from one btree block to another.
1222 */
1223STATIC void
1224xfs_btree_copy_ptrs(
1225 struct xfs_btree_cur *cur,
1226 union xfs_btree_ptr *dst_ptr,
1227 union xfs_btree_ptr *src_ptr,
1228 int numptrs)
1229{
1230 ASSERT(numptrs >= 0);
1231 memcpy(dst_ptr, src_ptr, numptrs * xfs_btree_ptr_len(cur));
1232}
1233
1234/*
1235 * Shift keys one index left/right inside a single btree block.
1236 */
1237STATIC void
1238xfs_btree_shift_keys(
1239 struct xfs_btree_cur *cur,
1240 union xfs_btree_key *key,
1241 int dir,
1242 int numkeys)
1243{
1244 char *dst_key;
1245
1246 ASSERT(numkeys >= 0);
1247 ASSERT(dir == 1 || dir == -1);
1248
1249 dst_key = (char *)key + (dir * cur->bc_ops->key_len);
1250 memmove(dst_key, key, numkeys * cur->bc_ops->key_len);
1251}
1252
1253/*
1254 * Shift records one index left/right inside a single btree block.
1255 */
1256STATIC void
1257xfs_btree_shift_recs(
1258 struct xfs_btree_cur *cur,
1259 union xfs_btree_rec *rec,
1260 int dir,
1261 int numrecs)
1262{
1263 char *dst_rec;
1264
1265 ASSERT(numrecs >= 0);
1266 ASSERT(dir == 1 || dir == -1);
1267
1268 dst_rec = (char *)rec + (dir * cur->bc_ops->rec_len);
1269 memmove(dst_rec, rec, numrecs * cur->bc_ops->rec_len);
1270}
1271
1272/*
1273 * Shift block pointers one index left/right inside a single btree block.
1274 */
1275STATIC void
1276xfs_btree_shift_ptrs(
1277 struct xfs_btree_cur *cur,
1278 union xfs_btree_ptr *ptr,
1279 int dir,
1280 int numptrs)
1281{
1282 char *dst_ptr;
1283
1284 ASSERT(numptrs >= 0);
1285 ASSERT(dir == 1 || dir == -1);
1286
1287 dst_ptr = (char *)ptr + (dir * xfs_btree_ptr_len(cur));
1288 memmove(dst_ptr, ptr, numptrs * xfs_btree_ptr_len(cur));
1289}
1290
38bb7423
CH		 1291/*
1292 * Log key values from the btree block.
1293 */
1294STATIC void
1295xfs_btree_log_keys(
1296 struct xfs_btree_cur *cur,
1297 struct xfs_buf *bp,
1298 int first,
1299 int last)
1300{
1301 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1302 XFS_BTREE_TRACE_ARGBII(cur, bp, first, last);
1303
1304 if (bp) {
61fe135c 1305 xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
38bb7423
CH		 1306 xfs_trans_log_buf(cur->bc_tp, bp,
1307 xfs_btree_key_offset(cur, first),
1308 xfs_btree_key_offset(cur, last + 1) - 1);
1309 } else {
1310 xfs_trans_log_inode(cur->bc_tp, cur->bc_private.b.ip,
1311 xfs_ilog_fbroot(cur->bc_private.b.whichfork));
1312 }
1313
1314 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1315}
1316
278d0ca1
CH		 1317/*
1318 * Log record values from the btree block.
1319 */
fd6bcc5b 1320void
278d0ca1
CH		 1321xfs_btree_log_recs(
1322 struct xfs_btree_cur *cur,
1323 struct xfs_buf *bp,
1324 int first,
1325 int last)
1326{
1327 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1328 XFS_BTREE_TRACE_ARGBII(cur, bp, first, last);
1329
61fe135c 1330 xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
278d0ca1
CH		 1331 xfs_trans_log_buf(cur->bc_tp, bp,
1332 xfs_btree_rec_offset(cur, first),
1333 xfs_btree_rec_offset(cur, last + 1) - 1);
1334
1335 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1336}
1337
9eaead51
CH		 1338/*
1339 * Log block pointer fields from a btree block (nonleaf).
1340 */
1341STATIC void
1342xfs_btree_log_ptrs(
1343 struct xfs_btree_cur *cur, /* btree cursor */
1344 struct xfs_buf *bp, /* buffer containing btree block */
1345 int first, /* index of first pointer to log */
1346 int last) /* index of last pointer to log */
1347{
1348 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1349 XFS_BTREE_TRACE_ARGBII(cur, bp, first, last);
1350
1351 if (bp) {
1352 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
1353 int level = xfs_btree_get_level(block);
1354
61fe135c 1355 xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
9eaead51
CH		 1356 xfs_trans_log_buf(cur->bc_tp, bp,
1357 xfs_btree_ptr_offset(cur, first, level),
1358 xfs_btree_ptr_offset(cur, last + 1, level) - 1);
1359 } else {
1360 xfs_trans_log_inode(cur->bc_tp, cur->bc_private.b.ip,
1361 xfs_ilog_fbroot(cur->bc_private.b.whichfork));
1362 }
1363
1364 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1365}
1366
1367/*
1368 * Log fields from a btree block header.
1369 */
fd6bcc5b 1370void
9eaead51
CH		 1371xfs_btree_log_block(
1372 struct xfs_btree_cur *cur, /* btree cursor */
1373 struct xfs_buf *bp, /* buffer containing btree block */
1374 int fields) /* mask of fields: XFS_BB_... */
1375{
1376 int first; /* first byte offset logged */
1377 int last; /* last byte offset logged */
1378 static const short soffsets[] = { /* table of offsets (short) */
7cc95a82
CH		 1379 offsetof(struct xfs_btree_block, bb_magic),
1380 offsetof(struct xfs_btree_block, bb_level),
1381 offsetof(struct xfs_btree_block, bb_numrecs),
1382 offsetof(struct xfs_btree_block, bb_u.s.bb_leftsib),
1383 offsetof(struct xfs_btree_block, bb_u.s.bb_rightsib),
ee1a47ab
CH		 1384 offsetof(struct xfs_btree_block, bb_u.s.bb_blkno),
1385 offsetof(struct xfs_btree_block, bb_u.s.bb_lsn),
1386 offsetof(struct xfs_btree_block, bb_u.s.bb_uuid),
1387 offsetof(struct xfs_btree_block, bb_u.s.bb_owner),
1388 offsetof(struct xfs_btree_block, bb_u.s.bb_crc),
1389 XFS_BTREE_SBLOCK_CRC_LEN
9eaead51
CH		 1390 };
1391 static const short loffsets[] = { /* table of offsets (long) */
7cc95a82
CH		 1392 offsetof(struct xfs_btree_block, bb_magic),
1393 offsetof(struct xfs_btree_block, bb_level),
1394 offsetof(struct xfs_btree_block, bb_numrecs),
1395 offsetof(struct xfs_btree_block, bb_u.l.bb_leftsib),
1396 offsetof(struct xfs_btree_block, bb_u.l.bb_rightsib),
ee1a47ab
CH		 1397 offsetof(struct xfs_btree_block, bb_u.l.bb_blkno),
1398 offsetof(struct xfs_btree_block, bb_u.l.bb_lsn),
1399 offsetof(struct xfs_btree_block, bb_u.l.bb_uuid),
1400 offsetof(struct xfs_btree_block, bb_u.l.bb_owner),
1401 offsetof(struct xfs_btree_block, bb_u.l.bb_crc),
1402 offsetof(struct xfs_btree_block, bb_u.l.bb_pad),
1403 XFS_BTREE_LBLOCK_CRC_LEN
9eaead51
CH		 1404 };
1405
1406 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1407 XFS_BTREE_TRACE_ARGBI(cur, bp, fields);
1408
1409 if (bp) {
ee1a47ab
CH		 1410 int nbits;
1411
1412 if (cur->bc_flags & XFS_BTREE_CRC_BLOCKS) {
1413 /*
1414 * We don't log the CRC when updating a btree
1415 * block but instead recreate it during log
1416 * recovery. As the log buffers have checksums
1417 * of their own this is safe and avoids logging a crc
1418 * update in a lot of places.
1419 */
1420 if (fields == XFS_BB_ALL_BITS)
1421 fields = XFS_BB_ALL_BITS_CRC;
1422 nbits = XFS_BB_NUM_BITS_CRC;
1423 } else {
1424 nbits = XFS_BB_NUM_BITS;
1425 }
9eaead51
CH		 1426 xfs_btree_offsets(fields,
1427 (cur->bc_flags & XFS_BTREE_LONG_PTRS) ?
1428 loffsets : soffsets,
ee1a47ab 1429 nbits, &first, &last);
61fe135c 1430 xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
9eaead51
CH		 1431 xfs_trans_log_buf(cur->bc_tp, bp, first, last);
1432 } else {
1433 xfs_trans_log_inode(cur->bc_tp, cur->bc_private.b.ip,
1434 xfs_ilog_fbroot(cur->bc_private.b.whichfork));
1435 }
1436
1437 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1438}
1439
637aa50f
CH		 1440/*
1441 * Increment cursor by one record at the level.
1442 * For nonzero levels the leaf-ward information is untouched.
1443 */
1444int /* error */
1445xfs_btree_increment(
1446 struct xfs_btree_cur *cur,
1447 int level,
1448 int *stat) /* success/failure */
1449{
1450 struct xfs_btree_block *block;
1451 union xfs_btree_ptr ptr;
1452 struct xfs_buf *bp;
1453 int error; /* error return value */
1454 int lev;
1455
1456 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1457 XFS_BTREE_TRACE_ARGI(cur, level);
1458
1459 ASSERT(level < cur->bc_nlevels);
1460
1461 /* Read-ahead to the right at this level. */
1462 xfs_btree_readahead(cur, level, XFS_BTCUR_RIGHTRA);
1463
1464 /* Get a pointer to the btree block. */
1465 block = xfs_btree_get_block(cur, level, &bp);
1466
1467#ifdef DEBUG
1468 error = xfs_btree_check_block(cur, block, level, bp);
1469 if (error)
1470 goto error0;
1471#endif
1472
1473 /* We're done if we remain in the block after the increment. */
1474 if (++cur->bc_ptrs[level] <= xfs_btree_get_numrecs(block))
1475 goto out1;
1476
1477 /* Fail if we just went off the right edge of the tree. */
1478 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
1479 if (xfs_btree_ptr_is_null(cur, &ptr))
1480 goto out0;
1481
1482 XFS_BTREE_STATS_INC(cur, increment);
1483
1484 /*
1485 * March up the tree incrementing pointers.
1486 * Stop when we don't go off the right edge of a block.
1487 */
1488 for (lev = level + 1; lev < cur->bc_nlevels; lev++) {
1489 block = xfs_btree_get_block(cur, lev, &bp);
1490
1491#ifdef DEBUG
1492 error = xfs_btree_check_block(cur, block, lev, bp);
1493 if (error)
1494 goto error0;
1495#endif
1496
1497 if (++cur->bc_ptrs[lev] <= xfs_btree_get_numrecs(block))
1498 break;
1499
1500 /* Read-ahead the right block for the next loop. */
1501 xfs_btree_readahead(cur, lev, XFS_BTCUR_RIGHTRA);
1502 }
1503
1504 /*
1505 * If we went off the root then we are either seriously
1506 * confused or have the tree root in an inode.
1507 */
1508 if (lev == cur->bc_nlevels) {
1509 if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE)
1510 goto out0;
1511 ASSERT(0);
1512 error = EFSCORRUPTED;
1513 goto error0;
1514 }
1515 ASSERT(lev < cur->bc_nlevels);
1516
1517 /*
1518 * Now walk back down the tree, fixing up the cursor's buffer
1519 * pointers and key numbers.
1520 */
1521 for (block = xfs_btree_get_block(cur, lev, &bp); lev > level; ) {
1522 union xfs_btree_ptr *ptrp;
1523
1524 ptrp = xfs_btree_ptr_addr(cur, cur->bc_ptrs[lev], block);
1525 error = xfs_btree_read_buf_block(cur, ptrp, --lev,
1526 0, &block, &bp);
1527 if (error)
1528 goto error0;
1529
1530 xfs_btree_setbuf(cur, lev, bp);
1531 cur->bc_ptrs[lev] = 1;
1532 }
1533out1:
1534 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1535 *stat = 1;
1536 return 0;
1537
1538out0:
1539 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1540 *stat = 0;
1541 return 0;
1542
1543error0:
1544 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
1545 return error;
1546}
8df4da4a
CH		 1547
1548/*
1549 * Decrement cursor by one record at the level.
1550 * For nonzero levels the leaf-ward information is untouched.
1551 */
1552int /* error */
1553xfs_btree_decrement(
1554 struct xfs_btree_cur *cur,
1555 int level,
1556 int *stat) /* success/failure */
1557{
1558 struct xfs_btree_block *block;
1559 xfs_buf_t *bp;
1560 int error; /* error return value */
1561 int lev;
1562 union xfs_btree_ptr ptr;
1563
1564 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1565 XFS_BTREE_TRACE_ARGI(cur, level);
1566
1567 ASSERT(level < cur->bc_nlevels);
1568
1569 /* Read-ahead to the left at this level. */
1570 xfs_btree_readahead(cur, level, XFS_BTCUR_LEFTRA);
1571
1572 /* We're done if we remain in the block after the decrement. */
1573 if (--cur->bc_ptrs[level] > 0)
1574 goto out1;
1575
1576 /* Get a pointer to the btree block. */
1577 block = xfs_btree_get_block(cur, level, &bp);
1578
1579#ifdef DEBUG
1580 error = xfs_btree_check_block(cur, block, level, bp);
1581 if (error)
1582 goto error0;
1583#endif
1584
1585 /* Fail if we just went off the left edge of the tree. */
1586 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_LEFTSIB);
1587 if (xfs_btree_ptr_is_null(cur, &ptr))
1588 goto out0;
1589
1590 XFS_BTREE_STATS_INC(cur, decrement);
1591
1592 /*
1593 * March up the tree decrementing pointers.
1594 * Stop when we don't go off the left edge of a block.
1595 */
1596 for (lev = level + 1; lev < cur->bc_nlevels; lev++) {
1597 if (--cur->bc_ptrs[lev] > 0)
1598 break;
1599 /* Read-ahead the left block for the next loop. */
1600 xfs_btree_readahead(cur, lev, XFS_BTCUR_LEFTRA);
1601 }
1602
1603 /*
1604 * If we went off the root then we are seriously confused.
1605 * or the root of the tree is in an inode.
1606 */
1607 if (lev == cur->bc_nlevels) {
1608 if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE)
1609 goto out0;
1610 ASSERT(0);
1611 error = EFSCORRUPTED;
1612 goto error0;
1613 }
1614 ASSERT(lev < cur->bc_nlevels);
1615
1616 /*
1617 * Now walk back down the tree, fixing up the cursor's buffer
1618 * pointers and key numbers.
1619 */
1620 for (block = xfs_btree_get_block(cur, lev, &bp); lev > level; ) {
1621 union xfs_btree_ptr *ptrp;
1622
1623 ptrp = xfs_btree_ptr_addr(cur, cur->bc_ptrs[lev], block);
1624 error = xfs_btree_read_buf_block(cur, ptrp, --lev,
1625 0, &block, &bp);
1626 if (error)
1627 goto error0;
1628 xfs_btree_setbuf(cur, lev, bp);
1629 cur->bc_ptrs[lev] = xfs_btree_get_numrecs(block);
1630 }
1631out1:
1632 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1633 *stat = 1;
1634 return 0;
1635
1636out0:
1637 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1638 *stat = 0;
1639 return 0;
1640
1641error0:
1642 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
1643 return error;
1644}
1645
fe033cc8
CH		 1646STATIC int
1647xfs_btree_lookup_get_block(
1648 struct xfs_btree_cur *cur, /* btree cursor */
1649 int level, /* level in the btree */
1650 union xfs_btree_ptr *pp, /* ptr to btree block */
1651 struct xfs_btree_block **blkp) /* return btree block */
1652{
1653 struct xfs_buf *bp; /* buffer pointer for btree block */
1654 int error = 0;
1655
1656 /* special case the root block if in an inode */
1657 if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
1658 (level == cur->bc_nlevels - 1)) {
1659 *blkp = xfs_btree_get_iroot(cur);
1660 return 0;
1661 }
1662
1663 /*
1664 * If the old buffer at this level for the disk address we are
1665 * looking for re-use it.
1666 *
1667 * Otherwise throw it away and get a new one.
1668 */
1669 bp = cur->bc_bufs[level];
1670 if (bp && XFS_BUF_ADDR(bp) == xfs_btree_ptr_to_daddr(cur, pp)) {
1671 *blkp = XFS_BUF_TO_BLOCK(bp);
1672 return 0;
1673 }
1674
1675 error = xfs_btree_read_buf_block(cur, pp, level, 0, blkp, &bp);
1676 if (error)
1677 return error;
1678
1679 xfs_btree_setbuf(cur, level, bp);
1680 return 0;
1681}
1682
1683/*
1684 * Get current search key. For level 0 we don't actually have a key
1685 * structure so we make one up from the record. For all other levels
1686 * we just return the right key.
1687 */
1688STATIC union xfs_btree_key *
1689xfs_lookup_get_search_key(
1690 struct xfs_btree_cur *cur,
1691 int level,
1692 int keyno,
1693 struct xfs_btree_block *block,
1694 union xfs_btree_key *kp)
1695{
1696 if (level == 0) {
1697 cur->bc_ops->init_key_from_rec(kp,
1698 xfs_btree_rec_addr(cur, keyno, block));
1699 return kp;
1700 }
1701
1702 return xfs_btree_key_addr(cur, keyno, block);
1703}
1704
1705/*
1706 * Lookup the record. The cursor is made to point to it, based on dir.
49d3da14 1707 * stat is set to 0 if can't find any such record, 1 for success.
fe033cc8
CH		 1708 */
1709int /* error */
1710xfs_btree_lookup(
1711 struct xfs_btree_cur *cur, /* btree cursor */
1712 xfs_lookup_t dir, /* <=, ==, or >= */
1713 int *stat) /* success/failure */
1714{
1715 struct xfs_btree_block *block; /* current btree block */
1716 __int64_t diff; /* difference for the current key */
1717 int error; /* error return value */
1718 int keyno; /* current key number */
1719 int level; /* level in the btree */
1720 union xfs_btree_ptr *pp; /* ptr to btree block */
1721 union xfs_btree_ptr ptr; /* ptr to btree block */
1722
1723 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1724 XFS_BTREE_TRACE_ARGI(cur, dir);
1725
1726 XFS_BTREE_STATS_INC(cur, lookup);
1727
1728 block = NULL;
1729 keyno = 0;
1730
1731 /* initialise start pointer from cursor */
1732 cur->bc_ops->init_ptr_from_cur(cur, &ptr);
1733 pp = &ptr;
1734
1735 /*
1736 * Iterate over each level in the btree, starting at the root.
1737 * For each level above the leaves, find the key we need, based
1738 * on the lookup record, then follow the corresponding block
1739 * pointer down to the next level.
1740 */
1741 for (level = cur->bc_nlevels - 1, diff = 1; level >= 0; level--) {
1742 /* Get the block we need to do the lookup on. */
1743 error = xfs_btree_lookup_get_block(cur, level, pp, &block);
1744 if (error)
1745 goto error0;
1746
1747 if (diff == 0) {
1748 /*
1749 * If we already had a key match at a higher level, we
1750 * know we need to use the first entry in this block.
1751 */
1752 keyno = 1;
1753 } else {
1754 /* Otherwise search this block. Do a binary search. */
1755
1756 int high; /* high entry number */
1757 int low; /* low entry number */
1758
1759 /* Set low and high entry numbers, 1-based. */
1760 low = 1;
1761 high = xfs_btree_get_numrecs(block);
1762 if (!high) {
1763 /* Block is empty, must be an empty leaf. */
1764 ASSERT(level == 0 && cur->bc_nlevels == 1);
1765
1766 cur->bc_ptrs[0] = dir != XFS_LOOKUP_LE;
1767 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1768 *stat = 0;
1769 return 0;
1770 }
1771
1772 /* Binary search the block. */
1773 while (low <= high) {
1774 union xfs_btree_key key;
1775 union xfs_btree_key *kp;
1776
1777 XFS_BTREE_STATS_INC(cur, compare);
1778
1779 /* keyno is average of low and high. */
1780 keyno = (low + high) >> 1;
1781
1782 /* Get current search key */
1783 kp = xfs_lookup_get_search_key(cur, level,
1784 keyno, block, &key);
1785
1786 /*
1787 * Compute difference to get next direction:
1788 * - less than, move right
1789 * - greater than, move left
1790 * - equal, we're done
1791 */
1792 diff = cur->bc_ops->key_diff(cur, kp);
1793 if (diff < 0)
1794 low = keyno + 1;
1795 else if (diff > 0)
1796 high = keyno - 1;
1797 else
1798 break;
1799 }
1800 }
1801
1802 /*
1803 * If there are more levels, set up for the next level
1804 * by getting the block number and filling in the cursor.
1805 */
1806 if (level > 0) {
1807 /*
1808 * If we moved left, need the previous key number,
1809 * unless there isn't one.
1810 */
1811 if (diff > 0 && --keyno < 1)
1812 keyno = 1;
1813 pp = xfs_btree_ptr_addr(cur, keyno, block);
1814
1815#ifdef DEBUG
1816 error = xfs_btree_check_ptr(cur, pp, 0, level);
1817 if (error)
1818 goto error0;
1819#endif
1820 cur->bc_ptrs[level] = keyno;
1821 }
1822 }
1823
1824 /* Done with the search. See if we need to adjust the results. */
1825 if (dir != XFS_LOOKUP_LE && diff < 0) {
1826 keyno++;
1827 /*
1828 * If ge search and we went off the end of the block, but it's
1829 * not the last block, we're in the wrong block.
1830 */
1831 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
1832 if (dir == XFS_LOOKUP_GE &&
1833 keyno > xfs_btree_get_numrecs(block) &&
1834 !xfs_btree_ptr_is_null(cur, &ptr)) {
1835 int i;
1836
1837 cur->bc_ptrs[0] = keyno;
1838 error = xfs_btree_increment(cur, 0, &i);
1839 if (error)
1840 goto error0;
1841 XFS_WANT_CORRUPTED_RETURN(i == 1);
1842 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1843 *stat = 1;
1844 return 0;
1845 }
1846 } else if (dir == XFS_LOOKUP_LE && diff > 0)
1847 keyno--;
1848 cur->bc_ptrs[0] = keyno;
1849
1850 /* Return if we succeeded or not. */
1851 if (keyno == 0 || keyno > xfs_btree_get_numrecs(block))
1852 *stat = 0;
1853 else if (dir != XFS_LOOKUP_EQ || diff == 0)
1854 *stat = 1;
1855 else
1856 *stat = 0;
1857 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1858 return 0;
1859
1860error0:
1861 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
1862 return error;
1863}
38bb7423
CH		 1864
1865/*
1866 * Update keys at all levels from here to the root along the cursor's path.
1867 */
3cc7524c 1868STATIC int
38bb7423
CH		 1869xfs_btree_updkey(
1870 struct xfs_btree_cur *cur,
1871 union xfs_btree_key *keyp,
1872 int level)
1873{
1874 struct xfs_btree_block *block;
1875 struct xfs_buf *bp;
1876 union xfs_btree_key *kp;
1877 int ptr;
1878
1879 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1880 XFS_BTREE_TRACE_ARGIK(cur, level, keyp);
1881
1882 ASSERT(!(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) || level >= 1);
1883
1884 /*
1885 * Go up the tree from this level toward the root.
1886 * At each level, update the key value to the value input.
1887 * Stop when we reach a level where the cursor isn't pointing
1888 * at the first entry in the block.
1889 */
1890 for (ptr = 1; ptr == 1 && level < cur->bc_nlevels; level++) {
1891#ifdef DEBUG
1892 int error;
1893#endif
1894 block = xfs_btree_get_block(cur, level, &bp);
1895#ifdef DEBUG
1896 error = xfs_btree_check_block(cur, block, level, bp);
1897 if (error) {
1898 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
1899 return error;
1900 }
1901#endif
1902 ptr = cur->bc_ptrs[level];
1903 kp = xfs_btree_key_addr(cur, ptr, block);
1904 xfs_btree_copy_keys(cur, kp, keyp, 1);
1905 xfs_btree_log_keys(cur, bp, ptr, ptr);
1906 }
1907
1908 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1909 return 0;
1910}
278d0ca1
CH		 1911
1912/*
1913 * Update the record referred to by cur to the value in the
1914 * given record. This either works (return 0) or gets an
1915 * EFSCORRUPTED error.
1916 */
1917int
1918xfs_btree_update(
1919 struct xfs_btree_cur *cur,
1920 union xfs_btree_rec *rec)
1921{
1922 struct xfs_btree_block *block;
1923 struct xfs_buf *bp;
1924 int error;
1925 int ptr;
1926 union xfs_btree_rec *rp;
1927
1928 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1929 XFS_BTREE_TRACE_ARGR(cur, rec);
1930
1931 /* Pick up the current block. */
1932 block = xfs_btree_get_block(cur, 0, &bp);
1933
1934#ifdef DEBUG
1935 error = xfs_btree_check_block(cur, block, 0, bp);
1936 if (error)
1937 goto error0;
1938#endif
1939 /* Get the address of the rec to be updated. */
1940 ptr = cur->bc_ptrs[0];
1941 rp = xfs_btree_rec_addr(cur, ptr, block);
1942
1943 /* Fill in the new contents and log them. */
1944 xfs_btree_copy_recs(cur, rp, rec, 1);
1945 xfs_btree_log_recs(cur, bp, ptr, ptr);
1946
1947 /*
1948 * If we are tracking the last record in the tree and
1949 * we are at the far right edge of the tree, update it.
1950 */
1951 if (xfs_btree_is_lastrec(cur, block, 0)) {
1952 cur->bc_ops->update_lastrec(cur, block, rec,
1953 ptr, LASTREC_UPDATE);
1954 }
1955
1956 /* Updating first rec in leaf. Pass new key value up to our parent. */
1957 if (ptr == 1) {
1958 union xfs_btree_key key;
1959
1960 cur->bc_ops->init_key_from_rec(&key, rec);
1961 error = xfs_btree_updkey(cur, &key, 1);
1962 if (error)
1963 goto error0;
1964 }
1965
1966 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1967 return 0;
1968
1969error0:
1970 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
1971 return error;
1972}
1973
687b890a
CH		 1974/*
1975 * Move 1 record left from cur/level if possible.
1976 * Update cur to reflect the new path.
1977 */
3cc7524c 1978STATIC int /* error */
687b890a
CH		 1979xfs_btree_lshift(
1980 struct xfs_btree_cur *cur,
1981 int level,
1982 int *stat) /* success/failure */
1983{
1984 union xfs_btree_key key; /* btree key */
1985 struct xfs_buf *lbp; /* left buffer pointer */
1986 struct xfs_btree_block *left; /* left btree block */
1987 int lrecs; /* left record count */
1988 struct xfs_buf *rbp; /* right buffer pointer */
1989 struct xfs_btree_block *right; /* right btree block */
1990 int rrecs; /* right record count */
1991 union xfs_btree_ptr lptr; /* left btree pointer */
1992 union xfs_btree_key *rkp = NULL; /* right btree key */
1993 union xfs_btree_ptr *rpp = NULL; /* right address pointer */
1994 union xfs_btree_rec *rrp = NULL; /* right record pointer */
1995 int error; /* error return value */
1996
1997 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1998 XFS_BTREE_TRACE_ARGI(cur, level);
1999
2000 if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
2001 level == cur->bc_nlevels - 1)
2002 goto out0;
2003
2004 /* Set up variables for this block as "right". */
2005 right = xfs_btree_get_block(cur, level, &rbp);
2006
2007#ifdef DEBUG
2008 error = xfs_btree_check_block(cur, right, level, rbp);
2009 if (error)
2010 goto error0;
2011#endif
2012
2013 /* If we've got no left sibling then we can't shift an entry left. */
2014 xfs_btree_get_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
2015 if (xfs_btree_ptr_is_null(cur, &lptr))
2016 goto out0;
2017
2018 /*
2019 * If the cursor entry is the one that would be moved, don't
2020 * do it... it's too complicated.
2021 */
2022 if (cur->bc_ptrs[level] <= 1)
2023 goto out0;
2024
2025 /* Set up the left neighbor as "left". */
2026 error = xfs_btree_read_buf_block(cur, &lptr, level, 0, &left, &lbp);
2027 if (error)
2028 goto error0;
2029
2030 /* If it's full, it can't take another entry. */
2031 lrecs = xfs_btree_get_numrecs(left);
2032 if (lrecs == cur->bc_ops->get_maxrecs(cur, level))
2033 goto out0;
2034
2035 rrecs = xfs_btree_get_numrecs(right);
2036
2037 /*
2038 * We add one entry to the left side and remove one for the right side.
9da096fd 2039 * Account for it here, the changes will be updated on disk and logged
687b890a
CH		 2040 * later.
2041 */
2042 lrecs++;
2043 rrecs--;
2044
2045 XFS_BTREE_STATS_INC(cur, lshift);
2046 XFS_BTREE_STATS_ADD(cur, moves, 1);
2047
2048 /*
2049 * If non-leaf, copy a key and a ptr to the left block.
2050 * Log the changes to the left block.
2051 */
2052 if (level > 0) {
2053 /* It's a non-leaf. Move keys and pointers. */
2054 union xfs_btree_key *lkp; /* left btree key */
2055 union xfs_btree_ptr *lpp; /* left address pointer */
2056
2057 lkp = xfs_btree_key_addr(cur, lrecs, left);
2058 rkp = xfs_btree_key_addr(cur, 1, right);
2059
2060 lpp = xfs_btree_ptr_addr(cur, lrecs, left);
2061 rpp = xfs_btree_ptr_addr(cur, 1, right);
2062#ifdef DEBUG
2063 error = xfs_btree_check_ptr(cur, rpp, 0, level);
2064 if (error)
2065 goto error0;
2066#endif
2067 xfs_btree_copy_keys(cur, lkp, rkp, 1);
2068 xfs_btree_copy_ptrs(cur, lpp, rpp, 1);
2069
2070 xfs_btree_log_keys(cur, lbp, lrecs, lrecs);
2071 xfs_btree_log_ptrs(cur, lbp, lrecs, lrecs);
2072
4a26e66e
CH		 2073 ASSERT(cur->bc_ops->keys_inorder(cur,
2074 xfs_btree_key_addr(cur, lrecs - 1, left), lkp));
687b890a
CH		 2075 } else {
2076 /* It's a leaf. Move records. */
2077 union xfs_btree_rec *lrp; /* left record pointer */
2078
2079 lrp = xfs_btree_rec_addr(cur, lrecs, left);
2080 rrp = xfs_btree_rec_addr(cur, 1, right);
2081
2082 xfs_btree_copy_recs(cur, lrp, rrp, 1);
2083 xfs_btree_log_recs(cur, lbp, lrecs, lrecs);
2084
4a26e66e
CH		 2085 ASSERT(cur->bc_ops->recs_inorder(cur,
2086 xfs_btree_rec_addr(cur, lrecs - 1, left), lrp));
687b890a
CH		 2087 }
2088
2089 xfs_btree_set_numrecs(left, lrecs);
2090 xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS);
2091
2092 xfs_btree_set_numrecs(right, rrecs);
2093 xfs_btree_log_block(cur, rbp, XFS_BB_NUMRECS);
2094
2095 /*
2096 * Slide the contents of right down one entry.
2097 */
2098 XFS_BTREE_STATS_ADD(cur, moves, rrecs - 1);
2099 if (level > 0) {
2100 /* It's a nonleaf. operate on keys and ptrs */
2101#ifdef DEBUG
2102 int i; /* loop index */
2103
2104 for (i = 0; i < rrecs; i++) {
2105 error = xfs_btree_check_ptr(cur, rpp, i + 1, level);
2106 if (error)
2107 goto error0;
2108 }
2109#endif
2110 xfs_btree_shift_keys(cur,
2111 xfs_btree_key_addr(cur, 2, right),
2112 -1, rrecs);
2113 xfs_btree_shift_ptrs(cur,
2114 xfs_btree_ptr_addr(cur, 2, right),
2115 -1, rrecs);
2116
2117 xfs_btree_log_keys(cur, rbp, 1, rrecs);
2118 xfs_btree_log_ptrs(cur, rbp, 1, rrecs);
2119 } else {
2120 /* It's a leaf. operate on records */
2121 xfs_btree_shift_recs(cur,
2122 xfs_btree_rec_addr(cur, 2, right),
2123 -1, rrecs);
2124 xfs_btree_log_recs(cur, rbp, 1, rrecs);
2125
2126 /*
2127 * If it's the first record in the block, we'll need a key
2128 * structure to pass up to the next level (updkey).
2129 */
2130 cur->bc_ops->init_key_from_rec(&key,
2131 xfs_btree_rec_addr(cur, 1, right));
2132 rkp = &key;
2133 }
2134
2135 /* Update the parent key values of right. */
2136 error = xfs_btree_updkey(cur, rkp, level + 1);
2137 if (error)
2138 goto error0;
2139
2140 /* Slide the cursor value left one. */
2141 cur->bc_ptrs[level]--;
2142
2143 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2144 *stat = 1;
2145 return 0;
2146
2147out0:
2148 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2149 *stat = 0;
2150 return 0;
2151
2152error0:
2153 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2154 return error;
2155}
2156
9eaead51
CH		 2157/*
2158 * Move 1 record right from cur/level if possible.
2159 * Update cur to reflect the new path.
2160 */
3cc7524c 2161STATIC int /* error */
9eaead51
CH		 2162xfs_btree_rshift(
2163 struct xfs_btree_cur *cur,
2164 int level,
2165 int *stat) /* success/failure */
2166{
2167 union xfs_btree_key key; /* btree key */
2168 struct xfs_buf *lbp; /* left buffer pointer */
2169 struct xfs_btree_block *left; /* left btree block */
2170 struct xfs_buf *rbp; /* right buffer pointer */
2171 struct xfs_btree_block *right; /* right btree block */
2172 struct xfs_btree_cur *tcur; /* temporary btree cursor */
2173 union xfs_btree_ptr rptr; /* right block pointer */
2174 union xfs_btree_key *rkp; /* right btree key */
2175 int rrecs; /* right record count */
2176 int lrecs; /* left record count */
2177 int error; /* error return value */
2178 int i; /* loop counter */
2179
2180 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2181 XFS_BTREE_TRACE_ARGI(cur, level);
2182
2183 if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
2184 (level == cur->bc_nlevels - 1))
2185 goto out0;
2186
2187 /* Set up variables for this block as "left". */
2188 left = xfs_btree_get_block(cur, level, &lbp);
2189
2190#ifdef DEBUG
2191 error = xfs_btree_check_block(cur, left, level, lbp);
2192 if (error)
2193 goto error0;
2194#endif
2195
2196 /* If we've got no right sibling then we can't shift an entry right. */
2197 xfs_btree_get_sibling(cur, left, &rptr, XFS_BB_RIGHTSIB);
2198 if (xfs_btree_ptr_is_null(cur, &rptr))
2199 goto out0;
2200
2201 /*
2202 * If the cursor entry is the one that would be moved, don't
2203 * do it... it's too complicated.
2204 */
2205 lrecs = xfs_btree_get_numrecs(left);
2206 if (cur->bc_ptrs[level] >= lrecs)
2207 goto out0;
2208
2209 /* Set up the right neighbor as "right". */
2210 error = xfs_btree_read_buf_block(cur, &rptr, level, 0, &right, &rbp);
2211 if (error)
2212 goto error0;
2213
2214 /* If it's full, it can't take another entry. */
2215 rrecs = xfs_btree_get_numrecs(right);
2216 if (rrecs == cur->bc_ops->get_maxrecs(cur, level))
2217 goto out0;
2218
2219 XFS_BTREE_STATS_INC(cur, rshift);
2220 XFS_BTREE_STATS_ADD(cur, moves, rrecs);
2221
2222 /*
2223 * Make a hole at the start of the right neighbor block, then
2224 * copy the last left block entry to the hole.
2225 */
2226 if (level > 0) {
2227 /* It's a nonleaf. make a hole in the keys and ptrs */
2228 union xfs_btree_key *lkp;
2229 union xfs_btree_ptr *lpp;
2230 union xfs_btree_ptr *rpp;
2231
2232 lkp = xfs_btree_key_addr(cur, lrecs, left);
2233 lpp = xfs_btree_ptr_addr(cur, lrecs, left);
2234 rkp = xfs_btree_key_addr(cur, 1, right);
2235 rpp = xfs_btree_ptr_addr(cur, 1, right);
2236
2237#ifdef DEBUG
2238 for (i = rrecs - 1; i >= 0; i--) {
2239 error = xfs_btree_check_ptr(cur, rpp, i, level);
2240 if (error)
2241 goto error0;
2242 }
2243#endif
2244
2245 xfs_btree_shift_keys(cur, rkp, 1, rrecs);
2246 xfs_btree_shift_ptrs(cur, rpp, 1, rrecs);
2247
2248#ifdef DEBUG
2249 error = xfs_btree_check_ptr(cur, lpp, 0, level);
2250 if (error)
2251 goto error0;
2252#endif
2253
2254 /* Now put the new data in, and log it. */
2255 xfs_btree_copy_keys(cur, rkp, lkp, 1);
2256 xfs_btree_copy_ptrs(cur, rpp, lpp, 1);
2257
2258 xfs_btree_log_keys(cur, rbp, 1, rrecs + 1);
2259 xfs_btree_log_ptrs(cur, rbp, 1, rrecs + 1);
2260
4a26e66e
CH		 2261 ASSERT(cur->bc_ops->keys_inorder(cur, rkp,
2262 xfs_btree_key_addr(cur, 2, right)));
9eaead51
CH		 2263 } else {
2264 /* It's a leaf. make a hole in the records */
2265 union xfs_btree_rec *lrp;
2266 union xfs_btree_rec *rrp;
2267
2268 lrp = xfs_btree_rec_addr(cur, lrecs, left);
2269 rrp = xfs_btree_rec_addr(cur, 1, right);
2270
2271 xfs_btree_shift_recs(cur, rrp, 1, rrecs);
2272
2273 /* Now put the new data in, and log it. */
2274 xfs_btree_copy_recs(cur, rrp, lrp, 1);
2275 xfs_btree_log_recs(cur, rbp, 1, rrecs + 1);
2276
2277 cur->bc_ops->init_key_from_rec(&key, rrp);
2278 rkp = &key;
2279
4a26e66e
CH		 2280 ASSERT(cur->bc_ops->recs_inorder(cur, rrp,
2281 xfs_btree_rec_addr(cur, 2, right)));
9eaead51
CH		 2282 }
2283
2284 /*
2285 * Decrement and log left's numrecs, bump and log right's numrecs.
2286 */
2287 xfs_btree_set_numrecs(left, --lrecs);
2288 xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS);
2289
2290 xfs_btree_set_numrecs(right, ++rrecs);
2291 xfs_btree_log_block(cur, rbp, XFS_BB_NUMRECS);
2292
2293 /*
2294 * Using a temporary cursor, update the parent key values of the
2295 * block on the right.
2296 */
2297 error = xfs_btree_dup_cursor(cur, &tcur);
2298 if (error)
2299 goto error0;
2300 i = xfs_btree_lastrec(tcur, level);
2301 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
2302
2303 error = xfs_btree_increment(tcur, level, &i);
2304 if (error)
2305 goto error1;
2306
2307 error = xfs_btree_updkey(tcur, rkp, level + 1);
2308 if (error)
2309 goto error1;
2310
2311 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
2312
2313 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2314 *stat = 1;
2315 return 0;
2316
2317out0:
2318 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2319 *stat = 0;
2320 return 0;
2321
2322error0:
2323 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2324 return error;
2325
2326error1:
2327 XFS_BTREE_TRACE_CURSOR(tcur, XBT_ERROR);
2328 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
2329 return error;
2330}
f5eb8e7c
CH		 2331
2332/*
2333 * Split cur/level block in half.
2334 * Return new block number and the key to its first
2335 * record (to be inserted into parent).
2336 */
3cc7524c 2337STATIC int /* error */
f5eb8e7c
CH		 2338xfs_btree_split(
2339 struct xfs_btree_cur *cur,
2340 int level,
2341 union xfs_btree_ptr *ptrp,
2342 union xfs_btree_key *key,
2343 struct xfs_btree_cur **curp,
2344 int *stat) /* success/failure */
2345{
2346 union xfs_btree_ptr lptr; /* left sibling block ptr */
2347 struct xfs_buf *lbp; /* left buffer pointer */
2348 struct xfs_btree_block *left; /* left btree block */
2349 union xfs_btree_ptr rptr; /* right sibling block ptr */
2350 struct xfs_buf *rbp; /* right buffer pointer */
2351 struct xfs_btree_block *right; /* right btree block */
2352 union xfs_btree_ptr rrptr; /* right-right sibling ptr */
2353 struct xfs_buf *rrbp; /* right-right buffer pointer */
2354 struct xfs_btree_block *rrblock; /* right-right btree block */
2355 int lrecs;
2356 int rrecs;
2357 int src_index;
2358 int error; /* error return value */
2359#ifdef DEBUG
2360 int i;
2361#endif
2362
2363 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2364 XFS_BTREE_TRACE_ARGIPK(cur, level, *ptrp, key);
2365
2366 XFS_BTREE_STATS_INC(cur, split);
2367
2368 /* Set up left block (current one). */
2369 left = xfs_btree_get_block(cur, level, &lbp);
2370
2371#ifdef DEBUG
2372 error = xfs_btree_check_block(cur, left, level, lbp);
2373 if (error)
2374 goto error0;
2375#endif
2376
2377 xfs_btree_buf_to_ptr(cur, lbp, &lptr);
2378
2379 /* Allocate the new block. If we can't do it, we're toast. Give up. */
2380 error = cur->bc_ops->alloc_block(cur, &lptr, &rptr, 1, stat);
2381 if (error)
2382 goto error0;
2383 if (*stat == 0)
2384 goto out0;
2385 XFS_BTREE_STATS_INC(cur, alloc);
2386
2387 /* Set up the new block as "right". */
2388 error = xfs_btree_get_buf_block(cur, &rptr, 0, &right, &rbp);
2389 if (error)
2390 goto error0;
2391
2392 /* Fill in the btree header for the new right block. */
ee1a47ab 2393 xfs_btree_init_block_cur(cur, rbp, xfs_btree_get_level(left), 0);
f5eb8e7c
CH		 2394
2395 /*
2396 * Split the entries between the old and the new block evenly.
2397 * Make sure that if there's an odd number of entries now, that
2398 * each new block will have the same number of entries.
2399 */
2400 lrecs = xfs_btree_get_numrecs(left);
2401 rrecs = lrecs / 2;
2402 if ((lrecs & 1) && cur->bc_ptrs[level] <= rrecs + 1)
2403 rrecs++;
2404 src_index = (lrecs - rrecs + 1);
2405
2406 XFS_BTREE_STATS_ADD(cur, moves, rrecs);
2407
2408 /*
2409 * Copy btree block entries from the left block over to the
2410 * new block, the right. Update the right block and log the
2411 * changes.
2412 */
2413 if (level > 0) {
2414 /* It's a non-leaf. Move keys and pointers. */
2415 union xfs_btree_key *lkp; /* left btree key */
2416 union xfs_btree_ptr *lpp; /* left address pointer */
2417 union xfs_btree_key *rkp; /* right btree key */
2418 union xfs_btree_ptr *rpp; /* right address pointer */
2419
2420 lkp = xfs_btree_key_addr(cur, src_index, left);
2421 lpp = xfs_btree_ptr_addr(cur, src_index, left);
2422 rkp = xfs_btree_key_addr(cur, 1, right);
2423 rpp = xfs_btree_ptr_addr(cur, 1, right);
2424
2425#ifdef DEBUG
2426 for (i = src_index; i < rrecs; i++) {
2427 error = xfs_btree_check_ptr(cur, lpp, i, level);
2428 if (error)
2429 goto error0;
2430 }
2431#endif
2432
2433 xfs_btree_copy_keys(cur, rkp, lkp, rrecs);
2434 xfs_btree_copy_ptrs(cur, rpp, lpp, rrecs);
2435
2436 xfs_btree_log_keys(cur, rbp, 1, rrecs);
2437 xfs_btree_log_ptrs(cur, rbp, 1, rrecs);
2438
2439 /* Grab the keys to the entries moved to the right block */
2440 xfs_btree_copy_keys(cur, key, rkp, 1);
2441 } else {
2442 /* It's a leaf. Move records. */
2443 union xfs_btree_rec *lrp; /* left record pointer */
2444 union xfs_btree_rec *rrp; /* right record pointer */
2445
2446 lrp = xfs_btree_rec_addr(cur, src_index, left);
2447 rrp = xfs_btree_rec_addr(cur, 1, right);
2448
2449 xfs_btree_copy_recs(cur, rrp, lrp, rrecs);
2450 xfs_btree_log_recs(cur, rbp, 1, rrecs);
2451
2452 cur->bc_ops->init_key_from_rec(key,
2453 xfs_btree_rec_addr(cur, 1, right));
2454 }
2455
2456
2457 /*
2458 * Find the left block number by looking in the buffer.
2459 * Adjust numrecs, sibling pointers.
2460 */
2461 xfs_btree_get_sibling(cur, left, &rrptr, XFS_BB_RIGHTSIB);
2462 xfs_btree_set_sibling(cur, right, &rrptr, XFS_BB_RIGHTSIB);
2463 xfs_btree_set_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
2464 xfs_btree_set_sibling(cur, left, &rptr, XFS_BB_RIGHTSIB);
2465
2466 lrecs -= rrecs;
2467 xfs_btree_set_numrecs(left, lrecs);
2468 xfs_btree_set_numrecs(right, xfs_btree_get_numrecs(right) + rrecs);
2469
2470 xfs_btree_log_block(cur, rbp, XFS_BB_ALL_BITS);
2471 xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS | XFS_BB_RIGHTSIB);
2472
2473 /*
2474 * If there's a block to the new block's right, make that block
2475 * point back to right instead of to left.
2476 */
2477 if (!xfs_btree_ptr_is_null(cur, &rrptr)) {
2478 error = xfs_btree_read_buf_block(cur, &rrptr, level,
2479 0, &rrblock, &rrbp);
2480 if (error)
2481 goto error0;
2482 xfs_btree_set_sibling(cur, rrblock, &rptr, XFS_BB_LEFTSIB);
2483 xfs_btree_log_block(cur, rrbp, XFS_BB_LEFTSIB);
2484 }
2485 /*
2486 * If the cursor is really in the right block, move it there.
2487 * If it's just pointing past the last entry in left, then we'll
2488 * insert there, so don't change anything in that case.
2489 */
2490 if (cur->bc_ptrs[level] > lrecs + 1) {
2491 xfs_btree_setbuf(cur, level, rbp);
2492 cur->bc_ptrs[level] -= lrecs;
2493 }
2494 /*
2495 * If there are more levels, we'll need another cursor which refers
2496 * the right block, no matter where this cursor was.
2497 */
2498 if (level + 1 < cur->bc_nlevels) {
2499 error = xfs_btree_dup_cursor(cur, curp);
2500 if (error)
2501 goto error0;
2502 (*curp)->bc_ptrs[level + 1]++;
2503 }
2504 *ptrp = rptr;
2505 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2506 *stat = 1;
2507 return 0;
2508out0:
2509 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2510 *stat = 0;
2511 return 0;
2512
2513error0:
2514 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2515 return error;
2516}
344207ce 2517
ea77b0a6
CH		 2518/*
2519 * Copy the old inode root contents into a real block and make the
2520 * broot point to it.
2521 */
2522int /* error */
2523xfs_btree_new_iroot(
2524 struct xfs_btree_cur *cur, /* btree cursor */
2525 int *logflags, /* logging flags for inode */
2526 int *stat) /* return status - 0 fail */
2527{
2528 struct xfs_buf *cbp; /* buffer for cblock */
2529 struct xfs_btree_block *block; /* btree block */
2530 struct xfs_btree_block *cblock; /* child btree block */
2531 union xfs_btree_key *ckp; /* child key pointer */
2532 union xfs_btree_ptr *cpp; /* child ptr pointer */
2533 union xfs_btree_key *kp; /* pointer to btree key */
2534 union xfs_btree_ptr *pp; /* pointer to block addr */
2535 union xfs_btree_ptr nptr; /* new block addr */
2536 int level; /* btree level */
2537 int error; /* error return code */
2538#ifdef DEBUG
2539 int i; /* loop counter */
2540#endif
2541
2542 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2543 XFS_BTREE_STATS_INC(cur, newroot);
2544
2545 ASSERT(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE);
2546
2547 level = cur->bc_nlevels - 1;
2548
2549 block = xfs_btree_get_iroot(cur);
2550 pp = xfs_btree_ptr_addr(cur, 1, block);
2551
2552 /* Allocate the new block. If we can't do it, we're toast. Give up. */
2553 error = cur->bc_ops->alloc_block(cur, pp, &nptr, 1, stat);
2554 if (error)
2555 goto error0;
2556 if (*stat == 0) {
2557 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2558 return 0;
2559 }
2560 XFS_BTREE_STATS_INC(cur, alloc);
2561
2562 /* Copy the root into a real block. */
2563 error = xfs_btree_get_buf_block(cur, &nptr, 0, &cblock, &cbp);
2564 if (error)
2565 goto error0;
2566
088c9f67
DC		 2567 /*
2568 * we can't just memcpy() the root in for CRC enabled btree blocks.
2569 * In that case have to also ensure the blkno remains correct
2570 */
ea77b0a6 2571 memcpy(cblock, block, xfs_btree_block_len(cur));
088c9f67
DC		 2572 if (cur->bc_flags & XFS_BTREE_CRC_BLOCKS) {
2573 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
2574 cblock->bb_u.l.bb_blkno = cpu_to_be64(cbp->b_bn);
2575 else
2576 cblock->bb_u.s.bb_blkno = cpu_to_be64(cbp->b_bn);
2577 }
ea77b0a6
CH		 2578
2579 be16_add_cpu(&block->bb_level, 1);
2580 xfs_btree_set_numrecs(block, 1);
2581 cur->bc_nlevels++;
2582 cur->bc_ptrs[level + 1] = 1;
2583
2584 kp = xfs_btree_key_addr(cur, 1, block);
2585 ckp = xfs_btree_key_addr(cur, 1, cblock);
2586 xfs_btree_copy_keys(cur, ckp, kp, xfs_btree_get_numrecs(cblock));
2587
2588 cpp = xfs_btree_ptr_addr(cur, 1, cblock);
2589#ifdef DEBUG
2590 for (i = 0; i < be16_to_cpu(cblock->bb_numrecs); i++) {
2591 error = xfs_btree_check_ptr(cur, pp, i, level);
2592 if (error)
2593 goto error0;
2594 }
2595#endif
2596 xfs_btree_copy_ptrs(cur, cpp, pp, xfs_btree_get_numrecs(cblock));
2597
2598#ifdef DEBUG
2599 error = xfs_btree_check_ptr(cur, &nptr, 0, level);
2600 if (error)
2601 goto error0;
2602#endif
2603 xfs_btree_copy_ptrs(cur, pp, &nptr, 1);
2604
2605 xfs_iroot_realloc(cur->bc_private.b.ip,
2606 1 - xfs_btree_get_numrecs(cblock),
2607 cur->bc_private.b.whichfork);
2608
2609 xfs_btree_setbuf(cur, level, cbp);
2610
2611 /*
2612 * Do all this logging at the end so that
2613 * the root is at the right level.
2614 */
2615 xfs_btree_log_block(cur, cbp, XFS_BB_ALL_BITS);
2616 xfs_btree_log_keys(cur, cbp, 1, be16_to_cpu(cblock->bb_numrecs));
2617 xfs_btree_log_ptrs(cur, cbp, 1, be16_to_cpu(cblock->bb_numrecs));
2618
2619 *logflags |=
9d87c319 2620 XFS_ILOG_CORE | xfs_ilog_fbroot(cur->bc_private.b.whichfork);
ea77b0a6
CH		 2621 *stat = 1;
2622 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2623 return 0;
2624error0:
2625 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2626 return error;
2627}
2628
344207ce
CH		 2629/*
2630 * Allocate a new root block, fill it in.
2631 */
3cc7524c 2632STATIC int /* error */
344207ce
CH		 2633xfs_btree_new_root(
2634 struct xfs_btree_cur *cur, /* btree cursor */
2635 int *stat) /* success/failure */
2636{
2637 struct xfs_btree_block *block; /* one half of the old root block */
2638 struct xfs_buf *bp; /* buffer containing block */
2639 int error; /* error return value */
2640 struct xfs_buf *lbp; /* left buffer pointer */
2641 struct xfs_btree_block *left; /* left btree block */
2642 struct xfs_buf *nbp; /* new (root) buffer */
2643 struct xfs_btree_block *new; /* new (root) btree block */
2644 int nptr; /* new value for key index, 1 or 2 */
2645 struct xfs_buf *rbp; /* right buffer pointer */
2646 struct xfs_btree_block *right; /* right btree block */
2647 union xfs_btree_ptr rptr;
2648 union xfs_btree_ptr lptr;
2649
2650 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2651 XFS_BTREE_STATS_INC(cur, newroot);
2652
2653 /* initialise our start point from the cursor */
2654 cur->bc_ops->init_ptr_from_cur(cur, &rptr);
2655
2656 /* Allocate the new block. If we can't do it, we're toast. Give up. */
2657 error = cur->bc_ops->alloc_block(cur, &rptr, &lptr, 1, stat);
2658 if (error)
2659 goto error0;
2660 if (*stat == 0)
2661 goto out0;
2662 XFS_BTREE_STATS_INC(cur, alloc);
2663
2664 /* Set up the new block. */
2665 error = xfs_btree_get_buf_block(cur, &lptr, 0, &new, &nbp);
2666 if (error)
2667 goto error0;
2668
2669 /* Set the root in the holding structure increasing the level by 1. */
2670 cur->bc_ops->set_root(cur, &lptr, 1);
2671
2672 /*
2673 * At the previous root level there are now two blocks: the old root,
2674 * and the new block generated when it was split. We don't know which
2675 * one the cursor is pointing at, so we set up variables "left" and
2676 * "right" for each case.
2677 */
2678 block = xfs_btree_get_block(cur, cur->bc_nlevels - 1, &bp);
2679
2680#ifdef DEBUG
2681 error = xfs_btree_check_block(cur, block, cur->bc_nlevels - 1, bp);
2682 if (error)
2683 goto error0;
2684#endif
2685
2686 xfs_btree_get_sibling(cur, block, &rptr, XFS_BB_RIGHTSIB);
2687 if (!xfs_btree_ptr_is_null(cur, &rptr)) {
2688 /* Our block is left, pick up the right block. */
2689 lbp = bp;
2690 xfs_btree_buf_to_ptr(cur, lbp, &lptr);
2691 left = block;
2692 error = xfs_btree_read_buf_block(cur, &rptr,
2693 cur->bc_nlevels - 1, 0, &right, &rbp);
2694 if (error)
2695 goto error0;
2696 bp = rbp;
2697 nptr = 1;
2698 } else {
2699 /* Our block is right, pick up the left block. */
2700 rbp = bp;
2701 xfs_btree_buf_to_ptr(cur, rbp, &rptr);
2702 right = block;
2703 xfs_btree_get_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
2704 error = xfs_btree_read_buf_block(cur, &lptr,
2705 cur->bc_nlevels - 1, 0, &left, &lbp);
2706 if (error)
2707 goto error0;
2708 bp = lbp;
2709 nptr = 2;
2710 }
2711 /* Fill in the new block's btree header and log it. */
ee1a47ab 2712 xfs_btree_init_block_cur(cur, nbp, cur->bc_nlevels, 2);
344207ce
CH		 2713 xfs_btree_log_block(cur, nbp, XFS_BB_ALL_BITS);
2714 ASSERT(!xfs_btree_ptr_is_null(cur, &lptr) &&
2715 !xfs_btree_ptr_is_null(cur, &rptr));
2716
2717 /* Fill in the key data in the new root. */
2718 if (xfs_btree_get_level(left) > 0) {
2719 xfs_btree_copy_keys(cur,
2720 xfs_btree_key_addr(cur, 1, new),
2721 xfs_btree_key_addr(cur, 1, left), 1);
2722 xfs_btree_copy_keys(cur,
2723 xfs_btree_key_addr(cur, 2, new),
2724 xfs_btree_key_addr(cur, 1, right), 1);
2725 } else {
2726 cur->bc_ops->init_key_from_rec(
2727 xfs_btree_key_addr(cur, 1, new),
2728 xfs_btree_rec_addr(cur, 1, left));
2729 cur->bc_ops->init_key_from_rec(
2730 xfs_btree_key_addr(cur, 2, new),
2731 xfs_btree_rec_addr(cur, 1, right));
2732 }
2733 xfs_btree_log_keys(cur, nbp, 1, 2);
2734
2735 /* Fill in the pointer data in the new root. */
2736 xfs_btree_copy_ptrs(cur,
2737 xfs_btree_ptr_addr(cur, 1, new), &lptr, 1);
2738 xfs_btree_copy_ptrs(cur,
2739 xfs_btree_ptr_addr(cur, 2, new), &rptr, 1);
2740 xfs_btree_log_ptrs(cur, nbp, 1, 2);
2741
2742 /* Fix up the cursor. */
2743 xfs_btree_setbuf(cur, cur->bc_nlevels, nbp);
2744 cur->bc_ptrs[cur->bc_nlevels] = nptr;
2745 cur->bc_nlevels++;
2746 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2747 *stat = 1;
2748 return 0;
2749error0:
2750 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2751 return error;
2752out0:
2753 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2754 *stat = 0;
2755 return 0;
2756}
4b22a571
CH		 2757
2758STATIC int
2759xfs_btree_make_block_unfull(
2760 struct xfs_btree_cur *cur, /* btree cursor */
2761 int level, /* btree level */
2762 int numrecs,/* # of recs in block */
2763 int *oindex,/* old tree index */
2764 int *index, /* new tree index */
2765 union xfs_btree_ptr *nptr, /* new btree ptr */
2766 struct xfs_btree_cur **ncur, /* new btree cursor */
2767 union xfs_btree_rec *nrec, /* new record */
2768 int *stat)
2769{
2770 union xfs_btree_key key; /* new btree key value */
2771 int error = 0;
2772
2773 if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
2774 level == cur->bc_nlevels - 1) {
2775 struct xfs_inode *ip = cur->bc_private.b.ip;
2776
2777 if (numrecs < cur->bc_ops->get_dmaxrecs(cur, level)) {
2778 /* A root block that can be made bigger. */
4b22a571
CH		 2779 xfs_iroot_realloc(ip, 1, cur->bc_private.b.whichfork);
2780 } else {
2781 /* A root block that needs replacing */
2782 int logflags = 0;
2783
2784 error = xfs_btree_new_iroot(cur, &logflags, stat);
2785 if (error || *stat == 0)
2786 return error;
2787
2788 xfs_trans_log_inode(cur->bc_tp, ip, logflags);
2789 }
2790
2791 return 0;
2792 }
2793
2794 /* First, try shifting an entry to the right neighbor. */
2795 error = xfs_btree_rshift(cur, level, stat);
2796 if (error || *stat)
2797 return error;
2798
2799 /* Next, try shifting an entry to the left neighbor. */
2800 error = xfs_btree_lshift(cur, level, stat);
2801 if (error)
2802 return error;
2803
2804 if (*stat) {
2805 *oindex = *index = cur->bc_ptrs[level];
2806 return 0;
2807 }
2808
2809 /*
2810 * Next, try splitting the current block in half.
2811 *
2812 * If this works we have to re-set our variables because we
2813 * could be in a different block now.
2814 */
2815 error = xfs_btree_split(cur, level, nptr, &key, ncur, stat);
2816 if (error || *stat == 0)
2817 return error;
2818
2819
2820 *index = cur->bc_ptrs[level];
2821 cur->bc_ops->init_rec_from_key(&key, nrec);
2822 return 0;
2823}
2824
2825/*
2826 * Insert one record/level. Return information to the caller
2827 * allowing the next level up to proceed if necessary.
2828 */
2829STATIC int
2830xfs_btree_insrec(
2831 struct xfs_btree_cur *cur, /* btree cursor */
2832 int level, /* level to insert record at */
2833 union xfs_btree_ptr *ptrp, /* i/o: block number inserted */
2834 union xfs_btree_rec *recp, /* i/o: record data inserted */
2835 struct xfs_btree_cur **curp, /* output: new cursor replacing cur */
2836 int *stat) /* success/failure */
2837{
2838 struct xfs_btree_block *block; /* btree block */
2839 struct xfs_buf *bp; /* buffer for block */
2840 union xfs_btree_key key; /* btree key */
2841 union xfs_btree_ptr nptr; /* new block ptr */
2842 struct xfs_btree_cur *ncur; /* new btree cursor */
2843 union xfs_btree_rec nrec; /* new record count */
2844 int optr; /* old key/record index */
2845 int ptr; /* key/record index */
2846 int numrecs;/* number of records */
2847 int error; /* error return value */
2848#ifdef DEBUG
2849 int i;
2850#endif
2851
2852 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2853 XFS_BTREE_TRACE_ARGIPR(cur, level, *ptrp, recp);
2854
2855 ncur = NULL;
2856
2857 /*
2858 * If we have an external root pointer, and we've made it to the
2859 * root level, allocate a new root block and we're done.
2860 */
2861 if (!(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
2862 (level >= cur->bc_nlevels)) {
2863 error = xfs_btree_new_root(cur, stat);
2864 xfs_btree_set_ptr_null(cur, ptrp);
2865
2866 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2867 return error;
2868 }
2869
2870 /* If we're off the left edge, return failure. */
2871 ptr = cur->bc_ptrs[level];
2872 if (ptr == 0) {
2873 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2874 *stat = 0;
2875 return 0;
2876 }
2877
2878 /* Make a key out of the record data to be inserted, and save it. */
2879 cur->bc_ops->init_key_from_rec(&key, recp);
2880
2881 optr = ptr;
2882
2883 XFS_BTREE_STATS_INC(cur, insrec);
2884
2885 /* Get pointers to the btree buffer and block. */
2886 block = xfs_btree_get_block(cur, level, &bp);
2887 numrecs = xfs_btree_get_numrecs(block);
2888
2889#ifdef DEBUG
2890 error = xfs_btree_check_block(cur, block, level, bp);
2891 if (error)
2892 goto error0;
2893
2894 /* Check that the new entry is being inserted in the right place. */
2895 if (ptr <= numrecs) {
2896 if (level == 0) {
4a26e66e
CH		 2897 ASSERT(cur->bc_ops->recs_inorder(cur, recp,
2898 xfs_btree_rec_addr(cur, ptr, block)));
4b22a571 2899 } else {
4a26e66e
CH		 2900 ASSERT(cur->bc_ops->keys_inorder(cur, &key,
2901 xfs_btree_key_addr(cur, ptr, block)));
4b22a571
CH		 2902 }
2903 }
2904#endif
2905
2906 /*
2907 * If the block is full, we can't insert the new entry until we
2908 * make the block un-full.
2909 */
2910 xfs_btree_set_ptr_null(cur, &nptr);
2911 if (numrecs == cur->bc_ops->get_maxrecs(cur, level)) {
2912 error = xfs_btree_make_block_unfull(cur, level, numrecs,
2913 &optr, &ptr, &nptr, &ncur, &nrec, stat);
2914 if (error || *stat == 0)
2915 goto error0;
2916 }
2917
2918 /*
2919 * The current block may have changed if the block was
2920 * previously full and we have just made space in it.
2921 */
2922 block = xfs_btree_get_block(cur, level, &bp);
2923 numrecs = xfs_btree_get_numrecs(block);
2924
2925#ifdef DEBUG
2926 error = xfs_btree_check_block(cur, block, level, bp);
2927 if (error)
2928 return error;
2929#endif
2930
2931 /*
2932 * At this point we know there's room for our new entry in the block
2933 * we're pointing at.
2934 */
2935 XFS_BTREE_STATS_ADD(cur, moves, numrecs - ptr + 1);
2936
2937 if (level > 0) {
2938 /* It's a nonleaf. make a hole in the keys and ptrs */
2939 union xfs_btree_key *kp;
2940 union xfs_btree_ptr *pp;
2941
2942 kp = xfs_btree_key_addr(cur, ptr, block);
2943 pp = xfs_btree_ptr_addr(cur, ptr, block);
2944
2945#ifdef DEBUG
2946 for (i = numrecs - ptr; i >= 0; i--) {
2947 error = xfs_btree_check_ptr(cur, pp, i, level);
2948 if (error)
2949 return error;
2950 }
2951#endif
2952
2953 xfs_btree_shift_keys(cur, kp, 1, numrecs - ptr + 1);
2954 xfs_btree_shift_ptrs(cur, pp, 1, numrecs - ptr + 1);
2955
2956#ifdef DEBUG
2957 error = xfs_btree_check_ptr(cur, ptrp, 0, level);
2958 if (error)
2959 goto error0;
2960#endif
2961
2962 /* Now put the new data in, bump numrecs and log it. */
2963 xfs_btree_copy_keys(cur, kp, &key, 1);
2964 xfs_btree_copy_ptrs(cur, pp, ptrp, 1);
2965 numrecs++;
2966 xfs_btree_set_numrecs(block, numrecs);
2967 xfs_btree_log_ptrs(cur, bp, ptr, numrecs);
2968 xfs_btree_log_keys(cur, bp, ptr, numrecs);
2969#ifdef DEBUG
2970 if (ptr < numrecs) {
4a26e66e
CH		 2971 ASSERT(cur->bc_ops->keys_inorder(cur, kp,
2972 xfs_btree_key_addr(cur, ptr + 1, block)));
4b22a571
CH		 2973 }
2974#endif
2975 } else {
2976 /* It's a leaf. make a hole in the records */
2977 union xfs_btree_rec *rp;
2978
2979 rp = xfs_btree_rec_addr(cur, ptr, block);
2980
2981 xfs_btree_shift_recs(cur, rp, 1, numrecs - ptr + 1);
2982
2983 /* Now put the new data in, bump numrecs and log it. */
2984 xfs_btree_copy_recs(cur, rp, recp, 1);
2985 xfs_btree_set_numrecs(block, ++numrecs);
2986 xfs_btree_log_recs(cur, bp, ptr, numrecs);
2987#ifdef DEBUG
2988 if (ptr < numrecs) {
4a26e66e
CH		 2989 ASSERT(cur->bc_ops->recs_inorder(cur, rp,
2990 xfs_btree_rec_addr(cur, ptr + 1, block)));
4b22a571
CH		 2991 }
2992#endif
2993 }
2994
2995 /* Log the new number of records in the btree header. */
2996 xfs_btree_log_block(cur, bp, XFS_BB_NUMRECS);
2997
2998 /* If we inserted at the start of a block, update the parents' keys. */
2999 if (optr == 1) {
3000 error = xfs_btree_updkey(cur, &key, level + 1);
3001 if (error)
3002 goto error0;
3003 }
3004
3005 /*
3006 * If we are tracking the last record in the tree and
3007 * we are at the far right edge of the tree, update it.
3008 */
3009 if (xfs_btree_is_lastrec(cur, block, level)) {
3010 cur->bc_ops->update_lastrec(cur, block, recp,
3011 ptr, LASTREC_INSREC);
3012 }
3013
3014 /*
3015 * Return the new block number, if any.
3016 * If there is one, give back a record value and a cursor too.
3017 */
3018 *ptrp = nptr;
3019 if (!xfs_btree_ptr_is_null(cur, &nptr)) {
3020 *recp = nrec;
3021 *curp = ncur;
3022 }
3023
3024 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3025 *stat = 1;
3026 return 0;
3027
3028error0:
3029 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3030 return error;
3031}
3032
3033/*
3034 * Insert the record at the point referenced by cur.
3035 *
3036 * A multi-level split of the tree on insert will invalidate the original
3037 * cursor. All callers of this function should assume that the cursor is
3038 * no longer valid and revalidate it.
3039 */
3040int
3041xfs_btree_insert(
3042 struct xfs_btree_cur *cur,
3043 int *stat)
3044{
3045 int error; /* error return value */
3046 int i; /* result value, 0 for failure */
3047 int level; /* current level number in btree */
3048 union xfs_btree_ptr nptr; /* new block number (split result) */
3049 struct xfs_btree_cur *ncur; /* new cursor (split result) */
3050 struct xfs_btree_cur *pcur; /* previous level's cursor */
3051 union xfs_btree_rec rec; /* record to insert */
3052
3053 level = 0;
3054 ncur = NULL;
3055 pcur = cur;
3056
3057 xfs_btree_set_ptr_null(cur, &nptr);
3058 cur->bc_ops->init_rec_from_cur(cur, &rec);
3059
3060 /*
3061 * Loop going up the tree, starting at the leaf level.
3062 * Stop when we don't get a split block, that must mean that
3063 * the insert is finished with this level.
3064 */
3065 do {
3066 /*
3067 * Insert nrec/nptr into this level of the tree.
3068 * Note if we fail, nptr will be null.
3069 */
3070 error = xfs_btree_insrec(pcur, level, &nptr, &rec, &ncur, &i);
3071 if (error) {
3072 if (pcur != cur)
3073 xfs_btree_del_cursor(pcur, XFS_BTREE_ERROR);
3074 goto error0;
3075 }
3076
3077 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
3078 level++;
3079
3080 /*
3081 * See if the cursor we just used is trash.
3082 * Can't trash the caller's cursor, but otherwise we should
3083 * if ncur is a new cursor or we're about to be done.
3084 */
3085 if (pcur != cur &&
3086 (ncur || xfs_btree_ptr_is_null(cur, &nptr))) {
3087 /* Save the state from the cursor before we trash it */
3088 if (cur->bc_ops->update_cursor)
3089 cur->bc_ops->update_cursor(pcur, cur);
3090 cur->bc_nlevels = pcur->bc_nlevels;
3091 xfs_btree_del_cursor(pcur, XFS_BTREE_NOERROR);
3092 }
3093 /* If we got a new cursor, switch to it. */
3094 if (ncur) {
3095 pcur = ncur;
3096 ncur = NULL;
3097 }
3098 } while (!xfs_btree_ptr_is_null(cur, &nptr));
3099
3100 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3101 *stat = i;
3102 return 0;
3103error0:
3104 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3105 return error;
3106}
d4b3a4b7
CH		 3107
3108/*
3109 * Try to merge a non-leaf block back into the inode root.
3110 *
3111 * Note: the killroot names comes from the fact that we're effectively
3112 * killing the old root block. But because we can't just delete the
3113 * inode we have to copy the single block it was pointing to into the
3114 * inode.
3115 */
d96f8f89 3116STATIC int
d4b3a4b7
CH		 3117xfs_btree_kill_iroot(
3118 struct xfs_btree_cur *cur)
3119{
3120 int whichfork = cur->bc_private.b.whichfork;
3121 struct xfs_inode *ip = cur->bc_private.b.ip;
3122 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
3123 struct xfs_btree_block *block;
3124 struct xfs_btree_block *cblock;
3125 union xfs_btree_key *kp;
3126 union xfs_btree_key *ckp;
3127 union xfs_btree_ptr *pp;
3128 union xfs_btree_ptr *cpp;
3129 struct xfs_buf *cbp;
3130 int level;
3131 int index;
3132 int numrecs;
3133#ifdef DEBUG
3134 union xfs_btree_ptr ptr;
3135 int i;
3136#endif
3137
3138 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
3139
3140 ASSERT(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE);
3141 ASSERT(cur->bc_nlevels > 1);
3142
3143 /*
3144 * Don't deal with the root block needs to be a leaf case.
3145 * We're just going to turn the thing back into extents anyway.
3146 */
3147 level = cur->bc_nlevels - 1;
3148 if (level == 1)
3149 goto out0;
3150
3151 /*
3152 * Give up if the root has multiple children.
3153 */
3154 block = xfs_btree_get_iroot(cur);
3155 if (xfs_btree_get_numrecs(block) != 1)
3156 goto out0;
3157
3158 cblock = xfs_btree_get_block(cur, level - 1, &cbp);
3159 numrecs = xfs_btree_get_numrecs(cblock);
3160
3161 /*
3162 * Only do this if the next level will fit.
3163 * Then the data must be copied up to the inode,
3164 * instead of freeing the root you free the next level.
3165 */
3166 if (numrecs > cur->bc_ops->get_dmaxrecs(cur, level))
3167 goto out0;
3168
3169 XFS_BTREE_STATS_INC(cur, killroot);
3170
3171#ifdef DEBUG
3172 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_LEFTSIB);
3173 ASSERT(xfs_btree_ptr_is_null(cur, &ptr));
3174 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
3175 ASSERT(xfs_btree_ptr_is_null(cur, &ptr));
3176#endif
3177
3178 index = numrecs - cur->bc_ops->get_maxrecs(cur, level);
3179 if (index) {
3180 xfs_iroot_realloc(cur->bc_private.b.ip, index,
3181 cur->bc_private.b.whichfork);
7cc95a82 3182 block = ifp->if_broot;
d4b3a4b7
CH		 3183 }
3184
3185 be16_add_cpu(&block->bb_numrecs, index);
3186 ASSERT(block->bb_numrecs == cblock->bb_numrecs);
3187
3188 kp = xfs_btree_key_addr(cur, 1, block);
3189 ckp = xfs_btree_key_addr(cur, 1, cblock);
3190 xfs_btree_copy_keys(cur, kp, ckp, numrecs);
3191
3192 pp = xfs_btree_ptr_addr(cur, 1, block);
3193 cpp = xfs_btree_ptr_addr(cur, 1, cblock);
3194#ifdef DEBUG
3195 for (i = 0; i < numrecs; i++) {
3196 int error;
3197
3198 error = xfs_btree_check_ptr(cur, cpp, i, level - 1);
3199 if (error) {
3200 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3201 return error;
3202 }
3203 }
3204#endif
3205 xfs_btree_copy_ptrs(cur, pp, cpp, numrecs);
3206
3207 cur->bc_ops->free_block(cur, cbp);
3208 XFS_BTREE_STATS_INC(cur, free);
3209
3210 cur->bc_bufs[level - 1] = NULL;
3211 be16_add_cpu(&block->bb_level, -1);
3212 xfs_trans_log_inode(cur->bc_tp, ip,
9d87c319 3213 XFS_ILOG_CORE | xfs_ilog_fbroot(cur->bc_private.b.whichfork));
d4b3a4b7
CH		 3214 cur->bc_nlevels--;
3215out0:
3216 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3217 return 0;
3218}
91cca5df 3219
c0e59e1a
CH		 3220/*
3221 * Kill the current root node, and replace it with it's only child node.
3222 */
3223STATIC int
3224xfs_btree_kill_root(
3225 struct xfs_btree_cur *cur,
3226 struct xfs_buf *bp,
3227 int level,
3228 union xfs_btree_ptr *newroot)
3229{
3230 int error;
3231
3232 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
3233 XFS_BTREE_STATS_INC(cur, killroot);
3234
3235 /*
3236 * Update the root pointer, decreasing the level by 1 and then
3237 * free the old root.
3238 */
3239 cur->bc_ops->set_root(cur, newroot, -1);
3240
3241 error = cur->bc_ops->free_block(cur, bp);
3242 if (error) {
3243 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3244 return error;
3245 }
3246
3247 XFS_BTREE_STATS_INC(cur, free);
3248
3249 cur->bc_bufs[level] = NULL;
3250 cur->bc_ra[level] = 0;
3251 cur->bc_nlevels--;
3252
3253 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3254 return 0;
3255}
3256
91cca5df
CH		 3257STATIC int
3258xfs_btree_dec_cursor(
3259 struct xfs_btree_cur *cur,
3260 int level,
3261 int *stat)
3262{
3263 int error;
3264 int i;
3265
3266 if (level > 0) {
3267 error = xfs_btree_decrement(cur, level, &i);
3268 if (error)
3269 return error;
3270 }
3271
3272 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3273 *stat = 1;
3274 return 0;
3275}
3276
3277/*
3278 * Single level of the btree record deletion routine.
3279 * Delete record pointed to by cur/level.
3280 * Remove the record from its block then rebalance the tree.
3281 * Return 0 for error, 1 for done, 2 to go on to the next level.
3282 */
3283STATIC int /* error */
3284xfs_btree_delrec(
3285 struct xfs_btree_cur *cur, /* btree cursor */
3286 int level, /* level removing record from */
3287 int *stat) /* fail/done/go-on */
3288{
3289 struct xfs_btree_block *block; /* btree block */
3290 union xfs_btree_ptr cptr; /* current block ptr */
3291 struct xfs_buf *bp; /* buffer for block */
3292 int error; /* error return value */
3293 int i; /* loop counter */
3294 union xfs_btree_key key; /* storage for keyp */
3295 union xfs_btree_key *keyp = &key; /* passed to the next level */
3296 union xfs_btree_ptr lptr; /* left sibling block ptr */
3297 struct xfs_buf *lbp; /* left buffer pointer */
3298 struct xfs_btree_block *left; /* left btree block */
3299 int lrecs = 0; /* left record count */
3300 int ptr; /* key/record index */
3301 union xfs_btree_ptr rptr; /* right sibling block ptr */
3302 struct xfs_buf *rbp; /* right buffer pointer */
3303 struct xfs_btree_block *right; /* right btree block */
3304 struct xfs_btree_block *rrblock; /* right-right btree block */
3305 struct xfs_buf *rrbp; /* right-right buffer pointer */
3306 int rrecs = 0; /* right record count */
3307 struct xfs_btree_cur *tcur; /* temporary btree cursor */
3308 int numrecs; /* temporary numrec count */
3309
3310 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
3311 XFS_BTREE_TRACE_ARGI(cur, level);
3312
3313 tcur = NULL;
3314
3315 /* Get the index of the entry being deleted, check for nothing there. */
3316 ptr = cur->bc_ptrs[level];
3317 if (ptr == 0) {
3318 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3319 *stat = 0;
3320 return 0;
3321 }
3322
3323 /* Get the buffer & block containing the record or key/ptr. */
3324 block = xfs_btree_get_block(cur, level, &bp);
3325 numrecs = xfs_btree_get_numrecs(block);
3326
3327#ifdef DEBUG
3328 error = xfs_btree_check_block(cur, block, level, bp);
3329 if (error)
3330 goto error0;
3331#endif
3332
3333 /* Fail if we're off the end of the block. */
3334 if (ptr > numrecs) {
3335 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3336 *stat = 0;
3337 return 0;
3338 }
3339
3340 XFS_BTREE_STATS_INC(cur, delrec);
3341 XFS_BTREE_STATS_ADD(cur, moves, numrecs - ptr);
3342
3343 /* Excise the entries being deleted. */
3344 if (level > 0) {
3345 /* It's a nonleaf. operate on keys and ptrs */
3346 union xfs_btree_key *lkp;
3347 union xfs_btree_ptr *lpp;
3348
3349 lkp = xfs_btree_key_addr(cur, ptr + 1, block);
3350 lpp = xfs_btree_ptr_addr(cur, ptr + 1, block);
3351
3352#ifdef DEBUG
3353 for (i = 0; i < numrecs - ptr; i++) {
3354 error = xfs_btree_check_ptr(cur, lpp, i, level);
3355 if (error)
3356 goto error0;
3357 }
3358#endif
3359
3360 if (ptr < numrecs) {
3361 xfs_btree_shift_keys(cur, lkp, -1, numrecs - ptr);
3362 xfs_btree_shift_ptrs(cur, lpp, -1, numrecs - ptr);
3363 xfs_btree_log_keys(cur, bp, ptr, numrecs - 1);
3364 xfs_btree_log_ptrs(cur, bp, ptr, numrecs - 1);
3365 }
3366
3367 /*
3368 * If it's the first record in the block, we'll need to pass a
3369 * key up to the next level (updkey).
3370 */
3371 if (ptr == 1)
3372 keyp = xfs_btree_key_addr(cur, 1, block);
3373 } else {
3374 /* It's a leaf. operate on records */
3375 if (ptr < numrecs) {
3376 xfs_btree_shift_recs(cur,
3377 xfs_btree_rec_addr(cur, ptr + 1, block),
3378 -1, numrecs - ptr);
3379 xfs_btree_log_recs(cur, bp, ptr, numrecs - 1);
3380 }
3381
3382 /*
3383 * If it's the first record in the block, we'll need a key
3384 * structure to pass up to the next level (updkey).
3385 */
3386 if (ptr == 1) {
3387 cur->bc_ops->init_key_from_rec(&key,
3388 xfs_btree_rec_addr(cur, 1, block));
3389 keyp = &key;
3390 }
3391 }
3392
3393 /*
3394 * Decrement and log the number of entries in the block.
3395 */
3396 xfs_btree_set_numrecs(block, --numrecs);
3397 xfs_btree_log_block(cur, bp, XFS_BB_NUMRECS);
3398
3399 /*
3400 * If we are tracking the last record in the tree and
3401 * we are at the far right edge of the tree, update it.
3402 */
3403 if (xfs_btree_is_lastrec(cur, block, level)) {
3404 cur->bc_ops->update_lastrec(cur, block, NULL,
3405 ptr, LASTREC_DELREC);
3406 }
3407
3408 /*
3409 * We're at the root level. First, shrink the root block in-memory.
3410 * Try to get rid of the next level down. If we can't then there's
3411 * nothing left to do.
3412 */
3413 if (level == cur->bc_nlevels - 1) {
3414 if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) {
3415 xfs_iroot_realloc(cur->bc_private.b.ip, -1,
3416 cur->bc_private.b.whichfork);
3417
3418 error = xfs_btree_kill_iroot(cur);
3419 if (error)
3420 goto error0;
3421
3422 error = xfs_btree_dec_cursor(cur, level, stat);
3423 if (error)
3424 goto error0;
3425 *stat = 1;
3426 return 0;
3427 }
3428
3429 /*
3430 * If this is the root level, and there's only one entry left,
3431 * and it's NOT the leaf level, then we can get rid of this
3432 * level.
3433 */
3434 if (numrecs == 1 && level > 0) {
3435 union xfs_btree_ptr *pp;
3436 /*
3437 * pp is still set to the first pointer in the block.
3438 * Make it the new root of the btree.
3439 */
3440 pp = xfs_btree_ptr_addr(cur, 1, block);
c0e59e1a 3441 error = xfs_btree_kill_root(cur, bp, level, pp);
91cca5df
CH		 3442 if (error)
3443 goto error0;
3444 } else if (level > 0) {
3445 error = xfs_btree_dec_cursor(cur, level, stat);
3446 if (error)
3447 goto error0;
3448 }
3449 *stat = 1;
3450 return 0;
3451 }
3452
3453 /*
3454 * If we deleted the leftmost entry in the block, update the
3455 * key values above us in the tree.
3456 */
3457 if (ptr == 1) {
3458 error = xfs_btree_updkey(cur, keyp, level + 1);
3459 if (error)
3460 goto error0;
3461 }
3462
3463 /*
3464 * If the number of records remaining in the block is at least
3465 * the minimum, we're done.
3466 */
3467 if (numrecs >= cur->bc_ops->get_minrecs(cur, level)) {
3468 error = xfs_btree_dec_cursor(cur, level, stat);
3469 if (error)
3470 goto error0;
3471 return 0;
3472 }
3473
3474 /*
3475 * Otherwise, we have to move some records around to keep the
3476 * tree balanced. Look at the left and right sibling blocks to
3477 * see if we can re-balance by moving only one record.
3478 */
3479 xfs_btree_get_sibling(cur, block, &rptr, XFS_BB_RIGHTSIB);
3480 xfs_btree_get_sibling(cur, block, &lptr, XFS_BB_LEFTSIB);
3481
3482 if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) {
3483 /*
3484 * One child of root, need to get a chance to copy its contents
3485 * into the root and delete it. Can't go up to next level,
3486 * there's nothing to delete there.
3487 */
3488 if (xfs_btree_ptr_is_null(cur, &rptr) &&
3489 xfs_btree_ptr_is_null(cur, &lptr) &&
3490 level == cur->bc_nlevels - 2) {
3491 error = xfs_btree_kill_iroot(cur);
3492 if (!error)
3493 error = xfs_btree_dec_cursor(cur, level, stat);
3494 if (error)
3495 goto error0;
3496 return 0;
3497 }
3498 }
3499
3500 ASSERT(!xfs_btree_ptr_is_null(cur, &rptr) ||
3501 !xfs_btree_ptr_is_null(cur, &lptr));
3502
3503 /*
3504 * Duplicate the cursor so our btree manipulations here won't
3505 * disrupt the next level up.
3506 */
3507 error = xfs_btree_dup_cursor(cur, &tcur);
3508 if (error)
3509 goto error0;
3510
3511 /*
3512 * If there's a right sibling, see if it's ok to shift an entry
3513 * out of it.
3514 */
3515 if (!xfs_btree_ptr_is_null(cur, &rptr)) {
3516 /*
3517 * Move the temp cursor to the last entry in the next block.
3518 * Actually any entry but the first would suffice.
3519 */
3520 i = xfs_btree_lastrec(tcur, level);
3521 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
3522
3523 error = xfs_btree_increment(tcur, level, &i);
3524 if (error)
3525 goto error0;
3526 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
3527
3528 i = xfs_btree_lastrec(tcur, level);
3529 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
3530
3531 /* Grab a pointer to the block. */
3532 right = xfs_btree_get_block(tcur, level, &rbp);
3533#ifdef DEBUG
3534 error = xfs_btree_check_block(tcur, right, level, rbp);
3535 if (error)
3536 goto error0;
3537#endif
3538 /* Grab the current block number, for future use. */
3539 xfs_btree_get_sibling(tcur, right, &cptr, XFS_BB_LEFTSIB);
3540
3541 /*
3542 * If right block is full enough so that removing one entry
3543 * won't make it too empty, and left-shifting an entry out
3544 * of right to us works, we're done.
3545 */
3546 if (xfs_btree_get_numrecs(right) - 1 >=
3547 cur->bc_ops->get_minrecs(tcur, level)) {
3548 error = xfs_btree_lshift(tcur, level, &i);
3549 if (error)
3550 goto error0;
3551 if (i) {
3552 ASSERT(xfs_btree_get_numrecs(block) >=
3553 cur->bc_ops->get_minrecs(tcur, level));
3554
3555 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
3556 tcur = NULL;
3557
3558 error = xfs_btree_dec_cursor(cur, level, stat);
3559 if (error)
3560 goto error0;
3561 return 0;
3562 }
3563 }
3564
3565 /*
3566 * Otherwise, grab the number of records in right for
3567 * future reference, and fix up the temp cursor to point
3568 * to our block again (last record).
3569 */
3570 rrecs = xfs_btree_get_numrecs(right);
3571 if (!xfs_btree_ptr_is_null(cur, &lptr)) {
3572 i = xfs_btree_firstrec(tcur, level);
3573 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
3574
3575 error = xfs_btree_decrement(tcur, level, &i);
3576 if (error)
3577 goto error0;
3578 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
3579 }
3580 }
3581
3582 /*
3583 * If there's a left sibling, see if it's ok to shift an entry
3584 * out of it.
3585 */
3586 if (!xfs_btree_ptr_is_null(cur, &lptr)) {
3587 /*
3588 * Move the temp cursor to the first entry in the
3589 * previous block.
3590 */
3591 i = xfs_btree_firstrec(tcur, level);
3592 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
3593
3594 error = xfs_btree_decrement(tcur, level, &i);
3595 if (error)
3596 goto error0;
3597 i = xfs_btree_firstrec(tcur, level);
3598 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
3599
3600 /* Grab a pointer to the block. */
3601 left = xfs_btree_get_block(tcur, level, &lbp);
3602#ifdef DEBUG
3603 error = xfs_btree_check_block(cur, left, level, lbp);
3604 if (error)
3605 goto error0;
3606#endif
3607 /* Grab the current block number, for future use. */
3608 xfs_btree_get_sibling(tcur, left, &cptr, XFS_BB_RIGHTSIB);
3609
3610 /*
3611 * If left block is full enough so that removing one entry
3612 * won't make it too empty, and right-shifting an entry out
3613 * of left to us works, we're done.
3614 */
3615 if (xfs_btree_get_numrecs(left) - 1 >=
3616 cur->bc_ops->get_minrecs(tcur, level)) {
3617 error = xfs_btree_rshift(tcur, level, &i);
3618 if (error)
3619 goto error0;
3620 if (i) {
3621 ASSERT(xfs_btree_get_numrecs(block) >=
3622 cur->bc_ops->get_minrecs(tcur, level));
3623 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
3624 tcur = NULL;
3625 if (level == 0)
3626 cur->bc_ptrs[0]++;
3627 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3628 *stat = 1;
3629 return 0;
3630 }
3631 }
3632
3633 /*
3634 * Otherwise, grab the number of records in right for
3635 * future reference.
3636 */
3637 lrecs = xfs_btree_get_numrecs(left);
3638 }
3639
3640 /* Delete the temp cursor, we're done with it. */
3641 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
3642 tcur = NULL;
3643
3644 /* If here, we need to do a join to keep the tree balanced. */
3645 ASSERT(!xfs_btree_ptr_is_null(cur, &cptr));
3646
3647 if (!xfs_btree_ptr_is_null(cur, &lptr) &&
3648 lrecs + xfs_btree_get_numrecs(block) <=
3649 cur->bc_ops->get_maxrecs(cur, level)) {
3650 /*
3651 * Set "right" to be the starting block,
3652 * "left" to be the left neighbor.
3653 */
3654 rptr = cptr;
3655 right = block;
3656 rbp = bp;
3657 error = xfs_btree_read_buf_block(cur, &lptr, level,
3658 0, &left, &lbp);
3659 if (error)
3660 goto error0;
3661
3662 /*
3663 * If that won't work, see if we can join with the right neighbor block.
3664 */
3665 } else if (!xfs_btree_ptr_is_null(cur, &rptr) &&
3666 rrecs + xfs_btree_get_numrecs(block) <=
3667 cur->bc_ops->get_maxrecs(cur, level)) {
3668 /*
3669 * Set "left" to be the starting block,
3670 * "right" to be the right neighbor.
3671 */
3672 lptr = cptr;
3673 left = block;
3674 lbp = bp;
3675 error = xfs_btree_read_buf_block(cur, &rptr, level,
3676 0, &right, &rbp);
3677 if (error)
3678 goto error0;
3679
3680 /*
3681 * Otherwise, we can't fix the imbalance.
3682 * Just return. This is probably a logic error, but it's not fatal.
3683 */
3684 } else {
3685 error = xfs_btree_dec_cursor(cur, level, stat);
3686 if (error)
3687 goto error0;
3688 return 0;
3689 }
3690
3691 rrecs = xfs_btree_get_numrecs(right);
3692 lrecs = xfs_btree_get_numrecs(left);
3693
3694 /*
3695 * We're now going to join "left" and "right" by moving all the stuff
3696 * in "right" to "left" and deleting "right".
3697 */
3698 XFS_BTREE_STATS_ADD(cur, moves, rrecs);
3699 if (level > 0) {
3700 /* It's a non-leaf. Move keys and pointers. */
3701 union xfs_btree_key *lkp; /* left btree key */
3702 union xfs_btree_ptr *lpp; /* left address pointer */
3703 union xfs_btree_key *rkp; /* right btree key */
3704 union xfs_btree_ptr *rpp; /* right address pointer */
3705
3706 lkp = xfs_btree_key_addr(cur, lrecs + 1, left);
3707 lpp = xfs_btree_ptr_addr(cur, lrecs + 1, left);
3708 rkp = xfs_btree_key_addr(cur, 1, right);
3709 rpp = xfs_btree_ptr_addr(cur, 1, right);
3710#ifdef DEBUG
3711 for (i = 1; i < rrecs; i++) {
3712 error = xfs_btree_check_ptr(cur, rpp, i, level);
3713 if (error)
3714 goto error0;
3715 }
3716#endif
3717 xfs_btree_copy_keys(cur, lkp, rkp, rrecs);
3718 xfs_btree_copy_ptrs(cur, lpp, rpp, rrecs);
3719
3720 xfs_btree_log_keys(cur, lbp, lrecs + 1, lrecs + rrecs);
3721 xfs_btree_log_ptrs(cur, lbp, lrecs + 1, lrecs + rrecs);
3722 } else {
3723 /* It's a leaf. Move records. */
3724 union xfs_btree_rec *lrp; /* left record pointer */
3725 union xfs_btree_rec *rrp; /* right record pointer */
3726
3727 lrp = xfs_btree_rec_addr(cur, lrecs + 1, left);
3728 rrp = xfs_btree_rec_addr(cur, 1, right);
3729
3730 xfs_btree_copy_recs(cur, lrp, rrp, rrecs);
3731 xfs_btree_log_recs(cur, lbp, lrecs + 1, lrecs + rrecs);
3732 }
3733
3734 XFS_BTREE_STATS_INC(cur, join);
3735
3736 /*
9da096fd 3737 * Fix up the number of records and right block pointer in the
91cca5df
CH		 3738 * surviving block, and log it.
3739 */
3740 xfs_btree_set_numrecs(left, lrecs + rrecs);
3741 xfs_btree_get_sibling(cur, right, &cptr, XFS_BB_RIGHTSIB),
3742 xfs_btree_set_sibling(cur, left, &cptr, XFS_BB_RIGHTSIB);
3743 xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS | XFS_BB_RIGHTSIB);
3744
3745 /* If there is a right sibling, point it to the remaining block. */
3746 xfs_btree_get_sibling(cur, left, &cptr, XFS_BB_RIGHTSIB);
3747 if (!xfs_btree_ptr_is_null(cur, &cptr)) {
3748 error = xfs_btree_read_buf_block(cur, &cptr, level,
3749 0, &rrblock, &rrbp);
3750 if (error)
3751 goto error0;
3752 xfs_btree_set_sibling(cur, rrblock, &lptr, XFS_BB_LEFTSIB);
3753 xfs_btree_log_block(cur, rrbp, XFS_BB_LEFTSIB);
3754 }
3755
3756 /* Free the deleted block. */
3757 error = cur->bc_ops->free_block(cur, rbp);
3758 if (error)
3759 goto error0;
3760 XFS_BTREE_STATS_INC(cur, free);
3761
3762 /*
3763 * If we joined with the left neighbor, set the buffer in the
3764 * cursor to the left block, and fix up the index.
3765 */
3766 if (bp != lbp) {
3767 cur->bc_bufs[level] = lbp;
3768 cur->bc_ptrs[level] += lrecs;
3769 cur->bc_ra[level] = 0;
3770 }
3771 /*
3772 * If we joined with the right neighbor and there's a level above
3773 * us, increment the cursor at that level.
3774 */
3775 else if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) ||
3776 (level + 1 < cur->bc_nlevels)) {
3777 error = xfs_btree_increment(cur, level + 1, &i);
3778 if (error)
3779 goto error0;
3780 }
3781
3782 /*
3783 * Readjust the ptr at this level if it's not a leaf, since it's
3784 * still pointing at the deletion point, which makes the cursor
3785 * inconsistent. If this makes the ptr 0, the caller fixes it up.
3786 * We can't use decrement because it would change the next level up.
3787 */
3788 if (level > 0)
3789 cur->bc_ptrs[level]--;
3790
3791 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3792 /* Return value means the next level up has something to do. */
3793 *stat = 2;
3794 return 0;
3795
3796error0:
3797 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3798 if (tcur)
3799 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
3800 return error;
3801}
3802
3803/*
3804 * Delete the record pointed to by cur.
3805 * The cursor refers to the place where the record was (could be inserted)
3806 * when the operation returns.
3807 */
3808int /* error */
3809xfs_btree_delete(
3810 struct xfs_btree_cur *cur,
3811 int *stat) /* success/failure */
3812{
3813 int error; /* error return value */
3814 int level;
3815 int i;
3816
3817 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
3818
3819 /*
3820 * Go up the tree, starting at leaf level.
3821 *
3822 * If 2 is returned then a join was done; go to the next level.
3823 * Otherwise we are done.
3824 */
3825 for (level = 0, i = 2; i == 2; level++) {
3826 error = xfs_btree_delrec(cur, level, &i);
3827 if (error)
3828 goto error0;
3829 }
3830
3831 if (i == 0) {
3832 for (level = 1; level < cur->bc_nlevels; level++) {
3833 if (cur->bc_ptrs[level] == 0) {
3834 error = xfs_btree_decrement(cur, level, &i);
3835 if (error)
3836 goto error0;
3837 break;
3838 }
3839 }
3840 }
3841
3842 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3843 *stat = i;
3844 return 0;
3845error0:
3846 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3847 return error;
3848}
8cc938fe
CH		 3849
3850/*
3851 * Get the data from the pointed-to record.
3852 */
3853int /* error */
3854xfs_btree_get_rec(
3855 struct xfs_btree_cur *cur, /* btree cursor */
3856 union xfs_btree_rec **recp, /* output: btree record */
3857 int *stat) /* output: success/failure */
3858{
3859 struct xfs_btree_block *block; /* btree block */
3860 struct xfs_buf *bp; /* buffer pointer */
3861 int ptr; /* record number */
3862#ifdef DEBUG
3863 int error; /* error return value */
3864#endif
3865
3866 ptr = cur->bc_ptrs[0];
3867 block = xfs_btree_get_block(cur, 0, &bp);
3868
3869#ifdef DEBUG
3870 error = xfs_btree_check_block(cur, block, 0, bp);
3871 if (error)
3872 return error;
3873#endif
3874
3875 /*
3876 * Off the right end or left end, return failure.
3877 */
3878 if (ptr > xfs_btree_get_numrecs(block) || ptr <= 0) {
3879 *stat = 0;
3880 return 0;
3881 }
3882
3883 /*
3884 * Point to the record and extract its data.
3885 */
3886 *recp = xfs_btree_rec_addr(cur, ptr, block);
3887 *stat = 1;
3888 return 0;
3889}
21b5c978
DC		 3890
3891/*
3892 * Change the owner of a btree.
3893 *
3894 * The mechanism we use here is ordered buffer logging. Because we don't know
3895 * how many buffers were are going to need to modify, we don't really want to
3896 * have to make transaction reservations for the worst case of every buffer in a
3897 * full size btree as that may be more space that we can fit in the log....
3898 *
3899 * We do the btree walk in the most optimal manner possible - we have sibling
3900 * pointers so we can just walk all the blocks on each level from left to right
3901 * in a single pass, and then move to the next level and do the same. We can
3902 * also do readahead on the sibling pointers to get IO moving more quickly,
3903 * though for slow disks this is unlikely to make much difference to performance
3904 * as the amount of CPU work we have to do before moving to the next block is
3905 * relatively small.
3906 *
3907 * For each btree block that we load, modify the owner appropriately, set the
3908 * buffer as an ordered buffer and log it appropriately. We need to ensure that
3909 * we mark the region we change dirty so that if the buffer is relogged in
3910 * a subsequent transaction the changes we make here as an ordered buffer are
638f4416
DC		 3911 * correctly relogged in that transaction. If we are in recovery context, then
3912 * just queue the modified buffer as delayed write buffer so the transaction
3913 * recovery completion writes the changes to disk.
21b5c978
DC		 3914 */
3915static int
3916xfs_btree_block_change_owner(
3917 struct xfs_btree_cur *cur,
3918 int level,
638f4416
DC		 3919 __uint64_t new_owner,
3920 struct list_head *buffer_list)
21b5c978
DC		 3921{
3922 struct xfs_btree_block *block;
3923 struct xfs_buf *bp;
3924 union xfs_btree_ptr rptr;
3925
3926 /* do right sibling readahead */
3927 xfs_btree_readahead(cur, level, XFS_BTCUR_RIGHTRA);
3928
3929 /* modify the owner */
3930 block = xfs_btree_get_block(cur, level, &bp);
3931 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
3932 block->bb_u.l.bb_owner = cpu_to_be64(new_owner);
3933 else
3934 block->bb_u.s.bb_owner = cpu_to_be32(new_owner);
3935
3936 /*
638f4416
DC		 3937 * If the block is a root block hosted in an inode, we might not have a
3938 * buffer pointer here and we shouldn't attempt to log the change as the
3939 * information is already held in the inode and discarded when the root
3940 * block is formatted into the on-disk inode fork. We still change it,
3941 * though, so everything is consistent in memory.
21b5c978
DC		 3942 */
3943 if (bp) {
638f4416
DC		 3944 if (cur->bc_tp) {
3945 xfs_trans_ordered_buf(cur->bc_tp, bp);
3946 xfs_btree_log_block(cur, bp, XFS_BB_OWNER);
3947 } else {
3948 xfs_buf_delwri_queue(bp, buffer_list);
3949 }
21b5c978
DC		 3950 } else {
3951 ASSERT(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE);
3952 ASSERT(level == cur->bc_nlevels - 1);
3953 }
3954
3955 /* now read rh sibling block for next iteration */
3956 xfs_btree_get_sibling(cur, block, &rptr, XFS_BB_RIGHTSIB);
3957 if (xfs_btree_ptr_is_null(cur, &rptr))
3958 return ENOENT;
3959
3960 return xfs_btree_lookup_get_block(cur, level, &rptr, &block);
3961}
3962
3963int
3964xfs_btree_change_owner(
3965 struct xfs_btree_cur *cur,
638f4416
DC		 3966 __uint64_t new_owner,
3967 struct list_head *buffer_list)
21b5c978
DC		 3968{
3969 union xfs_btree_ptr lptr;
3970 int level;
3971 struct xfs_btree_block *block = NULL;
3972 int error = 0;
3973
3974 cur->bc_ops->init_ptr_from_cur(cur, &lptr);
3975
3976 /* for each level */
3977 for (level = cur->bc_nlevels - 1; level >= 0; level--) {
3978 /* grab the left hand block */
3979 error = xfs_btree_lookup_get_block(cur, level, &lptr, &block);
3980 if (error)
3981 return error;
3982
3983 /* readahead the left most block for the next level down */
3984 if (level > 0) {
3985 union xfs_btree_ptr *ptr;
3986
3987 ptr = xfs_btree_ptr_addr(cur, 1, block);
3988 xfs_btree_readahead_ptr(cur, ptr, 1);
3989
3990 /* save for the next iteration of the loop */
3991 lptr = *ptr;
3992 }
3993
3994 /* for each buffer in the level */
3995 do {
3996 error = xfs_btree_block_change_owner(cur, level,
638f4416
DC		 3997 new_owner,
3998 buffer_list);
21b5c978
DC		 3999 } while (!error);
4000
4001 if (error != ENOENT)
4002 return error;
4003 }
4004
4005 return 0;
4006}
ubuntu-bionic-kernel mirror