1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
4 * All Rights Reserved.
5 */
6 #include "xfs.h"
7 #include "xfs_fs.h"
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
12 #include "xfs_bit.h"
13 #include "xfs_mount.h"
14 #include "xfs_inode.h"
15 #include "xfs_trans.h"
16 #include "xfs_buf_item.h"
17 #include "xfs_btree.h"
18 #include "xfs_errortag.h"
19 #include "xfs_error.h"
20 #include "xfs_trace.h"
21 #include "xfs_alloc.h"
22 #include "xfs_log.h"
23 #include "xfs_btree_staging.h"
24 #include "xfs_ag.h"
25 #include "xfs_alloc_btree.h"
26 #include "xfs_ialloc_btree.h"
27 #include "xfs_bmap_btree.h"
28 #include "xfs_rmap_btree.h"
29 #include "xfs_refcount_btree.h"
30 #include "xfs_health.h"
31 #include "xfs_buf_mem.h"
32 #include "xfs_btree_mem.h"
33
34 /*
35 * Btree magic numbers.
36 */
37 uint32_t
xfs_btree_magic(struct xfs_mount * mp,const struct xfs_btree_ops * ops)38 xfs_btree_magic(
39 struct xfs_mount *mp,
40 const struct xfs_btree_ops *ops)
41 {
42 int idx = xfs_has_crc(mp) ? 1 : 0;
43 __be32 magic = ops->buf_ops->magic[idx];
44
45 /* Ensure we asked for crc for crc-only magics. */
46 ASSERT(magic != 0);
47 return be32_to_cpu(magic);
48 }
49
50 /*
51 * These sibling pointer checks are optimised for null sibling pointers. This
52 * happens a lot, and we don't need to byte swap at runtime if the sibling
53 * pointer is NULL.
54 *
55 * These are explicitly marked at inline because the cost of calling them as
56 * functions instead of inlining them is about 36 bytes extra code per call site
57 * on x86-64. Yes, gcc-11 fails to inline them, and explicit inlining of these
58 * two sibling check functions reduces the compiled code size by over 300
59 * bytes.
60 */
61 static inline xfs_failaddr_t
xfs_btree_check_fsblock_siblings(struct xfs_mount * mp,xfs_fsblock_t fsb,__be64 dsibling)62 xfs_btree_check_fsblock_siblings(
63 struct xfs_mount *mp,
64 xfs_fsblock_t fsb,
65 __be64 dsibling)
66 {
67 xfs_fsblock_t sibling;
68
69 if (dsibling == cpu_to_be64(NULLFSBLOCK))
70 return NULL;
71
72 sibling = be64_to_cpu(dsibling);
73 if (sibling == fsb)
74 return __this_address;
75 if (!xfs_verify_fsbno(mp, sibling))
76 return __this_address;
77 return NULL;
78 }
79
80 static inline xfs_failaddr_t
xfs_btree_check_memblock_siblings(struct xfs_buftarg * btp,xfbno_t bno,__be64 dsibling)81 xfs_btree_check_memblock_siblings(
82 struct xfs_buftarg *btp,
83 xfbno_t bno,
84 __be64 dsibling)
85 {
86 xfbno_t sibling;
87
88 if (dsibling == cpu_to_be64(NULLFSBLOCK))
89 return NULL;
90
91 sibling = be64_to_cpu(dsibling);
92 if (sibling == bno)
93 return __this_address;
94 if (!xmbuf_verify_daddr(btp, xfbno_to_daddr(sibling)))
95 return __this_address;
96 return NULL;
97 }
98
99 static inline xfs_failaddr_t
xfs_btree_check_agblock_siblings(struct xfs_perag * pag,xfs_agblock_t agbno,__be32 dsibling)100 xfs_btree_check_agblock_siblings(
101 struct xfs_perag *pag,
102 xfs_agblock_t agbno,
103 __be32 dsibling)
104 {
105 xfs_agblock_t sibling;
106
107 if (dsibling == cpu_to_be32(NULLAGBLOCK))
108 return NULL;
109
110 sibling = be32_to_cpu(dsibling);
111 if (sibling == agbno)
112 return __this_address;
113 if (!xfs_verify_agbno(pag, sibling))
114 return __this_address;
115 return NULL;
116 }
117
118 static xfs_failaddr_t
__xfs_btree_check_lblock_hdr(struct xfs_btree_cur * cur,struct xfs_btree_block * block,int level,struct xfs_buf * bp)119 __xfs_btree_check_lblock_hdr(
120 struct xfs_btree_cur *cur,
121 struct xfs_btree_block *block,
122 int level,
123 struct xfs_buf *bp)
124 {
125 struct xfs_mount *mp = cur->bc_mp;
126
127 if (xfs_has_crc(mp)) {
128 if (!uuid_equal(&block->bb_u.l.bb_uuid, &mp->m_sb.sb_meta_uuid))
129 return __this_address;
130 if (block->bb_u.l.bb_blkno !=
131 cpu_to_be64(bp ? xfs_buf_daddr(bp) : XFS_BUF_DADDR_NULL))
132 return __this_address;
133 if (block->bb_u.l.bb_pad != cpu_to_be32(0))
134 return __this_address;
135 }
136
137 if (be32_to_cpu(block->bb_magic) != xfs_btree_magic(mp, cur->bc_ops))
138 return __this_address;
139 if (be16_to_cpu(block->bb_level) != level)
140 return __this_address;
141 if (be16_to_cpu(block->bb_numrecs) >
142 cur->bc_ops->get_maxrecs(cur, level))
143 return __this_address;
144
145 return NULL;
146 }
147
148 /*
149 * Check a long btree block header. Return the address of the failing check,
150 * or NULL if everything is ok.
151 */
152 static xfs_failaddr_t
__xfs_btree_check_fsblock(struct xfs_btree_cur * cur,struct xfs_btree_block * block,int level,struct xfs_buf * bp)153 __xfs_btree_check_fsblock(
154 struct xfs_btree_cur *cur,
155 struct xfs_btree_block *block,
156 int level,
157 struct xfs_buf *bp)
158 {
159 struct xfs_mount *mp = cur->bc_mp;
160 xfs_failaddr_t fa;
161 xfs_fsblock_t fsb;
162
163 fa = __xfs_btree_check_lblock_hdr(cur, block, level, bp);
164 if (fa)
165 return fa;
166
167 /*
168 * For inode-rooted btrees, the root block sits in the inode fork. In
169 * that case bp is NULL, and the block must not have any siblings.
170 */
171 if (!bp) {
172 if (block->bb_u.l.bb_leftsib != cpu_to_be64(NULLFSBLOCK))
173 return __this_address;
174 if (block->bb_u.l.bb_rightsib != cpu_to_be64(NULLFSBLOCK))
175 return __this_address;
176 return NULL;
177 }
178
179 fsb = XFS_DADDR_TO_FSB(mp, xfs_buf_daddr(bp));
180 fa = xfs_btree_check_fsblock_siblings(mp, fsb,
181 block->bb_u.l.bb_leftsib);
182 if (!fa)
183 fa = xfs_btree_check_fsblock_siblings(mp, fsb,
184 block->bb_u.l.bb_rightsib);
185 return fa;
186 }
187
188 /*
189 * Check an in-memory btree block header. Return the address of the failing
190 * check, or NULL if everything is ok.
191 */
192 static xfs_failaddr_t
__xfs_btree_check_memblock(struct xfs_btree_cur * cur,struct xfs_btree_block * block,int level,struct xfs_buf * bp)193 __xfs_btree_check_memblock(
194 struct xfs_btree_cur *cur,
195 struct xfs_btree_block *block,
196 int level,
197 struct xfs_buf *bp)
198 {
199 struct xfs_buftarg *btp = cur->bc_mem.xfbtree->target;
200 xfs_failaddr_t fa;
201 xfbno_t bno;
202
203 fa = __xfs_btree_check_lblock_hdr(cur, block, level, bp);
204 if (fa)
205 return fa;
206
207 bno = xfs_daddr_to_xfbno(xfs_buf_daddr(bp));
208 fa = xfs_btree_check_memblock_siblings(btp, bno,
209 block->bb_u.l.bb_leftsib);
210 if (!fa)
211 fa = xfs_btree_check_memblock_siblings(btp, bno,
212 block->bb_u.l.bb_rightsib);
213 return fa;
214 }
215
216 /*
217 * Check a short btree block header. Return the address of the failing check,
218 * or NULL if everything is ok.
219 */
220 static xfs_failaddr_t
__xfs_btree_check_agblock(struct xfs_btree_cur * cur,struct xfs_btree_block * block,int level,struct xfs_buf * bp)221 __xfs_btree_check_agblock(
222 struct xfs_btree_cur *cur,
223 struct xfs_btree_block *block,
224 int level,
225 struct xfs_buf *bp)
226 {
227 struct xfs_mount *mp = cur->bc_mp;
228 struct xfs_perag *pag = cur->bc_ag.pag;
229 xfs_failaddr_t fa;
230 xfs_agblock_t agbno;
231
232 if (xfs_has_crc(mp)) {
233 if (!uuid_equal(&block->bb_u.s.bb_uuid, &mp->m_sb.sb_meta_uuid))
234 return __this_address;
235 if (block->bb_u.s.bb_blkno != cpu_to_be64(xfs_buf_daddr(bp)))
236 return __this_address;
237 }
238
239 if (be32_to_cpu(block->bb_magic) != xfs_btree_magic(mp, cur->bc_ops))
240 return __this_address;
241 if (be16_to_cpu(block->bb_level) != level)
242 return __this_address;
243 if (be16_to_cpu(block->bb_numrecs) >
244 cur->bc_ops->get_maxrecs(cur, level))
245 return __this_address;
246
247 agbno = xfs_daddr_to_agbno(mp, xfs_buf_daddr(bp));
248 fa = xfs_btree_check_agblock_siblings(pag, agbno,
249 block->bb_u.s.bb_leftsib);
250 if (!fa)
251 fa = xfs_btree_check_agblock_siblings(pag, agbno,
252 block->bb_u.s.bb_rightsib);
253 return fa;
254 }
255
256 /*
257 * Internal btree block check.
258 *
259 * Return NULL if the block is ok or the address of the failed check otherwise.
260 */
261 xfs_failaddr_t
__xfs_btree_check_block(struct xfs_btree_cur * cur,struct xfs_btree_block * block,int level,struct xfs_buf * bp)262 __xfs_btree_check_block(
263 struct xfs_btree_cur *cur,
264 struct xfs_btree_block *block,
265 int level,
266 struct xfs_buf *bp)
267 {
268 switch (cur->bc_ops->type) {
269 case XFS_BTREE_TYPE_MEM:
270 return __xfs_btree_check_memblock(cur, block, level, bp);
271 case XFS_BTREE_TYPE_AG:
272 return __xfs_btree_check_agblock(cur, block, level, bp);
273 case XFS_BTREE_TYPE_INODE:
274 return __xfs_btree_check_fsblock(cur, block, level, bp);
275 default:
276 ASSERT(0);
277 return __this_address;
278 }
279 }
280
xfs_btree_block_errtag(struct xfs_btree_cur * cur)281 static inline unsigned int xfs_btree_block_errtag(struct xfs_btree_cur *cur)
282 {
283 if (cur->bc_ops->ptr_len == XFS_BTREE_SHORT_PTR_LEN)
284 return XFS_ERRTAG_BTREE_CHECK_SBLOCK;
285 return XFS_ERRTAG_BTREE_CHECK_LBLOCK;
286 }
287
288 /*
289 * Debug routine: check that block header is ok.
290 */
291 int
xfs_btree_check_block(struct xfs_btree_cur * cur,struct xfs_btree_block * block,int level,struct xfs_buf * bp)292 xfs_btree_check_block(
293 struct xfs_btree_cur *cur, /* btree cursor */
294 struct xfs_btree_block *block, /* generic btree block pointer */
295 int level, /* level of the btree block */
296 struct xfs_buf *bp) /* buffer containing block, if any */
297 {
298 struct xfs_mount *mp = cur->bc_mp;
299 xfs_failaddr_t fa;
300
301 fa = __xfs_btree_check_block(cur, block, level, bp);
302 if (XFS_IS_CORRUPT(mp, fa != NULL) ||
303 XFS_TEST_ERROR(false, mp, xfs_btree_block_errtag(cur))) {
304 if (bp)
305 trace_xfs_btree_corrupt(bp, _RET_IP_);
306 xfs_btree_mark_sick(cur);
307 return -EFSCORRUPTED;
308 }
309 return 0;
310 }
311
312 int
__xfs_btree_check_ptr(struct xfs_btree_cur * cur,const union xfs_btree_ptr * ptr,int index,int level)313 __xfs_btree_check_ptr(
314 struct xfs_btree_cur *cur,
315 const union xfs_btree_ptr *ptr,
316 int index,
317 int level)
318 {
319 if (level <= 0)
320 return -EFSCORRUPTED;
321
322 switch (cur->bc_ops->type) {
323 case XFS_BTREE_TYPE_MEM:
324 if (!xfbtree_verify_bno(cur->bc_mem.xfbtree,
325 be64_to_cpu((&ptr->l)[index])))
326 return -EFSCORRUPTED;
327 break;
328 case XFS_BTREE_TYPE_INODE:
329 if (!xfs_verify_fsbno(cur->bc_mp,
330 be64_to_cpu((&ptr->l)[index])))
331 return -EFSCORRUPTED;
332 break;
333 case XFS_BTREE_TYPE_AG:
334 if (!xfs_verify_agbno(cur->bc_ag.pag,
335 be32_to_cpu((&ptr->s)[index])))
336 return -EFSCORRUPTED;
337 break;
338 }
339
340 return 0;
341 }
342
343 /*
344 * Check that a given (indexed) btree pointer at a certain level of a
345 * btree is valid and doesn't point past where it should.
346 */
347 static int
xfs_btree_check_ptr(struct xfs_btree_cur * cur,const union xfs_btree_ptr * ptr,int index,int level)348 xfs_btree_check_ptr(
349 struct xfs_btree_cur *cur,
350 const union xfs_btree_ptr *ptr,
351 int index,
352 int level)
353 {
354 int error;
355
356 error = __xfs_btree_check_ptr(cur, ptr, index, level);
357 if (error) {
358 switch (cur->bc_ops->type) {
359 case XFS_BTREE_TYPE_MEM:
360 xfs_err(cur->bc_mp,
361 "In-memory: Corrupt %sbt flags 0x%x pointer at level %d index %d fa %pS.",
362 cur->bc_ops->name, cur->bc_flags, level, index,
363 __this_address);
364 break;
365 case XFS_BTREE_TYPE_INODE:
366 xfs_err(cur->bc_mp,
367 "Inode %llu fork %d: Corrupt %sbt pointer at level %d index %d.",
368 cur->bc_ino.ip->i_ino,
369 cur->bc_ino.whichfork, cur->bc_ops->name,
370 level, index);
371 break;
372 case XFS_BTREE_TYPE_AG:
373 xfs_err(cur->bc_mp,
374 "AG %u: Corrupt %sbt pointer at level %d index %d.",
375 cur->bc_ag.pag->pag_agno, cur->bc_ops->name,
376 level, index);
377 break;
378 }
379 xfs_btree_mark_sick(cur);
380 }
381
382 return error;
383 }
384
385 #ifdef DEBUG
386 # define xfs_btree_debug_check_ptr xfs_btree_check_ptr
387 #else
388 # define xfs_btree_debug_check_ptr(...) (0)
389 #endif
390
391 /*
392 * Calculate CRC on the whole btree block and stuff it into the
393 * long-form btree header.
394 *
395 * Prior to calculting the CRC, pull the LSN out of the buffer log item and put
396 * it into the buffer so recovery knows what the last modification was that made
397 * it to disk.
398 */
399 void
xfs_btree_fsblock_calc_crc(struct xfs_buf * bp)400 xfs_btree_fsblock_calc_crc(
401 struct xfs_buf *bp)
402 {
403 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
404 struct xfs_buf_log_item *bip = bp->b_log_item;
405
406 if (!xfs_has_crc(bp->b_mount))
407 return;
408 if (bip)
409 block->bb_u.l.bb_lsn = cpu_to_be64(bip->bli_item.li_lsn);
410 xfs_buf_update_cksum(bp, XFS_BTREE_LBLOCK_CRC_OFF);
411 }
412
413 bool
xfs_btree_fsblock_verify_crc(struct xfs_buf * bp)414 xfs_btree_fsblock_verify_crc(
415 struct xfs_buf *bp)
416 {
417 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
418 struct xfs_mount *mp = bp->b_mount;
419
420 if (xfs_has_crc(mp)) {
421 if (!xfs_log_check_lsn(mp, be64_to_cpu(block->bb_u.l.bb_lsn)))
422 return false;
423 return xfs_buf_verify_cksum(bp, XFS_BTREE_LBLOCK_CRC_OFF);
424 }
425
426 return true;
427 }
428
429 /*
430 * Calculate CRC on the whole btree block and stuff it into the
431 * short-form btree header.
432 *
433 * Prior to calculting the CRC, pull the LSN out of the buffer log item and put
434 * it into the buffer so recovery knows what the last modification was that made
435 * it to disk.
436 */
437 void
xfs_btree_agblock_calc_crc(struct xfs_buf * bp)438 xfs_btree_agblock_calc_crc(
439 struct xfs_buf *bp)
440 {
441 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
442 struct xfs_buf_log_item *bip = bp->b_log_item;
443
444 if (!xfs_has_crc(bp->b_mount))
445 return;
446 if (bip)
447 block->bb_u.s.bb_lsn = cpu_to_be64(bip->bli_item.li_lsn);
448 xfs_buf_update_cksum(bp, XFS_BTREE_SBLOCK_CRC_OFF);
449 }
450
451 bool
xfs_btree_agblock_verify_crc(struct xfs_buf * bp)452 xfs_btree_agblock_verify_crc(
453 struct xfs_buf *bp)
454 {
455 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
456 struct xfs_mount *mp = bp->b_mount;
457
458 if (xfs_has_crc(mp)) {
459 if (!xfs_log_check_lsn(mp, be64_to_cpu(block->bb_u.s.bb_lsn)))
460 return false;
461 return xfs_buf_verify_cksum(bp, XFS_BTREE_SBLOCK_CRC_OFF);
462 }
463
464 return true;
465 }
466
467 static int
xfs_btree_free_block(struct xfs_btree_cur * cur,struct xfs_buf * bp)468 xfs_btree_free_block(
469 struct xfs_btree_cur *cur,
470 struct xfs_buf *bp)
471 {
472 int error;
473
474 trace_xfs_btree_free_block(cur, bp);
475
476 /*
477 * Don't allow block freeing for a staging cursor, because staging
478 * cursors do not support regular btree modifications.
479 */
480 if (unlikely(cur->bc_flags & XFS_BTREE_STAGING)) {
481 ASSERT(0);
482 return -EFSCORRUPTED;
483 }
484
485 error = cur->bc_ops->free_block(cur, bp);
486 if (!error) {
487 xfs_trans_binval(cur->bc_tp, bp);
488 XFS_BTREE_STATS_INC(cur, free);
489 }
490 return error;
491 }
492
493 /*
494 * Delete the btree cursor.
495 */
496 void
xfs_btree_del_cursor(struct xfs_btree_cur * cur,int error)497 xfs_btree_del_cursor(
498 struct xfs_btree_cur *cur, /* btree cursor */
499 int error) /* del because of error */
500 {
501 int i; /* btree level */
502
503 /*
504 * Clear the buffer pointers and release the buffers. If we're doing
505 * this because of an error, inspect all of the entries in the bc_bufs
506 * array for buffers to be unlocked. This is because some of the btree
507 * code works from level n down to 0, and if we get an error along the
508 * way we won't have initialized all the entries down to 0.
509 */
510 for (i = 0; i < cur->bc_nlevels; i++) {
511 if (cur->bc_levels[i].bp)
512 xfs_trans_brelse(cur->bc_tp, cur->bc_levels[i].bp);
513 else if (!error)
514 break;
515 }
516
517 /*
518 * If we are doing a BMBT update, the number of unaccounted blocks
519 * allocated during this cursor life time should be zero. If it's not
520 * zero, then we should be shut down or on our way to shutdown due to
521 * cancelling a dirty transaction on error.
522 */
523 ASSERT(!xfs_btree_is_bmap(cur->bc_ops) || cur->bc_bmap.allocated == 0 ||
524 xfs_is_shutdown(cur->bc_mp) || error != 0);
525
526 switch (cur->bc_ops->type) {
527 case XFS_BTREE_TYPE_AG:
528 if (cur->bc_ag.pag)
529 xfs_perag_put(cur->bc_ag.pag);
530 break;
531 case XFS_BTREE_TYPE_INODE:
532 /* nothing to do */
533 break;
534 case XFS_BTREE_TYPE_MEM:
535 if (cur->bc_mem.pag)
536 xfs_perag_put(cur->bc_mem.pag);
537 break;
538 }
539
540 kmem_cache_free(cur->bc_cache, cur);
541 }
542
543 /* Return the buffer target for this btree's buffer. */
544 static inline struct xfs_buftarg *
xfs_btree_buftarg(struct xfs_btree_cur * cur)545 xfs_btree_buftarg(
546 struct xfs_btree_cur *cur)
547 {
548 if (cur->bc_ops->type == XFS_BTREE_TYPE_MEM)
549 return cur->bc_mem.xfbtree->target;
550 return cur->bc_mp->m_ddev_targp;
551 }
552
553 /* Return the block size (in units of 512b sectors) for this btree. */
554 static inline unsigned int
xfs_btree_bbsize(struct xfs_btree_cur * cur)555 xfs_btree_bbsize(
556 struct xfs_btree_cur *cur)
557 {
558 if (cur->bc_ops->type == XFS_BTREE_TYPE_MEM)
559 return XFBNO_BBSIZE;
560 return cur->bc_mp->m_bsize;
561 }
562
563 /*
564 * Duplicate the btree cursor.
565 * Allocate a new one, copy the record, re-get the buffers.
566 */
567 int /* error */
xfs_btree_dup_cursor(struct xfs_btree_cur * cur,struct xfs_btree_cur ** ncur)568 xfs_btree_dup_cursor(
569 struct xfs_btree_cur *cur, /* input cursor */
570 struct xfs_btree_cur **ncur) /* output cursor */
571 {
572 struct xfs_mount *mp = cur->bc_mp;
573 struct xfs_trans *tp = cur->bc_tp;
574 struct xfs_buf *bp;
575 struct xfs_btree_cur *new;
576 int error;
577 int i;
578
579 /*
580 * Don't allow staging cursors to be duplicated because they're supposed
581 * to be kept private to a single thread.
582 */
583 if (unlikely(cur->bc_flags & XFS_BTREE_STAGING)) {
584 ASSERT(0);
585 return -EFSCORRUPTED;
586 }
587
588 /*
589 * Allocate a new cursor like the old one.
590 */
591 new = cur->bc_ops->dup_cursor(cur);
592
593 /*
594 * Copy the record currently in the cursor.
595 */
596 new->bc_rec = cur->bc_rec;
597
598 /*
599 * For each level current, re-get the buffer and copy the ptr value.
600 */
601 for (i = 0; i < new->bc_nlevels; i++) {
602 new->bc_levels[i].ptr = cur->bc_levels[i].ptr;
603 new->bc_levels[i].ra = cur->bc_levels[i].ra;
604 bp = cur->bc_levels[i].bp;
605 if (bp) {
606 error = xfs_trans_read_buf(mp, tp,
607 xfs_btree_buftarg(cur),
608 xfs_buf_daddr(bp),
609 xfs_btree_bbsize(cur), 0, &bp,
610 cur->bc_ops->buf_ops);
611 if (xfs_metadata_is_sick(error))
612 xfs_btree_mark_sick(new);
613 if (error) {
614 xfs_btree_del_cursor(new, error);
615 *ncur = NULL;
616 return error;
617 }
618 }
619 new->bc_levels[i].bp = bp;
620 }
621 *ncur = new;
622 return 0;
623 }
624
625 /*
626 * XFS btree block layout and addressing:
627 *
628 * There are two types of blocks in the btree: leaf and non-leaf blocks.
629 *
630 * The leaf record start with a header then followed by records containing
631 * the values. A non-leaf block also starts with the same header, and
632 * then first contains lookup keys followed by an equal number of pointers
633 * to the btree blocks at the previous level.
634 *
635 * +--------+-------+-------+-------+-------+-------+-------+
636 * Leaf: | header | rec 1 | rec 2 | rec 3 | rec 4 | rec 5 | rec N |
637 * +--------+-------+-------+-------+-------+-------+-------+
638 *
639 * +--------+-------+-------+-------+-------+-------+-------+
640 * Non-Leaf: | header | key 1 | key 2 | key N | ptr 1 | ptr 2 | ptr N |
641 * +--------+-------+-------+-------+-------+-------+-------+
642 *
643 * The header is called struct xfs_btree_block for reasons better left unknown
644 * and comes in different versions for short (32bit) and long (64bit) block
645 * pointers. The record and key structures are defined by the btree instances
646 * and opaque to the btree core. The block pointers are simple disk endian
647 * integers, available in a short (32bit) and long (64bit) variant.
648 *
649 * The helpers below calculate the offset of a given record, key or pointer
650 * into a btree block (xfs_btree_*_offset) or return a pointer to the given
651 * record, key or pointer (xfs_btree_*_addr). Note that all addressing
652 * inside the btree block is done using indices starting at one, not zero!
653 *
654 * If XFS_BTGEO_OVERLAPPING is set, then this btree supports keys containing
655 * overlapping intervals. In such a tree, records are still sorted lowest to
656 * highest and indexed by the smallest key value that refers to the record.
657 * However, nodes are different: each pointer has two associated keys -- one
658 * indexing the lowest key available in the block(s) below (the same behavior
659 * as the key in a regular btree) and another indexing the highest key
660 * available in the block(s) below. Because records are /not/ sorted by the
661 * highest key, all leaf block updates require us to compute the highest key
662 * that matches any record in the leaf and to recursively update the high keys
663 * in the nodes going further up in the tree, if necessary. Nodes look like
664 * this:
665 *
666 * +--------+-----+-----+-----+-----+-----+-------+-------+-----+
667 * Non-Leaf: | header | lo1 | hi1 | lo2 | hi2 | ... | ptr 1 | ptr 2 | ... |
668 * +--------+-----+-----+-----+-----+-----+-------+-------+-----+
669 *
670 * To perform an interval query on an overlapped tree, perform the usual
671 * depth-first search and use the low and high keys to decide if we can skip
672 * that particular node. If a leaf node is reached, return the records that
673 * intersect the interval. Note that an interval query may return numerous
674 * entries. For a non-overlapped tree, simply search for the record associated
675 * with the lowest key and iterate forward until a non-matching record is
676 * found. Section 14.3 ("Interval Trees") of _Introduction to Algorithms_ by
677 * Cormen, Leiserson, Rivest, and Stein (2nd or 3rd ed. only) discuss this in
678 * more detail.
679 *
680 * Why do we care about overlapping intervals? Let's say you have a bunch of
681 * reverse mapping records on a reflink filesystem:
682 *
683 * 1: +- file A startblock B offset C length D -----------+
684 * 2: +- file E startblock F offset G length H --------------+
685 * 3: +- file I startblock F offset J length K --+
686 * 4: +- file L... --+
687 *
688 * Now say we want to map block (B+D) into file A at offset (C+D). Ideally,
689 * we'd simply increment the length of record 1. But how do we find the record
690 * that ends at (B+D-1) (i.e. record 1)? A LE lookup of (B+D-1) would return
691 * record 3 because the keys are ordered first by startblock. An interval
692 * query would return records 1 and 2 because they both overlap (B+D-1), and
693 * from that we can pick out record 1 as the appropriate left neighbor.
694 *
695 * In the non-overlapped case you can do a LE lookup and decrement the cursor
696 * because a record's interval must end before the next record.
697 */
698
699 /*
700 * Return size of the btree block header for this btree instance.
701 */
xfs_btree_block_len(struct xfs_btree_cur * cur)702 static inline size_t xfs_btree_block_len(struct xfs_btree_cur *cur)
703 {
704 if (cur->bc_ops->ptr_len == XFS_BTREE_LONG_PTR_LEN) {
705 if (xfs_has_crc(cur->bc_mp))
706 return XFS_BTREE_LBLOCK_CRC_LEN;
707 return XFS_BTREE_LBLOCK_LEN;
708 }
709 if (xfs_has_crc(cur->bc_mp))
710 return XFS_BTREE_SBLOCK_CRC_LEN;
711 return XFS_BTREE_SBLOCK_LEN;
712 }
713
714 /*
715 * Calculate offset of the n-th record in a btree block.
716 */
717 STATIC size_t
xfs_btree_rec_offset(struct xfs_btree_cur * cur,int n)718 xfs_btree_rec_offset(
719 struct xfs_btree_cur *cur,
720 int n)
721 {
722 return xfs_btree_block_len(cur) +
723 (n - 1) * cur->bc_ops->rec_len;
724 }
725
726 /*
727 * Calculate offset of the n-th key in a btree block.
728 */
729 STATIC size_t
xfs_btree_key_offset(struct xfs_btree_cur * cur,int n)730 xfs_btree_key_offset(
731 struct xfs_btree_cur *cur,
732 int n)
733 {
734 return xfs_btree_block_len(cur) +
735 (n - 1) * cur->bc_ops->key_len;
736 }
737
738 /*
739 * Calculate offset of the n-th high key in a btree block.
740 */
741 STATIC size_t
xfs_btree_high_key_offset(struct xfs_btree_cur * cur,int n)742 xfs_btree_high_key_offset(
743 struct xfs_btree_cur *cur,
744 int n)
745 {
746 return xfs_btree_block_len(cur) +
747 (n - 1) * cur->bc_ops->key_len + (cur->bc_ops->key_len / 2);
748 }
749
750 /*
751 * Calculate offset of the n-th block pointer in a btree block.
752 */
753 STATIC size_t
xfs_btree_ptr_offset(struct xfs_btree_cur * cur,int n,int level)754 xfs_btree_ptr_offset(
755 struct xfs_btree_cur *cur,
756 int n,
757 int level)
758 {
759 return xfs_btree_block_len(cur) +
760 cur->bc_ops->get_maxrecs(cur, level) * cur->bc_ops->key_len +
761 (n - 1) * cur->bc_ops->ptr_len;
762 }
763
764 /*
765 * Return a pointer to the n-th record in the btree block.
766 */
767 union xfs_btree_rec *
xfs_btree_rec_addr(struct xfs_btree_cur * cur,int n,struct xfs_btree_block * block)768 xfs_btree_rec_addr(
769 struct xfs_btree_cur *cur,
770 int n,
771 struct xfs_btree_block *block)
772 {
773 return (union xfs_btree_rec *)
774 ((char *)block + xfs_btree_rec_offset(cur, n));
775 }
776
777 /*
778 * Return a pointer to the n-th key in the btree block.
779 */
780 union xfs_btree_key *
xfs_btree_key_addr(struct xfs_btree_cur * cur,int n,struct xfs_btree_block * block)781 xfs_btree_key_addr(
782 struct xfs_btree_cur *cur,
783 int n,
784 struct xfs_btree_block *block)
785 {
786 return (union xfs_btree_key *)
787 ((char *)block + xfs_btree_key_offset(cur, n));
788 }
789
790 /*
791 * Return a pointer to the n-th high key in the btree block.
792 */
793 union xfs_btree_key *
xfs_btree_high_key_addr(struct xfs_btree_cur * cur,int n,struct xfs_btree_block * block)794 xfs_btree_high_key_addr(
795 struct xfs_btree_cur *cur,
796 int n,
797 struct xfs_btree_block *block)
798 {
799 return (union xfs_btree_key *)
800 ((char *)block + xfs_btree_high_key_offset(cur, n));
801 }
802
803 /*
804 * Return a pointer to the n-th block pointer in the btree block.
805 */
806 union xfs_btree_ptr *
xfs_btree_ptr_addr(struct xfs_btree_cur * cur,int n,struct xfs_btree_block * block)807 xfs_btree_ptr_addr(
808 struct xfs_btree_cur *cur,
809 int n,
810 struct xfs_btree_block *block)
811 {
812 int level = xfs_btree_get_level(block);
813
814 ASSERT(block->bb_level != 0);
815
816 return (union xfs_btree_ptr *)
817 ((char *)block + xfs_btree_ptr_offset(cur, n, level));
818 }
819
820 struct xfs_ifork *
xfs_btree_ifork_ptr(struct xfs_btree_cur * cur)821 xfs_btree_ifork_ptr(
822 struct xfs_btree_cur *cur)
823 {
824 ASSERT(cur->bc_ops->type == XFS_BTREE_TYPE_INODE);
825
826 if (cur->bc_flags & XFS_BTREE_STAGING)
827 return cur->bc_ino.ifake->if_fork;
828 return xfs_ifork_ptr(cur->bc_ino.ip, cur->bc_ino.whichfork);
829 }
830
831 /*
832 * Get the root block which is stored in the inode.
833 *
834 * For now this btree implementation assumes the btree root is always
835 * stored in the if_broot field of an inode fork.
836 */
837 STATIC struct xfs_btree_block *
xfs_btree_get_iroot(struct xfs_btree_cur * cur)838 xfs_btree_get_iroot(
839 struct xfs_btree_cur *cur)
840 {
841 struct xfs_ifork *ifp = xfs_btree_ifork_ptr(cur);
842
843 return (struct xfs_btree_block *)ifp->if_broot;
844 }
845
846 /*
847 * Retrieve the block pointer from the cursor at the given level.
848 * This may be an inode btree root or from a buffer.
849 */
850 struct xfs_btree_block * /* generic btree block pointer */
xfs_btree_get_block(struct xfs_btree_cur * cur,int level,struct xfs_buf ** bpp)851 xfs_btree_get_block(
852 struct xfs_btree_cur *cur, /* btree cursor */
853 int level, /* level in btree */
854 struct xfs_buf **bpp) /* buffer containing the block */
855 {
856 if (xfs_btree_at_iroot(cur, level)) {
857 *bpp = NULL;
858 return xfs_btree_get_iroot(cur);
859 }
860
861 *bpp = cur->bc_levels[level].bp;
862 return XFS_BUF_TO_BLOCK(*bpp);
863 }
864
865 /*
866 * Change the cursor to point to the first record at the given level.
867 * Other levels are unaffected.
868 */
869 STATIC int /* success=1, failure=0 */
xfs_btree_firstrec(struct xfs_btree_cur * cur,int level)870 xfs_btree_firstrec(
871 struct xfs_btree_cur *cur, /* btree cursor */
872 int level) /* level to change */
873 {
874 struct xfs_btree_block *block; /* generic btree block pointer */
875 struct xfs_buf *bp; /* buffer containing block */
876
877 /*
878 * Get the block pointer for this level.
879 */
880 block = xfs_btree_get_block(cur, level, &bp);
881 if (xfs_btree_check_block(cur, block, level, bp))
882 return 0;
883 /*
884 * It's empty, there is no such record.
885 */
886 if (!block->bb_numrecs)
887 return 0;
888 /*
889 * Set the ptr value to 1, that's the first record/key.
890 */
891 cur->bc_levels[level].ptr = 1;
892 return 1;
893 }
894
895 /*
896 * Change the cursor to point to the last record in the current block
897 * at the given level. Other levels are unaffected.
898 */
899 STATIC int /* success=1, failure=0 */
xfs_btree_lastrec(struct xfs_btree_cur * cur,int level)900 xfs_btree_lastrec(
901 struct xfs_btree_cur *cur, /* btree cursor */
902 int level) /* level to change */
903 {
904 struct xfs_btree_block *block; /* generic btree block pointer */
905 struct xfs_buf *bp; /* buffer containing block */
906
907 /*
908 * Get the block pointer for this level.
909 */
910 block = xfs_btree_get_block(cur, level, &bp);
911 if (xfs_btree_check_block(cur, block, level, bp))
912 return 0;
913 /*
914 * It's empty, there is no such record.
915 */
916 if (!block->bb_numrecs)
917 return 0;
918 /*
919 * Set the ptr value to numrecs, that's the last record/key.
920 */
921 cur->bc_levels[level].ptr = be16_to_cpu(block->bb_numrecs);
922 return 1;
923 }
924
925 /*
926 * Compute first and last byte offsets for the fields given.
927 * Interprets the offsets table, which contains struct field offsets.
928 */
929 void
xfs_btree_offsets(uint32_t fields,const short * offsets,int nbits,int * first,int * last)930 xfs_btree_offsets(
931 uint32_t fields, /* bitmask of fields */
932 const short *offsets, /* table of field offsets */
933 int nbits, /* number of bits to inspect */
934 int *first, /* output: first byte offset */
935 int *last) /* output: last byte offset */
936 {
937 int i; /* current bit number */
938 uint32_t imask; /* mask for current bit number */
939
940 ASSERT(fields != 0);
941 /*
942 * Find the lowest bit, so the first byte offset.
943 */
944 for (i = 0, imask = 1u; ; i++, imask <<= 1) {
945 if (imask & fields) {
946 *first = offsets[i];
947 break;
948 }
949 }
950 /*
951 * Find the highest bit, so the last byte offset.
952 */
953 for (i = nbits - 1, imask = 1u << i; ; i--, imask >>= 1) {
954 if (imask & fields) {
955 *last = offsets[i + 1] - 1;
956 break;
957 }
958 }
959 }
960
961 STATIC int
xfs_btree_readahead_fsblock(struct xfs_btree_cur * cur,int lr,struct xfs_btree_block * block)962 xfs_btree_readahead_fsblock(
963 struct xfs_btree_cur *cur,
964 int lr,
965 struct xfs_btree_block *block)
966 {
967 struct xfs_mount *mp = cur->bc_mp;
968 xfs_fsblock_t left = be64_to_cpu(block->bb_u.l.bb_leftsib);
969 xfs_fsblock_t right = be64_to_cpu(block->bb_u.l.bb_rightsib);
970 int rval = 0;
971
972 if ((lr & XFS_BTCUR_LEFTRA) && left != NULLFSBLOCK) {
973 xfs_buf_readahead(mp->m_ddev_targp, XFS_FSB_TO_DADDR(mp, left),
974 mp->m_bsize, cur->bc_ops->buf_ops);
975 rval++;
976 }
977
978 if ((lr & XFS_BTCUR_RIGHTRA) && right != NULLFSBLOCK) {
979 xfs_buf_readahead(mp->m_ddev_targp, XFS_FSB_TO_DADDR(mp, right),
980 mp->m_bsize, cur->bc_ops->buf_ops);
981 rval++;
982 }
983
984 return rval;
985 }
986
987 STATIC int
xfs_btree_readahead_memblock(struct xfs_btree_cur * cur,int lr,struct xfs_btree_block * block)988 xfs_btree_readahead_memblock(
989 struct xfs_btree_cur *cur,
990 int lr,
991 struct xfs_btree_block *block)
992 {
993 struct xfs_buftarg *btp = cur->bc_mem.xfbtree->target;
994 xfbno_t left = be64_to_cpu(block->bb_u.l.bb_leftsib);
995 xfbno_t right = be64_to_cpu(block->bb_u.l.bb_rightsib);
996 int rval = 0;
997
998 if ((lr & XFS_BTCUR_LEFTRA) && left != NULLFSBLOCK) {
999 xfs_buf_readahead(btp, xfbno_to_daddr(left), XFBNO_BBSIZE,
1000 cur->bc_ops->buf_ops);
1001 rval++;
1002 }
1003
1004 if ((lr & XFS_BTCUR_RIGHTRA) && right != NULLFSBLOCK) {
1005 xfs_buf_readahead(btp, xfbno_to_daddr(right), XFBNO_BBSIZE,
1006 cur->bc_ops->buf_ops);
1007 rval++;
1008 }
1009
1010 return rval;
1011 }
1012
1013 STATIC int
xfs_btree_readahead_agblock(struct xfs_btree_cur * cur,int lr,struct xfs_btree_block * block)1014 xfs_btree_readahead_agblock(
1015 struct xfs_btree_cur *cur,
1016 int lr,
1017 struct xfs_btree_block *block)
1018 {
1019 struct xfs_mount *mp = cur->bc_mp;
1020 xfs_agnumber_t agno = cur->bc_ag.pag->pag_agno;
1021 xfs_agblock_t left = be32_to_cpu(block->bb_u.s.bb_leftsib);
1022 xfs_agblock_t right = be32_to_cpu(block->bb_u.s.bb_rightsib);
1023 int rval = 0;
1024
1025 if ((lr & XFS_BTCUR_LEFTRA) && left != NULLAGBLOCK) {
1026 xfs_buf_readahead(mp->m_ddev_targp,
1027 XFS_AGB_TO_DADDR(mp, agno, left),
1028 mp->m_bsize, cur->bc_ops->buf_ops);
1029 rval++;
1030 }
1031
1032 if ((lr & XFS_BTCUR_RIGHTRA) && right != NULLAGBLOCK) {
1033 xfs_buf_readahead(mp->m_ddev_targp,
1034 XFS_AGB_TO_DADDR(mp, agno, right),
1035 mp->m_bsize, cur->bc_ops->buf_ops);
1036 rval++;
1037 }
1038
1039 return rval;
1040 }
1041
1042 /*
1043 * Read-ahead btree blocks, at the given level.
1044 * Bits in lr are set from XFS_BTCUR_{LEFT,RIGHT}RA.
1045 */
1046 STATIC int
xfs_btree_readahead(struct xfs_btree_cur * cur,int lev,int lr)1047 xfs_btree_readahead(
1048 struct xfs_btree_cur *cur, /* btree cursor */
1049 int lev, /* level in btree */
1050 int lr) /* left/right bits */
1051 {
1052 struct xfs_btree_block *block;
1053
1054 /*
1055 * No readahead needed if we are at the root level and the
1056 * btree root is stored in the inode.
1057 */
1058 if (xfs_btree_at_iroot(cur, lev))
1059 return 0;
1060
1061 if ((cur->bc_levels[lev].ra | lr) == cur->bc_levels[lev].ra)
1062 return 0;
1063
1064 cur->bc_levels[lev].ra |= lr;
1065 block = XFS_BUF_TO_BLOCK(cur->bc_levels[lev].bp);
1066
1067 switch (cur->bc_ops->type) {
1068 case XFS_BTREE_TYPE_AG:
1069 return xfs_btree_readahead_agblock(cur, lr, block);
1070 case XFS_BTREE_TYPE_INODE:
1071 return xfs_btree_readahead_fsblock(cur, lr, block);
1072 case XFS_BTREE_TYPE_MEM:
1073 return xfs_btree_readahead_memblock(cur, lr, block);
1074 default:
1075 ASSERT(0);
1076 return 0;
1077 }
1078 }
1079
1080 STATIC int
xfs_btree_ptr_to_daddr(struct xfs_btree_cur * cur,const union xfs_btree_ptr * ptr,xfs_daddr_t * daddr)1081 xfs_btree_ptr_to_daddr(
1082 struct xfs_btree_cur *cur,
1083 const union xfs_btree_ptr *ptr,
1084 xfs_daddr_t *daddr)
1085 {
1086 int error;
1087
1088 error = xfs_btree_check_ptr(cur, ptr, 0, 1);
1089 if (error)
1090 return error;
1091
1092 switch (cur->bc_ops->type) {
1093 case XFS_BTREE_TYPE_AG:
1094 *daddr = XFS_AGB_TO_DADDR(cur->bc_mp, cur->bc_ag.pag->pag_agno,
1095 be32_to_cpu(ptr->s));
1096 break;
1097 case XFS_BTREE_TYPE_INODE:
1098 *daddr = XFS_FSB_TO_DADDR(cur->bc_mp, be64_to_cpu(ptr->l));
1099 break;
1100 case XFS_BTREE_TYPE_MEM:
1101 *daddr = xfbno_to_daddr(be64_to_cpu(ptr->l));
1102 break;
1103 }
1104 return 0;
1105 }
1106
1107 /*
1108 * Readahead @count btree blocks at the given @ptr location.
1109 *
1110 * We don't need to care about long or short form btrees here as we have a
1111 * method of converting the ptr directly to a daddr available to us.
1112 */
1113 STATIC void
xfs_btree_readahead_ptr(struct xfs_btree_cur * cur,union xfs_btree_ptr * ptr,xfs_extlen_t count)1114 xfs_btree_readahead_ptr(
1115 struct xfs_btree_cur *cur,
1116 union xfs_btree_ptr *ptr,
1117 xfs_extlen_t count)
1118 {
1119 xfs_daddr_t daddr;
1120
1121 if (xfs_btree_ptr_to_daddr(cur, ptr, &daddr))
1122 return;
1123 xfs_buf_readahead(xfs_btree_buftarg(cur), daddr,
1124 xfs_btree_bbsize(cur) * count,
1125 cur->bc_ops->buf_ops);
1126 }
1127
1128 /*
1129 * Set the buffer for level "lev" in the cursor to bp, releasing
1130 * any previous buffer.
1131 */
1132 STATIC void
xfs_btree_setbuf(struct xfs_btree_cur * cur,int lev,struct xfs_buf * bp)1133 xfs_btree_setbuf(
1134 struct xfs_btree_cur *cur, /* btree cursor */
1135 int lev, /* level in btree */
1136 struct xfs_buf *bp) /* new buffer to set */
1137 {
1138 struct xfs_btree_block *b; /* btree block */
1139
1140 if (cur->bc_levels[lev].bp)
1141 xfs_trans_brelse(cur->bc_tp, cur->bc_levels[lev].bp);
1142 cur->bc_levels[lev].bp = bp;
1143 cur->bc_levels[lev].ra = 0;
1144
1145 b = XFS_BUF_TO_BLOCK(bp);
1146 if (cur->bc_ops->ptr_len == XFS_BTREE_LONG_PTR_LEN) {
1147 if (b->bb_u.l.bb_leftsib == cpu_to_be64(NULLFSBLOCK))
1148 cur->bc_levels[lev].ra |= XFS_BTCUR_LEFTRA;
1149 if (b->bb_u.l.bb_rightsib == cpu_to_be64(NULLFSBLOCK))
1150 cur->bc_levels[lev].ra |= XFS_BTCUR_RIGHTRA;
1151 } else {
1152 if (b->bb_u.s.bb_leftsib == cpu_to_be32(NULLAGBLOCK))
1153 cur->bc_levels[lev].ra |= XFS_BTCUR_LEFTRA;
1154 if (b->bb_u.s.bb_rightsib == cpu_to_be32(NULLAGBLOCK))
1155 cur->bc_levels[lev].ra |= XFS_BTCUR_RIGHTRA;
1156 }
1157 }
1158
1159 bool
xfs_btree_ptr_is_null(struct xfs_btree_cur * cur,const union xfs_btree_ptr * ptr)1160 xfs_btree_ptr_is_null(
1161 struct xfs_btree_cur *cur,
1162 const union xfs_btree_ptr *ptr)
1163 {
1164 if (cur->bc_ops->ptr_len == XFS_BTREE_LONG_PTR_LEN)
1165 return ptr->l == cpu_to_be64(NULLFSBLOCK);
1166 else
1167 return ptr->s == cpu_to_be32(NULLAGBLOCK);
1168 }
1169
1170 void
xfs_btree_set_ptr_null(struct xfs_btree_cur * cur,union xfs_btree_ptr * ptr)1171 xfs_btree_set_ptr_null(
1172 struct xfs_btree_cur *cur,
1173 union xfs_btree_ptr *ptr)
1174 {
1175 if (cur->bc_ops->ptr_len == XFS_BTREE_LONG_PTR_LEN)
1176 ptr->l = cpu_to_be64(NULLFSBLOCK);
1177 else
1178 ptr->s = cpu_to_be32(NULLAGBLOCK);
1179 }
1180
1181 static inline bool
xfs_btree_ptrs_equal(struct xfs_btree_cur * cur,union xfs_btree_ptr * ptr1,union xfs_btree_ptr * ptr2)1182 xfs_btree_ptrs_equal(
1183 struct xfs_btree_cur *cur,
1184 union xfs_btree_ptr *ptr1,
1185 union xfs_btree_ptr *ptr2)
1186 {
1187 if (cur->bc_ops->ptr_len == XFS_BTREE_LONG_PTR_LEN)
1188 return ptr1->l == ptr2->l;
1189 return ptr1->s == ptr2->s;
1190 }
1191
1192 /*
1193 * Get/set/init sibling pointers
1194 */
1195 void
xfs_btree_get_sibling(struct xfs_btree_cur * cur,struct xfs_btree_block * block,union xfs_btree_ptr * ptr,int lr)1196 xfs_btree_get_sibling(
1197 struct xfs_btree_cur *cur,
1198 struct xfs_btree_block *block,
1199 union xfs_btree_ptr *ptr,
1200 int lr)
1201 {
1202 ASSERT(lr == XFS_BB_LEFTSIB || lr == XFS_BB_RIGHTSIB);
1203
1204 if (cur->bc_ops->ptr_len == XFS_BTREE_LONG_PTR_LEN) {
1205 if (lr == XFS_BB_RIGHTSIB)
1206 ptr->l = block->bb_u.l.bb_rightsib;
1207 else
1208 ptr->l = block->bb_u.l.bb_leftsib;
1209 } else {
1210 if (lr == XFS_BB_RIGHTSIB)
1211 ptr->s = block->bb_u.s.bb_rightsib;
1212 else
1213 ptr->s = block->bb_u.s.bb_leftsib;
1214 }
1215 }
1216
1217 void
xfs_btree_set_sibling(struct xfs_btree_cur * cur,struct xfs_btree_block * block,const union xfs_btree_ptr * ptr,int lr)1218 xfs_btree_set_sibling(
1219 struct xfs_btree_cur *cur,
1220 struct xfs_btree_block *block,
1221 const union xfs_btree_ptr *ptr,
1222 int lr)
1223 {
1224 ASSERT(lr == XFS_BB_LEFTSIB || lr == XFS_BB_RIGHTSIB);
1225
1226 if (cur->bc_ops->ptr_len == XFS_BTREE_LONG_PTR_LEN) {
1227 if (lr == XFS_BB_RIGHTSIB)
1228 block->bb_u.l.bb_rightsib = ptr->l;
1229 else
1230 block->bb_u.l.bb_leftsib = ptr->l;
1231 } else {
1232 if (lr == XFS_BB_RIGHTSIB)
1233 block->bb_u.s.bb_rightsib = ptr->s;
1234 else
1235 block->bb_u.s.bb_leftsib = ptr->s;
1236 }
1237 }
1238
1239 static void
__xfs_btree_init_block(struct xfs_mount * mp,struct xfs_btree_block * buf,const struct xfs_btree_ops * ops,xfs_daddr_t blkno,__u16 level,__u16 numrecs,__u64 owner)1240 __xfs_btree_init_block(
1241 struct xfs_mount *mp,
1242 struct xfs_btree_block *buf,
1243 const struct xfs_btree_ops *ops,
1244 xfs_daddr_t blkno,
1245 __u16 level,
1246 __u16 numrecs,
1247 __u64 owner)
1248 {
1249 bool crc = xfs_has_crc(mp);
1250 __u32 magic = xfs_btree_magic(mp, ops);
1251
1252 buf->bb_magic = cpu_to_be32(magic);
1253 buf->bb_level = cpu_to_be16(level);
1254 buf->bb_numrecs = cpu_to_be16(numrecs);
1255
1256 if (ops->ptr_len == XFS_BTREE_LONG_PTR_LEN) {
1257 buf->bb_u.l.bb_leftsib = cpu_to_be64(NULLFSBLOCK);
1258 buf->bb_u.l.bb_rightsib = cpu_to_be64(NULLFSBLOCK);
1259 if (crc) {
1260 buf->bb_u.l.bb_blkno = cpu_to_be64(blkno);
1261 buf->bb_u.l.bb_owner = cpu_to_be64(owner);
1262 uuid_copy(&buf->bb_u.l.bb_uuid, &mp->m_sb.sb_meta_uuid);
1263 buf->bb_u.l.bb_pad = 0;
1264 buf->bb_u.l.bb_lsn = 0;
1265 }
1266 } else {
1267 buf->bb_u.s.bb_leftsib = cpu_to_be32(NULLAGBLOCK);
1268 buf->bb_u.s.bb_rightsib = cpu_to_be32(NULLAGBLOCK);
1269 if (crc) {
1270 buf->bb_u.s.bb_blkno = cpu_to_be64(blkno);
1271 /* owner is a 32 bit value on short blocks */
1272 buf->bb_u.s.bb_owner = cpu_to_be32((__u32)owner);
1273 uuid_copy(&buf->bb_u.s.bb_uuid, &mp->m_sb.sb_meta_uuid);
1274 buf->bb_u.s.bb_lsn = 0;
1275 }
1276 }
1277 }
1278
1279 void
xfs_btree_init_block(struct xfs_mount * mp,struct xfs_btree_block * block,const struct xfs_btree_ops * ops,__u16 level,__u16 numrecs,__u64 owner)1280 xfs_btree_init_block(
1281 struct xfs_mount *mp,
1282 struct xfs_btree_block *block,
1283 const struct xfs_btree_ops *ops,
1284 __u16 level,
1285 __u16 numrecs,
1286 __u64 owner)
1287 {
1288 __xfs_btree_init_block(mp, block, ops, XFS_BUF_DADDR_NULL, level,
1289 numrecs, owner);
1290 }
1291
1292 void
xfs_btree_init_buf(struct xfs_mount * mp,struct xfs_buf * bp,const struct xfs_btree_ops * ops,__u16 level,__u16 numrecs,__u64 owner)1293 xfs_btree_init_buf(
1294 struct xfs_mount *mp,
1295 struct xfs_buf *bp,
1296 const struct xfs_btree_ops *ops,
1297 __u16 level,
1298 __u16 numrecs,
1299 __u64 owner)
1300 {
1301 __xfs_btree_init_block(mp, XFS_BUF_TO_BLOCK(bp), ops,
1302 xfs_buf_daddr(bp), level, numrecs, owner);
1303 bp->b_ops = ops->buf_ops;
1304 }
1305
1306 static inline __u64
xfs_btree_owner(struct xfs_btree_cur * cur)1307 xfs_btree_owner(
1308 struct xfs_btree_cur *cur)
1309 {
1310 switch (cur->bc_ops->type) {
1311 case XFS_BTREE_TYPE_MEM:
1312 return cur->bc_mem.xfbtree->owner;
1313 case XFS_BTREE_TYPE_INODE:
1314 return cur->bc_ino.ip->i_ino;
1315 case XFS_BTREE_TYPE_AG:
1316 return cur->bc_ag.pag->pag_agno;
1317 default:
1318 ASSERT(0);
1319 return 0;
1320 }
1321 }
1322
1323 void
xfs_btree_init_block_cur(struct xfs_btree_cur * cur,struct xfs_buf * bp,int level,int numrecs)1324 xfs_btree_init_block_cur(
1325 struct xfs_btree_cur *cur,
1326 struct xfs_buf *bp,
1327 int level,
1328 int numrecs)
1329 {
1330 xfs_btree_init_buf(cur->bc_mp, bp, cur->bc_ops, level, numrecs,
1331 xfs_btree_owner(cur));
1332 }
1333
1334 STATIC void
xfs_btree_buf_to_ptr(struct xfs_btree_cur * cur,struct xfs_buf * bp,union xfs_btree_ptr * ptr)1335 xfs_btree_buf_to_ptr(
1336 struct xfs_btree_cur *cur,
1337 struct xfs_buf *bp,
1338 union xfs_btree_ptr *ptr)
1339 {
1340 switch (cur->bc_ops->type) {
1341 case XFS_BTREE_TYPE_AG:
1342 ptr->s = cpu_to_be32(xfs_daddr_to_agbno(cur->bc_mp,
1343 xfs_buf_daddr(bp)));
1344 break;
1345 case XFS_BTREE_TYPE_INODE:
1346 ptr->l = cpu_to_be64(XFS_DADDR_TO_FSB(cur->bc_mp,
1347 xfs_buf_daddr(bp)));
1348 break;
1349 case XFS_BTREE_TYPE_MEM:
1350 ptr->l = cpu_to_be64(xfs_daddr_to_xfbno(xfs_buf_daddr(bp)));
1351 break;
1352 }
1353 }
1354
1355 static inline void
xfs_btree_set_refs(struct xfs_btree_cur * cur,struct xfs_buf * bp)1356 xfs_btree_set_refs(
1357 struct xfs_btree_cur *cur,
1358 struct xfs_buf *bp)
1359 {
1360 xfs_buf_set_ref(bp, cur->bc_ops->lru_refs);
1361 }
1362
1363 int
xfs_btree_get_buf_block(struct xfs_btree_cur * cur,const union xfs_btree_ptr * ptr,struct xfs_btree_block ** block,struct xfs_buf ** bpp)1364 xfs_btree_get_buf_block(
1365 struct xfs_btree_cur *cur,
1366 const union xfs_btree_ptr *ptr,
1367 struct xfs_btree_block **block,
1368 struct xfs_buf **bpp)
1369 {
1370 xfs_daddr_t d;
1371 int error;
1372
1373 error = xfs_btree_ptr_to_daddr(cur, ptr, &d);
1374 if (error)
1375 return error;
1376 error = xfs_trans_get_buf(cur->bc_tp, xfs_btree_buftarg(cur), d,
1377 xfs_btree_bbsize(cur), 0, bpp);
1378 if (error)
1379 return error;
1380
1381 (*bpp)->b_ops = cur->bc_ops->buf_ops;
1382 *block = XFS_BUF_TO_BLOCK(*bpp);
1383 return 0;
1384 }
1385
1386 /*
1387 * Read in the buffer at the given ptr and return the buffer and
1388 * the block pointer within the buffer.
1389 */
1390 int
xfs_btree_read_buf_block(struct xfs_btree_cur * cur,const union xfs_btree_ptr * ptr,int flags,struct xfs_btree_block ** block,struct xfs_buf ** bpp)1391 xfs_btree_read_buf_block(
1392 struct xfs_btree_cur *cur,
1393 const union xfs_btree_ptr *ptr,
1394 int flags,
1395 struct xfs_btree_block **block,
1396 struct xfs_buf **bpp)
1397 {
1398 struct xfs_mount *mp = cur->bc_mp;
1399 xfs_daddr_t d;
1400 int error;
1401
1402 /* need to sort out how callers deal with failures first */
1403 ASSERT(!(flags & XBF_TRYLOCK));
1404
1405 error = xfs_btree_ptr_to_daddr(cur, ptr, &d);
1406 if (error)
1407 return error;
1408 error = xfs_trans_read_buf(mp, cur->bc_tp, xfs_btree_buftarg(cur), d,
1409 xfs_btree_bbsize(cur), flags, bpp,
1410 cur->bc_ops->buf_ops);
1411 if (xfs_metadata_is_sick(error))
1412 xfs_btree_mark_sick(cur);
1413 if (error)
1414 return error;
1415
1416 xfs_btree_set_refs(cur, *bpp);
1417 *block = XFS_BUF_TO_BLOCK(*bpp);
1418 return 0;
1419 }
1420
1421 /*
1422 * Copy keys from one btree block to another.
1423 */
1424 void
xfs_btree_copy_keys(struct xfs_btree_cur * cur,union xfs_btree_key * dst_key,const union xfs_btree_key * src_key,int numkeys)1425 xfs_btree_copy_keys(
1426 struct xfs_btree_cur *cur,
1427 union xfs_btree_key *dst_key,
1428 const union xfs_btree_key *src_key,
1429 int numkeys)
1430 {
1431 ASSERT(numkeys >= 0);
1432 memcpy(dst_key, src_key, numkeys * cur->bc_ops->key_len);
1433 }
1434
1435 /*
1436 * Copy records from one btree block to another.
1437 */
1438 STATIC void
xfs_btree_copy_recs(struct xfs_btree_cur * cur,union xfs_btree_rec * dst_rec,union xfs_btree_rec * src_rec,int numrecs)1439 xfs_btree_copy_recs(
1440 struct xfs_btree_cur *cur,
1441 union xfs_btree_rec *dst_rec,
1442 union xfs_btree_rec *src_rec,
1443 int numrecs)
1444 {
1445 ASSERT(numrecs >= 0);
1446 memcpy(dst_rec, src_rec, numrecs * cur->bc_ops->rec_len);
1447 }
1448
1449 /*
1450 * Copy block pointers from one btree block to another.
1451 */
1452 void
xfs_btree_copy_ptrs(struct xfs_btree_cur * cur,union xfs_btree_ptr * dst_ptr,const union xfs_btree_ptr * src_ptr,int numptrs)1453 xfs_btree_copy_ptrs(
1454 struct xfs_btree_cur *cur,
1455 union xfs_btree_ptr *dst_ptr,
1456 const union xfs_btree_ptr *src_ptr,
1457 int numptrs)
1458 {
1459 ASSERT(numptrs >= 0);
1460 memcpy(dst_ptr, src_ptr, numptrs * cur->bc_ops->ptr_len);
1461 }
1462
1463 /*
1464 * Shift keys one index left/right inside a single btree block.
1465 */
1466 STATIC void
xfs_btree_shift_keys(struct xfs_btree_cur * cur,union xfs_btree_key * key,int dir,int numkeys)1467 xfs_btree_shift_keys(
1468 struct xfs_btree_cur *cur,
1469 union xfs_btree_key *key,
1470 int dir,
1471 int numkeys)
1472 {
1473 char *dst_key;
1474
1475 ASSERT(numkeys >= 0);
1476 ASSERT(dir == 1 || dir == -1);
1477
1478 dst_key = (char *)key + (dir * cur->bc_ops->key_len);
1479 memmove(dst_key, key, numkeys * cur->bc_ops->key_len);
1480 }
1481
1482 /*
1483 * Shift records one index left/right inside a single btree block.
1484 */
1485 STATIC void
xfs_btree_shift_recs(struct xfs_btree_cur * cur,union xfs_btree_rec * rec,int dir,int numrecs)1486 xfs_btree_shift_recs(
1487 struct xfs_btree_cur *cur,
1488 union xfs_btree_rec *rec,
1489 int dir,
1490 int numrecs)
1491 {
1492 char *dst_rec;
1493
1494 ASSERT(numrecs >= 0);
1495 ASSERT(dir == 1 || dir == -1);
1496
1497 dst_rec = (char *)rec + (dir * cur->bc_ops->rec_len);
1498 memmove(dst_rec, rec, numrecs * cur->bc_ops->rec_len);
1499 }
1500
1501 /*
1502 * Shift block pointers one index left/right inside a single btree block.
1503 */
1504 STATIC void
xfs_btree_shift_ptrs(struct xfs_btree_cur * cur,union xfs_btree_ptr * ptr,int dir,int numptrs)1505 xfs_btree_shift_ptrs(
1506 struct xfs_btree_cur *cur,
1507 union xfs_btree_ptr *ptr,
1508 int dir,
1509 int numptrs)
1510 {
1511 char *dst_ptr;
1512
1513 ASSERT(numptrs >= 0);
1514 ASSERT(dir == 1 || dir == -1);
1515
1516 dst_ptr = (char *)ptr + (dir * cur->bc_ops->ptr_len);
1517 memmove(dst_ptr, ptr, numptrs * cur->bc_ops->ptr_len);
1518 }
1519
1520 /*
1521 * Log key values from the btree block.
1522 */
1523 STATIC void
xfs_btree_log_keys(struct xfs_btree_cur * cur,struct xfs_buf * bp,int first,int last)1524 xfs_btree_log_keys(
1525 struct xfs_btree_cur *cur,
1526 struct xfs_buf *bp,
1527 int first,
1528 int last)
1529 {
1530
1531 if (bp) {
1532 xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
1533 xfs_trans_log_buf(cur->bc_tp, bp,
1534 xfs_btree_key_offset(cur, first),
1535 xfs_btree_key_offset(cur, last + 1) - 1);
1536 } else {
1537 xfs_trans_log_inode(cur->bc_tp, cur->bc_ino.ip,
1538 xfs_ilog_fbroot(cur->bc_ino.whichfork));
1539 }
1540 }
1541
1542 /*
1543 * Log record values from the btree block.
1544 */
1545 void
xfs_btree_log_recs(struct xfs_btree_cur * cur,struct xfs_buf * bp,int first,int last)1546 xfs_btree_log_recs(
1547 struct xfs_btree_cur *cur,
1548 struct xfs_buf *bp,
1549 int first,
1550 int last)
1551 {
1552
1553 xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
1554 xfs_trans_log_buf(cur->bc_tp, bp,
1555 xfs_btree_rec_offset(cur, first),
1556 xfs_btree_rec_offset(cur, last + 1) - 1);
1557
1558 }
1559
1560 /*
1561 * Log block pointer fields from a btree block (nonleaf).
1562 */
1563 STATIC void
xfs_btree_log_ptrs(struct xfs_btree_cur * cur,struct xfs_buf * bp,int first,int last)1564 xfs_btree_log_ptrs(
1565 struct xfs_btree_cur *cur, /* btree cursor */
1566 struct xfs_buf *bp, /* buffer containing btree block */
1567 int first, /* index of first pointer to log */
1568 int last) /* index of last pointer to log */
1569 {
1570
1571 if (bp) {
1572 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
1573 int level = xfs_btree_get_level(block);
1574
1575 xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
1576 xfs_trans_log_buf(cur->bc_tp, bp,
1577 xfs_btree_ptr_offset(cur, first, level),
1578 xfs_btree_ptr_offset(cur, last + 1, level) - 1);
1579 } else {
1580 xfs_trans_log_inode(cur->bc_tp, cur->bc_ino.ip,
1581 xfs_ilog_fbroot(cur->bc_ino.whichfork));
1582 }
1583
1584 }
1585
1586 /*
1587 * Log fields from a btree block header.
1588 */
1589 void
xfs_btree_log_block(struct xfs_btree_cur * cur,struct xfs_buf * bp,uint32_t fields)1590 xfs_btree_log_block(
1591 struct xfs_btree_cur *cur, /* btree cursor */
1592 struct xfs_buf *bp, /* buffer containing btree block */
1593 uint32_t fields) /* mask of fields: XFS_BB_... */
1594 {
1595 int first; /* first byte offset logged */
1596 int last; /* last byte offset logged */
1597 static const short soffsets[] = { /* table of offsets (short) */
1598 offsetof(struct xfs_btree_block, bb_magic),
1599 offsetof(struct xfs_btree_block, bb_level),
1600 offsetof(struct xfs_btree_block, bb_numrecs),
1601 offsetof(struct xfs_btree_block, bb_u.s.bb_leftsib),
1602 offsetof(struct xfs_btree_block, bb_u.s.bb_rightsib),
1603 offsetof(struct xfs_btree_block, bb_u.s.bb_blkno),
1604 offsetof(struct xfs_btree_block, bb_u.s.bb_lsn),
1605 offsetof(struct xfs_btree_block, bb_u.s.bb_uuid),
1606 offsetof(struct xfs_btree_block, bb_u.s.bb_owner),
1607 offsetof(struct xfs_btree_block, bb_u.s.bb_crc),
1608 XFS_BTREE_SBLOCK_CRC_LEN
1609 };
1610 static const short loffsets[] = { /* table of offsets (long) */
1611 offsetof(struct xfs_btree_block, bb_magic),
1612 offsetof(struct xfs_btree_block, bb_level),
1613 offsetof(struct xfs_btree_block, bb_numrecs),
1614 offsetof(struct xfs_btree_block, bb_u.l.bb_leftsib),
1615 offsetof(struct xfs_btree_block, bb_u.l.bb_rightsib),
1616 offsetof(struct xfs_btree_block, bb_u.l.bb_blkno),
1617 offsetof(struct xfs_btree_block, bb_u.l.bb_lsn),
1618 offsetof(struct xfs_btree_block, bb_u.l.bb_uuid),
1619 offsetof(struct xfs_btree_block, bb_u.l.bb_owner),
1620 offsetof(struct xfs_btree_block, bb_u.l.bb_crc),
1621 offsetof(struct xfs_btree_block, bb_u.l.bb_pad),
1622 XFS_BTREE_LBLOCK_CRC_LEN
1623 };
1624
1625 if (bp) {
1626 int nbits;
1627
1628 if (xfs_has_crc(cur->bc_mp)) {
1629 /*
1630 * We don't log the CRC when updating a btree
1631 * block but instead recreate it during log
1632 * recovery. As the log buffers have checksums
1633 * of their own this is safe and avoids logging a crc
1634 * update in a lot of places.
1635 */
1636 if (fields == XFS_BB_ALL_BITS)
1637 fields = XFS_BB_ALL_BITS_CRC;
1638 nbits = XFS_BB_NUM_BITS_CRC;
1639 } else {
1640 nbits = XFS_BB_NUM_BITS;
1641 }
1642 xfs_btree_offsets(fields,
1643 (cur->bc_ops->ptr_len == XFS_BTREE_LONG_PTR_LEN) ?
1644 loffsets : soffsets,
1645 nbits, &first, &last);
1646 xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
1647 xfs_trans_log_buf(cur->bc_tp, bp, first, last);
1648 } else {
1649 xfs_trans_log_inode(cur->bc_tp, cur->bc_ino.ip,
1650 xfs_ilog_fbroot(cur->bc_ino.whichfork));
1651 }
1652 }
1653
1654 /*
1655 * Increment cursor by one record at the level.
1656 * For nonzero levels the leaf-ward information is untouched.
1657 */
1658 int /* error */
xfs_btree_increment(struct xfs_btree_cur * cur,int level,int * stat)1659 xfs_btree_increment(
1660 struct xfs_btree_cur *cur,
1661 int level,
1662 int *stat) /* success/failure */
1663 {
1664 struct xfs_btree_block *block;
1665 union xfs_btree_ptr ptr;
1666 struct xfs_buf *bp;
1667 int error; /* error return value */
1668 int lev;
1669
1670 ASSERT(level < cur->bc_nlevels);
1671
1672 /* Read-ahead to the right at this level. */
1673 xfs_btree_readahead(cur, level, XFS_BTCUR_RIGHTRA);
1674
1675 /* Get a pointer to the btree block. */
1676 block = xfs_btree_get_block(cur, level, &bp);
1677
1678 #ifdef DEBUG
1679 error = xfs_btree_check_block(cur, block, level, bp);
1680 if (error)
1681 goto error0;
1682 #endif
1683
1684 /* We're done if we remain in the block after the increment. */
1685 if (++cur->bc_levels[level].ptr <= xfs_btree_get_numrecs(block))
1686 goto out1;
1687
1688 /* Fail if we just went off the right edge of the tree. */
1689 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
1690 if (xfs_btree_ptr_is_null(cur, &ptr))
1691 goto out0;
1692
1693 XFS_BTREE_STATS_INC(cur, increment);
1694
1695 /*
1696 * March up the tree incrementing pointers.
1697 * Stop when we don't go off the right edge of a block.
1698 */
1699 for (lev = level + 1; lev < cur->bc_nlevels; lev++) {
1700 block = xfs_btree_get_block(cur, lev, &bp);
1701
1702 #ifdef DEBUG
1703 error = xfs_btree_check_block(cur, block, lev, bp);
1704 if (error)
1705 goto error0;
1706 #endif
1707
1708 if (++cur->bc_levels[lev].ptr <= xfs_btree_get_numrecs(block))
1709 break;
1710
1711 /* Read-ahead the right block for the next loop. */
1712 xfs_btree_readahead(cur, lev, XFS_BTCUR_RIGHTRA);
1713 }
1714
1715 /*
1716 * If we went off the root then we are either seriously
1717 * confused or have the tree root in an inode.
1718 */
1719 if (lev == cur->bc_nlevels) {
1720 if (cur->bc_ops->type == XFS_BTREE_TYPE_INODE)
1721 goto out0;
1722 ASSERT(0);
1723 xfs_btree_mark_sick(cur);
1724 error = -EFSCORRUPTED;
1725 goto error0;
1726 }
1727 ASSERT(lev < cur->bc_nlevels);
1728
1729 /*
1730 * Now walk back down the tree, fixing up the cursor's buffer
1731 * pointers and key numbers.
1732 */
1733 for (block = xfs_btree_get_block(cur, lev, &bp); lev > level; ) {
1734 union xfs_btree_ptr *ptrp;
1735
1736 ptrp = xfs_btree_ptr_addr(cur, cur->bc_levels[lev].ptr, block);
1737 --lev;
1738 error = xfs_btree_read_buf_block(cur, ptrp, 0, &block, &bp);
1739 if (error)
1740 goto error0;
1741
1742 xfs_btree_setbuf(cur, lev, bp);
1743 cur->bc_levels[lev].ptr = 1;
1744 }
1745 out1:
1746 *stat = 1;
1747 return 0;
1748
1749 out0:
1750 *stat = 0;
1751 return 0;
1752
1753 error0:
1754 return error;
1755 }
1756
1757 /*
1758 * Decrement cursor by one record at the level.
1759 * For nonzero levels the leaf-ward information is untouched.
1760 */
1761 int /* error */
xfs_btree_decrement(struct xfs_btree_cur * cur,int level,int * stat)1762 xfs_btree_decrement(
1763 struct xfs_btree_cur *cur,
1764 int level,
1765 int *stat) /* success/failure */
1766 {
1767 struct xfs_btree_block *block;
1768 struct xfs_buf *bp;
1769 int error; /* error return value */
1770 int lev;
1771 union xfs_btree_ptr ptr;
1772
1773 ASSERT(level < cur->bc_nlevels);
1774
1775 /* Read-ahead to the left at this level. */
1776 xfs_btree_readahead(cur, level, XFS_BTCUR_LEFTRA);
1777
1778 /* We're done if we remain in the block after the decrement. */
1779 if (--cur->bc_levels[level].ptr > 0)
1780 goto out1;
1781
1782 /* Get a pointer to the btree block. */
1783 block = xfs_btree_get_block(cur, level, &bp);
1784
1785 #ifdef DEBUG
1786 error = xfs_btree_check_block(cur, block, level, bp);
1787 if (error)
1788 goto error0;
1789 #endif
1790
1791 /* Fail if we just went off the left edge of the tree. */
1792 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_LEFTSIB);
1793 if (xfs_btree_ptr_is_null(cur, &ptr))
1794 goto out0;
1795
1796 XFS_BTREE_STATS_INC(cur, decrement);
1797
1798 /*
1799 * March up the tree decrementing pointers.
1800 * Stop when we don't go off the left edge of a block.
1801 */
1802 for (lev = level + 1; lev < cur->bc_nlevels; lev++) {
1803 if (--cur->bc_levels[lev].ptr > 0)
1804 break;
1805 /* Read-ahead the left block for the next loop. */
1806 xfs_btree_readahead(cur, lev, XFS_BTCUR_LEFTRA);
1807 }
1808
1809 /*
1810 * If we went off the root then we are seriously confused.
1811 * or the root of the tree is in an inode.
1812 */
1813 if (lev == cur->bc_nlevels) {
1814 if (cur->bc_ops->type == XFS_BTREE_TYPE_INODE)
1815 goto out0;
1816 ASSERT(0);
1817 xfs_btree_mark_sick(cur);
1818 error = -EFSCORRUPTED;
1819 goto error0;
1820 }
1821 ASSERT(lev < cur->bc_nlevels);
1822
1823 /*
1824 * Now walk back down the tree, fixing up the cursor's buffer
1825 * pointers and key numbers.
1826 */
1827 for (block = xfs_btree_get_block(cur, lev, &bp); lev > level; ) {
1828 union xfs_btree_ptr *ptrp;
1829
1830 ptrp = xfs_btree_ptr_addr(cur, cur->bc_levels[lev].ptr, block);
1831 --lev;
1832 error = xfs_btree_read_buf_block(cur, ptrp, 0, &block, &bp);
1833 if (error)
1834 goto error0;
1835 xfs_btree_setbuf(cur, lev, bp);
1836 cur->bc_levels[lev].ptr = xfs_btree_get_numrecs(block);
1837 }
1838 out1:
1839 *stat = 1;
1840 return 0;
1841
1842 out0:
1843 *stat = 0;
1844 return 0;
1845
1846 error0:
1847 return error;
1848 }
1849
1850 /*
1851 * Check the btree block owner now that we have the context to know who the
1852 * real owner is.
1853 */
1854 static inline xfs_failaddr_t
xfs_btree_check_block_owner(struct xfs_btree_cur * cur,struct xfs_btree_block * block)1855 xfs_btree_check_block_owner(
1856 struct xfs_btree_cur *cur,
1857 struct xfs_btree_block *block)
1858 {
1859 __u64 owner;
1860
1861 if (!xfs_has_crc(cur->bc_mp) ||
1862 (cur->bc_flags & XFS_BTREE_BMBT_INVALID_OWNER))
1863 return NULL;
1864
1865 owner = xfs_btree_owner(cur);
1866 if (cur->bc_ops->ptr_len == XFS_BTREE_LONG_PTR_LEN) {
1867 if (be64_to_cpu(block->bb_u.l.bb_owner) != owner)
1868 return __this_address;
1869 } else {
1870 if (be32_to_cpu(block->bb_u.s.bb_owner) != owner)
1871 return __this_address;
1872 }
1873
1874 return NULL;
1875 }
1876
1877 int
xfs_btree_lookup_get_block(struct xfs_btree_cur * cur,int level,const union xfs_btree_ptr * pp,struct xfs_btree_block ** blkp)1878 xfs_btree_lookup_get_block(
1879 struct xfs_btree_cur *cur, /* btree cursor */
1880 int level, /* level in the btree */
1881 const union xfs_btree_ptr *pp, /* ptr to btree block */
1882 struct xfs_btree_block **blkp) /* return btree block */
1883 {
1884 struct xfs_buf *bp; /* buffer pointer for btree block */
1885 xfs_daddr_t daddr;
1886 int error = 0;
1887
1888 /* special case the root block if in an inode */
1889 if (xfs_btree_at_iroot(cur, level)) {
1890 *blkp = xfs_btree_get_iroot(cur);
1891 return 0;
1892 }
1893
1894 /*
1895 * If the old buffer at this level for the disk address we are
1896 * looking for re-use it.
1897 *
1898 * Otherwise throw it away and get a new one.
1899 */
1900 bp = cur->bc_levels[level].bp;
1901 error = xfs_btree_ptr_to_daddr(cur, pp, &daddr);
1902 if (error)
1903 return error;
1904 if (bp && xfs_buf_daddr(bp) == daddr) {
1905 *blkp = XFS_BUF_TO_BLOCK(bp);
1906 return 0;
1907 }
1908
1909 error = xfs_btree_read_buf_block(cur, pp, 0, blkp, &bp);
1910 if (error)
1911 return error;
1912
1913 /* Check the inode owner since the verifiers don't. */
1914 if (xfs_btree_check_block_owner(cur, *blkp) != NULL)
1915 goto out_bad;
1916
1917 /* Did we get the level we were looking for? */
1918 if (be16_to_cpu((*blkp)->bb_level) != level)
1919 goto out_bad;
1920
1921 /* Check that internal nodes have at least one record. */
1922 if (level != 0 && be16_to_cpu((*blkp)->bb_numrecs) == 0)
1923 goto out_bad;
1924
1925 xfs_btree_setbuf(cur, level, bp);
1926 return 0;
1927
1928 out_bad:
1929 *blkp = NULL;
1930 xfs_buf_mark_corrupt(bp);
1931 xfs_trans_brelse(cur->bc_tp, bp);
1932 xfs_btree_mark_sick(cur);
1933 return -EFSCORRUPTED;
1934 }
1935
1936 /*
1937 * Get current search key. For level 0 we don't actually have a key
1938 * structure so we make one up from the record. For all other levels
1939 * we just return the right key.
1940 */
1941 STATIC union xfs_btree_key *
xfs_lookup_get_search_key(struct xfs_btree_cur * cur,int level,int keyno,struct xfs_btree_block * block,union xfs_btree_key * kp)1942 xfs_lookup_get_search_key(
1943 struct xfs_btree_cur *cur,
1944 int level,
1945 int keyno,
1946 struct xfs_btree_block *block,
1947 union xfs_btree_key *kp)
1948 {
1949 if (level == 0) {
1950 cur->bc_ops->init_key_from_rec(kp,
1951 xfs_btree_rec_addr(cur, keyno, block));
1952 return kp;
1953 }
1954
1955 return xfs_btree_key_addr(cur, keyno, block);
1956 }
1957
1958 /*
1959 * Initialize a pointer to the root block.
1960 */
1961 void
xfs_btree_init_ptr_from_cur(struct xfs_btree_cur * cur,union xfs_btree_ptr * ptr)1962 xfs_btree_init_ptr_from_cur(
1963 struct xfs_btree_cur *cur,
1964 union xfs_btree_ptr *ptr)
1965 {
1966 if (cur->bc_ops->type == XFS_BTREE_TYPE_INODE) {
1967 /*
1968 * Inode-rooted btrees call xfs_btree_get_iroot to find the root
1969 * in xfs_btree_lookup_get_block and don't need a pointer here.
1970 */
1971 ptr->l = 0;
1972 } else if (cur->bc_flags & XFS_BTREE_STAGING) {
1973 ptr->s = cpu_to_be32(cur->bc_ag.afake->af_root);
1974 } else {
1975 cur->bc_ops->init_ptr_from_cur(cur, ptr);
1976 }
1977 }
1978
1979 /*
1980 * Lookup the record. The cursor is made to point to it, based on dir.
1981 * stat is set to 0 if can't find any such record, 1 for success.
1982 */
1983 int /* error */
xfs_btree_lookup(struct xfs_btree_cur * cur,xfs_lookup_t dir,int * stat)1984 xfs_btree_lookup(
1985 struct xfs_btree_cur *cur, /* btree cursor */
1986 xfs_lookup_t dir, /* <=, ==, or >= */
1987 int *stat) /* success/failure */
1988 {
1989 struct xfs_btree_block *block; /* current btree block */
1990 int64_t diff; /* difference for the current key */
1991 int error; /* error return value */
1992 int keyno; /* current key number */
1993 int level; /* level in the btree */
1994 union xfs_btree_ptr *pp; /* ptr to btree block */
1995 union xfs_btree_ptr ptr; /* ptr to btree block */
1996
1997 XFS_BTREE_STATS_INC(cur, lookup);
1998
1999 /* No such thing as a zero-level tree. */
2000 if (XFS_IS_CORRUPT(cur->bc_mp, cur->bc_nlevels == 0)) {
2001 xfs_btree_mark_sick(cur);
2002 return -EFSCORRUPTED;
2003 }
2004
2005 block = NULL;
2006 keyno = 0;
2007
2008 /* initialise start pointer from cursor */
2009 xfs_btree_init_ptr_from_cur(cur, &ptr);
2010 pp = &ptr;
2011
2012 /*
2013 * Iterate over each level in the btree, starting at the root.
2014 * For each level above the leaves, find the key we need, based
2015 * on the lookup record, then follow the corresponding block
2016 * pointer down to the next level.
2017 */
2018 for (level = cur->bc_nlevels - 1, diff = 1; level >= 0; level--) {
2019 /* Get the block we need to do the lookup on. */
2020 error = xfs_btree_lookup_get_block(cur, level, pp, &block);
2021 if (error)
2022 goto error0;
2023
2024 if (diff == 0) {
2025 /*
2026 * If we already had a key match at a higher level, we
2027 * know we need to use the first entry in this block.
2028 */
2029 keyno = 1;
2030 } else {
2031 /* Otherwise search this block. Do a binary search. */
2032
2033 int high; /* high entry number */
2034 int low; /* low entry number */
2035
2036 /* Set low and high entry numbers, 1-based. */
2037 low = 1;
2038 high = xfs_btree_get_numrecs(block);
2039 if (!high) {
2040 /* Block is empty, must be an empty leaf. */
2041 if (level != 0 || cur->bc_nlevels != 1) {
2042 XFS_CORRUPTION_ERROR(__func__,
2043 XFS_ERRLEVEL_LOW,
2044 cur->bc_mp, block,
2045 sizeof(*block));
2046 xfs_btree_mark_sick(cur);
2047 return -EFSCORRUPTED;
2048 }
2049
2050 cur->bc_levels[0].ptr = dir != XFS_LOOKUP_LE;
2051 *stat = 0;
2052 return 0;
2053 }
2054
2055 /* Binary search the block. */
2056 while (low <= high) {
2057 union xfs_btree_key key;
2058 union xfs_btree_key *kp;
2059
2060 XFS_BTREE_STATS_INC(cur, compare);
2061
2062 /* keyno is average of low and high. */
2063 keyno = (low + high) >> 1;
2064
2065 /* Get current search key */
2066 kp = xfs_lookup_get_search_key(cur, level,
2067 keyno, block, &key);
2068
2069 /*
2070 * Compute difference to get next direction:
2071 * - less than, move right
2072 * - greater than, move left
2073 * - equal, we're done
2074 */
2075 diff = cur->bc_ops->key_diff(cur, kp);
2076 if (diff < 0)
2077 low = keyno + 1;
2078 else if (diff > 0)
2079 high = keyno - 1;
2080 else
2081 break;
2082 }
2083 }
2084
2085 /*
2086 * If there are more levels, set up for the next level
2087 * by getting the block number and filling in the cursor.
2088 */
2089 if (level > 0) {
2090 /*
2091 * If we moved left, need the previous key number,
2092 * unless there isn't one.
2093 */
2094 if (diff > 0 && --keyno < 1)
2095 keyno = 1;
2096 pp = xfs_btree_ptr_addr(cur, keyno, block);
2097
2098 error = xfs_btree_debug_check_ptr(cur, pp, 0, level);
2099 if (error)
2100 goto error0;
2101
2102 cur->bc_levels[level].ptr = keyno;
2103 }
2104 }
2105
2106 /* Done with the search. See if we need to adjust the results. */
2107 if (dir != XFS_LOOKUP_LE && diff < 0) {
2108 keyno++;
2109 /*
2110 * If ge search and we went off the end of the block, but it's
2111 * not the last block, we're in the wrong block.
2112 */
2113 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
2114 if (dir == XFS_LOOKUP_GE &&
2115 keyno > xfs_btree_get_numrecs(block) &&
2116 !xfs_btree_ptr_is_null(cur, &ptr)) {
2117 int i;
2118
2119 cur->bc_levels[0].ptr = keyno;
2120 error = xfs_btree_increment(cur, 0, &i);
2121 if (error)
2122 goto error0;
2123 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
2124 xfs_btree_mark_sick(cur);
2125 return -EFSCORRUPTED;
2126 }
2127 *stat = 1;
2128 return 0;
2129 }
2130 } else if (dir == XFS_LOOKUP_LE && diff > 0)
2131 keyno--;
2132 cur->bc_levels[0].ptr = keyno;
2133
2134 /* Return if we succeeded or not. */
2135 if (keyno == 0 || keyno > xfs_btree_get_numrecs(block))
2136 *stat = 0;
2137 else if (dir != XFS_LOOKUP_EQ || diff == 0)
2138 *stat = 1;
2139 else
2140 *stat = 0;
2141 return 0;
2142
2143 error0:
2144 return error;
2145 }
2146
2147 /* Find the high key storage area from a regular key. */
2148 union xfs_btree_key *
xfs_btree_high_key_from_key(struct xfs_btree_cur * cur,union xfs_btree_key * key)2149 xfs_btree_high_key_from_key(
2150 struct xfs_btree_cur *cur,
2151 union xfs_btree_key *key)
2152 {
2153 ASSERT(cur->bc_ops->geom_flags & XFS_BTGEO_OVERLAPPING);
2154 return (union xfs_btree_key *)((char *)key +
2155 (cur->bc_ops->key_len / 2));
2156 }
2157
2158 /* Determine the low (and high if overlapped) keys of a leaf block */
2159 STATIC void
xfs_btree_get_leaf_keys(struct xfs_btree_cur * cur,struct xfs_btree_block * block,union xfs_btree_key * key)2160 xfs_btree_get_leaf_keys(
2161 struct xfs_btree_cur *cur,
2162 struct xfs_btree_block *block,
2163 union xfs_btree_key *key)
2164 {
2165 union xfs_btree_key max_hkey;
2166 union xfs_btree_key hkey;
2167 union xfs_btree_rec *rec;
2168 union xfs_btree_key *high;
2169 int n;
2170
2171 rec = xfs_btree_rec_addr(cur, 1, block);
2172 cur->bc_ops->init_key_from_rec(key, rec);
2173
2174 if (cur->bc_ops->geom_flags & XFS_BTGEO_OVERLAPPING) {
2175
2176 cur->bc_ops->init_high_key_from_rec(&max_hkey, rec);
2177 for (n = 2; n <= xfs_btree_get_numrecs(block); n++) {
2178 rec = xfs_btree_rec_addr(cur, n, block);
2179 cur->bc_ops->init_high_key_from_rec(&hkey, rec);
2180 if (xfs_btree_keycmp_gt(cur, &hkey, &max_hkey))
2181 max_hkey = hkey;
2182 }
2183
2184 high = xfs_btree_high_key_from_key(cur, key);
2185 memcpy(high, &max_hkey, cur->bc_ops->key_len / 2);
2186 }
2187 }
2188
2189 /* Determine the low (and high if overlapped) keys of a node block */
2190 STATIC void
xfs_btree_get_node_keys(struct xfs_btree_cur * cur,struct xfs_btree_block * block,union xfs_btree_key * key)2191 xfs_btree_get_node_keys(
2192 struct xfs_btree_cur *cur,
2193 struct xfs_btree_block *block,
2194 union xfs_btree_key *key)
2195 {
2196 union xfs_btree_key *hkey;
2197 union xfs_btree_key *max_hkey;
2198 union xfs_btree_key *high;
2199 int n;
2200
2201 if (cur->bc_ops->geom_flags & XFS_BTGEO_OVERLAPPING) {
2202 memcpy(key, xfs_btree_key_addr(cur, 1, block),
2203 cur->bc_ops->key_len / 2);
2204
2205 max_hkey = xfs_btree_high_key_addr(cur, 1, block);
2206 for (n = 2; n <= xfs_btree_get_numrecs(block); n++) {
2207 hkey = xfs_btree_high_key_addr(cur, n, block);
2208 if (xfs_btree_keycmp_gt(cur, hkey, max_hkey))
2209 max_hkey = hkey;
2210 }
2211
2212 high = xfs_btree_high_key_from_key(cur, key);
2213 memcpy(high, max_hkey, cur->bc_ops->key_len / 2);
2214 } else {
2215 memcpy(key, xfs_btree_key_addr(cur, 1, block),
2216 cur->bc_ops->key_len);
2217 }
2218 }
2219
2220 /* Derive the keys for any btree block. */
2221 void
xfs_btree_get_keys(struct xfs_btree_cur * cur,struct xfs_btree_block * block,union xfs_btree_key * key)2222 xfs_btree_get_keys(
2223 struct xfs_btree_cur *cur,
2224 struct xfs_btree_block *block,
2225 union xfs_btree_key *key)
2226 {
2227 if (be16_to_cpu(block->bb_level) == 0)
2228 xfs_btree_get_leaf_keys(cur, block, key);
2229 else
2230 xfs_btree_get_node_keys(cur, block, key);
2231 }
2232
2233 /*
2234 * Decide if we need to update the parent keys of a btree block. For
2235 * a standard btree this is only necessary if we're updating the first
2236 * record/key. For an overlapping btree, we must always update the
2237 * keys because the highest key can be in any of the records or keys
2238 * in the block.
2239 */
2240 static inline bool
xfs_btree_needs_key_update(struct xfs_btree_cur * cur,int ptr)2241 xfs_btree_needs_key_update(
2242 struct xfs_btree_cur *cur,
2243 int ptr)
2244 {
2245 return (cur->bc_ops->geom_flags & XFS_BTGEO_OVERLAPPING) || ptr == 1;
2246 }
2247
2248 /*
2249 * Update the low and high parent keys of the given level, progressing
2250 * towards the root. If force_all is false, stop if the keys for a given
2251 * level do not need updating.
2252 */
2253 STATIC int
__xfs_btree_updkeys(struct xfs_btree_cur * cur,int level,struct xfs_btree_block * block,struct xfs_buf * bp0,bool force_all)2254 __xfs_btree_updkeys(
2255 struct xfs_btree_cur *cur,
2256 int level,
2257 struct xfs_btree_block *block,
2258 struct xfs_buf *bp0,
2259 bool force_all)
2260 {
2261 union xfs_btree_key key; /* keys from current level */
2262 union xfs_btree_key *lkey; /* keys from the next level up */
2263 union xfs_btree_key *hkey;
2264 union xfs_btree_key *nlkey; /* keys from the next level up */
2265 union xfs_btree_key *nhkey;
2266 struct xfs_buf *bp;
2267 int ptr;
2268
2269 ASSERT(cur->bc_ops->geom_flags & XFS_BTGEO_OVERLAPPING);
2270
2271 /* Exit if there aren't any parent levels to update. */
2272 if (level + 1 >= cur->bc_nlevels)
2273 return 0;
2274
2275 trace_xfs_btree_updkeys(cur, level, bp0);
2276
2277 lkey = &key;
2278 hkey = xfs_btree_high_key_from_key(cur, lkey);
2279 xfs_btree_get_keys(cur, block, lkey);
2280 for (level++; level < cur->bc_nlevels; level++) {
2281 #ifdef DEBUG
2282 int error;
2283 #endif
2284 block = xfs_btree_get_block(cur, level, &bp);
2285 trace_xfs_btree_updkeys(cur, level, bp);
2286 #ifdef DEBUG
2287 error = xfs_btree_check_block(cur, block, level, bp);
2288 if (error)
2289 return error;
2290 #endif
2291 ptr = cur->bc_levels[level].ptr;
2292 nlkey = xfs_btree_key_addr(cur, ptr, block);
2293 nhkey = xfs_btree_high_key_addr(cur, ptr, block);
2294 if (!force_all &&
2295 xfs_btree_keycmp_eq(cur, nlkey, lkey) &&
2296 xfs_btree_keycmp_eq(cur, nhkey, hkey))
2297 break;
2298 xfs_btree_copy_keys(cur, nlkey, lkey, 1);
2299 xfs_btree_log_keys(cur, bp, ptr, ptr);
2300 if (level + 1 >= cur->bc_nlevels)
2301 break;
2302 xfs_btree_get_node_keys(cur, block, lkey);
2303 }
2304
2305 return 0;
2306 }
2307
2308 /* Update all the keys from some level in cursor back to the root. */
2309 STATIC int
xfs_btree_updkeys_force(struct xfs_btree_cur * cur,int level)2310 xfs_btree_updkeys_force(
2311 struct xfs_btree_cur *cur,
2312 int level)
2313 {
2314 struct xfs_buf *bp;
2315 struct xfs_btree_block *block;
2316
2317 block = xfs_btree_get_block(cur, level, &bp);
2318 return __xfs_btree_updkeys(cur, level, block, bp, true);
2319 }
2320
2321 /*
2322 * Update the parent keys of the given level, progressing towards the root.
2323 */
2324 STATIC int
xfs_btree_update_keys(struct xfs_btree_cur * cur,int level)2325 xfs_btree_update_keys(
2326 struct xfs_btree_cur *cur,
2327 int level)
2328 {
2329 struct xfs_btree_block *block;
2330 struct xfs_buf *bp;
2331 union xfs_btree_key *kp;
2332 union xfs_btree_key key;
2333 int ptr;
2334
2335 ASSERT(level >= 0);
2336
2337 block = xfs_btree_get_block(cur, level, &bp);
2338 if (cur->bc_ops->geom_flags & XFS_BTGEO_OVERLAPPING)
2339 return __xfs_btree_updkeys(cur, level, block, bp, false);
2340
2341 /*
2342 * Go up the tree from this level toward the root.
2343 * At each level, update the key value to the value input.
2344 * Stop when we reach a level where the cursor isn't pointing
2345 * at the first entry in the block.
2346 */
2347 xfs_btree_get_keys(cur, block, &key);
2348 for (level++, ptr = 1; ptr == 1 && level < cur->bc_nlevels; level++) {
2349 #ifdef DEBUG
2350 int error;
2351 #endif
2352 block = xfs_btree_get_block(cur, level, &bp);
2353 #ifdef DEBUG
2354 error = xfs_btree_check_block(cur, block, level, bp);
2355 if (error)
2356 return error;
2357 #endif
2358 ptr = cur->bc_levels[level].ptr;
2359 kp = xfs_btree_key_addr(cur, ptr, block);
2360 xfs_btree_copy_keys(cur, kp, &key, 1);
2361 xfs_btree_log_keys(cur, bp, ptr, ptr);
2362 }
2363
2364 return 0;
2365 }
2366
2367 /*
2368 * Update the record referred to by cur to the value in the
2369 * given record. This either works (return 0) or gets an
2370 * EFSCORRUPTED error.
2371 */
2372 int
xfs_btree_update(struct xfs_btree_cur * cur,union xfs_btree_rec * rec)2373 xfs_btree_update(
2374 struct xfs_btree_cur *cur,
2375 union xfs_btree_rec *rec)
2376 {
2377 struct xfs_btree_block *block;
2378 struct xfs_buf *bp;
2379 int error;
2380 int ptr;
2381 union xfs_btree_rec *rp;
2382
2383 /* Pick up the current block. */
2384 block = xfs_btree_get_block(cur, 0, &bp);
2385
2386 #ifdef DEBUG
2387 error = xfs_btree_check_block(cur, block, 0, bp);
2388 if (error)
2389 goto error0;
2390 #endif
2391 /* Get the address of the rec to be updated. */
2392 ptr = cur->bc_levels[0].ptr;
2393 rp = xfs_btree_rec_addr(cur, ptr, block);
2394
2395 /* Fill in the new contents and log them. */
2396 xfs_btree_copy_recs(cur, rp, rec, 1);
2397 xfs_btree_log_recs(cur, bp, ptr, ptr);
2398
2399 /* Pass new key value up to our parent. */
2400 if (xfs_btree_needs_key_update(cur, ptr)) {
2401 error = xfs_btree_update_keys(cur, 0);
2402 if (error)
2403 goto error0;
2404 }
2405
2406 return 0;
2407
2408 error0:
2409 return error;
2410 }
2411
2412 /*
2413 * Move 1 record left from cur/level if possible.
2414 * Update cur to reflect the new path.
2415 */
2416 STATIC int /* error */
xfs_btree_lshift(struct xfs_btree_cur * cur,int level,int * stat)2417 xfs_btree_lshift(
2418 struct xfs_btree_cur *cur,
2419 int level,
2420 int *stat) /* success/failure */
2421 {
2422 struct xfs_buf *lbp; /* left buffer pointer */
2423 struct xfs_btree_block *left; /* left btree block */
2424 int lrecs; /* left record count */
2425 struct xfs_buf *rbp; /* right buffer pointer */
2426 struct xfs_btree_block *right; /* right btree block */
2427 struct xfs_btree_cur *tcur; /* temporary btree cursor */
2428 int rrecs; /* right record count */
2429 union xfs_btree_ptr lptr; /* left btree pointer */
2430 union xfs_btree_key *rkp = NULL; /* right btree key */
2431 union xfs_btree_ptr *rpp = NULL; /* right address pointer */
2432 union xfs_btree_rec *rrp = NULL; /* right record pointer */
2433 int error; /* error return value */
2434 int i;
2435
2436 if (xfs_btree_at_iroot(cur, level))
2437 goto out0;
2438
2439 /* Set up variables for this block as "right". */
2440 right = xfs_btree_get_block(cur, level, &rbp);
2441
2442 #ifdef DEBUG
2443 error = xfs_btree_check_block(cur, right, level, rbp);
2444 if (error)
2445 goto error0;
2446 #endif
2447
2448 /* If we've got no left sibling then we can't shift an entry left. */
2449 xfs_btree_get_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
2450 if (xfs_btree_ptr_is_null(cur, &lptr))
2451 goto out0;
2452
2453 /*
2454 * If the cursor entry is the one that would be moved, don't
2455 * do it... it's too complicated.
2456 */
2457 if (cur->bc_levels[level].ptr <= 1)
2458 goto out0;
2459
2460 /* Set up the left neighbor as "left". */
2461 error = xfs_btree_read_buf_block(cur, &lptr, 0, &left, &lbp);
2462 if (error)
2463 goto error0;
2464
2465 /* If it's full, it can't take another entry. */
2466 lrecs = xfs_btree_get_numrecs(left);
2467 if (lrecs == cur->bc_ops->get_maxrecs(cur, level))
2468 goto out0;
2469
2470 rrecs = xfs_btree_get_numrecs(right);
2471
2472 /*
2473 * We add one entry to the left side and remove one for the right side.
2474 * Account for it here, the changes will be updated on disk and logged
2475 * later.
2476 */
2477 lrecs++;
2478 rrecs--;
2479
2480 XFS_BTREE_STATS_INC(cur, lshift);
2481 XFS_BTREE_STATS_ADD(cur, moves, 1);
2482
2483 /*
2484 * If non-leaf, copy a key and a ptr to the left block.
2485 * Log the changes to the left block.
2486 */
2487 if (level > 0) {
2488 /* It's a non-leaf. Move keys and pointers. */
2489 union xfs_btree_key *lkp; /* left btree key */
2490 union xfs_btree_ptr *lpp; /* left address pointer */
2491
2492 lkp = xfs_btree_key_addr(cur, lrecs, left);
2493 rkp = xfs_btree_key_addr(cur, 1, right);
2494
2495 lpp = xfs_btree_ptr_addr(cur, lrecs, left);
2496 rpp = xfs_btree_ptr_addr(cur, 1, right);
2497
2498 error = xfs_btree_debug_check_ptr(cur, rpp, 0, level);
2499 if (error)
2500 goto error0;
2501
2502 xfs_btree_copy_keys(cur, lkp, rkp, 1);
2503 xfs_btree_copy_ptrs(cur, lpp, rpp, 1);
2504
2505 xfs_btree_log_keys(cur, lbp, lrecs, lrecs);
2506 xfs_btree_log_ptrs(cur, lbp, lrecs, lrecs);
2507
2508 ASSERT(cur->bc_ops->keys_inorder(cur,
2509 xfs_btree_key_addr(cur, lrecs - 1, left), lkp));
2510 } else {
2511 /* It's a leaf. Move records. */
2512 union xfs_btree_rec *lrp; /* left record pointer */
2513
2514 lrp = xfs_btree_rec_addr(cur, lrecs, left);
2515 rrp = xfs_btree_rec_addr(cur, 1, right);
2516
2517 xfs_btree_copy_recs(cur, lrp, rrp, 1);
2518 xfs_btree_log_recs(cur, lbp, lrecs, lrecs);
2519
2520 ASSERT(cur->bc_ops->recs_inorder(cur,
2521 xfs_btree_rec_addr(cur, lrecs - 1, left), lrp));
2522 }
2523
2524 xfs_btree_set_numrecs(left, lrecs);
2525 xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS);
2526
2527 xfs_btree_set_numrecs(right, rrecs);
2528 xfs_btree_log_block(cur, rbp, XFS_BB_NUMRECS);
2529
2530 /*
2531 * Slide the contents of right down one entry.
2532 */
2533 XFS_BTREE_STATS_ADD(cur, moves, rrecs - 1);
2534 if (level > 0) {
2535 /* It's a nonleaf. operate on keys and ptrs */
2536 for (i = 0; i < rrecs; i++) {
2537 error = xfs_btree_debug_check_ptr(cur, rpp, i + 1, level);
2538 if (error)
2539 goto error0;
2540 }
2541
2542 xfs_btree_shift_keys(cur,
2543 xfs_btree_key_addr(cur, 2, right),
2544 -1, rrecs);
2545 xfs_btree_shift_ptrs(cur,
2546 xfs_btree_ptr_addr(cur, 2, right),
2547 -1, rrecs);
2548
2549 xfs_btree_log_keys(cur, rbp, 1, rrecs);
2550 xfs_btree_log_ptrs(cur, rbp, 1, rrecs);
2551 } else {
2552 /* It's a leaf. operate on records */
2553 xfs_btree_shift_recs(cur,
2554 xfs_btree_rec_addr(cur, 2, right),
2555 -1, rrecs);
2556 xfs_btree_log_recs(cur, rbp, 1, rrecs);
2557 }
2558
2559 /*
2560 * Using a temporary cursor, update the parent key values of the
2561 * block on the left.
2562 */
2563 if (cur->bc_ops->geom_flags & XFS_BTGEO_OVERLAPPING) {
2564 error = xfs_btree_dup_cursor(cur, &tcur);
2565 if (error)
2566 goto error0;
2567 i = xfs_btree_firstrec(tcur, level);
2568 if (XFS_IS_CORRUPT(tcur->bc_mp, i != 1)) {
2569 xfs_btree_mark_sick(cur);
2570 error = -EFSCORRUPTED;
2571 goto error0;
2572 }
2573
2574 error = xfs_btree_decrement(tcur, level, &i);
2575 if (error)
2576 goto error1;
2577
2578 /* Update the parent high keys of the left block, if needed. */
2579 error = xfs_btree_update_keys(tcur, level);
2580 if (error)
2581 goto error1;
2582
2583 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
2584 }
2585
2586 /* Update the parent keys of the right block. */
2587 error = xfs_btree_update_keys(cur, level);
2588 if (error)
2589 goto error0;
2590
2591 /* Slide the cursor value left one. */
2592 cur->bc_levels[level].ptr--;
2593
2594 *stat = 1;
2595 return 0;
2596
2597 out0:
2598 *stat = 0;
2599 return 0;
2600
2601 error0:
2602 return error;
2603
2604 error1:
2605 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
2606 return error;
2607 }
2608
2609 /*
2610 * Move 1 record right from cur/level if possible.
2611 * Update cur to reflect the new path.
2612 */
2613 STATIC int /* error */
xfs_btree_rshift(struct xfs_btree_cur * cur,int level,int * stat)2614 xfs_btree_rshift(
2615 struct xfs_btree_cur *cur,
2616 int level,
2617 int *stat) /* success/failure */
2618 {
2619 struct xfs_buf *lbp; /* left buffer pointer */
2620 struct xfs_btree_block *left; /* left btree block */
2621 struct xfs_buf *rbp; /* right buffer pointer */
2622 struct xfs_btree_block *right; /* right btree block */
2623 struct xfs_btree_cur *tcur; /* temporary btree cursor */
2624 union xfs_btree_ptr rptr; /* right block pointer */
2625 union xfs_btree_key *rkp; /* right btree key */
2626 int rrecs; /* right record count */
2627 int lrecs; /* left record count */
2628 int error; /* error return value */
2629 int i; /* loop counter */
2630
2631 if (xfs_btree_at_iroot(cur, level))
2632 goto out0;
2633
2634 /* Set up variables for this block as "left". */
2635 left = xfs_btree_get_block(cur, level, &lbp);
2636
2637 #ifdef DEBUG
2638 error = xfs_btree_check_block(cur, left, level, lbp);
2639 if (error)
2640 goto error0;
2641 #endif
2642
2643 /* If we've got no right sibling then we can't shift an entry right. */
2644 xfs_btree_get_sibling(cur, left, &rptr, XFS_BB_RIGHTSIB);
2645 if (xfs_btree_ptr_is_null(cur, &rptr))
2646 goto out0;
2647
2648 /*
2649 * If the cursor entry is the one that would be moved, don't
2650 * do it... it's too complicated.
2651 */
2652 lrecs = xfs_btree_get_numrecs(left);
2653 if (cur->bc_levels[level].ptr >= lrecs)
2654 goto out0;
2655
2656 /* Set up the right neighbor as "right". */
2657 error = xfs_btree_read_buf_block(cur, &rptr, 0, &right, &rbp);
2658 if (error)
2659 goto error0;
2660
2661 /* If it's full, it can't take another entry. */
2662 rrecs = xfs_btree_get_numrecs(right);
2663 if (rrecs == cur->bc_ops->get_maxrecs(cur, level))
2664 goto out0;
2665
2666 XFS_BTREE_STATS_INC(cur, rshift);
2667 XFS_BTREE_STATS_ADD(cur, moves, rrecs);
2668
2669 /*
2670 * Make a hole at the start of the right neighbor block, then
2671 * copy the last left block entry to the hole.
2672 */
2673 if (level > 0) {
2674 /* It's a nonleaf. make a hole in the keys and ptrs */
2675 union xfs_btree_key *lkp;
2676 union xfs_btree_ptr *lpp;
2677 union xfs_btree_ptr *rpp;
2678
2679 lkp = xfs_btree_key_addr(cur, lrecs, left);
2680 lpp = xfs_btree_ptr_addr(cur, lrecs, left);
2681 rkp = xfs_btree_key_addr(cur, 1, right);
2682 rpp = xfs_btree_ptr_addr(cur, 1, right);
2683
2684 for (i = rrecs - 1; i >= 0; i--) {
2685 error = xfs_btree_debug_check_ptr(cur, rpp, i, level);
2686 if (error)
2687 goto error0;
2688 }
2689
2690 xfs_btree_shift_keys(cur, rkp, 1, rrecs);
2691 xfs_btree_shift_ptrs(cur, rpp, 1, rrecs);
2692
2693 error = xfs_btree_debug_check_ptr(cur, lpp, 0, level);
2694 if (error)
2695 goto error0;
2696
2697 /* Now put the new data in, and log it. */
2698 xfs_btree_copy_keys(cur, rkp, lkp, 1);
2699 xfs_btree_copy_ptrs(cur, rpp, lpp, 1);
2700
2701 xfs_btree_log_keys(cur, rbp, 1, rrecs + 1);
2702 xfs_btree_log_ptrs(cur, rbp, 1, rrecs + 1);
2703
2704 ASSERT(cur->bc_ops->keys_inorder(cur, rkp,
2705 xfs_btree_key_addr(cur, 2, right)));
2706 } else {
2707 /* It's a leaf. make a hole in the records */
2708 union xfs_btree_rec *lrp;
2709 union xfs_btree_rec *rrp;
2710
2711 lrp = xfs_btree_rec_addr(cur, lrecs, left);
2712 rrp = xfs_btree_rec_addr(cur, 1, right);
2713
2714 xfs_btree_shift_recs(cur, rrp, 1, rrecs);
2715
2716 /* Now put the new data in, and log it. */
2717 xfs_btree_copy_recs(cur, rrp, lrp, 1);
2718 xfs_btree_log_recs(cur, rbp, 1, rrecs + 1);
2719 }
2720
2721 /*
2722 * Decrement and log left's numrecs, bump and log right's numrecs.
2723 */
2724 xfs_btree_set_numrecs(left, --lrecs);
2725 xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS);
2726
2727 xfs_btree_set_numrecs(right, ++rrecs);
2728 xfs_btree_log_block(cur, rbp, XFS_BB_NUMRECS);
2729
2730 /*
2731 * Using a temporary cursor, update the parent key values of the
2732 * block on the right.
2733 */
2734 error = xfs_btree_dup_cursor(cur, &tcur);
2735 if (error)
2736 goto error0;
2737 i = xfs_btree_lastrec(tcur, level);
2738 if (XFS_IS_CORRUPT(tcur->bc_mp, i != 1)) {
2739 xfs_btree_mark_sick(cur);
2740 error = -EFSCORRUPTED;
2741 goto error0;
2742 }
2743
2744 error = xfs_btree_increment(tcur, level, &i);
2745 if (error)
2746 goto error1;
2747
2748 /* Update the parent high keys of the left block, if needed. */
2749 if (cur->bc_ops->geom_flags & XFS_BTGEO_OVERLAPPING) {
2750 error = xfs_btree_update_keys(cur, level);
2751 if (error)
2752 goto error1;
2753 }
2754
2755 /* Update the parent keys of the right block. */
2756 error = xfs_btree_update_keys(tcur, level);
2757 if (error)
2758 goto error1;
2759
2760 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
2761
2762 *stat = 1;
2763 return 0;
2764
2765 out0:
2766 *stat = 0;
2767 return 0;
2768
2769 error0:
2770 return error;
2771
2772 error1:
2773 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
2774 return error;
2775 }
2776
2777 static inline int
xfs_btree_alloc_block(struct xfs_btree_cur * cur,const union xfs_btree_ptr * hint_block,union xfs_btree_ptr * new_block,int * stat)2778 xfs_btree_alloc_block(
2779 struct xfs_btree_cur *cur,
2780 const union xfs_btree_ptr *hint_block,
2781 union xfs_btree_ptr *new_block,
2782 int *stat)
2783 {
2784 int error;
2785
2786 /*
2787 * Don't allow block allocation for a staging cursor, because staging
2788 * cursors do not support regular btree modifications.
2789 *
2790 * Bulk loading uses a separate callback to obtain new blocks from a
2791 * preallocated list, which prevents ENOSPC failures during loading.
2792 */
2793 if (unlikely(cur->bc_flags & XFS_BTREE_STAGING)) {
2794 ASSERT(0);
2795 return -EFSCORRUPTED;
2796 }
2797
2798 error = cur->bc_ops->alloc_block(cur, hint_block, new_block, stat);
2799 trace_xfs_btree_alloc_block(cur, new_block, *stat, error);
2800 return error;
2801 }
2802
2803 /*
2804 * Split cur/level block in half.
2805 * Return new block number and the key to its first
2806 * record (to be inserted into parent).
2807 */
2808 STATIC int /* error */
__xfs_btree_split(struct xfs_btree_cur * cur,int level,union xfs_btree_ptr * ptrp,union xfs_btree_key * key,struct xfs_btree_cur ** curp,int * stat)2809 __xfs_btree_split(
2810 struct xfs_btree_cur *cur,
2811 int level,
2812 union xfs_btree_ptr *ptrp,
2813 union xfs_btree_key *key,
2814 struct xfs_btree_cur **curp,
2815 int *stat) /* success/failure */
2816 {
2817 union xfs_btree_ptr lptr; /* left sibling block ptr */
2818 struct xfs_buf *lbp; /* left buffer pointer */
2819 struct xfs_btree_block *left; /* left btree block */
2820 union xfs_btree_ptr rptr; /* right sibling block ptr */
2821 struct xfs_buf *rbp; /* right buffer pointer */
2822 struct xfs_btree_block *right; /* right btree block */
2823 union xfs_btree_ptr rrptr; /* right-right sibling ptr */
2824 struct xfs_buf *rrbp; /* right-right buffer pointer */
2825 struct xfs_btree_block *rrblock; /* right-right btree block */
2826 int lrecs;
2827 int rrecs;
2828 int src_index;
2829 int error; /* error return value */
2830 int i;
2831
2832 XFS_BTREE_STATS_INC(cur, split);
2833
2834 /* Set up left block (current one). */
2835 left = xfs_btree_get_block(cur, level, &lbp);
2836
2837 #ifdef DEBUG
2838 error = xfs_btree_check_block(cur, left, level, lbp);
2839 if (error)
2840 goto error0;
2841 #endif
2842
2843 xfs_btree_buf_to_ptr(cur, lbp, &lptr);
2844
2845 /* Allocate the new block. If we can't do it, we're toast. Give up. */
2846 error = xfs_btree_alloc_block(cur, &lptr, &rptr, stat);
2847 if (error)
2848 goto error0;
2849 if (*stat == 0)
2850 goto out0;
2851 XFS_BTREE_STATS_INC(cur, alloc);
2852
2853 /* Set up the new block as "right". */
2854 error = xfs_btree_get_buf_block(cur, &rptr, &right, &rbp);
2855 if (error)
2856 goto error0;
2857
2858 /* Fill in the btree header for the new right block. */
2859 xfs_btree_init_block_cur(cur, rbp, xfs_btree_get_level(left), 0);
2860
2861 /*
2862 * Split the entries between the old and the new block evenly.
2863 * Make sure that if there's an odd number of entries now, that
2864 * each new block will have the same number of entries.
2865 */
2866 lrecs = xfs_btree_get_numrecs(left);
2867 rrecs = lrecs / 2;
2868 if ((lrecs & 1) && cur->bc_levels[level].ptr <= rrecs + 1)
2869 rrecs++;
2870 src_index = (lrecs - rrecs + 1);
2871
2872 XFS_BTREE_STATS_ADD(cur, moves, rrecs);
2873
2874 /* Adjust numrecs for the later get_*_keys() calls. */
2875 lrecs -= rrecs;
2876 xfs_btree_set_numrecs(left, lrecs);
2877 xfs_btree_set_numrecs(right, xfs_btree_get_numrecs(right) + rrecs);
2878
2879 /*
2880 * Copy btree block entries from the left block over to the
2881 * new block, the right. Update the right block and log the
2882 * changes.
2883 */
2884 if (level > 0) {
2885 /* It's a non-leaf. Move keys and pointers. */
2886 union xfs_btree_key *lkp; /* left btree key */
2887 union xfs_btree_ptr *lpp; /* left address pointer */
2888 union xfs_btree_key *rkp; /* right btree key */
2889 union xfs_btree_ptr *rpp; /* right address pointer */
2890
2891 lkp = xfs_btree_key_addr(cur, src_index, left);
2892 lpp = xfs_btree_ptr_addr(cur, src_index, left);
2893 rkp = xfs_btree_key_addr(cur, 1, right);
2894 rpp = xfs_btree_ptr_addr(cur, 1, right);
2895
2896 for (i = src_index; i < rrecs; i++) {
2897 error = xfs_btree_debug_check_ptr(cur, lpp, i, level);
2898 if (error)
2899 goto error0;
2900 }
2901
2902 /* Copy the keys & pointers to the new block. */
2903 xfs_btree_copy_keys(cur, rkp, lkp, rrecs);
2904 xfs_btree_copy_ptrs(cur, rpp, lpp, rrecs);
2905
2906 xfs_btree_log_keys(cur, rbp, 1, rrecs);
2907 xfs_btree_log_ptrs(cur, rbp, 1, rrecs);
2908
2909 /* Stash the keys of the new block for later insertion. */
2910 xfs_btree_get_node_keys(cur, right, key);
2911 } else {
2912 /* It's a leaf. Move records. */
2913 union xfs_btree_rec *lrp; /* left record pointer */
2914 union xfs_btree_rec *rrp; /* right record pointer */
2915
2916 lrp = xfs_btree_rec_addr(cur, src_index, left);
2917 rrp = xfs_btree_rec_addr(cur, 1, right);
2918
2919 /* Copy records to the new block. */
2920 xfs_btree_copy_recs(cur, rrp, lrp, rrecs);
2921 xfs_btree_log_recs(cur, rbp, 1, rrecs);
2922
2923 /* Stash the keys of the new block for later insertion. */
2924 xfs_btree_get_leaf_keys(cur, right, key);
2925 }
2926
2927 /*
2928 * Find the left block number by looking in the buffer.
2929 * Adjust sibling pointers.
2930 */
2931 xfs_btree_get_sibling(cur, left, &rrptr, XFS_BB_RIGHTSIB);
2932 xfs_btree_set_sibling(cur, right, &rrptr, XFS_BB_RIGHTSIB);
2933 xfs_btree_set_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
2934 xfs_btree_set_sibling(cur, left, &rptr, XFS_BB_RIGHTSIB);
2935
2936 xfs_btree_log_block(cur, rbp, XFS_BB_ALL_BITS);
2937 xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS | XFS_BB_RIGHTSIB);
2938
2939 /*
2940 * If there's a block to the new block's right, make that block
2941 * point back to right instead of to left.
2942 */
2943 if (!xfs_btree_ptr_is_null(cur, &rrptr)) {
2944 error = xfs_btree_read_buf_block(cur, &rrptr,
2945 0, &rrblock, &rrbp);
2946 if (error)
2947 goto error0;
2948 xfs_btree_set_sibling(cur, rrblock, &rptr, XFS_BB_LEFTSIB);
2949 xfs_btree_log_block(cur, rrbp, XFS_BB_LEFTSIB);
2950 }
2951
2952 /* Update the parent high keys of the left block, if needed. */
2953 if (cur->bc_ops->geom_flags & XFS_BTGEO_OVERLAPPING) {
2954 error = xfs_btree_update_keys(cur, level);
2955 if (error)
2956 goto error0;
2957 }
2958
2959 /*
2960 * If the cursor is really in the right block, move it there.
2961 * If it's just pointing past the last entry in left, then we'll
2962 * insert there, so don't change anything in that case.
2963 */
2964 if (cur->bc_levels[level].ptr > lrecs + 1) {
2965 xfs_btree_setbuf(cur, level, rbp);
2966 cur->bc_levels[level].ptr -= lrecs;
2967 }
2968 /*
2969 * If there are more levels, we'll need another cursor which refers
2970 * the right block, no matter where this cursor was.
2971 */
2972 if (level + 1 < cur->bc_nlevels) {
2973 error = xfs_btree_dup_cursor(cur, curp);
2974 if (error)
2975 goto error0;
2976 (*curp)->bc_levels[level + 1].ptr++;
2977 }
2978 *ptrp = rptr;
2979 *stat = 1;
2980 return 0;
2981 out0:
2982 *stat = 0;
2983 return 0;
2984
2985 error0:
2986 return error;
2987 }
2988
2989 #ifdef __KERNEL__
2990 struct xfs_btree_split_args {
2991 struct xfs_btree_cur *cur;
2992 int level;
2993 union xfs_btree_ptr *ptrp;
2994 union xfs_btree_key *key;
2995 struct xfs_btree_cur **curp;
2996 int *stat; /* success/failure */
2997 int result;
2998 bool kswapd; /* allocation in kswapd context */
2999 struct completion *done;
3000 struct work_struct work;
3001 };
3002
3003 /*
3004 * Stack switching interfaces for allocation
3005 */
3006 static void
xfs_btree_split_worker(struct work_struct * work)3007 xfs_btree_split_worker(
3008 struct work_struct *work)
3009 {
3010 struct xfs_btree_split_args *args = container_of(work,
3011 struct xfs_btree_split_args, work);
3012 unsigned long pflags;
3013 unsigned long new_pflags = 0;
3014
3015 /*
3016 * we are in a transaction context here, but may also be doing work
3017 * in kswapd context, and hence we may need to inherit that state
3018 * temporarily to ensure that we don't block waiting for memory reclaim
3019 * in any way.
3020 */
3021 if (args->kswapd)
3022 new_pflags |= PF_MEMALLOC | PF_KSWAPD;
3023
3024 current_set_flags_nested(&pflags, new_pflags);
3025 xfs_trans_set_context(args->cur->bc_tp);
3026
3027 args->result = __xfs_btree_split(args->cur, args->level, args->ptrp,
3028 args->key, args->curp, args->stat);
3029
3030 xfs_trans_clear_context(args->cur->bc_tp);
3031 current_restore_flags_nested(&pflags, new_pflags);
3032
3033 /*
3034 * Do not access args after complete() has run here. We don't own args
3035 * and the owner may run and free args before we return here.
3036 */
3037 complete(args->done);
3038
3039 }
3040
3041 /*
3042 * BMBT split requests often come in with little stack to work on so we push
3043 * them off to a worker thread so there is lots of stack to use. For the other
3044 * btree types, just call directly to avoid the context switch overhead here.
3045 *
3046 * Care must be taken here - the work queue rescuer thread introduces potential
3047 * AGF <> worker queue deadlocks if the BMBT block allocation has to lock new
3048 * AGFs to allocate blocks. A task being run by the rescuer could attempt to
3049 * lock an AGF that is already locked by a task queued to run by the rescuer,
3050 * resulting in an ABBA deadlock as the rescuer cannot run the lock holder to
3051 * release it until the current thread it is running gains the lock.
3052 *
3053 * To avoid this issue, we only ever queue BMBT splits that don't have an AGF
3054 * already locked to allocate from. The only place that doesn't hold an AGF
3055 * locked is unwritten extent conversion at IO completion, but that has already
3056 * been offloaded to a worker thread and hence has no stack consumption issues
3057 * we have to worry about.
3058 */
3059 STATIC int /* error */
xfs_btree_split(struct xfs_btree_cur * cur,int level,union xfs_btree_ptr * ptrp,union xfs_btree_key * key,struct xfs_btree_cur ** curp,int * stat)3060 xfs_btree_split(
3061 struct xfs_btree_cur *cur,
3062 int level,
3063 union xfs_btree_ptr *ptrp,
3064 union xfs_btree_key *key,
3065 struct xfs_btree_cur **curp,
3066 int *stat) /* success/failure */
3067 {
3068 struct xfs_btree_split_args args;
3069 DECLARE_COMPLETION_ONSTACK(done);
3070
3071 if (!xfs_btree_is_bmap(cur->bc_ops) ||
3072 cur->bc_tp->t_highest_agno == NULLAGNUMBER)
3073 return __xfs_btree_split(cur, level, ptrp, key, curp, stat);
3074
3075 args.cur = cur;
3076 args.level = level;
3077 args.ptrp = ptrp;
3078 args.key = key;
3079 args.curp = curp;
3080 args.stat = stat;
3081 args.done = &done;
3082 args.kswapd = current_is_kswapd();
3083 INIT_WORK_ONSTACK(&args.work, xfs_btree_split_worker);
3084 queue_work(xfs_alloc_wq, &args.work);
3085 wait_for_completion(&done);
3086 destroy_work_on_stack(&args.work);
3087 return args.result;
3088 }
3089 #else
3090 #define xfs_btree_split __xfs_btree_split
3091 #endif /* __KERNEL__ */
3092
3093 /*
3094 * Copy the old inode root contents into a real block and make the
3095 * broot point to it.
3096 */
3097 int /* error */
xfs_btree_new_iroot(struct xfs_btree_cur * cur,int * logflags,int * stat)3098 xfs_btree_new_iroot(
3099 struct xfs_btree_cur *cur, /* btree cursor */
3100 int *logflags, /* logging flags for inode */
3101 int *stat) /* return status - 0 fail */
3102 {
3103 struct xfs_buf *cbp; /* buffer for cblock */
3104 struct xfs_btree_block *block; /* btree block */
3105 struct xfs_btree_block *cblock; /* child btree block */
3106 union xfs_btree_key *ckp; /* child key pointer */
3107 union xfs_btree_ptr *cpp; /* child ptr pointer */
3108 union xfs_btree_key *kp; /* pointer to btree key */
3109 union xfs_btree_ptr *pp; /* pointer to block addr */
3110 union xfs_btree_ptr nptr; /* new block addr */
3111 int level; /* btree level */
3112 int error; /* error return code */
3113 int i; /* loop counter */
3114
3115 XFS_BTREE_STATS_INC(cur, newroot);
3116
3117 ASSERT(cur->bc_ops->type == XFS_BTREE_TYPE_INODE);
3118
3119 level = cur->bc_nlevels - 1;
3120
3121 block = xfs_btree_get_iroot(cur);
3122 pp = xfs_btree_ptr_addr(cur, 1, block);
3123
3124 /* Allocate the new block. If we can't do it, we're toast. Give up. */
3125 error = xfs_btree_alloc_block(cur, pp, &nptr, stat);
3126 if (error)
3127 goto error0;
3128 if (*stat == 0)
3129 return 0;
3130
3131 XFS_BTREE_STATS_INC(cur, alloc);
3132
3133 /* Copy the root into a real block. */
3134 error = xfs_btree_get_buf_block(cur, &nptr, &cblock, &cbp);
3135 if (error)
3136 goto error0;
3137
3138 /*
3139 * we can't just memcpy() the root in for CRC enabled btree blocks.
3140 * In that case have to also ensure the blkno remains correct
3141 */
3142 memcpy(cblock, block, xfs_btree_block_len(cur));
3143 if (xfs_has_crc(cur->bc_mp)) {
3144 __be64 bno = cpu_to_be64(xfs_buf_daddr(cbp));
3145 if (cur->bc_ops->ptr_len == XFS_BTREE_LONG_PTR_LEN)
3146 cblock->bb_u.l.bb_blkno = bno;
3147 else
3148 cblock->bb_u.s.bb_blkno = bno;
3149 }
3150
3151 be16_add_cpu(&block->bb_level, 1);
3152 xfs_btree_set_numrecs(block, 1);
3153 cur->bc_nlevels++;
3154 ASSERT(cur->bc_nlevels <= cur->bc_maxlevels);
3155 cur->bc_levels[level + 1].ptr = 1;
3156
3157 kp = xfs_btree_key_addr(cur, 1, block);
3158 ckp = xfs_btree_key_addr(cur, 1, cblock);
3159 xfs_btree_copy_keys(cur, ckp, kp, xfs_btree_get_numrecs(cblock));
3160
3161 cpp = xfs_btree_ptr_addr(cur, 1, cblock);
3162 for (i = 0; i < be16_to_cpu(cblock->bb_numrecs); i++) {
3163 error = xfs_btree_debug_check_ptr(cur, pp, i, level);
3164 if (error)
3165 goto error0;
3166 }
3167
3168 xfs_btree_copy_ptrs(cur, cpp, pp, xfs_btree_get_numrecs(cblock));
3169
3170 error = xfs_btree_debug_check_ptr(cur, &nptr, 0, level);
3171 if (error)
3172 goto error0;
3173
3174 xfs_btree_copy_ptrs(cur, pp, &nptr, 1);
3175
3176 xfs_iroot_realloc(cur->bc_ino.ip,
3177 1 - xfs_btree_get_numrecs(cblock),
3178 cur->bc_ino.whichfork);
3179
3180 xfs_btree_setbuf(cur, level, cbp);
3181
3182 /*
3183 * Do all this logging at the end so that
3184 * the root is at the right level.
3185 */
3186 xfs_btree_log_block(cur, cbp, XFS_BB_ALL_BITS);
3187 xfs_btree_log_keys(cur, cbp, 1, be16_to_cpu(cblock->bb_numrecs));
3188 xfs_btree_log_ptrs(cur, cbp, 1, be16_to_cpu(cblock->bb_numrecs));
3189
3190 *logflags |=
3191 XFS_ILOG_CORE | xfs_ilog_fbroot(cur->bc_ino.whichfork);
3192 *stat = 1;
3193 return 0;
3194 error0:
3195 return error;
3196 }
3197
3198 static void
xfs_btree_set_root(struct xfs_btree_cur * cur,const union xfs_btree_ptr * ptr,int inc)3199 xfs_btree_set_root(
3200 struct xfs_btree_cur *cur,
3201 const union xfs_btree_ptr *ptr,
3202 int inc)
3203 {
3204 if (cur->bc_flags & XFS_BTREE_STAGING) {
3205 /* Update the btree root information for a per-AG fake root. */
3206 cur->bc_ag.afake->af_root = be32_to_cpu(ptr->s);
3207 cur->bc_ag.afake->af_levels += inc;
3208 } else {
3209 cur->bc_ops->set_root(cur, ptr, inc);
3210 }
3211 }
3212
3213 /*
3214 * Allocate a new root block, fill it in.
3215 */
3216 STATIC int /* error */
xfs_btree_new_root(struct xfs_btree_cur * cur,int * stat)3217 xfs_btree_new_root(
3218 struct xfs_btree_cur *cur, /* btree cursor */
3219 int *stat) /* success/failure */
3220 {
3221 struct xfs_btree_block *block; /* one half of the old root block */
3222 struct xfs_buf *bp; /* buffer containing block */
3223 int error; /* error return value */
3224 struct xfs_buf *lbp; /* left buffer pointer */
3225 struct xfs_btree_block *left; /* left btree block */
3226 struct xfs_buf *nbp; /* new (root) buffer */
3227 struct xfs_btree_block *new; /* new (root) btree block */
3228 int nptr; /* new value for key index, 1 or 2 */
3229 struct xfs_buf *rbp; /* right buffer pointer */
3230 struct xfs_btree_block *right; /* right btree block */
3231 union xfs_btree_ptr rptr;
3232 union xfs_btree_ptr lptr;
3233
3234 XFS_BTREE_STATS_INC(cur, newroot);
3235
3236 /* initialise our start point from the cursor */
3237 xfs_btree_init_ptr_from_cur(cur, &rptr);
3238
3239 /* Allocate the new block. If we can't do it, we're toast. Give up. */
3240 error = xfs_btree_alloc_block(cur, &rptr, &lptr, stat);
3241 if (error)
3242 goto error0;
3243 if (*stat == 0)
3244 goto out0;
3245 XFS_BTREE_STATS_INC(cur, alloc);
3246
3247 /* Set up the new block. */
3248 error = xfs_btree_get_buf_block(cur, &lptr, &new, &nbp);
3249 if (error)
3250 goto error0;
3251
3252 /* Set the root in the holding structure increasing the level by 1. */
3253 xfs_btree_set_root(cur, &lptr, 1);
3254
3255 /*
3256 * At the previous root level there are now two blocks: the old root,
3257 * and the new block generated when it was split. We don't know which
3258 * one the cursor is pointing at, so we set up variables "left" and
3259 * "right" for each case.
3260 */
3261 block = xfs_btree_get_block(cur, cur->bc_nlevels - 1, &bp);
3262
3263 #ifdef DEBUG
3264 error = xfs_btree_check_block(cur, block, cur->bc_nlevels - 1, bp);
3265 if (error)
3266 goto error0;
3267 #endif
3268
3269 xfs_btree_get_sibling(cur, block, &rptr, XFS_BB_RIGHTSIB);
3270 if (!xfs_btree_ptr_is_null(cur, &rptr)) {
3271 /* Our block is left, pick up the right block. */
3272 lbp = bp;
3273 xfs_btree_buf_to_ptr(cur, lbp, &lptr);
3274 left = block;
3275 error = xfs_btree_read_buf_block(cur, &rptr, 0, &right, &rbp);
3276 if (error)
3277 goto error0;
3278 bp = rbp;
3279 nptr = 1;
3280 } else {
3281 /* Our block is right, pick up the left block. */
3282 rbp = bp;
3283 xfs_btree_buf_to_ptr(cur, rbp, &rptr);
3284 right = block;
3285 xfs_btree_get_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
3286 error = xfs_btree_read_buf_block(cur, &lptr, 0, &left, &lbp);
3287 if (error)
3288 goto error0;
3289 bp = lbp;
3290 nptr = 2;
3291 }
3292
3293 /* Fill in the new block's btree header and log it. */
3294 xfs_btree_init_block_cur(cur, nbp, cur->bc_nlevels, 2);
3295 xfs_btree_log_block(cur, nbp, XFS_BB_ALL_BITS);
3296 ASSERT(!xfs_btree_ptr_is_null(cur, &lptr) &&
3297 !xfs_btree_ptr_is_null(cur, &rptr));
3298
3299 /* Fill in the key data in the new root. */
3300 if (xfs_btree_get_level(left) > 0) {
3301 /*
3302 * Get the keys for the left block's keys and put them directly
3303 * in the parent block. Do the same for the right block.
3304 */
3305 xfs_btree_get_node_keys(cur, left,
3306 xfs_btree_key_addr(cur, 1, new));
3307 xfs_btree_get_node_keys(cur, right,
3308 xfs_btree_key_addr(cur, 2, new));
3309 } else {
3310 /*
3311 * Get the keys for the left block's records and put them
3312 * directly in the parent block. Do the same for the right
3313 * block.
3314 */
3315 xfs_btree_get_leaf_keys(cur, left,
3316 xfs_btree_key_addr(cur, 1, new));
3317 xfs_btree_get_leaf_keys(cur, right,
3318 xfs_btree_key_addr(cur, 2, new));
3319 }
3320 xfs_btree_log_keys(cur, nbp, 1, 2);
3321
3322 /* Fill in the pointer data in the new root. */
3323 xfs_btree_copy_ptrs(cur,
3324 xfs_btree_ptr_addr(cur, 1, new), &lptr, 1);
3325 xfs_btree_copy_ptrs(cur,
3326 xfs_btree_ptr_addr(cur, 2, new), &rptr, 1);
3327 xfs_btree_log_ptrs(cur, nbp, 1, 2);
3328
3329 /* Fix up the cursor. */
3330 xfs_btree_setbuf(cur, cur->bc_nlevels, nbp);
3331 cur->bc_levels[cur->bc_nlevels].ptr = nptr;
3332 cur->bc_nlevels++;
3333 ASSERT(cur->bc_nlevels <= cur->bc_maxlevels);
3334 *stat = 1;
3335 return 0;
3336 error0:
3337 return error;
3338 out0:
3339 *stat = 0;
3340 return 0;
3341 }
3342
3343 STATIC int
xfs_btree_make_block_unfull(struct xfs_btree_cur * cur,int level,int numrecs,int * oindex,int * index,union xfs_btree_ptr * nptr,struct xfs_btree_cur ** ncur,union xfs_btree_key * key,int * stat)3344 xfs_btree_make_block_unfull(
3345 struct xfs_btree_cur *cur, /* btree cursor */
3346 int level, /* btree level */
3347 int numrecs,/* # of recs in block */
3348 int *oindex,/* old tree index */
3349 int *index, /* new tree index */
3350 union xfs_btree_ptr *nptr, /* new btree ptr */
3351 struct xfs_btree_cur **ncur, /* new btree cursor */
3352 union xfs_btree_key *key, /* key of new block */
3353 int *stat)
3354 {
3355 int error = 0;
3356
3357 if (xfs_btree_at_iroot(cur, level)) {
3358 struct xfs_inode *ip = cur->bc_ino.ip;
3359
3360 if (numrecs < cur->bc_ops->get_dmaxrecs(cur, level)) {
3361 /* A root block that can be made bigger. */
3362 xfs_iroot_realloc(ip, 1, cur->bc_ino.whichfork);
3363 *stat = 1;
3364 } else {
3365 /* A root block that needs replacing */
3366 int logflags = 0;
3367
3368 error = xfs_btree_new_iroot(cur, &logflags, stat);
3369 if (error || *stat == 0)
3370 return error;
3371
3372 xfs_trans_log_inode(cur->bc_tp, ip, logflags);
3373 }
3374
3375 return 0;
3376 }
3377
3378 /* First, try shifting an entry to the right neighbor. */
3379 error = xfs_btree_rshift(cur, level, stat);
3380 if (error || *stat)
3381 return error;
3382
3383 /* Next, try shifting an entry to the left neighbor. */
3384 error = xfs_btree_lshift(cur, level, stat);
3385 if (error)
3386 return error;
3387
3388 if (*stat) {
3389 *oindex = *index = cur->bc_levels[level].ptr;
3390 return 0;
3391 }
3392
3393 /*
3394 * Next, try splitting the current block in half.
3395 *
3396 * If this works we have to re-set our variables because we
3397 * could be in a different block now.
3398 */
3399 error = xfs_btree_split(cur, level, nptr, key, ncur, stat);
3400 if (error || *stat == 0)
3401 return error;
3402
3403
3404 *index = cur->bc_levels[level].ptr;
3405 return 0;
3406 }
3407
3408 /*
3409 * Insert one record/level. Return information to the caller
3410 * allowing the next level up to proceed if necessary.
3411 */
3412 STATIC int
xfs_btree_insrec(struct xfs_btree_cur * cur,int level,union xfs_btree_ptr * ptrp,union xfs_btree_rec * rec,union xfs_btree_key * key,struct xfs_btree_cur ** curp,int * stat)3413 xfs_btree_insrec(
3414 struct xfs_btree_cur *cur, /* btree cursor */
3415 int level, /* level to insert record at */
3416 union xfs_btree_ptr *ptrp, /* i/o: block number inserted */
3417 union xfs_btree_rec *rec, /* record to insert */
3418 union xfs_btree_key *key, /* i/o: block key for ptrp */
3419 struct xfs_btree_cur **curp, /* output: new cursor replacing cur */
3420 int *stat) /* success/failure */
3421 {
3422 struct xfs_btree_block *block; /* btree block */
3423 struct xfs_buf *bp; /* buffer for block */
3424 union xfs_btree_ptr nptr; /* new block ptr */
3425 struct xfs_btree_cur *ncur = NULL; /* new btree cursor */
3426 union xfs_btree_key nkey; /* new block key */
3427 union xfs_btree_key *lkey;
3428 int optr; /* old key/record index */
3429 int ptr; /* key/record index */
3430 int numrecs;/* number of records */
3431 int error; /* error return value */
3432 int i;
3433 xfs_daddr_t old_bn;
3434
3435 ncur = NULL;
3436 lkey = &nkey;
3437
3438 /*
3439 * If we have an external root pointer, and we've made it to the
3440 * root level, allocate a new root block and we're done.
3441 */
3442 if (cur->bc_ops->type != XFS_BTREE_TYPE_INODE &&
3443 level >= cur->bc_nlevels) {
3444 error = xfs_btree_new_root(cur, stat);
3445 xfs_btree_set_ptr_null(cur, ptrp);
3446
3447 return error;
3448 }
3449
3450 /* If we're off the left edge, return failure. */
3451 ptr = cur->bc_levels[level].ptr;
3452 if (ptr == 0) {
3453 *stat = 0;
3454 return 0;
3455 }
3456
3457 optr = ptr;
3458
3459 XFS_BTREE_STATS_INC(cur, insrec);
3460
3461 /* Get pointers to the btree buffer and block. */
3462 block = xfs_btree_get_block(cur, level, &bp);
3463 old_bn = bp ? xfs_buf_daddr(bp) : XFS_BUF_DADDR_NULL;
3464 numrecs = xfs_btree_get_numrecs(block);
3465
3466 #ifdef DEBUG
3467 error = xfs_btree_check_block(cur, block, level, bp);
3468 if (error)
3469 goto error0;
3470
3471 /* Check that the new entry is being inserted in the right place. */
3472 if (ptr <= numrecs) {
3473 if (level == 0) {
3474 ASSERT(cur->bc_ops->recs_inorder(cur, rec,
3475 xfs_btree_rec_addr(cur, ptr, block)));
3476 } else {
3477 ASSERT(cur->bc_ops->keys_inorder(cur, key,
3478 xfs_btree_key_addr(cur, ptr, block)));
3479 }
3480 }
3481 #endif
3482
3483 /*
3484 * If the block is full, we can't insert the new entry until we
3485 * make the block un-full.
3486 */
3487 xfs_btree_set_ptr_null(cur, &nptr);
3488 if (numrecs == cur->bc_ops->get_maxrecs(cur, level)) {
3489 error = xfs_btree_make_block_unfull(cur, level, numrecs,
3490 &optr, &ptr, &nptr, &ncur, lkey, stat);
3491 if (error || *stat == 0)
3492 goto error0;
3493 }
3494
3495 /*
3496 * The current block may have changed if the block was
3497 * previously full and we have just made space in it.
3498 */
3499 block = xfs_btree_get_block(cur, level, &bp);
3500 numrecs = xfs_btree_get_numrecs(block);
3501
3502 #ifdef DEBUG
3503 error = xfs_btree_check_block(cur, block, level, bp);
3504 if (error)
3505 goto error0;
3506 #endif
3507
3508 /*
3509 * At this point we know there's room for our new entry in the block
3510 * we're pointing at.
3511 */
3512 XFS_BTREE_STATS_ADD(cur, moves, numrecs - ptr + 1);
3513
3514 if (level > 0) {
3515 /* It's a nonleaf. make a hole in the keys and ptrs */
3516 union xfs_btree_key *kp;
3517 union xfs_btree_ptr *pp;
3518
3519 kp = xfs_btree_key_addr(cur, ptr, block);
3520 pp = xfs_btree_ptr_addr(cur, ptr, block);
3521
3522 for (i = numrecs - ptr; i >= 0; i--) {
3523 error = xfs_btree_debug_check_ptr(cur, pp, i, level);
3524 if (error)
3525 goto error0;
3526 }
3527
3528 xfs_btree_shift_keys(cur, kp, 1, numrecs - ptr + 1);
3529 xfs_btree_shift_ptrs(cur, pp, 1, numrecs - ptr + 1);
3530
3531 error = xfs_btree_debug_check_ptr(cur, ptrp, 0, level);
3532 if (error)
3533 goto error0;
3534
3535 /* Now put the new data in, bump numrecs and log it. */
3536 xfs_btree_copy_keys(cur, kp, key, 1);
3537 xfs_btree_copy_ptrs(cur, pp, ptrp, 1);
3538 numrecs++;
3539 xfs_btree_set_numrecs(block, numrecs);
3540 xfs_btree_log_ptrs(cur, bp, ptr, numrecs);
3541 xfs_btree_log_keys(cur, bp, ptr, numrecs);
3542 #ifdef DEBUG
3543 if (ptr < numrecs) {
3544 ASSERT(cur->bc_ops->keys_inorder(cur, kp,
3545 xfs_btree_key_addr(cur, ptr + 1, block)));
3546 }
3547 #endif
3548 } else {
3549 /* It's a leaf. make a hole in the records */
3550 union xfs_btree_rec *rp;
3551
3552 rp = xfs_btree_rec_addr(cur, ptr, block);
3553
3554 xfs_btree_shift_recs(cur, rp, 1, numrecs - ptr + 1);
3555
3556 /* Now put the new data in, bump numrecs and log it. */
3557 xfs_btree_copy_recs(cur, rp, rec, 1);
3558 xfs_btree_set_numrecs(block, ++numrecs);
3559 xfs_btree_log_recs(cur, bp, ptr, numrecs);
3560 #ifdef DEBUG
3561 if (ptr < numrecs) {
3562 ASSERT(cur->bc_ops->recs_inorder(cur, rp,
3563 xfs_btree_rec_addr(cur, ptr + 1, block)));
3564 }
3565 #endif
3566 }
3567
3568 /* Log the new number of records in the btree header. */
3569 xfs_btree_log_block(cur, bp, XFS_BB_NUMRECS);
3570
3571 /*
3572 * If we just inserted into a new tree block, we have to
3573 * recalculate nkey here because nkey is out of date.
3574 *
3575 * Otherwise we're just updating an existing block (having shoved
3576 * some records into the new tree block), so use the regular key
3577 * update mechanism.
3578 */
3579 if (bp && xfs_buf_daddr(bp) != old_bn) {
3580 xfs_btree_get_keys(cur, block, lkey);
3581 } else if (xfs_btree_needs_key_update(cur, optr)) {
3582 error = xfs_btree_update_keys(cur, level);
3583 if (error)
3584 goto error0;
3585 }
3586
3587 /*
3588 * Return the new block number, if any.
3589 * If there is one, give back a record value and a cursor too.
3590 */
3591 *ptrp = nptr;
3592 if (!xfs_btree_ptr_is_null(cur, &nptr)) {
3593 xfs_btree_copy_keys(cur, key, lkey, 1);
3594 *curp = ncur;
3595 }
3596
3597 *stat = 1;
3598 return 0;
3599
3600 error0:
3601 if (ncur)
3602 xfs_btree_del_cursor(ncur, error);
3603 return error;
3604 }
3605
3606 /*
3607 * Insert the record at the point referenced by cur.
3608 *
3609 * A multi-level split of the tree on insert will invalidate the original
3610 * cursor. All callers of this function should assume that the cursor is
3611 * no longer valid and revalidate it.
3612 */
3613 int
xfs_btree_insert(struct xfs_btree_cur * cur,int * stat)3614 xfs_btree_insert(
3615 struct xfs_btree_cur *cur,
3616 int *stat)
3617 {
3618 int error; /* error return value */
3619 int i; /* result value, 0 for failure */
3620 int level; /* current level number in btree */
3621 union xfs_btree_ptr nptr; /* new block number (split result) */
3622 struct xfs_btree_cur *ncur; /* new cursor (split result) */
3623 struct xfs_btree_cur *pcur; /* previous level's cursor */
3624 union xfs_btree_key bkey; /* key of block to insert */
3625 union xfs_btree_key *key;
3626 union xfs_btree_rec rec; /* record to insert */
3627
3628 level = 0;
3629 ncur = NULL;
3630 pcur = cur;
3631 key = &bkey;
3632
3633 xfs_btree_set_ptr_null(cur, &nptr);
3634
3635 /* Make a key out of the record data to be inserted, and save it. */
3636 cur->bc_ops->init_rec_from_cur(cur, &rec);
3637 cur->bc_ops->init_key_from_rec(key, &rec);
3638
3639 /*
3640 * Loop going up the tree, starting at the leaf level.
3641 * Stop when we don't get a split block, that must mean that
3642 * the insert is finished with this level.
3643 */
3644 do {
3645 /*
3646 * Insert nrec/nptr into this level of the tree.
3647 * Note if we fail, nptr will be null.
3648 */
3649 error = xfs_btree_insrec(pcur, level, &nptr, &rec, key,
3650 &ncur, &i);
3651 if (error) {
3652 if (pcur != cur)
3653 xfs_btree_del_cursor(pcur, XFS_BTREE_ERROR);
3654 goto error0;
3655 }
3656
3657 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
3658 xfs_btree_mark_sick(cur);
3659 error = -EFSCORRUPTED;
3660 goto error0;
3661 }
3662 level++;
3663
3664 /*
3665 * See if the cursor we just used is trash.
3666 * Can't trash the caller's cursor, but otherwise we should
3667 * if ncur is a new cursor or we're about to be done.
3668 */
3669 if (pcur != cur &&
3670 (ncur || xfs_btree_ptr_is_null(cur, &nptr))) {
3671 /* Save the state from the cursor before we trash it */
3672 if (cur->bc_ops->update_cursor &&
3673 !(cur->bc_flags & XFS_BTREE_STAGING))
3674 cur->bc_ops->update_cursor(pcur, cur);
3675 cur->bc_nlevels = pcur->bc_nlevels;
3676 xfs_btree_del_cursor(pcur, XFS_BTREE_NOERROR);
3677 }
3678 /* If we got a new cursor, switch to it. */
3679 if (ncur) {
3680 pcur = ncur;
3681 ncur = NULL;
3682 }
3683 } while (!xfs_btree_ptr_is_null(cur, &nptr));
3684
3685 *stat = i;
3686 return 0;
3687 error0:
3688 return error;
3689 }
3690
3691 /*
3692 * Try to merge a non-leaf block back into the inode root.
3693 *
3694 * Note: the killroot names comes from the fact that we're effectively
3695 * killing the old root block. But because we can't just delete the
3696 * inode we have to copy the single block it was pointing to into the
3697 * inode.
3698 */
3699 STATIC int
xfs_btree_kill_iroot(struct xfs_btree_cur * cur)3700 xfs_btree_kill_iroot(
3701 struct xfs_btree_cur *cur)
3702 {
3703 int whichfork = cur->bc_ino.whichfork;
3704 struct xfs_inode *ip = cur->bc_ino.ip;
3705 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
3706 struct xfs_btree_block *block;
3707 struct xfs_btree_block *cblock;
3708 union xfs_btree_key *kp;
3709 union xfs_btree_key *ckp;
3710 union xfs_btree_ptr *pp;
3711 union xfs_btree_ptr *cpp;
3712 struct xfs_buf *cbp;
3713 int level;
3714 int index;
3715 int numrecs;
3716 int error;
3717 #ifdef DEBUG
3718 union xfs_btree_ptr ptr;
3719 #endif
3720 int i;
3721
3722 ASSERT(cur->bc_ops->type == XFS_BTREE_TYPE_INODE);
3723 ASSERT(cur->bc_nlevels > 1);
3724
3725 /*
3726 * Don't deal with the root block needs to be a leaf case.
3727 * We're just going to turn the thing back into extents anyway.
3728 */
3729 level = cur->bc_nlevels - 1;
3730 if (level == 1)
3731 goto out0;
3732
3733 /*
3734 * Give up if the root has multiple children.
3735 */
3736 block = xfs_btree_get_iroot(cur);
3737 if (xfs_btree_get_numrecs(block) != 1)
3738 goto out0;
3739
3740 cblock = xfs_btree_get_block(cur, level - 1, &cbp);
3741 numrecs = xfs_btree_get_numrecs(cblock);
3742
3743 /*
3744 * Only do this if the next level will fit.
3745 * Then the data must be copied up to the inode,
3746 * instead of freeing the root you free the next level.
3747 */
3748 if (numrecs > cur->bc_ops->get_dmaxrecs(cur, level))
3749 goto out0;
3750
3751 XFS_BTREE_STATS_INC(cur, killroot);
3752
3753 #ifdef DEBUG
3754 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_LEFTSIB);
3755 ASSERT(xfs_btree_ptr_is_null(cur, &ptr));
3756 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
3757 ASSERT(xfs_btree_ptr_is_null(cur, &ptr));
3758 #endif
3759
3760 index = numrecs - cur->bc_ops->get_maxrecs(cur, level);
3761 if (index) {
3762 xfs_iroot_realloc(cur->bc_ino.ip, index,
3763 cur->bc_ino.whichfork);
3764 block = ifp->if_broot;
3765 }
3766
3767 be16_add_cpu(&block->bb_numrecs, index);
3768 ASSERT(block->bb_numrecs == cblock->bb_numrecs);
3769
3770 kp = xfs_btree_key_addr(cur, 1, block);
3771 ckp = xfs_btree_key_addr(cur, 1, cblock);
3772 xfs_btree_copy_keys(cur, kp, ckp, numrecs);
3773
3774 pp = xfs_btree_ptr_addr(cur, 1, block);
3775 cpp = xfs_btree_ptr_addr(cur, 1, cblock);
3776
3777 for (i = 0; i < numrecs; i++) {
3778 error = xfs_btree_debug_check_ptr(cur, cpp, i, level - 1);
3779 if (error)
3780 return error;
3781 }
3782
3783 xfs_btree_copy_ptrs(cur, pp, cpp, numrecs);
3784
3785 error = xfs_btree_free_block(cur, cbp);
3786 if (error)
3787 return error;
3788
3789 cur->bc_levels[level - 1].bp = NULL;
3790 be16_add_cpu(&block->bb_level, -1);
3791 xfs_trans_log_inode(cur->bc_tp, ip,
3792 XFS_ILOG_CORE | xfs_ilog_fbroot(cur->bc_ino.whichfork));
3793 cur->bc_nlevels--;
3794 out0:
3795 return 0;
3796 }
3797
3798 /*
3799 * Kill the current root node, and replace it with it's only child node.
3800 */
3801 STATIC int
xfs_btree_kill_root(struct xfs_btree_cur * cur,struct xfs_buf * bp,int level,union xfs_btree_ptr * newroot)3802 xfs_btree_kill_root(
3803 struct xfs_btree_cur *cur,
3804 struct xfs_buf *bp,
3805 int level,
3806 union xfs_btree_ptr *newroot)
3807 {
3808 int error;
3809
3810 XFS_BTREE_STATS_INC(cur, killroot);
3811
3812 /*
3813 * Update the root pointer, decreasing the level by 1 and then
3814 * free the old root.
3815 */
3816 xfs_btree_set_root(cur, newroot, -1);
3817
3818 error = xfs_btree_free_block(cur, bp);
3819 if (error)
3820 return error;
3821
3822 cur->bc_levels[level].bp = NULL;
3823 cur->bc_levels[level].ra = 0;
3824 cur->bc_nlevels--;
3825
3826 return 0;
3827 }
3828
3829 STATIC int
xfs_btree_dec_cursor(struct xfs_btree_cur * cur,int level,int * stat)3830 xfs_btree_dec_cursor(
3831 struct xfs_btree_cur *cur,
3832 int level,
3833 int *stat)
3834 {
3835 int error;
3836 int i;
3837
3838 if (level > 0) {
3839 error = xfs_btree_decrement(cur, level, &i);
3840 if (error)
3841 return error;
3842 }
3843
3844 *stat = 1;
3845 return 0;
3846 }
3847
3848 /*
3849 * Single level of the btree record deletion routine.
3850 * Delete record pointed to by cur/level.
3851 * Remove the record from its block then rebalance the tree.
3852 * Return 0 for error, 1 for done, 2 to go on to the next level.
3853 */
3854 STATIC int /* error */
xfs_btree_delrec(struct xfs_btree_cur * cur,int level,int * stat)3855 xfs_btree_delrec(
3856 struct xfs_btree_cur *cur, /* btree cursor */
3857 int level, /* level removing record from */
3858 int *stat) /* fail/done/go-on */
3859 {
3860 struct xfs_btree_block *block; /* btree block */
3861 union xfs_btree_ptr cptr; /* current block ptr */
3862 struct xfs_buf *bp; /* buffer for block */
3863 int error; /* error return value */
3864 int i; /* loop counter */
3865 union xfs_btree_ptr lptr; /* left sibling block ptr */
3866 struct xfs_buf *lbp; /* left buffer pointer */
3867 struct xfs_btree_block *left; /* left btree block */
3868 int lrecs = 0; /* left record count */
3869 int ptr; /* key/record index */
3870 union xfs_btree_ptr rptr; /* right sibling block ptr */
3871 struct xfs_buf *rbp; /* right buffer pointer */
3872 struct xfs_btree_block *right; /* right btree block */
3873 struct xfs_btree_block *rrblock; /* right-right btree block */
3874 struct xfs_buf *rrbp; /* right-right buffer pointer */
3875 int rrecs = 0; /* right record count */
3876 struct xfs_btree_cur *tcur; /* temporary btree cursor */
3877 int numrecs; /* temporary numrec count */
3878
3879 tcur = NULL;
3880
3881 /* Get the index of the entry being deleted, check for nothing there. */
3882 ptr = cur->bc_levels[level].ptr;
3883 if (ptr == 0) {
3884 *stat = 0;
3885 return 0;
3886 }
3887
3888 /* Get the buffer & block containing the record or key/ptr. */
3889 block = xfs_btree_get_block(cur, level, &bp);
3890 numrecs = xfs_btree_get_numrecs(block);
3891
3892 #ifdef DEBUG
3893 error = xfs_btree_check_block(cur, block, level, bp);
3894 if (error)
3895 goto error0;
3896 #endif
3897
3898 /* Fail if we're off the end of the block. */
3899 if (ptr > numrecs) {
3900 *stat = 0;
3901 return 0;
3902 }
3903
3904 XFS_BTREE_STATS_INC(cur, delrec);
3905 XFS_BTREE_STATS_ADD(cur, moves, numrecs - ptr);
3906
3907 /* Excise the entries being deleted. */
3908 if (level > 0) {
3909 /* It's a nonleaf. operate on keys and ptrs */
3910 union xfs_btree_key *lkp;
3911 union xfs_btree_ptr *lpp;
3912
3913 lkp = xfs_btree_key_addr(cur, ptr + 1, block);
3914 lpp = xfs_btree_ptr_addr(cur, ptr + 1, block);
3915
3916 for (i = 0; i < numrecs - ptr; i++) {
3917 error = xfs_btree_debug_check_ptr(cur, lpp, i, level);
3918 if (error)
3919 goto error0;
3920 }
3921
3922 if (ptr < numrecs) {
3923 xfs_btree_shift_keys(cur, lkp, -1, numrecs - ptr);
3924 xfs_btree_shift_ptrs(cur, lpp, -1, numrecs - ptr);
3925 xfs_btree_log_keys(cur, bp, ptr, numrecs - 1);
3926 xfs_btree_log_ptrs(cur, bp, ptr, numrecs - 1);
3927 }
3928 } else {
3929 /* It's a leaf. operate on records */
3930 if (ptr < numrecs) {
3931 xfs_btree_shift_recs(cur,
3932 xfs_btree_rec_addr(cur, ptr + 1, block),
3933 -1, numrecs - ptr);
3934 xfs_btree_log_recs(cur, bp, ptr, numrecs - 1);
3935 }
3936 }
3937
3938 /*
3939 * Decrement and log the number of entries in the block.
3940 */
3941 xfs_btree_set_numrecs(block, --numrecs);
3942 xfs_btree_log_block(cur, bp, XFS_BB_NUMRECS);
3943
3944 /*
3945 * We're at the root level. First, shrink the root block in-memory.
3946 * Try to get rid of the next level down. If we can't then there's
3947 * nothing left to do.
3948 */
3949 if (xfs_btree_at_iroot(cur, level)) {
3950 xfs_iroot_realloc(cur->bc_ino.ip, -1, cur->bc_ino.whichfork);
3951
3952 error = xfs_btree_kill_iroot(cur);
3953 if (error)
3954 goto error0;
3955
3956 error = xfs_btree_dec_cursor(cur, level, stat);
3957 if (error)
3958 goto error0;
3959 *stat = 1;
3960 return 0;
3961 }
3962
3963 /*
3964 * If this is the root level, and there's only one entry left, and it's
3965 * NOT the leaf level, then we can get rid of this level.
3966 */
3967 if (level == cur->bc_nlevels - 1) {
3968 if (numrecs == 1 && level > 0) {
3969 union xfs_btree_ptr *pp;
3970 /*
3971 * pp is still set to the first pointer in the block.
3972 * Make it the new root of the btree.
3973 */
3974 pp = xfs_btree_ptr_addr(cur, 1, block);
3975 error = xfs_btree_kill_root(cur, bp, level, pp);
3976 if (error)
3977 goto error0;
3978 } else if (level > 0) {
3979 error = xfs_btree_dec_cursor(cur, level, stat);
3980 if (error)
3981 goto error0;
3982 }
3983 *stat = 1;
3984 return 0;
3985 }
3986
3987 /*
3988 * If we deleted the leftmost entry in the block, update the
3989 * key values above us in the tree.
3990 */
3991 if (xfs_btree_needs_key_update(cur, ptr)) {
3992 error = xfs_btree_update_keys(cur, level);
3993 if (error)
3994 goto error0;
3995 }
3996
3997 /*
3998 * If the number of records remaining in the block is at least
3999 * the minimum, we're done.
4000 */
4001 if (numrecs >= cur->bc_ops->get_minrecs(cur, level)) {
4002 error = xfs_btree_dec_cursor(cur, level, stat);
4003 if (error)
4004 goto error0;
4005 return 0;
4006 }
4007
4008 /*
4009 * Otherwise, we have to move some records around to keep the
4010 * tree balanced. Look at the left and right sibling blocks to
4011 * see if we can re-balance by moving only one record.
4012 */
4013 xfs_btree_get_sibling(cur, block, &rptr, XFS_BB_RIGHTSIB);
4014 xfs_btree_get_sibling(cur, block, &lptr, XFS_BB_LEFTSIB);
4015
4016 if (cur->bc_ops->type == XFS_BTREE_TYPE_INODE) {
4017 /*
4018 * One child of root, need to get a chance to copy its contents
4019 * into the root and delete it. Can't go up to next level,
4020 * there's nothing to delete there.
4021 */
4022 if (xfs_btree_ptr_is_null(cur, &rptr) &&
4023 xfs_btree_ptr_is_null(cur, &lptr) &&
4024 level == cur->bc_nlevels - 2) {
4025 error = xfs_btree_kill_iroot(cur);
4026 if (!error)
4027 error = xfs_btree_dec_cursor(cur, level, stat);
4028 if (error)
4029 goto error0;
4030 return 0;
4031 }
4032 }
4033
4034 ASSERT(!xfs_btree_ptr_is_null(cur, &rptr) ||
4035 !xfs_btree_ptr_is_null(cur, &lptr));
4036
4037 /*
4038 * Duplicate the cursor so our btree manipulations here won't
4039 * disrupt the next level up.
4040 */
4041 error = xfs_btree_dup_cursor(cur, &tcur);
4042 if (error)
4043 goto error0;
4044
4045 /*
4046 * If there's a right sibling, see if it's ok to shift an entry
4047 * out of it.
4048 */
4049 if (!xfs_btree_ptr_is_null(cur, &rptr)) {
4050 /*
4051 * Move the temp cursor to the last entry in the next block.
4052 * Actually any entry but the first would suffice.
4053 */
4054 i = xfs_btree_lastrec(tcur, level);
4055 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
4056 xfs_btree_mark_sick(cur);
4057 error = -EFSCORRUPTED;
4058 goto error0;
4059 }
4060
4061 error = xfs_btree_increment(tcur, level, &i);
4062 if (error)
4063 goto error0;
4064 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
4065 xfs_btree_mark_sick(cur);
4066 error = -EFSCORRUPTED;
4067 goto error0;
4068 }
4069
4070 i = xfs_btree_lastrec(tcur, level);
4071 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
4072 xfs_btree_mark_sick(cur);
4073 error = -EFSCORRUPTED;
4074 goto error0;
4075 }
4076
4077 /* Grab a pointer to the block. */
4078 right = xfs_btree_get_block(tcur, level, &rbp);
4079 #ifdef DEBUG
4080 error = xfs_btree_check_block(tcur, right, level, rbp);
4081 if (error)
4082 goto error0;
4083 #endif
4084 /* Grab the current block number, for future use. */
4085 xfs_btree_get_sibling(tcur, right, &cptr, XFS_BB_LEFTSIB);
4086
4087 /*
4088 * If right block is full enough so that removing one entry
4089 * won't make it too empty, and left-shifting an entry out
4090 * of right to us works, we're done.
4091 */
4092 if (xfs_btree_get_numrecs(right) - 1 >=
4093 cur->bc_ops->get_minrecs(tcur, level)) {
4094 error = xfs_btree_lshift(tcur, level, &i);
4095 if (error)
4096 goto error0;
4097 if (i) {
4098 ASSERT(xfs_btree_get_numrecs(block) >=
4099 cur->bc_ops->get_minrecs(tcur, level));
4100
4101 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
4102 tcur = NULL;
4103
4104 error = xfs_btree_dec_cursor(cur, level, stat);
4105 if (error)
4106 goto error0;
4107 return 0;
4108 }
4109 }
4110
4111 /*
4112 * Otherwise, grab the number of records in right for
4113 * future reference, and fix up the temp cursor to point
4114 * to our block again (last record).
4115 */
4116 rrecs = xfs_btree_get_numrecs(right);
4117 if (!xfs_btree_ptr_is_null(cur, &lptr)) {
4118 i = xfs_btree_firstrec(tcur, level);
4119 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
4120 xfs_btree_mark_sick(cur);
4121 error = -EFSCORRUPTED;
4122 goto error0;
4123 }
4124
4125 error = xfs_btree_decrement(tcur, level, &i);
4126 if (error)
4127 goto error0;
4128 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
4129 xfs_btree_mark_sick(cur);
4130 error = -EFSCORRUPTED;
4131 goto error0;
4132 }
4133 }
4134 }
4135
4136 /*
4137 * If there's a left sibling, see if it's ok to shift an entry
4138 * out of it.
4139 */
4140 if (!xfs_btree_ptr_is_null(cur, &lptr)) {
4141 /*
4142 * Move the temp cursor to the first entry in the
4143 * previous block.
4144 */
4145 i = xfs_btree_firstrec(tcur, level);
4146 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
4147 xfs_btree_mark_sick(cur);
4148 error = -EFSCORRUPTED;
4149 goto error0;
4150 }
4151
4152 error = xfs_btree_decrement(tcur, level, &i);
4153 if (error)
4154 goto error0;
4155 i = xfs_btree_firstrec(tcur, level);
4156 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
4157 xfs_btree_mark_sick(cur);
4158 error = -EFSCORRUPTED;
4159 goto error0;
4160 }
4161
4162 /* Grab a pointer to the block. */
4163 left = xfs_btree_get_block(tcur, level, &lbp);
4164 #ifdef DEBUG
4165 error = xfs_btree_check_block(cur, left, level, lbp);
4166 if (error)
4167 goto error0;
4168 #endif
4169 /* Grab the current block number, for future use. */
4170 xfs_btree_get_sibling(tcur, left, &cptr, XFS_BB_RIGHTSIB);
4171
4172 /*
4173 * If left block is full enough so that removing one entry
4174 * won't make it too empty, and right-shifting an entry out
4175 * of left to us works, we're done.
4176 */
4177 if (xfs_btree_get_numrecs(left) - 1 >=
4178 cur->bc_ops->get_minrecs(tcur, level)) {
4179 error = xfs_btree_rshift(tcur, level, &i);
4180 if (error)
4181 goto error0;
4182 if (i) {
4183 ASSERT(xfs_btree_get_numrecs(block) >=
4184 cur->bc_ops->get_minrecs(tcur, level));
4185 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
4186 tcur = NULL;
4187 if (level == 0)
4188 cur->bc_levels[0].ptr++;
4189
4190 *stat = 1;
4191 return 0;
4192 }
4193 }
4194
4195 /*
4196 * Otherwise, grab the number of records in right for
4197 * future reference.
4198 */
4199 lrecs = xfs_btree_get_numrecs(left);
4200 }
4201
4202 /* Delete the temp cursor, we're done with it. */
4203 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
4204 tcur = NULL;
4205
4206 /* If here, we need to do a join to keep the tree balanced. */
4207 ASSERT(!xfs_btree_ptr_is_null(cur, &cptr));
4208
4209 if (!xfs_btree_ptr_is_null(cur, &lptr) &&
4210 lrecs + xfs_btree_get_numrecs(block) <=
4211 cur->bc_ops->get_maxrecs(cur, level)) {
4212 /*
4213 * Set "right" to be the starting block,
4214 * "left" to be the left neighbor.
4215 */
4216 rptr = cptr;
4217 right = block;
4218 rbp = bp;
4219 error = xfs_btree_read_buf_block(cur, &lptr, 0, &left, &lbp);
4220 if (error)
4221 goto error0;
4222
4223 /*
4224 * If that won't work, see if we can join with the right neighbor block.
4225 */
4226 } else if (!xfs_btree_ptr_is_null(cur, &rptr) &&
4227 rrecs + xfs_btree_get_numrecs(block) <=
4228 cur->bc_ops->get_maxrecs(cur, level)) {
4229 /*
4230 * Set "left" to be the starting block,
4231 * "right" to be the right neighbor.
4232 */
4233 lptr = cptr;
4234 left = block;
4235 lbp = bp;
4236 error = xfs_btree_read_buf_block(cur, &rptr, 0, &right, &rbp);
4237 if (error)
4238 goto error0;
4239
4240 /*
4241 * Otherwise, we can't fix the imbalance.
4242 * Just return. This is probably a logic error, but it's not fatal.
4243 */
4244 } else {
4245 error = xfs_btree_dec_cursor(cur, level, stat);
4246 if (error)
4247 goto error0;
4248 return 0;
4249 }
4250
4251 rrecs = xfs_btree_get_numrecs(right);
4252 lrecs = xfs_btree_get_numrecs(left);
4253
4254 /*
4255 * We're now going to join "left" and "right" by moving all the stuff
4256 * in "right" to "left" and deleting "right".
4257 */
4258 XFS_BTREE_STATS_ADD(cur, moves, rrecs);
4259 if (level > 0) {
4260 /* It's a non-leaf. Move keys and pointers. */
4261 union xfs_btree_key *lkp; /* left btree key */
4262 union xfs_btree_ptr *lpp; /* left address pointer */
4263 union xfs_btree_key *rkp; /* right btree key */
4264 union xfs_btree_ptr *rpp; /* right address pointer */
4265
4266 lkp = xfs_btree_key_addr(cur, lrecs + 1, left);
4267 lpp = xfs_btree_ptr_addr(cur, lrecs + 1, left);
4268 rkp = xfs_btree_key_addr(cur, 1, right);
4269 rpp = xfs_btree_ptr_addr(cur, 1, right);
4270
4271 for (i = 1; i < rrecs; i++) {
4272 error = xfs_btree_debug_check_ptr(cur, rpp, i, level);
4273 if (error)
4274 goto error0;
4275 }
4276
4277 xfs_btree_copy_keys(cur, lkp, rkp, rrecs);
4278 xfs_btree_copy_ptrs(cur, lpp, rpp, rrecs);
4279
4280 xfs_btree_log_keys(cur, lbp, lrecs + 1, lrecs + rrecs);
4281 xfs_btree_log_ptrs(cur, lbp, lrecs + 1, lrecs + rrecs);
4282 } else {
4283 /* It's a leaf. Move records. */
4284 union xfs_btree_rec *lrp; /* left record pointer */
4285 union xfs_btree_rec *rrp; /* right record pointer */
4286
4287 lrp = xfs_btree_rec_addr(cur, lrecs + 1, left);
4288 rrp = xfs_btree_rec_addr(cur, 1, right);
4289
4290 xfs_btree_copy_recs(cur, lrp, rrp, rrecs);
4291 xfs_btree_log_recs(cur, lbp, lrecs + 1, lrecs + rrecs);
4292 }
4293
4294 XFS_BTREE_STATS_INC(cur, join);
4295
4296 /*
4297 * Fix up the number of records and right block pointer in the
4298 * surviving block, and log it.
4299 */
4300 xfs_btree_set_numrecs(left, lrecs + rrecs);
4301 xfs_btree_get_sibling(cur, right, &cptr, XFS_BB_RIGHTSIB);
4302 xfs_btree_set_sibling(cur, left, &cptr, XFS_BB_RIGHTSIB);
4303 xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS | XFS_BB_RIGHTSIB);
4304
4305 /* If there is a right sibling, point it to the remaining block. */
4306 xfs_btree_get_sibling(cur, left, &cptr, XFS_BB_RIGHTSIB);
4307 if (!xfs_btree_ptr_is_null(cur, &cptr)) {
4308 error = xfs_btree_read_buf_block(cur, &cptr, 0, &rrblock, &rrbp);
4309 if (error)
4310 goto error0;
4311 xfs_btree_set_sibling(cur, rrblock, &lptr, XFS_BB_LEFTSIB);
4312 xfs_btree_log_block(cur, rrbp, XFS_BB_LEFTSIB);
4313 }
4314
4315 /* Free the deleted block. */
4316 error = xfs_btree_free_block(cur, rbp);
4317 if (error)
4318 goto error0;
4319
4320 /*
4321 * If we joined with the left neighbor, set the buffer in the
4322 * cursor to the left block, and fix up the index.
4323 */
4324 if (bp != lbp) {
4325 cur->bc_levels[level].bp = lbp;
4326 cur->bc_levels[level].ptr += lrecs;
4327 cur->bc_levels[level].ra = 0;
4328 }
4329 /*
4330 * If we joined with the right neighbor and there's a level above
4331 * us, increment the cursor at that level.
4332 */
4333 else if (cur->bc_ops->type == XFS_BTREE_TYPE_INODE ||
4334 level + 1 < cur->bc_nlevels) {
4335 error = xfs_btree_increment(cur, level + 1, &i);
4336 if (error)
4337 goto error0;
4338 }
4339
4340 /*
4341 * Readjust the ptr at this level if it's not a leaf, since it's
4342 * still pointing at the deletion point, which makes the cursor
4343 * inconsistent. If this makes the ptr 0, the caller fixes it up.
4344 * We can't use decrement because it would change the next level up.
4345 */
4346 if (level > 0)
4347 cur->bc_levels[level].ptr--;
4348
4349 /*
4350 * We combined blocks, so we have to update the parent keys if the
4351 * btree supports overlapped intervals. However,
4352 * bc_levels[level + 1].ptr points to the old block so that the caller
4353 * knows which record to delete. Therefore, the caller must be savvy
4354 * enough to call updkeys for us if we return stat == 2. The other
4355 * exit points from this function don't require deletions further up
4356 * the tree, so they can call updkeys directly.
4357 */
4358
4359 /* Return value means the next level up has something to do. */
4360 *stat = 2;
4361 return 0;
4362
4363 error0:
4364 if (tcur)
4365 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
4366 return error;
4367 }
4368
4369 /*
4370 * Delete the record pointed to by cur.
4371 * The cursor refers to the place where the record was (could be inserted)
4372 * when the operation returns.
4373 */
4374 int /* error */
xfs_btree_delete(struct xfs_btree_cur * cur,int * stat)4375 xfs_btree_delete(
4376 struct xfs_btree_cur *cur,
4377 int *stat) /* success/failure */
4378 {
4379 int error; /* error return value */
4380 int level;
4381 int i;
4382 bool joined = false;
4383
4384 /*
4385 * Go up the tree, starting at leaf level.
4386 *
4387 * If 2 is returned then a join was done; go to the next level.
4388 * Otherwise we are done.
4389 */
4390 for (level = 0, i = 2; i == 2; level++) {
4391 error = xfs_btree_delrec(cur, level, &i);
4392 if (error)
4393 goto error0;
4394 if (i == 2)
4395 joined = true;
4396 }
4397
4398 /*
4399 * If we combined blocks as part of deleting the record, delrec won't
4400 * have updated the parent high keys so we have to do that here.
4401 */
4402 if (joined && (cur->bc_ops->geom_flags & XFS_BTGEO_OVERLAPPING)) {
4403 error = xfs_btree_updkeys_force(cur, 0);
4404 if (error)
4405 goto error0;
4406 }
4407
4408 if (i == 0) {
4409 for (level = 1; level < cur->bc_nlevels; level++) {
4410 if (cur->bc_levels[level].ptr == 0) {
4411 error = xfs_btree_decrement(cur, level, &i);
4412 if (error)
4413 goto error0;
4414 break;
4415 }
4416 }
4417 }
4418
4419 *stat = i;
4420 return 0;
4421 error0:
4422 return error;
4423 }
4424
4425 /*
4426 * Get the data from the pointed-to record.
4427 */
4428 int /* error */
xfs_btree_get_rec(struct xfs_btree_cur * cur,union xfs_btree_rec ** recp,int * stat)4429 xfs_btree_get_rec(
4430 struct xfs_btree_cur *cur, /* btree cursor */
4431 union xfs_btree_rec **recp, /* output: btree record */
4432 int *stat) /* output: success/failure */
4433 {
4434 struct xfs_btree_block *block; /* btree block */
4435 struct xfs_buf *bp; /* buffer pointer */
4436 int ptr; /* record number */
4437 #ifdef DEBUG
4438 int error; /* error return value */
4439 #endif
4440
4441 ptr = cur->bc_levels[0].ptr;
4442 block = xfs_btree_get_block(cur, 0, &bp);
4443
4444 #ifdef DEBUG
4445 error = xfs_btree_check_block(cur, block, 0, bp);
4446 if (error)
4447 return error;
4448 #endif
4449
4450 /*
4451 * Off the right end or left end, return failure.
4452 */
4453 if (ptr > xfs_btree_get_numrecs(block) || ptr <= 0) {
4454 *stat = 0;
4455 return 0;
4456 }
4457
4458 /*
4459 * Point to the record and extract its data.
4460 */
4461 *recp = xfs_btree_rec_addr(cur, ptr, block);
4462 *stat = 1;
4463 return 0;
4464 }
4465
4466 /* Visit a block in a btree. */
4467 STATIC int
xfs_btree_visit_block(struct xfs_btree_cur * cur,int level,xfs_btree_visit_blocks_fn fn,void * data)4468 xfs_btree_visit_block(
4469 struct xfs_btree_cur *cur,
4470 int level,
4471 xfs_btree_visit_blocks_fn fn,
4472 void *data)
4473 {
4474 struct xfs_btree_block *block;
4475 struct xfs_buf *bp;
4476 union xfs_btree_ptr rptr, bufptr;
4477 int error;
4478
4479 /* do right sibling readahead */
4480 xfs_btree_readahead(cur, level, XFS_BTCUR_RIGHTRA);
4481 block = xfs_btree_get_block(cur, level, &bp);
4482
4483 /* process the block */
4484 error = fn(cur, level, data);
4485 if (error)
4486 return error;
4487
4488 /* now read rh sibling block for next iteration */
4489 xfs_btree_get_sibling(cur, block, &rptr, XFS_BB_RIGHTSIB);
4490 if (xfs_btree_ptr_is_null(cur, &rptr))
4491 return -ENOENT;
4492
4493 /*
4494 * We only visit blocks once in this walk, so we have to avoid the
4495 * internal xfs_btree_lookup_get_block() optimisation where it will
4496 * return the same block without checking if the right sibling points
4497 * back to us and creates a cyclic reference in the btree.
4498 */
4499 xfs_btree_buf_to_ptr(cur, bp, &bufptr);
4500 if (xfs_btree_ptrs_equal(cur, &rptr, &bufptr)) {
4501 xfs_btree_mark_sick(cur);
4502 return -EFSCORRUPTED;
4503 }
4504
4505 return xfs_btree_lookup_get_block(cur, level, &rptr, &block);
4506 }
4507
4508
4509 /* Visit every block in a btree. */
4510 int
xfs_btree_visit_blocks(struct xfs_btree_cur * cur,xfs_btree_visit_blocks_fn fn,unsigned int flags,void * data)4511 xfs_btree_visit_blocks(
4512 struct xfs_btree_cur *cur,
4513 xfs_btree_visit_blocks_fn fn,
4514 unsigned int flags,
4515 void *data)
4516 {
4517 union xfs_btree_ptr lptr;
4518 int level;
4519 struct xfs_btree_block *block = NULL;
4520 int error = 0;
4521
4522 xfs_btree_init_ptr_from_cur(cur, &lptr);
4523
4524 /* for each level */
4525 for (level = cur->bc_nlevels - 1; level >= 0; level--) {
4526 /* grab the left hand block */
4527 error = xfs_btree_lookup_get_block(cur, level, &lptr, &block);
4528 if (error)
4529 return error;
4530
4531 /* readahead the left most block for the next level down */
4532 if (level > 0) {
4533 union xfs_btree_ptr *ptr;
4534
4535 ptr = xfs_btree_ptr_addr(cur, 1, block);
4536 xfs_btree_readahead_ptr(cur, ptr, 1);
4537
4538 /* save for the next iteration of the loop */
4539 xfs_btree_copy_ptrs(cur, &lptr, ptr, 1);
4540
4541 if (!(flags & XFS_BTREE_VISIT_LEAVES))
4542 continue;
4543 } else if (!(flags & XFS_BTREE_VISIT_RECORDS)) {
4544 continue;
4545 }
4546
4547 /* for each buffer in the level */
4548 do {
4549 error = xfs_btree_visit_block(cur, level, fn, data);
4550 } while (!error);
4551
4552 if (error != -ENOENT)
4553 return error;
4554 }
4555
4556 return 0;
4557 }
4558
4559 /*
4560 * Change the owner of a btree.
4561 *
4562 * The mechanism we use here is ordered buffer logging. Because we don't know
4563 * how many buffers were are going to need to modify, we don't really want to
4564 * have to make transaction reservations for the worst case of every buffer in a
4565 * full size btree as that may be more space that we can fit in the log....
4566 *
4567 * We do the btree walk in the most optimal manner possible - we have sibling
4568 * pointers so we can just walk all the blocks on each level from left to right
4569 * in a single pass, and then move to the next level and do the same. We can
4570 * also do readahead on the sibling pointers to get IO moving more quickly,
4571 * though for slow disks this is unlikely to make much difference to performance
4572 * as the amount of CPU work we have to do before moving to the next block is
4573 * relatively small.
4574 *
4575 * For each btree block that we load, modify the owner appropriately, set the
4576 * buffer as an ordered buffer and log it appropriately. We need to ensure that
4577 * we mark the region we change dirty so that if the buffer is relogged in
4578 * a subsequent transaction the changes we make here as an ordered buffer are
4579 * correctly relogged in that transaction. If we are in recovery context, then
4580 * just queue the modified buffer as delayed write buffer so the transaction
4581 * recovery completion writes the changes to disk.
4582 */
4583 struct xfs_btree_block_change_owner_info {
4584 uint64_t new_owner;
4585 struct list_head *buffer_list;
4586 };
4587
4588 static int
xfs_btree_block_change_owner(struct xfs_btree_cur * cur,int level,void * data)4589 xfs_btree_block_change_owner(
4590 struct xfs_btree_cur *cur,
4591 int level,
4592 void *data)
4593 {
4594 struct xfs_btree_block_change_owner_info *bbcoi = data;
4595 struct xfs_btree_block *block;
4596 struct xfs_buf *bp;
4597
4598 /* modify the owner */
4599 block = xfs_btree_get_block(cur, level, &bp);
4600 if (cur->bc_ops->ptr_len == XFS_BTREE_LONG_PTR_LEN) {
4601 if (block->bb_u.l.bb_owner == cpu_to_be64(bbcoi->new_owner))
4602 return 0;
4603 block->bb_u.l.bb_owner = cpu_to_be64(bbcoi->new_owner);
4604 } else {
4605 if (block->bb_u.s.bb_owner == cpu_to_be32(bbcoi->new_owner))
4606 return 0;
4607 block->bb_u.s.bb_owner = cpu_to_be32(bbcoi->new_owner);
4608 }
4609
4610 /*
4611 * If the block is a root block hosted in an inode, we might not have a
4612 * buffer pointer here and we shouldn't attempt to log the change as the
4613 * information is already held in the inode and discarded when the root
4614 * block is formatted into the on-disk inode fork. We still change it,
4615 * though, so everything is consistent in memory.
4616 */
4617 if (!bp) {
4618 ASSERT(cur->bc_ops->type == XFS_BTREE_TYPE_INODE);
4619 ASSERT(level == cur->bc_nlevels - 1);
4620 return 0;
4621 }
4622
4623 if (cur->bc_tp) {
4624 if (!xfs_trans_ordered_buf(cur->bc_tp, bp)) {
4625 xfs_btree_log_block(cur, bp, XFS_BB_OWNER);
4626 return -EAGAIN;
4627 }
4628 } else {
4629 xfs_buf_delwri_queue(bp, bbcoi->buffer_list);
4630 }
4631
4632 return 0;
4633 }
4634
4635 int
xfs_btree_change_owner(struct xfs_btree_cur * cur,uint64_t new_owner,struct list_head * buffer_list)4636 xfs_btree_change_owner(
4637 struct xfs_btree_cur *cur,
4638 uint64_t new_owner,
4639 struct list_head *buffer_list)
4640 {
4641 struct xfs_btree_block_change_owner_info bbcoi;
4642
4643 bbcoi.new_owner = new_owner;
4644 bbcoi.buffer_list = buffer_list;
4645
4646 return xfs_btree_visit_blocks(cur, xfs_btree_block_change_owner,
4647 XFS_BTREE_VISIT_ALL, &bbcoi);
4648 }
4649
4650 /* Verify the v5 fields of a long-format btree block. */
4651 xfs_failaddr_t
xfs_btree_fsblock_v5hdr_verify(struct xfs_buf * bp,uint64_t owner)4652 xfs_btree_fsblock_v5hdr_verify(
4653 struct xfs_buf *bp,
4654 uint64_t owner)
4655 {
4656 struct xfs_mount *mp = bp->b_mount;
4657 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
4658
4659 if (!xfs_has_crc(mp))
4660 return __this_address;
4661 if (!uuid_equal(&block->bb_u.l.bb_uuid, &mp->m_sb.sb_meta_uuid))
4662 return __this_address;
4663 if (block->bb_u.l.bb_blkno != cpu_to_be64(xfs_buf_daddr(bp)))
4664 return __this_address;
4665 if (owner != XFS_RMAP_OWN_UNKNOWN &&
4666 be64_to_cpu(block->bb_u.l.bb_owner) != owner)
4667 return __this_address;
4668 return NULL;
4669 }
4670
4671 /* Verify a long-format btree block. */
4672 xfs_failaddr_t
xfs_btree_fsblock_verify(struct xfs_buf * bp,unsigned int max_recs)4673 xfs_btree_fsblock_verify(
4674 struct xfs_buf *bp,
4675 unsigned int max_recs)
4676 {
4677 struct xfs_mount *mp = bp->b_mount;
4678 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
4679 xfs_fsblock_t fsb;
4680 xfs_failaddr_t fa;
4681
4682 ASSERT(!xfs_buftarg_is_mem(bp->b_target));
4683
4684 /* numrecs verification */
4685 if (be16_to_cpu(block->bb_numrecs) > max_recs)
4686 return __this_address;
4687
4688 /* sibling pointer verification */
4689 fsb = XFS_DADDR_TO_FSB(mp, xfs_buf_daddr(bp));
4690 fa = xfs_btree_check_fsblock_siblings(mp, fsb,
4691 block->bb_u.l.bb_leftsib);
4692 if (!fa)
4693 fa = xfs_btree_check_fsblock_siblings(mp, fsb,
4694 block->bb_u.l.bb_rightsib);
4695 return fa;
4696 }
4697
4698 /* Verify an in-memory btree block. */
4699 xfs_failaddr_t
xfs_btree_memblock_verify(struct xfs_buf * bp,unsigned int max_recs)4700 xfs_btree_memblock_verify(
4701 struct xfs_buf *bp,
4702 unsigned int max_recs)
4703 {
4704 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
4705 struct xfs_buftarg *btp = bp->b_target;
4706 xfs_failaddr_t fa;
4707 xfbno_t bno;
4708
4709 ASSERT(xfs_buftarg_is_mem(bp->b_target));
4710
4711 /* numrecs verification */
4712 if (be16_to_cpu(block->bb_numrecs) > max_recs)
4713 return __this_address;
4714
4715 /* sibling pointer verification */
4716 bno = xfs_daddr_to_xfbno(xfs_buf_daddr(bp));
4717 fa = xfs_btree_check_memblock_siblings(btp, bno,
4718 block->bb_u.l.bb_leftsib);
4719 if (fa)
4720 return fa;
4721 fa = xfs_btree_check_memblock_siblings(btp, bno,
4722 block->bb_u.l.bb_rightsib);
4723 if (fa)
4724 return fa;
4725
4726 return NULL;
4727 }
4728 /**
4729 * xfs_btree_agblock_v5hdr_verify() -- verify the v5 fields of a short-format
4730 * btree block
4731 *
4732 * @bp: buffer containing the btree block
4733 */
4734 xfs_failaddr_t
xfs_btree_agblock_v5hdr_verify(struct xfs_buf * bp)4735 xfs_btree_agblock_v5hdr_verify(
4736 struct xfs_buf *bp)
4737 {
4738 struct xfs_mount *mp = bp->b_mount;
4739 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
4740 struct xfs_perag *pag = bp->b_pag;
4741
4742 if (!xfs_has_crc(mp))
4743 return __this_address;
4744 if (!uuid_equal(&block->bb_u.s.bb_uuid, &mp->m_sb.sb_meta_uuid))
4745 return __this_address;
4746 if (block->bb_u.s.bb_blkno != cpu_to_be64(xfs_buf_daddr(bp)))
4747 return __this_address;
4748 if (pag && be32_to_cpu(block->bb_u.s.bb_owner) != pag->pag_agno)
4749 return __this_address;
4750 return NULL;
4751 }
4752
4753 /**
4754 * xfs_btree_agblock_verify() -- verify a short-format btree block
4755 *
4756 * @bp: buffer containing the btree block
4757 * @max_recs: maximum records allowed in this btree node
4758 */
4759 xfs_failaddr_t
xfs_btree_agblock_verify(struct xfs_buf * bp,unsigned int max_recs)4760 xfs_btree_agblock_verify(
4761 struct xfs_buf *bp,
4762 unsigned int max_recs)
4763 {
4764 struct xfs_mount *mp = bp->b_mount;
4765 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
4766 xfs_agblock_t agbno;
4767 xfs_failaddr_t fa;
4768
4769 ASSERT(!xfs_buftarg_is_mem(bp->b_target));
4770
4771 /* numrecs verification */
4772 if (be16_to_cpu(block->bb_numrecs) > max_recs)
4773 return __this_address;
4774
4775 /* sibling pointer verification */
4776 agbno = xfs_daddr_to_agbno(mp, xfs_buf_daddr(bp));
4777 fa = xfs_btree_check_agblock_siblings(bp->b_pag, agbno,
4778 block->bb_u.s.bb_leftsib);
4779 if (!fa)
4780 fa = xfs_btree_check_agblock_siblings(bp->b_pag, agbno,
4781 block->bb_u.s.bb_rightsib);
4782 return fa;
4783 }
4784
4785 /*
4786 * For the given limits on leaf and keyptr records per block, calculate the
4787 * height of the tree needed to index the number of leaf records.
4788 */
4789 unsigned int
xfs_btree_compute_maxlevels(const unsigned int * limits,unsigned long long records)4790 xfs_btree_compute_maxlevels(
4791 const unsigned int *limits,
4792 unsigned long long records)
4793 {
4794 unsigned long long level_blocks = howmany_64(records, limits[0]);
4795 unsigned int height = 1;
4796
4797 while (level_blocks > 1) {
4798 level_blocks = howmany_64(level_blocks, limits[1]);
4799 height++;
4800 }
4801
4802 return height;
4803 }
4804
4805 /*
4806 * For the given limits on leaf and keyptr records per block, calculate the
4807 * number of blocks needed to index the given number of leaf records.
4808 */
4809 unsigned long long
xfs_btree_calc_size(const unsigned int * limits,unsigned long long records)4810 xfs_btree_calc_size(
4811 const unsigned int *limits,
4812 unsigned long long records)
4813 {
4814 unsigned long long level_blocks = howmany_64(records, limits[0]);
4815 unsigned long long blocks = level_blocks;
4816
4817 while (level_blocks > 1) {
4818 level_blocks = howmany_64(level_blocks, limits[1]);
4819 blocks += level_blocks;
4820 }
4821
4822 return blocks;
4823 }
4824
4825 /*
4826 * Given a number of available blocks for the btree to consume with records and
4827 * pointers, calculate the height of the tree needed to index all the records
4828 * that space can hold based on the number of pointers each interior node
4829 * holds.
4830 *
4831 * We start by assuming a single level tree consumes a single block, then track
4832 * the number of blocks each node level consumes until we no longer have space
4833 * to store the next node level. At this point, we are indexing all the leaf
4834 * blocks in the space, and there's no more free space to split the tree any
4835 * further. That's our maximum btree height.
4836 */
4837 unsigned int
xfs_btree_space_to_height(const unsigned int * limits,unsigned long long leaf_blocks)4838 xfs_btree_space_to_height(
4839 const unsigned int *limits,
4840 unsigned long long leaf_blocks)
4841 {
4842 /*
4843 * The root btree block can have fewer than minrecs pointers in it
4844 * because the tree might not be big enough to require that amount of
4845 * fanout. Hence it has a minimum size of 2 pointers, not limits[1].
4846 */
4847 unsigned long long node_blocks = 2;
4848 unsigned long long blocks_left = leaf_blocks - 1;
4849 unsigned int height = 1;
4850
4851 if (leaf_blocks < 1)
4852 return 0;
4853
4854 while (node_blocks < blocks_left) {
4855 blocks_left -= node_blocks;
4856 node_blocks *= limits[1];
4857 height++;
4858 }
4859
4860 return height;
4861 }
4862
4863 /*
4864 * Query a regular btree for all records overlapping a given interval.
4865 * Start with a LE lookup of the key of low_rec and return all records
4866 * until we find a record with a key greater than the key of high_rec.
4867 */
4868 STATIC int
xfs_btree_simple_query_range(struct xfs_btree_cur * cur,const union xfs_btree_key * low_key,const union xfs_btree_key * high_key,xfs_btree_query_range_fn fn,void * priv)4869 xfs_btree_simple_query_range(
4870 struct xfs_btree_cur *cur,
4871 const union xfs_btree_key *low_key,
4872 const union xfs_btree_key *high_key,
4873 xfs_btree_query_range_fn fn,
4874 void *priv)
4875 {
4876 union xfs_btree_rec *recp;
4877 union xfs_btree_key rec_key;
4878 int stat;
4879 bool firstrec = true;
4880 int error;
4881
4882 ASSERT(cur->bc_ops->init_high_key_from_rec);
4883 ASSERT(cur->bc_ops->diff_two_keys);
4884
4885 /*
4886 * Find the leftmost record. The btree cursor must be set
4887 * to the low record used to generate low_key.
4888 */
4889 stat = 0;
4890 error = xfs_btree_lookup(cur, XFS_LOOKUP_LE, &stat);
4891 if (error)
4892 goto out;
4893
4894 /* Nothing? See if there's anything to the right. */
4895 if (!stat) {
4896 error = xfs_btree_increment(cur, 0, &stat);
4897 if (error)
4898 goto out;
4899 }
4900
4901 while (stat) {
4902 /* Find the record. */
4903 error = xfs_btree_get_rec(cur, &recp, &stat);
4904 if (error || !stat)
4905 break;
4906
4907 /* Skip if low_key > high_key(rec). */
4908 if (firstrec) {
4909 cur->bc_ops->init_high_key_from_rec(&rec_key, recp);
4910 firstrec = false;
4911 if (xfs_btree_keycmp_gt(cur, low_key, &rec_key))
4912 goto advloop;
4913 }
4914
4915 /* Stop if low_key(rec) > high_key. */
4916 cur->bc_ops->init_key_from_rec(&rec_key, recp);
4917 if (xfs_btree_keycmp_gt(cur, &rec_key, high_key))
4918 break;
4919
4920 /* Callback */
4921 error = fn(cur, recp, priv);
4922 if (error)
4923 break;
4924
4925 advloop:
4926 /* Move on to the next record. */
4927 error = xfs_btree_increment(cur, 0, &stat);
4928 if (error)
4929 break;
4930 }
4931
4932 out:
4933 return error;
4934 }
4935
4936 /*
4937 * Query an overlapped interval btree for all records overlapping a given
4938 * interval. This function roughly follows the algorithm given in
4939 * "Interval Trees" of _Introduction to Algorithms_, which is section
4940 * 14.3 in the 2nd and 3rd editions.
4941 *
4942 * First, generate keys for the low and high records passed in.
4943 *
4944 * For any leaf node, generate the high and low keys for the record.
4945 * If the record keys overlap with the query low/high keys, pass the
4946 * record to the function iterator.
4947 *
4948 * For any internal node, compare the low and high keys of each
4949 * pointer against the query low/high keys. If there's an overlap,
4950 * follow the pointer.
4951 *
4952 * As an optimization, we stop scanning a block when we find a low key
4953 * that is greater than the query's high key.
4954 */
4955 STATIC int
xfs_btree_overlapped_query_range(struct xfs_btree_cur * cur,const union xfs_btree_key * low_key,const union xfs_btree_key * high_key,xfs_btree_query_range_fn fn,void * priv)4956 xfs_btree_overlapped_query_range(
4957 struct xfs_btree_cur *cur,
4958 const union xfs_btree_key *low_key,
4959 const union xfs_btree_key *high_key,
4960 xfs_btree_query_range_fn fn,
4961 void *priv)
4962 {
4963 union xfs_btree_ptr ptr;
4964 union xfs_btree_ptr *pp;
4965 union xfs_btree_key rec_key;
4966 union xfs_btree_key rec_hkey;
4967 union xfs_btree_key *lkp;
4968 union xfs_btree_key *hkp;
4969 union xfs_btree_rec *recp;
4970 struct xfs_btree_block *block;
4971 int level;
4972 struct xfs_buf *bp;
4973 int i;
4974 int error;
4975
4976 /* Load the root of the btree. */
4977 level = cur->bc_nlevels - 1;
4978 xfs_btree_init_ptr_from_cur(cur, &ptr);
4979 error = xfs_btree_lookup_get_block(cur, level, &ptr, &block);
4980 if (error)
4981 return error;
4982 xfs_btree_get_block(cur, level, &bp);
4983 trace_xfs_btree_overlapped_query_range(cur, level, bp);
4984 #ifdef DEBUG
4985 error = xfs_btree_check_block(cur, block, level, bp);
4986 if (error)
4987 goto out;
4988 #endif
4989 cur->bc_levels[level].ptr = 1;
4990
4991 while (level < cur->bc_nlevels) {
4992 block = xfs_btree_get_block(cur, level, &bp);
4993
4994 /* End of node, pop back towards the root. */
4995 if (cur->bc_levels[level].ptr >
4996 be16_to_cpu(block->bb_numrecs)) {
4997 pop_up:
4998 if (level < cur->bc_nlevels - 1)
4999 cur->bc_levels[level + 1].ptr++;
5000 level++;
5001 continue;
5002 }
5003
5004 if (level == 0) {
5005 /* Handle a leaf node. */
5006 recp = xfs_btree_rec_addr(cur, cur->bc_levels[0].ptr,
5007 block);
5008
5009 cur->bc_ops->init_high_key_from_rec(&rec_hkey, recp);
5010 cur->bc_ops->init_key_from_rec(&rec_key, recp);
5011
5012 /*
5013 * If (query's high key < record's low key), then there
5014 * are no more interesting records in this block. Pop
5015 * up to the leaf level to find more record blocks.
5016 *
5017 * If (record's high key >= query's low key) and
5018 * (query's high key >= record's low key), then
5019 * this record overlaps the query range; callback.
5020 */
5021 if (xfs_btree_keycmp_lt(cur, high_key, &rec_key))
5022 goto pop_up;
5023 if (xfs_btree_keycmp_ge(cur, &rec_hkey, low_key)) {
5024 error = fn(cur, recp, priv);
5025 if (error)
5026 break;
5027 }
5028 cur->bc_levels[level].ptr++;
5029 continue;
5030 }
5031
5032 /* Handle an internal node. */
5033 lkp = xfs_btree_key_addr(cur, cur->bc_levels[level].ptr, block);
5034 hkp = xfs_btree_high_key_addr(cur, cur->bc_levels[level].ptr,
5035 block);
5036 pp = xfs_btree_ptr_addr(cur, cur->bc_levels[level].ptr, block);
5037
5038 /*
5039 * If (query's high key < pointer's low key), then there are no
5040 * more interesting keys in this block. Pop up one leaf level
5041 * to continue looking for records.
5042 *
5043 * If (pointer's high key >= query's low key) and
5044 * (query's high key >= pointer's low key), then
5045 * this record overlaps the query range; follow pointer.
5046 */
5047 if (xfs_btree_keycmp_lt(cur, high_key, lkp))
5048 goto pop_up;
5049 if (xfs_btree_keycmp_ge(cur, hkp, low_key)) {
5050 level--;
5051 error = xfs_btree_lookup_get_block(cur, level, pp,
5052 &block);
5053 if (error)
5054 goto out;
5055 xfs_btree_get_block(cur, level, &bp);
5056 trace_xfs_btree_overlapped_query_range(cur, level, bp);
5057 #ifdef DEBUG
5058 error = xfs_btree_check_block(cur, block, level, bp);
5059 if (error)
5060 goto out;
5061 #endif
5062 cur->bc_levels[level].ptr = 1;
5063 continue;
5064 }
5065 cur->bc_levels[level].ptr++;
5066 }
5067
5068 out:
5069 /*
5070 * If we don't end this function with the cursor pointing at a record
5071 * block, a subsequent non-error cursor deletion will not release
5072 * node-level buffers, causing a buffer leak. This is quite possible
5073 * with a zero-results range query, so release the buffers if we
5074 * failed to return any results.
5075 */
5076 if (cur->bc_levels[0].bp == NULL) {
5077 for (i = 0; i < cur->bc_nlevels; i++) {
5078 if (cur->bc_levels[i].bp) {
5079 xfs_trans_brelse(cur->bc_tp,
5080 cur->bc_levels[i].bp);
5081 cur->bc_levels[i].bp = NULL;
5082 cur->bc_levels[i].ptr = 0;
5083 cur->bc_levels[i].ra = 0;
5084 }
5085 }
5086 }
5087
5088 return error;
5089 }
5090
5091 static inline void
xfs_btree_key_from_irec(struct xfs_btree_cur * cur,union xfs_btree_key * key,const union xfs_btree_irec * irec)5092 xfs_btree_key_from_irec(
5093 struct xfs_btree_cur *cur,
5094 union xfs_btree_key *key,
5095 const union xfs_btree_irec *irec)
5096 {
5097 union xfs_btree_rec rec;
5098
5099 cur->bc_rec = *irec;
5100 cur->bc_ops->init_rec_from_cur(cur, &rec);
5101 cur->bc_ops->init_key_from_rec(key, &rec);
5102 }
5103
5104 /*
5105 * Query a btree for all records overlapping a given interval of keys. The
5106 * supplied function will be called with each record found; return one of the
5107 * XFS_BTREE_QUERY_RANGE_{CONTINUE,ABORT} values or the usual negative error
5108 * code. This function returns -ECANCELED, zero, or a negative error code.
5109 */
5110 int
xfs_btree_query_range(struct xfs_btree_cur * cur,const union xfs_btree_irec * low_rec,const union xfs_btree_irec * high_rec,xfs_btree_query_range_fn fn,void * priv)5111 xfs_btree_query_range(
5112 struct xfs_btree_cur *cur,
5113 const union xfs_btree_irec *low_rec,
5114 const union xfs_btree_irec *high_rec,
5115 xfs_btree_query_range_fn fn,
5116 void *priv)
5117 {
5118 union xfs_btree_key low_key;
5119 union xfs_btree_key high_key;
5120
5121 /* Find the keys of both ends of the interval. */
5122 xfs_btree_key_from_irec(cur, &high_key, high_rec);
5123 xfs_btree_key_from_irec(cur, &low_key, low_rec);
5124
5125 /* Enforce low key <= high key. */
5126 if (!xfs_btree_keycmp_le(cur, &low_key, &high_key))
5127 return -EINVAL;
5128
5129 if (!(cur->bc_ops->geom_flags & XFS_BTGEO_OVERLAPPING))
5130 return xfs_btree_simple_query_range(cur, &low_key,
5131 &high_key, fn, priv);
5132 return xfs_btree_overlapped_query_range(cur, &low_key, &high_key,
5133 fn, priv);
5134 }
5135
5136 /* Query a btree for all records. */
5137 int
xfs_btree_query_all(struct xfs_btree_cur * cur,xfs_btree_query_range_fn fn,void * priv)5138 xfs_btree_query_all(
5139 struct xfs_btree_cur *cur,
5140 xfs_btree_query_range_fn fn,
5141 void *priv)
5142 {
5143 union xfs_btree_key low_key;
5144 union xfs_btree_key high_key;
5145
5146 memset(&cur->bc_rec, 0, sizeof(cur->bc_rec));
5147 memset(&low_key, 0, sizeof(low_key));
5148 memset(&high_key, 0xFF, sizeof(high_key));
5149
5150 return xfs_btree_simple_query_range(cur, &low_key, &high_key, fn, priv);
5151 }
5152
5153 static int
xfs_btree_count_blocks_helper(struct xfs_btree_cur * cur,int level,void * data)5154 xfs_btree_count_blocks_helper(
5155 struct xfs_btree_cur *cur,
5156 int level,
5157 void *data)
5158 {
5159 xfs_extlen_t *blocks = data;
5160 (*blocks)++;
5161
5162 return 0;
5163 }
5164
5165 /* Count the blocks in a btree and return the result in *blocks. */
5166 int
xfs_btree_count_blocks(struct xfs_btree_cur * cur,xfs_extlen_t * blocks)5167 xfs_btree_count_blocks(
5168 struct xfs_btree_cur *cur,
5169 xfs_extlen_t *blocks)
5170 {
5171 *blocks = 0;
5172 return xfs_btree_visit_blocks(cur, xfs_btree_count_blocks_helper,
5173 XFS_BTREE_VISIT_ALL, blocks);
5174 }
5175
5176 /* Compare two btree pointers. */
5177 int64_t
xfs_btree_diff_two_ptrs(struct xfs_btree_cur * cur,const union xfs_btree_ptr * a,const union xfs_btree_ptr * b)5178 xfs_btree_diff_two_ptrs(
5179 struct xfs_btree_cur *cur,
5180 const union xfs_btree_ptr *a,
5181 const union xfs_btree_ptr *b)
5182 {
5183 if (cur->bc_ops->ptr_len == XFS_BTREE_LONG_PTR_LEN)
5184 return (int64_t)be64_to_cpu(a->l) - be64_to_cpu(b->l);
5185 return (int64_t)be32_to_cpu(a->s) - be32_to_cpu(b->s);
5186 }
5187
5188 struct xfs_btree_has_records {
5189 /* Keys for the start and end of the range we want to know about. */
5190 union xfs_btree_key start_key;
5191 union xfs_btree_key end_key;
5192
5193 /* Mask for key comparisons, if desired. */
5194 const union xfs_btree_key *key_mask;
5195
5196 /* Highest record key we've seen so far. */
5197 union xfs_btree_key high_key;
5198
5199 enum xbtree_recpacking outcome;
5200 };
5201
5202 STATIC int
xfs_btree_has_records_helper(struct xfs_btree_cur * cur,const union xfs_btree_rec * rec,void * priv)5203 xfs_btree_has_records_helper(
5204 struct xfs_btree_cur *cur,
5205 const union xfs_btree_rec *rec,
5206 void *priv)
5207 {
5208 union xfs_btree_key rec_key;
5209 union xfs_btree_key rec_high_key;
5210 struct xfs_btree_has_records *info = priv;
5211 enum xbtree_key_contig key_contig;
5212
5213 cur->bc_ops->init_key_from_rec(&rec_key, rec);
5214
5215 if (info->outcome == XBTREE_RECPACKING_EMPTY) {
5216 info->outcome = XBTREE_RECPACKING_SPARSE;
5217
5218 /*
5219 * If the first record we find does not overlap the start key,
5220 * then there is a hole at the start of the search range.
5221 * Classify this as sparse and stop immediately.
5222 */
5223 if (xfs_btree_masked_keycmp_lt(cur, &info->start_key, &rec_key,
5224 info->key_mask))
5225 return -ECANCELED;
5226 } else {
5227 /*
5228 * If a subsequent record does not overlap with the any record
5229 * we've seen so far, there is a hole in the middle of the
5230 * search range. Classify this as sparse and stop.
5231 * If the keys overlap and this btree does not allow overlap,
5232 * signal corruption.
5233 */
5234 key_contig = cur->bc_ops->keys_contiguous(cur, &info->high_key,
5235 &rec_key, info->key_mask);
5236 if (key_contig == XBTREE_KEY_OVERLAP &&
5237 !(cur->bc_ops->geom_flags & XFS_BTGEO_OVERLAPPING))
5238 return -EFSCORRUPTED;
5239 if (key_contig == XBTREE_KEY_GAP)
5240 return -ECANCELED;
5241 }
5242
5243 /*
5244 * If high_key(rec) is larger than any other high key we've seen,
5245 * remember it for later.
5246 */
5247 cur->bc_ops->init_high_key_from_rec(&rec_high_key, rec);
5248 if (xfs_btree_masked_keycmp_gt(cur, &rec_high_key, &info->high_key,
5249 info->key_mask))
5250 info->high_key = rec_high_key; /* struct copy */
5251
5252 return 0;
5253 }
5254
5255 /*
5256 * Scan part of the keyspace of a btree and tell us if that keyspace does not
5257 * map to any records; is fully mapped to records; or is partially mapped to
5258 * records. This is the btree record equivalent to determining if a file is
5259 * sparse.
5260 *
5261 * For most btree types, the record scan should use all available btree key
5262 * fields to compare the keys encountered. These callers should pass NULL for
5263 * @mask. However, some callers (e.g. scanning physical space in the rmapbt)
5264 * want to ignore some part of the btree record keyspace when performing the
5265 * comparison. These callers should pass in a union xfs_btree_key object with
5266 * the fields that *should* be a part of the comparison set to any nonzero
5267 * value, and the rest zeroed.
5268 */
5269 int
xfs_btree_has_records(struct xfs_btree_cur * cur,const union xfs_btree_irec * low,const union xfs_btree_irec * high,const union xfs_btree_key * mask,enum xbtree_recpacking * outcome)5270 xfs_btree_has_records(
5271 struct xfs_btree_cur *cur,
5272 const union xfs_btree_irec *low,
5273 const union xfs_btree_irec *high,
5274 const union xfs_btree_key *mask,
5275 enum xbtree_recpacking *outcome)
5276 {
5277 struct xfs_btree_has_records info = {
5278 .outcome = XBTREE_RECPACKING_EMPTY,
5279 .key_mask = mask,
5280 };
5281 int error;
5282
5283 /* Not all btrees support this operation. */
5284 if (!cur->bc_ops->keys_contiguous) {
5285 ASSERT(0);
5286 return -EOPNOTSUPP;
5287 }
5288
5289 xfs_btree_key_from_irec(cur, &info.start_key, low);
5290 xfs_btree_key_from_irec(cur, &info.end_key, high);
5291
5292 error = xfs_btree_query_range(cur, low, high,
5293 xfs_btree_has_records_helper, &info);
5294 if (error == -ECANCELED)
5295 goto out;
5296 if (error)
5297 return error;
5298
5299 if (info.outcome == XBTREE_RECPACKING_EMPTY)
5300 goto out;
5301
5302 /*
5303 * If the largest high_key(rec) we saw during the walk is greater than
5304 * the end of the search range, classify this as full. Otherwise,
5305 * there is a hole at the end of the search range.
5306 */
5307 if (xfs_btree_masked_keycmp_ge(cur, &info.high_key, &info.end_key,
5308 mask))
5309 info.outcome = XBTREE_RECPACKING_FULL;
5310
5311 out:
5312 *outcome = info.outcome;
5313 return 0;
5314 }
5315
5316 /* Are there more records in this btree? */
5317 bool
xfs_btree_has_more_records(struct xfs_btree_cur * cur)5318 xfs_btree_has_more_records(
5319 struct xfs_btree_cur *cur)
5320 {
5321 struct xfs_btree_block *block;
5322 struct xfs_buf *bp;
5323
5324 block = xfs_btree_get_block(cur, 0, &bp);
5325
5326 /* There are still records in this block. */
5327 if (cur->bc_levels[0].ptr < xfs_btree_get_numrecs(block))
5328 return true;
5329
5330 /* There are more record blocks. */
5331 if (cur->bc_ops->ptr_len == XFS_BTREE_LONG_PTR_LEN)
5332 return block->bb_u.l.bb_rightsib != cpu_to_be64(NULLFSBLOCK);
5333 else
5334 return block->bb_u.s.bb_rightsib != cpu_to_be32(NULLAGBLOCK);
5335 }
5336
5337 /* Set up all the btree cursor caches. */
5338 int __init
xfs_btree_init_cur_caches(void)5339 xfs_btree_init_cur_caches(void)
5340 {
5341 int error;
5342
5343 error = xfs_allocbt_init_cur_cache();
5344 if (error)
5345 return error;
5346 error = xfs_inobt_init_cur_cache();
5347 if (error)
5348 goto err;
5349 error = xfs_bmbt_init_cur_cache();
5350 if (error)
5351 goto err;
5352 error = xfs_rmapbt_init_cur_cache();
5353 if (error)
5354 goto err;
5355 error = xfs_refcountbt_init_cur_cache();
5356 if (error)
5357 goto err;
5358
5359 return 0;
5360 err:
5361 xfs_btree_destroy_cur_caches();
5362 return error;
5363 }
5364
5365 /* Destroy all the btree cursor caches, if they've been allocated. */
5366 void
xfs_btree_destroy_cur_caches(void)5367 xfs_btree_destroy_cur_caches(void)
5368 {
5369 xfs_allocbt_destroy_cur_cache();
5370 xfs_inobt_destroy_cur_cache();
5371 xfs_bmbt_destroy_cur_cache();
5372 xfs_rmapbt_destroy_cur_cache();
5373 xfs_refcountbt_destroy_cur_cache();
5374 }
5375
5376 /* Move the btree cursor before the first record. */
5377 int
xfs_btree_goto_left_edge(struct xfs_btree_cur * cur)5378 xfs_btree_goto_left_edge(
5379 struct xfs_btree_cur *cur)
5380 {
5381 int stat = 0;
5382 int error;
5383
5384 memset(&cur->bc_rec, 0, sizeof(cur->bc_rec));
5385 error = xfs_btree_lookup(cur, XFS_LOOKUP_LE, &stat);
5386 if (error)
5387 return error;
5388 if (!stat)
5389 return 0;
5390
5391 error = xfs_btree_decrement(cur, 0, &stat);
5392 if (error)
5393 return error;
5394 if (stat != 0) {
5395 ASSERT(0);
5396 xfs_btree_mark_sick(cur);
5397 return -EFSCORRUPTED;
5398 }
5399
5400 return 0;
5401 }
5402