xref: /linux/fs/xfs/libxfs/xfs_btree.c (revision e9c4d8bfb26c13c41b73fdf4183d3df2d392101e)
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
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
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
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
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
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
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
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
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
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 
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
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
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
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 				pag_agno(cur->bc_ag.pag), 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
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
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
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
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
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
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 *
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
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 */
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  */
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
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
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
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
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 *
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 *
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 *
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 *
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 *
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 *
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 */
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 */
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 */
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
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
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
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
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_agblock_t		left = be32_to_cpu(block->bb_u.s.bb_leftsib);
1021 	xfs_agblock_t		right = be32_to_cpu(block->bb_u.s.bb_rightsib);
1022 	int			rval = 0;
1023 
1024 	if ((lr & XFS_BTCUR_LEFTRA) && left != NULLAGBLOCK) {
1025 		xfs_buf_readahead(mp->m_ddev_targp,
1026 				xfs_agbno_to_daddr(cur->bc_ag.pag, left),
1027 				mp->m_bsize, cur->bc_ops->buf_ops);
1028 		rval++;
1029 	}
1030 
1031 	if ((lr & XFS_BTCUR_RIGHTRA) && right != NULLAGBLOCK) {
1032 		xfs_buf_readahead(mp->m_ddev_targp,
1033 				xfs_agbno_to_daddr(cur->bc_ag.pag, right),
1034 				mp->m_bsize, cur->bc_ops->buf_ops);
1035 		rval++;
1036 	}
1037 
1038 	return rval;
1039 }
1040 
1041 /*
1042  * Read-ahead btree blocks, at the given level.
1043  * Bits in lr are set from XFS_BTCUR_{LEFT,RIGHT}RA.
1044  */
1045 STATIC int
1046 xfs_btree_readahead(
1047 	struct xfs_btree_cur	*cur,		/* btree cursor */
1048 	int			lev,		/* level in btree */
1049 	int			lr)		/* left/right bits */
1050 {
1051 	struct xfs_btree_block	*block;
1052 
1053 	/*
1054 	 * No readahead needed if we are at the root level and the
1055 	 * btree root is stored in the inode.
1056 	 */
1057 	if (xfs_btree_at_iroot(cur, lev))
1058 		return 0;
1059 
1060 	if ((cur->bc_levels[lev].ra | lr) == cur->bc_levels[lev].ra)
1061 		return 0;
1062 
1063 	cur->bc_levels[lev].ra |= lr;
1064 	block = XFS_BUF_TO_BLOCK(cur->bc_levels[lev].bp);
1065 
1066 	switch (cur->bc_ops->type) {
1067 	case XFS_BTREE_TYPE_AG:
1068 		return xfs_btree_readahead_agblock(cur, lr, block);
1069 	case XFS_BTREE_TYPE_INODE:
1070 		return xfs_btree_readahead_fsblock(cur, lr, block);
1071 	case XFS_BTREE_TYPE_MEM:
1072 		return xfs_btree_readahead_memblock(cur, lr, block);
1073 	default:
1074 		ASSERT(0);
1075 		return 0;
1076 	}
1077 }
1078 
1079 STATIC int
1080 xfs_btree_ptr_to_daddr(
1081 	struct xfs_btree_cur		*cur,
1082 	const union xfs_btree_ptr	*ptr,
1083 	xfs_daddr_t			*daddr)
1084 {
1085 	int			error;
1086 
1087 	error = xfs_btree_check_ptr(cur, ptr, 0, 1);
1088 	if (error)
1089 		return error;
1090 
1091 	switch (cur->bc_ops->type) {
1092 	case XFS_BTREE_TYPE_AG:
1093 		*daddr = xfs_agbno_to_daddr(cur->bc_ag.pag,
1094 				be32_to_cpu(ptr->s));
1095 		break;
1096 	case XFS_BTREE_TYPE_INODE:
1097 		*daddr = XFS_FSB_TO_DADDR(cur->bc_mp, be64_to_cpu(ptr->l));
1098 		break;
1099 	case XFS_BTREE_TYPE_MEM:
1100 		*daddr = xfbno_to_daddr(be64_to_cpu(ptr->l));
1101 		break;
1102 	}
1103 	return 0;
1104 }
1105 
1106 /*
1107  * Readahead @count btree blocks at the given @ptr location.
1108  *
1109  * We don't need to care about long or short form btrees here as we have a
1110  * method of converting the ptr directly to a daddr available to us.
1111  */
1112 STATIC void
1113 xfs_btree_readahead_ptr(
1114 	struct xfs_btree_cur	*cur,
1115 	union xfs_btree_ptr	*ptr,
1116 	xfs_extlen_t		count)
1117 {
1118 	xfs_daddr_t		daddr;
1119 
1120 	if (xfs_btree_ptr_to_daddr(cur, ptr, &daddr))
1121 		return;
1122 	xfs_buf_readahead(xfs_btree_buftarg(cur), daddr,
1123 			xfs_btree_bbsize(cur) * count,
1124 			cur->bc_ops->buf_ops);
1125 }
1126 
1127 /*
1128  * Set the buffer for level "lev" in the cursor to bp, releasing
1129  * any previous buffer.
1130  */
1131 STATIC void
1132 xfs_btree_setbuf(
1133 	struct xfs_btree_cur	*cur,	/* btree cursor */
1134 	int			lev,	/* level in btree */
1135 	struct xfs_buf		*bp)	/* new buffer to set */
1136 {
1137 	struct xfs_btree_block	*b;	/* btree block */
1138 
1139 	if (cur->bc_levels[lev].bp)
1140 		xfs_trans_brelse(cur->bc_tp, cur->bc_levels[lev].bp);
1141 	cur->bc_levels[lev].bp = bp;
1142 	cur->bc_levels[lev].ra = 0;
1143 
1144 	b = XFS_BUF_TO_BLOCK(bp);
1145 	if (cur->bc_ops->ptr_len == XFS_BTREE_LONG_PTR_LEN) {
1146 		if (b->bb_u.l.bb_leftsib == cpu_to_be64(NULLFSBLOCK))
1147 			cur->bc_levels[lev].ra |= XFS_BTCUR_LEFTRA;
1148 		if (b->bb_u.l.bb_rightsib == cpu_to_be64(NULLFSBLOCK))
1149 			cur->bc_levels[lev].ra |= XFS_BTCUR_RIGHTRA;
1150 	} else {
1151 		if (b->bb_u.s.bb_leftsib == cpu_to_be32(NULLAGBLOCK))
1152 			cur->bc_levels[lev].ra |= XFS_BTCUR_LEFTRA;
1153 		if (b->bb_u.s.bb_rightsib == cpu_to_be32(NULLAGBLOCK))
1154 			cur->bc_levels[lev].ra |= XFS_BTCUR_RIGHTRA;
1155 	}
1156 }
1157 
1158 bool
1159 xfs_btree_ptr_is_null(
1160 	struct xfs_btree_cur		*cur,
1161 	const union xfs_btree_ptr	*ptr)
1162 {
1163 	if (cur->bc_ops->ptr_len == XFS_BTREE_LONG_PTR_LEN)
1164 		return ptr->l == cpu_to_be64(NULLFSBLOCK);
1165 	else
1166 		return ptr->s == cpu_to_be32(NULLAGBLOCK);
1167 }
1168 
1169 void
1170 xfs_btree_set_ptr_null(
1171 	struct xfs_btree_cur	*cur,
1172 	union xfs_btree_ptr	*ptr)
1173 {
1174 	if (cur->bc_ops->ptr_len == XFS_BTREE_LONG_PTR_LEN)
1175 		ptr->l = cpu_to_be64(NULLFSBLOCK);
1176 	else
1177 		ptr->s = cpu_to_be32(NULLAGBLOCK);
1178 }
1179 
1180 static inline bool
1181 xfs_btree_ptrs_equal(
1182 	struct xfs_btree_cur		*cur,
1183 	union xfs_btree_ptr		*ptr1,
1184 	union xfs_btree_ptr		*ptr2)
1185 {
1186 	if (cur->bc_ops->ptr_len == XFS_BTREE_LONG_PTR_LEN)
1187 		return ptr1->l == ptr2->l;
1188 	return ptr1->s == ptr2->s;
1189 }
1190 
1191 /*
1192  * Get/set/init sibling pointers
1193  */
1194 void
1195 xfs_btree_get_sibling(
1196 	struct xfs_btree_cur	*cur,
1197 	struct xfs_btree_block	*block,
1198 	union xfs_btree_ptr	*ptr,
1199 	int			lr)
1200 {
1201 	ASSERT(lr == XFS_BB_LEFTSIB || lr == XFS_BB_RIGHTSIB);
1202 
1203 	if (cur->bc_ops->ptr_len == XFS_BTREE_LONG_PTR_LEN) {
1204 		if (lr == XFS_BB_RIGHTSIB)
1205 			ptr->l = block->bb_u.l.bb_rightsib;
1206 		else
1207 			ptr->l = block->bb_u.l.bb_leftsib;
1208 	} else {
1209 		if (lr == XFS_BB_RIGHTSIB)
1210 			ptr->s = block->bb_u.s.bb_rightsib;
1211 		else
1212 			ptr->s = block->bb_u.s.bb_leftsib;
1213 	}
1214 }
1215 
1216 void
1217 xfs_btree_set_sibling(
1218 	struct xfs_btree_cur		*cur,
1219 	struct xfs_btree_block		*block,
1220 	const union xfs_btree_ptr	*ptr,
1221 	int				lr)
1222 {
1223 	ASSERT(lr == XFS_BB_LEFTSIB || lr == XFS_BB_RIGHTSIB);
1224 
1225 	if (cur->bc_ops->ptr_len == XFS_BTREE_LONG_PTR_LEN) {
1226 		if (lr == XFS_BB_RIGHTSIB)
1227 			block->bb_u.l.bb_rightsib = ptr->l;
1228 		else
1229 			block->bb_u.l.bb_leftsib = ptr->l;
1230 	} else {
1231 		if (lr == XFS_BB_RIGHTSIB)
1232 			block->bb_u.s.bb_rightsib = ptr->s;
1233 		else
1234 			block->bb_u.s.bb_leftsib = ptr->s;
1235 	}
1236 }
1237 
1238 static void
1239 __xfs_btree_init_block(
1240 	struct xfs_mount	*mp,
1241 	struct xfs_btree_block	*buf,
1242 	const struct xfs_btree_ops *ops,
1243 	xfs_daddr_t		blkno,
1244 	__u16			level,
1245 	__u16			numrecs,
1246 	__u64			owner)
1247 {
1248 	bool			crc = xfs_has_crc(mp);
1249 	__u32			magic = xfs_btree_magic(mp, ops);
1250 
1251 	buf->bb_magic = cpu_to_be32(magic);
1252 	buf->bb_level = cpu_to_be16(level);
1253 	buf->bb_numrecs = cpu_to_be16(numrecs);
1254 
1255 	if (ops->ptr_len == XFS_BTREE_LONG_PTR_LEN) {
1256 		buf->bb_u.l.bb_leftsib = cpu_to_be64(NULLFSBLOCK);
1257 		buf->bb_u.l.bb_rightsib = cpu_to_be64(NULLFSBLOCK);
1258 		if (crc) {
1259 			buf->bb_u.l.bb_blkno = cpu_to_be64(blkno);
1260 			buf->bb_u.l.bb_owner = cpu_to_be64(owner);
1261 			uuid_copy(&buf->bb_u.l.bb_uuid, &mp->m_sb.sb_meta_uuid);
1262 			buf->bb_u.l.bb_pad = 0;
1263 			buf->bb_u.l.bb_lsn = 0;
1264 		}
1265 	} else {
1266 		buf->bb_u.s.bb_leftsib = cpu_to_be32(NULLAGBLOCK);
1267 		buf->bb_u.s.bb_rightsib = cpu_to_be32(NULLAGBLOCK);
1268 		if (crc) {
1269 			buf->bb_u.s.bb_blkno = cpu_to_be64(blkno);
1270 			/* owner is a 32 bit value on short blocks */
1271 			buf->bb_u.s.bb_owner = cpu_to_be32((__u32)owner);
1272 			uuid_copy(&buf->bb_u.s.bb_uuid, &mp->m_sb.sb_meta_uuid);
1273 			buf->bb_u.s.bb_lsn = 0;
1274 		}
1275 	}
1276 }
1277 
1278 void
1279 xfs_btree_init_block(
1280 	struct xfs_mount	*mp,
1281 	struct xfs_btree_block	*block,
1282 	const struct xfs_btree_ops *ops,
1283 	__u16			level,
1284 	__u16			numrecs,
1285 	__u64			owner)
1286 {
1287 	__xfs_btree_init_block(mp, block, ops, XFS_BUF_DADDR_NULL, level,
1288 			numrecs, owner);
1289 }
1290 
1291 void
1292 xfs_btree_init_buf(
1293 	struct xfs_mount		*mp,
1294 	struct xfs_buf			*bp,
1295 	const struct xfs_btree_ops	*ops,
1296 	__u16				level,
1297 	__u16				numrecs,
1298 	__u64				owner)
1299 {
1300 	__xfs_btree_init_block(mp, XFS_BUF_TO_BLOCK(bp), ops,
1301 			xfs_buf_daddr(bp), level, numrecs, owner);
1302 	bp->b_ops = ops->buf_ops;
1303 }
1304 
1305 static inline __u64
1306 xfs_btree_owner(
1307 	struct xfs_btree_cur    *cur)
1308 {
1309 	switch (cur->bc_ops->type) {
1310 	case XFS_BTREE_TYPE_MEM:
1311 		return cur->bc_mem.xfbtree->owner;
1312 	case XFS_BTREE_TYPE_INODE:
1313 		return cur->bc_ino.ip->i_ino;
1314 	case XFS_BTREE_TYPE_AG:
1315 		return pag_agno(cur->bc_ag.pag);
1316 	default:
1317 		ASSERT(0);
1318 		return 0;
1319 	}
1320 }
1321 
1322 void
1323 xfs_btree_init_block_cur(
1324 	struct xfs_btree_cur	*cur,
1325 	struct xfs_buf		*bp,
1326 	int			level,
1327 	int			numrecs)
1328 {
1329 	xfs_btree_init_buf(cur->bc_mp, bp, cur->bc_ops, level, numrecs,
1330 			xfs_btree_owner(cur));
1331 }
1332 
1333 STATIC void
1334 xfs_btree_buf_to_ptr(
1335 	struct xfs_btree_cur	*cur,
1336 	struct xfs_buf		*bp,
1337 	union xfs_btree_ptr	*ptr)
1338 {
1339 	switch (cur->bc_ops->type) {
1340 	case XFS_BTREE_TYPE_AG:
1341 		ptr->s = cpu_to_be32(xfs_daddr_to_agbno(cur->bc_mp,
1342 					xfs_buf_daddr(bp)));
1343 		break;
1344 	case XFS_BTREE_TYPE_INODE:
1345 		ptr->l = cpu_to_be64(XFS_DADDR_TO_FSB(cur->bc_mp,
1346 					xfs_buf_daddr(bp)));
1347 		break;
1348 	case XFS_BTREE_TYPE_MEM:
1349 		ptr->l = cpu_to_be64(xfs_daddr_to_xfbno(xfs_buf_daddr(bp)));
1350 		break;
1351 	}
1352 }
1353 
1354 static inline void
1355 xfs_btree_set_refs(
1356 	struct xfs_btree_cur	*cur,
1357 	struct xfs_buf		*bp)
1358 {
1359 	xfs_buf_set_ref(bp, cur->bc_ops->lru_refs);
1360 }
1361 
1362 int
1363 xfs_btree_get_buf_block(
1364 	struct xfs_btree_cur		*cur,
1365 	const union xfs_btree_ptr	*ptr,
1366 	struct xfs_btree_block		**block,
1367 	struct xfs_buf			**bpp)
1368 {
1369 	xfs_daddr_t			d;
1370 	int				error;
1371 
1372 	error = xfs_btree_ptr_to_daddr(cur, ptr, &d);
1373 	if (error)
1374 		return error;
1375 	error = xfs_trans_get_buf(cur->bc_tp, xfs_btree_buftarg(cur), d,
1376 			xfs_btree_bbsize(cur), 0, bpp);
1377 	if (error)
1378 		return error;
1379 
1380 	(*bpp)->b_ops = cur->bc_ops->buf_ops;
1381 	*block = XFS_BUF_TO_BLOCK(*bpp);
1382 	return 0;
1383 }
1384 
1385 /*
1386  * Read in the buffer at the given ptr and return the buffer and
1387  * the block pointer within the buffer.
1388  */
1389 int
1390 xfs_btree_read_buf_block(
1391 	struct xfs_btree_cur		*cur,
1392 	const union xfs_btree_ptr	*ptr,
1393 	int				flags,
1394 	struct xfs_btree_block		**block,
1395 	struct xfs_buf			**bpp)
1396 {
1397 	struct xfs_mount	*mp = cur->bc_mp;
1398 	xfs_daddr_t		d;
1399 	int			error;
1400 
1401 	/* need to sort out how callers deal with failures first */
1402 	ASSERT(!(flags & XBF_TRYLOCK));
1403 
1404 	error = xfs_btree_ptr_to_daddr(cur, ptr, &d);
1405 	if (error)
1406 		return error;
1407 	error = xfs_trans_read_buf(mp, cur->bc_tp, xfs_btree_buftarg(cur), d,
1408 			xfs_btree_bbsize(cur), flags, bpp,
1409 			cur->bc_ops->buf_ops);
1410 	if (xfs_metadata_is_sick(error))
1411 		xfs_btree_mark_sick(cur);
1412 	if (error)
1413 		return error;
1414 
1415 	xfs_btree_set_refs(cur, *bpp);
1416 	*block = XFS_BUF_TO_BLOCK(*bpp);
1417 	return 0;
1418 }
1419 
1420 /*
1421  * Copy keys from one btree block to another.
1422  */
1423 void
1424 xfs_btree_copy_keys(
1425 	struct xfs_btree_cur		*cur,
1426 	union xfs_btree_key		*dst_key,
1427 	const union xfs_btree_key	*src_key,
1428 	int				numkeys)
1429 {
1430 	ASSERT(numkeys >= 0);
1431 	memcpy(dst_key, src_key, numkeys * cur->bc_ops->key_len);
1432 }
1433 
1434 /*
1435  * Copy records from one btree block to another.
1436  */
1437 STATIC void
1438 xfs_btree_copy_recs(
1439 	struct xfs_btree_cur	*cur,
1440 	union xfs_btree_rec	*dst_rec,
1441 	union xfs_btree_rec	*src_rec,
1442 	int			numrecs)
1443 {
1444 	ASSERT(numrecs >= 0);
1445 	memcpy(dst_rec, src_rec, numrecs * cur->bc_ops->rec_len);
1446 }
1447 
1448 /*
1449  * Copy block pointers from one btree block to another.
1450  */
1451 void
1452 xfs_btree_copy_ptrs(
1453 	struct xfs_btree_cur	*cur,
1454 	union xfs_btree_ptr	*dst_ptr,
1455 	const union xfs_btree_ptr *src_ptr,
1456 	int			numptrs)
1457 {
1458 	ASSERT(numptrs >= 0);
1459 	memcpy(dst_ptr, src_ptr, numptrs * cur->bc_ops->ptr_len);
1460 }
1461 
1462 /*
1463  * Shift keys one index left/right inside a single btree block.
1464  */
1465 STATIC void
1466 xfs_btree_shift_keys(
1467 	struct xfs_btree_cur	*cur,
1468 	union xfs_btree_key	*key,
1469 	int			dir,
1470 	int			numkeys)
1471 {
1472 	char			*dst_key;
1473 
1474 	ASSERT(numkeys >= 0);
1475 	ASSERT(dir == 1 || dir == -1);
1476 
1477 	dst_key = (char *)key + (dir * cur->bc_ops->key_len);
1478 	memmove(dst_key, key, numkeys * cur->bc_ops->key_len);
1479 }
1480 
1481 /*
1482  * Shift records one index left/right inside a single btree block.
1483  */
1484 STATIC void
1485 xfs_btree_shift_recs(
1486 	struct xfs_btree_cur	*cur,
1487 	union xfs_btree_rec	*rec,
1488 	int			dir,
1489 	int			numrecs)
1490 {
1491 	char			*dst_rec;
1492 
1493 	ASSERT(numrecs >= 0);
1494 	ASSERT(dir == 1 || dir == -1);
1495 
1496 	dst_rec = (char *)rec + (dir * cur->bc_ops->rec_len);
1497 	memmove(dst_rec, rec, numrecs * cur->bc_ops->rec_len);
1498 }
1499 
1500 /*
1501  * Shift block pointers one index left/right inside a single btree block.
1502  */
1503 STATIC void
1504 xfs_btree_shift_ptrs(
1505 	struct xfs_btree_cur	*cur,
1506 	union xfs_btree_ptr	*ptr,
1507 	int			dir,
1508 	int			numptrs)
1509 {
1510 	char			*dst_ptr;
1511 
1512 	ASSERT(numptrs >= 0);
1513 	ASSERT(dir == 1 || dir == -1);
1514 
1515 	dst_ptr = (char *)ptr + (dir * cur->bc_ops->ptr_len);
1516 	memmove(dst_ptr, ptr, numptrs * cur->bc_ops->ptr_len);
1517 }
1518 
1519 /*
1520  * Log key values from the btree block.
1521  */
1522 STATIC void
1523 xfs_btree_log_keys(
1524 	struct xfs_btree_cur	*cur,
1525 	struct xfs_buf		*bp,
1526 	int			first,
1527 	int			last)
1528 {
1529 
1530 	if (bp) {
1531 		xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
1532 		xfs_trans_log_buf(cur->bc_tp, bp,
1533 				  xfs_btree_key_offset(cur, first),
1534 				  xfs_btree_key_offset(cur, last + 1) - 1);
1535 	} else {
1536 		xfs_trans_log_inode(cur->bc_tp, cur->bc_ino.ip,
1537 				xfs_ilog_fbroot(cur->bc_ino.whichfork));
1538 	}
1539 }
1540 
1541 /*
1542  * Log record values from the btree block.
1543  */
1544 void
1545 xfs_btree_log_recs(
1546 	struct xfs_btree_cur	*cur,
1547 	struct xfs_buf		*bp,
1548 	int			first,
1549 	int			last)
1550 {
1551 
1552 	xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
1553 	xfs_trans_log_buf(cur->bc_tp, bp,
1554 			  xfs_btree_rec_offset(cur, first),
1555 			  xfs_btree_rec_offset(cur, last + 1) - 1);
1556 
1557 }
1558 
1559 /*
1560  * Log block pointer fields from a btree block (nonleaf).
1561  */
1562 STATIC void
1563 xfs_btree_log_ptrs(
1564 	struct xfs_btree_cur	*cur,	/* btree cursor */
1565 	struct xfs_buf		*bp,	/* buffer containing btree block */
1566 	int			first,	/* index of first pointer to log */
1567 	int			last)	/* index of last pointer to log */
1568 {
1569 
1570 	if (bp) {
1571 		struct xfs_btree_block	*block = XFS_BUF_TO_BLOCK(bp);
1572 		int			level = xfs_btree_get_level(block);
1573 
1574 		xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
1575 		xfs_trans_log_buf(cur->bc_tp, bp,
1576 				xfs_btree_ptr_offset(cur, first, level),
1577 				xfs_btree_ptr_offset(cur, last + 1, level) - 1);
1578 	} else {
1579 		xfs_trans_log_inode(cur->bc_tp, cur->bc_ino.ip,
1580 			xfs_ilog_fbroot(cur->bc_ino.whichfork));
1581 	}
1582 
1583 }
1584 
1585 /*
1586  * Log fields from a btree block header.
1587  */
1588 void
1589 xfs_btree_log_block(
1590 	struct xfs_btree_cur	*cur,	/* btree cursor */
1591 	struct xfs_buf		*bp,	/* buffer containing btree block */
1592 	uint32_t		fields)	/* mask of fields: XFS_BB_... */
1593 {
1594 	int			first;	/* first byte offset logged */
1595 	int			last;	/* last byte offset logged */
1596 	static const short	soffsets[] = {	/* table of offsets (short) */
1597 		offsetof(struct xfs_btree_block, bb_magic),
1598 		offsetof(struct xfs_btree_block, bb_level),
1599 		offsetof(struct xfs_btree_block, bb_numrecs),
1600 		offsetof(struct xfs_btree_block, bb_u.s.bb_leftsib),
1601 		offsetof(struct xfs_btree_block, bb_u.s.bb_rightsib),
1602 		offsetof(struct xfs_btree_block, bb_u.s.bb_blkno),
1603 		offsetof(struct xfs_btree_block, bb_u.s.bb_lsn),
1604 		offsetof(struct xfs_btree_block, bb_u.s.bb_uuid),
1605 		offsetof(struct xfs_btree_block, bb_u.s.bb_owner),
1606 		offsetof(struct xfs_btree_block, bb_u.s.bb_crc),
1607 		XFS_BTREE_SBLOCK_CRC_LEN
1608 	};
1609 	static const short	loffsets[] = {	/* table of offsets (long) */
1610 		offsetof(struct xfs_btree_block, bb_magic),
1611 		offsetof(struct xfs_btree_block, bb_level),
1612 		offsetof(struct xfs_btree_block, bb_numrecs),
1613 		offsetof(struct xfs_btree_block, bb_u.l.bb_leftsib),
1614 		offsetof(struct xfs_btree_block, bb_u.l.bb_rightsib),
1615 		offsetof(struct xfs_btree_block, bb_u.l.bb_blkno),
1616 		offsetof(struct xfs_btree_block, bb_u.l.bb_lsn),
1617 		offsetof(struct xfs_btree_block, bb_u.l.bb_uuid),
1618 		offsetof(struct xfs_btree_block, bb_u.l.bb_owner),
1619 		offsetof(struct xfs_btree_block, bb_u.l.bb_crc),
1620 		offsetof(struct xfs_btree_block, bb_u.l.bb_pad),
1621 		XFS_BTREE_LBLOCK_CRC_LEN
1622 	};
1623 
1624 	if (bp) {
1625 		int nbits;
1626 
1627 		if (xfs_has_crc(cur->bc_mp)) {
1628 			/*
1629 			 * We don't log the CRC when updating a btree
1630 			 * block but instead recreate it during log
1631 			 * recovery.  As the log buffers have checksums
1632 			 * of their own this is safe and avoids logging a crc
1633 			 * update in a lot of places.
1634 			 */
1635 			if (fields == XFS_BB_ALL_BITS)
1636 				fields = XFS_BB_ALL_BITS_CRC;
1637 			nbits = XFS_BB_NUM_BITS_CRC;
1638 		} else {
1639 			nbits = XFS_BB_NUM_BITS;
1640 		}
1641 		xfs_btree_offsets(fields,
1642 				  (cur->bc_ops->ptr_len == XFS_BTREE_LONG_PTR_LEN) ?
1643 					loffsets : soffsets,
1644 				  nbits, &first, &last);
1645 		xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
1646 		xfs_trans_log_buf(cur->bc_tp, bp, first, last);
1647 	} else {
1648 		xfs_trans_log_inode(cur->bc_tp, cur->bc_ino.ip,
1649 			xfs_ilog_fbroot(cur->bc_ino.whichfork));
1650 	}
1651 }
1652 
1653 /*
1654  * Increment cursor by one record at the level.
1655  * For nonzero levels the leaf-ward information is untouched.
1656  */
1657 int						/* error */
1658 xfs_btree_increment(
1659 	struct xfs_btree_cur	*cur,
1660 	int			level,
1661 	int			*stat)		/* success/failure */
1662 {
1663 	struct xfs_btree_block	*block;
1664 	union xfs_btree_ptr	ptr;
1665 	struct xfs_buf		*bp;
1666 	int			error;		/* error return value */
1667 	int			lev;
1668 
1669 	ASSERT(level < cur->bc_nlevels);
1670 
1671 	/* Read-ahead to the right at this level. */
1672 	xfs_btree_readahead(cur, level, XFS_BTCUR_RIGHTRA);
1673 
1674 	/* Get a pointer to the btree block. */
1675 	block = xfs_btree_get_block(cur, level, &bp);
1676 
1677 #ifdef DEBUG
1678 	error = xfs_btree_check_block(cur, block, level, bp);
1679 	if (error)
1680 		goto error0;
1681 #endif
1682 
1683 	/* We're done if we remain in the block after the increment. */
1684 	if (++cur->bc_levels[level].ptr <= xfs_btree_get_numrecs(block))
1685 		goto out1;
1686 
1687 	/* Fail if we just went off the right edge of the tree. */
1688 	xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
1689 	if (xfs_btree_ptr_is_null(cur, &ptr))
1690 		goto out0;
1691 
1692 	XFS_BTREE_STATS_INC(cur, increment);
1693 
1694 	/*
1695 	 * March up the tree incrementing pointers.
1696 	 * Stop when we don't go off the right edge of a block.
1697 	 */
1698 	for (lev = level + 1; lev < cur->bc_nlevels; lev++) {
1699 		block = xfs_btree_get_block(cur, lev, &bp);
1700 
1701 #ifdef DEBUG
1702 		error = xfs_btree_check_block(cur, block, lev, bp);
1703 		if (error)
1704 			goto error0;
1705 #endif
1706 
1707 		if (++cur->bc_levels[lev].ptr <= xfs_btree_get_numrecs(block))
1708 			break;
1709 
1710 		/* Read-ahead the right block for the next loop. */
1711 		xfs_btree_readahead(cur, lev, XFS_BTCUR_RIGHTRA);
1712 	}
1713 
1714 	/*
1715 	 * If we went off the root then we are either seriously
1716 	 * confused or have the tree root in an inode.
1717 	 */
1718 	if (lev == cur->bc_nlevels) {
1719 		if (cur->bc_ops->type == XFS_BTREE_TYPE_INODE)
1720 			goto out0;
1721 		ASSERT(0);
1722 		xfs_btree_mark_sick(cur);
1723 		error = -EFSCORRUPTED;
1724 		goto error0;
1725 	}
1726 	ASSERT(lev < cur->bc_nlevels);
1727 
1728 	/*
1729 	 * Now walk back down the tree, fixing up the cursor's buffer
1730 	 * pointers and key numbers.
1731 	 */
1732 	for (block = xfs_btree_get_block(cur, lev, &bp); lev > level; ) {
1733 		union xfs_btree_ptr	*ptrp;
1734 
1735 		ptrp = xfs_btree_ptr_addr(cur, cur->bc_levels[lev].ptr, block);
1736 		--lev;
1737 		error = xfs_btree_read_buf_block(cur, ptrp, 0, &block, &bp);
1738 		if (error)
1739 			goto error0;
1740 
1741 		xfs_btree_setbuf(cur, lev, bp);
1742 		cur->bc_levels[lev].ptr = 1;
1743 	}
1744 out1:
1745 	*stat = 1;
1746 	return 0;
1747 
1748 out0:
1749 	*stat = 0;
1750 	return 0;
1751 
1752 error0:
1753 	return error;
1754 }
1755 
1756 /*
1757  * Decrement cursor by one record at the level.
1758  * For nonzero levels the leaf-ward information is untouched.
1759  */
1760 int						/* error */
1761 xfs_btree_decrement(
1762 	struct xfs_btree_cur	*cur,
1763 	int			level,
1764 	int			*stat)		/* success/failure */
1765 {
1766 	struct xfs_btree_block	*block;
1767 	struct xfs_buf		*bp;
1768 	int			error;		/* error return value */
1769 	int			lev;
1770 	union xfs_btree_ptr	ptr;
1771 
1772 	ASSERT(level < cur->bc_nlevels);
1773 
1774 	/* Read-ahead to the left at this level. */
1775 	xfs_btree_readahead(cur, level, XFS_BTCUR_LEFTRA);
1776 
1777 	/* We're done if we remain in the block after the decrement. */
1778 	if (--cur->bc_levels[level].ptr > 0)
1779 		goto out1;
1780 
1781 	/* Get a pointer to the btree block. */
1782 	block = xfs_btree_get_block(cur, level, &bp);
1783 
1784 #ifdef DEBUG
1785 	error = xfs_btree_check_block(cur, block, level, bp);
1786 	if (error)
1787 		goto error0;
1788 #endif
1789 
1790 	/* Fail if we just went off the left edge of the tree. */
1791 	xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_LEFTSIB);
1792 	if (xfs_btree_ptr_is_null(cur, &ptr))
1793 		goto out0;
1794 
1795 	XFS_BTREE_STATS_INC(cur, decrement);
1796 
1797 	/*
1798 	 * March up the tree decrementing pointers.
1799 	 * Stop when we don't go off the left edge of a block.
1800 	 */
1801 	for (lev = level + 1; lev < cur->bc_nlevels; lev++) {
1802 		if (--cur->bc_levels[lev].ptr > 0)
1803 			break;
1804 		/* Read-ahead the left block for the next loop. */
1805 		xfs_btree_readahead(cur, lev, XFS_BTCUR_LEFTRA);
1806 	}
1807 
1808 	/*
1809 	 * If we went off the root then we are seriously confused.
1810 	 * or the root of the tree is in an inode.
1811 	 */
1812 	if (lev == cur->bc_nlevels) {
1813 		if (cur->bc_ops->type == XFS_BTREE_TYPE_INODE)
1814 			goto out0;
1815 		ASSERT(0);
1816 		xfs_btree_mark_sick(cur);
1817 		error = -EFSCORRUPTED;
1818 		goto error0;
1819 	}
1820 	ASSERT(lev < cur->bc_nlevels);
1821 
1822 	/*
1823 	 * Now walk back down the tree, fixing up the cursor's buffer
1824 	 * pointers and key numbers.
1825 	 */
1826 	for (block = xfs_btree_get_block(cur, lev, &bp); lev > level; ) {
1827 		union xfs_btree_ptr	*ptrp;
1828 
1829 		ptrp = xfs_btree_ptr_addr(cur, cur->bc_levels[lev].ptr, block);
1830 		--lev;
1831 		error = xfs_btree_read_buf_block(cur, ptrp, 0, &block, &bp);
1832 		if (error)
1833 			goto error0;
1834 		xfs_btree_setbuf(cur, lev, bp);
1835 		cur->bc_levels[lev].ptr = xfs_btree_get_numrecs(block);
1836 	}
1837 out1:
1838 	*stat = 1;
1839 	return 0;
1840 
1841 out0:
1842 	*stat = 0;
1843 	return 0;
1844 
1845 error0:
1846 	return error;
1847 }
1848 
1849 /*
1850  * Check the btree block owner now that we have the context to know who the
1851  * real owner is.
1852  */
1853 static inline xfs_failaddr_t
1854 xfs_btree_check_block_owner(
1855 	struct xfs_btree_cur	*cur,
1856 	struct xfs_btree_block	*block)
1857 {
1858 	__u64			owner;
1859 
1860 	if (!xfs_has_crc(cur->bc_mp) ||
1861 	    (cur->bc_flags & XFS_BTREE_BMBT_INVALID_OWNER))
1862 		return NULL;
1863 
1864 	owner = xfs_btree_owner(cur);
1865 	if (cur->bc_ops->ptr_len == XFS_BTREE_LONG_PTR_LEN) {
1866 		if (be64_to_cpu(block->bb_u.l.bb_owner) != owner)
1867 			return __this_address;
1868 	} else {
1869 		if (be32_to_cpu(block->bb_u.s.bb_owner) != owner)
1870 			return __this_address;
1871 	}
1872 
1873 	return NULL;
1874 }
1875 
1876 int
1877 xfs_btree_lookup_get_block(
1878 	struct xfs_btree_cur		*cur,	/* btree cursor */
1879 	int				level,	/* level in the btree */
1880 	const union xfs_btree_ptr	*pp,	/* ptr to btree block */
1881 	struct xfs_btree_block		**blkp) /* return btree block */
1882 {
1883 	struct xfs_buf		*bp;	/* buffer pointer for btree block */
1884 	xfs_daddr_t		daddr;
1885 	int			error = 0;
1886 
1887 	/* special case the root block if in an inode */
1888 	if (xfs_btree_at_iroot(cur, level)) {
1889 		*blkp = xfs_btree_get_iroot(cur);
1890 		return 0;
1891 	}
1892 
1893 	/*
1894 	 * If the old buffer at this level for the disk address we are
1895 	 * looking for re-use it.
1896 	 *
1897 	 * Otherwise throw it away and get a new one.
1898 	 */
1899 	bp = cur->bc_levels[level].bp;
1900 	error = xfs_btree_ptr_to_daddr(cur, pp, &daddr);
1901 	if (error)
1902 		return error;
1903 	if (bp && xfs_buf_daddr(bp) == daddr) {
1904 		*blkp = XFS_BUF_TO_BLOCK(bp);
1905 		return 0;
1906 	}
1907 
1908 	error = xfs_btree_read_buf_block(cur, pp, 0, blkp, &bp);
1909 	if (error)
1910 		return error;
1911 
1912 	/* Check the inode owner since the verifiers don't. */
1913 	if (xfs_btree_check_block_owner(cur, *blkp) != NULL)
1914 		goto out_bad;
1915 
1916 	/* Did we get the level we were looking for? */
1917 	if (be16_to_cpu((*blkp)->bb_level) != level)
1918 		goto out_bad;
1919 
1920 	/* Check that internal nodes have at least one record. */
1921 	if (level != 0 && be16_to_cpu((*blkp)->bb_numrecs) == 0)
1922 		goto out_bad;
1923 
1924 	xfs_btree_setbuf(cur, level, bp);
1925 	return 0;
1926 
1927 out_bad:
1928 	*blkp = NULL;
1929 	xfs_buf_mark_corrupt(bp);
1930 	xfs_trans_brelse(cur->bc_tp, bp);
1931 	xfs_btree_mark_sick(cur);
1932 	return -EFSCORRUPTED;
1933 }
1934 
1935 /*
1936  * Get current search key.  For level 0 we don't actually have a key
1937  * structure so we make one up from the record.  For all other levels
1938  * we just return the right key.
1939  */
1940 STATIC union xfs_btree_key *
1941 xfs_lookup_get_search_key(
1942 	struct xfs_btree_cur	*cur,
1943 	int			level,
1944 	int			keyno,
1945 	struct xfs_btree_block	*block,
1946 	union xfs_btree_key	*kp)
1947 {
1948 	if (level == 0) {
1949 		cur->bc_ops->init_key_from_rec(kp,
1950 				xfs_btree_rec_addr(cur, keyno, block));
1951 		return kp;
1952 	}
1953 
1954 	return xfs_btree_key_addr(cur, keyno, block);
1955 }
1956 
1957 /*
1958  * Initialize a pointer to the root block.
1959  */
1960 void
1961 xfs_btree_init_ptr_from_cur(
1962 	struct xfs_btree_cur	*cur,
1963 	union xfs_btree_ptr	*ptr)
1964 {
1965 	if (cur->bc_ops->type == XFS_BTREE_TYPE_INODE) {
1966 		/*
1967 		 * Inode-rooted btrees call xfs_btree_get_iroot to find the root
1968 		 * in xfs_btree_lookup_get_block and don't need a pointer here.
1969 		 */
1970 		ptr->l = 0;
1971 	} else if (cur->bc_flags & XFS_BTREE_STAGING) {
1972 		ptr->s = cpu_to_be32(cur->bc_ag.afake->af_root);
1973 	} else {
1974 		cur->bc_ops->init_ptr_from_cur(cur, ptr);
1975 	}
1976 }
1977 
1978 /*
1979  * Lookup the record.  The cursor is made to point to it, based on dir.
1980  * stat is set to 0 if can't find any such record, 1 for success.
1981  */
1982 int					/* error */
1983 xfs_btree_lookup(
1984 	struct xfs_btree_cur	*cur,	/* btree cursor */
1985 	xfs_lookup_t		dir,	/* <=, ==, or >= */
1986 	int			*stat)	/* success/failure */
1987 {
1988 	struct xfs_btree_block	*block;	/* current btree block */
1989 	int64_t			diff;	/* difference for the current key */
1990 	int			error;	/* error return value */
1991 	int			keyno;	/* current key number */
1992 	int			level;	/* level in the btree */
1993 	union xfs_btree_ptr	*pp;	/* ptr to btree block */
1994 	union xfs_btree_ptr	ptr;	/* ptr to btree block */
1995 
1996 	XFS_BTREE_STATS_INC(cur, lookup);
1997 
1998 	/* No such thing as a zero-level tree. */
1999 	if (XFS_IS_CORRUPT(cur->bc_mp, cur->bc_nlevels == 0)) {
2000 		xfs_btree_mark_sick(cur);
2001 		return -EFSCORRUPTED;
2002 	}
2003 
2004 	block = NULL;
2005 	keyno = 0;
2006 
2007 	/* initialise start pointer from cursor */
2008 	xfs_btree_init_ptr_from_cur(cur, &ptr);
2009 	pp = &ptr;
2010 
2011 	/*
2012 	 * Iterate over each level in the btree, starting at the root.
2013 	 * For each level above the leaves, find the key we need, based
2014 	 * on the lookup record, then follow the corresponding block
2015 	 * pointer down to the next level.
2016 	 */
2017 	for (level = cur->bc_nlevels - 1, diff = 1; level >= 0; level--) {
2018 		/* Get the block we need to do the lookup on. */
2019 		error = xfs_btree_lookup_get_block(cur, level, pp, &block);
2020 		if (error)
2021 			goto error0;
2022 
2023 		if (diff == 0) {
2024 			/*
2025 			 * If we already had a key match at a higher level, we
2026 			 * know we need to use the first entry in this block.
2027 			 */
2028 			keyno = 1;
2029 		} else {
2030 			/* Otherwise search this block. Do a binary search. */
2031 
2032 			int	high;	/* high entry number */
2033 			int	low;	/* low entry number */
2034 
2035 			/* Set low and high entry numbers, 1-based. */
2036 			low = 1;
2037 			high = xfs_btree_get_numrecs(block);
2038 			if (!high) {
2039 				/* Block is empty, must be an empty leaf. */
2040 				if (level != 0 || cur->bc_nlevels != 1) {
2041 					XFS_CORRUPTION_ERROR(__func__,
2042 							XFS_ERRLEVEL_LOW,
2043 							cur->bc_mp, block,
2044 							sizeof(*block));
2045 					xfs_btree_mark_sick(cur);
2046 					return -EFSCORRUPTED;
2047 				}
2048 
2049 				cur->bc_levels[0].ptr = dir != XFS_LOOKUP_LE;
2050 				*stat = 0;
2051 				return 0;
2052 			}
2053 
2054 			/* Binary search the block. */
2055 			while (low <= high) {
2056 				union xfs_btree_key	key;
2057 				union xfs_btree_key	*kp;
2058 
2059 				XFS_BTREE_STATS_INC(cur, compare);
2060 
2061 				/* keyno is average of low and high. */
2062 				keyno = (low + high) >> 1;
2063 
2064 				/* Get current search key */
2065 				kp = xfs_lookup_get_search_key(cur, level,
2066 						keyno, block, &key);
2067 
2068 				/*
2069 				 * Compute difference to get next direction:
2070 				 *  - less than, move right
2071 				 *  - greater than, move left
2072 				 *  - equal, we're done
2073 				 */
2074 				diff = cur->bc_ops->key_diff(cur, kp);
2075 				if (diff < 0)
2076 					low = keyno + 1;
2077 				else if (diff > 0)
2078 					high = keyno - 1;
2079 				else
2080 					break;
2081 			}
2082 		}
2083 
2084 		/*
2085 		 * If there are more levels, set up for the next level
2086 		 * by getting the block number and filling in the cursor.
2087 		 */
2088 		if (level > 0) {
2089 			/*
2090 			 * If we moved left, need the previous key number,
2091 			 * unless there isn't one.
2092 			 */
2093 			if (diff > 0 && --keyno < 1)
2094 				keyno = 1;
2095 			pp = xfs_btree_ptr_addr(cur, keyno, block);
2096 
2097 			error = xfs_btree_debug_check_ptr(cur, pp, 0, level);
2098 			if (error)
2099 				goto error0;
2100 
2101 			cur->bc_levels[level].ptr = keyno;
2102 		}
2103 	}
2104 
2105 	/* Done with the search. See if we need to adjust the results. */
2106 	if (dir != XFS_LOOKUP_LE && diff < 0) {
2107 		keyno++;
2108 		/*
2109 		 * If ge search and we went off the end of the block, but it's
2110 		 * not the last block, we're in the wrong block.
2111 		 */
2112 		xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
2113 		if (dir == XFS_LOOKUP_GE &&
2114 		    keyno > xfs_btree_get_numrecs(block) &&
2115 		    !xfs_btree_ptr_is_null(cur, &ptr)) {
2116 			int	i;
2117 
2118 			cur->bc_levels[0].ptr = keyno;
2119 			error = xfs_btree_increment(cur, 0, &i);
2120 			if (error)
2121 				goto error0;
2122 			if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
2123 				xfs_btree_mark_sick(cur);
2124 				return -EFSCORRUPTED;
2125 			}
2126 			*stat = 1;
2127 			return 0;
2128 		}
2129 	} else if (dir == XFS_LOOKUP_LE && diff > 0)
2130 		keyno--;
2131 	cur->bc_levels[0].ptr = keyno;
2132 
2133 	/* Return if we succeeded or not. */
2134 	if (keyno == 0 || keyno > xfs_btree_get_numrecs(block))
2135 		*stat = 0;
2136 	else if (dir != XFS_LOOKUP_EQ || diff == 0)
2137 		*stat = 1;
2138 	else
2139 		*stat = 0;
2140 	return 0;
2141 
2142 error0:
2143 	return error;
2144 }
2145 
2146 /* Find the high key storage area from a regular key. */
2147 union xfs_btree_key *
2148 xfs_btree_high_key_from_key(
2149 	struct xfs_btree_cur	*cur,
2150 	union xfs_btree_key	*key)
2151 {
2152 	ASSERT(cur->bc_ops->geom_flags & XFS_BTGEO_OVERLAPPING);
2153 	return (union xfs_btree_key *)((char *)key +
2154 			(cur->bc_ops->key_len / 2));
2155 }
2156 
2157 /* Determine the low (and high if overlapped) keys of a leaf block */
2158 STATIC void
2159 xfs_btree_get_leaf_keys(
2160 	struct xfs_btree_cur	*cur,
2161 	struct xfs_btree_block	*block,
2162 	union xfs_btree_key	*key)
2163 {
2164 	union xfs_btree_key	max_hkey;
2165 	union xfs_btree_key	hkey;
2166 	union xfs_btree_rec	*rec;
2167 	union xfs_btree_key	*high;
2168 	int			n;
2169 
2170 	rec = xfs_btree_rec_addr(cur, 1, block);
2171 	cur->bc_ops->init_key_from_rec(key, rec);
2172 
2173 	if (cur->bc_ops->geom_flags & XFS_BTGEO_OVERLAPPING) {
2174 
2175 		cur->bc_ops->init_high_key_from_rec(&max_hkey, rec);
2176 		for (n = 2; n <= xfs_btree_get_numrecs(block); n++) {
2177 			rec = xfs_btree_rec_addr(cur, n, block);
2178 			cur->bc_ops->init_high_key_from_rec(&hkey, rec);
2179 			if (xfs_btree_keycmp_gt(cur, &hkey, &max_hkey))
2180 				max_hkey = hkey;
2181 		}
2182 
2183 		high = xfs_btree_high_key_from_key(cur, key);
2184 		memcpy(high, &max_hkey, cur->bc_ops->key_len / 2);
2185 	}
2186 }
2187 
2188 /* Determine the low (and high if overlapped) keys of a node block */
2189 STATIC void
2190 xfs_btree_get_node_keys(
2191 	struct xfs_btree_cur	*cur,
2192 	struct xfs_btree_block	*block,
2193 	union xfs_btree_key	*key)
2194 {
2195 	union xfs_btree_key	*hkey;
2196 	union xfs_btree_key	*max_hkey;
2197 	union xfs_btree_key	*high;
2198 	int			n;
2199 
2200 	if (cur->bc_ops->geom_flags & XFS_BTGEO_OVERLAPPING) {
2201 		memcpy(key, xfs_btree_key_addr(cur, 1, block),
2202 				cur->bc_ops->key_len / 2);
2203 
2204 		max_hkey = xfs_btree_high_key_addr(cur, 1, block);
2205 		for (n = 2; n <= xfs_btree_get_numrecs(block); n++) {
2206 			hkey = xfs_btree_high_key_addr(cur, n, block);
2207 			if (xfs_btree_keycmp_gt(cur, hkey, max_hkey))
2208 				max_hkey = hkey;
2209 		}
2210 
2211 		high = xfs_btree_high_key_from_key(cur, key);
2212 		memcpy(high, max_hkey, cur->bc_ops->key_len / 2);
2213 	} else {
2214 		memcpy(key, xfs_btree_key_addr(cur, 1, block),
2215 				cur->bc_ops->key_len);
2216 	}
2217 }
2218 
2219 /* Derive the keys for any btree block. */
2220 void
2221 xfs_btree_get_keys(
2222 	struct xfs_btree_cur	*cur,
2223 	struct xfs_btree_block	*block,
2224 	union xfs_btree_key	*key)
2225 {
2226 	if (be16_to_cpu(block->bb_level) == 0)
2227 		xfs_btree_get_leaf_keys(cur, block, key);
2228 	else
2229 		xfs_btree_get_node_keys(cur, block, key);
2230 }
2231 
2232 /*
2233  * Decide if we need to update the parent keys of a btree block.  For
2234  * a standard btree this is only necessary if we're updating the first
2235  * record/key.  For an overlapping btree, we must always update the
2236  * keys because the highest key can be in any of the records or keys
2237  * in the block.
2238  */
2239 static inline bool
2240 xfs_btree_needs_key_update(
2241 	struct xfs_btree_cur	*cur,
2242 	int			ptr)
2243 {
2244 	return (cur->bc_ops->geom_flags & XFS_BTGEO_OVERLAPPING) || ptr == 1;
2245 }
2246 
2247 /*
2248  * Update the low and high parent keys of the given level, progressing
2249  * towards the root.  If force_all is false, stop if the keys for a given
2250  * level do not need updating.
2251  */
2252 STATIC int
2253 __xfs_btree_updkeys(
2254 	struct xfs_btree_cur	*cur,
2255 	int			level,
2256 	struct xfs_btree_block	*block,
2257 	struct xfs_buf		*bp0,
2258 	bool			force_all)
2259 {
2260 	union xfs_btree_key	key;	/* keys from current level */
2261 	union xfs_btree_key	*lkey;	/* keys from the next level up */
2262 	union xfs_btree_key	*hkey;
2263 	union xfs_btree_key	*nlkey;	/* keys from the next level up */
2264 	union xfs_btree_key	*nhkey;
2265 	struct xfs_buf		*bp;
2266 	int			ptr;
2267 
2268 	ASSERT(cur->bc_ops->geom_flags & XFS_BTGEO_OVERLAPPING);
2269 
2270 	/* Exit if there aren't any parent levels to update. */
2271 	if (level + 1 >= cur->bc_nlevels)
2272 		return 0;
2273 
2274 	trace_xfs_btree_updkeys(cur, level, bp0);
2275 
2276 	lkey = &key;
2277 	hkey = xfs_btree_high_key_from_key(cur, lkey);
2278 	xfs_btree_get_keys(cur, block, lkey);
2279 	for (level++; level < cur->bc_nlevels; level++) {
2280 #ifdef DEBUG
2281 		int		error;
2282 #endif
2283 		block = xfs_btree_get_block(cur, level, &bp);
2284 		trace_xfs_btree_updkeys(cur, level, bp);
2285 #ifdef DEBUG
2286 		error = xfs_btree_check_block(cur, block, level, bp);
2287 		if (error)
2288 			return error;
2289 #endif
2290 		ptr = cur->bc_levels[level].ptr;
2291 		nlkey = xfs_btree_key_addr(cur, ptr, block);
2292 		nhkey = xfs_btree_high_key_addr(cur, ptr, block);
2293 		if (!force_all &&
2294 		    xfs_btree_keycmp_eq(cur, nlkey, lkey) &&
2295 		    xfs_btree_keycmp_eq(cur, nhkey, hkey))
2296 			break;
2297 		xfs_btree_copy_keys(cur, nlkey, lkey, 1);
2298 		xfs_btree_log_keys(cur, bp, ptr, ptr);
2299 		if (level + 1 >= cur->bc_nlevels)
2300 			break;
2301 		xfs_btree_get_node_keys(cur, block, lkey);
2302 	}
2303 
2304 	return 0;
2305 }
2306 
2307 /* Update all the keys from some level in cursor back to the root. */
2308 STATIC int
2309 xfs_btree_updkeys_force(
2310 	struct xfs_btree_cur	*cur,
2311 	int			level)
2312 {
2313 	struct xfs_buf		*bp;
2314 	struct xfs_btree_block	*block;
2315 
2316 	block = xfs_btree_get_block(cur, level, &bp);
2317 	return __xfs_btree_updkeys(cur, level, block, bp, true);
2318 }
2319 
2320 /*
2321  * Update the parent keys of the given level, progressing towards the root.
2322  */
2323 STATIC int
2324 xfs_btree_update_keys(
2325 	struct xfs_btree_cur	*cur,
2326 	int			level)
2327 {
2328 	struct xfs_btree_block	*block;
2329 	struct xfs_buf		*bp;
2330 	union xfs_btree_key	*kp;
2331 	union xfs_btree_key	key;
2332 	int			ptr;
2333 
2334 	ASSERT(level >= 0);
2335 
2336 	block = xfs_btree_get_block(cur, level, &bp);
2337 	if (cur->bc_ops->geom_flags & XFS_BTGEO_OVERLAPPING)
2338 		return __xfs_btree_updkeys(cur, level, block, bp, false);
2339 
2340 	/*
2341 	 * Go up the tree from this level toward the root.
2342 	 * At each level, update the key value to the value input.
2343 	 * Stop when we reach a level where the cursor isn't pointing
2344 	 * at the first entry in the block.
2345 	 */
2346 	xfs_btree_get_keys(cur, block, &key);
2347 	for (level++, ptr = 1; ptr == 1 && level < cur->bc_nlevels; level++) {
2348 #ifdef DEBUG
2349 		int		error;
2350 #endif
2351 		block = xfs_btree_get_block(cur, level, &bp);
2352 #ifdef DEBUG
2353 		error = xfs_btree_check_block(cur, block, level, bp);
2354 		if (error)
2355 			return error;
2356 #endif
2357 		ptr = cur->bc_levels[level].ptr;
2358 		kp = xfs_btree_key_addr(cur, ptr, block);
2359 		xfs_btree_copy_keys(cur, kp, &key, 1);
2360 		xfs_btree_log_keys(cur, bp, ptr, ptr);
2361 	}
2362 
2363 	return 0;
2364 }
2365 
2366 /*
2367  * Update the record referred to by cur to the value in the
2368  * given record. This either works (return 0) or gets an
2369  * EFSCORRUPTED error.
2370  */
2371 int
2372 xfs_btree_update(
2373 	struct xfs_btree_cur	*cur,
2374 	union xfs_btree_rec	*rec)
2375 {
2376 	struct xfs_btree_block	*block;
2377 	struct xfs_buf		*bp;
2378 	int			error;
2379 	int			ptr;
2380 	union xfs_btree_rec	*rp;
2381 
2382 	/* Pick up the current block. */
2383 	block = xfs_btree_get_block(cur, 0, &bp);
2384 
2385 #ifdef DEBUG
2386 	error = xfs_btree_check_block(cur, block, 0, bp);
2387 	if (error)
2388 		goto error0;
2389 #endif
2390 	/* Get the address of the rec to be updated. */
2391 	ptr = cur->bc_levels[0].ptr;
2392 	rp = xfs_btree_rec_addr(cur, ptr, block);
2393 
2394 	/* Fill in the new contents and log them. */
2395 	xfs_btree_copy_recs(cur, rp, rec, 1);
2396 	xfs_btree_log_recs(cur, bp, ptr, ptr);
2397 
2398 	/* Pass new key value up to our parent. */
2399 	if (xfs_btree_needs_key_update(cur, ptr)) {
2400 		error = xfs_btree_update_keys(cur, 0);
2401 		if (error)
2402 			goto error0;
2403 	}
2404 
2405 	return 0;
2406 
2407 error0:
2408 	return error;
2409 }
2410 
2411 /*
2412  * Move 1 record left from cur/level if possible.
2413  * Update cur to reflect the new path.
2414  */
2415 STATIC int					/* error */
2416 xfs_btree_lshift(
2417 	struct xfs_btree_cur	*cur,
2418 	int			level,
2419 	int			*stat)		/* success/failure */
2420 {
2421 	struct xfs_buf		*lbp;		/* left buffer pointer */
2422 	struct xfs_btree_block	*left;		/* left btree block */
2423 	int			lrecs;		/* left record count */
2424 	struct xfs_buf		*rbp;		/* right buffer pointer */
2425 	struct xfs_btree_block	*right;		/* right btree block */
2426 	struct xfs_btree_cur	*tcur;		/* temporary btree cursor */
2427 	int			rrecs;		/* right record count */
2428 	union xfs_btree_ptr	lptr;		/* left btree pointer */
2429 	union xfs_btree_key	*rkp = NULL;	/* right btree key */
2430 	union xfs_btree_ptr	*rpp = NULL;	/* right address pointer */
2431 	union xfs_btree_rec	*rrp = NULL;	/* right record pointer */
2432 	int			error;		/* error return value */
2433 	int			i;
2434 
2435 	if (xfs_btree_at_iroot(cur, level))
2436 		goto out0;
2437 
2438 	/* Set up variables for this block as "right". */
2439 	right = xfs_btree_get_block(cur, level, &rbp);
2440 
2441 #ifdef DEBUG
2442 	error = xfs_btree_check_block(cur, right, level, rbp);
2443 	if (error)
2444 		goto error0;
2445 #endif
2446 
2447 	/* If we've got no left sibling then we can't shift an entry left. */
2448 	xfs_btree_get_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
2449 	if (xfs_btree_ptr_is_null(cur, &lptr))
2450 		goto out0;
2451 
2452 	/*
2453 	 * If the cursor entry is the one that would be moved, don't
2454 	 * do it... it's too complicated.
2455 	 */
2456 	if (cur->bc_levels[level].ptr <= 1)
2457 		goto out0;
2458 
2459 	/* Set up the left neighbor as "left". */
2460 	error = xfs_btree_read_buf_block(cur, &lptr, 0, &left, &lbp);
2461 	if (error)
2462 		goto error0;
2463 
2464 	/* If it's full, it can't take another entry. */
2465 	lrecs = xfs_btree_get_numrecs(left);
2466 	if (lrecs == cur->bc_ops->get_maxrecs(cur, level))
2467 		goto out0;
2468 
2469 	rrecs = xfs_btree_get_numrecs(right);
2470 
2471 	/*
2472 	 * We add one entry to the left side and remove one for the right side.
2473 	 * Account for it here, the changes will be updated on disk and logged
2474 	 * later.
2475 	 */
2476 	lrecs++;
2477 	rrecs--;
2478 
2479 	XFS_BTREE_STATS_INC(cur, lshift);
2480 	XFS_BTREE_STATS_ADD(cur, moves, 1);
2481 
2482 	/*
2483 	 * If non-leaf, copy a key and a ptr to the left block.
2484 	 * Log the changes to the left block.
2485 	 */
2486 	if (level > 0) {
2487 		/* It's a non-leaf.  Move keys and pointers. */
2488 		union xfs_btree_key	*lkp;	/* left btree key */
2489 		union xfs_btree_ptr	*lpp;	/* left address pointer */
2490 
2491 		lkp = xfs_btree_key_addr(cur, lrecs, left);
2492 		rkp = xfs_btree_key_addr(cur, 1, right);
2493 
2494 		lpp = xfs_btree_ptr_addr(cur, lrecs, left);
2495 		rpp = xfs_btree_ptr_addr(cur, 1, right);
2496 
2497 		error = xfs_btree_debug_check_ptr(cur, rpp, 0, level);
2498 		if (error)
2499 			goto error0;
2500 
2501 		xfs_btree_copy_keys(cur, lkp, rkp, 1);
2502 		xfs_btree_copy_ptrs(cur, lpp, rpp, 1);
2503 
2504 		xfs_btree_log_keys(cur, lbp, lrecs, lrecs);
2505 		xfs_btree_log_ptrs(cur, lbp, lrecs, lrecs);
2506 
2507 		ASSERT(cur->bc_ops->keys_inorder(cur,
2508 			xfs_btree_key_addr(cur, lrecs - 1, left), lkp));
2509 	} else {
2510 		/* It's a leaf.  Move records.  */
2511 		union xfs_btree_rec	*lrp;	/* left record pointer */
2512 
2513 		lrp = xfs_btree_rec_addr(cur, lrecs, left);
2514 		rrp = xfs_btree_rec_addr(cur, 1, right);
2515 
2516 		xfs_btree_copy_recs(cur, lrp, rrp, 1);
2517 		xfs_btree_log_recs(cur, lbp, lrecs, lrecs);
2518 
2519 		ASSERT(cur->bc_ops->recs_inorder(cur,
2520 			xfs_btree_rec_addr(cur, lrecs - 1, left), lrp));
2521 	}
2522 
2523 	xfs_btree_set_numrecs(left, lrecs);
2524 	xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS);
2525 
2526 	xfs_btree_set_numrecs(right, rrecs);
2527 	xfs_btree_log_block(cur, rbp, XFS_BB_NUMRECS);
2528 
2529 	/*
2530 	 * Slide the contents of right down one entry.
2531 	 */
2532 	XFS_BTREE_STATS_ADD(cur, moves, rrecs - 1);
2533 	if (level > 0) {
2534 		/* It's a nonleaf. operate on keys and ptrs */
2535 		for (i = 0; i < rrecs; i++) {
2536 			error = xfs_btree_debug_check_ptr(cur, rpp, i + 1, level);
2537 			if (error)
2538 				goto error0;
2539 		}
2540 
2541 		xfs_btree_shift_keys(cur,
2542 				xfs_btree_key_addr(cur, 2, right),
2543 				-1, rrecs);
2544 		xfs_btree_shift_ptrs(cur,
2545 				xfs_btree_ptr_addr(cur, 2, right),
2546 				-1, rrecs);
2547 
2548 		xfs_btree_log_keys(cur, rbp, 1, rrecs);
2549 		xfs_btree_log_ptrs(cur, rbp, 1, rrecs);
2550 	} else {
2551 		/* It's a leaf. operate on records */
2552 		xfs_btree_shift_recs(cur,
2553 			xfs_btree_rec_addr(cur, 2, right),
2554 			-1, rrecs);
2555 		xfs_btree_log_recs(cur, rbp, 1, rrecs);
2556 	}
2557 
2558 	/*
2559 	 * Using a temporary cursor, update the parent key values of the
2560 	 * block on the left.
2561 	 */
2562 	if (cur->bc_ops->geom_flags & XFS_BTGEO_OVERLAPPING) {
2563 		error = xfs_btree_dup_cursor(cur, &tcur);
2564 		if (error)
2565 			goto error0;
2566 		i = xfs_btree_firstrec(tcur, level);
2567 		if (XFS_IS_CORRUPT(tcur->bc_mp, i != 1)) {
2568 			xfs_btree_mark_sick(cur);
2569 			error = -EFSCORRUPTED;
2570 			goto error0;
2571 		}
2572 
2573 		error = xfs_btree_decrement(tcur, level, &i);
2574 		if (error)
2575 			goto error1;
2576 
2577 		/* Update the parent high keys of the left block, if needed. */
2578 		error = xfs_btree_update_keys(tcur, level);
2579 		if (error)
2580 			goto error1;
2581 
2582 		xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
2583 	}
2584 
2585 	/* Update the parent keys of the right block. */
2586 	error = xfs_btree_update_keys(cur, level);
2587 	if (error)
2588 		goto error0;
2589 
2590 	/* Slide the cursor value left one. */
2591 	cur->bc_levels[level].ptr--;
2592 
2593 	*stat = 1;
2594 	return 0;
2595 
2596 out0:
2597 	*stat = 0;
2598 	return 0;
2599 
2600 error0:
2601 	return error;
2602 
2603 error1:
2604 	xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
2605 	return error;
2606 }
2607 
2608 /*
2609  * Move 1 record right from cur/level if possible.
2610  * Update cur to reflect the new path.
2611  */
2612 STATIC int					/* error */
2613 xfs_btree_rshift(
2614 	struct xfs_btree_cur	*cur,
2615 	int			level,
2616 	int			*stat)		/* success/failure */
2617 {
2618 	struct xfs_buf		*lbp;		/* left buffer pointer */
2619 	struct xfs_btree_block	*left;		/* left btree block */
2620 	struct xfs_buf		*rbp;		/* right buffer pointer */
2621 	struct xfs_btree_block	*right;		/* right btree block */
2622 	struct xfs_btree_cur	*tcur;		/* temporary btree cursor */
2623 	union xfs_btree_ptr	rptr;		/* right block pointer */
2624 	union xfs_btree_key	*rkp;		/* right btree key */
2625 	int			rrecs;		/* right record count */
2626 	int			lrecs;		/* left record count */
2627 	int			error;		/* error return value */
2628 	int			i;		/* loop counter */
2629 
2630 	if (xfs_btree_at_iroot(cur, level))
2631 		goto out0;
2632 
2633 	/* Set up variables for this block as "left". */
2634 	left = xfs_btree_get_block(cur, level, &lbp);
2635 
2636 #ifdef DEBUG
2637 	error = xfs_btree_check_block(cur, left, level, lbp);
2638 	if (error)
2639 		goto error0;
2640 #endif
2641 
2642 	/* If we've got no right sibling then we can't shift an entry right. */
2643 	xfs_btree_get_sibling(cur, left, &rptr, XFS_BB_RIGHTSIB);
2644 	if (xfs_btree_ptr_is_null(cur, &rptr))
2645 		goto out0;
2646 
2647 	/*
2648 	 * If the cursor entry is the one that would be moved, don't
2649 	 * do it... it's too complicated.
2650 	 */
2651 	lrecs = xfs_btree_get_numrecs(left);
2652 	if (cur->bc_levels[level].ptr >= lrecs)
2653 		goto out0;
2654 
2655 	/* Set up the right neighbor as "right". */
2656 	error = xfs_btree_read_buf_block(cur, &rptr, 0, &right, &rbp);
2657 	if (error)
2658 		goto error0;
2659 
2660 	/* If it's full, it can't take another entry. */
2661 	rrecs = xfs_btree_get_numrecs(right);
2662 	if (rrecs == cur->bc_ops->get_maxrecs(cur, level))
2663 		goto out0;
2664 
2665 	XFS_BTREE_STATS_INC(cur, rshift);
2666 	XFS_BTREE_STATS_ADD(cur, moves, rrecs);
2667 
2668 	/*
2669 	 * Make a hole at the start of the right neighbor block, then
2670 	 * copy the last left block entry to the hole.
2671 	 */
2672 	if (level > 0) {
2673 		/* It's a nonleaf. make a hole in the keys and ptrs */
2674 		union xfs_btree_key	*lkp;
2675 		union xfs_btree_ptr	*lpp;
2676 		union xfs_btree_ptr	*rpp;
2677 
2678 		lkp = xfs_btree_key_addr(cur, lrecs, left);
2679 		lpp = xfs_btree_ptr_addr(cur, lrecs, left);
2680 		rkp = xfs_btree_key_addr(cur, 1, right);
2681 		rpp = xfs_btree_ptr_addr(cur, 1, right);
2682 
2683 		for (i = rrecs - 1; i >= 0; i--) {
2684 			error = xfs_btree_debug_check_ptr(cur, rpp, i, level);
2685 			if (error)
2686 				goto error0;
2687 		}
2688 
2689 		xfs_btree_shift_keys(cur, rkp, 1, rrecs);
2690 		xfs_btree_shift_ptrs(cur, rpp, 1, rrecs);
2691 
2692 		error = xfs_btree_debug_check_ptr(cur, lpp, 0, level);
2693 		if (error)
2694 			goto error0;
2695 
2696 		/* Now put the new data in, and log it. */
2697 		xfs_btree_copy_keys(cur, rkp, lkp, 1);
2698 		xfs_btree_copy_ptrs(cur, rpp, lpp, 1);
2699 
2700 		xfs_btree_log_keys(cur, rbp, 1, rrecs + 1);
2701 		xfs_btree_log_ptrs(cur, rbp, 1, rrecs + 1);
2702 
2703 		ASSERT(cur->bc_ops->keys_inorder(cur, rkp,
2704 			xfs_btree_key_addr(cur, 2, right)));
2705 	} else {
2706 		/* It's a leaf. make a hole in the records */
2707 		union xfs_btree_rec	*lrp;
2708 		union xfs_btree_rec	*rrp;
2709 
2710 		lrp = xfs_btree_rec_addr(cur, lrecs, left);
2711 		rrp = xfs_btree_rec_addr(cur, 1, right);
2712 
2713 		xfs_btree_shift_recs(cur, rrp, 1, rrecs);
2714 
2715 		/* Now put the new data in, and log it. */
2716 		xfs_btree_copy_recs(cur, rrp, lrp, 1);
2717 		xfs_btree_log_recs(cur, rbp, 1, rrecs + 1);
2718 	}
2719 
2720 	/*
2721 	 * Decrement and log left's numrecs, bump and log right's numrecs.
2722 	 */
2723 	xfs_btree_set_numrecs(left, --lrecs);
2724 	xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS);
2725 
2726 	xfs_btree_set_numrecs(right, ++rrecs);
2727 	xfs_btree_log_block(cur, rbp, XFS_BB_NUMRECS);
2728 
2729 	/*
2730 	 * Using a temporary cursor, update the parent key values of the
2731 	 * block on the right.
2732 	 */
2733 	error = xfs_btree_dup_cursor(cur, &tcur);
2734 	if (error)
2735 		goto error0;
2736 	i = xfs_btree_lastrec(tcur, level);
2737 	if (XFS_IS_CORRUPT(tcur->bc_mp, i != 1)) {
2738 		xfs_btree_mark_sick(cur);
2739 		error = -EFSCORRUPTED;
2740 		goto error0;
2741 	}
2742 
2743 	error = xfs_btree_increment(tcur, level, &i);
2744 	if (error)
2745 		goto error1;
2746 
2747 	/* Update the parent high keys of the left block, if needed. */
2748 	if (cur->bc_ops->geom_flags & XFS_BTGEO_OVERLAPPING) {
2749 		error = xfs_btree_update_keys(cur, level);
2750 		if (error)
2751 			goto error1;
2752 	}
2753 
2754 	/* Update the parent keys of the right block. */
2755 	error = xfs_btree_update_keys(tcur, level);
2756 	if (error)
2757 		goto error1;
2758 
2759 	xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
2760 
2761 	*stat = 1;
2762 	return 0;
2763 
2764 out0:
2765 	*stat = 0;
2766 	return 0;
2767 
2768 error0:
2769 	return error;
2770 
2771 error1:
2772 	xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
2773 	return error;
2774 }
2775 
2776 static inline int
2777 xfs_btree_alloc_block(
2778 	struct xfs_btree_cur		*cur,
2779 	const union xfs_btree_ptr	*hint_block,
2780 	union xfs_btree_ptr		*new_block,
2781 	int				*stat)
2782 {
2783 	int				error;
2784 
2785 	/*
2786 	 * Don't allow block allocation for a staging cursor, because staging
2787 	 * cursors do not support regular btree modifications.
2788 	 *
2789 	 * Bulk loading uses a separate callback to obtain new blocks from a
2790 	 * preallocated list, which prevents ENOSPC failures during loading.
2791 	 */
2792 	if (unlikely(cur->bc_flags & XFS_BTREE_STAGING)) {
2793 		ASSERT(0);
2794 		return -EFSCORRUPTED;
2795 	}
2796 
2797 	error = cur->bc_ops->alloc_block(cur, hint_block, new_block, stat);
2798 	trace_xfs_btree_alloc_block(cur, new_block, *stat, error);
2799 	return error;
2800 }
2801 
2802 /*
2803  * Split cur/level block in half.
2804  * Return new block number and the key to its first
2805  * record (to be inserted into parent).
2806  */
2807 STATIC int					/* error */
2808 __xfs_btree_split(
2809 	struct xfs_btree_cur	*cur,
2810 	int			level,
2811 	union xfs_btree_ptr	*ptrp,
2812 	union xfs_btree_key	*key,
2813 	struct xfs_btree_cur	**curp,
2814 	int			*stat)		/* success/failure */
2815 {
2816 	union xfs_btree_ptr	lptr;		/* left sibling block ptr */
2817 	struct xfs_buf		*lbp;		/* left buffer pointer */
2818 	struct xfs_btree_block	*left;		/* left btree block */
2819 	union xfs_btree_ptr	rptr;		/* right sibling block ptr */
2820 	struct xfs_buf		*rbp;		/* right buffer pointer */
2821 	struct xfs_btree_block	*right;		/* right btree block */
2822 	union xfs_btree_ptr	rrptr;		/* right-right sibling ptr */
2823 	struct xfs_buf		*rrbp;		/* right-right buffer pointer */
2824 	struct xfs_btree_block	*rrblock;	/* right-right btree block */
2825 	int			lrecs;
2826 	int			rrecs;
2827 	int			src_index;
2828 	int			error;		/* error return value */
2829 	int			i;
2830 
2831 	XFS_BTREE_STATS_INC(cur, split);
2832 
2833 	/* Set up left block (current one). */
2834 	left = xfs_btree_get_block(cur, level, &lbp);
2835 
2836 #ifdef DEBUG
2837 	error = xfs_btree_check_block(cur, left, level, lbp);
2838 	if (error)
2839 		goto error0;
2840 #endif
2841 
2842 	xfs_btree_buf_to_ptr(cur, lbp, &lptr);
2843 
2844 	/* Allocate the new block. If we can't do it, we're toast. Give up. */
2845 	error = xfs_btree_alloc_block(cur, &lptr, &rptr, stat);
2846 	if (error)
2847 		goto error0;
2848 	if (*stat == 0)
2849 		goto out0;
2850 	XFS_BTREE_STATS_INC(cur, alloc);
2851 
2852 	/* Set up the new block as "right". */
2853 	error = xfs_btree_get_buf_block(cur, &rptr, &right, &rbp);
2854 	if (error)
2855 		goto error0;
2856 
2857 	/* Fill in the btree header for the new right block. */
2858 	xfs_btree_init_block_cur(cur, rbp, xfs_btree_get_level(left), 0);
2859 
2860 	/*
2861 	 * Split the entries between the old and the new block evenly.
2862 	 * Make sure that if there's an odd number of entries now, that
2863 	 * each new block will have the same number of entries.
2864 	 */
2865 	lrecs = xfs_btree_get_numrecs(left);
2866 	rrecs = lrecs / 2;
2867 	if ((lrecs & 1) && cur->bc_levels[level].ptr <= rrecs + 1)
2868 		rrecs++;
2869 	src_index = (lrecs - rrecs + 1);
2870 
2871 	XFS_BTREE_STATS_ADD(cur, moves, rrecs);
2872 
2873 	/* Adjust numrecs for the later get_*_keys() calls. */
2874 	lrecs -= rrecs;
2875 	xfs_btree_set_numrecs(left, lrecs);
2876 	xfs_btree_set_numrecs(right, xfs_btree_get_numrecs(right) + rrecs);
2877 
2878 	/*
2879 	 * Copy btree block entries from the left block over to the
2880 	 * new block, the right. Update the right block and log the
2881 	 * changes.
2882 	 */
2883 	if (level > 0) {
2884 		/* It's a non-leaf.  Move keys and pointers. */
2885 		union xfs_btree_key	*lkp;	/* left btree key */
2886 		union xfs_btree_ptr	*lpp;	/* left address pointer */
2887 		union xfs_btree_key	*rkp;	/* right btree key */
2888 		union xfs_btree_ptr	*rpp;	/* right address pointer */
2889 
2890 		lkp = xfs_btree_key_addr(cur, src_index, left);
2891 		lpp = xfs_btree_ptr_addr(cur, src_index, left);
2892 		rkp = xfs_btree_key_addr(cur, 1, right);
2893 		rpp = xfs_btree_ptr_addr(cur, 1, right);
2894 
2895 		for (i = src_index; i < rrecs; i++) {
2896 			error = xfs_btree_debug_check_ptr(cur, lpp, i, level);
2897 			if (error)
2898 				goto error0;
2899 		}
2900 
2901 		/* Copy the keys & pointers to the new block. */
2902 		xfs_btree_copy_keys(cur, rkp, lkp, rrecs);
2903 		xfs_btree_copy_ptrs(cur, rpp, lpp, rrecs);
2904 
2905 		xfs_btree_log_keys(cur, rbp, 1, rrecs);
2906 		xfs_btree_log_ptrs(cur, rbp, 1, rrecs);
2907 
2908 		/* Stash the keys of the new block for later insertion. */
2909 		xfs_btree_get_node_keys(cur, right, key);
2910 	} else {
2911 		/* It's a leaf.  Move records.  */
2912 		union xfs_btree_rec	*lrp;	/* left record pointer */
2913 		union xfs_btree_rec	*rrp;	/* right record pointer */
2914 
2915 		lrp = xfs_btree_rec_addr(cur, src_index, left);
2916 		rrp = xfs_btree_rec_addr(cur, 1, right);
2917 
2918 		/* Copy records to the new block. */
2919 		xfs_btree_copy_recs(cur, rrp, lrp, rrecs);
2920 		xfs_btree_log_recs(cur, rbp, 1, rrecs);
2921 
2922 		/* Stash the keys of the new block for later insertion. */
2923 		xfs_btree_get_leaf_keys(cur, right, key);
2924 	}
2925 
2926 	/*
2927 	 * Find the left block number by looking in the buffer.
2928 	 * Adjust sibling pointers.
2929 	 */
2930 	xfs_btree_get_sibling(cur, left, &rrptr, XFS_BB_RIGHTSIB);
2931 	xfs_btree_set_sibling(cur, right, &rrptr, XFS_BB_RIGHTSIB);
2932 	xfs_btree_set_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
2933 	xfs_btree_set_sibling(cur, left, &rptr, XFS_BB_RIGHTSIB);
2934 
2935 	xfs_btree_log_block(cur, rbp, XFS_BB_ALL_BITS);
2936 	xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS | XFS_BB_RIGHTSIB);
2937 
2938 	/*
2939 	 * If there's a block to the new block's right, make that block
2940 	 * point back to right instead of to left.
2941 	 */
2942 	if (!xfs_btree_ptr_is_null(cur, &rrptr)) {
2943 		error = xfs_btree_read_buf_block(cur, &rrptr,
2944 							0, &rrblock, &rrbp);
2945 		if (error)
2946 			goto error0;
2947 		xfs_btree_set_sibling(cur, rrblock, &rptr, XFS_BB_LEFTSIB);
2948 		xfs_btree_log_block(cur, rrbp, XFS_BB_LEFTSIB);
2949 	}
2950 
2951 	/* Update the parent high keys of the left block, if needed. */
2952 	if (cur->bc_ops->geom_flags & XFS_BTGEO_OVERLAPPING) {
2953 		error = xfs_btree_update_keys(cur, level);
2954 		if (error)
2955 			goto error0;
2956 	}
2957 
2958 	/*
2959 	 * If the cursor is really in the right block, move it there.
2960 	 * If it's just pointing past the last entry in left, then we'll
2961 	 * insert there, so don't change anything in that case.
2962 	 */
2963 	if (cur->bc_levels[level].ptr > lrecs + 1) {
2964 		xfs_btree_setbuf(cur, level, rbp);
2965 		cur->bc_levels[level].ptr -= lrecs;
2966 	}
2967 	/*
2968 	 * If there are more levels, we'll need another cursor which refers
2969 	 * the right block, no matter where this cursor was.
2970 	 */
2971 	if (level + 1 < cur->bc_nlevels) {
2972 		error = xfs_btree_dup_cursor(cur, curp);
2973 		if (error)
2974 			goto error0;
2975 		(*curp)->bc_levels[level + 1].ptr++;
2976 	}
2977 	*ptrp = rptr;
2978 	*stat = 1;
2979 	return 0;
2980 out0:
2981 	*stat = 0;
2982 	return 0;
2983 
2984 error0:
2985 	return error;
2986 }
2987 
2988 #ifdef __KERNEL__
2989 struct xfs_btree_split_args {
2990 	struct xfs_btree_cur	*cur;
2991 	int			level;
2992 	union xfs_btree_ptr	*ptrp;
2993 	union xfs_btree_key	*key;
2994 	struct xfs_btree_cur	**curp;
2995 	int			*stat;		/* success/failure */
2996 	int			result;
2997 	bool			kswapd;	/* allocation in kswapd context */
2998 	struct completion	*done;
2999 	struct work_struct	work;
3000 };
3001 
3002 /*
3003  * Stack switching interfaces for allocation
3004  */
3005 static void
3006 xfs_btree_split_worker(
3007 	struct work_struct	*work)
3008 {
3009 	struct xfs_btree_split_args	*args = container_of(work,
3010 						struct xfs_btree_split_args, work);
3011 	unsigned long		pflags;
3012 	unsigned long		new_pflags = 0;
3013 
3014 	/*
3015 	 * we are in a transaction context here, but may also be doing work
3016 	 * in kswapd context, and hence we may need to inherit that state
3017 	 * temporarily to ensure that we don't block waiting for memory reclaim
3018 	 * in any way.
3019 	 */
3020 	if (args->kswapd)
3021 		new_pflags |= PF_MEMALLOC | PF_KSWAPD;
3022 
3023 	current_set_flags_nested(&pflags, new_pflags);
3024 	xfs_trans_set_context(args->cur->bc_tp);
3025 
3026 	args->result = __xfs_btree_split(args->cur, args->level, args->ptrp,
3027 					 args->key, args->curp, args->stat);
3028 
3029 	xfs_trans_clear_context(args->cur->bc_tp);
3030 	current_restore_flags_nested(&pflags, new_pflags);
3031 
3032 	/*
3033 	 * Do not access args after complete() has run here. We don't own args
3034 	 * and the owner may run and free args before we return here.
3035 	 */
3036 	complete(args->done);
3037 
3038 }
3039 
3040 /*
3041  * BMBT split requests often come in with little stack to work on so we push
3042  * them off to a worker thread so there is lots of stack to use. For the other
3043  * btree types, just call directly to avoid the context switch overhead here.
3044  *
3045  * Care must be taken here - the work queue rescuer thread introduces potential
3046  * AGF <> worker queue deadlocks if the BMBT block allocation has to lock new
3047  * AGFs to allocate blocks. A task being run by the rescuer could attempt to
3048  * lock an AGF that is already locked by a task queued to run by the rescuer,
3049  * resulting in an ABBA deadlock as the rescuer cannot run the lock holder to
3050  * release it until the current thread it is running gains the lock.
3051  *
3052  * To avoid this issue, we only ever queue BMBT splits that don't have an AGF
3053  * already locked to allocate from. The only place that doesn't hold an AGF
3054  * locked is unwritten extent conversion at IO completion, but that has already
3055  * been offloaded to a worker thread and hence has no stack consumption issues
3056  * we have to worry about.
3057  */
3058 STATIC int					/* error */
3059 xfs_btree_split(
3060 	struct xfs_btree_cur	*cur,
3061 	int			level,
3062 	union xfs_btree_ptr	*ptrp,
3063 	union xfs_btree_key	*key,
3064 	struct xfs_btree_cur	**curp,
3065 	int			*stat)		/* success/failure */
3066 {
3067 	struct xfs_btree_split_args	args;
3068 	DECLARE_COMPLETION_ONSTACK(done);
3069 
3070 	if (!xfs_btree_is_bmap(cur->bc_ops) ||
3071 	    cur->bc_tp->t_highest_agno == NULLAGNUMBER)
3072 		return __xfs_btree_split(cur, level, ptrp, key, curp, stat);
3073 
3074 	args.cur = cur;
3075 	args.level = level;
3076 	args.ptrp = ptrp;
3077 	args.key = key;
3078 	args.curp = curp;
3079 	args.stat = stat;
3080 	args.done = &done;
3081 	args.kswapd = current_is_kswapd();
3082 	INIT_WORK_ONSTACK(&args.work, xfs_btree_split_worker);
3083 	queue_work(xfs_alloc_wq, &args.work);
3084 	wait_for_completion(&done);
3085 	destroy_work_on_stack(&args.work);
3086 	return args.result;
3087 }
3088 #else
3089 #define xfs_btree_split	__xfs_btree_split
3090 #endif /* __KERNEL__ */
3091 
3092 /*
3093  * Copy the old inode root contents into a real block and make the
3094  * broot point to it.
3095  */
3096 int						/* error */
3097 xfs_btree_new_iroot(
3098 	struct xfs_btree_cur	*cur,		/* btree cursor */
3099 	int			*logflags,	/* logging flags for inode */
3100 	int			*stat)		/* return status - 0 fail */
3101 {
3102 	struct xfs_buf		*cbp;		/* buffer for cblock */
3103 	struct xfs_btree_block	*block;		/* btree block */
3104 	struct xfs_btree_block	*cblock;	/* child btree block */
3105 	union xfs_btree_key	*ckp;		/* child key pointer */
3106 	union xfs_btree_ptr	*cpp;		/* child ptr pointer */
3107 	union xfs_btree_key	*kp;		/* pointer to btree key */
3108 	union xfs_btree_ptr	*pp;		/* pointer to block addr */
3109 	union xfs_btree_ptr	nptr;		/* new block addr */
3110 	int			level;		/* btree level */
3111 	int			error;		/* error return code */
3112 	int			i;		/* loop counter */
3113 
3114 	XFS_BTREE_STATS_INC(cur, newroot);
3115 
3116 	ASSERT(cur->bc_ops->type == XFS_BTREE_TYPE_INODE);
3117 
3118 	level = cur->bc_nlevels - 1;
3119 
3120 	block = xfs_btree_get_iroot(cur);
3121 	pp = xfs_btree_ptr_addr(cur, 1, block);
3122 
3123 	/* Allocate the new block. If we can't do it, we're toast. Give up. */
3124 	error = xfs_btree_alloc_block(cur, pp, &nptr, stat);
3125 	if (error)
3126 		goto error0;
3127 	if (*stat == 0)
3128 		return 0;
3129 
3130 	XFS_BTREE_STATS_INC(cur, alloc);
3131 
3132 	/* Copy the root into a real block. */
3133 	error = xfs_btree_get_buf_block(cur, &nptr, &cblock, &cbp);
3134 	if (error)
3135 		goto error0;
3136 
3137 	/*
3138 	 * we can't just memcpy() the root in for CRC enabled btree blocks.
3139 	 * In that case have to also ensure the blkno remains correct
3140 	 */
3141 	memcpy(cblock, block, xfs_btree_block_len(cur));
3142 	if (xfs_has_crc(cur->bc_mp)) {
3143 		__be64 bno = cpu_to_be64(xfs_buf_daddr(cbp));
3144 		if (cur->bc_ops->ptr_len == XFS_BTREE_LONG_PTR_LEN)
3145 			cblock->bb_u.l.bb_blkno = bno;
3146 		else
3147 			cblock->bb_u.s.bb_blkno = bno;
3148 	}
3149 
3150 	be16_add_cpu(&block->bb_level, 1);
3151 	xfs_btree_set_numrecs(block, 1);
3152 	cur->bc_nlevels++;
3153 	ASSERT(cur->bc_nlevels <= cur->bc_maxlevels);
3154 	cur->bc_levels[level + 1].ptr = 1;
3155 
3156 	kp = xfs_btree_key_addr(cur, 1, block);
3157 	ckp = xfs_btree_key_addr(cur, 1, cblock);
3158 	xfs_btree_copy_keys(cur, ckp, kp, xfs_btree_get_numrecs(cblock));
3159 
3160 	cpp = xfs_btree_ptr_addr(cur, 1, cblock);
3161 	for (i = 0; i < be16_to_cpu(cblock->bb_numrecs); i++) {
3162 		error = xfs_btree_debug_check_ptr(cur, pp, i, level);
3163 		if (error)
3164 			goto error0;
3165 	}
3166 
3167 	xfs_btree_copy_ptrs(cur, cpp, pp, xfs_btree_get_numrecs(cblock));
3168 
3169 	error = xfs_btree_debug_check_ptr(cur, &nptr, 0, level);
3170 	if (error)
3171 		goto error0;
3172 
3173 	xfs_btree_copy_ptrs(cur, pp, &nptr, 1);
3174 
3175 	xfs_iroot_realloc(cur->bc_ino.ip,
3176 			  1 - xfs_btree_get_numrecs(cblock),
3177 			  cur->bc_ino.whichfork);
3178 
3179 	xfs_btree_setbuf(cur, level, cbp);
3180 
3181 	/*
3182 	 * Do all this logging at the end so that
3183 	 * the root is at the right level.
3184 	 */
3185 	xfs_btree_log_block(cur, cbp, XFS_BB_ALL_BITS);
3186 	xfs_btree_log_keys(cur, cbp, 1, be16_to_cpu(cblock->bb_numrecs));
3187 	xfs_btree_log_ptrs(cur, cbp, 1, be16_to_cpu(cblock->bb_numrecs));
3188 
3189 	*logflags |=
3190 		XFS_ILOG_CORE | xfs_ilog_fbroot(cur->bc_ino.whichfork);
3191 	*stat = 1;
3192 	return 0;
3193 error0:
3194 	return error;
3195 }
3196 
3197 static void
3198 xfs_btree_set_root(
3199 	struct xfs_btree_cur		*cur,
3200 	const union xfs_btree_ptr	*ptr,
3201 	int				inc)
3202 {
3203 	if (cur->bc_flags & XFS_BTREE_STAGING) {
3204 		/* Update the btree root information for a per-AG fake root. */
3205 		cur->bc_ag.afake->af_root = be32_to_cpu(ptr->s);
3206 		cur->bc_ag.afake->af_levels += inc;
3207 	} else {
3208 		cur->bc_ops->set_root(cur, ptr, inc);
3209 	}
3210 }
3211 
3212 /*
3213  * Allocate a new root block, fill it in.
3214  */
3215 STATIC int				/* error */
3216 xfs_btree_new_root(
3217 	struct xfs_btree_cur	*cur,	/* btree cursor */
3218 	int			*stat)	/* success/failure */
3219 {
3220 	struct xfs_btree_block	*block;	/* one half of the old root block */
3221 	struct xfs_buf		*bp;	/* buffer containing block */
3222 	int			error;	/* error return value */
3223 	struct xfs_buf		*lbp;	/* left buffer pointer */
3224 	struct xfs_btree_block	*left;	/* left btree block */
3225 	struct xfs_buf		*nbp;	/* new (root) buffer */
3226 	struct xfs_btree_block	*new;	/* new (root) btree block */
3227 	int			nptr;	/* new value for key index, 1 or 2 */
3228 	struct xfs_buf		*rbp;	/* right buffer pointer */
3229 	struct xfs_btree_block	*right;	/* right btree block */
3230 	union xfs_btree_ptr	rptr;
3231 	union xfs_btree_ptr	lptr;
3232 
3233 	XFS_BTREE_STATS_INC(cur, newroot);
3234 
3235 	/* initialise our start point from the cursor */
3236 	xfs_btree_init_ptr_from_cur(cur, &rptr);
3237 
3238 	/* Allocate the new block. If we can't do it, we're toast. Give up. */
3239 	error = xfs_btree_alloc_block(cur, &rptr, &lptr, stat);
3240 	if (error)
3241 		goto error0;
3242 	if (*stat == 0)
3243 		goto out0;
3244 	XFS_BTREE_STATS_INC(cur, alloc);
3245 
3246 	/* Set up the new block. */
3247 	error = xfs_btree_get_buf_block(cur, &lptr, &new, &nbp);
3248 	if (error)
3249 		goto error0;
3250 
3251 	/* Set the root in the holding structure  increasing the level by 1. */
3252 	xfs_btree_set_root(cur, &lptr, 1);
3253 
3254 	/*
3255 	 * At the previous root level there are now two blocks: the old root,
3256 	 * and the new block generated when it was split.  We don't know which
3257 	 * one the cursor is pointing at, so we set up variables "left" and
3258 	 * "right" for each case.
3259 	 */
3260 	block = xfs_btree_get_block(cur, cur->bc_nlevels - 1, &bp);
3261 
3262 #ifdef DEBUG
3263 	error = xfs_btree_check_block(cur, block, cur->bc_nlevels - 1, bp);
3264 	if (error)
3265 		goto error0;
3266 #endif
3267 
3268 	xfs_btree_get_sibling(cur, block, &rptr, XFS_BB_RIGHTSIB);
3269 	if (!xfs_btree_ptr_is_null(cur, &rptr)) {
3270 		/* Our block is left, pick up the right block. */
3271 		lbp = bp;
3272 		xfs_btree_buf_to_ptr(cur, lbp, &lptr);
3273 		left = block;
3274 		error = xfs_btree_read_buf_block(cur, &rptr, 0, &right, &rbp);
3275 		if (error)
3276 			goto error0;
3277 		bp = rbp;
3278 		nptr = 1;
3279 	} else {
3280 		/* Our block is right, pick up the left block. */
3281 		rbp = bp;
3282 		xfs_btree_buf_to_ptr(cur, rbp, &rptr);
3283 		right = block;
3284 		xfs_btree_get_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
3285 		error = xfs_btree_read_buf_block(cur, &lptr, 0, &left, &lbp);
3286 		if (error)
3287 			goto error0;
3288 		bp = lbp;
3289 		nptr = 2;
3290 	}
3291 
3292 	/* Fill in the new block's btree header and log it. */
3293 	xfs_btree_init_block_cur(cur, nbp, cur->bc_nlevels, 2);
3294 	xfs_btree_log_block(cur, nbp, XFS_BB_ALL_BITS);
3295 	ASSERT(!xfs_btree_ptr_is_null(cur, &lptr) &&
3296 			!xfs_btree_ptr_is_null(cur, &rptr));
3297 
3298 	/* Fill in the key data in the new root. */
3299 	if (xfs_btree_get_level(left) > 0) {
3300 		/*
3301 		 * Get the keys for the left block's keys and put them directly
3302 		 * in the parent block.  Do the same for the right block.
3303 		 */
3304 		xfs_btree_get_node_keys(cur, left,
3305 				xfs_btree_key_addr(cur, 1, new));
3306 		xfs_btree_get_node_keys(cur, right,
3307 				xfs_btree_key_addr(cur, 2, new));
3308 	} else {
3309 		/*
3310 		 * Get the keys for the left block's records and put them
3311 		 * directly in the parent block.  Do the same for the right
3312 		 * block.
3313 		 */
3314 		xfs_btree_get_leaf_keys(cur, left,
3315 			xfs_btree_key_addr(cur, 1, new));
3316 		xfs_btree_get_leaf_keys(cur, right,
3317 			xfs_btree_key_addr(cur, 2, new));
3318 	}
3319 	xfs_btree_log_keys(cur, nbp, 1, 2);
3320 
3321 	/* Fill in the pointer data in the new root. */
3322 	xfs_btree_copy_ptrs(cur,
3323 		xfs_btree_ptr_addr(cur, 1, new), &lptr, 1);
3324 	xfs_btree_copy_ptrs(cur,
3325 		xfs_btree_ptr_addr(cur, 2, new), &rptr, 1);
3326 	xfs_btree_log_ptrs(cur, nbp, 1, 2);
3327 
3328 	/* Fix up the cursor. */
3329 	xfs_btree_setbuf(cur, cur->bc_nlevels, nbp);
3330 	cur->bc_levels[cur->bc_nlevels].ptr = nptr;
3331 	cur->bc_nlevels++;
3332 	ASSERT(cur->bc_nlevels <= cur->bc_maxlevels);
3333 	*stat = 1;
3334 	return 0;
3335 error0:
3336 	return error;
3337 out0:
3338 	*stat = 0;
3339 	return 0;
3340 }
3341 
3342 STATIC int
3343 xfs_btree_make_block_unfull(
3344 	struct xfs_btree_cur	*cur,	/* btree cursor */
3345 	int			level,	/* btree level */
3346 	int			numrecs,/* # of recs in block */
3347 	int			*oindex,/* old tree index */
3348 	int			*index,	/* new tree index */
3349 	union xfs_btree_ptr	*nptr,	/* new btree ptr */
3350 	struct xfs_btree_cur	**ncur,	/* new btree cursor */
3351 	union xfs_btree_key	*key,	/* key of new block */
3352 	int			*stat)
3353 {
3354 	int			error = 0;
3355 
3356 	if (xfs_btree_at_iroot(cur, level)) {
3357 		struct xfs_inode *ip = cur->bc_ino.ip;
3358 
3359 		if (numrecs < cur->bc_ops->get_dmaxrecs(cur, level)) {
3360 			/* A root block that can be made bigger. */
3361 			xfs_iroot_realloc(ip, 1, cur->bc_ino.whichfork);
3362 			*stat = 1;
3363 		} else {
3364 			/* A root block that needs replacing */
3365 			int	logflags = 0;
3366 
3367 			error = xfs_btree_new_iroot(cur, &logflags, stat);
3368 			if (error || *stat == 0)
3369 				return error;
3370 
3371 			xfs_trans_log_inode(cur->bc_tp, ip, logflags);
3372 		}
3373 
3374 		return 0;
3375 	}
3376 
3377 	/* First, try shifting an entry to the right neighbor. */
3378 	error = xfs_btree_rshift(cur, level, stat);
3379 	if (error || *stat)
3380 		return error;
3381 
3382 	/* Next, try shifting an entry to the left neighbor. */
3383 	error = xfs_btree_lshift(cur, level, stat);
3384 	if (error)
3385 		return error;
3386 
3387 	if (*stat) {
3388 		*oindex = *index = cur->bc_levels[level].ptr;
3389 		return 0;
3390 	}
3391 
3392 	/*
3393 	 * Next, try splitting the current block in half.
3394 	 *
3395 	 * If this works we have to re-set our variables because we
3396 	 * could be in a different block now.
3397 	 */
3398 	error = xfs_btree_split(cur, level, nptr, key, ncur, stat);
3399 	if (error || *stat == 0)
3400 		return error;
3401 
3402 
3403 	*index = cur->bc_levels[level].ptr;
3404 	return 0;
3405 }
3406 
3407 /*
3408  * Insert one record/level.  Return information to the caller
3409  * allowing the next level up to proceed if necessary.
3410  */
3411 STATIC int
3412 xfs_btree_insrec(
3413 	struct xfs_btree_cur	*cur,	/* btree cursor */
3414 	int			level,	/* level to insert record at */
3415 	union xfs_btree_ptr	*ptrp,	/* i/o: block number inserted */
3416 	union xfs_btree_rec	*rec,	/* record to insert */
3417 	union xfs_btree_key	*key,	/* i/o: block key for ptrp */
3418 	struct xfs_btree_cur	**curp,	/* output: new cursor replacing cur */
3419 	int			*stat)	/* success/failure */
3420 {
3421 	struct xfs_btree_block	*block;	/* btree block */
3422 	struct xfs_buf		*bp;	/* buffer for block */
3423 	union xfs_btree_ptr	nptr;	/* new block ptr */
3424 	struct xfs_btree_cur	*ncur = NULL;	/* new btree cursor */
3425 	union xfs_btree_key	nkey;	/* new block key */
3426 	union xfs_btree_key	*lkey;
3427 	int			optr;	/* old key/record index */
3428 	int			ptr;	/* key/record index */
3429 	int			numrecs;/* number of records */
3430 	int			error;	/* error return value */
3431 	int			i;
3432 	xfs_daddr_t		old_bn;
3433 
3434 	ncur = NULL;
3435 	lkey = &nkey;
3436 
3437 	/*
3438 	 * If we have an external root pointer, and we've made it to the
3439 	 * root level, allocate a new root block and we're done.
3440 	 */
3441 	if (cur->bc_ops->type != XFS_BTREE_TYPE_INODE &&
3442 	    level >= cur->bc_nlevels) {
3443 		error = xfs_btree_new_root(cur, stat);
3444 		xfs_btree_set_ptr_null(cur, ptrp);
3445 
3446 		return error;
3447 	}
3448 
3449 	/* If we're off the left edge, return failure. */
3450 	ptr = cur->bc_levels[level].ptr;
3451 	if (ptr == 0) {
3452 		*stat = 0;
3453 		return 0;
3454 	}
3455 
3456 	optr = ptr;
3457 
3458 	XFS_BTREE_STATS_INC(cur, insrec);
3459 
3460 	/* Get pointers to the btree buffer and block. */
3461 	block = xfs_btree_get_block(cur, level, &bp);
3462 	old_bn = bp ? xfs_buf_daddr(bp) : XFS_BUF_DADDR_NULL;
3463 	numrecs = xfs_btree_get_numrecs(block);
3464 
3465 #ifdef DEBUG
3466 	error = xfs_btree_check_block(cur, block, level, bp);
3467 	if (error)
3468 		goto error0;
3469 
3470 	/* Check that the new entry is being inserted in the right place. */
3471 	if (ptr <= numrecs) {
3472 		if (level == 0) {
3473 			ASSERT(cur->bc_ops->recs_inorder(cur, rec,
3474 				xfs_btree_rec_addr(cur, ptr, block)));
3475 		} else {
3476 			ASSERT(cur->bc_ops->keys_inorder(cur, key,
3477 				xfs_btree_key_addr(cur, ptr, block)));
3478 		}
3479 	}
3480 #endif
3481 
3482 	/*
3483 	 * If the block is full, we can't insert the new entry until we
3484 	 * make the block un-full.
3485 	 */
3486 	xfs_btree_set_ptr_null(cur, &nptr);
3487 	if (numrecs == cur->bc_ops->get_maxrecs(cur, level)) {
3488 		error = xfs_btree_make_block_unfull(cur, level, numrecs,
3489 					&optr, &ptr, &nptr, &ncur, lkey, stat);
3490 		if (error || *stat == 0)
3491 			goto error0;
3492 	}
3493 
3494 	/*
3495 	 * The current block may have changed if the block was
3496 	 * previously full and we have just made space in it.
3497 	 */
3498 	block = xfs_btree_get_block(cur, level, &bp);
3499 	numrecs = xfs_btree_get_numrecs(block);
3500 
3501 #ifdef DEBUG
3502 	error = xfs_btree_check_block(cur, block, level, bp);
3503 	if (error)
3504 		goto error0;
3505 #endif
3506 
3507 	/*
3508 	 * At this point we know there's room for our new entry in the block
3509 	 * we're pointing at.
3510 	 */
3511 	XFS_BTREE_STATS_ADD(cur, moves, numrecs - ptr + 1);
3512 
3513 	if (level > 0) {
3514 		/* It's a nonleaf. make a hole in the keys and ptrs */
3515 		union xfs_btree_key	*kp;
3516 		union xfs_btree_ptr	*pp;
3517 
3518 		kp = xfs_btree_key_addr(cur, ptr, block);
3519 		pp = xfs_btree_ptr_addr(cur, ptr, block);
3520 
3521 		for (i = numrecs - ptr; i >= 0; i--) {
3522 			error = xfs_btree_debug_check_ptr(cur, pp, i, level);
3523 			if (error)
3524 				goto error0;
3525 		}
3526 
3527 		xfs_btree_shift_keys(cur, kp, 1, numrecs - ptr + 1);
3528 		xfs_btree_shift_ptrs(cur, pp, 1, numrecs - ptr + 1);
3529 
3530 		error = xfs_btree_debug_check_ptr(cur, ptrp, 0, level);
3531 		if (error)
3532 			goto error0;
3533 
3534 		/* Now put the new data in, bump numrecs and log it. */
3535 		xfs_btree_copy_keys(cur, kp, key, 1);
3536 		xfs_btree_copy_ptrs(cur, pp, ptrp, 1);
3537 		numrecs++;
3538 		xfs_btree_set_numrecs(block, numrecs);
3539 		xfs_btree_log_ptrs(cur, bp, ptr, numrecs);
3540 		xfs_btree_log_keys(cur, bp, ptr, numrecs);
3541 #ifdef DEBUG
3542 		if (ptr < numrecs) {
3543 			ASSERT(cur->bc_ops->keys_inorder(cur, kp,
3544 				xfs_btree_key_addr(cur, ptr + 1, block)));
3545 		}
3546 #endif
3547 	} else {
3548 		/* It's a leaf. make a hole in the records */
3549 		union xfs_btree_rec             *rp;
3550 
3551 		rp = xfs_btree_rec_addr(cur, ptr, block);
3552 
3553 		xfs_btree_shift_recs(cur, rp, 1, numrecs - ptr + 1);
3554 
3555 		/* Now put the new data in, bump numrecs and log it. */
3556 		xfs_btree_copy_recs(cur, rp, rec, 1);
3557 		xfs_btree_set_numrecs(block, ++numrecs);
3558 		xfs_btree_log_recs(cur, bp, ptr, numrecs);
3559 #ifdef DEBUG
3560 		if (ptr < numrecs) {
3561 			ASSERT(cur->bc_ops->recs_inorder(cur, rp,
3562 				xfs_btree_rec_addr(cur, ptr + 1, block)));
3563 		}
3564 #endif
3565 	}
3566 
3567 	/* Log the new number of records in the btree header. */
3568 	xfs_btree_log_block(cur, bp, XFS_BB_NUMRECS);
3569 
3570 	/*
3571 	 * If we just inserted into a new tree block, we have to
3572 	 * recalculate nkey here because nkey is out of date.
3573 	 *
3574 	 * Otherwise we're just updating an existing block (having shoved
3575 	 * some records into the new tree block), so use the regular key
3576 	 * update mechanism.
3577 	 */
3578 	if (bp && xfs_buf_daddr(bp) != old_bn) {
3579 		xfs_btree_get_keys(cur, block, lkey);
3580 	} else if (xfs_btree_needs_key_update(cur, optr)) {
3581 		error = xfs_btree_update_keys(cur, level);
3582 		if (error)
3583 			goto error0;
3584 	}
3585 
3586 	/*
3587 	 * Return the new block number, if any.
3588 	 * If there is one, give back a record value and a cursor too.
3589 	 */
3590 	*ptrp = nptr;
3591 	if (!xfs_btree_ptr_is_null(cur, &nptr)) {
3592 		xfs_btree_copy_keys(cur, key, lkey, 1);
3593 		*curp = ncur;
3594 	}
3595 
3596 	*stat = 1;
3597 	return 0;
3598 
3599 error0:
3600 	if (ncur)
3601 		xfs_btree_del_cursor(ncur, error);
3602 	return error;
3603 }
3604 
3605 /*
3606  * Insert the record at the point referenced by cur.
3607  *
3608  * A multi-level split of the tree on insert will invalidate the original
3609  * cursor.  All callers of this function should assume that the cursor is
3610  * no longer valid and revalidate it.
3611  */
3612 int
3613 xfs_btree_insert(
3614 	struct xfs_btree_cur	*cur,
3615 	int			*stat)
3616 {
3617 	int			error;	/* error return value */
3618 	int			i;	/* result value, 0 for failure */
3619 	int			level;	/* current level number in btree */
3620 	union xfs_btree_ptr	nptr;	/* new block number (split result) */
3621 	struct xfs_btree_cur	*ncur;	/* new cursor (split result) */
3622 	struct xfs_btree_cur	*pcur;	/* previous level's cursor */
3623 	union xfs_btree_key	bkey;	/* key of block to insert */
3624 	union xfs_btree_key	*key;
3625 	union xfs_btree_rec	rec;	/* record to insert */
3626 
3627 	level = 0;
3628 	ncur = NULL;
3629 	pcur = cur;
3630 	key = &bkey;
3631 
3632 	xfs_btree_set_ptr_null(cur, &nptr);
3633 
3634 	/* Make a key out of the record data to be inserted, and save it. */
3635 	cur->bc_ops->init_rec_from_cur(cur, &rec);
3636 	cur->bc_ops->init_key_from_rec(key, &rec);
3637 
3638 	/*
3639 	 * Loop going up the tree, starting at the leaf level.
3640 	 * Stop when we don't get a split block, that must mean that
3641 	 * the insert is finished with this level.
3642 	 */
3643 	do {
3644 		/*
3645 		 * Insert nrec/nptr into this level of the tree.
3646 		 * Note if we fail, nptr will be null.
3647 		 */
3648 		error = xfs_btree_insrec(pcur, level, &nptr, &rec, key,
3649 				&ncur, &i);
3650 		if (error) {
3651 			if (pcur != cur)
3652 				xfs_btree_del_cursor(pcur, XFS_BTREE_ERROR);
3653 			goto error0;
3654 		}
3655 
3656 		if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
3657 			xfs_btree_mark_sick(cur);
3658 			error = -EFSCORRUPTED;
3659 			goto error0;
3660 		}
3661 		level++;
3662 
3663 		/*
3664 		 * See if the cursor we just used is trash.
3665 		 * Can't trash the caller's cursor, but otherwise we should
3666 		 * if ncur is a new cursor or we're about to be done.
3667 		 */
3668 		if (pcur != cur &&
3669 		    (ncur || xfs_btree_ptr_is_null(cur, &nptr))) {
3670 			/* Save the state from the cursor before we trash it */
3671 			if (cur->bc_ops->update_cursor &&
3672 			    !(cur->bc_flags & XFS_BTREE_STAGING))
3673 				cur->bc_ops->update_cursor(pcur, cur);
3674 			cur->bc_nlevels = pcur->bc_nlevels;
3675 			xfs_btree_del_cursor(pcur, XFS_BTREE_NOERROR);
3676 		}
3677 		/* If we got a new cursor, switch to it. */
3678 		if (ncur) {
3679 			pcur = ncur;
3680 			ncur = NULL;
3681 		}
3682 	} while (!xfs_btree_ptr_is_null(cur, &nptr));
3683 
3684 	*stat = i;
3685 	return 0;
3686 error0:
3687 	return error;
3688 }
3689 
3690 /*
3691  * Try to merge a non-leaf block back into the inode root.
3692  *
3693  * Note: the killroot names comes from the fact that we're effectively
3694  * killing the old root block.  But because we can't just delete the
3695  * inode we have to copy the single block it was pointing to into the
3696  * inode.
3697  */
3698 STATIC int
3699 xfs_btree_kill_iroot(
3700 	struct xfs_btree_cur	*cur)
3701 {
3702 	int			whichfork = cur->bc_ino.whichfork;
3703 	struct xfs_inode	*ip = cur->bc_ino.ip;
3704 	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, whichfork);
3705 	struct xfs_btree_block	*block;
3706 	struct xfs_btree_block	*cblock;
3707 	union xfs_btree_key	*kp;
3708 	union xfs_btree_key	*ckp;
3709 	union xfs_btree_ptr	*pp;
3710 	union xfs_btree_ptr	*cpp;
3711 	struct xfs_buf		*cbp;
3712 	int			level;
3713 	int			index;
3714 	int			numrecs;
3715 	int			error;
3716 #ifdef DEBUG
3717 	union xfs_btree_ptr	ptr;
3718 #endif
3719 	int			i;
3720 
3721 	ASSERT(cur->bc_ops->type == XFS_BTREE_TYPE_INODE);
3722 	ASSERT(cur->bc_nlevels > 1);
3723 
3724 	/*
3725 	 * Don't deal with the root block needs to be a leaf case.
3726 	 * We're just going to turn the thing back into extents anyway.
3727 	 */
3728 	level = cur->bc_nlevels - 1;
3729 	if (level == 1)
3730 		goto out0;
3731 
3732 	/*
3733 	 * Give up if the root has multiple children.
3734 	 */
3735 	block = xfs_btree_get_iroot(cur);
3736 	if (xfs_btree_get_numrecs(block) != 1)
3737 		goto out0;
3738 
3739 	cblock = xfs_btree_get_block(cur, level - 1, &cbp);
3740 	numrecs = xfs_btree_get_numrecs(cblock);
3741 
3742 	/*
3743 	 * Only do this if the next level will fit.
3744 	 * Then the data must be copied up to the inode,
3745 	 * instead of freeing the root you free the next level.
3746 	 */
3747 	if (numrecs > cur->bc_ops->get_dmaxrecs(cur, level))
3748 		goto out0;
3749 
3750 	XFS_BTREE_STATS_INC(cur, killroot);
3751 
3752 #ifdef DEBUG
3753 	xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_LEFTSIB);
3754 	ASSERT(xfs_btree_ptr_is_null(cur, &ptr));
3755 	xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
3756 	ASSERT(xfs_btree_ptr_is_null(cur, &ptr));
3757 #endif
3758 
3759 	index = numrecs - cur->bc_ops->get_maxrecs(cur, level);
3760 	if (index) {
3761 		xfs_iroot_realloc(cur->bc_ino.ip, index,
3762 				  cur->bc_ino.whichfork);
3763 		block = ifp->if_broot;
3764 	}
3765 
3766 	be16_add_cpu(&block->bb_numrecs, index);
3767 	ASSERT(block->bb_numrecs == cblock->bb_numrecs);
3768 
3769 	kp = xfs_btree_key_addr(cur, 1, block);
3770 	ckp = xfs_btree_key_addr(cur, 1, cblock);
3771 	xfs_btree_copy_keys(cur, kp, ckp, numrecs);
3772 
3773 	pp = xfs_btree_ptr_addr(cur, 1, block);
3774 	cpp = xfs_btree_ptr_addr(cur, 1, cblock);
3775 
3776 	for (i = 0; i < numrecs; i++) {
3777 		error = xfs_btree_debug_check_ptr(cur, cpp, i, level - 1);
3778 		if (error)
3779 			return error;
3780 	}
3781 
3782 	xfs_btree_copy_ptrs(cur, pp, cpp, numrecs);
3783 
3784 	error = xfs_btree_free_block(cur, cbp);
3785 	if (error)
3786 		return error;
3787 
3788 	cur->bc_levels[level - 1].bp = NULL;
3789 	be16_add_cpu(&block->bb_level, -1);
3790 	xfs_trans_log_inode(cur->bc_tp, ip,
3791 		XFS_ILOG_CORE | xfs_ilog_fbroot(cur->bc_ino.whichfork));
3792 	cur->bc_nlevels--;
3793 out0:
3794 	return 0;
3795 }
3796 
3797 /*
3798  * Kill the current root node, and replace it with it's only child node.
3799  */
3800 STATIC int
3801 xfs_btree_kill_root(
3802 	struct xfs_btree_cur	*cur,
3803 	struct xfs_buf		*bp,
3804 	int			level,
3805 	union xfs_btree_ptr	*newroot)
3806 {
3807 	int			error;
3808 
3809 	XFS_BTREE_STATS_INC(cur, killroot);
3810 
3811 	/*
3812 	 * Update the root pointer, decreasing the level by 1 and then
3813 	 * free the old root.
3814 	 */
3815 	xfs_btree_set_root(cur, newroot, -1);
3816 
3817 	error = xfs_btree_free_block(cur, bp);
3818 	if (error)
3819 		return error;
3820 
3821 	cur->bc_levels[level].bp = NULL;
3822 	cur->bc_levels[level].ra = 0;
3823 	cur->bc_nlevels--;
3824 
3825 	return 0;
3826 }
3827 
3828 STATIC int
3829 xfs_btree_dec_cursor(
3830 	struct xfs_btree_cur	*cur,
3831 	int			level,
3832 	int			*stat)
3833 {
3834 	int			error;
3835 	int			i;
3836 
3837 	if (level > 0) {
3838 		error = xfs_btree_decrement(cur, level, &i);
3839 		if (error)
3840 			return error;
3841 	}
3842 
3843 	*stat = 1;
3844 	return 0;
3845 }
3846 
3847 /*
3848  * Single level of the btree record deletion routine.
3849  * Delete record pointed to by cur/level.
3850  * Remove the record from its block then rebalance the tree.
3851  * Return 0 for error, 1 for done, 2 to go on to the next level.
3852  */
3853 STATIC int					/* error */
3854 xfs_btree_delrec(
3855 	struct xfs_btree_cur	*cur,		/* btree cursor */
3856 	int			level,		/* level removing record from */
3857 	int			*stat)		/* fail/done/go-on */
3858 {
3859 	struct xfs_btree_block	*block;		/* btree block */
3860 	union xfs_btree_ptr	cptr;		/* current block ptr */
3861 	struct xfs_buf		*bp;		/* buffer for block */
3862 	int			error;		/* error return value */
3863 	int			i;		/* loop counter */
3864 	union xfs_btree_ptr	lptr;		/* left sibling block ptr */
3865 	struct xfs_buf		*lbp;		/* left buffer pointer */
3866 	struct xfs_btree_block	*left;		/* left btree block */
3867 	int			lrecs = 0;	/* left record count */
3868 	int			ptr;		/* key/record index */
3869 	union xfs_btree_ptr	rptr;		/* right sibling block ptr */
3870 	struct xfs_buf		*rbp;		/* right buffer pointer */
3871 	struct xfs_btree_block	*right;		/* right btree block */
3872 	struct xfs_btree_block	*rrblock;	/* right-right btree block */
3873 	struct xfs_buf		*rrbp;		/* right-right buffer pointer */
3874 	int			rrecs = 0;	/* right record count */
3875 	struct xfs_btree_cur	*tcur;		/* temporary btree cursor */
3876 	int			numrecs;	/* temporary numrec count */
3877 
3878 	tcur = NULL;
3879 
3880 	/* Get the index of the entry being deleted, check for nothing there. */
3881 	ptr = cur->bc_levels[level].ptr;
3882 	if (ptr == 0) {
3883 		*stat = 0;
3884 		return 0;
3885 	}
3886 
3887 	/* Get the buffer & block containing the record or key/ptr. */
3888 	block = xfs_btree_get_block(cur, level, &bp);
3889 	numrecs = xfs_btree_get_numrecs(block);
3890 
3891 #ifdef DEBUG
3892 	error = xfs_btree_check_block(cur, block, level, bp);
3893 	if (error)
3894 		goto error0;
3895 #endif
3896 
3897 	/* Fail if we're off the end of the block. */
3898 	if (ptr > numrecs) {
3899 		*stat = 0;
3900 		return 0;
3901 	}
3902 
3903 	XFS_BTREE_STATS_INC(cur, delrec);
3904 	XFS_BTREE_STATS_ADD(cur, moves, numrecs - ptr);
3905 
3906 	/* Excise the entries being deleted. */
3907 	if (level > 0) {
3908 		/* It's a nonleaf. operate on keys and ptrs */
3909 		union xfs_btree_key	*lkp;
3910 		union xfs_btree_ptr	*lpp;
3911 
3912 		lkp = xfs_btree_key_addr(cur, ptr + 1, block);
3913 		lpp = xfs_btree_ptr_addr(cur, ptr + 1, block);
3914 
3915 		for (i = 0; i < numrecs - ptr; i++) {
3916 			error = xfs_btree_debug_check_ptr(cur, lpp, i, level);
3917 			if (error)
3918 				goto error0;
3919 		}
3920 
3921 		if (ptr < numrecs) {
3922 			xfs_btree_shift_keys(cur, lkp, -1, numrecs - ptr);
3923 			xfs_btree_shift_ptrs(cur, lpp, -1, numrecs - ptr);
3924 			xfs_btree_log_keys(cur, bp, ptr, numrecs - 1);
3925 			xfs_btree_log_ptrs(cur, bp, ptr, numrecs - 1);
3926 		}
3927 	} else {
3928 		/* It's a leaf. operate on records */
3929 		if (ptr < numrecs) {
3930 			xfs_btree_shift_recs(cur,
3931 				xfs_btree_rec_addr(cur, ptr + 1, block),
3932 				-1, numrecs - ptr);
3933 			xfs_btree_log_recs(cur, bp, ptr, numrecs - 1);
3934 		}
3935 	}
3936 
3937 	/*
3938 	 * Decrement and log the number of entries in the block.
3939 	 */
3940 	xfs_btree_set_numrecs(block, --numrecs);
3941 	xfs_btree_log_block(cur, bp, XFS_BB_NUMRECS);
3942 
3943 	/*
3944 	 * We're at the root level.  First, shrink the root block in-memory.
3945 	 * Try to get rid of the next level down.  If we can't then there's
3946 	 * nothing left to do.
3947 	 */
3948 	if (xfs_btree_at_iroot(cur, level)) {
3949 		xfs_iroot_realloc(cur->bc_ino.ip, -1, cur->bc_ino.whichfork);
3950 
3951 		error = xfs_btree_kill_iroot(cur);
3952 		if (error)
3953 			goto error0;
3954 
3955 		error = xfs_btree_dec_cursor(cur, level, stat);
3956 		if (error)
3957 			goto error0;
3958 		*stat = 1;
3959 		return 0;
3960 	}
3961 
3962 	/*
3963 	 * If this is the root level, and there's only one entry left, and it's
3964 	 * NOT the leaf level, then we can get rid of this level.
3965 	 */
3966 	if (level == cur->bc_nlevels - 1) {
3967 		if (numrecs == 1 && level > 0) {
3968 			union xfs_btree_ptr	*pp;
3969 			/*
3970 			 * pp is still set to the first pointer in the block.
3971 			 * Make it the new root of the btree.
3972 			 */
3973 			pp = xfs_btree_ptr_addr(cur, 1, block);
3974 			error = xfs_btree_kill_root(cur, bp, level, pp);
3975 			if (error)
3976 				goto error0;
3977 		} else if (level > 0) {
3978 			error = xfs_btree_dec_cursor(cur, level, stat);
3979 			if (error)
3980 				goto error0;
3981 		}
3982 		*stat = 1;
3983 		return 0;
3984 	}
3985 
3986 	/*
3987 	 * If we deleted the leftmost entry in the block, update the
3988 	 * key values above us in the tree.
3989 	 */
3990 	if (xfs_btree_needs_key_update(cur, ptr)) {
3991 		error = xfs_btree_update_keys(cur, level);
3992 		if (error)
3993 			goto error0;
3994 	}
3995 
3996 	/*
3997 	 * If the number of records remaining in the block is at least
3998 	 * the minimum, we're done.
3999 	 */
4000 	if (numrecs >= cur->bc_ops->get_minrecs(cur, level)) {
4001 		error = xfs_btree_dec_cursor(cur, level, stat);
4002 		if (error)
4003 			goto error0;
4004 		return 0;
4005 	}
4006 
4007 	/*
4008 	 * Otherwise, we have to move some records around to keep the
4009 	 * tree balanced.  Look at the left and right sibling blocks to
4010 	 * see if we can re-balance by moving only one record.
4011 	 */
4012 	xfs_btree_get_sibling(cur, block, &rptr, XFS_BB_RIGHTSIB);
4013 	xfs_btree_get_sibling(cur, block, &lptr, XFS_BB_LEFTSIB);
4014 
4015 	if (cur->bc_ops->type == XFS_BTREE_TYPE_INODE) {
4016 		/*
4017 		 * One child of root, need to get a chance to copy its contents
4018 		 * into the root and delete it. Can't go up to next level,
4019 		 * there's nothing to delete there.
4020 		 */
4021 		if (xfs_btree_ptr_is_null(cur, &rptr) &&
4022 		    xfs_btree_ptr_is_null(cur, &lptr) &&
4023 		    level == cur->bc_nlevels - 2) {
4024 			error = xfs_btree_kill_iroot(cur);
4025 			if (!error)
4026 				error = xfs_btree_dec_cursor(cur, level, stat);
4027 			if (error)
4028 				goto error0;
4029 			return 0;
4030 		}
4031 	}
4032 
4033 	ASSERT(!xfs_btree_ptr_is_null(cur, &rptr) ||
4034 	       !xfs_btree_ptr_is_null(cur, &lptr));
4035 
4036 	/*
4037 	 * Duplicate the cursor so our btree manipulations here won't
4038 	 * disrupt the next level up.
4039 	 */
4040 	error = xfs_btree_dup_cursor(cur, &tcur);
4041 	if (error)
4042 		goto error0;
4043 
4044 	/*
4045 	 * If there's a right sibling, see if it's ok to shift an entry
4046 	 * out of it.
4047 	 */
4048 	if (!xfs_btree_ptr_is_null(cur, &rptr)) {
4049 		/*
4050 		 * Move the temp cursor to the last entry in the next block.
4051 		 * Actually any entry but the first would suffice.
4052 		 */
4053 		i = xfs_btree_lastrec(tcur, level);
4054 		if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
4055 			xfs_btree_mark_sick(cur);
4056 			error = -EFSCORRUPTED;
4057 			goto error0;
4058 		}
4059 
4060 		error = xfs_btree_increment(tcur, level, &i);
4061 		if (error)
4062 			goto error0;
4063 		if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
4064 			xfs_btree_mark_sick(cur);
4065 			error = -EFSCORRUPTED;
4066 			goto error0;
4067 		}
4068 
4069 		i = xfs_btree_lastrec(tcur, level);
4070 		if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
4071 			xfs_btree_mark_sick(cur);
4072 			error = -EFSCORRUPTED;
4073 			goto error0;
4074 		}
4075 
4076 		/* Grab a pointer to the block. */
4077 		right = xfs_btree_get_block(tcur, level, &rbp);
4078 #ifdef DEBUG
4079 		error = xfs_btree_check_block(tcur, right, level, rbp);
4080 		if (error)
4081 			goto error0;
4082 #endif
4083 		/* Grab the current block number, for future use. */
4084 		xfs_btree_get_sibling(tcur, right, &cptr, XFS_BB_LEFTSIB);
4085 
4086 		/*
4087 		 * If right block is full enough so that removing one entry
4088 		 * won't make it too empty, and left-shifting an entry out
4089 		 * of right to us works, we're done.
4090 		 */
4091 		if (xfs_btree_get_numrecs(right) - 1 >=
4092 		    cur->bc_ops->get_minrecs(tcur, level)) {
4093 			error = xfs_btree_lshift(tcur, level, &i);
4094 			if (error)
4095 				goto error0;
4096 			if (i) {
4097 				ASSERT(xfs_btree_get_numrecs(block) >=
4098 				       cur->bc_ops->get_minrecs(tcur, level));
4099 
4100 				xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
4101 				tcur = NULL;
4102 
4103 				error = xfs_btree_dec_cursor(cur, level, stat);
4104 				if (error)
4105 					goto error0;
4106 				return 0;
4107 			}
4108 		}
4109 
4110 		/*
4111 		 * Otherwise, grab the number of records in right for
4112 		 * future reference, and fix up the temp cursor to point
4113 		 * to our block again (last record).
4114 		 */
4115 		rrecs = xfs_btree_get_numrecs(right);
4116 		if (!xfs_btree_ptr_is_null(cur, &lptr)) {
4117 			i = xfs_btree_firstrec(tcur, level);
4118 			if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
4119 				xfs_btree_mark_sick(cur);
4120 				error = -EFSCORRUPTED;
4121 				goto error0;
4122 			}
4123 
4124 			error = xfs_btree_decrement(tcur, level, &i);
4125 			if (error)
4126 				goto error0;
4127 			if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
4128 				xfs_btree_mark_sick(cur);
4129 				error = -EFSCORRUPTED;
4130 				goto error0;
4131 			}
4132 		}
4133 	}
4134 
4135 	/*
4136 	 * If there's a left sibling, see if it's ok to shift an entry
4137 	 * out of it.
4138 	 */
4139 	if (!xfs_btree_ptr_is_null(cur, &lptr)) {
4140 		/*
4141 		 * Move the temp cursor to the first entry in the
4142 		 * previous block.
4143 		 */
4144 		i = xfs_btree_firstrec(tcur, level);
4145 		if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
4146 			xfs_btree_mark_sick(cur);
4147 			error = -EFSCORRUPTED;
4148 			goto error0;
4149 		}
4150 
4151 		error = xfs_btree_decrement(tcur, level, &i);
4152 		if (error)
4153 			goto error0;
4154 		i = xfs_btree_firstrec(tcur, level);
4155 		if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
4156 			xfs_btree_mark_sick(cur);
4157 			error = -EFSCORRUPTED;
4158 			goto error0;
4159 		}
4160 
4161 		/* Grab a pointer to the block. */
4162 		left = xfs_btree_get_block(tcur, level, &lbp);
4163 #ifdef DEBUG
4164 		error = xfs_btree_check_block(cur, left, level, lbp);
4165 		if (error)
4166 			goto error0;
4167 #endif
4168 		/* Grab the current block number, for future use. */
4169 		xfs_btree_get_sibling(tcur, left, &cptr, XFS_BB_RIGHTSIB);
4170 
4171 		/*
4172 		 * If left block is full enough so that removing one entry
4173 		 * won't make it too empty, and right-shifting an entry out
4174 		 * of left to us works, we're done.
4175 		 */
4176 		if (xfs_btree_get_numrecs(left) - 1 >=
4177 		    cur->bc_ops->get_minrecs(tcur, level)) {
4178 			error = xfs_btree_rshift(tcur, level, &i);
4179 			if (error)
4180 				goto error0;
4181 			if (i) {
4182 				ASSERT(xfs_btree_get_numrecs(block) >=
4183 				       cur->bc_ops->get_minrecs(tcur, level));
4184 				xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
4185 				tcur = NULL;
4186 				if (level == 0)
4187 					cur->bc_levels[0].ptr++;
4188 
4189 				*stat = 1;
4190 				return 0;
4191 			}
4192 		}
4193 
4194 		/*
4195 		 * Otherwise, grab the number of records in right for
4196 		 * future reference.
4197 		 */
4198 		lrecs = xfs_btree_get_numrecs(left);
4199 	}
4200 
4201 	/* Delete the temp cursor, we're done with it. */
4202 	xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
4203 	tcur = NULL;
4204 
4205 	/* If here, we need to do a join to keep the tree balanced. */
4206 	ASSERT(!xfs_btree_ptr_is_null(cur, &cptr));
4207 
4208 	if (!xfs_btree_ptr_is_null(cur, &lptr) &&
4209 	    lrecs + xfs_btree_get_numrecs(block) <=
4210 			cur->bc_ops->get_maxrecs(cur, level)) {
4211 		/*
4212 		 * Set "right" to be the starting block,
4213 		 * "left" to be the left neighbor.
4214 		 */
4215 		rptr = cptr;
4216 		right = block;
4217 		rbp = bp;
4218 		error = xfs_btree_read_buf_block(cur, &lptr, 0, &left, &lbp);
4219 		if (error)
4220 			goto error0;
4221 
4222 	/*
4223 	 * If that won't work, see if we can join with the right neighbor block.
4224 	 */
4225 	} else if (!xfs_btree_ptr_is_null(cur, &rptr) &&
4226 		   rrecs + xfs_btree_get_numrecs(block) <=
4227 			cur->bc_ops->get_maxrecs(cur, level)) {
4228 		/*
4229 		 * Set "left" to be the starting block,
4230 		 * "right" to be the right neighbor.
4231 		 */
4232 		lptr = cptr;
4233 		left = block;
4234 		lbp = bp;
4235 		error = xfs_btree_read_buf_block(cur, &rptr, 0, &right, &rbp);
4236 		if (error)
4237 			goto error0;
4238 
4239 	/*
4240 	 * Otherwise, we can't fix the imbalance.
4241 	 * Just return.  This is probably a logic error, but it's not fatal.
4242 	 */
4243 	} else {
4244 		error = xfs_btree_dec_cursor(cur, level, stat);
4245 		if (error)
4246 			goto error0;
4247 		return 0;
4248 	}
4249 
4250 	rrecs = xfs_btree_get_numrecs(right);
4251 	lrecs = xfs_btree_get_numrecs(left);
4252 
4253 	/*
4254 	 * We're now going to join "left" and "right" by moving all the stuff
4255 	 * in "right" to "left" and deleting "right".
4256 	 */
4257 	XFS_BTREE_STATS_ADD(cur, moves, rrecs);
4258 	if (level > 0) {
4259 		/* It's a non-leaf.  Move keys and pointers. */
4260 		union xfs_btree_key	*lkp;	/* left btree key */
4261 		union xfs_btree_ptr	*lpp;	/* left address pointer */
4262 		union xfs_btree_key	*rkp;	/* right btree key */
4263 		union xfs_btree_ptr	*rpp;	/* right address pointer */
4264 
4265 		lkp = xfs_btree_key_addr(cur, lrecs + 1, left);
4266 		lpp = xfs_btree_ptr_addr(cur, lrecs + 1, left);
4267 		rkp = xfs_btree_key_addr(cur, 1, right);
4268 		rpp = xfs_btree_ptr_addr(cur, 1, right);
4269 
4270 		for (i = 1; i < rrecs; i++) {
4271 			error = xfs_btree_debug_check_ptr(cur, rpp, i, level);
4272 			if (error)
4273 				goto error0;
4274 		}
4275 
4276 		xfs_btree_copy_keys(cur, lkp, rkp, rrecs);
4277 		xfs_btree_copy_ptrs(cur, lpp, rpp, rrecs);
4278 
4279 		xfs_btree_log_keys(cur, lbp, lrecs + 1, lrecs + rrecs);
4280 		xfs_btree_log_ptrs(cur, lbp, lrecs + 1, lrecs + rrecs);
4281 	} else {
4282 		/* It's a leaf.  Move records.  */
4283 		union xfs_btree_rec	*lrp;	/* left record pointer */
4284 		union xfs_btree_rec	*rrp;	/* right record pointer */
4285 
4286 		lrp = xfs_btree_rec_addr(cur, lrecs + 1, left);
4287 		rrp = xfs_btree_rec_addr(cur, 1, right);
4288 
4289 		xfs_btree_copy_recs(cur, lrp, rrp, rrecs);
4290 		xfs_btree_log_recs(cur, lbp, lrecs + 1, lrecs + rrecs);
4291 	}
4292 
4293 	XFS_BTREE_STATS_INC(cur, join);
4294 
4295 	/*
4296 	 * Fix up the number of records and right block pointer in the
4297 	 * surviving block, and log it.
4298 	 */
4299 	xfs_btree_set_numrecs(left, lrecs + rrecs);
4300 	xfs_btree_get_sibling(cur, right, &cptr, XFS_BB_RIGHTSIB);
4301 	xfs_btree_set_sibling(cur, left, &cptr, XFS_BB_RIGHTSIB);
4302 	xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS | XFS_BB_RIGHTSIB);
4303 
4304 	/* If there is a right sibling, point it to the remaining block. */
4305 	xfs_btree_get_sibling(cur, left, &cptr, XFS_BB_RIGHTSIB);
4306 	if (!xfs_btree_ptr_is_null(cur, &cptr)) {
4307 		error = xfs_btree_read_buf_block(cur, &cptr, 0, &rrblock, &rrbp);
4308 		if (error)
4309 			goto error0;
4310 		xfs_btree_set_sibling(cur, rrblock, &lptr, XFS_BB_LEFTSIB);
4311 		xfs_btree_log_block(cur, rrbp, XFS_BB_LEFTSIB);
4312 	}
4313 
4314 	/* Free the deleted block. */
4315 	error = xfs_btree_free_block(cur, rbp);
4316 	if (error)
4317 		goto error0;
4318 
4319 	/*
4320 	 * If we joined with the left neighbor, set the buffer in the
4321 	 * cursor to the left block, and fix up the index.
4322 	 */
4323 	if (bp != lbp) {
4324 		cur->bc_levels[level].bp = lbp;
4325 		cur->bc_levels[level].ptr += lrecs;
4326 		cur->bc_levels[level].ra = 0;
4327 	}
4328 	/*
4329 	 * If we joined with the right neighbor and there's a level above
4330 	 * us, increment the cursor at that level.
4331 	 */
4332 	else if (cur->bc_ops->type == XFS_BTREE_TYPE_INODE ||
4333 		 level + 1 < cur->bc_nlevels) {
4334 		error = xfs_btree_increment(cur, level + 1, &i);
4335 		if (error)
4336 			goto error0;
4337 	}
4338 
4339 	/*
4340 	 * Readjust the ptr at this level if it's not a leaf, since it's
4341 	 * still pointing at the deletion point, which makes the cursor
4342 	 * inconsistent.  If this makes the ptr 0, the caller fixes it up.
4343 	 * We can't use decrement because it would change the next level up.
4344 	 */
4345 	if (level > 0)
4346 		cur->bc_levels[level].ptr--;
4347 
4348 	/*
4349 	 * We combined blocks, so we have to update the parent keys if the
4350 	 * btree supports overlapped intervals.  However,
4351 	 * bc_levels[level + 1].ptr points to the old block so that the caller
4352 	 * knows which record to delete.  Therefore, the caller must be savvy
4353 	 * enough to call updkeys for us if we return stat == 2.  The other
4354 	 * exit points from this function don't require deletions further up
4355 	 * the tree, so they can call updkeys directly.
4356 	 */
4357 
4358 	/* Return value means the next level up has something to do. */
4359 	*stat = 2;
4360 	return 0;
4361 
4362 error0:
4363 	if (tcur)
4364 		xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
4365 	return error;
4366 }
4367 
4368 /*
4369  * Delete the record pointed to by cur.
4370  * The cursor refers to the place where the record was (could be inserted)
4371  * when the operation returns.
4372  */
4373 int					/* error */
4374 xfs_btree_delete(
4375 	struct xfs_btree_cur	*cur,
4376 	int			*stat)	/* success/failure */
4377 {
4378 	int			error;	/* error return value */
4379 	int			level;
4380 	int			i;
4381 	bool			joined = false;
4382 
4383 	/*
4384 	 * Go up the tree, starting at leaf level.
4385 	 *
4386 	 * If 2 is returned then a join was done; go to the next level.
4387 	 * Otherwise we are done.
4388 	 */
4389 	for (level = 0, i = 2; i == 2; level++) {
4390 		error = xfs_btree_delrec(cur, level, &i);
4391 		if (error)
4392 			goto error0;
4393 		if (i == 2)
4394 			joined = true;
4395 	}
4396 
4397 	/*
4398 	 * If we combined blocks as part of deleting the record, delrec won't
4399 	 * have updated the parent high keys so we have to do that here.
4400 	 */
4401 	if (joined && (cur->bc_ops->geom_flags & XFS_BTGEO_OVERLAPPING)) {
4402 		error = xfs_btree_updkeys_force(cur, 0);
4403 		if (error)
4404 			goto error0;
4405 	}
4406 
4407 	if (i == 0) {
4408 		for (level = 1; level < cur->bc_nlevels; level++) {
4409 			if (cur->bc_levels[level].ptr == 0) {
4410 				error = xfs_btree_decrement(cur, level, &i);
4411 				if (error)
4412 					goto error0;
4413 				break;
4414 			}
4415 		}
4416 	}
4417 
4418 	*stat = i;
4419 	return 0;
4420 error0:
4421 	return error;
4422 }
4423 
4424 /*
4425  * Get the data from the pointed-to record.
4426  */
4427 int					/* error */
4428 xfs_btree_get_rec(
4429 	struct xfs_btree_cur	*cur,	/* btree cursor */
4430 	union xfs_btree_rec	**recp,	/* output: btree record */
4431 	int			*stat)	/* output: success/failure */
4432 {
4433 	struct xfs_btree_block	*block;	/* btree block */
4434 	struct xfs_buf		*bp;	/* buffer pointer */
4435 	int			ptr;	/* record number */
4436 #ifdef DEBUG
4437 	int			error;	/* error return value */
4438 #endif
4439 
4440 	ptr = cur->bc_levels[0].ptr;
4441 	block = xfs_btree_get_block(cur, 0, &bp);
4442 
4443 #ifdef DEBUG
4444 	error = xfs_btree_check_block(cur, block, 0, bp);
4445 	if (error)
4446 		return error;
4447 #endif
4448 
4449 	/*
4450 	 * Off the right end or left end, return failure.
4451 	 */
4452 	if (ptr > xfs_btree_get_numrecs(block) || ptr <= 0) {
4453 		*stat = 0;
4454 		return 0;
4455 	}
4456 
4457 	/*
4458 	 * Point to the record and extract its data.
4459 	 */
4460 	*recp = xfs_btree_rec_addr(cur, ptr, block);
4461 	*stat = 1;
4462 	return 0;
4463 }
4464 
4465 /* Visit a block in a btree. */
4466 STATIC int
4467 xfs_btree_visit_block(
4468 	struct xfs_btree_cur		*cur,
4469 	int				level,
4470 	xfs_btree_visit_blocks_fn	fn,
4471 	void				*data)
4472 {
4473 	struct xfs_btree_block		*block;
4474 	struct xfs_buf			*bp;
4475 	union xfs_btree_ptr		rptr, bufptr;
4476 	int				error;
4477 
4478 	/* do right sibling readahead */
4479 	xfs_btree_readahead(cur, level, XFS_BTCUR_RIGHTRA);
4480 	block = xfs_btree_get_block(cur, level, &bp);
4481 
4482 	/* process the block */
4483 	error = fn(cur, level, data);
4484 	if (error)
4485 		return error;
4486 
4487 	/* now read rh sibling block for next iteration */
4488 	xfs_btree_get_sibling(cur, block, &rptr, XFS_BB_RIGHTSIB);
4489 	if (xfs_btree_ptr_is_null(cur, &rptr))
4490 		return -ENOENT;
4491 
4492 	/*
4493 	 * We only visit blocks once in this walk, so we have to avoid the
4494 	 * internal xfs_btree_lookup_get_block() optimisation where it will
4495 	 * return the same block without checking if the right sibling points
4496 	 * back to us and creates a cyclic reference in the btree.
4497 	 */
4498 	xfs_btree_buf_to_ptr(cur, bp, &bufptr);
4499 	if (xfs_btree_ptrs_equal(cur, &rptr, &bufptr)) {
4500 		xfs_btree_mark_sick(cur);
4501 		return -EFSCORRUPTED;
4502 	}
4503 
4504 	return xfs_btree_lookup_get_block(cur, level, &rptr, &block);
4505 }
4506 
4507 
4508 /* Visit every block in a btree. */
4509 int
4510 xfs_btree_visit_blocks(
4511 	struct xfs_btree_cur		*cur,
4512 	xfs_btree_visit_blocks_fn	fn,
4513 	unsigned int			flags,
4514 	void				*data)
4515 {
4516 	union xfs_btree_ptr		lptr;
4517 	int				level;
4518 	struct xfs_btree_block		*block = NULL;
4519 	int				error = 0;
4520 
4521 	xfs_btree_init_ptr_from_cur(cur, &lptr);
4522 
4523 	/* for each level */
4524 	for (level = cur->bc_nlevels - 1; level >= 0; level--) {
4525 		/* grab the left hand block */
4526 		error = xfs_btree_lookup_get_block(cur, level, &lptr, &block);
4527 		if (error)
4528 			return error;
4529 
4530 		/* readahead the left most block for the next level down */
4531 		if (level > 0) {
4532 			union xfs_btree_ptr     *ptr;
4533 
4534 			ptr = xfs_btree_ptr_addr(cur, 1, block);
4535 			xfs_btree_readahead_ptr(cur, ptr, 1);
4536 
4537 			/* save for the next iteration of the loop */
4538 			xfs_btree_copy_ptrs(cur, &lptr, ptr, 1);
4539 
4540 			if (!(flags & XFS_BTREE_VISIT_LEAVES))
4541 				continue;
4542 		} else if (!(flags & XFS_BTREE_VISIT_RECORDS)) {
4543 			continue;
4544 		}
4545 
4546 		/* for each buffer in the level */
4547 		do {
4548 			error = xfs_btree_visit_block(cur, level, fn, data);
4549 		} while (!error);
4550 
4551 		if (error != -ENOENT)
4552 			return error;
4553 	}
4554 
4555 	return 0;
4556 }
4557 
4558 /*
4559  * Change the owner of a btree.
4560  *
4561  * The mechanism we use here is ordered buffer logging. Because we don't know
4562  * how many buffers were are going to need to modify, we don't really want to
4563  * have to make transaction reservations for the worst case of every buffer in a
4564  * full size btree as that may be more space that we can fit in the log....
4565  *
4566  * We do the btree walk in the most optimal manner possible - we have sibling
4567  * pointers so we can just walk all the blocks on each level from left to right
4568  * in a single pass, and then move to the next level and do the same. We can
4569  * also do readahead on the sibling pointers to get IO moving more quickly,
4570  * though for slow disks this is unlikely to make much difference to performance
4571  * as the amount of CPU work we have to do before moving to the next block is
4572  * relatively small.
4573  *
4574  * For each btree block that we load, modify the owner appropriately, set the
4575  * buffer as an ordered buffer and log it appropriately. We need to ensure that
4576  * we mark the region we change dirty so that if the buffer is relogged in
4577  * a subsequent transaction the changes we make here as an ordered buffer are
4578  * correctly relogged in that transaction.  If we are in recovery context, then
4579  * just queue the modified buffer as delayed write buffer so the transaction
4580  * recovery completion writes the changes to disk.
4581  */
4582 struct xfs_btree_block_change_owner_info {
4583 	uint64_t		new_owner;
4584 	struct list_head	*buffer_list;
4585 };
4586 
4587 static int
4588 xfs_btree_block_change_owner(
4589 	struct xfs_btree_cur	*cur,
4590 	int			level,
4591 	void			*data)
4592 {
4593 	struct xfs_btree_block_change_owner_info	*bbcoi = data;
4594 	struct xfs_btree_block	*block;
4595 	struct xfs_buf		*bp;
4596 
4597 	/* modify the owner */
4598 	block = xfs_btree_get_block(cur, level, &bp);
4599 	if (cur->bc_ops->ptr_len == XFS_BTREE_LONG_PTR_LEN) {
4600 		if (block->bb_u.l.bb_owner == cpu_to_be64(bbcoi->new_owner))
4601 			return 0;
4602 		block->bb_u.l.bb_owner = cpu_to_be64(bbcoi->new_owner);
4603 	} else {
4604 		if (block->bb_u.s.bb_owner == cpu_to_be32(bbcoi->new_owner))
4605 			return 0;
4606 		block->bb_u.s.bb_owner = cpu_to_be32(bbcoi->new_owner);
4607 	}
4608 
4609 	/*
4610 	 * If the block is a root block hosted in an inode, we might not have a
4611 	 * buffer pointer here and we shouldn't attempt to log the change as the
4612 	 * information is already held in the inode and discarded when the root
4613 	 * block is formatted into the on-disk inode fork. We still change it,
4614 	 * though, so everything is consistent in memory.
4615 	 */
4616 	if (!bp) {
4617 		ASSERT(cur->bc_ops->type == XFS_BTREE_TYPE_INODE);
4618 		ASSERT(level == cur->bc_nlevels - 1);
4619 		return 0;
4620 	}
4621 
4622 	if (cur->bc_tp) {
4623 		if (!xfs_trans_ordered_buf(cur->bc_tp, bp)) {
4624 			xfs_btree_log_block(cur, bp, XFS_BB_OWNER);
4625 			return -EAGAIN;
4626 		}
4627 	} else {
4628 		xfs_buf_delwri_queue(bp, bbcoi->buffer_list);
4629 	}
4630 
4631 	return 0;
4632 }
4633 
4634 int
4635 xfs_btree_change_owner(
4636 	struct xfs_btree_cur	*cur,
4637 	uint64_t		new_owner,
4638 	struct list_head	*buffer_list)
4639 {
4640 	struct xfs_btree_block_change_owner_info	bbcoi;
4641 
4642 	bbcoi.new_owner = new_owner;
4643 	bbcoi.buffer_list = buffer_list;
4644 
4645 	return xfs_btree_visit_blocks(cur, xfs_btree_block_change_owner,
4646 			XFS_BTREE_VISIT_ALL, &bbcoi);
4647 }
4648 
4649 /* Verify the v5 fields of a long-format btree block. */
4650 xfs_failaddr_t
4651 xfs_btree_fsblock_v5hdr_verify(
4652 	struct xfs_buf		*bp,
4653 	uint64_t		owner)
4654 {
4655 	struct xfs_mount	*mp = bp->b_mount;
4656 	struct xfs_btree_block	*block = XFS_BUF_TO_BLOCK(bp);
4657 
4658 	if (!xfs_has_crc(mp))
4659 		return __this_address;
4660 	if (!uuid_equal(&block->bb_u.l.bb_uuid, &mp->m_sb.sb_meta_uuid))
4661 		return __this_address;
4662 	if (block->bb_u.l.bb_blkno != cpu_to_be64(xfs_buf_daddr(bp)))
4663 		return __this_address;
4664 	if (owner != XFS_RMAP_OWN_UNKNOWN &&
4665 	    be64_to_cpu(block->bb_u.l.bb_owner) != owner)
4666 		return __this_address;
4667 	return NULL;
4668 }
4669 
4670 /* Verify a long-format btree block. */
4671 xfs_failaddr_t
4672 xfs_btree_fsblock_verify(
4673 	struct xfs_buf		*bp,
4674 	unsigned int		max_recs)
4675 {
4676 	struct xfs_mount	*mp = bp->b_mount;
4677 	struct xfs_btree_block	*block = XFS_BUF_TO_BLOCK(bp);
4678 	xfs_fsblock_t		fsb;
4679 	xfs_failaddr_t		fa;
4680 
4681 	ASSERT(!xfs_buftarg_is_mem(bp->b_target));
4682 
4683 	/* numrecs verification */
4684 	if (be16_to_cpu(block->bb_numrecs) > max_recs)
4685 		return __this_address;
4686 
4687 	/* sibling pointer verification */
4688 	fsb = XFS_DADDR_TO_FSB(mp, xfs_buf_daddr(bp));
4689 	fa = xfs_btree_check_fsblock_siblings(mp, fsb,
4690 			block->bb_u.l.bb_leftsib);
4691 	if (!fa)
4692 		fa = xfs_btree_check_fsblock_siblings(mp, fsb,
4693 				block->bb_u.l.bb_rightsib);
4694 	return fa;
4695 }
4696 
4697 /* Verify an in-memory btree block. */
4698 xfs_failaddr_t
4699 xfs_btree_memblock_verify(
4700 	struct xfs_buf		*bp,
4701 	unsigned int		max_recs)
4702 {
4703 	struct xfs_btree_block	*block = XFS_BUF_TO_BLOCK(bp);
4704 	struct xfs_buftarg	*btp = bp->b_target;
4705 	xfs_failaddr_t		fa;
4706 	xfbno_t			bno;
4707 
4708 	ASSERT(xfs_buftarg_is_mem(bp->b_target));
4709 
4710 	/* numrecs verification */
4711 	if (be16_to_cpu(block->bb_numrecs) > max_recs)
4712 		return __this_address;
4713 
4714 	/* sibling pointer verification */
4715 	bno = xfs_daddr_to_xfbno(xfs_buf_daddr(bp));
4716 	fa = xfs_btree_check_memblock_siblings(btp, bno,
4717 			block->bb_u.l.bb_leftsib);
4718 	if (fa)
4719 		return fa;
4720 	fa = xfs_btree_check_memblock_siblings(btp, bno,
4721 			block->bb_u.l.bb_rightsib);
4722 	if (fa)
4723 		return fa;
4724 
4725 	return NULL;
4726 }
4727 /**
4728  * xfs_btree_agblock_v5hdr_verify() -- verify the v5 fields of a short-format
4729  *				      btree block
4730  *
4731  * @bp: buffer containing the btree block
4732  */
4733 xfs_failaddr_t
4734 xfs_btree_agblock_v5hdr_verify(
4735 	struct xfs_buf		*bp)
4736 {
4737 	struct xfs_mount	*mp = bp->b_mount;
4738 	struct xfs_btree_block	*block = XFS_BUF_TO_BLOCK(bp);
4739 	struct xfs_perag	*pag = bp->b_pag;
4740 
4741 	if (!xfs_has_crc(mp))
4742 		return __this_address;
4743 	if (!uuid_equal(&block->bb_u.s.bb_uuid, &mp->m_sb.sb_meta_uuid))
4744 		return __this_address;
4745 	if (block->bb_u.s.bb_blkno != cpu_to_be64(xfs_buf_daddr(bp)))
4746 		return __this_address;
4747 	if (pag && be32_to_cpu(block->bb_u.s.bb_owner) != pag_agno(pag))
4748 		return __this_address;
4749 	return NULL;
4750 }
4751 
4752 /**
4753  * xfs_btree_agblock_verify() -- verify a short-format btree block
4754  *
4755  * @bp: buffer containing the btree block
4756  * @max_recs: maximum records allowed in this btree node
4757  */
4758 xfs_failaddr_t
4759 xfs_btree_agblock_verify(
4760 	struct xfs_buf		*bp,
4761 	unsigned int		max_recs)
4762 {
4763 	struct xfs_mount	*mp = bp->b_mount;
4764 	struct xfs_btree_block	*block = XFS_BUF_TO_BLOCK(bp);
4765 	xfs_agblock_t		agbno;
4766 	xfs_failaddr_t		fa;
4767 
4768 	ASSERT(!xfs_buftarg_is_mem(bp->b_target));
4769 
4770 	/* numrecs verification */
4771 	if (be16_to_cpu(block->bb_numrecs) > max_recs)
4772 		return __this_address;
4773 
4774 	/* sibling pointer verification */
4775 	agbno = xfs_daddr_to_agbno(mp, xfs_buf_daddr(bp));
4776 	fa = xfs_btree_check_agblock_siblings(bp->b_pag, agbno,
4777 			block->bb_u.s.bb_leftsib);
4778 	if (!fa)
4779 		fa = xfs_btree_check_agblock_siblings(bp->b_pag, agbno,
4780 				block->bb_u.s.bb_rightsib);
4781 	return fa;
4782 }
4783 
4784 /*
4785  * For the given limits on leaf and keyptr records per block, calculate the
4786  * height of the tree needed to index the number of leaf records.
4787  */
4788 unsigned int
4789 xfs_btree_compute_maxlevels(
4790 	const unsigned int	*limits,
4791 	unsigned long long	records)
4792 {
4793 	unsigned long long	level_blocks = howmany_64(records, limits[0]);
4794 	unsigned int		height = 1;
4795 
4796 	while (level_blocks > 1) {
4797 		level_blocks = howmany_64(level_blocks, limits[1]);
4798 		height++;
4799 	}
4800 
4801 	return height;
4802 }
4803 
4804 /*
4805  * For the given limits on leaf and keyptr records per block, calculate the
4806  * number of blocks needed to index the given number of leaf records.
4807  */
4808 unsigned long long
4809 xfs_btree_calc_size(
4810 	const unsigned int	*limits,
4811 	unsigned long long	records)
4812 {
4813 	unsigned long long	level_blocks = howmany_64(records, limits[0]);
4814 	unsigned long long	blocks = level_blocks;
4815 
4816 	while (level_blocks > 1) {
4817 		level_blocks = howmany_64(level_blocks, limits[1]);
4818 		blocks += level_blocks;
4819 	}
4820 
4821 	return blocks;
4822 }
4823 
4824 /*
4825  * Given a number of available blocks for the btree to consume with records and
4826  * pointers, calculate the height of the tree needed to index all the records
4827  * that space can hold based on the number of pointers each interior node
4828  * holds.
4829  *
4830  * We start by assuming a single level tree consumes a single block, then track
4831  * the number of blocks each node level consumes until we no longer have space
4832  * to store the next node level. At this point, we are indexing all the leaf
4833  * blocks in the space, and there's no more free space to split the tree any
4834  * further. That's our maximum btree height.
4835  */
4836 unsigned int
4837 xfs_btree_space_to_height(
4838 	const unsigned int	*limits,
4839 	unsigned long long	leaf_blocks)
4840 {
4841 	/*
4842 	 * The root btree block can have fewer than minrecs pointers in it
4843 	 * because the tree might not be big enough to require that amount of
4844 	 * fanout. Hence it has a minimum size of 2 pointers, not limits[1].
4845 	 */
4846 	unsigned long long	node_blocks = 2;
4847 	unsigned long long	blocks_left = leaf_blocks - 1;
4848 	unsigned int		height = 1;
4849 
4850 	if (leaf_blocks < 1)
4851 		return 0;
4852 
4853 	while (node_blocks < blocks_left) {
4854 		blocks_left -= node_blocks;
4855 		node_blocks *= limits[1];
4856 		height++;
4857 	}
4858 
4859 	return height;
4860 }
4861 
4862 /*
4863  * Query a regular btree for all records overlapping a given interval.
4864  * Start with a LE lookup of the key of low_rec and return all records
4865  * until we find a record with a key greater than the key of high_rec.
4866  */
4867 STATIC int
4868 xfs_btree_simple_query_range(
4869 	struct xfs_btree_cur		*cur,
4870 	const union xfs_btree_key	*low_key,
4871 	const union xfs_btree_key	*high_key,
4872 	xfs_btree_query_range_fn	fn,
4873 	void				*priv)
4874 {
4875 	union xfs_btree_rec		*recp;
4876 	union xfs_btree_key		rec_key;
4877 	int				stat;
4878 	bool				firstrec = true;
4879 	int				error;
4880 
4881 	ASSERT(cur->bc_ops->init_high_key_from_rec);
4882 	ASSERT(cur->bc_ops->diff_two_keys);
4883 
4884 	/*
4885 	 * Find the leftmost record.  The btree cursor must be set
4886 	 * to the low record used to generate low_key.
4887 	 */
4888 	stat = 0;
4889 	error = xfs_btree_lookup(cur, XFS_LOOKUP_LE, &stat);
4890 	if (error)
4891 		goto out;
4892 
4893 	/* Nothing?  See if there's anything to the right. */
4894 	if (!stat) {
4895 		error = xfs_btree_increment(cur, 0, &stat);
4896 		if (error)
4897 			goto out;
4898 	}
4899 
4900 	while (stat) {
4901 		/* Find the record. */
4902 		error = xfs_btree_get_rec(cur, &recp, &stat);
4903 		if (error || !stat)
4904 			break;
4905 
4906 		/* Skip if low_key > high_key(rec). */
4907 		if (firstrec) {
4908 			cur->bc_ops->init_high_key_from_rec(&rec_key, recp);
4909 			firstrec = false;
4910 			if (xfs_btree_keycmp_gt(cur, low_key, &rec_key))
4911 				goto advloop;
4912 		}
4913 
4914 		/* Stop if low_key(rec) > high_key. */
4915 		cur->bc_ops->init_key_from_rec(&rec_key, recp);
4916 		if (xfs_btree_keycmp_gt(cur, &rec_key, high_key))
4917 			break;
4918 
4919 		/* Callback */
4920 		error = fn(cur, recp, priv);
4921 		if (error)
4922 			break;
4923 
4924 advloop:
4925 		/* Move on to the next record. */
4926 		error = xfs_btree_increment(cur, 0, &stat);
4927 		if (error)
4928 			break;
4929 	}
4930 
4931 out:
4932 	return error;
4933 }
4934 
4935 /*
4936  * Query an overlapped interval btree for all records overlapping a given
4937  * interval.  This function roughly follows the algorithm given in
4938  * "Interval Trees" of _Introduction to Algorithms_, which is section
4939  * 14.3 in the 2nd and 3rd editions.
4940  *
4941  * First, generate keys for the low and high records passed in.
4942  *
4943  * For any leaf node, generate the high and low keys for the record.
4944  * If the record keys overlap with the query low/high keys, pass the
4945  * record to the function iterator.
4946  *
4947  * For any internal node, compare the low and high keys of each
4948  * pointer against the query low/high keys.  If there's an overlap,
4949  * follow the pointer.
4950  *
4951  * As an optimization, we stop scanning a block when we find a low key
4952  * that is greater than the query's high key.
4953  */
4954 STATIC int
4955 xfs_btree_overlapped_query_range(
4956 	struct xfs_btree_cur		*cur,
4957 	const union xfs_btree_key	*low_key,
4958 	const union xfs_btree_key	*high_key,
4959 	xfs_btree_query_range_fn	fn,
4960 	void				*priv)
4961 {
4962 	union xfs_btree_ptr		ptr;
4963 	union xfs_btree_ptr		*pp;
4964 	union xfs_btree_key		rec_key;
4965 	union xfs_btree_key		rec_hkey;
4966 	union xfs_btree_key		*lkp;
4967 	union xfs_btree_key		*hkp;
4968 	union xfs_btree_rec		*recp;
4969 	struct xfs_btree_block		*block;
4970 	int				level;
4971 	struct xfs_buf			*bp;
4972 	int				i;
4973 	int				error;
4974 
4975 	/* Load the root of the btree. */
4976 	level = cur->bc_nlevels - 1;
4977 	xfs_btree_init_ptr_from_cur(cur, &ptr);
4978 	error = xfs_btree_lookup_get_block(cur, level, &ptr, &block);
4979 	if (error)
4980 		return error;
4981 	xfs_btree_get_block(cur, level, &bp);
4982 	trace_xfs_btree_overlapped_query_range(cur, level, bp);
4983 #ifdef DEBUG
4984 	error = xfs_btree_check_block(cur, block, level, bp);
4985 	if (error)
4986 		goto out;
4987 #endif
4988 	cur->bc_levels[level].ptr = 1;
4989 
4990 	while (level < cur->bc_nlevels) {
4991 		block = xfs_btree_get_block(cur, level, &bp);
4992 
4993 		/* End of node, pop back towards the root. */
4994 		if (cur->bc_levels[level].ptr >
4995 					be16_to_cpu(block->bb_numrecs)) {
4996 pop_up:
4997 			if (level < cur->bc_nlevels - 1)
4998 				cur->bc_levels[level + 1].ptr++;
4999 			level++;
5000 			continue;
5001 		}
5002 
5003 		if (level == 0) {
5004 			/* Handle a leaf node. */
5005 			recp = xfs_btree_rec_addr(cur, cur->bc_levels[0].ptr,
5006 					block);
5007 
5008 			cur->bc_ops->init_high_key_from_rec(&rec_hkey, recp);
5009 			cur->bc_ops->init_key_from_rec(&rec_key, recp);
5010 
5011 			/*
5012 			 * If (query's high key < record's low key), then there
5013 			 * are no more interesting records in this block.  Pop
5014 			 * up to the leaf level to find more record blocks.
5015 			 *
5016 			 * If (record's high key >= query's low key) and
5017 			 *    (query's high key >= record's low key), then
5018 			 * this record overlaps the query range; callback.
5019 			 */
5020 			if (xfs_btree_keycmp_lt(cur, high_key, &rec_key))
5021 				goto pop_up;
5022 			if (xfs_btree_keycmp_ge(cur, &rec_hkey, low_key)) {
5023 				error = fn(cur, recp, priv);
5024 				if (error)
5025 					break;
5026 			}
5027 			cur->bc_levels[level].ptr++;
5028 			continue;
5029 		}
5030 
5031 		/* Handle an internal node. */
5032 		lkp = xfs_btree_key_addr(cur, cur->bc_levels[level].ptr, block);
5033 		hkp = xfs_btree_high_key_addr(cur, cur->bc_levels[level].ptr,
5034 				block);
5035 		pp = xfs_btree_ptr_addr(cur, cur->bc_levels[level].ptr, block);
5036 
5037 		/*
5038 		 * If (query's high key < pointer's low key), then there are no
5039 		 * more interesting keys in this block.  Pop up one leaf level
5040 		 * to continue looking for records.
5041 		 *
5042 		 * If (pointer's high key >= query's low key) and
5043 		 *    (query's high key >= pointer's low key), then
5044 		 * this record overlaps the query range; follow pointer.
5045 		 */
5046 		if (xfs_btree_keycmp_lt(cur, high_key, lkp))
5047 			goto pop_up;
5048 		if (xfs_btree_keycmp_ge(cur, hkp, low_key)) {
5049 			level--;
5050 			error = xfs_btree_lookup_get_block(cur, level, pp,
5051 					&block);
5052 			if (error)
5053 				goto out;
5054 			xfs_btree_get_block(cur, level, &bp);
5055 			trace_xfs_btree_overlapped_query_range(cur, level, bp);
5056 #ifdef DEBUG
5057 			error = xfs_btree_check_block(cur, block, level, bp);
5058 			if (error)
5059 				goto out;
5060 #endif
5061 			cur->bc_levels[level].ptr = 1;
5062 			continue;
5063 		}
5064 		cur->bc_levels[level].ptr++;
5065 	}
5066 
5067 out:
5068 	/*
5069 	 * If we don't end this function with the cursor pointing at a record
5070 	 * block, a subsequent non-error cursor deletion will not release
5071 	 * node-level buffers, causing a buffer leak.  This is quite possible
5072 	 * with a zero-results range query, so release the buffers if we
5073 	 * failed to return any results.
5074 	 */
5075 	if (cur->bc_levels[0].bp == NULL) {
5076 		for (i = 0; i < cur->bc_nlevels; i++) {
5077 			if (cur->bc_levels[i].bp) {
5078 				xfs_trans_brelse(cur->bc_tp,
5079 						cur->bc_levels[i].bp);
5080 				cur->bc_levels[i].bp = NULL;
5081 				cur->bc_levels[i].ptr = 0;
5082 				cur->bc_levels[i].ra = 0;
5083 			}
5084 		}
5085 	}
5086 
5087 	return error;
5088 }
5089 
5090 static inline void
5091 xfs_btree_key_from_irec(
5092 	struct xfs_btree_cur		*cur,
5093 	union xfs_btree_key		*key,
5094 	const union xfs_btree_irec	*irec)
5095 {
5096 	union xfs_btree_rec		rec;
5097 
5098 	cur->bc_rec = *irec;
5099 	cur->bc_ops->init_rec_from_cur(cur, &rec);
5100 	cur->bc_ops->init_key_from_rec(key, &rec);
5101 }
5102 
5103 /*
5104  * Query a btree for all records overlapping a given interval of keys.  The
5105  * supplied function will be called with each record found; return one of the
5106  * XFS_BTREE_QUERY_RANGE_{CONTINUE,ABORT} values or the usual negative error
5107  * code.  This function returns -ECANCELED, zero, or a negative error code.
5108  */
5109 int
5110 xfs_btree_query_range(
5111 	struct xfs_btree_cur		*cur,
5112 	const union xfs_btree_irec	*low_rec,
5113 	const union xfs_btree_irec	*high_rec,
5114 	xfs_btree_query_range_fn	fn,
5115 	void				*priv)
5116 {
5117 	union xfs_btree_key		low_key;
5118 	union xfs_btree_key		high_key;
5119 
5120 	/* Find the keys of both ends of the interval. */
5121 	xfs_btree_key_from_irec(cur, &high_key, high_rec);
5122 	xfs_btree_key_from_irec(cur, &low_key, low_rec);
5123 
5124 	/* Enforce low key <= high key. */
5125 	if (!xfs_btree_keycmp_le(cur, &low_key, &high_key))
5126 		return -EINVAL;
5127 
5128 	if (!(cur->bc_ops->geom_flags & XFS_BTGEO_OVERLAPPING))
5129 		return xfs_btree_simple_query_range(cur, &low_key,
5130 				&high_key, fn, priv);
5131 	return xfs_btree_overlapped_query_range(cur, &low_key, &high_key,
5132 			fn, priv);
5133 }
5134 
5135 /* Query a btree for all records. */
5136 int
5137 xfs_btree_query_all(
5138 	struct xfs_btree_cur		*cur,
5139 	xfs_btree_query_range_fn	fn,
5140 	void				*priv)
5141 {
5142 	union xfs_btree_key		low_key;
5143 	union xfs_btree_key		high_key;
5144 
5145 	memset(&cur->bc_rec, 0, sizeof(cur->bc_rec));
5146 	memset(&low_key, 0, sizeof(low_key));
5147 	memset(&high_key, 0xFF, sizeof(high_key));
5148 
5149 	return xfs_btree_simple_query_range(cur, &low_key, &high_key, fn, priv);
5150 }
5151 
5152 static int
5153 xfs_btree_count_blocks_helper(
5154 	struct xfs_btree_cur	*cur,
5155 	int			level,
5156 	void			*data)
5157 {
5158 	xfs_extlen_t		*blocks = data;
5159 	(*blocks)++;
5160 
5161 	return 0;
5162 }
5163 
5164 /* Count the blocks in a btree and return the result in *blocks. */
5165 int
5166 xfs_btree_count_blocks(
5167 	struct xfs_btree_cur	*cur,
5168 	xfs_extlen_t		*blocks)
5169 {
5170 	*blocks = 0;
5171 	return xfs_btree_visit_blocks(cur, xfs_btree_count_blocks_helper,
5172 			XFS_BTREE_VISIT_ALL, blocks);
5173 }
5174 
5175 /* Compare two btree pointers. */
5176 int64_t
5177 xfs_btree_diff_two_ptrs(
5178 	struct xfs_btree_cur		*cur,
5179 	const union xfs_btree_ptr	*a,
5180 	const union xfs_btree_ptr	*b)
5181 {
5182 	if (cur->bc_ops->ptr_len == XFS_BTREE_LONG_PTR_LEN)
5183 		return (int64_t)be64_to_cpu(a->l) - be64_to_cpu(b->l);
5184 	return (int64_t)be32_to_cpu(a->s) - be32_to_cpu(b->s);
5185 }
5186 
5187 struct xfs_btree_has_records {
5188 	/* Keys for the start and end of the range we want to know about. */
5189 	union xfs_btree_key		start_key;
5190 	union xfs_btree_key		end_key;
5191 
5192 	/* Mask for key comparisons, if desired. */
5193 	const union xfs_btree_key	*key_mask;
5194 
5195 	/* Highest record key we've seen so far. */
5196 	union xfs_btree_key		high_key;
5197 
5198 	enum xbtree_recpacking		outcome;
5199 };
5200 
5201 STATIC int
5202 xfs_btree_has_records_helper(
5203 	struct xfs_btree_cur		*cur,
5204 	const union xfs_btree_rec	*rec,
5205 	void				*priv)
5206 {
5207 	union xfs_btree_key		rec_key;
5208 	union xfs_btree_key		rec_high_key;
5209 	struct xfs_btree_has_records	*info = priv;
5210 	enum xbtree_key_contig		key_contig;
5211 
5212 	cur->bc_ops->init_key_from_rec(&rec_key, rec);
5213 
5214 	if (info->outcome == XBTREE_RECPACKING_EMPTY) {
5215 		info->outcome = XBTREE_RECPACKING_SPARSE;
5216 
5217 		/*
5218 		 * If the first record we find does not overlap the start key,
5219 		 * then there is a hole at the start of the search range.
5220 		 * Classify this as sparse and stop immediately.
5221 		 */
5222 		if (xfs_btree_masked_keycmp_lt(cur, &info->start_key, &rec_key,
5223 					info->key_mask))
5224 			return -ECANCELED;
5225 	} else {
5226 		/*
5227 		 * If a subsequent record does not overlap with the any record
5228 		 * we've seen so far, there is a hole in the middle of the
5229 		 * search range.  Classify this as sparse and stop.
5230 		 * If the keys overlap and this btree does not allow overlap,
5231 		 * signal corruption.
5232 		 */
5233 		key_contig = cur->bc_ops->keys_contiguous(cur, &info->high_key,
5234 					&rec_key, info->key_mask);
5235 		if (key_contig == XBTREE_KEY_OVERLAP &&
5236 				!(cur->bc_ops->geom_flags & XFS_BTGEO_OVERLAPPING))
5237 			return -EFSCORRUPTED;
5238 		if (key_contig == XBTREE_KEY_GAP)
5239 			return -ECANCELED;
5240 	}
5241 
5242 	/*
5243 	 * If high_key(rec) is larger than any other high key we've seen,
5244 	 * remember it for later.
5245 	 */
5246 	cur->bc_ops->init_high_key_from_rec(&rec_high_key, rec);
5247 	if (xfs_btree_masked_keycmp_gt(cur, &rec_high_key, &info->high_key,
5248 				info->key_mask))
5249 		info->high_key = rec_high_key; /* struct copy */
5250 
5251 	return 0;
5252 }
5253 
5254 /*
5255  * Scan part of the keyspace of a btree and tell us if that keyspace does not
5256  * map to any records; is fully mapped to records; or is partially mapped to
5257  * records.  This is the btree record equivalent to determining if a file is
5258  * sparse.
5259  *
5260  * For most btree types, the record scan should use all available btree key
5261  * fields to compare the keys encountered.  These callers should pass NULL for
5262  * @mask.  However, some callers (e.g.  scanning physical space in the rmapbt)
5263  * want to ignore some part of the btree record keyspace when performing the
5264  * comparison.  These callers should pass in a union xfs_btree_key object with
5265  * the fields that *should* be a part of the comparison set to any nonzero
5266  * value, and the rest zeroed.
5267  */
5268 int
5269 xfs_btree_has_records(
5270 	struct xfs_btree_cur		*cur,
5271 	const union xfs_btree_irec	*low,
5272 	const union xfs_btree_irec	*high,
5273 	const union xfs_btree_key	*mask,
5274 	enum xbtree_recpacking		*outcome)
5275 {
5276 	struct xfs_btree_has_records	info = {
5277 		.outcome		= XBTREE_RECPACKING_EMPTY,
5278 		.key_mask		= mask,
5279 	};
5280 	int				error;
5281 
5282 	/* Not all btrees support this operation. */
5283 	if (!cur->bc_ops->keys_contiguous) {
5284 		ASSERT(0);
5285 		return -EOPNOTSUPP;
5286 	}
5287 
5288 	xfs_btree_key_from_irec(cur, &info.start_key, low);
5289 	xfs_btree_key_from_irec(cur, &info.end_key, high);
5290 
5291 	error = xfs_btree_query_range(cur, low, high,
5292 			xfs_btree_has_records_helper, &info);
5293 	if (error == -ECANCELED)
5294 		goto out;
5295 	if (error)
5296 		return error;
5297 
5298 	if (info.outcome == XBTREE_RECPACKING_EMPTY)
5299 		goto out;
5300 
5301 	/*
5302 	 * If the largest high_key(rec) we saw during the walk is greater than
5303 	 * the end of the search range, classify this as full.  Otherwise,
5304 	 * there is a hole at the end of the search range.
5305 	 */
5306 	if (xfs_btree_masked_keycmp_ge(cur, &info.high_key, &info.end_key,
5307 				mask))
5308 		info.outcome = XBTREE_RECPACKING_FULL;
5309 
5310 out:
5311 	*outcome = info.outcome;
5312 	return 0;
5313 }
5314 
5315 /* Are there more records in this btree? */
5316 bool
5317 xfs_btree_has_more_records(
5318 	struct xfs_btree_cur	*cur)
5319 {
5320 	struct xfs_btree_block	*block;
5321 	struct xfs_buf		*bp;
5322 
5323 	block = xfs_btree_get_block(cur, 0, &bp);
5324 
5325 	/* There are still records in this block. */
5326 	if (cur->bc_levels[0].ptr < xfs_btree_get_numrecs(block))
5327 		return true;
5328 
5329 	/* There are more record blocks. */
5330 	if (cur->bc_ops->ptr_len == XFS_BTREE_LONG_PTR_LEN)
5331 		return block->bb_u.l.bb_rightsib != cpu_to_be64(NULLFSBLOCK);
5332 	else
5333 		return block->bb_u.s.bb_rightsib != cpu_to_be32(NULLAGBLOCK);
5334 }
5335 
5336 /* Set up all the btree cursor caches. */
5337 int __init
5338 xfs_btree_init_cur_caches(void)
5339 {
5340 	int		error;
5341 
5342 	error = xfs_allocbt_init_cur_cache();
5343 	if (error)
5344 		return error;
5345 	error = xfs_inobt_init_cur_cache();
5346 	if (error)
5347 		goto err;
5348 	error = xfs_bmbt_init_cur_cache();
5349 	if (error)
5350 		goto err;
5351 	error = xfs_rmapbt_init_cur_cache();
5352 	if (error)
5353 		goto err;
5354 	error = xfs_refcountbt_init_cur_cache();
5355 	if (error)
5356 		goto err;
5357 
5358 	return 0;
5359 err:
5360 	xfs_btree_destroy_cur_caches();
5361 	return error;
5362 }
5363 
5364 /* Destroy all the btree cursor caches, if they've been allocated. */
5365 void
5366 xfs_btree_destroy_cur_caches(void)
5367 {
5368 	xfs_allocbt_destroy_cur_cache();
5369 	xfs_inobt_destroy_cur_cache();
5370 	xfs_bmbt_destroy_cur_cache();
5371 	xfs_rmapbt_destroy_cur_cache();
5372 	xfs_refcountbt_destroy_cur_cache();
5373 }
5374 
5375 /* Move the btree cursor before the first record. */
5376 int
5377 xfs_btree_goto_left_edge(
5378 	struct xfs_btree_cur	*cur)
5379 {
5380 	int			stat = 0;
5381 	int			error;
5382 
5383 	memset(&cur->bc_rec, 0, sizeof(cur->bc_rec));
5384 	error = xfs_btree_lookup(cur, XFS_LOOKUP_LE, &stat);
5385 	if (error)
5386 		return error;
5387 	if (!stat)
5388 		return 0;
5389 
5390 	error = xfs_btree_decrement(cur, 0, &stat);
5391 	if (error)
5392 		return error;
5393 	if (stat != 0) {
5394 		ASSERT(0);
5395 		xfs_btree_mark_sick(cur);
5396 		return -EFSCORRUPTED;
5397 	}
5398 
5399 	return 0;
5400 }
5401