/usr/src/usr.sbin/ldapd/btree.c

Bug Summary

File:	src/usr.sbin/ldapd/btree.c
Warning:	line 1027, column 19 Division by zero
Annotated Source Code

Press '?' to see keyboard shortcuts
Show analyzer invocation
clang -cc1 -cc1 -triple amd64-unknown-openbsd7.0 -analyze -disable-free -disable-llvm-verifier -discard-value-names -main-file-name btree.c -analyzer-store=region -analyzer-opt-analyze-nested-blocks -analyzer-checker=core -analyzer-checker=apiModeling -analyzer-checker=unix -analyzer-checker=deadcode -analyzer-checker=security.insecureAPI.UncheckedReturn -analyzer-checker=security.insecureAPI.getpw -analyzer-checker=security.insecureAPI.gets -analyzer-checker=security.insecureAPI.mktemp -analyzer-checker=security.insecureAPI.mkstemp -analyzer-checker=security.insecureAPI.vfork -analyzer-checker=nullability.NullPassedToNonnull -analyzer-checker=nullability.NullReturnedFromNonnull -analyzer-output plist -w -setup-static-analyzer -mrelocation-model pic -pic-level 1 -pic-is-pie -mframe-pointer=all -relaxed-aliasing -fno-rounding-math -mconstructor-aliases -munwind-tables -target-cpu x86-64 -target-feature +retpoline-indirect-calls -target-feature +retpoline-indirect-branches -tune-cpu generic -debugger-tuning=gdb -fcoverage-compilation-dir=/usr/src/usr.sbin/ldapd/obj -resource-dir /usr/local/lib/clang/13.0.0 -I /usr/src/usr.sbin/ldapd -internal-isystem /usr/local/lib/clang/13.0.0/include -internal-externc-isystem /usr/include -O2 -fdebug-compilation-dir=/usr/src/usr.sbin/ldapd/obj -ferror-limit 19 -fwrapv -D_RET_PROTECTOR -ret-protector -fgnuc-version=4.2.1 -vectorize-loops -vectorize-slp -fno-builtin-malloc -fno-builtin-calloc -fno-builtin-realloc -fno-builtin-valloc -fno-builtin-free -fno-builtin-strdup -fno-builtin-strndup -analyzer-output=html -faddrsig -D__GCC_HAVE_DWARF2_CFI_ASM=1 -o /home/ben/Projects/vmm/scan-build/2022-01-12-194120-40624-1 -x c /usr/src/usr.sbin/ldapd/btree.c
1/*	$OpenBSD: btree.c,v 1.38 2017/05/26 21:23:14 sthen Exp $ */

3/*
* Copyright (c) 2009, 2010 Martin Hedenfalk <martin@bzero.se>
*
* Permission to use, copy, modify, and distribute this software for any
* purpose with or without fee is hereby granted, provided that the above
* copyright notice and this permission notice appear in all copies.
*
* THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
* WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
* MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
* ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
* WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
* ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
* OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
*/

19#include <sys/types.h>
20#include <sys/tree.h>
21#include <sys/stat.h>
22#include <sys/queue.h>
23#include <sys/uio.h>

25#include <assert.h>
26#include <err.h>
27#include <errno(*__errno()).h>
28#include <fcntl.h>
29#include <stddef.h>
30#include <stdint.h>
31#include <stdio.h>
32#include <stdlib.h>
33#include <string.h>
34#include <time.h>
35#include <unistd.h>

37#include "btree.h"

39/* #define DEBUG */

41#ifdef DEBUG
42# define DPRINTF(...)	do { fprintf(stderr(&__sF[2]), "%s:%d: ", __func__, __LINE__42); \
	     fprintf(stderr(&__sF[2]), __VA_ARGS__); \
	     fprintf(stderr(&__sF[2]), "\n"); } while(0)
45#else
46# define DPRINTF(...)
47#endif

49#define PAGESIZE4096	 4096
50#define BT_MINKEYS4	 4
51#define BT_MAGIC0xB3DBB3DB	 0xB3DBB3DB
52#define BT_VERSION4	 4
53#define MAXKEYSIZE255	 255

55#define P_INVALID0xFFFFFFFF	 0xFFFFFFFF

57#define F_ISSET(w, f)(((w) & (f)) == (f))	 (((w) & (f)) == (f))
58#define MINIMUM(a,b)((a) < (b) ? (a) : (b))	 ((a) < (b) ? (a) : (b))

60typedef uint32_t	 pgno_t;
61typedef uint16_t	 indx_t;

63/* There are four page types: meta, index, leaf and overflow.
* They all share the same page header.
*/
66struct page {				/* represents an on-disk page */
pgno_t		 pgno;		/* page number */
68#define	P_BRANCH0x01	 0x01		/* branch page */
69#define	P_LEAF0x02		 0x02		/* leaf page */
70#define	P_OVERFLOW0x04	 0x04		/* overflow page */
71#define	P_META0x08		 0x08		/* meta page */
72#define	P_HEAD0x10		 0x10		/* header page */
uint32_t	 flags;
74#define lowerb.fb.fb_lower		 b.fb.fb_lower
75#define upperb.fb.fb_upper		 b.fb.fb_upper
76#define p_next_pgnob.pb_next_pgno	 b.pb_next_pgno
union page_bounds {
struct {
	indx_t	 fb_lower;	/* lower bound of free space */
	indx_t	 fb_upper;	/* upper bound of free space */
} fb;
pgno_t		 pb_next_pgno;	/* overflow page linked list */
} b;
indx_t		 ptrs[1];		/* dynamic size */
85} __packed__attribute__((__packed__));

87#define PAGEHDRSZ__builtin_offsetof(struct page, ptrs)	 offsetof(struct page, ptrs)__builtin_offsetof(struct page, ptrs)

89#define NUMKEYSP(p)(((p)->b.fb.fb_lower - __builtin_offsetof(struct page, ptrs
)) >> 1)	 (((p)->lowerb.fb.fb_lower - PAGEHDRSZ__builtin_offsetof(struct page, ptrs)) >> 1)
90#define NUMKEYS(mp)(((mp)->page->b.fb.fb_lower - __builtin_offsetof(struct
 page, ptrs)) >> 1)	 (((mp)->page->lowerb.fb.fb_lower - PAGEHDRSZ__builtin_offsetof(struct page, ptrs)) >> 1)
91#define SIZELEFT(mp)(indx_t)((mp)->page->b.fb.fb_upper - (mp)->page->
b.fb.fb_lower)	 (indx_t)((mp)->page->upperb.fb.fb_upper - (mp)->page->lowerb.fb.fb_lower)
92#define PAGEFILL(bt, mp)(1000 * ((bt)->head.psize - __builtin_offsetof(struct page
, ptrs) - (indx_t)((mp)->page->b.fb.fb_upper - (mp)->
page->b.fb.fb_lower)) / ((bt)->head.psize - __builtin_offsetof
(struct page, ptrs))) (1000 * ((bt)->head.psize - PAGEHDRSZ__builtin_offsetof(struct page, ptrs) - SIZELEFT(mp)(indx_t)((mp)->page->b.fb.fb_upper - (mp)->page->
b.fb.fb_lower)) / \
		((bt)->head.psize - PAGEHDRSZ__builtin_offsetof(struct page, ptrs)))
94#define IS_LEAF(mp)((((mp)->page->flags) & (0x02)) == (0x02))	 F_ISSET((mp)->page->flags, P_LEAF)((((mp)->page->flags) & (0x02)) == (0x02))
95#define IS_BRANCH(mp)((((mp)->page->flags) & (0x01)) == (0x01))	 F_ISSET((mp)->page->flags, P_BRANCH)((((mp)->page->flags) & (0x01)) == (0x01))
96#define IS_OVERFLOW(mp)((((mp)->page->flags) & (0x04)) == (0x04))	 F_ISSET((mp)->page->flags, P_OVERFLOW)((((mp)->page->flags) & (0x04)) == (0x04))

98struct bt_head {				/* header page content */
uint32_t	 magic;
uint32_t	 version;
uint32_t	 flags;
uint32_t	 psize;			/* page size */
103} __packed__attribute__((__packed__));

105struct bt_meta {				/* meta (footer) page content */
106#define BT_TOMBSTONE0x01	 0x01			/* file is replaced */
uint32_t	 flags;
pgno_t		 root;			/* page number of root page */
pgno_t		 prev_meta;		/* previous meta page number */
time_t		 created_at;
uint32_t	 branch_pages;
uint32_t	 leaf_pages;
uint32_t	 overflow_pages;
uint32_t	 revisions;
uint32_t	 depth;
uint64_t	 entries;
unsigned char	 hash[SHA_DIGEST_LENGTH20];
118} __packed__attribute__((__packed__));

120struct btkey {
size_t			 len;
char			 str[MAXKEYSIZE255];
123};

125struct mpage {					/* an in-memory cached page */
RB_ENTRY(mpage)struct { struct mpage *rbe_left; struct mpage *rbe_right; struct
 mpage *rbe_parent; int rbe_color; }		 entry;		/* page cache entry */
SIMPLEQ_ENTRY(mpage)struct { struct mpage *sqe_next; }	 next;		/* queue of dirty pages */
TAILQ_ENTRY(mpage)struct { struct mpage *tqe_next; struct mpage **tqe_prev; }	 lru_next;	/* LRU queue */
struct mpage		*parent;	/* NULL if root */
unsigned int		 parent_index;	/* keep track of node index */
struct btkey		 prefix;
struct page		*page;
pgno_t			 pgno;		/* copy of page->pgno */
short			 ref;		/* increased by cursors */
short			 dirty;		/* 1 if on dirty queue */
136};
137RB_HEAD(page_cache, mpage)struct page_cache { struct mpage *rbh_root; };
138SIMPLEQ_HEAD(dirty_queue, mpage)struct dirty_queue { struct mpage *sqh_first; struct mpage **
sqh_last; };
139TAILQ_HEAD(lru_queue, mpage)struct lru_queue { struct mpage *tqh_first; struct mpage **tqh_last
; };

141static int		 mpage_cmp(struct mpage *a, struct mpage *b);
142static struct mpage	*mpage_lookup(struct btree *bt, pgno_t pgno);
143static void		 mpage_add(struct btree *bt, struct mpage *mp);
144static void		 mpage_free(struct mpage *mp);
145static void		 mpage_del(struct btree *bt, struct mpage *mp);
146static void		 mpage_flush(struct btree *bt);
147static struct mpage	*mpage_copy(struct btree *bt, struct mpage *mp);
148static void		 mpage_prune(struct btree *bt);
149static void		 mpage_dirty(struct btree *bt, struct mpage *mp);
150static struct mpage	*mpage_touch(struct btree *bt, struct mpage *mp);

152RB_PROTOTYPE(page_cache, mpage, entry, mpage_cmp)void page_cache_RB_INSERT_COLOR(struct page_cache *, struct mpage
 *); void page_cache_RB_REMOVE_COLOR(struct page_cache *, struct
 mpage *, struct mpage *); struct mpage *page_cache_RB_REMOVE
(struct page_cache *, struct mpage *); struct mpage *page_cache_RB_INSERT
(struct page_cache *, struct mpage *); struct mpage *page_cache_RB_FIND
(struct page_cache *, struct mpage *); struct mpage *page_cache_RB_NFIND
(struct page_cache *, struct mpage *); struct mpage *page_cache_RB_NEXT
(struct mpage *); struct mpage *page_cache_RB_PREV(struct mpage
 *); struct mpage *page_cache_RB_MINMAX(struct page_cache *, int
);;
153RB_GENERATE(page_cache, mpage, entry, mpage_cmp)void page_cache_RB_INSERT_COLOR(struct page_cache *head, struct
 mpage *elm) { struct mpage *parent, *gparent, *tmp; while ((
parent = (elm)->entry.rbe_parent) && (parent)->
entry.rbe_color == 1) { gparent = (parent)->entry.rbe_parent
; if (parent == (gparent)->entry.rbe_left) { tmp = (gparent
)->entry.rbe_right; if (tmp && (tmp)->entry.rbe_color
 == 1) { (tmp)->entry.rbe_color = 0; do { (parent)->entry
.rbe_color = 0; (gparent)->entry.rbe_color = 1; } while (0
); elm = gparent; continue; } if ((parent)->entry.rbe_right
 == elm) { do { (tmp) = (parent)->entry.rbe_right; if (((parent
)->entry.rbe_right = (tmp)->entry.rbe_left)) { ((tmp)->
entry.rbe_left)->entry.rbe_parent = (parent); } do {} while
 (0); if (((tmp)->entry.rbe_parent = (parent)->entry.rbe_parent
)) { if ((parent) == ((parent)->entry.rbe_parent)->entry
.rbe_left) ((parent)->entry.rbe_parent)->entry.rbe_left
 = (tmp); else ((parent)->entry.rbe_parent)->entry.rbe_right
 = (tmp); } else (head)->rbh_root = (tmp); (tmp)->entry
.rbe_left = (parent); (parent)->entry.rbe_parent = (tmp); do
 {} while (0); if (((tmp)->entry.rbe_parent)) do {} while (
0); } while (0); tmp = parent; parent = elm; elm = tmp; } do {
 (parent)->entry.rbe_color = 0; (gparent)->entry.rbe_color
 = 1; } while (0); do { (tmp) = (gparent)->entry.rbe_left;
 if (((gparent)->entry.rbe_left = (tmp)->entry.rbe_right
)) { ((tmp)->entry.rbe_right)->entry.rbe_parent = (gparent
); } do {} while (0); if (((tmp)->entry.rbe_parent = (gparent
)->entry.rbe_parent)) { if ((gparent) == ((gparent)->entry
.rbe_parent)->entry.rbe_left) ((gparent)->entry.rbe_parent
)->entry.rbe_left = (tmp); else ((gparent)->entry.rbe_parent
)->entry.rbe_right = (tmp); } else (head)->rbh_root = (
tmp); (tmp)->entry.rbe_right = (gparent); (gparent)->entry
.rbe_parent = (tmp); do {} while (0); if (((tmp)->entry.rbe_parent
)) do {} while (0); } while (0); } else { tmp = (gparent)->
entry.rbe_left; if (tmp && (tmp)->entry.rbe_color ==
 1) { (tmp)->entry.rbe_color = 0; do { (parent)->entry.
rbe_color = 0; (gparent)->entry.rbe_color = 1; } while (0)
; elm = gparent; continue; } if ((parent)->entry.rbe_left ==
 elm) { do { (tmp) = (parent)->entry.rbe_left; if (((parent
)->entry.rbe_left = (tmp)->entry.rbe_right)) { ((tmp)->
entry.rbe_right)->entry.rbe_parent = (parent); } do {} while
 (0); if (((tmp)->entry.rbe_parent = (parent)->entry.rbe_parent
)) { if ((parent) == ((parent)->entry.rbe_parent)->entry
.rbe_left) ((parent)->entry.rbe_parent)->entry.rbe_left
 = (tmp); else ((parent)->entry.rbe_parent)->entry.rbe_right
 = (tmp); } else (head)->rbh_root = (tmp); (tmp)->entry
.rbe_right = (parent); (parent)->entry.rbe_parent = (tmp);
 do {} while (0); if (((tmp)->entry.rbe_parent)) do {} while
 (0); } while (0); tmp = parent; parent = elm; elm = tmp; } do
 { (parent)->entry.rbe_color = 0; (gparent)->entry.rbe_color
 = 1; } while (0); do { (tmp) = (gparent)->entry.rbe_right
; if (((gparent)->entry.rbe_right = (tmp)->entry.rbe_left
)) { ((tmp)->entry.rbe_left)->entry.rbe_parent = (gparent
); } do {} while (0); if (((tmp)->entry.rbe_parent = (gparent
)->entry.rbe_parent)) { if ((gparent) == ((gparent)->entry
.rbe_parent)->entry.rbe_left) ((gparent)->entry.rbe_parent
)->entry.rbe_left = (tmp); else ((gparent)->entry.rbe_parent
)->entry.rbe_right = (tmp); } else (head)->rbh_root = (
tmp); (tmp)->entry.rbe_left = (gparent); (gparent)->entry
.rbe_parent = (tmp); do {} while (0); if (((tmp)->entry.rbe_parent
)) do {} while (0); } while (0); } } (head->rbh_root)->
entry.rbe_color = 0; } void page_cache_RB_REMOVE_COLOR(struct
 page_cache *head, struct mpage *parent, struct mpage *elm) {
 struct mpage *tmp; while ((elm == ((void*)0) || (elm)->entry
.rbe_color == 0) && elm != (head)->rbh_root) { if (
(parent)->entry.rbe_left == elm) { tmp = (parent)->entry
.rbe_right; if ((tmp)->entry.rbe_color == 1) { do { (tmp)->
entry.rbe_color = 0; (parent)->entry.rbe_color = 1; } while
 (0); do { (tmp) = (parent)->entry.rbe_right; if (((parent
)->entry.rbe_right = (tmp)->entry.rbe_left)) { ((tmp)->
entry.rbe_left)->entry.rbe_parent = (parent); } do {} while
 (0); if (((tmp)->entry.rbe_parent = (parent)->entry.rbe_parent
)) { if ((parent) == ((parent)->entry.rbe_parent)->entry
.rbe_left) ((parent)->entry.rbe_parent)->entry.rbe_left
 = (tmp); else ((parent)->entry.rbe_parent)->entry.rbe_right
 = (tmp); } else (head)->rbh_root = (tmp); (tmp)->entry
.rbe_left = (parent); (parent)->entry.rbe_parent = (tmp); do
 {} while (0); if (((tmp)->entry.rbe_parent)) do {} while (
0); } while (0); tmp = (parent)->entry.rbe_right; } if (((
tmp)->entry.rbe_left == ((void*)0) || ((tmp)->entry.rbe_left
)->entry.rbe_color == 0) && ((tmp)->entry.rbe_right
 == ((void*)0) || ((tmp)->entry.rbe_right)->entry.rbe_color
 == 0)) { (tmp)->entry.rbe_color = 1; elm = parent; parent
 = (elm)->entry.rbe_parent; } else { if ((tmp)->entry.rbe_right
 == ((void*)0) || ((tmp)->entry.rbe_right)->entry.rbe_color
 == 0) { struct mpage *oleft; if ((oleft = (tmp)->entry.rbe_left
)) (oleft)->entry.rbe_color = 0; (tmp)->entry.rbe_color
 = 1; do { (oleft) = (tmp)->entry.rbe_left; if (((tmp)->
entry.rbe_left = (oleft)->entry.rbe_right)) { ((oleft)->
entry.rbe_right)->entry.rbe_parent = (tmp); } do {} while (
0); if (((oleft)->entry.rbe_parent = (tmp)->entry.rbe_parent
)) { if ((tmp) == ((tmp)->entry.rbe_parent)->entry.rbe_left
) ((tmp)->entry.rbe_parent)->entry.rbe_left = (oleft); else
 ((tmp)->entry.rbe_parent)->entry.rbe_right = (oleft); }
 else (head)->rbh_root = (oleft); (oleft)->entry.rbe_right
 = (tmp); (tmp)->entry.rbe_parent = (oleft); do {} while (
0); if (((oleft)->entry.rbe_parent)) do {} while (0); } while
 (0); tmp = (parent)->entry.rbe_right; } (tmp)->entry.rbe_color
 = (parent)->entry.rbe_color; (parent)->entry.rbe_color
 = 0; if ((tmp)->entry.rbe_right) ((tmp)->entry.rbe_right
)->entry.rbe_color = 0; do { (tmp) = (parent)->entry.rbe_right
; if (((parent)->entry.rbe_right = (tmp)->entry.rbe_left
)) { ((tmp)->entry.rbe_left)->entry.rbe_parent = (parent
); } do {} while (0); if (((tmp)->entry.rbe_parent = (parent
)->entry.rbe_parent)) { if ((parent) == ((parent)->entry
.rbe_parent)->entry.rbe_left) ((parent)->entry.rbe_parent
)->entry.rbe_left = (tmp); else ((parent)->entry.rbe_parent
)->entry.rbe_right = (tmp); } else (head)->rbh_root = (
tmp); (tmp)->entry.rbe_left = (parent); (parent)->entry
.rbe_parent = (tmp); do {} while (0); if (((tmp)->entry.rbe_parent
)) do {} while (0); } while (0); elm = (head)->rbh_root; break
; } } else { tmp = (parent)->entry.rbe_left; if ((tmp)->
entry.rbe_color == 1) { do { (tmp)->entry.rbe_color = 0; (
parent)->entry.rbe_color = 1; } while (0); do { (tmp) = (parent
)->entry.rbe_left; if (((parent)->entry.rbe_left = (tmp
)->entry.rbe_right)) { ((tmp)->entry.rbe_right)->entry
.rbe_parent = (parent); } do {} while (0); if (((tmp)->entry
.rbe_parent = (parent)->entry.rbe_parent)) { if ((parent) ==
 ((parent)->entry.rbe_parent)->entry.rbe_left) ((parent
)->entry.rbe_parent)->entry.rbe_left = (tmp); else ((parent
)->entry.rbe_parent)->entry.rbe_right = (tmp); } else (
head)->rbh_root = (tmp); (tmp)->entry.rbe_right = (parent
); (parent)->entry.rbe_parent = (tmp); do {} while (0); if
 (((tmp)->entry.rbe_parent)) do {} while (0); } while (0);
 tmp = (parent)->entry.rbe_left; } if (((tmp)->entry.rbe_left
 == ((void*)0) || ((tmp)->entry.rbe_left)->entry.rbe_color
 == 0) && ((tmp)->entry.rbe_right == ((void*)0) ||
 ((tmp)->entry.rbe_right)->entry.rbe_color == 0)) { (tmp
)->entry.rbe_color = 1; elm = parent; parent = (elm)->entry
.rbe_parent; } else { if ((tmp)->entry.rbe_left == ((void*
)0) || ((tmp)->entry.rbe_left)->entry.rbe_color == 0) {
 struct mpage *oright; if ((oright = (tmp)->entry.rbe_right
)) (oright)->entry.rbe_color = 0; (tmp)->entry.rbe_color
 = 1; do { (oright) = (tmp)->entry.rbe_right; if (((tmp)->
entry.rbe_right = (oright)->entry.rbe_left)) { ((oright)->
entry.rbe_left)->entry.rbe_parent = (tmp); } do {} while (
0); if (((oright)->entry.rbe_parent = (tmp)->entry.rbe_parent
)) { if ((tmp) == ((tmp)->entry.rbe_parent)->entry.rbe_left
) ((tmp)->entry.rbe_parent)->entry.rbe_left = (oright);
 else ((tmp)->entry.rbe_parent)->entry.rbe_right = (oright
); } else (head)->rbh_root = (oright); (oright)->entry.
rbe_left = (tmp); (tmp)->entry.rbe_parent = (oright); do {
} while (0); if (((oright)->entry.rbe_parent)) do {} while
 (0); } while (0); tmp = (parent)->entry.rbe_left; } (tmp)
->entry.rbe_color = (parent)->entry.rbe_color; (parent)
->entry.rbe_color = 0; if ((tmp)->entry.rbe_left) ((tmp
)->entry.rbe_left)->entry.rbe_color = 0; do { (tmp) = (
parent)->entry.rbe_left; if (((parent)->entry.rbe_left =
 (tmp)->entry.rbe_right)) { ((tmp)->entry.rbe_right)->
entry.rbe_parent = (parent); } do {} while (0); if (((tmp)->
entry.rbe_parent = (parent)->entry.rbe_parent)) { if ((parent
) == ((parent)->entry.rbe_parent)->entry.rbe_left) ((parent
)->entry.rbe_parent)->entry.rbe_left = (tmp); else ((parent
)->entry.rbe_parent)->entry.rbe_right = (tmp); } else (
head)->rbh_root = (tmp); (tmp)->entry.rbe_right = (parent
); (parent)->entry.rbe_parent = (tmp); do {} while (0); if
 (((tmp)->entry.rbe_parent)) do {} while (0); } while (0);
 elm = (head)->rbh_root; break; } } } if (elm) (elm)->entry
.rbe_color = 0; } struct mpage * page_cache_RB_REMOVE(struct page_cache
 *head, struct mpage *elm) { struct mpage *child, *parent, *old
 = elm; int color; if ((elm)->entry.rbe_left == ((void*)0)
) child = (elm)->entry.rbe_right; else if ((elm)->entry
.rbe_right == ((void*)0)) child = (elm)->entry.rbe_left; else
 { struct mpage *left; elm = (elm)->entry.rbe_right; while
 ((left = (elm)->entry.rbe_left)) elm = left; child = (elm
)->entry.rbe_right; parent = (elm)->entry.rbe_parent; color
 = (elm)->entry.rbe_color; if (child) (child)->entry.rbe_parent
 = parent; if (parent) { if ((parent)->entry.rbe_left == elm
) (parent)->entry.rbe_left = child; else (parent)->entry
.rbe_right = child; do {} while (0); } else (head)->rbh_root
 = child; if ((elm)->entry.rbe_parent == old) parent = elm
; (elm)->entry = (old)->entry; if ((old)->entry.rbe_parent
) { if (((old)->entry.rbe_parent)->entry.rbe_left == old
) ((old)->entry.rbe_parent)->entry.rbe_left = elm; else
 ((old)->entry.rbe_parent)->entry.rbe_right = elm; do {
} while (0); } else (head)->rbh_root = elm; ((old)->entry
.rbe_left)->entry.rbe_parent = elm; if ((old)->entry.rbe_right
) ((old)->entry.rbe_right)->entry.rbe_parent = elm; if (
parent) { left = parent; do { do {} while (0); } while ((left
 = (left)->entry.rbe_parent)); } goto color; } parent = (elm
)->entry.rbe_parent; color = (elm)->entry.rbe_color; if
 (child) (child)->entry.rbe_parent = parent; if (parent) {
 if ((parent)->entry.rbe_left == elm) (parent)->entry.rbe_left
 = child; else (parent)->entry.rbe_right = child; do {} while
 (0); } else (head)->rbh_root = child; color: if (color ==
 0) page_cache_RB_REMOVE_COLOR(head, parent, child); return (
old); } struct mpage * page_cache_RB_INSERT(struct page_cache
 *head, struct mpage *elm) { struct mpage *tmp; struct mpage *
parent = ((void*)0); int comp = 0; tmp = (head)->rbh_root;
 while (tmp) { parent = tmp; comp = (mpage_cmp)(elm, parent);
 if (comp < 0) tmp = (tmp)->entry.rbe_left; else if (comp
 > 0) tmp = (tmp)->entry.rbe_right; else return (tmp); }
 do { (elm)->entry.rbe_parent = parent; (elm)->entry.rbe_left
 = (elm)->entry.rbe_right = ((void*)0); (elm)->entry.rbe_color
 = 1; } while (0); if (parent != ((void*)0)) { if (comp < 0
) (parent)->entry.rbe_left = elm; else (parent)->entry.
rbe_right = elm; do {} while (0); } else (head)->rbh_root =
 elm; page_cache_RB_INSERT_COLOR(head, elm); return (((void*)
0)); } struct mpage * page_cache_RB_FIND(struct page_cache *head
, struct mpage *elm) { struct mpage *tmp = (head)->rbh_root
; int comp; while (tmp) { comp = mpage_cmp(elm, tmp); if (comp
 < 0) tmp = (tmp)->entry.rbe_left; else if (comp > 0
) tmp = (tmp)->entry.rbe_right; else return (tmp); } return
 (((void*)0)); } struct mpage * page_cache_RB_NFIND(struct page_cache
 *head, struct mpage *elm) { struct mpage *tmp = (head)->rbh_root
; struct mpage *res = ((void*)0); int comp; while (tmp) { comp
 = mpage_cmp(elm, tmp); if (comp < 0) { res = tmp; tmp = (
tmp)->entry.rbe_left; } else if (comp > 0) tmp = (tmp)->
entry.rbe_right; else return (tmp); } return (res); } struct mpage
 * page_cache_RB_NEXT(struct mpage *elm) { if ((elm)->entry
.rbe_right) { elm = (elm)->entry.rbe_right; while ((elm)->
entry.rbe_left) elm = (elm)->entry.rbe_left; } else { if (
(elm)->entry.rbe_parent && (elm == ((elm)->entry
.rbe_parent)->entry.rbe_left)) elm = (elm)->entry.rbe_parent
; else { while ((elm)->entry.rbe_parent && (elm ==
 ((elm)->entry.rbe_parent)->entry.rbe_right)) elm = (elm
)->entry.rbe_parent; elm = (elm)->entry.rbe_parent; } }
 return (elm); } struct mpage * page_cache_RB_PREV(struct mpage
 *elm) { if ((elm)->entry.rbe_left) { elm = (elm)->entry
.rbe_left; while ((elm)->entry.rbe_right) elm = (elm)->
entry.rbe_right; } else { if ((elm)->entry.rbe_parent &&
 (elm == ((elm)->entry.rbe_parent)->entry.rbe_right)) elm
 = (elm)->entry.rbe_parent; else { while ((elm)->entry.
rbe_parent && (elm == ((elm)->entry.rbe_parent)->
entry.rbe_left)) elm = (elm)->entry.rbe_parent; elm = (elm
)->entry.rbe_parent; } } return (elm); } struct mpage * page_cache_RB_MINMAX
(struct page_cache *head, int val) { struct mpage *tmp = (head
)->rbh_root; struct mpage *parent = ((void*)0); while (tmp
) { parent = tmp; if (val < 0) tmp = (tmp)->entry.rbe_left
; else tmp = (tmp)->entry.rbe_right; } return (parent); };

155struct ppage {					/* ordered list of pages */
SLIST_ENTRY(ppage)struct { struct ppage *sle_next; }	 entry;
struct mpage		*mpage;
unsigned int		 ki;		/* cursor index on page */
159};
160SLIST_HEAD(page_stack, ppage)struct page_stack { struct ppage *slh_first; };

162#define CURSOR_EMPTY(c)(((&(c)->stack)->slh_first) == ((void*)0))		 SLIST_EMPTY(&(c)->stack)(((&(c)->stack)->slh_first) == ((void*)0))
163#define CURSOR_TOP(c)((&(c)->stack)->slh_first)		 SLIST_FIRST(&(c)->stack)((&(c)->stack)->slh_first)
164#define CURSOR_POP(c)do { (&(c)->stack)->slh_first = (&(c)->stack
)->slh_first->entry.sle_next; } while (0)		 SLIST_REMOVE_HEAD(&(c)->stack, entry)do { (&(c)->stack)->slh_first = (&(c)->stack
)->slh_first->entry.sle_next; } while (0)
165#define CURSOR_PUSH(c,p)do { (p)->entry.sle_next = (&(c)->stack)->slh_first
; (&(c)->stack)->slh_first = (p); } while (0)	 SLIST_INSERT_HEAD(&(c)->stack, p, entry)do { (p)->entry.sle_next = (&(c)->stack)->slh_first
; (&(c)->stack)->slh_first = (p); } while (0)

167struct cursor {
struct btree		*bt;
struct btree_txn	*txn;
struct page_stack	 stack;		/* stack of parent pages */
short			 initialized;	/* 1 if initialized */
short			 eof;		/* 1 if end is reached */
173};

175#define METAHASHLEN__builtin_offsetof(struct bt_meta, hash)	 offsetof(struct bt_meta, hash)__builtin_offsetof(struct bt_meta, hash)
176#define METADATA(p)((void *)((char *)p + __builtin_offsetof(struct page, ptrs)))	 ((void *)((char *)p + PAGEHDRSZ__builtin_offsetof(struct page, ptrs)))

178struct node {
179#define n_pgnop.np_pgno		 p.np_pgno
180#define n_dsizep.np_dsize		 p.np_dsize
union {
pgno_t		 np_pgno;	/* child page number */
uint32_t	 np_dsize;	/* leaf data size */
}		 p;
uint16_t	 ksize;			/* key size */
186#define F_BIGDATA0x01	 0x01			/* data put on overflow page */
uint8_t		 flags;
char		 data[1];
189} __packed__attribute__((__packed__));

191struct btree_txn {
pgno_t			 root;		/* current / new root page */
pgno_t			 next_pgno;	/* next unallocated page */
struct btree		*bt;		/* btree is ref'd */
struct dirty_queue	*dirty_queue;	/* modified pages */
196#define BT_TXN_RDONLY0x01		 0x01		/* read-only transaction */
197#define BT_TXN_ERROR0x02		 0x02		/* an error has occurred */
unsigned int		 flags;
199};

201struct btree {
int			 fd;
char			*path;
204#define BT_FIXPADDING0x01		 0x01		/* internal */
unsigned int		 flags;
bt_cmp_func		 cmp;		/* user compare function */
struct bt_head		 head;
struct bt_meta		 meta;
struct page_cache	*page_cache;
struct lru_queue	*lru_queue;
struct btree_txn	*txn;		/* current write transaction */
int			 ref;		/* increased by cursors & txn */
struct btree_stat	 stat;
off_t			 size;		/* current file size */
215};

217#define NODESIZE__builtin_offsetof(struct node, data)	 offsetof(struct node, data)__builtin_offsetof(struct node, data)

219#define INDXSIZE(k)(__builtin_offsetof(struct node, data) + ((k) == ((void*)0) ?
: (k)->size))	 (NODESIZE__builtin_offsetof(struct node, data) + ((k) == NULL((void*)0) ? 0 : (k)->size))
220#define LEAFSIZE(k, d)(__builtin_offsetof(struct node, data) + (k)->size + (d)->
size)	 (NODESIZE__builtin_offsetof(struct node, data) + (k)->size + (d)->size)
221#define NODEPTRP(p, i)((struct node *)((char *)(p) + (p)->ptrs[i]))	 ((struct node *)((char *)(p) + (p)->ptrs[i]))
222#define NODEPTR(mp, i)((struct node *)((char *)((mp)->page) + ((mp)->page)->
ptrs[i]))	 NODEPTRP((mp)->page, i)((struct node *)((char *)((mp)->page) + ((mp)->page)->
ptrs[i]))
223#define NODEKEY(node)(void *)((node)->data)	 (void *)((node)->data)
224#define NODEDATA(node)(void *)((char *)(node)->data + (node)->ksize)	 (void *)((char *)(node)->data + (node)->ksize)
225#define NODEPGNO(node)((node)->p.np_pgno)	 ((node)->p.np_pgno)
226#define NODEDSZ(node)((node)->p.np_dsize)	 ((node)->p.np_dsize)

228#define BT_COMMIT_PAGES64	 64	/* max number of pages to write in one commit */
229#define BT_MAXCACHE_DEF1024	 1024	/* max number of pages to keep in cache  */

231static int		 btree_read_page(struct btree *bt, pgno_t pgno,
	    struct page *page);
233static struct mpage	*btree_get_mpage(struct btree *bt, pgno_t pgno);
234static int		 btree_search_page_root(struct btree *bt,
	    struct mpage *root, struct btval *key,
	    struct cursor *cursor, int modify,
	    struct mpage **mpp);
238static int		 btree_search_page(struct btree *bt,
	    struct btree_txn *txn, struct btval *key,
	    struct cursor *cursor, int modify,
	    struct mpage **mpp);

243static int		 btree_write_header(struct btree *bt, int fd);
244static int		 btree_read_header(struct btree *bt);
245static int		 btree_is_meta_page(struct page *p);
246static int		 btree_read_meta(struct btree *bt, pgno_t *p_next);
247static int		 btree_write_meta(struct btree *bt, pgno_t root,
	    unsigned int flags);
249static void		 btree_ref(struct btree *bt);

251static struct node	*btree_search_node(struct btree *bt, struct mpage *mp,
	    struct btval *key, int *exactp, unsigned int *kip);
253static int		 btree_add_node(struct btree *bt, struct mpage *mp,
	    indx_t indx, struct btval *key, struct btval *data,
	    pgno_t pgno, uint8_t flags);
256static void		 btree_del_node(struct btree *bt, struct mpage *mp,
	    indx_t indx);
258static int		 btree_read_data(struct btree *bt, struct mpage *mp,
	    struct node *leaf, struct btval *data);

261static int		 btree_rebalance(struct btree *bt, struct mpage *mp);
262static int		 btree_update_key(struct btree *bt, struct mpage *mp,
	    indx_t indx, struct btval *key);
264static int		 btree_adjust_prefix(struct btree *bt,
	    struct mpage *src, int delta);
266static int		 btree_move_node(struct btree *bt, struct mpage *src,
	    indx_t srcindx, struct mpage *dst, indx_t dstindx);
268static int		 btree_merge(struct btree *bt, struct mpage *src,
	    struct mpage *dst);
270static int		 btree_split(struct btree *bt, struct mpage **mpp,
	    unsigned int *newindxp, struct btval *newkey,
	    struct btval *newdata, pgno_t newpgno);
273static struct mpage	*btree_new_page(struct btree *bt, uint32_t flags);
274static int		 btree_write_overflow_data(struct btree *bt,
	    struct page *p, struct btval *data);

277static void		 cursor_pop_page(struct cursor *cursor);
278static struct ppage	 *cursor_push_page(struct cursor *cursor,
	    struct mpage *mp);

281static int		 bt_set_key(struct btree *bt, struct mpage *mp,
	    struct node *node, struct btval *key);
283static int		 btree_sibling(struct cursor *cursor, int move_right);
284static int		 btree_cursor_next(struct cursor *cursor,
	    struct btval *key, struct btval *data);
286static int		 btree_cursor_set(struct cursor *cursor,
	    struct btval *key, struct btval *data, int *exactp);
288static int		 btree_cursor_first(struct cursor *cursor,
	    struct btval *key, struct btval *data);

291static void		 bt_reduce_separator(struct btree *bt, struct node *min,
	    struct btval *sep);
293static void		 remove_prefix(struct btree *bt, struct btval *key,
	    size_t pfxlen);
295static void		 expand_prefix(struct btree *bt, struct mpage *mp,
	    indx_t indx, struct btkey *expkey);
297static void		 concat_prefix(struct btree *bt, char *s1, size_t n1,
	    char *s2, size_t n2, char *cs, size_t *cn);
299static void		 common_prefix(struct btree *bt, struct btkey *min,
	    struct btkey *max, struct btkey *pfx);
301static void		 find_common_prefix(struct btree *bt, struct mpage *mp);

303static size_t		 bt_leaf_size(struct btree *bt, struct btval *key,
	    struct btval *data);
305static size_t		 bt_branch_size(struct btree *bt, struct btval *key);

307static pgno_t		 btree_compact_tree(struct btree *bt, pgno_t pgno,
	    struct btree *btc);

310static int		 memncmp(const void *s1, size_t n1,
		 const void *s2, size_t n2);
312static int		 memnrcmp(const void *s1, size_t n1,
		  const void *s2, size_t n2);

315static int
316memncmp(const void *s1, size_t n1, const void *s2, size_t n2)
317{
if (n1 < n2) {
if (memcmp(s1, s2, n1) == 0)
	return -1;
}
else if (n1 > n2) {
if (memcmp(s1, s2, n2) == 0)
	return 1;
}
return memcmp(s1, s2, n1);
327}

329static int
330memnrcmp(const void *s1, size_t n1, const void *s2, size_t n2)
331{
const unsigned char	*p1;
const unsigned char	*p2;

if (n1 == 0)
return n2 == 0 ? 0 : -1;

if (n2 == 0)
return n1 == 0 ? 0 : 1;

p1 = (const unsigned char *)s1 + n1 - 1;
p2 = (const unsigned char *)s2 + n2 - 1;

while (*p1 == *p2) {
if (p1 == s1)
	return (p2 == s2) ? 0 : -1;
if (p2 == s2)
	return (p1 == p2) ? 0 : 1;
p1--;
p2--;
}
return *p1 - *p2;
353}

355int
356btree_cmp(struct btree *bt, const struct btval *a, const struct btval *b)
357{
return bt->cmp(a, b);
359}

361static void
362common_prefix(struct btree *bt, struct btkey *min, struct btkey *max,
  struct btkey *pfx)
364{
size_t		 n = 0;
char		*p1;
char		*p2;

if (min->len == 0 || max->len == 0) {
pfx->len = 0;
return;
}

if (F_ISSET(bt->flags, BT_REVERSEKEY)(((bt->flags) & (0x08)) == (0x08))) {
p1 = min->str + min->len - 1;
p2 = max->str + max->len - 1;

while (*p1 == *p2) {
	if (p1 < min->str || p2 < max->str)
		break;
	p1--;
	p2--;
	n++;
}

assert(n <= (int)sizeof(pfx->str))((n <= (int)sizeof(pfx->str)) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 386, __func__, "n <= (int)sizeof(pfx->str)"));
pfx->len = n;
bcopy(p2 + 1, pfx->str, n);
} else {
p1 = min->str;
p2 = max->str;

while (*p1 == *p2) {
	if (n == min->len || n == max->len)
		break;
	p1++;
	p2++;
	n++;
}

assert(n <= (int)sizeof(pfx->str))((n <= (int)sizeof(pfx->str)) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 401, __func__, "n <= (int)sizeof(pfx->str)"));
pfx->len = n;
bcopy(max->str, pfx->str, n);
}
405}

407static void
408remove_prefix(struct btree *bt, struct btval *key, size_t pfxlen)
409{
if (pfxlen == 0 || bt->cmp != NULL((void*)0))
return;

DPRINTF("removing %zu bytes of prefix from key [%.*s]", pfxlen,
   (int)key->size, (char *)key->data);
assert(pfxlen <= key->size)((pfxlen <= key->size) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 415, __func__, "pfxlen <= key->size"));
key->size -= pfxlen;
if (!F_ISSET(bt->flags, BT_REVERSEKEY)(((bt->flags) & (0x08)) == (0x08)))
key->data = (char *)key->data + pfxlen;
419}

421static void
422expand_prefix(struct btree *bt, struct mpage *mp, indx_t indx,
  struct btkey *expkey)
424{
struct node	*node;

node = NODEPTR(mp, indx)((struct node *)((char *)((mp)->page) + ((mp)->page)->
ptrs[indx]));
expkey->len = sizeof(expkey->str);
concat_prefix(bt, mp->prefix.str, mp->prefix.len,
   NODEKEY(node)(void *)((node)->data), node->ksize, expkey->str, &expkey->len);
431}

433static int
434bt_cmp(struct btree *bt, const struct btval *key1, const struct btval *key2,
  struct btkey *pfx)
436{
if (F_ISSET(bt->flags, BT_REVERSEKEY)(((bt->flags) & (0x08)) == (0x08)))
return memnrcmp(key1->data, key1->size - pfx->len,
    key2->data, key2->size);
else
return memncmp((char *)key1->data + pfx->len, key1->size - pfx->len,
    key2->data, key2->size);
443}

445void
446btval_reset(struct btval *btv)
447{
if (btv) {
if (btv->mp)
	btv->mp->ref--;
if (btv->free_data)
	free(btv->data);
memset(btv, 0, sizeof(*btv));
}
455}

457static int
458mpage_cmp(struct mpage *a, struct mpage *b)
459{
if (a->pgno > b->pgno)
return 1;
if (a->pgno < b->pgno)
return -1;
return 0;
465}

467static struct mpage *
468mpage_lookup(struct btree *bt, pgno_t pgno)
469{
struct mpage	 find, *mp;

find.pgno = pgno;
mp = RB_FIND(page_cache, bt->page_cache, &find)page_cache_RB_FIND(bt->page_cache, &find);
if (mp) {
bt->stat.hits++;
/* Update LRU queue. Move page to the end. */
TAILQ_REMOVE(bt->lru_queue, mp, lru_next)do { if (((mp)->lru_next.tqe_next) != ((void*)0)) (mp)->
lru_next.tqe_next->lru_next.tqe_prev = (mp)->lru_next.tqe_prev
; else (bt->lru_queue)->tqh_last = (mp)->lru_next.tqe_prev
; *(mp)->lru_next.tqe_prev = (mp)->lru_next.tqe_next; ;
 ; } while (0);
TAILQ_INSERT_TAIL(bt->lru_queue, mp, lru_next)do { (mp)->lru_next.tqe_next = ((void*)0); (mp)->lru_next
.tqe_prev = (bt->lru_queue)->tqh_last; *(bt->lru_queue
)->tqh_last = (mp); (bt->lru_queue)->tqh_last = &
(mp)->lru_next.tqe_next; } while (0);
}
return mp;
481}

483static void
484mpage_add(struct btree *bt, struct mpage *mp)
485{
assert(RB_INSERT(page_cache, bt->page_cache, mp) == NULL)((page_cache_RB_INSERT(bt->page_cache, mp) == ((void*)0)) ?
 (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c", 486, __func__
, "RB_INSERT(page_cache, bt->page_cache, mp) == NULL"));
bt->stat.cache_size++;
TAILQ_INSERT_TAIL(bt->lru_queue, mp, lru_next)do { (mp)->lru_next.tqe_next = ((void*)0); (mp)->lru_next
.tqe_prev = (bt->lru_queue)->tqh_last; *(bt->lru_queue
)->tqh_last = (mp); (bt->lru_queue)->tqh_last = &
(mp)->lru_next.tqe_next; } while (0);
489}

491static void
492mpage_free(struct mpage *mp)
493{
if (mp != NULL((void*)0)) {
free(mp->page);
free(mp);
}
498}

500static void
501mpage_del(struct btree *bt, struct mpage *mp)
502{
assert(RB_REMOVE(page_cache, bt->page_cache, mp) == mp)((page_cache_RB_REMOVE(bt->page_cache, mp) == mp) ? (void)
: __assert2("/usr/src/usr.sbin/ldapd/btree.c", 503, __func__
, "RB_REMOVE(page_cache, bt->page_cache, mp) == mp"));
assert(bt->stat.cache_size > 0)((bt->stat.cache_size > 0) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 504, __func__, "bt->stat.cache_size > 0"));
bt->stat.cache_size--;
TAILQ_REMOVE(bt->lru_queue, mp, lru_next)do { if (((mp)->lru_next.tqe_next) != ((void*)0)) (mp)->
lru_next.tqe_next->lru_next.tqe_prev = (mp)->lru_next.tqe_prev
; else (bt->lru_queue)->tqh_last = (mp)->lru_next.tqe_prev
; *(mp)->lru_next.tqe_prev = (mp)->lru_next.tqe_next; ;
 ; } while (0);
507}

509static void
510mpage_flush(struct btree *bt)
511{
struct mpage	*mp;

while ((mp = RB_MIN(page_cache, bt->page_cache)page_cache_RB_MINMAX(bt->page_cache, -1)) != NULL((void*)0)) {
mpage_del(bt, mp);
mpage_free(mp);
}
518}

520static struct mpage *
521mpage_copy(struct btree *bt, struct mpage *mp)
522{
struct mpage	*copy;

if ((copy = calloc(1, sizeof(*copy))) == NULL((void*)0))
return NULL((void*)0);
if ((copy->page = malloc(bt->head.psize)) == NULL((void*)0)) {
free(copy);
return NULL((void*)0);
}
bcopy(mp->page, copy->page, bt->head.psize);
bcopy(&mp->prefix, &copy->prefix, sizeof(mp->prefix));
copy->parent = mp->parent;
copy->parent_index = mp->parent_index;
copy->pgno = mp->pgno;

return copy;
538}

540/* Remove the least recently used memory pages until the cache size is
* within the configured bounds. Pages referenced by cursors or returned
* key/data are not pruned.
*/
544static void
545mpage_prune(struct btree *bt)
546{
struct mpage	*mp, *next;

for (mp = TAILQ_FIRST(bt->lru_queue)((bt->lru_queue)->tqh_first); mp; mp = next) {
if (bt->stat.cache_size <= bt->stat.max_cache)
	break;
next = TAILQ_NEXT(mp, lru_next)((mp)->lru_next.tqe_next);
if (!mp->dirty && mp->ref <= 0) {
	mpage_del(bt, mp);
	mpage_free(mp);
}
}
558}

560/* Mark a page as dirty and push it on the dirty queue.
*/
562static void
563mpage_dirty(struct btree *bt, struct mpage *mp)
564{
assert(bt != NULL)((bt != ((void*)0)) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 565, __func__, "bt != NULL"));
assert(bt->txn != NULL)((bt->txn != ((void*)0)) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 566, __func__, "bt->txn != NULL"));

if (!mp->dirty) {
mp->dirty = 1;
SIMPLEQ_INSERT_TAIL(bt->txn->dirty_queue, mp, next)do { (mp)->next.sqe_next = ((void*)0); *(bt->txn->dirty_queue
)->sqh_last = (mp); (bt->txn->dirty_queue)->sqh_last
 = &(mp)->next.sqe_next; } while (0);
}
572}

574/* Touch a page: make it dirty and re-insert into tree with updated pgno.
*/
576static struct mpage *
577mpage_touch(struct btree *bt, struct mpage *mp)
578{
assert(bt != NULL)((bt != ((void*)0)) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 579, __func__, "bt != NULL"));
assert(bt->txn != NULL)((bt->txn != ((void*)0)) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 580, __func__, "bt->txn != NULL"));
assert(mp != NULL)((mp != ((void*)0)) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 581, __func__, "mp != NULL"));

if (!mp->dirty) {
DPRINTF("touching page %u -> %u", mp->pgno, bt->txn->next_pgno);
if (mp->ref == 0)
	mpage_del(bt, mp);
else {
	if ((mp = mpage_copy(bt, mp)) == NULL((void*)0))
		return NULL((void*)0);
}
mp->pgno = mp->page->pgno = bt->txn->next_pgno++;
mpage_dirty(bt, mp);
mpage_add(bt, mp);

/* Update the page number to new touched page. */
if (mp->parent != NULL((void*)0))
	NODEPGNO(NODEPTR(mp->parent,((((struct node *)((char *)((mp->parent)->page) + ((mp->
parent)->page)->ptrs[mp->parent_index])))->p.np_pgno
)
	    mp->parent_index))((((struct node *)((char *)((mp->parent)->page) + ((mp->
parent)->page)->ptrs[mp->parent_index])))->p.np_pgno
) = mp->pgno;
}

return mp;
602}

604static int
605btree_read_page(struct btree *bt, pgno_t pgno, struct page *page)
606{
ssize_t		 rc;

DPRINTF("reading page %u", pgno);
bt->stat.reads++;
if ((rc = pread(bt->fd, page, bt->head.psize, (off_t)pgno*bt->head.psize)) == 0) {
DPRINTF("page %u doesn't exist", pgno);
errno(*__errno()) = ENOENT2;
return BT_FAIL-1;
} else if (rc != (ssize_t)bt->head.psize) {
if (rc > 0)
	errno(*__errno()) = EINVAL22;
DPRINTF("read: %s", strerror(errno));
return BT_FAIL-1;
}

if (page->pgno != pgno) {
DPRINTF("page numbers don't match: %u != %u", pgno, page->pgno);
errno(*__errno()) = EINVAL22;
return BT_FAIL-1;
}

DPRINTF("page %u has flags 0x%X", pgno, page->flags);

return BT_SUCCESS0;
631}

633int
634btree_sync(struct btree *bt)
635{
if (!F_ISSET(bt->flags, BT_NOSYNC)(((bt->flags) & (0x02)) == (0x02)))
return fsync(bt->fd);
return 0;
639}

641struct btree_txn *
642btree_txn_begin(struct btree *bt, int rdonly)
643{
struct btree_txn	*txn;

if (!rdonly && bt->txn != NULL((void*)0)) {
DPRINTF("write transaction already begun");
errno(*__errno()) = EBUSY16;
return NULL((void*)0);
}

if ((txn = calloc(1, sizeof(*txn))) == NULL((void*)0)) {
DPRINTF("calloc: %s", strerror(errno));
return NULL((void*)0);
}

if (rdonly) {
txn->flags |= BT_TXN_RDONLY0x01;
} else {
txn->dirty_queue = calloc(1, sizeof(*txn->dirty_queue));
if (txn->dirty_queue == NULL((void*)0)) {
	free(txn);
	return NULL((void*)0);
}
SIMPLEQ_INIT(txn->dirty_queue)do { (txn->dirty_queue)->sqh_first = ((void*)0); (txn->
dirty_queue)->sqh_last = &(txn->dirty_queue)->sqh_first
; } while (0);

DPRINTF("taking write lock on txn %p", txn);
if (flock(bt->fd, LOCK_EX0x02 | LOCK_NB0x04) != 0) {
	DPRINTF("flock: %s", strerror(errno));
	errno(*__errno()) = EBUSY16;
	free(txn->dirty_queue);
	free(txn);
	return NULL((void*)0);
}
bt->txn = txn;
}

txn->bt = bt;
btree_ref(bt);

if (btree_read_meta(bt, &txn->next_pgno) != BT_SUCCESS0) {
btree_txn_abort(txn);
return NULL((void*)0);
}

txn->root = bt->meta.root;
DPRINTF("begin transaction on btree %p, root page %u", bt, txn->root);

return txn;
690}

692void
693btree_txn_abort(struct btree_txn *txn)
694{
struct mpage	*mp;
struct btree	*bt;

if (txn == NULL((void*)0))
return;

bt = txn->bt;
DPRINTF("abort transaction on btree %p, root page %u", bt, txn->root);

if (!F_ISSET(txn->flags, BT_TXN_RDONLY)(((txn->flags) & (0x01)) == (0x01))) {
/* Discard all dirty pages.
 */
while (!SIMPLEQ_EMPTY(txn->dirty_queue)(((txn->dirty_queue)->sqh_first) == ((void*)0))) {
	mp = SIMPLEQ_FIRST(txn->dirty_queue)((txn->dirty_queue)->sqh_first);
	assert(mp->ref == 0)((mp->ref == 0) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 709, __func__, "mp->ref == 0"));	/* cursors should be closed */
	mpage_del(bt, mp);
	SIMPLEQ_REMOVE_HEAD(txn->dirty_queue, next)do { if (((txn->dirty_queue)->sqh_first = (txn->dirty_queue
)->sqh_first->next.sqe_next) == ((void*)0)) (txn->dirty_queue
)->sqh_last = &(txn->dirty_queue)->sqh_first; } while
 (0);
	mpage_free(mp);
}

DPRINTF("releasing write lock on txn %p", txn);
txn->bt->txn = NULL((void*)0);
if (flock(txn->bt->fd, LOCK_UN0x08) != 0) {
	DPRINTF("failed to unlock fd %d: %s",
	    txn->bt->fd, strerror(errno));
}
free(txn->dirty_queue);
}

btree_close(txn->bt);
free(txn);
726}

728int
729btree_txn_commit(struct btree_txn *txn)
730{
int		 n, done;
ssize_t		 rc;
off_t		 size;
struct mpage	*mp;
struct btree	*bt;
struct iovec	 iov[BT_COMMIT_PAGES64];

assert(txn != NULL)((txn != ((void*)0)) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 738, __func__, "txn != NULL"));
assert(txn->bt != NULL)((txn->bt != ((void*)0)) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 739, __func__, "txn->bt != NULL"));

bt = txn->bt;

if (F_ISSET(txn->flags, BT_TXN_RDONLY)(((txn->flags) & (0x01)) == (0x01))) {
DPRINTF("attempt to commit read-only transaction");
btree_txn_abort(txn);
errno(*__errno()) = EPERM1;
return BT_FAIL-1;
}

if (txn != bt->txn) {
DPRINTF("attempt to commit unknown transaction");
btree_txn_abort(txn);
errno(*__errno()) = EINVAL22;
return BT_FAIL-1;
}

if (F_ISSET(txn->flags, BT_TXN_ERROR)(((txn->flags) & (0x02)) == (0x02))) {
DPRINTF("error flag is set, can't commit");
btree_txn_abort(txn);
errno(*__errno()) = EINVAL22;
return BT_FAIL-1;
}

if (SIMPLEQ_EMPTY(txn->dirty_queue)(((txn->dirty_queue)->sqh_first) == ((void*)0)))
goto done;

if (F_ISSET(bt->flags, BT_FIXPADDING)(((bt->flags) & (0x01)) == (0x01))) {
size = lseek(bt->fd, 0, SEEK_END2);
size += bt->head.psize - (size % bt->head.psize);
DPRINTF("extending to multiple of page size: %llu", size);
if (ftruncate(bt->fd, size) != 0) {
	DPRINTF("ftruncate: %s", strerror(errno));
	btree_txn_abort(txn);
	return BT_FAIL-1;
}
bt->flags &= ~BT_FIXPADDING0x01;
}

DPRINTF("committing transaction on btree %p, root page %u",
   bt, txn->root);

/* Commit up to BT_COMMIT_PAGES dirty pages to disk until done.
*/
do {
n = 0;
done = 1;
SIMPLEQ_FOREACH(mp, txn->dirty_queue, next)for((mp) = ((txn->dirty_queue)->sqh_first); (mp) != ((void
*)0); (mp) = ((mp)->next.sqe_next)) {
	DPRINTF("committing page %u", mp->pgno);
	iov[n].iov_len = bt->head.psize;
	iov[n].iov_base = mp->page;
	if (++n >= BT_COMMIT_PAGES64) {
		done = 0;
		break;
	}
}

if (n == 0)
	break;

DPRINTF("committing %u dirty pages", n);
rc = writev(bt->fd, iov, n);
if (rc != (ssize_t)bt->head.psize*n) {
	if (rc > 0)
		DPRINTF("short write, filesystem full?");
	else
		DPRINTF("writev: %s", strerror(errno));
	btree_txn_abort(txn);
	return BT_FAIL-1;
}

/* Remove the dirty flag from the written pages.
 */
while (!SIMPLEQ_EMPTY(txn->dirty_queue)(((txn->dirty_queue)->sqh_first) == ((void*)0))) {
	mp = SIMPLEQ_FIRST(txn->dirty_queue)((txn->dirty_queue)->sqh_first);
	mp->dirty = 0;
	SIMPLEQ_REMOVE_HEAD(txn->dirty_queue, next)do { if (((txn->dirty_queue)->sqh_first = (txn->dirty_queue
)->sqh_first->next.sqe_next) == ((void*)0)) (txn->dirty_queue
)->sqh_last = &(txn->dirty_queue)->sqh_first; } while
 (0);
	if (--n == 0)
		break;
}
} while (!done);

if (btree_sync(bt) != 0 ||
   btree_write_meta(bt, txn->root, 0) != BT_SUCCESS0 ||
   btree_sync(bt) != 0) {
btree_txn_abort(txn);
return BT_FAIL-1;
}

829done:
mpage_prune(bt);
btree_txn_abort(txn);

return BT_SUCCESS0;
834}

836static int
837btree_write_header(struct btree *bt, int fd)
838{
struct stat	 sb;
struct bt_head	*h;
struct page	*p;
ssize_t		 rc;
unsigned int	 psize;

DPRINTF("writing header page");
assert(bt != NULL)((bt != ((void*)0)) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 846, __func__, "bt != NULL"));
27
←
'?' condition is true→

/*
* Ask stat for 'optimal blocksize for I/O', but cap to fit in indx_t.
*/
if (fstat(fd, &sb) == 0)
28
←
Assuming the condition is false→
29
←
Taking false branch→
psize = MINIMUM(32*1024, sb.st_blksize)((32*1024) < (sb.st_blksize) ? (32*1024) : (sb.st_blksize)
);
else
psize = PAGESIZE4096;

if ((p = calloc(1, psize)) == NULL((void*)0))
30
←
Assuming the condition is true→
31
←
Taking true branch→
return -1;
32
←
Returning without writing to 'bt->head.psize'→
p->flags = P_HEAD0x10;

h = METADATA(p)((void *)((char *)p + __builtin_offsetof(struct page, ptrs)));
h->magic = BT_MAGIC0xB3DBB3DB;
h->version = BT_VERSION4;
h->psize = psize;
bcopy(h, &bt->head, sizeof(*h));

rc = write(fd, p, bt->head.psize);
free(p);
if (rc != (ssize_t)bt->head.psize) {
if (rc > 0)
	DPRINTF("short write, filesystem full?");
return BT_FAIL-1;
}

return BT_SUCCESS0;
875}

877static int
878btree_read_header(struct btree *bt)
879{
char		 page[PAGESIZE4096];
struct page	*p;
struct bt_head	*h;
int		 rc;

assert(bt != NULL)((bt != ((void*)0)) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 885, __func__, "bt != NULL"));
18
←
'?' condition is true→

/* We don't know the page size yet, so use a minimum value.
*/

if ((rc = pread(bt->fd, page, PAGESIZE4096, 0)) == 0) {
19
←
Assuming the condition is true→
20
←
Taking true branch→
errno(*__errno()) = ENOENT2;
return -1;
21
←
Returning without writing to 'bt->head.psize'→
} else if (rc != PAGESIZE4096) {
if (rc > 0)
	errno(*__errno()) = EINVAL22;
DPRINTF("read: %s", strerror(errno));
return -1;
}

p = (struct page *)page;

if (!F_ISSET(p->flags, P_HEAD)(((p->flags) & (0x10)) == (0x10))) {
DPRINTF("page %d not a header page", p->pgno);
errno(*__errno()) = EINVAL22;
return -1;
}

h = METADATA(p)((void *)((char *)p + __builtin_offsetof(struct page, ptrs)));
if (h->magic != BT_MAGIC0xB3DBB3DB) {
DPRINTF("header has invalid magic");
errno(*__errno()) = EINVAL22;
return -1;
}

if (h->version != BT_VERSION4) {
DPRINTF("database is version %u, expected version %u",
    bt->head.version, BT_VERSION);
errno(*__errno()) = EINVAL22;
return -1;
}

bcopy(h, &bt->head, sizeof(*h));
return 0;
924}

926static int
927btree_write_meta(struct btree *bt, pgno_t root, unsigned int flags)
928{
struct mpage	*mp;
struct bt_meta	*meta;
ssize_t		 rc;

DPRINTF("writing meta page for root page %u", root);

assert(bt != NULL)((bt != ((void*)0)) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 935, __func__, "bt != NULL"));
assert(bt->txn != NULL)((bt->txn != ((void*)0)) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 936, __func__, "bt->txn != NULL"));

if ((mp = btree_new_page(bt, P_META0x08)) == NULL((void*)0))
return -1;

bt->meta.prev_meta = bt->meta.root;
bt->meta.root = root;
bt->meta.flags = flags;
bt->meta.created_at = time(0);
bt->meta.revisions++;
SHA1((unsigned char *)&bt->meta, METAHASHLEN__builtin_offsetof(struct bt_meta, hash), bt->meta.hash);

/* Copy the meta data changes to the new meta page. */
meta = METADATA(mp->page)((void *)((char *)mp->page + __builtin_offsetof(struct page
, ptrs)));
bcopy(&bt->meta, meta, sizeof(*meta));

rc = write(bt->fd, mp->page, bt->head.psize);
mp->dirty = 0;
SIMPLEQ_REMOVE_HEAD(bt->txn->dirty_queue, next)do { if (((bt->txn->dirty_queue)->sqh_first = (bt->
txn->dirty_queue)->sqh_first->next.sqe_next) == ((void
*)0)) (bt->txn->dirty_queue)->sqh_last = &(bt->
txn->dirty_queue)->sqh_first; } while (0);
if (rc != (ssize_t)bt->head.psize) {
if (rc > 0)
	DPRINTF("short write, filesystem full?");
return BT_FAIL-1;
}

if ((bt->size = lseek(bt->fd, 0, SEEK_END2)) == -1) {
DPRINTF("failed to update file size: %s", strerror(errno));
bt->size = 0;
}

return BT_SUCCESS0;
967}

969/* Returns true if page p is a valid meta page, false otherwise.
*/
971static int
972btree_is_meta_page(struct page *p)
973{
struct bt_meta	*m;
unsigned char	 hash[SHA_DIGEST_LENGTH20];

m = METADATA(p)((void *)((char *)p + __builtin_offsetof(struct page, ptrs)));
if (!F_ISSET(p->flags, P_META)(((p->flags) & (0x08)) == (0x08))) {
DPRINTF("page %d not a meta page", p->pgno);
errno(*__errno()) = EINVAL22;
return 0;
}

if (m->root >= p->pgno && m->root != P_INVALID0xFFFFFFFF) {
DPRINTF("page %d points to an invalid root page", p->pgno);
errno(*__errno()) = EINVAL22;
return 0;
}

SHA1((unsigned char *)m, METAHASHLEN__builtin_offsetof(struct bt_meta, hash), hash);
if (bcmp(hash, m->hash, SHA_DIGEST_LENGTH20) != 0) {
DPRINTF("page %d has an invalid digest", p->pgno);
errno(*__errno()) = EINVAL22;
return 0;
}

return 1;
998}

1000static int
1001btree_read_meta(struct btree *bt, pgno_t *p_next)
1002{
struct mpage	*mp;
struct bt_meta	*meta;
pgno_t		 meta_pgno, next_pgno;
off_t		 size;

assert(bt != NULL)((bt != ((void*)0)) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 1008, __func__, "bt != NULL"));
35
←
'?' condition is true→

if ((size = lseek(bt->fd, 0, SEEK_END2)) == -1)
36
←
Assuming the condition is false→
37
←
Taking false branch→
goto fail;

DPRINTF("btree_read_meta: size = %llu", size);

if (size < bt->size) {
38
←
Assuming 'size' is >= field 'size'→
39
←
Taking false branch→
DPRINTF("file has shrunk!");
errno(*__errno()) = EIO5;
goto fail;
}

if (size == bt->head.psize) {		/* there is only the header */
40
←
Assuming 'size' is not equal to field 'psize'→
41
←
Taking false branch→
if (p_next != NULL((void*)0))
	*p_next = 1;
return BT_SUCCESS0;		/* new file */
}

next_pgno = size / bt->head.psize;
42
←
Division by zero
if (next_pgno == 0) {
DPRINTF("corrupt file");
errno(*__errno()) = EIO5;
goto fail;
}

meta_pgno = next_pgno - 1;

if (size % bt->head.psize != 0) {
DPRINTF("filesize not a multiple of the page size!");
bt->flags |= BT_FIXPADDING0x01;
next_pgno++;
}

if (p_next != NULL((void*)0))
*p_next = next_pgno;

if (size == bt->size) {
DPRINTF("size unchanged, keeping current meta page");
if (F_ISSET(bt->meta.flags, BT_TOMBSTONE)(((bt->meta.flags) & (0x01)) == (0x01))) {
	DPRINTF("file is dead");
	errno(*__errno()) = ESTALE70;
	return BT_FAIL-1;
} else
	return BT_SUCCESS0;
}
bt->size = size;

while (meta_pgno > 0) {
if ((mp = btree_get_mpage(bt, meta_pgno)) == NULL((void*)0))
	break;
if (btree_is_meta_page(mp->page)) {
	meta = METADATA(mp->page)((void *)((char *)mp->page + __builtin_offsetof(struct page
, ptrs)));
	DPRINTF("flags = 0x%x", meta->flags);
	if (F_ISSET(meta->flags, BT_TOMBSTONE)(((meta->flags) & (0x01)) == (0x01))) {
		DPRINTF("file is dead");
		errno(*__errno()) = ESTALE70;
		return BT_FAIL-1;
	} else {
		/* Make copy of last meta page. */
		bcopy(meta, &bt->meta, sizeof(bt->meta));
		return BT_SUCCESS0;
	}
}
--meta_pgno;	/* scan backwards to first valid meta page */
}

errno(*__errno()) = EIO5;
1076fail:
if (p_next != NULL((void*)0))
*p_next = P_INVALID0xFFFFFFFF;
return BT_FAIL-1;
1080}

1082struct btree *
1083btree_open_fd(int fd, unsigned int flags)
1084{
struct btree	*bt;
int		 fl;

fl = fcntl(fd, F_GETFL3);
if (fcntl(fd, F_SETFL4, fl | O_APPEND0x0008) == -1)
6
←
Assuming the condition is false→
7
←
Taking false branch→
return NULL((void*)0);

if ((bt = calloc(1, sizeof(*bt))) == NULL((void*)0))
8
←
The value 0 is assigned to field 'psize'→
9
←
Assuming the condition is false→
10
←
Taking false branch→
return NULL((void*)0);
bt->fd = fd;
bt->flags = flags;
bt->flags &= ~BT_FIXPADDING0x01;
bt->ref = 1;
bt->meta.root = P_INVALID0xFFFFFFFF;

if ((bt->page_cache = calloc(1, sizeof(*bt->page_cache))) == NULL((void*)0))
11
←
Assuming the condition is false→
12
←
Taking false branch→
goto fail;
bt->stat.max_cache = BT_MAXCACHE_DEF1024;
RB_INIT(bt->page_cache)do { (bt->page_cache)->rbh_root = ((void*)0); } while (
0);
13
←
Loop condition is false.  Exiting loop→

if ((bt->lru_queue = calloc(1, sizeof(*bt->lru_queue))) == NULL((void*)0))
14
←
Assuming the condition is false→
15
←
Taking false branch→
goto fail;
TAILQ_INIT(bt->lru_queue)do { (bt->lru_queue)->tqh_first = ((void*)0); (bt->lru_queue
)->tqh_last = &(bt->lru_queue)->tqh_first; } while
 (0);
16
←
Loop condition is false.  Exiting loop→

if (btree_read_header(bt) != 0) {
17
←
Calling 'btree_read_header'→
22
←
Returning from 'btree_read_header'→
23
←
Taking true branch→
if (errno(*__errno()) != ENOENT2)
24
←
Assuming the condition is false→
25
←
Taking false branch→
	goto fail;
DPRINTF("new database");
btree_write_header(bt, bt->fd);
26
←
Calling 'btree_write_header'→
33
←
Returning from 'btree_write_header'→
}

if (btree_read_meta(bt, NULL((void*)0)) != 0)
34
←
Calling 'btree_read_meta'→
goto fail;

DPRINTF("opened database version %u, pagesize %u",
   bt->head.version, bt->head.psize);
DPRINTF("timestamp: %s", ctime(&bt->meta.created_at));
DPRINTF("depth: %u", bt->meta.depth);
DPRINTF("entries: %llu", bt->meta.entries);
DPRINTF("revisions: %u", bt->meta.revisions);
DPRINTF("branch pages: %u", bt->meta.branch_pages);
DPRINTF("leaf pages: %u", bt->meta.leaf_pages);
DPRINTF("overflow pages: %u", bt->meta.overflow_pages);
DPRINTF("root: %u", bt->meta.root);
DPRINTF("previous meta page: %u", bt->meta.prev_meta);

return bt;

1133fail:
free(bt->lru_queue);
free(bt->page_cache);
free(bt);
return NULL((void*)0);
1138}

1140struct btree *
1141btree_open(const char *path, unsigned int flags, mode_t mode)
1142{
int		 fd, oflags;
struct btree	*bt;

if (F_ISSET(flags, BT_RDONLY)(((flags) & (0x04)) == (0x04)))
1
Assuming the condition is false→
2
←
Taking false branch→
oflags = O_RDONLY0x0000;
else
oflags = O_RDWR0x0002 | O_CREAT0x0200 | O_APPEND0x0008;

if ((fd = open(path, oflags, mode)) == -1)
3
←
Assuming the condition is false→
4
←
Taking false branch→
return NULL((void*)0);

if ((bt = btree_open_fd(fd, flags)) == NULL((void*)0))
5
←
Calling 'btree_open_fd'→
close(fd);
else {
bt->path = strdup(path);
DPRINTF("opened btree %p", bt);
}

return bt;
1162}

1164static void
1165btree_ref(struct btree *bt)
1166{
bt->ref++;
DPRINTF("ref is now %d on btree %p", bt->ref, bt);
1169}

1171void
1172btree_close(struct btree *bt)
1173{
if (bt == NULL((void*)0))
return;

if (--bt->ref == 0) {
DPRINTF("ref is zero, closing btree %p", bt);
close(bt->fd);
mpage_flush(bt);
free(bt->lru_queue);
free(bt->path);
free(bt->page_cache);
free(bt);
} else
DPRINTF("ref is now %d on btree %p", bt->ref, bt);
1187}

1189/* Search for key within a leaf page, using binary search.
* Returns the smallest entry larger or equal to the key.
* If exactp is non-null, stores whether the found entry was an exact match
* in *exactp (1 or 0).
* If kip is non-null, stores the index of the found entry in *kip.
* If no entry larger of equal to the key is found, returns NULL.
*/
1196static struct node *
1197btree_search_node(struct btree *bt, struct mpage *mp, struct btval *key,
  int *exactp, unsigned int *kip)
1199{
unsigned int	 i = 0;
int		 low, high;
int		 rc = 0;
struct node	*node;
struct btval	 nodekey;

DPRINTF("searching %lu keys in %s page %u with prefix [%.*s]",
   NUMKEYS(mp),
   IS_LEAF(mp) ? "leaf" : "branch",
   mp->pgno, (int)mp->prefix.len, (char *)mp->prefix.str);

assert(NUMKEYS(mp) > 0)(((((mp)->page->b.fb.fb_lower - __builtin_offsetof(struct
 page, ptrs)) >> 1) > 0) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 1211, __func__, "NUMKEYS(mp) > 0"));

memset(&nodekey, 0, sizeof(nodekey));

low = IS_LEAF(mp)((((mp)->page->flags) & (0x02)) == (0x02)) ? 0 : 1;
high = NUMKEYS(mp)(((mp)->page->b.fb.fb_lower - __builtin_offsetof(struct
 page, ptrs)) >> 1) - 1;
while (low <= high) {
i = (low + high) >> 1;
node = NODEPTR(mp, i)((struct node *)((char *)((mp)->page) + ((mp)->page)->
ptrs[i]));

nodekey.size = node->ksize;
nodekey.data = NODEKEY(node)(void *)((node)->data);

if (bt->cmp)
	rc = bt->cmp(key, &nodekey);
else
	rc = bt_cmp(bt, key, &nodekey, &mp->prefix);

if (IS_LEAF(mp)((((mp)->page->flags) & (0x02)) == (0x02)))
	DPRINTF("found leaf index %u [%.*s], rc = %d",
	    i, (int)nodekey.size, (char *)nodekey.data, rc);
else
	DPRINTF("found branch index %u [%.*s -> %u], rc = %d",
	    i, (int)node->ksize, (char *)NODEKEY(node),
	    node->n_pgno, rc);

if (rc == 0)
	break;
if (rc > 0)
	low = i + 1;
else
	high = i - 1;
}

if (rc > 0) {	/* Found entry is less than the key. */
i++;	/* Skip to get the smallest entry larger than key. */
if (i >= NUMKEYS(mp)(((mp)->page->b.fb.fb_lower - __builtin_offsetof(struct
 page, ptrs)) >> 1))
	/* There is no entry larger or equal to the key. */
	return NULL((void*)0);
}
if (exactp)
*exactp = (rc == 0);
if (kip)	/* Store the key index if requested. */
*kip = i;

return NODEPTR(mp, i)((struct node *)((char *)((mp)->page) + ((mp)->page)->
ptrs[i]));
1257}

1259static void
1260cursor_pop_page(struct cursor *cursor)
1261{
struct ppage	*top;

top = CURSOR_TOP(cursor)((&(cursor)->stack)->slh_first);
CURSOR_POP(cursor)do { (&(cursor)->stack)->slh_first = (&(cursor)
->stack)->slh_first->entry.sle_next; } while (0);
top->mpage->ref--;

DPRINTF("popped page %u off cursor %p", top->mpage->pgno, cursor);

free(top);
1271}

1273static struct ppage *
1274cursor_push_page(struct cursor *cursor, struct mpage *mp)
1275{
struct ppage	*ppage;

DPRINTF("pushing page %u on cursor %p", mp->pgno, cursor);

if ((ppage = calloc(1, sizeof(*ppage))) == NULL((void*)0))
return NULL((void*)0);
ppage->mpage = mp;
mp->ref++;
CURSOR_PUSH(cursor, ppage)do { (ppage)->entry.sle_next = (&(cursor)->stack)->
slh_first; (&(cursor)->stack)->slh_first = (ppage);
 } while (0);
return ppage;
1286}

1288static struct mpage *
1289btree_get_mpage(struct btree *bt, pgno_t pgno)
1290{
struct mpage	*mp;

mp = mpage_lookup(bt, pgno);
if (mp == NULL((void*)0)) {
if ((mp = calloc(1, sizeof(*mp))) == NULL((void*)0))
	return NULL((void*)0);
if ((mp->page = malloc(bt->head.psize)) == NULL((void*)0)) {
	free(mp);
	return NULL((void*)0);
}
if (btree_read_page(bt, pgno, mp->page) != BT_SUCCESS0) {
	mpage_free(mp);
	return NULL((void*)0);
}
mp->pgno = pgno;
mpage_add(bt, mp);
} else
DPRINTF("returning page %u from cache", pgno);

return mp;
1311}

1313static void
1314concat_prefix(struct btree *bt, char *s1, size_t n1, char *s2, size_t n2,
  char *cs, size_t *cn)
1316{
assert(*cn >= n1 + n2)((*cn >= n1 + n2) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 1317, __func__, "*cn >= n1 + n2"));
if (F_ISSET(bt->flags, BT_REVERSEKEY)(((bt->flags) & (0x08)) == (0x08))) {
bcopy(s2, cs, n2);
bcopy(s1, cs + n2, n1);
} else {
bcopy(s1, cs, n1);
bcopy(s2, cs + n1, n2);
}
*cn = n1 + n2;
1326}

1328static void
1329find_common_prefix(struct btree *bt, struct mpage *mp)
1330{
indx_t			 lbound = 0, ubound = 0;
struct mpage		*lp, *up;
struct btkey		 lprefix, uprefix;

mp->prefix.len = 0;
if (bt->cmp != NULL((void*)0))
return;

lp = mp;
while (lp->parent != NULL((void*)0)) {
if (lp->parent_index > 0) {
	lbound = lp->parent_index;
	break;
}
lp = lp->parent;
}

up = mp;
while (up->parent != NULL((void*)0)) {
if (up->parent_index + 1 < (indx_t)NUMKEYS(up->parent)(((up->parent)->page->b.fb.fb_lower - __builtin_offsetof
(struct page, ptrs)) >> 1)) {
	ubound = up->parent_index + 1;
	break;
}
up = up->parent;
}

if (lp->parent != NULL((void*)0) && up->parent != NULL((void*)0)) {
expand_prefix(bt, lp->parent, lbound, &lprefix);
expand_prefix(bt, up->parent, ubound, &uprefix);
common_prefix(bt, &lprefix, &uprefix, &mp->prefix);
}
else if (mp->parent)
bcopy(&mp->parent->prefix, &mp->prefix, sizeof(mp->prefix));

DPRINTF("found common prefix [%.*s] (len %zu) for page %u",
   (int)mp->prefix.len, mp->prefix.str, mp->prefix.len, mp->pgno);
1367}

1369static int
1370btree_search_page_root(struct btree *bt, struct mpage *root, struct btval *key,
  struct cursor *cursor, int modify, struct mpage **mpp)
1372{
struct mpage	*mp, *parent;

if (cursor && cursor_push_page(cursor, root) == NULL((void*)0))
return BT_FAIL-1;

mp = root;
while (IS_BRANCH(mp)((((mp)->page->flags) & (0x01)) == (0x01))) {
unsigned int	 i = 0;
struct node	*node;

DPRINTF("branch page %u has %lu keys", mp->pgno, NUMKEYS(mp));
assert(NUMKEYS(mp) > 1)(((((mp)->page->b.fb.fb_lower - __builtin_offsetof(struct
 page, ptrs)) >> 1) > 1) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 1384, __func__, "NUMKEYS(mp) > 1"));
DPRINTF("found index 0 to page %u", NODEPGNO(NODEPTR(mp, 0)));

if (key == NULL((void*)0))	/* Initialize cursor to first page. */
	i = 0;
else {
	int	 exact;
	node = btree_search_node(bt, mp, key, &exact, &i);
	if (node == NULL((void*)0))
		i = NUMKEYS(mp)(((mp)->page->b.fb.fb_lower - __builtin_offsetof(struct
 page, ptrs)) >> 1) - 1;
	else if (!exact) {
		assert(i > 0)((i > 0) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 1395, __func__, "i > 0"));
		i--;
	}
}

if (key)
	DPRINTF("following index %u for key %.*s",
	    i, (int)key->size, (char *)key->data);
assert(i >= 0 && i < NUMKEYS(mp))((i >= 0 && i < (((mp)->page->b.fb.fb_lower
 - __builtin_offsetof(struct page, ptrs)) >> 1)) ? (void
)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c", 1403, __func__
, "i >= 0 && i < NUMKEYS(mp)"));
node = NODEPTR(mp, i)((struct node *)((char *)((mp)->page) + ((mp)->page)->
ptrs[i]));

if (cursor)
	CURSOR_TOP(cursor)((&(cursor)->stack)->slh_first)->ki = i;

parent = mp;
if ((mp = btree_get_mpage(bt, NODEPGNO(node)((node)->p.np_pgno))) == NULL((void*)0))
	return BT_FAIL-1;
mp->parent = parent;
mp->parent_index = i;
find_common_prefix(bt, mp);

if (cursor && cursor_push_page(cursor, mp) == NULL((void*)0))
	return BT_FAIL-1;

if (modify && (mp = mpage_touch(bt, mp)) == NULL((void*)0))
	return BT_FAIL-1;
}

if (!IS_LEAF(mp)((((mp)->page->flags) & (0x02)) == (0x02))) {
DPRINTF("internal error, index points to a %02X page!?",
    mp->page->flags);
return BT_FAIL-1;
}

DPRINTF("found leaf page %u for key %.*s", mp->pgno,
   key ? (int)key->size : 0, key ? (char *)key->data : NULL);

*mpp = mp;
return BT_SUCCESS0;
1434}

1436/* Search for the page a given key should be in.
* Stores a pointer to the found page in *mpp.
* If key is NULL, search for the lowest page (used by btree_cursor_first).
* If cursor is non-null, pushes parent pages on the cursor stack.
* If modify is true, visited pages are updated with new page numbers.
*/
1442static int
1443btree_search_page(struct btree *bt, struct btree_txn *txn, struct btval *key,
  struct cursor *cursor, int modify, struct mpage **mpp)
1445{
int		 rc;
pgno_t		 root;
struct mpage	*mp;

/* Can't modify pages outside a transaction. */
if (txn == NULL((void*)0) && modify) {
errno(*__errno()) = EINVAL22;
return BT_FAIL-1;
}

/* Choose which root page to start with. If a transaction is given
       * use the root page from the transaction, otherwise read the last
       * committed root page.
*/
if (txn == NULL((void*)0)) {
if ((rc = btree_read_meta(bt, NULL((void*)0))) != BT_SUCCESS0)
	return rc;
root = bt->meta.root;
} else if (F_ISSET(txn->flags, BT_TXN_ERROR)(((txn->flags) & (0x02)) == (0x02))) {
DPRINTF("transaction has failed, must abort");
errno(*__errno()) = EINVAL22;
return BT_FAIL-1;
} else
root = txn->root;

if (root == P_INVALID0xFFFFFFFF) {		/* Tree is empty. */
DPRINTF("tree is empty");
errno(*__errno()) = ENOENT2;
return BT_FAIL-1;
}

if ((mp = btree_get_mpage(bt, root)) == NULL((void*)0))
return BT_FAIL-1;

DPRINTF("root page has flags 0x%X", mp->page->flags);

assert(mp->parent == NULL)((mp->parent == ((void*)0)) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 1482, __func__, "mp->parent == NULL"));
assert(mp->prefix.len == 0)((mp->prefix.len == 0) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 1483, __func__, "mp->prefix.len == 0"));

if (modify && !mp->dirty) {
if ((mp = mpage_touch(bt, mp)) == NULL((void*)0))
	return BT_FAIL-1;
txn->root = mp->pgno;
}

return btree_search_page_root(bt, mp, key, cursor, modify, mpp);
1492}

1494static int
1495btree_read_data(struct btree *bt, struct mpage *mp, struct node *leaf,
  struct btval *data)
1497{
struct mpage	*omp;		/* overflow mpage */
size_t		 psz;
size_t		 max;
size_t		 sz = 0;
pgno_t		 pgno;

memset(data, 0, sizeof(*data));
max = bt->head.psize - PAGEHDRSZ__builtin_offsetof(struct page, ptrs);

if (!F_ISSET(leaf->flags, F_BIGDATA)(((leaf->flags) & (0x01)) == (0x01))) {
data->size = leaf->n_dsizep.np_dsize;
if (data->size > 0) {
	if (mp == NULL((void*)0)) {
		if ((data->data = malloc(data->size)) == NULL((void*)0))
			return BT_FAIL-1;
		bcopy(NODEDATA(leaf)(void *)((char *)(leaf)->data + (leaf)->ksize), data->data, data->size);
		data->free_data = 1;
		data->mp = NULL((void*)0);
	} else {
		data->data = NODEDATA(leaf)(void *)((char *)(leaf)->data + (leaf)->ksize);
		data->free_data = 0;
		data->mp = mp;
		mp->ref++;
	}
}
return BT_SUCCESS0;
}

/* Read overflow data.
*/
DPRINTF("allocating %u byte for overflow data", leaf->n_dsize);
if ((data->data = malloc(leaf->n_dsizep.np_dsize)) == NULL((void*)0))
return BT_FAIL-1;
data->size = leaf->n_dsizep.np_dsize;
data->free_data = 1;
data->mp = NULL((void*)0);
bcopy(NODEDATA(leaf)(void *)((char *)(leaf)->data + (leaf)->ksize), &pgno, sizeof(pgno));
for (sz = 0; sz < data->size; ) {
if ((omp = btree_get_mpage(bt, pgno)) == NULL((void*)0) ||
    !F_ISSET(omp->page->flags, P_OVERFLOW)(((omp->page->flags) & (0x04)) == (0x04))) {
	DPRINTF("read overflow page %u failed", pgno);
	free(data->data);
	mpage_free(omp);
	return BT_FAIL-1;
}
psz = data->size - sz;
if (psz > max)
	psz = max;
bcopy(omp->page->ptrs, (char *)data->data + sz, psz);
sz += psz;
pgno = omp->page->p_next_pgnob.pb_next_pgno;
}

return BT_SUCCESS0;
1552}

1554int
1555btree_txn_get(struct btree *bt, struct btree_txn *txn,
  struct btval *key, struct btval *data)
1557{
int		 rc, exact;
struct node	*leaf;
struct mpage	*mp;

assert(key)((key) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 1562, __func__, "key"));
assert(data)((data) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 1563, __func__, "data"));
DPRINTF("===> get key [%.*s]", (int)key->size, (char *)key->data);

if (bt != NULL((void*)0) && txn != NULL((void*)0) && bt != txn->bt) {
errno(*__errno()) = EINVAL22;
return BT_FAIL-1;
}

if (bt == NULL((void*)0)) {
if (txn == NULL((void*)0)) {
	errno(*__errno()) = EINVAL22;
	return BT_FAIL-1;
}
bt = txn->bt;
}

if (key->size == 0 || key->size > MAXKEYSIZE255) {
errno(*__errno()) = EINVAL22;
return BT_FAIL-1;
}

if ((rc = btree_search_page(bt, txn, key, NULL((void*)0), 0, &mp)) != BT_SUCCESS0)
return rc;

leaf = btree_search_node(bt, mp, key, &exact, NULL((void*)0));
if (leaf && exact)
rc = btree_read_data(bt, mp, leaf, data);
else {
errno(*__errno()) = ENOENT2;
rc = BT_FAIL-1;
}

mpage_prune(bt);
return rc;
1597}

1599static int
1600btree_sibling(struct cursor *cursor, int move_right)
1601{
int		 rc;
struct node	*indx;
struct ppage	*parent, *top;
struct mpage	*mp;

top = CURSOR_TOP(cursor)((&(cursor)->stack)->slh_first);
if ((parent = SLIST_NEXT(top, entry)((top)->entry.sle_next)) == NULL((void*)0)) {
errno(*__errno()) = ENOENT2;
return BT_FAIL-1;			/* root has no siblings */
}

DPRINTF("parent page is page %u, index %u",
   parent->mpage->pgno, parent->ki);

cursor_pop_page(cursor);
if (move_right ? (parent->ki + 1 >= NUMKEYS(parent->mpage)(((parent->mpage)->page->b.fb.fb_lower - __builtin_offsetof
(struct page, ptrs)) >> 1))
       : (parent->ki == 0)) {
DPRINTF("no more keys left, moving to %s sibling",
    move_right ? "right" : "left");
if ((rc = btree_sibling(cursor, move_right)) != BT_SUCCESS0)
	return rc;
parent = CURSOR_TOP(cursor)((&(cursor)->stack)->slh_first);
} else {
if (move_right)
	parent->ki++;
else
	parent->ki--;
DPRINTF("just moving to %s index key %u",
    move_right ? "right" : "left", parent->ki);
}
assert(IS_BRANCH(parent->mpage))((((((parent->mpage)->page->flags) & (0x01)) == (
0x01))) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 1632, __func__, "IS_BRANCH(parent->mpage)"));

indx = NODEPTR(parent->mpage, parent->ki)((struct node *)((char *)((parent->mpage)->page) + ((parent
->mpage)->page)->ptrs[parent->ki]));
if ((mp = btree_get_mpage(cursor->bt, indx->n_pgnop.np_pgno)) == NULL((void*)0))
return BT_FAIL-1;
mp->parent = parent->mpage;
mp->parent_index = parent->ki;

cursor_push_page(cursor, mp);
find_common_prefix(cursor->bt, mp);

return BT_SUCCESS0;
1644}

1646static int
1647bt_set_key(struct btree *bt, struct mpage *mp, struct node *node,
  struct btval *key)
1649{
if (key == NULL((void*)0))
return 0;

if (mp->prefix.len > 0) {
key->size = node->ksize + mp->prefix.len;
key->data = malloc(key->size);
if (key->data == NULL((void*)0))
	return -1;
concat_prefix(bt,
    mp->prefix.str, mp->prefix.len,
    NODEKEY(node)(void *)((node)->data), node->ksize,
    key->data, &key->size);
key->free_data = 1;
} else {
key->size = node->ksize;
key->data = NODEKEY(node)(void *)((node)->data);
key->free_data = 0;
key->mp = mp;
mp->ref++;
}

return 0;
1672}

1674static int
1675btree_cursor_next(struct cursor *cursor, struct btval *key, struct btval *data)
1676{
struct ppage	*top;
struct mpage	*mp;
struct node	*leaf;

if (cursor->eof) {
errno(*__errno()) = ENOENT2;
return BT_FAIL-1;
}

assert(cursor->initialized)((cursor->initialized) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 1686, __func__, "cursor->initialized"));

top = CURSOR_TOP(cursor)((&(cursor)->stack)->slh_first);
mp = top->mpage;

DPRINTF("cursor_next: top page is %u in cursor %p", mp->pgno, cursor);

if (top->ki + 1 >= NUMKEYS(mp)(((mp)->page->b.fb.fb_lower - __builtin_offsetof(struct
 page, ptrs)) >> 1)) {
DPRINTF("=====> move to next sibling page");
if (btree_sibling(cursor, 1) != BT_SUCCESS0) {
	cursor->eof = 1;
	return BT_FAIL-1;
}
top = CURSOR_TOP(cursor)((&(cursor)->stack)->slh_first);
mp = top->mpage;
DPRINTF("next page is %u, key index %u", mp->pgno, top->ki);
} else
top->ki++;

DPRINTF("==> cursor points to page %u with %lu keys, key index %u",
   mp->pgno, NUMKEYS(mp), top->ki);

assert(IS_LEAF(mp))((((((mp)->page->flags) & (0x02)) == (0x02))) ? (void
)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c", 1708, __func__
, "IS_LEAF(mp)"));
leaf = NODEPTR(mp, top->ki)((struct node *)((char *)((mp)->page) + ((mp)->page)->
ptrs[top->ki]));

if (data && btree_read_data(cursor->bt, mp, leaf, data) != BT_SUCCESS0)
return BT_FAIL-1;

if (bt_set_key(cursor->bt, mp, leaf, key) != 0)
return BT_FAIL-1;

return BT_SUCCESS0;
1718}

1720static int
1721btree_cursor_set(struct cursor *cursor, struct btval *key, struct btval *data,
  int *exactp)
1723{
int		 rc;
struct node	*leaf;
struct mpage	*mp;
struct ppage	*top;

assert(cursor)((cursor) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 1729, __func__, "cursor"));
assert(key)((key) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 1730, __func__, "key"));
assert(key->size > 0)((key->size > 0) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 1731, __func__, "key->size > 0"));

rc = btree_search_page(cursor->bt, cursor->txn, key, cursor, 0, &mp);
if (rc != BT_SUCCESS0)
return rc;
assert(IS_LEAF(mp))((((((mp)->page->flags) & (0x02)) == (0x02))) ? (void
)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c", 1736, __func__
, "IS_LEAF(mp)"));

top = CURSOR_TOP(cursor)((&(cursor)->stack)->slh_first);
leaf = btree_search_node(cursor->bt, mp, key, exactp, &top->ki);
if (exactp != NULL((void*)0) && !*exactp) {
/* BT_CURSOR_EXACT specified and not an exact match. */
errno(*__errno()) = ENOENT2;
return BT_FAIL-1;
}

if (leaf == NULL((void*)0)) {
DPRINTF("===> inexact leaf not found, goto sibling");
if (btree_sibling(cursor, 1) != BT_SUCCESS0)
	return BT_FAIL-1;		/* no entries matched */
top = CURSOR_TOP(cursor)((&(cursor)->stack)->slh_first);
top->ki = 0;
mp = top->mpage;
assert(IS_LEAF(mp))((((((mp)->page->flags) & (0x02)) == (0x02))) ? (void
)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c", 1753, __func__
, "IS_LEAF(mp)"));
leaf = NODEPTR(mp, 0)((struct node *)((char *)((mp)->page) + ((mp)->page)->
ptrs[0]));
}

cursor->initialized = 1;
cursor->eof = 0;

if (data && btree_read_data(cursor->bt, mp, leaf, data) != BT_SUCCESS0)
return BT_FAIL-1;

if (bt_set_key(cursor->bt, mp, leaf, key) != 0)
return BT_FAIL-1;
DPRINTF("==> cursor placed on key %.*s",
   (int)key->size, (char *)key->data);

return BT_SUCCESS0;
1769}

1771static int
1772btree_cursor_first(struct cursor *cursor, struct btval *key, struct btval *data)
1773{
int		 rc;
struct mpage	*mp;
struct node	*leaf;

rc = btree_search_page(cursor->bt, cursor->txn, NULL((void*)0), cursor, 0, &mp);
if (rc != BT_SUCCESS0)
return rc;
assert(IS_LEAF(mp))((((((mp)->page->flags) & (0x02)) == (0x02))) ? (void
)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c", 1781, __func__
, "IS_LEAF(mp)"));

leaf = NODEPTR(mp, 0)((struct node *)((char *)((mp)->page) + ((mp)->page)->
ptrs[0]));
cursor->initialized = 1;
cursor->eof = 0;

if (data && btree_read_data(cursor->bt, mp, leaf, data) != BT_SUCCESS0)
return BT_FAIL-1;

if (bt_set_key(cursor->bt, mp, leaf, key) != 0)
return BT_FAIL-1;

return BT_SUCCESS0;
1794}

1796int
1797btree_cursor_get(struct cursor *cursor, struct btval *key, struct btval *data,
  enum cursor_op op)
1799{
int		 rc;
int		 exact = 0;

assert(cursor)((cursor) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 1803, __func__, "cursor"));

switch (op) {
case BT_CURSOR:
case BT_CURSOR_EXACT:
while (CURSOR_TOP(cursor)((&(cursor)->stack)->slh_first) != NULL((void*)0))
	cursor_pop_page(cursor);
if (key == NULL((void*)0) || key->size == 0 || key->size > MAXKEYSIZE255) {
	errno(*__errno()) = EINVAL22;
	rc = BT_FAIL-1;
} else if (op == BT_CURSOR_EXACT)
	rc = btree_cursor_set(cursor, key, data, &exact);
else
	rc = btree_cursor_set(cursor, key, data, NULL((void*)0));
break;
case BT_NEXT:
if (!cursor->initialized)
	rc = btree_cursor_first(cursor, key, data);
else
	rc = btree_cursor_next(cursor, key, data);
break;
case BT_FIRST:
while (CURSOR_TOP(cursor)((&(cursor)->stack)->slh_first) != NULL((void*)0))
	cursor_pop_page(cursor);
rc = btree_cursor_first(cursor, key, data);
break;
default:
DPRINTF("unhandled/unimplemented cursor operation %u", op);
rc = BT_FAIL-1;
break;
}

mpage_prune(cursor->bt);

return rc;
1838}

1840static struct mpage *
1841btree_new_page(struct btree *bt, uint32_t flags)
1842{
struct mpage	*mp;

assert(bt != NULL)((bt != ((void*)0)) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 1845, __func__, "bt != NULL"));
assert(bt->txn != NULL)((bt->txn != ((void*)0)) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 1846, __func__, "bt->txn != NULL"));

DPRINTF("allocating new mpage %u, page size %u",
   bt->txn->next_pgno, bt->head.psize);
if ((mp = calloc(1, sizeof(*mp))) == NULL((void*)0))
return NULL((void*)0);
if ((mp->page = malloc(bt->head.psize)) == NULL((void*)0)) {
free(mp);
return NULL((void*)0);
}
mp->pgno = mp->page->pgno = bt->txn->next_pgno++;
mp->page->flags = flags;
mp->page->lowerb.fb.fb_lower = PAGEHDRSZ__builtin_offsetof(struct page, ptrs);
mp->page->upperb.fb.fb_upper = bt->head.psize;

if (IS_BRANCH(mp)((((mp)->page->flags) & (0x01)) == (0x01)))
bt->meta.branch_pages++;
else if (IS_LEAF(mp)((((mp)->page->flags) & (0x02)) == (0x02)))
bt->meta.leaf_pages++;
else if (IS_OVERFLOW(mp)((((mp)->page->flags) & (0x04)) == (0x04)))
bt->meta.overflow_pages++;

mpage_add(bt, mp);
mpage_dirty(bt, mp);

return mp;
1872}

1874static size_t
1875bt_leaf_size(struct btree *bt, struct btval *key, struct btval *data)
1876{
size_t		 sz;

sz = LEAFSIZE(key, data)(__builtin_offsetof(struct node, data) + (key)->size + (data
)->size);
if (data->size >= bt->head.psize / BT_MINKEYS4) {
/* put on overflow page */
sz -= data->size - sizeof(pgno_t);
}

return sz + sizeof(indx_t);
1886}

1888static size_t
1889bt_branch_size(struct btree *bt, struct btval *key)
1890{
size_t		 sz;

sz = INDXSIZE(key)(__builtin_offsetof(struct node, data) + ((key) == ((void*)0)
 ? 0 : (key)->size));
if (sz >= bt->head.psize / BT_MINKEYS4) {
/* put on overflow page */
/* not implemented */
/* sz -= key->size - sizeof(pgno_t); */
}

return sz + sizeof(indx_t);
1901}

1903static int
1904btree_write_overflow_data(struct btree *bt, struct page *p, struct btval *data)
1905{
size_t		 done = 0;
size_t		 sz;
size_t		 max;
pgno_t		*linkp;			/* linked page stored here */
struct mpage	*next = NULL((void*)0);

max = bt->head.psize - PAGEHDRSZ__builtin_offsetof(struct page, ptrs);

while (done < data->size) {
if (next != NULL((void*)0))
	p = next->page;
linkp = &p->p_next_pgnob.pb_next_pgno;
if (data->size - done > max) {
	/* need another overflow page */
	if ((next = btree_new_page(bt, P_OVERFLOW0x04)) == NULL((void*)0))
		return BT_FAIL-1;
	*linkp = next->pgno;
	DPRINTF("linking overflow page %u", next->pgno);
} else
	*linkp = 0;		/* indicates end of list */
sz = data->size - done;
if (sz > max)
	sz = max;
DPRINTF("copying %zu bytes to overflow page %u", sz, p->pgno);
bcopy((char *)data->data + done, p->ptrs, sz);
done += sz;
}

return BT_SUCCESS0;
1935}

1937/* Key prefix should already be stripped.
*/
1939static int
1940btree_add_node(struct btree *bt, struct mpage *mp, indx_t indx,
  struct btval *key, struct btval *data, pgno_t pgno, uint8_t flags)
1942{
unsigned int	 i;
size_t		 node_size = NODESIZE__builtin_offsetof(struct node, data);
indx_t		 ofs;
struct node	*node;
struct page	*p;
struct mpage	*ofp = NULL((void*)0);		/* overflow page */

p = mp->page;
assert(p->upper >= p->lower)((p->b.fb.fb_upper >= p->b.fb.fb_lower) ? (void)0 : __assert2
("/usr/src/usr.sbin/ldapd/btree.c", 1951, __func__, "p->upper >= p->lower"
));

DPRINTF("add node [%.*s] to %s page %u at index %d, key size %zu",
   key ? (int)key->size : 0, key ? (char *)key->data : NULL,
   IS_LEAF(mp) ? "leaf" : "branch",
   mp->pgno, indx, key ? key->size : 0);

if (key != NULL((void*)0))
node_size += key->size;

if (IS_LEAF(mp)((((mp)->page->flags) & (0x02)) == (0x02))) {
assert(data)((data) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 1962, __func__, "data"));
node_size += data->size;
if (F_ISSET(flags, F_BIGDATA)(((flags) & (0x01)) == (0x01))) {
	/* Data already on overflow page. */
	node_size -= data->size - sizeof(pgno_t);
} else if (data->size >= bt->head.psize / BT_MINKEYS4) {
	/* Put data on overflow page. */
	DPRINTF("data size is %zu, put on overflow page",
	    data->size);
	node_size -= data->size - sizeof(pgno_t);
	if ((ofp = btree_new_page(bt, P_OVERFLOW0x04)) == NULL((void*)0))
		return BT_FAIL-1;
	DPRINTF("allocated overflow page %u", ofp->pgno);
	flags |= F_BIGDATA0x01;
}
}

if (node_size + sizeof(indx_t) > SIZELEFT(mp)(indx_t)((mp)->page->b.fb.fb_upper - (mp)->page->
b.fb.fb_lower)) {
DPRINTF("not enough room in page %u, got %lu ptrs",
    mp->pgno, NUMKEYS(mp));
DPRINTF("upper - lower = %u - %u = %u", p->upper, p->lower,
    p->upper - p->lower);
DPRINTF("node size = %zu", node_size);
return BT_FAIL-1;
}

/* Move higher pointers up one slot. */
for (i = NUMKEYS(mp)(((mp)->page->b.fb.fb_lower - __builtin_offsetof(struct
 page, ptrs)) >> 1); i > indx; i--)
p->ptrs[i] = p->ptrs[i - 1];

/* Adjust free space offsets. */
ofs = p->upperb.fb.fb_upper - node_size;
assert(ofs >= p->lower + sizeof(indx_t))((ofs >= p->b.fb.fb_lower + sizeof(indx_t)) ? (void)0 :
 __assert2("/usr/src/usr.sbin/ldapd/btree.c", 1994, __func__,
 "ofs >= p->lower + sizeof(indx_t)"));
p->ptrs[indx] = ofs;
p->upperb.fb.fb_upper = ofs;
p->lowerb.fb.fb_lower += sizeof(indx_t);

/* Write the node data. */
node = NODEPTR(mp, indx)((struct node *)((char *)((mp)->page) + ((mp)->page)->
ptrs[indx]));
node->ksize = (key == NULL((void*)0)) ? 0 : key->size;
node->flags = flags;
if (IS_LEAF(mp)((((mp)->page->flags) & (0x02)) == (0x02)))
node->n_dsizep.np_dsize = data->size;
else
node->n_pgnop.np_pgno = pgno;

if (key)
bcopy(key->data, NODEKEY(node)(void *)((node)->data), key->size);

if (IS_LEAF(mp)((((mp)->page->flags) & (0x02)) == (0x02))) {
assert(key)((key) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 2012, __func__, "key"));
if (ofp == NULL((void*)0)) {
	if (F_ISSET(flags, F_BIGDATA)(((flags) & (0x01)) == (0x01)))
		bcopy(data->data, node->data + key->size,
		    sizeof(pgno_t));
	else
		bcopy(data->data, node->data + key->size,
		    data->size);
} else {
	bcopy(&ofp->pgno, node->data + key->size,
	    sizeof(pgno_t));
	if (btree_write_overflow_data(bt, ofp->page,
	    data) == BT_FAIL-1)
		return BT_FAIL-1;
}
}

return BT_SUCCESS0;
2030}

2032static void
2033btree_del_node(struct btree *bt, struct mpage *mp, indx_t indx)
2034{
unsigned int	 sz;
indx_t		 i, j, numkeys, ptr;
struct node	*node;
char		*base;

DPRINTF("delete node %u on %s page %u", indx,
   IS_LEAF(mp) ? "leaf" : "branch", mp->pgno);
assert(indx < NUMKEYS(mp))((indx < (((mp)->page->b.fb.fb_lower - __builtin_offsetof
(struct page, ptrs)) >> 1)) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 2042, __func__, "indx < NUMKEYS(mp)"));

node = NODEPTR(mp, indx)((struct node *)((char *)((mp)->page) + ((mp)->page)->
ptrs[indx]));
sz = NODESIZE__builtin_offsetof(struct node, data) + node->ksize;
if (IS_LEAF(mp)((((mp)->page->flags) & (0x02)) == (0x02))) {
if (F_ISSET(node->flags, F_BIGDATA)(((node->flags) & (0x01)) == (0x01)))
	sz += sizeof(pgno_t);
else
	sz += NODEDSZ(node)((node)->p.np_dsize);
}

ptr = mp->page->ptrs[indx];
numkeys = NUMKEYS(mp)(((mp)->page->b.fb.fb_lower - __builtin_offsetof(struct
 page, ptrs)) >> 1);
for (i = j = 0; i < numkeys; i++) {
if (i != indx) {
	mp->page->ptrs[j] = mp->page->ptrs[i];
	if (mp->page->ptrs[i] < ptr)
		mp->page->ptrs[j] += sz;
	j++;
}
}

base = (char *)mp->page + mp->page->upperb.fb.fb_upper;
bcopy(base, base + sz, ptr - mp->page->upperb.fb.fb_upper);

mp->page->lowerb.fb.fb_lower -= sizeof(indx_t);
mp->page->upperb.fb.fb_upper += sz;
2069}

2071struct cursor *
2072btree_txn_cursor_open(struct btree *bt, struct btree_txn *txn)
2073{
struct cursor	*cursor;

if (bt != NULL((void*)0) && txn != NULL((void*)0) && bt != txn->bt) {
errno(*__errno()) = EINVAL22;
return NULL((void*)0);
}

if (bt == NULL((void*)0)) {
if (txn == NULL((void*)0)) {
	errno(*__errno()) = EINVAL22;
	return NULL((void*)0);
}
bt = txn->bt;
}

if ((cursor = calloc(1, sizeof(*cursor))) != NULL((void*)0)) {
SLIST_INIT(&cursor->stack){ ((&cursor->stack)->slh_first) = ((void*)0); };
cursor->bt = bt;
cursor->txn = txn;
btree_ref(bt);
}

return cursor;
2097}

2099void
2100btree_cursor_close(struct cursor *cursor)
2101{
if (cursor != NULL((void*)0)) {
while (!CURSOR_EMPTY(cursor)(((&(cursor)->stack)->slh_first) == ((void*)0)))
	cursor_pop_page(cursor);

btree_close(cursor->bt);
free(cursor);
}
2109}

2111static int
2112btree_update_key(struct btree *bt, struct mpage *mp, indx_t indx,
  struct btval *key)
2114{
indx_t			 ptr, i, numkeys;
int			 delta;
size_t			 len;
struct node		*node;
char			*base;

node = NODEPTR(mp, indx)((struct node *)((char *)((mp)->page) + ((mp)->page)->
ptrs[indx]));
ptr = mp->page->ptrs[indx];
DPRINTF("update key %u (ofs %u) [%.*s] to [%.*s] on page %u",
   indx, ptr,
   (int)node->ksize, (char *)NODEKEY(node),
   (int)key->size, (char *)key->data,
   mp->pgno);

if (key->size != node->ksize) {
delta = key->size - node->ksize;
if (delta > 0 && SIZELEFT(mp)(indx_t)((mp)->page->b.fb.fb_upper - (mp)->page->
b.fb.fb_lower) < delta) {
	DPRINTF("OUCH! Not enough room, delta = %d", delta);
	return BT_FAIL-1;
}

numkeys = NUMKEYS(mp)(((mp)->page->b.fb.fb_lower - __builtin_offsetof(struct
 page, ptrs)) >> 1);
for (i = 0; i < numkeys; i++) {
	if (mp->page->ptrs[i] <= ptr)
		mp->page->ptrs[i] -= delta;
}

base = (char *)mp->page + mp->page->upperb.fb.fb_upper;
len = ptr - mp->page->upperb.fb.fb_upper + NODESIZE__builtin_offsetof(struct node, data);
bcopy(base, base - delta, len);
mp->page->upperb.fb.fb_upper -= delta;

node = NODEPTR(mp, indx)((struct node *)((char *)((mp)->page) + ((mp)->page)->
ptrs[indx]));
node->ksize = key->size;
}

bcopy(key->data, NODEKEY(node)(void *)((node)->data), key->size);

return BT_SUCCESS0;
2154}

2156static int
2157btree_adjust_prefix(struct btree *bt, struct mpage *src, int delta)
2158{
indx_t		 i;
struct node	*node;
struct btkey	 tmpkey;
struct btval	 key;

DPRINTF("adjusting prefix lengths on page %u with delta %d",
   src->pgno, delta);
assert(delta != 0)((delta != 0) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 2166, __func__, "delta != 0"));

for (i = 0; i < NUMKEYS(src)(((src)->page->b.fb.fb_lower - __builtin_offsetof(struct
 page, ptrs)) >> 1); i++) {
node = NODEPTR(src, i)((struct node *)((char *)((src)->page) + ((src)->page)->
ptrs[i]));
tmpkey.len = node->ksize - delta;
if (delta > 0) {
	if (F_ISSET(bt->flags, BT_REVERSEKEY)(((bt->flags) & (0x08)) == (0x08)))
		bcopy(NODEKEY(node)(void *)((node)->data), tmpkey.str, tmpkey.len);
	else
		bcopy((char *)NODEKEY(node)(void *)((node)->data) + delta, tmpkey.str,
		    tmpkey.len);
} else {
	if (F_ISSET(bt->flags, BT_REVERSEKEY)(((bt->flags) & (0x08)) == (0x08))) {
		bcopy(NODEKEY(node)(void *)((node)->data), tmpkey.str, node->ksize);
		bcopy(src->prefix.str, tmpkey.str + node->ksize,
		    -delta);
	} else {
		bcopy(src->prefix.str + src->prefix.len + delta,
		    tmpkey.str, -delta);
		bcopy(NODEKEY(node)(void *)((node)->data), tmpkey.str - delta,
		    node->ksize);
	}
}
key.size = tmpkey.len;
key.data = tmpkey.str;
if (btree_update_key(bt, src, i, &key) != BT_SUCCESS0)
	return BT_FAIL-1;
}

return BT_SUCCESS0;
2196}

2198/* Move a node from src to dst.
*/
2200static int
2201btree_move_node(struct btree *bt, struct mpage *src, indx_t srcindx,
  struct mpage *dst, indx_t dstindx)
2203{
int			 rc;
unsigned int		 pfxlen, mp_pfxlen = 0;
struct node		*srcnode;
struct mpage		*mp = NULL((void*)0);
struct btkey		 tmpkey, srckey;
struct btval		 key, data;

assert(src->parent)((src->parent) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 2211, __func__, "src->parent"));
assert(dst->parent)((dst->parent) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 2212, __func__, "dst->parent"));

srcnode = NODEPTR(src, srcindx)((struct node *)((char *)((src)->page) + ((src)->page)->
ptrs[srcindx]));
DPRINTF("moving %s node %u [%.*s] on page %u to node %u on page %u",
   IS_LEAF(src) ? "leaf" : "branch",
   srcindx,
   (int)srcnode->ksize, (char *)NODEKEY(srcnode),
   src->pgno,
   dstindx, dst->pgno);

find_common_prefix(bt, src);

if (IS_BRANCH(src)((((src)->page->flags) & (0x01)) == (0x01))) {
/* Need to check if the page the moved node points to
 * changes prefix.
 */
if ((mp = btree_get_mpage(bt, NODEPGNO(srcnode)((srcnode)->p.np_pgno))) == NULL((void*)0))
	return BT_FAIL-1;
mp->parent = src;
mp->parent_index = srcindx;
find_common_prefix(bt, mp);
mp_pfxlen = mp->prefix.len;
}

/* Mark src and dst as dirty. */
if ((src = mpage_touch(bt, src)) == NULL((void*)0) ||
   (dst = mpage_touch(bt, dst)) == NULL((void*)0))
return BT_FAIL-1;

find_common_prefix(bt, dst);

/* Check if src node has destination page prefix. Otherwise the
* destination page must expand its prefix on all its nodes.
*/
srckey.len = srcnode->ksize;
bcopy(NODEKEY(srcnode)(void *)((srcnode)->data), srckey.str, srckey.len);
common_prefix(bt, &srckey, &dst->prefix, &tmpkey);
if (tmpkey.len != dst->prefix.len) {
if (btree_adjust_prefix(bt, dst,
    tmpkey.len - dst->prefix.len) != BT_SUCCESS0)
	return BT_FAIL-1;
bcopy(&tmpkey, &dst->prefix, sizeof(tmpkey));
}

if (srcindx == 0 && IS_BRANCH(src)((((src)->page->flags) & (0x01)) == (0x01))) {
struct mpage	*low;

/* must find the lowest key below src
 */
assert(btree_search_page_root(bt, src, NULL, NULL, 0,((btree_search_page_root(bt, src, ((void*)0), ((void*)0), 0, &
low) == 0) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 2262, __func__, "btree_search_page_root(bt, src, NULL, NULL, 0, &low) == BT_SUCCESS"
))
    &low) == BT_SUCCESS)((btree_search_page_root(bt, src, ((void*)0), ((void*)0), 0, &
low) == 0) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 2262, __func__, "btree_search_page_root(bt, src, NULL, NULL, 0, &low) == BT_SUCCESS"
));
expand_prefix(bt, low, 0, &srckey);
DPRINTF("found lowest key [%.*s] on leaf page %u",
    (int)srckey.len, srckey.str, low->pgno);
} else {
srckey.len = srcnode->ksize;
bcopy(NODEKEY(srcnode)(void *)((srcnode)->data), srckey.str, srcnode->ksize);
}
find_common_prefix(bt, src);

/* expand the prefix */
tmpkey.len = sizeof(tmpkey.str);
concat_prefix(bt, src->prefix.str, src->prefix.len,
   srckey.str, srckey.len, tmpkey.str, &tmpkey.len);

/* Add the node to the destination page. Adjust prefix for
* destination page.
*/
key.size = tmpkey.len;
key.data = tmpkey.str;
remove_prefix(bt, &key, dst->prefix.len);
data.size = NODEDSZ(srcnode)((srcnode)->p.np_dsize);
data.data = NODEDATA(srcnode)(void *)((char *)(srcnode)->data + (srcnode)->ksize);
rc = btree_add_node(bt, dst, dstindx, &key, &data, NODEPGNO(srcnode)((srcnode)->p.np_pgno),
   srcnode->flags);
if (rc != BT_SUCCESS0)
return rc;

/* Delete the node from the source page.
*/
btree_del_node(bt, src, srcindx);

/* Update the parent separators.
*/
if (srcindx == 0 && src->parent_index != 0) {
expand_prefix(bt, src, 0, &tmpkey);
key.size = tmpkey.len;
key.data = tmpkey.str;
remove_prefix(bt, &key, src->parent->prefix.len);

DPRINTF("update separator for source page %u to [%.*s]",
    src->pgno, (int)key.size, (char *)key.data);
if (btree_update_key(bt, src->parent, src->parent_index,
    &key) != BT_SUCCESS0)
	return BT_FAIL-1;
}

if (srcindx == 0 && IS_BRANCH(src)((((src)->page->flags) & (0x01)) == (0x01))) {
struct btval	 nullkey;
nullkey.size = 0;
assert(btree_update_key(bt, src, 0, &nullkey) == BT_SUCCESS)((btree_update_key(bt, src, 0, &nullkey) == 0) ? (void)0 :
 __assert2("/usr/src/usr.sbin/ldapd/btree.c", 2312, __func__,
 "btree_update_key(bt, src, 0, &nullkey) == BT_SUCCESS"));
}

if (dstindx == 0 && dst->parent_index != 0) {
expand_prefix(bt, dst, 0, &tmpkey);
key.size = tmpkey.len;
key.data = tmpkey.str;
remove_prefix(bt, &key, dst->parent->prefix.len);

DPRINTF("update separator for destination page %u to [%.*s]",
    dst->pgno, (int)key.size, (char *)key.data);
if (btree_update_key(bt, dst->parent, dst->parent_index,
    &key) != BT_SUCCESS0)
	return BT_FAIL-1;
}

if (dstindx == 0 && IS_BRANCH(dst)((((dst)->page->flags) & (0x01)) == (0x01))) {
struct btval	 nullkey;
nullkey.size = 0;
assert(btree_update_key(bt, dst, 0, &nullkey) == BT_SUCCESS)((btree_update_key(bt, dst, 0, &nullkey) == 0) ? (void)0 :
 __assert2("/usr/src/usr.sbin/ldapd/btree.c", 2331, __func__,
 "btree_update_key(bt, dst, 0, &nullkey) == BT_SUCCESS"));
}

/* We can get a new page prefix here!
* Must update keys in all nodes of this page!
*/
pfxlen = src->prefix.len;
find_common_prefix(bt, src);
if (src->prefix.len != pfxlen) {
if (btree_adjust_prefix(bt, src,
    src->prefix.len - pfxlen) != BT_SUCCESS0)
	return BT_FAIL-1;
}

pfxlen = dst->prefix.len;
find_common_prefix(bt, dst);
if (dst->prefix.len != pfxlen) {
if (btree_adjust_prefix(bt, dst,
    dst->prefix.len - pfxlen) != BT_SUCCESS0)
	return BT_FAIL-1;
}

if (IS_BRANCH(dst)((((dst)->page->flags) & (0x01)) == (0x01))) {
assert(mp)((mp) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 2354, __func__, "mp"));
mp->parent = dst;
mp->parent_index = dstindx;
find_common_prefix(bt, mp);
if (mp->prefix.len != mp_pfxlen) {
	DPRINTF("moved branch node has changed prefix");
	if ((mp = mpage_touch(bt, mp)) == NULL((void*)0))
		return BT_FAIL-1;
	if (btree_adjust_prefix(bt, mp,
	    mp->prefix.len - mp_pfxlen) != BT_SUCCESS0)
		return BT_FAIL-1;
}
}

return BT_SUCCESS0;
2369}

2371static int
2372btree_merge(struct btree *bt, struct mpage *src, struct mpage *dst)
2373{
int			 rc;
indx_t			 i;
unsigned int		 pfxlen;
struct node		*srcnode;
struct btkey		 tmpkey, dstpfx;
struct btval		 key, data;

DPRINTF("merging page %u and %u", src->pgno, dst->pgno);

assert(src->parent)((src->parent) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 2383, __func__, "src->parent"));	/* can't merge root page */
assert(dst->parent)((dst->parent) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 2384, __func__, "dst->parent"));
assert(bt->txn != NULL)((bt->txn != ((void*)0)) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 2385, __func__, "bt->txn != NULL"));

/* Mark src and dst as dirty. */
if ((src = mpage_touch(bt, src)) == NULL((void*)0) ||
   (dst = mpage_touch(bt, dst)) == NULL((void*)0))
return BT_FAIL-1;

find_common_prefix(bt, src);
find_common_prefix(bt, dst);

/* Check if source nodes has destination page prefix. Otherwise
* the destination page must expand its prefix on all its nodes.
*/
common_prefix(bt, &src->prefix, &dst->prefix, &dstpfx);
if (dstpfx.len != dst->prefix.len) {
if (btree_adjust_prefix(bt, dst,
    dstpfx.len - dst->prefix.len) != BT_SUCCESS0)
	return BT_FAIL-1;
bcopy(&dstpfx, &dst->prefix, sizeof(dstpfx));
}

/* Move all nodes from src to dst.
*/
for (i = 0; i < NUMKEYS(src)(((src)->page->b.fb.fb_lower - __builtin_offsetof(struct
 page, ptrs)) >> 1); i++) {
srcnode = NODEPTR(src, i)((struct node *)((char *)((src)->page) + ((src)->page)->
ptrs[i]));

/* If branch node 0 (implicit key), find the real key.
 */
if (i == 0 && IS_BRANCH(src)((((src)->page->flags) & (0x01)) == (0x01))) {
	struct mpage	*low;

	/* must find the lowest key below src
	 */
	assert(btree_search_page_root(bt, src, NULL, NULL, 0,((btree_search_page_root(bt, src, ((void*)0), ((void*)0), 0, &
low) == 0) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 2419, __func__, "btree_search_page_root(bt, src, NULL, NULL, 0, &low) == BT_SUCCESS"
))
	    &low) == BT_SUCCESS)((btree_search_page_root(bt, src, ((void*)0), ((void*)0), 0, &
low) == 0) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 2419, __func__, "btree_search_page_root(bt, src, NULL, NULL, 0, &low) == BT_SUCCESS"
));
	expand_prefix(bt, low, 0, &tmpkey);
	DPRINTF("found lowest key [%.*s] on leaf page %u",
	    (int)tmpkey.len, tmpkey.str, low->pgno);
} else {
	expand_prefix(bt, src, i, &tmpkey);
}

key.size = tmpkey.len;
key.data = tmpkey.str;

remove_prefix(bt, &key, dst->prefix.len);
data.size = NODEDSZ(srcnode)((srcnode)->p.np_dsize);
data.data = NODEDATA(srcnode)(void *)((char *)(srcnode)->data + (srcnode)->ksize);
rc = btree_add_node(bt, dst, NUMKEYS(dst)(((dst)->page->b.fb.fb_lower - __builtin_offsetof(struct
 page, ptrs)) >> 1), &key,
    &data, NODEPGNO(srcnode)((srcnode)->p.np_pgno), srcnode->flags);
if (rc != BT_SUCCESS0)
	return rc;
}

DPRINTF("dst page %u now has %lu keys (%.1f%% filled)",
   dst->pgno, NUMKEYS(dst), (float)PAGEFILL(bt, dst) / 10);

/* Unlink the src page from parent.
*/
btree_del_node(bt, src->parent, src->parent_index);
if (src->parent_index == 0) {
key.size = 0;
if (btree_update_key(bt, src->parent, 0, &key) != BT_SUCCESS0)
	return BT_FAIL-1;

pfxlen = src->prefix.len;
find_common_prefix(bt, src);
assert (src->prefix.len == pfxlen)((src->prefix.len == pfxlen) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 2452, __func__, "src->prefix.len == pfxlen"));
}

if (IS_LEAF(src)((((src)->page->flags) & (0x02)) == (0x02)))
bt->meta.leaf_pages--;
else
bt->meta.branch_pages--;

return btree_rebalance(bt, src->parent);
2461}

2463#define FILL_THRESHOLD250	 250

2465static int
2466btree_rebalance(struct btree *bt, struct mpage *mp)
2467{
struct node	*node;
struct mpage	*parent;
struct mpage	*root;
struct mpage	*neighbor = NULL((void*)0);
indx_t		 si = 0, di = 0;

assert(bt != NULL)((bt != ((void*)0)) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 2474, __func__, "bt != NULL"));
assert(bt->txn != NULL)((bt->txn != ((void*)0)) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 2475, __func__, "bt->txn != NULL"));
assert(mp != NULL)((mp != ((void*)0)) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 2476, __func__, "mp != NULL"));

DPRINTF("rebalancing %s page %u (has %lu keys, %.1f%% full)",
   IS_LEAF(mp) ? "leaf" : "branch",
   mp->pgno, NUMKEYS(mp), (float)PAGEFILL(bt, mp) / 10);

if (PAGEFILL(bt, mp)(1000 * ((bt)->head.psize - __builtin_offsetof(struct page
, ptrs) - (indx_t)((mp)->page->b.fb.fb_upper - (mp)->
page->b.fb.fb_lower)) / ((bt)->head.psize - __builtin_offsetof
(struct page, ptrs))) >= FILL_THRESHOLD250) {
DPRINTF("no need to rebalance page %u, above fill threshold",
    mp->pgno);
return BT_SUCCESS0;
}

parent = mp->parent;

if (parent == NULL((void*)0)) {
if (NUMKEYS(mp)(((mp)->page->b.fb.fb_lower - __builtin_offsetof(struct
 page, ptrs)) >> 1) == 0) {
	DPRINTF("tree is completely empty");
	bt->txn->root = P_INVALID0xFFFFFFFF;
	bt->meta.depth--;
	bt->meta.leaf_pages--;
} else if (IS_BRANCH(mp)((((mp)->page->flags) & (0x01)) == (0x01)) && NUMKEYS(mp)(((mp)->page->b.fb.fb_lower - __builtin_offsetof(struct
 page, ptrs)) >> 1) == 1) {
	DPRINTF("collapsing root page!");
	bt->txn->root = NODEPGNO(NODEPTR(mp, 0))((((struct node *)((char *)((mp)->page) + ((mp)->page)->
ptrs[0])))->p.np_pgno);
	if ((root = btree_get_mpage(bt, bt->txn->root)) == NULL((void*)0))
		return BT_FAIL-1;
	root->parent = NULL((void*)0);
	bt->meta.depth--;
	bt->meta.branch_pages--;
} else
	DPRINTF("root page doesn't need rebalancing");
return BT_SUCCESS0;
}

/* The parent (branch page) must have at least 2 pointers,
* otherwise the tree is invalid.
*/
assert(NUMKEYS(parent) > 1)(((((parent)->page->b.fb.fb_lower - __builtin_offsetof(
struct page, ptrs)) >> 1) > 1) ? (void)0 : __assert2
("/usr/src/usr.sbin/ldapd/btree.c", 2512, __func__, "NUMKEYS(parent) > 1"
));

/* Leaf page fill factor is below the threshold.
* Try to move keys from left or right neighbor, or
* merge with a neighbor page.
*/

/* Find neighbors.
*/
if (mp->parent_index == 0) {
/* We're the leftmost leaf in our parent.
 */
DPRINTF("reading right neighbor");
node = NODEPTR(parent, mp->parent_index + 1)((struct node *)((char *)((parent)->page) + ((parent)->
page)->ptrs[mp->parent_index + 1]));
if ((neighbor = btree_get_mpage(bt, NODEPGNO(node)((node)->p.np_pgno))) == NULL((void*)0))
	return BT_FAIL-1;
neighbor->parent_index = mp->parent_index + 1;
si = 0;
di = NUMKEYS(mp)(((mp)->page->b.fb.fb_lower - __builtin_offsetof(struct
 page, ptrs)) >> 1);
} else {
/* There is at least one neighbor to the left.
 */
DPRINTF("reading left neighbor");
node = NODEPTR(parent, mp->parent_index - 1)((struct node *)((char *)((parent)->page) + ((parent)->
page)->ptrs[mp->parent_index - 1]));
if ((neighbor = btree_get_mpage(bt, NODEPGNO(node)((node)->p.np_pgno))) == NULL((void*)0))
	return BT_FAIL-1;
neighbor->parent_index = mp->parent_index - 1;
si = NUMKEYS(neighbor)(((neighbor)->page->b.fb.fb_lower - __builtin_offsetof(
struct page, ptrs)) >> 1) - 1;
di = 0;
}
neighbor->parent = parent;

DPRINTF("found neighbor page %u (%lu keys, %.1f%% full)",
   neighbor->pgno, NUMKEYS(neighbor), (float)PAGEFILL(bt, neighbor) / 10);

/* If the neighbor page is above threshold and has at least two
* keys, move one key from it.
*
* Otherwise we should try to merge them, but that might not be
* possible, even if both are below threshold, as prefix expansion
* might make keys larger. FIXME: detect this
*/
if (PAGEFILL(bt, neighbor)(1000 * ((bt)->head.psize - __builtin_offsetof(struct page
, ptrs) - (indx_t)((neighbor)->page->b.fb.fb_upper - (neighbor
)->page->b.fb.fb_lower)) / ((bt)->head.psize - __builtin_offsetof
(struct page, ptrs))) >= FILL_THRESHOLD250 && NUMKEYS(neighbor)(((neighbor)->page->b.fb.fb_lower - __builtin_offsetof(
struct page, ptrs)) >> 1) >= 2)
return btree_move_node(bt, neighbor, si, mp, di);
else { /* FIXME: if (has_enough_room()) */
if (mp->parent_index == 0)
	return btree_merge(bt, neighbor, mp);
else
	return btree_merge(bt, mp, neighbor);
}
2562}

2564int
2565btree_txn_del(struct btree *bt, struct btree_txn *txn,
  struct btval *key, struct btval *data)
2567{
int		 rc, exact, close_txn = 0;
unsigned int	 ki;
struct node	*leaf;
struct mpage	*mp;

DPRINTF("========> delete key %.*s", (int)key->size, (char *)key->data);

assert(key != NULL)((key != ((void*)0)) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 2575, __func__, "key != NULL"));

if (bt != NULL((void*)0) && txn != NULL((void*)0) && bt != txn->bt) {
errno(*__errno()) = EINVAL22;
return BT_FAIL-1;
}

if (txn != NULL((void*)0) && F_ISSET(txn->flags, BT_TXN_RDONLY)(((txn->flags) & (0x01)) == (0x01))) {
errno(*__errno()) = EINVAL22;
return BT_FAIL-1;
}

if (bt == NULL((void*)0)) {
if (txn == NULL((void*)0)) {
	errno(*__errno()) = EINVAL22;
	return BT_FAIL-1;
}
bt = txn->bt;
}

if (key->size == 0 || key->size > MAXKEYSIZE255) {
errno(*__errno()) = EINVAL22;
return BT_FAIL-1;
}

if (txn == NULL((void*)0)) {
close_txn = 1;
if ((txn = btree_txn_begin(bt, 0)) == NULL((void*)0))
	return BT_FAIL-1;
}

if ((rc = btree_search_page(bt, txn, key, NULL((void*)0), 1, &mp)) != BT_SUCCESS0)
goto done;

leaf = btree_search_node(bt, mp, key, &exact, &ki);
if (leaf == NULL((void*)0) || !exact) {
errno(*__errno()) = ENOENT2;
rc = BT_FAIL-1;
goto done;
}

if (data && (rc = btree_read_data(bt, NULL((void*)0), leaf, data)) != BT_SUCCESS0)
goto done;

btree_del_node(bt, mp, ki);
bt->meta.entries--;
rc = btree_rebalance(bt, mp);
if (rc != BT_SUCCESS0)
txn->flags |= BT_TXN_ERROR0x02;

2625done:
if (close_txn) {
if (rc == BT_SUCCESS0)
	rc = btree_txn_commit(txn);
else
	btree_txn_abort(txn);
}
mpage_prune(bt);
return rc;
2634}

2636/* Reduce the length of the prefix separator <sep> to the minimum length that
* still makes it uniquely distinguishable from <min>.
*
* <min> is guaranteed to be sorted less than <sep>
*
* On return, <sep> is modified to the minimum length.
*/
2643static void
2644bt_reduce_separator(struct btree *bt, struct node *min, struct btval *sep)
2645{
size_t		 n = 0;
char		*p1;
char		*p2;

if (F_ISSET(bt->flags, BT_REVERSEKEY)(((bt->flags) & (0x08)) == (0x08))) {

assert(sep->size > 0)((sep->size > 0) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 2652, __func__, "sep->size > 0"));

p1 = (char *)NODEKEY(min)(void *)((min)->data) + min->ksize - 1;
p2 = (char *)sep->data + sep->size - 1;

while (p1 >= (char *)NODEKEY(min)(void *)((min)->data) && *p1 == *p2) {
	assert(p2 > (char *)sep->data)((p2 > (char *)sep->data) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 2658, __func__, "p2 > (char *)sep->data"));
	p1--;
	p2--;
	n++;
}

sep->data = p2;
sep->size = n + 1;
} else {

assert(min->ksize > 0)((min->ksize > 0) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 2668, __func__, "min->ksize > 0"));
assert(sep->size > 0)((sep->size > 0) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 2669, __func__, "sep->size > 0"));

p1 = (char *)NODEKEY(min)(void *)((min)->data);
p2 = (char *)sep->data;

while (*p1 == *p2) {
	p1++;
	p2++;
	n++;
	if (n == min->ksize || n == sep->size)
		break;
}

sep->size = n + 1;
}

DPRINTF("reduced separator to [%.*s] > [%.*s]",
   (int)sep->size, (char *)sep->data,
   (int)min->ksize, (char *)NODEKEY(min));
2688}

2690/* Split page <*mpp>, and insert <key,(data|newpgno)> in either left or
* right sibling, at index <*newindxp> (as if unsplit). Updates *mpp and
* *newindxp with the actual values after split, ie if *mpp and *newindxp
* refer to a node in the new right sibling page.
*/
2695static int
2696btree_split(struct btree *bt, struct mpage **mpp, unsigned int *newindxp,
  struct btval *newkey, struct btval *newdata, pgno_t newpgno)
2698{
uint8_t		 flags;
int		 rc = BT_SUCCESS0, ins_new = 0;
indx_t		 newindx;
pgno_t		 pgno = 0;
size_t		 orig_pfx_len, left_pfx_diff, right_pfx_diff, pfx_diff;
unsigned int	 i, j, split_indx;
struct node	*node;
struct mpage	*pright, *p, *mp;
struct btval	 sepkey, rkey, rdata;
struct btkey	 tmpkey;
struct page	*copy;

assert(bt != NULL)((bt != ((void*)0)) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 2711, __func__, "bt != NULL"));
assert(bt->txn != NULL)((bt->txn != ((void*)0)) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 2712, __func__, "bt->txn != NULL"));

mp = *mpp;
newindx = *newindxp;

DPRINTF("-----> splitting %s page %u and adding [%.*s] at index %d",
   IS_LEAF(mp) ? "leaf" : "branch", mp->pgno,
   (int)newkey->size, (char *)newkey->data, *newindxp);
DPRINTF("page %u has prefix [%.*s]", mp->pgno,
   (int)mp->prefix.len, (char *)mp->prefix.str);
orig_pfx_len = mp->prefix.len;

if (mp->parent == NULL((void*)0)) {
if ((mp->parent = btree_new_page(bt, P_BRANCH0x01)) == NULL((void*)0))
	return BT_FAIL-1;
mp->parent_index = 0;
bt->txn->root = mp->parent->pgno;
DPRINTF("root split! new root = %u", mp->parent->pgno);
bt->meta.depth++;

/* Add left (implicit) pointer. */
if (btree_add_node(bt, mp->parent, 0, NULL((void*)0), NULL((void*)0),
    mp->pgno, 0) != BT_SUCCESS0)
	return BT_FAIL-1;
} else {
DPRINTF("parent branch page is %u", mp->parent->pgno);
}

/* Create a right sibling. */
if ((pright = btree_new_page(bt, mp->page->flags)) == NULL((void*)0))
return BT_FAIL-1;
pright->parent = mp->parent;
pright->parent_index = mp->parent_index + 1;
DPRINTF("new right sibling: page %u", pright->pgno);

/* Move half of the keys to the right sibling. */
if ((copy = malloc(bt->head.psize)) == NULL((void*)0))
return BT_FAIL-1;
bcopy(mp->page, copy, bt->head.psize);
assert(mp->ref == 0)((mp->ref == 0) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 2751, __func__, "mp->ref == 0"));				/* XXX */
memset(&mp->page->ptrs, 0, bt->head.psize - PAGEHDRSZ__builtin_offsetof(struct page, ptrs));
mp->page->lowerb.fb.fb_lower = PAGEHDRSZ__builtin_offsetof(struct page, ptrs);
mp->page->upperb.fb.fb_upper = bt->head.psize;

split_indx = NUMKEYSP(copy)(((copy)->b.fb.fb_lower - __builtin_offsetof(struct page, ptrs
)) >> 1) / 2 + 1;

/* First find the separating key between the split pages.
*/
memset(&sepkey, 0, sizeof(sepkey));
if (newindx == split_indx) {
sepkey.size = newkey->size;
sepkey.data = newkey->data;
remove_prefix(bt, &sepkey, mp->prefix.len);
} else {
node = NODEPTRP(copy, split_indx)((struct node *)((char *)(copy) + (copy)->ptrs[split_indx]
));
sepkey.size = node->ksize;
sepkey.data = NODEKEY(node)(void *)((node)->data);
}

if (IS_LEAF(mp)((((mp)->page->flags) & (0x02)) == (0x02)) && bt->cmp == NULL((void*)0)) {
/* Find the smallest separator. */
/* Ref: Prefix B-trees, R. Bayer, K. Unterauer, 1977 */
node = NODEPTRP(copy, split_indx - 1)((struct node *)((char *)(copy) + (copy)->ptrs[split_indx -
 1]));
bt_reduce_separator(bt, node, &sepkey);
}

/* Fix separator wrt parent prefix. */
if (bt->cmp == NULL((void*)0)) {
tmpkey.len = sizeof(tmpkey.str);
concat_prefix(bt, mp->prefix.str, mp->prefix.len,
    sepkey.data, sepkey.size, tmpkey.str, &tmpkey.len);
sepkey.data = tmpkey.str;
sepkey.size = tmpkey.len;
}

DPRINTF("separator is [%.*s]", (int)sepkey.size, (char *)sepkey.data);

/* Copy separator key to the parent.
*/
if (SIZELEFT(pright->parent)(indx_t)((pright->parent)->page->b.fb.fb_upper - (pright
->parent)->page->b.fb.fb_lower) < bt_branch_size(bt, &sepkey)) {
rc = btree_split(bt, &pright->parent, &pright->parent_index,
    &sepkey, NULL((void*)0), pright->pgno);

/* Right page might now have changed parent.
 * Check if left page also changed parent.
 */
if (pright->parent != mp->parent &&
    mp->parent_index >= NUMKEYS(mp->parent)(((mp->parent)->page->b.fb.fb_lower - __builtin_offsetof
(struct page, ptrs)) >> 1)) {
	mp->parent = pright->parent;
	mp->parent_index = pright->parent_index - 1;
}
} else {
remove_prefix(bt, &sepkey, pright->parent->prefix.len);
rc = btree_add_node(bt, pright->parent, pright->parent_index,
    &sepkey, NULL((void*)0), pright->pgno, 0);
}
if (rc != BT_SUCCESS0) {
free(copy);
return BT_FAIL-1;
}

/* Update prefix for right and left page, if the parent was split.
*/
find_common_prefix(bt, pright);
assert(orig_pfx_len <= pright->prefix.len)((orig_pfx_len <= pright->prefix.len) ? (void)0 : __assert2
("/usr/src/usr.sbin/ldapd/btree.c", 2816, __func__, "orig_pfx_len <= pright->prefix.len"
));
right_pfx_diff = pright->prefix.len - orig_pfx_len;

find_common_prefix(bt, mp);
assert(orig_pfx_len <= mp->prefix.len)((orig_pfx_len <= mp->prefix.len) ? (void)0 : __assert2
("/usr/src/usr.sbin/ldapd/btree.c", 2820, __func__, "orig_pfx_len <= mp->prefix.len"
));
left_pfx_diff = mp->prefix.len - orig_pfx_len;

for (i = j = 0; i <= NUMKEYSP(copy)(((copy)->b.fb.fb_lower - __builtin_offsetof(struct page, ptrs
)) >> 1); j++) {
if (i < split_indx) {
	/* Re-insert in left sibling. */
	p = mp;
	pfx_diff = left_pfx_diff;
} else {
	/* Insert in right sibling. */
	if (i == split_indx)
		/* Reset insert index for right sibling. */
		j = (i == newindx && ins_new);
	p = pright;
	pfx_diff = right_pfx_diff;
}

if (i == newindx && !ins_new) {
	/* Insert the original entry that caused the split. */
	rkey.data = newkey->data;
	rkey.size = newkey->size;
	if (IS_LEAF(mp)((((mp)->page->flags) & (0x02)) == (0x02))) {
		rdata.data = newdata->data;
		rdata.size = newdata->size;
	} else
		pgno = newpgno;
	flags = 0;
	pfx_diff = p->prefix.len;

	ins_new = 1;

	/* Update page and index for the new key. */
	*newindxp = j;
	*mpp = p;
} else if (i == NUMKEYSP(copy)(((copy)->b.fb.fb_lower - __builtin_offsetof(struct page, ptrs
)) >> 1)) {
	break;
} else {
	node = NODEPTRP(copy, i)((struct node *)((char *)(copy) + (copy)->ptrs[i]));
	rkey.data = NODEKEY(node)(void *)((node)->data);
	rkey.size = node->ksize;
	if (IS_LEAF(mp)((((mp)->page->flags) & (0x02)) == (0x02))) {
		rdata.data = NODEDATA(node)(void *)((char *)(node)->data + (node)->ksize);
		rdata.size = node->n_dsizep.np_dsize;
	} else
		pgno = node->n_pgnop.np_pgno;
	flags = node->flags;

	i++;
}

if (!IS_LEAF(mp)((((mp)->page->flags) & (0x02)) == (0x02)) && j == 0) {
	/* First branch index doesn't need key data. */
	rkey.size = 0;
} else
	remove_prefix(bt, &rkey, pfx_diff);

rc = btree_add_node(bt, p, j, &rkey, &rdata, pgno,flags);
}

free(copy);
return rc;
2881}

2883int
2884btree_txn_put(struct btree *bt, struct btree_txn *txn,
  struct btval *key, struct btval *data, unsigned int flags)
2886{
int		 rc = BT_SUCCESS0, exact, close_txn = 0;
unsigned int	 ki;
struct node	*leaf;
struct mpage	*mp;
struct btval	 xkey;

assert(key != NULL)((key != ((void*)0)) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 2893, __func__, "key != NULL"));
assert(data != NULL)((data != ((void*)0)) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 2894, __func__, "data != NULL"));

if (bt != NULL((void*)0) && txn != NULL((void*)0) && bt != txn->bt) {
errno(*__errno()) = EINVAL22;
return BT_FAIL-1;
}

if (txn != NULL((void*)0) && F_ISSET(txn->flags, BT_TXN_RDONLY)(((txn->flags) & (0x01)) == (0x01))) {
errno(*__errno()) = EINVAL22;
return BT_FAIL-1;
}

if (bt == NULL((void*)0)) {
if (txn == NULL((void*)0)) {
	errno(*__errno()) = EINVAL22;
	return BT_FAIL-1;
}
bt = txn->bt;
}

if (key->size == 0 || key->size > MAXKEYSIZE255) {
errno(*__errno()) = EINVAL22;
return BT_FAIL-1;
}

DPRINTF("==> put key %.*s, size %zu, data size %zu",
(int)key->size, (char *)key->data, key->size, data->size);

if (txn == NULL((void*)0)) {
close_txn = 1;
if ((txn = btree_txn_begin(bt, 0)) == NULL((void*)0))
	return BT_FAIL-1;
}

rc = btree_search_page(bt, txn, key, NULL((void*)0), 1, &mp);
if (rc == BT_SUCCESS0) {
leaf = btree_search_node(bt, mp, key, &exact, &ki);
if (leaf && exact) {
	if (F_ISSET(flags, BT_NOOVERWRITE)(((flags) & (1)) == (1))) {
		DPRINTF("duplicate key %.*s",
		    (int)key->size, (char *)key->data);
		errno(*__errno()) = EEXIST17;
		rc = BT_FAIL-1;
		goto done;
	}
	btree_del_node(bt, mp, ki);
}
if (leaf == NULL((void*)0)) {		/* append if not found */
	ki = NUMKEYS(mp)(((mp)->page->b.fb.fb_lower - __builtin_offsetof(struct
 page, ptrs)) >> 1);
	DPRINTF("appending key at index %d", ki);
}
} else if (errno(*__errno()) == ENOENT2) {
/* new file, just write a root leaf page */
DPRINTF("allocating new root leaf page");
if ((mp = btree_new_page(bt, P_LEAF0x02)) == NULL((void*)0)) {
	rc = BT_FAIL-1;
	goto done;
}
txn->root = mp->pgno;
bt->meta.depth++;
ki = 0;
}
else
goto done;

assert(IS_LEAF(mp))((((((mp)->page->flags) & (0x02)) == (0x02))) ? (void
)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c", 2959, __func__
, "IS_LEAF(mp)"));
DPRINTF("there are %lu keys, should insert new key at index %u",
NUMKEYS(mp), ki);

/* Copy the key pointer as it is modified by the prefix code. The
* caller might have malloc'ed the data.
*/
xkey.data = key->data;
xkey.size = key->size;

if (SIZELEFT(mp)(indx_t)((mp)->page->b.fb.fb_upper - (mp)->page->
b.fb.fb_lower) < bt_leaf_size(bt, key, data)) {
rc = btree_split(bt, &mp, &ki, &xkey, data, P_INVALID0xFFFFFFFF);
} else {
/* There is room already in this leaf page. */
remove_prefix(bt, &xkey, mp->prefix.len);
rc = btree_add_node(bt, mp, ki, &xkey, data, 0, 0);
}

if (rc != BT_SUCCESS0)
txn->flags |= BT_TXN_ERROR0x02;
else
bt->meta.entries++;

2982done:
if (close_txn) {
if (rc == BT_SUCCESS0)
	rc = btree_txn_commit(txn);
else
	btree_txn_abort(txn);
}
mpage_prune(bt);
return rc;
2991}

2993static pgno_t
2994btree_compact_tree(struct btree *bt, pgno_t pgno, struct btree *btc)
2995{
ssize_t		 rc;
indx_t		 i;
pgno_t		*pnext, next;
struct node	*node;
struct page	*p;
struct mpage	*mp;

/* Get the page and make a copy of it.
*/
if ((mp = btree_get_mpage(bt, pgno)) == NULL((void*)0))
return P_INVALID0xFFFFFFFF;
if ((p = malloc(bt->head.psize)) == NULL((void*)0))
return P_INVALID0xFFFFFFFF;
bcopy(mp->page, p, bt->head.psize);

      /* Go through all nodes in the (copied) page and update the
       * page pointers.
*/
if (F_ISSET(p->flags, P_BRANCH)(((p->flags) & (0x01)) == (0x01))) {
for (i = 0; i < NUMKEYSP(p)(((p)->b.fb.fb_lower - __builtin_offsetof(struct page, ptrs
)) >> 1); i++) {
	node = NODEPTRP(p, i)((struct node *)((char *)(p) + (p)->ptrs[i]));
	node->n_pgnop.np_pgno = btree_compact_tree(bt, node->n_pgnop.np_pgno, btc);
	if (node->n_pgnop.np_pgno == P_INVALID0xFFFFFFFF) {
		free(p);
		return P_INVALID0xFFFFFFFF;
	}
}
} else if (F_ISSET(p->flags, P_LEAF)(((p->flags) & (0x02)) == (0x02))) {
for (i = 0; i < NUMKEYSP(p)(((p)->b.fb.fb_lower - __builtin_offsetof(struct page, ptrs
)) >> 1); i++) {
	node = NODEPTRP(p, i)((struct node *)((char *)(p) + (p)->ptrs[i]));
	if (F_ISSET(node->flags, F_BIGDATA)(((node->flags) & (0x01)) == (0x01))) {
		bcopy(NODEDATA(node)(void *)((char *)(node)->data + (node)->ksize), &next, sizeof(next));
		next = btree_compact_tree(bt, next, btc);
		if (next == P_INVALID0xFFFFFFFF) {
			free(p);
			return P_INVALID0xFFFFFFFF;
		}
		bcopy(&next, NODEDATA(node)(void *)((char *)(node)->data + (node)->ksize), sizeof(next));
	}
}
} else if (F_ISSET(p->flags, P_OVERFLOW)(((p->flags) & (0x04)) == (0x04))) {
pnext = &p->p_next_pgnob.pb_next_pgno;
if (*pnext > 0) {
	*pnext = btree_compact_tree(bt, *pnext, btc);
	if (*pnext == P_INVALID0xFFFFFFFF) {
		free(p);
		return P_INVALID0xFFFFFFFF;
	}
}
} else
assert(0)((0) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c",
 3046, __func__, "0"));

pgno = p->pgno = btc->txn->next_pgno++;
rc = write(btc->fd, p, bt->head.psize);
free(p);
if (rc != (ssize_t)bt->head.psize)
return P_INVALID0xFFFFFFFF;
mpage_prune(bt);
return pgno;
3055}

3057int
3058btree_compact(struct btree *bt)
3059{
char			*compact_path = NULL((void*)0);
struct btree		*btc;
struct btree_txn	*txn, *txnc = NULL((void*)0);
int			 fd;
pgno_t			 root;

assert(bt != NULL)((bt != ((void*)0)) ? (void)0 : __assert2("/usr/src/usr.sbin/ldapd/btree.c"
, 3066, __func__, "bt != NULL"));

DPRINTF("compacting btree %p with path %s", bt, bt->path);

if (bt->path == NULL((void*)0)) {
errno(*__errno()) = EINVAL22;
return BT_FAIL-1;
}

if ((txn = btree_txn_begin(bt, 0)) == NULL((void*)0))
return BT_FAIL-1;

if (asprintf(&compact_path, "%s.compact.XXXXXX", bt->path) == -1) {
btree_txn_abort(txn);
return BT_FAIL-1;
}
fd = mkstemp(compact_path);
if (fd == -1) {
free(compact_path);
btree_txn_abort(txn);
return BT_FAIL-1;
}

if ((btc = btree_open_fd(fd, 0)) == NULL((void*)0))
goto failed;
bcopy(&bt->meta, &btc->meta, sizeof(bt->meta));
btc->meta.revisions = 0;

if ((txnc = btree_txn_begin(btc, 0)) == NULL((void*)0))
goto failed;

if (bt->meta.root != P_INVALID0xFFFFFFFF) {
root = btree_compact_tree(bt, bt->meta.root, btc);
if (root == P_INVALID0xFFFFFFFF)
	goto failed;
if (btree_write_meta(btc, root, 0) != BT_SUCCESS0)
	goto failed;
}

fsync(fd);

DPRINTF("renaming %s to %s", compact_path, bt->path);
if (rename(compact_path, bt->path) != 0)
goto failed;

/* Write a "tombstone" meta page so other processes can pick up
* the change and re-open the file.
*/
if (btree_write_meta(bt, P_INVALID0xFFFFFFFF, BT_TOMBSTONE0x01) != BT_SUCCESS0)
goto failed;

btree_txn_abort(txn);
btree_txn_abort(txnc);
free(compact_path);
btree_close(btc);
mpage_prune(bt);
return 0;

3124failed:
btree_txn_abort(txn);
btree_txn_abort(txnc);
unlink(compact_path);
free(compact_path);
btree_close(btc);
mpage_prune(bt);
return BT_FAIL-1;
3132}

3134/* Reverts the last change. Truncates the file at the last root page.
*/
3136int
3137btree_revert(struct btree *bt)
3138{
if (btree_read_meta(bt, NULL((void*)0)) != 0)
return -1;

DPRINTF("truncating file at page %u", bt->meta.root);
return ftruncate(bt->fd, bt->head.psize * bt->meta.root);
3144}

3146void
3147btree_set_cache_size(struct btree *bt, unsigned int cache_size)
3148{
bt->stat.max_cache = cache_size;
3150}

3152unsigned int
3153btree_get_flags(struct btree *bt)
3154{
return (bt->flags & ~BT_FIXPADDING0x01);
3156}

3158const char *
3159btree_get_path(struct btree *bt)
3160{
return bt->path;
3162}

3164const struct btree_stat *
3165btree_stat(struct btree *bt)
3166{
if (bt == NULL((void*)0))
return NULL((void*)0);

bt->stat.branch_pages = bt->meta.branch_pages;
bt->stat.leaf_pages = bt->meta.leaf_pages;
bt->stat.overflow_pages = bt->meta.overflow_pages;
bt->stat.revisions = bt->meta.revisions;
bt->stat.depth = bt->meta.depth;
bt->stat.entries = bt->meta.entries;
bt->stat.psize = bt->head.psize;
bt->stat.created_at = bt->meta.created_at;

return &bt->stat;
3180}
