1 /* mdb.c - memory-mapped database library */
3 * Copyright 2011 Howard Chu, Symas Corp.
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted only as authorized by the OpenLDAP
10 * A copy of this license is available in the file LICENSE in the
11 * top-level directory of the distribution or, alternatively, at
12 * <http://www.OpenLDAP.org/license.html>.
14 * This code is derived from btree.c written by Martin Hedenfalk.
16 * Copyright (c) 2009, 2010 Martin Hedenfalk <martin@bzero.se>
18 * Permission to use, copy, modify, and distribute this software for any
19 * purpose with or without fee is hereby granted, provided that the above
20 * copyright notice and this permission notice appear in all copies.
22 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
23 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
24 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
25 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
26 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
27 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
28 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
30 #include <sys/types.h>
32 #include <sys/param.h>
35 #ifdef HAVE_SYS_FILE_H
53 #define ULONG unsigned long
62 #if DEBUG && defined(__GNUC__)
63 # define DPRINTF(fmt, ...) \
64 fprintf(stderr, "%s:%d: " fmt "\n", __func__, __LINE__, ##__VA_ARGS__)
66 # define DPRINTF(...) ((void) 0)
71 #define MDB_MAGIC 0xBEEFC0DE
73 #define MAXKEYSIZE 511
75 #define P_INVALID (~0UL)
77 #define F_ISSET(w, f) (((w) & (f)) == (f))
79 typedef uint16_t indx_t;
81 #define DEFAULT_READERS 126
82 #define DEFAULT_MAPSIZE 1048576
84 /* Lock descriptor stuff */
86 #define CACHELINE 64 /* most CPUs. Itanium uses 128 */
89 typedef struct MDB_rxbody {
95 typedef struct MDB_reader {
98 #define mr_txnid mru.mrx.mrb_txnid
99 #define mr_pid mru.mrx.mrb_pid
100 #define mr_tid mru.mrx.mrb_tid
101 /* cache line alignment */
102 char pad[(sizeof(MDB_rxbody)+CACHELINE-1) & ~(CACHELINE-1)];
106 typedef struct MDB_txbody {
108 uint32_t mtb_version;
109 pthread_mutex_t mtb_mutex;
111 uint32_t mtb_numreaders;
112 uint32_t mtb_me_toggle;
115 typedef struct MDB_txninfo {
118 #define mti_magic mt1.mtb.mtb_magic
119 #define mti_version mt1.mtb.mtb_version
120 #define mti_mutex mt1.mtb.mtb_mutex
121 #define mti_txnid mt1.mtb.mtb_txnid
122 #define mti_numreaders mt1.mtb.mtb_numreaders
123 #define mti_me_toggle mt1.mtb.mtb_me_toggle
124 char pad[(sizeof(MDB_txbody)+CACHELINE-1) & ~(CACHELINE-1)];
127 pthread_mutex_t mt2_wmutex;
128 #define mti_wmutex mt2.mt2_wmutex
129 char pad[(sizeof(pthread_mutex_t)+CACHELINE-1) & ~(CACHELINE-1)];
131 MDB_reader mti_readers[1];
134 /* Common header for all page types. Overflow pages
135 * occupy a number of contiguous pages with no
136 * headers on any page after the first.
138 typedef struct MDB_page { /* represents a page of storage */
139 #define mp_pgno mp_p.p_pgno
141 pgno_t p_pgno; /* page number */
142 void * p_align; /* for IL32P64 */
144 #define P_BRANCH 0x01 /* branch page */
145 #define P_LEAF 0x02 /* leaf page */
146 #define P_OVERFLOW 0x04 /* overflow page */
147 #define P_META 0x08 /* meta page */
148 #define P_DIRTY 0x10 /* dirty page */
149 #define P_LEAF2 0x20 /* DB with small, fixed size keys and no data */
151 #define mp_lower mp_pb.pb.pb_lower
152 #define mp_upper mp_pb.pb.pb_upper
153 #define mp_pages mp_pb.pb_pages
156 indx_t pb_lower; /* lower bound of free space */
157 indx_t pb_upper; /* upper bound of free space */
159 uint32_t pb_pages; /* number of overflow pages */
161 indx_t pb_ksize; /* on a LEAF2 page */
165 indx_t mp_ptrs[1]; /* dynamic size */
168 #define PAGEHDRSZ ((unsigned) offsetof(MDB_page, mp_ptrs))
170 #define NUMKEYS(p) (((p)->mp_lower - PAGEHDRSZ) >> 1)
171 #define SIZELEFT(p) (indx_t)((p)->mp_upper - (p)->mp_lower)
172 #define PAGEFILL(env, p) (1000L * ((env)->me_psize - PAGEHDRSZ - SIZELEFT(p)) / \
173 ((env)->me_psize - PAGEHDRSZ))
174 #define IS_LEAF(p) F_ISSET((p)->mp_flags, P_LEAF)
175 #define IS_BRANCH(p) F_ISSET((p)->mp_flags, P_BRANCH)
176 #define IS_OVERFLOW(p) F_ISSET((p)->mp_flags, P_OVERFLOW)
178 #define OVPAGES(size, psize) (PAGEHDRSZ + size + psize - 1) / psize;
180 typedef struct MDB_db {
184 ULONG md_branch_pages;
186 ULONG md_overflow_pages;
194 typedef struct MDB_meta { /* meta (footer) page content */
197 void *mm_address; /* address for fixed mapping */
198 size_t mm_mapsize; /* size of mmap region */
199 MDB_db mm_dbs[2]; /* first is free space, 2nd is main db */
200 #define mm_psize mm_dbs[0].md_pad
201 #define mm_flags mm_dbs[0].md_flags
202 pgno_t mm_last_pg; /* last used page in file */
203 ULONG mm_txnid; /* txnid that committed this page */
206 typedef struct MDB_dhead { /* a dirty page */
208 unsigned md_pi; /* parent index */
212 typedef struct MDB_dpage {
217 typedef struct MDB_oldpages {
218 struct MDB_oldpages *mo_next;
220 pgno_t mo_pages[1]; /* dynamic */
223 typedef struct MDB_pageparent {
229 static MDB_dpage *mdb_alloc_page(MDB_txn *txn, MDB_page *parent, unsigned int parent_idx, int num);
230 static int mdb_touch(MDB_txn *txn, MDB_pageparent *mp);
232 typedef struct MDB_ppage { /* ordered list of pages */
234 unsigned int mp_ki; /* cursor index on page */
237 #define CURSOR_TOP(c) (&(c)->mc_stack[(c)->mc_snum-1])
238 #define CURSOR_PARENT(c) (&(c)->mc_stack[(c)->mc_snum-2])
244 MDB_ppage mc_stack[32]; /* stack of parent pages */
245 unsigned int mc_snum; /* number of pushed pages */
247 short mc_initialized; /* 1 if initialized */
248 short mc_eof; /* 1 if end is reached */
249 struct MDB_xcursor *mc_xcursor;
252 #define METADATA(p) ((void *)((char *)p + PAGEHDRSZ))
254 typedef struct MDB_node {
255 #define mn_pgno mn_p.np_pgno
256 #define mn_dsize mn_p.np_dsize
258 pgno_t np_pgno; /* child page number */
259 uint32_t np_dsize; /* leaf data size */
261 unsigned int mn_flags:4;
262 unsigned int mn_ksize:12; /* key size */
263 #define F_BIGDATA 0x01 /* data put on overflow page */
264 #define F_SUBDATA 0x02 /* data is a sub-database */
268 typedef struct MDB_dbx {
270 MDB_cmp_func *md_cmp; /* user compare function */
271 MDB_cmp_func *md_dcmp; /* user dupsort function */
272 MDB_rel_func *md_rel; /* user relocate function */
274 unsigned int md_dirty;
278 pgno_t mt_next_pgno; /* next unallocated page */
282 pgno_t *mt_free_pgs; /* this is an IDL */
284 MIDL2 *dirty_list; /* modified pages */
287 MDB_dbx *mt_dbxs; /* array */
289 unsigned int mt_numdbs;
291 #define MDB_TXN_RDONLY 0x01 /* read-only transaction */
292 #define MDB_TXN_ERROR 0x02 /* an error has occurred */
293 #define MDB_TXN_METOGGLE 0x04 /* used meta page 1 */
294 unsigned int mt_flags;
297 /* Context for sorted-dup records */
298 typedef struct MDB_xcursor {
299 MDB_cursor mx_cursor;
308 int me_mfd; /* just for writing the meta pages */
309 #define MDB_FATAL_ERROR 0x80000000U
311 uint32_t me_extrapad; /* unused for now */
312 unsigned int me_maxreaders;
313 unsigned int me_numdbs;
314 unsigned int me_maxdbs;
317 MDB_txninfo *me_txns;
318 MDB_meta *me_metas[2];
320 MDB_txn *me_txn; /* current write transaction */
322 off_t me_size; /* current file size */
323 pgno_t me_maxpg; /* me_mapsize / me_psize */
324 unsigned int me_psize;
325 unsigned int me_db_toggle;
326 MDB_dbx *me_dbxs; /* array */
328 MDB_oldpages *me_pghead;
329 pthread_key_t me_txkey; /* thread-key for readers */
330 pgno_t me_free_pgs[MDB_IDL_UM_SIZE];
331 MIDL2 me_dirty_list[MDB_IDL_DB_SIZE];
334 #define NODESIZE offsetof(MDB_node, mn_data)
336 #define INDXSIZE(k) (NODESIZE + ((k) == NULL ? 0 : (k)->mv_size))
337 #define LEAFSIZE(k, d) (NODESIZE + (k)->mv_size + (d)->mv_size)
338 #define NODEPTR(p, i) ((MDB_node *)((char *)(p) + (p)->mp_ptrs[i]))
339 #define NODEKEY(node) (void *)((node)->mn_data)
340 #define NODEDATA(node) (void *)((char *)(node)->mn_data + (node)->mn_ksize)
341 #define NODEPGNO(node) ((node)->mn_pgno)
342 #define NODEDSZ(node) ((node)->mn_dsize)
344 #define MDB_COMMIT_PAGES 64 /* max number of pages to write in one commit */
346 static int mdb_search_page_root(MDB_txn *txn,
347 MDB_dbi dbi, MDB_val *key,
348 MDB_cursor *cursor, int modify,
349 MDB_pageparent *mpp);
350 static int mdb_search_page(MDB_txn *txn,
351 MDB_dbi dbi, MDB_val *key,
352 MDB_cursor *cursor, int modify,
353 MDB_pageparent *mpp);
355 static int mdb_env_read_header(MDB_env *env, MDB_meta *meta);
356 static int mdb_env_read_meta(MDB_env *env, int *which);
357 static int mdb_env_write_meta(MDB_txn *txn);
358 static MDB_page *mdb_get_page(MDB_txn *txn, pgno_t pgno);
360 static MDB_node *mdb_search_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp,
361 MDB_val *key, int *exactp, unsigned int *kip);
362 static int mdb_add_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp,
363 indx_t indx, MDB_val *key, MDB_val *data,
364 pgno_t pgno, uint8_t flags);
365 static void mdb_del_node(MDB_page *mp, indx_t indx);
366 static int mdb_del0(MDB_txn *txn, MDB_dbi dbi, unsigned int ki,
367 MDB_pageparent *mpp, MDB_node *leaf);
368 static int mdb_put0(MDB_txn *txn, MDB_dbi dbi,
369 MDB_val *key, MDB_val *data, unsigned int flags);
370 static int mdb_read_data(MDB_txn *txn, MDB_node *leaf, MDB_val *data);
372 static int mdb_rebalance(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *mp);
373 static int mdb_update_key(MDB_page *mp, indx_t indx, MDB_val *key);
374 static int mdb_move_node(MDB_txn *txn, MDB_dbi dbi,
375 MDB_pageparent *src, indx_t srcindx,
376 MDB_pageparent *dst, indx_t dstindx);
377 static int mdb_merge(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *src,
378 MDB_pageparent *dst);
379 static int mdb_split(MDB_txn *txn, MDB_dbi dbi, MDB_page **mpp,
380 unsigned int *newindxp, MDB_val *newkey,
381 MDB_val *newdata, pgno_t newpgno);
382 static MDB_dpage *mdb_new_page(MDB_txn *txn, MDB_dbi dbi, uint32_t flags, int num);
384 static void cursor_pop_page(MDB_cursor *cursor);
385 static MDB_ppage *cursor_push_page(MDB_cursor *cursor,
388 static int mdb_set_key(MDB_node *node, MDB_val *key);
389 static int mdb_sibling(MDB_cursor *cursor, int move_right);
390 static int mdb_cursor_next(MDB_cursor *cursor,
391 MDB_val *key, MDB_val *data, MDB_cursor_op op);
392 static int mdb_cursor_prev(MDB_cursor *cursor,
393 MDB_val *key, MDB_val *data, MDB_cursor_op op);
394 static int mdb_cursor_set(MDB_cursor *cursor,
395 MDB_val *key, MDB_val *data, MDB_cursor_op op, int *exactp);
396 static int mdb_cursor_first(MDB_cursor *cursor,
397 MDB_val *key, MDB_val *data);
398 static int mdb_cursor_last(MDB_cursor *cursor,
399 MDB_val *key, MDB_val *data);
401 static void mdb_xcursor_init0(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx);
402 static void mdb_xcursor_init1(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx, MDB_node *node);
403 static void mdb_xcursor_fini(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx);
405 static size_t mdb_leaf_size(MDB_env *env, MDB_val *key,
407 static size_t mdb_branch_size(MDB_env *env, MDB_val *key);
409 static int memncmp(const void *s1, size_t n1,
410 const void *s2, size_t n2);
411 static int memnrcmp(const void *s1, size_t n1,
412 const void *s2, size_t n2);
415 memncmp(const void *s1, size_t n1, const void *s2, size_t n2)
417 int diff, len_diff = -1;
420 len_diff = (n1 > n2);
423 diff = memcmp(s1, s2, n1);
424 return diff ? diff : len_diff;
428 memnrcmp(const void *s1, size_t n1, const void *s2, size_t n2)
430 const unsigned char *p1, *p2, *p1_lim;
437 p1 = (const unsigned char *)s1 + n1 - 1;
438 p2 = (const unsigned char *)s2 + n2 - 1;
440 for (p1_lim = (n1 <= n2 ? s1 : s2); *p1 == *p2; p1--, p2--) {
442 return (p1 != s1) ? (p1 != p2) : (p2 != s2) ? -1 : 0;
448 mdb_version(int *maj, int *min, int *pat)
450 *maj = MDB_VERSION_MAJOR;
451 *min = MDB_VERSION_MINOR;
452 *pat = MDB_VERSION_PATCH;
453 return MDB_VERSION_STRING;
456 static const char *errstr[] = {
457 "MDB_KEYEXIST: Key/data pair already exists",
458 "MDB_NOTFOUND: No matching key/data pair found",
459 "MDB_PAGE_NOTFOUND: Requested page not found",
460 "MDB_CORRUPTED: Located page was wrong type",
461 "MDB_PANIC: Update of meta page failed",
462 "MDB_VERSION_MISMATCH: Database environment version mismatch"
466 mdb_strerror(int err)
469 return ("Successful return: 0");
471 if (err >= MDB_KEYEXIST && err <= MDB_VERSION_MISMATCH)
472 return (char *)errstr[err - MDB_KEYEXIST];
474 return strerror(err);
478 mdb_cmp(MDB_txn *txn, MDB_dbi dbi, const MDB_val *a, const MDB_val *b)
480 return txn->mt_dbxs[dbi].md_cmp(a, b);
484 _mdb_cmp(MDB_txn *txn, MDB_dbi dbi, const MDB_val *key1, const MDB_val *key2)
486 if (txn->mt_dbs[dbi].md_flags & (MDB_REVERSEKEY
487 #if __BYTE_ORDER == __LITTLE_ENDIAN
491 return memnrcmp(key1->mv_data, key1->mv_size, key2->mv_data, key2->mv_size);
493 return memncmp((char *)key1->mv_data, key1->mv_size, key2->mv_data, key2->mv_size);
496 /* Allocate new page(s) for writing */
498 mdb_alloc_page(MDB_txn *txn, MDB_page *parent, unsigned int parent_idx, int num)
501 pgno_t pgno = P_INVALID;
505 if (txn->mt_txnid > 2) {
507 oldest = txn->mt_txnid - 2;
508 if (!txn->mt_env->me_pghead && txn->mt_dbs[FREE_DBI].md_root != P_INVALID) {
509 /* See if there's anything in the free DB */
514 mpp.mp_parent = NULL;
516 mdb_search_page(txn, FREE_DBI, NULL, NULL, 0, &mpp);
517 leaf = NODEPTR(mpp.mp_page, 0);
518 kptr = (ULONG *)NODEKEY(leaf);
520 /* It's potentially usable, unless there are still
521 * older readers outstanding. Grab it.
523 if (oldest > *kptr) {
528 mdb_read_data(txn, leaf, &data);
529 idl = (ULONG *)data.mv_data;
530 mop = malloc(sizeof(MDB_oldpages) + MDB_IDL_SIZEOF(idl) - sizeof(pgno_t));
531 mop->mo_next = txn->mt_env->me_pghead;
532 mop->mo_txnid = *kptr;
533 txn->mt_env->me_pghead = mop;
534 memcpy(mop->mo_pages, idl, MDB_IDL_SIZEOF(idl));
539 DPRINTF("IDL read txn %lu root %lu num %lu",
540 mop->mo_txnid, txn->mt_dbs[FREE_DBI].md_root, idl[0]);
541 for (i=0; i<idl[0]; i++) {
542 DPRINTF("IDL %lu", idl[i+1]);
546 /* drop this IDL from the DB */
547 mpp.mp_parent = NULL;
549 mdb_search_page(txn, FREE_DBI, NULL, NULL, 1, &mpp);
550 leaf = NODEPTR(mpp.mp_page, 0);
551 mdb_del0(txn, FREE_DBI, 0, &mpp, leaf);
554 if (txn->mt_env->me_pghead) {
556 for (i=0; i<txn->mt_env->me_txns->mti_numreaders; i++) {
557 ULONG mr = txn->mt_env->me_txns->mti_readers[i].mr_txnid;
560 oldest = txn->mt_env->me_txns->mti_readers[i].mr_txnid;
562 if (oldest > txn->mt_env->me_pghead->mo_txnid) {
563 MDB_oldpages *mop = txn->mt_env->me_pghead;
564 txn->mt_oldest = oldest;
566 /* FIXME: For now, always use fresh pages. We
567 * really ought to search the free list for a
572 /* peel pages off tail, so we only have to truncate the list */
573 pgno = MDB_IDL_LAST(mop->mo_pages);
574 if (MDB_IDL_IS_RANGE(mop->mo_pages)) {
576 if (mop->mo_pages[2] > mop->mo_pages[1])
577 mop->mo_pages[0] = 0;
581 if (MDB_IDL_IS_ZERO(mop->mo_pages)) {
582 txn->mt_env->me_pghead = mop->mo_next;
590 if (pgno == P_INVALID) {
591 /* DB size is maxed out */
592 if (txn->mt_next_pgno + num >= txn->mt_env->me_maxpg)
595 if ((dp = malloc(txn->mt_env->me_psize * num + sizeof(MDB_dhead))) == NULL)
598 dp->h.md_parent = parent;
599 dp->h.md_pi = parent_idx;
600 if (pgno == P_INVALID) {
601 dp->p.mp_pgno = txn->mt_next_pgno;
602 txn->mt_next_pgno += num;
604 dp->p.mp_pgno = pgno;
606 mid.mid = dp->p.mp_pgno;
608 mdb_midl2_insert(txn->mt_u.dirty_list, &mid);
613 /* Touch a page: make it dirty and re-insert into tree with updated pgno.
616 mdb_touch(MDB_txn *txn, MDB_pageparent *pp)
618 MDB_page *mp = pp->mp_page;
623 if (!F_ISSET(mp->mp_flags, P_DIRTY)) {
625 if ((dp = mdb_alloc_page(txn, pp->mp_parent, pp->mp_pi, 1)) == NULL)
627 DPRINTF("touched page %lu -> %lu", mp->mp_pgno, dp->p.mp_pgno);
628 mdb_midl_insert(txn->mt_free_pgs, mp->mp_pgno);
629 pgno = dp->p.mp_pgno;
630 memcpy(&dp->p, mp, txn->mt_env->me_psize);
633 mp->mp_flags |= P_DIRTY;
635 /* Update the page number to new touched page. */
636 if (pp->mp_parent != NULL)
637 NODEPGNO(NODEPTR(pp->mp_parent, pp->mp_pi)) = mp->mp_pgno;
644 mdb_env_sync(MDB_env *env, int force)
647 if (force || !F_ISSET(env->me_flags, MDB_NOSYNC)) {
648 if (fdatasync(env->me_fd))
655 mdb_txn_begin(MDB_env *env, int rdonly, MDB_txn **ret)
660 if (env->me_flags & MDB_FATAL_ERROR) {
661 DPRINTF("mdb_txn_begin: environment had fatal error, must shutdown!");
664 if ((txn = calloc(1, sizeof(MDB_txn))) == NULL) {
665 DPRINTF("calloc: %s", strerror(errno));
670 txn->mt_flags |= MDB_TXN_RDONLY;
672 txn->mt_u.dirty_list = env->me_dirty_list;
673 txn->mt_u.dirty_list[0].mid = 0;
674 txn->mt_free_pgs = env->me_free_pgs;
675 txn->mt_free_pgs[0] = 0;
677 pthread_mutex_lock(&env->me_txns->mti_wmutex);
678 env->me_txns->mti_txnid++;
681 txn->mt_txnid = env->me_txns->mti_txnid;
683 MDB_reader *r = pthread_getspecific(env->me_txkey);
686 pthread_mutex_lock(&env->me_txns->mti_mutex);
687 for (i=0; i<env->me_txns->mti_numreaders; i++)
688 if (env->me_txns->mti_readers[i].mr_pid == 0)
690 if (i == env->me_maxreaders) {
691 pthread_mutex_unlock(&env->me_txns->mti_mutex);
694 env->me_txns->mti_readers[i].mr_pid = getpid();
695 env->me_txns->mti_readers[i].mr_tid = pthread_self();
696 r = &env->me_txns->mti_readers[i];
697 pthread_setspecific(env->me_txkey, r);
698 if (i >= env->me_txns->mti_numreaders)
699 env->me_txns->mti_numreaders = i+1;
700 pthread_mutex_unlock(&env->me_txns->mti_mutex);
702 r->mr_txnid = txn->mt_txnid;
703 txn->mt_u.reader = r;
710 toggle = env->me_txns->mti_me_toggle;
711 if ((rc = mdb_env_read_meta(env, &toggle)) != MDB_SUCCESS) {
716 /* Copy the DB arrays */
717 txn->mt_numdbs = env->me_numdbs;
718 txn->mt_dbxs = env->me_dbxs; /* mostly static anyway */
719 txn->mt_dbs = malloc(env->me_maxdbs * sizeof(MDB_db));
720 memcpy(txn->mt_dbs, env->me_meta->mm_dbs, 2 * sizeof(MDB_db));
721 if (txn->mt_numdbs > 2)
722 memcpy(txn->mt_dbs+2, env->me_dbs[env->me_db_toggle]+2,
723 (txn->mt_numdbs - 2) * sizeof(MDB_db));
727 txn->mt_flags |= MDB_TXN_METOGGLE;
728 txn->mt_next_pgno = env->me_meta->mm_last_pg+1;
731 DPRINTF("begin transaction %lu on mdbenv %p, root page %lu",
732 txn->mt_txnid, (void *) env, txn->mt_dbs[MAIN_DBI].md_root);
739 mdb_txn_abort(MDB_txn *txn)
747 DPRINTF("abort transaction %lu on mdbenv %p, root page %lu",
748 txn->mt_txnid, (void *) env, txn->mt_dbs[MAIN_DBI].md_root);
752 if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
753 txn->mt_u.reader->mr_txnid = 0;
758 /* Discard all dirty pages. */
759 for (i=1; i<=txn->mt_u.dirty_list[0].mid; i++)
760 free(txn->mt_u.dirty_list[i].mptr);
762 while ((mop = txn->mt_env->me_pghead)) {
763 txn->mt_env->me_pghead = mop->mo_next;
768 env->me_txns->mti_txnid--;
769 for (i=2; i<env->me_numdbs; i++)
770 env->me_dbxs[i].md_dirty = 0;
771 pthread_mutex_unlock(&env->me_txns->mti_wmutex);
778 mdb_txn_commit(MDB_txn *txn)
787 struct iovec iov[MDB_COMMIT_PAGES];
790 assert(txn->mt_env != NULL);
794 if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
799 if (txn != env->me_txn) {
800 DPRINTF("attempt to commit unknown transaction");
805 if (F_ISSET(txn->mt_flags, MDB_TXN_ERROR)) {
806 DPRINTF("error flag is set, can't commit");
811 if (!txn->mt_u.dirty_list[0].mid)
814 DPRINTF("committing transaction %lu on mdbenv %p, root page %lu",
815 txn->mt_txnid, (void *) env, txn->mt_dbs[MAIN_DBI].md_root);
817 /* should only be one record now */
818 if (env->me_pghead) {
822 mop = env->me_pghead;
823 key.mv_size = sizeof(pgno_t);
824 key.mv_data = (char *)&mop->mo_txnid;
825 data.mv_size = MDB_IDL_SIZEOF(mop->mo_pages);
826 data.mv_data = mop->mo_pages;
827 mdb_put0(txn, FREE_DBI, &key, &data, 0);
828 free(env->me_pghead);
829 env->me_pghead = NULL;
831 /* save to free list */
832 if (!MDB_IDL_IS_ZERO(txn->mt_free_pgs)) {
836 /* make sure last page of freeDB is touched and on freelist */
837 key.mv_size = MAXKEYSIZE+1;
839 mpp.mp_parent = NULL;
841 mdb_search_page(txn, FREE_DBI, &key, NULL, 1, &mpp);
846 ULONG *idl = txn->mt_free_pgs;
847 DPRINTF("IDL write txn %lu root %lu num %lu",
848 txn->mt_txnid, txn->mt_dbs[FREE_DBI].md_root, idl[0]);
849 for (i=0; i<idl[0]; i++) {
850 DPRINTF("IDL %lu", idl[i+1]);
854 /* write to last page of freeDB */
855 key.mv_size = sizeof(pgno_t);
856 key.mv_data = (char *)&txn->mt_txnid;
857 data.mv_size = MDB_IDL_SIZEOF(txn->mt_free_pgs);
858 data.mv_data = txn->mt_free_pgs;
859 mdb_put0(txn, FREE_DBI, &key, &data, 0);
862 /* Update DB root pointers. Their pages have already been
863 * touched so this is all in-place and cannot fail.
867 data.mv_size = sizeof(MDB_db);
869 for (i = 2; i < txn->mt_numdbs; i++) {
870 if (txn->mt_dbxs[i].md_dirty) {
871 data.mv_data = &txn->mt_dbs[i];
872 mdb_put0(txn, MAIN_DBI, &txn->mt_dbxs[i].md_name, &data, 0);
877 /* Commit up to MDB_COMMIT_PAGES dirty pages to disk until done.
885 for (; i<=txn->mt_u.dirty_list[0].mid; i++) {
886 dp = txn->mt_u.dirty_list[i].mptr;
887 if (dp->p.mp_pgno != next) {
889 DPRINTF("committing %u dirty pages", n);
890 rc = writev(env->me_fd, iov, n);
894 DPRINTF("short write, filesystem full?");
896 DPRINTF("writev: %s", strerror(errno));
903 lseek(env->me_fd, dp->p.mp_pgno * env->me_psize, SEEK_SET);
904 next = dp->p.mp_pgno;
906 DPRINTF("committing page %lu", dp->p.mp_pgno);
907 iov[n].iov_len = env->me_psize * dp->h.md_num;
908 iov[n].iov_base = &dp->p;
909 size += iov[n].iov_len;
910 next = dp->p.mp_pgno + dp->h.md_num;
911 /* clear dirty flag */
912 dp->p.mp_flags &= ~P_DIRTY;
913 if (++n >= MDB_COMMIT_PAGES) {
922 DPRINTF("committing %u dirty pages", n);
923 rc = writev(env->me_fd, iov, n);
927 DPRINTF("short write, filesystem full?");
929 DPRINTF("writev: %s", strerror(errno));
936 /* Drop the dirty pages.
938 for (i=1; i<=txn->mt_u.dirty_list[0].mid; i++)
939 free(txn->mt_u.dirty_list[i].mptr);
941 if ((n = mdb_env_sync(env, 0)) != 0 ||
942 (n = mdb_env_write_meta(txn)) != MDB_SUCCESS) {
949 /* update the DB tables */
951 int toggle = !env->me_db_toggle;
953 for (i = 2; i < env->me_numdbs; i++) {
954 if (txn->mt_dbxs[i].md_dirty) {
955 env->me_dbs[toggle][i] = txn->mt_dbs[i];
956 txn->mt_dbxs[i].md_dirty = 0;
959 for (i = env->me_numdbs; i < txn->mt_numdbs; i++) {
960 txn->mt_dbxs[i].md_dirty = 0;
961 env->me_dbxs[i] = txn->mt_dbxs[i];
962 env->me_dbs[toggle][i] = txn->mt_dbs[i];
964 env->me_db_toggle = toggle;
965 env->me_numdbs = txn->mt_numdbs;
970 pthread_mutex_unlock(&env->me_txns->mti_wmutex);
977 mdb_env_read_header(MDB_env *env, MDB_meta *meta)
986 /* We don't know the page size yet, so use a minimum value.
989 if ((rc = pread(env->me_fd, page, PAGESIZE, 0)) == 0) {
991 } else if (rc != PAGESIZE) {
994 DPRINTF("read: %s", strerror(errno));
998 p = (MDB_page *)page;
1000 if (!F_ISSET(p->mp_flags, P_META)) {
1001 DPRINTF("page %lu not a meta page", p->mp_pgno);
1006 if (m->mm_magic != MDB_MAGIC) {
1007 DPRINTF("meta has invalid magic");
1011 if (m->mm_version != MDB_VERSION) {
1012 DPRINTF("database is version %u, expected version %u",
1013 m->mm_version, MDB_VERSION);
1014 return MDB_VERSION_MISMATCH;
1017 memcpy(meta, m, sizeof(*m));
1022 mdb_env_init_meta(MDB_env *env, MDB_meta *meta)
1029 DPRINTF("writing new meta page");
1030 psize = sysconf(_SC_PAGE_SIZE);
1032 meta->mm_magic = MDB_MAGIC;
1033 meta->mm_version = MDB_VERSION;
1034 meta->mm_psize = psize;
1035 meta->mm_last_pg = 1;
1036 meta->mm_flags = env->me_flags & 0xffff;
1037 meta->mm_flags |= MDB_INTEGERKEY;
1038 meta->mm_dbs[0].md_root = P_INVALID;
1039 meta->mm_dbs[1].md_root = P_INVALID;
1041 p = calloc(2, psize);
1043 p->mp_flags = P_META;
1046 memcpy(m, meta, sizeof(*meta));
1048 q = (MDB_page *)((char *)p + psize);
1051 q->mp_flags = P_META;
1054 memcpy(m, meta, sizeof(*meta));
1056 rc = write(env->me_fd, p, psize * 2);
1058 return (rc == (int)psize * 2) ? MDB_SUCCESS : errno;
1062 mdb_env_write_meta(MDB_txn *txn)
1065 MDB_meta meta, metab;
1067 int rc, len, toggle;
1070 assert(txn != NULL);
1071 assert(txn->mt_env != NULL);
1073 toggle = !F_ISSET(txn->mt_flags, MDB_TXN_METOGGLE);
1074 DPRINTF("writing meta page %d for root page %lu",
1075 toggle, txn->mt_dbs[MAIN_DBI].md_root);
1079 metab.mm_txnid = env->me_metas[toggle]->mm_txnid;
1080 metab.mm_last_pg = env->me_metas[toggle]->mm_last_pg;
1082 ptr = (char *)&meta;
1083 off = offsetof(MDB_meta, mm_dbs[0].md_depth);
1084 len = sizeof(MDB_meta) - off;
1087 meta.mm_dbs[0] = txn->mt_dbs[0];
1088 meta.mm_dbs[1] = txn->mt_dbs[1];
1089 meta.mm_last_pg = txn->mt_next_pgno - 1;
1090 meta.mm_txnid = txn->mt_txnid;
1093 off += env->me_psize;
1096 /* Write to the SYNC fd */
1097 rc = pwrite(env->me_mfd, ptr, len, off);
1099 DPRINTF("write failed, disk error?");
1100 /* On a failure, the pagecache still contains the new data.
1101 * Write some old data back, to prevent it from being used.
1102 * Use the non-SYNC fd; we know it will fail anyway.
1104 meta.mm_last_pg = metab.mm_last_pg;
1105 meta.mm_txnid = metab.mm_txnid;
1106 rc = pwrite(env->me_fd, ptr, len, off);
1107 env->me_flags |= MDB_FATAL_ERROR;
1110 txn->mt_env->me_txns->mti_me_toggle = toggle;
1116 mdb_env_read_meta(MDB_env *env, int *which)
1120 assert(env != NULL);
1124 else if (env->me_metas[0]->mm_txnid < env->me_metas[1]->mm_txnid)
1127 if (env->me_meta != env->me_metas[toggle])
1128 env->me_meta = env->me_metas[toggle];
1130 DPRINTF("Using meta page %d", toggle);
1136 mdb_env_create(MDB_env **env)
1140 e = calloc(1, sizeof(MDB_env));
1141 if (!e) return ENOMEM;
1143 e->me_maxreaders = DEFAULT_READERS;
1153 mdb_env_set_mapsize(MDB_env *env, size_t size)
1157 env->me_mapsize = size;
1162 mdb_env_set_maxdbs(MDB_env *env, int dbs)
1164 env->me_maxdbs = dbs;
1169 mdb_env_set_maxreaders(MDB_env *env, int readers)
1171 env->me_maxreaders = readers;
1176 mdb_env_get_maxreaders(MDB_env *env, int *readers)
1178 if (!env || !readers)
1180 *readers = env->me_maxreaders;
1185 mdb_env_open2(MDB_env *env, unsigned int flags)
1191 env->me_flags = flags;
1193 memset(&meta, 0, sizeof(meta));
1195 if ((i = mdb_env_read_header(env, &meta)) != 0) {
1198 DPRINTF("new mdbenv");
1202 if (!env->me_mapsize) {
1203 env->me_mapsize = newenv ? DEFAULT_MAPSIZE : meta.mm_mapsize;
1207 if (meta.mm_address && (flags & MDB_FIXEDMAP))
1209 env->me_map = mmap(meta.mm_address, env->me_mapsize, PROT_READ, i,
1211 if (env->me_map == MAP_FAILED)
1215 meta.mm_mapsize = env->me_mapsize;
1216 if (flags & MDB_FIXEDMAP)
1217 meta.mm_address = env->me_map;
1218 i = mdb_env_init_meta(env, &meta);
1219 if (i != MDB_SUCCESS) {
1220 munmap(env->me_map, env->me_mapsize);
1224 env->me_psize = meta.mm_psize;
1226 env->me_maxpg = env->me_mapsize / env->me_psize;
1228 p = (MDB_page *)env->me_map;
1229 env->me_metas[0] = METADATA(p);
1230 env->me_metas[1] = (MDB_meta *)((char *)env->me_metas[0] + meta.mm_psize);
1232 if ((i = mdb_env_read_meta(env, NULL)) != 0)
1235 DPRINTF("opened database version %u, pagesize %u",
1236 env->me_meta->mm_version, env->me_psize);
1237 DPRINTF("depth: %u", env->me_meta->mm_dbs[MAIN_DBI].md_depth);
1238 DPRINTF("entries: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_entries);
1239 DPRINTF("branch pages: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_branch_pages);
1240 DPRINTF("leaf pages: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_leaf_pages);
1241 DPRINTF("overflow pages: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_overflow_pages);
1242 DPRINTF("root: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_root);
1248 mdb_env_reader_dest(void *ptr)
1250 MDB_reader *reader = ptr;
1252 reader->mr_txnid = 0;
1257 /* downgrade the exclusive lock on the region back to shared */
1259 mdb_env_share_locks(MDB_env *env)
1261 struct flock lock_info;
1263 env->me_txns->mti_txnid = env->me_meta->mm_txnid;
1264 if (env->me_metas[0]->mm_txnid < env->me_metas[1]->mm_txnid)
1265 env->me_txns->mti_me_toggle = 1;
1267 memset((void *)&lock_info, 0, sizeof(lock_info));
1268 lock_info.l_type = F_RDLCK;
1269 lock_info.l_whence = SEEK_SET;
1270 lock_info.l_start = 0;
1271 lock_info.l_len = 1;
1272 fcntl(env->me_lfd, F_SETLK, &lock_info);
1276 mdb_env_setup_locks(MDB_env *env, char *lpath, int mode, int *excl)
1280 struct flock lock_info;
1284 if ((env->me_lfd = open(lpath, O_RDWR|O_CREAT, mode)) == -1) {
1288 /* Try to get exclusive lock. If we succeed, then
1289 * nobody is using the lock region and we should initialize it.
1291 memset((void *)&lock_info, 0, sizeof(lock_info));
1292 lock_info.l_type = F_WRLCK;
1293 lock_info.l_whence = SEEK_SET;
1294 lock_info.l_start = 0;
1295 lock_info.l_len = 1;
1296 rc = fcntl(env->me_lfd, F_SETLK, &lock_info);
1300 lock_info.l_type = F_RDLCK;
1301 rc = fcntl(env->me_lfd, F_SETLK, &lock_info);
1307 size = lseek(env->me_lfd, 0, SEEK_END);
1308 rsize = (env->me_maxreaders-1) * sizeof(MDB_reader) + sizeof(MDB_txninfo);
1309 if (size < rsize && *excl) {
1310 if (ftruncate(env->me_lfd, rsize) != 0) {
1316 size = rsize - sizeof(MDB_txninfo);
1317 env->me_maxreaders = size/sizeof(MDB_reader) + 1;
1319 env->me_txns = mmap(0, rsize, PROT_READ|PROT_WRITE, MAP_SHARED,
1321 if (env->me_txns == MAP_FAILED) {
1326 pthread_mutexattr_t mattr;
1328 pthread_mutexattr_init(&mattr);
1329 rc = pthread_mutexattr_setpshared(&mattr, PTHREAD_PROCESS_SHARED);
1333 pthread_mutex_init(&env->me_txns->mti_mutex, &mattr);
1334 pthread_mutex_init(&env->me_txns->mti_wmutex, &mattr);
1335 env->me_txns->mti_version = MDB_VERSION;
1336 env->me_txns->mti_magic = MDB_MAGIC;
1337 env->me_txns->mti_txnid = 0;
1338 env->me_txns->mti_numreaders = 0;
1339 env->me_txns->mti_me_toggle = 0;
1342 if (env->me_txns->mti_magic != MDB_MAGIC) {
1343 DPRINTF("lock region has invalid magic");
1347 if (env->me_txns->mti_version != MDB_VERSION) {
1348 DPRINTF("lock region is version %u, expected version %u",
1349 env->me_txns->mti_version, MDB_VERSION);
1350 rc = MDB_VERSION_MISMATCH;
1353 if (errno != EACCES && errno != EAGAIN) {
1367 #define LOCKNAME "/lock.mdb"
1368 #define DATANAME "/data.mdb"
1370 mdb_env_open(MDB_env *env, const char *path, unsigned int flags, mode_t mode)
1372 int oflags, rc, len, excl;
1373 char *lpath, *dpath;
1376 lpath = malloc(len + sizeof(LOCKNAME) + len + sizeof(DATANAME));
1379 dpath = lpath + len + sizeof(LOCKNAME);
1380 sprintf(lpath, "%s" LOCKNAME, path);
1381 sprintf(dpath, "%s" DATANAME, path);
1383 rc = mdb_env_setup_locks(env, lpath, mode, &excl);
1387 if (F_ISSET(flags, MDB_RDONLY))
1390 oflags = O_RDWR | O_CREAT;
1392 if ((env->me_fd = open(dpath, oflags, mode)) == -1) {
1397 if ((rc = mdb_env_open2(env, flags)) == MDB_SUCCESS) {
1398 /* synchronous fd for meta writes */
1399 if (!(flags & (MDB_RDONLY|MDB_NOSYNC)))
1401 if ((env->me_mfd = open(dpath, oflags, mode)) == -1) {
1406 env->me_path = strdup(path);
1407 DPRINTF("opened dbenv %p", (void *) env);
1408 pthread_key_create(&env->me_txkey, mdb_env_reader_dest);
1410 mdb_env_share_locks(env);
1411 env->me_dbxs = calloc(env->me_maxdbs, sizeof(MDB_dbx));
1412 env->me_dbs[0] = calloc(env->me_maxdbs, sizeof(MDB_db));
1413 env->me_dbs[1] = calloc(env->me_maxdbs, sizeof(MDB_db));
1419 if (env->me_fd >= 0) {
1423 if (env->me_lfd >= 0) {
1433 mdb_env_close(MDB_env *env)
1438 free(env->me_dbs[1]);
1439 free(env->me_dbs[0]);
1443 pthread_key_delete(env->me_txkey);
1446 munmap(env->me_map, env->me_mapsize);
1451 pid_t pid = getpid();
1452 size_t size = (env->me_maxreaders-1) * sizeof(MDB_reader) + sizeof(MDB_txninfo);
1454 for (i=0; i<env->me_txns->mti_numreaders; i++)
1455 if (env->me_txns->mti_readers[i].mr_pid == pid)
1456 env->me_txns->mti_readers[i].mr_pid = 0;
1457 munmap(env->me_txns, size);
1463 /* Search for key within a leaf page, using binary search.
1464 * Returns the smallest entry larger or equal to the key.
1465 * If exactp is non-null, stores whether the found entry was an exact match
1466 * in *exactp (1 or 0).
1467 * If kip is non-null, stores the index of the found entry in *kip.
1468 * If no entry larger or equal to the key is found, returns NULL.
1471 mdb_search_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp, MDB_val *key,
1472 int *exactp, unsigned int *kip)
1480 DPRINTF("searching %u keys in %s page %lu",
1482 IS_LEAF(mp) ? "leaf" : "branch",
1485 assert(NUMKEYS(mp) > 0);
1487 memset(&nodekey, 0, sizeof(nodekey));
1489 low = IS_LEAF(mp) ? 0 : 1;
1490 high = NUMKEYS(mp) - 1;
1491 while (low <= high) {
1492 i = (low + high) >> 1;
1493 node = NODEPTR(mp, i);
1495 nodekey.mv_size = node->mn_ksize;
1496 nodekey.mv_data = NODEKEY(node);
1498 if (txn->mt_dbxs[dbi].md_cmp)
1499 rc = txn->mt_dbxs[dbi].md_cmp(key, &nodekey);
1501 rc = _mdb_cmp(txn, dbi, key, &nodekey);
1504 DPRINTF("found leaf index %u [%.*s], rc = %i",
1505 i, (int)nodekey.mv_size, (char *)nodekey.mv_data, rc);
1507 DPRINTF("found branch index %u [%.*s -> %lu], rc = %i",
1508 i, (int)node->mn_ksize, (char *)NODEKEY(node),
1519 if (rc > 0) { /* Found entry is less than the key. */
1520 i++; /* Skip to get the smallest entry larger than key. */
1521 if (i >= NUMKEYS(mp))
1522 /* There is no entry larger or equal to the key. */
1526 *exactp = (rc == 0);
1527 if (kip) /* Store the key index if requested. */
1530 return NODEPTR(mp, i);
1534 cursor_pop_page(MDB_cursor *cursor)
1538 if (cursor->mc_snum) {
1539 top = CURSOR_TOP(cursor);
1542 DPRINTF("popped page %lu off cursor %p", top->mp_page->mp_pgno, (void *) cursor);
1547 cursor_push_page(MDB_cursor *cursor, MDB_page *mp)
1551 DPRINTF("pushing page %lu on cursor %p", mp->mp_pgno, (void *) cursor);
1553 ppage = &cursor->mc_stack[cursor->mc_snum++];
1554 ppage->mp_page = mp;
1560 mdb_get_page(MDB_txn *txn, pgno_t pgno)
1565 if (!F_ISSET(txn->mt_flags, MDB_TXN_RDONLY) && txn->mt_u.dirty_list[0].mid) {
1570 x = mdb_midl2_search(txn->mt_u.dirty_list, &id);
1571 if (x <= txn->mt_u.dirty_list[0].mid && txn->mt_u.dirty_list[x].mid == pgno) {
1572 dp = txn->mt_u.dirty_list[x].mptr;
1578 if (pgno > txn->mt_env->me_meta->mm_last_pg)
1580 p = (MDB_page *)(txn->mt_env->me_map + txn->mt_env->me_psize * pgno);
1586 mdb_search_page_root(MDB_txn *txn, MDB_dbi dbi, MDB_val *key,
1587 MDB_cursor *cursor, int modify, MDB_pageparent *mpp)
1589 MDB_page *mp = mpp->mp_page;
1592 if (cursor && cursor_push_page(cursor, mp) == NULL)
1595 while (IS_BRANCH(mp)) {
1599 DPRINTF("branch page %lu has %u keys", mp->mp_pgno, NUMKEYS(mp));
1600 assert(NUMKEYS(mp) > 1);
1601 DPRINTF("found index 0 to page %lu", NODEPGNO(NODEPTR(mp, 0)));
1603 if (key == NULL) /* Initialize cursor to first page. */
1605 else if (key->mv_size > MAXKEYSIZE && key->mv_data == NULL) {
1606 /* cursor to last page */
1610 node = mdb_search_node(txn, dbi, mp, key, &exact, &i);
1612 i = NUMKEYS(mp) - 1;
1620 DPRINTF("following index %u for key %.*s",
1621 i, (int)key->mv_size, (char *)key->mv_data);
1622 assert(i < NUMKEYS(mp));
1623 node = NODEPTR(mp, i);
1626 CURSOR_TOP(cursor)->mp_ki = i;
1628 mpp->mp_parent = mp;
1629 if ((mp = mdb_get_page(txn, NODEPGNO(node))) == NULL)
1630 return MDB_PAGE_NOTFOUND;
1634 if (cursor && cursor_push_page(cursor, mp) == NULL)
1638 MDB_dhead *dh = ((MDB_dhead *)mp)-1;
1639 if ((rc = mdb_touch(txn, mpp)) != 0)
1641 dh = ((MDB_dhead *)mpp->mp_page)-1;
1642 dh->md_parent = mpp->mp_parent;
1643 dh->md_pi = mpp->mp_pi;
1650 DPRINTF("internal error, index points to a %02X page!?",
1652 return MDB_CORRUPTED;
1655 DPRINTF("found leaf page %lu for key %.*s", mp->mp_pgno,
1656 key ? (int)key->mv_size : 0, key ? (char *)key->mv_data : NULL);
1661 /* Search for the page a given key should be in.
1662 * Stores a pointer to the found page in *mpp.
1663 * If key is NULL, search for the lowest page (used by mdb_cursor_first).
1664 * If cursor is non-null, pushes parent pages on the cursor stack.
1665 * If modify is true, visited pages are updated with new page numbers.
1668 mdb_search_page(MDB_txn *txn, MDB_dbi dbi, MDB_val *key,
1669 MDB_cursor *cursor, int modify, MDB_pageparent *mpp)
1674 /* Choose which root page to start with. If a transaction is given
1675 * use the root page from the transaction, otherwise read the last
1676 * committed root page.
1678 if (F_ISSET(txn->mt_flags, MDB_TXN_ERROR)) {
1679 DPRINTF("transaction has failed, must abort");
1682 root = txn->mt_dbs[dbi].md_root;
1684 if (root == P_INVALID) { /* Tree is empty. */
1685 DPRINTF("tree is empty");
1686 return MDB_NOTFOUND;
1689 if ((mpp->mp_page = mdb_get_page(txn, root)) == NULL)
1690 return MDB_PAGE_NOTFOUND;
1692 DPRINTF("root page has flags 0x%X", mpp->mp_page->mp_flags);
1695 /* For sub-databases, update main root first */
1696 if (dbi > MAIN_DBI && !txn->mt_dbxs[dbi].md_dirty) {
1698 rc = mdb_search_page(txn, MAIN_DBI, &txn->mt_dbxs[dbi].md_name,
1702 txn->mt_dbxs[dbi].md_dirty = 1;
1704 if (!F_ISSET(mpp->mp_page->mp_flags, P_DIRTY)) {
1705 mpp->mp_parent = NULL;
1707 if ((rc = mdb_touch(txn, mpp)))
1709 txn->mt_dbs[dbi].md_root = mpp->mp_page->mp_pgno;
1713 return mdb_search_page_root(txn, dbi, key, cursor, modify, mpp);
1717 mdb_read_data(MDB_txn *txn, MDB_node *leaf, MDB_val *data)
1719 MDB_page *omp; /* overflow mpage */
1722 if (!F_ISSET(leaf->mn_flags, F_BIGDATA)) {
1723 data->mv_size = leaf->mn_dsize;
1724 data->mv_data = NODEDATA(leaf);
1728 /* Read overflow data.
1730 data->mv_size = leaf->mn_dsize;
1731 memcpy(&pgno, NODEDATA(leaf), sizeof(pgno));
1732 if ((omp = mdb_get_page(txn, pgno)) == NULL) {
1733 DPRINTF("read overflow page %lu failed", pgno);
1734 return MDB_PAGE_NOTFOUND;
1736 data->mv_data = METADATA(omp);
1742 mdb_get(MDB_txn *txn, MDB_dbi dbi,
1743 MDB_val *key, MDB_val *data)
1751 DPRINTF("===> get key [%.*s]", (int)key->mv_size, (char *)key->mv_data);
1753 if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
1756 if (key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
1760 if ((rc = mdb_search_page(txn, dbi, key, NULL, 0, &mpp)) != MDB_SUCCESS)
1763 leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, NULL);
1764 if (leaf && exact) {
1765 /* Return first duplicate data item */
1766 if (F_ISSET(txn->mt_dbs[dbi].md_flags, MDB_DUPSORT)) {
1769 mdb_xcursor_init0(txn, dbi, &mx);
1770 mdb_xcursor_init1(txn, dbi, &mx, leaf);
1771 rc = mdb_search_page(&mx.mx_txn, mx.mx_cursor.mc_dbi, NULL, NULL, 0, &mpp);
1772 if (rc != MDB_SUCCESS)
1774 leaf = NODEPTR(mpp.mp_page, 0);
1776 rc = mdb_read_data(txn, leaf, data);
1785 mdb_sibling(MDB_cursor *cursor, int move_right)
1792 if (cursor->mc_snum < 2) {
1793 return MDB_NOTFOUND; /* root has no siblings */
1795 parent = CURSOR_PARENT(cursor);
1797 DPRINTF("parent page is page %lu, index %u",
1798 parent->mp_page->mp_pgno, parent->mp_ki);
1800 cursor_pop_page(cursor);
1801 if (move_right ? (parent->mp_ki + 1 >= NUMKEYS(parent->mp_page))
1802 : (parent->mp_ki == 0)) {
1803 DPRINTF("no more keys left, moving to %s sibling",
1804 move_right ? "right" : "left");
1805 if ((rc = mdb_sibling(cursor, move_right)) != MDB_SUCCESS)
1807 parent = CURSOR_TOP(cursor);
1813 DPRINTF("just moving to %s index key %u",
1814 move_right ? "right" : "left", parent->mp_ki);
1816 assert(IS_BRANCH(parent->mp_page));
1818 indx = NODEPTR(parent->mp_page, parent->mp_ki);
1819 if ((mp = mdb_get_page(cursor->mc_txn, indx->mn_pgno)) == NULL)
1820 return MDB_PAGE_NOTFOUND;
1822 mp->parent = parent->mp_page;
1823 mp->parent_index = parent->mp_ki;
1826 cursor_push_page(cursor, mp);
1832 mdb_set_key(MDB_node *node, MDB_val *key)
1837 key->mv_size = node->mn_ksize;
1838 key->mv_data = NODEKEY(node);
1844 mdb_cursor_next(MDB_cursor *cursor, MDB_val *key, MDB_val *data, MDB_cursor_op op)
1851 if (cursor->mc_eof) {
1852 return MDB_NOTFOUND;
1855 assert(cursor->mc_initialized);
1857 if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
1858 if (op == MDB_NEXT || op == MDB_NEXT_DUP) {
1859 rc = mdb_cursor_next(&cursor->mc_xcursor->mx_cursor, data, NULL, MDB_NEXT);
1860 if (op != MDB_NEXT || rc == MDB_SUCCESS)
1865 top = CURSOR_TOP(cursor);
1868 DPRINTF("cursor_next: top page is %lu in cursor %p", mp->mp_pgno, (void *) cursor);
1870 if (top->mp_ki + 1 >= NUMKEYS(mp)) {
1871 DPRINTF("=====> move to next sibling page");
1872 if (mdb_sibling(cursor, 1) != MDB_SUCCESS) {
1874 return MDB_NOTFOUND;
1876 top = CURSOR_TOP(cursor);
1878 DPRINTF("next page is %lu, key index %u", mp->mp_pgno, top->mp_ki);
1882 DPRINTF("==> cursor points to page %lu with %u keys, key index %u",
1883 mp->mp_pgno, NUMKEYS(mp), top->mp_ki);
1885 assert(IS_LEAF(mp));
1886 leaf = NODEPTR(mp, top->mp_ki);
1889 if ((rc = mdb_read_data(cursor->mc_txn, leaf, data) != MDB_SUCCESS))
1892 if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
1893 mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, leaf);
1894 rc = mdb_cursor_first(&cursor->mc_xcursor->mx_cursor, data, NULL);
1895 if (rc != MDB_SUCCESS)
1900 return mdb_set_key(leaf, key);
1904 mdb_cursor_prev(MDB_cursor *cursor, MDB_val *key, MDB_val *data, MDB_cursor_op op)
1911 assert(cursor->mc_initialized);
1913 if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
1914 if (op == MDB_PREV || op == MDB_PREV_DUP) {
1915 rc = mdb_cursor_prev(&cursor->mc_xcursor->mx_cursor, data, NULL, MDB_PREV);
1916 if (op != MDB_PREV || rc == MDB_SUCCESS)
1921 top = CURSOR_TOP(cursor);
1924 DPRINTF("cursor_prev: top page is %lu in cursor %p", mp->mp_pgno, (void *) cursor);
1926 if (top->mp_ki == 0) {
1927 DPRINTF("=====> move to prev sibling page");
1928 if (mdb_sibling(cursor, 0) != MDB_SUCCESS) {
1929 return MDB_NOTFOUND;
1931 top = CURSOR_TOP(cursor);
1933 top->mp_ki = NUMKEYS(mp) - 1;
1934 DPRINTF("prev page is %lu, key index %u", mp->mp_pgno, top->mp_ki);
1940 DPRINTF("==> cursor points to page %lu with %u keys, key index %u",
1941 mp->mp_pgno, NUMKEYS(mp), top->mp_ki);
1943 assert(IS_LEAF(mp));
1944 leaf = NODEPTR(mp, top->mp_ki);
1947 if ((rc = mdb_read_data(cursor->mc_txn, leaf, data) != MDB_SUCCESS))
1950 if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
1951 mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, leaf);
1952 rc = mdb_cursor_last(&cursor->mc_xcursor->mx_cursor, data, NULL);
1953 if (rc != MDB_SUCCESS)
1958 return mdb_set_key(leaf, key);
1962 mdb_cursor_set(MDB_cursor *cursor, MDB_val *key, MDB_val *data,
1963 MDB_cursor_op op, int *exactp)
1972 assert(key->mv_size > 0);
1974 cursor->mc_snum = 0;
1976 rc = mdb_search_page(cursor->mc_txn, cursor->mc_dbi, key, cursor, 0, &mpp);
1977 if (rc != MDB_SUCCESS)
1979 assert(IS_LEAF(mpp.mp_page));
1981 top = CURSOR_TOP(cursor);
1982 leaf = mdb_search_node(cursor->mc_txn, cursor->mc_dbi, mpp.mp_page, key, exactp, &top->mp_ki);
1983 if (exactp != NULL && !*exactp) {
1984 /* MDB_SET specified and not an exact match. */
1985 return MDB_NOTFOUND;
1989 DPRINTF("===> inexact leaf not found, goto sibling");
1990 if ((rc = mdb_sibling(cursor, 1)) != MDB_SUCCESS)
1991 return rc; /* no entries matched */
1992 top = CURSOR_TOP(cursor);
1994 mpp.mp_page = top->mp_page;
1995 assert(IS_LEAF(mpp.mp_page));
1996 leaf = NODEPTR(mpp.mp_page, 0);
1999 cursor->mc_initialized = 1;
2004 if ((rc = mdb_read_data(cursor->mc_txn, leaf, &d2)) != MDB_SUCCESS)
2007 if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
2008 mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, leaf);
2009 if (op == MDB_SET || op == MDB_SET_RANGE) {
2010 rc = mdb_cursor_first(&cursor->mc_xcursor->mx_cursor, data, NULL);
2014 if (op == MDB_GET_BOTH) {
2019 op2 = MDB_SET_RANGE;
2021 rc = mdb_cursor_set(&cursor->mc_xcursor->mx_cursor, data, NULL, op2, ex2p);
2022 if (rc != MDB_SUCCESS)
2031 rc = mdb_set_key(leaf, key);
2032 if (rc == MDB_SUCCESS) {
2033 DPRINTF("==> cursor placed on key %.*s",
2034 (int)key->mv_size, (char *)key->mv_data);
2042 mdb_cursor_first(MDB_cursor *cursor, MDB_val *key, MDB_val *data)
2048 cursor->mc_snum = 0;
2050 rc = mdb_search_page(cursor->mc_txn, cursor->mc_dbi, NULL, cursor, 0, &mpp);
2051 if (rc != MDB_SUCCESS)
2053 assert(IS_LEAF(mpp.mp_page));
2055 leaf = NODEPTR(mpp.mp_page, 0);
2056 cursor->mc_initialized = 1;
2060 if ((rc = mdb_read_data(cursor->mc_txn, leaf, data)) != MDB_SUCCESS)
2063 if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
2064 mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, leaf);
2065 rc = mdb_cursor_first(&cursor->mc_xcursor->mx_cursor, data, NULL);
2070 return mdb_set_key(leaf, key);
2074 mdb_cursor_last(MDB_cursor *cursor, MDB_val *key, MDB_val *data)
2082 cursor->mc_snum = 0;
2084 lkey.mv_size = MAXKEYSIZE+1;
2085 lkey.mv_data = NULL;
2087 rc = mdb_search_page(cursor->mc_txn, cursor->mc_dbi, &lkey, cursor, 0, &mpp);
2088 if (rc != MDB_SUCCESS)
2090 assert(IS_LEAF(mpp.mp_page));
2092 leaf = NODEPTR(mpp.mp_page, NUMKEYS(mpp.mp_page)-1);
2093 cursor->mc_initialized = 1;
2096 top = CURSOR_TOP(cursor);
2097 top->mp_ki = NUMKEYS(top->mp_page) - 1;
2100 if ((rc = mdb_read_data(cursor->mc_txn, leaf, data)) != MDB_SUCCESS)
2103 if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
2104 mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, leaf);
2105 rc = mdb_cursor_last(&cursor->mc_xcursor->mx_cursor, data, NULL);
2111 return mdb_set_key(leaf, key);
2115 mdb_cursor_get(MDB_cursor *cursor, MDB_val *key, MDB_val *data,
2125 case MDB_GET_BOTH_RANGE:
2133 if (key == NULL || key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
2135 } else if (op != MDB_SET_RANGE)
2136 rc = mdb_cursor_set(cursor, key, data, op, NULL);
2138 rc = mdb_cursor_set(cursor, key, data, op, &exact);
2142 case MDB_NEXT_NODUP:
2143 if (!cursor->mc_initialized)
2144 rc = mdb_cursor_first(cursor, key, data);
2146 rc = mdb_cursor_next(cursor, key, data, op);
2150 case MDB_PREV_NODUP:
2151 if (!cursor->mc_initialized || cursor->mc_eof)
2152 rc = mdb_cursor_last(cursor, key, data);
2154 rc = mdb_cursor_prev(cursor, key, data, op);
2157 rc = mdb_cursor_first(cursor, key, data);
2160 rc = mdb_cursor_last(cursor, key, data);
2163 DPRINTF("unhandled/unimplemented cursor operation %u", op);
2171 /* Allocate a page and initialize it
2174 mdb_new_page(MDB_txn *txn, MDB_dbi dbi, uint32_t flags, int num)
2178 if ((dp = mdb_alloc_page(txn, NULL, 0, num)) == NULL)
2180 DPRINTF("allocated new mpage %lu, page size %u",
2181 dp->p.mp_pgno, txn->mt_env->me_psize);
2182 dp->p.mp_flags = flags | P_DIRTY;
2183 dp->p.mp_lower = PAGEHDRSZ;
2184 dp->p.mp_upper = txn->mt_env->me_psize;
2186 if (IS_BRANCH(&dp->p))
2187 txn->mt_dbs[dbi].md_branch_pages++;
2188 else if (IS_LEAF(&dp->p))
2189 txn->mt_dbs[dbi].md_leaf_pages++;
2190 else if (IS_OVERFLOW(&dp->p)) {
2191 txn->mt_dbs[dbi].md_overflow_pages += num;
2192 dp->p.mp_pages = num;
2199 mdb_leaf_size(MDB_env *env, MDB_val *key, MDB_val *data)
2203 sz = LEAFSIZE(key, data);
2204 if (data->mv_size >= env->me_psize / MDB_MINKEYS) {
2205 /* put on overflow page */
2206 sz -= data->mv_size - sizeof(pgno_t);
2209 return sz + sizeof(indx_t);
2213 mdb_branch_size(MDB_env *env, MDB_val *key)
2218 if (sz >= env->me_psize / MDB_MINKEYS) {
2219 /* put on overflow page */
2220 /* not implemented */
2221 /* sz -= key->size - sizeof(pgno_t); */
2224 return sz + sizeof(indx_t);
2228 mdb_add_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp, indx_t indx,
2229 MDB_val *key, MDB_val *data, pgno_t pgno, uint8_t flags)
2232 size_t node_size = NODESIZE;
2235 MDB_dpage *ofp = NULL; /* overflow page */
2237 assert(mp->mp_upper >= mp->mp_lower);
2239 DPRINTF("add node [%.*s] to %s page %lu at index %i, key size %zu",
2240 key ? (int)key->mv_size : 0, key ? (char *)key->mv_data : NULL,
2241 IS_LEAF(mp) ? "leaf" : "branch",
2242 mp->mp_pgno, indx, key ? key->mv_size : 0);
2245 node_size += key->mv_size;
2249 if (F_ISSET(flags, F_BIGDATA)) {
2250 /* Data already on overflow page. */
2251 node_size += sizeof(pgno_t);
2252 } else if (data->mv_size >= txn->mt_env->me_psize / MDB_MINKEYS) {
2253 int ovpages = OVPAGES(data->mv_size, txn->mt_env->me_psize);
2254 /* Put data on overflow page. */
2255 DPRINTF("data size is %zu, put on overflow page",
2257 node_size += sizeof(pgno_t);
2258 if ((ofp = mdb_new_page(txn, dbi, P_OVERFLOW, ovpages)) == NULL)
2260 DPRINTF("allocated overflow page %lu", ofp->p.mp_pgno);
2263 node_size += data->mv_size;
2267 if (node_size + sizeof(indx_t) > SIZELEFT(mp)) {
2268 DPRINTF("not enough room in page %lu, got %u ptrs",
2269 mp->mp_pgno, NUMKEYS(mp));
2270 DPRINTF("upper - lower = %u - %u = %u", mp->mp_upper, mp->mp_lower,
2271 mp->mp_upper - mp->mp_lower);
2272 DPRINTF("node size = %zu", node_size);
2276 /* Move higher pointers up one slot. */
2277 for (i = NUMKEYS(mp); i > indx; i--)
2278 mp->mp_ptrs[i] = mp->mp_ptrs[i - 1];
2280 /* Adjust free space offsets. */
2281 ofs = mp->mp_upper - node_size;
2282 assert(ofs >= mp->mp_lower + sizeof(indx_t));
2283 mp->mp_ptrs[indx] = ofs;
2285 mp->mp_lower += sizeof(indx_t);
2287 /* Write the node data. */
2288 node = NODEPTR(mp, indx);
2289 node->mn_ksize = (key == NULL) ? 0 : key->mv_size;
2290 node->mn_flags = flags;
2292 node->mn_dsize = data->mv_size;
2294 node->mn_pgno = pgno;
2297 memcpy(NODEKEY(node), key->mv_data, key->mv_size);
2302 if (F_ISSET(flags, F_BIGDATA))
2303 memcpy(node->mn_data + key->mv_size, data->mv_data,
2306 memcpy(node->mn_data + key->mv_size, data->mv_data,
2309 memcpy(node->mn_data + key->mv_size, &ofp->p.mp_pgno,
2311 memcpy(METADATA(&ofp->p), data->mv_data, data->mv_size);
2319 mdb_del_node(MDB_page *mp, indx_t indx)
2322 indx_t i, j, numkeys, ptr;
2326 DPRINTF("delete node %u on %s page %lu", indx,
2327 IS_LEAF(mp) ? "leaf" : "branch", mp->mp_pgno);
2328 assert(indx < NUMKEYS(mp));
2330 node = NODEPTR(mp, indx);
2331 sz = NODESIZE + node->mn_ksize;
2333 if (F_ISSET(node->mn_flags, F_BIGDATA))
2334 sz += sizeof(pgno_t);
2336 sz += NODEDSZ(node);
2339 ptr = mp->mp_ptrs[indx];
2340 numkeys = NUMKEYS(mp);
2341 for (i = j = 0; i < numkeys; i++) {
2343 mp->mp_ptrs[j] = mp->mp_ptrs[i];
2344 if (mp->mp_ptrs[i] < ptr)
2345 mp->mp_ptrs[j] += sz;
2350 base = (char *)mp + mp->mp_upper;
2351 memmove(base + sz, base, ptr - mp->mp_upper);
2353 mp->mp_lower -= sizeof(indx_t);
2358 mdb_xcursor_init0(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx)
2363 mx->mx_txn.mt_dbxs = mx->mx_dbxs;
2364 mx->mx_txn.mt_dbs = mx->mx_dbs;
2365 mx->mx_dbxs[0] = txn->mt_dbxs[0];
2366 mx->mx_dbxs[1] = txn->mt_dbxs[1];
2368 mx->mx_dbxs[2] = txn->mt_dbxs[dbi];
2373 mx->mx_dbxs[dbn+1].md_parent = dbn;
2374 mx->mx_dbxs[dbn+1].md_cmp = mx->mx_dbxs[dbn].md_dcmp;
2375 mx->mx_dbxs[dbn+1].md_rel = mx->mx_dbxs[dbn].md_rel;
2376 mx->mx_dbxs[dbn+1].md_dirty = 0;
2377 mx->mx_txn.mt_numdbs = dbn+2;
2379 mx->mx_cursor.mc_snum = 0;
2380 mx->mx_cursor.mc_txn = &mx->mx_txn;
2381 mx->mx_cursor.mc_dbi = dbn+1;
2385 mdb_xcursor_init1(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx, MDB_node *node)
2387 MDB_db *db = NODEDATA(node);
2389 mx->mx_dbs[0] = txn->mt_dbs[0];
2390 mx->mx_dbs[1] = txn->mt_dbs[1];
2392 mx->mx_dbs[2] = txn->mt_dbs[dbi];
2397 mx->mx_dbs[dbn] = *db;
2398 mx->mx_dbxs[dbn].md_name.mv_data = NODEKEY(node);
2399 mx->mx_dbxs[dbn].md_name.mv_size = node->mn_ksize;
2400 mx->mx_txn.mt_next_pgno = txn->mt_next_pgno;
2401 mx->mx_txn.mt_oldest = txn->mt_oldest;
2402 mx->mx_txn.mt_u = txn->mt_u;
2406 mdb_xcursor_fini(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx)
2408 txn->mt_next_pgno = mx->mx_txn.mt_next_pgno;
2409 txn->mt_oldest = mx->mx_txn.mt_oldest;
2410 txn->mt_u = mx->mx_txn.mt_u;
2411 txn->mt_dbs[0] = mx->mx_dbs[0];
2412 txn->mt_dbs[1] = mx->mx_dbs[1];
2413 txn->mt_dbxs[0].md_dirty = mx->mx_dbxs[0].md_dirty;
2414 txn->mt_dbxs[1].md_dirty = mx->mx_dbxs[1].md_dirty;
2416 txn->mt_dbs[dbi] = mx->mx_dbs[2];
2417 txn->mt_dbxs[dbi].md_dirty = mx->mx_dbxs[2].md_dirty;
2422 mdb_cursor_open(MDB_txn *txn, MDB_dbi dbi, MDB_cursor **ret)
2425 size_t size = sizeof(MDB_cursor);
2427 if (txn == NULL || ret == NULL || !dbi || dbi >= txn->mt_numdbs)
2430 if (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT)
2431 size += sizeof(MDB_xcursor);
2433 if ((cursor = calloc(1, size)) != NULL) {
2434 cursor->mc_dbi = dbi;
2435 cursor->mc_txn = txn;
2436 if (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT) {
2437 MDB_xcursor *mx = (MDB_xcursor *)(cursor + 1);
2438 cursor->mc_xcursor = mx;
2439 mdb_xcursor_init0(txn, dbi, mx);
2450 /* Return the count of duplicate data items for the current key */
2452 mdb_cursor_count(MDB_cursor *mc, unsigned long *countp)
2454 if (mc == NULL || countp == NULL)
2457 if (!(mc->mc_txn->mt_dbs[mc->mc_dbi].md_flags & MDB_DUPSORT))
2460 if (!mc->mc_xcursor->mx_cursor.mc_initialized)
2463 *countp = mc->mc_xcursor->mx_txn.mt_dbs[mc->mc_xcursor->mx_cursor.mc_dbi].md_entries;
2468 mdb_cursor_close(MDB_cursor *cursor)
2470 if (cursor != NULL) {
2476 mdb_update_key(MDB_page *mp, indx_t indx, MDB_val *key)
2478 indx_t ptr, i, numkeys;
2484 node = NODEPTR(mp, indx);
2485 ptr = mp->mp_ptrs[indx];
2486 DPRINTF("update key %u (ofs %u) [%.*s] to [%.*s] on page %lu",
2488 (int)node->mn_ksize, (char *)NODEKEY(node),
2489 (int)key->mv_size, (char *)key->mv_data,
2492 delta = key->mv_size - node->mn_ksize;
2494 if (delta > 0 && SIZELEFT(mp) < delta) {
2495 DPRINTF("OUCH! Not enough room, delta = %d", delta);
2499 numkeys = NUMKEYS(mp);
2500 for (i = 0; i < numkeys; i++) {
2501 if (mp->mp_ptrs[i] <= ptr)
2502 mp->mp_ptrs[i] -= delta;
2505 base = (char *)mp + mp->mp_upper;
2506 len = ptr - mp->mp_upper + NODESIZE;
2507 memmove(base - delta, base, len);
2508 mp->mp_upper -= delta;
2510 node = NODEPTR(mp, indx);
2511 node->mn_ksize = key->mv_size;
2514 memcpy(NODEKEY(node), key->mv_data, key->mv_size);
2519 /* Move a node from src to dst.
2522 mdb_move_node(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *src, indx_t srcindx,
2523 MDB_pageparent *dst, indx_t dstindx)
2529 srcnode = NODEPTR(src->mp_page, srcindx);
2530 DPRINTF("moving %s node %u [%.*s] on page %lu to node %u on page %lu",
2531 IS_LEAF(src->mp_page) ? "leaf" : "branch",
2533 (int)srcnode->mn_ksize, (char *)NODEKEY(srcnode),
2534 src->mp_page->mp_pgno,
2535 dstindx, dst->mp_page->mp_pgno);
2537 /* Mark src and dst as dirty. */
2538 if ((rc = mdb_touch(txn, src)) ||
2539 (rc = mdb_touch(txn, dst)))
2542 /* Add the node to the destination page.
2544 key.mv_size = srcnode->mn_ksize;
2545 key.mv_data = NODEKEY(srcnode);
2546 data.mv_size = NODEDSZ(srcnode);
2547 data.mv_data = NODEDATA(srcnode);
2548 rc = mdb_add_node(txn, dbi, dst->mp_page, dstindx, &key, &data, NODEPGNO(srcnode),
2550 if (rc != MDB_SUCCESS)
2553 /* Delete the node from the source page.
2555 mdb_del_node(src->mp_page, srcindx);
2557 /* Update the parent separators.
2559 if (srcindx == 0 && src->mp_pi != 0) {
2560 DPRINTF("update separator for source page %lu to [%.*s]",
2561 src->mp_page->mp_pgno, (int)key.mv_size, (char *)key.mv_data);
2562 if ((rc = mdb_update_key(src->mp_parent, src->mp_pi,
2563 &key)) != MDB_SUCCESS)
2567 if (srcindx == 0 && IS_BRANCH(src->mp_page)) {
2569 nullkey.mv_size = 0;
2570 assert(mdb_update_key(src->mp_page, 0, &nullkey) == MDB_SUCCESS);
2573 if (dstindx == 0 && dst->mp_pi != 0) {
2574 DPRINTF("update separator for destination page %lu to [%.*s]",
2575 dst->mp_page->mp_pgno, (int)key.mv_size, (char *)key.mv_data);
2576 if ((rc = mdb_update_key(dst->mp_parent, dst->mp_pi,
2577 &key)) != MDB_SUCCESS)
2581 if (dstindx == 0 && IS_BRANCH(dst->mp_page)) {
2583 nullkey.mv_size = 0;
2584 assert(mdb_update_key(dst->mp_page, 0, &nullkey) == MDB_SUCCESS);
2591 mdb_merge(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *src, MDB_pageparent *dst)
2600 DPRINTF("merging page %lu and %lu", src->mp_page->mp_pgno, dst->mp_page->mp_pgno);
2602 assert(txn != NULL);
2603 assert(src->mp_parent); /* can't merge root page */
2604 assert(dst->mp_parent);
2606 /* Mark src and dst as dirty. */
2607 if ((rc = mdb_touch(txn, src)) ||
2608 (rc = mdb_touch(txn, dst)))
2611 /* Move all nodes from src to dst.
2613 for (i = 0; i < NUMKEYS(src->mp_page); i++) {
2614 srcnode = NODEPTR(src->mp_page, i);
2616 key.mv_size = srcnode->mn_ksize;
2617 key.mv_data = NODEKEY(srcnode);
2618 data.mv_size = NODEDSZ(srcnode);
2619 data.mv_data = NODEDATA(srcnode);
2620 rc = mdb_add_node(txn, dbi, dst->mp_page, NUMKEYS(dst->mp_page), &key,
2621 &data, NODEPGNO(srcnode), srcnode->mn_flags);
2622 if (rc != MDB_SUCCESS)
2626 DPRINTF("dst page %lu now has %u keys (%.1f%% filled)",
2627 dst->mp_page->mp_pgno, NUMKEYS(dst->mp_page), (float)PAGEFILL(txn->mt_env, dst->mp_page) / 10);
2629 /* Unlink the src page from parent.
2631 mdb_del_node(src->mp_parent, src->mp_pi);
2632 if (src->mp_pi == 0) {
2634 if ((rc = mdb_update_key(src->mp_parent, 0, &key)) != MDB_SUCCESS)
2638 if (IS_LEAF(src->mp_page))
2639 txn->mt_dbs[dbi].md_leaf_pages--;
2641 txn->mt_dbs[dbi].md_branch_pages--;
2643 mpp.mp_page = src->mp_parent;
2644 dh = (MDB_dhead *)src->mp_parent;
2646 mpp.mp_parent = dh->md_parent;
2647 mpp.mp_pi = dh->md_pi;
2649 return mdb_rebalance(txn, dbi, &mpp);
2652 #define FILL_THRESHOLD 250
2655 mdb_rebalance(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *mpp)
2660 indx_t si = 0, di = 0;
2662 assert(txn != NULL);
2663 assert(mpp != NULL);
2665 DPRINTF("rebalancing %s page %lu (has %u keys, %.1f%% full)",
2666 IS_LEAF(mpp->mp_page) ? "leaf" : "branch",
2667 mpp->mp_page->mp_pgno, NUMKEYS(mpp->mp_page), (float)PAGEFILL(txn->mt_env, mpp->mp_page) / 10);
2669 if (PAGEFILL(txn->mt_env, mpp->mp_page) >= FILL_THRESHOLD) {
2670 DPRINTF("no need to rebalance page %lu, above fill threshold",
2671 mpp->mp_page->mp_pgno);
2675 if (mpp->mp_parent == NULL) {
2676 if (NUMKEYS(mpp->mp_page) == 0) {
2677 DPRINTF("tree is completely empty");
2678 txn->mt_dbs[dbi].md_root = P_INVALID;
2679 txn->mt_dbs[dbi].md_depth--;
2680 txn->mt_dbs[dbi].md_leaf_pages--;
2681 } else if (IS_BRANCH(mpp->mp_page) && NUMKEYS(mpp->mp_page) == 1) {
2682 DPRINTF("collapsing root page!");
2683 txn->mt_dbs[dbi].md_root = NODEPGNO(NODEPTR(mpp->mp_page, 0));
2684 if ((root = mdb_get_page(txn, txn->mt_dbs[dbi].md_root)) == NULL)
2685 return MDB_PAGE_NOTFOUND;
2686 txn->mt_dbs[dbi].md_depth--;
2687 txn->mt_dbs[dbi].md_branch_pages--;
2689 DPRINTF("root page doesn't need rebalancing");
2693 /* The parent (branch page) must have at least 2 pointers,
2694 * otherwise the tree is invalid.
2696 assert(NUMKEYS(mpp->mp_parent) > 1);
2698 /* Leaf page fill factor is below the threshold.
2699 * Try to move keys from left or right neighbor, or
2700 * merge with a neighbor page.
2705 if (mpp->mp_pi == 0) {
2706 /* We're the leftmost leaf in our parent.
2708 DPRINTF("reading right neighbor");
2709 node = NODEPTR(mpp->mp_parent, mpp->mp_pi + 1);
2710 if ((npp.mp_page = mdb_get_page(txn, NODEPGNO(node))) == NULL)
2711 return MDB_PAGE_NOTFOUND;
2712 npp.mp_pi = mpp->mp_pi + 1;
2714 di = NUMKEYS(mpp->mp_page);
2716 /* There is at least one neighbor to the left.
2718 DPRINTF("reading left neighbor");
2719 node = NODEPTR(mpp->mp_parent, mpp->mp_pi - 1);
2720 if ((npp.mp_page = mdb_get_page(txn, NODEPGNO(node))) == NULL)
2721 return MDB_PAGE_NOTFOUND;
2722 npp.mp_pi = mpp->mp_pi - 1;
2723 si = NUMKEYS(npp.mp_page) - 1;
2726 npp.mp_parent = mpp->mp_parent;
2728 DPRINTF("found neighbor page %lu (%u keys, %.1f%% full)",
2729 npp.mp_page->mp_pgno, NUMKEYS(npp.mp_page), (float)PAGEFILL(txn->mt_env, npp.mp_page) / 10);
2731 /* If the neighbor page is above threshold and has at least two
2732 * keys, move one key from it.
2734 * Otherwise we should try to merge them.
2736 if (PAGEFILL(txn->mt_env, npp.mp_page) >= FILL_THRESHOLD && NUMKEYS(npp.mp_page) >= 2)
2737 return mdb_move_node(txn, dbi, &npp, si, mpp, di);
2738 else { /* FIXME: if (has_enough_room()) */
2739 if (mpp->mp_pi == 0)
2740 return mdb_merge(txn, dbi, &npp, mpp);
2742 return mdb_merge(txn, dbi, mpp, &npp);
2747 mdb_del0(MDB_txn *txn, MDB_dbi dbi, unsigned int ki, MDB_pageparent *mpp, MDB_node *leaf)
2751 /* add overflow pages to free list */
2752 if (F_ISSET(leaf->mn_flags, F_BIGDATA)) {
2756 memcpy(&pg, NODEDATA(leaf), sizeof(pg));
2757 ovpages = OVPAGES(NODEDSZ(leaf), txn->mt_env->me_psize);
2758 for (i=0; i<ovpages; i++) {
2759 DPRINTF("freed ov page %lu", pg);
2760 mdb_midl_insert(txn->mt_free_pgs, pg);
2764 mdb_del_node(mpp->mp_page, ki);
2765 txn->mt_dbs[dbi].md_entries--;
2766 rc = mdb_rebalance(txn, dbi, mpp);
2767 if (rc != MDB_SUCCESS)
2768 txn->mt_flags |= MDB_TXN_ERROR;
2774 mdb_del(MDB_txn *txn, MDB_dbi dbi,
2775 MDB_val *key, MDB_val *data,
2783 DPRINTF("========> delete key %.*s", (int)key->mv_size, (char *)key->mv_data);
2785 assert(key != NULL);
2787 if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
2790 if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
2794 if (key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
2798 mpp.mp_parent = NULL;
2800 if ((rc = mdb_search_page(txn, dbi, key, NULL, 1, &mpp)) != MDB_SUCCESS)
2803 leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, &ki);
2804 if (leaf == NULL || !exact) {
2805 return MDB_NOTFOUND;
2808 if (F_ISSET(txn->mt_dbs[dbi].md_flags, MDB_DUPSORT)) {
2812 mdb_xcursor_init0(txn, dbi, &mx);
2813 mdb_xcursor_init1(txn, dbi, &mx, leaf);
2814 if (flags == MDB_DEL_DUP) {
2815 rc = mdb_del(&mx.mx_txn, mx.mx_cursor.mc_dbi, data, NULL, 0);
2816 mdb_xcursor_fini(txn, dbi, &mx);
2817 if (rc != MDB_SUCCESS)
2819 /* If sub-DB still has entries, we're done */
2820 if (mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi].md_root != P_INVALID) {
2821 memcpy(NODEDATA(leaf), &mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi],
2825 /* otherwise fall thru and delete the sub-DB */
2827 /* add all the child DB's pages to the free list */
2828 rc = mdb_search_page(&mx.mx_txn, mx.mx_cursor.mc_dbi,
2829 NULL, &mx.mx_cursor, 0, &mp2);
2830 if (rc == MDB_SUCCESS) {
2831 MDB_ppage *top, *parent;
2835 cursor_pop_page(&mx.mx_cursor);
2836 if (mx.mx_cursor.mc_snum) {
2837 top = CURSOR_TOP(&mx.mx_cursor);
2838 while (mx.mx_cursor.mc_snum > 1) {
2839 parent = CURSOR_PARENT(&mx.mx_cursor);
2840 for (i=0; i<NUMKEYS(top->mp_page); i++) {
2841 ni = NODEPTR(top->mp_page, i);
2842 mdb_midl_insert(txn->mt_free_pgs, ni->mn_pgno);
2845 if (parent->mp_ki >= NUMKEYS(parent->mp_page)) {
2846 cursor_pop_page(&mx.mx_cursor);
2849 ni = NODEPTR(parent->mp_page, parent->mp_ki);
2850 top->mp_page = mdb_get_page(&mx.mx_txn, ni->mn_pgno);
2854 mdb_midl_insert(txn->mt_free_pgs, mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi].md_root);
2859 if (data && (rc = mdb_read_data(txn, leaf, data)) != MDB_SUCCESS)
2862 return mdb_del0(txn, dbi, ki, &mpp, leaf);
2865 /* Split page <*mpp>, and insert <key,(data|newpgno)> in either left or
2866 * right sibling, at index <*newindxp> (as if unsplit). Updates *mpp and
2867 * *newindxp with the actual values after split, ie if *mpp and *newindxp
2868 * refer to a node in the new right sibling page.
2871 mdb_split(MDB_txn *txn, MDB_dbi dbi, MDB_page **mpp, unsigned int *newindxp,
2872 MDB_val *newkey, MDB_val *newdata, pgno_t newpgno)
2875 int rc = MDB_SUCCESS, ins_new = 0;
2878 unsigned int i, j, split_indx;
2880 MDB_val sepkey, rkey, rdata;
2882 MDB_dpage *mdp, *rdp, *pdp;
2885 assert(txn != NULL);
2887 dh = ((MDB_dhead *)*mpp) - 1;
2888 mdp = (MDB_dpage *)dh;
2889 newindx = *newindxp;
2891 DPRINTF("-----> splitting %s page %lu and adding [%.*s] at index %i",
2892 IS_LEAF(&mdp->p) ? "leaf" : "branch", mdp->p.mp_pgno,
2893 (int)newkey->mv_size, (char *)newkey->mv_data, *newindxp);
2895 if (mdp->h.md_parent == NULL) {
2896 if ((pdp = mdb_new_page(txn, dbi, P_BRANCH, 1)) == NULL)
2899 mdp->h.md_parent = &pdp->p;
2900 txn->mt_dbs[dbi].md_root = pdp->p.mp_pgno;
2901 DPRINTF("root split! new root = %lu", pdp->p.mp_pgno);
2902 txn->mt_dbs[dbi].md_depth++;
2904 /* Add left (implicit) pointer. */
2905 if ((rc = mdb_add_node(txn, dbi, &pdp->p, 0, NULL, NULL,
2906 mdp->p.mp_pgno, 0)) != MDB_SUCCESS)
2909 DPRINTF("parent branch page is %lu", mdp->h.md_parent->mp_pgno);
2912 /* Create a right sibling. */
2913 if ((rdp = mdb_new_page(txn, dbi, mdp->p.mp_flags, 1)) == NULL)
2915 rdp->h.md_parent = mdp->h.md_parent;
2916 rdp->h.md_pi = mdp->h.md_pi + 1;
2917 DPRINTF("new right sibling: page %lu", rdp->p.mp_pgno);
2919 /* Move half of the keys to the right sibling. */
2920 if ((copy = malloc(txn->mt_env->me_psize)) == NULL)
2922 memcpy(copy, &mdp->p, txn->mt_env->me_psize);
2923 memset(&mdp->p.mp_ptrs, 0, txn->mt_env->me_psize - PAGEHDRSZ);
2924 mdp->p.mp_lower = PAGEHDRSZ;
2925 mdp->p.mp_upper = txn->mt_env->me_psize;
2927 split_indx = NUMKEYS(copy) / 2 + 1;
2929 /* First find the separating key between the split pages.
2931 memset(&sepkey, 0, sizeof(sepkey));
2932 if (newindx == split_indx) {
2933 sepkey.mv_size = newkey->mv_size;
2934 sepkey.mv_data = newkey->mv_data;
2936 node = NODEPTR(copy, split_indx);
2937 sepkey.mv_size = node->mn_ksize;
2938 sepkey.mv_data = NODEKEY(node);
2941 DPRINTF("separator is [%.*s]", (int)sepkey.mv_size, (char *)sepkey.mv_data);
2943 /* Copy separator key to the parent.
2945 if (SIZELEFT(rdp->h.md_parent) < mdb_branch_size(txn->mt_env, &sepkey)) {
2946 rc = mdb_split(txn, dbi, &rdp->h.md_parent, &rdp->h.md_pi,
2947 &sepkey, NULL, rdp->p.mp_pgno);
2949 /* Right page might now have changed parent.
2950 * Check if left page also changed parent.
2952 if (rdp->h.md_parent != mdp->h.md_parent &&
2953 mdp->h.md_pi >= NUMKEYS(mdp->h.md_parent)) {
2954 mdp->h.md_parent = rdp->h.md_parent;
2955 mdp->h.md_pi = rdp->h.md_pi - 1;
2958 rc = mdb_add_node(txn, dbi, rdp->h.md_parent, rdp->h.md_pi,
2959 &sepkey, NULL, rdp->p.mp_pgno, 0);
2961 if (rc != MDB_SUCCESS) {
2966 for (i = j = 0; i <= NUMKEYS(copy); j++) {
2967 if (i < split_indx) {
2968 /* Re-insert in left sibling. */
2971 /* Insert in right sibling. */
2972 if (i == split_indx)
2973 /* Reset insert index for right sibling. */
2974 j = (i == newindx && ins_new);
2978 if (i == newindx && !ins_new) {
2979 /* Insert the original entry that caused the split. */
2980 rkey.mv_data = newkey->mv_data;
2981 rkey.mv_size = newkey->mv_size;
2982 if (IS_LEAF(&mdp->p)) {
2983 rdata.mv_data = newdata->mv_data;
2984 rdata.mv_size = newdata->mv_size;
2991 /* Update page and index for the new key. */
2994 } else if (i == NUMKEYS(copy)) {
2997 node = NODEPTR(copy, i);
2998 rkey.mv_data = NODEKEY(node);
2999 rkey.mv_size = node->mn_ksize;
3000 if (IS_LEAF(&mdp->p)) {
3001 rdata.mv_data = NODEDATA(node);
3002 rdata.mv_size = node->mn_dsize;
3004 pgno = node->mn_pgno;
3005 flags = node->mn_flags;
3010 if (!IS_LEAF(&mdp->p) && j == 0) {
3011 /* First branch index doesn't need key data. */
3015 rc = mdb_add_node(txn, dbi, &pdp->p, j, &rkey, &rdata, pgno,flags);
3023 mdb_put0(MDB_txn *txn, MDB_dbi dbi,
3024 MDB_val *key, MDB_val *data, unsigned int flags)
3026 int rc = MDB_SUCCESS, exact;
3030 MDB_val xdata, *rdata;
3033 DPRINTF("==> put key %.*s, size %zu, data size %zu",
3034 (int)key->mv_size, (char *)key->mv_data, key->mv_size, data->mv_size);
3036 mpp.mp_parent = NULL;
3038 rc = mdb_search_page(txn, dbi, key, NULL, 1, &mpp);
3039 if (rc == MDB_SUCCESS) {
3040 leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, &ki);
3041 if (leaf && exact) {
3042 if (F_ISSET(txn->mt_dbs[dbi].md_flags, MDB_DUPSORT)) {
3045 if (flags == MDB_NOOVERWRITE) {
3046 DPRINTF("duplicate key %.*s",
3047 (int)key->mv_size, (char *)key->mv_data);
3048 return MDB_KEYEXIST;
3050 /* same size, just replace it */
3051 if (NODEDSZ(leaf) == data->mv_size) {
3052 memcpy(NODEDATA(leaf), data->mv_data, data->mv_size);
3055 mdb_del_node(mpp.mp_page, ki);
3057 if (leaf == NULL) { /* append if not found */
3058 ki = NUMKEYS(mpp.mp_page);
3059 DPRINTF("appending key at index %i", ki);
3061 } else if (rc == MDB_NOTFOUND) {
3063 /* new file, just write a root leaf page */
3064 DPRINTF("allocating new root leaf page");
3065 if ((dp = mdb_new_page(txn, dbi, P_LEAF, 1)) == NULL) {
3068 mpp.mp_page = &dp->p;
3069 txn->mt_dbs[dbi].md_root = mpp.mp_page->mp_pgno;
3070 txn->mt_dbs[dbi].md_depth++;
3071 txn->mt_dbxs[dbi].md_dirty = 1;
3077 assert(IS_LEAF(mpp.mp_page));
3078 DPRINTF("there are %u keys, should insert new key at index %i",
3079 NUMKEYS(mpp.mp_page), ki);
3081 /* For sorted dups, the data item at this level is a DB record
3082 * for a child DB; the actual data elements are stored as keys
3085 if (F_ISSET(txn->mt_dbs[dbi].md_flags, MDB_DUPSORT)) {
3087 xdata.mv_size = sizeof(MDB_db);
3088 xdata.mv_data = &dummy;
3089 memset(&dummy, 0, sizeof(dummy));
3090 dummy.md_root = P_INVALID;
3095 if (SIZELEFT(mpp.mp_page) < mdb_leaf_size(txn->mt_env, key, rdata)) {
3096 rc = mdb_split(txn, dbi, &mpp.mp_page, &ki, key, rdata, P_INVALID);
3098 /* There is room already in this leaf page. */
3099 rc = mdb_add_node(txn, dbi, mpp.mp_page, ki, key, rdata, 0, 0);
3102 if (rc != MDB_SUCCESS)
3103 txn->mt_flags |= MDB_TXN_ERROR;
3105 txn->mt_dbs[dbi].md_entries++;
3107 /* Remember if we just added a subdatabase */
3108 if (flags & F_SUBDATA) {
3109 leaf = NODEPTR(mpp.mp_page, ki);
3110 leaf->mn_flags |= F_SUBDATA;
3113 /* Now store the actual data in the child DB. Note that we're
3114 * storing the user data in the keys field, so there are strict
3115 * size limits on dupdata. The actual data fields of the child
3116 * DB are all zero size.
3118 if (F_ISSET(txn->mt_dbs[dbi].md_flags, MDB_DUPSORT)) {
3121 leaf = NODEPTR(mpp.mp_page, ki);
3123 mdb_xcursor_init0(txn, dbi, &mx);
3124 mdb_xcursor_init1(txn, dbi, &mx, leaf);
3127 if (flags == MDB_NODUPDATA)
3128 flags = MDB_NOOVERWRITE;
3129 rc = mdb_put0(&mx.mx_txn, mx.mx_cursor.mc_dbi, data, &xdata, flags);
3130 mdb_xcursor_fini(txn, dbi, &mx);
3131 memcpy(NODEDATA(leaf), &mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi],
3141 mdb_put(MDB_txn *txn, MDB_dbi dbi,
3142 MDB_val *key, MDB_val *data, unsigned int flags)
3144 assert(key != NULL);
3145 assert(data != NULL);
3147 if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
3150 if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
3154 if (key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
3158 if ((flags & (MDB_NOOVERWRITE|MDB_NODUPDATA)) != flags)
3161 return mdb_put0(txn, dbi, key, data, flags);
3165 mdb_env_set_flags(MDB_env *env, unsigned int flag, int onoff)
3167 #define CHANGEABLE (MDB_NOSYNC)
3168 if ((flag & CHANGEABLE) != flag)
3171 env->me_flags |= flag;
3173 env->me_flags &= ~flag;
3178 mdb_env_get_flags(MDB_env *env, unsigned int *arg)
3183 *arg = env->me_flags;
3188 mdb_env_get_path(MDB_env *env, const char **arg)
3193 *arg = env->me_path;
3198 mdb_stat0(MDB_env *env, MDB_db *db, MDB_stat *arg)
3200 arg->ms_psize = env->me_psize;
3201 arg->ms_depth = db->md_depth;
3202 arg->ms_branch_pages = db->md_branch_pages;
3203 arg->ms_leaf_pages = db->md_leaf_pages;
3204 arg->ms_overflow_pages = db->md_overflow_pages;
3205 arg->ms_entries = db->md_entries;
3210 mdb_env_stat(MDB_env *env, MDB_stat *arg)
3212 if (env == NULL || arg == NULL)
3215 return mdb_stat0(env, &env->me_meta->mm_dbs[MAIN_DBI], arg);
3218 int mdb_open(MDB_txn *txn, const char *name, unsigned int flags, MDB_dbi *dbi)
3228 if (flags & (MDB_DUPSORT|MDB_REVERSEKEY|MDB_INTEGERKEY))
3229 txn->mt_dbs[MAIN_DBI].md_flags |= (flags & (MDB_DUPSORT|MDB_REVERSEKEY|MDB_INTEGERKEY));
3233 /* Is the DB already open? */
3235 for (i=2; i<txn->mt_numdbs; i++) {
3236 if (len == txn->mt_dbxs[i].md_name.mv_size &&
3237 !strncmp(name, txn->mt_dbxs[i].md_name.mv_data, len)) {
3243 if (txn->mt_numdbs >= txn->mt_env->me_maxdbs - 1)
3246 /* Find the DB info */
3248 key.mv_data = (void *)name;
3249 rc = mdb_get(txn, MAIN_DBI, &key, &data);
3251 /* Create if requested */
3252 if (rc == MDB_NOTFOUND && (flags & MDB_CREATE)) {
3254 data.mv_size = sizeof(MDB_db);
3255 data.mv_data = &dummy;
3256 memset(&dummy, 0, sizeof(dummy));
3257 dummy.md_root = P_INVALID;
3258 dummy.md_flags = flags & 0xffff;
3259 rc = mdb_put0(txn, MAIN_DBI, &key, &data, F_SUBDATA);
3263 /* OK, got info, add to table */
3264 if (rc == MDB_SUCCESS) {
3265 txn->mt_dbxs[txn->mt_numdbs].md_name.mv_data = strdup(name);
3266 txn->mt_dbxs[txn->mt_numdbs].md_name.mv_size = len;
3267 txn->mt_dbxs[txn->mt_numdbs].md_cmp = NULL;
3268 txn->mt_dbxs[txn->mt_numdbs].md_dcmp = NULL;
3269 txn->mt_dbxs[txn->mt_numdbs].md_rel = NULL;
3270 txn->mt_dbxs[txn->mt_numdbs].md_parent = MAIN_DBI;
3271 txn->mt_dbxs[txn->mt_numdbs].md_dirty = dirty;
3272 memcpy(&txn->mt_dbs[txn->mt_numdbs], data.mv_data, sizeof(MDB_db));
3273 *dbi = txn->mt_numdbs;
3274 txn->mt_env->me_dbs[0][txn->mt_numdbs] = txn->mt_dbs[txn->mt_numdbs];
3275 txn->mt_env->me_dbs[1][txn->mt_numdbs] = txn->mt_dbs[txn->mt_numdbs];
3282 int mdb_stat(MDB_txn *txn, MDB_dbi dbi, MDB_stat *arg)
3284 if (txn == NULL || arg == NULL || dbi >= txn->mt_numdbs)
3287 return mdb_stat0(txn->mt_env, &txn->mt_dbs[dbi], arg);
3290 void mdb_close(MDB_txn *txn, MDB_dbi dbi)
3293 if (dbi <= MAIN_DBI || dbi >= txn->mt_numdbs)
3295 ptr = txn->mt_dbxs[dbi].md_name.mv_data;
3296 txn->mt_dbxs[dbi].md_name.mv_data = NULL;
3297 txn->mt_dbxs[dbi].md_name.mv_size = 0;
3301 int mdb_set_compare(MDB_txn *txn, MDB_dbi dbi, MDB_cmp_func *cmp)
3303 if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
3306 txn->mt_dbxs[dbi].md_cmp = cmp;
3310 int mdb_set_dupsort(MDB_txn *txn, MDB_dbi dbi, MDB_cmp_func *cmp)
3312 if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
3315 txn->mt_dbxs[dbi].md_dcmp = cmp;
3319 int mdb_set_relfunc(MDB_txn *txn, MDB_dbi dbi, MDB_rel_func *rel)
3321 if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
3324 txn->mt_dbxs[dbi].md_rel = rel;