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/queue.h>
33 #include <sys/param.h>
36 #ifdef HAVE_SYS_FILE_H
55 #define ULONG unsigned long
64 #if DEBUG && defined(__GNUC__)
65 # define DPRINTF(fmt, ...) \
66 fprintf(stderr, "%s:%d: " fmt "\n", __func__, __LINE__, ##__VA_ARGS__)
68 # define DPRINTF(...) ((void) 0)
73 #define MDB_MAGIC 0xBEEFC0DE
75 #define MAXKEYSIZE 511
77 #define P_INVALID (~0UL)
79 #define F_ISSET(w, f) (((w) & (f)) == (f))
81 typedef uint16_t indx_t;
83 #define DEFAULT_READERS 126
84 #define DEFAULT_MAPSIZE 1048576
86 /* Lock descriptor stuff */
91 typedef struct MDB_rxbody {
96 #define CACHELINE 64 /* most CPUs. Itanium uses 128 */
99 typedef struct MDB_reader {
101 /* cache line alignment */
102 char pad[CACHELINE-sizeof(MDB_rxbody)];
107 uint32_t mt_version; \
108 pthread_mutex_t mt_mutex; \
110 uint32_t mt_numreaders
111 typedef struct MDB_txbody {
115 typedef struct MDB_txninfo {
117 char pad[CACHELINE-sizeof(MDB_txbody)];
118 pthread_mutex_t mt_wmutex;
119 char pad2[CACHELINE-sizeof(pthread_mutex_t)];
120 MDB_reader mt_readers[1];
123 /* Common header for all page types. Overflow pages
124 * occupy a number of contiguous pages with no
125 * headers on any page after the first.
127 typedef struct MDB_page { /* represents a page of storage */
128 #define mp_pgno mp_p.p_pgno
130 pgno_t p_pgno; /* page number */
133 #define P_BRANCH 0x01 /* branch page */
134 #define P_LEAF 0x02 /* leaf page */
135 #define P_OVERFLOW 0x04 /* overflow page */
136 #define P_META 0x08 /* meta page */
137 #define P_DIRTY 0x10 /* dirty page */
139 #define mp_lower mp_pb.pb.pb_lower
140 #define mp_upper mp_pb.pb.pb_upper
141 #define mp_pages mp_pb.pb_pages
144 indx_t pb_lower; /* lower bound of free space */
145 indx_t pb_upper; /* upper bound of free space */
147 uint32_t pb_pages; /* number of overflow pages */
149 indx_t mp_ptrs[1]; /* dynamic size */
152 #define PAGEHDRSZ ((unsigned) offsetof(MDB_page, mp_ptrs))
154 #define NUMKEYS(p) (((p)->mp_lower - PAGEHDRSZ) >> 1)
155 #define SIZELEFT(p) (indx_t)((p)->mp_upper - (p)->mp_lower)
156 #define PAGEFILL(env, p) (1000L * ((env)->me_psize - PAGEHDRSZ - SIZELEFT(p)) / \
157 ((env)->me_psize - PAGEHDRSZ))
158 #define IS_LEAF(p) F_ISSET((p)->mp_flags, P_LEAF)
159 #define IS_BRANCH(p) F_ISSET((p)->mp_flags, P_BRANCH)
160 #define IS_OVERFLOW(p) F_ISSET((p)->mp_flags, P_OVERFLOW)
162 #define OVPAGES(size, psize) (PAGEHDRSZ + size + psize - 1) / psize;
164 typedef struct MDB_db {
168 ULONG md_branch_pages;
170 ULONG md_overflow_pages;
178 typedef struct MDB_meta { /* meta (footer) page content */
181 void *mm_address; /* address for fixed mapping */
182 size_t mm_mapsize; /* size of mmap region */
183 MDB_db mm_dbs[2]; /* first is free space, 2nd is main db */
184 #define mm_psize mm_dbs[0].md_pad
185 #define mm_flags mm_dbs[0].md_flags
186 pgno_t mm_last_pg; /* last used page in file */
187 ULONG mm_txnid; /* txnid that committed this page */
190 typedef struct MDB_dhead { /* a dirty page */
191 STAILQ_ENTRY(MDB_dpage) md_next; /* queue of dirty pages */
193 unsigned md_pi; /* parent index */
197 typedef struct MDB_dpage {
202 STAILQ_HEAD(dirty_queue, MDB_dpage); /* FIXME: use a sorted data structure */
204 typedef struct MDB_oldpages {
205 struct MDB_oldpages *mo_next;
207 pgno_t mo_pages[1]; /* dynamic */
210 typedef struct MDB_pageparent {
216 static MDB_dpage *mdb_alloc_page(MDB_txn *txn, MDB_page *parent, unsigned int parent_idx, int num);
217 static int mdb_touch(MDB_txn *txn, MDB_pageparent *mp);
219 typedef struct MDB_ppage { /* ordered list of pages */
220 SLIST_ENTRY(MDB_ppage) mp_entry;
222 unsigned int mp_ki; /* cursor index on page */
224 SLIST_HEAD(page_stack, MDB_ppage);
226 #define CURSOR_EMPTY(c) SLIST_EMPTY(&(c)->mc_stack)
227 #define CURSOR_TOP(c) SLIST_FIRST(&(c)->mc_stack)
228 #define CURSOR_POP(c) SLIST_REMOVE_HEAD(&(c)->mc_stack, mp_entry)
229 #define CURSOR_PUSH(c,p) SLIST_INSERT_HEAD(&(c)->mc_stack, p, mp_entry)
235 struct page_stack mc_stack; /* stack of parent pages */
237 short mc_initialized; /* 1 if initialized */
238 short mc_eof; /* 1 if end is reached */
239 struct MDB_xcursor *mc_xcursor;
242 #define METAHASHLEN offsetof(MDB_meta, mm_hash)
243 #define METADATA(p) ((void *)((char *)p + PAGEHDRSZ))
245 typedef struct MDB_node {
246 #define mn_pgno mn_p.np_pgno
247 #define mn_dsize mn_p.np_dsize
249 pgno_t np_pgno; /* child page number */
250 uint32_t np_dsize; /* leaf data size */
252 unsigned int mn_flags:4;
253 unsigned int mn_ksize:12; /* key size */
254 #define F_BIGDATA 0x01 /* data put on overflow page */
255 #define F_SUBDATA 0x02 /* data is a sub-database */
259 typedef struct MDB_dbx {
261 MDB_cmp_func *md_cmp; /* user compare function */
262 MDB_cmp_func *md_dcmp; /* user dupsort function */
263 MDB_rel_func *md_rel; /* user relocate function */
265 unsigned int md_dirty;
269 pgno_t mt_next_pgno; /* next unallocated page */
273 pgno_t *mt_free_pgs; /* this is an IDL */
275 struct dirty_queue *dirty_queue; /* modified pages */
278 MDB_dbx *mt_dbxs; /* array */
280 unsigned int mt_numdbs;
282 #define MDB_TXN_RDONLY 0x01 /* read-only transaction */
283 #define MDB_TXN_ERROR 0x02 /* an error has occurred */
284 #define MDB_TXN_METOGGLE 0x04 /* used meta page 1 */
285 unsigned int mt_flags;
288 /* Context for sorted-dup records */
289 typedef struct MDB_xcursor {
290 MDB_cursor mx_cursor;
300 unsigned int me_maxreaders;
301 unsigned int me_numdbs;
302 unsigned int me_maxdbs;
305 MDB_txninfo *me_txns;
306 MDB_meta *me_metas[2];
308 MDB_txn *me_txn; /* current write transaction */
310 off_t me_size; /* current file size */
311 unsigned int me_psize;
313 MDB_dbx *me_dbxs; /* array */
315 MDB_oldpages *me_pghead;
316 pthread_key_t me_txkey; /* thread-key for readers */
317 pgno_t me_free_pgs[MDB_IDL_UM_SIZE];
320 #define NODESIZE offsetof(MDB_node, mn_data)
322 #define INDXSIZE(k) (NODESIZE + ((k) == NULL ? 0 : (k)->mv_size))
323 #define LEAFSIZE(k, d) (NODESIZE + (k)->mv_size + (d)->mv_size)
324 #define NODEPTR(p, i) ((MDB_node *)((char *)(p) + (p)->mp_ptrs[i]))
325 #define NODEKEY(node) (void *)((node)->mn_data)
326 #define NODEDATA(node) (void *)((char *)(node)->mn_data + (node)->mn_ksize)
327 #define NODEPGNO(node) ((node)->mn_pgno)
328 #define NODEDSZ(node) ((node)->mn_dsize)
330 #define MDB_COMMIT_PAGES 64 /* max number of pages to write in one commit */
331 #define MDB_MAXCACHE_DEF 1024 /* max number of pages to keep in cache */
333 static int mdb_search_page_root(MDB_txn *txn,
334 MDB_dbi dbi, MDB_val *key,
335 MDB_cursor *cursor, int modify,
336 MDB_pageparent *mpp);
337 static int mdb_search_page(MDB_txn *txn,
338 MDB_dbi dbi, MDB_val *key,
339 MDB_cursor *cursor, int modify,
340 MDB_pageparent *mpp);
342 static int mdbenv_read_header(MDB_env *env, MDB_meta *meta);
343 static int mdbenv_read_meta(MDB_env *env, int *which);
344 static int mdbenv_write_meta(MDB_txn *txn);
345 static MDB_page *mdb_get_page(MDB_txn *txn, pgno_t pgno);
347 static MDB_node *mdb_search_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp,
348 MDB_val *key, int *exactp, unsigned int *kip);
349 static int mdb_add_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp,
350 indx_t indx, MDB_val *key, MDB_val *data,
351 pgno_t pgno, uint8_t flags);
352 static void mdb_del_node(MDB_page *mp, indx_t indx);
353 static int mdb_del0(MDB_txn *txn, MDB_dbi dbi, unsigned int ki,
354 MDB_pageparent *mpp, MDB_node *leaf);
355 static int mdb_put0(MDB_txn *txn, MDB_dbi dbi,
356 MDB_val *key, MDB_val *data, unsigned int flags);
357 static int mdb_read_data(MDB_txn *txn, MDB_node *leaf, MDB_val *data);
359 static int mdb_rebalance(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *mp);
360 static int mdb_update_key(MDB_page *mp, indx_t indx, MDB_val *key);
361 static int mdb_move_node(MDB_txn *txn, MDB_dbi dbi,
362 MDB_pageparent *src, indx_t srcindx,
363 MDB_pageparent *dst, indx_t dstindx);
364 static int mdb_merge(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *src,
365 MDB_pageparent *dst);
366 static int mdb_split(MDB_txn *txn, MDB_dbi dbi, MDB_page **mpp,
367 unsigned int *newindxp, MDB_val *newkey,
368 MDB_val *newdata, pgno_t newpgno);
369 static MDB_dpage *mdb_new_page(MDB_txn *txn, MDB_dbi dbi, uint32_t flags, int num);
371 static void cursor_pop_page(MDB_cursor *cursor);
372 static MDB_ppage *cursor_push_page(MDB_cursor *cursor,
375 static int mdb_set_key(MDB_node *node, MDB_val *key);
376 static int mdb_sibling(MDB_cursor *cursor, int move_right);
377 static int mdb_cursor_next(MDB_cursor *cursor,
378 MDB_val *key, MDB_val *data, MDB_cursor_op op);
379 static int mdb_cursor_prev(MDB_cursor *cursor,
380 MDB_val *key, MDB_val *data, MDB_cursor_op op);
381 static int mdb_cursor_set(MDB_cursor *cursor,
382 MDB_val *key, MDB_val *data, MDB_cursor_op op, int *exactp);
383 static int mdb_cursor_first(MDB_cursor *cursor,
384 MDB_val *key, MDB_val *data);
385 static int mdb_cursor_last(MDB_cursor *cursor,
386 MDB_val *key, MDB_val *data);
388 static void mdb_xcursor_init0(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx);
389 static void mdb_xcursor_init1(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx, MDB_node *node);
390 static void mdb_xcursor_fini(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx);
392 static size_t mdb_leaf_size(MDB_env *env, MDB_val *key,
394 static size_t mdb_branch_size(MDB_env *env, MDB_val *key);
396 static int memncmp(const void *s1, size_t n1,
397 const void *s2, size_t n2);
398 static int memnrcmp(const void *s1, size_t n1,
399 const void *s2, size_t n2);
402 memncmp(const void *s1, size_t n1, const void *s2, size_t n2)
404 int diff, len_diff = -1;
407 len_diff = (n1 > n2);
410 diff = memcmp(s1, s2, n1);
411 return diff ? diff : len_diff;
415 memnrcmp(const void *s1, size_t n1, const void *s2, size_t n2)
417 const unsigned char *p1, *p2, *p1_lim;
424 p1 = (const unsigned char *)s1 + n1 - 1;
425 p2 = (const unsigned char *)s2 + n2 - 1;
427 for (p1_lim = (n1 <= n2 ? s1 : s2); *p1 == *p2; p1--, p2--) {
429 return (p1 != s1) ? (p1 != p2) : (p2 != s2) ? -1 : 0;
435 mdb_cmp(MDB_txn *txn, MDB_dbi dbi, const MDB_val *a, const MDB_val *b)
437 return txn->mt_dbxs[dbi].md_cmp(a, b);
441 _mdb_cmp(MDB_txn *txn, MDB_dbi dbi, const MDB_val *key1, const MDB_val *key2)
443 if (txn->mt_dbs[dbi].md_flags & (MDB_REVERSEKEY
444 #if __BYTE_ORDER == __LITTLE_ENDIAN
448 return memnrcmp(key1->mv_data, key1->mv_size, key2->mv_data, key2->mv_size);
450 return memncmp((char *)key1->mv_data, key1->mv_size, key2->mv_data, key2->mv_size);
453 /* Allocate new page(s) for writing */
455 mdb_alloc_page(MDB_txn *txn, MDB_page *parent, unsigned int parent_idx, int num)
458 pgno_t pgno = P_INVALID;
461 if (txn->mt_txnid > 2) {
463 oldest = txn->mt_txnid - 2;
464 if (!txn->mt_env->me_pghead && txn->mt_dbs[FREE_DBI].md_root != P_INVALID) {
465 /* See if there's anything in the free DB */
470 mpp.mp_parent = NULL;
472 mdb_search_page(txn, FREE_DBI, NULL, NULL, 0, &mpp);
473 leaf = NODEPTR(mpp.mp_page, 0);
474 kptr = (ULONG *)NODEKEY(leaf);
476 /* It's potentially usable, unless there are still
477 * older readers outstanding. Grab it.
479 if (oldest > *kptr) {
484 mdb_read_data(txn, leaf, &data);
485 idl = (ULONG *)data.mv_data;
486 mop = malloc(sizeof(MDB_oldpages) + MDB_IDL_SIZEOF(idl) - sizeof(pgno_t));
487 mop->mo_next = txn->mt_env->me_pghead;
488 mop->mo_txnid = *kptr;
489 txn->mt_env->me_pghead = mop;
490 memcpy(mop->mo_pages, idl, MDB_IDL_SIZEOF(idl));
495 DPRINTF("IDL read txn %lu root %lu num %lu",
496 mop->mo_txnid, txn->mt_dbs[FREE_DBI].md_root, idl[0]);
497 for (i=0; i<idl[0]; i++) {
498 DPRINTF("IDL %lu", idl[i+1]);
502 /* drop this IDL from the DB */
503 mpp.mp_parent = NULL;
505 mdb_search_page(txn, FREE_DBI, NULL, NULL, 1, &mpp);
506 leaf = NODEPTR(mpp.mp_page, 0);
507 mdb_del0(txn, FREE_DBI, 0, &mpp, leaf);
510 if (txn->mt_env->me_pghead) {
512 for (i=0; i<txn->mt_env->me_txns->mt_numreaders; i++) {
513 ULONG mr = txn->mt_env->me_txns->mt_readers[i].mr_txnid;
516 oldest = txn->mt_env->me_txns->mt_readers[i].mr_txnid;
518 if (oldest > txn->mt_env->me_pghead->mo_txnid) {
519 MDB_oldpages *mop = txn->mt_env->me_pghead;
520 txn->mt_oldest = oldest;
522 /* FIXME: For now, always use fresh pages. We
523 * really ought to search the free list for a
528 /* peel pages off tail, so we only have to truncate the list */
529 pgno = MDB_IDL_LAST(mop->mo_pages);
530 if (MDB_IDL_IS_RANGE(mop->mo_pages)) {
532 if (mop->mo_pages[2] > mop->mo_pages[1])
533 mop->mo_pages[0] = 0;
537 if (MDB_IDL_IS_ZERO(mop->mo_pages)) {
538 txn->mt_env->me_pghead = mop->mo_next;
546 if ((dp = malloc(txn->mt_env->me_psize * num + sizeof(MDB_dhead))) == NULL)
549 dp->h.md_parent = parent;
550 dp->h.md_pi = parent_idx;
551 STAILQ_INSERT_TAIL(txn->mt_u.dirty_queue, dp, h.md_next);
552 if (pgno == P_INVALID) {
553 dp->p.mp_pgno = txn->mt_next_pgno;
554 txn->mt_next_pgno += num;
556 dp->p.mp_pgno = pgno;
562 /* Touch a page: make it dirty and re-insert into tree with updated pgno.
565 mdb_touch(MDB_txn *txn, MDB_pageparent *pp)
567 MDB_page *mp = pp->mp_page;
572 if (!F_ISSET(mp->mp_flags, P_DIRTY)) {
574 if ((dp = mdb_alloc_page(txn, pp->mp_parent, pp->mp_pi, 1)) == NULL)
576 DPRINTF("touched page %lu -> %lu", mp->mp_pgno, dp->p.mp_pgno);
577 mdb_idl_insert(txn->mt_free_pgs, mp->mp_pgno);
578 pgno = dp->p.mp_pgno;
579 memcpy(&dp->p, mp, txn->mt_env->me_psize);
582 mp->mp_flags |= P_DIRTY;
584 /* Update the page number to new touched page. */
585 if (pp->mp_parent != NULL)
586 NODEPGNO(NODEPTR(pp->mp_parent, pp->mp_pi)) = mp->mp_pgno;
593 mdbenv_sync(MDB_env *env)
596 if (!F_ISSET(env->me_flags, MDB_NOSYNC)) {
597 if (fsync(env->me_fd))
604 mdb_txn_begin(MDB_env *env, int rdonly, MDB_txn **ret)
609 if ((txn = calloc(1, sizeof(MDB_txn))) == NULL) {
610 DPRINTF("calloc: %s", strerror(errno));
615 txn->mt_flags |= MDB_TXN_RDONLY;
617 txn->mt_u.dirty_queue = calloc(1, sizeof(*txn->mt_u.dirty_queue));
618 if (txn->mt_u.dirty_queue == NULL) {
622 STAILQ_INIT(txn->mt_u.dirty_queue);
624 pthread_mutex_lock(&env->me_txns->mt_wmutex);
625 env->me_txns->mt_txnid++;
626 txn->mt_free_pgs = env->me_free_pgs;
627 txn->mt_free_pgs[0] = 0;
630 txn->mt_txnid = env->me_txns->mt_txnid;
632 MDB_reader *r = pthread_getspecific(env->me_txkey);
635 pthread_mutex_lock(&env->me_txns->mt_mutex);
636 for (i=0; i<env->me_txns->mt_numreaders; i++)
637 if (env->me_txns->mt_readers[i].mr_pid == 0)
639 if (i == env->me_maxreaders) {
640 pthread_mutex_unlock(&env->me_txns->mti_mutex);
643 env->me_txns->mt_readers[i].mr_pid = getpid();
644 env->me_txns->mt_readers[i].mr_tid = pthread_self();
645 r = &env->me_txns->mt_readers[i];
646 pthread_setspecific(env->me_txkey, r);
647 if (i >= env->me_txns->mt_numreaders)
648 env->me_txns->mt_numreaders = i+1;
649 pthread_mutex_unlock(&env->me_txns->mt_mutex);
651 r->mr_txnid = txn->mt_txnid;
652 txn->mt_u.reader = r;
659 if ((rc = mdbenv_read_meta(env, &toggle)) != MDB_SUCCESS) {
664 /* Copy the DB arrays */
665 txn->mt_numdbs = env->me_numdbs;
666 txn->mt_dbxs = env->me_dbxs; /* mostly static anyway */
667 txn->mt_dbs = malloc(env->me_maxdbs * sizeof(MDB_db));
668 memcpy(txn->mt_dbs, env->me_meta->mm_dbs, 2 * sizeof(MDB_db));
669 if (txn->mt_numdbs > 2)
670 memcpy(txn->mt_dbs+2, env->me_dbs[env->me_db_toggle]+2,
671 (txn->mt_numdbs - 2) * sizeof(MDB_db));
675 txn->mt_flags |= MDB_TXN_METOGGLE;
676 txn->mt_next_pgno = env->me_meta->mm_last_pg+1;
679 DPRINTF("begin transaction %lu on mdbenv %p, root page %lu",
680 txn->mt_txnid, (void *) env, txn->mt_dbs[MAIN_DBI].md_root);
687 mdb_txn_abort(MDB_txn *txn)
696 DPRINTF("abort transaction %lu on mdbenv %p, root page %lu",
697 txn->mt_txnid, (void *) env, txn->mt_dbs[MAIN_DBI].md_root);
701 if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
702 txn->mt_u.reader->mr_txnid = 0;
707 /* Discard all dirty pages. */
708 while (!STAILQ_EMPTY(txn->mt_u.dirty_queue)) {
709 dp = STAILQ_FIRST(txn->mt_u.dirty_queue);
710 STAILQ_REMOVE_HEAD(txn->mt_u.dirty_queue, h.md_next);
713 free(txn->mt_u.dirty_queue);
715 while ((mop = txn->mt_env->me_pghead)) {
716 txn->mt_env->me_pghead = mop->mo_next;
721 env->me_txns->mt_txnid--;
722 for (i=2; i<env->me_numdbs; i++)
723 env->me_dbxs[i].md_dirty = 0;
724 pthread_mutex_unlock(&env->me_txns->mt_wmutex);
731 mdb_txn_commit(MDB_txn *txn)
740 struct iovec iov[MDB_COMMIT_PAGES];
743 assert(txn->mt_env != NULL);
747 if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
748 DPRINTF("attempt to commit read-only transaction");
753 if (txn != env->me_txn) {
754 DPRINTF("attempt to commit unknown transaction");
759 if (F_ISSET(txn->mt_flags, MDB_TXN_ERROR)) {
760 DPRINTF("error flag is set, can't commit");
765 if (STAILQ_EMPTY(txn->mt_u.dirty_queue))
768 DPRINTF("committing transaction %lu on mdbenv %p, root page %lu",
769 txn->mt_txnid, (void *) env, txn->mt_dbs[MAIN_DBI].md_root);
771 /* should only be one record now */
772 if (env->me_pghead) {
776 mop = env->me_pghead;
777 key.mv_size = sizeof(pgno_t);
778 key.mv_data = (char *)&mop->mo_txnid;
779 data.mv_size = MDB_IDL_SIZEOF(mop->mo_pages);
780 data.mv_data = mop->mo_pages;
781 mdb_put0(txn, FREE_DBI, &key, &data, 0);
782 free(env->me_pghead);
783 env->me_pghead = NULL;
785 /* save to free list */
786 if (!MDB_IDL_IS_ZERO(txn->mt_free_pgs)) {
790 /* make sure last page of freeDB is touched and on freelist */
791 key.mv_size = MAXKEYSIZE+1;
793 mpp.mp_parent = NULL;
795 mdb_search_page(txn, FREE_DBI, &key, NULL, 1, &mpp);
800 ULONG *idl = txn->mt_free_pgs;
801 DPRINTF("IDL write txn %lu root %lu num %lu",
802 txn->mt_txnid, txn->mt_dbs[FREE_DBI].md_root, idl[0]);
803 for (i=0; i<idl[0]; i++) {
804 DPRINTF("IDL %lu", idl[i+1]);
808 /* write to last page of freeDB */
809 key.mv_size = sizeof(pgno_t);
810 key.mv_data = (char *)&txn->mt_txnid;
811 data.mv_size = MDB_IDL_SIZEOF(txn->mt_free_pgs);
812 data.mv_data = txn->mt_free_pgs;
813 mdb_put0(txn, FREE_DBI, &key, &data, 0);
816 /* Update DB root pointers. Their pages have already been
817 * touched so this is all in-place and cannot fail.
821 data.mv_size = sizeof(MDB_db);
823 for (i = 2; i < txn->mt_numdbs; i++) {
824 if (txn->mt_dbxs[i].md_dirty) {
825 data.mv_data = &txn->mt_dbs[i];
826 mdb_put0(txn, MAIN_DBI, &txn->mt_dbxs[i].md_name, &data, 0);
831 /* Commit up to MDB_COMMIT_PAGES dirty pages to disk until done.
838 STAILQ_FOREACH(dp, txn->mt_u.dirty_queue, h.md_next) {
839 if (dp->p.mp_pgno != next) {
841 DPRINTF("committing %u dirty pages", n);
842 rc = writev(env->me_fd, iov, n);
846 DPRINTF("short write, filesystem full?");
848 DPRINTF("writev: %s", strerror(errno));
855 lseek(env->me_fd, dp->p.mp_pgno * env->me_psize, SEEK_SET);
856 next = dp->p.mp_pgno;
858 DPRINTF("committing page %lu", dp->p.mp_pgno);
859 iov[n].iov_len = env->me_psize * dp->h.md_num;
860 iov[n].iov_base = &dp->p;
861 size += iov[n].iov_len;
862 next = dp->p.mp_pgno + dp->h.md_num;
863 /* clear dirty flag */
864 dp->p.mp_flags &= ~P_DIRTY;
865 if (++n >= MDB_COMMIT_PAGES) {
874 DPRINTF("committing %u dirty pages", n);
875 rc = writev(env->me_fd, iov, n);
879 DPRINTF("short write, filesystem full?");
881 DPRINTF("writev: %s", strerror(errno));
888 /* Drop the dirty pages.
890 while (!STAILQ_EMPTY(txn->mt_u.dirty_queue)) {
891 dp = STAILQ_FIRST(txn->mt_u.dirty_queue);
892 STAILQ_REMOVE_HEAD(txn->mt_u.dirty_queue, h.md_next);
896 if ((n = mdbenv_sync(env)) != 0 ||
897 (n = mdbenv_write_meta(txn)) != MDB_SUCCESS ||
898 (n = mdbenv_sync(env)) != 0) {
904 /* update the DB tables */
906 int toggle = !env->me_db_toggle;
908 for (i = 2; i < env->me_numdbs; i++) {
909 if (txn->mt_dbxs[i].md_dirty) {
910 env->me_dbs[toggle][i] = txn->mt_dbs[i];
911 txn->mt_dbxs[i].md_dirty = 0;
914 for (i = env->me_numdbs; i < txn->mt_numdbs; i++) {
915 txn->mt_dbxs[i].md_dirty = 0;
916 env->me_dbxs[i] = txn->mt_dbxs[i];
917 env->me_dbs[toggle][i] = txn->mt_dbs[i];
919 env->me_db_toggle = toggle;
920 env->me_numdbs = txn->mt_numdbs;
925 pthread_mutex_unlock(&env->me_txns->mt_wmutex);
926 free(txn->mt_u.dirty_queue);
937 mdbenv_read_header(MDB_env *env, MDB_meta *meta)
946 /* We don't know the page size yet, so use a minimum value.
949 if ((rc = pread(env->me_fd, page, PAGESIZE, 0)) == 0) {
951 } else if (rc != PAGESIZE) {
954 DPRINTF("read: %s", strerror(errno));
958 p = (MDB_page *)page;
960 if (!F_ISSET(p->mp_flags, P_META)) {
961 DPRINTF("page %lu not a meta page", p->mp_pgno);
966 if (m->mm_magic != MDB_MAGIC) {
967 DPRINTF("meta has invalid magic");
971 if (m->mm_version != MDB_VERSION) {
972 DPRINTF("database is version %u, expected version %u",
973 m->mm_version, MDB_VERSION);
974 return MDB_VERSION_MISMATCH;
977 memcpy(meta, m, sizeof(*m));
982 mdbenv_init_meta(MDB_env *env, MDB_meta *meta)
989 DPRINTF("writing new meta page");
990 psize = sysconf(_SC_PAGE_SIZE);
992 meta->mm_magic = MDB_MAGIC;
993 meta->mm_version = MDB_VERSION;
994 meta->mm_psize = psize;
995 meta->mm_last_pg = 1;
996 meta->mm_flags = env->me_flags & 0xffff;
997 meta->mm_flags |= MDB_INTEGERKEY;
998 meta->mm_dbs[0].md_root = P_INVALID;
999 meta->mm_dbs[1].md_root = P_INVALID;
1001 p = calloc(2, psize);
1003 p->mp_flags = P_META;
1006 memcpy(m, meta, sizeof(*meta));
1008 q = (MDB_page *)((char *)p + psize);
1011 q->mp_flags = P_META;
1014 memcpy(m, meta, sizeof(*meta));
1016 rc = write(env->me_fd, p, psize * 2);
1018 return (rc == (int)psize * 2) ? MDB_SUCCESS : errno;
1022 mdbenv_write_meta(MDB_txn *txn)
1030 assert(txn != NULL);
1031 assert(txn->mt_env != NULL);
1033 DPRINTF("writing meta page %d for root page %lu",
1034 !F_ISSET(txn->mt_flags, MDB_TXN_METOGGLE), txn->mt_dbs[MAIN_DBI].md_root);
1038 ptr = (char *)&meta;
1039 off = offsetof(MDB_meta, mm_dbs[0].md_depth);
1040 len = sizeof(MDB_meta) - off;
1043 meta.mm_dbs[0] = txn->mt_dbs[0];
1044 meta.mm_dbs[1] = txn->mt_dbs[1];
1045 meta.mm_last_pg = txn->mt_next_pgno - 1;
1046 meta.mm_txnid = txn->mt_txnid;
1048 if (!F_ISSET(txn->mt_flags, MDB_TXN_METOGGLE))
1049 off += env->me_psize;
1052 lseek(env->me_fd, off, SEEK_SET);
1053 rc = write(env->me_fd, ptr, len);
1055 DPRINTF("write failed, disk error?");
1063 mdbenv_read_meta(MDB_env *env, int *which)
1067 assert(env != NULL);
1069 if (env->me_metas[0]->mm_txnid < env->me_metas[1]->mm_txnid)
1072 if (env->me_meta != env->me_metas[toggle])
1073 env->me_meta = env->me_metas[toggle];
1077 DPRINTF("Using meta page %d", toggle);
1083 mdbenv_create(MDB_env **env)
1087 e = calloc(1, sizeof(MDB_env));
1088 if (!e) return ENOMEM;
1090 e->me_maxreaders = DEFAULT_READERS;
1099 mdbenv_set_mapsize(MDB_env *env, size_t size)
1103 env->me_mapsize = size;
1108 mdbenv_set_maxdbs(MDB_env *env, int dbs)
1110 env->me_maxdbs = dbs;
1115 mdbenv_set_maxreaders(MDB_env *env, int readers)
1117 env->me_maxreaders = readers;
1122 mdbenv_get_maxreaders(MDB_env *env, int *readers)
1124 if (!env || !readers)
1126 *readers = env->me_maxreaders;
1131 mdbenv_open2(MDB_env *env, unsigned int flags)
1137 env->me_flags = flags;
1139 memset(&meta, 0, sizeof(meta));
1141 if ((i = mdbenv_read_header(env, &meta)) != 0) {
1144 DPRINTF("new mdbenv");
1148 if (!env->me_mapsize) {
1149 env->me_mapsize = newenv ? DEFAULT_MAPSIZE : meta.mm_mapsize;
1153 if (meta.mm_address && (flags & MDB_FIXEDMAP))
1155 env->me_map = mmap(meta.mm_address, env->me_mapsize, PROT_READ, i,
1157 if (env->me_map == MAP_FAILED)
1161 meta.mm_mapsize = env->me_mapsize;
1162 if (flags & MDB_FIXEDMAP)
1163 meta.mm_address = env->me_map;
1164 i = mdbenv_init_meta(env, &meta);
1165 if (i != MDB_SUCCESS) {
1166 munmap(env->me_map, env->me_mapsize);
1170 env->me_psize = meta.mm_psize;
1172 p = (MDB_page *)(MDB_page *)(MDB_page *)(MDB_page *)(MDB_page *)(MDB_page *)(MDB_page *)(MDB_page *)(MDB_page *)env->me_map;
1173 env->me_metas[0] = METADATA(p);
1174 env->me_metas[1] = (MDB_meta *)((char *)env->me_metas[0] + meta.mm_psize);
1176 if ((i = mdbenv_read_meta(env, NULL)) != 0)
1179 DPRINTF("opened database version %u, pagesize %u",
1180 env->me_meta->mm_version, env->me_psize);
1181 DPRINTF("depth: %u", env->me_meta->mm_dbs[MAIN_DBI].md_depth);
1182 DPRINTF("entries: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_entries);
1183 DPRINTF("branch pages: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_branch_pages);
1184 DPRINTF("leaf pages: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_leaf_pages);
1185 DPRINTF("overflow pages: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_overflow_pages);
1186 DPRINTF("root: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_root);
1192 mdbenv_reader_dest(void *ptr)
1194 MDB_reader *reader = ptr;
1196 reader->mr_txnid = 0;
1201 /* downgrade the exclusive lock on the region back to shared */
1203 mdbenv_share_locks(MDB_env *env)
1205 struct flock lock_info;
1207 env->me_txns->mt_txnid = env->me_meta->mm_txnid;
1209 memset((void *)&lock_info, 0, sizeof(lock_info));
1210 lock_info.l_type = F_RDLCK;
1211 lock_info.l_whence = SEEK_SET;
1212 lock_info.l_start = 0;
1213 lock_info.l_len = 1;
1214 fcntl(env->me_lfd, F_SETLK, &lock_info);
1218 mdbenv_setup_locks(MDB_env *env, char *lpath, int mode, int *excl)
1222 struct flock lock_info;
1226 if ((env->me_lfd = open(lpath, O_RDWR|O_CREAT, mode)) == -1) {
1230 /* Try to get exclusive lock. If we succeed, then
1231 * nobody is using the lock region and we should initialize it.
1233 memset((void *)&lock_info, 0, sizeof(lock_info));
1234 lock_info.l_type = F_WRLCK;
1235 lock_info.l_whence = SEEK_SET;
1236 lock_info.l_start = 0;
1237 lock_info.l_len = 1;
1238 rc = fcntl(env->me_lfd, F_SETLK, &lock_info);
1242 lock_info.l_type = F_RDLCK;
1243 rc = fcntl(env->me_lfd, F_SETLK, &lock_info);
1249 size = lseek(env->me_lfd, 0, SEEK_END);
1250 rsize = (env->me_maxreaders-1) * sizeof(MDB_reader) + sizeof(MDB_txninfo);
1251 if (size < rsize && *excl) {
1252 if (ftruncate(env->me_lfd, rsize) != 0) {
1258 size = rsize - sizeof(MDB_txninfo);
1259 env->me_maxreaders = size/sizeof(MDB_reader) + 1;
1261 env->me_txns = mmap(0, rsize, PROT_READ|PROT_WRITE, MAP_SHARED,
1263 if (env->me_txns == MAP_FAILED) {
1268 pthread_mutexattr_t mattr;
1270 pthread_mutexattr_init(&mattr);
1271 pthread_mutexattr_setpshared(&mattr, PTHREAD_PROCESS_SHARED);
1272 pthread_mutex_init(&env->me_txns->mt_mutex, &mattr);
1273 pthread_mutex_init(&env->me_txns->mt_wmutex, &mattr);
1274 env->me_txns->mt_version = MDB_VERSION;
1275 env->me_txns->mt_magic = MDB_MAGIC;
1276 env->me_txns->mt_txnid = 0;
1277 env->me_txns->mt_numreaders = 0;
1280 if (env->me_txns->mt_magic != MDB_MAGIC) {
1281 DPRINTF("lock region has invalid magic");
1285 if (env->me_txns->mt_version != MDB_VERSION) {
1286 DPRINTF("lock region is version %u, expected version %u",
1287 env->me_txns->mt_version, MDB_VERSION);
1288 rc = MDB_VERSION_MISMATCH;
1291 if (errno != EACCES && errno != EAGAIN) {
1304 #define LOCKNAME "/lock.mdb"
1305 #define DATANAME "/data.mdb"
1307 mdbenv_open(MDB_env *env, const char *path, unsigned int flags, mode_t mode)
1309 int oflags, rc, len, excl;
1310 char *lpath, *dpath;
1313 lpath = malloc(len + sizeof(LOCKNAME) + len + sizeof(DATANAME));
1316 dpath = lpath + len + sizeof(LOCKNAME);
1317 sprintf(lpath, "%s" LOCKNAME, path);
1318 sprintf(dpath, "%s" DATANAME, path);
1320 rc = mdbenv_setup_locks(env, lpath, mode, &excl);
1324 if (F_ISSET(flags, MDB_RDONLY))
1327 oflags = O_RDWR | O_CREAT;
1329 if ((env->me_fd = open(dpath, oflags, mode)) == -1)
1332 if ((rc = mdbenv_open2(env, flags)) != MDB_SUCCESS) {
1336 env->me_path = strdup(path);
1337 DPRINTF("opened dbenv %p", (void *) env);
1338 pthread_key_create(&env->me_txkey, mdbenv_reader_dest);
1340 mdbenv_share_locks(env);
1341 env->me_dbxs = calloc(env->me_maxdbs, sizeof(MDB_dbx));
1342 env->me_dbs[0] = calloc(env->me_maxdbs, sizeof(MDB_db));
1343 env->me_dbs[1] = calloc(env->me_maxdbs, sizeof(MDB_db));
1353 mdbenv_close(MDB_env *env)
1358 free(env->me_dbs[1]);
1359 free(env->me_dbs[0]);
1364 munmap(env->me_map, env->me_mapsize);
1368 size_t size = (env->me_maxreaders-1) * sizeof(MDB_reader) + sizeof(MDB_txninfo);
1369 munmap(env->me_txns, size);
1375 /* Search for key within a leaf page, using binary search.
1376 * Returns the smallest entry larger or equal to the key.
1377 * If exactp is non-null, stores whether the found entry was an exact match
1378 * in *exactp (1 or 0).
1379 * If kip is non-null, stores the index of the found entry in *kip.
1380 * If no entry larger or equal to the key is found, returns NULL.
1383 mdb_search_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp, MDB_val *key,
1384 int *exactp, unsigned int *kip)
1392 DPRINTF("searching %u keys in %s page %lu",
1394 IS_LEAF(mp) ? "leaf" : "branch",
1397 assert(NUMKEYS(mp) > 0);
1399 memset(&nodekey, 0, sizeof(nodekey));
1401 low = IS_LEAF(mp) ? 0 : 1;
1402 high = NUMKEYS(mp) - 1;
1403 while (low <= high) {
1404 i = (low + high) >> 1;
1405 node = NODEPTR(mp, i);
1407 nodekey.mv_size = node->mn_ksize;
1408 nodekey.mv_data = NODEKEY(node);
1410 if (txn->mt_dbxs[dbi].md_cmp)
1411 rc = txn->mt_dbxs[dbi].md_cmp(key, &nodekey);
1413 rc = _mdb_cmp(txn, dbi, key, &nodekey);
1416 DPRINTF("found leaf index %u [%.*s], rc = %i",
1417 i, (int)nodekey.mv_size, (char *)nodekey.mv_data, rc);
1419 DPRINTF("found branch index %u [%.*s -> %lu], rc = %i",
1420 i, (int)node->mn_ksize, (char *)NODEKEY(node),
1431 if (rc > 0) { /* Found entry is less than the key. */
1432 i++; /* Skip to get the smallest entry larger than key. */
1433 if (i >= NUMKEYS(mp))
1434 /* There is no entry larger or equal to the key. */
1438 *exactp = (rc == 0);
1439 if (kip) /* Store the key index if requested. */
1442 return NODEPTR(mp, i);
1446 cursor_pop_page(MDB_cursor *cursor)
1450 top = CURSOR_TOP(cursor);
1453 DPRINTF("popped page %lu off cursor %p", top->mp_page->mp_pgno, (void *) cursor);
1459 cursor_push_page(MDB_cursor *cursor, MDB_page *mp)
1463 DPRINTF("pushing page %lu on cursor %p", mp->mp_pgno, (void *) cursor);
1465 if ((ppage = calloc(1, sizeof(MDB_ppage))) == NULL)
1467 ppage->mp_page = mp;
1468 CURSOR_PUSH(cursor, ppage);
1473 mdb_get_page(MDB_txn *txn, pgno_t pgno)
1478 if (!F_ISSET(txn->mt_flags, MDB_TXN_RDONLY) && !STAILQ_EMPTY(txn->mt_u.dirty_queue)) {
1480 STAILQ_FOREACH(dp, txn->mt_u.dirty_queue, h.md_next) {
1481 if (dp->p.mp_pgno == pgno) {
1489 if (pgno > txn->mt_env->me_meta->mm_last_pg)
1491 p = (MDB_page *)(txn->mt_env->me_map + txn->mt_env->me_psize * pgno);
1497 mdb_search_page_root(MDB_txn *txn, MDB_dbi dbi, MDB_val *key,
1498 MDB_cursor *cursor, int modify, MDB_pageparent *mpp)
1500 MDB_page *mp = mpp->mp_page;
1503 if (cursor && cursor_push_page(cursor, mp) == NULL)
1506 while (IS_BRANCH(mp)) {
1510 DPRINTF("branch page %lu has %u keys", mp->mp_pgno, NUMKEYS(mp));
1511 assert(NUMKEYS(mp) > 1);
1512 DPRINTF("found index 0 to page %lu", NODEPGNO(NODEPTR(mp, 0)));
1514 if (key == NULL) /* Initialize cursor to first page. */
1516 else if (key->mv_size > MAXKEYSIZE && key->mv_data == NULL) {
1517 /* cursor to last page */
1521 node = mdb_search_node(txn, dbi, mp, key, &exact, &i);
1523 i = NUMKEYS(mp) - 1;
1531 DPRINTF("following index %u for key %.*s",
1532 i, (int)key->mv_size, (char *)key->mv_data);
1533 assert(i < NUMKEYS(mp));
1534 node = NODEPTR(mp, i);
1537 CURSOR_TOP(cursor)->mp_ki = i;
1539 mpp->mp_parent = mp;
1540 if ((mp = mdb_get_page(txn, NODEPGNO(node))) == NULL)
1545 if (cursor && cursor_push_page(cursor, mp) == NULL)
1549 MDB_dhead *dh = ((MDB_dhead *)mp)-1;
1550 if ((rc = mdb_touch(txn, mpp)) != 0)
1552 dh = ((MDB_dhead *)mpp->mp_page)-1;
1553 dh->md_parent = mpp->mp_parent;
1554 dh->md_pi = mpp->mp_pi;
1561 DPRINTF("internal error, index points to a %02X page!?",
1566 DPRINTF("found leaf page %lu for key %.*s", mp->mp_pgno,
1567 key ? (int)key->mv_size : 0, key ? (char *)key->mv_data : NULL);
1572 /* Search for the page a given key should be in.
1573 * Stores a pointer to the found page in *mpp.
1574 * If key is NULL, search for the lowest page (used by mdb_cursor_first).
1575 * If cursor is non-null, pushes parent pages on the cursor stack.
1576 * If modify is true, visited pages are updated with new page numbers.
1579 mdb_search_page(MDB_txn *txn, MDB_dbi dbi, MDB_val *key,
1580 MDB_cursor *cursor, int modify, MDB_pageparent *mpp)
1585 /* Choose which root page to start with. If a transaction is given
1586 * use the root page from the transaction, otherwise read the last
1587 * committed root page.
1589 if (F_ISSET(txn->mt_flags, MDB_TXN_ERROR)) {
1590 DPRINTF("transaction has failed, must abort");
1593 root = txn->mt_dbs[dbi].md_root;
1595 if (root == P_INVALID) { /* Tree is empty. */
1596 DPRINTF("tree is empty");
1597 return MDB_NOTFOUND;
1600 if ((mpp->mp_page = mdb_get_page(txn, root)) == NULL)
1603 DPRINTF("root page has flags 0x%X", mpp->mp_page->mp_flags);
1606 /* For sub-databases, update main root first */
1607 if (dbi > MAIN_DBI && !txn->mt_dbxs[dbi].md_dirty) {
1609 rc = mdb_search_page(txn, MAIN_DBI, &txn->mt_dbxs[dbi].md_name,
1613 txn->mt_dbxs[dbi].md_dirty = 1;
1615 if (!F_ISSET(mpp->mp_page->mp_flags, P_DIRTY)) {
1616 mpp->mp_parent = NULL;
1618 if ((rc = mdb_touch(txn, mpp)))
1620 txn->mt_dbs[dbi].md_root = mpp->mp_page->mp_pgno;
1624 return mdb_search_page_root(txn, dbi, key, cursor, modify, mpp);
1628 mdb_read_data(MDB_txn *txn, MDB_node *leaf, MDB_val *data)
1630 MDB_page *omp; /* overflow mpage */
1633 if (!F_ISSET(leaf->mn_flags, F_BIGDATA)) {
1634 data->mv_size = leaf->mn_dsize;
1635 data->mv_data = NODEDATA(leaf);
1639 /* Read overflow data.
1641 data->mv_size = leaf->mn_dsize;
1642 memcpy(&pgno, NODEDATA(leaf), sizeof(pgno));
1643 if ((omp = mdb_get_page(txn, pgno)) == NULL) {
1644 DPRINTF("read overflow page %lu failed", pgno);
1647 data->mv_data = omp;
1653 mdb_get(MDB_txn *txn, MDB_dbi dbi,
1654 MDB_val *key, MDB_val *data)
1662 DPRINTF("===> get key [%.*s]", (int)key->mv_size, (char *)key->mv_data);
1664 if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
1667 if (key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
1671 if ((rc = mdb_search_page(txn, dbi, key, NULL, 0, &mpp)) != MDB_SUCCESS)
1674 leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, NULL);
1675 if (leaf && exact) {
1676 /* Return first duplicate data item */
1677 if (F_ISSET(txn->mt_dbs[dbi].md_flags, MDB_DUPSORT)) {
1680 mdb_xcursor_init0(txn, dbi, &mx);
1681 mdb_xcursor_init1(txn, dbi, &mx, leaf);
1682 rc = mdb_search_page(&mx.mx_txn, mx.mx_cursor.mc_dbi, NULL, NULL, 0, &mpp);
1683 if (rc != MDB_SUCCESS)
1685 leaf = NODEPTR(mpp.mp_page, 0);
1687 rc = mdb_read_data(txn, leaf, data);
1696 mdb_sibling(MDB_cursor *cursor, int move_right)
1700 MDB_ppage *parent, *top;
1703 top = CURSOR_TOP(cursor);
1704 if ((parent = SLIST_NEXT(top, mp_entry)) == NULL) {
1705 return MDB_NOTFOUND; /* root has no siblings */
1708 DPRINTF("parent page is page %lu, index %u",
1709 parent->mp_page->mp_pgno, parent->mp_ki);
1711 cursor_pop_page(cursor);
1712 if (move_right ? (parent->mp_ki + 1 >= NUMKEYS(parent->mp_page))
1713 : (parent->mp_ki == 0)) {
1714 DPRINTF("no more keys left, moving to %s sibling",
1715 move_right ? "right" : "left");
1716 if ((rc = mdb_sibling(cursor, move_right)) != MDB_SUCCESS)
1718 parent = CURSOR_TOP(cursor);
1724 DPRINTF("just moving to %s index key %u",
1725 move_right ? "right" : "left", parent->mp_ki);
1727 assert(IS_BRANCH(parent->mp_page));
1729 indx = NODEPTR(parent->mp_page, parent->mp_ki);
1730 if ((mp = mdb_get_page(cursor->mc_txn, indx->mn_pgno)) == NULL)
1733 mp->parent = parent->mp_page;
1734 mp->parent_index = parent->mp_ki;
1737 cursor_push_page(cursor, mp);
1743 mdb_set_key(MDB_node *node, MDB_val *key)
1748 key->mv_size = node->mn_ksize;
1749 key->mv_data = NODEKEY(node);
1755 mdb_cursor_next(MDB_cursor *cursor, MDB_val *key, MDB_val *data, MDB_cursor_op op)
1762 if (cursor->mc_eof) {
1763 return MDB_NOTFOUND;
1766 assert(cursor->mc_initialized);
1768 if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
1769 if (op == MDB_NEXT || op == MDB_NEXT_DUP) {
1770 rc = mdb_cursor_next(&cursor->mc_xcursor->mx_cursor, data, NULL, MDB_NEXT);
1771 if (op != MDB_NEXT || rc == MDB_SUCCESS)
1776 top = CURSOR_TOP(cursor);
1779 DPRINTF("cursor_next: top page is %lu in cursor %p", mp->mp_pgno, (void *) cursor);
1781 if (top->mp_ki + 1 >= NUMKEYS(mp)) {
1782 DPRINTF("=====> move to next sibling page");
1783 if (mdb_sibling(cursor, 1) != MDB_SUCCESS) {
1785 return MDB_NOTFOUND;
1787 top = CURSOR_TOP(cursor);
1789 DPRINTF("next page is %lu, key index %u", mp->mp_pgno, top->mp_ki);
1793 DPRINTF("==> cursor points to page %lu with %u keys, key index %u",
1794 mp->mp_pgno, NUMKEYS(mp), top->mp_ki);
1796 assert(IS_LEAF(mp));
1797 leaf = NODEPTR(mp, top->mp_ki);
1800 if ((rc = mdb_read_data(cursor->mc_txn, leaf, data) != MDB_SUCCESS))
1803 if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
1804 mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, leaf);
1805 rc = mdb_cursor_first(&cursor->mc_xcursor->mx_cursor, data, NULL);
1806 if (rc != MDB_SUCCESS)
1811 return mdb_set_key(leaf, key);
1815 mdb_cursor_prev(MDB_cursor *cursor, MDB_val *key, MDB_val *data, MDB_cursor_op op)
1822 assert(cursor->mc_initialized);
1824 if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
1825 if (op == MDB_PREV || op == MDB_PREV_DUP) {
1826 rc = mdb_cursor_prev(&cursor->mc_xcursor->mx_cursor, data, NULL, MDB_PREV);
1827 if (op != MDB_PREV || rc == MDB_SUCCESS)
1832 top = CURSOR_TOP(cursor);
1835 DPRINTF("cursor_prev: top page is %lu in cursor %p", mp->mp_pgno, (void *) cursor);
1837 if (top->mp_ki == 0) {
1838 DPRINTF("=====> move to prev sibling page");
1839 if (mdb_sibling(cursor, 0) != MDB_SUCCESS) {
1840 return MDB_NOTFOUND;
1842 top = CURSOR_TOP(cursor);
1844 top->mp_ki = NUMKEYS(mp) - 1;
1845 DPRINTF("prev page is %lu, key index %u", mp->mp_pgno, top->mp_ki);
1851 DPRINTF("==> cursor points to page %lu with %u keys, key index %u",
1852 mp->mp_pgno, NUMKEYS(mp), top->mp_ki);
1854 assert(IS_LEAF(mp));
1855 leaf = NODEPTR(mp, top->mp_ki);
1858 if ((rc = mdb_read_data(cursor->mc_txn, leaf, data) != MDB_SUCCESS))
1861 if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
1862 mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, leaf);
1863 rc = mdb_cursor_last(&cursor->mc_xcursor->mx_cursor, data, NULL);
1864 if (rc != MDB_SUCCESS)
1869 return mdb_set_key(leaf, key);
1873 mdb_cursor_set(MDB_cursor *cursor, MDB_val *key, MDB_val *data,
1874 MDB_cursor_op op, int *exactp)
1883 assert(key->mv_size > 0);
1885 while (CURSOR_TOP(cursor) != NULL)
1886 cursor_pop_page(cursor);
1888 rc = mdb_search_page(cursor->mc_txn, cursor->mc_dbi, key, cursor, 0, &mpp);
1889 if (rc != MDB_SUCCESS)
1891 assert(IS_LEAF(mpp.mp_page));
1893 top = CURSOR_TOP(cursor);
1894 leaf = mdb_search_node(cursor->mc_txn, cursor->mc_dbi, mpp.mp_page, key, exactp, &top->mp_ki);
1895 if (exactp != NULL && !*exactp) {
1896 /* MDB_SET specified and not an exact match. */
1897 return MDB_NOTFOUND;
1901 DPRINTF("===> inexact leaf not found, goto sibling");
1902 if ((rc = mdb_sibling(cursor, 1)) != MDB_SUCCESS)
1903 return rc; /* no entries matched */
1904 top = CURSOR_TOP(cursor);
1906 mpp.mp_page = top->mp_page;
1907 assert(IS_LEAF(mpp.mp_page));
1908 leaf = NODEPTR(mpp.mp_page, 0);
1911 cursor->mc_initialized = 1;
1915 if ((rc = mdb_read_data(cursor->mc_txn, leaf, data)) != MDB_SUCCESS)
1918 if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
1919 mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, leaf);
1920 if (op == MDB_SET || op == MDB_SET_RANGE) {
1921 rc = mdb_cursor_first(&cursor->mc_xcursor->mx_cursor, data, NULL);
1925 if (op == MDB_GET_BOTH) {
1930 op2 = MDB_SET_RANGE;
1932 rc = mdb_cursor_set(&cursor->mc_xcursor->mx_cursor, data, NULL, op2, ex2p);
1933 if (rc != MDB_SUCCESS)
1939 rc = mdb_set_key(leaf, key);
1940 if (rc == MDB_SUCCESS) {
1941 DPRINTF("==> cursor placed on key %.*s",
1942 (int)key->mv_size, (char *)key->mv_data);
1950 mdb_cursor_first(MDB_cursor *cursor, MDB_val *key, MDB_val *data)
1956 while (CURSOR_TOP(cursor) != NULL)
1957 cursor_pop_page(cursor);
1959 rc = mdb_search_page(cursor->mc_txn, cursor->mc_dbi, NULL, cursor, 0, &mpp);
1960 if (rc != MDB_SUCCESS)
1962 assert(IS_LEAF(mpp.mp_page));
1964 leaf = NODEPTR(mpp.mp_page, 0);
1965 cursor->mc_initialized = 1;
1969 if ((rc = mdb_read_data(cursor->mc_txn, leaf, data)) != MDB_SUCCESS)
1972 if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
1973 mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, leaf);
1974 rc = mdb_cursor_first(&cursor->mc_xcursor->mx_cursor, data, NULL);
1979 return mdb_set_key(leaf, key);
1983 mdb_cursor_last(MDB_cursor *cursor, MDB_val *key, MDB_val *data)
1991 while (CURSOR_TOP(cursor) != NULL)
1992 cursor_pop_page(cursor);
1994 lkey.mv_size = MAXKEYSIZE+1;
1995 lkey.mv_data = NULL;
1997 rc = mdb_search_page(cursor->mc_txn, cursor->mc_dbi, &lkey, cursor, 0, &mpp);
1998 if (rc != MDB_SUCCESS)
2000 assert(IS_LEAF(mpp.mp_page));
2002 leaf = NODEPTR(mpp.mp_page, NUMKEYS(mpp.mp_page)-1);
2003 cursor->mc_initialized = 1;
2006 top = CURSOR_TOP(cursor);
2007 top->mp_ki = NUMKEYS(top->mp_page) - 1;
2010 if ((rc = mdb_read_data(cursor->mc_txn, leaf, data)) != MDB_SUCCESS)
2013 if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
2014 mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, leaf);
2015 rc = mdb_cursor_last(&cursor->mc_xcursor->mx_cursor, data, NULL);
2021 return mdb_set_key(leaf, key);
2025 mdb_cursor_get(MDB_cursor *cursor, MDB_val *key, MDB_val *data,
2035 case MDB_GET_BOTH_RANGE:
2043 if (key == NULL || key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
2045 } else if (op != MDB_SET_RANGE)
2046 rc = mdb_cursor_set(cursor, key, data, op, NULL);
2048 rc = mdb_cursor_set(cursor, key, data, op, &exact);
2052 case MDB_NEXT_NODUP:
2053 if (!cursor->mc_initialized)
2054 rc = mdb_cursor_first(cursor, key, data);
2056 rc = mdb_cursor_next(cursor, key, data, op);
2060 case MDB_PREV_NODUP:
2061 if (!cursor->mc_initialized || cursor->mc_eof)
2062 rc = mdb_cursor_last(cursor, key, data);
2064 rc = mdb_cursor_prev(cursor, key, data, op);
2067 rc = mdb_cursor_first(cursor, key, data);
2070 rc = mdb_cursor_last(cursor, key, data);
2073 DPRINTF("unhandled/unimplemented cursor operation %u", op);
2081 /* Allocate a page and initialize it
2084 mdb_new_page(MDB_txn *txn, MDB_dbi dbi, uint32_t flags, int num)
2088 if ((dp = mdb_alloc_page(txn, NULL, 0, num)) == NULL)
2090 DPRINTF("allocated new mpage %lu, page size %u",
2091 dp->p.mp_pgno, txn->mt_env->me_psize);
2092 dp->p.mp_flags = flags | P_DIRTY;
2093 dp->p.mp_lower = PAGEHDRSZ;
2094 dp->p.mp_upper = txn->mt_env->me_psize;
2096 if (IS_BRANCH(&dp->p))
2097 txn->mt_dbs[dbi].md_branch_pages++;
2098 else if (IS_LEAF(&dp->p))
2099 txn->mt_dbs[dbi].md_leaf_pages++;
2100 else if (IS_OVERFLOW(&dp->p)) {
2101 txn->mt_dbs[dbi].md_overflow_pages += num;
2102 dp->p.mp_pages = num;
2109 mdb_leaf_size(MDB_env *env, MDB_val *key, MDB_val *data)
2113 sz = LEAFSIZE(key, data);
2114 if (data->mv_size >= env->me_psize / MDB_MINKEYS) {
2115 /* put on overflow page */
2116 sz -= data->mv_size - sizeof(pgno_t);
2119 return sz + sizeof(indx_t);
2123 mdb_branch_size(MDB_env *env, MDB_val *key)
2128 if (sz >= env->me_psize / MDB_MINKEYS) {
2129 /* put on overflow page */
2130 /* not implemented */
2131 /* sz -= key->size - sizeof(pgno_t); */
2134 return sz + sizeof(indx_t);
2138 mdb_add_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp, indx_t indx,
2139 MDB_val *key, MDB_val *data, pgno_t pgno, uint8_t flags)
2142 size_t node_size = NODESIZE;
2145 MDB_dpage *ofp = NULL; /* overflow page */
2147 assert(mp->mp_upper >= mp->mp_lower);
2149 DPRINTF("add node [%.*s] to %s page %lu at index %i, key size %zu",
2150 key ? (int)key->mv_size : 0, key ? (char *)key->mv_data : NULL,
2151 IS_LEAF(mp) ? "leaf" : "branch",
2152 mp->mp_pgno, indx, key ? key->mv_size : 0);
2155 node_size += key->mv_size;
2159 if (F_ISSET(flags, F_BIGDATA)) {
2160 /* Data already on overflow page. */
2161 node_size += sizeof(pgno_t);
2162 } else if (data->mv_size >= txn->mt_env->me_psize / MDB_MINKEYS) {
2163 int ovpages = OVPAGES(data->mv_size, txn->mt_env->me_psize);
2164 /* Put data on overflow page. */
2165 DPRINTF("data size is %zu, put on overflow page",
2167 node_size += sizeof(pgno_t);
2168 if ((ofp = mdb_new_page(txn, dbi, P_OVERFLOW, ovpages)) == NULL)
2170 DPRINTF("allocated overflow page %lu", ofp->p.mp_pgno);
2173 node_size += data->mv_size;
2177 if (node_size + sizeof(indx_t) > SIZELEFT(mp)) {
2178 DPRINTF("not enough room in page %lu, got %u ptrs",
2179 mp->mp_pgno, NUMKEYS(mp));
2180 DPRINTF("upper - lower = %u - %u = %u", mp->mp_upper, mp->mp_lower,
2181 mp->mp_upper - mp->mp_lower);
2182 DPRINTF("node size = %zu", node_size);
2186 /* Move higher pointers up one slot. */
2187 for (i = NUMKEYS(mp); i > indx; i--)
2188 mp->mp_ptrs[i] = mp->mp_ptrs[i - 1];
2190 /* Adjust free space offsets. */
2191 ofs = mp->mp_upper - node_size;
2192 assert(ofs >= mp->mp_lower + sizeof(indx_t));
2193 mp->mp_ptrs[indx] = ofs;
2195 mp->mp_lower += sizeof(indx_t);
2197 /* Write the node data. */
2198 node = NODEPTR(mp, indx);
2199 node->mn_ksize = (key == NULL) ? 0 : key->mv_size;
2200 node->mn_flags = flags;
2202 node->mn_dsize = data->mv_size;
2204 node->mn_pgno = pgno;
2207 memcpy(NODEKEY(node), key->mv_data, key->mv_size);
2212 if (F_ISSET(flags, F_BIGDATA))
2213 memcpy(node->mn_data + key->mv_size, data->mv_data,
2216 memcpy(node->mn_data + key->mv_size, data->mv_data,
2219 memcpy(node->mn_data + key->mv_size, &ofp->p.mp_pgno,
2221 memcpy(METADATA(&ofp->p), data->mv_data, data->mv_size);
2229 mdb_del_node(MDB_page *mp, indx_t indx)
2232 indx_t i, j, numkeys, ptr;
2236 DPRINTF("delete node %u on %s page %lu", indx,
2237 IS_LEAF(mp) ? "leaf" : "branch", mp->mp_pgno);
2238 assert(indx < NUMKEYS(mp));
2240 node = NODEPTR(mp, indx);
2241 sz = NODESIZE + node->mn_ksize;
2243 if (F_ISSET(node->mn_flags, F_BIGDATA))
2244 sz += sizeof(pgno_t);
2246 sz += NODEDSZ(node);
2249 ptr = mp->mp_ptrs[indx];
2250 numkeys = NUMKEYS(mp);
2251 for (i = j = 0; i < numkeys; i++) {
2253 mp->mp_ptrs[j] = mp->mp_ptrs[i];
2254 if (mp->mp_ptrs[i] < ptr)
2255 mp->mp_ptrs[j] += sz;
2260 base = (char *)mp + mp->mp_upper;
2261 memmove(base + sz, base, ptr - mp->mp_upper);
2263 mp->mp_lower -= sizeof(indx_t);
2268 mdb_xcursor_init0(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx)
2273 mx->mx_txn.mt_dbxs = mx->mx_dbxs;
2274 mx->mx_txn.mt_dbs = mx->mx_dbs;
2275 mx->mx_dbxs[0] = txn->mt_dbxs[0];
2276 mx->mx_dbxs[1] = txn->mt_dbxs[1];
2278 mx->mx_dbxs[2] = txn->mt_dbxs[dbi];
2283 mx->mx_dbxs[dbn+1].md_parent = dbn;
2284 mx->mx_dbxs[dbn+1].md_cmp = mx->mx_dbxs[dbn].md_dcmp;
2285 mx->mx_dbxs[dbn+1].md_rel = mx->mx_dbxs[dbn].md_rel;
2286 mx->mx_dbxs[dbn+1].md_dirty = 0;
2287 mx->mx_txn.mt_numdbs = dbn+2;
2289 SLIST_INIT(&mx->mx_cursor.mc_stack);
2290 mx->mx_cursor.mc_txn = &mx->mx_txn;
2291 mx->mx_cursor.mc_dbi = dbn+1;
2295 mdb_xcursor_init1(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx, MDB_node *node)
2297 MDB_db *db = NODEDATA(node);
2299 mx->mx_dbs[0] = txn->mt_dbs[0];
2300 mx->mx_dbs[1] = txn->mt_dbs[1];
2302 mx->mx_dbs[2] = txn->mt_dbs[dbi];
2307 mx->mx_dbs[dbn] = *db;
2308 mx->mx_dbxs[dbn].md_name.mv_data = NODEKEY(node);
2309 mx->mx_dbxs[dbn].md_name.mv_size = node->mn_ksize;
2310 mx->mx_txn.mt_next_pgno = txn->mt_next_pgno;
2311 mx->mx_txn.mt_oldest = txn->mt_oldest;
2312 mx->mx_txn.mt_u = txn->mt_u;
2316 mdb_xcursor_fini(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx)
2318 txn->mt_next_pgno = mx->mx_txn.mt_next_pgno;
2319 txn->mt_oldest = mx->mx_txn.mt_oldest;
2320 txn->mt_u = mx->mx_txn.mt_u;
2321 txn->mt_dbs[0] = mx->mx_dbs[0];
2322 txn->mt_dbs[1] = mx->mx_dbs[1];
2323 txn->mt_dbxs[0].md_dirty = mx->mx_dbxs[0].md_dirty;
2324 txn->mt_dbxs[1].md_dirty = mx->mx_dbxs[1].md_dirty;
2326 txn->mt_dbs[dbi] = mx->mx_dbs[2];
2327 txn->mt_dbxs[dbi].md_dirty = mx->mx_dbxs[2].md_dirty;
2332 mdb_cursor_open(MDB_txn *txn, MDB_dbi dbi, MDB_cursor **ret)
2335 size_t size = sizeof(MDB_cursor);
2337 if (txn == NULL || ret == NULL || !dbi || dbi >= txn->mt_numdbs)
2340 if (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT)
2341 size += sizeof(MDB_xcursor);
2343 if ((cursor = calloc(1, size)) != NULL) {
2344 SLIST_INIT(&cursor->mc_stack);
2345 cursor->mc_dbi = dbi;
2346 cursor->mc_txn = txn;
2347 if (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT) {
2348 MDB_xcursor *mx = (MDB_xcursor *)(cursor + 1);
2349 cursor->mc_xcursor = mx;
2350 mdb_xcursor_init0(txn, dbi, mx);
2361 /* Return the count of duplicate data items for the current key */
2363 mdb_cursor_count(MDB_cursor *mc, unsigned long *countp)
2365 if (mc == NULL || countp == NULL)
2368 if (!(mc->mc_txn->mt_dbs[mc->mc_dbi].md_flags & MDB_DUPSORT))
2371 if (!mc->mc_xcursor->mx_cursor.mc_initialized)
2374 *countp = mc->mc_xcursor->mx_txn.mt_dbs[mc->mc_xcursor->mx_cursor.mc_dbi].md_entries;
2379 mdb_cursor_close(MDB_cursor *cursor)
2381 if (cursor != NULL) {
2382 while(!CURSOR_EMPTY(cursor))
2383 cursor_pop_page(cursor);
2384 if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
2385 mdb_xcursor_fini(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor);
2386 while(!CURSOR_EMPTY(&cursor->mc_xcursor->mx_cursor))
2387 cursor_pop_page(&cursor->mc_xcursor->mx_cursor);
2395 mdb_update_key(MDB_page *mp, indx_t indx, MDB_val *key)
2397 indx_t ptr, i, numkeys;
2403 node = NODEPTR(mp, indx);
2404 ptr = mp->mp_ptrs[indx];
2405 DPRINTF("update key %u (ofs %u) [%.*s] to [%.*s] on page %lu",
2407 (int)node->mn_ksize, (char *)NODEKEY(node),
2408 (int)key->mv_size, (char *)key->mv_data,
2411 delta = key->mv_size - node->mn_ksize;
2413 if (delta > 0 && SIZELEFT(mp) < delta) {
2414 DPRINTF("OUCH! Not enough room, delta = %d", delta);
2418 numkeys = NUMKEYS(mp);
2419 for (i = 0; i < numkeys; i++) {
2420 if (mp->mp_ptrs[i] <= ptr)
2421 mp->mp_ptrs[i] -= delta;
2424 base = (char *)mp + mp->mp_upper;
2425 len = ptr - mp->mp_upper + NODESIZE;
2426 memmove(base - delta, base, len);
2427 mp->mp_upper -= delta;
2429 node = NODEPTR(mp, indx);
2430 node->mn_ksize = key->mv_size;
2433 memcpy(NODEKEY(node), key->mv_data, key->mv_size);
2438 /* Move a node from src to dst.
2441 mdb_move_node(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *src, indx_t srcindx,
2442 MDB_pageparent *dst, indx_t dstindx)
2448 srcnode = NODEPTR(src->mp_page, srcindx);
2449 DPRINTF("moving %s node %u [%.*s] on page %lu to node %u on page %lu",
2450 IS_LEAF(src->mp_page) ? "leaf" : "branch",
2452 (int)srcnode->mn_ksize, (char *)NODEKEY(srcnode),
2453 src->mp_page->mp_pgno,
2454 dstindx, dst->mp_page->mp_pgno);
2456 /* Mark src and dst as dirty. */
2457 if ((rc = mdb_touch(txn, src)) ||
2458 (rc = mdb_touch(txn, dst)))
2461 /* Add the node to the destination page.
2463 key.mv_size = srcnode->mn_ksize;
2464 key.mv_data = NODEKEY(srcnode);
2465 data.mv_size = NODEDSZ(srcnode);
2466 data.mv_data = NODEDATA(srcnode);
2467 rc = mdb_add_node(txn, dbi, dst->mp_page, dstindx, &key, &data, NODEPGNO(srcnode),
2469 if (rc != MDB_SUCCESS)
2472 /* Delete the node from the source page.
2474 mdb_del_node(src->mp_page, srcindx);
2476 /* Update the parent separators.
2478 if (srcindx == 0 && src->mp_pi != 0) {
2479 DPRINTF("update separator for source page %lu to [%.*s]",
2480 src->mp_page->mp_pgno, (int)key.mv_size, (char *)key.mv_data);
2481 if ((rc = mdb_update_key(src->mp_parent, src->mp_pi,
2482 &key)) != MDB_SUCCESS)
2486 if (srcindx == 0 && IS_BRANCH(src->mp_page)) {
2488 nullkey.mv_size = 0;
2489 assert(mdb_update_key(src->mp_page, 0, &nullkey) == MDB_SUCCESS);
2492 if (dstindx == 0 && dst->mp_pi != 0) {
2493 DPRINTF("update separator for destination page %lu to [%.*s]",
2494 dst->mp_page->mp_pgno, (int)key.mv_size, (char *)key.mv_data);
2495 if ((rc = mdb_update_key(dst->mp_parent, dst->mp_pi,
2496 &key)) != MDB_SUCCESS)
2500 if (dstindx == 0 && IS_BRANCH(dst->mp_page)) {
2502 nullkey.mv_size = 0;
2503 assert(mdb_update_key(dst->mp_page, 0, &nullkey) == MDB_SUCCESS);
2510 mdb_merge(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *src, MDB_pageparent *dst)
2519 DPRINTF("merging page %lu and %lu", src->mp_page->mp_pgno, dst->mp_page->mp_pgno);
2521 assert(txn != NULL);
2522 assert(src->mp_parent); /* can't merge root page */
2523 assert(dst->mp_parent);
2525 /* Mark src and dst as dirty. */
2526 if ((rc = mdb_touch(txn, src)) ||
2527 (rc = mdb_touch(txn, dst)))
2530 /* Move all nodes from src to dst.
2532 for (i = 0; i < NUMKEYS(src->mp_page); i++) {
2533 srcnode = NODEPTR(src->mp_page, i);
2535 key.mv_size = srcnode->mn_ksize;
2536 key.mv_data = NODEKEY(srcnode);
2537 data.mv_size = NODEDSZ(srcnode);
2538 data.mv_data = NODEDATA(srcnode);
2539 rc = mdb_add_node(txn, dbi, dst->mp_page, NUMKEYS(dst->mp_page), &key,
2540 &data, NODEPGNO(srcnode), srcnode->mn_flags);
2541 if (rc != MDB_SUCCESS)
2545 DPRINTF("dst page %lu now has %u keys (%.1f%% filled)",
2546 dst->mp_page->mp_pgno, NUMKEYS(dst->mp_page), (float)PAGEFILL(txn->mt_env, dst->mp_page) / 10);
2548 /* Unlink the src page from parent.
2550 mdb_del_node(src->mp_parent, src->mp_pi);
2551 if (src->mp_pi == 0) {
2553 if ((rc = mdb_update_key(src->mp_parent, 0, &key)) != MDB_SUCCESS)
2557 if (IS_LEAF(src->mp_page))
2558 txn->mt_dbs[dbi].md_leaf_pages--;
2560 txn->mt_dbs[dbi].md_branch_pages--;
2562 mpp.mp_page = src->mp_parent;
2563 dh = (MDB_dhead *)src->mp_parent;
2565 mpp.mp_parent = dh->md_parent;
2566 mpp.mp_pi = dh->md_pi;
2568 return mdb_rebalance(txn, dbi, &mpp);
2571 #define FILL_THRESHOLD 250
2574 mdb_rebalance(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *mpp)
2579 indx_t si = 0, di = 0;
2581 assert(txn != NULL);
2582 assert(mpp != NULL);
2584 DPRINTF("rebalancing %s page %lu (has %u keys, %.1f%% full)",
2585 IS_LEAF(mpp->mp_page) ? "leaf" : "branch",
2586 mpp->mp_page->mp_pgno, NUMKEYS(mpp->mp_page), (float)PAGEFILL(txn->mt_env, mpp->mp_page) / 10);
2588 if (PAGEFILL(txn->mt_env, mpp->mp_page) >= FILL_THRESHOLD) {
2589 DPRINTF("no need to rebalance page %lu, above fill threshold",
2590 mpp->mp_page->mp_pgno);
2594 if (mpp->mp_parent == NULL) {
2595 if (NUMKEYS(mpp->mp_page) == 0) {
2596 DPRINTF("tree is completely empty");
2597 txn->mt_dbs[dbi].md_root = P_INVALID;
2598 txn->mt_dbs[dbi].md_depth--;
2599 txn->mt_dbs[dbi].md_leaf_pages--;
2600 } else if (IS_BRANCH(mpp->mp_page) && NUMKEYS(mpp->mp_page) == 1) {
2601 DPRINTF("collapsing root page!");
2602 txn->mt_dbs[dbi].md_root = NODEPGNO(NODEPTR(mpp->mp_page, 0));
2603 if ((root = mdb_get_page(txn, txn->mt_dbs[dbi].md_root)) == NULL)
2605 txn->mt_dbs[dbi].md_depth--;
2606 txn->mt_dbs[dbi].md_branch_pages--;
2608 DPRINTF("root page doesn't need rebalancing");
2612 /* The parent (branch page) must have at least 2 pointers,
2613 * otherwise the tree is invalid.
2615 assert(NUMKEYS(mpp->mp_parent) > 1);
2617 /* Leaf page fill factor is below the threshold.
2618 * Try to move keys from left or right neighbor, or
2619 * merge with a neighbor page.
2624 if (mpp->mp_pi == 0) {
2625 /* We're the leftmost leaf in our parent.
2627 DPRINTF("reading right neighbor");
2628 node = NODEPTR(mpp->mp_parent, mpp->mp_pi + 1);
2629 if ((npp.mp_page = mdb_get_page(txn, NODEPGNO(node))) == NULL)
2631 npp.mp_pi = mpp->mp_pi + 1;
2633 di = NUMKEYS(mpp->mp_page);
2635 /* There is at least one neighbor to the left.
2637 DPRINTF("reading left neighbor");
2638 node = NODEPTR(mpp->mp_parent, mpp->mp_pi - 1);
2639 if ((npp.mp_page = mdb_get_page(txn, NODEPGNO(node))) == NULL)
2641 npp.mp_pi = mpp->mp_pi - 1;
2642 si = NUMKEYS(npp.mp_page) - 1;
2645 npp.mp_parent = mpp->mp_parent;
2647 DPRINTF("found neighbor page %lu (%u keys, %.1f%% full)",
2648 npp.mp_page->mp_pgno, NUMKEYS(npp.mp_page), (float)PAGEFILL(txn->mt_env, npp.mp_page) / 10);
2650 /* If the neighbor page is above threshold and has at least two
2651 * keys, move one key from it.
2653 * Otherwise we should try to merge them.
2655 if (PAGEFILL(txn->mt_env, npp.mp_page) >= FILL_THRESHOLD && NUMKEYS(npp.mp_page) >= 2)
2656 return mdb_move_node(txn, dbi, &npp, si, mpp, di);
2657 else { /* FIXME: if (has_enough_room()) */
2658 if (mpp->mp_pi == 0)
2659 return mdb_merge(txn, dbi, &npp, mpp);
2661 return mdb_merge(txn, dbi, mpp, &npp);
2666 mdb_del0(MDB_txn *txn, MDB_dbi dbi, unsigned int ki, MDB_pageparent *mpp, MDB_node *leaf)
2670 /* add overflow pages to free list */
2671 if (F_ISSET(leaf->mn_flags, F_BIGDATA)) {
2675 memcpy(&pg, NODEDATA(leaf), sizeof(pg));
2676 ovpages = OVPAGES(NODEDSZ(leaf), txn->mt_env->me_psize);
2677 for (i=0; i<ovpages; i++) {
2678 DPRINTF("freed ov page %lu", pg);
2679 mdb_idl_insert(txn->mt_free_pgs, pg);
2683 mdb_del_node(mpp->mp_page, ki);
2684 txn->mt_dbs[dbi].md_entries--;
2685 rc = mdb_rebalance(txn, dbi, mpp);
2686 if (rc != MDB_SUCCESS)
2687 txn->mt_flags |= MDB_TXN_ERROR;
2693 mdb_del(MDB_txn *txn, MDB_dbi dbi,
2694 MDB_val *key, MDB_val *data,
2702 DPRINTF("========> delete key %.*s", (int)key->mv_size, (char *)key->mv_data);
2704 assert(key != NULL);
2706 if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
2709 if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
2713 if (key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
2717 mpp.mp_parent = NULL;
2719 if ((rc = mdb_search_page(txn, dbi, key, NULL, 1, &mpp)) != MDB_SUCCESS)
2722 leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, &ki);
2723 if (leaf == NULL || !exact) {
2724 return MDB_NOTFOUND;
2727 if (F_ISSET(txn->mt_dbs[dbi].md_flags, MDB_DUPSORT)) {
2731 mdb_xcursor_init0(txn, dbi, &mx);
2732 mdb_xcursor_init1(txn, dbi, &mx, leaf);
2733 if (flags == MDB_DEL_DUP) {
2734 rc = mdb_del(&mx.mx_txn, mx.mx_cursor.mc_dbi, data, NULL, 0);
2735 mdb_xcursor_fini(txn, dbi, &mx);
2736 if (rc != MDB_SUCCESS)
2738 /* If sub-DB still has entries, we're done */
2739 if (mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi].md_root != P_INVALID) {
2740 memcpy(NODEDATA(leaf), &mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi],
2744 /* otherwise fall thru and delete the sub-DB */
2746 /* add all the child DB's pages to the free list */
2747 rc = mdb_search_page(&mx.mx_txn, mx.mx_cursor.mc_dbi,
2748 NULL, &mx.mx_cursor, 0, &mp2);
2749 if (rc == MDB_SUCCESS) {
2750 MDB_ppage *top, *parent;
2754 cursor_pop_page(&mx.mx_cursor);
2755 top = CURSOR_TOP(&mx.mx_cursor);
2757 parent = SLIST_NEXT(top, mp_entry);
2758 while (parent != NULL) {
2759 for (i=0; i<NUMKEYS(top->mp_page); i++) {
2760 ni = NODEPTR(top->mp_page, i);
2761 mdb_idl_insert(txn->mt_free_pgs, ni->mn_pgno);
2765 if (parent->mp_ki >= NUMKEYS(parent->mp_page)) {
2766 cursor_pop_page(&mx.mx_cursor);
2767 top = CURSOR_TOP(&mx.mx_cursor);
2768 parent = SLIST_NEXT(top, mp_entry);
2770 ni = NODEPTR(parent->mp_page, parent->mp_ki);
2771 top->mp_page = mdb_get_page(&mx.mx_txn, ni->mn_pgno);
2776 mdb_idl_insert(txn->mt_free_pgs, mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi].md_root);
2781 if (data && (rc = mdb_read_data(txn, leaf, data)) != MDB_SUCCESS)
2784 return mdb_del0(txn, dbi, ki, &mpp, leaf);
2787 /* Split page <*mpp>, and insert <key,(data|newpgno)> in either left or
2788 * right sibling, at index <*newindxp> (as if unsplit). Updates *mpp and
2789 * *newindxp with the actual values after split, ie if *mpp and *newindxp
2790 * refer to a node in the new right sibling page.
2793 mdb_split(MDB_txn *txn, MDB_dbi dbi, MDB_page **mpp, unsigned int *newindxp,
2794 MDB_val *newkey, MDB_val *newdata, pgno_t newpgno)
2797 int rc = MDB_SUCCESS, ins_new = 0;
2800 unsigned int i, j, split_indx;
2802 MDB_val sepkey, rkey, rdata;
2804 MDB_dpage *mdp, *rdp, *pdp;
2807 assert(txn != NULL);
2809 dh = ((MDB_dhead *)*mpp) - 1;
2810 mdp = (MDB_dpage *)dh;
2811 newindx = *newindxp;
2813 DPRINTF("-----> splitting %s page %lu and adding [%.*s] at index %i",
2814 IS_LEAF(&mdp->p) ? "leaf" : "branch", mdp->p.mp_pgno,
2815 (int)newkey->mv_size, (char *)newkey->mv_data, *newindxp);
2817 if (mdp->h.md_parent == NULL) {
2818 if ((pdp = mdb_new_page(txn, dbi, P_BRANCH, 1)) == NULL)
2821 mdp->h.md_parent = &pdp->p;
2822 txn->mt_dbs[dbi].md_root = pdp->p.mp_pgno;
2823 DPRINTF("root split! new root = %lu", pdp->p.mp_pgno);
2824 txn->mt_dbs[dbi].md_depth++;
2826 /* Add left (implicit) pointer. */
2827 if (mdb_add_node(txn, dbi, &pdp->p, 0, NULL, NULL,
2828 mdp->p.mp_pgno, 0) != MDB_SUCCESS)
2831 DPRINTF("parent branch page is %lu", mdp->h.md_parent->mp_pgno);
2834 /* Create a right sibling. */
2835 if ((rdp = mdb_new_page(txn, dbi, mdp->p.mp_flags, 1)) == NULL)
2837 rdp->h.md_parent = mdp->h.md_parent;
2838 rdp->h.md_pi = mdp->h.md_pi + 1;
2839 DPRINTF("new right sibling: page %lu", rdp->p.mp_pgno);
2841 /* Move half of the keys to the right sibling. */
2842 if ((copy = malloc(txn->mt_env->me_psize)) == NULL)
2844 memcpy(copy, &mdp->p, txn->mt_env->me_psize);
2845 memset(&mdp->p.mp_ptrs, 0, txn->mt_env->me_psize - PAGEHDRSZ);
2846 mdp->p.mp_lower = PAGEHDRSZ;
2847 mdp->p.mp_upper = txn->mt_env->me_psize;
2849 split_indx = NUMKEYS(copy) / 2 + 1;
2851 /* First find the separating key between the split pages.
2853 memset(&sepkey, 0, sizeof(sepkey));
2854 if (newindx == split_indx) {
2855 sepkey.mv_size = newkey->mv_size;
2856 sepkey.mv_data = newkey->mv_data;
2858 node = NODEPTR(copy, split_indx);
2859 sepkey.mv_size = node->mn_ksize;
2860 sepkey.mv_data = NODEKEY(node);
2863 DPRINTF("separator is [%.*s]", (int)sepkey.mv_size, (char *)sepkey.mv_data);
2865 /* Copy separator key to the parent.
2867 if (SIZELEFT(rdp->h.md_parent) < mdb_branch_size(txn->mt_env, &sepkey)) {
2868 rc = mdb_split(txn, dbi, &rdp->h.md_parent, &rdp->h.md_pi,
2869 &sepkey, NULL, rdp->p.mp_pgno);
2871 /* Right page might now have changed parent.
2872 * Check if left page also changed parent.
2874 if (rdp->h.md_parent != mdp->h.md_parent &&
2875 mdp->h.md_pi >= NUMKEYS(mdp->h.md_parent)) {
2876 mdp->h.md_parent = rdp->h.md_parent;
2877 mdp->h.md_pi = rdp->h.md_pi - 1;
2880 rc = mdb_add_node(txn, dbi, rdp->h.md_parent, rdp->h.md_pi,
2881 &sepkey, NULL, rdp->p.mp_pgno, 0);
2883 if (rc != MDB_SUCCESS) {
2888 for (i = j = 0; i <= NUMKEYS(copy); j++) {
2889 if (i < split_indx) {
2890 /* Re-insert in left sibling. */
2893 /* Insert in right sibling. */
2894 if (i == split_indx)
2895 /* Reset insert index for right sibling. */
2896 j = (i == newindx && ins_new);
2900 if (i == newindx && !ins_new) {
2901 /* Insert the original entry that caused the split. */
2902 rkey.mv_data = newkey->mv_data;
2903 rkey.mv_size = newkey->mv_size;
2904 if (IS_LEAF(&mdp->p)) {
2905 rdata.mv_data = newdata->mv_data;
2906 rdata.mv_size = newdata->mv_size;
2913 /* Update page and index for the new key. */
2916 } else if (i == NUMKEYS(copy)) {
2919 node = NODEPTR(copy, i);
2920 rkey.mv_data = NODEKEY(node);
2921 rkey.mv_size = node->mn_ksize;
2922 if (IS_LEAF(&mdp->p)) {
2923 rdata.mv_data = NODEDATA(node);
2924 rdata.mv_size = node->mn_dsize;
2926 pgno = node->mn_pgno;
2927 flags = node->mn_flags;
2932 if (!IS_LEAF(&mdp->p) && j == 0) {
2933 /* First branch index doesn't need key data. */
2937 rc = mdb_add_node(txn, dbi, &pdp->p, j, &rkey, &rdata, pgno,flags);
2945 mdb_put0(MDB_txn *txn, MDB_dbi dbi,
2946 MDB_val *key, MDB_val *data, unsigned int flags)
2948 int rc = MDB_SUCCESS, exact;
2952 MDB_val xdata, *rdata;
2955 DPRINTF("==> put key %.*s, size %zu, data size %zu",
2956 (int)key->mv_size, (char *)key->mv_data, key->mv_size, data->mv_size);
2958 mpp.mp_parent = NULL;
2960 rc = mdb_search_page(txn, dbi, key, NULL, 1, &mpp);
2961 if (rc == MDB_SUCCESS) {
2962 leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, &ki);
2963 if (leaf && exact) {
2964 if (F_ISSET(txn->mt_dbs[dbi].md_flags, MDB_DUPSORT)) {
2967 if (flags == MDB_NOOVERWRITE) {
2968 DPRINTF("duplicate key %.*s",
2969 (int)key->mv_size, (char *)key->mv_data);
2970 return MDB_KEYEXIST;
2972 /* same size, just replace it */
2973 if (NODEDSZ(leaf) == data->mv_size) {
2974 memcpy(NODEDATA(leaf), data->mv_data, data->mv_size);
2977 mdb_del_node(mpp.mp_page, ki);
2979 if (leaf == NULL) { /* append if not found */
2980 ki = NUMKEYS(mpp.mp_page);
2981 DPRINTF("appending key at index %i", ki);
2983 } else if (rc == MDB_NOTFOUND) {
2985 /* new file, just write a root leaf page */
2986 DPRINTF("allocating new root leaf page");
2987 if ((dp = mdb_new_page(txn, dbi, P_LEAF, 1)) == NULL) {
2990 mpp.mp_page = &dp->p;
2991 txn->mt_dbs[dbi].md_root = mpp.mp_page->mp_pgno;
2992 txn->mt_dbs[dbi].md_depth++;
2998 assert(IS_LEAF(mpp.mp_page));
2999 DPRINTF("there are %u keys, should insert new key at index %i",
3000 NUMKEYS(mpp.mp_page), ki);
3002 /* For sorted dups, the data item at this level is a DB record
3003 * for a child DB; the actual data elements are stored as keys
3006 if (F_ISSET(txn->mt_dbs[dbi].md_flags, MDB_DUPSORT)) {
3008 xdata.mv_size = sizeof(MDB_db);
3009 xdata.mv_data = &dummy;
3010 memset(&dummy, 0, sizeof(dummy));
3011 dummy.md_root = P_INVALID;
3016 if (SIZELEFT(mpp.mp_page) < mdb_leaf_size(txn->mt_env, key, rdata)) {
3017 rc = mdb_split(txn, dbi, &mpp.mp_page, &ki, key, rdata, P_INVALID);
3019 /* There is room already in this leaf page. */
3020 rc = mdb_add_node(txn, dbi, mpp.mp_page, ki, key, rdata, 0, 0);
3023 if (rc != MDB_SUCCESS)
3024 txn->mt_flags |= MDB_TXN_ERROR;
3026 txn->mt_dbs[dbi].md_entries++;
3028 /* Remember if we just added a subdatabase */
3029 if (flags & F_SUBDATA) {
3030 leaf = NODEPTR(mpp.mp_page, ki);
3031 leaf->mn_flags |= F_SUBDATA;
3034 /* Now store the actual data in the child DB. Note that we're
3035 * storing the user data in the keys field, so there are strict
3036 * size limits on dupdata. The actual data fields of the child
3037 * DB are all zero size.
3039 if (F_ISSET(txn->mt_dbs[dbi].md_flags, MDB_DUPSORT)) {
3042 leaf = NODEPTR(mpp.mp_page, ki);
3044 mdb_xcursor_init0(txn, dbi, &mx);
3045 mdb_xcursor_init1(txn, dbi, &mx, leaf);
3048 if (flags == MDB_NODUPDATA)
3049 flags = MDB_NOOVERWRITE;
3050 rc = mdb_put0(&mx.mx_txn, mx.mx_cursor.mc_dbi, data, &xdata, flags);
3051 mdb_xcursor_fini(txn, dbi, &mx);
3052 memcpy(NODEDATA(leaf), &mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi],
3062 mdb_put(MDB_txn *txn, MDB_dbi dbi,
3063 MDB_val *key, MDB_val *data, unsigned int flags)
3065 assert(key != NULL);
3066 assert(data != NULL);
3068 if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
3071 if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
3075 if (key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
3079 if ((flags & (MDB_NOOVERWRITE|MDB_NODUPDATA)) != flags)
3082 return mdb_put0(txn, dbi, key, data, flags);
3086 mdbenv_get_flags(MDB_env *env, unsigned int *arg)
3091 *arg = env->me_flags;
3096 mdbenv_get_path(MDB_env *env, const char **arg)
3101 *arg = env->me_path;
3106 mdb_stat0(MDB_env *env, MDB_db *db, MDB_stat *arg)
3108 arg->ms_psize = env->me_psize;
3109 arg->ms_depth = db->md_depth;
3110 arg->ms_branch_pages = db->md_branch_pages;
3111 arg->ms_leaf_pages = db->md_leaf_pages;
3112 arg->ms_overflow_pages = db->md_overflow_pages;
3113 arg->ms_entries = db->md_entries;
3118 mdbenv_stat(MDB_env *env, MDB_stat *arg)
3120 if (env == NULL || arg == NULL)
3123 return mdb_stat0(env, &env->me_meta->mm_dbs[MAIN_DBI], arg);
3126 int mdb_open(MDB_txn *txn, const char *name, unsigned int flags, MDB_dbi *dbi)
3136 if (flags & (MDB_DUPSORT|MDB_REVERSEKEY|MDB_INTEGERKEY))
3137 txn->mt_dbs[MAIN_DBI].md_flags |= (flags & (MDB_DUPSORT|MDB_REVERSEKEY|MDB_INTEGERKEY));
3141 /* Is the DB already open? */
3143 for (i=2; i<txn->mt_numdbs; i++) {
3144 if (len == txn->mt_dbxs[i].md_name.mv_size &&
3145 !strncmp(name, txn->mt_dbxs[i].md_name.mv_data, len)) {
3151 if (txn->mt_numdbs >= txn->mt_env->me_maxdbs - 1)
3154 /* Find the DB info */
3156 key.mv_data = (void *)name;
3157 rc = mdb_get(txn, MAIN_DBI, &key, &data);
3159 /* Create if requested */
3160 if (rc == MDB_NOTFOUND && (flags & MDB_CREATE)) {
3162 data.mv_size = sizeof(MDB_db);
3163 data.mv_data = &dummy;
3164 memset(&dummy, 0, sizeof(dummy));
3165 dummy.md_root = P_INVALID;
3166 dummy.md_flags = flags & 0xffff;
3167 rc = mdb_put0(txn, MAIN_DBI, &key, &data, F_SUBDATA);
3171 /* OK, got info, add to table */
3172 if (rc == MDB_SUCCESS) {
3173 txn->mt_dbxs[txn->mt_numdbs].md_name.mv_data = strdup(name);
3174 txn->mt_dbxs[txn->mt_numdbs].md_name.mv_size = len;
3175 txn->mt_dbxs[txn->mt_numdbs].md_cmp = NULL;
3176 txn->mt_dbxs[txn->mt_numdbs].md_dcmp = NULL;
3177 txn->mt_dbxs[txn->mt_numdbs].md_rel = NULL;
3178 txn->mt_dbxs[txn->mt_numdbs].md_parent = MAIN_DBI;
3179 txn->mt_dbxs[txn->mt_numdbs].md_dirty = dirty;
3180 memcpy(&txn->mt_dbs[txn->mt_numdbs], data.mv_data, sizeof(MDB_db));
3181 *dbi = txn->mt_numdbs;
3188 int mdb_stat(MDB_txn *txn, MDB_dbi dbi, MDB_stat *arg)
3190 if (txn == NULL || arg == NULL || dbi >= txn->mt_numdbs)
3193 return mdb_stat0(txn->mt_env, &txn->mt_dbs[dbi], arg);
3196 void mdb_close(MDB_txn *txn, MDB_dbi dbi)
3199 if (dbi <= MAIN_DBI || dbi >= txn->mt_numdbs)
3201 ptr = txn->mt_dbxs[dbi].md_name.mv_data;
3202 txn->mt_dbxs[dbi].md_name.mv_data = NULL;
3203 txn->mt_dbxs[dbi].md_name.mv_size = 0;
3207 int mdb_set_compare(MDB_txn *txn, MDB_dbi dbi, MDB_cmp_func *cmp)
3209 if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
3212 txn->mt_dbxs[dbi].md_cmp = cmp;
3216 int mdb_set_dupsort(MDB_txn *txn, MDB_dbi dbi, MDB_cmp_func *cmp)
3218 if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
3221 txn->mt_dbxs[dbi].md_dcmp = cmp;
3225 int mdb_set_relfunc(MDB_txn *txn, MDB_dbi dbi, MDB_rel_func *rel)
3227 if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
3230 txn->mt_dbxs[dbi].md_rel = rel;