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 */
265 #define F_DUPDATA 0x04 /* data has duplicates */
269 typedef struct MDB_dbx {
271 MDB_cmp_func *md_cmp; /* user compare function */
272 MDB_cmp_func *md_dcmp; /* user dupsort function */
273 MDB_rel_func *md_rel; /* user relocate function */
275 unsigned int md_dirty;
279 pgno_t mt_next_pgno; /* next unallocated page */
283 pgno_t *mt_free_pgs; /* this is an IDL */
285 MIDL2 *dirty_list; /* modified pages */
288 MDB_dbx *mt_dbxs; /* array */
290 unsigned int mt_numdbs;
292 #define MDB_TXN_RDONLY 0x01 /* read-only transaction */
293 #define MDB_TXN_ERROR 0x02 /* an error has occurred */
294 #define MDB_TXN_METOGGLE 0x04 /* used meta page 1 */
295 unsigned int mt_flags;
298 /* Context for sorted-dup records */
299 typedef struct MDB_xcursor {
300 MDB_cursor mx_cursor;
309 int me_mfd; /* just for writing the meta pages */
310 #define MDB_FATAL_ERROR 0x80000000U
312 uint32_t me_extrapad; /* unused for now */
313 unsigned int me_maxreaders;
314 unsigned int me_numdbs;
315 unsigned int me_maxdbs;
318 MDB_txninfo *me_txns;
319 MDB_meta *me_metas[2];
321 MDB_txn *me_txn; /* current write transaction */
323 off_t me_size; /* current file size */
324 pgno_t me_maxpg; /* me_mapsize / me_psize */
325 unsigned int me_psize;
326 unsigned int me_db_toggle;
327 MDB_dbx *me_dbxs; /* array */
329 MDB_oldpages *me_pghead;
330 pthread_key_t me_txkey; /* thread-key for readers */
331 pgno_t me_free_pgs[MDB_IDL_UM_SIZE];
332 MIDL2 me_dirty_list[MDB_IDL_DB_SIZE];
335 #define NODESIZE offsetof(MDB_node, mn_data)
337 #define INDXSIZE(k) (NODESIZE + ((k) == NULL ? 0 : (k)->mv_size))
338 #define LEAFSIZE(k, d) (NODESIZE + (k)->mv_size + (d)->mv_size)
339 #define NODEPTR(p, i) ((MDB_node *)((char *)(p) + (p)->mp_ptrs[i]))
340 #define NODEKEY(node) (void *)((node)->mn_data)
341 #define NODEDATA(node) (void *)((char *)(node)->mn_data + (node)->mn_ksize)
342 #define NODEPGNO(node) ((node)->mn_pgno)
343 #define NODEDSZ(node) ((node)->mn_dsize)
345 #define MDB_COMMIT_PAGES 64 /* max number of pages to write in one commit */
347 static int mdb_search_page_root(MDB_txn *txn,
348 MDB_dbi dbi, MDB_val *key,
349 MDB_cursor *cursor, int modify,
350 MDB_pageparent *mpp);
351 static int mdb_search_page(MDB_txn *txn,
352 MDB_dbi dbi, MDB_val *key,
353 MDB_cursor *cursor, int modify,
354 MDB_pageparent *mpp);
356 static int mdb_env_read_header(MDB_env *env, MDB_meta *meta);
357 static int mdb_env_read_meta(MDB_env *env, int *which);
358 static int mdb_env_write_meta(MDB_txn *txn);
359 static MDB_page *mdb_get_page(MDB_txn *txn, pgno_t pgno);
361 static MDB_node *mdb_search_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp,
362 MDB_val *key, int *exactp, unsigned int *kip);
363 static int mdb_add_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp,
364 indx_t indx, MDB_val *key, MDB_val *data,
365 pgno_t pgno, uint8_t flags);
366 static void mdb_del_node(MDB_page *mp, indx_t indx);
367 static int mdb_del0(MDB_txn *txn, MDB_dbi dbi, unsigned int ki,
368 MDB_pageparent *mpp, MDB_node *leaf);
369 static int mdb_put0(MDB_txn *txn, MDB_dbi dbi,
370 MDB_val *key, MDB_val *data, unsigned int flags);
371 static int mdb_read_data(MDB_txn *txn, MDB_node *leaf, MDB_val *data);
373 static int mdb_rebalance(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *mp);
374 static int mdb_update_key(MDB_page *mp, indx_t indx, MDB_val *key);
375 static int mdb_move_node(MDB_txn *txn, MDB_dbi dbi,
376 MDB_pageparent *src, indx_t srcindx,
377 MDB_pageparent *dst, indx_t dstindx);
378 static int mdb_merge(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *src,
379 MDB_pageparent *dst);
380 static int mdb_split(MDB_txn *txn, MDB_dbi dbi, MDB_page **mpp,
381 unsigned int *newindxp, MDB_val *newkey,
382 MDB_val *newdata, pgno_t newpgno);
383 static MDB_dpage *mdb_new_page(MDB_txn *txn, MDB_dbi dbi, uint32_t flags, int num);
385 static void cursor_pop_page(MDB_cursor *cursor);
386 static MDB_ppage *cursor_push_page(MDB_cursor *cursor,
389 static int mdb_set_key(MDB_node *node, MDB_val *key);
390 static int mdb_sibling(MDB_cursor *cursor, int move_right);
391 static int mdb_cursor_next(MDB_cursor *cursor,
392 MDB_val *key, MDB_val *data, MDB_cursor_op op);
393 static int mdb_cursor_prev(MDB_cursor *cursor,
394 MDB_val *key, MDB_val *data, MDB_cursor_op op);
395 static int mdb_cursor_set(MDB_cursor *cursor,
396 MDB_val *key, MDB_val *data, MDB_cursor_op op, int *exactp);
397 static int mdb_cursor_first(MDB_cursor *cursor,
398 MDB_val *key, MDB_val *data);
399 static int mdb_cursor_last(MDB_cursor *cursor,
400 MDB_val *key, MDB_val *data);
402 static void mdb_xcursor_init0(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx);
403 static void mdb_xcursor_init1(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx, MDB_node *node);
404 static void mdb_xcursor_fini(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx);
406 static size_t mdb_leaf_size(MDB_env *env, MDB_val *key,
408 static size_t mdb_branch_size(MDB_env *env, MDB_val *key);
410 static int memncmp(const void *s1, size_t n1,
411 const void *s2, size_t n2);
412 static int memnrcmp(const void *s1, size_t n1,
413 const void *s2, size_t n2);
416 memncmp(const void *s1, size_t n1, const void *s2, size_t n2)
418 int diff, len_diff = -1;
421 len_diff = (n1 > n2);
424 diff = memcmp(s1, s2, n1);
425 return diff ? diff : len_diff;
429 memnrcmp(const void *s1, size_t n1, const void *s2, size_t n2)
431 const unsigned char *p1, *p2, *p1_lim;
438 p1 = (const unsigned char *)s1 + n1 - 1;
439 p2 = (const unsigned char *)s2 + n2 - 1;
441 for (p1_lim = (n1 <= n2 ? s1 : s2); *p1 == *p2; p1--, p2--) {
443 return (p1 != s1) ? (p1 != p2) : (p2 != s2) ? -1 : 0;
449 mdb_version(int *maj, int *min, int *pat)
451 *maj = MDB_VERSION_MAJOR;
452 *min = MDB_VERSION_MINOR;
453 *pat = MDB_VERSION_PATCH;
454 return MDB_VERSION_STRING;
457 static const char *errstr[] = {
458 "MDB_KEYEXIST: Key/data pair already exists",
459 "MDB_NOTFOUND: No matching key/data pair found",
460 "MDB_PAGE_NOTFOUND: Requested page not found",
461 "MDB_CORRUPTED: Located page was wrong type",
462 "MDB_PANIC: Update of meta page failed",
463 "MDB_VERSION_MISMATCH: Database environment version mismatch"
467 mdb_strerror(int err)
470 return ("Successful return: 0");
472 if (err >= MDB_KEYEXIST && err <= MDB_VERSION_MISMATCH)
473 return (char *)errstr[err - MDB_KEYEXIST];
475 return strerror(err);
479 mdb_cmp(MDB_txn *txn, MDB_dbi dbi, const MDB_val *a, const MDB_val *b)
481 return txn->mt_dbxs[dbi].md_cmp(a, b);
485 _mdb_cmp(MDB_txn *txn, MDB_dbi dbi, const MDB_val *key1, const MDB_val *key2)
487 if (txn->mt_dbs[dbi].md_flags & (MDB_REVERSEKEY
488 #if __BYTE_ORDER == __LITTLE_ENDIAN
492 return memnrcmp(key1->mv_data, key1->mv_size, key2->mv_data, key2->mv_size);
494 return memncmp((char *)key1->mv_data, key1->mv_size, key2->mv_data, key2->mv_size);
497 /* Allocate new page(s) for writing */
499 mdb_alloc_page(MDB_txn *txn, MDB_page *parent, unsigned int parent_idx, int num)
502 pgno_t pgno = P_INVALID;
506 if (txn->mt_txnid > 2) {
508 oldest = txn->mt_txnid - 2;
509 if (!txn->mt_env->me_pghead && txn->mt_dbs[FREE_DBI].md_root != P_INVALID) {
510 /* See if there's anything in the free DB */
515 mpp.mp_parent = NULL;
517 mdb_search_page(txn, FREE_DBI, NULL, NULL, 0, &mpp);
518 leaf = NODEPTR(mpp.mp_page, 0);
519 kptr = (ULONG *)NODEKEY(leaf);
521 /* It's potentially usable, unless there are still
522 * older readers outstanding. Grab it.
524 if (oldest > *kptr) {
529 mdb_read_data(txn, leaf, &data);
530 idl = (ULONG *)data.mv_data;
531 mop = malloc(sizeof(MDB_oldpages) + MDB_IDL_SIZEOF(idl) - sizeof(pgno_t));
532 mop->mo_next = txn->mt_env->me_pghead;
533 mop->mo_txnid = *kptr;
534 txn->mt_env->me_pghead = mop;
535 memcpy(mop->mo_pages, idl, MDB_IDL_SIZEOF(idl));
540 DPRINTF("IDL read txn %lu root %lu num %lu",
541 mop->mo_txnid, txn->mt_dbs[FREE_DBI].md_root, idl[0]);
542 for (i=0; i<idl[0]; i++) {
543 DPRINTF("IDL %lu", idl[i+1]);
547 /* drop this IDL from the DB */
548 mpp.mp_parent = NULL;
550 mdb_search_page(txn, FREE_DBI, NULL, NULL, 1, &mpp);
551 leaf = NODEPTR(mpp.mp_page, 0);
552 mdb_del0(txn, FREE_DBI, 0, &mpp, leaf);
555 if (txn->mt_env->me_pghead) {
557 for (i=0; i<txn->mt_env->me_txns->mti_numreaders; i++) {
558 ULONG mr = txn->mt_env->me_txns->mti_readers[i].mr_txnid;
561 oldest = txn->mt_env->me_txns->mti_readers[i].mr_txnid;
563 if (oldest > txn->mt_env->me_pghead->mo_txnid) {
564 MDB_oldpages *mop = txn->mt_env->me_pghead;
565 txn->mt_oldest = oldest;
567 /* FIXME: For now, always use fresh pages. We
568 * really ought to search the free list for a
573 /* peel pages off tail, so we only have to truncate the list */
574 pgno = MDB_IDL_LAST(mop->mo_pages);
575 if (MDB_IDL_IS_RANGE(mop->mo_pages)) {
577 if (mop->mo_pages[2] > mop->mo_pages[1])
578 mop->mo_pages[0] = 0;
582 if (MDB_IDL_IS_ZERO(mop->mo_pages)) {
583 txn->mt_env->me_pghead = mop->mo_next;
591 if (pgno == P_INVALID) {
592 /* DB size is maxed out */
593 if (txn->mt_next_pgno + num >= txn->mt_env->me_maxpg)
596 if ((dp = malloc(txn->mt_env->me_psize * num + sizeof(MDB_dhead))) == NULL)
599 dp->h.md_parent = parent;
600 dp->h.md_pi = parent_idx;
601 if (pgno == P_INVALID) {
602 dp->p.mp_pgno = txn->mt_next_pgno;
603 txn->mt_next_pgno += num;
605 dp->p.mp_pgno = pgno;
607 mid.mid = dp->p.mp_pgno;
609 mdb_midl2_insert(txn->mt_u.dirty_list, &mid);
614 /* Touch a page: make it dirty and re-insert into tree with updated pgno.
617 mdb_touch(MDB_txn *txn, MDB_pageparent *pp)
619 MDB_page *mp = pp->mp_page;
624 if (!F_ISSET(mp->mp_flags, P_DIRTY)) {
626 if ((dp = mdb_alloc_page(txn, pp->mp_parent, pp->mp_pi, 1)) == NULL)
628 DPRINTF("touched page %lu -> %lu", mp->mp_pgno, dp->p.mp_pgno);
629 mdb_midl_insert(txn->mt_free_pgs, mp->mp_pgno);
630 pgno = dp->p.mp_pgno;
631 memcpy(&dp->p, mp, txn->mt_env->me_psize);
634 mp->mp_flags |= P_DIRTY;
636 /* Update the page number to new touched page. */
637 if (pp->mp_parent != NULL)
638 NODEPGNO(NODEPTR(pp->mp_parent, pp->mp_pi)) = mp->mp_pgno;
645 mdb_env_sync(MDB_env *env, int force)
648 if (force || !F_ISSET(env->me_flags, MDB_NOSYNC)) {
649 if (fdatasync(env->me_fd))
656 mdb_txn_begin(MDB_env *env, int rdonly, MDB_txn **ret)
661 if (env->me_flags & MDB_FATAL_ERROR) {
662 DPRINTF("mdb_txn_begin: environment had fatal error, must shutdown!");
665 if ((txn = calloc(1, sizeof(MDB_txn))) == NULL) {
666 DPRINTF("calloc: %s", strerror(errno));
671 txn->mt_flags |= MDB_TXN_RDONLY;
673 txn->mt_u.dirty_list = env->me_dirty_list;
674 txn->mt_u.dirty_list[0].mid = 0;
675 txn->mt_free_pgs = env->me_free_pgs;
676 txn->mt_free_pgs[0] = 0;
678 pthread_mutex_lock(&env->me_txns->mti_wmutex);
679 env->me_txns->mti_txnid++;
682 txn->mt_txnid = env->me_txns->mti_txnid;
684 MDB_reader *r = pthread_getspecific(env->me_txkey);
687 pthread_mutex_lock(&env->me_txns->mti_mutex);
688 for (i=0; i<env->me_txns->mti_numreaders; i++)
689 if (env->me_txns->mti_readers[i].mr_pid == 0)
691 if (i == env->me_maxreaders) {
692 pthread_mutex_unlock(&env->me_txns->mti_mutex);
695 env->me_txns->mti_readers[i].mr_pid = getpid();
696 env->me_txns->mti_readers[i].mr_tid = pthread_self();
697 r = &env->me_txns->mti_readers[i];
698 pthread_setspecific(env->me_txkey, r);
699 if (i >= env->me_txns->mti_numreaders)
700 env->me_txns->mti_numreaders = i+1;
701 pthread_mutex_unlock(&env->me_txns->mti_mutex);
703 r->mr_txnid = txn->mt_txnid;
704 txn->mt_u.reader = r;
711 toggle = env->me_txns->mti_me_toggle;
712 if ((rc = mdb_env_read_meta(env, &toggle)) != MDB_SUCCESS) {
717 /* Copy the DB arrays */
718 txn->mt_numdbs = env->me_numdbs;
719 txn->mt_dbxs = env->me_dbxs; /* mostly static anyway */
720 txn->mt_dbs = malloc(env->me_maxdbs * sizeof(MDB_db));
721 memcpy(txn->mt_dbs, env->me_meta->mm_dbs, 2 * sizeof(MDB_db));
722 if (txn->mt_numdbs > 2)
723 memcpy(txn->mt_dbs+2, env->me_dbs[env->me_db_toggle]+2,
724 (txn->mt_numdbs - 2) * sizeof(MDB_db));
728 txn->mt_flags |= MDB_TXN_METOGGLE;
729 txn->mt_next_pgno = env->me_meta->mm_last_pg+1;
732 DPRINTF("begin transaction %lu on mdbenv %p, root page %lu",
733 txn->mt_txnid, (void *) env, txn->mt_dbs[MAIN_DBI].md_root);
740 mdb_txn_abort(MDB_txn *txn)
748 DPRINTF("abort transaction %lu on mdbenv %p, root page %lu",
749 txn->mt_txnid, (void *) env, txn->mt_dbs[MAIN_DBI].md_root);
753 if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
754 txn->mt_u.reader->mr_txnid = 0;
759 /* Discard all dirty pages. */
760 for (i=1; i<=txn->mt_u.dirty_list[0].mid; i++)
761 free(txn->mt_u.dirty_list[i].mptr);
763 while ((mop = txn->mt_env->me_pghead)) {
764 txn->mt_env->me_pghead = mop->mo_next;
769 env->me_txns->mti_txnid--;
770 for (i=2; i<env->me_numdbs; i++)
771 env->me_dbxs[i].md_dirty = 0;
772 pthread_mutex_unlock(&env->me_txns->mti_wmutex);
779 mdb_txn_commit(MDB_txn *txn)
788 struct iovec iov[MDB_COMMIT_PAGES];
791 assert(txn->mt_env != NULL);
795 if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
800 if (txn != env->me_txn) {
801 DPRINTF("attempt to commit unknown transaction");
806 if (F_ISSET(txn->mt_flags, MDB_TXN_ERROR)) {
807 DPRINTF("error flag is set, can't commit");
812 if (!txn->mt_u.dirty_list[0].mid)
815 DPRINTF("committing transaction %lu on mdbenv %p, root page %lu",
816 txn->mt_txnid, (void *) env, txn->mt_dbs[MAIN_DBI].md_root);
818 /* should only be one record now */
819 if (env->me_pghead) {
823 mop = env->me_pghead;
824 key.mv_size = sizeof(pgno_t);
825 key.mv_data = (char *)&mop->mo_txnid;
826 data.mv_size = MDB_IDL_SIZEOF(mop->mo_pages);
827 data.mv_data = mop->mo_pages;
828 mdb_put0(txn, FREE_DBI, &key, &data, 0);
829 free(env->me_pghead);
830 env->me_pghead = NULL;
832 /* save to free list */
833 if (!MDB_IDL_IS_ZERO(txn->mt_free_pgs)) {
837 /* make sure last page of freeDB is touched and on freelist */
838 key.mv_size = MAXKEYSIZE+1;
840 mpp.mp_parent = NULL;
842 mdb_search_page(txn, FREE_DBI, &key, NULL, 1, &mpp);
847 ULONG *idl = txn->mt_free_pgs;
848 DPRINTF("IDL write txn %lu root %lu num %lu",
849 txn->mt_txnid, txn->mt_dbs[FREE_DBI].md_root, idl[0]);
850 for (i=0; i<idl[0]; i++) {
851 DPRINTF("IDL %lu", idl[i+1]);
855 /* write to last page of freeDB */
856 key.mv_size = sizeof(pgno_t);
857 key.mv_data = (char *)&txn->mt_txnid;
858 data.mv_size = MDB_IDL_SIZEOF(txn->mt_free_pgs);
859 data.mv_data = txn->mt_free_pgs;
860 mdb_put0(txn, FREE_DBI, &key, &data, 0);
863 /* Update DB root pointers. Their pages have already been
864 * touched so this is all in-place and cannot fail.
868 data.mv_size = sizeof(MDB_db);
870 for (i = 2; i < txn->mt_numdbs; i++) {
871 if (txn->mt_dbxs[i].md_dirty) {
872 data.mv_data = &txn->mt_dbs[i];
873 mdb_put0(txn, MAIN_DBI, &txn->mt_dbxs[i].md_name, &data, 0);
878 /* Commit up to MDB_COMMIT_PAGES dirty pages to disk until done.
886 for (; i<=txn->mt_u.dirty_list[0].mid; i++) {
887 dp = txn->mt_u.dirty_list[i].mptr;
888 if (dp->p.mp_pgno != next) {
890 DPRINTF("committing %u dirty pages", n);
891 rc = writev(env->me_fd, iov, n);
895 DPRINTF("short write, filesystem full?");
897 DPRINTF("writev: %s", strerror(errno));
904 lseek(env->me_fd, dp->p.mp_pgno * env->me_psize, SEEK_SET);
905 next = dp->p.mp_pgno;
907 DPRINTF("committing page %lu", dp->p.mp_pgno);
908 iov[n].iov_len = env->me_psize * dp->h.md_num;
909 iov[n].iov_base = &dp->p;
910 size += iov[n].iov_len;
911 next = dp->p.mp_pgno + dp->h.md_num;
912 /* clear dirty flag */
913 dp->p.mp_flags &= ~P_DIRTY;
914 if (++n >= MDB_COMMIT_PAGES) {
923 DPRINTF("committing %u dirty pages", n);
924 rc = writev(env->me_fd, iov, n);
928 DPRINTF("short write, filesystem full?");
930 DPRINTF("writev: %s", strerror(errno));
937 /* Drop the dirty pages.
939 for (i=1; i<=txn->mt_u.dirty_list[0].mid; i++)
940 free(txn->mt_u.dirty_list[i].mptr);
942 if ((n = mdb_env_sync(env, 0)) != 0 ||
943 (n = mdb_env_write_meta(txn)) != MDB_SUCCESS) {
950 /* update the DB tables */
952 int toggle = !env->me_db_toggle;
954 for (i = 2; i < env->me_numdbs; i++) {
955 if (txn->mt_dbxs[i].md_dirty) {
956 env->me_dbs[toggle][i] = txn->mt_dbs[i];
957 txn->mt_dbxs[i].md_dirty = 0;
960 for (i = env->me_numdbs; i < txn->mt_numdbs; i++) {
961 txn->mt_dbxs[i].md_dirty = 0;
962 env->me_dbxs[i] = txn->mt_dbxs[i];
963 env->me_dbs[toggle][i] = txn->mt_dbs[i];
965 env->me_db_toggle = toggle;
966 env->me_numdbs = txn->mt_numdbs;
971 pthread_mutex_unlock(&env->me_txns->mti_wmutex);
978 mdb_env_read_header(MDB_env *env, MDB_meta *meta)
987 /* We don't know the page size yet, so use a minimum value.
990 if ((rc = pread(env->me_fd, page, PAGESIZE, 0)) == 0) {
992 } else if (rc != PAGESIZE) {
995 DPRINTF("read: %s", strerror(errno));
999 p = (MDB_page *)page;
1001 if (!F_ISSET(p->mp_flags, P_META)) {
1002 DPRINTF("page %lu not a meta page", p->mp_pgno);
1007 if (m->mm_magic != MDB_MAGIC) {
1008 DPRINTF("meta has invalid magic");
1012 if (m->mm_version != MDB_VERSION) {
1013 DPRINTF("database is version %u, expected version %u",
1014 m->mm_version, MDB_VERSION);
1015 return MDB_VERSION_MISMATCH;
1018 memcpy(meta, m, sizeof(*m));
1023 mdb_env_init_meta(MDB_env *env, MDB_meta *meta)
1030 DPRINTF("writing new meta page");
1031 psize = sysconf(_SC_PAGE_SIZE);
1033 meta->mm_magic = MDB_MAGIC;
1034 meta->mm_version = MDB_VERSION;
1035 meta->mm_psize = psize;
1036 meta->mm_last_pg = 1;
1037 meta->mm_flags = env->me_flags & 0xffff;
1038 meta->mm_flags |= MDB_INTEGERKEY;
1039 meta->mm_dbs[0].md_root = P_INVALID;
1040 meta->mm_dbs[1].md_root = P_INVALID;
1042 p = calloc(2, psize);
1044 p->mp_flags = P_META;
1047 memcpy(m, meta, sizeof(*meta));
1049 q = (MDB_page *)((char *)p + psize);
1052 q->mp_flags = P_META;
1055 memcpy(m, meta, sizeof(*meta));
1057 rc = write(env->me_fd, p, psize * 2);
1059 return (rc == (int)psize * 2) ? MDB_SUCCESS : errno;
1063 mdb_env_write_meta(MDB_txn *txn)
1066 MDB_meta meta, metab;
1068 int rc, len, toggle;
1071 assert(txn != NULL);
1072 assert(txn->mt_env != NULL);
1074 toggle = !F_ISSET(txn->mt_flags, MDB_TXN_METOGGLE);
1075 DPRINTF("writing meta page %d for root page %lu",
1076 toggle, txn->mt_dbs[MAIN_DBI].md_root);
1080 metab.mm_txnid = env->me_metas[toggle]->mm_txnid;
1081 metab.mm_last_pg = env->me_metas[toggle]->mm_last_pg;
1083 ptr = (char *)&meta;
1084 off = offsetof(MDB_meta, mm_dbs[0].md_depth);
1085 len = sizeof(MDB_meta) - off;
1088 meta.mm_dbs[0] = txn->mt_dbs[0];
1089 meta.mm_dbs[1] = txn->mt_dbs[1];
1090 meta.mm_last_pg = txn->mt_next_pgno - 1;
1091 meta.mm_txnid = txn->mt_txnid;
1094 off += env->me_psize;
1097 /* Write to the SYNC fd */
1098 rc = pwrite(env->me_mfd, ptr, len, off);
1100 DPRINTF("write failed, disk error?");
1101 /* On a failure, the pagecache still contains the new data.
1102 * Write some old data back, to prevent it from being used.
1103 * Use the non-SYNC fd; we know it will fail anyway.
1105 meta.mm_last_pg = metab.mm_last_pg;
1106 meta.mm_txnid = metab.mm_txnid;
1107 rc = pwrite(env->me_fd, ptr, len, off);
1108 env->me_flags |= MDB_FATAL_ERROR;
1111 txn->mt_env->me_txns->mti_me_toggle = toggle;
1117 mdb_env_read_meta(MDB_env *env, int *which)
1121 assert(env != NULL);
1125 else if (env->me_metas[0]->mm_txnid < env->me_metas[1]->mm_txnid)
1128 if (env->me_meta != env->me_metas[toggle])
1129 env->me_meta = env->me_metas[toggle];
1131 DPRINTF("Using meta page %d", toggle);
1137 mdb_env_create(MDB_env **env)
1141 e = calloc(1, sizeof(MDB_env));
1142 if (!e) return ENOMEM;
1144 e->me_maxreaders = DEFAULT_READERS;
1154 mdb_env_set_mapsize(MDB_env *env, size_t size)
1158 env->me_mapsize = size;
1163 mdb_env_set_maxdbs(MDB_env *env, int dbs)
1165 env->me_maxdbs = dbs;
1170 mdb_env_set_maxreaders(MDB_env *env, int readers)
1172 env->me_maxreaders = readers;
1177 mdb_env_get_maxreaders(MDB_env *env, int *readers)
1179 if (!env || !readers)
1181 *readers = env->me_maxreaders;
1186 mdb_env_open2(MDB_env *env, unsigned int flags)
1192 env->me_flags = flags;
1194 memset(&meta, 0, sizeof(meta));
1196 if ((i = mdb_env_read_header(env, &meta)) != 0) {
1199 DPRINTF("new mdbenv");
1203 if (!env->me_mapsize) {
1204 env->me_mapsize = newenv ? DEFAULT_MAPSIZE : meta.mm_mapsize;
1208 if (meta.mm_address && (flags & MDB_FIXEDMAP))
1210 env->me_map = mmap(meta.mm_address, env->me_mapsize, PROT_READ, i,
1212 if (env->me_map == MAP_FAILED)
1216 meta.mm_mapsize = env->me_mapsize;
1217 if (flags & MDB_FIXEDMAP)
1218 meta.mm_address = env->me_map;
1219 i = mdb_env_init_meta(env, &meta);
1220 if (i != MDB_SUCCESS) {
1221 munmap(env->me_map, env->me_mapsize);
1225 env->me_psize = meta.mm_psize;
1227 env->me_maxpg = env->me_mapsize / env->me_psize;
1229 p = (MDB_page *)env->me_map;
1230 env->me_metas[0] = METADATA(p);
1231 env->me_metas[1] = (MDB_meta *)((char *)env->me_metas[0] + meta.mm_psize);
1233 if ((i = mdb_env_read_meta(env, NULL)) != 0)
1236 DPRINTF("opened database version %u, pagesize %u",
1237 env->me_meta->mm_version, env->me_psize);
1238 DPRINTF("depth: %u", env->me_meta->mm_dbs[MAIN_DBI].md_depth);
1239 DPRINTF("entries: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_entries);
1240 DPRINTF("branch pages: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_branch_pages);
1241 DPRINTF("leaf pages: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_leaf_pages);
1242 DPRINTF("overflow pages: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_overflow_pages);
1243 DPRINTF("root: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_root);
1249 mdb_env_reader_dest(void *ptr)
1251 MDB_reader *reader = ptr;
1253 reader->mr_txnid = 0;
1258 /* downgrade the exclusive lock on the region back to shared */
1260 mdb_env_share_locks(MDB_env *env)
1262 struct flock lock_info;
1264 env->me_txns->mti_txnid = env->me_meta->mm_txnid;
1265 if (env->me_metas[0]->mm_txnid < env->me_metas[1]->mm_txnid)
1266 env->me_txns->mti_me_toggle = 1;
1268 memset((void *)&lock_info, 0, sizeof(lock_info));
1269 lock_info.l_type = F_RDLCK;
1270 lock_info.l_whence = SEEK_SET;
1271 lock_info.l_start = 0;
1272 lock_info.l_len = 1;
1273 fcntl(env->me_lfd, F_SETLK, &lock_info);
1277 mdb_env_setup_locks(MDB_env *env, char *lpath, int mode, int *excl)
1281 struct flock lock_info;
1285 if ((env->me_lfd = open(lpath, O_RDWR|O_CREAT, mode)) == -1) {
1289 /* Try to get exclusive lock. If we succeed, then
1290 * nobody is using the lock region and we should initialize it.
1292 memset((void *)&lock_info, 0, sizeof(lock_info));
1293 lock_info.l_type = F_WRLCK;
1294 lock_info.l_whence = SEEK_SET;
1295 lock_info.l_start = 0;
1296 lock_info.l_len = 1;
1297 rc = fcntl(env->me_lfd, F_SETLK, &lock_info);
1301 lock_info.l_type = F_RDLCK;
1302 rc = fcntl(env->me_lfd, F_SETLK, &lock_info);
1308 size = lseek(env->me_lfd, 0, SEEK_END);
1309 rsize = (env->me_maxreaders-1) * sizeof(MDB_reader) + sizeof(MDB_txninfo);
1310 if (size < rsize && *excl) {
1311 if (ftruncate(env->me_lfd, rsize) != 0) {
1317 size = rsize - sizeof(MDB_txninfo);
1318 env->me_maxreaders = size/sizeof(MDB_reader) + 1;
1320 env->me_txns = mmap(0, rsize, PROT_READ|PROT_WRITE, MAP_SHARED,
1322 if (env->me_txns == MAP_FAILED) {
1327 pthread_mutexattr_t mattr;
1329 pthread_mutexattr_init(&mattr);
1330 rc = pthread_mutexattr_setpshared(&mattr, PTHREAD_PROCESS_SHARED);
1334 pthread_mutex_init(&env->me_txns->mti_mutex, &mattr);
1335 pthread_mutex_init(&env->me_txns->mti_wmutex, &mattr);
1336 env->me_txns->mti_version = MDB_VERSION;
1337 env->me_txns->mti_magic = MDB_MAGIC;
1338 env->me_txns->mti_txnid = 0;
1339 env->me_txns->mti_numreaders = 0;
1340 env->me_txns->mti_me_toggle = 0;
1343 if (env->me_txns->mti_magic != MDB_MAGIC) {
1344 DPRINTF("lock region has invalid magic");
1348 if (env->me_txns->mti_version != MDB_VERSION) {
1349 DPRINTF("lock region is version %u, expected version %u",
1350 env->me_txns->mti_version, MDB_VERSION);
1351 rc = MDB_VERSION_MISMATCH;
1354 if (errno != EACCES && errno != EAGAIN) {
1368 #define LOCKNAME "/lock.mdb"
1369 #define DATANAME "/data.mdb"
1371 mdb_env_open(MDB_env *env, const char *path, unsigned int flags, mode_t mode)
1373 int oflags, rc, len, excl;
1374 char *lpath, *dpath;
1377 lpath = malloc(len + sizeof(LOCKNAME) + len + sizeof(DATANAME));
1380 dpath = lpath + len + sizeof(LOCKNAME);
1381 sprintf(lpath, "%s" LOCKNAME, path);
1382 sprintf(dpath, "%s" DATANAME, path);
1384 rc = mdb_env_setup_locks(env, lpath, mode, &excl);
1388 if (F_ISSET(flags, MDB_RDONLY))
1391 oflags = O_RDWR | O_CREAT;
1393 if ((env->me_fd = open(dpath, oflags, mode)) == -1) {
1398 if ((rc = mdb_env_open2(env, flags)) == MDB_SUCCESS) {
1399 /* synchronous fd for meta writes */
1400 if (!(flags & (MDB_RDONLY|MDB_NOSYNC)))
1402 if ((env->me_mfd = open(dpath, oflags, mode)) == -1) {
1407 env->me_path = strdup(path);
1408 DPRINTF("opened dbenv %p", (void *) env);
1409 pthread_key_create(&env->me_txkey, mdb_env_reader_dest);
1411 mdb_env_share_locks(env);
1412 env->me_dbxs = calloc(env->me_maxdbs, sizeof(MDB_dbx));
1413 env->me_dbs[0] = calloc(env->me_maxdbs, sizeof(MDB_db));
1414 env->me_dbs[1] = calloc(env->me_maxdbs, sizeof(MDB_db));
1420 if (env->me_fd >= 0) {
1424 if (env->me_lfd >= 0) {
1434 mdb_env_close(MDB_env *env)
1439 free(env->me_dbs[1]);
1440 free(env->me_dbs[0]);
1444 pthread_key_delete(env->me_txkey);
1447 munmap(env->me_map, env->me_mapsize);
1452 pid_t pid = getpid();
1453 size_t size = (env->me_maxreaders-1) * sizeof(MDB_reader) + sizeof(MDB_txninfo);
1455 for (i=0; i<env->me_txns->mti_numreaders; i++)
1456 if (env->me_txns->mti_readers[i].mr_pid == pid)
1457 env->me_txns->mti_readers[i].mr_pid = 0;
1458 munmap(env->me_txns, size);
1464 /* Search for key within a leaf page, using binary search.
1465 * Returns the smallest entry larger or equal to the key.
1466 * If exactp is non-null, stores whether the found entry was an exact match
1467 * in *exactp (1 or 0).
1468 * If kip is non-null, stores the index of the found entry in *kip.
1469 * If no entry larger or equal to the key is found, returns NULL.
1472 mdb_search_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp, MDB_val *key,
1473 int *exactp, unsigned int *kip)
1481 DPRINTF("searching %u keys in %s page %lu",
1483 IS_LEAF(mp) ? "leaf" : "branch",
1486 assert(NUMKEYS(mp) > 0);
1488 memset(&nodekey, 0, sizeof(nodekey));
1490 low = IS_LEAF(mp) ? 0 : 1;
1491 high = NUMKEYS(mp) - 1;
1492 while (low <= high) {
1493 i = (low + high) >> 1;
1494 node = NODEPTR(mp, i);
1496 nodekey.mv_size = node->mn_ksize;
1497 nodekey.mv_data = NODEKEY(node);
1499 if (txn->mt_dbxs[dbi].md_cmp)
1500 rc = txn->mt_dbxs[dbi].md_cmp(key, &nodekey);
1502 rc = _mdb_cmp(txn, dbi, key, &nodekey);
1505 DPRINTF("found leaf index %u [%.*s], rc = %i",
1506 i, (int)nodekey.mv_size, (char *)nodekey.mv_data, rc);
1508 DPRINTF("found branch index %u [%.*s -> %lu], rc = %i",
1509 i, (int)node->mn_ksize, (char *)NODEKEY(node),
1520 if (rc > 0) { /* Found entry is less than the key. */
1521 i++; /* Skip to get the smallest entry larger than key. */
1522 if (i >= NUMKEYS(mp))
1523 /* There is no entry larger or equal to the key. */
1527 *exactp = (rc == 0);
1528 if (kip) /* Store the key index if requested. */
1531 return NODEPTR(mp, i);
1535 cursor_pop_page(MDB_cursor *cursor)
1539 if (cursor->mc_snum) {
1540 top = CURSOR_TOP(cursor);
1543 DPRINTF("popped page %lu off cursor %p", top->mp_page->mp_pgno, (void *) cursor);
1548 cursor_push_page(MDB_cursor *cursor, MDB_page *mp)
1552 DPRINTF("pushing page %lu on cursor %p", mp->mp_pgno, (void *) cursor);
1554 ppage = &cursor->mc_stack[cursor->mc_snum++];
1555 ppage->mp_page = mp;
1561 mdb_get_page(MDB_txn *txn, pgno_t pgno)
1566 if (!F_ISSET(txn->mt_flags, MDB_TXN_RDONLY) && txn->mt_u.dirty_list[0].mid) {
1571 x = mdb_midl2_search(txn->mt_u.dirty_list, &id);
1572 if (x <= txn->mt_u.dirty_list[0].mid && txn->mt_u.dirty_list[x].mid == pgno) {
1573 dp = txn->mt_u.dirty_list[x].mptr;
1579 if (pgno > txn->mt_env->me_meta->mm_last_pg)
1581 p = (MDB_page *)(txn->mt_env->me_map + txn->mt_env->me_psize * pgno);
1587 mdb_search_page_root(MDB_txn *txn, MDB_dbi dbi, MDB_val *key,
1588 MDB_cursor *cursor, int modify, MDB_pageparent *mpp)
1590 MDB_page *mp = mpp->mp_page;
1593 if (cursor && cursor_push_page(cursor, mp) == NULL)
1596 while (IS_BRANCH(mp)) {
1600 DPRINTF("branch page %lu has %u keys", mp->mp_pgno, NUMKEYS(mp));
1601 assert(NUMKEYS(mp) > 1);
1602 DPRINTF("found index 0 to page %lu", NODEPGNO(NODEPTR(mp, 0)));
1604 if (key == NULL) /* Initialize cursor to first page. */
1606 else if (key->mv_size > MAXKEYSIZE && key->mv_data == NULL) {
1607 /* cursor to last page */
1611 node = mdb_search_node(txn, dbi, mp, key, &exact, &i);
1613 i = NUMKEYS(mp) - 1;
1621 DPRINTF("following index %u for key %.*s",
1622 i, (int)key->mv_size, (char *)key->mv_data);
1623 assert(i < NUMKEYS(mp));
1624 node = NODEPTR(mp, i);
1627 CURSOR_TOP(cursor)->mp_ki = i;
1629 mpp->mp_parent = mp;
1630 if ((mp = mdb_get_page(txn, NODEPGNO(node))) == NULL)
1631 return MDB_PAGE_NOTFOUND;
1635 if (cursor && cursor_push_page(cursor, mp) == NULL)
1639 MDB_dhead *dh = ((MDB_dhead *)mp)-1;
1640 if ((rc = mdb_touch(txn, mpp)) != 0)
1642 dh = ((MDB_dhead *)mpp->mp_page)-1;
1643 dh->md_parent = mpp->mp_parent;
1644 dh->md_pi = mpp->mp_pi;
1651 DPRINTF("internal error, index points to a %02X page!?",
1653 return MDB_CORRUPTED;
1656 DPRINTF("found leaf page %lu for key %.*s", mp->mp_pgno,
1657 key ? (int)key->mv_size : 0, key ? (char *)key->mv_data : NULL);
1662 /* Search for the page a given key should be in.
1663 * Stores a pointer to the found page in *mpp.
1664 * If key is NULL, search for the lowest page (used by mdb_cursor_first).
1665 * If cursor is non-null, pushes parent pages on the cursor stack.
1666 * If modify is true, visited pages are updated with new page numbers.
1669 mdb_search_page(MDB_txn *txn, MDB_dbi dbi, MDB_val *key,
1670 MDB_cursor *cursor, int modify, MDB_pageparent *mpp)
1675 /* Choose which root page to start with. If a transaction is given
1676 * use the root page from the transaction, otherwise read the last
1677 * committed root page.
1679 if (F_ISSET(txn->mt_flags, MDB_TXN_ERROR)) {
1680 DPRINTF("transaction has failed, must abort");
1683 root = txn->mt_dbs[dbi].md_root;
1685 if (root == P_INVALID) { /* Tree is empty. */
1686 DPRINTF("tree is empty");
1687 return MDB_NOTFOUND;
1690 if ((mpp->mp_page = mdb_get_page(txn, root)) == NULL)
1691 return MDB_PAGE_NOTFOUND;
1693 DPRINTF("root page has flags 0x%X", mpp->mp_page->mp_flags);
1696 /* For sub-databases, update main root first */
1697 if (dbi > MAIN_DBI && !txn->mt_dbxs[dbi].md_dirty) {
1699 rc = mdb_search_page(txn, MAIN_DBI, &txn->mt_dbxs[dbi].md_name,
1703 txn->mt_dbxs[dbi].md_dirty = 1;
1705 if (!F_ISSET(mpp->mp_page->mp_flags, P_DIRTY)) {
1706 mpp->mp_parent = NULL;
1708 if ((rc = mdb_touch(txn, mpp)))
1710 txn->mt_dbs[dbi].md_root = mpp->mp_page->mp_pgno;
1714 return mdb_search_page_root(txn, dbi, key, cursor, modify, mpp);
1718 mdb_read_data(MDB_txn *txn, MDB_node *leaf, MDB_val *data)
1720 MDB_page *omp; /* overflow mpage */
1723 if (!F_ISSET(leaf->mn_flags, F_BIGDATA)) {
1724 data->mv_size = leaf->mn_dsize;
1725 data->mv_data = NODEDATA(leaf);
1729 /* Read overflow data.
1731 data->mv_size = leaf->mn_dsize;
1732 memcpy(&pgno, NODEDATA(leaf), sizeof(pgno));
1733 if ((omp = mdb_get_page(txn, pgno)) == NULL) {
1734 DPRINTF("read overflow page %lu failed", pgno);
1735 return MDB_PAGE_NOTFOUND;
1737 data->mv_data = METADATA(omp);
1743 mdb_get(MDB_txn *txn, MDB_dbi dbi,
1744 MDB_val *key, MDB_val *data)
1752 DPRINTF("===> get key [%.*s]", (int)key->mv_size, (char *)key->mv_data);
1754 if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
1757 if (key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
1761 if ((rc = mdb_search_page(txn, dbi, key, NULL, 0, &mpp)) != MDB_SUCCESS)
1764 leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, NULL);
1765 if (leaf && exact) {
1766 /* Return first duplicate data item */
1767 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
1770 mdb_xcursor_init0(txn, dbi, &mx);
1771 mdb_xcursor_init1(txn, dbi, &mx, leaf);
1772 rc = mdb_search_page(&mx.mx_txn, mx.mx_cursor.mc_dbi, NULL, NULL, 0, &mpp);
1773 if (rc != MDB_SUCCESS)
1775 leaf = NODEPTR(mpp.mp_page, 0);
1777 rc = mdb_read_data(txn, leaf, data);
1786 mdb_sibling(MDB_cursor *cursor, int move_right)
1793 if (cursor->mc_snum < 2) {
1794 return MDB_NOTFOUND; /* root has no siblings */
1796 parent = CURSOR_PARENT(cursor);
1798 DPRINTF("parent page is page %lu, index %u",
1799 parent->mp_page->mp_pgno, parent->mp_ki);
1801 cursor_pop_page(cursor);
1802 if (move_right ? (parent->mp_ki + 1 >= NUMKEYS(parent->mp_page))
1803 : (parent->mp_ki == 0)) {
1804 DPRINTF("no more keys left, moving to %s sibling",
1805 move_right ? "right" : "left");
1806 if ((rc = mdb_sibling(cursor, move_right)) != MDB_SUCCESS)
1808 parent = CURSOR_TOP(cursor);
1814 DPRINTF("just moving to %s index key %u",
1815 move_right ? "right" : "left", parent->mp_ki);
1817 assert(IS_BRANCH(parent->mp_page));
1819 indx = NODEPTR(parent->mp_page, parent->mp_ki);
1820 if ((mp = mdb_get_page(cursor->mc_txn, indx->mn_pgno)) == NULL)
1821 return MDB_PAGE_NOTFOUND;
1823 mp->parent = parent->mp_page;
1824 mp->parent_index = parent->mp_ki;
1827 cursor_push_page(cursor, mp);
1833 mdb_set_key(MDB_node *node, MDB_val *key)
1838 key->mv_size = node->mn_ksize;
1839 key->mv_data = NODEKEY(node);
1845 mdb_cursor_next(MDB_cursor *cursor, MDB_val *key, MDB_val *data, MDB_cursor_op op)
1852 if (cursor->mc_eof) {
1853 return MDB_NOTFOUND;
1856 assert(cursor->mc_initialized);
1858 top = CURSOR_TOP(cursor);
1861 if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
1862 leaf = NODEPTR(mp, top->mp_ki);
1863 if ((op == MDB_NEXT || op == MDB_NEXT_DUP) && F_ISSET(leaf->mn_flags, F_DUPDATA)) {
1864 rc = mdb_cursor_next(&cursor->mc_xcursor->mx_cursor, data, NULL, MDB_NEXT);
1865 if (op != MDB_NEXT || rc == MDB_SUCCESS)
1870 DPRINTF("cursor_next: top page is %lu in cursor %p", mp->mp_pgno, (void *) cursor);
1872 if (top->mp_ki + 1 >= NUMKEYS(mp)) {
1873 DPRINTF("=====> move to next sibling page");
1874 if (mdb_sibling(cursor, 1) != MDB_SUCCESS) {
1876 return MDB_NOTFOUND;
1878 top = CURSOR_TOP(cursor);
1880 DPRINTF("next page is %lu, key index %u", mp->mp_pgno, top->mp_ki);
1884 DPRINTF("==> cursor points to page %lu with %u keys, key index %u",
1885 mp->mp_pgno, NUMKEYS(mp), top->mp_ki);
1887 assert(IS_LEAF(mp));
1888 leaf = NODEPTR(mp, top->mp_ki);
1890 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
1891 mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, leaf);
1894 if ((rc = mdb_read_data(cursor->mc_txn, leaf, data) != MDB_SUCCESS))
1897 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
1898 rc = mdb_cursor_first(&cursor->mc_xcursor->mx_cursor, data, NULL);
1899 if (rc != MDB_SUCCESS)
1904 return mdb_set_key(leaf, key);
1908 mdb_cursor_prev(MDB_cursor *cursor, MDB_val *key, MDB_val *data, MDB_cursor_op op)
1915 assert(cursor->mc_initialized);
1917 top = CURSOR_TOP(cursor);
1920 if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
1921 leaf = NODEPTR(mp, top->mp_ki);
1922 if ((op == MDB_PREV || op == MDB_PREV_DUP) && F_ISSET(leaf->mn_flags, F_DUPDATA)) {
1923 rc = mdb_cursor_prev(&cursor->mc_xcursor->mx_cursor, data, NULL, MDB_PREV);
1924 if (op != MDB_PREV || rc == MDB_SUCCESS)
1929 DPRINTF("cursor_prev: top page is %lu in cursor %p", mp->mp_pgno, (void *) cursor);
1931 if (top->mp_ki == 0) {
1932 DPRINTF("=====> move to prev sibling page");
1933 if (mdb_sibling(cursor, 0) != MDB_SUCCESS) {
1934 return MDB_NOTFOUND;
1936 top = CURSOR_TOP(cursor);
1938 top->mp_ki = NUMKEYS(mp) - 1;
1939 DPRINTF("prev page is %lu, key index %u", mp->mp_pgno, top->mp_ki);
1945 DPRINTF("==> cursor points to page %lu with %u keys, key index %u",
1946 mp->mp_pgno, NUMKEYS(mp), top->mp_ki);
1948 assert(IS_LEAF(mp));
1949 leaf = NODEPTR(mp, top->mp_ki);
1951 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
1952 mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, leaf);
1955 if ((rc = mdb_read_data(cursor->mc_txn, leaf, data) != MDB_SUCCESS))
1958 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
1959 rc = mdb_cursor_last(&cursor->mc_xcursor->mx_cursor, data, NULL);
1960 if (rc != MDB_SUCCESS)
1965 return mdb_set_key(leaf, key);
1969 mdb_cursor_set(MDB_cursor *cursor, MDB_val *key, MDB_val *data,
1970 MDB_cursor_op op, int *exactp)
1979 assert(key->mv_size > 0);
1981 cursor->mc_snum = 0;
1983 rc = mdb_search_page(cursor->mc_txn, cursor->mc_dbi, key, cursor, 0, &mpp);
1984 if (rc != MDB_SUCCESS)
1986 assert(IS_LEAF(mpp.mp_page));
1988 top = CURSOR_TOP(cursor);
1989 leaf = mdb_search_node(cursor->mc_txn, cursor->mc_dbi, mpp.mp_page, key, exactp, &top->mp_ki);
1990 if (exactp != NULL && !*exactp) {
1991 /* MDB_SET specified and not an exact match. */
1992 return MDB_NOTFOUND;
1996 DPRINTF("===> inexact leaf not found, goto sibling");
1997 if ((rc = mdb_sibling(cursor, 1)) != MDB_SUCCESS)
1998 return rc; /* no entries matched */
1999 top = CURSOR_TOP(cursor);
2001 mpp.mp_page = top->mp_page;
2002 assert(IS_LEAF(mpp.mp_page));
2003 leaf = NODEPTR(mpp.mp_page, 0);
2006 cursor->mc_initialized = 1;
2009 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
2010 mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, leaf);
2013 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
2014 if (op == MDB_SET || op == MDB_SET_RANGE) {
2015 rc = mdb_cursor_first(&cursor->mc_xcursor->mx_cursor, data, NULL);
2019 if (op == MDB_GET_BOTH) {
2024 op2 = MDB_SET_RANGE;
2026 rc = mdb_cursor_set(&cursor->mc_xcursor->mx_cursor, data, NULL, op2, ex2p);
2027 if (rc != MDB_SUCCESS)
2031 if ((rc = mdb_read_data(cursor->mc_txn, leaf, data)) != MDB_SUCCESS)
2036 rc = mdb_set_key(leaf, key);
2037 if (rc == MDB_SUCCESS) {
2038 DPRINTF("==> cursor placed on key %.*s",
2039 (int)key->mv_size, (char *)key->mv_data);
2047 mdb_cursor_first(MDB_cursor *cursor, MDB_val *key, MDB_val *data)
2053 cursor->mc_snum = 0;
2055 rc = mdb_search_page(cursor->mc_txn, cursor->mc_dbi, NULL, cursor, 0, &mpp);
2056 if (rc != MDB_SUCCESS)
2058 assert(IS_LEAF(mpp.mp_page));
2060 leaf = NODEPTR(mpp.mp_page, 0);
2061 cursor->mc_initialized = 1;
2065 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
2066 mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, leaf);
2067 rc = mdb_cursor_first(&cursor->mc_xcursor->mx_cursor, data, NULL);
2071 if ((rc = mdb_read_data(cursor->mc_txn, leaf, data)) != MDB_SUCCESS)
2075 return mdb_set_key(leaf, key);
2079 mdb_cursor_last(MDB_cursor *cursor, MDB_val *key, MDB_val *data)
2087 cursor->mc_snum = 0;
2089 lkey.mv_size = MAXKEYSIZE+1;
2090 lkey.mv_data = NULL;
2092 rc = mdb_search_page(cursor->mc_txn, cursor->mc_dbi, &lkey, cursor, 0, &mpp);
2093 if (rc != MDB_SUCCESS)
2095 assert(IS_LEAF(mpp.mp_page));
2097 leaf = NODEPTR(mpp.mp_page, NUMKEYS(mpp.mp_page)-1);
2098 cursor->mc_initialized = 1;
2101 top = CURSOR_TOP(cursor);
2102 top->mp_ki = NUMKEYS(top->mp_page) - 1;
2105 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
2106 mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, leaf);
2107 rc = mdb_cursor_last(&cursor->mc_xcursor->mx_cursor, data, NULL);
2111 if ((rc = mdb_read_data(cursor->mc_txn, leaf, data)) != MDB_SUCCESS)
2116 return mdb_set_key(leaf, key);
2120 mdb_cursor_get(MDB_cursor *cursor, MDB_val *key, MDB_val *data,
2130 case MDB_GET_BOTH_RANGE:
2138 if (key == NULL || key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
2140 } else if (op != MDB_SET_RANGE)
2141 rc = mdb_cursor_set(cursor, key, data, op, NULL);
2143 rc = mdb_cursor_set(cursor, key, data, op, &exact);
2147 case MDB_NEXT_NODUP:
2148 if (!cursor->mc_initialized)
2149 rc = mdb_cursor_first(cursor, key, data);
2151 rc = mdb_cursor_next(cursor, key, data, op);
2155 case MDB_PREV_NODUP:
2156 if (!cursor->mc_initialized || cursor->mc_eof)
2157 rc = mdb_cursor_last(cursor, key, data);
2159 rc = mdb_cursor_prev(cursor, key, data, op);
2162 rc = mdb_cursor_first(cursor, key, data);
2165 rc = mdb_cursor_last(cursor, key, data);
2168 DPRINTF("unhandled/unimplemented cursor operation %u", op);
2176 /* Allocate a page and initialize it
2179 mdb_new_page(MDB_txn *txn, MDB_dbi dbi, uint32_t flags, int num)
2183 if ((dp = mdb_alloc_page(txn, NULL, 0, num)) == NULL)
2185 DPRINTF("allocated new mpage %lu, page size %u",
2186 dp->p.mp_pgno, txn->mt_env->me_psize);
2187 dp->p.mp_flags = flags | P_DIRTY;
2188 dp->p.mp_lower = PAGEHDRSZ;
2189 dp->p.mp_upper = txn->mt_env->me_psize;
2191 if (IS_BRANCH(&dp->p))
2192 txn->mt_dbs[dbi].md_branch_pages++;
2193 else if (IS_LEAF(&dp->p))
2194 txn->mt_dbs[dbi].md_leaf_pages++;
2195 else if (IS_OVERFLOW(&dp->p)) {
2196 txn->mt_dbs[dbi].md_overflow_pages += num;
2197 dp->p.mp_pages = num;
2204 mdb_leaf_size(MDB_env *env, MDB_val *key, MDB_val *data)
2208 sz = LEAFSIZE(key, data);
2209 if (data->mv_size >= env->me_psize / MDB_MINKEYS) {
2210 /* put on overflow page */
2211 sz -= data->mv_size - sizeof(pgno_t);
2214 return sz + sizeof(indx_t);
2218 mdb_branch_size(MDB_env *env, MDB_val *key)
2223 if (sz >= env->me_psize / MDB_MINKEYS) {
2224 /* put on overflow page */
2225 /* not implemented */
2226 /* sz -= key->size - sizeof(pgno_t); */
2229 return sz + sizeof(indx_t);
2233 mdb_add_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp, indx_t indx,
2234 MDB_val *key, MDB_val *data, pgno_t pgno, uint8_t flags)
2237 size_t node_size = NODESIZE;
2240 MDB_dpage *ofp = NULL; /* overflow page */
2242 assert(mp->mp_upper >= mp->mp_lower);
2244 DPRINTF("add node [%.*s] to %s page %lu at index %i, key size %zu",
2245 key ? (int)key->mv_size : 0, key ? (char *)key->mv_data : NULL,
2246 IS_LEAF(mp) ? "leaf" : "branch",
2247 mp->mp_pgno, indx, key ? key->mv_size : 0);
2250 node_size += key->mv_size;
2254 if (F_ISSET(flags, F_BIGDATA)) {
2255 /* Data already on overflow page. */
2256 node_size += sizeof(pgno_t);
2257 } else if (data->mv_size >= txn->mt_env->me_psize / MDB_MINKEYS) {
2258 int ovpages = OVPAGES(data->mv_size, txn->mt_env->me_psize);
2259 /* Put data on overflow page. */
2260 DPRINTF("data size is %zu, put on overflow page",
2262 node_size += sizeof(pgno_t);
2263 if ((ofp = mdb_new_page(txn, dbi, P_OVERFLOW, ovpages)) == NULL)
2265 DPRINTF("allocated overflow page %lu", ofp->p.mp_pgno);
2268 node_size += data->mv_size;
2272 if (node_size + sizeof(indx_t) > SIZELEFT(mp)) {
2273 DPRINTF("not enough room in page %lu, got %u ptrs",
2274 mp->mp_pgno, NUMKEYS(mp));
2275 DPRINTF("upper - lower = %u - %u = %u", mp->mp_upper, mp->mp_lower,
2276 mp->mp_upper - mp->mp_lower);
2277 DPRINTF("node size = %zu", node_size);
2281 /* Move higher pointers up one slot. */
2282 for (i = NUMKEYS(mp); i > indx; i--)
2283 mp->mp_ptrs[i] = mp->mp_ptrs[i - 1];
2285 /* Adjust free space offsets. */
2286 ofs = mp->mp_upper - node_size;
2287 assert(ofs >= mp->mp_lower + sizeof(indx_t));
2288 mp->mp_ptrs[indx] = ofs;
2290 mp->mp_lower += sizeof(indx_t);
2292 /* Write the node data. */
2293 node = NODEPTR(mp, indx);
2294 node->mn_ksize = (key == NULL) ? 0 : key->mv_size;
2295 node->mn_flags = flags;
2297 node->mn_dsize = data->mv_size;
2299 node->mn_pgno = pgno;
2302 memcpy(NODEKEY(node), key->mv_data, key->mv_size);
2307 if (F_ISSET(flags, F_BIGDATA))
2308 memcpy(node->mn_data + key->mv_size, data->mv_data,
2311 memcpy(node->mn_data + key->mv_size, data->mv_data,
2314 memcpy(node->mn_data + key->mv_size, &ofp->p.mp_pgno,
2316 memcpy(METADATA(&ofp->p), data->mv_data, data->mv_size);
2324 mdb_del_node(MDB_page *mp, indx_t indx)
2327 indx_t i, j, numkeys, ptr;
2331 DPRINTF("delete node %u on %s page %lu", indx,
2332 IS_LEAF(mp) ? "leaf" : "branch", mp->mp_pgno);
2333 assert(indx < NUMKEYS(mp));
2335 node = NODEPTR(mp, indx);
2336 sz = NODESIZE + node->mn_ksize;
2338 if (F_ISSET(node->mn_flags, F_BIGDATA))
2339 sz += sizeof(pgno_t);
2341 sz += NODEDSZ(node);
2344 ptr = mp->mp_ptrs[indx];
2345 numkeys = NUMKEYS(mp);
2346 for (i = j = 0; i < numkeys; i++) {
2348 mp->mp_ptrs[j] = mp->mp_ptrs[i];
2349 if (mp->mp_ptrs[i] < ptr)
2350 mp->mp_ptrs[j] += sz;
2355 base = (char *)mp + mp->mp_upper;
2356 memmove(base + sz, base, ptr - mp->mp_upper);
2358 mp->mp_lower -= sizeof(indx_t);
2363 mdb_xcursor_init0(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx)
2368 mx->mx_txn.mt_dbxs = mx->mx_dbxs;
2369 mx->mx_txn.mt_dbs = mx->mx_dbs;
2370 mx->mx_dbxs[0] = txn->mt_dbxs[0];
2371 mx->mx_dbxs[1] = txn->mt_dbxs[1];
2373 mx->mx_dbxs[2] = txn->mt_dbxs[dbi];
2378 mx->mx_dbxs[dbn+1].md_parent = dbn;
2379 mx->mx_dbxs[dbn+1].md_cmp = mx->mx_dbxs[dbn].md_dcmp;
2380 mx->mx_dbxs[dbn+1].md_rel = mx->mx_dbxs[dbn].md_rel;
2381 mx->mx_dbxs[dbn+1].md_dirty = 0;
2382 mx->mx_txn.mt_numdbs = dbn+2;
2384 mx->mx_cursor.mc_snum = 0;
2385 mx->mx_cursor.mc_txn = &mx->mx_txn;
2386 mx->mx_cursor.mc_dbi = dbn+1;
2390 mdb_xcursor_init1(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx, MDB_node *node)
2392 MDB_db *db = NODEDATA(node);
2394 mx->mx_dbs[0] = txn->mt_dbs[0];
2395 mx->mx_dbs[1] = txn->mt_dbs[1];
2397 mx->mx_dbs[2] = txn->mt_dbs[dbi];
2402 mx->mx_dbs[dbn] = *db;
2403 mx->mx_dbxs[dbn].md_name.mv_data = NODEKEY(node);
2404 mx->mx_dbxs[dbn].md_name.mv_size = node->mn_ksize;
2405 mx->mx_txn.mt_next_pgno = txn->mt_next_pgno;
2406 mx->mx_txn.mt_oldest = txn->mt_oldest;
2407 mx->mx_txn.mt_u = txn->mt_u;
2411 mdb_xcursor_fini(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx)
2413 txn->mt_next_pgno = mx->mx_txn.mt_next_pgno;
2414 txn->mt_oldest = mx->mx_txn.mt_oldest;
2415 txn->mt_u = mx->mx_txn.mt_u;
2416 txn->mt_dbs[0] = mx->mx_dbs[0];
2417 txn->mt_dbs[1] = mx->mx_dbs[1];
2418 txn->mt_dbxs[0].md_dirty = mx->mx_dbxs[0].md_dirty;
2419 txn->mt_dbxs[1].md_dirty = mx->mx_dbxs[1].md_dirty;
2421 txn->mt_dbs[dbi] = mx->mx_dbs[2];
2422 txn->mt_dbxs[dbi].md_dirty = mx->mx_dbxs[2].md_dirty;
2427 mdb_cursor_open(MDB_txn *txn, MDB_dbi dbi, MDB_cursor **ret)
2430 size_t size = sizeof(MDB_cursor);
2432 if (txn == NULL || ret == NULL || !dbi || dbi >= txn->mt_numdbs)
2435 if (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT)
2436 size += sizeof(MDB_xcursor);
2438 if ((cursor = calloc(1, size)) != NULL) {
2439 cursor->mc_dbi = dbi;
2440 cursor->mc_txn = txn;
2441 if (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT) {
2442 MDB_xcursor *mx = (MDB_xcursor *)(cursor + 1);
2443 cursor->mc_xcursor = mx;
2444 mdb_xcursor_init0(txn, dbi, mx);
2455 /* Return the count of duplicate data items for the current key */
2457 mdb_cursor_count(MDB_cursor *mc, unsigned long *countp)
2462 if (mc == NULL || countp == NULL)
2465 if (!(mc->mc_txn->mt_dbs[mc->mc_dbi].md_flags & MDB_DUPSORT))
2468 top = CURSOR_TOP(mc);
2469 leaf = NODEPTR(top->mp_page, top->mp_ki);
2470 if (!F_ISSET(leaf->mn_flags, F_DUPDATA)) {
2473 if (!mc->mc_xcursor->mx_cursor.mc_initialized)
2476 *countp = mc->mc_xcursor->mx_txn.mt_dbs[mc->mc_xcursor->mx_cursor.mc_dbi].md_entries;
2482 mdb_cursor_close(MDB_cursor *cursor)
2484 if (cursor != NULL) {
2490 mdb_update_key(MDB_page *mp, indx_t indx, MDB_val *key)
2492 indx_t ptr, i, numkeys;
2498 node = NODEPTR(mp, indx);
2499 ptr = mp->mp_ptrs[indx];
2500 DPRINTF("update key %u (ofs %u) [%.*s] to [%.*s] on page %lu",
2502 (int)node->mn_ksize, (char *)NODEKEY(node),
2503 (int)key->mv_size, (char *)key->mv_data,
2506 delta = key->mv_size - node->mn_ksize;
2508 if (delta > 0 && SIZELEFT(mp) < delta) {
2509 DPRINTF("OUCH! Not enough room, delta = %d", delta);
2513 numkeys = NUMKEYS(mp);
2514 for (i = 0; i < numkeys; i++) {
2515 if (mp->mp_ptrs[i] <= ptr)
2516 mp->mp_ptrs[i] -= delta;
2519 base = (char *)mp + mp->mp_upper;
2520 len = ptr - mp->mp_upper + NODESIZE;
2521 memmove(base - delta, base, len);
2522 mp->mp_upper -= delta;
2524 node = NODEPTR(mp, indx);
2525 node->mn_ksize = key->mv_size;
2528 memcpy(NODEKEY(node), key->mv_data, key->mv_size);
2533 /* Move a node from src to dst.
2536 mdb_move_node(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *src, indx_t srcindx,
2537 MDB_pageparent *dst, indx_t dstindx)
2543 srcnode = NODEPTR(src->mp_page, srcindx);
2544 DPRINTF("moving %s node %u [%.*s] on page %lu to node %u on page %lu",
2545 IS_LEAF(src->mp_page) ? "leaf" : "branch",
2547 (int)srcnode->mn_ksize, (char *)NODEKEY(srcnode),
2548 src->mp_page->mp_pgno,
2549 dstindx, dst->mp_page->mp_pgno);
2551 /* Mark src and dst as dirty. */
2552 if ((rc = mdb_touch(txn, src)) ||
2553 (rc = mdb_touch(txn, dst)))
2556 /* Add the node to the destination page.
2558 key.mv_size = srcnode->mn_ksize;
2559 key.mv_data = NODEKEY(srcnode);
2560 data.mv_size = NODEDSZ(srcnode);
2561 data.mv_data = NODEDATA(srcnode);
2562 rc = mdb_add_node(txn, dbi, dst->mp_page, dstindx, &key, &data, NODEPGNO(srcnode),
2564 if (rc != MDB_SUCCESS)
2567 /* Delete the node from the source page.
2569 mdb_del_node(src->mp_page, srcindx);
2571 /* Update the parent separators.
2573 if (srcindx == 0 && src->mp_pi != 0) {
2574 DPRINTF("update separator for source page %lu to [%.*s]",
2575 src->mp_page->mp_pgno, (int)key.mv_size, (char *)key.mv_data);
2576 if ((rc = mdb_update_key(src->mp_parent, src->mp_pi,
2577 &key)) != MDB_SUCCESS)
2581 if (srcindx == 0 && IS_BRANCH(src->mp_page)) {
2583 nullkey.mv_size = 0;
2584 assert(mdb_update_key(src->mp_page, 0, &nullkey) == MDB_SUCCESS);
2587 if (dstindx == 0 && dst->mp_pi != 0) {
2588 DPRINTF("update separator for destination page %lu to [%.*s]",
2589 dst->mp_page->mp_pgno, (int)key.mv_size, (char *)key.mv_data);
2590 if ((rc = mdb_update_key(dst->mp_parent, dst->mp_pi,
2591 &key)) != MDB_SUCCESS)
2595 if (dstindx == 0 && IS_BRANCH(dst->mp_page)) {
2597 nullkey.mv_size = 0;
2598 assert(mdb_update_key(dst->mp_page, 0, &nullkey) == MDB_SUCCESS);
2605 mdb_merge(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *src, MDB_pageparent *dst)
2614 DPRINTF("merging page %lu and %lu", src->mp_page->mp_pgno, dst->mp_page->mp_pgno);
2616 assert(txn != NULL);
2617 assert(src->mp_parent); /* can't merge root page */
2618 assert(dst->mp_parent);
2620 /* Mark src and dst as dirty. */
2621 if ((rc = mdb_touch(txn, src)) ||
2622 (rc = mdb_touch(txn, dst)))
2625 /* Move all nodes from src to dst.
2627 for (i = 0; i < NUMKEYS(src->mp_page); i++) {
2628 srcnode = NODEPTR(src->mp_page, i);
2630 key.mv_size = srcnode->mn_ksize;
2631 key.mv_data = NODEKEY(srcnode);
2632 data.mv_size = NODEDSZ(srcnode);
2633 data.mv_data = NODEDATA(srcnode);
2634 rc = mdb_add_node(txn, dbi, dst->mp_page, NUMKEYS(dst->mp_page), &key,
2635 &data, NODEPGNO(srcnode), srcnode->mn_flags);
2636 if (rc != MDB_SUCCESS)
2640 DPRINTF("dst page %lu now has %u keys (%.1f%% filled)",
2641 dst->mp_page->mp_pgno, NUMKEYS(dst->mp_page), (float)PAGEFILL(txn->mt_env, dst->mp_page) / 10);
2643 /* Unlink the src page from parent.
2645 mdb_del_node(src->mp_parent, src->mp_pi);
2646 if (src->mp_pi == 0) {
2648 if ((rc = mdb_update_key(src->mp_parent, 0, &key)) != MDB_SUCCESS)
2652 if (IS_LEAF(src->mp_page))
2653 txn->mt_dbs[dbi].md_leaf_pages--;
2655 txn->mt_dbs[dbi].md_branch_pages--;
2657 mpp.mp_page = src->mp_parent;
2658 dh = (MDB_dhead *)src->mp_parent;
2660 mpp.mp_parent = dh->md_parent;
2661 mpp.mp_pi = dh->md_pi;
2663 return mdb_rebalance(txn, dbi, &mpp);
2666 #define FILL_THRESHOLD 250
2669 mdb_rebalance(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *mpp)
2674 indx_t si = 0, di = 0;
2676 assert(txn != NULL);
2677 assert(mpp != NULL);
2679 DPRINTF("rebalancing %s page %lu (has %u keys, %.1f%% full)",
2680 IS_LEAF(mpp->mp_page) ? "leaf" : "branch",
2681 mpp->mp_page->mp_pgno, NUMKEYS(mpp->mp_page), (float)PAGEFILL(txn->mt_env, mpp->mp_page) / 10);
2683 if (PAGEFILL(txn->mt_env, mpp->mp_page) >= FILL_THRESHOLD) {
2684 DPRINTF("no need to rebalance page %lu, above fill threshold",
2685 mpp->mp_page->mp_pgno);
2689 if (mpp->mp_parent == NULL) {
2690 if (NUMKEYS(mpp->mp_page) == 0) {
2691 DPRINTF("tree is completely empty");
2692 txn->mt_dbs[dbi].md_root = P_INVALID;
2693 txn->mt_dbs[dbi].md_depth--;
2694 txn->mt_dbs[dbi].md_leaf_pages--;
2695 } else if (IS_BRANCH(mpp->mp_page) && NUMKEYS(mpp->mp_page) == 1) {
2696 DPRINTF("collapsing root page!");
2697 txn->mt_dbs[dbi].md_root = NODEPGNO(NODEPTR(mpp->mp_page, 0));
2698 if ((root = mdb_get_page(txn, txn->mt_dbs[dbi].md_root)) == NULL)
2699 return MDB_PAGE_NOTFOUND;
2700 txn->mt_dbs[dbi].md_depth--;
2701 txn->mt_dbs[dbi].md_branch_pages--;
2703 DPRINTF("root page doesn't need rebalancing");
2707 /* The parent (branch page) must have at least 2 pointers,
2708 * otherwise the tree is invalid.
2710 assert(NUMKEYS(mpp->mp_parent) > 1);
2712 /* Leaf page fill factor is below the threshold.
2713 * Try to move keys from left or right neighbor, or
2714 * merge with a neighbor page.
2719 if (mpp->mp_pi == 0) {
2720 /* We're the leftmost leaf in our parent.
2722 DPRINTF("reading right neighbor");
2723 node = NODEPTR(mpp->mp_parent, mpp->mp_pi + 1);
2724 if ((npp.mp_page = mdb_get_page(txn, NODEPGNO(node))) == NULL)
2725 return MDB_PAGE_NOTFOUND;
2726 npp.mp_pi = mpp->mp_pi + 1;
2728 di = NUMKEYS(mpp->mp_page);
2730 /* There is at least one neighbor to the left.
2732 DPRINTF("reading left neighbor");
2733 node = NODEPTR(mpp->mp_parent, mpp->mp_pi - 1);
2734 if ((npp.mp_page = mdb_get_page(txn, NODEPGNO(node))) == NULL)
2735 return MDB_PAGE_NOTFOUND;
2736 npp.mp_pi = mpp->mp_pi - 1;
2737 si = NUMKEYS(npp.mp_page) - 1;
2740 npp.mp_parent = mpp->mp_parent;
2742 DPRINTF("found neighbor page %lu (%u keys, %.1f%% full)",
2743 npp.mp_page->mp_pgno, NUMKEYS(npp.mp_page), (float)PAGEFILL(txn->mt_env, npp.mp_page) / 10);
2745 /* If the neighbor page is above threshold and has at least two
2746 * keys, move one key from it.
2748 * Otherwise we should try to merge them.
2750 if (PAGEFILL(txn->mt_env, npp.mp_page) >= FILL_THRESHOLD && NUMKEYS(npp.mp_page) >= 2)
2751 return mdb_move_node(txn, dbi, &npp, si, mpp, di);
2752 else { /* FIXME: if (has_enough_room()) */
2753 if (mpp->mp_pi == 0)
2754 return mdb_merge(txn, dbi, &npp, mpp);
2756 return mdb_merge(txn, dbi, mpp, &npp);
2761 mdb_del0(MDB_txn *txn, MDB_dbi dbi, unsigned int ki, MDB_pageparent *mpp, MDB_node *leaf)
2765 /* add overflow pages to free list */
2766 if (F_ISSET(leaf->mn_flags, F_BIGDATA)) {
2770 memcpy(&pg, NODEDATA(leaf), sizeof(pg));
2771 ovpages = OVPAGES(NODEDSZ(leaf), txn->mt_env->me_psize);
2772 for (i=0; i<ovpages; i++) {
2773 DPRINTF("freed ov page %lu", pg);
2774 mdb_midl_insert(txn->mt_free_pgs, pg);
2778 mdb_del_node(mpp->mp_page, ki);
2779 txn->mt_dbs[dbi].md_entries--;
2780 rc = mdb_rebalance(txn, dbi, mpp);
2781 if (rc != MDB_SUCCESS)
2782 txn->mt_flags |= MDB_TXN_ERROR;
2788 mdb_del(MDB_txn *txn, MDB_dbi dbi,
2789 MDB_val *key, MDB_val *data,
2797 DPRINTF("========> delete key %.*s", (int)key->mv_size, (char *)key->mv_data);
2799 assert(key != NULL);
2801 if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
2804 if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
2808 if (key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
2812 mpp.mp_parent = NULL;
2814 if ((rc = mdb_search_page(txn, dbi, key, NULL, 1, &mpp)) != MDB_SUCCESS)
2817 leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, &ki);
2818 if (leaf == NULL || !exact) {
2819 return MDB_NOTFOUND;
2822 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
2826 mdb_xcursor_init0(txn, dbi, &mx);
2827 mdb_xcursor_init1(txn, dbi, &mx, leaf);
2828 if (flags == MDB_DEL_DUP) {
2829 rc = mdb_del(&mx.mx_txn, mx.mx_cursor.mc_dbi, data, NULL, 0);
2830 mdb_xcursor_fini(txn, dbi, &mx);
2831 if (rc != MDB_SUCCESS)
2833 /* If sub-DB still has entries, we're done */
2834 if (mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi].md_root != P_INVALID) {
2835 memcpy(NODEDATA(leaf), &mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi],
2839 /* otherwise fall thru and delete the sub-DB */
2841 /* add all the child DB's pages to the free list */
2842 rc = mdb_search_page(&mx.mx_txn, mx.mx_cursor.mc_dbi,
2843 NULL, &mx.mx_cursor, 0, &mp2);
2844 if (rc == MDB_SUCCESS) {
2845 MDB_ppage *top, *parent;
2849 cursor_pop_page(&mx.mx_cursor);
2850 if (mx.mx_cursor.mc_snum) {
2851 top = CURSOR_TOP(&mx.mx_cursor);
2852 while (mx.mx_cursor.mc_snum > 1) {
2853 parent = CURSOR_PARENT(&mx.mx_cursor);
2854 for (i=0; i<NUMKEYS(top->mp_page); i++) {
2855 ni = NODEPTR(top->mp_page, i);
2856 mdb_midl_insert(txn->mt_free_pgs, ni->mn_pgno);
2859 if (parent->mp_ki >= NUMKEYS(parent->mp_page)) {
2860 cursor_pop_page(&mx.mx_cursor);
2863 ni = NODEPTR(parent->mp_page, parent->mp_ki);
2864 top->mp_page = mdb_get_page(&mx.mx_txn, ni->mn_pgno);
2868 mdb_midl_insert(txn->mt_free_pgs, mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi].md_root);
2873 if (data && (rc = mdb_read_data(txn, leaf, data)) != MDB_SUCCESS)
2876 return mdb_del0(txn, dbi, ki, &mpp, leaf);
2879 /* Split page <*mpp>, and insert <key,(data|newpgno)> in either left or
2880 * right sibling, at index <*newindxp> (as if unsplit). Updates *mpp and
2881 * *newindxp with the actual values after split, ie if *mpp and *newindxp
2882 * refer to a node in the new right sibling page.
2885 mdb_split(MDB_txn *txn, MDB_dbi dbi, MDB_page **mpp, unsigned int *newindxp,
2886 MDB_val *newkey, MDB_val *newdata, pgno_t newpgno)
2889 int rc = MDB_SUCCESS, ins_new = 0;
2892 unsigned int i, j, split_indx;
2894 MDB_val sepkey, rkey, rdata;
2896 MDB_dpage *mdp, *rdp, *pdp;
2899 assert(txn != NULL);
2901 dh = ((MDB_dhead *)*mpp) - 1;
2902 mdp = (MDB_dpage *)dh;
2903 newindx = *newindxp;
2905 DPRINTF("-----> splitting %s page %lu and adding [%.*s] at index %i",
2906 IS_LEAF(&mdp->p) ? "leaf" : "branch", mdp->p.mp_pgno,
2907 (int)newkey->mv_size, (char *)newkey->mv_data, *newindxp);
2909 if (mdp->h.md_parent == NULL) {
2910 if ((pdp = mdb_new_page(txn, dbi, P_BRANCH, 1)) == NULL)
2913 mdp->h.md_parent = &pdp->p;
2914 txn->mt_dbs[dbi].md_root = pdp->p.mp_pgno;
2915 DPRINTF("root split! new root = %lu", pdp->p.mp_pgno);
2916 txn->mt_dbs[dbi].md_depth++;
2918 /* Add left (implicit) pointer. */
2919 if ((rc = mdb_add_node(txn, dbi, &pdp->p, 0, NULL, NULL,
2920 mdp->p.mp_pgno, 0)) != MDB_SUCCESS)
2923 DPRINTF("parent branch page is %lu", mdp->h.md_parent->mp_pgno);
2926 /* Create a right sibling. */
2927 if ((rdp = mdb_new_page(txn, dbi, mdp->p.mp_flags, 1)) == NULL)
2929 rdp->h.md_parent = mdp->h.md_parent;
2930 rdp->h.md_pi = mdp->h.md_pi + 1;
2931 DPRINTF("new right sibling: page %lu", rdp->p.mp_pgno);
2933 /* Move half of the keys to the right sibling. */
2934 if ((copy = malloc(txn->mt_env->me_psize)) == NULL)
2936 memcpy(copy, &mdp->p, txn->mt_env->me_psize);
2937 memset(&mdp->p.mp_ptrs, 0, txn->mt_env->me_psize - PAGEHDRSZ);
2938 mdp->p.mp_lower = PAGEHDRSZ;
2939 mdp->p.mp_upper = txn->mt_env->me_psize;
2941 split_indx = NUMKEYS(copy) / 2 + 1;
2943 /* First find the separating key between the split pages.
2945 memset(&sepkey, 0, sizeof(sepkey));
2946 if (newindx == split_indx) {
2947 sepkey.mv_size = newkey->mv_size;
2948 sepkey.mv_data = newkey->mv_data;
2950 node = NODEPTR(copy, split_indx);
2951 sepkey.mv_size = node->mn_ksize;
2952 sepkey.mv_data = NODEKEY(node);
2955 DPRINTF("separator is [%.*s]", (int)sepkey.mv_size, (char *)sepkey.mv_data);
2957 /* Copy separator key to the parent.
2959 if (SIZELEFT(rdp->h.md_parent) < mdb_branch_size(txn->mt_env, &sepkey)) {
2960 rc = mdb_split(txn, dbi, &rdp->h.md_parent, &rdp->h.md_pi,
2961 &sepkey, NULL, rdp->p.mp_pgno);
2963 /* Right page might now have changed parent.
2964 * Check if left page also changed parent.
2966 if (rdp->h.md_parent != mdp->h.md_parent &&
2967 mdp->h.md_pi >= NUMKEYS(mdp->h.md_parent)) {
2968 mdp->h.md_parent = rdp->h.md_parent;
2969 mdp->h.md_pi = rdp->h.md_pi - 1;
2972 rc = mdb_add_node(txn, dbi, rdp->h.md_parent, rdp->h.md_pi,
2973 &sepkey, NULL, rdp->p.mp_pgno, 0);
2975 if (rc != MDB_SUCCESS) {
2980 for (i = j = 0; i <= NUMKEYS(copy); j++) {
2981 if (i < split_indx) {
2982 /* Re-insert in left sibling. */
2985 /* Insert in right sibling. */
2986 if (i == split_indx)
2987 /* Reset insert index for right sibling. */
2988 j = (i == newindx && ins_new);
2992 if (i == newindx && !ins_new) {
2993 /* Insert the original entry that caused the split. */
2994 rkey.mv_data = newkey->mv_data;
2995 rkey.mv_size = newkey->mv_size;
2996 if (IS_LEAF(&mdp->p)) {
2997 rdata.mv_data = newdata->mv_data;
2998 rdata.mv_size = newdata->mv_size;
3005 /* Update page and index for the new key. */
3008 } else if (i == NUMKEYS(copy)) {
3011 node = NODEPTR(copy, i);
3012 rkey.mv_data = NODEKEY(node);
3013 rkey.mv_size = node->mn_ksize;
3014 if (IS_LEAF(&mdp->p)) {
3015 rdata.mv_data = NODEDATA(node);
3016 rdata.mv_size = node->mn_dsize;
3018 pgno = node->mn_pgno;
3019 flags = node->mn_flags;
3024 if (!IS_LEAF(&mdp->p) && j == 0) {
3025 /* First branch index doesn't need key data. */
3029 rc = mdb_add_node(txn, dbi, &pdp->p, j, &rkey, &rdata, pgno,flags);
3037 mdb_put0(MDB_txn *txn, MDB_dbi dbi,
3038 MDB_val *key, MDB_val *data, unsigned int flags)
3040 int rc = MDB_SUCCESS, exact;
3044 MDB_val xdata, *rdata, dkey;
3046 char dbuf[PAGESIZE];
3049 DPRINTF("==> put key %.*s, size %zu, data size %zu",
3050 (int)key->mv_size, (char *)key->mv_data, key->mv_size, data->mv_size);
3053 mpp.mp_parent = NULL;
3055 rc = mdb_search_page(txn, dbi, key, NULL, 1, &mpp);
3056 if (rc == MDB_SUCCESS) {
3057 leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, &ki);
3058 if (leaf && exact) {
3059 if (flags == MDB_NOOVERWRITE) {
3060 DPRINTF("duplicate key %.*s",
3061 (int)key->mv_size, (char *)key->mv_data);
3062 return MDB_KEYEXIST;
3064 if (F_ISSET(txn->mt_dbs[dbi].md_flags, MDB_DUPSORT)) {
3065 /* Was a single item before, must convert now */
3066 if (!F_ISSET(leaf->mn_flags, F_DUPDATA)) {
3067 dkey.mv_size = NODEDSZ(leaf);
3068 memcpy(dbuf, NODEDATA(leaf), dkey.mv_size);
3069 memset(&dummy, 0, sizeof(dummy));
3070 dummy.md_root = P_INVALID;
3071 if (dkey.mv_size == sizeof(MDB_db)) {
3072 memcpy(NODEDATA(leaf), &dummy, sizeof(dummy));
3075 mdb_del_node(mpp.mp_page, ki);
3078 xdata.mv_size = sizeof(MDB_db);
3079 xdata.mv_data = &dummy;
3084 /* same size, just replace it */
3085 if (NODEDSZ(leaf) == data->mv_size) {
3086 memcpy(NODEDATA(leaf), data->mv_data, data->mv_size);
3089 mdb_del_node(mpp.mp_page, ki);
3091 if (leaf == NULL) { /* append if not found */
3092 ki = NUMKEYS(mpp.mp_page);
3093 DPRINTF("appending key at index %i", ki);
3095 } else if (rc == MDB_NOTFOUND) {
3097 /* new file, just write a root leaf page */
3098 DPRINTF("allocating new root leaf page");
3099 if ((dp = mdb_new_page(txn, dbi, P_LEAF, 1)) == NULL) {
3102 mpp.mp_page = &dp->p;
3103 txn->mt_dbs[dbi].md_root = mpp.mp_page->mp_pgno;
3104 txn->mt_dbs[dbi].md_depth++;
3105 txn->mt_dbxs[dbi].md_dirty = 1;
3111 assert(IS_LEAF(mpp.mp_page));
3112 DPRINTF("there are %u keys, should insert new key at index %i",
3113 NUMKEYS(mpp.mp_page), ki);
3118 if (SIZELEFT(mpp.mp_page) < mdb_leaf_size(txn->mt_env, key, rdata)) {
3119 rc = mdb_split(txn, dbi, &mpp.mp_page, &ki, key, rdata, P_INVALID);
3121 /* There is room already in this leaf page. */
3122 rc = mdb_add_node(txn, dbi, mpp.mp_page, ki, key, rdata, 0, 0);
3125 if (rc != MDB_SUCCESS)
3126 txn->mt_flags |= MDB_TXN_ERROR;
3128 txn->mt_dbs[dbi].md_entries++;
3130 /* Remember if we just added a subdatabase */
3131 if (flags & F_SUBDATA) {
3132 leaf = NODEPTR(mpp.mp_page, ki);
3133 leaf->mn_flags |= F_SUBDATA;
3136 /* Now store the actual data in the child DB. Note that we're
3137 * storing the user data in the keys field, so there are strict
3138 * size limits on dupdata. The actual data fields of the child
3139 * DB are all zero size.
3144 leaf = NODEPTR(mpp.mp_page, ki);
3146 mdb_xcursor_init0(txn, dbi, &mx);
3147 mdb_xcursor_init1(txn, dbi, &mx, leaf);
3150 if (flags == MDB_NODUPDATA)
3151 flags = MDB_NOOVERWRITE;
3152 /* converted, write the original data first */
3154 dkey.mv_data = dbuf;
3155 rc = mdb_put0(&mx.mx_txn, mx.mx_cursor.mc_dbi, &dkey, &xdata, flags);
3157 leaf->mn_flags |= F_DUPDATA;
3159 rc = mdb_put0(&mx.mx_txn, mx.mx_cursor.mc_dbi, data, &xdata, flags);
3160 mdb_xcursor_fini(txn, dbi, &mx);
3161 memcpy(NODEDATA(leaf), &mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi],
3171 mdb_put(MDB_txn *txn, MDB_dbi dbi,
3172 MDB_val *key, MDB_val *data, unsigned int flags)
3174 assert(key != NULL);
3175 assert(data != NULL);
3177 if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
3180 if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
3184 if (key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
3188 if ((flags & (MDB_NOOVERWRITE|MDB_NODUPDATA)) != flags)
3191 return mdb_put0(txn, dbi, key, data, flags);
3195 mdb_env_set_flags(MDB_env *env, unsigned int flag, int onoff)
3197 #define CHANGEABLE (MDB_NOSYNC)
3198 if ((flag & CHANGEABLE) != flag)
3201 env->me_flags |= flag;
3203 env->me_flags &= ~flag;
3208 mdb_env_get_flags(MDB_env *env, unsigned int *arg)
3213 *arg = env->me_flags;
3218 mdb_env_get_path(MDB_env *env, const char **arg)
3223 *arg = env->me_path;
3228 mdb_stat0(MDB_env *env, MDB_db *db, MDB_stat *arg)
3230 arg->ms_psize = env->me_psize;
3231 arg->ms_depth = db->md_depth;
3232 arg->ms_branch_pages = db->md_branch_pages;
3233 arg->ms_leaf_pages = db->md_leaf_pages;
3234 arg->ms_overflow_pages = db->md_overflow_pages;
3235 arg->ms_entries = db->md_entries;
3240 mdb_env_stat(MDB_env *env, MDB_stat *arg)
3242 if (env == NULL || arg == NULL)
3245 return mdb_stat0(env, &env->me_meta->mm_dbs[MAIN_DBI], arg);
3248 int mdb_open(MDB_txn *txn, const char *name, unsigned int flags, MDB_dbi *dbi)
3258 if (flags & (MDB_DUPSORT|MDB_REVERSEKEY|MDB_INTEGERKEY))
3259 txn->mt_dbs[MAIN_DBI].md_flags |= (flags & (MDB_DUPSORT|MDB_REVERSEKEY|MDB_INTEGERKEY));
3263 /* Is the DB already open? */
3265 for (i=2; i<txn->mt_numdbs; i++) {
3266 if (len == txn->mt_dbxs[i].md_name.mv_size &&
3267 !strncmp(name, txn->mt_dbxs[i].md_name.mv_data, len)) {
3273 if (txn->mt_numdbs >= txn->mt_env->me_maxdbs - 1)
3276 /* Find the DB info */
3278 key.mv_data = (void *)name;
3279 rc = mdb_get(txn, MAIN_DBI, &key, &data);
3281 /* Create if requested */
3282 if (rc == MDB_NOTFOUND && (flags & MDB_CREATE)) {
3284 data.mv_size = sizeof(MDB_db);
3285 data.mv_data = &dummy;
3286 memset(&dummy, 0, sizeof(dummy));
3287 dummy.md_root = P_INVALID;
3288 dummy.md_flags = flags & 0xffff;
3289 rc = mdb_put0(txn, MAIN_DBI, &key, &data, F_SUBDATA);
3293 /* OK, got info, add to table */
3294 if (rc == MDB_SUCCESS) {
3295 txn->mt_dbxs[txn->mt_numdbs].md_name.mv_data = strdup(name);
3296 txn->mt_dbxs[txn->mt_numdbs].md_name.mv_size = len;
3297 txn->mt_dbxs[txn->mt_numdbs].md_cmp = NULL;
3298 txn->mt_dbxs[txn->mt_numdbs].md_dcmp = NULL;
3299 txn->mt_dbxs[txn->mt_numdbs].md_rel = NULL;
3300 txn->mt_dbxs[txn->mt_numdbs].md_parent = MAIN_DBI;
3301 txn->mt_dbxs[txn->mt_numdbs].md_dirty = dirty;
3302 memcpy(&txn->mt_dbs[txn->mt_numdbs], data.mv_data, sizeof(MDB_db));
3303 *dbi = txn->mt_numdbs;
3304 txn->mt_env->me_dbs[0][txn->mt_numdbs] = txn->mt_dbs[txn->mt_numdbs];
3305 txn->mt_env->me_dbs[1][txn->mt_numdbs] = txn->mt_dbs[txn->mt_numdbs];
3312 int mdb_stat(MDB_txn *txn, MDB_dbi dbi, MDB_stat *arg)
3314 if (txn == NULL || arg == NULL || dbi >= txn->mt_numdbs)
3317 return mdb_stat0(txn->mt_env, &txn->mt_dbs[dbi], arg);
3320 void mdb_close(MDB_txn *txn, MDB_dbi dbi)
3323 if (dbi <= MAIN_DBI || dbi >= txn->mt_numdbs)
3325 ptr = txn->mt_dbxs[dbi].md_name.mv_data;
3326 txn->mt_dbxs[dbi].md_name.mv_data = NULL;
3327 txn->mt_dbxs[dbi].md_name.mv_size = 0;
3331 int mdb_set_compare(MDB_txn *txn, MDB_dbi dbi, MDB_cmp_func *cmp)
3333 if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
3336 txn->mt_dbxs[dbi].md_cmp = cmp;
3340 int mdb_set_dupsort(MDB_txn *txn, MDB_dbi dbi, MDB_cmp_func *cmp)
3342 if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
3345 txn->mt_dbxs[dbi].md_dcmp = cmp;
3349 int mdb_set_relfunc(MDB_txn *txn, MDB_dbi dbi, MDB_rel_func *rel)
3351 if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
3354 txn->mt_dbxs[dbi].md_rel = rel;