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
54 #define ULONG unsigned long
63 #if DEBUG && defined(__GNUC__)
64 # define DPRINTF(fmt, ...) \
65 fprintf(stderr, "%s:%d: " fmt "\n", __func__, __LINE__, ##__VA_ARGS__)
67 # define DPRINTF(...) ((void) 0)
72 #define MDB_MAGIC 0xBEEFC0DE
74 #define MAXKEYSIZE 511
76 #define P_INVALID (~0UL)
78 #define F_ISSET(w, f) (((w) & (f)) == (f))
80 typedef uint16_t indx_t;
82 #define DEFAULT_READERS 126
83 #define DEFAULT_MAPSIZE 1048576
85 /* Lock descriptor stuff */
87 #define CACHELINE 64 /* most CPUs. Itanium uses 128 */
90 typedef struct MDB_rxbody {
96 typedef struct MDB_reader {
99 #define mr_txnid mru.mrx.mrb_txnid
100 #define mr_pid mru.mrx.mrb_pid
101 #define mr_tid mru.mrx.mrb_tid
102 /* cache line alignment */
103 char pad[(sizeof(MDB_rxbody)+CACHELINE-1) & ~(CACHELINE-1)];
107 typedef struct MDB_txbody {
109 uint32_t mtb_version;
110 pthread_mutex_t mtb_mutex;
112 uint32_t mtb_numreaders;
113 uint32_t mtb_me_toggle;
116 typedef struct MDB_txninfo {
119 #define mti_magic mt1.mtb.mtb_magic
120 #define mti_version mt1.mtb.mtb_version
121 #define mti_mutex mt1.mtb.mtb_mutex
122 #define mti_txnid mt1.mtb.mtb_txnid
123 #define mti_numreaders mt1.mtb.mtb_numreaders
124 #define mti_me_toggle mt1.mtb.mtb_me_toggle
125 char pad[(sizeof(MDB_txbody)+CACHELINE-1) & ~(CACHELINE-1)];
128 pthread_mutex_t mt2_wmutex;
129 #define mti_wmutex mt2.mt2_wmutex
130 char pad[(sizeof(pthread_mutex_t)+CACHELINE-1) & ~(CACHELINE-1)];
132 MDB_reader mti_readers[1];
135 /* Common header for all page types. Overflow pages
136 * occupy a number of contiguous pages with no
137 * headers on any page after the first.
139 typedef struct MDB_page { /* represents a page of storage */
140 #define mp_pgno mp_p.p_pgno
142 pgno_t p_pgno; /* page number */
145 #define P_BRANCH 0x01 /* branch page */
146 #define P_LEAF 0x02 /* leaf page */
147 #define P_OVERFLOW 0x04 /* overflow page */
148 #define P_META 0x08 /* meta page */
149 #define P_DIRTY 0x10 /* dirty page */
150 #define P_LEAF2 0x20 /* DB with small, fixed size keys and no data */
152 #define mp_lower mp_pb.pb.pb_lower
153 #define mp_upper mp_pb.pb.pb_upper
154 #define mp_pages mp_pb.pb_pages
157 indx_t pb_lower; /* lower bound of free space */
158 indx_t pb_upper; /* upper bound of free space */
160 uint32_t pb_pages; /* number of overflow pages */
162 indx_t pb_ksize; /* on a LEAF2 page */
166 indx_t mp_ptrs[1]; /* dynamic size */
169 #define PAGEHDRSZ ((unsigned) offsetof(MDB_page, mp_ptrs))
171 #define NUMKEYS(p) (((p)->mp_lower - PAGEHDRSZ) >> 1)
172 #define SIZELEFT(p) (indx_t)((p)->mp_upper - (p)->mp_lower)
173 #define PAGEFILL(env, p) (1000L * ((env)->me_psize - PAGEHDRSZ - SIZELEFT(p)) / \
174 ((env)->me_psize - PAGEHDRSZ))
175 #define IS_LEAF(p) F_ISSET((p)->mp_flags, P_LEAF)
176 #define IS_BRANCH(p) F_ISSET((p)->mp_flags, P_BRANCH)
177 #define IS_OVERFLOW(p) F_ISSET((p)->mp_flags, P_OVERFLOW)
179 #define OVPAGES(size, psize) (PAGEHDRSZ + size + psize - 1) / psize;
181 typedef struct MDB_db {
185 ULONG md_branch_pages;
187 ULONG md_overflow_pages;
195 typedef struct MDB_meta { /* meta (footer) page content */
198 void *mm_address; /* address for fixed mapping */
199 size_t mm_mapsize; /* size of mmap region */
200 MDB_db mm_dbs[2]; /* first is free space, 2nd is main db */
201 #define mm_psize mm_dbs[0].md_pad
202 #define mm_flags mm_dbs[0].md_flags
203 pgno_t mm_last_pg; /* last used page in file */
204 ULONG mm_txnid; /* txnid that committed this page */
207 typedef struct MDB_dhead { /* a dirty page */
209 unsigned md_pi; /* parent index */
213 typedef struct MDB_dpage {
218 typedef struct MDB_oldpages {
219 struct MDB_oldpages *mo_next;
221 pgno_t mo_pages[1]; /* dynamic */
224 typedef struct MDB_pageparent {
230 static MDB_dpage *mdb_alloc_page(MDB_txn *txn, MDB_page *parent, unsigned int parent_idx, int num);
231 static int mdb_touch(MDB_txn *txn, MDB_pageparent *mp);
233 typedef struct MDB_ppage { /* ordered list of pages */
234 SLIST_ENTRY(MDB_ppage) mp_entry;
236 unsigned int mp_ki; /* cursor index on page */
238 SLIST_HEAD(page_stack, MDB_ppage);
240 /* FIXME: tree depth is mostly bounded, we should just
241 * use a fixed array and avoid malloc/pointer chasing
243 #define CURSOR_EMPTY(c) SLIST_EMPTY(&(c)->mc_stack)
244 #define CURSOR_TOP(c) SLIST_FIRST(&(c)->mc_stack)
245 #define CURSOR_POP(c) SLIST_REMOVE_HEAD(&(c)->mc_stack, mp_entry)
246 #define CURSOR_PUSH(c,p) SLIST_INSERT_HEAD(&(c)->mc_stack, p, mp_entry)
252 struct page_stack mc_stack; /* stack of parent pages */
254 short mc_initialized; /* 1 if initialized */
255 short mc_eof; /* 1 if end is reached */
256 struct MDB_xcursor *mc_xcursor;
259 #define METADATA(p) ((void *)((char *)p + PAGEHDRSZ))
261 typedef struct MDB_node {
262 #define mn_pgno mn_p.np_pgno
263 #define mn_dsize mn_p.np_dsize
265 pgno_t np_pgno; /* child page number */
266 uint32_t np_dsize; /* leaf data size */
268 unsigned int mn_flags:4;
269 unsigned int mn_ksize:12; /* key size */
270 #define F_BIGDATA 0x01 /* data put on overflow page */
271 #define F_SUBDATA 0x02 /* data is a sub-database */
275 typedef struct MDB_dbx {
277 MDB_cmp_func *md_cmp; /* user compare function */
278 MDB_cmp_func *md_dcmp; /* user dupsort function */
279 MDB_rel_func *md_rel; /* user relocate function */
281 unsigned int md_dirty;
285 pgno_t mt_next_pgno; /* next unallocated page */
289 pgno_t *mt_free_pgs; /* this is an IDL */
291 MIDL2 *dirty_list; /* modified pages */
294 MDB_dbx *mt_dbxs; /* array */
296 unsigned int mt_numdbs;
298 #define MDB_TXN_RDONLY 0x01 /* read-only transaction */
299 #define MDB_TXN_ERROR 0x02 /* an error has occurred */
300 #define MDB_TXN_METOGGLE 0x04 /* used meta page 1 */
301 unsigned int mt_flags;
304 /* Context for sorted-dup records */
305 typedef struct MDB_xcursor {
306 MDB_cursor mx_cursor;
315 int me_mfd; /* just for writing the meta pages */
316 #define MDB_FATAL_ERROR 0x80000000U
318 uint32_t me_extrapad; /* unused for now */
319 unsigned int me_maxreaders;
320 unsigned int me_numdbs;
321 unsigned int me_maxdbs;
324 MDB_txninfo *me_txns;
325 MDB_meta *me_metas[2];
327 MDB_txn *me_txn; /* current write transaction */
329 off_t me_size; /* current file size */
330 pgno_t me_maxpg; /* me_mapsize / me_psize */
331 unsigned int me_psize;
332 unsigned int me_db_toggle;
333 MDB_dbx *me_dbxs; /* array */
335 MDB_oldpages *me_pghead;
336 pthread_key_t me_txkey; /* thread-key for readers */
337 pgno_t me_free_pgs[MDB_IDL_UM_SIZE];
338 MIDL2 me_dirty_list[MDB_IDL_DB_SIZE];
341 #define NODESIZE offsetof(MDB_node, mn_data)
343 #define INDXSIZE(k) (NODESIZE + ((k) == NULL ? 0 : (k)->mv_size))
344 #define LEAFSIZE(k, d) (NODESIZE + (k)->mv_size + (d)->mv_size)
345 #define NODEPTR(p, i) ((MDB_node *)((char *)(p) + (p)->mp_ptrs[i]))
346 #define NODEKEY(node) (void *)((node)->mn_data)
347 #define NODEDATA(node) (void *)((char *)(node)->mn_data + (node)->mn_ksize)
348 #define NODEPGNO(node) ((node)->mn_pgno)
349 #define NODEDSZ(node) ((node)->mn_dsize)
351 #define MDB_COMMIT_PAGES 64 /* max number of pages to write in one commit */
353 static int mdb_search_page_root(MDB_txn *txn,
354 MDB_dbi dbi, MDB_val *key,
355 MDB_cursor *cursor, int modify,
356 MDB_pageparent *mpp);
357 static int mdb_search_page(MDB_txn *txn,
358 MDB_dbi dbi, MDB_val *key,
359 MDB_cursor *cursor, int modify,
360 MDB_pageparent *mpp);
362 static int mdb_env_read_header(MDB_env *env, MDB_meta *meta);
363 static int mdb_env_read_meta(MDB_env *env, int *which);
364 static int mdb_env_write_meta(MDB_txn *txn);
365 static MDB_page *mdb_get_page(MDB_txn *txn, pgno_t pgno);
367 static MDB_node *mdb_search_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp,
368 MDB_val *key, int *exactp, unsigned int *kip);
369 static int mdb_add_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp,
370 indx_t indx, MDB_val *key, MDB_val *data,
371 pgno_t pgno, uint8_t flags);
372 static void mdb_del_node(MDB_page *mp, indx_t indx);
373 static int mdb_del0(MDB_txn *txn, MDB_dbi dbi, unsigned int ki,
374 MDB_pageparent *mpp, MDB_node *leaf);
375 static int mdb_put0(MDB_txn *txn, MDB_dbi dbi,
376 MDB_val *key, MDB_val *data, unsigned int flags);
377 static int mdb_read_data(MDB_txn *txn, MDB_node *leaf, MDB_val *data);
379 static int mdb_rebalance(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *mp);
380 static int mdb_update_key(MDB_page *mp, indx_t indx, MDB_val *key);
381 static int mdb_move_node(MDB_txn *txn, MDB_dbi dbi,
382 MDB_pageparent *src, indx_t srcindx,
383 MDB_pageparent *dst, indx_t dstindx);
384 static int mdb_merge(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *src,
385 MDB_pageparent *dst);
386 static int mdb_split(MDB_txn *txn, MDB_dbi dbi, MDB_page **mpp,
387 unsigned int *newindxp, MDB_val *newkey,
388 MDB_val *newdata, pgno_t newpgno);
389 static MDB_dpage *mdb_new_page(MDB_txn *txn, MDB_dbi dbi, uint32_t flags, int num);
391 static void cursor_pop_page(MDB_cursor *cursor);
392 static MDB_ppage *cursor_push_page(MDB_cursor *cursor,
395 static int mdb_set_key(MDB_node *node, MDB_val *key);
396 static int mdb_sibling(MDB_cursor *cursor, int move_right);
397 static int mdb_cursor_next(MDB_cursor *cursor,
398 MDB_val *key, MDB_val *data, MDB_cursor_op op);
399 static int mdb_cursor_prev(MDB_cursor *cursor,
400 MDB_val *key, MDB_val *data, MDB_cursor_op op);
401 static int mdb_cursor_set(MDB_cursor *cursor,
402 MDB_val *key, MDB_val *data, MDB_cursor_op op, int *exactp);
403 static int mdb_cursor_first(MDB_cursor *cursor,
404 MDB_val *key, MDB_val *data);
405 static int mdb_cursor_last(MDB_cursor *cursor,
406 MDB_val *key, MDB_val *data);
408 static void mdb_xcursor_init0(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx);
409 static void mdb_xcursor_init1(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx, MDB_node *node);
410 static void mdb_xcursor_fini(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx);
412 static size_t mdb_leaf_size(MDB_env *env, MDB_val *key,
414 static size_t mdb_branch_size(MDB_env *env, MDB_val *key);
416 static int memncmp(const void *s1, size_t n1,
417 const void *s2, size_t n2);
418 static int memnrcmp(const void *s1, size_t n1,
419 const void *s2, size_t n2);
422 memncmp(const void *s1, size_t n1, const void *s2, size_t n2)
424 int diff, len_diff = -1;
427 len_diff = (n1 > n2);
430 diff = memcmp(s1, s2, n1);
431 return diff ? diff : len_diff;
435 memnrcmp(const void *s1, size_t n1, const void *s2, size_t n2)
437 const unsigned char *p1, *p2, *p1_lim;
444 p1 = (const unsigned char *)s1 + n1 - 1;
445 p2 = (const unsigned char *)s2 + n2 - 1;
447 for (p1_lim = (n1 <= n2 ? s1 : s2); *p1 == *p2; p1--, p2--) {
449 return (p1 != s1) ? (p1 != p2) : (p2 != s2) ? -1 : 0;
455 mdb_version(int *maj, int *min, int *pat)
457 *maj = MDB_VERSION_MAJOR;
458 *min = MDB_VERSION_MINOR;
459 *pat = MDB_VERSION_PATCH;
460 return MDB_VERSION_STRING;
464 mdb_cmp(MDB_txn *txn, MDB_dbi dbi, const MDB_val *a, const MDB_val *b)
466 return txn->mt_dbxs[dbi].md_cmp(a, b);
470 _mdb_cmp(MDB_txn *txn, MDB_dbi dbi, const MDB_val *key1, const MDB_val *key2)
472 if (txn->mt_dbs[dbi].md_flags & (MDB_REVERSEKEY
473 #if __BYTE_ORDER == __LITTLE_ENDIAN
477 return memnrcmp(key1->mv_data, key1->mv_size, key2->mv_data, key2->mv_size);
479 return memncmp((char *)key1->mv_data, key1->mv_size, key2->mv_data, key2->mv_size);
482 /* Allocate new page(s) for writing */
484 mdb_alloc_page(MDB_txn *txn, MDB_page *parent, unsigned int parent_idx, int num)
487 pgno_t pgno = P_INVALID;
491 if (txn->mt_txnid > 2) {
493 oldest = txn->mt_txnid - 2;
494 if (!txn->mt_env->me_pghead && txn->mt_dbs[FREE_DBI].md_root != P_INVALID) {
495 /* See if there's anything in the free DB */
500 mpp.mp_parent = NULL;
502 mdb_search_page(txn, FREE_DBI, NULL, NULL, 0, &mpp);
503 leaf = NODEPTR(mpp.mp_page, 0);
504 kptr = (ULONG *)NODEKEY(leaf);
506 /* It's potentially usable, unless there are still
507 * older readers outstanding. Grab it.
509 if (oldest > *kptr) {
514 mdb_read_data(txn, leaf, &data);
515 idl = (ULONG *)data.mv_data;
516 mop = malloc(sizeof(MDB_oldpages) + MDB_IDL_SIZEOF(idl) - sizeof(pgno_t));
517 mop->mo_next = txn->mt_env->me_pghead;
518 mop->mo_txnid = *kptr;
519 txn->mt_env->me_pghead = mop;
520 memcpy(mop->mo_pages, idl, MDB_IDL_SIZEOF(idl));
525 DPRINTF("IDL read txn %lu root %lu num %lu",
526 mop->mo_txnid, txn->mt_dbs[FREE_DBI].md_root, idl[0]);
527 for (i=0; i<idl[0]; i++) {
528 DPRINTF("IDL %lu", idl[i+1]);
532 /* drop this IDL from the DB */
533 mpp.mp_parent = NULL;
535 mdb_search_page(txn, FREE_DBI, NULL, NULL, 1, &mpp);
536 leaf = NODEPTR(mpp.mp_page, 0);
537 mdb_del0(txn, FREE_DBI, 0, &mpp, leaf);
540 if (txn->mt_env->me_pghead) {
542 for (i=0; i<txn->mt_env->me_txns->mti_numreaders; i++) {
543 ULONG mr = txn->mt_env->me_txns->mti_readers[i].mr_txnid;
546 oldest = txn->mt_env->me_txns->mti_readers[i].mr_txnid;
548 if (oldest > txn->mt_env->me_pghead->mo_txnid) {
549 MDB_oldpages *mop = txn->mt_env->me_pghead;
550 txn->mt_oldest = oldest;
552 /* FIXME: For now, always use fresh pages. We
553 * really ought to search the free list for a
558 /* peel pages off tail, so we only have to truncate the list */
559 pgno = MDB_IDL_LAST(mop->mo_pages);
560 if (MDB_IDL_IS_RANGE(mop->mo_pages)) {
562 if (mop->mo_pages[2] > mop->mo_pages[1])
563 mop->mo_pages[0] = 0;
567 if (MDB_IDL_IS_ZERO(mop->mo_pages)) {
568 txn->mt_env->me_pghead = mop->mo_next;
576 if (pgno == P_INVALID) {
577 /* DB size is maxed out */
578 if (txn->mt_next_pgno + num >= txn->mt_env->me_maxpg)
581 if ((dp = malloc(txn->mt_env->me_psize * num + sizeof(MDB_dhead))) == NULL)
584 dp->h.md_parent = parent;
585 dp->h.md_pi = parent_idx;
586 if (pgno == P_INVALID) {
587 dp->p.mp_pgno = txn->mt_next_pgno;
588 txn->mt_next_pgno += num;
590 dp->p.mp_pgno = pgno;
592 mid.mid = dp->p.mp_pgno;
594 mdb_midl2_insert(txn->mt_u.dirty_list, &mid);
599 /* Touch a page: make it dirty and re-insert into tree with updated pgno.
602 mdb_touch(MDB_txn *txn, MDB_pageparent *pp)
604 MDB_page *mp = pp->mp_page;
609 if (!F_ISSET(mp->mp_flags, P_DIRTY)) {
611 if ((dp = mdb_alloc_page(txn, pp->mp_parent, pp->mp_pi, 1)) == NULL)
613 DPRINTF("touched page %lu -> %lu", mp->mp_pgno, dp->p.mp_pgno);
614 mdb_midl_insert(txn->mt_free_pgs, mp->mp_pgno);
615 pgno = dp->p.mp_pgno;
616 memcpy(&dp->p, mp, txn->mt_env->me_psize);
619 mp->mp_flags |= P_DIRTY;
621 /* Update the page number to new touched page. */
622 if (pp->mp_parent != NULL)
623 NODEPGNO(NODEPTR(pp->mp_parent, pp->mp_pi)) = mp->mp_pgno;
630 mdb_env_sync(MDB_env *env, int force)
633 if (force || !F_ISSET(env->me_flags, MDB_NOSYNC)) {
634 if (fdatasync(env->me_fd))
641 mdb_txn_begin(MDB_env *env, int rdonly, MDB_txn **ret)
646 if (env->me_flags & MDB_FATAL_ERROR) {
647 DPRINTF("mdb_txn_begin: environment had fatal error, must shutdown!");
650 if ((txn = calloc(1, sizeof(MDB_txn))) == NULL) {
651 DPRINTF("calloc: %s", strerror(errno));
656 txn->mt_flags |= MDB_TXN_RDONLY;
658 txn->mt_u.dirty_list = env->me_dirty_list;
659 txn->mt_u.dirty_list[0].mid = 0;
660 txn->mt_free_pgs = env->me_free_pgs;
661 txn->mt_free_pgs[0] = 0;
663 pthread_mutex_lock(&env->me_txns->mti_wmutex);
664 env->me_txns->mti_txnid++;
667 txn->mt_txnid = env->me_txns->mti_txnid;
669 MDB_reader *r = pthread_getspecific(env->me_txkey);
672 pthread_mutex_lock(&env->me_txns->mti_mutex);
673 for (i=0; i<env->me_txns->mti_numreaders; i++)
674 if (env->me_txns->mti_readers[i].mr_pid == 0)
676 if (i == env->me_maxreaders) {
677 pthread_mutex_unlock(&env->me_txns->mti_mutex);
680 env->me_txns->mti_readers[i].mr_pid = getpid();
681 env->me_txns->mti_readers[i].mr_tid = pthread_self();
682 r = &env->me_txns->mti_readers[i];
683 pthread_setspecific(env->me_txkey, r);
684 if (i >= env->me_txns->mti_numreaders)
685 env->me_txns->mti_numreaders = i+1;
686 pthread_mutex_unlock(&env->me_txns->mti_mutex);
688 r->mr_txnid = txn->mt_txnid;
689 txn->mt_u.reader = r;
696 toggle = env->me_txns->mti_me_toggle;
697 if ((rc = mdb_env_read_meta(env, &toggle)) != MDB_SUCCESS) {
702 /* Copy the DB arrays */
703 txn->mt_numdbs = env->me_numdbs;
704 txn->mt_dbxs = env->me_dbxs; /* mostly static anyway */
705 txn->mt_dbs = malloc(env->me_maxdbs * sizeof(MDB_db));
706 memcpy(txn->mt_dbs, env->me_meta->mm_dbs, 2 * sizeof(MDB_db));
707 if (txn->mt_numdbs > 2)
708 memcpy(txn->mt_dbs+2, env->me_dbs[env->me_db_toggle]+2,
709 (txn->mt_numdbs - 2) * sizeof(MDB_db));
713 txn->mt_flags |= MDB_TXN_METOGGLE;
714 txn->mt_next_pgno = env->me_meta->mm_last_pg+1;
717 DPRINTF("begin transaction %lu on mdbenv %p, root page %lu",
718 txn->mt_txnid, (void *) env, txn->mt_dbs[MAIN_DBI].md_root);
725 mdb_txn_abort(MDB_txn *txn)
733 DPRINTF("abort transaction %lu on mdbenv %p, root page %lu",
734 txn->mt_txnid, (void *) env, txn->mt_dbs[MAIN_DBI].md_root);
738 if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
739 txn->mt_u.reader->mr_txnid = 0;
744 /* Discard all dirty pages. */
745 for (i=1; i<=txn->mt_u.dirty_list[0].mid; i++)
746 free(txn->mt_u.dirty_list[i].mptr);
748 while ((mop = txn->mt_env->me_pghead)) {
749 txn->mt_env->me_pghead = mop->mo_next;
754 env->me_txns->mti_txnid--;
755 for (i=2; i<env->me_numdbs; i++)
756 env->me_dbxs[i].md_dirty = 0;
757 pthread_mutex_unlock(&env->me_txns->mti_wmutex);
764 mdb_txn_commit(MDB_txn *txn)
773 struct iovec iov[MDB_COMMIT_PAGES];
776 assert(txn->mt_env != NULL);
780 if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
785 if (txn != env->me_txn) {
786 DPRINTF("attempt to commit unknown transaction");
791 if (F_ISSET(txn->mt_flags, MDB_TXN_ERROR)) {
792 DPRINTF("error flag is set, can't commit");
797 if (!txn->mt_u.dirty_list[0].mid)
800 DPRINTF("committing transaction %lu on mdbenv %p, root page %lu",
801 txn->mt_txnid, (void *) env, txn->mt_dbs[MAIN_DBI].md_root);
803 /* should only be one record now */
804 if (env->me_pghead) {
808 mop = env->me_pghead;
809 key.mv_size = sizeof(pgno_t);
810 key.mv_data = (char *)&mop->mo_txnid;
811 data.mv_size = MDB_IDL_SIZEOF(mop->mo_pages);
812 data.mv_data = mop->mo_pages;
813 mdb_put0(txn, FREE_DBI, &key, &data, 0);
814 free(env->me_pghead);
815 env->me_pghead = NULL;
817 /* save to free list */
818 if (!MDB_IDL_IS_ZERO(txn->mt_free_pgs)) {
822 /* make sure last page of freeDB is touched and on freelist */
823 key.mv_size = MAXKEYSIZE+1;
825 mpp.mp_parent = NULL;
827 mdb_search_page(txn, FREE_DBI, &key, NULL, 1, &mpp);
832 ULONG *idl = txn->mt_free_pgs;
833 DPRINTF("IDL write txn %lu root %lu num %lu",
834 txn->mt_txnid, txn->mt_dbs[FREE_DBI].md_root, idl[0]);
835 for (i=0; i<idl[0]; i++) {
836 DPRINTF("IDL %lu", idl[i+1]);
840 /* write to last page of freeDB */
841 key.mv_size = sizeof(pgno_t);
842 key.mv_data = (char *)&txn->mt_txnid;
843 data.mv_size = MDB_IDL_SIZEOF(txn->mt_free_pgs);
844 data.mv_data = txn->mt_free_pgs;
845 mdb_put0(txn, FREE_DBI, &key, &data, 0);
848 /* Update DB root pointers. Their pages have already been
849 * touched so this is all in-place and cannot fail.
853 data.mv_size = sizeof(MDB_db);
855 for (i = 2; i < txn->mt_numdbs; i++) {
856 if (txn->mt_dbxs[i].md_dirty) {
857 data.mv_data = &txn->mt_dbs[i];
858 mdb_put0(txn, MAIN_DBI, &txn->mt_dbxs[i].md_name, &data, 0);
863 /* Commit up to MDB_COMMIT_PAGES dirty pages to disk until done.
870 for (i=1; i<=txn->mt_u.dirty_list[0].mid; i++) {
871 dp = txn->mt_u.dirty_list[i].mptr;
872 if (dp->p.mp_pgno != next) {
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 lseek(env->me_fd, dp->p.mp_pgno * env->me_psize, SEEK_SET);
889 next = dp->p.mp_pgno;
891 DPRINTF("committing page %lu", dp->p.mp_pgno);
892 iov[n].iov_len = env->me_psize * dp->h.md_num;
893 iov[n].iov_base = &dp->p;
894 size += iov[n].iov_len;
895 next = dp->p.mp_pgno + dp->h.md_num;
896 /* clear dirty flag */
897 dp->p.mp_flags &= ~P_DIRTY;
898 if (++n >= MDB_COMMIT_PAGES) {
907 DPRINTF("committing %u dirty pages", n);
908 rc = writev(env->me_fd, iov, n);
912 DPRINTF("short write, filesystem full?");
914 DPRINTF("writev: %s", strerror(errno));
921 /* Drop the dirty pages.
923 for (i=1; i<=txn->mt_u.dirty_list[0].mid; i++)
924 free(txn->mt_u.dirty_list[i].mptr);
926 if ((n = mdb_env_sync(env, 0)) != 0 ||
927 (n = mdb_env_write_meta(txn)) != MDB_SUCCESS) {
933 /* update the DB tables */
935 int toggle = !env->me_db_toggle;
937 for (i = 2; i < env->me_numdbs; i++) {
938 if (txn->mt_dbxs[i].md_dirty) {
939 env->me_dbs[toggle][i] = txn->mt_dbs[i];
940 txn->mt_dbxs[i].md_dirty = 0;
943 for (i = env->me_numdbs; i < txn->mt_numdbs; i++) {
944 txn->mt_dbxs[i].md_dirty = 0;
945 env->me_dbxs[i] = txn->mt_dbxs[i];
946 env->me_dbs[toggle][i] = txn->mt_dbs[i];
948 env->me_db_toggle = toggle;
949 env->me_numdbs = txn->mt_numdbs;
954 pthread_mutex_unlock(&env->me_txns->mti_wmutex);
965 mdb_env_read_header(MDB_env *env, MDB_meta *meta)
974 /* We don't know the page size yet, so use a minimum value.
977 if ((rc = pread(env->me_fd, page, PAGESIZE, 0)) == 0) {
979 } else if (rc != PAGESIZE) {
982 DPRINTF("read: %s", strerror(errno));
986 p = (MDB_page *)page;
988 if (!F_ISSET(p->mp_flags, P_META)) {
989 DPRINTF("page %lu not a meta page", p->mp_pgno);
994 if (m->mm_magic != MDB_MAGIC) {
995 DPRINTF("meta has invalid magic");
999 if (m->mm_version != MDB_VERSION) {
1000 DPRINTF("database is version %u, expected version %u",
1001 m->mm_version, MDB_VERSION);
1002 return MDB_VERSION_MISMATCH;
1005 memcpy(meta, m, sizeof(*m));
1010 mdb_env_init_meta(MDB_env *env, MDB_meta *meta)
1017 DPRINTF("writing new meta page");
1018 psize = sysconf(_SC_PAGE_SIZE);
1020 meta->mm_magic = MDB_MAGIC;
1021 meta->mm_version = MDB_VERSION;
1022 meta->mm_psize = psize;
1023 meta->mm_last_pg = 1;
1024 meta->mm_flags = env->me_flags & 0xffff;
1025 meta->mm_flags |= MDB_INTEGERKEY;
1026 meta->mm_dbs[0].md_root = P_INVALID;
1027 meta->mm_dbs[1].md_root = P_INVALID;
1029 p = calloc(2, psize);
1031 p->mp_flags = P_META;
1034 memcpy(m, meta, sizeof(*meta));
1036 q = (MDB_page *)((char *)p + psize);
1039 q->mp_flags = P_META;
1042 memcpy(m, meta, sizeof(*meta));
1044 rc = write(env->me_fd, p, psize * 2);
1046 return (rc == (int)psize * 2) ? MDB_SUCCESS : errno;
1050 mdb_env_write_meta(MDB_txn *txn)
1053 MDB_meta meta, metab;
1055 int rc, len, toggle;
1058 assert(txn != NULL);
1059 assert(txn->mt_env != NULL);
1061 toggle = !F_ISSET(txn->mt_flags, MDB_TXN_METOGGLE);
1062 DPRINTF("writing meta page %d for root page %lu",
1063 toggle, txn->mt_dbs[MAIN_DBI].md_root);
1067 metab.mm_txnid = env->me_metas[toggle]->mm_txnid;
1068 metab.mm_last_pg = env->me_metas[toggle]->mm_last_pg;
1070 ptr = (char *)&meta;
1071 off = offsetof(MDB_meta, mm_dbs[0].md_depth);
1072 len = sizeof(MDB_meta) - off;
1075 meta.mm_dbs[0] = txn->mt_dbs[0];
1076 meta.mm_dbs[1] = txn->mt_dbs[1];
1077 meta.mm_last_pg = txn->mt_next_pgno - 1;
1078 meta.mm_txnid = txn->mt_txnid;
1081 off += env->me_psize;
1084 /* Write to the SYNC fd */
1085 rc = pwrite(env->me_mfd, ptr, len, off);
1087 DPRINTF("write failed, disk error?");
1088 /* On a failure, the pagecache still contains the new data.
1089 * Write some old data back, to prevent it from being used.
1090 * Use the non-SYNC fd; we know it will fail anyway.
1092 meta.mm_last_pg = metab.mm_last_pg;
1093 meta.mm_txnid = metab.mm_txnid;
1094 rc = pwrite(env->me_fd, ptr, len, off);
1095 env->me_flags |= MDB_FATAL_ERROR;
1098 txn->mt_env->me_txns->mti_me_toggle = toggle;
1104 mdb_env_read_meta(MDB_env *env, int *which)
1108 assert(env != NULL);
1112 else if (env->me_metas[0]->mm_txnid < env->me_metas[1]->mm_txnid)
1115 if (env->me_meta != env->me_metas[toggle])
1116 env->me_meta = env->me_metas[toggle];
1118 DPRINTF("Using meta page %d", toggle);
1124 mdb_env_create(MDB_env **env)
1128 e = calloc(1, sizeof(MDB_env));
1129 if (!e) return ENOMEM;
1131 e->me_maxreaders = DEFAULT_READERS;
1141 mdb_env_set_mapsize(MDB_env *env, size_t size)
1145 env->me_mapsize = size;
1150 mdb_env_set_maxdbs(MDB_env *env, int dbs)
1152 env->me_maxdbs = dbs;
1157 mdb_env_set_maxreaders(MDB_env *env, int readers)
1159 env->me_maxreaders = readers;
1164 mdb_env_get_maxreaders(MDB_env *env, int *readers)
1166 if (!env || !readers)
1168 *readers = env->me_maxreaders;
1173 mdb_env_open2(MDB_env *env, unsigned int flags)
1179 env->me_flags = flags;
1181 memset(&meta, 0, sizeof(meta));
1183 if ((i = mdb_env_read_header(env, &meta)) != 0) {
1186 DPRINTF("new mdbenv");
1190 if (!env->me_mapsize) {
1191 env->me_mapsize = newenv ? DEFAULT_MAPSIZE : meta.mm_mapsize;
1195 if (meta.mm_address && (flags & MDB_FIXEDMAP))
1197 env->me_map = mmap(meta.mm_address, env->me_mapsize, PROT_READ, i,
1199 if (env->me_map == MAP_FAILED)
1203 meta.mm_mapsize = env->me_mapsize;
1204 if (flags & MDB_FIXEDMAP)
1205 meta.mm_address = env->me_map;
1206 i = mdb_env_init_meta(env, &meta);
1207 if (i != MDB_SUCCESS) {
1208 munmap(env->me_map, env->me_mapsize);
1212 env->me_psize = meta.mm_psize;
1214 env->me_maxpg = env->me_mapsize / env->me_psize;
1216 p = (MDB_page *)env->me_map;
1217 env->me_metas[0] = METADATA(p);
1218 env->me_metas[1] = (MDB_meta *)((char *)env->me_metas[0] + meta.mm_psize);
1220 if ((i = mdb_env_read_meta(env, NULL)) != 0)
1223 DPRINTF("opened database version %u, pagesize %u",
1224 env->me_meta->mm_version, env->me_psize);
1225 DPRINTF("depth: %u", env->me_meta->mm_dbs[MAIN_DBI].md_depth);
1226 DPRINTF("entries: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_entries);
1227 DPRINTF("branch pages: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_branch_pages);
1228 DPRINTF("leaf pages: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_leaf_pages);
1229 DPRINTF("overflow pages: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_overflow_pages);
1230 DPRINTF("root: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_root);
1236 mdb_env_reader_dest(void *ptr)
1238 MDB_reader *reader = ptr;
1240 reader->mr_txnid = 0;
1245 /* downgrade the exclusive lock on the region back to shared */
1247 mdb_env_share_locks(MDB_env *env)
1249 struct flock lock_info;
1251 env->me_txns->mti_txnid = env->me_meta->mm_txnid;
1252 if (env->me_metas[0]->mm_txnid < env->me_metas[1]->mm_txnid)
1253 env->me_txns->mti_me_toggle = 1;
1255 memset((void *)&lock_info, 0, sizeof(lock_info));
1256 lock_info.l_type = F_RDLCK;
1257 lock_info.l_whence = SEEK_SET;
1258 lock_info.l_start = 0;
1259 lock_info.l_len = 1;
1260 fcntl(env->me_lfd, F_SETLK, &lock_info);
1264 mdb_env_setup_locks(MDB_env *env, char *lpath, int mode, int *excl)
1268 struct flock lock_info;
1272 if ((env->me_lfd = open(lpath, O_RDWR|O_CREAT, mode)) == -1) {
1276 /* Try to get exclusive lock. If we succeed, then
1277 * nobody is using the lock region and we should initialize it.
1279 memset((void *)&lock_info, 0, sizeof(lock_info));
1280 lock_info.l_type = F_WRLCK;
1281 lock_info.l_whence = SEEK_SET;
1282 lock_info.l_start = 0;
1283 lock_info.l_len = 1;
1284 rc = fcntl(env->me_lfd, F_SETLK, &lock_info);
1288 lock_info.l_type = F_RDLCK;
1289 rc = fcntl(env->me_lfd, F_SETLK, &lock_info);
1295 size = lseek(env->me_lfd, 0, SEEK_END);
1296 rsize = (env->me_maxreaders-1) * sizeof(MDB_reader) + sizeof(MDB_txninfo);
1297 if (size < rsize && *excl) {
1298 if (ftruncate(env->me_lfd, rsize) != 0) {
1304 size = rsize - sizeof(MDB_txninfo);
1305 env->me_maxreaders = size/sizeof(MDB_reader) + 1;
1307 env->me_txns = mmap(0, rsize, PROT_READ|PROT_WRITE, MAP_SHARED,
1309 if (env->me_txns == MAP_FAILED) {
1314 pthread_mutexattr_t mattr;
1316 pthread_mutexattr_init(&mattr);
1317 rc = pthread_mutexattr_setpshared(&mattr, PTHREAD_PROCESS_SHARED);
1321 pthread_mutex_init(&env->me_txns->mti_mutex, &mattr);
1322 pthread_mutex_init(&env->me_txns->mti_wmutex, &mattr);
1323 env->me_txns->mti_version = MDB_VERSION;
1324 env->me_txns->mti_magic = MDB_MAGIC;
1325 env->me_txns->mti_txnid = 0;
1326 env->me_txns->mti_numreaders = 0;
1327 env->me_txns->mti_me_toggle = 0;
1330 if (env->me_txns->mti_magic != MDB_MAGIC) {
1331 DPRINTF("lock region has invalid magic");
1335 if (env->me_txns->mti_version != MDB_VERSION) {
1336 DPRINTF("lock region is version %u, expected version %u",
1337 env->me_txns->mti_version, MDB_VERSION);
1338 rc = MDB_VERSION_MISMATCH;
1341 if (errno != EACCES && errno != EAGAIN) {
1355 #define LOCKNAME "/lock.mdb"
1356 #define DATANAME "/data.mdb"
1358 mdb_env_open(MDB_env *env, const char *path, unsigned int flags, mode_t mode)
1360 int oflags, rc, len, excl;
1361 char *lpath, *dpath;
1364 lpath = malloc(len + sizeof(LOCKNAME) + len + sizeof(DATANAME));
1367 dpath = lpath + len + sizeof(LOCKNAME);
1368 sprintf(lpath, "%s" LOCKNAME, path);
1369 sprintf(dpath, "%s" DATANAME, path);
1371 rc = mdb_env_setup_locks(env, lpath, mode, &excl);
1375 if (F_ISSET(flags, MDB_RDONLY))
1378 oflags = O_RDWR | O_CREAT;
1380 if ((env->me_fd = open(dpath, oflags, mode)) == -1) {
1385 if ((rc = mdb_env_open2(env, flags)) == MDB_SUCCESS) {
1386 /* synchronous fd for meta writes */
1387 if (!(flags & (MDB_RDONLY|MDB_NOSYNC)))
1389 if ((env->me_mfd = open(dpath, oflags, mode)) == -1) {
1394 env->me_path = strdup(path);
1395 DPRINTF("opened dbenv %p", (void *) env);
1396 pthread_key_create(&env->me_txkey, mdb_env_reader_dest);
1398 mdb_env_share_locks(env);
1399 env->me_dbxs = calloc(env->me_maxdbs, sizeof(MDB_dbx));
1400 env->me_dbs[0] = calloc(env->me_maxdbs, sizeof(MDB_db));
1401 env->me_dbs[1] = calloc(env->me_maxdbs, sizeof(MDB_db));
1407 if (env->me_fd >= 0) {
1411 if (env->me_lfd >= 0) {
1421 mdb_env_close(MDB_env *env)
1426 free(env->me_dbs[1]);
1427 free(env->me_dbs[0]);
1431 pthread_key_delete(env->me_txkey);
1434 munmap(env->me_map, env->me_mapsize);
1439 pid_t pid = getpid();
1440 size_t size = (env->me_maxreaders-1) * sizeof(MDB_reader) + sizeof(MDB_txninfo);
1442 for (i=0; i<env->me_txns->mti_numreaders; i++)
1443 if (env->me_txns->mti_readers[i].mr_pid == pid)
1444 env->me_txns->mti_readers[i].mr_pid = 0;
1445 munmap(env->me_txns, size);
1451 /* Search for key within a leaf page, using binary search.
1452 * Returns the smallest entry larger or equal to the key.
1453 * If exactp is non-null, stores whether the found entry was an exact match
1454 * in *exactp (1 or 0).
1455 * If kip is non-null, stores the index of the found entry in *kip.
1456 * If no entry larger or equal to the key is found, returns NULL.
1459 mdb_search_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp, MDB_val *key,
1460 int *exactp, unsigned int *kip)
1468 DPRINTF("searching %u keys in %s page %lu",
1470 IS_LEAF(mp) ? "leaf" : "branch",
1473 assert(NUMKEYS(mp) > 0);
1475 memset(&nodekey, 0, sizeof(nodekey));
1477 low = IS_LEAF(mp) ? 0 : 1;
1478 high = NUMKEYS(mp) - 1;
1479 while (low <= high) {
1480 i = (low + high) >> 1;
1481 node = NODEPTR(mp, i);
1483 nodekey.mv_size = node->mn_ksize;
1484 nodekey.mv_data = NODEKEY(node);
1486 if (txn->mt_dbxs[dbi].md_cmp)
1487 rc = txn->mt_dbxs[dbi].md_cmp(key, &nodekey);
1489 rc = _mdb_cmp(txn, dbi, key, &nodekey);
1492 DPRINTF("found leaf index %u [%.*s], rc = %i",
1493 i, (int)nodekey.mv_size, (char *)nodekey.mv_data, rc);
1495 DPRINTF("found branch index %u [%.*s -> %lu], rc = %i",
1496 i, (int)node->mn_ksize, (char *)NODEKEY(node),
1507 if (rc > 0) { /* Found entry is less than the key. */
1508 i++; /* Skip to get the smallest entry larger than key. */
1509 if (i >= NUMKEYS(mp))
1510 /* There is no entry larger or equal to the key. */
1514 *exactp = (rc == 0);
1515 if (kip) /* Store the key index if requested. */
1518 return NODEPTR(mp, i);
1522 cursor_pop_page(MDB_cursor *cursor)
1526 top = CURSOR_TOP(cursor);
1529 DPRINTF("popped page %lu off cursor %p", top->mp_page->mp_pgno, (void *) cursor);
1535 cursor_push_page(MDB_cursor *cursor, MDB_page *mp)
1539 DPRINTF("pushing page %lu on cursor %p", mp->mp_pgno, (void *) cursor);
1541 if ((ppage = calloc(1, sizeof(MDB_ppage))) == NULL)
1543 ppage->mp_page = mp;
1544 CURSOR_PUSH(cursor, ppage);
1549 mdb_get_page(MDB_txn *txn, pgno_t pgno)
1554 if (!F_ISSET(txn->mt_flags, MDB_TXN_RDONLY) && txn->mt_u.dirty_list[0].mid) {
1559 x = mdb_midl2_search(txn->mt_u.dirty_list, &id);
1560 if (x <= txn->mt_u.dirty_list[0].mid && txn->mt_u.dirty_list[x].mid == pgno) {
1561 dp = txn->mt_u.dirty_list[x].mptr;
1567 if (pgno > txn->mt_env->me_meta->mm_last_pg)
1569 p = (MDB_page *)(txn->mt_env->me_map + txn->mt_env->me_psize * pgno);
1575 mdb_search_page_root(MDB_txn *txn, MDB_dbi dbi, MDB_val *key,
1576 MDB_cursor *cursor, int modify, MDB_pageparent *mpp)
1578 MDB_page *mp = mpp->mp_page;
1581 if (cursor && cursor_push_page(cursor, mp) == NULL)
1584 while (IS_BRANCH(mp)) {
1588 DPRINTF("branch page %lu has %u keys", mp->mp_pgno, NUMKEYS(mp));
1589 assert(NUMKEYS(mp) > 1);
1590 DPRINTF("found index 0 to page %lu", NODEPGNO(NODEPTR(mp, 0)));
1592 if (key == NULL) /* Initialize cursor to first page. */
1594 else if (key->mv_size > MAXKEYSIZE && key->mv_data == NULL) {
1595 /* cursor to last page */
1599 node = mdb_search_node(txn, dbi, mp, key, &exact, &i);
1601 i = NUMKEYS(mp) - 1;
1609 DPRINTF("following index %u for key %.*s",
1610 i, (int)key->mv_size, (char *)key->mv_data);
1611 assert(i < NUMKEYS(mp));
1612 node = NODEPTR(mp, i);
1615 CURSOR_TOP(cursor)->mp_ki = i;
1617 mpp->mp_parent = mp;
1618 if ((mp = mdb_get_page(txn, NODEPGNO(node))) == NULL)
1619 return MDB_PAGE_NOTFOUND;
1623 if (cursor && cursor_push_page(cursor, mp) == NULL)
1627 MDB_dhead *dh = ((MDB_dhead *)mp)-1;
1628 if ((rc = mdb_touch(txn, mpp)) != 0)
1630 dh = ((MDB_dhead *)mpp->mp_page)-1;
1631 dh->md_parent = mpp->mp_parent;
1632 dh->md_pi = mpp->mp_pi;
1639 DPRINTF("internal error, index points to a %02X page!?",
1641 return MDB_CORRUPTED;
1644 DPRINTF("found leaf page %lu for key %.*s", mp->mp_pgno,
1645 key ? (int)key->mv_size : 0, key ? (char *)key->mv_data : NULL);
1650 /* Search for the page a given key should be in.
1651 * Stores a pointer to the found page in *mpp.
1652 * If key is NULL, search for the lowest page (used by mdb_cursor_first).
1653 * If cursor is non-null, pushes parent pages on the cursor stack.
1654 * If modify is true, visited pages are updated with new page numbers.
1657 mdb_search_page(MDB_txn *txn, MDB_dbi dbi, MDB_val *key,
1658 MDB_cursor *cursor, int modify, MDB_pageparent *mpp)
1663 /* Choose which root page to start with. If a transaction is given
1664 * use the root page from the transaction, otherwise read the last
1665 * committed root page.
1667 if (F_ISSET(txn->mt_flags, MDB_TXN_ERROR)) {
1668 DPRINTF("transaction has failed, must abort");
1671 root = txn->mt_dbs[dbi].md_root;
1673 if (root == P_INVALID) { /* Tree is empty. */
1674 DPRINTF("tree is empty");
1675 return MDB_NOTFOUND;
1678 if ((mpp->mp_page = mdb_get_page(txn, root)) == NULL)
1679 return MDB_PAGE_NOTFOUND;
1681 DPRINTF("root page has flags 0x%X", mpp->mp_page->mp_flags);
1684 /* For sub-databases, update main root first */
1685 if (dbi > MAIN_DBI && !txn->mt_dbxs[dbi].md_dirty) {
1687 rc = mdb_search_page(txn, MAIN_DBI, &txn->mt_dbxs[dbi].md_name,
1691 txn->mt_dbxs[dbi].md_dirty = 1;
1693 if (!F_ISSET(mpp->mp_page->mp_flags, P_DIRTY)) {
1694 mpp->mp_parent = NULL;
1696 if ((rc = mdb_touch(txn, mpp)))
1698 txn->mt_dbs[dbi].md_root = mpp->mp_page->mp_pgno;
1702 return mdb_search_page_root(txn, dbi, key, cursor, modify, mpp);
1706 mdb_read_data(MDB_txn *txn, MDB_node *leaf, MDB_val *data)
1708 MDB_page *omp; /* overflow mpage */
1711 if (!F_ISSET(leaf->mn_flags, F_BIGDATA)) {
1712 data->mv_size = leaf->mn_dsize;
1713 data->mv_data = NODEDATA(leaf);
1717 /* Read overflow data.
1719 data->mv_size = leaf->mn_dsize;
1720 memcpy(&pgno, NODEDATA(leaf), sizeof(pgno));
1721 if ((omp = mdb_get_page(txn, pgno)) == NULL) {
1722 DPRINTF("read overflow page %lu failed", pgno);
1723 return MDB_PAGE_NOTFOUND;
1725 data->mv_data = omp;
1731 mdb_get(MDB_txn *txn, MDB_dbi dbi,
1732 MDB_val *key, MDB_val *data)
1740 DPRINTF("===> get key [%.*s]", (int)key->mv_size, (char *)key->mv_data);
1742 if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
1745 if (key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
1749 if ((rc = mdb_search_page(txn, dbi, key, NULL, 0, &mpp)) != MDB_SUCCESS)
1752 leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, NULL);
1753 if (leaf && exact) {
1754 /* Return first duplicate data item */
1755 if (F_ISSET(txn->mt_dbs[dbi].md_flags, MDB_DUPSORT)) {
1758 mdb_xcursor_init0(txn, dbi, &mx);
1759 mdb_xcursor_init1(txn, dbi, &mx, leaf);
1760 rc = mdb_search_page(&mx.mx_txn, mx.mx_cursor.mc_dbi, NULL, NULL, 0, &mpp);
1761 if (rc != MDB_SUCCESS)
1763 leaf = NODEPTR(mpp.mp_page, 0);
1765 rc = mdb_read_data(txn, leaf, data);
1774 mdb_sibling(MDB_cursor *cursor, int move_right)
1778 MDB_ppage *parent, *top;
1781 top = CURSOR_TOP(cursor);
1782 if ((parent = SLIST_NEXT(top, mp_entry)) == NULL) {
1783 return MDB_NOTFOUND; /* root has no siblings */
1786 DPRINTF("parent page is page %lu, index %u",
1787 parent->mp_page->mp_pgno, parent->mp_ki);
1789 cursor_pop_page(cursor);
1790 if (move_right ? (parent->mp_ki + 1 >= NUMKEYS(parent->mp_page))
1791 : (parent->mp_ki == 0)) {
1792 DPRINTF("no more keys left, moving to %s sibling",
1793 move_right ? "right" : "left");
1794 if ((rc = mdb_sibling(cursor, move_right)) != MDB_SUCCESS)
1796 parent = CURSOR_TOP(cursor);
1802 DPRINTF("just moving to %s index key %u",
1803 move_right ? "right" : "left", parent->mp_ki);
1805 assert(IS_BRANCH(parent->mp_page));
1807 indx = NODEPTR(parent->mp_page, parent->mp_ki);
1808 if ((mp = mdb_get_page(cursor->mc_txn, indx->mn_pgno)) == NULL)
1809 return MDB_PAGE_NOTFOUND;
1811 mp->parent = parent->mp_page;
1812 mp->parent_index = parent->mp_ki;
1815 cursor_push_page(cursor, mp);
1821 mdb_set_key(MDB_node *node, MDB_val *key)
1826 key->mv_size = node->mn_ksize;
1827 key->mv_data = NODEKEY(node);
1833 mdb_cursor_next(MDB_cursor *cursor, MDB_val *key, MDB_val *data, MDB_cursor_op op)
1840 if (cursor->mc_eof) {
1841 return MDB_NOTFOUND;
1844 assert(cursor->mc_initialized);
1846 if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
1847 if (op == MDB_NEXT || op == MDB_NEXT_DUP) {
1848 rc = mdb_cursor_next(&cursor->mc_xcursor->mx_cursor, data, NULL, MDB_NEXT);
1849 if (op != MDB_NEXT || rc == MDB_SUCCESS)
1854 top = CURSOR_TOP(cursor);
1857 DPRINTF("cursor_next: top page is %lu in cursor %p", mp->mp_pgno, (void *) cursor);
1859 if (top->mp_ki + 1 >= NUMKEYS(mp)) {
1860 DPRINTF("=====> move to next sibling page");
1861 if (mdb_sibling(cursor, 1) != MDB_SUCCESS) {
1863 return MDB_NOTFOUND;
1865 top = CURSOR_TOP(cursor);
1867 DPRINTF("next page is %lu, key index %u", mp->mp_pgno, top->mp_ki);
1871 DPRINTF("==> cursor points to page %lu with %u keys, key index %u",
1872 mp->mp_pgno, NUMKEYS(mp), top->mp_ki);
1874 assert(IS_LEAF(mp));
1875 leaf = NODEPTR(mp, top->mp_ki);
1878 if ((rc = mdb_read_data(cursor->mc_txn, leaf, data) != MDB_SUCCESS))
1881 if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
1882 mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, leaf);
1883 rc = mdb_cursor_first(&cursor->mc_xcursor->mx_cursor, data, NULL);
1884 if (rc != MDB_SUCCESS)
1889 return mdb_set_key(leaf, key);
1893 mdb_cursor_prev(MDB_cursor *cursor, MDB_val *key, MDB_val *data, MDB_cursor_op op)
1900 assert(cursor->mc_initialized);
1902 if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
1903 if (op == MDB_PREV || op == MDB_PREV_DUP) {
1904 rc = mdb_cursor_prev(&cursor->mc_xcursor->mx_cursor, data, NULL, MDB_PREV);
1905 if (op != MDB_PREV || rc == MDB_SUCCESS)
1910 top = CURSOR_TOP(cursor);
1913 DPRINTF("cursor_prev: top page is %lu in cursor %p", mp->mp_pgno, (void *) cursor);
1915 if (top->mp_ki == 0) {
1916 DPRINTF("=====> move to prev sibling page");
1917 if (mdb_sibling(cursor, 0) != MDB_SUCCESS) {
1918 return MDB_NOTFOUND;
1920 top = CURSOR_TOP(cursor);
1922 top->mp_ki = NUMKEYS(mp) - 1;
1923 DPRINTF("prev page is %lu, key index %u", mp->mp_pgno, top->mp_ki);
1929 DPRINTF("==> cursor points to page %lu with %u keys, key index %u",
1930 mp->mp_pgno, NUMKEYS(mp), top->mp_ki);
1932 assert(IS_LEAF(mp));
1933 leaf = NODEPTR(mp, top->mp_ki);
1936 if ((rc = mdb_read_data(cursor->mc_txn, leaf, data) != MDB_SUCCESS))
1939 if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
1940 mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, leaf);
1941 rc = mdb_cursor_last(&cursor->mc_xcursor->mx_cursor, data, NULL);
1942 if (rc != MDB_SUCCESS)
1947 return mdb_set_key(leaf, key);
1951 mdb_cursor_set(MDB_cursor *cursor, MDB_val *key, MDB_val *data,
1952 MDB_cursor_op op, int *exactp)
1961 assert(key->mv_size > 0);
1963 while (CURSOR_TOP(cursor) != NULL)
1964 cursor_pop_page(cursor);
1966 rc = mdb_search_page(cursor->mc_txn, cursor->mc_dbi, key, cursor, 0, &mpp);
1967 if (rc != MDB_SUCCESS)
1969 assert(IS_LEAF(mpp.mp_page));
1971 top = CURSOR_TOP(cursor);
1972 leaf = mdb_search_node(cursor->mc_txn, cursor->mc_dbi, mpp.mp_page, key, exactp, &top->mp_ki);
1973 if (exactp != NULL && !*exactp) {
1974 /* MDB_SET specified and not an exact match. */
1975 return MDB_NOTFOUND;
1979 DPRINTF("===> inexact leaf not found, goto sibling");
1980 if ((rc = mdb_sibling(cursor, 1)) != MDB_SUCCESS)
1981 return rc; /* no entries matched */
1982 top = CURSOR_TOP(cursor);
1984 mpp.mp_page = top->mp_page;
1985 assert(IS_LEAF(mpp.mp_page));
1986 leaf = NODEPTR(mpp.mp_page, 0);
1989 cursor->mc_initialized = 1;
1993 if ((rc = mdb_read_data(cursor->mc_txn, leaf, data)) != MDB_SUCCESS)
1996 if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
1997 mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, leaf);
1998 if (op == MDB_SET || op == MDB_SET_RANGE) {
1999 rc = mdb_cursor_first(&cursor->mc_xcursor->mx_cursor, data, NULL);
2003 if (op == MDB_GET_BOTH) {
2008 op2 = MDB_SET_RANGE;
2010 rc = mdb_cursor_set(&cursor->mc_xcursor->mx_cursor, data, NULL, op2, ex2p);
2011 if (rc != MDB_SUCCESS)
2017 rc = mdb_set_key(leaf, key);
2018 if (rc == MDB_SUCCESS) {
2019 DPRINTF("==> cursor placed on key %.*s",
2020 (int)key->mv_size, (char *)key->mv_data);
2028 mdb_cursor_first(MDB_cursor *cursor, MDB_val *key, MDB_val *data)
2034 while (CURSOR_TOP(cursor) != NULL)
2035 cursor_pop_page(cursor);
2037 rc = mdb_search_page(cursor->mc_txn, cursor->mc_dbi, NULL, cursor, 0, &mpp);
2038 if (rc != MDB_SUCCESS)
2040 assert(IS_LEAF(mpp.mp_page));
2042 leaf = NODEPTR(mpp.mp_page, 0);
2043 cursor->mc_initialized = 1;
2047 if ((rc = mdb_read_data(cursor->mc_txn, leaf, data)) != MDB_SUCCESS)
2050 if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
2051 mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, leaf);
2052 rc = mdb_cursor_first(&cursor->mc_xcursor->mx_cursor, data, NULL);
2057 return mdb_set_key(leaf, key);
2061 mdb_cursor_last(MDB_cursor *cursor, MDB_val *key, MDB_val *data)
2069 while (CURSOR_TOP(cursor) != NULL)
2070 cursor_pop_page(cursor);
2072 lkey.mv_size = MAXKEYSIZE+1;
2073 lkey.mv_data = NULL;
2075 rc = mdb_search_page(cursor->mc_txn, cursor->mc_dbi, &lkey, cursor, 0, &mpp);
2076 if (rc != MDB_SUCCESS)
2078 assert(IS_LEAF(mpp.mp_page));
2080 leaf = NODEPTR(mpp.mp_page, NUMKEYS(mpp.mp_page)-1);
2081 cursor->mc_initialized = 1;
2084 top = CURSOR_TOP(cursor);
2085 top->mp_ki = NUMKEYS(top->mp_page) - 1;
2088 if ((rc = mdb_read_data(cursor->mc_txn, leaf, data)) != MDB_SUCCESS)
2091 if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
2092 mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, leaf);
2093 rc = mdb_cursor_last(&cursor->mc_xcursor->mx_cursor, data, NULL);
2099 return mdb_set_key(leaf, key);
2103 mdb_cursor_get(MDB_cursor *cursor, MDB_val *key, MDB_val *data,
2113 case MDB_GET_BOTH_RANGE:
2121 if (key == NULL || key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
2123 } else if (op != MDB_SET_RANGE)
2124 rc = mdb_cursor_set(cursor, key, data, op, NULL);
2126 rc = mdb_cursor_set(cursor, key, data, op, &exact);
2130 case MDB_NEXT_NODUP:
2131 if (!cursor->mc_initialized)
2132 rc = mdb_cursor_first(cursor, key, data);
2134 rc = mdb_cursor_next(cursor, key, data, op);
2138 case MDB_PREV_NODUP:
2139 if (!cursor->mc_initialized || cursor->mc_eof)
2140 rc = mdb_cursor_last(cursor, key, data);
2142 rc = mdb_cursor_prev(cursor, key, data, op);
2145 rc = mdb_cursor_first(cursor, key, data);
2148 rc = mdb_cursor_last(cursor, key, data);
2151 DPRINTF("unhandled/unimplemented cursor operation %u", op);
2159 /* Allocate a page and initialize it
2162 mdb_new_page(MDB_txn *txn, MDB_dbi dbi, uint32_t flags, int num)
2166 if ((dp = mdb_alloc_page(txn, NULL, 0, num)) == NULL)
2168 DPRINTF("allocated new mpage %lu, page size %u",
2169 dp->p.mp_pgno, txn->mt_env->me_psize);
2170 dp->p.mp_flags = flags | P_DIRTY;
2171 dp->p.mp_lower = PAGEHDRSZ;
2172 dp->p.mp_upper = txn->mt_env->me_psize;
2174 if (IS_BRANCH(&dp->p))
2175 txn->mt_dbs[dbi].md_branch_pages++;
2176 else if (IS_LEAF(&dp->p))
2177 txn->mt_dbs[dbi].md_leaf_pages++;
2178 else if (IS_OVERFLOW(&dp->p)) {
2179 txn->mt_dbs[dbi].md_overflow_pages += num;
2180 dp->p.mp_pages = num;
2187 mdb_leaf_size(MDB_env *env, MDB_val *key, MDB_val *data)
2191 sz = LEAFSIZE(key, data);
2192 if (data->mv_size >= env->me_psize / MDB_MINKEYS) {
2193 /* put on overflow page */
2194 sz -= data->mv_size - sizeof(pgno_t);
2197 return sz + sizeof(indx_t);
2201 mdb_branch_size(MDB_env *env, MDB_val *key)
2206 if (sz >= env->me_psize / MDB_MINKEYS) {
2207 /* put on overflow page */
2208 /* not implemented */
2209 /* sz -= key->size - sizeof(pgno_t); */
2212 return sz + sizeof(indx_t);
2216 mdb_add_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp, indx_t indx,
2217 MDB_val *key, MDB_val *data, pgno_t pgno, uint8_t flags)
2220 size_t node_size = NODESIZE;
2223 MDB_dpage *ofp = NULL; /* overflow page */
2225 assert(mp->mp_upper >= mp->mp_lower);
2227 DPRINTF("add node [%.*s] to %s page %lu at index %i, key size %zu",
2228 key ? (int)key->mv_size : 0, key ? (char *)key->mv_data : NULL,
2229 IS_LEAF(mp) ? "leaf" : "branch",
2230 mp->mp_pgno, indx, key ? key->mv_size : 0);
2233 node_size += key->mv_size;
2237 if (F_ISSET(flags, F_BIGDATA)) {
2238 /* Data already on overflow page. */
2239 node_size += sizeof(pgno_t);
2240 } else if (data->mv_size >= txn->mt_env->me_psize / MDB_MINKEYS) {
2241 int ovpages = OVPAGES(data->mv_size, txn->mt_env->me_psize);
2242 /* Put data on overflow page. */
2243 DPRINTF("data size is %zu, put on overflow page",
2245 node_size += sizeof(pgno_t);
2246 if ((ofp = mdb_new_page(txn, dbi, P_OVERFLOW, ovpages)) == NULL)
2248 DPRINTF("allocated overflow page %lu", ofp->p.mp_pgno);
2251 node_size += data->mv_size;
2255 if (node_size + sizeof(indx_t) > SIZELEFT(mp)) {
2256 DPRINTF("not enough room in page %lu, got %u ptrs",
2257 mp->mp_pgno, NUMKEYS(mp));
2258 DPRINTF("upper - lower = %u - %u = %u", mp->mp_upper, mp->mp_lower,
2259 mp->mp_upper - mp->mp_lower);
2260 DPRINTF("node size = %zu", node_size);
2264 /* Move higher pointers up one slot. */
2265 for (i = NUMKEYS(mp); i > indx; i--)
2266 mp->mp_ptrs[i] = mp->mp_ptrs[i - 1];
2268 /* Adjust free space offsets. */
2269 ofs = mp->mp_upper - node_size;
2270 assert(ofs >= mp->mp_lower + sizeof(indx_t));
2271 mp->mp_ptrs[indx] = ofs;
2273 mp->mp_lower += sizeof(indx_t);
2275 /* Write the node data. */
2276 node = NODEPTR(mp, indx);
2277 node->mn_ksize = (key == NULL) ? 0 : key->mv_size;
2278 node->mn_flags = flags;
2280 node->mn_dsize = data->mv_size;
2282 node->mn_pgno = pgno;
2285 memcpy(NODEKEY(node), key->mv_data, key->mv_size);
2290 if (F_ISSET(flags, F_BIGDATA))
2291 memcpy(node->mn_data + key->mv_size, data->mv_data,
2294 memcpy(node->mn_data + key->mv_size, data->mv_data,
2297 memcpy(node->mn_data + key->mv_size, &ofp->p.mp_pgno,
2299 memcpy(METADATA(&ofp->p), data->mv_data, data->mv_size);
2307 mdb_del_node(MDB_page *mp, indx_t indx)
2310 indx_t i, j, numkeys, ptr;
2314 DPRINTF("delete node %u on %s page %lu", indx,
2315 IS_LEAF(mp) ? "leaf" : "branch", mp->mp_pgno);
2316 assert(indx < NUMKEYS(mp));
2318 node = NODEPTR(mp, indx);
2319 sz = NODESIZE + node->mn_ksize;
2321 if (F_ISSET(node->mn_flags, F_BIGDATA))
2322 sz += sizeof(pgno_t);
2324 sz += NODEDSZ(node);
2327 ptr = mp->mp_ptrs[indx];
2328 numkeys = NUMKEYS(mp);
2329 for (i = j = 0; i < numkeys; i++) {
2331 mp->mp_ptrs[j] = mp->mp_ptrs[i];
2332 if (mp->mp_ptrs[i] < ptr)
2333 mp->mp_ptrs[j] += sz;
2338 base = (char *)mp + mp->mp_upper;
2339 memmove(base + sz, base, ptr - mp->mp_upper);
2341 mp->mp_lower -= sizeof(indx_t);
2346 mdb_xcursor_init0(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx)
2351 mx->mx_txn.mt_dbxs = mx->mx_dbxs;
2352 mx->mx_txn.mt_dbs = mx->mx_dbs;
2353 mx->mx_dbxs[0] = txn->mt_dbxs[0];
2354 mx->mx_dbxs[1] = txn->mt_dbxs[1];
2356 mx->mx_dbxs[2] = txn->mt_dbxs[dbi];
2361 mx->mx_dbxs[dbn+1].md_parent = dbn;
2362 mx->mx_dbxs[dbn+1].md_cmp = mx->mx_dbxs[dbn].md_dcmp;
2363 mx->mx_dbxs[dbn+1].md_rel = mx->mx_dbxs[dbn].md_rel;
2364 mx->mx_dbxs[dbn+1].md_dirty = 0;
2365 mx->mx_txn.mt_numdbs = dbn+2;
2367 SLIST_INIT(&mx->mx_cursor.mc_stack);
2368 mx->mx_cursor.mc_txn = &mx->mx_txn;
2369 mx->mx_cursor.mc_dbi = dbn+1;
2373 mdb_xcursor_init1(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx, MDB_node *node)
2375 MDB_db *db = NODEDATA(node);
2377 mx->mx_dbs[0] = txn->mt_dbs[0];
2378 mx->mx_dbs[1] = txn->mt_dbs[1];
2380 mx->mx_dbs[2] = txn->mt_dbs[dbi];
2385 mx->mx_dbs[dbn] = *db;
2386 mx->mx_dbxs[dbn].md_name.mv_data = NODEKEY(node);
2387 mx->mx_dbxs[dbn].md_name.mv_size = node->mn_ksize;
2388 mx->mx_txn.mt_next_pgno = txn->mt_next_pgno;
2389 mx->mx_txn.mt_oldest = txn->mt_oldest;
2390 mx->mx_txn.mt_u = txn->mt_u;
2394 mdb_xcursor_fini(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx)
2396 txn->mt_next_pgno = mx->mx_txn.mt_next_pgno;
2397 txn->mt_oldest = mx->mx_txn.mt_oldest;
2398 txn->mt_u = mx->mx_txn.mt_u;
2399 txn->mt_dbs[0] = mx->mx_dbs[0];
2400 txn->mt_dbs[1] = mx->mx_dbs[1];
2401 txn->mt_dbxs[0].md_dirty = mx->mx_dbxs[0].md_dirty;
2402 txn->mt_dbxs[1].md_dirty = mx->mx_dbxs[1].md_dirty;
2404 txn->mt_dbs[dbi] = mx->mx_dbs[2];
2405 txn->mt_dbxs[dbi].md_dirty = mx->mx_dbxs[2].md_dirty;
2410 mdb_cursor_open(MDB_txn *txn, MDB_dbi dbi, MDB_cursor **ret)
2413 size_t size = sizeof(MDB_cursor);
2415 if (txn == NULL || ret == NULL || !dbi || dbi >= txn->mt_numdbs)
2418 if (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT)
2419 size += sizeof(MDB_xcursor);
2421 if ((cursor = calloc(1, size)) != NULL) {
2422 SLIST_INIT(&cursor->mc_stack);
2423 cursor->mc_dbi = dbi;
2424 cursor->mc_txn = txn;
2425 if (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT) {
2426 MDB_xcursor *mx = (MDB_xcursor *)(cursor + 1);
2427 cursor->mc_xcursor = mx;
2428 mdb_xcursor_init0(txn, dbi, mx);
2439 /* Return the count of duplicate data items for the current key */
2441 mdb_cursor_count(MDB_cursor *mc, unsigned long *countp)
2443 if (mc == NULL || countp == NULL)
2446 if (!(mc->mc_txn->mt_dbs[mc->mc_dbi].md_flags & MDB_DUPSORT))
2449 if (!mc->mc_xcursor->mx_cursor.mc_initialized)
2452 *countp = mc->mc_xcursor->mx_txn.mt_dbs[mc->mc_xcursor->mx_cursor.mc_dbi].md_entries;
2457 mdb_cursor_close(MDB_cursor *cursor)
2459 if (cursor != NULL) {
2460 while(!CURSOR_EMPTY(cursor))
2461 cursor_pop_page(cursor);
2462 if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
2463 mdb_xcursor_fini(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor);
2464 while(!CURSOR_EMPTY(&cursor->mc_xcursor->mx_cursor))
2465 cursor_pop_page(&cursor->mc_xcursor->mx_cursor);
2473 mdb_update_key(MDB_page *mp, indx_t indx, MDB_val *key)
2475 indx_t ptr, i, numkeys;
2481 node = NODEPTR(mp, indx);
2482 ptr = mp->mp_ptrs[indx];
2483 DPRINTF("update key %u (ofs %u) [%.*s] to [%.*s] on page %lu",
2485 (int)node->mn_ksize, (char *)NODEKEY(node),
2486 (int)key->mv_size, (char *)key->mv_data,
2489 delta = key->mv_size - node->mn_ksize;
2491 if (delta > 0 && SIZELEFT(mp) < delta) {
2492 DPRINTF("OUCH! Not enough room, delta = %d", delta);
2496 numkeys = NUMKEYS(mp);
2497 for (i = 0; i < numkeys; i++) {
2498 if (mp->mp_ptrs[i] <= ptr)
2499 mp->mp_ptrs[i] -= delta;
2502 base = (char *)mp + mp->mp_upper;
2503 len = ptr - mp->mp_upper + NODESIZE;
2504 memmove(base - delta, base, len);
2505 mp->mp_upper -= delta;
2507 node = NODEPTR(mp, indx);
2508 node->mn_ksize = key->mv_size;
2511 memcpy(NODEKEY(node), key->mv_data, key->mv_size);
2516 /* Move a node from src to dst.
2519 mdb_move_node(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *src, indx_t srcindx,
2520 MDB_pageparent *dst, indx_t dstindx)
2526 srcnode = NODEPTR(src->mp_page, srcindx);
2527 DPRINTF("moving %s node %u [%.*s] on page %lu to node %u on page %lu",
2528 IS_LEAF(src->mp_page) ? "leaf" : "branch",
2530 (int)srcnode->mn_ksize, (char *)NODEKEY(srcnode),
2531 src->mp_page->mp_pgno,
2532 dstindx, dst->mp_page->mp_pgno);
2534 /* Mark src and dst as dirty. */
2535 if ((rc = mdb_touch(txn, src)) ||
2536 (rc = mdb_touch(txn, dst)))
2539 /* Add the node to the destination page.
2541 key.mv_size = srcnode->mn_ksize;
2542 key.mv_data = NODEKEY(srcnode);
2543 data.mv_size = NODEDSZ(srcnode);
2544 data.mv_data = NODEDATA(srcnode);
2545 rc = mdb_add_node(txn, dbi, dst->mp_page, dstindx, &key, &data, NODEPGNO(srcnode),
2547 if (rc != MDB_SUCCESS)
2550 /* Delete the node from the source page.
2552 mdb_del_node(src->mp_page, srcindx);
2554 /* Update the parent separators.
2556 if (srcindx == 0 && src->mp_pi != 0) {
2557 DPRINTF("update separator for source page %lu to [%.*s]",
2558 src->mp_page->mp_pgno, (int)key.mv_size, (char *)key.mv_data);
2559 if ((rc = mdb_update_key(src->mp_parent, src->mp_pi,
2560 &key)) != MDB_SUCCESS)
2564 if (srcindx == 0 && IS_BRANCH(src->mp_page)) {
2566 nullkey.mv_size = 0;
2567 assert(mdb_update_key(src->mp_page, 0, &nullkey) == MDB_SUCCESS);
2570 if (dstindx == 0 && dst->mp_pi != 0) {
2571 DPRINTF("update separator for destination page %lu to [%.*s]",
2572 dst->mp_page->mp_pgno, (int)key.mv_size, (char *)key.mv_data);
2573 if ((rc = mdb_update_key(dst->mp_parent, dst->mp_pi,
2574 &key)) != MDB_SUCCESS)
2578 if (dstindx == 0 && IS_BRANCH(dst->mp_page)) {
2580 nullkey.mv_size = 0;
2581 assert(mdb_update_key(dst->mp_page, 0, &nullkey) == MDB_SUCCESS);
2588 mdb_merge(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *src, MDB_pageparent *dst)
2597 DPRINTF("merging page %lu and %lu", src->mp_page->mp_pgno, dst->mp_page->mp_pgno);
2599 assert(txn != NULL);
2600 assert(src->mp_parent); /* can't merge root page */
2601 assert(dst->mp_parent);
2603 /* Mark src and dst as dirty. */
2604 if ((rc = mdb_touch(txn, src)) ||
2605 (rc = mdb_touch(txn, dst)))
2608 /* Move all nodes from src to dst.
2610 for (i = 0; i < NUMKEYS(src->mp_page); i++) {
2611 srcnode = NODEPTR(src->mp_page, i);
2613 key.mv_size = srcnode->mn_ksize;
2614 key.mv_data = NODEKEY(srcnode);
2615 data.mv_size = NODEDSZ(srcnode);
2616 data.mv_data = NODEDATA(srcnode);
2617 rc = mdb_add_node(txn, dbi, dst->mp_page, NUMKEYS(dst->mp_page), &key,
2618 &data, NODEPGNO(srcnode), srcnode->mn_flags);
2619 if (rc != MDB_SUCCESS)
2623 DPRINTF("dst page %lu now has %u keys (%.1f%% filled)",
2624 dst->mp_page->mp_pgno, NUMKEYS(dst->mp_page), (float)PAGEFILL(txn->mt_env, dst->mp_page) / 10);
2626 /* Unlink the src page from parent.
2628 mdb_del_node(src->mp_parent, src->mp_pi);
2629 if (src->mp_pi == 0) {
2631 if ((rc = mdb_update_key(src->mp_parent, 0, &key)) != MDB_SUCCESS)
2635 if (IS_LEAF(src->mp_page))
2636 txn->mt_dbs[dbi].md_leaf_pages--;
2638 txn->mt_dbs[dbi].md_branch_pages--;
2640 mpp.mp_page = src->mp_parent;
2641 dh = (MDB_dhead *)src->mp_parent;
2643 mpp.mp_parent = dh->md_parent;
2644 mpp.mp_pi = dh->md_pi;
2646 return mdb_rebalance(txn, dbi, &mpp);
2649 #define FILL_THRESHOLD 250
2652 mdb_rebalance(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *mpp)
2657 indx_t si = 0, di = 0;
2659 assert(txn != NULL);
2660 assert(mpp != NULL);
2662 DPRINTF("rebalancing %s page %lu (has %u keys, %.1f%% full)",
2663 IS_LEAF(mpp->mp_page) ? "leaf" : "branch",
2664 mpp->mp_page->mp_pgno, NUMKEYS(mpp->mp_page), (float)PAGEFILL(txn->mt_env, mpp->mp_page) / 10);
2666 if (PAGEFILL(txn->mt_env, mpp->mp_page) >= FILL_THRESHOLD) {
2667 DPRINTF("no need to rebalance page %lu, above fill threshold",
2668 mpp->mp_page->mp_pgno);
2672 if (mpp->mp_parent == NULL) {
2673 if (NUMKEYS(mpp->mp_page) == 0) {
2674 DPRINTF("tree is completely empty");
2675 txn->mt_dbs[dbi].md_root = P_INVALID;
2676 txn->mt_dbs[dbi].md_depth--;
2677 txn->mt_dbs[dbi].md_leaf_pages--;
2678 } else if (IS_BRANCH(mpp->mp_page) && NUMKEYS(mpp->mp_page) == 1) {
2679 DPRINTF("collapsing root page!");
2680 txn->mt_dbs[dbi].md_root = NODEPGNO(NODEPTR(mpp->mp_page, 0));
2681 if ((root = mdb_get_page(txn, txn->mt_dbs[dbi].md_root)) == NULL)
2682 return MDB_PAGE_NOTFOUND;
2683 txn->mt_dbs[dbi].md_depth--;
2684 txn->mt_dbs[dbi].md_branch_pages--;
2686 DPRINTF("root page doesn't need rebalancing");
2690 /* The parent (branch page) must have at least 2 pointers,
2691 * otherwise the tree is invalid.
2693 assert(NUMKEYS(mpp->mp_parent) > 1);
2695 /* Leaf page fill factor is below the threshold.
2696 * Try to move keys from left or right neighbor, or
2697 * merge with a neighbor page.
2702 if (mpp->mp_pi == 0) {
2703 /* We're the leftmost leaf in our parent.
2705 DPRINTF("reading right neighbor");
2706 node = NODEPTR(mpp->mp_parent, mpp->mp_pi + 1);
2707 if ((npp.mp_page = mdb_get_page(txn, NODEPGNO(node))) == NULL)
2708 return MDB_PAGE_NOTFOUND;
2709 npp.mp_pi = mpp->mp_pi + 1;
2711 di = NUMKEYS(mpp->mp_page);
2713 /* There is at least one neighbor to the left.
2715 DPRINTF("reading left neighbor");
2716 node = NODEPTR(mpp->mp_parent, mpp->mp_pi - 1);
2717 if ((npp.mp_page = mdb_get_page(txn, NODEPGNO(node))) == NULL)
2718 return MDB_PAGE_NOTFOUND;
2719 npp.mp_pi = mpp->mp_pi - 1;
2720 si = NUMKEYS(npp.mp_page) - 1;
2723 npp.mp_parent = mpp->mp_parent;
2725 DPRINTF("found neighbor page %lu (%u keys, %.1f%% full)",
2726 npp.mp_page->mp_pgno, NUMKEYS(npp.mp_page), (float)PAGEFILL(txn->mt_env, npp.mp_page) / 10);
2728 /* If the neighbor page is above threshold and has at least two
2729 * keys, move one key from it.
2731 * Otherwise we should try to merge them.
2733 if (PAGEFILL(txn->mt_env, npp.mp_page) >= FILL_THRESHOLD && NUMKEYS(npp.mp_page) >= 2)
2734 return mdb_move_node(txn, dbi, &npp, si, mpp, di);
2735 else { /* FIXME: if (has_enough_room()) */
2736 if (mpp->mp_pi == 0)
2737 return mdb_merge(txn, dbi, &npp, mpp);
2739 return mdb_merge(txn, dbi, mpp, &npp);
2744 mdb_del0(MDB_txn *txn, MDB_dbi dbi, unsigned int ki, MDB_pageparent *mpp, MDB_node *leaf)
2748 /* add overflow pages to free list */
2749 if (F_ISSET(leaf->mn_flags, F_BIGDATA)) {
2753 memcpy(&pg, NODEDATA(leaf), sizeof(pg));
2754 ovpages = OVPAGES(NODEDSZ(leaf), txn->mt_env->me_psize);
2755 for (i=0; i<ovpages; i++) {
2756 DPRINTF("freed ov page %lu", pg);
2757 mdb_midl_insert(txn->mt_free_pgs, pg);
2761 mdb_del_node(mpp->mp_page, ki);
2762 txn->mt_dbs[dbi].md_entries--;
2763 rc = mdb_rebalance(txn, dbi, mpp);
2764 if (rc != MDB_SUCCESS)
2765 txn->mt_flags |= MDB_TXN_ERROR;
2771 mdb_del(MDB_txn *txn, MDB_dbi dbi,
2772 MDB_val *key, MDB_val *data,
2780 DPRINTF("========> delete key %.*s", (int)key->mv_size, (char *)key->mv_data);
2782 assert(key != NULL);
2784 if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
2787 if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
2791 if (key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
2795 mpp.mp_parent = NULL;
2797 if ((rc = mdb_search_page(txn, dbi, key, NULL, 1, &mpp)) != MDB_SUCCESS)
2800 leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, &ki);
2801 if (leaf == NULL || !exact) {
2802 return MDB_NOTFOUND;
2805 if (F_ISSET(txn->mt_dbs[dbi].md_flags, MDB_DUPSORT)) {
2809 mdb_xcursor_init0(txn, dbi, &mx);
2810 mdb_xcursor_init1(txn, dbi, &mx, leaf);
2811 if (flags == MDB_DEL_DUP) {
2812 rc = mdb_del(&mx.mx_txn, mx.mx_cursor.mc_dbi, data, NULL, 0);
2813 mdb_xcursor_fini(txn, dbi, &mx);
2814 if (rc != MDB_SUCCESS)
2816 /* If sub-DB still has entries, we're done */
2817 if (mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi].md_root != P_INVALID) {
2818 memcpy(NODEDATA(leaf), &mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi],
2822 /* otherwise fall thru and delete the sub-DB */
2824 /* add all the child DB's pages to the free list */
2825 rc = mdb_search_page(&mx.mx_txn, mx.mx_cursor.mc_dbi,
2826 NULL, &mx.mx_cursor, 0, &mp2);
2827 if (rc == MDB_SUCCESS) {
2828 MDB_ppage *top, *parent;
2832 cursor_pop_page(&mx.mx_cursor);
2833 top = CURSOR_TOP(&mx.mx_cursor);
2835 parent = SLIST_NEXT(top, mp_entry);
2836 while (parent != NULL) {
2837 for (i=0; i<NUMKEYS(top->mp_page); i++) {
2838 ni = NODEPTR(top->mp_page, i);
2839 mdb_midl_insert(txn->mt_free_pgs, ni->mn_pgno);
2843 if (parent->mp_ki >= NUMKEYS(parent->mp_page)) {
2844 cursor_pop_page(&mx.mx_cursor);
2845 top = CURSOR_TOP(&mx.mx_cursor);
2846 parent = SLIST_NEXT(top, mp_entry);
2848 ni = NODEPTR(parent->mp_page, parent->mp_ki);
2849 top->mp_page = mdb_get_page(&mx.mx_txn, ni->mn_pgno);
2854 mdb_midl_insert(txn->mt_free_pgs, mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi].md_root);
2859 if (data && (rc = mdb_read_data(txn, leaf, data)) != MDB_SUCCESS)
2862 return mdb_del0(txn, dbi, ki, &mpp, leaf);
2865 /* Split page <*mpp>, and insert <key,(data|newpgno)> in either left or
2866 * right sibling, at index <*newindxp> (as if unsplit). Updates *mpp and
2867 * *newindxp with the actual values after split, ie if *mpp and *newindxp
2868 * refer to a node in the new right sibling page.
2871 mdb_split(MDB_txn *txn, MDB_dbi dbi, MDB_page **mpp, unsigned int *newindxp,
2872 MDB_val *newkey, MDB_val *newdata, pgno_t newpgno)
2875 int rc = MDB_SUCCESS, ins_new = 0;
2878 unsigned int i, j, split_indx;
2880 MDB_val sepkey, rkey, rdata;
2882 MDB_dpage *mdp, *rdp, *pdp;
2885 assert(txn != NULL);
2887 dh = ((MDB_dhead *)*mpp) - 1;
2888 mdp = (MDB_dpage *)dh;
2889 newindx = *newindxp;
2891 DPRINTF("-----> splitting %s page %lu and adding [%.*s] at index %i",
2892 IS_LEAF(&mdp->p) ? "leaf" : "branch", mdp->p.mp_pgno,
2893 (int)newkey->mv_size, (char *)newkey->mv_data, *newindxp);
2895 if (mdp->h.md_parent == NULL) {
2896 if ((pdp = mdb_new_page(txn, dbi, P_BRANCH, 1)) == NULL)
2899 mdp->h.md_parent = &pdp->p;
2900 txn->mt_dbs[dbi].md_root = pdp->p.mp_pgno;
2901 DPRINTF("root split! new root = %lu", pdp->p.mp_pgno);
2902 txn->mt_dbs[dbi].md_depth++;
2904 /* Add left (implicit) pointer. */
2905 if ((rc = mdb_add_node(txn, dbi, &pdp->p, 0, NULL, NULL,
2906 mdp->p.mp_pgno, 0)) != MDB_SUCCESS)
2909 DPRINTF("parent branch page is %lu", mdp->h.md_parent->mp_pgno);
2912 /* Create a right sibling. */
2913 if ((rdp = mdb_new_page(txn, dbi, mdp->p.mp_flags, 1)) == NULL)
2915 rdp->h.md_parent = mdp->h.md_parent;
2916 rdp->h.md_pi = mdp->h.md_pi + 1;
2917 DPRINTF("new right sibling: page %lu", rdp->p.mp_pgno);
2919 /* Move half of the keys to the right sibling. */
2920 if ((copy = malloc(txn->mt_env->me_psize)) == NULL)
2922 memcpy(copy, &mdp->p, txn->mt_env->me_psize);
2923 memset(&mdp->p.mp_ptrs, 0, txn->mt_env->me_psize - PAGEHDRSZ);
2924 mdp->p.mp_lower = PAGEHDRSZ;
2925 mdp->p.mp_upper = txn->mt_env->me_psize;
2927 split_indx = NUMKEYS(copy) / 2 + 1;
2929 /* First find the separating key between the split pages.
2931 memset(&sepkey, 0, sizeof(sepkey));
2932 if (newindx == split_indx) {
2933 sepkey.mv_size = newkey->mv_size;
2934 sepkey.mv_data = newkey->mv_data;
2936 node = NODEPTR(copy, split_indx);
2937 sepkey.mv_size = node->mn_ksize;
2938 sepkey.mv_data = NODEKEY(node);
2941 DPRINTF("separator is [%.*s]", (int)sepkey.mv_size, (char *)sepkey.mv_data);
2943 /* Copy separator key to the parent.
2945 if (SIZELEFT(rdp->h.md_parent) < mdb_branch_size(txn->mt_env, &sepkey)) {
2946 rc = mdb_split(txn, dbi, &rdp->h.md_parent, &rdp->h.md_pi,
2947 &sepkey, NULL, rdp->p.mp_pgno);
2949 /* Right page might now have changed parent.
2950 * Check if left page also changed parent.
2952 if (rdp->h.md_parent != mdp->h.md_parent &&
2953 mdp->h.md_pi >= NUMKEYS(mdp->h.md_parent)) {
2954 mdp->h.md_parent = rdp->h.md_parent;
2955 mdp->h.md_pi = rdp->h.md_pi - 1;
2958 rc = mdb_add_node(txn, dbi, rdp->h.md_parent, rdp->h.md_pi,
2959 &sepkey, NULL, rdp->p.mp_pgno, 0);
2961 if (rc != MDB_SUCCESS) {
2966 for (i = j = 0; i <= NUMKEYS(copy); j++) {
2967 if (i < split_indx) {
2968 /* Re-insert in left sibling. */
2971 /* Insert in right sibling. */
2972 if (i == split_indx)
2973 /* Reset insert index for right sibling. */
2974 j = (i == newindx && ins_new);
2978 if (i == newindx && !ins_new) {
2979 /* Insert the original entry that caused the split. */
2980 rkey.mv_data = newkey->mv_data;
2981 rkey.mv_size = newkey->mv_size;
2982 if (IS_LEAF(&mdp->p)) {
2983 rdata.mv_data = newdata->mv_data;
2984 rdata.mv_size = newdata->mv_size;
2991 /* Update page and index for the new key. */
2994 } else if (i == NUMKEYS(copy)) {
2997 node = NODEPTR(copy, i);
2998 rkey.mv_data = NODEKEY(node);
2999 rkey.mv_size = node->mn_ksize;
3000 if (IS_LEAF(&mdp->p)) {
3001 rdata.mv_data = NODEDATA(node);
3002 rdata.mv_size = node->mn_dsize;
3004 pgno = node->mn_pgno;
3005 flags = node->mn_flags;
3010 if (!IS_LEAF(&mdp->p) && j == 0) {
3011 /* First branch index doesn't need key data. */
3015 rc = mdb_add_node(txn, dbi, &pdp->p, j, &rkey, &rdata, pgno,flags);
3023 mdb_put0(MDB_txn *txn, MDB_dbi dbi,
3024 MDB_val *key, MDB_val *data, unsigned int flags)
3026 int rc = MDB_SUCCESS, exact;
3030 MDB_val xdata, *rdata;
3033 DPRINTF("==> put key %.*s, size %zu, data size %zu",
3034 (int)key->mv_size, (char *)key->mv_data, key->mv_size, data->mv_size);
3036 mpp.mp_parent = NULL;
3038 rc = mdb_search_page(txn, dbi, key, NULL, 1, &mpp);
3039 if (rc == MDB_SUCCESS) {
3040 leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, &ki);
3041 if (leaf && exact) {
3042 if (F_ISSET(txn->mt_dbs[dbi].md_flags, MDB_DUPSORT)) {
3045 if (flags == MDB_NOOVERWRITE) {
3046 DPRINTF("duplicate key %.*s",
3047 (int)key->mv_size, (char *)key->mv_data);
3048 return MDB_KEYEXIST;
3050 /* same size, just replace it */
3051 if (NODEDSZ(leaf) == data->mv_size) {
3052 memcpy(NODEDATA(leaf), data->mv_data, data->mv_size);
3055 mdb_del_node(mpp.mp_page, ki);
3057 if (leaf == NULL) { /* append if not found */
3058 ki = NUMKEYS(mpp.mp_page);
3059 DPRINTF("appending key at index %i", ki);
3061 } else if (rc == MDB_NOTFOUND) {
3063 /* new file, just write a root leaf page */
3064 DPRINTF("allocating new root leaf page");
3065 if ((dp = mdb_new_page(txn, dbi, P_LEAF, 1)) == NULL) {
3068 mpp.mp_page = &dp->p;
3069 txn->mt_dbs[dbi].md_root = mpp.mp_page->mp_pgno;
3070 txn->mt_dbs[dbi].md_depth++;
3076 assert(IS_LEAF(mpp.mp_page));
3077 DPRINTF("there are %u keys, should insert new key at index %i",
3078 NUMKEYS(mpp.mp_page), ki);
3080 /* For sorted dups, the data item at this level is a DB record
3081 * for a child DB; the actual data elements are stored as keys
3084 if (F_ISSET(txn->mt_dbs[dbi].md_flags, MDB_DUPSORT)) {
3086 xdata.mv_size = sizeof(MDB_db);
3087 xdata.mv_data = &dummy;
3088 memset(&dummy, 0, sizeof(dummy));
3089 dummy.md_root = P_INVALID;
3094 if (SIZELEFT(mpp.mp_page) < mdb_leaf_size(txn->mt_env, key, rdata)) {
3095 rc = mdb_split(txn, dbi, &mpp.mp_page, &ki, key, rdata, P_INVALID);
3097 /* There is room already in this leaf page. */
3098 rc = mdb_add_node(txn, dbi, mpp.mp_page, ki, key, rdata, 0, 0);
3101 if (rc != MDB_SUCCESS)
3102 txn->mt_flags |= MDB_TXN_ERROR;
3104 txn->mt_dbs[dbi].md_entries++;
3106 /* Remember if we just added a subdatabase */
3107 if (flags & F_SUBDATA) {
3108 leaf = NODEPTR(mpp.mp_page, ki);
3109 leaf->mn_flags |= F_SUBDATA;
3112 /* Now store the actual data in the child DB. Note that we're
3113 * storing the user data in the keys field, so there are strict
3114 * size limits on dupdata. The actual data fields of the child
3115 * DB are all zero size.
3117 if (F_ISSET(txn->mt_dbs[dbi].md_flags, MDB_DUPSORT)) {
3120 leaf = NODEPTR(mpp.mp_page, ki);
3122 mdb_xcursor_init0(txn, dbi, &mx);
3123 mdb_xcursor_init1(txn, dbi, &mx, leaf);
3126 if (flags == MDB_NODUPDATA)
3127 flags = MDB_NOOVERWRITE;
3128 rc = mdb_put0(&mx.mx_txn, mx.mx_cursor.mc_dbi, data, &xdata, flags);
3129 mdb_xcursor_fini(txn, dbi, &mx);
3130 memcpy(NODEDATA(leaf), &mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi],
3140 mdb_put(MDB_txn *txn, MDB_dbi dbi,
3141 MDB_val *key, MDB_val *data, unsigned int flags)
3143 assert(key != NULL);
3144 assert(data != NULL);
3146 if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
3149 if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
3153 if (key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
3157 if ((flags & (MDB_NOOVERWRITE|MDB_NODUPDATA)) != flags)
3160 return mdb_put0(txn, dbi, key, data, flags);
3164 mdb_env_get_flags(MDB_env *env, unsigned int *arg)
3169 *arg = env->me_flags;
3174 mdb_env_get_path(MDB_env *env, const char **arg)
3179 *arg = env->me_path;
3184 mdb_stat0(MDB_env *env, MDB_db *db, MDB_stat *arg)
3186 arg->ms_psize = env->me_psize;
3187 arg->ms_depth = db->md_depth;
3188 arg->ms_branch_pages = db->md_branch_pages;
3189 arg->ms_leaf_pages = db->md_leaf_pages;
3190 arg->ms_overflow_pages = db->md_overflow_pages;
3191 arg->ms_entries = db->md_entries;
3196 mdb_env_stat(MDB_env *env, MDB_stat *arg)
3198 if (env == NULL || arg == NULL)
3201 return mdb_stat0(env, &env->me_meta->mm_dbs[MAIN_DBI], arg);
3204 int mdb_open(MDB_txn *txn, const char *name, unsigned int flags, MDB_dbi *dbi)
3214 if (flags & (MDB_DUPSORT|MDB_REVERSEKEY|MDB_INTEGERKEY))
3215 txn->mt_dbs[MAIN_DBI].md_flags |= (flags & (MDB_DUPSORT|MDB_REVERSEKEY|MDB_INTEGERKEY));
3219 /* Is the DB already open? */
3221 for (i=2; i<txn->mt_numdbs; i++) {
3222 if (len == txn->mt_dbxs[i].md_name.mv_size &&
3223 !strncmp(name, txn->mt_dbxs[i].md_name.mv_data, len)) {
3229 if (txn->mt_numdbs >= txn->mt_env->me_maxdbs - 1)
3232 /* Find the DB info */
3234 key.mv_data = (void *)name;
3235 rc = mdb_get(txn, MAIN_DBI, &key, &data);
3237 /* Create if requested */
3238 if (rc == MDB_NOTFOUND && (flags & MDB_CREATE)) {
3240 data.mv_size = sizeof(MDB_db);
3241 data.mv_data = &dummy;
3242 memset(&dummy, 0, sizeof(dummy));
3243 dummy.md_root = P_INVALID;
3244 dummy.md_flags = flags & 0xffff;
3245 rc = mdb_put0(txn, MAIN_DBI, &key, &data, F_SUBDATA);
3249 /* OK, got info, add to table */
3250 if (rc == MDB_SUCCESS) {
3251 txn->mt_dbxs[txn->mt_numdbs].md_name.mv_data = strdup(name);
3252 txn->mt_dbxs[txn->mt_numdbs].md_name.mv_size = len;
3253 txn->mt_dbxs[txn->mt_numdbs].md_cmp = NULL;
3254 txn->mt_dbxs[txn->mt_numdbs].md_dcmp = NULL;
3255 txn->mt_dbxs[txn->mt_numdbs].md_rel = NULL;
3256 txn->mt_dbxs[txn->mt_numdbs].md_parent = MAIN_DBI;
3257 txn->mt_dbxs[txn->mt_numdbs].md_dirty = dirty;
3258 memcpy(&txn->mt_dbs[txn->mt_numdbs], data.mv_data, sizeof(MDB_db));
3259 *dbi = txn->mt_numdbs;
3266 int mdb_stat(MDB_txn *txn, MDB_dbi dbi, MDB_stat *arg)
3268 if (txn == NULL || arg == NULL || dbi >= txn->mt_numdbs)
3271 return mdb_stat0(txn->mt_env, &txn->mt_dbs[dbi], arg);
3274 void mdb_close(MDB_txn *txn, MDB_dbi dbi)
3277 if (dbi <= MAIN_DBI || dbi >= txn->mt_numdbs)
3279 ptr = txn->mt_dbxs[dbi].md_name.mv_data;
3280 txn->mt_dbxs[dbi].md_name.mv_data = NULL;
3281 txn->mt_dbxs[dbi].md_name.mv_size = 0;
3285 int mdb_set_compare(MDB_txn *txn, MDB_dbi dbi, MDB_cmp_func *cmp)
3287 if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
3290 txn->mt_dbxs[dbi].md_cmp = cmp;
3294 int mdb_set_dupsort(MDB_txn *txn, MDB_dbi dbi, MDB_cmp_func *cmp)
3296 if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
3299 txn->mt_dbxs[dbi].md_dcmp = cmp;
3303 int mdb_set_relfunc(MDB_txn *txn, MDB_dbi dbi, MDB_rel_func *rel)
3305 if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
3308 txn->mt_dbxs[dbi].md_rel = rel;