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 !(__STDC_VERSION__ >= 199901L || defined(__GNUC__))
63 # define DPRINTF (void) /* Vararg macros may be unsupported */
65 # define DPRINTF(fmt, ...) /* Requires 2 or more args */ \
66 fprintf(stderr, "%s:%d: " fmt "\n", __func__, __LINE__, __VA_ARGS__)
68 # define DPRINTF(fmt, ...) ((void) 0)
70 #define DPUTS(arg) DPRINTF("%s", arg)
74 #define MDB_MAGIC 0xBEEFC0DE
76 #define MAXKEYSIZE 511
78 #define P_INVALID (~0UL)
80 #define F_ISSET(w, f) (((w) & (f)) == (f))
82 typedef uint16_t indx_t;
84 #define DEFAULT_READERS 126
85 #define DEFAULT_MAPSIZE 1048576
87 /* Lock descriptor stuff */
89 #define CACHELINE 64 /* most CPUs. Itanium uses 128 */
92 typedef struct MDB_rxbody {
98 typedef struct MDB_reader {
101 #define mr_txnid mru.mrx.mrb_txnid
102 #define mr_pid mru.mrx.mrb_pid
103 #define mr_tid mru.mrx.mrb_tid
104 /* cache line alignment */
105 char pad[(sizeof(MDB_rxbody)+CACHELINE-1) & ~(CACHELINE-1)];
109 typedef struct MDB_txbody {
111 uint32_t mtb_version;
112 pthread_mutex_t mtb_mutex;
114 uint32_t mtb_numreaders;
115 uint32_t mtb_me_toggle;
118 typedef struct MDB_txninfo {
121 #define mti_magic mt1.mtb.mtb_magic
122 #define mti_version mt1.mtb.mtb_version
123 #define mti_mutex mt1.mtb.mtb_mutex
124 #define mti_txnid mt1.mtb.mtb_txnid
125 #define mti_numreaders mt1.mtb.mtb_numreaders
126 #define mti_me_toggle mt1.mtb.mtb_me_toggle
127 char pad[(sizeof(MDB_txbody)+CACHELINE-1) & ~(CACHELINE-1)];
130 pthread_mutex_t mt2_wmutex;
131 #define mti_wmutex mt2.mt2_wmutex
132 char pad[(sizeof(pthread_mutex_t)+CACHELINE-1) & ~(CACHELINE-1)];
134 MDB_reader mti_readers[1];
137 /* Common header for all page types. Overflow pages
138 * occupy a number of contiguous pages with no
139 * headers on any page after the first.
141 typedef struct MDB_page { /* represents a page of storage */
142 #define mp_pgno mp_p.p_pgno
144 pgno_t p_pgno; /* page number */
145 void * p_align; /* for IL32P64 */
147 #define P_BRANCH 0x01 /* branch page */
148 #define P_LEAF 0x02 /* leaf page */
149 #define P_OVERFLOW 0x04 /* overflow page */
150 #define P_META 0x08 /* meta page */
151 #define P_DIRTY 0x10 /* dirty page */
152 #define P_LEAF2 0x20 /* DB with small, fixed size keys and no data */
154 #define mp_lower mp_pb.pb.pb_lower
155 #define mp_upper mp_pb.pb.pb_upper
156 #define mp_pages mp_pb.pb_pages
159 indx_t pb_lower; /* lower bound of free space */
160 indx_t pb_upper; /* upper bound of free space */
162 uint32_t pb_pages; /* number of overflow pages */
164 indx_t pb_ksize; /* on a LEAF2 page */
168 indx_t mp_ptrs[1]; /* dynamic size */
171 #define PAGEHDRSZ ((unsigned) offsetof(MDB_page, mp_ptrs))
173 #define NUMKEYS(p) (((p)->mp_lower - PAGEHDRSZ) >> 1)
174 #define SIZELEFT(p) (indx_t)((p)->mp_upper - (p)->mp_lower)
175 #define PAGEFILL(env, p) (1000L * ((env)->me_psize - PAGEHDRSZ - SIZELEFT(p)) / \
176 ((env)->me_psize - PAGEHDRSZ))
177 #define IS_LEAF(p) F_ISSET((p)->mp_flags, P_LEAF)
178 #define IS_BRANCH(p) F_ISSET((p)->mp_flags, P_BRANCH)
179 #define IS_OVERFLOW(p) F_ISSET((p)->mp_flags, P_OVERFLOW)
181 #define OVPAGES(size, psize) (PAGEHDRSZ + size + psize - 1) / psize;
183 typedef struct MDB_db {
187 ULONG md_branch_pages;
189 ULONG md_overflow_pages;
197 typedef struct MDB_meta { /* meta (footer) page content */
200 void *mm_address; /* address for fixed mapping */
201 size_t mm_mapsize; /* size of mmap region */
202 MDB_db mm_dbs[2]; /* first is free space, 2nd is main db */
203 #define mm_psize mm_dbs[0].md_pad
204 #define mm_flags mm_dbs[0].md_flags
205 pgno_t mm_last_pg; /* last used page in file */
206 ULONG mm_txnid; /* txnid that committed this page */
209 typedef struct MDB_dhead { /* a dirty page */
211 unsigned md_pi; /* parent index */
215 typedef struct MDB_dpage {
220 typedef struct MDB_oldpages {
221 struct MDB_oldpages *mo_next;
223 pgno_t mo_pages[1]; /* dynamic */
226 typedef struct MDB_pageparent {
232 static MDB_dpage *mdb_alloc_page(MDB_txn *txn, MDB_page *parent, unsigned int parent_idx, int num);
233 static int mdb_touch(MDB_txn *txn, MDB_pageparent *mp);
235 typedef struct MDB_ppage { /* ordered list of pages */
237 unsigned int mp_ki; /* cursor index on page */
240 #define CURSOR_TOP(c) (&(c)->mc_stack[(c)->mc_snum-1])
241 #define CURSOR_PARENT(c) (&(c)->mc_stack[(c)->mc_snum-2])
247 MDB_ppage mc_stack[32]; /* stack of parent pages */
248 unsigned int mc_snum; /* number of pushed pages */
250 short mc_initialized; /* 1 if initialized */
251 short mc_eof; /* 1 if end is reached */
252 struct MDB_xcursor *mc_xcursor;
255 #define METADATA(p) ((void *)((char *)p + PAGEHDRSZ))
257 typedef struct MDB_node {
258 #define mn_pgno mn_p.np_pgno
259 #define mn_dsize mn_p.np_dsize
261 pgno_t np_pgno; /* child page number */
262 uint32_t np_dsize; /* leaf data size */
264 unsigned int mn_flags:4;
265 unsigned int mn_ksize:12; /* key size */
266 #define F_BIGDATA 0x01 /* data put on overflow page */
267 #define F_SUBDATA 0x02 /* data is a sub-database */
268 #define F_DUPDATA 0x04 /* data has duplicates */
272 typedef struct MDB_dbx {
274 MDB_cmp_func *md_cmp; /* user compare function */
275 MDB_cmp_func *md_dcmp; /* user dupsort function */
276 MDB_rel_func *md_rel; /* user relocate function */
278 unsigned int md_dirty;
282 pgno_t mt_next_pgno; /* next unallocated page */
286 pgno_t *mt_free_pgs; /* this is an IDL */
288 MIDL2 *dirty_list; /* modified pages */
291 MDB_dbx *mt_dbxs; /* array */
293 unsigned int mt_numdbs;
295 #define MDB_TXN_RDONLY 0x01 /* read-only transaction */
296 #define MDB_TXN_ERROR 0x02 /* an error has occurred */
297 #define MDB_TXN_METOGGLE 0x04 /* used meta page 1 */
298 unsigned int mt_flags;
301 /* Context for sorted-dup records */
302 typedef struct MDB_xcursor {
303 MDB_cursor mx_cursor;
312 int me_mfd; /* just for writing the meta pages */
313 #define MDB_FATAL_ERROR 0x80000000U
315 uint32_t me_extrapad; /* unused for now */
316 unsigned int me_maxreaders;
317 unsigned int me_numdbs;
318 unsigned int me_maxdbs;
321 MDB_txninfo *me_txns;
322 MDB_meta *me_metas[2];
324 MDB_txn *me_txn; /* current write transaction */
326 off_t me_size; /* current file size */
327 pgno_t me_maxpg; /* me_mapsize / me_psize */
328 unsigned int me_psize;
329 unsigned int me_db_toggle;
330 MDB_dbx *me_dbxs; /* array */
332 MDB_oldpages *me_pghead;
333 pthread_key_t me_txkey; /* thread-key for readers */
334 pgno_t me_free_pgs[MDB_IDL_UM_SIZE];
335 MIDL2 me_dirty_list[MDB_IDL_DB_SIZE];
338 #define NODESIZE offsetof(MDB_node, mn_data)
340 #define INDXSIZE(k) (NODESIZE + ((k) == NULL ? 0 : (k)->mv_size))
341 #define LEAFSIZE(k, d) (NODESIZE + (k)->mv_size + (d)->mv_size)
342 #define NODEPTR(p, i) ((MDB_node *)((char *)(p) + (p)->mp_ptrs[i]))
343 #define NODEKEY(node) (void *)((node)->mn_data)
344 #define NODEDATA(node) (void *)((char *)(node)->mn_data + (node)->mn_ksize)
345 #define NODEPGNO(node) ((node)->mn_pgno)
346 #define NODEDSZ(node) ((node)->mn_dsize)
348 #define MDB_COMMIT_PAGES 64 /* max number of pages to write in one commit */
350 static int mdb_search_page_root(MDB_txn *txn,
351 MDB_dbi dbi, MDB_val *key,
352 MDB_cursor *cursor, int modify,
353 MDB_pageparent *mpp);
354 static int mdb_search_page(MDB_txn *txn,
355 MDB_dbi dbi, MDB_val *key,
356 MDB_cursor *cursor, int modify,
357 MDB_pageparent *mpp);
359 static int mdb_env_read_header(MDB_env *env, MDB_meta *meta);
360 static int mdb_env_read_meta(MDB_env *env, int *which);
361 static int mdb_env_write_meta(MDB_txn *txn);
362 static MDB_page *mdb_get_page(MDB_txn *txn, pgno_t pgno);
364 static MDB_node *mdb_search_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp,
365 MDB_val *key, int *exactp, unsigned int *kip);
366 static int mdb_add_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp,
367 indx_t indx, MDB_val *key, MDB_val *data,
368 pgno_t pgno, uint8_t flags);
369 static void mdb_del_node(MDB_page *mp, indx_t indx);
370 static int mdb_del0(MDB_txn *txn, MDB_dbi dbi, unsigned int ki,
371 MDB_pageparent *mpp, MDB_node *leaf);
372 static int mdb_put0(MDB_txn *txn, MDB_dbi dbi,
373 MDB_val *key, MDB_val *data, unsigned int flags);
374 static int mdb_read_data(MDB_txn *txn, MDB_node *leaf, MDB_val *data);
376 static int mdb_rebalance(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *mp);
377 static int mdb_update_key(MDB_page *mp, indx_t indx, MDB_val *key);
378 static int mdb_move_node(MDB_txn *txn, MDB_dbi dbi,
379 MDB_pageparent *src, indx_t srcindx,
380 MDB_pageparent *dst, indx_t dstindx);
381 static int mdb_merge(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *src,
382 MDB_pageparent *dst);
383 static int mdb_split(MDB_txn *txn, MDB_dbi dbi, MDB_page **mpp,
384 unsigned int *newindxp, MDB_val *newkey,
385 MDB_val *newdata, pgno_t newpgno);
386 static MDB_dpage *mdb_new_page(MDB_txn *txn, MDB_dbi dbi, uint32_t flags, int num);
388 static void cursor_pop_page(MDB_cursor *cursor);
389 static MDB_ppage *cursor_push_page(MDB_cursor *cursor,
392 static int mdb_set_key(MDB_node *node, MDB_val *key);
393 static int mdb_sibling(MDB_cursor *cursor, int move_right);
394 static int mdb_cursor_next(MDB_cursor *cursor,
395 MDB_val *key, MDB_val *data, MDB_cursor_op op);
396 static int mdb_cursor_prev(MDB_cursor *cursor,
397 MDB_val *key, MDB_val *data, MDB_cursor_op op);
398 static int mdb_cursor_set(MDB_cursor *cursor,
399 MDB_val *key, MDB_val *data, MDB_cursor_op op, int *exactp);
400 static int mdb_cursor_first(MDB_cursor *cursor,
401 MDB_val *key, MDB_val *data);
402 static int mdb_cursor_last(MDB_cursor *cursor,
403 MDB_val *key, MDB_val *data);
405 static void mdb_xcursor_init0(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx);
406 static void mdb_xcursor_init1(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx, MDB_node *node);
407 static void mdb_xcursor_fini(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx);
409 static size_t mdb_leaf_size(MDB_env *env, MDB_val *key,
411 static size_t mdb_branch_size(MDB_env *env, MDB_val *key);
413 static int memncmp(const void *s1, size_t n1,
414 const void *s2, size_t n2);
415 static int memnrcmp(const void *s1, size_t n1,
416 const void *s2, size_t n2);
419 memncmp(const void *s1, size_t n1, const void *s2, size_t n2)
421 int diff, len_diff = -1;
424 len_diff = (n1 > n2);
427 diff = memcmp(s1, s2, n1);
428 return diff ? diff : len_diff;
432 memnrcmp(const void *s1, size_t n1, const void *s2, size_t n2)
434 const unsigned char *p1, *p2, *p1_lim;
441 p1 = (const unsigned char *)s1 + n1 - 1;
442 p2 = (const unsigned char *)s2 + n2 - 1;
444 for (p1_lim = (n1 <= n2 ? s1 : s2); *p1 == *p2; p1--, p2--) {
446 return (p1 != s1) ? (p1 != p2) : (p2 != s2) ? -1 : 0;
452 mdb_version(int *maj, int *min, int *pat)
454 *maj = MDB_VERSION_MAJOR;
455 *min = MDB_VERSION_MINOR;
456 *pat = MDB_VERSION_PATCH;
457 return MDB_VERSION_STRING;
460 static const char *errstr[] = {
461 "MDB_KEYEXIST: Key/data pair already exists",
462 "MDB_NOTFOUND: No matching key/data pair found",
463 "MDB_PAGE_NOTFOUND: Requested page not found",
464 "MDB_CORRUPTED: Located page was wrong type",
465 "MDB_PANIC: Update of meta page failed",
466 "MDB_VERSION_MISMATCH: Database environment version mismatch"
470 mdb_strerror(int err)
473 return ("Successful return: 0");
475 if (err >= MDB_KEYEXIST && err <= MDB_VERSION_MISMATCH)
476 return (char *)errstr[err - MDB_KEYEXIST];
478 return strerror(err);
482 mdb_cmp(MDB_txn *txn, MDB_dbi dbi, const MDB_val *a, const MDB_val *b)
484 if (txn->mt_dbxs[dbi].md_cmp)
485 return txn->mt_dbxs[dbi].md_cmp(a, b);
487 if (txn->mt_dbs[dbi].md_flags & (MDB_REVERSEKEY
488 #if __BYTE_ORDER == __LITTLE_ENDIAN
492 return memnrcmp(a->mv_data, a->mv_size, b->mv_data, b->mv_size);
494 return memncmp((char *)a->mv_data, a->mv_size, b->mv_data, b->mv_size);
498 mdb_dcmp(MDB_txn *txn, MDB_dbi dbi, const MDB_val *a, const MDB_val *b)
500 if (txn->mt_dbxs[dbi].md_dcmp)
501 return txn->mt_dbxs[dbi].md_dcmp(a, b);
503 return memncmp((char *)a->mv_data, a->mv_size, b->mv_data, b->mv_size);
506 /* Allocate new page(s) for writing */
508 mdb_alloc_page(MDB_txn *txn, MDB_page *parent, unsigned int parent_idx, int num)
511 pgno_t pgno = P_INVALID;
515 if (txn->mt_txnid > 2) {
517 oldest = txn->mt_txnid - 2;
518 if (!txn->mt_env->me_pghead && txn->mt_dbs[FREE_DBI].md_root != P_INVALID) {
519 /* See if there's anything in the free DB */
524 mpp.mp_parent = NULL;
526 mdb_search_page(txn, FREE_DBI, NULL, NULL, 0, &mpp);
527 leaf = NODEPTR(mpp.mp_page, 0);
528 kptr = (ULONG *)NODEKEY(leaf);
530 /* It's potentially usable, unless there are still
531 * older readers outstanding. Grab it.
533 if (oldest > *kptr) {
538 mdb_read_data(txn, leaf, &data);
539 idl = (ULONG *)data.mv_data;
540 mop = malloc(sizeof(MDB_oldpages) + MDB_IDL_SIZEOF(idl) - sizeof(pgno_t));
541 mop->mo_next = txn->mt_env->me_pghead;
542 mop->mo_txnid = *kptr;
543 txn->mt_env->me_pghead = mop;
544 memcpy(mop->mo_pages, idl, MDB_IDL_SIZEOF(idl));
549 DPRINTF("IDL read txn %lu root %lu num %lu",
550 mop->mo_txnid, txn->mt_dbs[FREE_DBI].md_root, idl[0]);
551 for (i=0; i<idl[0]; i++) {
552 DPRINTF("IDL %lu", idl[i+1]);
556 /* drop this IDL from the DB */
557 mpp.mp_parent = NULL;
559 mdb_search_page(txn, FREE_DBI, NULL, NULL, 1, &mpp);
560 leaf = NODEPTR(mpp.mp_page, 0);
561 mdb_del0(txn, FREE_DBI, 0, &mpp, leaf);
564 if (txn->mt_env->me_pghead) {
566 for (i=0; i<txn->mt_env->me_txns->mti_numreaders; i++) {
567 ULONG mr = txn->mt_env->me_txns->mti_readers[i].mr_txnid;
570 oldest = txn->mt_env->me_txns->mti_readers[i].mr_txnid;
572 if (oldest > txn->mt_env->me_pghead->mo_txnid) {
573 MDB_oldpages *mop = txn->mt_env->me_pghead;
574 txn->mt_oldest = oldest;
576 /* FIXME: For now, always use fresh pages. We
577 * really ought to search the free list for a
582 /* peel pages off tail, so we only have to truncate the list */
583 pgno = MDB_IDL_LAST(mop->mo_pages);
584 if (MDB_IDL_IS_RANGE(mop->mo_pages)) {
586 if (mop->mo_pages[2] > mop->mo_pages[1])
587 mop->mo_pages[0] = 0;
591 if (MDB_IDL_IS_ZERO(mop->mo_pages)) {
592 txn->mt_env->me_pghead = mop->mo_next;
600 if (pgno == P_INVALID) {
601 /* DB size is maxed out */
602 if (txn->mt_next_pgno + num >= txn->mt_env->me_maxpg)
605 if ((dp = malloc(txn->mt_env->me_psize * num + sizeof(MDB_dhead))) == NULL)
608 dp->h.md_parent = parent;
609 dp->h.md_pi = parent_idx;
610 if (pgno == P_INVALID) {
611 dp->p.mp_pgno = txn->mt_next_pgno;
612 txn->mt_next_pgno += num;
614 dp->p.mp_pgno = pgno;
616 mid.mid = dp->p.mp_pgno;
618 mdb_midl2_insert(txn->mt_u.dirty_list, &mid);
623 /* Touch a page: make it dirty and re-insert into tree with updated pgno.
626 mdb_touch(MDB_txn *txn, MDB_pageparent *pp)
628 MDB_page *mp = pp->mp_page;
633 if (!F_ISSET(mp->mp_flags, P_DIRTY)) {
635 if ((dp = mdb_alloc_page(txn, pp->mp_parent, pp->mp_pi, 1)) == NULL)
637 DPRINTF("touched page %lu -> %lu", mp->mp_pgno, dp->p.mp_pgno);
638 mdb_midl_insert(txn->mt_free_pgs, mp->mp_pgno);
639 pgno = dp->p.mp_pgno;
640 memcpy(&dp->p, mp, txn->mt_env->me_psize);
643 mp->mp_flags |= P_DIRTY;
645 /* Update the page number to new touched page. */
646 if (pp->mp_parent != NULL)
647 NODEPGNO(NODEPTR(pp->mp_parent, pp->mp_pi)) = mp->mp_pgno;
654 mdb_env_sync(MDB_env *env, int force)
657 if (force || !F_ISSET(env->me_flags, MDB_NOSYNC)) {
658 if (fdatasync(env->me_fd))
665 mdb_txn_begin(MDB_env *env, int rdonly, MDB_txn **ret)
670 if (env->me_flags & MDB_FATAL_ERROR) {
671 DPUTS("mdb_txn_begin: environment had fatal error, must shutdown!");
674 if ((txn = calloc(1, sizeof(MDB_txn))) == NULL) {
675 DPRINTF("calloc: %s", strerror(errno));
680 txn->mt_flags |= MDB_TXN_RDONLY;
682 txn->mt_u.dirty_list = env->me_dirty_list;
683 txn->mt_u.dirty_list[0].mid = 0;
684 txn->mt_free_pgs = env->me_free_pgs;
685 txn->mt_free_pgs[0] = 0;
687 pthread_mutex_lock(&env->me_txns->mti_wmutex);
688 env->me_txns->mti_txnid++;
691 txn->mt_txnid = env->me_txns->mti_txnid;
693 MDB_reader *r = pthread_getspecific(env->me_txkey);
696 pthread_mutex_lock(&env->me_txns->mti_mutex);
697 for (i=0; i<env->me_txns->mti_numreaders; i++)
698 if (env->me_txns->mti_readers[i].mr_pid == 0)
700 if (i == env->me_maxreaders) {
701 pthread_mutex_unlock(&env->me_txns->mti_mutex);
704 env->me_txns->mti_readers[i].mr_pid = getpid();
705 env->me_txns->mti_readers[i].mr_tid = pthread_self();
706 r = &env->me_txns->mti_readers[i];
707 pthread_setspecific(env->me_txkey, r);
708 if (i >= env->me_txns->mti_numreaders)
709 env->me_txns->mti_numreaders = i+1;
710 pthread_mutex_unlock(&env->me_txns->mti_mutex);
712 r->mr_txnid = txn->mt_txnid;
713 txn->mt_u.reader = r;
720 toggle = env->me_txns->mti_me_toggle;
721 if ((rc = mdb_env_read_meta(env, &toggle)) != MDB_SUCCESS) {
726 /* Copy the DB arrays */
727 txn->mt_numdbs = env->me_numdbs;
728 txn->mt_dbxs = env->me_dbxs; /* mostly static anyway */
729 txn->mt_dbs = malloc(env->me_maxdbs * sizeof(MDB_db));
730 memcpy(txn->mt_dbs, env->me_meta->mm_dbs, 2 * sizeof(MDB_db));
731 if (txn->mt_numdbs > 2)
732 memcpy(txn->mt_dbs+2, env->me_dbs[env->me_db_toggle]+2,
733 (txn->mt_numdbs - 2) * sizeof(MDB_db));
737 txn->mt_flags |= MDB_TXN_METOGGLE;
738 txn->mt_next_pgno = env->me_meta->mm_last_pg+1;
741 DPRINTF("begin transaction %lu on mdbenv %p, root page %lu",
742 txn->mt_txnid, (void *) env, txn->mt_dbs[MAIN_DBI].md_root);
749 mdb_txn_abort(MDB_txn *txn)
757 DPRINTF("abort transaction %lu on mdbenv %p, root page %lu",
758 txn->mt_txnid, (void *) env, txn->mt_dbs[MAIN_DBI].md_root);
762 if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
763 txn->mt_u.reader->mr_txnid = 0;
768 /* Discard all dirty pages. */
769 for (i=1; i<=txn->mt_u.dirty_list[0].mid; i++)
770 free(txn->mt_u.dirty_list[i].mptr);
772 while ((mop = txn->mt_env->me_pghead)) {
773 txn->mt_env->me_pghead = mop->mo_next;
778 env->me_txns->mti_txnid--;
779 for (i=2; i<env->me_numdbs; i++)
780 env->me_dbxs[i].md_dirty = 0;
781 pthread_mutex_unlock(&env->me_txns->mti_wmutex);
788 mdb_txn_commit(MDB_txn *txn)
797 struct iovec iov[MDB_COMMIT_PAGES];
800 assert(txn->mt_env != NULL);
804 if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
809 if (txn != env->me_txn) {
810 DPUTS("attempt to commit unknown transaction");
815 if (F_ISSET(txn->mt_flags, MDB_TXN_ERROR)) {
816 DPUTS("error flag is set, can't commit");
821 if (!txn->mt_u.dirty_list[0].mid)
824 DPRINTF("committing transaction %lu on mdbenv %p, root page %lu",
825 txn->mt_txnid, (void *) env, txn->mt_dbs[MAIN_DBI].md_root);
827 /* should only be one record now */
828 if (env->me_pghead) {
832 mop = env->me_pghead;
833 key.mv_size = sizeof(pgno_t);
834 key.mv_data = (char *)&mop->mo_txnid;
835 data.mv_size = MDB_IDL_SIZEOF(mop->mo_pages);
836 data.mv_data = mop->mo_pages;
837 mdb_put0(txn, FREE_DBI, &key, &data, 0);
838 free(env->me_pghead);
839 env->me_pghead = NULL;
841 /* save to free list */
842 if (!MDB_IDL_IS_ZERO(txn->mt_free_pgs)) {
846 /* make sure last page of freeDB is touched and on freelist */
847 key.mv_size = MAXKEYSIZE+1;
849 mpp.mp_parent = NULL;
851 mdb_search_page(txn, FREE_DBI, &key, NULL, 1, &mpp);
856 ULONG *idl = txn->mt_free_pgs;
857 DPRINTF("IDL write txn %lu root %lu num %lu",
858 txn->mt_txnid, txn->mt_dbs[FREE_DBI].md_root, idl[0]);
859 for (i=0; i<idl[0]; i++) {
860 DPRINTF("IDL %lu", idl[i+1]);
864 /* write to last page of freeDB */
865 key.mv_size = sizeof(pgno_t);
866 key.mv_data = (char *)&txn->mt_txnid;
867 data.mv_size = MDB_IDL_SIZEOF(txn->mt_free_pgs);
868 data.mv_data = txn->mt_free_pgs;
869 mdb_put0(txn, FREE_DBI, &key, &data, 0);
872 /* Update DB root pointers. Their pages have already been
873 * touched so this is all in-place and cannot fail.
877 data.mv_size = sizeof(MDB_db);
879 for (i = 2; i < txn->mt_numdbs; i++) {
880 if (txn->mt_dbxs[i].md_dirty) {
881 data.mv_data = &txn->mt_dbs[i];
882 mdb_put0(txn, MAIN_DBI, &txn->mt_dbxs[i].md_name, &data, 0);
887 /* Commit up to MDB_COMMIT_PAGES dirty pages to disk until done.
895 for (; i<=txn->mt_u.dirty_list[0].mid; i++) {
896 dp = txn->mt_u.dirty_list[i].mptr;
897 if (dp->p.mp_pgno != next) {
899 DPRINTF("committing %u dirty pages", n);
900 rc = writev(env->me_fd, iov, n);
904 DPUTS("short write, filesystem full?");
906 DPRINTF("writev: %s", strerror(errno));
913 lseek(env->me_fd, dp->p.mp_pgno * env->me_psize, SEEK_SET);
914 next = dp->p.mp_pgno;
916 DPRINTF("committing page %lu", dp->p.mp_pgno);
917 iov[n].iov_len = env->me_psize * dp->h.md_num;
918 iov[n].iov_base = &dp->p;
919 size += iov[n].iov_len;
920 next = dp->p.mp_pgno + dp->h.md_num;
921 /* clear dirty flag */
922 dp->p.mp_flags &= ~P_DIRTY;
923 if (++n >= MDB_COMMIT_PAGES) {
932 DPRINTF("committing %u dirty pages", n);
933 rc = writev(env->me_fd, iov, n);
937 DPUTS("short write, filesystem full?");
939 DPRINTF("writev: %s", strerror(errno));
946 /* Drop the dirty pages.
948 for (i=1; i<=txn->mt_u.dirty_list[0].mid; i++)
949 free(txn->mt_u.dirty_list[i].mptr);
951 if ((n = mdb_env_sync(env, 0)) != 0 ||
952 (n = mdb_env_write_meta(txn)) != MDB_SUCCESS) {
959 /* update the DB tables */
961 int toggle = !env->me_db_toggle;
963 for (i = 2; i < env->me_numdbs; i++) {
964 if (txn->mt_dbxs[i].md_dirty) {
965 env->me_dbs[toggle][i] = txn->mt_dbs[i];
966 txn->mt_dbxs[i].md_dirty = 0;
969 for (i = env->me_numdbs; i < txn->mt_numdbs; i++) {
970 txn->mt_dbxs[i].md_dirty = 0;
971 env->me_dbxs[i] = txn->mt_dbxs[i];
972 env->me_dbs[toggle][i] = txn->mt_dbs[i];
974 env->me_db_toggle = toggle;
975 env->me_numdbs = txn->mt_numdbs;
980 pthread_mutex_unlock(&env->me_txns->mti_wmutex);
987 mdb_env_read_header(MDB_env *env, MDB_meta *meta)
996 /* We don't know the page size yet, so use a minimum value.
999 if ((rc = pread(env->me_fd, page, PAGESIZE, 0)) == 0) {
1001 } else if (rc != PAGESIZE) {
1004 DPRINTF("read: %s", strerror(errno));
1008 p = (MDB_page *)page;
1010 if (!F_ISSET(p->mp_flags, P_META)) {
1011 DPRINTF("page %lu not a meta page", p->mp_pgno);
1016 if (m->mm_magic != MDB_MAGIC) {
1017 DPUTS("meta has invalid magic");
1021 if (m->mm_version != MDB_VERSION) {
1022 DPRINTF("database is version %u, expected version %u",
1023 m->mm_version, MDB_VERSION);
1024 return MDB_VERSION_MISMATCH;
1027 memcpy(meta, m, sizeof(*m));
1032 mdb_env_init_meta(MDB_env *env, MDB_meta *meta)
1039 DPUTS("writing new meta page");
1040 psize = sysconf(_SC_PAGE_SIZE);
1042 meta->mm_magic = MDB_MAGIC;
1043 meta->mm_version = MDB_VERSION;
1044 meta->mm_psize = psize;
1045 meta->mm_last_pg = 1;
1046 meta->mm_flags = env->me_flags & 0xffff;
1047 meta->mm_flags |= MDB_INTEGERKEY;
1048 meta->mm_dbs[0].md_root = P_INVALID;
1049 meta->mm_dbs[1].md_root = P_INVALID;
1051 p = calloc(2, psize);
1053 p->mp_flags = P_META;
1056 memcpy(m, meta, sizeof(*meta));
1058 q = (MDB_page *)((char *)p + psize);
1061 q->mp_flags = P_META;
1064 memcpy(m, meta, sizeof(*meta));
1066 rc = write(env->me_fd, p, psize * 2);
1068 return (rc == (int)psize * 2) ? MDB_SUCCESS : errno;
1072 mdb_env_write_meta(MDB_txn *txn)
1075 MDB_meta meta, metab;
1077 int rc, len, toggle;
1080 assert(txn != NULL);
1081 assert(txn->mt_env != NULL);
1083 toggle = !F_ISSET(txn->mt_flags, MDB_TXN_METOGGLE);
1084 DPRINTF("writing meta page %d for root page %lu",
1085 toggle, txn->mt_dbs[MAIN_DBI].md_root);
1089 metab.mm_txnid = env->me_metas[toggle]->mm_txnid;
1090 metab.mm_last_pg = env->me_metas[toggle]->mm_last_pg;
1092 ptr = (char *)&meta;
1093 off = offsetof(MDB_meta, mm_dbs[0].md_depth);
1094 len = sizeof(MDB_meta) - off;
1097 meta.mm_dbs[0] = txn->mt_dbs[0];
1098 meta.mm_dbs[1] = txn->mt_dbs[1];
1099 meta.mm_last_pg = txn->mt_next_pgno - 1;
1100 meta.mm_txnid = txn->mt_txnid;
1103 off += env->me_psize;
1106 /* Write to the SYNC fd */
1107 rc = pwrite(env->me_mfd, ptr, len, off);
1109 DPUTS("write failed, disk error?");
1110 /* On a failure, the pagecache still contains the new data.
1111 * Write some old data back, to prevent it from being used.
1112 * Use the non-SYNC fd; we know it will fail anyway.
1114 meta.mm_last_pg = metab.mm_last_pg;
1115 meta.mm_txnid = metab.mm_txnid;
1116 rc = pwrite(env->me_fd, ptr, len, off);
1117 env->me_flags |= MDB_FATAL_ERROR;
1120 txn->mt_env->me_txns->mti_me_toggle = toggle;
1126 mdb_env_read_meta(MDB_env *env, int *which)
1130 assert(env != NULL);
1134 else if (env->me_metas[0]->mm_txnid < env->me_metas[1]->mm_txnid)
1137 if (env->me_meta != env->me_metas[toggle])
1138 env->me_meta = env->me_metas[toggle];
1140 DPRINTF("Using meta page %d", toggle);
1146 mdb_env_create(MDB_env **env)
1150 e = calloc(1, sizeof(MDB_env));
1151 if (!e) return ENOMEM;
1153 e->me_maxreaders = DEFAULT_READERS;
1163 mdb_env_set_mapsize(MDB_env *env, size_t size)
1167 env->me_mapsize = size;
1172 mdb_env_set_maxdbs(MDB_env *env, int dbs)
1174 env->me_maxdbs = dbs;
1179 mdb_env_set_maxreaders(MDB_env *env, int readers)
1181 env->me_maxreaders = readers;
1186 mdb_env_get_maxreaders(MDB_env *env, int *readers)
1188 if (!env || !readers)
1190 *readers = env->me_maxreaders;
1195 mdb_env_open2(MDB_env *env, unsigned int flags)
1201 env->me_flags = flags;
1203 memset(&meta, 0, sizeof(meta));
1205 if ((i = mdb_env_read_header(env, &meta)) != 0) {
1208 DPUTS("new mdbenv");
1212 if (!env->me_mapsize) {
1213 env->me_mapsize = newenv ? DEFAULT_MAPSIZE : meta.mm_mapsize;
1217 if (meta.mm_address && (flags & MDB_FIXEDMAP))
1219 env->me_map = mmap(meta.mm_address, env->me_mapsize, PROT_READ, i,
1221 if (env->me_map == MAP_FAILED)
1225 meta.mm_mapsize = env->me_mapsize;
1226 if (flags & MDB_FIXEDMAP)
1227 meta.mm_address = env->me_map;
1228 i = mdb_env_init_meta(env, &meta);
1229 if (i != MDB_SUCCESS) {
1230 munmap(env->me_map, env->me_mapsize);
1234 env->me_psize = meta.mm_psize;
1236 env->me_maxpg = env->me_mapsize / env->me_psize;
1238 p = (MDB_page *)env->me_map;
1239 env->me_metas[0] = METADATA(p);
1240 env->me_metas[1] = (MDB_meta *)((char *)env->me_metas[0] + meta.mm_psize);
1242 if ((i = mdb_env_read_meta(env, NULL)) != 0)
1245 DPRINTF("opened database version %u, pagesize %u",
1246 env->me_meta->mm_version, env->me_psize);
1247 DPRINTF("depth: %u", env->me_meta->mm_dbs[MAIN_DBI].md_depth);
1248 DPRINTF("entries: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_entries);
1249 DPRINTF("branch pages: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_branch_pages);
1250 DPRINTF("leaf pages: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_leaf_pages);
1251 DPRINTF("overflow pages: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_overflow_pages);
1252 DPRINTF("root: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_root);
1258 mdb_env_reader_dest(void *ptr)
1260 MDB_reader *reader = ptr;
1262 reader->mr_txnid = 0;
1267 /* downgrade the exclusive lock on the region back to shared */
1269 mdb_env_share_locks(MDB_env *env)
1271 struct flock lock_info;
1273 env->me_txns->mti_txnid = env->me_meta->mm_txnid;
1274 if (env->me_metas[0]->mm_txnid < env->me_metas[1]->mm_txnid)
1275 env->me_txns->mti_me_toggle = 1;
1277 memset((void *)&lock_info, 0, sizeof(lock_info));
1278 lock_info.l_type = F_RDLCK;
1279 lock_info.l_whence = SEEK_SET;
1280 lock_info.l_start = 0;
1281 lock_info.l_len = 1;
1282 fcntl(env->me_lfd, F_SETLK, &lock_info);
1286 mdb_env_setup_locks(MDB_env *env, char *lpath, int mode, int *excl)
1290 struct flock lock_info;
1294 if ((env->me_lfd = open(lpath, O_RDWR|O_CREAT, mode)) == -1) {
1298 /* Try to get exclusive lock. If we succeed, then
1299 * nobody is using the lock region and we should initialize it.
1301 memset((void *)&lock_info, 0, sizeof(lock_info));
1302 lock_info.l_type = F_WRLCK;
1303 lock_info.l_whence = SEEK_SET;
1304 lock_info.l_start = 0;
1305 lock_info.l_len = 1;
1306 rc = fcntl(env->me_lfd, F_SETLK, &lock_info);
1310 lock_info.l_type = F_RDLCK;
1311 rc = fcntl(env->me_lfd, F_SETLK, &lock_info);
1317 size = lseek(env->me_lfd, 0, SEEK_END);
1318 rsize = (env->me_maxreaders-1) * sizeof(MDB_reader) + sizeof(MDB_txninfo);
1319 if (size < rsize && *excl) {
1320 if (ftruncate(env->me_lfd, rsize) != 0) {
1326 size = rsize - sizeof(MDB_txninfo);
1327 env->me_maxreaders = size/sizeof(MDB_reader) + 1;
1329 env->me_txns = mmap(0, rsize, PROT_READ|PROT_WRITE, MAP_SHARED,
1331 if (env->me_txns == MAP_FAILED) {
1336 pthread_mutexattr_t mattr;
1338 pthread_mutexattr_init(&mattr);
1339 rc = pthread_mutexattr_setpshared(&mattr, PTHREAD_PROCESS_SHARED);
1343 pthread_mutex_init(&env->me_txns->mti_mutex, &mattr);
1344 pthread_mutex_init(&env->me_txns->mti_wmutex, &mattr);
1345 env->me_txns->mti_version = MDB_VERSION;
1346 env->me_txns->mti_magic = MDB_MAGIC;
1347 env->me_txns->mti_txnid = 0;
1348 env->me_txns->mti_numreaders = 0;
1349 env->me_txns->mti_me_toggle = 0;
1352 if (env->me_txns->mti_magic != MDB_MAGIC) {
1353 DPUTS("lock region has invalid magic");
1357 if (env->me_txns->mti_version != MDB_VERSION) {
1358 DPRINTF("lock region is version %u, expected version %u",
1359 env->me_txns->mti_version, MDB_VERSION);
1360 rc = MDB_VERSION_MISMATCH;
1363 if (errno != EACCES && errno != EAGAIN) {
1377 #define LOCKNAME "/lock.mdb"
1378 #define DATANAME "/data.mdb"
1380 mdb_env_open(MDB_env *env, const char *path, unsigned int flags, mode_t mode)
1382 int oflags, rc, len, excl;
1383 char *lpath, *dpath;
1386 lpath = malloc(len + sizeof(LOCKNAME) + len + sizeof(DATANAME));
1389 dpath = lpath + len + sizeof(LOCKNAME);
1390 sprintf(lpath, "%s" LOCKNAME, path);
1391 sprintf(dpath, "%s" DATANAME, path);
1393 rc = mdb_env_setup_locks(env, lpath, mode, &excl);
1397 if (F_ISSET(flags, MDB_RDONLY))
1400 oflags = O_RDWR | O_CREAT;
1402 if ((env->me_fd = open(dpath, oflags, mode)) == -1) {
1407 if ((rc = mdb_env_open2(env, flags)) == MDB_SUCCESS) {
1408 /* synchronous fd for meta writes */
1409 if (!(flags & (MDB_RDONLY|MDB_NOSYNC)))
1411 if ((env->me_mfd = open(dpath, oflags, mode)) == -1) {
1416 env->me_path = strdup(path);
1417 DPRINTF("opened dbenv %p", (void *) env);
1418 pthread_key_create(&env->me_txkey, mdb_env_reader_dest);
1420 mdb_env_share_locks(env);
1421 env->me_dbxs = calloc(env->me_maxdbs, sizeof(MDB_dbx));
1422 env->me_dbs[0] = calloc(env->me_maxdbs, sizeof(MDB_db));
1423 env->me_dbs[1] = calloc(env->me_maxdbs, sizeof(MDB_db));
1429 if (env->me_fd >= 0) {
1433 if (env->me_lfd >= 0) {
1443 mdb_env_close(MDB_env *env)
1448 free(env->me_dbs[1]);
1449 free(env->me_dbs[0]);
1453 pthread_key_delete(env->me_txkey);
1456 munmap(env->me_map, env->me_mapsize);
1461 pid_t pid = getpid();
1462 size_t size = (env->me_maxreaders-1) * sizeof(MDB_reader) + sizeof(MDB_txninfo);
1464 for (i=0; i<env->me_txns->mti_numreaders; i++)
1465 if (env->me_txns->mti_readers[i].mr_pid == pid)
1466 env->me_txns->mti_readers[i].mr_pid = 0;
1467 munmap(env->me_txns, size);
1473 /* Search for key within a leaf page, using binary search.
1474 * Returns the smallest entry larger or equal to the key.
1475 * If exactp is non-null, stores whether the found entry was an exact match
1476 * in *exactp (1 or 0).
1477 * If kip is non-null, stores the index of the found entry in *kip.
1478 * If no entry larger or equal to the key is found, returns NULL.
1481 mdb_search_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp, MDB_val *key,
1482 int *exactp, unsigned int *kip)
1490 DPRINTF("searching %u keys in %s page %lu",
1492 IS_LEAF(mp) ? "leaf" : "branch",
1495 assert(NUMKEYS(mp) > 0);
1497 memset(&nodekey, 0, sizeof(nodekey));
1499 low = IS_LEAF(mp) ? 0 : 1;
1500 high = NUMKEYS(mp) - 1;
1501 while (low <= high) {
1502 i = (low + high) >> 1;
1503 node = NODEPTR(mp, i);
1505 nodekey.mv_size = node->mn_ksize;
1506 nodekey.mv_data = NODEKEY(node);
1508 rc = mdb_cmp(txn, dbi, key, &nodekey);
1511 DPRINTF("found leaf index %u [%.*s], rc = %i",
1512 i, (int)nodekey.mv_size, (char *)nodekey.mv_data, rc);
1514 DPRINTF("found branch index %u [%.*s -> %lu], rc = %i",
1515 i, (int)node->mn_ksize, (char *)NODEKEY(node),
1526 if (rc > 0) { /* Found entry is less than the key. */
1527 i++; /* Skip to get the smallest entry larger than key. */
1528 if (i >= NUMKEYS(mp))
1529 /* There is no entry larger or equal to the key. */
1533 *exactp = (rc == 0);
1534 if (kip) /* Store the key index if requested. */
1537 return NODEPTR(mp, i);
1541 cursor_pop_page(MDB_cursor *cursor)
1545 if (cursor->mc_snum) {
1546 top = CURSOR_TOP(cursor);
1549 DPRINTF("popped page %lu off cursor %p", top->mp_page->mp_pgno, (void *) cursor);
1554 cursor_push_page(MDB_cursor *cursor, MDB_page *mp)
1558 DPRINTF("pushing page %lu on cursor %p", mp->mp_pgno, (void *) cursor);
1560 ppage = &cursor->mc_stack[cursor->mc_snum++];
1561 ppage->mp_page = mp;
1567 mdb_get_page(MDB_txn *txn, pgno_t pgno)
1572 if (!F_ISSET(txn->mt_flags, MDB_TXN_RDONLY) && txn->mt_u.dirty_list[0].mid) {
1577 x = mdb_midl2_search(txn->mt_u.dirty_list, &id);
1578 if (x <= txn->mt_u.dirty_list[0].mid && txn->mt_u.dirty_list[x].mid == pgno) {
1579 dp = txn->mt_u.dirty_list[x].mptr;
1585 if (pgno > txn->mt_env->me_meta->mm_last_pg)
1587 p = (MDB_page *)(txn->mt_env->me_map + txn->mt_env->me_psize * pgno);
1593 mdb_search_page_root(MDB_txn *txn, MDB_dbi dbi, MDB_val *key,
1594 MDB_cursor *cursor, int modify, MDB_pageparent *mpp)
1596 MDB_page *mp = mpp->mp_page;
1599 if (cursor && cursor_push_page(cursor, mp) == NULL)
1602 while (IS_BRANCH(mp)) {
1606 DPRINTF("branch page %lu has %u keys", mp->mp_pgno, NUMKEYS(mp));
1607 assert(NUMKEYS(mp) > 1);
1608 DPRINTF("found index 0 to page %lu", NODEPGNO(NODEPTR(mp, 0)));
1610 if (key == NULL) /* Initialize cursor to first page. */
1612 else if (key->mv_size > MAXKEYSIZE && key->mv_data == NULL) {
1613 /* cursor to last page */
1617 node = mdb_search_node(txn, dbi, mp, key, &exact, &i);
1619 i = NUMKEYS(mp) - 1;
1627 DPRINTF("following index %u for key %.*s",
1628 i, (int)key->mv_size, (char *)key->mv_data);
1629 assert(i < NUMKEYS(mp));
1630 node = NODEPTR(mp, i);
1633 CURSOR_TOP(cursor)->mp_ki = i;
1635 mpp->mp_parent = mp;
1636 if ((mp = mdb_get_page(txn, NODEPGNO(node))) == NULL)
1637 return MDB_PAGE_NOTFOUND;
1641 if (cursor && cursor_push_page(cursor, mp) == NULL)
1645 MDB_dhead *dh = ((MDB_dhead *)mp)-1;
1646 if ((rc = mdb_touch(txn, mpp)) != 0)
1648 dh = ((MDB_dhead *)mpp->mp_page)-1;
1649 dh->md_parent = mpp->mp_parent;
1650 dh->md_pi = mpp->mp_pi;
1657 DPRINTF("internal error, index points to a %02X page!?",
1659 return MDB_CORRUPTED;
1662 DPRINTF("found leaf page %lu for key %.*s", mp->mp_pgno,
1663 key ? (int)key->mv_size : 0, key ? (char *)key->mv_data : NULL);
1668 /* Search for the page a given key should be in.
1669 * Stores a pointer to the found page in *mpp.
1670 * If key is NULL, search for the lowest page (used by mdb_cursor_first).
1671 * If cursor is non-null, pushes parent pages on the cursor stack.
1672 * If modify is true, visited pages are updated with new page numbers.
1675 mdb_search_page(MDB_txn *txn, MDB_dbi dbi, MDB_val *key,
1676 MDB_cursor *cursor, int modify, MDB_pageparent *mpp)
1681 /* Choose which root page to start with. If a transaction is given
1682 * use the root page from the transaction, otherwise read the last
1683 * committed root page.
1685 if (F_ISSET(txn->mt_flags, MDB_TXN_ERROR)) {
1686 DPUTS("transaction has failed, must abort");
1689 root = txn->mt_dbs[dbi].md_root;
1691 if (root == P_INVALID) { /* Tree is empty. */
1692 DPUTS("tree is empty");
1693 return MDB_NOTFOUND;
1696 if ((mpp->mp_page = mdb_get_page(txn, root)) == NULL)
1697 return MDB_PAGE_NOTFOUND;
1699 DPRINTF("root page has flags 0x%X", mpp->mp_page->mp_flags);
1702 /* For sub-databases, update main root first */
1703 if (dbi > MAIN_DBI && !txn->mt_dbxs[dbi].md_dirty) {
1705 rc = mdb_search_page(txn, MAIN_DBI, &txn->mt_dbxs[dbi].md_name,
1709 txn->mt_dbxs[dbi].md_dirty = 1;
1711 if (!F_ISSET(mpp->mp_page->mp_flags, P_DIRTY)) {
1712 mpp->mp_parent = NULL;
1714 if ((rc = mdb_touch(txn, mpp)))
1716 txn->mt_dbs[dbi].md_root = mpp->mp_page->mp_pgno;
1720 return mdb_search_page_root(txn, dbi, key, cursor, modify, mpp);
1724 mdb_read_data(MDB_txn *txn, MDB_node *leaf, MDB_val *data)
1726 MDB_page *omp; /* overflow mpage */
1729 if (!F_ISSET(leaf->mn_flags, F_BIGDATA)) {
1730 data->mv_size = leaf->mn_dsize;
1731 data->mv_data = NODEDATA(leaf);
1735 /* Read overflow data.
1737 data->mv_size = leaf->mn_dsize;
1738 memcpy(&pgno, NODEDATA(leaf), sizeof(pgno));
1739 if ((omp = mdb_get_page(txn, pgno)) == NULL) {
1740 DPRINTF("read overflow page %lu failed", pgno);
1741 return MDB_PAGE_NOTFOUND;
1743 data->mv_data = METADATA(omp);
1749 mdb_get(MDB_txn *txn, MDB_dbi dbi,
1750 MDB_val *key, MDB_val *data)
1758 DPRINTF("===> get key [%.*s]", (int)key->mv_size, (char *)key->mv_data);
1760 if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
1763 if (key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
1767 if ((rc = mdb_search_page(txn, dbi, key, NULL, 0, &mpp)) != MDB_SUCCESS)
1770 leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, NULL);
1771 if (leaf && exact) {
1772 /* Return first duplicate data item */
1773 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
1776 mdb_xcursor_init0(txn, dbi, &mx);
1777 mdb_xcursor_init1(txn, dbi, &mx, leaf);
1778 rc = mdb_search_page(&mx.mx_txn, mx.mx_cursor.mc_dbi, NULL, NULL, 0, &mpp);
1779 if (rc != MDB_SUCCESS)
1781 leaf = NODEPTR(mpp.mp_page, 0);
1783 rc = mdb_read_data(txn, leaf, data);
1792 mdb_sibling(MDB_cursor *cursor, int move_right)
1799 if (cursor->mc_snum < 2) {
1800 return MDB_NOTFOUND; /* root has no siblings */
1802 parent = CURSOR_PARENT(cursor);
1804 DPRINTF("parent page is page %lu, index %u",
1805 parent->mp_page->mp_pgno, parent->mp_ki);
1807 cursor_pop_page(cursor);
1808 if (move_right ? (parent->mp_ki + 1 >= NUMKEYS(parent->mp_page))
1809 : (parent->mp_ki == 0)) {
1810 DPRINTF("no more keys left, moving to %s sibling",
1811 move_right ? "right" : "left");
1812 if ((rc = mdb_sibling(cursor, move_right)) != MDB_SUCCESS)
1814 parent = CURSOR_TOP(cursor);
1820 DPRINTF("just moving to %s index key %u",
1821 move_right ? "right" : "left", parent->mp_ki);
1823 assert(IS_BRANCH(parent->mp_page));
1825 indx = NODEPTR(parent->mp_page, parent->mp_ki);
1826 if ((mp = mdb_get_page(cursor->mc_txn, indx->mn_pgno)) == NULL)
1827 return MDB_PAGE_NOTFOUND;
1829 mp->parent = parent->mp_page;
1830 mp->parent_index = parent->mp_ki;
1833 cursor_push_page(cursor, mp);
1839 mdb_set_key(MDB_node *node, MDB_val *key)
1844 key->mv_size = node->mn_ksize;
1845 key->mv_data = NODEKEY(node);
1851 mdb_cursor_next(MDB_cursor *cursor, MDB_val *key, MDB_val *data, MDB_cursor_op op)
1858 if (cursor->mc_eof) {
1859 return MDB_NOTFOUND;
1862 assert(cursor->mc_initialized);
1864 top = CURSOR_TOP(cursor);
1867 if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
1868 leaf = NODEPTR(mp, top->mp_ki);
1869 if ((op == MDB_NEXT || op == MDB_NEXT_DUP) && F_ISSET(leaf->mn_flags, F_DUPDATA)) {
1870 rc = mdb_cursor_next(&cursor->mc_xcursor->mx_cursor, data, NULL, MDB_NEXT);
1871 if (op != MDB_NEXT || rc == MDB_SUCCESS)
1876 DPRINTF("cursor_next: top page is %lu in cursor %p", mp->mp_pgno, (void *) cursor);
1878 if (top->mp_ki + 1 >= NUMKEYS(mp)) {
1879 DPUTS("=====> move to next sibling page");
1880 if (mdb_sibling(cursor, 1) != MDB_SUCCESS) {
1882 return MDB_NOTFOUND;
1884 top = CURSOR_TOP(cursor);
1886 DPRINTF("next page is %lu, key index %u", mp->mp_pgno, top->mp_ki);
1890 DPRINTF("==> cursor points to page %lu with %u keys, key index %u",
1891 mp->mp_pgno, NUMKEYS(mp), top->mp_ki);
1893 assert(IS_LEAF(mp));
1894 leaf = NODEPTR(mp, top->mp_ki);
1896 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
1897 mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, leaf);
1900 if ((rc = mdb_read_data(cursor->mc_txn, leaf, data) != MDB_SUCCESS))
1903 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
1904 rc = mdb_cursor_first(&cursor->mc_xcursor->mx_cursor, data, NULL);
1905 if (rc != MDB_SUCCESS)
1910 return mdb_set_key(leaf, key);
1914 mdb_cursor_prev(MDB_cursor *cursor, MDB_val *key, MDB_val *data, MDB_cursor_op op)
1921 assert(cursor->mc_initialized);
1923 top = CURSOR_TOP(cursor);
1926 if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
1927 leaf = NODEPTR(mp, top->mp_ki);
1928 if ((op == MDB_PREV || op == MDB_PREV_DUP) && F_ISSET(leaf->mn_flags, F_DUPDATA)) {
1929 rc = mdb_cursor_prev(&cursor->mc_xcursor->mx_cursor, data, NULL, MDB_PREV);
1930 if (op != MDB_PREV || rc == MDB_SUCCESS)
1935 DPRINTF("cursor_prev: top page is %lu in cursor %p", mp->mp_pgno, (void *) cursor);
1937 if (top->mp_ki == 0) {
1938 DPUTS("=====> move to prev sibling page");
1939 if (mdb_sibling(cursor, 0) != MDB_SUCCESS) {
1940 return MDB_NOTFOUND;
1942 top = CURSOR_TOP(cursor);
1944 top->mp_ki = NUMKEYS(mp) - 1;
1945 DPRINTF("prev page is %lu, key index %u", mp->mp_pgno, top->mp_ki);
1951 DPRINTF("==> cursor points to page %lu with %u keys, key index %u",
1952 mp->mp_pgno, NUMKEYS(mp), top->mp_ki);
1954 assert(IS_LEAF(mp));
1955 leaf = NODEPTR(mp, top->mp_ki);
1957 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
1958 mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, leaf);
1961 if ((rc = mdb_read_data(cursor->mc_txn, leaf, data) != MDB_SUCCESS))
1964 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
1965 rc = mdb_cursor_last(&cursor->mc_xcursor->mx_cursor, data, NULL);
1966 if (rc != MDB_SUCCESS)
1971 return mdb_set_key(leaf, key);
1975 mdb_cursor_set(MDB_cursor *cursor, MDB_val *key, MDB_val *data,
1976 MDB_cursor_op op, int *exactp)
1985 assert(key->mv_size > 0);
1987 cursor->mc_snum = 0;
1989 rc = mdb_search_page(cursor->mc_txn, cursor->mc_dbi, key, cursor, 0, &mpp);
1990 if (rc != MDB_SUCCESS)
1992 assert(IS_LEAF(mpp.mp_page));
1994 top = CURSOR_TOP(cursor);
1995 leaf = mdb_search_node(cursor->mc_txn, cursor->mc_dbi, mpp.mp_page, key, exactp, &top->mp_ki);
1996 if (exactp != NULL && !*exactp) {
1997 /* MDB_SET specified and not an exact match. */
1998 return MDB_NOTFOUND;
2002 DPUTS("===> inexact leaf not found, goto sibling");
2003 if ((rc = mdb_sibling(cursor, 1)) != MDB_SUCCESS)
2004 return rc; /* no entries matched */
2005 top = CURSOR_TOP(cursor);
2007 mpp.mp_page = top->mp_page;
2008 assert(IS_LEAF(mpp.mp_page));
2009 leaf = NODEPTR(mpp.mp_page, 0);
2012 cursor->mc_initialized = 1;
2015 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
2016 mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, leaf);
2019 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
2020 if (op == MDB_SET || op == MDB_SET_RANGE) {
2021 rc = mdb_cursor_first(&cursor->mc_xcursor->mx_cursor, data, NULL);
2025 if (op == MDB_GET_BOTH) {
2030 op2 = MDB_SET_RANGE;
2032 rc = mdb_cursor_set(&cursor->mc_xcursor->mx_cursor, data, NULL, op2, ex2p);
2033 if (rc != MDB_SUCCESS)
2036 } else if (op == MDB_GET_BOTH || op == MDB_GET_BOTH_RANGE) {
2038 if ((rc = mdb_read_data(cursor->mc_txn, leaf, &d2)) != MDB_SUCCESS)
2040 rc = mdb_dcmp(cursor->mc_txn, cursor->mc_dbi, data, &d2);
2042 if (op == MDB_GET_BOTH || rc > 0)
2043 return MDB_NOTFOUND;
2047 if ((rc = mdb_read_data(cursor->mc_txn, leaf, data)) != MDB_SUCCESS)
2052 rc = mdb_set_key(leaf, key);
2053 if (rc == MDB_SUCCESS) {
2054 DPRINTF("==> cursor placed on key %.*s",
2055 (int)key->mv_size, (char *)key->mv_data);
2063 mdb_cursor_first(MDB_cursor *cursor, MDB_val *key, MDB_val *data)
2069 cursor->mc_snum = 0;
2071 rc = mdb_search_page(cursor->mc_txn, cursor->mc_dbi, NULL, cursor, 0, &mpp);
2072 if (rc != MDB_SUCCESS)
2074 assert(IS_LEAF(mpp.mp_page));
2076 leaf = NODEPTR(mpp.mp_page, 0);
2077 cursor->mc_initialized = 1;
2081 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
2082 mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, leaf);
2083 rc = mdb_cursor_first(&cursor->mc_xcursor->mx_cursor, data, NULL);
2087 if ((rc = mdb_read_data(cursor->mc_txn, leaf, data)) != MDB_SUCCESS)
2091 return mdb_set_key(leaf, key);
2095 mdb_cursor_last(MDB_cursor *cursor, MDB_val *key, MDB_val *data)
2103 cursor->mc_snum = 0;
2105 lkey.mv_size = MAXKEYSIZE+1;
2106 lkey.mv_data = NULL;
2108 rc = mdb_search_page(cursor->mc_txn, cursor->mc_dbi, &lkey, cursor, 0, &mpp);
2109 if (rc != MDB_SUCCESS)
2111 assert(IS_LEAF(mpp.mp_page));
2113 leaf = NODEPTR(mpp.mp_page, NUMKEYS(mpp.mp_page)-1);
2114 cursor->mc_initialized = 1;
2117 top = CURSOR_TOP(cursor);
2118 top->mp_ki = NUMKEYS(top->mp_page) - 1;
2121 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
2122 mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, leaf);
2123 rc = mdb_cursor_last(&cursor->mc_xcursor->mx_cursor, data, NULL);
2127 if ((rc = mdb_read_data(cursor->mc_txn, leaf, data)) != MDB_SUCCESS)
2132 return mdb_set_key(leaf, key);
2136 mdb_cursor_get(MDB_cursor *cursor, MDB_val *key, MDB_val *data,
2146 case MDB_GET_BOTH_RANGE:
2147 if (data == NULL || cursor->mc_xcursor == NULL) {
2154 if (key == NULL || key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
2156 } else if (op == MDB_SET_RANGE)
2157 rc = mdb_cursor_set(cursor, key, data, op, NULL);
2159 rc = mdb_cursor_set(cursor, key, data, op, &exact);
2163 case MDB_NEXT_NODUP:
2164 if (!cursor->mc_initialized)
2165 rc = mdb_cursor_first(cursor, key, data);
2167 rc = mdb_cursor_next(cursor, key, data, op);
2171 case MDB_PREV_NODUP:
2172 if (!cursor->mc_initialized || cursor->mc_eof)
2173 rc = mdb_cursor_last(cursor, key, data);
2175 rc = mdb_cursor_prev(cursor, key, data, op);
2178 rc = mdb_cursor_first(cursor, key, data);
2181 rc = mdb_cursor_last(cursor, key, data);
2184 DPRINTF("unhandled/unimplemented cursor operation %u", op);
2192 /* Allocate a page and initialize it
2195 mdb_new_page(MDB_txn *txn, MDB_dbi dbi, uint32_t flags, int num)
2199 if ((dp = mdb_alloc_page(txn, NULL, 0, num)) == NULL)
2201 DPRINTF("allocated new mpage %lu, page size %u",
2202 dp->p.mp_pgno, txn->mt_env->me_psize);
2203 dp->p.mp_flags = flags | P_DIRTY;
2204 dp->p.mp_lower = PAGEHDRSZ;
2205 dp->p.mp_upper = txn->mt_env->me_psize;
2207 if (IS_BRANCH(&dp->p))
2208 txn->mt_dbs[dbi].md_branch_pages++;
2209 else if (IS_LEAF(&dp->p))
2210 txn->mt_dbs[dbi].md_leaf_pages++;
2211 else if (IS_OVERFLOW(&dp->p)) {
2212 txn->mt_dbs[dbi].md_overflow_pages += num;
2213 dp->p.mp_pages = num;
2220 mdb_leaf_size(MDB_env *env, MDB_val *key, MDB_val *data)
2224 sz = LEAFSIZE(key, data);
2225 if (data->mv_size >= env->me_psize / MDB_MINKEYS) {
2226 /* put on overflow page */
2227 sz -= data->mv_size - sizeof(pgno_t);
2230 return sz + sizeof(indx_t);
2234 mdb_branch_size(MDB_env *env, MDB_val *key)
2239 if (sz >= env->me_psize / MDB_MINKEYS) {
2240 /* put on overflow page */
2241 /* not implemented */
2242 /* sz -= key->size - sizeof(pgno_t); */
2245 return sz + sizeof(indx_t);
2249 mdb_add_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp, indx_t indx,
2250 MDB_val *key, MDB_val *data, pgno_t pgno, uint8_t flags)
2253 size_t node_size = NODESIZE;
2256 MDB_dpage *ofp = NULL; /* overflow page */
2258 assert(mp->mp_upper >= mp->mp_lower);
2260 DPRINTF("add node [%.*s] to %s page %lu at index %i, key size %zu",
2261 key ? (int)key->mv_size : 0, key ? (char *)key->mv_data : NULL,
2262 IS_LEAF(mp) ? "leaf" : "branch",
2263 mp->mp_pgno, indx, key ? key->mv_size : 0);
2266 node_size += key->mv_size;
2270 if (F_ISSET(flags, F_BIGDATA)) {
2271 /* Data already on overflow page. */
2272 node_size += sizeof(pgno_t);
2273 } else if (data->mv_size >= txn->mt_env->me_psize / MDB_MINKEYS) {
2274 int ovpages = OVPAGES(data->mv_size, txn->mt_env->me_psize);
2275 /* Put data on overflow page. */
2276 DPRINTF("data size is %zu, put on overflow page",
2278 node_size += sizeof(pgno_t);
2279 if ((ofp = mdb_new_page(txn, dbi, P_OVERFLOW, ovpages)) == NULL)
2281 DPRINTF("allocated overflow page %lu", ofp->p.mp_pgno);
2284 node_size += data->mv_size;
2288 if (node_size + sizeof(indx_t) > SIZELEFT(mp)) {
2289 DPRINTF("not enough room in page %lu, got %u ptrs",
2290 mp->mp_pgno, NUMKEYS(mp));
2291 DPRINTF("upper - lower = %u - %u = %u", mp->mp_upper, mp->mp_lower,
2292 mp->mp_upper - mp->mp_lower);
2293 DPRINTF("node size = %zu", node_size);
2297 /* Move higher pointers up one slot. */
2298 for (i = NUMKEYS(mp); i > indx; i--)
2299 mp->mp_ptrs[i] = mp->mp_ptrs[i - 1];
2301 /* Adjust free space offsets. */
2302 ofs = mp->mp_upper - node_size;
2303 assert(ofs >= mp->mp_lower + sizeof(indx_t));
2304 mp->mp_ptrs[indx] = ofs;
2306 mp->mp_lower += sizeof(indx_t);
2308 /* Write the node data. */
2309 node = NODEPTR(mp, indx);
2310 node->mn_ksize = (key == NULL) ? 0 : key->mv_size;
2311 node->mn_flags = flags;
2313 node->mn_dsize = data->mv_size;
2315 node->mn_pgno = pgno;
2318 memcpy(NODEKEY(node), key->mv_data, key->mv_size);
2323 if (F_ISSET(flags, F_BIGDATA))
2324 memcpy(node->mn_data + key->mv_size, data->mv_data,
2327 memcpy(node->mn_data + key->mv_size, data->mv_data,
2330 memcpy(node->mn_data + key->mv_size, &ofp->p.mp_pgno,
2332 memcpy(METADATA(&ofp->p), data->mv_data, data->mv_size);
2340 mdb_del_node(MDB_page *mp, indx_t indx)
2343 indx_t i, j, numkeys, ptr;
2347 DPRINTF("delete node %u on %s page %lu", indx,
2348 IS_LEAF(mp) ? "leaf" : "branch", mp->mp_pgno);
2349 assert(indx < NUMKEYS(mp));
2351 node = NODEPTR(mp, indx);
2352 sz = NODESIZE + node->mn_ksize;
2354 if (F_ISSET(node->mn_flags, F_BIGDATA))
2355 sz += sizeof(pgno_t);
2357 sz += NODEDSZ(node);
2360 ptr = mp->mp_ptrs[indx];
2361 numkeys = NUMKEYS(mp);
2362 for (i = j = 0; i < numkeys; i++) {
2364 mp->mp_ptrs[j] = mp->mp_ptrs[i];
2365 if (mp->mp_ptrs[i] < ptr)
2366 mp->mp_ptrs[j] += sz;
2371 base = (char *)mp + mp->mp_upper;
2372 memmove(base + sz, base, ptr - mp->mp_upper);
2374 mp->mp_lower -= sizeof(indx_t);
2379 mdb_xcursor_init0(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx)
2384 mx->mx_txn.mt_dbxs = mx->mx_dbxs;
2385 mx->mx_txn.mt_dbs = mx->mx_dbs;
2386 mx->mx_dbxs[0] = txn->mt_dbxs[0];
2387 mx->mx_dbxs[1] = txn->mt_dbxs[1];
2389 mx->mx_dbxs[2] = txn->mt_dbxs[dbi];
2394 mx->mx_dbxs[dbn+1].md_parent = dbn;
2395 mx->mx_dbxs[dbn+1].md_cmp = mx->mx_dbxs[dbn].md_dcmp;
2396 mx->mx_dbxs[dbn+1].md_rel = mx->mx_dbxs[dbn].md_rel;
2397 mx->mx_dbxs[dbn+1].md_dirty = 0;
2398 mx->mx_txn.mt_numdbs = dbn+2;
2400 mx->mx_cursor.mc_snum = 0;
2401 mx->mx_cursor.mc_txn = &mx->mx_txn;
2402 mx->mx_cursor.mc_dbi = dbn+1;
2406 mdb_xcursor_init1(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx, MDB_node *node)
2408 MDB_db *db = NODEDATA(node);
2410 mx->mx_dbs[0] = txn->mt_dbs[0];
2411 mx->mx_dbs[1] = txn->mt_dbs[1];
2413 mx->mx_dbs[2] = txn->mt_dbs[dbi];
2418 mx->mx_dbs[dbn] = *db;
2419 mx->mx_dbxs[dbn].md_name.mv_data = NODEKEY(node);
2420 mx->mx_dbxs[dbn].md_name.mv_size = node->mn_ksize;
2421 mx->mx_txn.mt_next_pgno = txn->mt_next_pgno;
2422 mx->mx_txn.mt_oldest = txn->mt_oldest;
2423 mx->mx_txn.mt_u = txn->mt_u;
2427 mdb_xcursor_fini(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx)
2429 txn->mt_next_pgno = mx->mx_txn.mt_next_pgno;
2430 txn->mt_oldest = mx->mx_txn.mt_oldest;
2431 txn->mt_u = mx->mx_txn.mt_u;
2432 txn->mt_dbs[0] = mx->mx_dbs[0];
2433 txn->mt_dbs[1] = mx->mx_dbs[1];
2434 txn->mt_dbxs[0].md_dirty = mx->mx_dbxs[0].md_dirty;
2435 txn->mt_dbxs[1].md_dirty = mx->mx_dbxs[1].md_dirty;
2437 txn->mt_dbs[dbi] = mx->mx_dbs[2];
2438 txn->mt_dbxs[dbi].md_dirty = mx->mx_dbxs[2].md_dirty;
2443 mdb_cursor_open(MDB_txn *txn, MDB_dbi dbi, MDB_cursor **ret)
2446 size_t size = sizeof(MDB_cursor);
2448 if (txn == NULL || ret == NULL || !dbi || dbi >= txn->mt_numdbs)
2451 if (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT)
2452 size += sizeof(MDB_xcursor);
2454 if ((cursor = calloc(1, size)) != NULL) {
2455 cursor->mc_dbi = dbi;
2456 cursor->mc_txn = txn;
2457 if (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT) {
2458 MDB_xcursor *mx = (MDB_xcursor *)(cursor + 1);
2459 cursor->mc_xcursor = mx;
2460 mdb_xcursor_init0(txn, dbi, mx);
2471 /* Return the count of duplicate data items for the current key */
2473 mdb_cursor_count(MDB_cursor *mc, unsigned long *countp)
2478 if (mc == NULL || countp == NULL)
2481 if (!(mc->mc_txn->mt_dbs[mc->mc_dbi].md_flags & MDB_DUPSORT))
2484 top = CURSOR_TOP(mc);
2485 leaf = NODEPTR(top->mp_page, top->mp_ki);
2486 if (!F_ISSET(leaf->mn_flags, F_DUPDATA)) {
2489 if (!mc->mc_xcursor->mx_cursor.mc_initialized)
2492 *countp = mc->mc_xcursor->mx_txn.mt_dbs[mc->mc_xcursor->mx_cursor.mc_dbi].md_entries;
2498 mdb_cursor_close(MDB_cursor *cursor)
2500 if (cursor != NULL) {
2506 mdb_update_key(MDB_page *mp, indx_t indx, MDB_val *key)
2508 indx_t ptr, i, numkeys;
2514 node = NODEPTR(mp, indx);
2515 ptr = mp->mp_ptrs[indx];
2516 DPRINTF("update key %u (ofs %u) [%.*s] to [%.*s] on page %lu",
2518 (int)node->mn_ksize, (char *)NODEKEY(node),
2519 (int)key->mv_size, (char *)key->mv_data,
2522 delta = key->mv_size - node->mn_ksize;
2524 if (delta > 0 && SIZELEFT(mp) < delta) {
2525 DPRINTF("OUCH! Not enough room, delta = %d", delta);
2529 numkeys = NUMKEYS(mp);
2530 for (i = 0; i < numkeys; i++) {
2531 if (mp->mp_ptrs[i] <= ptr)
2532 mp->mp_ptrs[i] -= delta;
2535 base = (char *)mp + mp->mp_upper;
2536 len = ptr - mp->mp_upper + NODESIZE;
2537 memmove(base - delta, base, len);
2538 mp->mp_upper -= delta;
2540 node = NODEPTR(mp, indx);
2541 node->mn_ksize = key->mv_size;
2544 memcpy(NODEKEY(node), key->mv_data, key->mv_size);
2549 /* Move a node from src to dst.
2552 mdb_move_node(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *src, indx_t srcindx,
2553 MDB_pageparent *dst, indx_t dstindx)
2559 srcnode = NODEPTR(src->mp_page, srcindx);
2560 DPRINTF("moving %s node %u [%.*s] on page %lu to node %u on page %lu",
2561 IS_LEAF(src->mp_page) ? "leaf" : "branch",
2563 (int)srcnode->mn_ksize, (char *)NODEKEY(srcnode),
2564 src->mp_page->mp_pgno,
2565 dstindx, dst->mp_page->mp_pgno);
2567 /* Mark src and dst as dirty. */
2568 if ((rc = mdb_touch(txn, src)) ||
2569 (rc = mdb_touch(txn, dst)))
2572 /* Add the node to the destination page.
2574 key.mv_size = srcnode->mn_ksize;
2575 key.mv_data = NODEKEY(srcnode);
2576 data.mv_size = NODEDSZ(srcnode);
2577 data.mv_data = NODEDATA(srcnode);
2578 rc = mdb_add_node(txn, dbi, dst->mp_page, dstindx, &key, &data, NODEPGNO(srcnode),
2580 if (rc != MDB_SUCCESS)
2583 /* Delete the node from the source page.
2585 mdb_del_node(src->mp_page, srcindx);
2587 /* Update the parent separators.
2589 if (srcindx == 0 && src->mp_pi != 0) {
2590 DPRINTF("update separator for source page %lu to [%.*s]",
2591 src->mp_page->mp_pgno, (int)key.mv_size, (char *)key.mv_data);
2592 if ((rc = mdb_update_key(src->mp_parent, src->mp_pi,
2593 &key)) != MDB_SUCCESS)
2597 if (srcindx == 0 && IS_BRANCH(src->mp_page)) {
2599 nullkey.mv_size = 0;
2600 assert(mdb_update_key(src->mp_page, 0, &nullkey) == MDB_SUCCESS);
2603 if (dstindx == 0 && dst->mp_pi != 0) {
2604 DPRINTF("update separator for destination page %lu to [%.*s]",
2605 dst->mp_page->mp_pgno, (int)key.mv_size, (char *)key.mv_data);
2606 if ((rc = mdb_update_key(dst->mp_parent, dst->mp_pi,
2607 &key)) != MDB_SUCCESS)
2611 if (dstindx == 0 && IS_BRANCH(dst->mp_page)) {
2613 nullkey.mv_size = 0;
2614 assert(mdb_update_key(dst->mp_page, 0, &nullkey) == MDB_SUCCESS);
2621 mdb_merge(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *src, MDB_pageparent *dst)
2630 DPRINTF("merging page %lu and %lu", src->mp_page->mp_pgno, dst->mp_page->mp_pgno);
2632 assert(txn != NULL);
2633 assert(src->mp_parent); /* can't merge root page */
2634 assert(dst->mp_parent);
2636 /* Mark src and dst as dirty. */
2637 if ((rc = mdb_touch(txn, src)) ||
2638 (rc = mdb_touch(txn, dst)))
2641 /* Move all nodes from src to dst.
2643 for (i = 0; i < NUMKEYS(src->mp_page); i++) {
2644 srcnode = NODEPTR(src->mp_page, i);
2646 key.mv_size = srcnode->mn_ksize;
2647 key.mv_data = NODEKEY(srcnode);
2648 data.mv_size = NODEDSZ(srcnode);
2649 data.mv_data = NODEDATA(srcnode);
2650 rc = mdb_add_node(txn, dbi, dst->mp_page, NUMKEYS(dst->mp_page), &key,
2651 &data, NODEPGNO(srcnode), srcnode->mn_flags);
2652 if (rc != MDB_SUCCESS)
2656 DPRINTF("dst page %lu now has %u keys (%.1f%% filled)",
2657 dst->mp_page->mp_pgno, NUMKEYS(dst->mp_page), (float)PAGEFILL(txn->mt_env, dst->mp_page) / 10);
2659 /* Unlink the src page from parent.
2661 mdb_del_node(src->mp_parent, src->mp_pi);
2662 if (src->mp_pi == 0) {
2664 if ((rc = mdb_update_key(src->mp_parent, 0, &key)) != MDB_SUCCESS)
2668 if (IS_LEAF(src->mp_page))
2669 txn->mt_dbs[dbi].md_leaf_pages--;
2671 txn->mt_dbs[dbi].md_branch_pages--;
2673 mpp.mp_page = src->mp_parent;
2674 dh = (MDB_dhead *)src->mp_parent;
2676 mpp.mp_parent = dh->md_parent;
2677 mpp.mp_pi = dh->md_pi;
2679 return mdb_rebalance(txn, dbi, &mpp);
2682 #define FILL_THRESHOLD 250
2685 mdb_rebalance(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *mpp)
2690 indx_t si = 0, di = 0;
2692 assert(txn != NULL);
2693 assert(mpp != NULL);
2695 DPRINTF("rebalancing %s page %lu (has %u keys, %.1f%% full)",
2696 IS_LEAF(mpp->mp_page) ? "leaf" : "branch",
2697 mpp->mp_page->mp_pgno, NUMKEYS(mpp->mp_page), (float)PAGEFILL(txn->mt_env, mpp->mp_page) / 10);
2699 if (PAGEFILL(txn->mt_env, mpp->mp_page) >= FILL_THRESHOLD) {
2700 DPRINTF("no need to rebalance page %lu, above fill threshold",
2701 mpp->mp_page->mp_pgno);
2705 if (mpp->mp_parent == NULL) {
2706 if (NUMKEYS(mpp->mp_page) == 0) {
2707 DPUTS("tree is completely empty");
2708 txn->mt_dbs[dbi].md_root = P_INVALID;
2709 txn->mt_dbs[dbi].md_depth--;
2710 txn->mt_dbs[dbi].md_leaf_pages--;
2711 } else if (IS_BRANCH(mpp->mp_page) && NUMKEYS(mpp->mp_page) == 1) {
2712 DPUTS("collapsing root page!");
2713 txn->mt_dbs[dbi].md_root = NODEPGNO(NODEPTR(mpp->mp_page, 0));
2714 if ((root = mdb_get_page(txn, txn->mt_dbs[dbi].md_root)) == NULL)
2715 return MDB_PAGE_NOTFOUND;
2716 txn->mt_dbs[dbi].md_depth--;
2717 txn->mt_dbs[dbi].md_branch_pages--;
2719 DPUTS("root page doesn't need rebalancing");
2723 /* The parent (branch page) must have at least 2 pointers,
2724 * otherwise the tree is invalid.
2726 assert(NUMKEYS(mpp->mp_parent) > 1);
2728 /* Leaf page fill factor is below the threshold.
2729 * Try to move keys from left or right neighbor, or
2730 * merge with a neighbor page.
2735 if (mpp->mp_pi == 0) {
2736 /* We're the leftmost leaf in our parent.
2738 DPUTS("reading right neighbor");
2739 node = NODEPTR(mpp->mp_parent, mpp->mp_pi + 1);
2740 if ((npp.mp_page = mdb_get_page(txn, NODEPGNO(node))) == NULL)
2741 return MDB_PAGE_NOTFOUND;
2742 npp.mp_pi = mpp->mp_pi + 1;
2744 di = NUMKEYS(mpp->mp_page);
2746 /* There is at least one neighbor to the left.
2748 DPUTS("reading left neighbor");
2749 node = NODEPTR(mpp->mp_parent, mpp->mp_pi - 1);
2750 if ((npp.mp_page = mdb_get_page(txn, NODEPGNO(node))) == NULL)
2751 return MDB_PAGE_NOTFOUND;
2752 npp.mp_pi = mpp->mp_pi - 1;
2753 si = NUMKEYS(npp.mp_page) - 1;
2756 npp.mp_parent = mpp->mp_parent;
2758 DPRINTF("found neighbor page %lu (%u keys, %.1f%% full)",
2759 npp.mp_page->mp_pgno, NUMKEYS(npp.mp_page), (float)PAGEFILL(txn->mt_env, npp.mp_page) / 10);
2761 /* If the neighbor page is above threshold and has at least two
2762 * keys, move one key from it.
2764 * Otherwise we should try to merge them.
2766 if (PAGEFILL(txn->mt_env, npp.mp_page) >= FILL_THRESHOLD && NUMKEYS(npp.mp_page) >= 2)
2767 return mdb_move_node(txn, dbi, &npp, si, mpp, di);
2768 else { /* FIXME: if (has_enough_room()) */
2769 if (mpp->mp_pi == 0)
2770 return mdb_merge(txn, dbi, &npp, mpp);
2772 return mdb_merge(txn, dbi, mpp, &npp);
2777 mdb_del0(MDB_txn *txn, MDB_dbi dbi, unsigned int ki, MDB_pageparent *mpp, MDB_node *leaf)
2781 /* add overflow pages to free list */
2782 if (F_ISSET(leaf->mn_flags, F_BIGDATA)) {
2786 memcpy(&pg, NODEDATA(leaf), sizeof(pg));
2787 ovpages = OVPAGES(NODEDSZ(leaf), txn->mt_env->me_psize);
2788 for (i=0; i<ovpages; i++) {
2789 DPRINTF("freed ov page %lu", pg);
2790 mdb_midl_insert(txn->mt_free_pgs, pg);
2794 mdb_del_node(mpp->mp_page, ki);
2795 txn->mt_dbs[dbi].md_entries--;
2796 rc = mdb_rebalance(txn, dbi, mpp);
2797 if (rc != MDB_SUCCESS)
2798 txn->mt_flags |= MDB_TXN_ERROR;
2804 mdb_del(MDB_txn *txn, MDB_dbi dbi,
2805 MDB_val *key, MDB_val *data,
2813 DPRINTF("========> delete key %.*s", (int)key->mv_size, (char *)key->mv_data);
2815 assert(key != NULL);
2817 if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
2820 if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
2824 if (key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
2828 mpp.mp_parent = NULL;
2830 if ((rc = mdb_search_page(txn, dbi, key, NULL, 1, &mpp)) != MDB_SUCCESS)
2833 leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, &ki);
2834 if (leaf == NULL || !exact) {
2835 return MDB_NOTFOUND;
2838 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
2842 mdb_xcursor_init0(txn, dbi, &mx);
2843 mdb_xcursor_init1(txn, dbi, &mx, leaf);
2844 if (flags == MDB_DEL_DUP) {
2845 rc = mdb_del(&mx.mx_txn, mx.mx_cursor.mc_dbi, data, NULL, 0);
2846 mdb_xcursor_fini(txn, dbi, &mx);
2847 if (rc != MDB_SUCCESS)
2849 /* If sub-DB still has entries, we're done */
2850 if (mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi].md_root != P_INVALID) {
2851 memcpy(NODEDATA(leaf), &mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi],
2853 txn->mt_dbs[dbi].md_entries--;
2856 /* otherwise fall thru and delete the sub-DB */
2858 /* add all the child DB's pages to the free list */
2859 rc = mdb_search_page(&mx.mx_txn, mx.mx_cursor.mc_dbi,
2860 NULL, &mx.mx_cursor, 0, &mp2);
2861 if (rc == MDB_SUCCESS) {
2862 MDB_ppage *top, *parent;
2866 cursor_pop_page(&mx.mx_cursor);
2867 if (mx.mx_cursor.mc_snum) {
2868 top = CURSOR_TOP(&mx.mx_cursor);
2869 while (mx.mx_cursor.mc_snum > 1) {
2870 parent = CURSOR_PARENT(&mx.mx_cursor);
2871 for (i=0; i<NUMKEYS(top->mp_page); i++) {
2872 ni = NODEPTR(top->mp_page, i);
2873 mdb_midl_insert(txn->mt_free_pgs, ni->mn_pgno);
2876 if (parent->mp_ki >= NUMKEYS(parent->mp_page)) {
2877 cursor_pop_page(&mx.mx_cursor);
2880 ni = NODEPTR(parent->mp_page, parent->mp_ki);
2881 top->mp_page = mdb_get_page(&mx.mx_txn, ni->mn_pgno);
2885 mdb_midl_insert(txn->mt_free_pgs, mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi].md_root);
2890 if (data && (rc = mdb_read_data(txn, leaf, data)) != MDB_SUCCESS)
2893 return mdb_del0(txn, dbi, ki, &mpp, leaf);
2896 /* Split page <*mpp>, and insert <key,(data|newpgno)> in either left or
2897 * right sibling, at index <*newindxp> (as if unsplit). Updates *mpp and
2898 * *newindxp with the actual values after split, ie if *mpp and *newindxp
2899 * refer to a node in the new right sibling page.
2902 mdb_split(MDB_txn *txn, MDB_dbi dbi, MDB_page **mpp, unsigned int *newindxp,
2903 MDB_val *newkey, MDB_val *newdata, pgno_t newpgno)
2906 int rc = MDB_SUCCESS, ins_new = 0;
2909 unsigned int i, j, split_indx;
2911 MDB_val sepkey, rkey, rdata;
2913 MDB_dpage *mdp, *rdp, *pdp;
2916 assert(txn != NULL);
2918 dh = ((MDB_dhead *)*mpp) - 1;
2919 mdp = (MDB_dpage *)dh;
2920 newindx = *newindxp;
2922 DPRINTF("-----> splitting %s page %lu and adding [%.*s] at index %i",
2923 IS_LEAF(&mdp->p) ? "leaf" : "branch", mdp->p.mp_pgno,
2924 (int)newkey->mv_size, (char *)newkey->mv_data, *newindxp);
2926 if (mdp->h.md_parent == NULL) {
2927 if ((pdp = mdb_new_page(txn, dbi, P_BRANCH, 1)) == NULL)
2930 mdp->h.md_parent = &pdp->p;
2931 txn->mt_dbs[dbi].md_root = pdp->p.mp_pgno;
2932 DPRINTF("root split! new root = %lu", pdp->p.mp_pgno);
2933 txn->mt_dbs[dbi].md_depth++;
2935 /* Add left (implicit) pointer. */
2936 if ((rc = mdb_add_node(txn, dbi, &pdp->p, 0, NULL, NULL,
2937 mdp->p.mp_pgno, 0)) != MDB_SUCCESS)
2940 DPRINTF("parent branch page is %lu", mdp->h.md_parent->mp_pgno);
2943 /* Create a right sibling. */
2944 if ((rdp = mdb_new_page(txn, dbi, mdp->p.mp_flags, 1)) == NULL)
2946 rdp->h.md_parent = mdp->h.md_parent;
2947 rdp->h.md_pi = mdp->h.md_pi + 1;
2948 DPRINTF("new right sibling: page %lu", rdp->p.mp_pgno);
2950 /* Move half of the keys to the right sibling. */
2951 if ((copy = malloc(txn->mt_env->me_psize)) == NULL)
2953 memcpy(copy, &mdp->p, txn->mt_env->me_psize);
2954 memset(&mdp->p.mp_ptrs, 0, txn->mt_env->me_psize - PAGEHDRSZ);
2955 mdp->p.mp_lower = PAGEHDRSZ;
2956 mdp->p.mp_upper = txn->mt_env->me_psize;
2958 split_indx = NUMKEYS(copy) / 2 + 1;
2960 /* First find the separating key between the split pages.
2962 memset(&sepkey, 0, sizeof(sepkey));
2963 if (newindx == split_indx) {
2964 sepkey.mv_size = newkey->mv_size;
2965 sepkey.mv_data = newkey->mv_data;
2967 node = NODEPTR(copy, split_indx);
2968 sepkey.mv_size = node->mn_ksize;
2969 sepkey.mv_data = NODEKEY(node);
2972 DPRINTF("separator is [%.*s]", (int)sepkey.mv_size, (char *)sepkey.mv_data);
2974 /* Copy separator key to the parent.
2976 if (SIZELEFT(rdp->h.md_parent) < mdb_branch_size(txn->mt_env, &sepkey)) {
2977 rc = mdb_split(txn, dbi, &rdp->h.md_parent, &rdp->h.md_pi,
2978 &sepkey, NULL, rdp->p.mp_pgno);
2980 /* Right page might now have changed parent.
2981 * Check if left page also changed parent.
2983 if (rdp->h.md_parent != mdp->h.md_parent &&
2984 mdp->h.md_pi >= NUMKEYS(mdp->h.md_parent)) {
2985 mdp->h.md_parent = rdp->h.md_parent;
2986 mdp->h.md_pi = rdp->h.md_pi - 1;
2989 rc = mdb_add_node(txn, dbi, rdp->h.md_parent, rdp->h.md_pi,
2990 &sepkey, NULL, rdp->p.mp_pgno, 0);
2992 if (rc != MDB_SUCCESS) {
2997 for (i = j = 0; i <= NUMKEYS(copy); j++) {
2998 if (i < split_indx) {
2999 /* Re-insert in left sibling. */
3002 /* Insert in right sibling. */
3003 if (i == split_indx)
3004 /* Reset insert index for right sibling. */
3005 j = (i == newindx && ins_new);
3009 if (i == newindx && !ins_new) {
3010 /* Insert the original entry that caused the split. */
3011 rkey.mv_data = newkey->mv_data;
3012 rkey.mv_size = newkey->mv_size;
3013 if (IS_LEAF(&mdp->p)) {
3014 rdata.mv_data = newdata->mv_data;
3015 rdata.mv_size = newdata->mv_size;
3022 /* Update page and index for the new key. */
3025 } else if (i == NUMKEYS(copy)) {
3028 node = NODEPTR(copy, i);
3029 rkey.mv_data = NODEKEY(node);
3030 rkey.mv_size = node->mn_ksize;
3031 if (IS_LEAF(&mdp->p)) {
3032 rdata.mv_data = NODEDATA(node);
3033 rdata.mv_size = node->mn_dsize;
3035 pgno = node->mn_pgno;
3036 flags = node->mn_flags;
3041 if (!IS_LEAF(&mdp->p) && j == 0) {
3042 /* First branch index doesn't need key data. */
3046 rc = mdb_add_node(txn, dbi, &pdp->p, j, &rkey, &rdata, pgno,flags);
3054 mdb_put0(MDB_txn *txn, MDB_dbi dbi,
3055 MDB_val *key, MDB_val *data, unsigned int flags)
3057 int rc = MDB_SUCCESS, exact;
3061 MDB_val xdata, *rdata, dkey;
3063 char dbuf[PAGESIZE];
3066 DPRINTF("==> put key %.*s, size %zu, data size %zu",
3067 (int)key->mv_size, (char *)key->mv_data, key->mv_size, data->mv_size);
3070 mpp.mp_parent = NULL;
3072 rc = mdb_search_page(txn, dbi, key, NULL, 1, &mpp);
3073 if (rc == MDB_SUCCESS) {
3074 leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, &ki);
3075 if (leaf && exact) {
3076 if (flags == MDB_NOOVERWRITE) {
3077 DPRINTF("duplicate key %.*s",
3078 (int)key->mv_size, (char *)key->mv_data);
3079 return MDB_KEYEXIST;
3081 if (F_ISSET(txn->mt_dbs[dbi].md_flags, MDB_DUPSORT)) {
3082 /* Was a single item before, must convert now */
3083 if (!F_ISSET(leaf->mn_flags, F_DUPDATA)) {
3084 dkey.mv_size = NODEDSZ(leaf);
3085 memcpy(dbuf, NODEDATA(leaf), dkey.mv_size);
3086 memset(&dummy, 0, sizeof(dummy));
3087 dummy.md_root = P_INVALID;
3088 if (dkey.mv_size == sizeof(MDB_db)) {
3089 memcpy(NODEDATA(leaf), &dummy, sizeof(dummy));
3092 mdb_del_node(mpp.mp_page, ki);
3095 xdata.mv_size = sizeof(MDB_db);
3096 xdata.mv_data = &dummy;
3101 /* same size, just replace it */
3102 if (NODEDSZ(leaf) == data->mv_size) {
3103 memcpy(NODEDATA(leaf), data->mv_data, data->mv_size);
3106 mdb_del_node(mpp.mp_page, ki);
3108 if (leaf == NULL) { /* append if not found */
3109 ki = NUMKEYS(mpp.mp_page);
3110 DPRINTF("appending key at index %i", ki);
3112 } else if (rc == MDB_NOTFOUND) {
3114 /* new file, just write a root leaf page */
3115 DPUTS("allocating new root leaf page");
3116 if ((dp = mdb_new_page(txn, dbi, P_LEAF, 1)) == NULL) {
3119 mpp.mp_page = &dp->p;
3120 txn->mt_dbs[dbi].md_root = mpp.mp_page->mp_pgno;
3121 txn->mt_dbs[dbi].md_depth++;
3122 txn->mt_dbxs[dbi].md_dirty = 1;
3128 assert(IS_LEAF(mpp.mp_page));
3129 DPRINTF("there are %u keys, should insert new key at index %i",
3130 NUMKEYS(mpp.mp_page), ki);
3135 if (SIZELEFT(mpp.mp_page) < mdb_leaf_size(txn->mt_env, key, rdata)) {
3136 rc = mdb_split(txn, dbi, &mpp.mp_page, &ki, key, rdata, P_INVALID);
3138 /* There is room already in this leaf page. */
3139 rc = mdb_add_node(txn, dbi, mpp.mp_page, ki, key, rdata, 0, 0);
3142 if (rc != MDB_SUCCESS)
3143 txn->mt_flags |= MDB_TXN_ERROR;
3145 /* Remember if we just added a subdatabase */
3146 if (flags & F_SUBDATA) {
3147 leaf = NODEPTR(mpp.mp_page, ki);
3148 leaf->mn_flags |= F_SUBDATA;
3151 /* Now store the actual data in the child DB. Note that we're
3152 * storing the user data in the keys field, so there are strict
3153 * size limits on dupdata. The actual data fields of the child
3154 * DB are all zero size.
3159 leaf = NODEPTR(mpp.mp_page, ki);
3161 mdb_xcursor_init0(txn, dbi, &mx);
3162 mdb_xcursor_init1(txn, dbi, &mx, leaf);
3165 if (flags == MDB_NODUPDATA)
3166 flags = MDB_NOOVERWRITE;
3167 /* converted, write the original data first */
3169 dkey.mv_data = dbuf;
3170 rc = mdb_put0(&mx.mx_txn, mx.mx_cursor.mc_dbi, &dkey, &xdata, flags);
3172 leaf->mn_flags |= F_DUPDATA;
3174 rc = mdb_put0(&mx.mx_txn, mx.mx_cursor.mc_dbi, data, &xdata, flags);
3175 mdb_xcursor_fini(txn, dbi, &mx);
3176 memcpy(NODEDATA(leaf), &mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi],
3179 txn->mt_dbs[dbi].md_entries++;
3187 mdb_put(MDB_txn *txn, MDB_dbi dbi,
3188 MDB_val *key, MDB_val *data, unsigned int flags)
3190 assert(key != NULL);
3191 assert(data != NULL);
3193 if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
3196 if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
3200 if (key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
3204 if ((flags & (MDB_NOOVERWRITE|MDB_NODUPDATA)) != flags)
3207 return mdb_put0(txn, dbi, key, data, flags);
3211 mdb_env_set_flags(MDB_env *env, unsigned int flag, int onoff)
3213 #define CHANGEABLE (MDB_NOSYNC)
3214 if ((flag & CHANGEABLE) != flag)
3217 env->me_flags |= flag;
3219 env->me_flags &= ~flag;
3224 mdb_env_get_flags(MDB_env *env, unsigned int *arg)
3229 *arg = env->me_flags;
3234 mdb_env_get_path(MDB_env *env, const char **arg)
3239 *arg = env->me_path;
3244 mdb_stat0(MDB_env *env, MDB_db *db, MDB_stat *arg)
3246 arg->ms_psize = env->me_psize;
3247 arg->ms_depth = db->md_depth;
3248 arg->ms_branch_pages = db->md_branch_pages;
3249 arg->ms_leaf_pages = db->md_leaf_pages;
3250 arg->ms_overflow_pages = db->md_overflow_pages;
3251 arg->ms_entries = db->md_entries;
3256 mdb_env_stat(MDB_env *env, MDB_stat *arg)
3258 if (env == NULL || arg == NULL)
3261 return mdb_stat0(env, &env->me_meta->mm_dbs[MAIN_DBI], arg);
3264 int mdb_open(MDB_txn *txn, const char *name, unsigned int flags, MDB_dbi *dbi)
3274 if (flags & (MDB_DUPSORT|MDB_REVERSEKEY|MDB_INTEGERKEY))
3275 txn->mt_dbs[MAIN_DBI].md_flags |= (flags & (MDB_DUPSORT|MDB_REVERSEKEY|MDB_INTEGERKEY));
3279 /* Is the DB already open? */
3281 for (i=2; i<txn->mt_numdbs; i++) {
3282 if (len == txn->mt_dbxs[i].md_name.mv_size &&
3283 !strncmp(name, txn->mt_dbxs[i].md_name.mv_data, len)) {
3289 if (txn->mt_numdbs >= txn->mt_env->me_maxdbs - 1)
3292 /* Find the DB info */
3294 key.mv_data = (void *)name;
3295 rc = mdb_get(txn, MAIN_DBI, &key, &data);
3297 /* Create if requested */
3298 if (rc == MDB_NOTFOUND && (flags & MDB_CREATE)) {
3300 data.mv_size = sizeof(MDB_db);
3301 data.mv_data = &dummy;
3302 memset(&dummy, 0, sizeof(dummy));
3303 dummy.md_root = P_INVALID;
3304 dummy.md_flags = flags & 0xffff;
3305 rc = mdb_put0(txn, MAIN_DBI, &key, &data, F_SUBDATA);
3309 /* OK, got info, add to table */
3310 if (rc == MDB_SUCCESS) {
3311 txn->mt_dbxs[txn->mt_numdbs].md_name.mv_data = strdup(name);
3312 txn->mt_dbxs[txn->mt_numdbs].md_name.mv_size = len;
3313 txn->mt_dbxs[txn->mt_numdbs].md_cmp = NULL;
3314 txn->mt_dbxs[txn->mt_numdbs].md_dcmp = NULL;
3315 txn->mt_dbxs[txn->mt_numdbs].md_rel = NULL;
3316 txn->mt_dbxs[txn->mt_numdbs].md_parent = MAIN_DBI;
3317 txn->mt_dbxs[txn->mt_numdbs].md_dirty = dirty;
3318 memcpy(&txn->mt_dbs[txn->mt_numdbs], data.mv_data, sizeof(MDB_db));
3319 *dbi = txn->mt_numdbs;
3320 txn->mt_env->me_dbs[0][txn->mt_numdbs] = txn->mt_dbs[txn->mt_numdbs];
3321 txn->mt_env->me_dbs[1][txn->mt_numdbs] = txn->mt_dbs[txn->mt_numdbs];
3328 int mdb_stat(MDB_txn *txn, MDB_dbi dbi, MDB_stat *arg)
3330 if (txn == NULL || arg == NULL || dbi >= txn->mt_numdbs)
3333 return mdb_stat0(txn->mt_env, &txn->mt_dbs[dbi], arg);
3336 void mdb_close(MDB_txn *txn, MDB_dbi dbi)
3339 if (dbi <= MAIN_DBI || dbi >= txn->mt_numdbs)
3341 ptr = txn->mt_dbxs[dbi].md_name.mv_data;
3342 txn->mt_dbxs[dbi].md_name.mv_data = NULL;
3343 txn->mt_dbxs[dbi].md_name.mv_size = 0;
3347 int mdb_set_compare(MDB_txn *txn, MDB_dbi dbi, MDB_cmp_func *cmp)
3349 if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
3352 txn->mt_dbxs[dbi].md_cmp = cmp;
3356 int mdb_set_dupsort(MDB_txn *txn, MDB_dbi dbi, MDB_cmp_func *cmp)
3358 if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
3361 txn->mt_dbxs[dbi].md_dcmp = cmp;
3365 int mdb_set_relfunc(MDB_txn *txn, MDB_dbi dbi, MDB_rel_func *rel)
3367 if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
3370 txn->mt_dbxs[dbi].md_rel = rel;