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
58 /* Note: If O_DSYNC is undefined but exists in /usr/include,
59 * preferably set some compiler flag to get the definition.
60 * Otherwise compile with the less efficient -DMDB_DSYNC=O_SYNC.
63 # define MDB_DSYNC O_DSYNC
70 #if !(__STDC_VERSION__ >= 199901L || defined(__GNUC__))
71 # define DPRINTF (void) /* Vararg macros may be unsupported */
73 # define DPRINTF(fmt, ...) /* Requires 2 or more args */ \
74 fprintf(stderr, "%s:%d: " fmt "\n", __func__, __LINE__, __VA_ARGS__)
76 # define DPRINTF(fmt, ...) ((void) 0)
78 #define DPUTS(arg) DPRINTF("%s", arg)
82 #define MDB_MAGIC 0xBEEFC0DE
84 #define MAXKEYSIZE 511
86 #define P_INVALID (~0UL)
88 #define F_ISSET(w, f) (((w) & (f)) == (f))
90 typedef uint16_t indx_t;
92 #define DEFAULT_READERS 126
93 #define DEFAULT_MAPSIZE 1048576
95 /* Lock descriptor stuff */
97 #define CACHELINE 64 /* most CPUs. Itanium uses 128 */
100 typedef struct MDB_rxbody {
106 typedef struct MDB_reader {
109 #define mr_txnid mru.mrx.mrb_txnid
110 #define mr_pid mru.mrx.mrb_pid
111 #define mr_tid mru.mrx.mrb_tid
112 /* cache line alignment */
113 char pad[(sizeof(MDB_rxbody)+CACHELINE-1) & ~(CACHELINE-1)];
117 typedef struct MDB_txbody {
119 uint32_t mtb_version;
120 pthread_mutex_t mtb_mutex;
122 uint32_t mtb_numreaders;
123 uint32_t mtb_me_toggle;
126 typedef struct MDB_txninfo {
129 #define mti_magic mt1.mtb.mtb_magic
130 #define mti_version mt1.mtb.mtb_version
131 #define mti_mutex mt1.mtb.mtb_mutex
132 #define mti_txnid mt1.mtb.mtb_txnid
133 #define mti_numreaders mt1.mtb.mtb_numreaders
134 #define mti_me_toggle mt1.mtb.mtb_me_toggle
135 char pad[(sizeof(MDB_txbody)+CACHELINE-1) & ~(CACHELINE-1)];
138 pthread_mutex_t mt2_wmutex;
139 #define mti_wmutex mt2.mt2_wmutex
140 char pad[(sizeof(pthread_mutex_t)+CACHELINE-1) & ~(CACHELINE-1)];
142 MDB_reader mti_readers[1];
145 /* Common header for all page types. Overflow pages
146 * occupy a number of contiguous pages with no
147 * headers on any page after the first.
149 typedef struct MDB_page { /* represents a page of storage */
150 #define mp_pgno mp_p.p_pgno
152 pgno_t p_pgno; /* page number */
153 void * p_align; /* for IL32P64 */
155 #define P_BRANCH 0x01 /* branch page */
156 #define P_LEAF 0x02 /* leaf page */
157 #define P_OVERFLOW 0x04 /* overflow page */
158 #define P_META 0x08 /* meta page */
159 #define P_DIRTY 0x10 /* dirty page */
160 #define P_LEAF2 0x20 /* DB with small, fixed size keys and no data */
162 #define mp_lower mp_pb.pb.pb_lower
163 #define mp_upper mp_pb.pb.pb_upper
164 #define mp_pages mp_pb.pb_pages
167 indx_t pb_lower; /* lower bound of free space */
168 indx_t pb_upper; /* upper bound of free space */
170 uint32_t pb_pages; /* number of overflow pages */
172 indx_t pb_ksize; /* on a LEAF2 page */
176 indx_t mp_ptrs[1]; /* dynamic size */
179 #define PAGEHDRSZ ((unsigned) offsetof(MDB_page, mp_ptrs))
181 #define NUMKEYS(p) (((p)->mp_lower - PAGEHDRSZ) >> 1)
182 #define SIZELEFT(p) (indx_t)((p)->mp_upper - (p)->mp_lower)
183 #define PAGEFILL(env, p) (1000L * ((env)->me_psize - PAGEHDRSZ - SIZELEFT(p)) / \
184 ((env)->me_psize - PAGEHDRSZ))
185 #define IS_LEAF(p) F_ISSET((p)->mp_flags, P_LEAF)
186 #define IS_BRANCH(p) F_ISSET((p)->mp_flags, P_BRANCH)
187 #define IS_OVERFLOW(p) F_ISSET((p)->mp_flags, P_OVERFLOW)
189 #define OVPAGES(size, psize) ((PAGEHDRSZ-1 + (size)) / (psize) + 1)
191 typedef struct MDB_db {
195 ULONG md_branch_pages;
197 ULONG md_overflow_pages;
205 typedef struct MDB_meta { /* meta (footer) page content */
208 void *mm_address; /* address for fixed mapping */
209 size_t mm_mapsize; /* size of mmap region */
210 MDB_db mm_dbs[2]; /* first is free space, 2nd is main db */
211 #define mm_psize mm_dbs[0].md_pad
212 #define mm_flags mm_dbs[0].md_flags
213 pgno_t mm_last_pg; /* last used page in file */
214 ULONG mm_txnid; /* txnid that committed this page */
217 typedef struct MDB_dhead { /* a dirty page */
219 unsigned md_pi; /* parent index */
223 typedef struct MDB_dpage {
228 typedef struct MDB_oldpages {
229 struct MDB_oldpages *mo_next;
231 pgno_t mo_pages[1]; /* dynamic */
234 typedef struct MDB_pageparent {
240 static MDB_dpage *mdb_alloc_page(MDB_txn *txn, MDB_page *parent, unsigned int parent_idx, int num);
241 static int mdb_touch(MDB_txn *txn, MDB_pageparent *mp);
243 typedef struct MDB_ppage { /* ordered list of pages */
245 unsigned int mp_ki; /* cursor index on page */
248 #define CURSOR_TOP(c) (&(c)->mc_stack[(c)->mc_snum-1])
249 #define CURSOR_PARENT(c) (&(c)->mc_stack[(c)->mc_snum-2])
255 MDB_ppage mc_stack[32]; /* stack of parent pages */
256 unsigned int mc_snum; /* number of pushed pages */
258 short mc_initialized; /* 1 if initialized */
259 short mc_eof; /* 1 if end is reached */
260 struct MDB_xcursor *mc_xcursor;
263 #define METADATA(p) ((void *)((char *)(p) + PAGEHDRSZ))
265 typedef struct MDB_node {
266 #define mn_pgno mn_p.np_pgno
267 #define mn_dsize mn_p.np_dsize
269 pgno_t np_pgno; /* child page number */
270 uint32_t np_dsize; /* leaf data size */
272 unsigned int mn_flags:4;
273 unsigned int mn_ksize:12; /* key size */
274 #define F_BIGDATA 0x01 /* data put on overflow page */
275 #define F_SUBDATA 0x02 /* data is a sub-database */
276 #define F_DUPDATA 0x04 /* data has duplicates */
280 typedef struct MDB_dbx {
282 MDB_cmp_func *md_cmp; /* user compare function */
283 MDB_cmp_func *md_dcmp; /* user dupsort function */
284 MDB_rel_func *md_rel; /* user relocate function */
286 unsigned int md_dirty;
290 pgno_t mt_next_pgno; /* next unallocated page */
294 pgno_t *mt_free_pgs; /* this is an IDL */
296 MIDL2 *dirty_list; /* modified pages */
299 MDB_dbx *mt_dbxs; /* array */
301 unsigned int mt_numdbs;
303 #define MDB_TXN_RDONLY 0x01 /* read-only transaction */
304 #define MDB_TXN_ERROR 0x02 /* an error has occurred */
305 #define MDB_TXN_METOGGLE 0x04 /* used meta page 1 */
306 unsigned int mt_flags;
309 /* Context for sorted-dup records */
310 typedef struct MDB_xcursor {
311 MDB_cursor mx_cursor;
320 int me_mfd; /* just for writing the meta pages */
321 #define MDB_FATAL_ERROR 0x80000000U
323 uint32_t me_extrapad; /* unused for now */
324 unsigned int me_maxreaders;
325 unsigned int me_numdbs;
326 unsigned int me_maxdbs;
329 MDB_txninfo *me_txns;
330 MDB_meta *me_metas[2];
332 MDB_txn *me_txn; /* current write transaction */
334 off_t me_size; /* current file size */
335 pgno_t me_maxpg; /* me_mapsize / me_psize */
336 unsigned int me_psize;
337 unsigned int me_db_toggle;
338 MDB_dbx *me_dbxs; /* array */
340 MDB_oldpages *me_pghead;
341 pthread_key_t me_txkey; /* thread-key for readers */
342 pgno_t me_free_pgs[MDB_IDL_UM_SIZE];
343 MIDL2 me_dirty_list[MDB_IDL_DB_SIZE];
346 #define NODESIZE offsetof(MDB_node, mn_data)
348 #define INDXSIZE(k) (NODESIZE + ((k) == NULL ? 0 : (k)->mv_size))
349 #define LEAFSIZE(k, d) (NODESIZE + (k)->mv_size + (d)->mv_size)
350 #define NODEPTR(p, i) ((MDB_node *)((char *)(p) + (p)->mp_ptrs[i]))
351 #define NODEKEY(node) (void *)((node)->mn_data)
352 #define NODEDATA(node) (void *)((char *)(node)->mn_data + (node)->mn_ksize)
353 #define NODEPGNO(node) ((node)->mn_pgno)
354 #define NODEDSZ(node) ((node)->mn_dsize)
356 #define MDB_COMMIT_PAGES 64 /* max number of pages to write in one commit */
358 static int mdb_search_page_root(MDB_txn *txn,
359 MDB_dbi dbi, MDB_val *key,
360 MDB_cursor *cursor, int modify,
361 MDB_pageparent *mpp);
362 static int mdb_search_page(MDB_txn *txn,
363 MDB_dbi dbi, MDB_val *key,
364 MDB_cursor *cursor, int modify,
365 MDB_pageparent *mpp);
367 static int mdb_env_read_header(MDB_env *env, MDB_meta *meta);
368 static int mdb_env_read_meta(MDB_env *env, int *which);
369 static int mdb_env_write_meta(MDB_txn *txn);
370 static MDB_page *mdb_get_page(MDB_txn *txn, pgno_t pgno);
372 static MDB_node *mdb_search_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp,
373 MDB_val *key, int *exactp, unsigned int *kip);
374 static int mdb_add_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp,
375 indx_t indx, MDB_val *key, MDB_val *data,
376 pgno_t pgno, uint8_t flags);
377 static void mdb_del_node(MDB_page *mp, indx_t indx);
378 static int mdb_del0(MDB_txn *txn, MDB_dbi dbi, unsigned int ki,
379 MDB_pageparent *mpp, MDB_node *leaf);
380 static int mdb_put0(MDB_txn *txn, MDB_dbi dbi,
381 MDB_val *key, MDB_val *data, unsigned int flags);
382 static int mdb_read_data(MDB_txn *txn, MDB_node *leaf, MDB_val *data);
384 static int mdb_rebalance(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *mp);
385 static int mdb_update_key(MDB_page *mp, indx_t indx, MDB_val *key);
386 static int mdb_move_node(MDB_txn *txn, MDB_dbi dbi,
387 MDB_pageparent *src, indx_t srcindx,
388 MDB_pageparent *dst, indx_t dstindx);
389 static int mdb_merge(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *src,
390 MDB_pageparent *dst);
391 static int mdb_split(MDB_txn *txn, MDB_dbi dbi, MDB_page **mpp,
392 unsigned int *newindxp, MDB_val *newkey,
393 MDB_val *newdata, pgno_t newpgno);
394 static MDB_dpage *mdb_new_page(MDB_txn *txn, MDB_dbi dbi, uint32_t flags, int num);
396 static void cursor_pop_page(MDB_cursor *cursor);
397 static MDB_ppage *cursor_push_page(MDB_cursor *cursor,
400 static int mdb_set_key(MDB_node *node, MDB_val *key);
401 static int mdb_sibling(MDB_cursor *cursor, int move_right);
402 static int mdb_cursor_next(MDB_cursor *cursor,
403 MDB_val *key, MDB_val *data, MDB_cursor_op op);
404 static int mdb_cursor_prev(MDB_cursor *cursor,
405 MDB_val *key, MDB_val *data, MDB_cursor_op op);
406 static int mdb_cursor_set(MDB_cursor *cursor,
407 MDB_val *key, MDB_val *data, MDB_cursor_op op, int *exactp);
408 static int mdb_cursor_first(MDB_cursor *cursor,
409 MDB_val *key, MDB_val *data);
410 static int mdb_cursor_last(MDB_cursor *cursor,
411 MDB_val *key, MDB_val *data);
413 static void mdb_xcursor_init0(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx);
414 static void mdb_xcursor_init1(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx, MDB_node *node);
415 static void mdb_xcursor_fini(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx);
417 static size_t mdb_leaf_size(MDB_env *env, MDB_val *key,
419 static size_t mdb_branch_size(MDB_env *env, MDB_val *key);
421 static int memncmp(const void *s1, size_t n1,
422 const void *s2, size_t n2);
423 static int memnrcmp(const void *s1, size_t n1,
424 const void *s2, size_t n2);
427 memncmp(const void *s1, size_t n1, const void *s2, size_t n2)
429 int diff, len_diff = -1;
432 len_diff = (n1 > n2);
435 diff = memcmp(s1, s2, n1);
436 return diff ? diff : len_diff;
440 memnrcmp(const void *s1, size_t n1, const void *s2, size_t n2)
442 const unsigned char *p1, *p2, *p1_lim;
449 p1 = (const unsigned char *)s1 + n1 - 1;
450 p2 = (const unsigned char *)s2 + n2 - 1;
452 for (p1_lim = (n1 <= n2 ? s1 : s2); *p1 == *p2; p1--, p2--) {
454 return (p1 != s1) ? (p1 != p2) : (p2 != s2) ? -1 : 0;
460 mdb_version(int *maj, int *min, int *pat)
462 *maj = MDB_VERSION_MAJOR;
463 *min = MDB_VERSION_MINOR;
464 *pat = MDB_VERSION_PATCH;
465 return MDB_VERSION_STRING;
468 static const char *errstr[] = {
469 "MDB_KEYEXIST: Key/data pair already exists",
470 "MDB_NOTFOUND: No matching key/data pair found",
471 "MDB_PAGE_NOTFOUND: Requested page not found",
472 "MDB_CORRUPTED: Located page was wrong type",
473 "MDB_PANIC: Update of meta page failed",
474 "MDB_VERSION_MISMATCH: Database environment version mismatch"
478 mdb_strerror(int err)
481 return ("Successful return: 0");
483 if (err >= MDB_KEYEXIST && err <= MDB_VERSION_MISMATCH)
484 return (char *)errstr[err - MDB_KEYEXIST];
486 return strerror(err);
490 mdb_cmp(MDB_txn *txn, MDB_dbi dbi, const MDB_val *a, const MDB_val *b)
492 if (txn->mt_dbxs[dbi].md_cmp)
493 return txn->mt_dbxs[dbi].md_cmp(a, b);
495 if (txn->mt_dbs[dbi].md_flags & (MDB_REVERSEKEY
496 #if __BYTE_ORDER == __LITTLE_ENDIAN
500 return memnrcmp(a->mv_data, a->mv_size, b->mv_data, b->mv_size);
502 return memncmp((char *)a->mv_data, a->mv_size, b->mv_data, b->mv_size);
506 mdb_dcmp(MDB_txn *txn, MDB_dbi dbi, const MDB_val *a, const MDB_val *b)
508 if (txn->mt_dbxs[dbi].md_dcmp)
509 return txn->mt_dbxs[dbi].md_dcmp(a, b);
511 return memncmp((char *)a->mv_data, a->mv_size, b->mv_data, b->mv_size);
514 /* Allocate new page(s) for writing */
516 mdb_alloc_page(MDB_txn *txn, MDB_page *parent, unsigned int parent_idx, int num)
519 pgno_t pgno = P_INVALID;
523 if (txn->mt_txnid > 2) {
525 oldest = txn->mt_txnid - 2;
526 if (!txn->mt_env->me_pghead && txn->mt_dbs[FREE_DBI].md_root != P_INVALID) {
527 /* See if there's anything in the free DB */
532 mpp.mp_parent = NULL;
534 mdb_search_page(txn, FREE_DBI, NULL, NULL, 0, &mpp);
535 leaf = NODEPTR(mpp.mp_page, 0);
536 kptr = (ULONG *)NODEKEY(leaf);
538 /* It's potentially usable, unless there are still
539 * older readers outstanding. Grab it.
541 if (oldest > *kptr) {
546 mdb_read_data(txn, leaf, &data);
547 idl = (ULONG *)data.mv_data;
548 mop = malloc(sizeof(MDB_oldpages) + MDB_IDL_SIZEOF(idl) - sizeof(pgno_t));
549 mop->mo_next = txn->mt_env->me_pghead;
550 mop->mo_txnid = *kptr;
551 txn->mt_env->me_pghead = mop;
552 memcpy(mop->mo_pages, idl, MDB_IDL_SIZEOF(idl));
557 DPRINTF("IDL read txn %lu root %lu num %lu",
558 mop->mo_txnid, txn->mt_dbs[FREE_DBI].md_root, idl[0]);
559 for (i=0; i<idl[0]; i++) {
560 DPRINTF("IDL %lu", idl[i+1]);
564 /* drop this IDL from the DB */
565 mpp.mp_parent = NULL;
567 mdb_search_page(txn, FREE_DBI, NULL, NULL, 1, &mpp);
568 leaf = NODEPTR(mpp.mp_page, 0);
569 mdb_del0(txn, FREE_DBI, 0, &mpp, leaf);
572 if (txn->mt_env->me_pghead) {
574 for (i=0; i<txn->mt_env->me_txns->mti_numreaders; i++) {
575 ULONG mr = txn->mt_env->me_txns->mti_readers[i].mr_txnid;
578 oldest = txn->mt_env->me_txns->mti_readers[i].mr_txnid;
580 if (oldest > txn->mt_env->me_pghead->mo_txnid) {
581 MDB_oldpages *mop = txn->mt_env->me_pghead;
582 txn->mt_oldest = oldest;
584 /* FIXME: For now, always use fresh pages. We
585 * really ought to search the free list for a
590 /* peel pages off tail, so we only have to truncate the list */
591 pgno = MDB_IDL_LAST(mop->mo_pages);
592 if (MDB_IDL_IS_RANGE(mop->mo_pages)) {
594 if (mop->mo_pages[2] > mop->mo_pages[1])
595 mop->mo_pages[0] = 0;
599 if (MDB_IDL_IS_ZERO(mop->mo_pages)) {
600 txn->mt_env->me_pghead = mop->mo_next;
608 if (pgno == P_INVALID) {
609 /* DB size is maxed out */
610 if (txn->mt_next_pgno + num >= txn->mt_env->me_maxpg)
613 if ((dp = malloc(txn->mt_env->me_psize * num + sizeof(MDB_dhead))) == NULL)
616 dp->h.md_parent = parent;
617 dp->h.md_pi = parent_idx;
618 if (pgno == P_INVALID) {
619 dp->p.mp_pgno = txn->mt_next_pgno;
620 txn->mt_next_pgno += num;
622 dp->p.mp_pgno = pgno;
624 mid.mid = dp->p.mp_pgno;
626 mdb_midl2_insert(txn->mt_u.dirty_list, &mid);
631 /* Touch a page: make it dirty and re-insert into tree with updated pgno.
634 mdb_touch(MDB_txn *txn, MDB_pageparent *pp)
636 MDB_page *mp = pp->mp_page;
641 if (!F_ISSET(mp->mp_flags, P_DIRTY)) {
643 if ((dp = mdb_alloc_page(txn, pp->mp_parent, pp->mp_pi, 1)) == NULL)
645 DPRINTF("touched page %lu -> %lu", mp->mp_pgno, dp->p.mp_pgno);
646 mdb_midl_insert(txn->mt_free_pgs, mp->mp_pgno);
647 pgno = dp->p.mp_pgno;
648 memcpy(&dp->p, mp, txn->mt_env->me_psize);
651 mp->mp_flags |= P_DIRTY;
653 /* Update the page number to new touched page. */
654 if (pp->mp_parent != NULL)
655 NODEPGNO(NODEPTR(pp->mp_parent, pp->mp_pi)) = mp->mp_pgno;
662 mdb_env_sync(MDB_env *env, int force)
665 if (force || !F_ISSET(env->me_flags, MDB_NOSYNC)) {
666 if (fdatasync(env->me_fd))
673 mdb_txn_begin(MDB_env *env, int rdonly, MDB_txn **ret)
678 if (env->me_flags & MDB_FATAL_ERROR) {
679 DPUTS("mdb_txn_begin: environment had fatal error, must shutdown!");
682 if ((txn = calloc(1, sizeof(MDB_txn))) == NULL) {
683 DPRINTF("calloc: %s", strerror(errno));
688 txn->mt_flags |= MDB_TXN_RDONLY;
690 txn->mt_u.dirty_list = env->me_dirty_list;
691 txn->mt_u.dirty_list[0].mid = 0;
692 txn->mt_free_pgs = env->me_free_pgs;
693 txn->mt_free_pgs[0] = 0;
695 pthread_mutex_lock(&env->me_txns->mti_wmutex);
696 env->me_txns->mti_txnid++;
699 txn->mt_txnid = env->me_txns->mti_txnid;
701 MDB_reader *r = pthread_getspecific(env->me_txkey);
704 pthread_mutex_lock(&env->me_txns->mti_mutex);
705 for (i=0; i<env->me_txns->mti_numreaders; i++)
706 if (env->me_txns->mti_readers[i].mr_pid == 0)
708 if (i == env->me_maxreaders) {
709 pthread_mutex_unlock(&env->me_txns->mti_mutex);
712 env->me_txns->mti_readers[i].mr_pid = getpid();
713 env->me_txns->mti_readers[i].mr_tid = pthread_self();
714 r = &env->me_txns->mti_readers[i];
715 pthread_setspecific(env->me_txkey, r);
716 if (i >= env->me_txns->mti_numreaders)
717 env->me_txns->mti_numreaders = i+1;
718 pthread_mutex_unlock(&env->me_txns->mti_mutex);
720 r->mr_txnid = txn->mt_txnid;
721 txn->mt_u.reader = r;
728 toggle = env->me_txns->mti_me_toggle;
729 if ((rc = mdb_env_read_meta(env, &toggle)) != MDB_SUCCESS) {
734 /* Copy the DB arrays */
735 txn->mt_numdbs = env->me_numdbs;
736 txn->mt_dbxs = env->me_dbxs; /* mostly static anyway */
737 txn->mt_dbs = malloc(env->me_maxdbs * sizeof(MDB_db));
738 memcpy(txn->mt_dbs, env->me_meta->mm_dbs, 2 * sizeof(MDB_db));
739 if (txn->mt_numdbs > 2)
740 memcpy(txn->mt_dbs+2, env->me_dbs[env->me_db_toggle]+2,
741 (txn->mt_numdbs - 2) * sizeof(MDB_db));
745 txn->mt_flags |= MDB_TXN_METOGGLE;
746 txn->mt_next_pgno = env->me_meta->mm_last_pg+1;
749 DPRINTF("begin transaction %lu on mdbenv %p, root page %lu",
750 txn->mt_txnid, (void *) env, txn->mt_dbs[MAIN_DBI].md_root);
757 mdb_txn_abort(MDB_txn *txn)
765 DPRINTF("abort transaction %lu on mdbenv %p, root page %lu",
766 txn->mt_txnid, (void *) env, txn->mt_dbs[MAIN_DBI].md_root);
770 if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
771 txn->mt_u.reader->mr_txnid = 0;
776 /* Discard all dirty pages. */
777 for (i=1; i<=txn->mt_u.dirty_list[0].mid; i++)
778 free(txn->mt_u.dirty_list[i].mptr);
780 while ((mop = txn->mt_env->me_pghead)) {
781 txn->mt_env->me_pghead = mop->mo_next;
786 env->me_txns->mti_txnid--;
787 for (i=2; i<env->me_numdbs; i++)
788 env->me_dbxs[i].md_dirty = 0;
789 pthread_mutex_unlock(&env->me_txns->mti_wmutex);
796 mdb_txn_commit(MDB_txn *txn)
805 struct iovec iov[MDB_COMMIT_PAGES];
808 assert(txn->mt_env != NULL);
812 if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
817 if (txn != env->me_txn) {
818 DPUTS("attempt to commit unknown transaction");
823 if (F_ISSET(txn->mt_flags, MDB_TXN_ERROR)) {
824 DPUTS("error flag is set, can't commit");
829 if (!txn->mt_u.dirty_list[0].mid)
832 DPRINTF("committing transaction %lu on mdbenv %p, root page %lu",
833 txn->mt_txnid, (void *) env, txn->mt_dbs[MAIN_DBI].md_root);
835 /* should only be one record now */
836 if (env->me_pghead) {
840 mop = env->me_pghead;
841 key.mv_size = sizeof(pgno_t);
842 key.mv_data = (char *)&mop->mo_txnid;
843 data.mv_size = MDB_IDL_SIZEOF(mop->mo_pages);
844 data.mv_data = mop->mo_pages;
845 mdb_put0(txn, FREE_DBI, &key, &data, 0);
846 free(env->me_pghead);
847 env->me_pghead = NULL;
849 /* save to free list */
850 if (!MDB_IDL_IS_ZERO(txn->mt_free_pgs)) {
854 /* make sure last page of freeDB is touched and on freelist */
855 key.mv_size = MAXKEYSIZE+1;
857 mpp.mp_parent = NULL;
859 mdb_search_page(txn, FREE_DBI, &key, NULL, 1, &mpp);
864 ULONG *idl = txn->mt_free_pgs;
865 DPRINTF("IDL write txn %lu root %lu num %lu",
866 txn->mt_txnid, txn->mt_dbs[FREE_DBI].md_root, idl[0]);
867 for (i=0; i<idl[0]; i++) {
868 DPRINTF("IDL %lu", idl[i+1]);
872 /* write to last page of freeDB */
873 key.mv_size = sizeof(pgno_t);
874 key.mv_data = (char *)&txn->mt_txnid;
875 data.mv_size = MDB_IDL_SIZEOF(txn->mt_free_pgs);
876 data.mv_data = txn->mt_free_pgs;
877 mdb_put0(txn, FREE_DBI, &key, &data, 0);
880 /* Update DB root pointers. Their pages have already been
881 * touched so this is all in-place and cannot fail.
885 data.mv_size = sizeof(MDB_db);
887 for (i = 2; i < txn->mt_numdbs; i++) {
888 if (txn->mt_dbxs[i].md_dirty) {
889 data.mv_data = &txn->mt_dbs[i];
890 mdb_put0(txn, MAIN_DBI, &txn->mt_dbxs[i].md_name, &data, 0);
895 /* Commit up to MDB_COMMIT_PAGES dirty pages to disk until done.
903 for (; i<=txn->mt_u.dirty_list[0].mid; i++) {
904 dp = txn->mt_u.dirty_list[i].mptr;
905 if (dp->p.mp_pgno != next) {
907 DPRINTF("committing %u dirty pages", n);
908 rc = writev(env->me_fd, iov, n);
912 DPUTS("short write, filesystem full?");
914 DPRINTF("writev: %s", strerror(errno));
921 lseek(env->me_fd, dp->p.mp_pgno * env->me_psize, SEEK_SET);
922 next = dp->p.mp_pgno;
924 DPRINTF("committing page %lu", dp->p.mp_pgno);
925 iov[n].iov_len = env->me_psize * dp->h.md_num;
926 iov[n].iov_base = &dp->p;
927 size += iov[n].iov_len;
928 next = dp->p.mp_pgno + dp->h.md_num;
929 /* clear dirty flag */
930 dp->p.mp_flags &= ~P_DIRTY;
931 if (++n >= MDB_COMMIT_PAGES) {
940 DPRINTF("committing %u dirty pages", n);
941 rc = writev(env->me_fd, iov, n);
945 DPUTS("short write, filesystem full?");
947 DPRINTF("writev: %s", strerror(errno));
954 /* Drop the dirty pages.
956 for (i=1; i<=txn->mt_u.dirty_list[0].mid; i++)
957 free(txn->mt_u.dirty_list[i].mptr);
959 if ((n = mdb_env_sync(env, 0)) != 0 ||
960 (n = mdb_env_write_meta(txn)) != MDB_SUCCESS) {
967 /* update the DB tables */
969 int toggle = !env->me_db_toggle;
971 for (i = 2; i < env->me_numdbs; i++) {
972 if (txn->mt_dbxs[i].md_dirty) {
973 env->me_dbs[toggle][i] = txn->mt_dbs[i];
974 txn->mt_dbxs[i].md_dirty = 0;
977 for (i = env->me_numdbs; i < txn->mt_numdbs; i++) {
978 txn->mt_dbxs[i].md_dirty = 0;
979 env->me_dbxs[i] = txn->mt_dbxs[i];
980 env->me_dbs[toggle][i] = txn->mt_dbs[i];
982 env->me_db_toggle = toggle;
983 env->me_numdbs = txn->mt_numdbs;
988 pthread_mutex_unlock(&env->me_txns->mti_wmutex);
995 mdb_env_read_header(MDB_env *env, MDB_meta *meta)
1002 assert(env != NULL);
1004 /* We don't know the page size yet, so use a minimum value.
1007 if ((rc = pread(env->me_fd, page, PAGESIZE, 0)) == 0) {
1009 } else if (rc != PAGESIZE) {
1012 DPRINTF("read: %s", strerror(errno));
1016 p = (MDB_page *)page;
1018 if (!F_ISSET(p->mp_flags, P_META)) {
1019 DPRINTF("page %lu not a meta page", p->mp_pgno);
1024 if (m->mm_magic != MDB_MAGIC) {
1025 DPUTS("meta has invalid magic");
1029 if (m->mm_version != MDB_VERSION) {
1030 DPRINTF("database is version %u, expected version %u",
1031 m->mm_version, MDB_VERSION);
1032 return MDB_VERSION_MISMATCH;
1035 memcpy(meta, m, sizeof(*m));
1040 mdb_env_init_meta(MDB_env *env, MDB_meta *meta)
1047 DPUTS("writing new meta page");
1048 psize = sysconf(_SC_PAGE_SIZE);
1050 meta->mm_magic = MDB_MAGIC;
1051 meta->mm_version = MDB_VERSION;
1052 meta->mm_psize = psize;
1053 meta->mm_last_pg = 1;
1054 meta->mm_flags = env->me_flags & 0xffff;
1055 meta->mm_flags |= MDB_INTEGERKEY;
1056 meta->mm_dbs[0].md_root = P_INVALID;
1057 meta->mm_dbs[1].md_root = P_INVALID;
1059 p = calloc(2, psize);
1061 p->mp_flags = P_META;
1064 memcpy(m, meta, sizeof(*meta));
1066 q = (MDB_page *)((char *)p + psize);
1069 q->mp_flags = P_META;
1072 memcpy(m, meta, sizeof(*meta));
1074 rc = write(env->me_fd, p, psize * 2);
1076 return (rc == (int)psize * 2) ? MDB_SUCCESS : errno;
1080 mdb_env_write_meta(MDB_txn *txn)
1083 MDB_meta meta, metab;
1085 int rc, len, toggle;
1088 assert(txn != NULL);
1089 assert(txn->mt_env != NULL);
1091 toggle = !F_ISSET(txn->mt_flags, MDB_TXN_METOGGLE);
1092 DPRINTF("writing meta page %d for root page %lu",
1093 toggle, txn->mt_dbs[MAIN_DBI].md_root);
1097 metab.mm_txnid = env->me_metas[toggle]->mm_txnid;
1098 metab.mm_last_pg = env->me_metas[toggle]->mm_last_pg;
1100 ptr = (char *)&meta;
1101 off = offsetof(MDB_meta, mm_dbs[0].md_depth);
1102 len = sizeof(MDB_meta) - off;
1105 meta.mm_dbs[0] = txn->mt_dbs[0];
1106 meta.mm_dbs[1] = txn->mt_dbs[1];
1107 meta.mm_last_pg = txn->mt_next_pgno - 1;
1108 meta.mm_txnid = txn->mt_txnid;
1111 off += env->me_psize;
1114 /* Write to the SYNC fd */
1115 rc = pwrite(env->me_mfd, ptr, len, off);
1117 DPUTS("write failed, disk error?");
1118 /* On a failure, the pagecache still contains the new data.
1119 * Write some old data back, to prevent it from being used.
1120 * Use the non-SYNC fd; we know it will fail anyway.
1122 meta.mm_last_pg = metab.mm_last_pg;
1123 meta.mm_txnid = metab.mm_txnid;
1124 rc = pwrite(env->me_fd, ptr, len, off);
1125 env->me_flags |= MDB_FATAL_ERROR;
1128 txn->mt_env->me_txns->mti_me_toggle = toggle;
1134 mdb_env_read_meta(MDB_env *env, int *which)
1138 assert(env != NULL);
1142 else if (env->me_metas[0]->mm_txnid < env->me_metas[1]->mm_txnid)
1145 if (env->me_meta != env->me_metas[toggle])
1146 env->me_meta = env->me_metas[toggle];
1148 DPRINTF("Using meta page %d", toggle);
1154 mdb_env_create(MDB_env **env)
1158 e = calloc(1, sizeof(MDB_env));
1159 if (!e) return ENOMEM;
1161 e->me_maxreaders = DEFAULT_READERS;
1171 mdb_env_set_mapsize(MDB_env *env, size_t size)
1175 env->me_mapsize = size;
1180 mdb_env_set_maxdbs(MDB_env *env, int dbs)
1182 env->me_maxdbs = dbs;
1187 mdb_env_set_maxreaders(MDB_env *env, int readers)
1189 env->me_maxreaders = readers;
1194 mdb_env_get_maxreaders(MDB_env *env, int *readers)
1196 if (!env || !readers)
1198 *readers = env->me_maxreaders;
1203 mdb_env_open2(MDB_env *env, unsigned int flags)
1209 env->me_flags = flags;
1211 memset(&meta, 0, sizeof(meta));
1213 if ((i = mdb_env_read_header(env, &meta)) != 0) {
1216 DPUTS("new mdbenv");
1220 if (!env->me_mapsize) {
1221 env->me_mapsize = newenv ? DEFAULT_MAPSIZE : meta.mm_mapsize;
1225 if (meta.mm_address && (flags & MDB_FIXEDMAP))
1227 env->me_map = mmap(meta.mm_address, env->me_mapsize, PROT_READ, i,
1229 if (env->me_map == MAP_FAILED)
1233 meta.mm_mapsize = env->me_mapsize;
1234 if (flags & MDB_FIXEDMAP)
1235 meta.mm_address = env->me_map;
1236 i = mdb_env_init_meta(env, &meta);
1237 if (i != MDB_SUCCESS) {
1238 munmap(env->me_map, env->me_mapsize);
1242 env->me_psize = meta.mm_psize;
1244 env->me_maxpg = env->me_mapsize / env->me_psize;
1246 p = (MDB_page *)env->me_map;
1247 env->me_metas[0] = METADATA(p);
1248 env->me_metas[1] = (MDB_meta *)((char *)env->me_metas[0] + meta.mm_psize);
1250 if ((i = mdb_env_read_meta(env, NULL)) != 0)
1253 DPRINTF("opened database version %u, pagesize %u",
1254 env->me_meta->mm_version, env->me_psize);
1255 DPRINTF("depth: %u", env->me_meta->mm_dbs[MAIN_DBI].md_depth);
1256 DPRINTF("entries: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_entries);
1257 DPRINTF("branch pages: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_branch_pages);
1258 DPRINTF("leaf pages: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_leaf_pages);
1259 DPRINTF("overflow pages: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_overflow_pages);
1260 DPRINTF("root: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_root);
1266 mdb_env_reader_dest(void *ptr)
1268 MDB_reader *reader = ptr;
1270 reader->mr_txnid = 0;
1275 /* downgrade the exclusive lock on the region back to shared */
1277 mdb_env_share_locks(MDB_env *env)
1279 struct flock lock_info;
1281 env->me_txns->mti_txnid = env->me_meta->mm_txnid;
1282 if (env->me_metas[0]->mm_txnid < env->me_metas[1]->mm_txnid)
1283 env->me_txns->mti_me_toggle = 1;
1285 memset((void *)&lock_info, 0, sizeof(lock_info));
1286 lock_info.l_type = F_RDLCK;
1287 lock_info.l_whence = SEEK_SET;
1288 lock_info.l_start = 0;
1289 lock_info.l_len = 1;
1290 fcntl(env->me_lfd, F_SETLK, &lock_info);
1294 mdb_env_setup_locks(MDB_env *env, char *lpath, int mode, int *excl)
1298 struct flock lock_info;
1302 if ((env->me_lfd = open(lpath, O_RDWR|O_CREAT, mode)) == -1) {
1306 /* Try to get exclusive lock. If we succeed, then
1307 * nobody is using the lock region and we should initialize it.
1309 memset((void *)&lock_info, 0, sizeof(lock_info));
1310 lock_info.l_type = F_WRLCK;
1311 lock_info.l_whence = SEEK_SET;
1312 lock_info.l_start = 0;
1313 lock_info.l_len = 1;
1314 rc = fcntl(env->me_lfd, F_SETLK, &lock_info);
1318 lock_info.l_type = F_RDLCK;
1319 rc = fcntl(env->me_lfd, F_SETLK, &lock_info);
1325 size = lseek(env->me_lfd, 0, SEEK_END);
1326 rsize = (env->me_maxreaders-1) * sizeof(MDB_reader) + sizeof(MDB_txninfo);
1327 if (size < rsize && *excl) {
1328 if (ftruncate(env->me_lfd, rsize) != 0) {
1334 size = rsize - sizeof(MDB_txninfo);
1335 env->me_maxreaders = size/sizeof(MDB_reader) + 1;
1337 env->me_txns = mmap(0, rsize, PROT_READ|PROT_WRITE, MAP_SHARED,
1339 if (env->me_txns == MAP_FAILED) {
1344 pthread_mutexattr_t mattr;
1346 pthread_mutexattr_init(&mattr);
1347 rc = pthread_mutexattr_setpshared(&mattr, PTHREAD_PROCESS_SHARED);
1351 pthread_mutex_init(&env->me_txns->mti_mutex, &mattr);
1352 pthread_mutex_init(&env->me_txns->mti_wmutex, &mattr);
1353 env->me_txns->mti_version = MDB_VERSION;
1354 env->me_txns->mti_magic = MDB_MAGIC;
1355 env->me_txns->mti_txnid = 0;
1356 env->me_txns->mti_numreaders = 0;
1357 env->me_txns->mti_me_toggle = 0;
1360 if (env->me_txns->mti_magic != MDB_MAGIC) {
1361 DPUTS("lock region has invalid magic");
1365 if (env->me_txns->mti_version != MDB_VERSION) {
1366 DPRINTF("lock region is version %u, expected version %u",
1367 env->me_txns->mti_version, MDB_VERSION);
1368 rc = MDB_VERSION_MISMATCH;
1371 if (errno != EACCES && errno != EAGAIN) {
1385 #define LOCKNAME "/lock.mdb"
1386 #define DATANAME "/data.mdb"
1388 mdb_env_open(MDB_env *env, const char *path, unsigned int flags, mode_t mode)
1390 int oflags, rc, len, excl;
1391 char *lpath, *dpath;
1394 lpath = malloc(len + sizeof(LOCKNAME) + len + sizeof(DATANAME));
1397 dpath = lpath + len + sizeof(LOCKNAME);
1398 sprintf(lpath, "%s" LOCKNAME, path);
1399 sprintf(dpath, "%s" DATANAME, path);
1401 rc = mdb_env_setup_locks(env, lpath, mode, &excl);
1405 if (F_ISSET(flags, MDB_RDONLY))
1408 oflags = O_RDWR | O_CREAT;
1410 if ((env->me_fd = open(dpath, oflags, mode)) == -1) {
1415 if ((rc = mdb_env_open2(env, flags)) == MDB_SUCCESS) {
1416 /* synchronous fd for meta writes */
1417 if (!(flags & (MDB_RDONLY|MDB_NOSYNC)))
1418 oflags |= MDB_DSYNC;
1419 if ((env->me_mfd = open(dpath, oflags, mode)) == -1) {
1424 env->me_path = strdup(path);
1425 DPRINTF("opened dbenv %p", (void *) env);
1426 pthread_key_create(&env->me_txkey, mdb_env_reader_dest);
1428 mdb_env_share_locks(env);
1429 env->me_dbxs = calloc(env->me_maxdbs, sizeof(MDB_dbx));
1430 env->me_dbs[0] = calloc(env->me_maxdbs, sizeof(MDB_db));
1431 env->me_dbs[1] = calloc(env->me_maxdbs, sizeof(MDB_db));
1437 if (env->me_fd >= 0) {
1441 if (env->me_lfd >= 0) {
1451 mdb_env_close(MDB_env *env)
1456 free(env->me_dbs[1]);
1457 free(env->me_dbs[0]);
1461 pthread_key_delete(env->me_txkey);
1464 munmap(env->me_map, env->me_mapsize);
1469 pid_t pid = getpid();
1470 size_t size = (env->me_maxreaders-1) * sizeof(MDB_reader) + sizeof(MDB_txninfo);
1472 for (i=0; i<env->me_txns->mti_numreaders; i++)
1473 if (env->me_txns->mti_readers[i].mr_pid == pid)
1474 env->me_txns->mti_readers[i].mr_pid = 0;
1475 munmap(env->me_txns, size);
1481 /* Search for key within a leaf page, using binary search.
1482 * Returns the smallest entry larger or equal to the key.
1483 * If exactp is non-null, stores whether the found entry was an exact match
1484 * in *exactp (1 or 0).
1485 * If kip is non-null, stores the index of the found entry in *kip.
1486 * If no entry larger or equal to the key is found, returns NULL.
1489 mdb_search_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp, MDB_val *key,
1490 int *exactp, unsigned int *kip)
1498 DPRINTF("searching %u keys in %s page %lu",
1500 IS_LEAF(mp) ? "leaf" : "branch",
1503 assert(NUMKEYS(mp) > 0);
1505 memset(&nodekey, 0, sizeof(nodekey));
1507 low = IS_LEAF(mp) ? 0 : 1;
1508 high = NUMKEYS(mp) - 1;
1509 while (low <= high) {
1510 i = (low + high) >> 1;
1511 node = NODEPTR(mp, i);
1513 nodekey.mv_size = node->mn_ksize;
1514 nodekey.mv_data = NODEKEY(node);
1516 rc = mdb_cmp(txn, dbi, key, &nodekey);
1519 DPRINTF("found leaf index %u [%.*s], rc = %i",
1520 i, (int)nodekey.mv_size, (char *)nodekey.mv_data, rc);
1522 DPRINTF("found branch index %u [%.*s -> %lu], rc = %i",
1523 i, (int)node->mn_ksize, (char *)NODEKEY(node),
1534 if (rc > 0) { /* Found entry is less than the key. */
1535 i++; /* Skip to get the smallest entry larger than key. */
1536 if (i >= NUMKEYS(mp))
1537 /* There is no entry larger or equal to the key. */
1541 *exactp = (rc == 0);
1542 if (kip) /* Store the key index if requested. */
1545 return NODEPTR(mp, i);
1549 cursor_pop_page(MDB_cursor *cursor)
1553 if (cursor->mc_snum) {
1554 top = CURSOR_TOP(cursor);
1557 DPRINTF("popped page %lu off cursor %p", top->mp_page->mp_pgno, (void *) cursor);
1562 cursor_push_page(MDB_cursor *cursor, MDB_page *mp)
1566 DPRINTF("pushing page %lu on cursor %p", mp->mp_pgno, (void *) cursor);
1568 ppage = &cursor->mc_stack[cursor->mc_snum++];
1569 ppage->mp_page = mp;
1575 mdb_get_page(MDB_txn *txn, pgno_t pgno)
1580 if (!F_ISSET(txn->mt_flags, MDB_TXN_RDONLY) && txn->mt_u.dirty_list[0].mid) {
1585 x = mdb_midl2_search(txn->mt_u.dirty_list, &id);
1586 if (x <= txn->mt_u.dirty_list[0].mid && txn->mt_u.dirty_list[x].mid == pgno) {
1587 dp = txn->mt_u.dirty_list[x].mptr;
1593 if (pgno > txn->mt_env->me_meta->mm_last_pg)
1595 p = (MDB_page *)(txn->mt_env->me_map + txn->mt_env->me_psize * pgno);
1601 mdb_search_page_root(MDB_txn *txn, MDB_dbi dbi, MDB_val *key,
1602 MDB_cursor *cursor, int modify, MDB_pageparent *mpp)
1604 MDB_page *mp = mpp->mp_page;
1607 if (cursor && cursor_push_page(cursor, mp) == NULL)
1610 while (IS_BRANCH(mp)) {
1614 DPRINTF("branch page %lu has %u keys", mp->mp_pgno, NUMKEYS(mp));
1615 assert(NUMKEYS(mp) > 1);
1616 DPRINTF("found index 0 to page %lu", NODEPGNO(NODEPTR(mp, 0)));
1618 if (key == NULL) /* Initialize cursor to first page. */
1620 else if (key->mv_size > MAXKEYSIZE && key->mv_data == NULL) {
1621 /* cursor to last page */
1625 node = mdb_search_node(txn, dbi, mp, key, &exact, &i);
1627 i = NUMKEYS(mp) - 1;
1635 DPRINTF("following index %u for key %.*s",
1636 i, (int)key->mv_size, (char *)key->mv_data);
1637 assert(i < NUMKEYS(mp));
1638 node = NODEPTR(mp, i);
1641 CURSOR_TOP(cursor)->mp_ki = i;
1643 mpp->mp_parent = mp;
1644 if ((mp = mdb_get_page(txn, NODEPGNO(node))) == NULL)
1645 return MDB_PAGE_NOTFOUND;
1649 if (cursor && cursor_push_page(cursor, mp) == NULL)
1653 MDB_dhead *dh = ((MDB_dhead *)mp)-1;
1654 if ((rc = mdb_touch(txn, mpp)) != 0)
1656 dh = ((MDB_dhead *)mpp->mp_page)-1;
1657 dh->md_parent = mpp->mp_parent;
1658 dh->md_pi = mpp->mp_pi;
1665 DPRINTF("internal error, index points to a %02X page!?",
1667 return MDB_CORRUPTED;
1670 DPRINTF("found leaf page %lu for key %.*s", mp->mp_pgno,
1671 key ? (int)key->mv_size : 0, key ? (char *)key->mv_data : NULL);
1676 /* Search for the page a given key should be in.
1677 * Stores a pointer to the found page in *mpp.
1678 * If key is NULL, search for the lowest page (used by mdb_cursor_first).
1679 * If cursor is non-null, pushes parent pages on the cursor stack.
1680 * If modify is true, visited pages are updated with new page numbers.
1683 mdb_search_page(MDB_txn *txn, MDB_dbi dbi, MDB_val *key,
1684 MDB_cursor *cursor, int modify, MDB_pageparent *mpp)
1689 /* Choose which root page to start with. If a transaction is given
1690 * use the root page from the transaction, otherwise read the last
1691 * committed root page.
1693 if (F_ISSET(txn->mt_flags, MDB_TXN_ERROR)) {
1694 DPUTS("transaction has failed, must abort");
1697 root = txn->mt_dbs[dbi].md_root;
1699 if (root == P_INVALID) { /* Tree is empty. */
1700 DPUTS("tree is empty");
1701 return MDB_NOTFOUND;
1704 if ((mpp->mp_page = mdb_get_page(txn, root)) == NULL)
1705 return MDB_PAGE_NOTFOUND;
1707 DPRINTF("root page has flags 0x%X", mpp->mp_page->mp_flags);
1710 /* For sub-databases, update main root first */
1711 if (dbi > MAIN_DBI && !txn->mt_dbxs[dbi].md_dirty) {
1713 rc = mdb_search_page(txn, MAIN_DBI, &txn->mt_dbxs[dbi].md_name,
1717 txn->mt_dbxs[dbi].md_dirty = 1;
1719 if (!F_ISSET(mpp->mp_page->mp_flags, P_DIRTY)) {
1720 mpp->mp_parent = NULL;
1722 if ((rc = mdb_touch(txn, mpp)))
1724 txn->mt_dbs[dbi].md_root = mpp->mp_page->mp_pgno;
1728 return mdb_search_page_root(txn, dbi, key, cursor, modify, mpp);
1732 mdb_read_data(MDB_txn *txn, MDB_node *leaf, MDB_val *data)
1734 MDB_page *omp; /* overflow mpage */
1737 if (!F_ISSET(leaf->mn_flags, F_BIGDATA)) {
1738 data->mv_size = leaf->mn_dsize;
1739 data->mv_data = NODEDATA(leaf);
1743 /* Read overflow data.
1745 data->mv_size = leaf->mn_dsize;
1746 memcpy(&pgno, NODEDATA(leaf), sizeof(pgno));
1747 if ((omp = mdb_get_page(txn, pgno)) == NULL) {
1748 DPRINTF("read overflow page %lu failed", pgno);
1749 return MDB_PAGE_NOTFOUND;
1751 data->mv_data = METADATA(omp);
1757 mdb_get(MDB_txn *txn, MDB_dbi dbi,
1758 MDB_val *key, MDB_val *data)
1766 DPRINTF("===> get key [%.*s]", (int)key->mv_size, (char *)key->mv_data);
1768 if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
1771 if (key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
1775 if ((rc = mdb_search_page(txn, dbi, key, NULL, 0, &mpp)) != MDB_SUCCESS)
1778 leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, NULL);
1779 if (leaf && exact) {
1780 /* Return first duplicate data item */
1781 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
1784 mdb_xcursor_init0(txn, dbi, &mx);
1785 mdb_xcursor_init1(txn, dbi, &mx, leaf);
1786 rc = mdb_search_page(&mx.mx_txn, mx.mx_cursor.mc_dbi, NULL, NULL, 0, &mpp);
1787 if (rc != MDB_SUCCESS)
1789 leaf = NODEPTR(mpp.mp_page, 0);
1791 rc = mdb_read_data(txn, leaf, data);
1800 mdb_sibling(MDB_cursor *cursor, int move_right)
1807 if (cursor->mc_snum < 2) {
1808 return MDB_NOTFOUND; /* root has no siblings */
1810 parent = CURSOR_PARENT(cursor);
1812 DPRINTF("parent page is page %lu, index %u",
1813 parent->mp_page->mp_pgno, parent->mp_ki);
1815 cursor_pop_page(cursor);
1816 if (move_right ? (parent->mp_ki + 1 >= NUMKEYS(parent->mp_page))
1817 : (parent->mp_ki == 0)) {
1818 DPRINTF("no more keys left, moving to %s sibling",
1819 move_right ? "right" : "left");
1820 if ((rc = mdb_sibling(cursor, move_right)) != MDB_SUCCESS)
1822 parent = CURSOR_TOP(cursor);
1828 DPRINTF("just moving to %s index key %u",
1829 move_right ? "right" : "left", parent->mp_ki);
1831 assert(IS_BRANCH(parent->mp_page));
1833 indx = NODEPTR(parent->mp_page, parent->mp_ki);
1834 if ((mp = mdb_get_page(cursor->mc_txn, indx->mn_pgno)) == NULL)
1835 return MDB_PAGE_NOTFOUND;
1837 mp->parent = parent->mp_page;
1838 mp->parent_index = parent->mp_ki;
1841 cursor_push_page(cursor, mp);
1847 mdb_set_key(MDB_node *node, MDB_val *key)
1852 key->mv_size = node->mn_ksize;
1853 key->mv_data = NODEKEY(node);
1859 mdb_cursor_next(MDB_cursor *cursor, MDB_val *key, MDB_val *data, MDB_cursor_op op)
1866 if (cursor->mc_eof) {
1867 return MDB_NOTFOUND;
1870 assert(cursor->mc_initialized);
1872 top = CURSOR_TOP(cursor);
1875 if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
1876 leaf = NODEPTR(mp, top->mp_ki);
1877 if ((op == MDB_NEXT || op == MDB_NEXT_DUP) && F_ISSET(leaf->mn_flags, F_DUPDATA)) {
1878 rc = mdb_cursor_next(&cursor->mc_xcursor->mx_cursor, data, NULL, MDB_NEXT);
1879 if (op != MDB_NEXT || rc == MDB_SUCCESS)
1884 DPRINTF("cursor_next: top page is %lu in cursor %p", mp->mp_pgno, (void *) cursor);
1886 if (top->mp_ki + 1 >= NUMKEYS(mp)) {
1887 DPUTS("=====> move to next sibling page");
1888 if (mdb_sibling(cursor, 1) != MDB_SUCCESS) {
1890 return MDB_NOTFOUND;
1892 top = CURSOR_TOP(cursor);
1894 DPRINTF("next page is %lu, key index %u", mp->mp_pgno, top->mp_ki);
1898 DPRINTF("==> cursor points to page %lu with %u keys, key index %u",
1899 mp->mp_pgno, NUMKEYS(mp), top->mp_ki);
1901 assert(IS_LEAF(mp));
1902 leaf = NODEPTR(mp, top->mp_ki);
1904 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
1905 mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, leaf);
1908 if ((rc = mdb_read_data(cursor->mc_txn, leaf, data) != MDB_SUCCESS))
1911 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
1912 rc = mdb_cursor_first(&cursor->mc_xcursor->mx_cursor, data, NULL);
1913 if (rc != MDB_SUCCESS)
1918 return mdb_set_key(leaf, key);
1922 mdb_cursor_prev(MDB_cursor *cursor, MDB_val *key, MDB_val *data, MDB_cursor_op op)
1929 assert(cursor->mc_initialized);
1931 top = CURSOR_TOP(cursor);
1934 if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
1935 leaf = NODEPTR(mp, top->mp_ki);
1936 if ((op == MDB_PREV || op == MDB_PREV_DUP) && F_ISSET(leaf->mn_flags, F_DUPDATA)) {
1937 rc = mdb_cursor_prev(&cursor->mc_xcursor->mx_cursor, data, NULL, MDB_PREV);
1938 if (op != MDB_PREV || rc == MDB_SUCCESS)
1943 DPRINTF("cursor_prev: top page is %lu in cursor %p", mp->mp_pgno, (void *) cursor);
1945 if (top->mp_ki == 0) {
1946 DPUTS("=====> move to prev sibling page");
1947 if (mdb_sibling(cursor, 0) != MDB_SUCCESS) {
1948 return MDB_NOTFOUND;
1950 top = CURSOR_TOP(cursor);
1952 top->mp_ki = NUMKEYS(mp) - 1;
1953 DPRINTF("prev page is %lu, key index %u", mp->mp_pgno, top->mp_ki);
1959 DPRINTF("==> cursor points to page %lu with %u keys, key index %u",
1960 mp->mp_pgno, NUMKEYS(mp), top->mp_ki);
1962 assert(IS_LEAF(mp));
1963 leaf = NODEPTR(mp, top->mp_ki);
1965 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
1966 mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, leaf);
1969 if ((rc = mdb_read_data(cursor->mc_txn, leaf, data) != MDB_SUCCESS))
1972 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
1973 rc = mdb_cursor_last(&cursor->mc_xcursor->mx_cursor, data, NULL);
1974 if (rc != MDB_SUCCESS)
1979 return mdb_set_key(leaf, key);
1983 mdb_cursor_set(MDB_cursor *cursor, MDB_val *key, MDB_val *data,
1984 MDB_cursor_op op, int *exactp)
1993 assert(key->mv_size > 0);
1995 cursor->mc_snum = 0;
1997 rc = mdb_search_page(cursor->mc_txn, cursor->mc_dbi, key, cursor, 0, &mpp);
1998 if (rc != MDB_SUCCESS)
2000 assert(IS_LEAF(mpp.mp_page));
2002 top = CURSOR_TOP(cursor);
2003 leaf = mdb_search_node(cursor->mc_txn, cursor->mc_dbi, mpp.mp_page, key, exactp, &top->mp_ki);
2004 if (exactp != NULL && !*exactp) {
2005 /* MDB_SET specified and not an exact match. */
2006 return MDB_NOTFOUND;
2010 DPUTS("===> inexact leaf not found, goto sibling");
2011 if ((rc = mdb_sibling(cursor, 1)) != MDB_SUCCESS)
2012 return rc; /* no entries matched */
2013 top = CURSOR_TOP(cursor);
2015 mpp.mp_page = top->mp_page;
2016 assert(IS_LEAF(mpp.mp_page));
2017 leaf = NODEPTR(mpp.mp_page, 0);
2020 cursor->mc_initialized = 1;
2023 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
2024 mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, leaf);
2027 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
2028 if (op == MDB_SET || op == MDB_SET_RANGE) {
2029 rc = mdb_cursor_first(&cursor->mc_xcursor->mx_cursor, data, NULL);
2033 if (op == MDB_GET_BOTH) {
2038 op2 = MDB_SET_RANGE;
2040 rc = mdb_cursor_set(&cursor->mc_xcursor->mx_cursor, data, NULL, op2, ex2p);
2041 if (rc != MDB_SUCCESS)
2044 } else if (op == MDB_GET_BOTH || op == MDB_GET_BOTH_RANGE) {
2046 if ((rc = mdb_read_data(cursor->mc_txn, leaf, &d2)) != MDB_SUCCESS)
2048 rc = mdb_dcmp(cursor->mc_txn, cursor->mc_dbi, data, &d2);
2050 if (op == MDB_GET_BOTH || rc > 0)
2051 return MDB_NOTFOUND;
2055 if ((rc = mdb_read_data(cursor->mc_txn, leaf, data)) != MDB_SUCCESS)
2060 rc = mdb_set_key(leaf, key);
2061 if (rc == MDB_SUCCESS) {
2062 DPRINTF("==> cursor placed on key %.*s",
2063 (int)key->mv_size, (char *)key->mv_data);
2071 mdb_cursor_first(MDB_cursor *cursor, MDB_val *key, MDB_val *data)
2077 cursor->mc_snum = 0;
2079 rc = mdb_search_page(cursor->mc_txn, cursor->mc_dbi, NULL, cursor, 0, &mpp);
2080 if (rc != MDB_SUCCESS)
2082 assert(IS_LEAF(mpp.mp_page));
2084 leaf = NODEPTR(mpp.mp_page, 0);
2085 cursor->mc_initialized = 1;
2089 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
2090 mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, leaf);
2091 rc = mdb_cursor_first(&cursor->mc_xcursor->mx_cursor, data, NULL);
2095 if ((rc = mdb_read_data(cursor->mc_txn, leaf, data)) != MDB_SUCCESS)
2099 return mdb_set_key(leaf, key);
2103 mdb_cursor_last(MDB_cursor *cursor, MDB_val *key, MDB_val *data)
2111 cursor->mc_snum = 0;
2113 lkey.mv_size = MAXKEYSIZE+1;
2114 lkey.mv_data = NULL;
2116 rc = mdb_search_page(cursor->mc_txn, cursor->mc_dbi, &lkey, cursor, 0, &mpp);
2117 if (rc != MDB_SUCCESS)
2119 assert(IS_LEAF(mpp.mp_page));
2121 leaf = NODEPTR(mpp.mp_page, NUMKEYS(mpp.mp_page)-1);
2122 cursor->mc_initialized = 1;
2125 top = CURSOR_TOP(cursor);
2126 top->mp_ki = NUMKEYS(top->mp_page) - 1;
2129 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
2130 mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, leaf);
2131 rc = mdb_cursor_last(&cursor->mc_xcursor->mx_cursor, data, NULL);
2135 if ((rc = mdb_read_data(cursor->mc_txn, leaf, data)) != MDB_SUCCESS)
2140 return mdb_set_key(leaf, key);
2144 mdb_cursor_get(MDB_cursor *cursor, MDB_val *key, MDB_val *data,
2154 case MDB_GET_BOTH_RANGE:
2155 if (data == NULL || cursor->mc_xcursor == NULL) {
2162 if (key == NULL || key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
2164 } else if (op == MDB_SET_RANGE)
2165 rc = mdb_cursor_set(cursor, key, data, op, NULL);
2167 rc = mdb_cursor_set(cursor, key, data, op, &exact);
2171 case MDB_NEXT_NODUP:
2172 if (!cursor->mc_initialized)
2173 rc = mdb_cursor_first(cursor, key, data);
2175 rc = mdb_cursor_next(cursor, key, data, op);
2179 case MDB_PREV_NODUP:
2180 if (!cursor->mc_initialized || cursor->mc_eof)
2181 rc = mdb_cursor_last(cursor, key, data);
2183 rc = mdb_cursor_prev(cursor, key, data, op);
2186 rc = mdb_cursor_first(cursor, key, data);
2189 rc = mdb_cursor_last(cursor, key, data);
2192 DPRINTF("unhandled/unimplemented cursor operation %u", op);
2200 /* Allocate a page and initialize it
2203 mdb_new_page(MDB_txn *txn, MDB_dbi dbi, uint32_t flags, int num)
2207 if ((dp = mdb_alloc_page(txn, NULL, 0, num)) == NULL)
2209 DPRINTF("allocated new mpage %lu, page size %u",
2210 dp->p.mp_pgno, txn->mt_env->me_psize);
2211 dp->p.mp_flags = flags | P_DIRTY;
2212 dp->p.mp_lower = PAGEHDRSZ;
2213 dp->p.mp_upper = txn->mt_env->me_psize;
2215 if (IS_BRANCH(&dp->p))
2216 txn->mt_dbs[dbi].md_branch_pages++;
2217 else if (IS_LEAF(&dp->p))
2218 txn->mt_dbs[dbi].md_leaf_pages++;
2219 else if (IS_OVERFLOW(&dp->p)) {
2220 txn->mt_dbs[dbi].md_overflow_pages += num;
2221 dp->p.mp_pages = num;
2228 mdb_leaf_size(MDB_env *env, MDB_val *key, MDB_val *data)
2232 sz = LEAFSIZE(key, data);
2233 if (data->mv_size >= env->me_psize / MDB_MINKEYS) {
2234 /* put on overflow page */
2235 sz -= data->mv_size - sizeof(pgno_t);
2238 return sz + sizeof(indx_t);
2242 mdb_branch_size(MDB_env *env, MDB_val *key)
2247 if (sz >= env->me_psize / MDB_MINKEYS) {
2248 /* put on overflow page */
2249 /* not implemented */
2250 /* sz -= key->size - sizeof(pgno_t); */
2253 return sz + sizeof(indx_t);
2257 mdb_add_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp, indx_t indx,
2258 MDB_val *key, MDB_val *data, pgno_t pgno, uint8_t flags)
2261 size_t node_size = NODESIZE;
2264 MDB_dpage *ofp = NULL; /* overflow page */
2266 assert(mp->mp_upper >= mp->mp_lower);
2268 DPRINTF("add node [%.*s] to %s page %lu at index %i, key size %zu",
2269 key ? (int)key->mv_size : 0, key ? (char *)key->mv_data : NULL,
2270 IS_LEAF(mp) ? "leaf" : "branch",
2271 mp->mp_pgno, indx, key ? key->mv_size : 0);
2274 node_size += key->mv_size;
2278 if (F_ISSET(flags, F_BIGDATA)) {
2279 /* Data already on overflow page. */
2280 node_size += sizeof(pgno_t);
2281 } else if (data->mv_size >= txn->mt_env->me_psize / MDB_MINKEYS) {
2282 int ovpages = OVPAGES(data->mv_size, txn->mt_env->me_psize);
2283 /* Put data on overflow page. */
2284 DPRINTF("data size is %zu, put on overflow page",
2286 node_size += sizeof(pgno_t);
2287 if ((ofp = mdb_new_page(txn, dbi, P_OVERFLOW, ovpages)) == NULL)
2289 DPRINTF("allocated overflow page %lu", ofp->p.mp_pgno);
2292 node_size += data->mv_size;
2296 if (node_size + sizeof(indx_t) > SIZELEFT(mp)) {
2297 DPRINTF("not enough room in page %lu, got %u ptrs",
2298 mp->mp_pgno, NUMKEYS(mp));
2299 DPRINTF("upper - lower = %u - %u = %u", mp->mp_upper, mp->mp_lower,
2300 mp->mp_upper - mp->mp_lower);
2301 DPRINTF("node size = %zu", node_size);
2305 /* Move higher pointers up one slot. */
2306 for (i = NUMKEYS(mp); i > indx; i--)
2307 mp->mp_ptrs[i] = mp->mp_ptrs[i - 1];
2309 /* Adjust free space offsets. */
2310 ofs = mp->mp_upper - node_size;
2311 assert(ofs >= mp->mp_lower + sizeof(indx_t));
2312 mp->mp_ptrs[indx] = ofs;
2314 mp->mp_lower += sizeof(indx_t);
2316 /* Write the node data. */
2317 node = NODEPTR(mp, indx);
2318 node->mn_ksize = (key == NULL) ? 0 : key->mv_size;
2319 node->mn_flags = flags;
2321 node->mn_dsize = data->mv_size;
2323 node->mn_pgno = pgno;
2326 memcpy(NODEKEY(node), key->mv_data, key->mv_size);
2331 if (F_ISSET(flags, F_BIGDATA))
2332 memcpy(node->mn_data + key->mv_size, data->mv_data,
2335 memcpy(node->mn_data + key->mv_size, data->mv_data,
2338 memcpy(node->mn_data + key->mv_size, &ofp->p.mp_pgno,
2340 memcpy(METADATA(&ofp->p), data->mv_data, data->mv_size);
2348 mdb_del_node(MDB_page *mp, indx_t indx)
2351 indx_t i, j, numkeys, ptr;
2355 DPRINTF("delete node %u on %s page %lu", indx,
2356 IS_LEAF(mp) ? "leaf" : "branch", mp->mp_pgno);
2357 assert(indx < NUMKEYS(mp));
2359 node = NODEPTR(mp, indx);
2360 sz = NODESIZE + node->mn_ksize;
2362 if (F_ISSET(node->mn_flags, F_BIGDATA))
2363 sz += sizeof(pgno_t);
2365 sz += NODEDSZ(node);
2368 ptr = mp->mp_ptrs[indx];
2369 numkeys = NUMKEYS(mp);
2370 for (i = j = 0; i < numkeys; i++) {
2372 mp->mp_ptrs[j] = mp->mp_ptrs[i];
2373 if (mp->mp_ptrs[i] < ptr)
2374 mp->mp_ptrs[j] += sz;
2379 base = (char *)mp + mp->mp_upper;
2380 memmove(base + sz, base, ptr - mp->mp_upper);
2382 mp->mp_lower -= sizeof(indx_t);
2387 mdb_xcursor_init0(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx)
2392 mx->mx_txn.mt_dbxs = mx->mx_dbxs;
2393 mx->mx_txn.mt_dbs = mx->mx_dbs;
2394 mx->mx_dbxs[0] = txn->mt_dbxs[0];
2395 mx->mx_dbxs[1] = txn->mt_dbxs[1];
2397 mx->mx_dbxs[2] = txn->mt_dbxs[dbi];
2402 mx->mx_dbxs[dbn+1].md_parent = dbn;
2403 mx->mx_dbxs[dbn+1].md_cmp = mx->mx_dbxs[dbn].md_dcmp;
2404 mx->mx_dbxs[dbn+1].md_rel = mx->mx_dbxs[dbn].md_rel;
2405 mx->mx_dbxs[dbn+1].md_dirty = 0;
2406 mx->mx_txn.mt_numdbs = dbn+2;
2408 mx->mx_cursor.mc_snum = 0;
2409 mx->mx_cursor.mc_txn = &mx->mx_txn;
2410 mx->mx_cursor.mc_dbi = dbn+1;
2414 mdb_xcursor_init1(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx, MDB_node *node)
2416 MDB_db *db = NODEDATA(node);
2418 mx->mx_dbs[0] = txn->mt_dbs[0];
2419 mx->mx_dbs[1] = txn->mt_dbs[1];
2421 mx->mx_dbs[2] = txn->mt_dbs[dbi];
2426 mx->mx_dbs[dbn] = *db;
2427 mx->mx_dbxs[dbn].md_name.mv_data = NODEKEY(node);
2428 mx->mx_dbxs[dbn].md_name.mv_size = node->mn_ksize;
2429 mx->mx_txn.mt_next_pgno = txn->mt_next_pgno;
2430 mx->mx_txn.mt_oldest = txn->mt_oldest;
2431 mx->mx_txn.mt_u = txn->mt_u;
2435 mdb_xcursor_fini(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx)
2437 txn->mt_next_pgno = mx->mx_txn.mt_next_pgno;
2438 txn->mt_oldest = mx->mx_txn.mt_oldest;
2439 txn->mt_u = mx->mx_txn.mt_u;
2440 txn->mt_dbs[0] = mx->mx_dbs[0];
2441 txn->mt_dbs[1] = mx->mx_dbs[1];
2442 txn->mt_dbxs[0].md_dirty = mx->mx_dbxs[0].md_dirty;
2443 txn->mt_dbxs[1].md_dirty = mx->mx_dbxs[1].md_dirty;
2445 txn->mt_dbs[dbi] = mx->mx_dbs[2];
2446 txn->mt_dbxs[dbi].md_dirty = mx->mx_dbxs[2].md_dirty;
2451 mdb_cursor_open(MDB_txn *txn, MDB_dbi dbi, MDB_cursor **ret)
2454 size_t size = sizeof(MDB_cursor);
2456 if (txn == NULL || ret == NULL || !dbi || dbi >= txn->mt_numdbs)
2459 if (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT)
2460 size += sizeof(MDB_xcursor);
2462 if ((cursor = calloc(1, size)) != NULL) {
2463 cursor->mc_dbi = dbi;
2464 cursor->mc_txn = txn;
2465 if (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT) {
2466 MDB_xcursor *mx = (MDB_xcursor *)(cursor + 1);
2467 cursor->mc_xcursor = mx;
2468 mdb_xcursor_init0(txn, dbi, mx);
2479 /* Return the count of duplicate data items for the current key */
2481 mdb_cursor_count(MDB_cursor *mc, unsigned long *countp)
2486 if (mc == NULL || countp == NULL)
2489 if (!(mc->mc_txn->mt_dbs[mc->mc_dbi].md_flags & MDB_DUPSORT))
2492 top = CURSOR_TOP(mc);
2493 leaf = NODEPTR(top->mp_page, top->mp_ki);
2494 if (!F_ISSET(leaf->mn_flags, F_DUPDATA)) {
2497 if (!mc->mc_xcursor->mx_cursor.mc_initialized)
2500 *countp = mc->mc_xcursor->mx_txn.mt_dbs[mc->mc_xcursor->mx_cursor.mc_dbi].md_entries;
2506 mdb_cursor_close(MDB_cursor *cursor)
2508 if (cursor != NULL) {
2514 mdb_update_key(MDB_page *mp, indx_t indx, MDB_val *key)
2516 indx_t ptr, i, numkeys;
2522 node = NODEPTR(mp, indx);
2523 ptr = mp->mp_ptrs[indx];
2524 DPRINTF("update key %u (ofs %u) [%.*s] to [%.*s] on page %lu",
2526 (int)node->mn_ksize, (char *)NODEKEY(node),
2527 (int)key->mv_size, (char *)key->mv_data,
2530 delta = key->mv_size - node->mn_ksize;
2532 if (delta > 0 && SIZELEFT(mp) < delta) {
2533 DPRINTF("OUCH! Not enough room, delta = %d", delta);
2537 numkeys = NUMKEYS(mp);
2538 for (i = 0; i < numkeys; i++) {
2539 if (mp->mp_ptrs[i] <= ptr)
2540 mp->mp_ptrs[i] -= delta;
2543 base = (char *)mp + mp->mp_upper;
2544 len = ptr - mp->mp_upper + NODESIZE;
2545 memmove(base - delta, base, len);
2546 mp->mp_upper -= delta;
2548 node = NODEPTR(mp, indx);
2549 node->mn_ksize = key->mv_size;
2552 memcpy(NODEKEY(node), key->mv_data, key->mv_size);
2557 /* Move a node from src to dst.
2560 mdb_move_node(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *src, indx_t srcindx,
2561 MDB_pageparent *dst, indx_t dstindx)
2567 srcnode = NODEPTR(src->mp_page, srcindx);
2568 DPRINTF("moving %s node %u [%.*s] on page %lu to node %u on page %lu",
2569 IS_LEAF(src->mp_page) ? "leaf" : "branch",
2571 (int)srcnode->mn_ksize, (char *)NODEKEY(srcnode),
2572 src->mp_page->mp_pgno,
2573 dstindx, dst->mp_page->mp_pgno);
2575 /* Mark src and dst as dirty. */
2576 if ((rc = mdb_touch(txn, src)) ||
2577 (rc = mdb_touch(txn, dst)))
2580 /* Add the node to the destination page.
2582 key.mv_size = srcnode->mn_ksize;
2583 key.mv_data = NODEKEY(srcnode);
2584 data.mv_size = NODEDSZ(srcnode);
2585 data.mv_data = NODEDATA(srcnode);
2586 rc = mdb_add_node(txn, dbi, dst->mp_page, dstindx, &key, &data, NODEPGNO(srcnode),
2588 if (rc != MDB_SUCCESS)
2591 /* Delete the node from the source page.
2593 mdb_del_node(src->mp_page, srcindx);
2595 /* Update the parent separators.
2597 if (srcindx == 0 && src->mp_pi != 0) {
2598 DPRINTF("update separator for source page %lu to [%.*s]",
2599 src->mp_page->mp_pgno, (int)key.mv_size, (char *)key.mv_data);
2600 if ((rc = mdb_update_key(src->mp_parent, src->mp_pi,
2601 &key)) != MDB_SUCCESS)
2605 if (srcindx == 0 && IS_BRANCH(src->mp_page)) {
2607 nullkey.mv_size = 0;
2608 assert(mdb_update_key(src->mp_page, 0, &nullkey) == MDB_SUCCESS);
2611 if (dstindx == 0 && dst->mp_pi != 0) {
2612 DPRINTF("update separator for destination page %lu to [%.*s]",
2613 dst->mp_page->mp_pgno, (int)key.mv_size, (char *)key.mv_data);
2614 if ((rc = mdb_update_key(dst->mp_parent, dst->mp_pi,
2615 &key)) != MDB_SUCCESS)
2619 if (dstindx == 0 && IS_BRANCH(dst->mp_page)) {
2621 nullkey.mv_size = 0;
2622 assert(mdb_update_key(dst->mp_page, 0, &nullkey) == MDB_SUCCESS);
2629 mdb_merge(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *src, MDB_pageparent *dst)
2638 DPRINTF("merging page %lu and %lu", src->mp_page->mp_pgno, dst->mp_page->mp_pgno);
2640 assert(txn != NULL);
2641 assert(src->mp_parent); /* can't merge root page */
2642 assert(dst->mp_parent);
2644 /* Mark src and dst as dirty. */
2645 if ((rc = mdb_touch(txn, src)) ||
2646 (rc = mdb_touch(txn, dst)))
2649 /* Move all nodes from src to dst.
2651 for (i = 0; i < NUMKEYS(src->mp_page); i++) {
2652 srcnode = NODEPTR(src->mp_page, i);
2654 key.mv_size = srcnode->mn_ksize;
2655 key.mv_data = NODEKEY(srcnode);
2656 data.mv_size = NODEDSZ(srcnode);
2657 data.mv_data = NODEDATA(srcnode);
2658 rc = mdb_add_node(txn, dbi, dst->mp_page, NUMKEYS(dst->mp_page), &key,
2659 &data, NODEPGNO(srcnode), srcnode->mn_flags);
2660 if (rc != MDB_SUCCESS)
2664 DPRINTF("dst page %lu now has %u keys (%.1f%% filled)",
2665 dst->mp_page->mp_pgno, NUMKEYS(dst->mp_page), (float)PAGEFILL(txn->mt_env, dst->mp_page) / 10);
2667 /* Unlink the src page from parent.
2669 mdb_del_node(src->mp_parent, src->mp_pi);
2670 if (src->mp_pi == 0) {
2672 if ((rc = mdb_update_key(src->mp_parent, 0, &key)) != MDB_SUCCESS)
2676 if (IS_LEAF(src->mp_page))
2677 txn->mt_dbs[dbi].md_leaf_pages--;
2679 txn->mt_dbs[dbi].md_branch_pages--;
2681 mpp.mp_page = src->mp_parent;
2682 dh = (MDB_dhead *)src->mp_parent;
2684 mpp.mp_parent = dh->md_parent;
2685 mpp.mp_pi = dh->md_pi;
2687 return mdb_rebalance(txn, dbi, &mpp);
2690 #define FILL_THRESHOLD 250
2693 mdb_rebalance(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *mpp)
2698 indx_t si = 0, di = 0;
2700 assert(txn != NULL);
2701 assert(mpp != NULL);
2703 DPRINTF("rebalancing %s page %lu (has %u keys, %.1f%% full)",
2704 IS_LEAF(mpp->mp_page) ? "leaf" : "branch",
2705 mpp->mp_page->mp_pgno, NUMKEYS(mpp->mp_page), (float)PAGEFILL(txn->mt_env, mpp->mp_page) / 10);
2707 if (PAGEFILL(txn->mt_env, mpp->mp_page) >= FILL_THRESHOLD) {
2708 DPRINTF("no need to rebalance page %lu, above fill threshold",
2709 mpp->mp_page->mp_pgno);
2713 if (mpp->mp_parent == NULL) {
2714 if (NUMKEYS(mpp->mp_page) == 0) {
2715 DPUTS("tree is completely empty");
2716 txn->mt_dbs[dbi].md_root = P_INVALID;
2717 txn->mt_dbs[dbi].md_depth--;
2718 txn->mt_dbs[dbi].md_leaf_pages--;
2719 } else if (IS_BRANCH(mpp->mp_page) && NUMKEYS(mpp->mp_page) == 1) {
2720 DPUTS("collapsing root page!");
2721 txn->mt_dbs[dbi].md_root = NODEPGNO(NODEPTR(mpp->mp_page, 0));
2722 if ((root = mdb_get_page(txn, txn->mt_dbs[dbi].md_root)) == NULL)
2723 return MDB_PAGE_NOTFOUND;
2724 txn->mt_dbs[dbi].md_depth--;
2725 txn->mt_dbs[dbi].md_branch_pages--;
2727 DPUTS("root page doesn't need rebalancing");
2731 /* The parent (branch page) must have at least 2 pointers,
2732 * otherwise the tree is invalid.
2734 assert(NUMKEYS(mpp->mp_parent) > 1);
2736 /* Leaf page fill factor is below the threshold.
2737 * Try to move keys from left or right neighbor, or
2738 * merge with a neighbor page.
2743 if (mpp->mp_pi == 0) {
2744 /* We're the leftmost leaf in our parent.
2746 DPUTS("reading right neighbor");
2747 node = NODEPTR(mpp->mp_parent, mpp->mp_pi + 1);
2748 if ((npp.mp_page = mdb_get_page(txn, NODEPGNO(node))) == NULL)
2749 return MDB_PAGE_NOTFOUND;
2750 npp.mp_pi = mpp->mp_pi + 1;
2752 di = NUMKEYS(mpp->mp_page);
2754 /* There is at least one neighbor to the left.
2756 DPUTS("reading left neighbor");
2757 node = NODEPTR(mpp->mp_parent, mpp->mp_pi - 1);
2758 if ((npp.mp_page = mdb_get_page(txn, NODEPGNO(node))) == NULL)
2759 return MDB_PAGE_NOTFOUND;
2760 npp.mp_pi = mpp->mp_pi - 1;
2761 si = NUMKEYS(npp.mp_page) - 1;
2764 npp.mp_parent = mpp->mp_parent;
2766 DPRINTF("found neighbor page %lu (%u keys, %.1f%% full)",
2767 npp.mp_page->mp_pgno, NUMKEYS(npp.mp_page), (float)PAGEFILL(txn->mt_env, npp.mp_page) / 10);
2769 /* If the neighbor page is above threshold and has at least two
2770 * keys, move one key from it.
2772 * Otherwise we should try to merge them.
2774 if (PAGEFILL(txn->mt_env, npp.mp_page) >= FILL_THRESHOLD && NUMKEYS(npp.mp_page) >= 2)
2775 return mdb_move_node(txn, dbi, &npp, si, mpp, di);
2776 else { /* FIXME: if (has_enough_room()) */
2777 if (mpp->mp_pi == 0)
2778 return mdb_merge(txn, dbi, &npp, mpp);
2780 return mdb_merge(txn, dbi, mpp, &npp);
2785 mdb_del0(MDB_txn *txn, MDB_dbi dbi, unsigned int ki, MDB_pageparent *mpp, MDB_node *leaf)
2789 /* add overflow pages to free list */
2790 if (F_ISSET(leaf->mn_flags, F_BIGDATA)) {
2794 memcpy(&pg, NODEDATA(leaf), sizeof(pg));
2795 ovpages = OVPAGES(NODEDSZ(leaf), txn->mt_env->me_psize);
2796 for (i=0; i<ovpages; i++) {
2797 DPRINTF("freed ov page %lu", pg);
2798 mdb_midl_insert(txn->mt_free_pgs, pg);
2802 mdb_del_node(mpp->mp_page, ki);
2803 txn->mt_dbs[dbi].md_entries--;
2804 rc = mdb_rebalance(txn, dbi, mpp);
2805 if (rc != MDB_SUCCESS)
2806 txn->mt_flags |= MDB_TXN_ERROR;
2812 mdb_del(MDB_txn *txn, MDB_dbi dbi,
2813 MDB_val *key, MDB_val *data,
2821 DPRINTF("========> delete key %.*s", (int)key->mv_size, (char *)key->mv_data);
2823 assert(key != NULL);
2825 if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
2828 if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
2832 if (key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
2836 mpp.mp_parent = NULL;
2838 if ((rc = mdb_search_page(txn, dbi, key, NULL, 1, &mpp)) != MDB_SUCCESS)
2841 leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, &ki);
2842 if (leaf == NULL || !exact) {
2843 return MDB_NOTFOUND;
2846 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
2850 mdb_xcursor_init0(txn, dbi, &mx);
2851 mdb_xcursor_init1(txn, dbi, &mx, leaf);
2852 if (flags == MDB_DEL_DUP) {
2853 rc = mdb_del(&mx.mx_txn, mx.mx_cursor.mc_dbi, data, NULL, 0);
2854 mdb_xcursor_fini(txn, dbi, &mx);
2855 if (rc != MDB_SUCCESS)
2857 /* If sub-DB still has entries, we're done */
2858 if (mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi].md_root != P_INVALID) {
2859 memcpy(NODEDATA(leaf), &mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi],
2861 txn->mt_dbs[dbi].md_entries--;
2864 /* otherwise fall thru and delete the sub-DB */
2866 /* add all the child DB's pages to the free list */
2867 rc = mdb_search_page(&mx.mx_txn, mx.mx_cursor.mc_dbi,
2868 NULL, &mx.mx_cursor, 0, &mp2);
2869 if (rc == MDB_SUCCESS) {
2870 MDB_ppage *top, *parent;
2874 cursor_pop_page(&mx.mx_cursor);
2875 if (mx.mx_cursor.mc_snum) {
2876 top = CURSOR_TOP(&mx.mx_cursor);
2877 while (mx.mx_cursor.mc_snum > 1) {
2878 parent = CURSOR_PARENT(&mx.mx_cursor);
2879 for (i=0; i<NUMKEYS(top->mp_page); i++) {
2880 ni = NODEPTR(top->mp_page, i);
2881 mdb_midl_insert(txn->mt_free_pgs, ni->mn_pgno);
2884 if (parent->mp_ki >= NUMKEYS(parent->mp_page)) {
2885 cursor_pop_page(&mx.mx_cursor);
2888 ni = NODEPTR(parent->mp_page, parent->mp_ki);
2889 top->mp_page = mdb_get_page(&mx.mx_txn, ni->mn_pgno);
2893 mdb_midl_insert(txn->mt_free_pgs, mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi].md_root);
2898 if (data && (rc = mdb_read_data(txn, leaf, data)) != MDB_SUCCESS)
2901 return mdb_del0(txn, dbi, ki, &mpp, leaf);
2904 /* Split page <*mpp>, and insert <key,(data|newpgno)> in either left or
2905 * right sibling, at index <*newindxp> (as if unsplit). Updates *mpp and
2906 * *newindxp with the actual values after split, ie if *mpp and *newindxp
2907 * refer to a node in the new right sibling page.
2910 mdb_split(MDB_txn *txn, MDB_dbi dbi, MDB_page **mpp, unsigned int *newindxp,
2911 MDB_val *newkey, MDB_val *newdata, pgno_t newpgno)
2914 int rc = MDB_SUCCESS, ins_new = 0;
2917 unsigned int i, j, split_indx;
2919 MDB_val sepkey, rkey, rdata;
2921 MDB_dpage *mdp, *rdp, *pdp;
2924 assert(txn != NULL);
2926 dh = ((MDB_dhead *)*mpp) - 1;
2927 mdp = (MDB_dpage *)dh;
2928 newindx = *newindxp;
2930 DPRINTF("-----> splitting %s page %lu and adding [%.*s] at index %i",
2931 IS_LEAF(&mdp->p) ? "leaf" : "branch", mdp->p.mp_pgno,
2932 (int)newkey->mv_size, (char *)newkey->mv_data, *newindxp);
2934 if (mdp->h.md_parent == NULL) {
2935 if ((pdp = mdb_new_page(txn, dbi, P_BRANCH, 1)) == NULL)
2938 mdp->h.md_parent = &pdp->p;
2939 txn->mt_dbs[dbi].md_root = pdp->p.mp_pgno;
2940 DPRINTF("root split! new root = %lu", pdp->p.mp_pgno);
2941 txn->mt_dbs[dbi].md_depth++;
2943 /* Add left (implicit) pointer. */
2944 if ((rc = mdb_add_node(txn, dbi, &pdp->p, 0, NULL, NULL,
2945 mdp->p.mp_pgno, 0)) != MDB_SUCCESS)
2948 DPRINTF("parent branch page is %lu", mdp->h.md_parent->mp_pgno);
2951 /* Create a right sibling. */
2952 if ((rdp = mdb_new_page(txn, dbi, mdp->p.mp_flags, 1)) == NULL)
2954 rdp->h.md_parent = mdp->h.md_parent;
2955 rdp->h.md_pi = mdp->h.md_pi + 1;
2956 DPRINTF("new right sibling: page %lu", rdp->p.mp_pgno);
2958 /* Move half of the keys to the right sibling. */
2959 if ((copy = malloc(txn->mt_env->me_psize)) == NULL)
2961 memcpy(copy, &mdp->p, txn->mt_env->me_psize);
2962 memset(&mdp->p.mp_ptrs, 0, txn->mt_env->me_psize - PAGEHDRSZ);
2963 mdp->p.mp_lower = PAGEHDRSZ;
2964 mdp->p.mp_upper = txn->mt_env->me_psize;
2966 split_indx = NUMKEYS(copy) / 2 + 1;
2968 /* First find the separating key between the split pages.
2970 memset(&sepkey, 0, sizeof(sepkey));
2971 if (newindx == split_indx) {
2972 sepkey.mv_size = newkey->mv_size;
2973 sepkey.mv_data = newkey->mv_data;
2975 node = NODEPTR(copy, split_indx);
2976 sepkey.mv_size = node->mn_ksize;
2977 sepkey.mv_data = NODEKEY(node);
2980 DPRINTF("separator is [%.*s]", (int)sepkey.mv_size, (char *)sepkey.mv_data);
2982 /* Copy separator key to the parent.
2984 if (SIZELEFT(rdp->h.md_parent) < mdb_branch_size(txn->mt_env, &sepkey)) {
2985 rc = mdb_split(txn, dbi, &rdp->h.md_parent, &rdp->h.md_pi,
2986 &sepkey, NULL, rdp->p.mp_pgno);
2988 /* Right page might now have changed parent.
2989 * Check if left page also changed parent.
2991 if (rdp->h.md_parent != mdp->h.md_parent &&
2992 mdp->h.md_pi >= NUMKEYS(mdp->h.md_parent)) {
2993 mdp->h.md_parent = rdp->h.md_parent;
2994 mdp->h.md_pi = rdp->h.md_pi - 1;
2997 rc = mdb_add_node(txn, dbi, rdp->h.md_parent, rdp->h.md_pi,
2998 &sepkey, NULL, rdp->p.mp_pgno, 0);
3000 if (rc != MDB_SUCCESS) {
3005 for (i = j = 0; i <= NUMKEYS(copy); j++) {
3006 if (i < split_indx) {
3007 /* Re-insert in left sibling. */
3010 /* Insert in right sibling. */
3011 if (i == split_indx)
3012 /* Reset insert index for right sibling. */
3013 j = (i == newindx && ins_new);
3017 if (i == newindx && !ins_new) {
3018 /* Insert the original entry that caused the split. */
3019 rkey.mv_data = newkey->mv_data;
3020 rkey.mv_size = newkey->mv_size;
3021 if (IS_LEAF(&mdp->p)) {
3022 rdata.mv_data = newdata->mv_data;
3023 rdata.mv_size = newdata->mv_size;
3030 /* Update page and index for the new key. */
3033 } else if (i == NUMKEYS(copy)) {
3036 node = NODEPTR(copy, i);
3037 rkey.mv_data = NODEKEY(node);
3038 rkey.mv_size = node->mn_ksize;
3039 if (IS_LEAF(&mdp->p)) {
3040 rdata.mv_data = NODEDATA(node);
3041 rdata.mv_size = node->mn_dsize;
3043 pgno = node->mn_pgno;
3044 flags = node->mn_flags;
3049 if (!IS_LEAF(&mdp->p) && j == 0) {
3050 /* First branch index doesn't need key data. */
3054 rc = mdb_add_node(txn, dbi, &pdp->p, j, &rkey, &rdata, pgno,flags);
3062 mdb_put0(MDB_txn *txn, MDB_dbi dbi,
3063 MDB_val *key, MDB_val *data, unsigned int flags)
3065 int rc = MDB_SUCCESS, exact;
3069 MDB_val xdata, *rdata, dkey;
3071 char dbuf[PAGESIZE];
3074 DPRINTF("==> put key %.*s, size %zu, data size %zu",
3075 (int)key->mv_size, (char *)key->mv_data, key->mv_size, data->mv_size);
3078 mpp.mp_parent = NULL;
3080 rc = mdb_search_page(txn, dbi, key, NULL, 1, &mpp);
3081 if (rc == MDB_SUCCESS) {
3082 leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, &ki);
3083 if (leaf && exact) {
3084 if (flags == MDB_NOOVERWRITE) {
3085 DPRINTF("duplicate key %.*s",
3086 (int)key->mv_size, (char *)key->mv_data);
3087 return MDB_KEYEXIST;
3089 if (F_ISSET(txn->mt_dbs[dbi].md_flags, MDB_DUPSORT)) {
3090 /* Was a single item before, must convert now */
3091 if (!F_ISSET(leaf->mn_flags, F_DUPDATA)) {
3092 dkey.mv_size = NODEDSZ(leaf);
3093 memcpy(dbuf, NODEDATA(leaf), dkey.mv_size);
3094 memset(&dummy, 0, sizeof(dummy));
3095 dummy.md_root = P_INVALID;
3096 if (dkey.mv_size == sizeof(MDB_db)) {
3097 memcpy(NODEDATA(leaf), &dummy, sizeof(dummy));
3100 mdb_del_node(mpp.mp_page, ki);
3103 xdata.mv_size = sizeof(MDB_db);
3104 xdata.mv_data = &dummy;
3109 /* same size, just replace it */
3110 if (NODEDSZ(leaf) == data->mv_size) {
3111 memcpy(NODEDATA(leaf), data->mv_data, data->mv_size);
3114 mdb_del_node(mpp.mp_page, ki);
3116 if (leaf == NULL) { /* append if not found */
3117 ki = NUMKEYS(mpp.mp_page);
3118 DPRINTF("appending key at index %i", ki);
3120 } else if (rc == MDB_NOTFOUND) {
3122 /* new file, just write a root leaf page */
3123 DPUTS("allocating new root leaf page");
3124 if ((dp = mdb_new_page(txn, dbi, P_LEAF, 1)) == NULL) {
3127 mpp.mp_page = &dp->p;
3128 txn->mt_dbs[dbi].md_root = mpp.mp_page->mp_pgno;
3129 txn->mt_dbs[dbi].md_depth++;
3130 txn->mt_dbxs[dbi].md_dirty = 1;
3136 assert(IS_LEAF(mpp.mp_page));
3137 DPRINTF("there are %u keys, should insert new key at index %i",
3138 NUMKEYS(mpp.mp_page), ki);
3143 if (SIZELEFT(mpp.mp_page) < mdb_leaf_size(txn->mt_env, key, rdata)) {
3144 rc = mdb_split(txn, dbi, &mpp.mp_page, &ki, key, rdata, P_INVALID);
3146 /* There is room already in this leaf page. */
3147 rc = mdb_add_node(txn, dbi, mpp.mp_page, ki, key, rdata, 0, 0);
3150 if (rc != MDB_SUCCESS)
3151 txn->mt_flags |= MDB_TXN_ERROR;
3153 /* Remember if we just added a subdatabase */
3154 if (flags & F_SUBDATA) {
3155 leaf = NODEPTR(mpp.mp_page, ki);
3156 leaf->mn_flags |= F_SUBDATA;
3159 /* Now store the actual data in the child DB. Note that we're
3160 * storing the user data in the keys field, so there are strict
3161 * size limits on dupdata. The actual data fields of the child
3162 * DB are all zero size.
3167 leaf = NODEPTR(mpp.mp_page, ki);
3169 mdb_xcursor_init0(txn, dbi, &mx);
3170 mdb_xcursor_init1(txn, dbi, &mx, leaf);
3173 if (flags == MDB_NODUPDATA)
3174 flags = MDB_NOOVERWRITE;
3175 /* converted, write the original data first */
3177 dkey.mv_data = dbuf;
3178 rc = mdb_put0(&mx.mx_txn, mx.mx_cursor.mc_dbi, &dkey, &xdata, flags);
3180 leaf->mn_flags |= F_DUPDATA;
3182 rc = mdb_put0(&mx.mx_txn, mx.mx_cursor.mc_dbi, data, &xdata, flags);
3183 mdb_xcursor_fini(txn, dbi, &mx);
3184 memcpy(NODEDATA(leaf), &mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi],
3187 txn->mt_dbs[dbi].md_entries++;
3195 mdb_put(MDB_txn *txn, MDB_dbi dbi,
3196 MDB_val *key, MDB_val *data, unsigned int flags)
3198 assert(key != NULL);
3199 assert(data != NULL);
3201 if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
3204 if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
3208 if (key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
3212 if ((flags & (MDB_NOOVERWRITE|MDB_NODUPDATA)) != flags)
3215 return mdb_put0(txn, dbi, key, data, flags);
3219 mdb_env_set_flags(MDB_env *env, unsigned int flag, int onoff)
3221 #define CHANGEABLE (MDB_NOSYNC)
3222 if ((flag & CHANGEABLE) != flag)
3225 env->me_flags |= flag;
3227 env->me_flags &= ~flag;
3232 mdb_env_get_flags(MDB_env *env, unsigned int *arg)
3237 *arg = env->me_flags;
3242 mdb_env_get_path(MDB_env *env, const char **arg)
3247 *arg = env->me_path;
3252 mdb_stat0(MDB_env *env, MDB_db *db, MDB_stat *arg)
3254 arg->ms_psize = env->me_psize;
3255 arg->ms_depth = db->md_depth;
3256 arg->ms_branch_pages = db->md_branch_pages;
3257 arg->ms_leaf_pages = db->md_leaf_pages;
3258 arg->ms_overflow_pages = db->md_overflow_pages;
3259 arg->ms_entries = db->md_entries;
3264 mdb_env_stat(MDB_env *env, MDB_stat *arg)
3266 if (env == NULL || arg == NULL)
3269 return mdb_stat0(env, &env->me_meta->mm_dbs[MAIN_DBI], arg);
3272 int mdb_open(MDB_txn *txn, const char *name, unsigned int flags, MDB_dbi *dbi)
3282 if (flags & (MDB_DUPSORT|MDB_REVERSEKEY|MDB_INTEGERKEY))
3283 txn->mt_dbs[MAIN_DBI].md_flags |= (flags & (MDB_DUPSORT|MDB_REVERSEKEY|MDB_INTEGERKEY));
3287 /* Is the DB already open? */
3289 for (i=2; i<txn->mt_numdbs; i++) {
3290 if (len == txn->mt_dbxs[i].md_name.mv_size &&
3291 !strncmp(name, txn->mt_dbxs[i].md_name.mv_data, len)) {
3297 if (txn->mt_numdbs >= txn->mt_env->me_maxdbs - 1)
3300 /* Find the DB info */
3302 key.mv_data = (void *)name;
3303 rc = mdb_get(txn, MAIN_DBI, &key, &data);
3305 /* Create if requested */
3306 if (rc == MDB_NOTFOUND && (flags & MDB_CREATE)) {
3308 data.mv_size = sizeof(MDB_db);
3309 data.mv_data = &dummy;
3310 memset(&dummy, 0, sizeof(dummy));
3311 dummy.md_root = P_INVALID;
3312 dummy.md_flags = flags & 0xffff;
3313 rc = mdb_put0(txn, MAIN_DBI, &key, &data, F_SUBDATA);
3317 /* OK, got info, add to table */
3318 if (rc == MDB_SUCCESS) {
3319 txn->mt_dbxs[txn->mt_numdbs].md_name.mv_data = strdup(name);
3320 txn->mt_dbxs[txn->mt_numdbs].md_name.mv_size = len;
3321 txn->mt_dbxs[txn->mt_numdbs].md_cmp = NULL;
3322 txn->mt_dbxs[txn->mt_numdbs].md_dcmp = NULL;
3323 txn->mt_dbxs[txn->mt_numdbs].md_rel = NULL;
3324 txn->mt_dbxs[txn->mt_numdbs].md_parent = MAIN_DBI;
3325 txn->mt_dbxs[txn->mt_numdbs].md_dirty = dirty;
3326 memcpy(&txn->mt_dbs[txn->mt_numdbs], data.mv_data, sizeof(MDB_db));
3327 *dbi = txn->mt_numdbs;
3328 txn->mt_env->me_dbs[0][txn->mt_numdbs] = txn->mt_dbs[txn->mt_numdbs];
3329 txn->mt_env->me_dbs[1][txn->mt_numdbs] = txn->mt_dbs[txn->mt_numdbs];
3336 int mdb_stat(MDB_txn *txn, MDB_dbi dbi, MDB_stat *arg)
3338 if (txn == NULL || arg == NULL || dbi >= txn->mt_numdbs)
3341 return mdb_stat0(txn->mt_env, &txn->mt_dbs[dbi], arg);
3344 void mdb_close(MDB_txn *txn, MDB_dbi dbi)
3347 if (dbi <= MAIN_DBI || dbi >= txn->mt_numdbs)
3349 ptr = txn->mt_dbxs[dbi].md_name.mv_data;
3350 txn->mt_dbxs[dbi].md_name.mv_data = NULL;
3351 txn->mt_dbxs[dbi].md_name.mv_size = 0;
3355 int mdb_set_compare(MDB_txn *txn, MDB_dbi dbi, MDB_cmp_func *cmp)
3357 if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
3360 txn->mt_dbxs[dbi].md_cmp = cmp;
3364 int mdb_set_dupsort(MDB_txn *txn, MDB_dbi dbi, MDB_cmp_func *cmp)
3366 if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
3369 txn->mt_dbxs[dbi].md_dcmp = cmp;
3373 int mdb_set_relfunc(MDB_txn *txn, MDB_dbi dbi, MDB_rel_func *rel)
3375 if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
3378 txn->mt_dbxs[dbi].md_rel = rel;