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 MDB_dpage *me_dpages;
343 pgno_t me_free_pgs[MDB_IDL_UM_SIZE];
344 MIDL2 me_dirty_list[MDB_IDL_DB_SIZE];
347 #define NODESIZE offsetof(MDB_node, mn_data)
349 #define INDXSIZE(k) (NODESIZE + ((k) == NULL ? 0 : (k)->mv_size))
350 #define LEAFSIZE(k, d) (NODESIZE + (k)->mv_size + (d)->mv_size)
351 #define NODEPTR(p, i) ((MDB_node *)((char *)(p) + (p)->mp_ptrs[i]))
352 #define NODEKEY(node) (void *)((node)->mn_data)
353 #define NODEDATA(node) (void *)((char *)(node)->mn_data + (node)->mn_ksize)
354 #define NODEPGNO(node) ((node)->mn_pgno)
355 #define NODEDSZ(node) ((node)->mn_dsize)
357 #define MDB_COMMIT_PAGES 64 /* max number of pages to write in one commit */
359 static int mdb_search_page_root(MDB_txn *txn,
360 MDB_dbi dbi, MDB_val *key,
361 MDB_cursor *cursor, int modify,
362 MDB_pageparent *mpp);
363 static int mdb_search_page(MDB_txn *txn,
364 MDB_dbi dbi, MDB_val *key,
365 MDB_cursor *cursor, int modify,
366 MDB_pageparent *mpp);
368 static int mdb_env_read_header(MDB_env *env, MDB_meta *meta);
369 static int mdb_env_read_meta(MDB_env *env, int *which);
370 static int mdb_env_write_meta(MDB_txn *txn);
371 static MDB_page *mdb_get_page(MDB_txn *txn, pgno_t pgno);
373 static MDB_node *mdb_search_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp,
374 MDB_val *key, int *exactp, unsigned int *kip);
375 static int mdb_add_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp,
376 indx_t indx, MDB_val *key, MDB_val *data,
377 pgno_t pgno, uint8_t flags);
378 static void mdb_del_node(MDB_page *mp, indx_t indx);
379 static int mdb_del0(MDB_txn *txn, MDB_dbi dbi, unsigned int ki,
380 MDB_pageparent *mpp, MDB_node *leaf);
381 static int mdb_put0(MDB_txn *txn, MDB_dbi dbi,
382 MDB_val *key, MDB_val *data, unsigned int flags);
383 static int mdb_read_data(MDB_txn *txn, MDB_node *leaf, MDB_val *data);
385 static int mdb_rebalance(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *mp);
386 static int mdb_update_key(MDB_page *mp, indx_t indx, MDB_val *key);
387 static int mdb_move_node(MDB_txn *txn, MDB_dbi dbi,
388 MDB_pageparent *src, indx_t srcindx,
389 MDB_pageparent *dst, indx_t dstindx);
390 static int mdb_merge(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *src,
391 MDB_pageparent *dst);
392 static int mdb_split(MDB_txn *txn, MDB_dbi dbi, MDB_page **mpp,
393 unsigned int *newindxp, MDB_val *newkey,
394 MDB_val *newdata, pgno_t newpgno);
395 static MDB_dpage *mdb_new_page(MDB_txn *txn, MDB_dbi dbi, uint32_t flags, int num);
397 static void cursor_pop_page(MDB_cursor *cursor);
398 static MDB_ppage *cursor_push_page(MDB_cursor *cursor,
401 static int mdb_set_key(MDB_node *node, MDB_val *key);
402 static int mdb_sibling(MDB_cursor *cursor, int move_right);
403 static int mdb_cursor_next(MDB_cursor *cursor,
404 MDB_val *key, MDB_val *data, MDB_cursor_op op);
405 static int mdb_cursor_prev(MDB_cursor *cursor,
406 MDB_val *key, MDB_val *data, MDB_cursor_op op);
407 static int mdb_cursor_set(MDB_cursor *cursor,
408 MDB_val *key, MDB_val *data, MDB_cursor_op op, int *exactp);
409 static int mdb_cursor_first(MDB_cursor *cursor,
410 MDB_val *key, MDB_val *data);
411 static int mdb_cursor_last(MDB_cursor *cursor,
412 MDB_val *key, MDB_val *data);
414 static void mdb_xcursor_init0(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx);
415 static void mdb_xcursor_init1(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx, MDB_node *node);
416 static void mdb_xcursor_fini(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx);
418 static size_t mdb_leaf_size(MDB_env *env, MDB_val *key,
420 static size_t mdb_branch_size(MDB_env *env, MDB_val *key);
422 static int memncmp(const void *s1, size_t n1,
423 const void *s2, size_t n2);
424 static int memnrcmp(const void *s1, size_t n1,
425 const void *s2, size_t n2);
428 memncmp(const void *s1, size_t n1, const void *s2, size_t n2)
430 int diff, len_diff = -1;
433 len_diff = (n1 > n2);
436 diff = memcmp(s1, s2, n1);
437 return diff ? diff : len_diff;
441 memnrcmp(const void *s1, size_t n1, const void *s2, size_t n2)
443 const unsigned char *p1, *p2, *p1_lim;
450 p1 = (const unsigned char *)s1 + n1 - 1;
451 p2 = (const unsigned char *)s2 + n2 - 1;
453 for (p1_lim = (n1 <= n2 ? s1 : s2); *p1 == *p2; p1--, p2--) {
455 return (p1 != s1) ? (p1 != p2) : (p2 != s2) ? -1 : 0;
461 mdb_version(int *maj, int *min, int *pat)
463 *maj = MDB_VERSION_MAJOR;
464 *min = MDB_VERSION_MINOR;
465 *pat = MDB_VERSION_PATCH;
466 return MDB_VERSION_STRING;
469 static char *const errstr[] = {
470 "MDB_KEYEXIST: Key/data pair already exists",
471 "MDB_NOTFOUND: No matching key/data pair found",
472 "MDB_PAGE_NOTFOUND: Requested page not found",
473 "MDB_CORRUPTED: Located page was wrong type",
474 "MDB_PANIC: Update of meta page failed",
475 "MDB_VERSION_MISMATCH: Database environment version mismatch"
479 mdb_strerror(int err)
482 return ("Successful return: 0");
484 if (err >= MDB_KEYEXIST && err <= MDB_VERSION_MISMATCH)
485 return errstr[err - MDB_KEYEXIST];
487 return strerror(err);
491 mdb_cmp(MDB_txn *txn, MDB_dbi dbi, const MDB_val *a, const MDB_val *b)
493 if (txn->mt_dbxs[dbi].md_cmp)
494 return txn->mt_dbxs[dbi].md_cmp(a, b);
496 if (txn->mt_dbs[dbi].md_flags & (MDB_REVERSEKEY
497 #if __BYTE_ORDER == __LITTLE_ENDIAN
501 return memnrcmp(a->mv_data, a->mv_size, b->mv_data, b->mv_size);
503 return memncmp((char *)a->mv_data, a->mv_size, b->mv_data, b->mv_size);
507 mdb_dcmp(MDB_txn *txn, MDB_dbi dbi, const MDB_val *a, const MDB_val *b)
509 if (txn->mt_dbxs[dbi].md_dcmp)
510 return txn->mt_dbxs[dbi].md_dcmp(a, b);
512 return memncmp((char *)a->mv_data, a->mv_size, b->mv_data, b->mv_size);
515 /* Allocate new page(s) for writing */
517 mdb_alloc_page(MDB_txn *txn, MDB_page *parent, unsigned int parent_idx, int num)
520 pgno_t pgno = P_INVALID;
524 if (txn->mt_txnid > 2) {
526 oldest = txn->mt_txnid - 2;
527 if (!txn->mt_env->me_pghead && txn->mt_dbs[FREE_DBI].md_root != P_INVALID) {
528 /* See if there's anything in the free DB */
533 mpp.mp_parent = NULL;
535 mdb_search_page(txn, FREE_DBI, NULL, NULL, 0, &mpp);
536 leaf = NODEPTR(mpp.mp_page, 0);
537 kptr = (ULONG *)NODEKEY(leaf);
539 /* It's potentially usable, unless there are still
540 * older readers outstanding. Grab it.
542 if (oldest > *kptr) {
547 mdb_read_data(txn, leaf, &data);
548 idl = (ULONG *)data.mv_data;
549 mop = malloc(sizeof(MDB_oldpages) + MDB_IDL_SIZEOF(idl) - sizeof(pgno_t));
550 mop->mo_next = txn->mt_env->me_pghead;
551 mop->mo_txnid = *kptr;
552 txn->mt_env->me_pghead = mop;
553 memcpy(mop->mo_pages, idl, MDB_IDL_SIZEOF(idl));
558 DPRINTF("IDL read txn %lu root %lu num %lu",
559 mop->mo_txnid, txn->mt_dbs[FREE_DBI].md_root, idl[0]);
560 for (i=0; i<idl[0]; i++) {
561 DPRINTF("IDL %lu", idl[i+1]);
565 /* drop this IDL from the DB */
566 mpp.mp_parent = NULL;
568 mdb_search_page(txn, FREE_DBI, NULL, NULL, 1, &mpp);
569 leaf = NODEPTR(mpp.mp_page, 0);
570 mdb_del0(txn, FREE_DBI, 0, &mpp, leaf);
573 if (txn->mt_env->me_pghead) {
575 for (i=0; i<txn->mt_env->me_txns->mti_numreaders; i++) {
576 ULONG mr = txn->mt_env->me_txns->mti_readers[i].mr_txnid;
579 oldest = txn->mt_env->me_txns->mti_readers[i].mr_txnid;
581 if (oldest > txn->mt_env->me_pghead->mo_txnid) {
582 MDB_oldpages *mop = txn->mt_env->me_pghead;
583 txn->mt_oldest = oldest;
585 /* FIXME: For now, always use fresh pages. We
586 * really ought to search the free list for a
591 /* peel pages off tail, so we only have to truncate the list */
592 pgno = MDB_IDL_LAST(mop->mo_pages);
593 if (MDB_IDL_IS_RANGE(mop->mo_pages)) {
595 if (mop->mo_pages[2] > mop->mo_pages[1])
596 mop->mo_pages[0] = 0;
600 if (MDB_IDL_IS_ZERO(mop->mo_pages)) {
601 txn->mt_env->me_pghead = mop->mo_next;
609 if (pgno == P_INVALID) {
610 /* DB size is maxed out */
611 if (txn->mt_next_pgno + num >= txn->mt_env->me_maxpg)
614 if (txn->mt_env->me_dpages && num == 1) {
615 dp = txn->mt_env->me_dpages;
616 txn->mt_env->me_dpages = (MDB_dpage *)dp->h.md_parent;
618 if ((dp = malloc(txn->mt_env->me_psize * num + sizeof(MDB_dhead))) == NULL)
622 dp->h.md_parent = parent;
623 dp->h.md_pi = parent_idx;
624 if (pgno == P_INVALID) {
625 dp->p.mp_pgno = txn->mt_next_pgno;
626 txn->mt_next_pgno += num;
628 dp->p.mp_pgno = pgno;
630 mid.mid = dp->p.mp_pgno;
632 mdb_midl2_insert(txn->mt_u.dirty_list, &mid);
637 /* Touch a page: make it dirty and re-insert into tree with updated pgno.
640 mdb_touch(MDB_txn *txn, MDB_pageparent *pp)
642 MDB_page *mp = pp->mp_page;
647 if (!F_ISSET(mp->mp_flags, P_DIRTY)) {
649 if ((dp = mdb_alloc_page(txn, pp->mp_parent, pp->mp_pi, 1)) == NULL)
651 DPRINTF("touched page %lu -> %lu", mp->mp_pgno, dp->p.mp_pgno);
652 mdb_midl_insert(txn->mt_free_pgs, mp->mp_pgno);
653 pgno = dp->p.mp_pgno;
654 memcpy(&dp->p, mp, txn->mt_env->me_psize);
657 mp->mp_flags |= P_DIRTY;
659 /* Update the page number to new touched page. */
660 if (pp->mp_parent != NULL)
661 NODEPGNO(NODEPTR(pp->mp_parent, pp->mp_pi)) = mp->mp_pgno;
668 mdb_env_sync(MDB_env *env, int force)
671 if (force || !F_ISSET(env->me_flags, MDB_NOSYNC)) {
672 if (fdatasync(env->me_fd))
679 mdb_txn_begin(MDB_env *env, int rdonly, MDB_txn **ret)
684 if (env->me_flags & MDB_FATAL_ERROR) {
685 DPUTS("mdb_txn_begin: environment had fatal error, must shutdown!");
688 if ((txn = calloc(1, sizeof(MDB_txn))) == NULL) {
689 DPRINTF("calloc: %s", strerror(errno));
694 txn->mt_flags |= MDB_TXN_RDONLY;
696 txn->mt_u.dirty_list = env->me_dirty_list;
697 txn->mt_u.dirty_list[0].mid = 0;
698 txn->mt_free_pgs = env->me_free_pgs;
699 txn->mt_free_pgs[0] = 0;
701 pthread_mutex_lock(&env->me_txns->mti_wmutex);
702 env->me_txns->mti_txnid++;
705 txn->mt_txnid = env->me_txns->mti_txnid;
707 MDB_reader *r = pthread_getspecific(env->me_txkey);
710 pthread_mutex_lock(&env->me_txns->mti_mutex);
711 for (i=0; i<env->me_txns->mti_numreaders; i++)
712 if (env->me_txns->mti_readers[i].mr_pid == 0)
714 if (i == env->me_maxreaders) {
715 pthread_mutex_unlock(&env->me_txns->mti_mutex);
718 env->me_txns->mti_readers[i].mr_pid = getpid();
719 env->me_txns->mti_readers[i].mr_tid = pthread_self();
720 r = &env->me_txns->mti_readers[i];
721 pthread_setspecific(env->me_txkey, r);
722 if (i >= env->me_txns->mti_numreaders)
723 env->me_txns->mti_numreaders = i+1;
724 pthread_mutex_unlock(&env->me_txns->mti_mutex);
726 r->mr_txnid = txn->mt_txnid;
727 txn->mt_u.reader = r;
734 toggle = env->me_txns->mti_me_toggle;
735 if ((rc = mdb_env_read_meta(env, &toggle)) != MDB_SUCCESS) {
740 /* Copy the DB arrays */
741 txn->mt_numdbs = env->me_numdbs;
742 txn->mt_dbxs = env->me_dbxs; /* mostly static anyway */
743 txn->mt_dbs = malloc(env->me_maxdbs * sizeof(MDB_db));
744 memcpy(txn->mt_dbs, env->me_meta->mm_dbs, 2 * sizeof(MDB_db));
745 if (txn->mt_numdbs > 2)
746 memcpy(txn->mt_dbs+2, env->me_dbs[env->me_db_toggle]+2,
747 (txn->mt_numdbs - 2) * sizeof(MDB_db));
751 txn->mt_flags |= MDB_TXN_METOGGLE;
752 txn->mt_next_pgno = env->me_meta->mm_last_pg+1;
755 DPRINTF("begin transaction %lu on mdbenv %p, root page %lu",
756 txn->mt_txnid, (void *) env, txn->mt_dbs[MAIN_DBI].md_root);
763 mdb_txn_abort(MDB_txn *txn)
771 DPRINTF("abort transaction %lu on mdbenv %p, root page %lu",
772 txn->mt_txnid, (void *) env, txn->mt_dbs[MAIN_DBI].md_root);
776 if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
777 txn->mt_u.reader->mr_txnid = 0;
783 /* return all dirty pages to dpage list */
784 for (i=1; i<=txn->mt_u.dirty_list[0].mid; i++) {
785 dp = txn->mt_u.dirty_list[i].mptr;
786 if (dp->h.md_num == 1) {
787 dp->h.md_parent = (MDB_page *)txn->mt_env->me_dpages;
788 txn->mt_env->me_dpages = dp;
790 /* large pages just get freed directly */
795 while ((mop = txn->mt_env->me_pghead)) {
796 txn->mt_env->me_pghead = mop->mo_next;
801 env->me_txns->mti_txnid--;
802 for (i=2; i<env->me_numdbs; i++)
803 env->me_dbxs[i].md_dirty = 0;
804 pthread_mutex_unlock(&env->me_txns->mti_wmutex);
811 mdb_txn_commit(MDB_txn *txn)
820 struct iovec iov[MDB_COMMIT_PAGES];
823 assert(txn->mt_env != NULL);
827 if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
832 if (txn != env->me_txn) {
833 DPUTS("attempt to commit unknown transaction");
838 if (F_ISSET(txn->mt_flags, MDB_TXN_ERROR)) {
839 DPUTS("error flag is set, can't commit");
844 if (!txn->mt_u.dirty_list[0].mid)
847 DPRINTF("committing transaction %lu on mdbenv %p, root page %lu",
848 txn->mt_txnid, (void *) env, txn->mt_dbs[MAIN_DBI].md_root);
850 /* should only be one record now */
851 if (env->me_pghead) {
855 mop = env->me_pghead;
856 key.mv_size = sizeof(pgno_t);
857 key.mv_data = (char *)&mop->mo_txnid;
858 data.mv_size = MDB_IDL_SIZEOF(mop->mo_pages);
859 data.mv_data = mop->mo_pages;
860 mdb_put0(txn, FREE_DBI, &key, &data, 0);
861 free(env->me_pghead);
862 env->me_pghead = NULL;
864 /* save to free list */
865 if (!MDB_IDL_IS_ZERO(txn->mt_free_pgs)) {
869 /* make sure last page of freeDB is touched and on freelist */
870 key.mv_size = MAXKEYSIZE+1;
872 mpp.mp_parent = NULL;
874 mdb_search_page(txn, FREE_DBI, &key, NULL, 1, &mpp);
879 ULONG *idl = txn->mt_free_pgs;
880 DPRINTF("IDL write txn %lu root %lu num %lu",
881 txn->mt_txnid, txn->mt_dbs[FREE_DBI].md_root, idl[0]);
882 for (i=0; i<idl[0]; i++) {
883 DPRINTF("IDL %lu", idl[i+1]);
887 /* write to last page of freeDB */
888 key.mv_size = sizeof(pgno_t);
889 key.mv_data = (char *)&txn->mt_txnid;
890 data.mv_size = MDB_IDL_SIZEOF(txn->mt_free_pgs);
891 data.mv_data = txn->mt_free_pgs;
892 mdb_put0(txn, FREE_DBI, &key, &data, 0);
895 /* Update DB root pointers. Their pages have already been
896 * touched so this is all in-place and cannot fail.
900 data.mv_size = sizeof(MDB_db);
902 for (i = 2; i < txn->mt_numdbs; i++) {
903 if (txn->mt_dbxs[i].md_dirty) {
904 data.mv_data = &txn->mt_dbs[i];
905 mdb_put0(txn, MAIN_DBI, &txn->mt_dbxs[i].md_name, &data, 0);
910 /* Commit up to MDB_COMMIT_PAGES dirty pages to disk until done.
918 for (; i<=txn->mt_u.dirty_list[0].mid; i++) {
919 dp = txn->mt_u.dirty_list[i].mptr;
920 if (dp->p.mp_pgno != next) {
922 DPRINTF("committing %u dirty pages", n);
923 rc = writev(env->me_fd, iov, n);
927 DPUTS("short write, filesystem full?");
929 DPRINTF("writev: %s", strerror(errno));
936 lseek(env->me_fd, dp->p.mp_pgno * env->me_psize, SEEK_SET);
937 next = dp->p.mp_pgno;
939 DPRINTF("committing page %lu", dp->p.mp_pgno);
940 iov[n].iov_len = env->me_psize * dp->h.md_num;
941 iov[n].iov_base = &dp->p;
942 size += iov[n].iov_len;
943 next = dp->p.mp_pgno + dp->h.md_num;
944 /* clear dirty flag */
945 dp->p.mp_flags &= ~P_DIRTY;
946 if (++n >= MDB_COMMIT_PAGES) {
955 DPRINTF("committing %u dirty pages", n);
956 rc = writev(env->me_fd, iov, n);
960 DPUTS("short write, filesystem full?");
962 DPRINTF("writev: %s", strerror(errno));
969 /* Drop the dirty pages.
971 for (i=1; i<=txn->mt_u.dirty_list[0].mid; i++) {
972 dp = txn->mt_u.dirty_list[i].mptr;
973 if (dp->h.md_num == 1) {
974 dp->h.md_parent = (MDB_page *)txn->mt_env->me_dpages;
975 txn->mt_env->me_dpages = dp;
979 txn->mt_u.dirty_list[i].mid = 0;
982 if ((n = mdb_env_sync(env, 0)) != 0 ||
983 (n = mdb_env_write_meta(txn)) != MDB_SUCCESS) {
990 /* update the DB tables */
992 int toggle = !env->me_db_toggle;
994 for (i = 2; i < env->me_numdbs; i++) {
995 if (txn->mt_dbxs[i].md_dirty) {
996 env->me_dbs[toggle][i] = txn->mt_dbs[i];
997 txn->mt_dbxs[i].md_dirty = 0;
1000 for (i = env->me_numdbs; i < txn->mt_numdbs; i++) {
1001 txn->mt_dbxs[i].md_dirty = 0;
1002 env->me_dbxs[i] = txn->mt_dbxs[i];
1003 env->me_dbs[toggle][i] = txn->mt_dbs[i];
1005 env->me_db_toggle = toggle;
1006 env->me_numdbs = txn->mt_numdbs;
1011 pthread_mutex_unlock(&env->me_txns->mti_wmutex);
1018 mdb_env_read_header(MDB_env *env, MDB_meta *meta)
1020 char page[PAGESIZE];
1025 assert(env != NULL);
1027 /* We don't know the page size yet, so use a minimum value.
1030 if ((rc = pread(env->me_fd, page, PAGESIZE, 0)) == 0) {
1032 } else if (rc != PAGESIZE) {
1035 DPRINTF("read: %s", strerror(errno));
1039 p = (MDB_page *)page;
1041 if (!F_ISSET(p->mp_flags, P_META)) {
1042 DPRINTF("page %lu not a meta page", p->mp_pgno);
1047 if (m->mm_magic != MDB_MAGIC) {
1048 DPUTS("meta has invalid magic");
1052 if (m->mm_version != MDB_VERSION) {
1053 DPRINTF("database is version %u, expected version %u",
1054 m->mm_version, MDB_VERSION);
1055 return MDB_VERSION_MISMATCH;
1058 memcpy(meta, m, sizeof(*m));
1063 mdb_env_init_meta(MDB_env *env, MDB_meta *meta)
1070 DPUTS("writing new meta page");
1071 psize = sysconf(_SC_PAGE_SIZE);
1073 meta->mm_magic = MDB_MAGIC;
1074 meta->mm_version = MDB_VERSION;
1075 meta->mm_psize = psize;
1076 meta->mm_last_pg = 1;
1077 meta->mm_flags = env->me_flags & 0xffff;
1078 meta->mm_flags |= MDB_INTEGERKEY;
1079 meta->mm_dbs[0].md_root = P_INVALID;
1080 meta->mm_dbs[1].md_root = P_INVALID;
1082 p = calloc(2, psize);
1084 p->mp_flags = P_META;
1087 memcpy(m, meta, sizeof(*meta));
1089 q = (MDB_page *)((char *)p + psize);
1092 q->mp_flags = P_META;
1095 memcpy(m, meta, sizeof(*meta));
1097 rc = write(env->me_fd, p, psize * 2);
1099 return (rc == (int)psize * 2) ? MDB_SUCCESS : errno;
1103 mdb_env_write_meta(MDB_txn *txn)
1106 MDB_meta meta, metab;
1108 int rc, len, toggle;
1111 assert(txn != NULL);
1112 assert(txn->mt_env != NULL);
1114 toggle = !F_ISSET(txn->mt_flags, MDB_TXN_METOGGLE);
1115 DPRINTF("writing meta page %d for root page %lu",
1116 toggle, txn->mt_dbs[MAIN_DBI].md_root);
1120 metab.mm_txnid = env->me_metas[toggle]->mm_txnid;
1121 metab.mm_last_pg = env->me_metas[toggle]->mm_last_pg;
1123 ptr = (char *)&meta;
1124 off = offsetof(MDB_meta, mm_dbs[0].md_depth);
1125 len = sizeof(MDB_meta) - off;
1128 meta.mm_dbs[0] = txn->mt_dbs[0];
1129 meta.mm_dbs[1] = txn->mt_dbs[1];
1130 meta.mm_last_pg = txn->mt_next_pgno - 1;
1131 meta.mm_txnid = txn->mt_txnid;
1134 off += env->me_psize;
1137 /* Write to the SYNC fd */
1138 rc = pwrite(env->me_mfd, ptr, len, off);
1140 DPUTS("write failed, disk error?");
1141 /* On a failure, the pagecache still contains the new data.
1142 * Write some old data back, to prevent it from being used.
1143 * Use the non-SYNC fd; we know it will fail anyway.
1145 meta.mm_last_pg = metab.mm_last_pg;
1146 meta.mm_txnid = metab.mm_txnid;
1147 rc = pwrite(env->me_fd, ptr, len, off);
1148 env->me_flags |= MDB_FATAL_ERROR;
1151 txn->mt_env->me_txns->mti_me_toggle = toggle;
1157 mdb_env_read_meta(MDB_env *env, int *which)
1161 assert(env != NULL);
1165 else if (env->me_metas[0]->mm_txnid < env->me_metas[1]->mm_txnid)
1168 if (env->me_meta != env->me_metas[toggle])
1169 env->me_meta = env->me_metas[toggle];
1171 DPRINTF("Using meta page %d", toggle);
1177 mdb_env_create(MDB_env **env)
1181 e = calloc(1, sizeof(MDB_env));
1182 if (!e) return ENOMEM;
1184 e->me_maxreaders = DEFAULT_READERS;
1194 mdb_env_set_mapsize(MDB_env *env, size_t size)
1198 env->me_mapsize = size;
1203 mdb_env_set_maxdbs(MDB_env *env, int dbs)
1205 env->me_maxdbs = dbs;
1210 mdb_env_set_maxreaders(MDB_env *env, int readers)
1212 env->me_maxreaders = readers;
1217 mdb_env_get_maxreaders(MDB_env *env, int *readers)
1219 if (!env || !readers)
1221 *readers = env->me_maxreaders;
1226 mdb_env_open2(MDB_env *env, unsigned int flags)
1232 env->me_flags = flags;
1234 memset(&meta, 0, sizeof(meta));
1236 if ((i = mdb_env_read_header(env, &meta)) != 0) {
1239 DPUTS("new mdbenv");
1243 if (!env->me_mapsize) {
1244 env->me_mapsize = newenv ? DEFAULT_MAPSIZE : meta.mm_mapsize;
1248 if (meta.mm_address && (flags & MDB_FIXEDMAP))
1250 env->me_map = mmap(meta.mm_address, env->me_mapsize, PROT_READ, i,
1252 if (env->me_map == MAP_FAILED)
1256 meta.mm_mapsize = env->me_mapsize;
1257 if (flags & MDB_FIXEDMAP)
1258 meta.mm_address = env->me_map;
1259 i = mdb_env_init_meta(env, &meta);
1260 if (i != MDB_SUCCESS) {
1261 munmap(env->me_map, env->me_mapsize);
1265 env->me_psize = meta.mm_psize;
1267 env->me_maxpg = env->me_mapsize / env->me_psize;
1269 p = (MDB_page *)env->me_map;
1270 env->me_metas[0] = METADATA(p);
1271 env->me_metas[1] = (MDB_meta *)((char *)env->me_metas[0] + meta.mm_psize);
1273 if ((i = mdb_env_read_meta(env, NULL)) != 0)
1276 DPRINTF("opened database version %u, pagesize %u",
1277 env->me_meta->mm_version, env->me_psize);
1278 DPRINTF("depth: %u", env->me_meta->mm_dbs[MAIN_DBI].md_depth);
1279 DPRINTF("entries: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_entries);
1280 DPRINTF("branch pages: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_branch_pages);
1281 DPRINTF("leaf pages: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_leaf_pages);
1282 DPRINTF("overflow pages: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_overflow_pages);
1283 DPRINTF("root: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_root);
1289 mdb_env_reader_dest(void *ptr)
1291 MDB_reader *reader = ptr;
1293 reader->mr_txnid = 0;
1298 /* downgrade the exclusive lock on the region back to shared */
1300 mdb_env_share_locks(MDB_env *env)
1302 struct flock lock_info;
1304 env->me_txns->mti_txnid = env->me_meta->mm_txnid;
1305 if (env->me_metas[0]->mm_txnid < env->me_metas[1]->mm_txnid)
1306 env->me_txns->mti_me_toggle = 1;
1308 memset((void *)&lock_info, 0, sizeof(lock_info));
1309 lock_info.l_type = F_RDLCK;
1310 lock_info.l_whence = SEEK_SET;
1311 lock_info.l_start = 0;
1312 lock_info.l_len = 1;
1313 fcntl(env->me_lfd, F_SETLK, &lock_info);
1317 mdb_env_setup_locks(MDB_env *env, char *lpath, int mode, int *excl)
1321 struct flock lock_info;
1325 if ((env->me_lfd = open(lpath, O_RDWR|O_CREAT, mode)) == -1) {
1329 /* Try to get exclusive lock. If we succeed, then
1330 * nobody is using the lock region and we should initialize it.
1332 memset((void *)&lock_info, 0, sizeof(lock_info));
1333 lock_info.l_type = F_WRLCK;
1334 lock_info.l_whence = SEEK_SET;
1335 lock_info.l_start = 0;
1336 lock_info.l_len = 1;
1337 rc = fcntl(env->me_lfd, F_SETLK, &lock_info);
1341 lock_info.l_type = F_RDLCK;
1342 rc = fcntl(env->me_lfd, F_SETLK, &lock_info);
1348 size = lseek(env->me_lfd, 0, SEEK_END);
1349 rsize = (env->me_maxreaders-1) * sizeof(MDB_reader) + sizeof(MDB_txninfo);
1350 if (size < rsize && *excl) {
1351 if (ftruncate(env->me_lfd, rsize) != 0) {
1357 size = rsize - sizeof(MDB_txninfo);
1358 env->me_maxreaders = size/sizeof(MDB_reader) + 1;
1360 env->me_txns = mmap(0, rsize, PROT_READ|PROT_WRITE, MAP_SHARED,
1362 if (env->me_txns == MAP_FAILED) {
1367 pthread_mutexattr_t mattr;
1369 pthread_mutexattr_init(&mattr);
1370 rc = pthread_mutexattr_setpshared(&mattr, PTHREAD_PROCESS_SHARED);
1374 pthread_mutex_init(&env->me_txns->mti_mutex, &mattr);
1375 pthread_mutex_init(&env->me_txns->mti_wmutex, &mattr);
1376 env->me_txns->mti_version = MDB_VERSION;
1377 env->me_txns->mti_magic = MDB_MAGIC;
1378 env->me_txns->mti_txnid = 0;
1379 env->me_txns->mti_numreaders = 0;
1380 env->me_txns->mti_me_toggle = 0;
1383 if (env->me_txns->mti_magic != MDB_MAGIC) {
1384 DPUTS("lock region has invalid magic");
1388 if (env->me_txns->mti_version != MDB_VERSION) {
1389 DPRINTF("lock region is version %u, expected version %u",
1390 env->me_txns->mti_version, MDB_VERSION);
1391 rc = MDB_VERSION_MISMATCH;
1394 if (errno != EACCES && errno != EAGAIN) {
1408 #define LOCKNAME "/lock.mdb"
1409 #define DATANAME "/data.mdb"
1411 mdb_env_open(MDB_env *env, const char *path, unsigned int flags, mode_t mode)
1413 int oflags, rc, len, excl;
1414 char *lpath, *dpath;
1417 lpath = malloc(len + sizeof(LOCKNAME) + len + sizeof(DATANAME));
1420 dpath = lpath + len + sizeof(LOCKNAME);
1421 sprintf(lpath, "%s" LOCKNAME, path);
1422 sprintf(dpath, "%s" DATANAME, path);
1424 rc = mdb_env_setup_locks(env, lpath, mode, &excl);
1428 if (F_ISSET(flags, MDB_RDONLY))
1431 oflags = O_RDWR | O_CREAT;
1433 if ((env->me_fd = open(dpath, oflags, mode)) == -1) {
1438 if ((rc = mdb_env_open2(env, flags)) == MDB_SUCCESS) {
1439 /* synchronous fd for meta writes */
1440 if (!(flags & (MDB_RDONLY|MDB_NOSYNC)))
1441 oflags |= MDB_DSYNC;
1442 if ((env->me_mfd = open(dpath, oflags, mode)) == -1) {
1447 env->me_path = strdup(path);
1448 DPRINTF("opened dbenv %p", (void *) env);
1449 pthread_key_create(&env->me_txkey, mdb_env_reader_dest);
1451 mdb_env_share_locks(env);
1452 env->me_dbxs = calloc(env->me_maxdbs, sizeof(MDB_dbx));
1453 env->me_dbs[0] = calloc(env->me_maxdbs, sizeof(MDB_db));
1454 env->me_dbs[1] = calloc(env->me_maxdbs, sizeof(MDB_db));
1460 if (env->me_fd >= 0) {
1464 if (env->me_lfd >= 0) {
1474 mdb_env_close(MDB_env *env)
1481 while (env->me_dpages) {
1482 dp = env->me_dpages;
1483 env->me_dpages = (MDB_dpage *)dp->h.md_parent;
1487 free(env->me_dbs[1]);
1488 free(env->me_dbs[0]);
1492 pthread_key_delete(env->me_txkey);
1495 munmap(env->me_map, env->me_mapsize);
1500 pid_t pid = getpid();
1501 size_t size = (env->me_maxreaders-1) * sizeof(MDB_reader) + sizeof(MDB_txninfo);
1503 for (i=0; i<env->me_txns->mti_numreaders; i++)
1504 if (env->me_txns->mti_readers[i].mr_pid == pid)
1505 env->me_txns->mti_readers[i].mr_pid = 0;
1506 munmap(env->me_txns, size);
1512 /* Search for key within a leaf page, using binary search.
1513 * Returns the smallest entry larger or equal to the key.
1514 * If exactp is non-null, stores whether the found entry was an exact match
1515 * in *exactp (1 or 0).
1516 * If kip is non-null, stores the index of the found entry in *kip.
1517 * If no entry larger or equal to the key is found, returns NULL.
1520 mdb_search_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp, MDB_val *key,
1521 int *exactp, unsigned int *kip)
1529 DPRINTF("searching %u keys in %s page %lu",
1531 IS_LEAF(mp) ? "leaf" : "branch",
1534 assert(NUMKEYS(mp) > 0);
1536 memset(&nodekey, 0, sizeof(nodekey));
1538 low = IS_LEAF(mp) ? 0 : 1;
1539 high = NUMKEYS(mp) - 1;
1540 while (low <= high) {
1541 i = (low + high) >> 1;
1542 node = NODEPTR(mp, i);
1544 nodekey.mv_size = node->mn_ksize;
1545 nodekey.mv_data = NODEKEY(node);
1547 rc = mdb_cmp(txn, dbi, key, &nodekey);
1550 DPRINTF("found leaf index %u [%.*s], rc = %i",
1551 i, (int)nodekey.mv_size, (char *)nodekey.mv_data, rc);
1553 DPRINTF("found branch index %u [%.*s -> %lu], rc = %i",
1554 i, (int)node->mn_ksize, (char *)NODEKEY(node),
1565 if (rc > 0) { /* Found entry is less than the key. */
1566 i++; /* Skip to get the smallest entry larger than key. */
1567 if (i >= NUMKEYS(mp))
1568 /* There is no entry larger or equal to the key. */
1572 *exactp = (rc == 0);
1573 if (kip) /* Store the key index if requested. */
1576 return NODEPTR(mp, i);
1580 cursor_pop_page(MDB_cursor *cursor)
1584 if (cursor->mc_snum) {
1585 top = CURSOR_TOP(cursor);
1588 DPRINTF("popped page %lu off cursor %p", top->mp_page->mp_pgno, (void *) cursor);
1593 cursor_push_page(MDB_cursor *cursor, MDB_page *mp)
1597 DPRINTF("pushing page %lu on cursor %p", mp->mp_pgno, (void *) cursor);
1599 ppage = &cursor->mc_stack[cursor->mc_snum++];
1600 ppage->mp_page = mp;
1606 mdb_get_page(MDB_txn *txn, pgno_t pgno)
1611 if (!F_ISSET(txn->mt_flags, MDB_TXN_RDONLY) && txn->mt_u.dirty_list[0].mid) {
1616 x = mdb_midl2_search(txn->mt_u.dirty_list, &id);
1617 if (x <= txn->mt_u.dirty_list[0].mid && txn->mt_u.dirty_list[x].mid == pgno) {
1618 dp = txn->mt_u.dirty_list[x].mptr;
1624 if (pgno > txn->mt_env->me_meta->mm_last_pg)
1626 p = (MDB_page *)(txn->mt_env->me_map + txn->mt_env->me_psize * pgno);
1632 mdb_search_page_root(MDB_txn *txn, MDB_dbi dbi, MDB_val *key,
1633 MDB_cursor *cursor, int modify, MDB_pageparent *mpp)
1635 MDB_page *mp = mpp->mp_page;
1638 if (cursor && cursor_push_page(cursor, mp) == NULL)
1641 while (IS_BRANCH(mp)) {
1645 DPRINTF("branch page %lu has %u keys", mp->mp_pgno, NUMKEYS(mp));
1646 assert(NUMKEYS(mp) > 1);
1647 DPRINTF("found index 0 to page %lu", NODEPGNO(NODEPTR(mp, 0)));
1649 if (key == NULL) /* Initialize cursor to first page. */
1651 else if (key->mv_size > MAXKEYSIZE && key->mv_data == NULL) {
1652 /* cursor to last page */
1656 node = mdb_search_node(txn, dbi, mp, key, &exact, &i);
1658 i = NUMKEYS(mp) - 1;
1666 DPRINTF("following index %u for key %.*s",
1667 i, (int)key->mv_size, (char *)key->mv_data);
1668 assert(i < NUMKEYS(mp));
1669 node = NODEPTR(mp, i);
1672 CURSOR_TOP(cursor)->mp_ki = i;
1674 mpp->mp_parent = mp;
1675 if ((mp = mdb_get_page(txn, NODEPGNO(node))) == NULL)
1676 return MDB_PAGE_NOTFOUND;
1680 if (cursor && cursor_push_page(cursor, mp) == NULL)
1684 MDB_dhead *dh = ((MDB_dhead *)mp)-1;
1685 if ((rc = mdb_touch(txn, mpp)) != 0)
1687 dh = ((MDB_dhead *)mpp->mp_page)-1;
1688 dh->md_parent = mpp->mp_parent;
1689 dh->md_pi = mpp->mp_pi;
1696 DPRINTF("internal error, index points to a %02X page!?",
1698 return MDB_CORRUPTED;
1701 DPRINTF("found leaf page %lu for key %.*s", mp->mp_pgno,
1702 key ? (int)key->mv_size : 0, key ? (char *)key->mv_data : NULL);
1707 /* Search for the page a given key should be in.
1708 * Stores a pointer to the found page in *mpp.
1709 * If key is NULL, search for the lowest page (used by mdb_cursor_first).
1710 * If cursor is non-null, pushes parent pages on the cursor stack.
1711 * If modify is true, visited pages are updated with new page numbers.
1714 mdb_search_page(MDB_txn *txn, MDB_dbi dbi, MDB_val *key,
1715 MDB_cursor *cursor, int modify, MDB_pageparent *mpp)
1720 /* Choose which root page to start with. If a transaction is given
1721 * use the root page from the transaction, otherwise read the last
1722 * committed root page.
1724 if (F_ISSET(txn->mt_flags, MDB_TXN_ERROR)) {
1725 DPUTS("transaction has failed, must abort");
1728 root = txn->mt_dbs[dbi].md_root;
1730 if (root == P_INVALID) { /* Tree is empty. */
1731 DPUTS("tree is empty");
1732 return MDB_NOTFOUND;
1735 if ((mpp->mp_page = mdb_get_page(txn, root)) == NULL)
1736 return MDB_PAGE_NOTFOUND;
1738 DPRINTF("root page has flags 0x%X", mpp->mp_page->mp_flags);
1741 /* For sub-databases, update main root first */
1742 if (dbi > MAIN_DBI && !txn->mt_dbxs[dbi].md_dirty) {
1744 rc = mdb_search_page(txn, MAIN_DBI, &txn->mt_dbxs[dbi].md_name,
1748 txn->mt_dbxs[dbi].md_dirty = 1;
1750 if (!F_ISSET(mpp->mp_page->mp_flags, P_DIRTY)) {
1751 mpp->mp_parent = NULL;
1753 if ((rc = mdb_touch(txn, mpp)))
1755 txn->mt_dbs[dbi].md_root = mpp->mp_page->mp_pgno;
1759 return mdb_search_page_root(txn, dbi, key, cursor, modify, mpp);
1763 mdb_read_data(MDB_txn *txn, MDB_node *leaf, MDB_val *data)
1765 MDB_page *omp; /* overflow mpage */
1768 if (!F_ISSET(leaf->mn_flags, F_BIGDATA)) {
1769 data->mv_size = leaf->mn_dsize;
1770 data->mv_data = NODEDATA(leaf);
1774 /* Read overflow data.
1776 data->mv_size = leaf->mn_dsize;
1777 memcpy(&pgno, NODEDATA(leaf), sizeof(pgno));
1778 if ((omp = mdb_get_page(txn, pgno)) == NULL) {
1779 DPRINTF("read overflow page %lu failed", pgno);
1780 return MDB_PAGE_NOTFOUND;
1782 data->mv_data = METADATA(omp);
1788 mdb_get(MDB_txn *txn, MDB_dbi dbi,
1789 MDB_val *key, MDB_val *data)
1797 DPRINTF("===> get key [%.*s]", (int)key->mv_size, (char *)key->mv_data);
1799 if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
1802 if (key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
1806 if ((rc = mdb_search_page(txn, dbi, key, NULL, 0, &mpp)) != MDB_SUCCESS)
1809 leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, NULL);
1810 if (leaf && exact) {
1811 /* Return first duplicate data item */
1812 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
1815 mdb_xcursor_init0(txn, dbi, &mx);
1816 mdb_xcursor_init1(txn, dbi, &mx, leaf);
1817 rc = mdb_search_page(&mx.mx_txn, mx.mx_cursor.mc_dbi, NULL, NULL, 0, &mpp);
1818 if (rc != MDB_SUCCESS)
1820 leaf = NODEPTR(mpp.mp_page, 0);
1822 rc = mdb_read_data(txn, leaf, data);
1831 mdb_sibling(MDB_cursor *cursor, int move_right)
1838 if (cursor->mc_snum < 2) {
1839 return MDB_NOTFOUND; /* root has no siblings */
1841 parent = CURSOR_PARENT(cursor);
1843 DPRINTF("parent page is page %lu, index %u",
1844 parent->mp_page->mp_pgno, parent->mp_ki);
1846 cursor_pop_page(cursor);
1847 if (move_right ? (parent->mp_ki + 1 >= NUMKEYS(parent->mp_page))
1848 : (parent->mp_ki == 0)) {
1849 DPRINTF("no more keys left, moving to %s sibling",
1850 move_right ? "right" : "left");
1851 if ((rc = mdb_sibling(cursor, move_right)) != MDB_SUCCESS)
1853 parent = CURSOR_TOP(cursor);
1859 DPRINTF("just moving to %s index key %u",
1860 move_right ? "right" : "left", parent->mp_ki);
1862 assert(IS_BRANCH(parent->mp_page));
1864 indx = NODEPTR(parent->mp_page, parent->mp_ki);
1865 if ((mp = mdb_get_page(cursor->mc_txn, indx->mn_pgno)) == NULL)
1866 return MDB_PAGE_NOTFOUND;
1868 mp->parent = parent->mp_page;
1869 mp->parent_index = parent->mp_ki;
1872 cursor_push_page(cursor, mp);
1878 mdb_set_key(MDB_node *node, MDB_val *key)
1883 key->mv_size = node->mn_ksize;
1884 key->mv_data = NODEKEY(node);
1890 mdb_cursor_next(MDB_cursor *cursor, MDB_val *key, MDB_val *data, MDB_cursor_op op)
1897 if (cursor->mc_eof) {
1898 return MDB_NOTFOUND;
1901 assert(cursor->mc_initialized);
1903 top = CURSOR_TOP(cursor);
1906 if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
1907 leaf = NODEPTR(mp, top->mp_ki);
1908 if ((op == MDB_NEXT || op == MDB_NEXT_DUP) && F_ISSET(leaf->mn_flags, F_DUPDATA)) {
1909 rc = mdb_cursor_next(&cursor->mc_xcursor->mx_cursor, data, NULL, MDB_NEXT);
1910 if (op != MDB_NEXT || rc == MDB_SUCCESS)
1915 DPRINTF("cursor_next: top page is %lu in cursor %p", mp->mp_pgno, (void *) cursor);
1917 if (top->mp_ki + 1 >= NUMKEYS(mp)) {
1918 DPUTS("=====> move to next sibling page");
1919 if (mdb_sibling(cursor, 1) != MDB_SUCCESS) {
1921 return MDB_NOTFOUND;
1923 top = CURSOR_TOP(cursor);
1925 DPRINTF("next page is %lu, key index %u", mp->mp_pgno, top->mp_ki);
1929 DPRINTF("==> cursor points to page %lu with %u keys, key index %u",
1930 mp->mp_pgno, NUMKEYS(mp), top->mp_ki);
1932 assert(IS_LEAF(mp));
1933 leaf = NODEPTR(mp, top->mp_ki);
1935 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
1936 mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, leaf);
1939 if ((rc = mdb_read_data(cursor->mc_txn, leaf, data) != MDB_SUCCESS))
1942 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
1943 rc = mdb_cursor_first(&cursor->mc_xcursor->mx_cursor, data, NULL);
1944 if (rc != MDB_SUCCESS)
1949 return mdb_set_key(leaf, key);
1953 mdb_cursor_prev(MDB_cursor *cursor, MDB_val *key, MDB_val *data, MDB_cursor_op op)
1960 assert(cursor->mc_initialized);
1962 top = CURSOR_TOP(cursor);
1965 if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
1966 leaf = NODEPTR(mp, top->mp_ki);
1967 if ((op == MDB_PREV || op == MDB_PREV_DUP) && F_ISSET(leaf->mn_flags, F_DUPDATA)) {
1968 rc = mdb_cursor_prev(&cursor->mc_xcursor->mx_cursor, data, NULL, MDB_PREV);
1969 if (op != MDB_PREV || rc == MDB_SUCCESS)
1974 DPRINTF("cursor_prev: top page is %lu in cursor %p", mp->mp_pgno, (void *) cursor);
1976 if (top->mp_ki == 0) {
1977 DPUTS("=====> move to prev sibling page");
1978 if (mdb_sibling(cursor, 0) != MDB_SUCCESS) {
1979 return MDB_NOTFOUND;
1981 top = CURSOR_TOP(cursor);
1983 top->mp_ki = NUMKEYS(mp) - 1;
1984 DPRINTF("prev page is %lu, key index %u", mp->mp_pgno, top->mp_ki);
1990 DPRINTF("==> cursor points to page %lu with %u keys, key index %u",
1991 mp->mp_pgno, NUMKEYS(mp), top->mp_ki);
1993 assert(IS_LEAF(mp));
1994 leaf = NODEPTR(mp, top->mp_ki);
1996 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
1997 mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, leaf);
2000 if ((rc = mdb_read_data(cursor->mc_txn, leaf, data) != MDB_SUCCESS))
2003 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
2004 rc = mdb_cursor_last(&cursor->mc_xcursor->mx_cursor, data, NULL);
2005 if (rc != MDB_SUCCESS)
2010 return mdb_set_key(leaf, key);
2014 mdb_cursor_set(MDB_cursor *cursor, MDB_val *key, MDB_val *data,
2015 MDB_cursor_op op, int *exactp)
2024 assert(key->mv_size > 0);
2026 cursor->mc_snum = 0;
2028 rc = mdb_search_page(cursor->mc_txn, cursor->mc_dbi, key, cursor, 0, &mpp);
2029 if (rc != MDB_SUCCESS)
2031 assert(IS_LEAF(mpp.mp_page));
2033 top = CURSOR_TOP(cursor);
2034 leaf = mdb_search_node(cursor->mc_txn, cursor->mc_dbi, mpp.mp_page, key, exactp, &top->mp_ki);
2035 if (exactp != NULL && !*exactp) {
2036 /* MDB_SET specified and not an exact match. */
2037 return MDB_NOTFOUND;
2041 DPUTS("===> inexact leaf not found, goto sibling");
2042 if ((rc = mdb_sibling(cursor, 1)) != MDB_SUCCESS)
2043 return rc; /* no entries matched */
2044 top = CURSOR_TOP(cursor);
2046 mpp.mp_page = top->mp_page;
2047 assert(IS_LEAF(mpp.mp_page));
2048 leaf = NODEPTR(mpp.mp_page, 0);
2051 cursor->mc_initialized = 1;
2054 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
2055 mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, leaf);
2058 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
2059 if (op == MDB_SET || op == MDB_SET_RANGE) {
2060 rc = mdb_cursor_first(&cursor->mc_xcursor->mx_cursor, data, NULL);
2064 if (op == MDB_GET_BOTH) {
2069 op2 = MDB_SET_RANGE;
2071 rc = mdb_cursor_set(&cursor->mc_xcursor->mx_cursor, data, NULL, op2, ex2p);
2072 if (rc != MDB_SUCCESS)
2075 } else if (op == MDB_GET_BOTH || op == MDB_GET_BOTH_RANGE) {
2077 if ((rc = mdb_read_data(cursor->mc_txn, leaf, &d2)) != MDB_SUCCESS)
2079 rc = mdb_dcmp(cursor->mc_txn, cursor->mc_dbi, data, &d2);
2081 if (op == MDB_GET_BOTH || rc > 0)
2082 return MDB_NOTFOUND;
2086 if ((rc = mdb_read_data(cursor->mc_txn, leaf, data)) != MDB_SUCCESS)
2091 rc = mdb_set_key(leaf, key);
2092 if (rc == MDB_SUCCESS) {
2093 DPRINTF("==> cursor placed on key %.*s",
2094 (int)key->mv_size, (char *)key->mv_data);
2102 mdb_cursor_first(MDB_cursor *cursor, MDB_val *key, MDB_val *data)
2108 cursor->mc_snum = 0;
2110 rc = mdb_search_page(cursor->mc_txn, cursor->mc_dbi, NULL, cursor, 0, &mpp);
2111 if (rc != MDB_SUCCESS)
2113 assert(IS_LEAF(mpp.mp_page));
2115 leaf = NODEPTR(mpp.mp_page, 0);
2116 cursor->mc_initialized = 1;
2120 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
2121 mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, leaf);
2122 rc = mdb_cursor_first(&cursor->mc_xcursor->mx_cursor, data, NULL);
2126 if ((rc = mdb_read_data(cursor->mc_txn, leaf, data)) != MDB_SUCCESS)
2130 return mdb_set_key(leaf, key);
2134 mdb_cursor_last(MDB_cursor *cursor, MDB_val *key, MDB_val *data)
2142 cursor->mc_snum = 0;
2144 lkey.mv_size = MAXKEYSIZE+1;
2145 lkey.mv_data = NULL;
2147 rc = mdb_search_page(cursor->mc_txn, cursor->mc_dbi, &lkey, cursor, 0, &mpp);
2148 if (rc != MDB_SUCCESS)
2150 assert(IS_LEAF(mpp.mp_page));
2152 leaf = NODEPTR(mpp.mp_page, NUMKEYS(mpp.mp_page)-1);
2153 cursor->mc_initialized = 1;
2156 top = CURSOR_TOP(cursor);
2157 top->mp_ki = NUMKEYS(top->mp_page) - 1;
2160 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
2161 mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, leaf);
2162 rc = mdb_cursor_last(&cursor->mc_xcursor->mx_cursor, data, NULL);
2166 if ((rc = mdb_read_data(cursor->mc_txn, leaf, data)) != MDB_SUCCESS)
2171 return mdb_set_key(leaf, key);
2175 mdb_cursor_get(MDB_cursor *cursor, MDB_val *key, MDB_val *data,
2185 case MDB_GET_BOTH_RANGE:
2186 if (data == NULL || cursor->mc_xcursor == NULL) {
2193 if (key == NULL || key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
2195 } else if (op == MDB_SET_RANGE)
2196 rc = mdb_cursor_set(cursor, key, data, op, NULL);
2198 rc = mdb_cursor_set(cursor, key, data, op, &exact);
2202 case MDB_NEXT_NODUP:
2203 if (!cursor->mc_initialized)
2204 rc = mdb_cursor_first(cursor, key, data);
2206 rc = mdb_cursor_next(cursor, key, data, op);
2210 case MDB_PREV_NODUP:
2211 if (!cursor->mc_initialized || cursor->mc_eof)
2212 rc = mdb_cursor_last(cursor, key, data);
2214 rc = mdb_cursor_prev(cursor, key, data, op);
2217 rc = mdb_cursor_first(cursor, key, data);
2220 rc = mdb_cursor_last(cursor, key, data);
2223 DPRINTF("unhandled/unimplemented cursor operation %u", op);
2231 /* Allocate a page and initialize it
2234 mdb_new_page(MDB_txn *txn, MDB_dbi dbi, uint32_t flags, int num)
2238 if ((dp = mdb_alloc_page(txn, NULL, 0, num)) == NULL)
2240 DPRINTF("allocated new mpage %lu, page size %u",
2241 dp->p.mp_pgno, txn->mt_env->me_psize);
2242 dp->p.mp_flags = flags | P_DIRTY;
2243 dp->p.mp_lower = PAGEHDRSZ;
2244 dp->p.mp_upper = txn->mt_env->me_psize;
2246 if (IS_BRANCH(&dp->p))
2247 txn->mt_dbs[dbi].md_branch_pages++;
2248 else if (IS_LEAF(&dp->p))
2249 txn->mt_dbs[dbi].md_leaf_pages++;
2250 else if (IS_OVERFLOW(&dp->p)) {
2251 txn->mt_dbs[dbi].md_overflow_pages += num;
2252 dp->p.mp_pages = num;
2259 mdb_leaf_size(MDB_env *env, MDB_val *key, MDB_val *data)
2263 sz = LEAFSIZE(key, data);
2264 if (data->mv_size >= env->me_psize / MDB_MINKEYS) {
2265 /* put on overflow page */
2266 sz -= data->mv_size - sizeof(pgno_t);
2269 return sz + sizeof(indx_t);
2273 mdb_branch_size(MDB_env *env, MDB_val *key)
2278 if (sz >= env->me_psize / MDB_MINKEYS) {
2279 /* put on overflow page */
2280 /* not implemented */
2281 /* sz -= key->size - sizeof(pgno_t); */
2284 return sz + sizeof(indx_t);
2288 mdb_add_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp, indx_t indx,
2289 MDB_val *key, MDB_val *data, pgno_t pgno, uint8_t flags)
2292 size_t node_size = NODESIZE;
2295 MDB_dpage *ofp = NULL; /* overflow page */
2297 assert(mp->mp_upper >= mp->mp_lower);
2299 DPRINTF("add node [%.*s] to %s page %lu at index %i, key size %zu",
2300 key ? (int)key->mv_size : 0, key ? (char *)key->mv_data : NULL,
2301 IS_LEAF(mp) ? "leaf" : "branch",
2302 mp->mp_pgno, indx, key ? key->mv_size : 0);
2305 node_size += key->mv_size;
2309 if (F_ISSET(flags, F_BIGDATA)) {
2310 /* Data already on overflow page. */
2311 node_size += sizeof(pgno_t);
2312 } else if (data->mv_size >= txn->mt_env->me_psize / MDB_MINKEYS) {
2313 int ovpages = OVPAGES(data->mv_size, txn->mt_env->me_psize);
2314 /* Put data on overflow page. */
2315 DPRINTF("data size is %zu, put on overflow page",
2317 node_size += sizeof(pgno_t);
2318 if ((ofp = mdb_new_page(txn, dbi, P_OVERFLOW, ovpages)) == NULL)
2320 DPRINTF("allocated overflow page %lu", ofp->p.mp_pgno);
2323 node_size += data->mv_size;
2327 if (node_size + sizeof(indx_t) > SIZELEFT(mp)) {
2328 DPRINTF("not enough room in page %lu, got %u ptrs",
2329 mp->mp_pgno, NUMKEYS(mp));
2330 DPRINTF("upper - lower = %u - %u = %u", mp->mp_upper, mp->mp_lower,
2331 mp->mp_upper - mp->mp_lower);
2332 DPRINTF("node size = %zu", node_size);
2336 /* Move higher pointers up one slot. */
2337 for (i = NUMKEYS(mp); i > indx; i--)
2338 mp->mp_ptrs[i] = mp->mp_ptrs[i - 1];
2340 /* Adjust free space offsets. */
2341 ofs = mp->mp_upper - node_size;
2342 assert(ofs >= mp->mp_lower + sizeof(indx_t));
2343 mp->mp_ptrs[indx] = ofs;
2345 mp->mp_lower += sizeof(indx_t);
2347 /* Write the node data. */
2348 node = NODEPTR(mp, indx);
2349 node->mn_ksize = (key == NULL) ? 0 : key->mv_size;
2350 node->mn_flags = flags;
2352 node->mn_dsize = data->mv_size;
2354 node->mn_pgno = pgno;
2357 memcpy(NODEKEY(node), key->mv_data, key->mv_size);
2362 if (F_ISSET(flags, F_BIGDATA))
2363 memcpy(node->mn_data + key->mv_size, data->mv_data,
2366 memcpy(node->mn_data + key->mv_size, data->mv_data,
2369 memcpy(node->mn_data + key->mv_size, &ofp->p.mp_pgno,
2371 memcpy(METADATA(&ofp->p), data->mv_data, data->mv_size);
2379 mdb_del_node(MDB_page *mp, indx_t indx)
2382 indx_t i, j, numkeys, ptr;
2386 DPRINTF("delete node %u on %s page %lu", indx,
2387 IS_LEAF(mp) ? "leaf" : "branch", mp->mp_pgno);
2388 assert(indx < NUMKEYS(mp));
2390 node = NODEPTR(mp, indx);
2391 sz = NODESIZE + node->mn_ksize;
2393 if (F_ISSET(node->mn_flags, F_BIGDATA))
2394 sz += sizeof(pgno_t);
2396 sz += NODEDSZ(node);
2399 ptr = mp->mp_ptrs[indx];
2400 numkeys = NUMKEYS(mp);
2401 for (i = j = 0; i < numkeys; i++) {
2403 mp->mp_ptrs[j] = mp->mp_ptrs[i];
2404 if (mp->mp_ptrs[i] < ptr)
2405 mp->mp_ptrs[j] += sz;
2410 base = (char *)mp + mp->mp_upper;
2411 memmove(base + sz, base, ptr - mp->mp_upper);
2413 mp->mp_lower -= sizeof(indx_t);
2418 mdb_xcursor_init0(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx)
2423 mx->mx_txn.mt_dbxs = mx->mx_dbxs;
2424 mx->mx_txn.mt_dbs = mx->mx_dbs;
2425 mx->mx_dbxs[0] = txn->mt_dbxs[0];
2426 mx->mx_dbxs[1] = txn->mt_dbxs[1];
2428 mx->mx_dbxs[2] = txn->mt_dbxs[dbi];
2433 mx->mx_dbxs[dbn+1].md_parent = dbn;
2434 mx->mx_dbxs[dbn+1].md_cmp = mx->mx_dbxs[dbn].md_dcmp;
2435 mx->mx_dbxs[dbn+1].md_rel = mx->mx_dbxs[dbn].md_rel;
2436 mx->mx_dbxs[dbn+1].md_dirty = 0;
2437 mx->mx_txn.mt_numdbs = dbn+2;
2439 mx->mx_cursor.mc_snum = 0;
2440 mx->mx_cursor.mc_txn = &mx->mx_txn;
2441 mx->mx_cursor.mc_dbi = dbn+1;
2445 mdb_xcursor_init1(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx, MDB_node *node)
2447 MDB_db *db = NODEDATA(node);
2449 mx->mx_dbs[0] = txn->mt_dbs[0];
2450 mx->mx_dbs[1] = txn->mt_dbs[1];
2452 mx->mx_dbs[2] = txn->mt_dbs[dbi];
2457 mx->mx_dbs[dbn] = *db;
2458 mx->mx_dbxs[dbn].md_name.mv_data = NODEKEY(node);
2459 mx->mx_dbxs[dbn].md_name.mv_size = node->mn_ksize;
2460 mx->mx_txn.mt_next_pgno = txn->mt_next_pgno;
2461 mx->mx_txn.mt_oldest = txn->mt_oldest;
2462 mx->mx_txn.mt_u = txn->mt_u;
2466 mdb_xcursor_fini(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx)
2468 txn->mt_next_pgno = mx->mx_txn.mt_next_pgno;
2469 txn->mt_oldest = mx->mx_txn.mt_oldest;
2470 txn->mt_u = mx->mx_txn.mt_u;
2471 txn->mt_dbs[0] = mx->mx_dbs[0];
2472 txn->mt_dbs[1] = mx->mx_dbs[1];
2473 txn->mt_dbxs[0].md_dirty = mx->mx_dbxs[0].md_dirty;
2474 txn->mt_dbxs[1].md_dirty = mx->mx_dbxs[1].md_dirty;
2476 txn->mt_dbs[dbi] = mx->mx_dbs[2];
2477 txn->mt_dbxs[dbi].md_dirty = mx->mx_dbxs[2].md_dirty;
2482 mdb_cursor_open(MDB_txn *txn, MDB_dbi dbi, MDB_cursor **ret)
2485 size_t size = sizeof(MDB_cursor);
2487 if (txn == NULL || ret == NULL || !dbi || dbi >= txn->mt_numdbs)
2490 if (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT)
2491 size += sizeof(MDB_xcursor);
2493 if ((cursor = calloc(1, size)) != NULL) {
2494 cursor->mc_dbi = dbi;
2495 cursor->mc_txn = txn;
2496 if (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT) {
2497 MDB_xcursor *mx = (MDB_xcursor *)(cursor + 1);
2498 cursor->mc_xcursor = mx;
2499 mdb_xcursor_init0(txn, dbi, mx);
2510 /* Return the count of duplicate data items for the current key */
2512 mdb_cursor_count(MDB_cursor *mc, unsigned long *countp)
2517 if (mc == NULL || countp == NULL)
2520 if (!(mc->mc_txn->mt_dbs[mc->mc_dbi].md_flags & MDB_DUPSORT))
2523 top = CURSOR_TOP(mc);
2524 leaf = NODEPTR(top->mp_page, top->mp_ki);
2525 if (!F_ISSET(leaf->mn_flags, F_DUPDATA)) {
2528 if (!mc->mc_xcursor->mx_cursor.mc_initialized)
2531 *countp = mc->mc_xcursor->mx_txn.mt_dbs[mc->mc_xcursor->mx_cursor.mc_dbi].md_entries;
2537 mdb_cursor_close(MDB_cursor *cursor)
2539 if (cursor != NULL) {
2545 mdb_update_key(MDB_page *mp, indx_t indx, MDB_val *key)
2547 indx_t ptr, i, numkeys;
2553 node = NODEPTR(mp, indx);
2554 ptr = mp->mp_ptrs[indx];
2555 DPRINTF("update key %u (ofs %u) [%.*s] to [%.*s] on page %lu",
2557 (int)node->mn_ksize, (char *)NODEKEY(node),
2558 (int)key->mv_size, (char *)key->mv_data,
2561 delta = key->mv_size - node->mn_ksize;
2563 if (delta > 0 && SIZELEFT(mp) < delta) {
2564 DPRINTF("OUCH! Not enough room, delta = %d", delta);
2568 numkeys = NUMKEYS(mp);
2569 for (i = 0; i < numkeys; i++) {
2570 if (mp->mp_ptrs[i] <= ptr)
2571 mp->mp_ptrs[i] -= delta;
2574 base = (char *)mp + mp->mp_upper;
2575 len = ptr - mp->mp_upper + NODESIZE;
2576 memmove(base - delta, base, len);
2577 mp->mp_upper -= delta;
2579 node = NODEPTR(mp, indx);
2580 node->mn_ksize = key->mv_size;
2583 memcpy(NODEKEY(node), key->mv_data, key->mv_size);
2588 /* Move a node from src to dst.
2591 mdb_move_node(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *src, indx_t srcindx,
2592 MDB_pageparent *dst, indx_t dstindx)
2598 srcnode = NODEPTR(src->mp_page, srcindx);
2599 DPRINTF("moving %s node %u [%.*s] on page %lu to node %u on page %lu",
2600 IS_LEAF(src->mp_page) ? "leaf" : "branch",
2602 (int)srcnode->mn_ksize, (char *)NODEKEY(srcnode),
2603 src->mp_page->mp_pgno,
2604 dstindx, dst->mp_page->mp_pgno);
2606 /* Mark src and dst as dirty. */
2607 if ((rc = mdb_touch(txn, src)) ||
2608 (rc = mdb_touch(txn, dst)))
2611 /* Add the node to the destination page.
2613 key.mv_size = srcnode->mn_ksize;
2614 key.mv_data = NODEKEY(srcnode);
2615 data.mv_size = NODEDSZ(srcnode);
2616 data.mv_data = NODEDATA(srcnode);
2617 rc = mdb_add_node(txn, dbi, dst->mp_page, dstindx, &key, &data, NODEPGNO(srcnode),
2619 if (rc != MDB_SUCCESS)
2622 /* Delete the node from the source page.
2624 mdb_del_node(src->mp_page, srcindx);
2626 /* Update the parent separators.
2628 if (srcindx == 0 && src->mp_pi != 0) {
2629 DPRINTF("update separator for source page %lu to [%.*s]",
2630 src->mp_page->mp_pgno, (int)key.mv_size, (char *)key.mv_data);
2631 if ((rc = mdb_update_key(src->mp_parent, src->mp_pi,
2632 &key)) != MDB_SUCCESS)
2636 if (srcindx == 0 && IS_BRANCH(src->mp_page)) {
2638 nullkey.mv_size = 0;
2639 assert(mdb_update_key(src->mp_page, 0, &nullkey) == MDB_SUCCESS);
2642 if (dstindx == 0 && dst->mp_pi != 0) {
2643 DPRINTF("update separator for destination page %lu to [%.*s]",
2644 dst->mp_page->mp_pgno, (int)key.mv_size, (char *)key.mv_data);
2645 if ((rc = mdb_update_key(dst->mp_parent, dst->mp_pi,
2646 &key)) != MDB_SUCCESS)
2650 if (dstindx == 0 && IS_BRANCH(dst->mp_page)) {
2652 nullkey.mv_size = 0;
2653 assert(mdb_update_key(dst->mp_page, 0, &nullkey) == MDB_SUCCESS);
2660 mdb_merge(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *src, MDB_pageparent *dst)
2669 DPRINTF("merging page %lu and %lu", src->mp_page->mp_pgno, dst->mp_page->mp_pgno);
2671 assert(txn != NULL);
2672 assert(src->mp_parent); /* can't merge root page */
2673 assert(dst->mp_parent);
2675 /* Mark src and dst as dirty. */
2676 if ((rc = mdb_touch(txn, src)) ||
2677 (rc = mdb_touch(txn, dst)))
2680 /* Move all nodes from src to dst.
2682 for (i = 0; i < NUMKEYS(src->mp_page); i++) {
2683 srcnode = NODEPTR(src->mp_page, i);
2685 key.mv_size = srcnode->mn_ksize;
2686 key.mv_data = NODEKEY(srcnode);
2687 data.mv_size = NODEDSZ(srcnode);
2688 data.mv_data = NODEDATA(srcnode);
2689 rc = mdb_add_node(txn, dbi, dst->mp_page, NUMKEYS(dst->mp_page), &key,
2690 &data, NODEPGNO(srcnode), srcnode->mn_flags);
2691 if (rc != MDB_SUCCESS)
2695 DPRINTF("dst page %lu now has %u keys (%.1f%% filled)",
2696 dst->mp_page->mp_pgno, NUMKEYS(dst->mp_page), (float)PAGEFILL(txn->mt_env, dst->mp_page) / 10);
2698 /* Unlink the src page from parent.
2700 mdb_del_node(src->mp_parent, src->mp_pi);
2701 if (src->mp_pi == 0) {
2703 if ((rc = mdb_update_key(src->mp_parent, 0, &key)) != MDB_SUCCESS)
2707 if (IS_LEAF(src->mp_page))
2708 txn->mt_dbs[dbi].md_leaf_pages--;
2710 txn->mt_dbs[dbi].md_branch_pages--;
2712 mpp.mp_page = src->mp_parent;
2713 dh = (MDB_dhead *)src->mp_parent;
2715 mpp.mp_parent = dh->md_parent;
2716 mpp.mp_pi = dh->md_pi;
2718 return mdb_rebalance(txn, dbi, &mpp);
2721 #define FILL_THRESHOLD 250
2724 mdb_rebalance(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *mpp)
2729 indx_t si = 0, di = 0;
2731 assert(txn != NULL);
2732 assert(mpp != NULL);
2734 DPRINTF("rebalancing %s page %lu (has %u keys, %.1f%% full)",
2735 IS_LEAF(mpp->mp_page) ? "leaf" : "branch",
2736 mpp->mp_page->mp_pgno, NUMKEYS(mpp->mp_page), (float)PAGEFILL(txn->mt_env, mpp->mp_page) / 10);
2738 if (PAGEFILL(txn->mt_env, mpp->mp_page) >= FILL_THRESHOLD) {
2739 DPRINTF("no need to rebalance page %lu, above fill threshold",
2740 mpp->mp_page->mp_pgno);
2744 if (mpp->mp_parent == NULL) {
2745 if (NUMKEYS(mpp->mp_page) == 0) {
2746 DPUTS("tree is completely empty");
2747 txn->mt_dbs[dbi].md_root = P_INVALID;
2748 txn->mt_dbs[dbi].md_depth--;
2749 txn->mt_dbs[dbi].md_leaf_pages--;
2750 } else if (IS_BRANCH(mpp->mp_page) && NUMKEYS(mpp->mp_page) == 1) {
2751 DPUTS("collapsing root page!");
2752 txn->mt_dbs[dbi].md_root = NODEPGNO(NODEPTR(mpp->mp_page, 0));
2753 if ((root = mdb_get_page(txn, txn->mt_dbs[dbi].md_root)) == NULL)
2754 return MDB_PAGE_NOTFOUND;
2755 txn->mt_dbs[dbi].md_depth--;
2756 txn->mt_dbs[dbi].md_branch_pages--;
2758 DPUTS("root page doesn't need rebalancing");
2762 /* The parent (branch page) must have at least 2 pointers,
2763 * otherwise the tree is invalid.
2765 assert(NUMKEYS(mpp->mp_parent) > 1);
2767 /* Leaf page fill factor is below the threshold.
2768 * Try to move keys from left or right neighbor, or
2769 * merge with a neighbor page.
2774 if (mpp->mp_pi == 0) {
2775 /* We're the leftmost leaf in our parent.
2777 DPUTS("reading right neighbor");
2778 node = NODEPTR(mpp->mp_parent, mpp->mp_pi + 1);
2779 if ((npp.mp_page = mdb_get_page(txn, NODEPGNO(node))) == NULL)
2780 return MDB_PAGE_NOTFOUND;
2781 npp.mp_pi = mpp->mp_pi + 1;
2783 di = NUMKEYS(mpp->mp_page);
2785 /* There is at least one neighbor to the left.
2787 DPUTS("reading left neighbor");
2788 node = NODEPTR(mpp->mp_parent, mpp->mp_pi - 1);
2789 if ((npp.mp_page = mdb_get_page(txn, NODEPGNO(node))) == NULL)
2790 return MDB_PAGE_NOTFOUND;
2791 npp.mp_pi = mpp->mp_pi - 1;
2792 si = NUMKEYS(npp.mp_page) - 1;
2795 npp.mp_parent = mpp->mp_parent;
2797 DPRINTF("found neighbor page %lu (%u keys, %.1f%% full)",
2798 npp.mp_page->mp_pgno, NUMKEYS(npp.mp_page), (float)PAGEFILL(txn->mt_env, npp.mp_page) / 10);
2800 /* If the neighbor page is above threshold and has at least two
2801 * keys, move one key from it.
2803 * Otherwise we should try to merge them.
2805 if (PAGEFILL(txn->mt_env, npp.mp_page) >= FILL_THRESHOLD && NUMKEYS(npp.mp_page) >= 2)
2806 return mdb_move_node(txn, dbi, &npp, si, mpp, di);
2807 else { /* FIXME: if (has_enough_room()) */
2808 if (mpp->mp_pi == 0)
2809 return mdb_merge(txn, dbi, &npp, mpp);
2811 return mdb_merge(txn, dbi, mpp, &npp);
2816 mdb_del0(MDB_txn *txn, MDB_dbi dbi, unsigned int ki, MDB_pageparent *mpp, MDB_node *leaf)
2820 /* add overflow pages to free list */
2821 if (F_ISSET(leaf->mn_flags, F_BIGDATA)) {
2825 memcpy(&pg, NODEDATA(leaf), sizeof(pg));
2826 ovpages = OVPAGES(NODEDSZ(leaf), txn->mt_env->me_psize);
2827 for (i=0; i<ovpages; i++) {
2828 DPRINTF("freed ov page %lu", pg);
2829 mdb_midl_insert(txn->mt_free_pgs, pg);
2833 mdb_del_node(mpp->mp_page, ki);
2834 txn->mt_dbs[dbi].md_entries--;
2835 rc = mdb_rebalance(txn, dbi, mpp);
2836 if (rc != MDB_SUCCESS)
2837 txn->mt_flags |= MDB_TXN_ERROR;
2843 mdb_del(MDB_txn *txn, MDB_dbi dbi,
2844 MDB_val *key, MDB_val *data,
2852 DPRINTF("========> delete key %.*s", (int)key->mv_size, (char *)key->mv_data);
2854 assert(key != NULL);
2856 if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
2859 if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
2863 if (key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
2867 mpp.mp_parent = NULL;
2869 if ((rc = mdb_search_page(txn, dbi, key, NULL, 1, &mpp)) != MDB_SUCCESS)
2872 leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, &ki);
2873 if (leaf == NULL || !exact) {
2874 return MDB_NOTFOUND;
2877 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
2881 mdb_xcursor_init0(txn, dbi, &mx);
2882 mdb_xcursor_init1(txn, dbi, &mx, leaf);
2883 if (flags == MDB_DEL_DUP) {
2884 rc = mdb_del(&mx.mx_txn, mx.mx_cursor.mc_dbi, data, NULL, 0);
2885 mdb_xcursor_fini(txn, dbi, &mx);
2886 if (rc != MDB_SUCCESS)
2888 /* If sub-DB still has entries, we're done */
2889 if (mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi].md_root != P_INVALID) {
2890 memcpy(NODEDATA(leaf), &mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi],
2892 txn->mt_dbs[dbi].md_entries--;
2895 /* otherwise fall thru and delete the sub-DB */
2897 /* add all the child DB's pages to the free list */
2898 rc = mdb_search_page(&mx.mx_txn, mx.mx_cursor.mc_dbi,
2899 NULL, &mx.mx_cursor, 0, &mp2);
2900 if (rc == MDB_SUCCESS) {
2901 MDB_ppage *top, *parent;
2905 cursor_pop_page(&mx.mx_cursor);
2906 if (mx.mx_cursor.mc_snum) {
2907 top = CURSOR_TOP(&mx.mx_cursor);
2908 while (mx.mx_cursor.mc_snum > 1) {
2909 parent = CURSOR_PARENT(&mx.mx_cursor);
2910 for (i=0; i<NUMKEYS(top->mp_page); i++) {
2911 ni = NODEPTR(top->mp_page, i);
2912 mdb_midl_insert(txn->mt_free_pgs, ni->mn_pgno);
2915 if (parent->mp_ki >= NUMKEYS(parent->mp_page)) {
2916 cursor_pop_page(&mx.mx_cursor);
2919 ni = NODEPTR(parent->mp_page, parent->mp_ki);
2920 top->mp_page = mdb_get_page(&mx.mx_txn, ni->mn_pgno);
2924 mdb_midl_insert(txn->mt_free_pgs, mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi].md_root);
2929 if (data && (rc = mdb_read_data(txn, leaf, data)) != MDB_SUCCESS)
2932 return mdb_del0(txn, dbi, ki, &mpp, leaf);
2935 /* Split page <*mpp>, and insert <key,(data|newpgno)> in either left or
2936 * right sibling, at index <*newindxp> (as if unsplit). Updates *mpp and
2937 * *newindxp with the actual values after split, ie if *mpp and *newindxp
2938 * refer to a node in the new right sibling page.
2941 mdb_split(MDB_txn *txn, MDB_dbi dbi, MDB_page **mpp, unsigned int *newindxp,
2942 MDB_val *newkey, MDB_val *newdata, pgno_t newpgno)
2945 int rc = MDB_SUCCESS, ins_new = 0;
2948 unsigned int i, j, split_indx;
2950 MDB_val sepkey, rkey, rdata;
2952 MDB_dpage *mdp, *rdp, *pdp;
2955 assert(txn != NULL);
2957 dh = ((MDB_dhead *)*mpp) - 1;
2958 mdp = (MDB_dpage *)dh;
2959 newindx = *newindxp;
2961 DPRINTF("-----> splitting %s page %lu and adding [%.*s] at index %i",
2962 IS_LEAF(&mdp->p) ? "leaf" : "branch", mdp->p.mp_pgno,
2963 (int)newkey->mv_size, (char *)newkey->mv_data, *newindxp);
2965 if (mdp->h.md_parent == NULL) {
2966 if ((pdp = mdb_new_page(txn, dbi, P_BRANCH, 1)) == NULL)
2969 mdp->h.md_parent = &pdp->p;
2970 txn->mt_dbs[dbi].md_root = pdp->p.mp_pgno;
2971 DPRINTF("root split! new root = %lu", pdp->p.mp_pgno);
2972 txn->mt_dbs[dbi].md_depth++;
2974 /* Add left (implicit) pointer. */
2975 if ((rc = mdb_add_node(txn, dbi, &pdp->p, 0, NULL, NULL,
2976 mdp->p.mp_pgno, 0)) != MDB_SUCCESS)
2979 DPRINTF("parent branch page is %lu", mdp->h.md_parent->mp_pgno);
2982 /* Create a right sibling. */
2983 if ((rdp = mdb_new_page(txn, dbi, mdp->p.mp_flags, 1)) == NULL)
2985 rdp->h.md_parent = mdp->h.md_parent;
2986 rdp->h.md_pi = mdp->h.md_pi + 1;
2987 DPRINTF("new right sibling: page %lu", rdp->p.mp_pgno);
2989 /* Move half of the keys to the right sibling. */
2990 if ((copy = malloc(txn->mt_env->me_psize)) == NULL)
2992 memcpy(copy, &mdp->p, txn->mt_env->me_psize);
2993 memset(&mdp->p.mp_ptrs, 0, txn->mt_env->me_psize - PAGEHDRSZ);
2994 mdp->p.mp_lower = PAGEHDRSZ;
2995 mdp->p.mp_upper = txn->mt_env->me_psize;
2997 split_indx = NUMKEYS(copy) / 2 + 1;
2999 /* First find the separating key between the split pages.
3001 memset(&sepkey, 0, sizeof(sepkey));
3002 if (newindx == split_indx) {
3003 sepkey.mv_size = newkey->mv_size;
3004 sepkey.mv_data = newkey->mv_data;
3006 node = NODEPTR(copy, split_indx);
3007 sepkey.mv_size = node->mn_ksize;
3008 sepkey.mv_data = NODEKEY(node);
3011 DPRINTF("separator is [%.*s]", (int)sepkey.mv_size, (char *)sepkey.mv_data);
3013 /* Copy separator key to the parent.
3015 if (SIZELEFT(rdp->h.md_parent) < mdb_branch_size(txn->mt_env, &sepkey)) {
3016 rc = mdb_split(txn, dbi, &rdp->h.md_parent, &rdp->h.md_pi,
3017 &sepkey, NULL, rdp->p.mp_pgno);
3019 /* Right page might now have changed parent.
3020 * Check if left page also changed parent.
3022 if (rdp->h.md_parent != mdp->h.md_parent &&
3023 mdp->h.md_pi >= NUMKEYS(mdp->h.md_parent)) {
3024 mdp->h.md_parent = rdp->h.md_parent;
3025 mdp->h.md_pi = rdp->h.md_pi - 1;
3028 rc = mdb_add_node(txn, dbi, rdp->h.md_parent, rdp->h.md_pi,
3029 &sepkey, NULL, rdp->p.mp_pgno, 0);
3031 if (rc != MDB_SUCCESS) {
3036 for (i = j = 0; i <= NUMKEYS(copy); j++) {
3037 if (i < split_indx) {
3038 /* Re-insert in left sibling. */
3041 /* Insert in right sibling. */
3042 if (i == split_indx)
3043 /* Reset insert index for right sibling. */
3044 j = (i == newindx && ins_new);
3048 if (i == newindx && !ins_new) {
3049 /* Insert the original entry that caused the split. */
3050 rkey.mv_data = newkey->mv_data;
3051 rkey.mv_size = newkey->mv_size;
3052 if (IS_LEAF(&mdp->p)) {
3053 rdata.mv_data = newdata->mv_data;
3054 rdata.mv_size = newdata->mv_size;
3061 /* Update page and index for the new key. */
3064 } else if (i == NUMKEYS(copy)) {
3067 node = NODEPTR(copy, i);
3068 rkey.mv_data = NODEKEY(node);
3069 rkey.mv_size = node->mn_ksize;
3070 if (IS_LEAF(&mdp->p)) {
3071 rdata.mv_data = NODEDATA(node);
3072 rdata.mv_size = node->mn_dsize;
3074 pgno = node->mn_pgno;
3075 flags = node->mn_flags;
3080 if (!IS_LEAF(&mdp->p) && j == 0) {
3081 /* First branch index doesn't need key data. */
3085 rc = mdb_add_node(txn, dbi, &pdp->p, j, &rkey, &rdata, pgno,flags);
3093 mdb_put0(MDB_txn *txn, MDB_dbi dbi,
3094 MDB_val *key, MDB_val *data, unsigned int flags)
3096 int rc = MDB_SUCCESS, exact;
3100 MDB_val xdata, *rdata, dkey;
3102 char dbuf[PAGESIZE];
3105 DPRINTF("==> put key %.*s, size %zu, data size %zu",
3106 (int)key->mv_size, (char *)key->mv_data, key->mv_size, data->mv_size);
3109 mpp.mp_parent = NULL;
3111 rc = mdb_search_page(txn, dbi, key, NULL, 1, &mpp);
3112 if (rc == MDB_SUCCESS) {
3113 leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, &ki);
3114 if (leaf && exact) {
3115 if (flags == MDB_NOOVERWRITE) {
3116 DPRINTF("duplicate key %.*s",
3117 (int)key->mv_size, (char *)key->mv_data);
3118 return MDB_KEYEXIST;
3120 if (F_ISSET(txn->mt_dbs[dbi].md_flags, MDB_DUPSORT)) {
3121 /* Was a single item before, must convert now */
3122 if (!F_ISSET(leaf->mn_flags, F_DUPDATA)) {
3123 dkey.mv_size = NODEDSZ(leaf);
3124 memcpy(dbuf, NODEDATA(leaf), dkey.mv_size);
3125 memset(&dummy, 0, sizeof(dummy));
3126 dummy.md_root = P_INVALID;
3127 if (dkey.mv_size == sizeof(MDB_db)) {
3128 memcpy(NODEDATA(leaf), &dummy, sizeof(dummy));
3131 mdb_del_node(mpp.mp_page, ki);
3134 xdata.mv_size = sizeof(MDB_db);
3135 xdata.mv_data = &dummy;
3140 /* same size, just replace it */
3141 if (NODEDSZ(leaf) == data->mv_size) {
3142 memcpy(NODEDATA(leaf), data->mv_data, data->mv_size);
3145 mdb_del_node(mpp.mp_page, ki);
3147 if (leaf == NULL) { /* append if not found */
3148 ki = NUMKEYS(mpp.mp_page);
3149 DPRINTF("appending key at index %i", ki);
3151 } else if (rc == MDB_NOTFOUND) {
3153 /* new file, just write a root leaf page */
3154 DPUTS("allocating new root leaf page");
3155 if ((dp = mdb_new_page(txn, dbi, P_LEAF, 1)) == NULL) {
3158 mpp.mp_page = &dp->p;
3159 txn->mt_dbs[dbi].md_root = mpp.mp_page->mp_pgno;
3160 txn->mt_dbs[dbi].md_depth++;
3161 txn->mt_dbxs[dbi].md_dirty = 1;
3167 assert(IS_LEAF(mpp.mp_page));
3168 DPRINTF("there are %u keys, should insert new key at index %i",
3169 NUMKEYS(mpp.mp_page), ki);
3174 if (SIZELEFT(mpp.mp_page) < mdb_leaf_size(txn->mt_env, key, rdata)) {
3175 rc = mdb_split(txn, dbi, &mpp.mp_page, &ki, key, rdata, P_INVALID);
3177 /* There is room already in this leaf page. */
3178 rc = mdb_add_node(txn, dbi, mpp.mp_page, ki, key, rdata, 0, 0);
3181 if (rc != MDB_SUCCESS)
3182 txn->mt_flags |= MDB_TXN_ERROR;
3184 /* Remember if we just added a subdatabase */
3185 if (flags & F_SUBDATA) {
3186 leaf = NODEPTR(mpp.mp_page, ki);
3187 leaf->mn_flags |= F_SUBDATA;
3190 /* Now store the actual data in the child DB. Note that we're
3191 * storing the user data in the keys field, so there are strict
3192 * size limits on dupdata. The actual data fields of the child
3193 * DB are all zero size.
3198 leaf = NODEPTR(mpp.mp_page, ki);
3200 mdb_xcursor_init0(txn, dbi, &mx);
3201 mdb_xcursor_init1(txn, dbi, &mx, leaf);
3204 if (flags == MDB_NODUPDATA)
3205 flags = MDB_NOOVERWRITE;
3206 /* converted, write the original data first */
3208 dkey.mv_data = dbuf;
3209 rc = mdb_put0(&mx.mx_txn, mx.mx_cursor.mc_dbi, &dkey, &xdata, flags);
3211 leaf->mn_flags |= F_DUPDATA;
3213 rc = mdb_put0(&mx.mx_txn, mx.mx_cursor.mc_dbi, data, &xdata, flags);
3214 mdb_xcursor_fini(txn, dbi, &mx);
3215 memcpy(NODEDATA(leaf), &mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi],
3218 txn->mt_dbs[dbi].md_entries++;
3226 mdb_put(MDB_txn *txn, MDB_dbi dbi,
3227 MDB_val *key, MDB_val *data, unsigned int flags)
3229 assert(key != NULL);
3230 assert(data != NULL);
3232 if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
3235 if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
3239 if (key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
3243 if ((flags & (MDB_NOOVERWRITE|MDB_NODUPDATA)) != flags)
3246 return mdb_put0(txn, dbi, key, data, flags);
3250 mdb_env_set_flags(MDB_env *env, unsigned int flag, int onoff)
3252 #define CHANGEABLE (MDB_NOSYNC)
3253 if ((flag & CHANGEABLE) != flag)
3256 env->me_flags |= flag;
3258 env->me_flags &= ~flag;
3263 mdb_env_get_flags(MDB_env *env, unsigned int *arg)
3268 *arg = env->me_flags;
3273 mdb_env_get_path(MDB_env *env, const char **arg)
3278 *arg = env->me_path;
3283 mdb_stat0(MDB_env *env, MDB_db *db, MDB_stat *arg)
3285 arg->ms_psize = env->me_psize;
3286 arg->ms_depth = db->md_depth;
3287 arg->ms_branch_pages = db->md_branch_pages;
3288 arg->ms_leaf_pages = db->md_leaf_pages;
3289 arg->ms_overflow_pages = db->md_overflow_pages;
3290 arg->ms_entries = db->md_entries;
3295 mdb_env_stat(MDB_env *env, MDB_stat *arg)
3297 if (env == NULL || arg == NULL)
3300 return mdb_stat0(env, &env->me_meta->mm_dbs[MAIN_DBI], arg);
3303 int mdb_open(MDB_txn *txn, const char *name, unsigned int flags, MDB_dbi *dbi)
3313 if (flags & (MDB_DUPSORT|MDB_REVERSEKEY|MDB_INTEGERKEY))
3314 txn->mt_dbs[MAIN_DBI].md_flags |= (flags & (MDB_DUPSORT|MDB_REVERSEKEY|MDB_INTEGERKEY));
3318 /* Is the DB already open? */
3320 for (i=2; i<txn->mt_numdbs; i++) {
3321 if (len == txn->mt_dbxs[i].md_name.mv_size &&
3322 !strncmp(name, txn->mt_dbxs[i].md_name.mv_data, len)) {
3328 if (txn->mt_numdbs >= txn->mt_env->me_maxdbs - 1)
3331 /* Find the DB info */
3333 key.mv_data = (void *)name;
3334 rc = mdb_get(txn, MAIN_DBI, &key, &data);
3336 /* Create if requested */
3337 if (rc == MDB_NOTFOUND && (flags & MDB_CREATE)) {
3339 data.mv_size = sizeof(MDB_db);
3340 data.mv_data = &dummy;
3341 memset(&dummy, 0, sizeof(dummy));
3342 dummy.md_root = P_INVALID;
3343 dummy.md_flags = flags & 0xffff;
3344 rc = mdb_put0(txn, MAIN_DBI, &key, &data, F_SUBDATA);
3348 /* OK, got info, add to table */
3349 if (rc == MDB_SUCCESS) {
3350 txn->mt_dbxs[txn->mt_numdbs].md_name.mv_data = strdup(name);
3351 txn->mt_dbxs[txn->mt_numdbs].md_name.mv_size = len;
3352 txn->mt_dbxs[txn->mt_numdbs].md_cmp = NULL;
3353 txn->mt_dbxs[txn->mt_numdbs].md_dcmp = NULL;
3354 txn->mt_dbxs[txn->mt_numdbs].md_rel = NULL;
3355 txn->mt_dbxs[txn->mt_numdbs].md_parent = MAIN_DBI;
3356 txn->mt_dbxs[txn->mt_numdbs].md_dirty = dirty;
3357 memcpy(&txn->mt_dbs[txn->mt_numdbs], data.mv_data, sizeof(MDB_db));
3358 *dbi = txn->mt_numdbs;
3359 txn->mt_env->me_dbs[0][txn->mt_numdbs] = txn->mt_dbs[txn->mt_numdbs];
3360 txn->mt_env->me_dbs[1][txn->mt_numdbs] = txn->mt_dbs[txn->mt_numdbs];
3367 int mdb_stat(MDB_txn *txn, MDB_dbi dbi, MDB_stat *arg)
3369 if (txn == NULL || arg == NULL || dbi >= txn->mt_numdbs)
3372 return mdb_stat0(txn->mt_env, &txn->mt_dbs[dbi], arg);
3375 void mdb_close(MDB_txn *txn, MDB_dbi dbi)
3378 if (dbi <= MAIN_DBI || dbi >= txn->mt_numdbs)
3380 ptr = txn->mt_dbxs[dbi].md_name.mv_data;
3381 txn->mt_dbxs[dbi].md_name.mv_data = NULL;
3382 txn->mt_dbxs[dbi].md_name.mv_size = 0;
3386 int mdb_set_compare(MDB_txn *txn, MDB_dbi dbi, MDB_cmp_func *cmp)
3388 if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
3391 txn->mt_dbxs[dbi].md_cmp = cmp;
3395 int mdb_set_dupsort(MDB_txn *txn, MDB_dbi dbi, MDB_cmp_func *cmp)
3397 if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
3400 txn->mt_dbxs[dbi].md_dcmp = cmp;
3404 int mdb_set_relfunc(MDB_txn *txn, MDB_dbi dbi, MDB_rel_func *rel)
3406 if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
3409 txn->mt_dbxs[dbi].md_rel = rel;