2 * @brief memory-mapped database library
4 * A Btree-based database management library modeled loosely on the
5 * BerkeleyDB API, but much simplified.
8 * Copyright 2011 Howard Chu, Symas Corp.
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted only as authorized by the OpenLDAP
15 * A copy of this license is available in the file LICENSE in the
16 * top-level directory of the distribution or, alternatively, at
17 * <http://www.OpenLDAP.org/license.html>.
19 * This code is derived from btree.c written by Martin Hedenfalk.
21 * Copyright (c) 2009, 2010 Martin Hedenfalk <martin@bzero.se>
23 * Permission to use, copy, modify, and distribute this software for any
24 * purpose with or without fee is hereby granted, provided that the above
25 * copyright notice and this permission notice appear in all copies.
27 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
28 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
29 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
30 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
31 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
32 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
33 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
35 #include <sys/types.h>
37 #include <sys/param.h>
43 #ifdef HAVE_SYS_FILE_H
66 /** @defgroup internal MDB Internals
69 /** @defgroup compat Windows Compatibility Macros
73 #define pthread_t DWORD
74 #define pthread_mutex_t HANDLE
75 #define pthread_key_t DWORD
76 #define pthread_self() GetCurrentThreadId()
77 #define pthread_key_create(x,y) *(x) = TlsAlloc()
78 #define pthread_key_delete(x) TlsFree(x)
79 #define pthread_getspecific(x) TlsGetValue(x)
80 #define pthread_setspecific(x,y) TlsSetValue(x,y)
81 #define pthread_mutex_unlock(x) ReleaseMutex(x)
82 #define pthread_mutex_lock(x) WaitForSingleObject(x, INFINITE)
83 #define LOCK_MUTEX_R(env) pthread_mutex_lock(env->me_rmutex)
84 #define UNLOCK_MUTEX_R(env) pthread_mutex_unlock(env->me_rmutex)
85 #define LOCK_MUTEX_W(env) pthread_mutex_lock(env->me_wmutex)
86 #define UNLOCK_MUTEX_W(env) pthread_mutex_unlock(env->me_wmutex)
87 #define getpid() GetCurrentProcessId()
88 #define fdatasync(fd) !FlushFileBuffers(fd)
89 #define ErrCode() GetLastError()
90 #define GetPageSize(x) {SYSTEM_INFO si; GetSystemInfo(&si); (x) = si.dwPageSize;}
91 #define close(fd) CloseHandle(fd)
92 #define munmap(ptr,len) UnmapViewOfFile(ptr)
94 #define LOCK_MUTEX_R(env) pthread_mutex_lock(&env->me_txns->mti_mutex)
95 #define UNLOCK_MUTEX_R(env) pthread_mutex_unlock(&env->me_txns->mti_mutex)
96 #define LOCK_MUTEX_W(env) pthread_mutex_lock(&env->me_txns->mti_wmutex)
97 #define UNLOCK_MUTEX_W(env) pthread_mutex_unlock(&env->me_txns->mti_wmutex)
98 #define ErrCode() errno
100 #define INVALID_HANDLE_VALUE -1
101 #define GetPageSize(x) (x) = sysconf(_SC_PAGE_SIZE)
107 /* Note: If O_DSYNC is undefined but exists in /usr/include,
108 * preferably set some compiler flag to get the definition.
109 * Otherwise compile with the less efficient -DMDB_DSYNC=O_SYNC.
112 # define MDB_DSYNC O_DSYNC
116 typedef ULONG pgno_t;
122 #if !(__STDC_VERSION__ >= 199901L || defined(__GNUC__))
123 # define DPRINTF (void) /* Vararg macros may be unsupported */
125 # define DPRINTF(fmt, ...) /* Requires 2 or more args */ \
126 fprintf(stderr, "%s:%d:(%p) " fmt "\n", __func__, __LINE__, pthread_self(), __VA_ARGS__)
128 # define DPRINTF(fmt, ...) ((void) 0)
130 #define DPUTS(arg) DPRINTF("%s", arg)
132 #define PAGESIZE 4096
133 #define MDB_MINKEYS 2
134 #define MDB_MAGIC 0xBEEFC0DE
135 #define MDB_VERSION 1
136 #define MAXKEYSIZE 511
138 #define KBUF (MAXKEYSIZE*2+1)
139 #define DKBUF char kbuf[KBUF]
140 #define DKEY(x) mdb_dkey(x, kbuf)
146 /* The DB view is always consistent because all writes are wrapped in
147 * the wmutex. Finer-grained locks aren't necessary.
153 #define LAZY_MUTEX_LOCK(x)
154 #define LAZY_MUTEX_UNLOCK(x)
155 #define LAZY_RWLOCK_UNLOCK(x)
156 #define LAZY_RWLOCK_WRLOCK(x)
157 #define LAZY_RWLOCK_RDLOCK(x)
158 #define LAZY_RWLOCK_DEF(x)
159 #define LAZY_RWLOCK_INIT(x,y)
160 #define LAZY_RWLOCK_DESTROY(x)
162 #define LAZY_MUTEX_LOCK(x) pthread_mutex_lock(x)
163 #define LAZY_MUTEX_UNLOCK(x) pthread_mutex_unlock(x)
164 #define LAZY_RWLOCK_UNLOCK(x) pthread_rwlock_unlock(x)
165 #define LAZY_RWLOCK_WRLOCK(x) pthread_rwlock_wrlock(x)
166 #define LAZY_RWLOCK_RDLOCK(x) pthread_rwlock_rdlock(x)
167 #define LAZY_RWLOCK_DEF(x) pthread_rwlock_t x
168 #define LAZY_RWLOCK_INIT(x,y) pthread_rwlock_init(x,y)
169 #define LAZY_RWLOCK_DESTROY(x) pthread_rwlock_destroy(x)
172 #define P_INVALID (~0UL)
174 #define F_ISSET(w, f) (((w) & (f)) == (f))
176 typedef uint16_t indx_t;
178 #define DEFAULT_READERS 126
179 #define DEFAULT_MAPSIZE 1048576
181 /* Lock descriptor stuff */
183 #define CACHELINE 64 /* most CPUs. Itanium uses 128 */
186 typedef struct MDB_rxbody {
192 typedef struct MDB_reader {
195 #define mr_txnid mru.mrx.mrb_txnid
196 #define mr_pid mru.mrx.mrb_pid
197 #define mr_tid mru.mrx.mrb_tid
198 /* cache line alignment */
199 char pad[(sizeof(MDB_rxbody)+CACHELINE-1) & ~(CACHELINE-1)];
203 typedef struct MDB_txbody {
205 uint32_t mtb_version;
206 /* For POSIX the actual mutexes reside in shared memory.
207 * On Windows, mutexes are allocated by the kernel; we store
208 * the name in shared memory so that other processes can
214 pthread_mutex_t mtb_mutex;
217 uint32_t mtb_numreaders;
218 uint32_t mtb_me_toggle;
221 typedef struct MDB_txninfo {
224 #define mti_magic mt1.mtb.mtb_magic
225 #define mti_version mt1.mtb.mtb_version
226 #define mti_mutex mt1.mtb.mtb_mutex
227 #define mti_rmname mt1.mtb.mtb_rmname
228 #define mti_txnid mt1.mtb.mtb_txnid
229 #define mti_numreaders mt1.mtb.mtb_numreaders
230 #define mti_me_toggle mt1.mtb.mtb_me_toggle
231 char pad[(sizeof(MDB_txbody)+CACHELINE-1) & ~(CACHELINE-1)];
236 #define mti_wmname mt2.mt2_wmname
238 pthread_mutex_t mt2_wmutex;
239 #define mti_wmutex mt2.mt2_wmutex
241 char pad[(sizeof(pthread_mutex_t)+CACHELINE-1) & ~(CACHELINE-1)];
243 MDB_reader mti_readers[1];
246 /* Common header for all page types. Overflow pages
247 * occupy a number of contiguous pages with no
248 * headers on any page after the first.
250 typedef struct MDB_page { /* represents a page of storage */
251 #define mp_pgno mp_p.p_pgno
253 pgno_t p_pgno; /* page number */
254 void * p_align; /* for IL32P64 */
256 #define P_BRANCH 0x01 /* branch page */
257 #define P_LEAF 0x02 /* leaf page */
258 #define P_OVERFLOW 0x04 /* overflow page */
259 #define P_META 0x08 /* meta page */
260 #define P_DIRTY 0x10 /* dirty page */
261 #define P_LEAF2 0x20 /* DB with small, fixed size keys and no data */
263 #define mp_lower mp_pb.pb.pb_lower
264 #define mp_upper mp_pb.pb.pb_upper
265 #define mp_pages mp_pb.pb_pages
268 indx_t pb_lower; /* lower bound of free space */
269 indx_t pb_upper; /* upper bound of free space */
271 uint32_t pb_pages; /* number of overflow pages */
273 indx_t mp_ptrs[1]; /* dynamic size */
276 #define PAGEHDRSZ ((unsigned) offsetof(MDB_page, mp_ptrs))
278 #define NUMKEYS(p) (((p)->mp_lower - PAGEHDRSZ) >> 1)
279 #define SIZELEFT(p) (indx_t)((p)->mp_upper - (p)->mp_lower)
280 #define PAGEFILL(env, p) (1000L * ((env)->me_psize - PAGEHDRSZ - SIZELEFT(p)) / \
281 ((env)->me_psize - PAGEHDRSZ))
282 #define IS_LEAF(p) F_ISSET((p)->mp_flags, P_LEAF)
283 #define IS_LEAF2(p) F_ISSET((p)->mp_flags, P_LEAF2)
284 #define IS_BRANCH(p) F_ISSET((p)->mp_flags, P_BRANCH)
285 #define IS_OVERFLOW(p) F_ISSET((p)->mp_flags, P_OVERFLOW)
287 #define OVPAGES(size, psize) ((PAGEHDRSZ-1 + (size)) / (psize) + 1)
289 typedef struct MDB_db {
290 uint32_t md_pad; /* also ksize for LEAF2 pages */
293 ULONG md_branch_pages;
295 ULONG md_overflow_pages;
303 typedef struct MDB_meta { /* meta (footer) page content */
306 void *mm_address; /* address for fixed mapping */
307 size_t mm_mapsize; /* size of mmap region */
308 MDB_db mm_dbs[2]; /* first is free space, 2nd is main db */
309 #define mm_psize mm_dbs[0].md_pad
310 #define mm_flags mm_dbs[0].md_flags
311 pgno_t mm_last_pg; /* last used page in file */
312 ULONG mm_txnid; /* txnid that committed this page */
315 typedef struct MDB_dhead { /* a dirty page */
317 unsigned md_pi; /* parent index */
321 typedef struct MDB_dpage {
326 typedef struct MDB_oldpages {
327 struct MDB_oldpages *mo_next;
329 pgno_t mo_pages[1]; /* dynamic */
332 typedef struct MDB_pageparent {
338 static MDB_dpage *mdb_alloc_page(MDB_txn *txn, MDB_dbi dbi, MDB_page *parent, unsigned int parent_idx, int num);
339 static int mdb_touch(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *mp);
341 typedef struct MDB_ppage { /* ordered list of pages */
343 unsigned int mp_ki; /* cursor index on page */
346 #define CURSOR_TOP(c) (&(c)->mc_stack[(c)->mc_snum-1])
347 #define CURSOR_PARENT(c) (&(c)->mc_stack[(c)->mc_snum-2])
348 #define CURSOR_STACK 32
353 struct MDB_xcursor *mc_xcursor;
356 unsigned int mc_snum; /* number of pushed pages */
357 unsigned int mc_flags;
358 #define C_INITIALIZED 0x01
360 #define C_XDIRTY 0x04
361 MDB_ppage mc_stack[CURSOR_STACK]; /* stack of parent pages */
364 #define METADATA(p) ((void *)((char *)(p) + PAGEHDRSZ))
366 typedef struct MDB_node {
367 #define mn_pgno mn_p.np_pgno
368 #define mn_dsize mn_p.np_dsize
370 pgno_t np_pgno; /* child page number */
371 uint32_t np_dsize; /* leaf data size */
373 unsigned int mn_flags:4;
374 unsigned int mn_ksize:12; /* key size */
375 #define F_BIGDATA 0x01 /* data put on overflow page */
376 #define F_SUBDATA 0x02 /* data is a sub-database */
377 #define F_DUPDATA 0x04 /* data has duplicates */
381 typedef struct MDB_dbx {
383 MDB_cmp_func *md_cmp; /* user compare function */
384 MDB_cmp_func *md_dcmp; /* user dupsort function */
385 MDB_rel_func *md_rel; /* user relocate function */
387 unsigned int md_dirty;
391 pgno_t mt_next_pgno; /* next unallocated page */
394 pgno_t *mt_free_pgs; /* this is an IDL */
396 ID2L dirty_list; /* modified pages */
399 MDB_dbx *mt_dbxs; /* array */
401 unsigned int mt_numdbs;
403 #define MDB_TXN_RDONLY 0x01 /* read-only transaction */
404 #define MDB_TXN_ERROR 0x02 /* an error has occurred */
405 unsigned int mt_flags;
406 unsigned int mt_toggle;
409 /* Context for sorted-dup records */
410 typedef struct MDB_xcursor {
411 MDB_cursor mx_cursor;
420 HANDLE me_mfd; /* just for writing the meta pages */
421 #define MDB_FATAL_ERROR 0x80000000U
423 uint32_t me_extrapad; /* unused for now */
424 unsigned int me_maxreaders;
425 unsigned int me_numdbs;
426 unsigned int me_maxdbs;
429 MDB_txninfo *me_txns;
430 MDB_meta *me_metas[2];
431 MDB_txn *me_txn; /* current write transaction */
433 off_t me_size; /* current file size */
434 pgno_t me_maxpg; /* me_mapsize / me_psize */
435 unsigned int me_psize;
436 unsigned int me_db_toggle;
437 MDB_dbx *me_dbxs; /* array */
439 MDB_oldpages *me_pghead;
440 pthread_key_t me_txkey; /* thread-key for readers */
441 MDB_dpage *me_dpages;
442 pgno_t me_free_pgs[MDB_IDL_UM_SIZE];
443 ID2 me_dirty_list[MDB_IDL_DB_SIZE];
444 LAZY_RWLOCK_DEF(me_dblock);
446 HANDLE me_rmutex; /* Windows mutexes don't reside in shared mem */
451 #define NODESIZE offsetof(MDB_node, mn_data)
453 #define INDXSIZE(k) (NODESIZE + ((k) == NULL ? 0 : (k)->mv_size))
454 #define LEAFSIZE(k, d) (NODESIZE + (k)->mv_size + (d)->mv_size)
455 #define NODEPTR(p, i) ((MDB_node *)((char *)(p) + (p)->mp_ptrs[i]))
456 #define NODEKEY(node) (void *)((node)->mn_data)
457 #define NODEDATA(node) (void *)((char *)(node)->mn_data + (node)->mn_ksize)
458 #define NODEPGNO(node) ((node)->mn_pgno)
459 #define NODEDSZ(node) ((node)->mn_dsize)
460 #define NODEKSZ(node) ((node)->mn_ksize)
461 #define LEAF2KEY(p, i, ks) ((char *)(p) + PAGEHDRSZ + ((i)*(ks)))
463 #define MDB_SET_KEY(node, key) if (key!=NULL) {(key)->mv_size = NODEKSZ(node); (key)->mv_data = NODEKEY(node);}
465 #define MDB_COMMIT_PAGES 64 /* max number of pages to write in one commit */
467 static int mdb_search_page_root(MDB_txn *txn,
468 MDB_dbi dbi, MDB_val *key,
469 MDB_cursor *cursor, int modify,
470 MDB_pageparent *mpp);
471 static int mdb_search_page(MDB_txn *txn,
472 MDB_dbi dbi, MDB_val *key,
473 MDB_cursor *cursor, int modify,
474 MDB_pageparent *mpp);
476 static int mdb_env_read_header(MDB_env *env, MDB_meta *meta);
477 static int mdb_env_read_meta(MDB_env *env, int *which);
478 static int mdb_env_write_meta(MDB_txn *txn);
479 static int mdb_get_page(MDB_txn *txn, pgno_t pgno, MDB_page **mp);
481 static MDB_node *mdb_search_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp,
482 MDB_val *key, int *exactp, unsigned int *kip);
483 static int mdb_add_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp,
484 indx_t indx, MDB_val *key, MDB_val *data,
485 pgno_t pgno, uint8_t flags);
486 static void mdb_del_node(MDB_page *mp, indx_t indx, int ksize);
487 static int mdb_del0(MDB_cursor *mc, unsigned int ki,
488 MDB_pageparent *mpp, MDB_node *leaf);
490 static int mdb_put0(MDB_txn *txn, MDB_dbi dbi,
491 MDB_val *key, MDB_val *data, unsigned int flags);
493 static int mdb_read_data(MDB_txn *txn, MDB_node *leaf, MDB_val *data);
495 static int mdb_rebalance(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *mp);
496 static int mdb_update_key(MDB_page *mp, indx_t indx, MDB_val *key);
497 static int mdb_move_node(MDB_txn *txn, MDB_dbi dbi,
498 MDB_pageparent *src, indx_t srcindx,
499 MDB_pageparent *dst, indx_t dstindx);
500 static int mdb_merge(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *src,
501 MDB_pageparent *dst);
502 static int mdb_split(MDB_txn *txn, MDB_dbi dbi, MDB_page **mpp,
503 unsigned int *newindxp, MDB_val *newkey,
504 MDB_val *newdata, pgno_t newpgno);
505 static MDB_dpage *mdb_new_page(MDB_txn *txn, MDB_dbi dbi, uint32_t flags, int num);
507 static void cursor_pop_page(MDB_cursor *cursor);
508 static MDB_ppage *cursor_push_page(MDB_cursor *cursor,
511 static int mdb_sibling(MDB_cursor *cursor, int move_right);
512 static int mdb_cursor_next(MDB_cursor *cursor,
513 MDB_val *key, MDB_val *data, MDB_cursor_op op);
514 static int mdb_cursor_prev(MDB_cursor *cursor,
515 MDB_val *key, MDB_val *data, MDB_cursor_op op);
516 static int mdb_cursor_set(MDB_cursor *cursor,
517 MDB_val *key, MDB_val *data, MDB_cursor_op op, int *exactp);
518 static int mdb_cursor_first(MDB_cursor *cursor,
519 MDB_val *key, MDB_val *data);
520 static int mdb_cursor_last(MDB_cursor *cursor,
521 MDB_val *key, MDB_val *data);
523 static void mdb_xcursor_init0(MDB_cursor *mc);
524 static void mdb_xcursor_init1(MDB_cursor *mc, MDB_page *mp, MDB_node *node);
525 static void mdb_xcursor_init2(MDB_cursor *mc);
526 static void mdb_xcursor_fini(MDB_cursor *mc);
528 static size_t mdb_leaf_size(MDB_env *env, MDB_val *key,
530 static size_t mdb_branch_size(MDB_env *env, MDB_val *key);
532 static int memncmp(const void *s1, size_t n1,
533 const void *s2, size_t n2);
534 static int memnrcmp(const void *s1, size_t n1,
535 const void *s2, size_t n2);
538 static SECURITY_DESCRIPTOR mdb_null_sd;
539 static SECURITY_ATTRIBUTES mdb_all_sa;
540 static int mdb_sec_inited;
544 memncmp(const void *s1, size_t n1, const void *s2, size_t n2)
546 int diff, len_diff = -1;
549 len_diff = (n1 > n2);
552 diff = memcmp(s1, s2, n1);
553 return diff ? diff : len_diff;
557 memnrcmp(const void *s1, size_t n1, const void *s2, size_t n2)
559 const unsigned char *p1, *p2, *p1_lim;
566 p1 = (const unsigned char *)s1 + n1 - 1;
567 p2 = (const unsigned char *)s2 + n2 - 1;
569 for (p1_lim = (n1 <= n2 ? s1 : s2); *p1 == *p2; p1--, p2--) {
571 return (p1 != s1) ? (p1 != p2) : (p2 != s2) ? -1 : 0;
577 mdb_version(int *maj, int *min, int *pat)
579 if (maj) *maj = MDB_VERSION_MAJOR;
580 if (min) *min = MDB_VERSION_MINOR;
581 if (pat) *pat = MDB_VERSION_PATCH;
582 return MDB_VERSION_STRING;
585 static char *const mdb_errstr[] = {
586 "MDB_KEYEXIST: Key/data pair already exists",
587 "MDB_NOTFOUND: No matching key/data pair found",
588 "MDB_PAGE_NOTFOUND: Requested page not found",
589 "MDB_CORRUPTED: Located page was wrong type",
590 "MDB_PANIC: Update of meta page failed",
591 "MDB_VERSION_MISMATCH: Database environment version mismatch"
595 mdb_strerror(int err)
598 return ("Successful return: 0");
600 if (err >= MDB_KEYEXIST && err <= MDB_VERSION_MISMATCH)
601 return mdb_errstr[err - MDB_KEYEXIST];
603 return strerror(err);
608 mdb_dkey(MDB_val *key, char *buf)
611 unsigned char *c = key->mv_data;
613 if (key->mv_size > MAXKEYSIZE)
616 for (i=0; i<key->mv_size; i++)
617 ptr += sprintf(ptr, "%02x", *c++);
619 sprintf(buf, "%.*s", key->mv_size, key->mv_data);
626 mdb_cmp(MDB_txn *txn, MDB_dbi dbi, const MDB_val *a, const MDB_val *b)
628 if (txn->mt_dbxs[dbi].md_cmp)
629 return txn->mt_dbxs[dbi].md_cmp(a, b);
631 if (txn->mt_dbs[dbi].md_flags & (MDB_REVERSEKEY
632 #if __BYTE_ORDER == __LITTLE_ENDIAN
636 return memnrcmp(a->mv_data, a->mv_size, b->mv_data, b->mv_size);
638 return memncmp((char *)a->mv_data, a->mv_size, b->mv_data, b->mv_size);
642 mdb_dcmp(MDB_txn *txn, MDB_dbi dbi, const MDB_val *a, const MDB_val *b)
644 if (txn->mt_dbxs[dbi].md_dcmp)
645 return txn->mt_dbxs[dbi].md_dcmp(a, b);
647 if (txn->mt_dbs[dbi].md_flags & (0
648 #if __BYTE_ORDER == __LITTLE_ENDIAN
652 return memnrcmp(a->mv_data, a->mv_size, b->mv_data, b->mv_size);
654 return memncmp((char *)a->mv_data, a->mv_size, b->mv_data, b->mv_size);
657 /* Allocate new page(s) for writing */
659 mdb_alloc_page(MDB_txn *txn, MDB_dbi dbi, MDB_page *parent, unsigned int parent_idx, int num)
662 pgno_t pgno = P_INVALID;
665 if (txn->mt_txnid > 2) {
667 if (!txn->mt_env->me_pghead && dbi != FREE_DBI &&
668 txn->mt_dbs[FREE_DBI].md_root != P_INVALID) {
669 /* See if there's anything in the free DB */
675 mpp.mp_parent = NULL;
678 mc.mc_dbi = FREE_DBI;
681 mdb_search_page(txn, FREE_DBI, NULL, &mc, 0, &mpp);
682 leaf = NODEPTR(mpp.mp_page, 0);
683 kptr = (ULONG *)NODEKEY(leaf);
687 oldest = txn->mt_txnid - 1;
688 for (i=0; i<txn->mt_env->me_txns->mti_numreaders; i++) {
689 ULONG mr = txn->mt_env->me_txns->mti_readers[i].mr_txnid;
690 if (mr && mr < oldest)
695 if (oldest > *kptr) {
696 /* It's usable, grab it.
703 mdb_read_data(txn, leaf, &data);
704 idl = (ULONG *)data.mv_data;
705 mop = malloc(sizeof(MDB_oldpages) + MDB_IDL_SIZEOF(idl) - sizeof(pgno_t));
706 mop->mo_next = txn->mt_env->me_pghead;
707 mop->mo_txnid = *kptr;
708 txn->mt_env->me_pghead = mop;
709 memcpy(mop->mo_pages, idl, MDB_IDL_SIZEOF(idl));
714 DPRINTF("IDL read txn %lu root %lu num %lu",
715 mop->mo_txnid, txn->mt_dbs[FREE_DBI].md_root, idl[0]);
716 for (i=0; i<idl[0]; i++) {
717 DPRINTF("IDL %lu", idl[i+1]);
721 /* drop this IDL from the DB */
722 top = CURSOR_TOP(&mc);
724 mc.mc_flags = C_INITIALIZED;
725 mdb_cursor_del(&mc, 0);
728 if (txn->mt_env->me_pghead) {
729 MDB_oldpages *mop = txn->mt_env->me_pghead;
731 /* FIXME: For now, always use fresh pages. We
732 * really ought to search the free list for a
737 /* peel pages off tail, so we only have to truncate the list */
738 pgno = MDB_IDL_LAST(mop->mo_pages);
739 if (MDB_IDL_IS_RANGE(mop->mo_pages)) {
741 if (mop->mo_pages[2] > mop->mo_pages[1])
742 mop->mo_pages[0] = 0;
746 if (MDB_IDL_IS_ZERO(mop->mo_pages)) {
747 txn->mt_env->me_pghead = mop->mo_next;
754 if (pgno == P_INVALID) {
755 /* DB size is maxed out */
756 if (txn->mt_next_pgno + num >= txn->mt_env->me_maxpg)
759 if (txn->mt_env->me_dpages && num == 1) {
760 dp = txn->mt_env->me_dpages;
761 txn->mt_env->me_dpages = (MDB_dpage *)dp->h.md_parent;
763 if ((dp = malloc(txn->mt_env->me_psize * num + sizeof(MDB_dhead))) == NULL)
767 dp->h.md_parent = parent;
768 dp->h.md_pi = parent_idx;
769 if (pgno == P_INVALID) {
770 dp->p.mp_pgno = txn->mt_next_pgno;
771 txn->mt_next_pgno += num;
773 dp->p.mp_pgno = pgno;
775 mid.mid = dp->p.mp_pgno;
777 mdb_mid2l_insert(txn->mt_u.dirty_list, &mid);
782 /* Touch a page: make it dirty and re-insert into tree with updated pgno.
785 mdb_touch(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *pp)
787 MDB_page *mp = pp->mp_page;
792 if (!F_ISSET(mp->mp_flags, P_DIRTY)) {
794 if ((dp = mdb_alloc_page(txn, dbi, pp->mp_parent, pp->mp_pi, 1)) == NULL)
796 DPRINTF("touched db %u page %lu -> %lu", dbi, mp->mp_pgno, dp->p.mp_pgno);
797 assert(mp->mp_pgno != dp->p.mp_pgno);
798 mdb_midl_insert(txn->mt_free_pgs, mp->mp_pgno);
799 pgno = dp->p.mp_pgno;
800 memcpy(&dp->p, mp, txn->mt_env->me_psize);
803 mp->mp_flags |= P_DIRTY;
805 /* Update the page number to new touched page. */
806 if (pp->mp_parent != NULL)
807 NODEPGNO(NODEPTR(pp->mp_parent, pp->mp_pi)) = mp->mp_pgno;
814 mdb_env_sync(MDB_env *env, int force)
817 if (force || !F_ISSET(env->me_flags, MDB_NOSYNC)) {
818 if (fdatasync(env->me_fd))
825 mdb_txn_reset0(MDB_txn *txn);
828 mdb_txn_renew0(MDB_txn *txn)
830 MDB_env *env = txn->mt_env;
832 if (txn->mt_flags & MDB_TXN_RDONLY) {
833 MDB_reader *r = pthread_getspecific(env->me_txkey);
836 pid_t pid = getpid();
837 pthread_t tid = pthread_self();
840 for (i=0; i<env->me_txns->mti_numreaders; i++)
841 if (env->me_txns->mti_readers[i].mr_pid == 0)
843 if (i == env->me_maxreaders) {
847 env->me_txns->mti_readers[i].mr_pid = pid;
848 env->me_txns->mti_readers[i].mr_tid = tid;
849 if (i >= env->me_txns->mti_numreaders)
850 env->me_txns->mti_numreaders = i+1;
852 r = &env->me_txns->mti_readers[i];
853 pthread_setspecific(env->me_txkey, r);
855 txn->mt_txnid = env->me_txns->mti_txnid;
856 txn->mt_toggle = env->me_txns->mti_me_toggle;
857 r->mr_txnid = txn->mt_txnid;
858 txn->mt_u.reader = r;
862 txn->mt_txnid = env->me_txns->mti_txnid+1;
863 txn->mt_toggle = env->me_txns->mti_me_toggle;
864 txn->mt_u.dirty_list = env->me_dirty_list;
865 txn->mt_u.dirty_list[0].mid = 0;
866 txn->mt_free_pgs = env->me_free_pgs;
867 txn->mt_free_pgs[0] = 0;
868 txn->mt_next_pgno = env->me_metas[txn->mt_toggle]->mm_last_pg+1;
872 /* Copy the DB arrays */
873 LAZY_RWLOCK_RDLOCK(&env->me_dblock);
874 txn->mt_numdbs = env->me_numdbs;
875 txn->mt_dbxs = env->me_dbxs; /* mostly static anyway */
876 memcpy(txn->mt_dbs, env->me_metas[txn->mt_toggle]->mm_dbs, 2 * sizeof(MDB_db));
877 if (txn->mt_numdbs > 2)
878 memcpy(txn->mt_dbs+2, env->me_dbs[env->me_db_toggle]+2,
879 (txn->mt_numdbs - 2) * sizeof(MDB_db));
880 LAZY_RWLOCK_UNLOCK(&env->me_dblock);
886 mdb_txn_renew(MDB_txn *txn)
893 if (txn->mt_env->me_flags & MDB_FATAL_ERROR) {
894 DPUTS("environment had fatal error, must shutdown!");
898 rc = mdb_txn_renew0(txn);
899 if (rc == MDB_SUCCESS) {
900 DPRINTF("renew txn %lu%c %p on mdbenv %p, root page %lu",
901 txn->mt_txnid, (txn->mt_flags & MDB_TXN_RDONLY) ? 'r' : 'w', txn,
902 (void *)txn->mt_env, txn->mt_dbs[MAIN_DBI].md_root);
908 mdb_txn_begin(MDB_env *env, unsigned int flags, MDB_txn **ret)
913 if (env->me_flags & MDB_FATAL_ERROR) {
914 DPUTS("environment had fatal error, must shutdown!");
917 if ((txn = calloc(1, sizeof(MDB_txn) + env->me_maxdbs * sizeof(MDB_db))) == NULL) {
918 DPRINTF("calloc: %s", strerror(ErrCode()));
921 txn->mt_dbs = (MDB_db *)(txn+1);
922 if (flags & MDB_RDONLY) {
923 txn->mt_flags |= MDB_TXN_RDONLY;
927 rc = mdb_txn_renew0(txn);
932 DPRINTF("begin txn %lu%c %p on mdbenv %p, root page %lu",
933 txn->mt_txnid, (txn->mt_flags & MDB_TXN_RDONLY) ? 'r' : 'w', txn,
934 (void *) env, txn->mt_dbs[MAIN_DBI].md_root);
941 mdb_txn_reset0(MDB_txn *txn)
943 MDB_env *env = txn->mt_env;
945 if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
946 txn->mt_u.reader->mr_txnid = 0;
952 /* return all dirty pages to dpage list */
953 for (i=1; i<=txn->mt_u.dirty_list[0].mid; i++) {
954 dp = txn->mt_u.dirty_list[i].mptr;
955 if (dp->h.md_num == 1) {
956 dp->h.md_parent = (MDB_page *)txn->mt_env->me_dpages;
957 txn->mt_env->me_dpages = dp;
959 /* large pages just get freed directly */
964 while ((mop = txn->mt_env->me_pghead)) {
965 txn->mt_env->me_pghead = mop->mo_next;
970 for (i=2; i<env->me_numdbs; i++)
971 env->me_dbxs[i].md_dirty = 0;
977 mdb_txn_reset(MDB_txn *txn)
982 DPRINTF("reset txn %lu%c %p on mdbenv %p, root page %lu",
983 txn->mt_txnid, (txn->mt_flags & MDB_TXN_RDONLY) ? 'r' : 'w', txn,
984 (void *)txn->mt_env, txn->mt_dbs[MAIN_DBI].md_root);
990 mdb_txn_abort(MDB_txn *txn)
995 DPRINTF("abort txn %lu%c %p on mdbenv %p, root page %lu",
996 txn->mt_txnid, (txn->mt_flags & MDB_TXN_RDONLY) ? 'r' : 'w', txn,
997 (void *)txn->mt_env, txn->mt_dbs[MAIN_DBI].md_root);
1004 mdb_txn_commit(MDB_txn *txn)
1015 assert(txn != NULL);
1016 assert(txn->mt_env != NULL);
1020 if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
1025 if (txn != env->me_txn) {
1026 DPUTS("attempt to commit unknown transaction");
1031 if (F_ISSET(txn->mt_flags, MDB_TXN_ERROR)) {
1032 DPUTS("error flag is set, can't commit");
1037 if (!txn->mt_u.dirty_list[0].mid)
1040 DPRINTF("committing txn %lu %p on mdbenv %p, root page %lu",
1041 txn->mt_txnid, txn, (void *)env, txn->mt_dbs[MAIN_DBI].md_root);
1044 mc.mc_dbi = FREE_DBI;
1047 /* should only be one record now */
1048 if (env->me_pghead) {
1051 /* make sure first page of freeDB is touched and on freelist */
1052 mpp.mp_parent = NULL;
1055 mdb_search_page(txn, FREE_DBI, NULL, &mc, 1, &mpp);
1057 /* save to free list */
1058 if (!MDB_IDL_IS_ZERO(txn->mt_free_pgs)) {
1063 /* make sure last page of freeDB is touched and on freelist */
1064 key.mv_size = MAXKEYSIZE+1;
1066 mpp.mp_parent = NULL;
1069 mdb_search_page(txn, FREE_DBI, &key, &mc, 1, &mpp);
1074 ULONG *idl = txn->mt_free_pgs;
1075 DPRINTF("IDL write txn %lu root %lu num %lu",
1076 txn->mt_txnid, txn->mt_dbs[FREE_DBI].md_root, idl[0]);
1077 for (i=0; i<idl[0]; i++) {
1078 DPRINTF("IDL %lu", idl[i+1]);
1082 /* write to last page of freeDB */
1083 key.mv_size = sizeof(pgno_t);
1084 key.mv_data = (char *)&txn->mt_txnid;
1085 data.mv_data = txn->mt_free_pgs;
1086 /* The free list can still grow during this call,
1087 * despite the pre-emptive touches above. So check
1088 * and make sure the entire thing got written.
1091 i = txn->mt_free_pgs[0];
1092 data.mv_size = MDB_IDL_SIZEOF(txn->mt_free_pgs);
1093 rc = mdb_cursor_put(&mc, &key, &data, 0);
1098 } while (i != txn->mt_free_pgs[0]);
1100 /* should only be one record now */
1101 if (env->me_pghead) {
1105 mop = env->me_pghead;
1106 key.mv_size = sizeof(pgno_t);
1107 key.mv_data = (char *)&mop->mo_txnid;
1108 data.mv_size = MDB_IDL_SIZEOF(mop->mo_pages);
1109 data.mv_data = mop->mo_pages;
1110 mdb_cursor_put(&mc, &key, &data, 0);
1111 free(env->me_pghead);
1112 env->me_pghead = NULL;
1115 /* Update DB root pointers. Their pages have already been
1116 * touched so this is all in-place and cannot fail.
1120 data.mv_size = sizeof(MDB_db);
1122 mc.mc_dbi = MAIN_DBI;
1124 for (i = 2; i < txn->mt_numdbs; i++) {
1125 if (txn->mt_dbxs[i].md_dirty) {
1126 data.mv_data = &txn->mt_dbs[i];
1127 mdb_cursor_put(&mc, &txn->mt_dbxs[i].md_name, &data, 0);
1132 /* Commit up to MDB_COMMIT_PAGES dirty pages to disk until done.
1138 /* Windows actually supports scatter/gather I/O, but only on
1139 * unbuffered file handles. Since we're relying on the OS page
1140 * cache for all our data, that's self-defeating. So we just
1141 * write pages one at a time. We use the ov structure to set
1142 * the write offset, to at least save the overhead of a Seek
1146 memset(&ov, 0, sizeof(ov));
1147 for (; i<=txn->mt_u.dirty_list[0].mid; i++) {
1148 dp = txn->mt_u.dirty_list[i].mptr;
1149 DPRINTF("committing page %lu", dp->p.mp_pgno);
1150 size = dp->p.mp_pgno * env->me_psize;
1151 ov.Offset = size & 0xffffffff;
1152 ov.OffsetHigh = size >> 16;
1153 ov.OffsetHigh >>= 16;
1154 /* clear dirty flag */
1155 dp->p.mp_flags &= ~P_DIRTY;
1156 rc = WriteFile(env->me_fd, &dp->p, env->me_psize * dp->h.md_num, NULL, &ov);
1159 DPRINTF("WriteFile: %d", n);
1166 struct iovec iov[MDB_COMMIT_PAGES];
1170 for (; i<=txn->mt_u.dirty_list[0].mid; i++) {
1171 dp = txn->mt_u.dirty_list[i].mptr;
1172 if (dp->p.mp_pgno != next) {
1174 DPRINTF("committing %u dirty pages", n);
1175 rc = writev(env->me_fd, iov, n);
1179 DPUTS("short write, filesystem full?");
1181 DPRINTF("writev: %s", strerror(n));
1188 lseek(env->me_fd, dp->p.mp_pgno * env->me_psize, SEEK_SET);
1189 next = dp->p.mp_pgno;
1191 DPRINTF("committing page %lu", dp->p.mp_pgno);
1192 iov[n].iov_len = env->me_psize * dp->h.md_num;
1193 iov[n].iov_base = &dp->p;
1194 size += iov[n].iov_len;
1195 next = dp->p.mp_pgno + dp->h.md_num;
1196 /* clear dirty flag */
1197 dp->p.mp_flags &= ~P_DIRTY;
1198 if (++n >= MDB_COMMIT_PAGES) {
1208 DPRINTF("committing %u dirty pages", n);
1209 rc = writev(env->me_fd, iov, n);
1213 DPUTS("short write, filesystem full?");
1215 DPRINTF("writev: %s", strerror(n));
1222 /* Drop the dirty pages.
1224 for (i=1; i<=txn->mt_u.dirty_list[0].mid; i++) {
1225 dp = txn->mt_u.dirty_list[i].mptr;
1226 if (dp->h.md_num == 1) {
1227 dp->h.md_parent = (MDB_page *)txn->mt_env->me_dpages;
1228 txn->mt_env->me_dpages = dp;
1232 txn->mt_u.dirty_list[i].mid = 0;
1234 txn->mt_u.dirty_list[0].mid = 0;
1236 if ((n = mdb_env_sync(env, 0)) != 0 ||
1237 (n = mdb_env_write_meta(txn)) != MDB_SUCCESS) {
1244 /* update the DB tables */
1246 int toggle = !env->me_db_toggle;
1249 ip = &env->me_dbs[toggle][2];
1250 jp = &txn->mt_dbs[2];
1251 LAZY_RWLOCK_WRLOCK(&env->me_dblock);
1252 for (i = 2; i < txn->mt_numdbs; i++) {
1253 if (ip->md_root != jp->md_root)
1258 for (i = 2; i < txn->mt_numdbs; i++) {
1259 if (txn->mt_dbxs[i].md_dirty)
1260 txn->mt_dbxs[i].md_dirty = 0;
1262 env->me_db_toggle = toggle;
1263 env->me_numdbs = txn->mt_numdbs;
1264 LAZY_RWLOCK_UNLOCK(&env->me_dblock);
1267 UNLOCK_MUTEX_W(env);
1274 mdb_env_read_header(MDB_env *env, MDB_meta *meta)
1276 char page[PAGESIZE];
1281 /* We don't know the page size yet, so use a minimum value.
1285 if (!ReadFile(env->me_fd, page, PAGESIZE, (DWORD *)&rc, NULL) || rc == 0)
1287 if ((rc = read(env->me_fd, page, PAGESIZE)) == 0)
1292 else if (rc != PAGESIZE) {
1296 DPRINTF("read: %s", strerror(err));
1300 p = (MDB_page *)page;
1302 if (!F_ISSET(p->mp_flags, P_META)) {
1303 DPRINTF("page %lu not a meta page", p->mp_pgno);
1308 if (m->mm_magic != MDB_MAGIC) {
1309 DPUTS("meta has invalid magic");
1313 if (m->mm_version != MDB_VERSION) {
1314 DPRINTF("database is version %u, expected version %u",
1315 m->mm_version, MDB_VERSION);
1316 return MDB_VERSION_MISMATCH;
1319 memcpy(meta, m, sizeof(*m));
1324 mdb_env_init_meta(MDB_env *env, MDB_meta *meta)
1331 DPUTS("writing new meta page");
1335 meta->mm_magic = MDB_MAGIC;
1336 meta->mm_version = MDB_VERSION;
1337 meta->mm_psize = psize;
1338 meta->mm_last_pg = 1;
1339 meta->mm_flags = env->me_flags & 0xffff;
1340 meta->mm_flags |= MDB_INTEGERKEY;
1341 meta->mm_dbs[0].md_root = P_INVALID;
1342 meta->mm_dbs[1].md_root = P_INVALID;
1344 p = calloc(2, psize);
1346 p->mp_flags = P_META;
1349 memcpy(m, meta, sizeof(*meta));
1351 q = (MDB_page *)((char *)p + psize);
1354 q->mp_flags = P_META;
1357 memcpy(m, meta, sizeof(*meta));
1362 rc = WriteFile(env->me_fd, p, psize * 2, &len, NULL);
1363 rc = (len == psize * 2) ? MDB_SUCCESS : ErrCode();
1366 rc = write(env->me_fd, p, psize * 2);
1367 rc = (rc == (int)psize * 2) ? MDB_SUCCESS : ErrCode();
1374 mdb_env_write_meta(MDB_txn *txn)
1377 MDB_meta meta, metab;
1379 int rc, len, toggle;
1385 assert(txn != NULL);
1386 assert(txn->mt_env != NULL);
1388 toggle = !txn->mt_toggle;
1389 DPRINTF("writing meta page %d for root page %lu",
1390 toggle, txn->mt_dbs[MAIN_DBI].md_root);
1394 metab.mm_txnid = env->me_metas[toggle]->mm_txnid;
1395 metab.mm_last_pg = env->me_metas[toggle]->mm_last_pg;
1397 ptr = (char *)&meta;
1398 off = offsetof(MDB_meta, mm_dbs[0].md_depth);
1399 len = sizeof(MDB_meta) - off;
1402 meta.mm_dbs[0] = txn->mt_dbs[0];
1403 meta.mm_dbs[1] = txn->mt_dbs[1];
1404 meta.mm_last_pg = txn->mt_next_pgno - 1;
1405 meta.mm_txnid = txn->mt_txnid;
1408 off += env->me_psize;
1411 /* Write to the SYNC fd */
1414 memset(&ov, 0, sizeof(ov));
1416 WriteFile(env->me_mfd, ptr, len, (DWORD *)&rc, &ov);
1419 rc = pwrite(env->me_mfd, ptr, len, off);
1424 DPUTS("write failed, disk error?");
1425 /* On a failure, the pagecache still contains the new data.
1426 * Write some old data back, to prevent it from being used.
1427 * Use the non-SYNC fd; we know it will fail anyway.
1429 meta.mm_last_pg = metab.mm_last_pg;
1430 meta.mm_txnid = metab.mm_txnid;
1432 WriteFile(env->me_fd, ptr, len, NULL, &ov);
1434 r2 = pwrite(env->me_fd, ptr, len, off);
1436 env->me_flags |= MDB_FATAL_ERROR;
1439 /* Memory ordering issues are irrelevant; since the entire writer
1440 * is wrapped by wmutex, all of these changes will become visible
1441 * after the wmutex is unlocked. Since the DB is multi-version,
1442 * readers will get consistent data regardless of how fresh or
1443 * how stale their view of these values is.
1445 LAZY_MUTEX_LOCK(&env->me_txns->mti_mutex);
1446 txn->mt_env->me_txns->mti_me_toggle = toggle;
1447 txn->mt_env->me_txns->mti_txnid = txn->mt_txnid;
1448 LAZY_MUTEX_UNLOCK(&env->me_txns->mti_mutex);
1454 mdb_env_read_meta(MDB_env *env, int *which)
1458 assert(env != NULL);
1460 if (env->me_metas[0]->mm_txnid < env->me_metas[1]->mm_txnid)
1463 DPRINTF("Using meta page %d", toggle);
1470 mdb_env_create(MDB_env **env)
1474 e = calloc(1, sizeof(MDB_env));
1475 if (!e) return ENOMEM;
1477 e->me_maxreaders = DEFAULT_READERS;
1479 e->me_fd = INVALID_HANDLE_VALUE;
1480 e->me_lfd = INVALID_HANDLE_VALUE;
1481 e->me_mfd = INVALID_HANDLE_VALUE;
1487 mdb_env_set_mapsize(MDB_env *env, size_t size)
1491 env->me_mapsize = size;
1496 mdb_env_set_maxdbs(MDB_env *env, int dbs)
1500 env->me_maxdbs = dbs;
1505 mdb_env_set_maxreaders(MDB_env *env, int readers)
1509 env->me_maxreaders = readers;
1514 mdb_env_get_maxreaders(MDB_env *env, int *readers)
1516 if (!env || !readers)
1518 *readers = env->me_maxreaders;
1523 mdb_env_open2(MDB_env *env, unsigned int flags)
1525 int i, newenv = 0, toggle;
1529 env->me_flags = flags;
1531 memset(&meta, 0, sizeof(meta));
1533 if ((i = mdb_env_read_header(env, &meta)) != 0) {
1536 DPUTS("new mdbenv");
1540 if (!env->me_mapsize) {
1541 env->me_mapsize = newenv ? DEFAULT_MAPSIZE : meta.mm_mapsize;
1547 LONG sizelo, sizehi;
1548 sizelo = env->me_mapsize & 0xffffffff;
1549 sizehi = env->me_mapsize >> 16; /* pointless on WIN32, only needed on W64 */
1551 /* Windows won't create mappings for zero length files.
1552 * Just allocate the maxsize right now.
1555 SetFilePointer(env->me_fd, sizelo, sizehi ? &sizehi : NULL, 0);
1556 if (!SetEndOfFile(env->me_fd))
1558 SetFilePointer(env->me_fd, 0, NULL, 0);
1560 mh = CreateFileMapping(env->me_fd, NULL, PAGE_READONLY,
1561 sizehi, sizelo, NULL);
1564 env->me_map = MapViewOfFileEx(mh, FILE_MAP_READ, 0, 0, env->me_mapsize,
1572 if (meta.mm_address && (flags & MDB_FIXEDMAP))
1574 env->me_map = mmap(meta.mm_address, env->me_mapsize, PROT_READ, i,
1576 if (env->me_map == MAP_FAILED)
1581 meta.mm_mapsize = env->me_mapsize;
1582 if (flags & MDB_FIXEDMAP)
1583 meta.mm_address = env->me_map;
1584 i = mdb_env_init_meta(env, &meta);
1585 if (i != MDB_SUCCESS) {
1586 munmap(env->me_map, env->me_mapsize);
1590 env->me_psize = meta.mm_psize;
1592 env->me_maxpg = env->me_mapsize / env->me_psize;
1594 p = (MDB_page *)env->me_map;
1595 env->me_metas[0] = METADATA(p);
1596 env->me_metas[1] = (MDB_meta *)((char *)env->me_metas[0] + meta.mm_psize);
1598 if ((i = mdb_env_read_meta(env, &toggle)) != 0)
1601 DPRINTF("opened database version %u, pagesize %u",
1602 env->me_metas[toggle]->mm_version, env->me_psize);
1603 DPRINTF("depth: %u", env->me_metas[toggle]->mm_dbs[MAIN_DBI].md_depth);
1604 DPRINTF("entries: %lu", env->me_metas[toggle]->mm_dbs[MAIN_DBI].md_entries);
1605 DPRINTF("branch pages: %lu", env->me_metas[toggle]->mm_dbs[MAIN_DBI].md_branch_pages);
1606 DPRINTF("leaf pages: %lu", env->me_metas[toggle]->mm_dbs[MAIN_DBI].md_leaf_pages);
1607 DPRINTF("overflow pages: %lu", env->me_metas[toggle]->mm_dbs[MAIN_DBI].md_overflow_pages);
1608 DPRINTF("root: %lu", env->me_metas[toggle]->mm_dbs[MAIN_DBI].md_root);
1614 /* Windows doesn't support destructor callbacks for thread-specific storage */
1616 mdb_env_reader_dest(void *ptr)
1618 MDB_reader *reader = ptr;
1620 reader->mr_txnid = 0;
1626 /* downgrade the exclusive lock on the region back to shared */
1628 mdb_env_share_locks(MDB_env *env)
1632 if (env->me_metas[0]->mm_txnid < env->me_metas[1]->mm_txnid)
1634 env->me_txns->mti_me_toggle = toggle;
1635 env->me_txns->mti_txnid = env->me_metas[toggle]->mm_txnid;
1640 /* First acquire a shared lock. The Unlock will
1641 * then release the existing exclusive lock.
1643 memset(&ov, 0, sizeof(ov));
1644 LockFileEx(env->me_lfd, 0, 0, 1, 0, &ov);
1645 UnlockFile(env->me_lfd, 0, 0, 1, 0);
1649 struct flock lock_info;
1650 /* The shared lock replaces the existing lock */
1651 memset((void *)&lock_info, 0, sizeof(lock_info));
1652 lock_info.l_type = F_RDLCK;
1653 lock_info.l_whence = SEEK_SET;
1654 lock_info.l_start = 0;
1655 lock_info.l_len = 1;
1656 fcntl(env->me_lfd, F_SETLK, &lock_info);
1662 mdb_env_setup_locks(MDB_env *env, char *lpath, int mode, int *excl)
1670 if ((env->me_lfd = CreateFile(lpath, GENERIC_READ|GENERIC_WRITE,
1671 FILE_SHARE_READ|FILE_SHARE_WRITE, NULL, OPEN_ALWAYS,
1672 FILE_ATTRIBUTE_NORMAL, NULL)) == INVALID_HANDLE_VALUE) {
1676 /* Try to get exclusive lock. If we succeed, then
1677 * nobody is using the lock region and we should initialize it.
1680 if (LockFile(env->me_lfd, 0, 0, 1, 0)) {
1684 memset(&ov, 0, sizeof(ov));
1685 if (!LockFileEx(env->me_lfd, 0, 0, 1, 0, &ov)) {
1691 size = GetFileSize(env->me_lfd, NULL);
1693 if ((env->me_lfd = open(lpath, O_RDWR|O_CREAT, mode)) == -1) {
1697 /* Try to get exclusive lock. If we succeed, then
1698 * nobody is using the lock region and we should initialize it.
1701 struct flock lock_info;
1702 memset((void *)&lock_info, 0, sizeof(lock_info));
1703 lock_info.l_type = F_WRLCK;
1704 lock_info.l_whence = SEEK_SET;
1705 lock_info.l_start = 0;
1706 lock_info.l_len = 1;
1707 rc = fcntl(env->me_lfd, F_SETLK, &lock_info);
1711 lock_info.l_type = F_RDLCK;
1712 rc = fcntl(env->me_lfd, F_SETLKW, &lock_info);
1719 size = lseek(env->me_lfd, 0, SEEK_END);
1721 rsize = (env->me_maxreaders-1) * sizeof(MDB_reader) + sizeof(MDB_txninfo);
1722 if (size < rsize && *excl) {
1724 SetFilePointer(env->me_lfd, rsize, NULL, 0);
1725 if (!SetEndOfFile(env->me_lfd)) {
1730 if (ftruncate(env->me_lfd, rsize) != 0) {
1737 size = rsize - sizeof(MDB_txninfo);
1738 env->me_maxreaders = size/sizeof(MDB_reader) + 1;
1743 mh = CreateFileMapping(env->me_lfd, NULL, PAGE_READWRITE,
1749 env->me_txns = MapViewOfFileEx(mh, FILE_MAP_WRITE, 0, 0, rsize, NULL);
1751 if (!env->me_txns) {
1757 env->me_txns = mmap(0, rsize, PROT_READ|PROT_WRITE, MAP_SHARED,
1759 if (env->me_txns == MAP_FAILED) {
1767 if (!mdb_sec_inited) {
1768 InitializeSecurityDescriptor(&mdb_null_sd,
1769 SECURITY_DESCRIPTOR_REVISION);
1770 SetSecurityDescriptorDacl(&mdb_null_sd, TRUE, 0, FALSE);
1771 mdb_all_sa.nLength = sizeof(SECURITY_ATTRIBUTES);
1772 mdb_all_sa.bInheritHandle = FALSE;
1773 mdb_all_sa.lpSecurityDescriptor = &mdb_null_sd;
1776 /* FIXME: only using up to 20 characters of the env path here,
1777 * probably not enough to assure uniqueness...
1779 sprintf(env->me_txns->mti_rmname, "Global\\MDBr%.20s", lpath);
1780 ptr = env->me_txns->mti_rmname + sizeof("Global\\MDBr");
1781 while ((ptr = strchr(ptr, '\\')))
1783 env->me_rmutex = CreateMutex(&mdb_all_sa, FALSE, env->me_txns->mti_rmname);
1784 if (!env->me_rmutex) {
1788 sprintf(env->me_txns->mti_rmname, "Global\\MDBw%.20s", lpath);
1789 ptr = env->me_txns->mti_rmname + sizeof("Global\\MDBw");
1790 while ((ptr = strchr(ptr, '\\')))
1792 env->me_wmutex = CreateMutex(&mdb_all_sa, FALSE, env->me_txns->mti_rmname);
1793 if (!env->me_wmutex) {
1798 pthread_mutexattr_t mattr;
1800 pthread_mutexattr_init(&mattr);
1801 rc = pthread_mutexattr_setpshared(&mattr, PTHREAD_PROCESS_SHARED);
1805 pthread_mutex_init(&env->me_txns->mti_mutex, &mattr);
1806 pthread_mutex_init(&env->me_txns->mti_wmutex, &mattr);
1808 env->me_txns->mti_version = MDB_VERSION;
1809 env->me_txns->mti_magic = MDB_MAGIC;
1810 env->me_txns->mti_txnid = 0;
1811 env->me_txns->mti_numreaders = 0;
1812 env->me_txns->mti_me_toggle = 0;
1815 if (env->me_txns->mti_magic != MDB_MAGIC) {
1816 DPUTS("lock region has invalid magic");
1820 if (env->me_txns->mti_version != MDB_VERSION) {
1821 DPRINTF("lock region is version %u, expected version %u",
1822 env->me_txns->mti_version, MDB_VERSION);
1823 rc = MDB_VERSION_MISMATCH;
1827 if (rc != EACCES && rc != EAGAIN) {
1831 env->me_rmutex = OpenMutex(SYNCHRONIZE, FALSE, env->me_txns->mti_rmname);
1832 if (!env->me_rmutex) {
1836 env->me_wmutex = OpenMutex(SYNCHRONIZE, FALSE, env->me_txns->mti_wmname);
1837 if (!env->me_wmutex) {
1847 env->me_lfd = INVALID_HANDLE_VALUE;
1852 #define LOCKNAME "/lock.mdb"
1853 #define DATANAME "/data.mdb"
1855 mdb_env_open(MDB_env *env, const char *path, unsigned int flags, mode_t mode)
1857 int oflags, rc, len, excl;
1858 char *lpath, *dpath;
1861 lpath = malloc(len + sizeof(LOCKNAME) + len + sizeof(DATANAME));
1864 dpath = lpath + len + sizeof(LOCKNAME);
1865 sprintf(lpath, "%s" LOCKNAME, path);
1866 sprintf(dpath, "%s" DATANAME, path);
1868 rc = mdb_env_setup_locks(env, lpath, mode, &excl);
1873 if (F_ISSET(flags, MDB_RDONLY)) {
1874 oflags = GENERIC_READ;
1875 len = OPEN_EXISTING;
1877 oflags = GENERIC_READ|GENERIC_WRITE;
1880 mode = FILE_ATTRIBUTE_NORMAL;
1881 if ((env->me_fd = CreateFile(dpath, oflags, FILE_SHARE_READ|FILE_SHARE_WRITE,
1882 NULL, len, mode, NULL)) == INVALID_HANDLE_VALUE) {
1887 if (F_ISSET(flags, MDB_RDONLY))
1890 oflags = O_RDWR | O_CREAT;
1892 if ((env->me_fd = open(dpath, oflags, mode)) == -1) {
1898 if ((rc = mdb_env_open2(env, flags)) == MDB_SUCCESS) {
1899 /* synchronous fd for meta writes */
1901 if (!(flags & (MDB_RDONLY|MDB_NOSYNC)))
1902 mode |= FILE_FLAG_WRITE_THROUGH;
1903 if ((env->me_mfd = CreateFile(dpath, oflags, FILE_SHARE_READ|FILE_SHARE_WRITE,
1904 NULL, len, mode, NULL)) == INVALID_HANDLE_VALUE) {
1909 if (!(flags & (MDB_RDONLY|MDB_NOSYNC)))
1910 oflags |= MDB_DSYNC;
1911 if ((env->me_mfd = open(dpath, oflags, mode)) == -1) {
1916 env->me_path = strdup(path);
1917 DPRINTF("opened dbenv %p", (void *) env);
1918 pthread_key_create(&env->me_txkey, mdb_env_reader_dest);
1919 LAZY_RWLOCK_INIT(&env->me_dblock, NULL);
1921 mdb_env_share_locks(env);
1922 env->me_dbxs = calloc(env->me_maxdbs, sizeof(MDB_dbx));
1923 env->me_dbs[0] = calloc(env->me_maxdbs, sizeof(MDB_db));
1924 env->me_dbs[1] = calloc(env->me_maxdbs, sizeof(MDB_db));
1930 if (env->me_fd != INVALID_HANDLE_VALUE) {
1932 env->me_fd = INVALID_HANDLE_VALUE;
1934 if (env->me_lfd != INVALID_HANDLE_VALUE) {
1936 env->me_lfd = INVALID_HANDLE_VALUE;
1944 mdb_env_close(MDB_env *env)
1951 while (env->me_dpages) {
1952 dp = env->me_dpages;
1953 env->me_dpages = (MDB_dpage *)dp->h.md_parent;
1957 free(env->me_dbs[1]);
1958 free(env->me_dbs[0]);
1962 LAZY_RWLOCK_DESTROY(&env->me_dblock);
1963 pthread_key_delete(env->me_txkey);
1966 munmap(env->me_map, env->me_mapsize);
1971 pid_t pid = getpid();
1973 for (i=0; i<env->me_txns->mti_numreaders; i++)
1974 if (env->me_txns->mti_readers[i].mr_pid == pid)
1975 env->me_txns->mti_readers[i].mr_pid = 0;
1976 munmap(env->me_txns, (env->me_maxreaders-1)*sizeof(MDB_reader)+sizeof(MDB_txninfo));
1982 /* Search for key within a leaf page, using binary search.
1983 * Returns the smallest entry larger or equal to the key.
1984 * If exactp is non-null, stores whether the found entry was an exact match
1985 * in *exactp (1 or 0).
1986 * If kip is non-null, stores the index of the found entry in *kip.
1987 * If no entry larger or equal to the key is found, returns NULL.
1990 mdb_search_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp, MDB_val *key,
1991 int *exactp, unsigned int *kip)
1996 MDB_node *node = NULL;
2000 DPRINTF("searching %u keys in %s page %lu",
2002 IS_LEAF(mp) ? "leaf" : "branch",
2005 assert(NUMKEYS(mp) > 0);
2007 memset(&nodekey, 0, sizeof(nodekey));
2009 low = IS_LEAF(mp) ? 0 : 1;
2010 high = NUMKEYS(mp) - 1;
2011 while (low <= high) {
2012 i = (low + high) >> 1;
2015 nodekey.mv_size = txn->mt_dbs[dbi].md_pad;
2016 nodekey.mv_data = LEAF2KEY(mp, i, nodekey.mv_size);
2018 node = NODEPTR(mp, i);
2020 nodekey.mv_size = node->mn_ksize;
2021 nodekey.mv_data = NODEKEY(node);
2024 rc = mdb_cmp(txn, dbi, key, &nodekey);
2027 DPRINTF("found leaf index %u [%s], rc = %i",
2028 i, DKEY(&nodekey), rc);
2030 DPRINTF("found branch index %u [%s -> %lu], rc = %i",
2031 i, DKEY(&nodekey), NODEPGNO(node), rc);
2041 if (rc > 0) { /* Found entry is less than the key. */
2042 i++; /* Skip to get the smallest entry larger than key. */
2045 *exactp = (rc == 0);
2046 if (kip) /* Store the key index if requested. */
2048 if (i >= NUMKEYS(mp))
2049 /* There is no entry larger or equal to the key. */
2052 /* nodeptr is fake for LEAF2 */
2053 return IS_LEAF2(mp) ? NODEPTR(mp, 0) : NODEPTR(mp, i);
2057 cursor_pop_page(MDB_cursor *cursor)
2061 if (cursor->mc_snum) {
2062 top = CURSOR_TOP(cursor);
2065 DPRINTF("popped page %lu off db %u cursor %p", top->mp_page->mp_pgno,
2066 cursor->mc_dbi, (void *) cursor);
2071 cursor_push_page(MDB_cursor *cursor, MDB_page *mp)
2075 DPRINTF("pushing page %lu on db %u cursor %p", mp->mp_pgno,
2076 cursor->mc_dbi, (void *) cursor);
2078 assert(cursor->mc_snum < CURSOR_STACK);
2080 ppage = &cursor->mc_stack[cursor->mc_snum++];
2081 ppage->mp_page = mp;
2087 mdb_get_page(MDB_txn *txn, pgno_t pgno, MDB_page **ret)
2091 if (!F_ISSET(txn->mt_flags, MDB_TXN_RDONLY) && txn->mt_u.dirty_list[0].mid) {
2094 x = mdb_mid2l_search(txn->mt_u.dirty_list, pgno);
2095 if (x <= txn->mt_u.dirty_list[0].mid && txn->mt_u.dirty_list[x].mid == pgno) {
2096 dp = txn->mt_u.dirty_list[x].mptr;
2101 if (pgno <= txn->mt_env->me_metas[txn->mt_toggle]->mm_last_pg)
2102 p = (MDB_page *)(txn->mt_env->me_map + txn->mt_env->me_psize * pgno);
2106 DPRINTF("page %lu not found", pgno);
2109 return (p != NULL) ? MDB_SUCCESS : MDB_PAGE_NOTFOUND;
2113 mdb_search_page_root(MDB_txn *txn, MDB_dbi dbi, MDB_val *key,
2114 MDB_cursor *cursor, int modify, MDB_pageparent *mpp)
2116 MDB_page *mp = mpp->mp_page;
2120 if (cursor && cursor_push_page(cursor, mp) == NULL)
2123 while (IS_BRANCH(mp)) {
2127 DPRINTF("branch page %lu has %u keys", mp->mp_pgno, NUMKEYS(mp));
2128 assert(NUMKEYS(mp) > 1);
2129 DPRINTF("found index 0 to page %lu", NODEPGNO(NODEPTR(mp, 0)));
2131 if (key == NULL) /* Initialize cursor to first page. */
2133 else if (key->mv_size > MAXKEYSIZE && key->mv_data == NULL) {
2134 /* cursor to last page */
2138 node = mdb_search_node(txn, dbi, mp, key, &exact, &i);
2140 i = NUMKEYS(mp) - 1;
2148 DPRINTF("following index %u for key [%s]",
2150 assert(i < NUMKEYS(mp));
2151 node = NODEPTR(mp, i);
2154 CURSOR_TOP(cursor)->mp_ki = i;
2156 mpp->mp_parent = mp;
2157 if ((rc = mdb_get_page(txn, NODEPGNO(node), &mp)))
2162 if (cursor && cursor_push_page(cursor, mp) == NULL)
2166 MDB_dhead *dh = ((MDB_dhead *)mp)-1;
2167 if ((rc = mdb_touch(txn, dbi, mpp)) != 0)
2169 dh = ((MDB_dhead *)mpp->mp_page)-1;
2170 dh->md_parent = mpp->mp_parent;
2171 dh->md_pi = mpp->mp_pi;
2178 DPRINTF("internal error, index points to a %02X page!?",
2180 return MDB_CORRUPTED;
2183 DPRINTF("found leaf page %lu for key [%s]", mp->mp_pgno,
2184 key ? DKEY(key) : NULL);
2189 /* Search for the page a given key should be in.
2190 * Stores a pointer to the found page in *mpp.
2191 * If key is NULL, search for the lowest page (used by mdb_cursor_first).
2192 * If cursor is non-null, pushes parent pages on the cursor stack.
2193 * If modify is true, visited pages are updated with new page numbers.
2196 mdb_search_page(MDB_txn *txn, MDB_dbi dbi, MDB_val *key,
2197 MDB_cursor *cursor, int modify, MDB_pageparent *mpp)
2202 /* Choose which root page to start with. If a transaction is given
2203 * use the root page from the transaction, otherwise read the last
2204 * committed root page.
2206 if (F_ISSET(txn->mt_flags, MDB_TXN_ERROR)) {
2207 DPUTS("transaction has failed, must abort");
2210 root = txn->mt_dbs[dbi].md_root;
2212 if (root == P_INVALID) { /* Tree is empty. */
2213 DPUTS("tree is empty");
2214 return MDB_NOTFOUND;
2217 if ((rc = mdb_get_page(txn, root, &mpp->mp_page)))
2220 DPRINTF("db %u root page %lu has flags 0x%X",
2221 dbi, root, mpp->mp_page->mp_flags);
2224 /* For sub-databases, update main root first */
2225 if (dbi > MAIN_DBI && !txn->mt_dbxs[dbi].md_dirty) {
2227 rc = mdb_search_page(txn, MAIN_DBI, &txn->mt_dbxs[dbi].md_name,
2231 txn->mt_dbxs[dbi].md_dirty = 1;
2233 if (!F_ISSET(mpp->mp_page->mp_flags, P_DIRTY)) {
2234 mpp->mp_parent = NULL;
2236 if ((rc = mdb_touch(txn, dbi, mpp)))
2238 txn->mt_dbs[dbi].md_root = mpp->mp_page->mp_pgno;
2242 return mdb_search_page_root(txn, dbi, key, cursor, modify, mpp);
2246 mdb_read_data(MDB_txn *txn, MDB_node *leaf, MDB_val *data)
2248 MDB_page *omp; /* overflow mpage */
2252 if (!F_ISSET(leaf->mn_flags, F_BIGDATA)) {
2253 data->mv_size = leaf->mn_dsize;
2254 data->mv_data = NODEDATA(leaf);
2258 /* Read overflow data.
2260 data->mv_size = leaf->mn_dsize;
2261 memcpy(&pgno, NODEDATA(leaf), sizeof(pgno));
2262 if ((rc = mdb_get_page(txn, pgno, &omp))) {
2263 DPRINTF("read overflow page %lu failed", pgno);
2266 data->mv_data = METADATA(omp);
2272 mdb_get(MDB_txn *txn, MDB_dbi dbi,
2273 MDB_val *key, MDB_val *data)
2282 DPRINTF("===> get db %u key [%s]", dbi, DKEY(key));
2284 if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
2287 if (key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
2294 if (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT) {
2295 mc.mc_xcursor = &mx;
2296 mdb_xcursor_init0(&mc);
2298 mc.mc_xcursor = NULL;
2300 return mdb_cursor_set(&mc, key, data, MDB_SET, &exact);
2304 mdb_sibling(MDB_cursor *cursor, int move_right)
2311 if (cursor->mc_snum < 2) {
2312 return MDB_NOTFOUND; /* root has no siblings */
2314 parent = CURSOR_PARENT(cursor);
2316 DPRINTF("parent page is page %lu, index %u",
2317 parent->mp_page->mp_pgno, parent->mp_ki);
2319 cursor_pop_page(cursor);
2320 if (move_right ? (parent->mp_ki + 1 >= NUMKEYS(parent->mp_page))
2321 : (parent->mp_ki == 0)) {
2322 DPRINTF("no more keys left, moving to %s sibling",
2323 move_right ? "right" : "left");
2324 if ((rc = mdb_sibling(cursor, move_right)) != MDB_SUCCESS)
2326 parent = CURSOR_TOP(cursor);
2332 DPRINTF("just moving to %s index key %u",
2333 move_right ? "right" : "left", parent->mp_ki);
2335 assert(IS_BRANCH(parent->mp_page));
2337 indx = NODEPTR(parent->mp_page, parent->mp_ki);
2338 if ((rc = mdb_get_page(cursor->mc_txn, NODEPGNO(indx), &mp)))
2341 mp->parent = parent->mp_page;
2342 mp->parent_index = parent->mp_ki;
2345 cursor_push_page(cursor, mp);
2351 mdb_cursor_next(MDB_cursor *cursor, MDB_val *key, MDB_val *data, MDB_cursor_op op)
2358 if (cursor->mc_flags & C_EOF) {
2359 return MDB_NOTFOUND;
2362 assert(cursor->mc_flags & C_INITIALIZED);
2364 top = CURSOR_TOP(cursor);
2367 if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
2368 leaf = NODEPTR(mp, top->mp_ki);
2369 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
2370 if (op == MDB_NEXT || op == MDB_NEXT_DUP) {
2371 rc = mdb_cursor_next(&cursor->mc_xcursor->mx_cursor, data, NULL, MDB_NEXT);
2372 if (op != MDB_NEXT || rc == MDB_SUCCESS)
2376 cursor->mc_xcursor->mx_cursor.mc_flags = 0;
2377 if (op == MDB_NEXT_DUP)
2378 return MDB_NOTFOUND;
2382 DPRINTF("cursor_next: top page is %lu in cursor %p", mp->mp_pgno, (void *) cursor);
2384 if (top->mp_ki + 1 >= NUMKEYS(mp)) {
2385 DPUTS("=====> move to next sibling page");
2386 if (mdb_sibling(cursor, 1) != MDB_SUCCESS) {
2387 cursor->mc_flags |= C_EOF;
2388 return MDB_NOTFOUND;
2390 top = CURSOR_TOP(cursor);
2392 DPRINTF("next page is %lu, key index %u", mp->mp_pgno, top->mp_ki);
2396 DPRINTF("==> cursor points to page %lu with %u keys, key index %u",
2397 mp->mp_pgno, NUMKEYS(mp), top->mp_ki);
2400 key->mv_size = cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_pad;
2401 key->mv_data = LEAF2KEY(mp, top->mp_ki, key->mv_size);
2405 assert(IS_LEAF(mp));
2406 leaf = NODEPTR(mp, top->mp_ki);
2408 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
2409 mdb_xcursor_init1(cursor, mp, leaf);
2412 if ((rc = mdb_read_data(cursor->mc_txn, leaf, data) != MDB_SUCCESS))
2415 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
2416 rc = mdb_cursor_first(&cursor->mc_xcursor->mx_cursor, data, NULL);
2417 if (rc != MDB_SUCCESS)
2422 MDB_SET_KEY(leaf, key);
2427 mdb_cursor_prev(MDB_cursor *cursor, MDB_val *key, MDB_val *data, MDB_cursor_op op)
2434 assert(cursor->mc_flags & C_INITIALIZED);
2436 top = CURSOR_TOP(cursor);
2439 if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
2440 leaf = NODEPTR(mp, top->mp_ki);
2441 if (op == MDB_PREV || op == MDB_PREV_DUP) {
2442 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
2443 rc = mdb_cursor_prev(&cursor->mc_xcursor->mx_cursor, data, NULL, MDB_PREV);
2444 if (op != MDB_PREV || rc == MDB_SUCCESS)
2447 cursor->mc_xcursor->mx_cursor.mc_flags = 0;
2448 if (op == MDB_PREV_DUP)
2449 return MDB_NOTFOUND;
2454 DPRINTF("cursor_prev: top page is %lu in cursor %p", mp->mp_pgno, (void *) cursor);
2456 if (top->mp_ki == 0) {
2457 DPUTS("=====> move to prev sibling page");
2458 if (mdb_sibling(cursor, 0) != MDB_SUCCESS) {
2459 cursor->mc_flags &= ~C_INITIALIZED;
2460 return MDB_NOTFOUND;
2462 top = CURSOR_TOP(cursor);
2464 top->mp_ki = NUMKEYS(mp) - 1;
2465 DPRINTF("prev page is %lu, key index %u", mp->mp_pgno, top->mp_ki);
2469 cursor->mc_flags &= ~C_EOF;
2471 DPRINTF("==> cursor points to page %lu with %u keys, key index %u",
2472 mp->mp_pgno, NUMKEYS(mp), top->mp_ki);
2475 key->mv_size = cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_pad;
2476 key->mv_data = LEAF2KEY(mp, top->mp_ki, key->mv_size);
2480 assert(IS_LEAF(mp));
2481 leaf = NODEPTR(mp, top->mp_ki);
2483 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
2484 mdb_xcursor_init1(cursor, mp, leaf);
2487 if ((rc = mdb_read_data(cursor->mc_txn, leaf, data) != MDB_SUCCESS))
2490 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
2491 rc = mdb_cursor_last(&cursor->mc_xcursor->mx_cursor, data, NULL);
2492 if (rc != MDB_SUCCESS)
2497 MDB_SET_KEY(leaf, key);
2502 mdb_cursor_set(MDB_cursor *cursor, MDB_val *key, MDB_val *data,
2503 MDB_cursor_op op, int *exactp)
2513 assert(key->mv_size > 0);
2515 /* See if we're already on the right page */
2516 if (cursor->mc_flags & C_INITIALIZED) {
2519 top = CURSOR_TOP(cursor);
2520 if (top->mp_page->mp_flags & P_LEAF2) {
2521 nodekey.mv_size = cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_pad;
2522 nodekey.mv_data = LEAF2KEY(top->mp_page, 0, nodekey.mv_size);
2524 leaf = NODEPTR(top->mp_page, 0);
2525 MDB_SET_KEY(leaf, &nodekey);
2527 rc = mdb_cmp(cursor->mc_txn, cursor->mc_dbi, key, &nodekey);
2530 /* we're already on the right page */
2531 mpp.mp_page = top->mp_page;
2537 if (NUMKEYS(top->mp_page) > 1) {
2538 if (top->mp_page->mp_flags & P_LEAF2) {
2539 nodekey.mv_data = LEAF2KEY(top->mp_page,
2540 NUMKEYS(top->mp_page)-1, nodekey.mv_size);
2542 leaf = NODEPTR(top->mp_page, NUMKEYS(top->mp_page)-1);
2543 MDB_SET_KEY(leaf, &nodekey);
2545 rc = mdb_cmp(cursor->mc_txn, cursor->mc_dbi, key, &nodekey);
2546 if (rc <= 0) goto set1;
2548 /* If any parents have right-sibs, search.
2549 * Otherwise, there's nothing further.
2551 for (i=0; i<cursor->mc_snum-1; i++)
2552 if (cursor->mc_stack[i].mp_ki <
2553 NUMKEYS(cursor->mc_stack[i].mp_page)-1)
2555 if (i == cursor->mc_snum - 1) {
2556 /* There are no other pages */
2562 cursor->mc_snum = 0;
2564 rc = mdb_search_page(cursor->mc_txn, cursor->mc_dbi, key, cursor, 0, &mpp);
2565 if (rc != MDB_SUCCESS)
2568 assert(IS_LEAF(mpp.mp_page));
2570 top = CURSOR_TOP(cursor);
2572 leaf = mdb_search_node(cursor->mc_txn, cursor->mc_dbi, mpp.mp_page, key, exactp, &top->mp_ki);
2573 if (exactp != NULL && !*exactp) {
2574 /* MDB_SET specified and not an exact match. */
2575 return MDB_NOTFOUND;
2579 DPUTS("===> inexact leaf not found, goto sibling");
2580 if ((rc = mdb_sibling(cursor, 1)) != MDB_SUCCESS)
2581 return rc; /* no entries matched */
2582 top = CURSOR_TOP(cursor);
2584 mpp.mp_page = top->mp_page;
2585 assert(IS_LEAF(mpp.mp_page));
2586 leaf = NODEPTR(mpp.mp_page, 0);
2589 cursor->mc_flags |= C_INITIALIZED;
2590 cursor->mc_flags &= ~C_EOF;
2592 if (IS_LEAF2(mpp.mp_page)) {
2593 key->mv_size = cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_pad;
2594 key->mv_data = LEAF2KEY(mpp.mp_page, top->mp_ki, key->mv_size);
2598 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
2599 mdb_xcursor_init1(cursor, mpp.mp_page, leaf);
2602 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
2603 if (op == MDB_SET || op == MDB_SET_RANGE) {
2604 rc = mdb_cursor_first(&cursor->mc_xcursor->mx_cursor, data, NULL);
2607 if (op == MDB_GET_BOTH) {
2613 rc = mdb_cursor_set(&cursor->mc_xcursor->mx_cursor, data, NULL, MDB_SET_RANGE, ex2p);
2614 if (rc != MDB_SUCCESS)
2617 } else if (op == MDB_GET_BOTH || op == MDB_GET_BOTH_RANGE) {
2619 if ((rc = mdb_read_data(cursor->mc_txn, leaf, &d2)) != MDB_SUCCESS)
2621 rc = mdb_dcmp(cursor->mc_txn, cursor->mc_dbi, data, &d2);
2623 if (op == MDB_GET_BOTH || rc > 0)
2624 return MDB_NOTFOUND;
2628 if ((rc = mdb_read_data(cursor->mc_txn, leaf, data)) != MDB_SUCCESS)
2633 /* The key already matches in all other cases */
2634 if (op == MDB_SET_RANGE)
2635 MDB_SET_KEY(leaf, key);
2636 DPRINTF("==> cursor placed on key [%s]", DKEY(key));
2642 mdb_cursor_first(MDB_cursor *cursor, MDB_val *key, MDB_val *data)
2648 cursor->mc_snum = 0;
2650 rc = mdb_search_page(cursor->mc_txn, cursor->mc_dbi, NULL, cursor, 0, &mpp);
2651 if (rc != MDB_SUCCESS)
2653 assert(IS_LEAF(mpp.mp_page));
2655 leaf = NODEPTR(mpp.mp_page, 0);
2656 cursor->mc_flags |= C_INITIALIZED;
2657 cursor->mc_flags &= ~C_EOF;
2659 if (IS_LEAF2(mpp.mp_page)) {
2660 key->mv_size = cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_pad;
2661 key->mv_data = LEAF2KEY(mpp.mp_page, 0, key->mv_size);
2666 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
2667 mdb_xcursor_init1(cursor, mpp.mp_page, leaf);
2668 rc = mdb_cursor_first(&cursor->mc_xcursor->mx_cursor, data, NULL);
2672 if (cursor->mc_xcursor)
2673 cursor->mc_xcursor->mx_cursor.mc_flags = 0;
2674 if ((rc = mdb_read_data(cursor->mc_txn, leaf, data)) != MDB_SUCCESS)
2678 MDB_SET_KEY(leaf, key);
2683 mdb_cursor_last(MDB_cursor *cursor, MDB_val *key, MDB_val *data)
2691 cursor->mc_snum = 0;
2693 lkey.mv_size = MAXKEYSIZE+1;
2694 lkey.mv_data = NULL;
2696 rc = mdb_search_page(cursor->mc_txn, cursor->mc_dbi, &lkey, cursor, 0, &mpp);
2697 if (rc != MDB_SUCCESS)
2699 assert(IS_LEAF(mpp.mp_page));
2701 leaf = NODEPTR(mpp.mp_page, NUMKEYS(mpp.mp_page)-1);
2702 cursor->mc_flags |= C_INITIALIZED;
2703 cursor->mc_flags &= ~C_EOF;
2705 top = CURSOR_TOP(cursor);
2706 top->mp_ki = NUMKEYS(top->mp_page) - 1;
2708 if (IS_LEAF2(mpp.mp_page)) {
2709 key->mv_size = cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_pad;
2710 key->mv_data = LEAF2KEY(mpp.mp_page, top->mp_ki, key->mv_size);
2715 if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
2716 mdb_xcursor_init1(cursor, mpp.mp_page, leaf);
2717 rc = mdb_cursor_last(&cursor->mc_xcursor->mx_cursor, data, NULL);
2721 if ((rc = mdb_read_data(cursor->mc_txn, leaf, data)) != MDB_SUCCESS)
2726 MDB_SET_KEY(leaf, key);
2731 mdb_cursor_get(MDB_cursor *cursor, MDB_val *key, MDB_val *data,
2741 case MDB_GET_BOTH_RANGE:
2742 if (data == NULL || cursor->mc_xcursor == NULL) {
2749 if (key == NULL || key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
2751 } else if (op == MDB_SET_RANGE)
2752 rc = mdb_cursor_set(cursor, key, data, op, NULL);
2754 rc = mdb_cursor_set(cursor, key, data, op, &exact);
2756 case MDB_GET_MULTIPLE:
2758 !(cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPFIXED) ||
2759 !(cursor->mc_flags & C_INITIALIZED)) {
2764 if (!(cursor->mc_xcursor->mx_cursor.mc_flags & C_INITIALIZED) ||
2765 (cursor->mc_xcursor->mx_cursor.mc_flags & C_EOF))
2768 case MDB_NEXT_MULTIPLE:
2770 !(cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPFIXED)) {
2774 if (!(cursor->mc_flags & C_INITIALIZED))
2775 rc = mdb_cursor_first(cursor, key, data);
2777 rc = mdb_cursor_next(cursor, key, data, MDB_NEXT_DUP);
2778 if (rc == MDB_SUCCESS) {
2779 if (cursor->mc_xcursor->mx_cursor.mc_flags & C_INITIALIZED) {
2782 top = CURSOR_TOP(&cursor->mc_xcursor->mx_cursor);
2783 data->mv_size = NUMKEYS(top->mp_page) *
2784 cursor->mc_xcursor->mx_txn.mt_dbs[cursor->mc_xcursor->mx_cursor.mc_dbi].md_pad;
2785 data->mv_data = METADATA(top->mp_page);
2786 top->mp_ki = NUMKEYS(top->mp_page)-1;
2794 case MDB_NEXT_NODUP:
2795 if (!(cursor->mc_flags & C_INITIALIZED))
2796 rc = mdb_cursor_first(cursor, key, data);
2798 rc = mdb_cursor_next(cursor, key, data, op);
2802 case MDB_PREV_NODUP:
2803 if (!(cursor->mc_flags & C_INITIALIZED) || (cursor->mc_flags & C_EOF))
2804 rc = mdb_cursor_last(cursor, key, data);
2806 rc = mdb_cursor_prev(cursor, key, data, op);
2809 rc = mdb_cursor_first(cursor, key, data);
2813 !(cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) ||
2814 !(cursor->mc_flags & C_INITIALIZED) ||
2815 !(cursor->mc_xcursor->mx_cursor.mc_flags & C_INITIALIZED)) {
2819 rc = mdb_cursor_first(&cursor->mc_xcursor->mx_cursor, data, NULL);
2822 rc = mdb_cursor_last(cursor, key, data);
2826 !(cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) ||
2827 !(cursor->mc_flags & C_INITIALIZED) ||
2828 !(cursor->mc_xcursor->mx_cursor.mc_flags & C_INITIALIZED)) {
2832 rc = mdb_cursor_last(&cursor->mc_xcursor->mx_cursor, data, NULL);
2835 DPRINTF("unhandled/unimplemented cursor operation %u", op);
2844 mdb_cursor_touch(MDB_cursor *mc)
2851 mpp.mp_parent = NULL;
2853 if (mc->mc_dbi > MAIN_DBI && !mc->mc_txn->mt_dbxs[mc->mc_dbi].md_dirty) {
2854 rc = mdb_search_page(mc->mc_txn, MAIN_DBI, &mc->mc_txn->mt_dbxs[mc->mc_dbi].md_name,
2857 mc->mc_txn->mt_dbxs[mc->mc_dbi].md_dirty = 1;
2859 for(i=0; i<mc->mc_snum; i++) {
2860 mpp.mp_page = mc->mc_stack[i].mp_page;
2861 if (!F_ISSET(mpp.mp_page->mp_flags, P_DIRTY)) {
2862 rc = mdb_touch(mc->mc_txn, mc->mc_dbi, &mpp);
2864 mc->mc_stack[i].mp_page = mpp.mp_page;
2866 mc->mc_txn->mt_dbs[mc->mc_dbi].md_root = mpp.mp_page->mp_pgno;
2869 dh = ((MDB_dhead *)mpp.mp_page)-1;
2870 dh->md_parent = mpp.mp_parent;
2871 dh->md_pi = mpp.mp_pi;
2872 mpp.mp_parent = mpp.mp_page;
2873 mpp.mp_pi = mc->mc_stack[i].mp_ki;
2879 mdb_cursor_put(MDB_cursor *mc, MDB_val *key, MDB_val *data,
2884 MDB_val xdata, *rdata, dkey;
2886 char dbuf[PAGESIZE];
2892 if (F_ISSET(mc->mc_txn->mt_flags, MDB_TXN_RDONLY))
2895 DPRINTF("==> put db %u key [%s], size %zu, data size %zu",
2896 mc->mc_dbi, DKEY(key), key->mv_size, data->mv_size);
2900 if (flags == MDB_CURRENT) {
2901 if (!(mc->mc_flags & C_INITIALIZED))
2904 } else if (mc->mc_txn->mt_dbs[mc->mc_dbi].md_root == P_INVALID) {
2906 /* new database, write a root leaf page */
2907 DPUTS("allocating new root leaf page");
2908 if ((dp = mdb_new_page(mc->mc_txn, mc->mc_dbi, P_LEAF, 1)) == NULL) {
2912 cursor_push_page(mc, &dp->p);
2913 mc->mc_txn->mt_dbs[mc->mc_dbi].md_root = dp->p.mp_pgno;
2914 mc->mc_txn->mt_dbs[mc->mc_dbi].md_depth++;
2915 mc->mc_txn->mt_dbxs[mc->mc_dbi].md_dirty = 1;
2916 if ((mc->mc_txn->mt_dbs[mc->mc_dbi].md_flags & (MDB_DUPSORT|MDB_DUPFIXED))
2918 dp->p.mp_flags |= P_LEAF2;
2919 mc->mc_flags |= C_INITIALIZED;
2924 rc = mdb_cursor_set(mc, key, NULL, MDB_SET, &exact);
2925 if (flags == MDB_NOOVERWRITE && rc == 0) {
2926 DPRINTF("duplicate key [%s]", DKEY(key));
2927 return MDB_KEYEXIST;
2929 if (rc && rc != MDB_NOTFOUND)
2933 /* Cursor is positioned, now make sure all pages are writable */
2934 rc2 = mdb_cursor_touch(mc);
2935 if (rc2) return rc2;
2938 top = CURSOR_TOP(mc);
2940 /* The key already exists */
2941 if (rc == MDB_SUCCESS) {
2942 /* there's only a key anyway, so this is a no-op */
2943 if (IS_LEAF2(top->mp_page)) {
2944 int ksize = mc->mc_txn->mt_dbs[mc->mc_dbi].md_pad;
2945 if (key->mv_size != ksize)
2947 if (flags == MDB_CURRENT) {
2948 char *ptr = LEAF2KEY(top->mp_page, top->mp_ki, ksize);
2949 memcpy(ptr, key->mv_data, ksize);
2954 leaf = NODEPTR(top->mp_page, top->mp_ki);
2957 if (F_ISSET(mc->mc_txn->mt_dbs[mc->mc_dbi].md_flags, MDB_DUPSORT)) {
2958 /* Was a single item before, must convert now */
2959 if (!F_ISSET(leaf->mn_flags, F_DUPDATA)) {
2960 dkey.mv_size = NODEDSZ(leaf);
2961 dkey.mv_data = dbuf;
2962 memcpy(dbuf, NODEDATA(leaf), dkey.mv_size);
2963 /* data matches, ignore it */
2964 if (!mdb_dcmp(mc->mc_txn, mc->mc_dbi, data, &dkey))
2965 return (flags == MDB_NODUPDATA) ? MDB_KEYEXIST : MDB_SUCCESS;
2966 memset(&dummy, 0, sizeof(dummy));
2967 if (mc->mc_txn->mt_dbs[mc->mc_dbi].md_flags & MDB_DUPFIXED) {
2968 dummy.md_pad = data->mv_size;
2969 dummy.md_flags = MDB_DUPFIXED;
2970 if (mc->mc_txn->mt_dbs[mc->mc_dbi].md_flags & MDB_INTEGERDUP)
2971 dummy.md_flags |= MDB_INTEGERKEY;
2973 dummy.md_root = P_INVALID;
2974 if (dkey.mv_size == sizeof(MDB_db)) {
2975 memcpy(NODEDATA(leaf), &dummy, sizeof(dummy));
2978 mdb_del_node(top->mp_page, top->mp_ki, 0);
2981 xdata.mv_size = sizeof(MDB_db);
2982 xdata.mv_data = &dummy;
2987 /* same size, just replace it */
2988 if (!F_ISSET(leaf->mn_flags, F_BIGDATA) &&
2989 NODEDSZ(leaf) == data->mv_size) {
2990 memcpy(NODEDATA(leaf), data->mv_data, data->mv_size);
2993 mdb_del_node(top->mp_page, top->mp_ki, 0);
2995 DPRINTF("inserting key at index %i", top->mp_ki);
3001 nsize = IS_LEAF2(top->mp_page) ? key->mv_size : mdb_leaf_size(mc->mc_txn->mt_env, key, rdata);
3002 if (SIZELEFT(top->mp_page) < nsize) {
3003 rc = mdb_split(mc->mc_txn, mc->mc_dbi, &top->mp_page, &top->mp_ki, key, rdata, P_INVALID);
3005 /* There is room already in this leaf page. */
3006 rc = mdb_add_node(mc->mc_txn, mc->mc_dbi, top->mp_page, top->mp_ki, key, rdata, 0, 0);
3009 if (rc != MDB_SUCCESS)
3010 mc->mc_txn->mt_flags |= MDB_TXN_ERROR;
3012 /* Remember if we just added a subdatabase */
3013 if (flags & F_SUBDATA) {
3014 leaf = NODEPTR(top->mp_page, top->mp_ki);
3015 leaf->mn_flags |= F_SUBDATA;
3018 /* Now store the actual data in the child DB. Note that we're
3019 * storing the user data in the keys field, so there are strict
3020 * size limits on dupdata. The actual data fields of the child
3021 * DB are all zero size.
3024 leaf = NODEPTR(top->mp_page, top->mp_ki);
3026 if (flags == MDB_CURRENT)
3027 mdb_xcursor_init2(mc);
3029 mdb_xcursor_init1(mc, top->mp_page, leaf);
3032 if (flags == MDB_NODUPDATA)
3033 flags = MDB_NOOVERWRITE;
3034 /* converted, write the original data first */
3036 rc = mdb_cursor_put(&mc->mc_xcursor->mx_cursor, &dkey, &xdata, flags);
3038 leaf->mn_flags |= F_DUPDATA;
3040 rc = mdb_cursor_put(&mc->mc_xcursor->mx_cursor, data, &xdata, flags);
3041 mdb_xcursor_fini(mc);
3042 memcpy(NODEDATA(leaf),
3043 &mc->mc_xcursor->mx_txn.mt_dbs[mc->mc_xcursor->mx_cursor.mc_dbi],
3046 mc->mc_txn->mt_dbs[mc->mc_dbi].md_entries++;
3053 mdb_cursor_del(MDB_cursor *mc, unsigned int flags)
3060 if (F_ISSET(mc->mc_txn->mt_flags, MDB_TXN_RDONLY))
3063 if (!mc->mc_flags & C_INITIALIZED)
3066 rc = mdb_cursor_touch(mc);
3069 top = CURSOR_TOP(mc);
3070 leaf = NODEPTR(top->mp_page, top->mp_ki);
3071 mpp.mp_page = top->mp_page;
3072 if (mc->mc_snum > 1) {
3073 MDB_ppage *parent = CURSOR_PARENT(mc);
3074 mpp.mp_parent = parent->mp_page;
3075 mpp.mp_pi = parent->mp_ki;
3077 mpp.mp_parent = NULL;
3081 if (!IS_LEAF2(top->mp_page) && F_ISSET(leaf->mn_flags, F_DUPDATA)) {
3084 if (flags != MDB_NODUPDATA) {
3085 /* mdb_xcursor_init2(mc); */
3086 rc = mdb_cursor_del(&mc->mc_xcursor->mx_cursor, 0);
3087 mdb_xcursor_fini(mc);
3088 /* If sub-DB still has entries, we're done */
3089 if (mc->mc_xcursor->mx_txn.mt_dbs[mc->mc_xcursor->mx_cursor.mc_dbi].md_root
3091 memcpy(NODEDATA(leaf),
3092 &mc->mc_xcursor->mx_txn.mt_dbs[mc->mc_xcursor->mx_cursor.mc_dbi],
3094 mc->mc_txn->mt_dbs[mc->mc_dbi].md_entries--;
3097 /* otherwise fall thru and delete the sub-DB */
3100 /* add all the child DB's pages to the free list */
3101 mc->mc_xcursor->mx_cursor.mc_snum = 0;
3102 rc = mdb_search_page(&mc->mc_xcursor->mx_txn, mc->mc_xcursor->mx_cursor.mc_dbi,
3103 NULL, &mc->mc_xcursor->mx_cursor, 0, &mp2);
3104 if (rc == MDB_SUCCESS) {
3105 MDB_ppage *top, *parent;
3112 mc->mc_txn->mt_dbs[mc->mc_dbi].md_entries -=
3113 mc->mc_xcursor->mx_txn.mt_dbs[mc->mc_xcursor->mx_cursor.mc_dbi].md_entries;
3115 cursor_pop_page(&mc->mc_xcursor->mx_cursor);
3116 if (mc->mc_xcursor->mx_cursor.mc_snum) {
3118 if (mc->mc_xcursor->mx_cursor.mc_stack[0].mp_page->mp_flags & P_DIRTY) {
3122 while (mc->mc_xcursor->mx_cursor.mc_snum > 1) {
3123 top = CURSOR_TOP(&mc->mc_xcursor->mx_cursor);
3124 parent = CURSOR_PARENT(&mc->mc_xcursor->mx_cursor);
3125 for (i=0; i<NUMKEYS(top->mp_page); i++) {
3128 ni = NODEPTR(top->mp_page, i);
3130 if ((rc = mdb_get_page(mc->mc_txn, pg, &mp)))
3133 if (mp->mp_flags & P_DIRTY) {
3136 mdb_mid2l_delete(mc->mc_txn->mt_u.dirty_list, &mid);
3138 dp->h.md_parent = (MDB_page *)mc->mc_txn->mt_env->me_dpages;
3139 mc->mc_txn->mt_env->me_dpages = dp;
3144 mdb_midl_insert(mc->mc_txn->mt_free_pgs, pg);
3147 rc = mdb_sibling(&mc->mc_xcursor->mx_cursor, 1);
3154 mid.mid = mc->mc_xcursor->mx_txn.mt_dbs[mc->mc_xcursor->mx_cursor.mc_dbi].md_root;
3155 mdb_mid2l_delete(mc->mc_txn->mt_u.dirty_list, &mid);
3157 dp->h.md_parent = (MDB_page *)mc->mc_txn->mt_env->me_dpages;
3158 mc->mc_txn->mt_env->me_dpages = dp;
3163 mdb_midl_insert(mc->mc_txn->mt_free_pgs,
3164 mc->mc_xcursor->mx_txn.mt_dbs[mc->mc_xcursor->mx_cursor.mc_dbi].md_root);
3169 return mdb_del0(mc, top->mp_ki, &mpp, leaf);
3172 /* Allocate a page and initialize it
3175 mdb_new_page(MDB_txn *txn, MDB_dbi dbi, uint32_t flags, int num)
3179 if ((dp = mdb_alloc_page(txn, dbi, NULL, 0, num)) == NULL)
3181 DPRINTF("allocated new mpage %lu, page size %u",
3182 dp->p.mp_pgno, txn->mt_env->me_psize);
3183 dp->p.mp_flags = flags | P_DIRTY;
3184 dp->p.mp_lower = PAGEHDRSZ;
3185 dp->p.mp_upper = txn->mt_env->me_psize;
3187 if (IS_BRANCH(&dp->p))
3188 txn->mt_dbs[dbi].md_branch_pages++;
3189 else if (IS_LEAF(&dp->p))
3190 txn->mt_dbs[dbi].md_leaf_pages++;
3191 else if (IS_OVERFLOW(&dp->p)) {
3192 txn->mt_dbs[dbi].md_overflow_pages += num;
3193 dp->p.mp_pages = num;
3200 mdb_leaf_size(MDB_env *env, MDB_val *key, MDB_val *data)
3204 sz = LEAFSIZE(key, data);
3205 if (data->mv_size >= env->me_psize / MDB_MINKEYS) {
3206 /* put on overflow page */
3207 sz -= data->mv_size - sizeof(pgno_t);
3210 return sz + sizeof(indx_t);
3214 mdb_branch_size(MDB_env *env, MDB_val *key)
3219 if (sz >= env->me_psize / MDB_MINKEYS) {
3220 /* put on overflow page */
3221 /* not implemented */
3222 /* sz -= key->size - sizeof(pgno_t); */
3225 return sz + sizeof(indx_t);
3229 mdb_add_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp, indx_t indx,
3230 MDB_val *key, MDB_val *data, pgno_t pgno, uint8_t flags)
3233 size_t node_size = NODESIZE;
3236 MDB_dpage *ofp = NULL; /* overflow page */
3239 assert(mp->mp_upper >= mp->mp_lower);
3241 DPRINTF("add to %s page %lu index %i, data size %zu key size %zu [%s]",
3242 IS_LEAF(mp) ? "leaf" : "branch",
3243 mp->mp_pgno, indx, data ? data->mv_size : 0,
3244 key ? key->mv_size : 0, key ? DKEY(key) : NULL);
3247 /* Move higher keys up one slot. */
3248 int ksize = txn->mt_dbs[dbi].md_pad, dif;
3249 char *ptr = LEAF2KEY(mp, indx, ksize);
3250 dif = NUMKEYS(mp) - indx;
3252 memmove(ptr+ksize, ptr, dif*ksize);
3253 /* insert new key */
3254 memcpy(ptr, key->mv_data, ksize);
3256 /* Just using these for counting */
3257 mp->mp_lower += sizeof(indx_t);
3258 mp->mp_upper -= ksize - sizeof(indx_t);
3263 node_size += key->mv_size;
3267 if (F_ISSET(flags, F_BIGDATA)) {
3268 /* Data already on overflow page. */
3269 node_size += sizeof(pgno_t);
3270 } else if (data->mv_size >= txn->mt_env->me_psize / MDB_MINKEYS) {
3271 int ovpages = OVPAGES(data->mv_size, txn->mt_env->me_psize);
3272 /* Put data on overflow page. */
3273 DPRINTF("data size is %zu, put on overflow page",
3275 node_size += sizeof(pgno_t);
3276 if ((ofp = mdb_new_page(txn, dbi, P_OVERFLOW, ovpages)) == NULL)
3278 DPRINTF("allocated overflow page %lu", ofp->p.mp_pgno);
3281 node_size += data->mv_size;
3285 if (node_size + sizeof(indx_t) > SIZELEFT(mp)) {
3286 DPRINTF("not enough room in page %lu, got %u ptrs",
3287 mp->mp_pgno, NUMKEYS(mp));
3288 DPRINTF("upper - lower = %u - %u = %u", mp->mp_upper, mp->mp_lower,
3289 mp->mp_upper - mp->mp_lower);
3290 DPRINTF("node size = %zu", node_size);
3294 /* Move higher pointers up one slot. */
3295 for (i = NUMKEYS(mp); i > indx; i--)
3296 mp->mp_ptrs[i] = mp->mp_ptrs[i - 1];
3298 /* Adjust free space offsets. */
3299 ofs = mp->mp_upper - node_size;
3300 assert(ofs >= mp->mp_lower + sizeof(indx_t));
3301 mp->mp_ptrs[indx] = ofs;
3303 mp->mp_lower += sizeof(indx_t);
3305 /* Write the node data. */
3306 node = NODEPTR(mp, indx);
3307 node->mn_ksize = (key == NULL) ? 0 : key->mv_size;
3308 node->mn_flags = flags;
3310 node->mn_dsize = data->mv_size;
3312 NODEPGNO(node) = pgno;
3315 memcpy(NODEKEY(node), key->mv_data, key->mv_size);
3320 if (F_ISSET(flags, F_BIGDATA))
3321 memcpy(node->mn_data + key->mv_size, data->mv_data,
3324 memcpy(node->mn_data + key->mv_size, data->mv_data,
3327 memcpy(node->mn_data + key->mv_size, &ofp->p.mp_pgno,
3329 memcpy(METADATA(&ofp->p), data->mv_data, data->mv_size);
3337 mdb_del_node(MDB_page *mp, indx_t indx, int ksize)
3340 indx_t i, j, numkeys, ptr;
3344 DPRINTF("delete node %u on %s page %lu", indx,
3345 IS_LEAF(mp) ? "leaf" : "branch", mp->mp_pgno);
3346 assert(indx < NUMKEYS(mp));
3349 int x = NUMKEYS(mp) - 1 - indx;
3350 base = LEAF2KEY(mp, indx, ksize);
3352 memmove(base, base + ksize, x * ksize);
3353 mp->mp_lower -= sizeof(indx_t);
3354 mp->mp_upper += ksize - sizeof(indx_t);
3358 node = NODEPTR(mp, indx);
3359 sz = NODESIZE + node->mn_ksize;
3361 if (F_ISSET(node->mn_flags, F_BIGDATA))
3362 sz += sizeof(pgno_t);
3364 sz += NODEDSZ(node);
3367 ptr = mp->mp_ptrs[indx];
3368 numkeys = NUMKEYS(mp);
3369 for (i = j = 0; i < numkeys; i++) {
3371 mp->mp_ptrs[j] = mp->mp_ptrs[i];
3372 if (mp->mp_ptrs[i] < ptr)
3373 mp->mp_ptrs[j] += sz;
3378 base = (char *)mp + mp->mp_upper;
3379 memmove(base + sz, base, ptr - mp->mp_upper);
3381 mp->mp_lower -= sizeof(indx_t);
3386 mdb_xcursor_init0(MDB_cursor *mc)
3388 MDB_xcursor *mx = mc->mc_xcursor;
3391 mx->mx_txn = *mc->mc_txn;
3392 mx->mx_txn.mt_dbxs = mx->mx_dbxs;
3393 mx->mx_txn.mt_dbs = mx->mx_dbs;
3394 mx->mx_dbxs[0] = mc->mc_txn->mt_dbxs[0];
3395 mx->mx_dbxs[1] = mc->mc_txn->mt_dbxs[1];
3396 if (mc->mc_dbi > 1) {
3397 mx->mx_dbxs[2] = mc->mc_txn->mt_dbxs[mc->mc_dbi];
3402 mx->mx_dbxs[dbn+1].md_parent = dbn;
3403 mx->mx_dbxs[dbn+1].md_cmp = mx->mx_dbxs[dbn].md_dcmp;
3404 mx->mx_dbxs[dbn+1].md_rel = mx->mx_dbxs[dbn].md_rel;
3405 mx->mx_dbxs[dbn+1].md_dirty = 0;
3406 mx->mx_txn.mt_numdbs = dbn+2;
3408 mx->mx_cursor.mc_xcursor = NULL;
3409 mx->mx_cursor.mc_txn = &mx->mx_txn;
3410 mx->mx_cursor.mc_dbi = dbn+1;
3414 mdb_xcursor_init1(MDB_cursor *mc, MDB_page *mp, MDB_node *node)
3416 MDB_db *db = NODEDATA(node);
3417 MDB_xcursor *mx = mc->mc_xcursor;
3419 mx->mx_dbs[0] = mc->mc_txn->mt_dbs[0];
3420 mx->mx_dbs[1] = mc->mc_txn->mt_dbs[1];
3421 if (mc->mc_dbi > 1) {
3422 mx->mx_dbs[2] = mc->mc_txn->mt_dbs[mc->mc_dbi];
3427 DPRINTF("Sub-db %u for db %u root page %lu", dbn, mc->mc_dbi, db->md_root);
3428 mx->mx_dbs[dbn] = *db;
3429 if (F_ISSET(mp->mp_flags, P_DIRTY))
3430 mx->mx_dbxs[dbn].md_dirty = 1;
3431 mx->mx_dbxs[dbn].md_name.mv_data = NODEKEY(node);
3432 mx->mx_dbxs[dbn].md_name.mv_size = node->mn_ksize;
3433 mx->mx_txn.mt_next_pgno = mc->mc_txn->mt_next_pgno;
3434 mx->mx_txn.mt_u = mc->mc_txn->mt_u;
3435 mx->mx_cursor.mc_snum = 0;
3436 mx->mx_cursor.mc_flags = 0;
3440 mdb_xcursor_init2(MDB_cursor *mc)
3442 MDB_xcursor *mx = mc->mc_xcursor;
3444 mx->mx_dbs[0] = mc->mc_txn->mt_dbs[0];
3445 mx->mx_dbs[1] = mc->mc_txn->mt_dbs[1];
3446 if (mc->mc_dbi > 1) {
3447 mx->mx_dbs[2] = mc->mc_txn->mt_dbs[mc->mc_dbi];
3452 DPRINTF("Sub-db %u for db %u root page %lu", dbn, mc->mc_dbi,
3453 mx->mx_dbs[dbn].md_root);
3454 mx->mx_txn.mt_next_pgno = mc->mc_txn->mt_next_pgno;
3458 mdb_xcursor_fini(MDB_cursor *mc)
3460 MDB_xcursor *mx = mc->mc_xcursor;
3461 mc->mc_txn->mt_next_pgno = mx->mx_txn.mt_next_pgno;
3462 mc->mc_txn->mt_dbs[0] = mx->mx_dbs[0];
3463 mc->mc_txn->mt_dbs[1] = mx->mx_dbs[1];
3464 mc->mc_txn->mt_dbxs[0].md_dirty = mx->mx_dbxs[0].md_dirty;
3465 mc->mc_txn->mt_dbxs[1].md_dirty = mx->mx_dbxs[1].md_dirty;
3466 if (mc->mc_dbi > 1) {
3467 mc->mc_txn->mt_dbs[mc->mc_dbi] = mx->mx_dbs[2];
3468 mc->mc_txn->mt_dbxs[mc->mc_dbi].md_dirty = mx->mx_dbxs[2].md_dirty;
3473 mdb_cursor_open(MDB_txn *txn, MDB_dbi dbi, MDB_cursor **ret)
3476 size_t size = sizeof(MDB_cursor);
3478 if (txn == NULL || ret == NULL || !dbi || dbi >= txn->mt_numdbs)
3481 if (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT)
3482 size += sizeof(MDB_xcursor);
3484 if ((cursor = calloc(1, size)) != NULL) {
3485 cursor->mc_dbi = dbi;
3486 cursor->mc_txn = txn;
3487 if (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT) {
3488 MDB_xcursor *mx = (MDB_xcursor *)(cursor + 1);
3489 cursor->mc_xcursor = mx;
3490 mdb_xcursor_init0(cursor);
3501 /* Return the count of duplicate data items for the current key */
3503 mdb_cursor_count(MDB_cursor *mc, unsigned long *countp)
3508 if (mc == NULL || countp == NULL)
3511 if (!(mc->mc_txn->mt_dbs[mc->mc_dbi].md_flags & MDB_DUPSORT))
3514 top = CURSOR_TOP(mc);
3515 leaf = NODEPTR(top->mp_page, top->mp_ki);
3516 if (!F_ISSET(leaf->mn_flags, F_DUPDATA)) {
3519 if (!(mc->mc_xcursor->mx_cursor.mc_flags & C_INITIALIZED))
3522 *countp = mc->mc_xcursor->mx_txn.mt_dbs[mc->mc_xcursor->mx_cursor.mc_dbi].md_entries;
3529 mdb_cursor_reset(MDB_cursor *mc)
3532 if (mc->mc_flags & C_XDIRTY) {
3534 mdb_xcursor_fini(mc);
3536 /* If sub-DB still has entries, update root record */
3537 if (mc->mc_xcursor->mx_txn.mt_dbs[mc->mc_xcursor->mx_cursor.mc_dbi].md_root
3541 top = CURSOR_TOP(mc);
3542 leaf = NODEPTR(top->mp_page, top->mp_ki);
3543 memcpy(NODEDATA(leaf),
3544 &mc->mc_xcursor->mx_txn.mt_dbs[mc->mc_xcursor->mx_cursor.mc_dbi],
3547 mc->mc_flags ^= C_XDIRTY;
3554 mdb_cursor_close(MDB_cursor *cursor)
3556 if (cursor != NULL) {
3562 mdb_update_key(MDB_page *mp, indx_t indx, MDB_val *key)
3564 indx_t ptr, i, numkeys;
3571 node = NODEPTR(mp, indx);
3572 ptr = mp->mp_ptrs[indx];
3573 DPRINTF("update key %u (ofs %u) [%.*s] to [%s] on page %lu",
3575 (int)node->mn_ksize, (char *)NODEKEY(node),
3579 delta = key->mv_size - node->mn_ksize;
3581 if (delta > 0 && SIZELEFT(mp) < delta) {
3582 DPRINTF("OUCH! Not enough room, delta = %d", delta);
3586 numkeys = NUMKEYS(mp);
3587 for (i = 0; i < numkeys; i++) {
3588 if (mp->mp_ptrs[i] <= ptr)
3589 mp->mp_ptrs[i] -= delta;
3592 base = (char *)mp + mp->mp_upper;
3593 len = ptr - mp->mp_upper + NODESIZE;
3594 memmove(base - delta, base, len);
3595 mp->mp_upper -= delta;
3597 node = NODEPTR(mp, indx);
3598 node->mn_ksize = key->mv_size;
3601 memcpy(NODEKEY(node), key->mv_data, key->mv_size);
3606 /* Move a node from src to dst.
3609 mdb_move_node(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *src, indx_t srcindx,
3610 MDB_pageparent *dst, indx_t dstindx)
3617 /* Mark src and dst as dirty. */
3618 if ((rc = mdb_touch(txn, dbi, src)) ||
3619 (rc = mdb_touch(txn, dbi, dst)))
3622 if (IS_LEAF2(src->mp_page)) {
3623 srcnode = NODEPTR(src->mp_page, 0); /* fake */
3624 key.mv_size = txn->mt_dbs[dbi].md_pad;
3625 key.mv_data = LEAF2KEY(src->mp_page, srcindx, key.mv_size);
3627 data.mv_data = NULL;
3629 srcnode = NODEPTR(src->mp_page, srcindx);
3630 key.mv_size = NODEKSZ(srcnode);
3631 key.mv_data = NODEKEY(srcnode);
3632 data.mv_size = NODEDSZ(srcnode);
3633 data.mv_data = NODEDATA(srcnode);
3635 DPRINTF("moving %s node %u [%s] on page %lu to node %u on page %lu",
3636 IS_LEAF(src->mp_page) ? "leaf" : "branch",
3639 src->mp_page->mp_pgno,
3640 dstindx, dst->mp_page->mp_pgno);
3642 /* Add the node to the destination page.
3644 rc = mdb_add_node(txn, dbi, dst->mp_page, dstindx, &key, &data, NODEPGNO(srcnode),
3646 if (rc != MDB_SUCCESS)
3649 /* Delete the node from the source page.
3651 mdb_del_node(src->mp_page, srcindx, key.mv_size);
3653 /* The key value just changed due to del_node, find it again.
3655 if (!IS_LEAF2(src->mp_page)) {
3656 srcnode = NODEPTR(src->mp_page, srcindx);
3657 key.mv_data = NODEKEY(srcnode);
3660 /* Update the parent separators.
3663 if (src->mp_pi != 0) {
3664 DPRINTF("update separator for source page %lu to [%s]",
3665 src->mp_page->mp_pgno, DKEY(&key));
3666 if ((rc = mdb_update_key(src->mp_parent, src->mp_pi,
3667 &key)) != MDB_SUCCESS)
3670 if (IS_BRANCH(src->mp_page)) {
3672 nullkey.mv_size = 0;
3673 assert(mdb_update_key(src->mp_page, 0, &nullkey) == MDB_SUCCESS);
3678 if (dst->mp_pi != 0) {
3679 DPRINTF("update separator for destination page %lu to [%s]",
3680 dst->mp_page->mp_pgno, DKEY(&key));
3681 if ((rc = mdb_update_key(dst->mp_parent, dst->mp_pi,
3682 &key)) != MDB_SUCCESS)
3685 if (IS_BRANCH(dst->mp_page)) {
3687 nullkey.mv_size = 0;
3688 assert(mdb_update_key(dst->mp_page, 0, &nullkey) == MDB_SUCCESS);
3696 mdb_merge(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *src, MDB_pageparent *dst)
3705 DPRINTF("merging page %lu and %lu", src->mp_page->mp_pgno, dst->mp_page->mp_pgno);
3707 assert(txn != NULL);
3708 assert(src->mp_parent); /* can't merge root page */
3709 assert(dst->mp_parent);
3711 /* Mark src and dst as dirty. */
3712 if ((rc = mdb_touch(txn, dbi, src)) ||
3713 (rc = mdb_touch(txn, dbi, dst)))
3716 /* Move all nodes from src to dst.
3718 if (IS_LEAF2(src->mp_page)) {
3719 key.mv_size = txn->mt_dbs[dbi].md_pad;
3720 key.mv_data = METADATA(src->mp_page);
3721 for (i = 0; i < NUMKEYS(src->mp_page); i++) {
3722 rc = mdb_add_node(txn, dbi, dst->mp_page, NUMKEYS(dst->mp_page), &key,
3724 if (rc != MDB_SUCCESS)
3726 key.mv_data = (char *)key.mv_data + key.mv_size;
3729 for (i = 0; i < NUMKEYS(src->mp_page); i++) {
3730 srcnode = NODEPTR(src->mp_page, i);
3732 key.mv_size = srcnode->mn_ksize;
3733 key.mv_data = NODEKEY(srcnode);
3734 data.mv_size = NODEDSZ(srcnode);
3735 data.mv_data = NODEDATA(srcnode);
3736 rc = mdb_add_node(txn, dbi, dst->mp_page, NUMKEYS(dst->mp_page), &key,
3737 &data, NODEPGNO(srcnode), srcnode->mn_flags);
3738 if (rc != MDB_SUCCESS)
3743 DPRINTF("dst page %lu now has %u keys (%.1f%% filled)",
3744 dst->mp_page->mp_pgno, NUMKEYS(dst->mp_page), (float)PAGEFILL(txn->mt_env, dst->mp_page) / 10);
3746 /* Unlink the src page from parent.
3748 mdb_del_node(src->mp_parent, src->mp_pi, 0);
3749 if (src->mp_pi == 0) {
3751 if ((rc = mdb_update_key(src->mp_parent, 0, &key)) != MDB_SUCCESS)
3755 if (IS_LEAF(src->mp_page))
3756 txn->mt_dbs[dbi].md_leaf_pages--;
3758 txn->mt_dbs[dbi].md_branch_pages--;
3760 mpp.mp_page = src->mp_parent;
3761 dh = (MDB_dhead *)src->mp_parent;
3763 mpp.mp_parent = dh->md_parent;
3764 mpp.mp_pi = dh->md_pi;
3766 return mdb_rebalance(txn, dbi, &mpp);
3769 #define FILL_THRESHOLD 250
3772 mdb_rebalance(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *mpp)
3777 indx_t si = 0, di = 0;
3780 assert(txn != NULL);
3781 assert(mpp != NULL);
3783 DPRINTF("rebalancing %s page %lu (has %u keys, %.1f%% full)",
3784 IS_LEAF(mpp->mp_page) ? "leaf" : "branch",
3785 mpp->mp_page->mp_pgno, NUMKEYS(mpp->mp_page), (float)PAGEFILL(txn->mt_env, mpp->mp_page) / 10);
3787 if (PAGEFILL(txn->mt_env, mpp->mp_page) >= FILL_THRESHOLD) {
3788 DPRINTF("no need to rebalance page %lu, above fill threshold",
3789 mpp->mp_page->mp_pgno);
3793 if (mpp->mp_parent == NULL) {
3794 if (NUMKEYS(mpp->mp_page) == 0) {
3795 DPUTS("tree is completely empty");
3796 txn->mt_dbs[dbi].md_root = P_INVALID;
3797 txn->mt_dbs[dbi].md_depth = 0;
3798 txn->mt_dbs[dbi].md_leaf_pages = 0;
3799 mdb_midl_insert(txn->mt_free_pgs, mpp->mp_page->mp_pgno);
3800 } else if (IS_BRANCH(mpp->mp_page) && NUMKEYS(mpp->mp_page) == 1) {
3801 DPUTS("collapsing root page!");
3802 mdb_midl_insert(txn->mt_free_pgs, mpp->mp_page->mp_pgno);
3803 txn->mt_dbs[dbi].md_root = NODEPGNO(NODEPTR(mpp->mp_page, 0));
3804 if ((rc = mdb_get_page(txn, txn->mt_dbs[dbi].md_root, &root)))
3806 txn->mt_dbs[dbi].md_depth--;
3807 txn->mt_dbs[dbi].md_branch_pages--;
3809 DPUTS("root page doesn't need rebalancing");
3813 /* The parent (branch page) must have at least 2 pointers,
3814 * otherwise the tree is invalid.
3816 assert(NUMKEYS(mpp->mp_parent) > 1);
3818 /* Leaf page fill factor is below the threshold.
3819 * Try to move keys from left or right neighbor, or
3820 * merge with a neighbor page.
3825 if (mpp->mp_pi == 0) {
3826 /* We're the leftmost leaf in our parent.
3828 DPUTS("reading right neighbor");
3829 node = NODEPTR(mpp->mp_parent, mpp->mp_pi + 1);
3830 if ((rc = mdb_get_page(txn, NODEPGNO(node), &npp.mp_page)))
3832 npp.mp_pi = mpp->mp_pi + 1;
3834 di = NUMKEYS(mpp->mp_page);
3836 /* There is at least one neighbor to the left.
3838 DPUTS("reading left neighbor");
3839 node = NODEPTR(mpp->mp_parent, mpp->mp_pi - 1);
3840 if ((rc = mdb_get_page(txn, NODEPGNO(node), &npp.mp_page)))
3842 npp.mp_pi = mpp->mp_pi - 1;
3843 si = NUMKEYS(npp.mp_page) - 1;
3846 npp.mp_parent = mpp->mp_parent;
3848 DPRINTF("found neighbor page %lu (%u keys, %.1f%% full)",
3849 npp.mp_page->mp_pgno, NUMKEYS(npp.mp_page), (float)PAGEFILL(txn->mt_env, npp.mp_page) / 10);
3851 /* If the neighbor page is above threshold and has at least two
3852 * keys, move one key from it.
3854 * Otherwise we should try to merge them.
3856 if (PAGEFILL(txn->mt_env, npp.mp_page) >= FILL_THRESHOLD && NUMKEYS(npp.mp_page) >= 2)
3857 return mdb_move_node(txn, dbi, &npp, si, mpp, di);
3858 else { /* FIXME: if (has_enough_room()) */
3859 if (mpp->mp_pi == 0)
3860 return mdb_merge(txn, dbi, &npp, mpp);
3862 return mdb_merge(txn, dbi, mpp, &npp);
3867 mdb_del0(MDB_cursor *mc, unsigned int ki, MDB_pageparent *mpp, MDB_node *leaf)
3871 /* add overflow pages to free list */
3872 if (!IS_LEAF2(mpp->mp_page) && F_ISSET(leaf->mn_flags, F_BIGDATA)) {
3876 memcpy(&pg, NODEDATA(leaf), sizeof(pg));
3877 ovpages = OVPAGES(NODEDSZ(leaf), mc->mc_txn->mt_env->me_psize);
3878 for (i=0; i<ovpages; i++) {
3879 DPRINTF("freed ov page %lu", pg);
3880 mdb_midl_insert(mc->mc_txn->mt_free_pgs, pg);
3884 mdb_del_node(mpp->mp_page, ki, mc->mc_txn->mt_dbs[mc->mc_dbi].md_pad);
3885 mc->mc_txn->mt_dbs[mc->mc_dbi].md_entries--;
3886 rc = mdb_rebalance(mc->mc_txn, mc->mc_dbi, mpp);
3887 if (rc != MDB_SUCCESS)
3888 mc->mc_txn->mt_flags |= MDB_TXN_ERROR;
3894 mdb_del(MDB_txn *txn, MDB_dbi dbi,
3895 MDB_val *key, MDB_val *data)
3900 MDB_val rdata, *xdata;
3904 assert(key != NULL);
3906 DPRINTF("====> delete db %u key [%s]", dbi, DKEY(key));
3908 if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
3911 if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
3915 if (key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
3922 if (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT) {
3923 mc.mc_xcursor = &mx;
3924 mdb_xcursor_init0(&mc);
3926 mc.mc_xcursor = NULL;
3938 rc = mdb_cursor_set(&mc, key, xdata, op, &exact);
3940 rc = mdb_cursor_del(&mc, data ? 0 : MDB_NODUPDATA);
3944 /* Split page <*mpp>, and insert <key,(data|newpgno)> in either left or
3945 * right sibling, at index <*newindxp> (as if unsplit). Updates *mpp and
3946 * *newindxp with the actual values after split, ie if *mpp and *newindxp
3947 * refer to a node in the new right sibling page.
3950 mdb_split(MDB_txn *txn, MDB_dbi dbi, MDB_page **mpp, unsigned int *newindxp,
3951 MDB_val *newkey, MDB_val *newdata, pgno_t newpgno)
3954 int rc = MDB_SUCCESS, ins_new = 0;
3957 unsigned int i, j, split_indx, nkeys, pmax;
3959 MDB_val sepkey, rkey, rdata;
3960 MDB_page *copy, *cptr;
3961 MDB_dpage *mdp, *rdp, *pdp;
3965 assert(txn != NULL);
3967 dh = ((MDB_dhead *)*mpp) - 1;
3968 mdp = (MDB_dpage *)dh;
3969 newindx = *newindxp;
3971 DPRINTF("-----> splitting %s page %lu and adding [%s] at index %i",
3972 IS_LEAF(&mdp->p) ? "leaf" : "branch", mdp->p.mp_pgno,
3973 DKEY(newkey), *newindxp);
3975 if (mdp->h.md_parent == NULL) {
3976 if ((pdp = mdb_new_page(txn, dbi, P_BRANCH, 1)) == NULL)
3979 mdp->h.md_parent = &pdp->p;
3980 txn->mt_dbs[dbi].md_root = pdp->p.mp_pgno;
3981 DPRINTF("root split! new root = %lu", pdp->p.mp_pgno);
3982 txn->mt_dbs[dbi].md_depth++;
3984 /* Add left (implicit) pointer. */
3985 if ((rc = mdb_add_node(txn, dbi, &pdp->p, 0, NULL, NULL,
3986 mdp->p.mp_pgno, 0)) != MDB_SUCCESS)
3989 DPRINTF("parent branch page is %lu", mdp->h.md_parent->mp_pgno);
3992 /* Create a right sibling. */
3993 if ((rdp = mdb_new_page(txn, dbi, mdp->p.mp_flags, 1)) == NULL)
3995 rdp->h.md_parent = mdp->h.md_parent;
3996 rdp->h.md_pi = mdp->h.md_pi + 1;
3997 DPRINTF("new right sibling: page %lu", rdp->p.mp_pgno);
3999 nkeys = NUMKEYS(&mdp->p);
4000 split_indx = nkeys / 2 + 1;
4002 if (IS_LEAF2(&rdp->p)) {
4005 unsigned int lsize, rsize, ksize;
4006 /* Move half of the keys to the right sibling */
4008 x = *newindxp - split_indx;
4009 ksize = txn->mt_dbs[dbi].md_pad;
4010 split = LEAF2KEY(&mdp->p, split_indx, ksize);
4011 rsize = (nkeys - split_indx) * ksize;
4012 lsize = (nkeys - split_indx) * sizeof(indx_t);
4013 mdp->p.mp_lower -= lsize;
4014 rdp->p.mp_lower += lsize;
4015 mdp->p.mp_upper += rsize - lsize;
4016 rdp->p.mp_upper -= rsize - lsize;
4017 sepkey.mv_size = ksize;
4018 if (newindx == split_indx) {
4019 sepkey.mv_data = newkey->mv_data;
4021 sepkey.mv_data = split;
4024 ins = LEAF2KEY(&mdp->p, *newindxp, ksize);
4025 memcpy(&rdp->p.mp_ptrs, split, rsize);
4026 sepkey.mv_data = &rdp->p.mp_ptrs;
4027 memmove(ins+ksize, ins, (split_indx - *newindxp) * ksize);
4028 memcpy(ins, newkey->mv_data, ksize);
4029 mdp->p.mp_lower += sizeof(indx_t);
4030 mdp->p.mp_upper -= ksize - sizeof(indx_t);
4033 memcpy(&rdp->p.mp_ptrs, split, x * ksize);
4034 ins = LEAF2KEY(&rdp->p, x, ksize);
4035 memcpy(ins, newkey->mv_data, ksize);
4036 memcpy(ins+ksize, split + x * ksize, rsize - x * ksize);
4037 rdp->p.mp_lower += sizeof(indx_t);
4038 rdp->p.mp_upper -= ksize - sizeof(indx_t);
4045 /* For leaf pages, check the split point based on what
4046 * fits where, since otherwise add_node can fail.
4048 if (IS_LEAF(&mdp->p)) {
4049 unsigned int psize, nsize;
4050 /* Maximum free space in an empty page */
4051 pmax = txn->mt_env->me_psize - PAGEHDRSZ;
4052 nsize = mdb_leaf_size(txn->mt_env, newkey, newdata);
4053 if (newindx < split_indx) {
4055 for (i=0; i<split_indx; i++) {
4056 node = NODEPTR(&mdp->p, i);
4057 psize += NODESIZE + NODEKSZ(node) + sizeof(indx_t);
4058 if (F_ISSET(node->mn_flags, F_BIGDATA))
4059 psize += sizeof(pgno_t);
4061 psize += NODEDSZ(node);
4069 for (i=nkeys-1; i>=split_indx; i--) {
4070 node = NODEPTR(&mdp->p, i);
4071 psize += NODESIZE + NODEKSZ(node) + sizeof(indx_t);
4072 if (F_ISSET(node->mn_flags, F_BIGDATA))
4073 psize += sizeof(pgno_t);
4075 psize += NODEDSZ(node);
4084 /* First find the separating key between the split pages.
4086 if (newindx == split_indx) {
4087 sepkey.mv_size = newkey->mv_size;
4088 sepkey.mv_data = newkey->mv_data;
4090 node = NODEPTR(&mdp->p, split_indx);
4091 sepkey.mv_size = node->mn_ksize;
4092 sepkey.mv_data = NODEKEY(node);
4096 DPRINTF("separator is [%s]", DKEY(&sepkey));
4098 /* Copy separator key to the parent.
4100 if (SIZELEFT(rdp->h.md_parent) < mdb_branch_size(txn->mt_env, &sepkey)) {
4101 rc = mdb_split(txn, dbi, &rdp->h.md_parent, &rdp->h.md_pi,
4102 &sepkey, NULL, rdp->p.mp_pgno);
4104 /* Right page might now have changed parent.
4105 * Check if left page also changed parent.
4107 if (rdp->h.md_parent != mdp->h.md_parent &&
4108 mdp->h.md_pi >= NUMKEYS(mdp->h.md_parent)) {
4109 mdp->h.md_parent = rdp->h.md_parent;
4110 mdp->h.md_pi = rdp->h.md_pi - 1;
4113 rc = mdb_add_node(txn, dbi, rdp->h.md_parent, rdp->h.md_pi,
4114 &sepkey, NULL, rdp->p.mp_pgno, 0);
4116 if (IS_LEAF2(&rdp->p)) {
4119 if (rc != MDB_SUCCESS) {
4123 /* Move half of the keys to the right sibling. */
4124 if ((copy = malloc(txn->mt_env->me_psize)) == NULL)
4127 copy->mp_pgno = mdp->p.mp_pgno;
4128 copy->mp_flags = mdp->p.mp_flags;
4129 copy->mp_lower = PAGEHDRSZ;
4130 copy->mp_upper = txn->mt_env->me_psize;
4132 for (i = j = 0; i <= nkeys; j++) {
4133 if (i == split_indx) {
4134 /* Insert in right sibling. */
4135 /* Reset insert index for right sibling. */
4136 j = (i == newindx && ins_new);
4140 if (i == newindx && !ins_new) {
4141 /* Insert the original entry that caused the split. */
4142 rkey.mv_data = newkey->mv_data;
4143 rkey.mv_size = newkey->mv_size;
4144 if (IS_LEAF(&mdp->p)) {
4145 rdata.mv_data = newdata->mv_data;
4146 rdata.mv_size = newdata->mv_size;
4153 /* Update page and index for the new key. */
4155 if (cptr == &rdp->p)
4157 } else if (i == nkeys) {
4160 node = NODEPTR(&mdp->p, i);
4161 rkey.mv_data = NODEKEY(node);
4162 rkey.mv_size = node->mn_ksize;
4163 if (IS_LEAF(&mdp->p)) {
4164 rdata.mv_data = NODEDATA(node);
4165 rdata.mv_size = node->mn_dsize;
4167 pgno = NODEPGNO(node);
4168 flags = node->mn_flags;
4173 if (!IS_LEAF(&mdp->p) && j == 0) {
4174 /* First branch index doesn't need key data. */
4178 rc = mdb_add_node(txn, dbi, cptr, j, &rkey, &rdata, pgno, flags);
4180 nkeys = NUMKEYS(copy);
4181 for (i=0; i<nkeys; i++)
4182 mdp->p.mp_ptrs[i] = copy->mp_ptrs[i];
4183 mdp->p.mp_lower = copy->mp_lower;
4184 mdp->p.mp_upper = copy->mp_upper;
4185 memcpy(NODEPTR(&mdp->p, nkeys-1), NODEPTR(copy, nkeys-1),
4186 txn->mt_env->me_psize - copy->mp_upper);
4194 mdb_put0(MDB_txn *txn, MDB_dbi dbi,
4195 MDB_val *key, MDB_val *data, unsigned int flags)
4197 int rc = MDB_SUCCESS, exact;
4201 MDB_val xdata, *rdata, dkey;
4203 char dbuf[PAGESIZE];
4208 DPRINTF("==> put db %u key [%s], size %zu, data size %zu",
4209 dbi, DKEY(key), key->mv_size, data->mv_size);
4212 mpp.mp_parent = NULL;
4214 rc = mdb_search_page(txn, dbi, key, NULL, 1, &mpp);
4215 if (rc == MDB_SUCCESS) {
4216 leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, &ki);
4217 if (leaf && exact) {
4218 if (flags == MDB_NOOVERWRITE) {
4219 DPRINTF("duplicate key [%s]", DKEY(key));
4220 return MDB_KEYEXIST;
4222 /* there's only a key anyway, so this is a no-op */
4223 if (IS_LEAF2(mpp.mp_page))
4226 if (F_ISSET(txn->mt_dbs[dbi].md_flags, MDB_DUPSORT)) {
4227 /* Was a single item before, must convert now */
4228 if (!F_ISSET(leaf->mn_flags, F_DUPDATA)) {
4229 dkey.mv_size = NODEDSZ(leaf);
4230 dkey.mv_data = dbuf;
4231 memcpy(dbuf, NODEDATA(leaf), dkey.mv_size);
4232 /* data matches, ignore it */
4233 if (!mdb_dcmp(txn, dbi, data, &dkey))
4234 return (flags == MDB_NODUPDATA) ? MDB_KEYEXIST : MDB_SUCCESS;
4235 memset(&dummy, 0, sizeof(dummy));
4236 if (txn->mt_dbs[dbi].md_flags & MDB_DUPFIXED) {
4237 dummy.md_pad = data->mv_size;
4238 dummy.md_flags = MDB_DUPFIXED;
4239 if (txn->mt_dbs[dbi].md_flags & MDB_INTEGERDUP)
4240 dummy.md_flags |= MDB_INTEGERKEY;
4242 dummy.md_root = P_INVALID;
4243 if (dkey.mv_size == sizeof(MDB_db)) {
4244 memcpy(NODEDATA(leaf), &dummy, sizeof(dummy));
4247 mdb_del_node(mpp.mp_page, ki, 0);
4250 xdata.mv_size = sizeof(MDB_db);
4251 xdata.mv_data = &dummy;
4256 /* same size, just replace it */
4257 if (!F_ISSET(leaf->mn_flags, F_BIGDATA) &&
4258 NODEDSZ(leaf) == data->mv_size) {
4259 memcpy(NODEDATA(leaf), data->mv_data, data->mv_size);
4262 mdb_del_node(mpp.mp_page, ki, 0);
4264 if (leaf == NULL) { /* append if not found */
4265 ki = NUMKEYS(mpp.mp_page);
4266 DPRINTF("appending key at index %i", ki);
4268 } else if (rc == MDB_NOTFOUND) {
4270 /* new file, just write a root leaf page */
4271 DPUTS("allocating new root leaf page");
4272 if ((dp = mdb_new_page(txn, dbi, P_LEAF, 1)) == NULL) {
4275 mpp.mp_page = &dp->p;
4276 txn->mt_dbs[dbi].md_root = mpp.mp_page->mp_pgno;
4277 txn->mt_dbs[dbi].md_depth++;
4278 txn->mt_dbxs[dbi].md_dirty = 1;
4279 if ((txn->mt_dbs[dbi].md_flags & (MDB_DUPSORT|MDB_DUPFIXED)) == MDB_DUPFIXED)
4280 mpp.mp_page->mp_flags |= P_LEAF2;
4286 assert(IS_LEAF(mpp.mp_page));
4287 DPRINTF("there are %u keys, should insert new key at index %i",
4288 NUMKEYS(mpp.mp_page), ki);
4293 nsize = IS_LEAF2(mpp.mp_page) ? key->mv_size : mdb_leaf_size(txn->mt_env, key, rdata);
4294 if (SIZELEFT(mpp.mp_page) < nsize) {
4295 rc = mdb_split(txn, dbi, &mpp.mp_page, &ki, key, rdata, P_INVALID);
4297 /* There is room already in this leaf page. */
4298 rc = mdb_add_node(txn, dbi, mpp.mp_page, ki, key, rdata, 0, 0);
4301 if (rc != MDB_SUCCESS)
4302 txn->mt_flags |= MDB_TXN_ERROR;
4304 /* Remember if we just added a subdatabase */
4305 if (flags & F_SUBDATA) {
4306 leaf = NODEPTR(mpp.mp_page, ki);
4307 leaf->mn_flags |= F_SUBDATA;
4310 /* Now store the actual data in the child DB. Note that we're
4311 * storing the user data in the keys field, so there are strict
4312 * size limits on dupdata. The actual data fields of the child
4313 * DB are all zero size.
4319 leaf = NODEPTR(mpp.mp_page, ki);
4324 mc.mc_xcursor = &mx;
4325 mdb_xcursor_init0(&mc);
4326 mdb_xcursor_init1(txn, dbi, &mx, mpp.mp_page, leaf);
4329 if (flags == MDB_NODUPDATA)
4330 flags = MDB_NOOVERWRITE;
4331 /* converted, write the original data first */
4333 rc = mdb_put0(&mx.mx_txn, mx.mx_cursor.mc_dbi, &dkey, &xdata, flags);
4335 leaf->mn_flags |= F_DUPDATA;
4337 rc = mdb_put0(&mx.mx_txn, mx.mx_cursor.mc_dbi, data, &xdata, flags);
4338 mdb_xcursor_fini(&mc);
4339 memcpy(NODEDATA(leaf), &mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi],
4342 txn->mt_dbs[dbi].md_entries++;
4351 mdb_put(MDB_txn *txn, MDB_dbi dbi,
4352 MDB_val *key, MDB_val *data, unsigned int flags)
4358 assert(key != NULL);
4359 assert(data != NULL);
4361 if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
4364 if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
4368 if (key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
4372 if ((flags & (MDB_NOOVERWRITE|MDB_NODUPDATA)) != flags)
4379 if (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT) {
4380 mc.mc_xcursor = &mx;
4381 mdb_xcursor_init0(&mc);
4383 mc.mc_xcursor = NULL;
4385 return mdb_cursor_put(&mc, key, data, flags);
4389 mdb_env_set_flags(MDB_env *env, unsigned int flag, int onoff)
4391 #define CHANGEABLE (MDB_NOSYNC)
4392 if ((flag & CHANGEABLE) != flag)
4395 env->me_flags |= flag;
4397 env->me_flags &= ~flag;
4402 mdb_env_get_flags(MDB_env *env, unsigned int *arg)
4407 *arg = env->me_flags;
4412 mdb_env_get_path(MDB_env *env, const char **arg)
4417 *arg = env->me_path;
4422 mdb_stat0(MDB_env *env, MDB_db *db, MDB_stat *arg)
4424 arg->ms_psize = env->me_psize;
4425 arg->ms_depth = db->md_depth;
4426 arg->ms_branch_pages = db->md_branch_pages;
4427 arg->ms_leaf_pages = db->md_leaf_pages;
4428 arg->ms_overflow_pages = db->md_overflow_pages;
4429 arg->ms_entries = db->md_entries;
4434 mdb_env_stat(MDB_env *env, MDB_stat *arg)
4438 if (env == NULL || arg == NULL)
4441 mdb_env_read_meta(env, &toggle);
4443 return mdb_stat0(env, &env->me_metas[toggle]->mm_dbs[MAIN_DBI], arg);
4446 int mdb_open(MDB_txn *txn, const char *name, unsigned int flags, MDB_dbi *dbi)
4456 if (flags & (MDB_DUPSORT|MDB_REVERSEKEY|MDB_INTEGERKEY))
4457 txn->mt_dbs[MAIN_DBI].md_flags |= (flags & (MDB_DUPSORT|MDB_REVERSEKEY|MDB_INTEGERKEY));
4461 /* Is the DB already open? */
4463 for (i=2; i<txn->mt_numdbs; i++) {
4464 if (len == txn->mt_dbxs[i].md_name.mv_size &&
4465 !strncmp(name, txn->mt_dbxs[i].md_name.mv_data, len)) {
4471 if (txn->mt_numdbs >= txn->mt_env->me_maxdbs - 1)
4474 /* Find the DB info */
4476 key.mv_data = (void *)name;
4477 rc = mdb_get(txn, MAIN_DBI, &key, &data);
4479 /* Create if requested */
4480 if (rc == MDB_NOTFOUND && (flags & MDB_CREATE)) {
4483 data.mv_size = sizeof(MDB_db);
4484 data.mv_data = &dummy;
4485 memset(&dummy, 0, sizeof(dummy));
4486 dummy.md_root = P_INVALID;
4487 dummy.md_flags = flags & 0xffff;
4489 mc.mc_dbi = MAIN_DBI;
4491 rc = mdb_cursor_put(&mc, &key, &data, F_SUBDATA);
4495 /* OK, got info, add to table */
4496 if (rc == MDB_SUCCESS) {
4497 txn->mt_dbxs[txn->mt_numdbs].md_name.mv_data = strdup(name);
4498 txn->mt_dbxs[txn->mt_numdbs].md_name.mv_size = len;
4499 txn->mt_dbxs[txn->mt_numdbs].md_cmp = NULL;
4500 txn->mt_dbxs[txn->mt_numdbs].md_dcmp = NULL;
4501 txn->mt_dbxs[txn->mt_numdbs].md_rel = NULL;
4502 txn->mt_dbxs[txn->mt_numdbs].md_parent = MAIN_DBI;
4503 txn->mt_dbxs[txn->mt_numdbs].md_dirty = dirty;
4504 memcpy(&txn->mt_dbs[txn->mt_numdbs], data.mv_data, sizeof(MDB_db));
4505 *dbi = txn->mt_numdbs;
4506 txn->mt_env->me_dbs[0][txn->mt_numdbs] = txn->mt_dbs[txn->mt_numdbs];
4507 txn->mt_env->me_dbs[1][txn->mt_numdbs] = txn->mt_dbs[txn->mt_numdbs];
4514 int mdb_stat(MDB_txn *txn, MDB_dbi dbi, MDB_stat *arg)
4516 if (txn == NULL || arg == NULL || dbi >= txn->mt_numdbs)
4519 return mdb_stat0(txn->mt_env, &txn->mt_dbs[dbi], arg);
4522 void mdb_close(MDB_txn *txn, MDB_dbi dbi)
4525 if (dbi <= MAIN_DBI || dbi >= txn->mt_numdbs)
4527 ptr = txn->mt_dbxs[dbi].md_name.mv_data;
4528 txn->mt_dbxs[dbi].md_name.mv_data = NULL;
4529 txn->mt_dbxs[dbi].md_name.mv_size = 0;
4533 int mdb_set_compare(MDB_txn *txn, MDB_dbi dbi, MDB_cmp_func *cmp)
4535 if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
4538 txn->mt_dbxs[dbi].md_cmp = cmp;
4542 int mdb_set_dupsort(MDB_txn *txn, MDB_dbi dbi, MDB_cmp_func *cmp)
4544 if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
4547 txn->mt_dbxs[dbi].md_dcmp = cmp;
4551 int mdb_set_relfunc(MDB_txn *txn, MDB_dbi dbi, MDB_rel_func *rel)
4553 if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
4556 txn->mt_dbxs[dbi].md_rel = rel;