28 #define ULONG unsigned long
37 #if DEBUG && defined(__GNUC__)
38 # define DPRINTF(fmt, ...) \
39 fprintf(stderr, "%s:%d: " fmt "\n", __func__, __LINE__, ##__VA_ARGS__)
41 # define DPRINTF(...) ((void) 0)
46 #define MDB_MAGIC 0xBEEFC0DE
48 #define MAXKEYSIZE 255
50 #define P_INVALID (~0UL)
52 #define F_ISSET(w, f) (((w) & (f)) == (f))
54 typedef uint16_t indx_t;
56 #define DEFAULT_READERS 126
57 #define DEFAULT_MAPSIZE 1048576
59 /* Lock descriptor stuff */
64 typedef struct MDB_rxbody {
69 #define CACHELINE 64 /* most CPUs. Itanium uses 128 */
72 typedef struct MDB_reader {
74 /* cache line alignment */
75 char pad[CACHELINE-sizeof(MDB_rxbody)];
80 uint32_t mt_version; \
81 pthread_mutex_t mt_mutex; \
83 uint32_t mt_numreaders
84 typedef struct MDB_txbody {
88 typedef struct MDB_txninfo {
90 char pad[CACHELINE-sizeof(MDB_txbody)];
91 pthread_mutex_t mt_wmutex;
92 char pad2[CACHELINE-sizeof(pthread_mutex_t)];
93 MDB_reader mt_readers[1];
96 /* Common header for all page types. Overflow pages
97 * occupy a number of contiguous pages with no
98 * headers on any page after the first.
100 typedef struct MDB_page { /* represents a page of storage */
101 #define mp_pgno mp_p.p_pgno
103 pgno_t p_pgno; /* page number */
106 #define P_BRANCH 0x01 /* branch page */
107 #define P_LEAF 0x02 /* leaf page */
108 #define P_OVERFLOW 0x04 /* overflow page */
109 #define P_META 0x08 /* meta page */
110 #define P_DIRTY 0x10 /* dirty page */
112 #define mp_lower mp_pb.pb.pb_lower
113 #define mp_upper mp_pb.pb.pb_upper
114 #define mp_pages mp_pb.pb_pages
117 indx_t pb_lower; /* lower bound of free space */
118 indx_t pb_upper; /* upper bound of free space */
120 uint32_t pb_pages; /* number of overflow pages */
122 indx_t mp_ptrs[1]; /* dynamic size */
125 #define PAGEHDRSZ ((unsigned) offsetof(MDB_page, mp_ptrs))
127 #define NUMKEYS(p) (((p)->mp_lower - PAGEHDRSZ) >> 1)
128 #define SIZELEFT(p) (indx_t)((p)->mp_upper - (p)->mp_lower)
129 #define PAGEFILL(env, p) (1000L * ((env)->me_psize - PAGEHDRSZ - SIZELEFT(p)) / \
130 ((env)->me_psize - PAGEHDRSZ))
131 #define IS_LEAF(p) F_ISSET((p)->mp_flags, P_LEAF)
132 #define IS_BRANCH(p) F_ISSET((p)->mp_flags, P_BRANCH)
133 #define IS_OVERFLOW(p) F_ISSET((p)->mp_flags, P_OVERFLOW)
135 #define OVPAGES(size, psize) (PAGEHDRSZ + size + psize - 1) / psize;
137 typedef struct MDB_db {
141 ULONG md_branch_pages;
143 ULONG md_overflow_pages;
151 typedef struct MDB_meta { /* meta (footer) page content */
154 void *mm_address; /* address for fixed mapping */
155 size_t mm_mapsize; /* size of mmap region */
156 MDB_db mm_dbs[2]; /* first is free space, 2nd is main db */
157 #define mm_psize mm_dbs[0].md_pad
158 #define mm_flags mm_dbs[0].md_flags
159 pgno_t mm_last_pg; /* last used page in file */
160 ULONG mm_txnid; /* txnid that committed this page */
163 typedef struct MDB_dhead { /* a dirty page */
164 STAILQ_ENTRY(MDB_dpage) md_next; /* queue of dirty pages */
166 unsigned md_pi; /* parent index */
170 typedef struct MDB_dpage {
175 STAILQ_HEAD(dirty_queue, MDB_dpage);
177 typedef struct MDB_oldpages {
178 struct MDB_oldpages *mo_next;
180 pgno_t mo_pages[1]; /* dynamic */
183 typedef struct MDB_pageparent {
189 static MDB_dpage *mdb_alloc_page(MDB_txn *txn, MDB_page *parent, unsigned int parent_idx, int num);
190 static int mdb_touch(MDB_txn *txn, MDB_pageparent *mp);
192 typedef struct MDB_ppage { /* ordered list of pages */
193 SLIST_ENTRY(MDB_ppage) mp_entry;
195 unsigned int mp_ki; /* cursor index on page */
197 SLIST_HEAD(page_stack, MDB_ppage);
199 #define CURSOR_EMPTY(c) SLIST_EMPTY(&(c)->mc_stack)
200 #define CURSOR_TOP(c) SLIST_FIRST(&(c)->mc_stack)
201 #define CURSOR_POP(c) SLIST_REMOVE_HEAD(&(c)->mc_stack, mp_entry)
202 #define CURSOR_PUSH(c,p) SLIST_INSERT_HEAD(&(c)->mc_stack, p, mp_entry)
206 struct page_stack mc_stack; /* stack of parent pages */
208 short mc_initialized; /* 1 if initialized */
209 short mc_eof; /* 1 if end is reached */
212 #define METAHASHLEN offsetof(MDB_meta, mm_hash)
213 #define METADATA(p) ((void *)((char *)p + PAGEHDRSZ))
215 typedef struct MDB_node {
216 #define mn_pgno mn_p.np_pgno
217 #define mn_dsize mn_p.np_dsize
219 pgno_t np_pgno; /* child page number */
220 uint32_t np_dsize; /* leaf data size */
222 unsigned int mn_flags:4;
223 unsigned int mn_ksize:12; /* key size */
224 #define F_BIGDATA 0x01 /* data put on overflow page */
229 typedef struct MDB_dbx {
231 MDB_cmp_func *md_cmp; /* user compare function */
232 MDB_cmp_func *md_dcmp; /* user dupsort function */
233 MDB_rel_func *md_rel; /* user relocate function */
237 pgno_t mt_next_pgno; /* next unallocated page */
241 pgno_t *mt_free_pgs; /* this is an IDL */
243 struct dirty_queue *dirty_queue; /* modified pages */
246 MDB_dbx *mt_dbxs; /* array */
248 unsigned int mt_numdbs;
250 #define MDB_TXN_RDONLY 0x01 /* read-only transaction */
251 #define MDB_TXN_ERROR 0x02 /* an error has occurred */
252 #define MDB_TXN_METOGGLE 0x04 /* used meta page 1 */
253 unsigned int mt_flags;
260 unsigned int me_maxreaders;
261 unsigned int me_numdbs;
262 unsigned int me_maxdbs;
265 MDB_txninfo *me_txns;
266 MDB_meta *me_metas[2];
268 MDB_txn *me_txn; /* current write transaction */
270 off_t me_size; /* current file size */
271 unsigned int me_psize;
273 MDB_dbx *me_dbxs; /* array */
275 MDB_oldpages *me_pghead;
276 pthread_key_t me_txkey; /* thread-key for readers */
279 #define NODESIZE offsetof(MDB_node, mn_data)
281 #define INDXSIZE(k) (NODESIZE + ((k) == NULL ? 0 : (k)->mv_size))
282 #define LEAFSIZE(k, d) (NODESIZE + (k)->mv_size + (d)->mv_size)
283 #define NODEPTR(p, i) ((MDB_node *)((char *)(p) + (p)->mp_ptrs[i]))
284 #define NODEKEY(node) (void *)((node)->mn_data)
285 #define NODEDATA(node) (void *)((char *)(node)->mn_data + (node)->mn_ksize)
286 #define NODEPGNO(node) ((node)->mn_pgno)
287 #define NODEDSZ(node) ((node)->mn_dsize)
289 #define MDB_COMMIT_PAGES 64 /* max number of pages to write in one commit */
290 #define MDB_MAXCACHE_DEF 1024 /* max number of pages to keep in cache */
292 static int mdb_search_page_root(MDB_txn *txn,
293 MDB_dbi dbi, MDB_val *key,
294 MDB_cursor *cursor, int modify,
295 MDB_pageparent *mpp);
296 static int mdb_search_page(MDB_txn *txn,
297 MDB_dbi dbi, MDB_val *key,
298 MDB_cursor *cursor, int modify,
299 MDB_pageparent *mpp);
301 static int mdbenv_read_header(MDB_env *env, MDB_meta *meta);
302 static int mdbenv_read_meta(MDB_env *env, int *which);
303 static int mdbenv_write_meta(MDB_txn *txn);
304 static MDB_page *mdbenv_get_page(MDB_env *env, pgno_t pgno);
306 static MDB_node *mdb_search_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp,
307 MDB_val *key, int *exactp, unsigned int *kip);
308 static int mdb_add_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp,
309 indx_t indx, MDB_val *key, MDB_val *data,
310 pgno_t pgno, uint8_t flags);
311 static void mdb_del_node(MDB_page *mp, indx_t indx);
312 static int mdb_del0(MDB_txn *txn, MDB_dbi dbi, unsigned int ki,
313 MDB_pageparent *mpp, MDB_node *leaf);
314 static int mdb_read_data(MDB_env *env, MDB_node *leaf, MDB_val *data);
316 static int mdb_rebalance(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *mp);
317 static int mdb_update_key(MDB_page *mp, indx_t indx, MDB_val *key);
318 static int mdb_move_node(MDB_txn *txn, MDB_dbi dbi,
319 MDB_pageparent *src, indx_t srcindx,
320 MDB_pageparent *dst, indx_t dstindx);
321 static int mdb_merge(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *src,
322 MDB_pageparent *dst);
323 static int mdb_split(MDB_txn *txn, MDB_dbi dbi, MDB_page **mpp,
324 unsigned int *newindxp, MDB_val *newkey,
325 MDB_val *newdata, pgno_t newpgno);
326 static MDB_dpage *mdb_new_page(MDB_txn *txn, MDB_dbi dbi, uint32_t flags, int num);
328 static void cursor_pop_page(MDB_cursor *cursor);
329 static MDB_ppage *cursor_push_page(MDB_cursor *cursor,
332 static int mdb_set_key(MDB_node *node, MDB_val *key);
333 static int mdb_sibling(MDB_cursor *cursor, int move_right);
334 static int mdb_cursor_next(MDB_cursor *cursor,
335 MDB_val *key, MDB_val *data);
336 static int mdb_cursor_set(MDB_cursor *cursor,
337 MDB_val *key, MDB_val *data, int *exactp);
338 static int mdb_cursor_first(MDB_cursor *cursor,
339 MDB_val *key, MDB_val *data);
340 static int mdb_cursor_last(MDB_cursor *cursor,
341 MDB_val *key, MDB_val *data);
343 static size_t mdb_leaf_size(MDB_env *env, MDB_val *key,
345 static size_t mdb_branch_size(MDB_env *env, MDB_val *key);
347 static int memncmp(const void *s1, size_t n1,
348 const void *s2, size_t n2);
349 static int memnrcmp(const void *s1, size_t n1,
350 const void *s2, size_t n2);
353 memncmp(const void *s1, size_t n1, const void *s2, size_t n2)
355 int diff, len_diff = -1;
358 len_diff = (n1 > n2);
361 diff = memcmp(s1, s2, n1);
362 return diff ? diff : len_diff;
366 memnrcmp(const void *s1, size_t n1, const void *s2, size_t n2)
368 const unsigned char *p1, *p2, *p1_lim;
375 p1 = (const unsigned char *)s1 + n1 - 1;
376 p2 = (const unsigned char *)s2 + n2 - 1;
378 for (p1_lim = (n1 <= n2 ? s1 : s2); *p1 == *p2; p1--, p2--) {
380 return (p1 != s1) ? (p1 != p2) : (p2 != s2) ? -1 : 0;
386 mdb_cmp(MDB_txn *txn, MDB_dbi dbi, const MDB_val *a, const MDB_val *b)
388 return txn->mt_dbxs[dbi].md_cmp(a, b);
392 _mdb_cmp(MDB_txn *txn, MDB_dbi dbi, const MDB_val *key1, const MDB_val *key2)
394 if (F_ISSET(txn->mt_dbs[dbi].md_flags, MDB_REVERSEKEY))
395 return memnrcmp(key1->mv_data, key1->mv_size, key2->mv_data, key2->mv_size);
397 return memncmp((char *)key1->mv_data, key1->mv_size, key2->mv_data, key2->mv_size);
400 /* Allocate new page(s) for writing */
402 mdb_alloc_page(MDB_txn *txn, MDB_page *parent, unsigned int parent_idx, int num)
405 pgno_t pgno = P_INVALID;
406 ULONG oldest = txn->mt_txnid - 2;
408 if (!txn->mt_env->me_pghead && txn->mt_dbs[FREE_DBI].md_root != P_INVALID) {
409 /* See if there's anything in the free DB */
414 mpp.mp_parent = NULL;
416 mdb_search_page(txn, FREE_DBI, NULL, NULL, 0, &mpp);
417 leaf = NODEPTR(mpp.mp_page, 0);
418 kptr = (ULONG *)NODEKEY(leaf);
420 /* It's potentially usable, unless there are still
421 * older readers outstanding. Grab it.
423 if (oldest > *kptr) {
428 mdb_read_data(txn->mt_env, leaf, &data);
429 idl = (ULONG *)data.mv_data;
430 mop = malloc(sizeof(MDB_oldpages) + MDB_IDL_SIZEOF(idl) - sizeof(pgno_t));
431 mop->mo_next = txn->mt_env->me_pghead;
432 mop->mo_txnid = *kptr;
433 txn->mt_env->me_pghead = mop;
434 memcpy(mop->mo_pages, idl, MDB_IDL_SIZEOF(idl));
439 DPRINTF("IDL read txn %lu root %lu num %lu",
440 mop->mo_txnid, txn->mt_dbs[FREE_DBI].md_root, idl[0]);
441 for (i=0; i<idl[0]; i++) {
442 DPRINTF("IDL %lu", idl[i+1]);
446 /* drop this IDL from the DB */
447 mpp.mp_parent = NULL;
449 mdb_search_page(txn, FREE_DBI, NULL, NULL, 1, &mpp);
450 leaf = NODEPTR(mpp.mp_page, 0);
451 mdb_del0(txn, FREE_DBI, 0, &mpp, leaf);
454 if (txn->mt_env->me_pghead) {
456 for (i=0; i<txn->mt_env->me_txns->mt_numreaders; i++) {
457 ULONG mr = txn->mt_env->me_txns->mt_readers[i].mr_txnid;
460 oldest = txn->mt_env->me_txns->mt_readers[i].mr_txnid;
462 if (oldest > txn->mt_env->me_pghead->mo_txnid) {
463 MDB_oldpages *mop = txn->mt_env->me_pghead;
464 txn->mt_oldest = oldest;
466 /* FIXME: For now, always use fresh pages. We
467 * really ought to search the free list for a
472 /* peel pages off tail, so we only have to truncate the list */
473 pgno = MDB_IDL_LAST(mop->mo_pages);
474 if (MDB_IDL_IS_RANGE(mop->mo_pages)) {
476 if (mop->mo_pages[2] > mop->mo_pages[1])
477 mop->mo_pages[0] = 0;
481 if (MDB_IDL_IS_ZERO(mop->mo_pages)) {
482 txn->mt_env->me_pghead = mop->mo_next;
489 if ((dp = malloc(txn->mt_env->me_psize * num + sizeof(MDB_dhead))) == NULL)
492 dp->h.md_parent = parent;
493 dp->h.md_pi = parent_idx;
494 STAILQ_INSERT_TAIL(txn->mt_u.dirty_queue, dp, h.md_next);
495 if (pgno == P_INVALID) {
496 dp->p.mp_pgno = txn->mt_next_pgno;
497 txn->mt_next_pgno += num;
499 dp->p.mp_pgno = pgno;
505 /* Touch a page: make it dirty and re-insert into tree with updated pgno.
508 mdb_touch(MDB_txn *txn, MDB_pageparent *pp)
510 MDB_page *mp = pp->mp_page;
515 if (!F_ISSET(mp->mp_flags, P_DIRTY)) {
517 if ((dp = mdb_alloc_page(txn, pp->mp_parent, pp->mp_pi, 1)) == NULL)
519 DPRINTF("touched page %lu -> %lu", mp->mp_pgno, dp->p.mp_pgno);
520 mdb_idl_insert(txn->mt_free_pgs, mp->mp_pgno);
521 pgno = dp->p.mp_pgno;
522 memcpy(&dp->p, mp, txn->mt_env->me_psize);
525 mp->mp_flags |= P_DIRTY;
527 /* Update the page number to new touched page. */
528 if (pp->mp_parent != NULL)
529 NODEPGNO(NODEPTR(pp->mp_parent, pp->mp_pi)) = mp->mp_pgno;
536 mdbenv_sync(MDB_env *env)
539 if (!F_ISSET(env->me_flags, MDB_NOSYNC)) {
540 if (fsync(env->me_fd))
547 mdb_txn_begin(MDB_env *env, int rdonly, MDB_txn **ret)
552 if ((txn = calloc(1, sizeof(*txn))) == NULL) {
553 DPRINTF("calloc: %s", strerror(errno));
558 txn->mt_flags |= MDB_TXN_RDONLY;
560 txn->mt_u.dirty_queue = calloc(1, sizeof(*txn->mt_u.dirty_queue));
561 if (txn->mt_u.dirty_queue == NULL) {
565 STAILQ_INIT(txn->mt_u.dirty_queue);
567 pthread_mutex_lock(&env->me_txns->mt_wmutex);
568 env->me_txns->mt_txnid++;
569 txn->mt_free_pgs = malloc(MDB_IDL_UM_SIZEOF);
570 if (txn->mt_free_pgs == NULL) {
571 free(txn->mt_u.dirty_queue);
575 txn->mt_free_pgs[0] = 0;
578 txn->mt_txnid = env->me_txns->mt_txnid;
580 MDB_reader *r = pthread_getspecific(env->me_txkey);
583 pthread_mutex_lock(&env->me_txns->mt_mutex);
584 for (i=0; i<env->me_maxreaders; i++) {
585 if (env->me_txns->mt_readers[i].mr_pid == 0) {
586 env->me_txns->mt_readers[i].mr_pid = getpid();
587 env->me_txns->mt_readers[i].mr_tid = pthread_self();
588 r = &env->me_txns->mt_readers[i];
589 pthread_setspecific(env->me_txkey, r);
590 if (i >= env->me_txns->mt_numreaders)
591 env->me_txns->mt_numreaders = i+1;
595 pthread_mutex_unlock(&env->me_txns->mt_mutex);
596 if (i == env->me_maxreaders) {
600 r->mr_txnid = txn->mt_txnid;
601 txn->mt_u.reader = r;
608 if ((rc = mdbenv_read_meta(env, &toggle)) != MDB_SUCCESS) {
613 /* Copy the DB arrays */
614 txn->mt_numdbs = env->me_numdbs;
615 txn->mt_dbxs = env->me_dbxs; /* mostly static anyway */
616 txn->mt_dbs = malloc(env->me_maxdbs * sizeof(MDB_db));
617 memcpy(txn->mt_dbs, env->me_meta->mm_dbs, 2 * sizeof(MDB_db));
618 if (txn->mt_numdbs > 2)
619 memcpy(txn->mt_dbs+2, env->me_dbs[env->me_db_toggle]+2,
620 (txn->mt_numdbs - 2) * sizeof(MDB_db));
624 txn->mt_flags |= MDB_TXN_METOGGLE;
625 txn->mt_next_pgno = env->me_meta->mm_last_pg+1;
628 DPRINTF("begin transaction %lu on mdbenv %p, root page %lu",
629 txn->mt_txnid, (void *) env, txn->mt_dbs[MAIN_DBI].md_root);
636 mdb_txn_abort(MDB_txn *txn)
645 DPRINTF("abort transaction %lu on mdbenv %p, root page %lu",
646 txn->mt_txnid, (void *) env, txn->mt_dbs[MAIN_DBI].md_root);
650 if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
651 txn->mt_u.reader->mr_txnid = 0;
654 /* Discard all dirty pages. */
655 while (!STAILQ_EMPTY(txn->mt_u.dirty_queue)) {
656 dp = STAILQ_FIRST(txn->mt_u.dirty_queue);
657 STAILQ_REMOVE_HEAD(txn->mt_u.dirty_queue, h.md_next);
660 free(txn->mt_free_pgs);
661 free(txn->mt_u.dirty_queue);
663 while ((mop = txn->mt_env->me_pghead)) {
664 txn->mt_env->me_pghead = mop->mo_next;
669 env->me_txns->mt_txnid--;
670 pthread_mutex_unlock(&env->me_txns->mt_wmutex);
677 mdb_txn_commit(MDB_txn *txn)
686 struct iovec iov[MDB_COMMIT_PAGES];
689 assert(txn->mt_env != NULL);
693 if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
694 DPRINTF("attempt to commit read-only transaction");
699 if (txn != env->me_txn) {
700 DPRINTF("attempt to commit unknown transaction");
705 if (F_ISSET(txn->mt_flags, MDB_TXN_ERROR)) {
706 DPRINTF("error flag is set, can't commit");
711 if (STAILQ_EMPTY(txn->mt_u.dirty_queue))
714 DPRINTF("committing transaction %lu on mdbenv %p, root page %lu",
715 txn->mt_txnid, (void *) env, txn->mt_dbs[MAIN_DBI].md_root);
717 /* should only be one record now */
718 if (env->me_pghead) {
722 mop = env->me_pghead;
723 key.mv_size = sizeof(pgno_t);
724 key.mv_data = (char *)&mop->mo_txnid;
725 data.mv_size = MDB_IDL_SIZEOF(mop->mo_pages);
726 data.mv_data = mop->mo_pages;
727 mdb_put(txn, FREE_DBI, &key, &data, 0);
728 free(env->me_pghead);
729 env->me_pghead = NULL;
731 /* save to free list */
732 if (!MDB_IDL_IS_ZERO(txn->mt_free_pgs)) {
736 /* make sure last page of freeDB is touched and on freelist */
737 key.mv_size = MAXKEYSIZE+1;
739 mpp.mp_parent = NULL;
741 mdb_search_page(txn, FREE_DBI, &key, NULL, 1, &mpp);
746 ULONG *idl = txn->mt_free_pgs;
747 DPRINTF("IDL write txn %lu root %lu num %lu",
748 txn->mt_txnid, txn->mt_dbs[FREE_DBI].md_root, idl[0]);
749 for (i=0; i<idl[0]; i++) {
750 DPRINTF("IDL %lu", idl[i+1]);
754 /* write to last page of freeDB */
755 key.mv_size = sizeof(pgno_t);
756 key.mv_data = (char *)&txn->mt_txnid;
757 data.mv_size = MDB_IDL_SIZEOF(txn->mt_free_pgs);
758 data.mv_data = txn->mt_free_pgs;
759 mdb_put(txn, FREE_DBI, &key, &data, 0);
762 /* Update DB root pointers. Their pages have already been
763 * touched so this is all in-place and cannot fail.
767 data.mv_size = sizeof(MDB_db);
769 for (i = 2; i < env->me_numdbs; i++) {
770 if (env->me_dbs[env->me_db_toggle][i].md_root != txn->mt_dbs[i].md_root) {
771 data.mv_data = &txn->mt_dbs[i];
772 mdb_put(txn, i, &txn->mt_dbxs[i].md_name, &data, 0);
775 for (i = env->me_numdbs; i < txn->mt_numdbs; i++) {
776 data.mv_data = &txn->mt_dbs[i];
777 mdb_put(txn, i, &txn->mt_dbxs[i].md_name, &data, 0);
781 /* Commit up to MDB_COMMIT_PAGES dirty pages to disk until done.
788 STAILQ_FOREACH(dp, txn->mt_u.dirty_queue, h.md_next) {
789 if (dp->p.mp_pgno != next) {
791 DPRINTF("committing %u dirty pages", n);
792 rc = writev(env->me_fd, iov, n);
796 DPRINTF("short write, filesystem full?");
798 DPRINTF("writev: %s", strerror(errno));
805 lseek(env->me_fd, dp->p.mp_pgno * env->me_psize, SEEK_SET);
806 next = dp->p.mp_pgno;
808 DPRINTF("committing page %lu", dp->p.mp_pgno);
809 iov[n].iov_len = env->me_psize * dp->h.md_num;
810 iov[n].iov_base = &dp->p;
811 size += iov[n].iov_len;
812 next = dp->p.mp_pgno + dp->h.md_num;
813 /* clear dirty flag */
814 dp->p.mp_flags &= ~P_DIRTY;
815 if (++n >= MDB_COMMIT_PAGES) {
824 DPRINTF("committing %u dirty pages", n);
825 rc = writev(env->me_fd, iov, n);
829 DPRINTF("short write, filesystem full?");
831 DPRINTF("writev: %s", strerror(errno));
838 /* Drop the dirty pages.
840 while (!STAILQ_EMPTY(txn->mt_u.dirty_queue)) {
841 dp = STAILQ_FIRST(txn->mt_u.dirty_queue);
842 STAILQ_REMOVE_HEAD(txn->mt_u.dirty_queue, h.md_next);
846 if ((n = mdbenv_sync(env)) != 0 ||
847 (n = mdbenv_write_meta(txn)) != MDB_SUCCESS ||
848 (n = mdbenv_sync(env)) != 0) {
854 /* update the DB tables */
856 int toggle = !env->me_db_toggle;
858 for (i = 0; i < env->me_numdbs; i++) {
859 if (env->me_dbs[toggle][i].md_root != txn->mt_dbs[i].md_root)
860 env->me_dbs[toggle][i] = txn->mt_dbs[i];
862 for (i = env->me_numdbs; i < txn->mt_numdbs; i++) {
863 env->me_dbxs[i] = txn->mt_dbxs[i];
864 env->me_dbs[toggle][i] = txn->mt_dbs[i];
866 env->me_db_toggle = toggle;
867 env->me_numdbs = txn->mt_numdbs;
872 pthread_mutex_unlock(&env->me_txns->mt_wmutex);
873 free(txn->mt_free_pgs);
874 free(txn->mt_u.dirty_queue);
885 mdbenv_read_header(MDB_env *env, MDB_meta *meta)
894 /* We don't know the page size yet, so use a minimum value.
897 if ((rc = pread(env->me_fd, page, PAGESIZE, 0)) == 0) {
899 } else if (rc != PAGESIZE) {
902 DPRINTF("read: %s", strerror(errno));
906 p = (MDB_page *)page;
908 if (!F_ISSET(p->mp_flags, P_META)) {
909 DPRINTF("page %lu not a meta page", p->mp_pgno);
914 if (m->mm_magic != MDB_MAGIC) {
915 DPRINTF("meta has invalid magic");
919 if (m->mm_version != MDB_VERSION) {
920 DPRINTF("database is version %u, expected version %u",
921 m->mm_version, MDB_VERSION);
925 memcpy(meta, m, sizeof(*m));
930 mdbenv_init_meta(MDB_env *env, MDB_meta *meta)
937 DPRINTF("writing new meta page");
938 psize = sysconf(_SC_PAGE_SIZE);
940 meta->mm_magic = MDB_MAGIC;
941 meta->mm_version = MDB_VERSION;
942 meta->mm_psize = psize;
943 meta->mm_last_pg = 1;
944 meta->mm_flags = env->me_flags & 0xffff;
945 #if __BYTE_ORDER == __LITTLE_ENDIAN
946 /* freeDB keys are pgno_t's, must compare in int order */
947 meta->mm_flags |= MDB_REVERSEKEY;
949 meta->mm_dbs[0].md_root = P_INVALID;
950 meta->mm_dbs[1].md_root = P_INVALID;
952 p = calloc(2, psize);
954 p->mp_flags = P_META;
957 memcpy(m, meta, sizeof(*meta));
959 q = (MDB_page *)((char *)p + psize);
962 q->mp_flags = P_META;
965 memcpy(m, meta, sizeof(*meta));
967 rc = write(env->me_fd, p, psize * 2);
969 return (rc == (int)psize * 2) ? MDB_SUCCESS : errno;
973 mdbenv_write_meta(MDB_txn *txn)
982 assert(txn->mt_env != NULL);
984 DPRINTF("writing meta page for root page %lu", txn->mt_dbs[MAIN_DBI].md_root);
989 off = offsetof(MDB_meta, mm_dbs[0].md_depth);
990 len = sizeof(MDB_meta) - off;
993 meta.mm_dbs[0] = txn->mt_dbs[0];
994 meta.mm_dbs[1] = txn->mt_dbs[1];
995 meta.mm_last_pg = txn->mt_next_pgno - 1;
996 meta.mm_txnid = txn->mt_txnid;
998 if (!F_ISSET(txn->mt_flags, MDB_TXN_METOGGLE))
999 off += env->me_psize;
1002 lseek(env->me_fd, off, SEEK_SET);
1003 rc = write(env->me_fd, ptr, len);
1005 DPRINTF("write failed, disk error?");
1013 mdbenv_read_meta(MDB_env *env, int *which)
1017 assert(env != NULL);
1019 if (env->me_metas[0]->mm_txnid < env->me_metas[1]->mm_txnid)
1022 if (env->me_meta != env->me_metas[toggle])
1023 env->me_meta = env->me_metas[toggle];
1027 DPRINTF("Using meta page %d", toggle);
1033 mdbenv_create(MDB_env **env)
1037 e = calloc(1, sizeof(*e));
1038 if (!e) return ENOMEM;
1040 e->me_maxreaders = DEFAULT_READERS;
1049 mdbenv_set_mapsize(MDB_env *env, size_t size)
1053 env->me_mapsize = size;
1058 mdbenv_set_maxdbs(MDB_env *env, int dbs)
1062 env->me_maxdbs = dbs;
1067 mdbenv_set_maxreaders(MDB_env *env, int readers)
1069 env->me_maxreaders = readers;
1074 mdbenv_get_maxreaders(MDB_env *env, int *readers)
1076 if (!env || !readers)
1078 *readers = env->me_maxreaders;
1083 mdbenv_open2(MDB_env *env, unsigned int flags)
1089 env->me_flags = flags;
1091 memset(&meta, 0, sizeof(meta));
1093 if ((i = mdbenv_read_header(env, &meta)) != 0) {
1096 DPRINTF("new mdbenv");
1100 if (!env->me_mapsize) {
1101 env->me_mapsize = newenv ? DEFAULT_MAPSIZE : meta.mm_mapsize;
1105 if (meta.mm_address && (flags & MDB_FIXEDMAP))
1107 env->me_map = mmap(meta.mm_address, env->me_mapsize, PROT_READ, i,
1109 if (env->me_map == MAP_FAILED)
1113 meta.mm_mapsize = env->me_mapsize;
1114 if (flags & MDB_FIXEDMAP)
1115 meta.mm_address = env->me_map;
1116 i = mdbenv_init_meta(env, &meta);
1117 if (i != MDB_SUCCESS) {
1118 munmap(env->me_map, env->me_mapsize);
1122 env->me_psize = meta.mm_psize;
1124 p = (MDB_page *)(MDB_page *)(MDB_page *)(MDB_page *)(MDB_page *)(MDB_page *)(MDB_page *)(MDB_page *)(MDB_page *)env->me_map;
1125 env->me_metas[0] = METADATA(p);
1126 env->me_metas[1] = (MDB_meta *)((char *)env->me_metas[0] + meta.mm_psize);
1128 if ((i = mdbenv_read_meta(env, NULL)) != 0)
1131 DPRINTF("opened database version %u, pagesize %u",
1132 env->me_meta->mm_version, env->me_psize);
1133 DPRINTF("depth: %u", env->me_meta->mm_dbs[MAIN_DBI].md_depth);
1134 DPRINTF("entries: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_entries);
1135 DPRINTF("branch pages: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_branch_pages);
1136 DPRINTF("leaf pages: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_leaf_pages);
1137 DPRINTF("overflow pages: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_overflow_pages);
1138 DPRINTF("root: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_root);
1144 mdbenv_reader_dest(void *ptr)
1146 MDB_reader *reader = ptr;
1148 reader->mr_txnid = 0;
1153 /* downgrade the exclusive lock on the region back to shared */
1155 mdbenv_share_locks(MDB_env *env)
1157 struct flock lock_info;
1159 env->me_txns->mt_txnid = env->me_meta->mm_txnid;
1161 memset((void *)&lock_info, 0, sizeof(lock_info));
1162 lock_info.l_type = F_RDLCK;
1163 lock_info.l_whence = SEEK_SET;
1164 lock_info.l_start = 0;
1165 lock_info.l_len = 1;
1166 fcntl(env->me_lfd, F_SETLK, &lock_info);
1170 mdbenv_setup_locks(MDB_env *env, char *lpath, int mode, int *excl)
1174 struct flock lock_info;
1178 if ((env->me_lfd = open(lpath, O_RDWR|O_CREAT, mode)) == -1) {
1182 /* Try to get exclusive lock. If we succeed, then
1183 * nobody is using the lock region and we should initialize it.
1185 memset((void *)&lock_info, 0, sizeof(lock_info));
1186 lock_info.l_type = F_WRLCK;
1187 lock_info.l_whence = SEEK_SET;
1188 lock_info.l_start = 0;
1189 lock_info.l_len = 1;
1190 rc = fcntl(env->me_lfd, F_SETLK, &lock_info);
1194 lock_info.l_type = F_RDLCK;
1195 rc = fcntl(env->me_lfd, F_SETLK, &lock_info);
1201 size = lseek(env->me_lfd, 0, SEEK_END);
1202 rsize = (env->me_maxreaders-1) * sizeof(MDB_reader) + sizeof(MDB_txninfo);
1203 if (size < rsize && *excl) {
1204 if (ftruncate(env->me_lfd, rsize) != 0) {
1210 size = rsize - sizeof(MDB_txninfo);
1211 env->me_maxreaders = size/sizeof(MDB_reader) + 1;
1213 env->me_txns = mmap(0, rsize, PROT_READ|PROT_WRITE, MAP_SHARED,
1215 if (env->me_txns == MAP_FAILED) {
1220 pthread_mutexattr_t mattr;
1222 pthread_mutexattr_init(&mattr);
1223 pthread_mutexattr_setpshared(&mattr, PTHREAD_PROCESS_SHARED);
1224 pthread_mutex_init(&env->me_txns->mt_mutex, &mattr);
1225 pthread_mutex_init(&env->me_txns->mt_wmutex, &mattr);
1226 env->me_txns->mt_version = MDB_VERSION;
1227 env->me_txns->mt_magic = MDB_MAGIC;
1228 env->me_txns->mt_txnid = 0;
1229 env->me_txns->mt_numreaders = 0;
1232 if (env->me_txns->mt_magic != MDB_MAGIC) {
1233 DPRINTF("lock region has invalid magic");
1236 if (env->me_txns->mt_version != MDB_VERSION) {
1237 DPRINTF("lock region is version %u, expected version %u",
1238 env->me_txns->mt_version, MDB_VERSION);
1241 if (errno != EACCES && errno != EAGAIN) {
1255 mdbenv_open(MDB_env *env, const char *path, unsigned int flags, mode_t mode)
1257 int oflags, rc, len, excl;
1258 char *lpath, *dpath;
1261 lpath = malloc(len + sizeof("/lock.mdb") + len + sizeof("/data.db"));
1264 dpath = lpath + len + sizeof("/lock.mdb");
1265 sprintf(lpath, "%s/lock.mdb", path);
1266 sprintf(dpath, "%s/data.mdb", path);
1268 rc = mdbenv_setup_locks(env, lpath, mode, &excl);
1272 if (F_ISSET(flags, MDB_RDONLY))
1275 oflags = O_RDWR | O_CREAT;
1277 if ((env->me_fd = open(dpath, oflags, mode)) == -1)
1280 if ((rc = mdbenv_open2(env, flags)) != MDB_SUCCESS) {
1284 env->me_path = strdup(path);
1285 DPRINTF("opened dbenv %p", (void *) env);
1286 pthread_key_create(&env->me_txkey, mdbenv_reader_dest);
1288 mdbenv_share_locks(env);
1289 env->me_dbxs = calloc(env->me_maxdbs, sizeof(MDB_dbx));
1290 env->me_dbs[0] = calloc(env->me_maxdbs, sizeof(MDB_db));
1291 env->me_dbs[1] = calloc(env->me_maxdbs, sizeof(MDB_db));
1301 mdbenv_close(MDB_env *env)
1310 munmap(env->me_map, env->me_mapsize);
1314 size_t size = (env->me_maxreaders-1) * sizeof(MDB_reader) + sizeof(MDB_txninfo);
1315 munmap(env->me_txns, size);
1321 /* Search for key within a leaf page, using binary search.
1322 * Returns the smallest entry larger or equal to the key.
1323 * If exactp is non-null, stores whether the found entry was an exact match
1324 * in *exactp (1 or 0).
1325 * If kip is non-null, stores the index of the found entry in *kip.
1326 * If no entry larger or equal to the key is found, returns NULL.
1329 mdb_search_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp, MDB_val *key,
1330 int *exactp, unsigned int *kip)
1338 DPRINTF("searching %u keys in %s page %lu",
1340 IS_LEAF(mp) ? "leaf" : "branch",
1343 assert(NUMKEYS(mp) > 0);
1345 memset(&nodekey, 0, sizeof(nodekey));
1347 low = IS_LEAF(mp) ? 0 : 1;
1348 high = NUMKEYS(mp) - 1;
1349 while (low <= high) {
1350 i = (low + high) >> 1;
1351 node = NODEPTR(mp, i);
1353 nodekey.mv_size = node->mn_ksize;
1354 nodekey.mv_data = NODEKEY(node);
1356 if (txn->mt_dbxs[dbi].md_cmp)
1357 rc = txn->mt_dbxs[dbi].md_cmp(key, &nodekey);
1359 rc = _mdb_cmp(txn, dbi, key, &nodekey);
1362 DPRINTF("found leaf index %u [%.*s], rc = %i",
1363 i, (int)nodekey.mv_size, (char *)nodekey.mv_data, rc);
1365 DPRINTF("found branch index %u [%.*s -> %lu], rc = %i",
1366 i, (int)node->mn_ksize, (char *)NODEKEY(node),
1377 if (rc > 0) { /* Found entry is less than the key. */
1378 i++; /* Skip to get the smallest entry larger than key. */
1379 if (i >= NUMKEYS(mp))
1380 /* There is no entry larger or equal to the key. */
1384 *exactp = (rc == 0);
1385 if (kip) /* Store the key index if requested. */
1388 return NODEPTR(mp, i);
1392 cursor_pop_page(MDB_cursor *cursor)
1396 top = CURSOR_TOP(cursor);
1399 DPRINTF("popped page %lu off cursor %p", top->mp_page->mp_pgno, (void *) cursor);
1405 cursor_push_page(MDB_cursor *cursor, MDB_page *mp)
1409 DPRINTF("pushing page %lu on cursor %p", mp->mp_pgno, (void *) cursor);
1411 if ((ppage = calloc(1, sizeof(*ppage))) == NULL)
1413 ppage->mp_page = mp;
1414 CURSOR_PUSH(cursor, ppage);
1419 mdbenv_get_page(MDB_env *env, pgno_t pgno)
1422 MDB_txn *txn = env->me_txn;
1425 if (txn && !STAILQ_EMPTY(txn->mt_u.dirty_queue)) {
1427 STAILQ_FOREACH(dp, txn->mt_u.dirty_queue, h.md_next) {
1428 if (dp->p.mp_pgno == pgno) {
1436 p = (MDB_page *)(env->me_map + env->me_psize * pgno);
1442 mdb_search_page_root(MDB_txn *txn, MDB_dbi dbi, MDB_val *key,
1443 MDB_cursor *cursor, int modify, MDB_pageparent *mpp)
1445 MDB_page *mp = mpp->mp_page;
1448 if (cursor && cursor_push_page(cursor, mp) == NULL)
1451 while (IS_BRANCH(mp)) {
1455 DPRINTF("branch page %lu has %u keys", mp->mp_pgno, NUMKEYS(mp));
1456 assert(NUMKEYS(mp) > 1);
1457 DPRINTF("found index 0 to page %lu", NODEPGNO(NODEPTR(mp, 0)));
1459 if (key == NULL) /* Initialize cursor to first page. */
1461 else if (key->mv_size > MAXKEYSIZE && key->mv_data == NULL) {
1462 /* cursor to last page */
1466 node = mdb_search_node(txn, dbi, mp, key, &exact, &i);
1468 i = NUMKEYS(mp) - 1;
1476 DPRINTF("following index %u for key %.*s",
1477 i, (int)key->mv_size, (char *)key->mv_data);
1478 assert(i < NUMKEYS(mp));
1479 node = NODEPTR(mp, i);
1482 CURSOR_TOP(cursor)->mp_ki = i;
1484 mpp->mp_parent = mp;
1485 if ((mp = mdbenv_get_page(txn->mt_env, NODEPGNO(node))) == NULL)
1490 if (cursor && cursor_push_page(cursor, mp) == NULL)
1494 MDB_dhead *dh = ((MDB_dhead *)mp)-1;
1495 if ((rc = mdb_touch(txn, mpp)) != 0)
1497 dh = ((MDB_dhead *)mpp->mp_page)-1;
1498 dh->md_parent = mpp->mp_parent;
1499 dh->md_pi = mpp->mp_pi;
1506 DPRINTF("internal error, index points to a %02X page!?",
1511 DPRINTF("found leaf page %lu for key %.*s", mp->mp_pgno,
1512 key ? (int)key->mv_size : 0, key ? (char *)key->mv_data : NULL);
1517 /* Search for the page a given key should be in.
1518 * Stores a pointer to the found page in *mpp.
1519 * If key is NULL, search for the lowest page (used by mdb_cursor_first).
1520 * If cursor is non-null, pushes parent pages on the cursor stack.
1521 * If modify is true, visited pages are updated with new page numbers.
1524 mdb_search_page(MDB_txn *txn, MDB_dbi dbi, MDB_val *key,
1525 MDB_cursor *cursor, int modify, MDB_pageparent *mpp)
1530 /* Choose which root page to start with. If a transaction is given
1531 * use the root page from the transaction, otherwise read the last
1532 * committed root page.
1534 if (F_ISSET(txn->mt_flags, MDB_TXN_ERROR)) {
1535 DPRINTF("transaction has failed, must abort");
1538 root = txn->mt_dbs[dbi].md_root;
1540 if (root == P_INVALID) { /* Tree is empty. */
1541 DPRINTF("tree is empty");
1545 if ((mpp->mp_page = mdbenv_get_page(txn->mt_env, root)) == NULL)
1548 DPRINTF("root page has flags 0x%X", mpp->mp_page->mp_flags);
1551 /* For sub-databases, update main root first */
1552 if (dbi > MAIN_DBI && txn->mt_dbs[dbi].md_root ==
1553 txn->mt_env->me_dbs[txn->mt_env->me_db_toggle][dbi].md_root) {
1555 rc = mdb_search_page(txn, 0, &txn->mt_dbxs[dbi].md_name,
1560 if (!F_ISSET(mpp->mp_page->mp_flags, P_DIRTY)) {
1561 mpp->mp_parent = NULL;
1563 if ((rc = mdb_touch(txn, mpp)))
1565 txn->mt_dbs[dbi].md_root = mpp->mp_page->mp_pgno;
1569 return mdb_search_page_root(txn, dbi, key, cursor, modify, mpp);
1573 mdb_read_data(MDB_env *env, MDB_node *leaf, MDB_val *data)
1575 MDB_page *omp; /* overflow mpage */
1578 if (!F_ISSET(leaf->mn_flags, F_BIGDATA)) {
1579 data->mv_size = leaf->mn_dsize;
1580 data->mv_data = NODEDATA(leaf);
1584 /* Read overflow data.
1586 data->mv_size = leaf->mn_dsize;
1587 memcpy(&pgno, NODEDATA(leaf), sizeof(pgno));
1588 if ((omp = mdbenv_get_page(env, pgno)) == NULL) {
1589 DPRINTF("read overflow page %lu failed", pgno);
1592 data->mv_data = omp;
1598 mdb_get(MDB_txn *txn, MDB_dbi dbi,
1599 MDB_val *key, MDB_val *data)
1607 DPRINTF("===> get key [%.*s]", (int)key->mv_size, (char *)key->mv_data);
1609 if (key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
1613 if ((rc = mdb_search_page(txn, dbi, key, NULL, 0, &mpp)) != MDB_SUCCESS)
1616 leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, NULL);
1618 rc = mdb_read_data(txn->mt_env, leaf, data);
1627 mdb_sibling(MDB_cursor *cursor, int move_right)
1631 MDB_ppage *parent, *top;
1634 top = CURSOR_TOP(cursor);
1635 if ((parent = SLIST_NEXT(top, mp_entry)) == NULL) {
1636 return ENOENT; /* root has no siblings */
1639 DPRINTF("parent page is page %lu, index %u",
1640 parent->mp_page->mp_pgno, parent->mp_ki);
1642 cursor_pop_page(cursor);
1643 if (move_right ? (parent->mp_ki + 1 >= NUMKEYS(parent->mp_page))
1644 : (parent->mp_ki == 0)) {
1645 DPRINTF("no more keys left, moving to %s sibling",
1646 move_right ? "right" : "left");
1647 if ((rc = mdb_sibling(cursor, move_right)) != MDB_SUCCESS)
1649 parent = CURSOR_TOP(cursor);
1655 DPRINTF("just moving to %s index key %u",
1656 move_right ? "right" : "left", parent->mp_ki);
1658 assert(IS_BRANCH(parent->mp_page));
1660 indx = NODEPTR(parent->mp_page, parent->mp_ki);
1661 if ((mp = mdbenv_get_page(cursor->mc_txn->mt_env, indx->mn_pgno)) == NULL)
1664 mp->parent = parent->mp_page;
1665 mp->parent_index = parent->mp_ki;
1668 cursor_push_page(cursor, mp);
1674 mdb_set_key(MDB_node *node, MDB_val *key)
1679 key->mv_size = node->mn_ksize;
1680 key->mv_data = NODEKEY(node);
1686 mdb_cursor_next(MDB_cursor *cursor, MDB_val *key, MDB_val *data)
1692 if (cursor->mc_eof) {
1696 assert(cursor->mc_initialized);
1698 top = CURSOR_TOP(cursor);
1701 DPRINTF("cursor_next: top page is %lu in cursor %p", mp->mp_pgno, (void *) cursor);
1703 if (top->mp_ki + 1 >= NUMKEYS(mp)) {
1704 DPRINTF("=====> move to next sibling page");
1705 if (mdb_sibling(cursor, 1) != MDB_SUCCESS) {
1709 top = CURSOR_TOP(cursor);
1711 DPRINTF("next page is %lu, key index %u", mp->mp_pgno, top->mp_ki);
1715 DPRINTF("==> cursor points to page %lu with %u keys, key index %u",
1716 mp->mp_pgno, NUMKEYS(mp), top->mp_ki);
1718 assert(IS_LEAF(mp));
1719 leaf = NODEPTR(mp, top->mp_ki);
1721 if (data && mdb_read_data(cursor->mc_txn->mt_env, leaf, data) != MDB_SUCCESS)
1724 return mdb_set_key(leaf, key);
1728 mdb_cursor_set(MDB_cursor *cursor, MDB_val *key, MDB_val *data,
1738 assert(key->mv_size > 0);
1740 rc = mdb_search_page(cursor->mc_txn, cursor->mc_dbi, key, cursor, 0, &mpp);
1741 if (rc != MDB_SUCCESS)
1743 assert(IS_LEAF(mpp.mp_page));
1745 top = CURSOR_TOP(cursor);
1746 leaf = mdb_search_node(cursor->mc_txn, cursor->mc_dbi, mpp.mp_page, key, exactp, &top->mp_ki);
1747 if (exactp != NULL && !*exactp) {
1748 /* MDB_CURSOR_EXACT specified and not an exact match. */
1753 DPRINTF("===> inexact leaf not found, goto sibling");
1754 if ((rc = mdb_sibling(cursor, 1)) != MDB_SUCCESS)
1755 return rc; /* no entries matched */
1756 top = CURSOR_TOP(cursor);
1758 mpp.mp_page = top->mp_page;
1759 assert(IS_LEAF(mpp.mp_page));
1760 leaf = NODEPTR(mpp.mp_page, 0);
1763 cursor->mc_initialized = 1;
1766 if (data && (rc = mdb_read_data(cursor->mc_txn->mt_env, leaf, data)) != MDB_SUCCESS)
1769 rc = mdb_set_key(leaf, key);
1770 if (rc == MDB_SUCCESS) {
1771 DPRINTF("==> cursor placed on key %.*s",
1772 (int)key->mv_size, (char *)key->mv_data);
1780 mdb_cursor_first(MDB_cursor *cursor, MDB_val *key, MDB_val *data)
1786 rc = mdb_search_page(cursor->mc_txn, cursor->mc_dbi, NULL, cursor, 0, &mpp);
1787 if (rc != MDB_SUCCESS)
1789 assert(IS_LEAF(mpp.mp_page));
1791 leaf = NODEPTR(mpp.mp_page, 0);
1792 cursor->mc_initialized = 1;
1795 if (data && (rc = mdb_read_data(cursor->mc_txn->mt_env, leaf, data)) != MDB_SUCCESS)
1798 return mdb_set_key(leaf, key);
1802 mdb_cursor_last(MDB_cursor *cursor, MDB_val *key, MDB_val *data)
1809 lkey.mv_size = MAXKEYSIZE+1;
1810 lkey.mv_data = NULL;
1812 rc = mdb_search_page(cursor->mc_txn, cursor->mc_dbi, &lkey, cursor, 0, &mpp);
1813 if (rc != MDB_SUCCESS)
1815 assert(IS_LEAF(mpp.mp_page));
1817 leaf = NODEPTR(mpp.mp_page, NUMKEYS(mpp.mp_page)-1);
1818 cursor->mc_initialized = 1;
1821 if (data && (rc = mdb_read_data(cursor->mc_txn->mt_env, leaf, data)) != MDB_SUCCESS)
1824 return mdb_set_key(leaf, key);
1828 mdb_cursor_get(MDB_cursor *cursor, MDB_val *key, MDB_val *data,
1838 case MDB_CURSOR_EXACT:
1839 while (CURSOR_TOP(cursor) != NULL)
1840 cursor_pop_page(cursor);
1841 if (key == NULL || key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
1843 } else if (op == MDB_CURSOR_EXACT)
1844 rc = mdb_cursor_set(cursor, key, data, &exact);
1846 rc = mdb_cursor_set(cursor, key, data, NULL);
1849 if (!cursor->mc_initialized)
1850 rc = mdb_cursor_first(cursor, key, data);
1852 rc = mdb_cursor_next(cursor, key, data);
1855 while (CURSOR_TOP(cursor) != NULL)
1856 cursor_pop_page(cursor);
1857 rc = mdb_cursor_first(cursor, key, data);
1860 while (CURSOR_TOP(cursor) != NULL)
1861 cursor_pop_page(cursor);
1862 rc = mdb_cursor_last(cursor, key, data);
1865 DPRINTF("unhandled/unimplemented cursor operation %u", op);
1873 /* Allocate a page and initialize it
1876 mdb_new_page(MDB_txn *txn, MDB_dbi dbi, uint32_t flags, int num)
1880 if ((dp = mdb_alloc_page(txn, NULL, 0, num)) == NULL)
1882 DPRINTF("allocated new mpage %lu, page size %u",
1883 dp->p.mp_pgno, txn->mt_env->me_psize);
1884 dp->p.mp_flags = flags | P_DIRTY;
1885 dp->p.mp_lower = PAGEHDRSZ;
1886 dp->p.mp_upper = txn->mt_env->me_psize;
1888 if (IS_BRANCH(&dp->p))
1889 txn->mt_dbs[dbi].md_branch_pages++;
1890 else if (IS_LEAF(&dp->p))
1891 txn->mt_dbs[dbi].md_leaf_pages++;
1892 else if (IS_OVERFLOW(&dp->p)) {
1893 txn->mt_dbs[dbi].md_overflow_pages += num;
1894 dp->p.mp_pages = num;
1901 mdb_leaf_size(MDB_env *env, MDB_val *key, MDB_val *data)
1905 sz = LEAFSIZE(key, data);
1906 if (data->mv_size >= env->me_psize / MDB_MINKEYS) {
1907 /* put on overflow page */
1908 sz -= data->mv_size - sizeof(pgno_t);
1911 return sz + sizeof(indx_t);
1915 mdb_branch_size(MDB_env *env, MDB_val *key)
1920 if (sz >= env->me_psize / MDB_MINKEYS) {
1921 /* put on overflow page */
1922 /* not implemented */
1923 /* sz -= key->size - sizeof(pgno_t); */
1926 return sz + sizeof(indx_t);
1930 mdb_add_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp, indx_t indx,
1931 MDB_val *key, MDB_val *data, pgno_t pgno, uint8_t flags)
1934 size_t node_size = NODESIZE;
1937 MDB_dpage *ofp = NULL; /* overflow page */
1939 assert(mp->mp_upper >= mp->mp_lower);
1941 DPRINTF("add node [%.*s] to %s page %lu at index %i, key size %zu",
1942 key ? (int)key->mv_size : 0, key ? (char *)key->mv_data : NULL,
1943 IS_LEAF(mp) ? "leaf" : "branch",
1944 mp->mp_pgno, indx, key ? key->mv_size : 0);
1947 node_size += key->mv_size;
1951 if (F_ISSET(flags, F_BIGDATA)) {
1952 /* Data already on overflow page. */
1953 node_size += sizeof(pgno_t);
1954 } else if (data->mv_size >= txn->mt_env->me_psize / MDB_MINKEYS) {
1955 int ovpages = OVPAGES(data->mv_size, txn->mt_env->me_psize);
1956 /* Put data on overflow page. */
1957 DPRINTF("data size is %zu, put on overflow page",
1959 node_size += sizeof(pgno_t);
1960 if ((ofp = mdb_new_page(txn, dbi, P_OVERFLOW, ovpages)) == NULL)
1962 DPRINTF("allocated overflow page %lu", ofp->p.mp_pgno);
1965 node_size += data->mv_size;
1969 if (node_size + sizeof(indx_t) > SIZELEFT(mp)) {
1970 DPRINTF("not enough room in page %lu, got %u ptrs",
1971 mp->mp_pgno, NUMKEYS(mp));
1972 DPRINTF("upper - lower = %u - %u = %u", mp->mp_upper, mp->mp_lower,
1973 mp->mp_upper - mp->mp_lower);
1974 DPRINTF("node size = %zu", node_size);
1978 /* Move higher pointers up one slot. */
1979 for (i = NUMKEYS(mp); i > indx; i--)
1980 mp->mp_ptrs[i] = mp->mp_ptrs[i - 1];
1982 /* Adjust free space offsets. */
1983 ofs = mp->mp_upper - node_size;
1984 assert(ofs >= mp->mp_lower + sizeof(indx_t));
1985 mp->mp_ptrs[indx] = ofs;
1987 mp->mp_lower += sizeof(indx_t);
1989 /* Write the node data. */
1990 node = NODEPTR(mp, indx);
1991 node->mn_ksize = (key == NULL) ? 0 : key->mv_size;
1992 node->mn_flags = flags;
1994 node->mn_dsize = data->mv_size;
1996 node->mn_pgno = pgno;
1999 memcpy(NODEKEY(node), key->mv_data, key->mv_size);
2004 if (F_ISSET(flags, F_BIGDATA))
2005 memcpy(node->mn_data + key->mv_size, data->mv_data,
2008 memcpy(node->mn_data + key->mv_size, data->mv_data,
2011 memcpy(node->mn_data + key->mv_size, &ofp->p.mp_pgno,
2013 memcpy(METADATA(&ofp->p), data->mv_data, data->mv_size);
2021 mdb_del_node(MDB_page *mp, indx_t indx)
2024 indx_t i, j, numkeys, ptr;
2028 DPRINTF("delete node %u on %s page %lu", indx,
2029 IS_LEAF(mp) ? "leaf" : "branch", mp->mp_pgno);
2030 assert(indx < NUMKEYS(mp));
2032 node = NODEPTR(mp, indx);
2033 sz = NODESIZE + node->mn_ksize;
2035 if (F_ISSET(node->mn_flags, F_BIGDATA))
2036 sz += sizeof(pgno_t);
2038 sz += NODEDSZ(node);
2041 ptr = mp->mp_ptrs[indx];
2042 numkeys = NUMKEYS(mp);
2043 for (i = j = 0; i < numkeys; i++) {
2045 mp->mp_ptrs[j] = mp->mp_ptrs[i];
2046 if (mp->mp_ptrs[i] < ptr)
2047 mp->mp_ptrs[j] += sz;
2052 base = (char *)mp + mp->mp_upper;
2053 memmove(base + sz, base, ptr - mp->mp_upper);
2055 mp->mp_lower -= sizeof(indx_t);
2060 mdb_cursor_open(MDB_txn *txn, MDB_dbi dbi, MDB_cursor **ret)
2064 if (txn == NULL || ret == NULL)
2067 if ((cursor = calloc(1, sizeof(*cursor))) != NULL) {
2068 SLIST_INIT(&cursor->mc_stack);
2069 cursor->mc_dbi = dbi;
2070 cursor->mc_txn = txn;
2079 mdb_cursor_close(MDB_cursor *cursor)
2081 if (cursor != NULL) {
2082 while (!CURSOR_EMPTY(cursor))
2083 cursor_pop_page(cursor);
2085 /* btree_close(cursor->bt); */
2091 mdb_update_key(MDB_page *mp, indx_t indx, MDB_val *key)
2093 indx_t ptr, i, numkeys;
2099 node = NODEPTR(mp, indx);
2100 ptr = mp->mp_ptrs[indx];
2101 DPRINTF("update key %u (ofs %u) [%.*s] to [%.*s] on page %lu",
2103 (int)node->mn_ksize, (char *)NODEKEY(node),
2104 (int)key->mv_size, (char *)key->mv_data,
2107 delta = key->mv_size - node->mn_ksize;
2109 if (delta > 0 && SIZELEFT(mp) < delta) {
2110 DPRINTF("OUCH! Not enough room, delta = %d", delta);
2114 numkeys = NUMKEYS(mp);
2115 for (i = 0; i < numkeys; i++) {
2116 if (mp->mp_ptrs[i] <= ptr)
2117 mp->mp_ptrs[i] -= delta;
2120 base = (char *)mp + mp->mp_upper;
2121 len = ptr - mp->mp_upper + NODESIZE;
2122 memmove(base - delta, base, len);
2123 mp->mp_upper -= delta;
2125 node = NODEPTR(mp, indx);
2126 node->mn_ksize = key->mv_size;
2129 memcpy(NODEKEY(node), key->mv_data, key->mv_size);
2134 /* Move a node from src to dst.
2137 mdb_move_node(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *src, indx_t srcindx,
2138 MDB_pageparent *dst, indx_t dstindx)
2144 srcnode = NODEPTR(src->mp_page, srcindx);
2145 DPRINTF("moving %s node %u [%.*s] on page %lu to node %u on page %lu",
2146 IS_LEAF(src->mp_page) ? "leaf" : "branch",
2148 (int)srcnode->mn_ksize, (char *)NODEKEY(srcnode),
2149 src->mp_page->mp_pgno,
2150 dstindx, dst->mp_page->mp_pgno);
2152 /* Mark src and dst as dirty. */
2153 if ((rc = mdb_touch(txn, src)) ||
2154 (rc = mdb_touch(txn, dst)))
2157 /* Add the node to the destination page.
2159 key.mv_size = srcnode->mn_ksize;
2160 key.mv_data = NODEKEY(srcnode);
2161 data.mv_size = NODEDSZ(srcnode);
2162 data.mv_data = NODEDATA(srcnode);
2163 rc = mdb_add_node(txn, dbi, dst->mp_page, dstindx, &key, &data, NODEPGNO(srcnode),
2165 if (rc != MDB_SUCCESS)
2168 /* Delete the node from the source page.
2170 mdb_del_node(src->mp_page, srcindx);
2172 /* Update the parent separators.
2174 if (srcindx == 0 && src->mp_pi != 0) {
2175 DPRINTF("update separator for source page %lu to [%.*s]",
2176 src->mp_page->mp_pgno, (int)key.mv_size, (char *)key.mv_data);
2177 if ((rc = mdb_update_key(src->mp_parent, src->mp_pi,
2178 &key)) != MDB_SUCCESS)
2182 if (srcindx == 0 && IS_BRANCH(src->mp_page)) {
2184 nullkey.mv_size = 0;
2185 assert(mdb_update_key(src->mp_page, 0, &nullkey) == MDB_SUCCESS);
2188 if (dstindx == 0 && dst->mp_pi != 0) {
2189 DPRINTF("update separator for destination page %lu to [%.*s]",
2190 dst->mp_page->mp_pgno, (int)key.mv_size, (char *)key.mv_data);
2191 if ((rc = mdb_update_key(dst->mp_parent, dst->mp_pi,
2192 &key)) != MDB_SUCCESS)
2196 if (dstindx == 0 && IS_BRANCH(dst->mp_page)) {
2198 nullkey.mv_size = 0;
2199 assert(mdb_update_key(dst->mp_page, 0, &nullkey) == MDB_SUCCESS);
2206 mdb_merge(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *src, MDB_pageparent *dst)
2215 DPRINTF("merging page %lu and %lu", src->mp_page->mp_pgno, dst->mp_page->mp_pgno);
2217 assert(txn != NULL);
2218 assert(src->mp_parent); /* can't merge root page */
2219 assert(dst->mp_parent);
2221 /* Mark src and dst as dirty. */
2222 if ((rc = mdb_touch(txn, src)) ||
2223 (rc = mdb_touch(txn, dst)))
2226 /* Move all nodes from src to dst.
2228 for (i = 0; i < NUMKEYS(src->mp_page); i++) {
2229 srcnode = NODEPTR(src->mp_page, i);
2231 key.mv_size = srcnode->mn_ksize;
2232 key.mv_data = NODEKEY(srcnode);
2233 data.mv_size = NODEDSZ(srcnode);
2234 data.mv_data = NODEDATA(srcnode);
2235 rc = mdb_add_node(txn, dbi, dst->mp_page, NUMKEYS(dst->mp_page), &key,
2236 &data, NODEPGNO(srcnode), srcnode->mn_flags);
2237 if (rc != MDB_SUCCESS)
2241 DPRINTF("dst page %lu now has %u keys (%.1f%% filled)",
2242 dst->mp_page->mp_pgno, NUMKEYS(dst->mp_page), (float)PAGEFILL(txn->mt_env, dst->mp_page) / 10);
2244 /* Unlink the src page from parent.
2246 mdb_del_node(src->mp_parent, src->mp_pi);
2247 if (src->mp_pi == 0) {
2249 if ((rc = mdb_update_key(src->mp_parent, 0, &key)) != MDB_SUCCESS)
2253 if (IS_LEAF(src->mp_page))
2254 txn->mt_dbs[dbi].md_leaf_pages--;
2256 txn->mt_dbs[dbi].md_branch_pages--;
2258 mpp.mp_page = src->mp_parent;
2259 dh = (MDB_dhead *)src->mp_parent;
2261 mpp.mp_parent = dh->md_parent;
2262 mpp.mp_pi = dh->md_pi;
2264 return mdb_rebalance(txn, dbi, &mpp);
2267 #define FILL_THRESHOLD 250
2270 mdb_rebalance(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *mpp)
2275 indx_t si = 0, di = 0;
2277 assert(txn != NULL);
2278 assert(mpp != NULL);
2280 DPRINTF("rebalancing %s page %lu (has %u keys, %.1f%% full)",
2281 IS_LEAF(mpp->mp_page) ? "leaf" : "branch",
2282 mpp->mp_page->mp_pgno, NUMKEYS(mpp->mp_page), (float)PAGEFILL(txn->mt_env, mpp->mp_page) / 10);
2284 if (PAGEFILL(txn->mt_env, mpp->mp_page) >= FILL_THRESHOLD) {
2285 DPRINTF("no need to rebalance page %lu, above fill threshold",
2286 mpp->mp_page->mp_pgno);
2290 if (mpp->mp_parent == NULL) {
2291 if (NUMKEYS(mpp->mp_page) == 0) {
2292 DPRINTF("tree is completely empty");
2293 txn->mt_dbs[dbi].md_root = P_INVALID;
2294 txn->mt_dbs[dbi].md_depth--;
2295 txn->mt_dbs[dbi].md_leaf_pages--;
2296 } else if (IS_BRANCH(mpp->mp_page) && NUMKEYS(mpp->mp_page) == 1) {
2297 DPRINTF("collapsing root page!");
2298 txn->mt_dbs[dbi].md_root = NODEPGNO(NODEPTR(mpp->mp_page, 0));
2299 if ((root = mdbenv_get_page(txn->mt_env, txn->mt_dbs[dbi].md_root)) == NULL)
2301 txn->mt_dbs[dbi].md_depth--;
2302 txn->mt_dbs[dbi].md_branch_pages--;
2304 DPRINTF("root page doesn't need rebalancing");
2308 /* The parent (branch page) must have at least 2 pointers,
2309 * otherwise the tree is invalid.
2311 assert(NUMKEYS(mpp->mp_parent) > 1);
2313 /* Leaf page fill factor is below the threshold.
2314 * Try to move keys from left or right neighbor, or
2315 * merge with a neighbor page.
2320 if (mpp->mp_pi == 0) {
2321 /* We're the leftmost leaf in our parent.
2323 DPRINTF("reading right neighbor");
2324 node = NODEPTR(mpp->mp_parent, mpp->mp_pi + 1);
2325 if ((npp.mp_page = mdbenv_get_page(txn->mt_env, NODEPGNO(node))) == NULL)
2327 npp.mp_pi = mpp->mp_pi + 1;
2329 di = NUMKEYS(mpp->mp_page);
2331 /* There is at least one neighbor to the left.
2333 DPRINTF("reading left neighbor");
2334 node = NODEPTR(mpp->mp_parent, mpp->mp_pi - 1);
2335 if ((npp.mp_page = mdbenv_get_page(txn->mt_env, NODEPGNO(node))) == NULL)
2337 npp.mp_pi = mpp->mp_pi - 1;
2338 si = NUMKEYS(npp.mp_page) - 1;
2341 npp.mp_parent = mpp->mp_parent;
2343 DPRINTF("found neighbor page %lu (%u keys, %.1f%% full)",
2344 npp.mp_page->mp_pgno, NUMKEYS(npp.mp_page), (float)PAGEFILL(txn->mt_env, npp.mp_page) / 10);
2346 /* If the neighbor page is above threshold and has at least two
2347 * keys, move one key from it.
2349 * Otherwise we should try to merge them.
2351 if (PAGEFILL(txn->mt_env, npp.mp_page) >= FILL_THRESHOLD && NUMKEYS(npp.mp_page) >= 2)
2352 return mdb_move_node(txn, dbi, &npp, si, mpp, di);
2353 else { /* FIXME: if (has_enough_room()) */
2354 if (mpp->mp_pi == 0)
2355 return mdb_merge(txn, dbi, &npp, mpp);
2357 return mdb_merge(txn, dbi, mpp, &npp);
2362 mdb_del0(MDB_txn *txn, MDB_dbi dbi, unsigned int ki, MDB_pageparent *mpp, MDB_node *leaf)
2366 /* add overflow pages to free list */
2367 if (F_ISSET(leaf->mn_flags, F_BIGDATA)) {
2371 memcpy(&pg, NODEDATA(leaf), sizeof(pg));
2372 ovpages = OVPAGES(NODEDSZ(leaf), txn->mt_env->me_psize);
2373 for (i=0; i<ovpages; i++) {
2374 DPRINTF("freed ov page %lu", pg);
2375 mdb_idl_insert(txn->mt_free_pgs, pg);
2379 mdb_del_node(mpp->mp_page, ki);
2380 txn->mt_dbs[dbi].md_entries--;
2381 rc = mdb_rebalance(txn, dbi, mpp);
2382 if (rc != MDB_SUCCESS)
2383 txn->mt_flags |= MDB_TXN_ERROR;
2389 mdb_del(MDB_txn *txn, MDB_dbi dbi,
2390 MDB_val *key, MDB_val *data)
2397 DPRINTF("========> delete key %.*s", (int)key->mv_size, (char *)key->mv_data);
2399 assert(key != NULL);
2401 if (txn == NULL || dbi >= txn->mt_numdbs)
2404 if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
2408 if (key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
2412 mpp.mp_parent = NULL;
2414 if ((rc = mdb_search_page(txn, dbi, key, NULL, 1, &mpp)) != MDB_SUCCESS)
2417 leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, &ki);
2418 if (leaf == NULL || !exact) {
2422 if (data && (rc = mdb_read_data(txn->mt_env, leaf, data)) != MDB_SUCCESS)
2425 return mdb_del0(txn, dbi, ki, &mpp, leaf);
2428 /* Split page <*mpp>, and insert <key,(data|newpgno)> in either left or
2429 * right sibling, at index <*newindxp> (as if unsplit). Updates *mpp and
2430 * *newindxp with the actual values after split, ie if *mpp and *newindxp
2431 * refer to a node in the new right sibling page.
2434 mdb_split(MDB_txn *txn, MDB_dbi dbi, MDB_page **mpp, unsigned int *newindxp,
2435 MDB_val *newkey, MDB_val *newdata, pgno_t newpgno)
2438 int rc = MDB_SUCCESS, ins_new = 0;
2441 unsigned int i, j, split_indx;
2443 MDB_val sepkey, rkey, rdata;
2445 MDB_dpage *mdp, *rdp, *pdp;
2448 assert(txn != NULL);
2450 dh = ((MDB_dhead *)*mpp) - 1;
2451 mdp = (MDB_dpage *)dh;
2452 newindx = *newindxp;
2454 DPRINTF("-----> splitting %s page %lu and adding [%.*s] at index %i",
2455 IS_LEAF(&mdp->p) ? "leaf" : "branch", mdp->p.mp_pgno,
2456 (int)newkey->mv_size, (char *)newkey->mv_data, *newindxp);
2458 if (mdp->h.md_parent == NULL) {
2459 if ((pdp = mdb_new_page(txn, dbi, P_BRANCH, 1)) == NULL)
2462 mdp->h.md_parent = &pdp->p;
2463 txn->mt_dbs[dbi].md_root = pdp->p.mp_pgno;
2464 DPRINTF("root split! new root = %lu", pdp->p.mp_pgno);
2465 txn->mt_dbs[dbi].md_depth++;
2467 /* Add left (implicit) pointer. */
2468 if (mdb_add_node(txn, dbi, &pdp->p, 0, NULL, NULL,
2469 mdp->p.mp_pgno, 0) != MDB_SUCCESS)
2472 DPRINTF("parent branch page is %lu", mdp->h.md_parent->mp_pgno);
2475 /* Create a right sibling. */
2476 if ((rdp = mdb_new_page(txn, dbi, mdp->p.mp_flags, 1)) == NULL)
2478 rdp->h.md_parent = mdp->h.md_parent;
2479 rdp->h.md_pi = mdp->h.md_pi + 1;
2480 DPRINTF("new right sibling: page %lu", rdp->p.mp_pgno);
2482 /* Move half of the keys to the right sibling. */
2483 if ((copy = malloc(txn->mt_env->me_psize)) == NULL)
2485 memcpy(copy, &mdp->p, txn->mt_env->me_psize);
2486 memset(&mdp->p.mp_ptrs, 0, txn->mt_env->me_psize - PAGEHDRSZ);
2487 mdp->p.mp_lower = PAGEHDRSZ;
2488 mdp->p.mp_upper = txn->mt_env->me_psize;
2490 split_indx = NUMKEYS(copy) / 2 + 1;
2492 /* First find the separating key between the split pages.
2494 memset(&sepkey, 0, sizeof(sepkey));
2495 if (newindx == split_indx) {
2496 sepkey.mv_size = newkey->mv_size;
2497 sepkey.mv_data = newkey->mv_data;
2499 node = NODEPTR(copy, split_indx);
2500 sepkey.mv_size = node->mn_ksize;
2501 sepkey.mv_data = NODEKEY(node);
2504 DPRINTF("separator is [%.*s]", (int)sepkey.mv_size, (char *)sepkey.mv_data);
2506 /* Copy separator key to the parent.
2508 if (SIZELEFT(rdp->h.md_parent) < mdb_branch_size(txn->mt_env, &sepkey)) {
2509 rc = mdb_split(txn, dbi, &rdp->h.md_parent, &rdp->h.md_pi,
2510 &sepkey, NULL, rdp->p.mp_pgno);
2512 /* Right page might now have changed parent.
2513 * Check if left page also changed parent.
2515 if (rdp->h.md_parent != mdp->h.md_parent &&
2516 mdp->h.md_pi >= NUMKEYS(mdp->h.md_parent)) {
2517 mdp->h.md_parent = rdp->h.md_parent;
2518 mdp->h.md_pi = rdp->h.md_pi - 1;
2521 rc = mdb_add_node(txn, dbi, rdp->h.md_parent, rdp->h.md_pi,
2522 &sepkey, NULL, rdp->p.mp_pgno, 0);
2524 if (rc != MDB_SUCCESS) {
2529 for (i = j = 0; i <= NUMKEYS(copy); j++) {
2530 if (i < split_indx) {
2531 /* Re-insert in left sibling. */
2534 /* Insert in right sibling. */
2535 if (i == split_indx)
2536 /* Reset insert index for right sibling. */
2537 j = (i == newindx && ins_new);
2541 if (i == newindx && !ins_new) {
2542 /* Insert the original entry that caused the split. */
2543 rkey.mv_data = newkey->mv_data;
2544 rkey.mv_size = newkey->mv_size;
2545 if (IS_LEAF(&mdp->p)) {
2546 rdata.mv_data = newdata->mv_data;
2547 rdata.mv_size = newdata->mv_size;
2554 /* Update page and index for the new key. */
2557 } else if (i == NUMKEYS(copy)) {
2560 node = NODEPTR(copy, i);
2561 rkey.mv_data = NODEKEY(node);
2562 rkey.mv_size = node->mn_ksize;
2563 if (IS_LEAF(&mdp->p)) {
2564 rdata.mv_data = NODEDATA(node);
2565 rdata.mv_size = node->mn_dsize;
2567 pgno = node->mn_pgno;
2568 flags = node->mn_flags;
2573 if (!IS_LEAF(&mdp->p) && j == 0) {
2574 /* First branch index doesn't need key data. */
2578 rc = mdb_add_node(txn, dbi, &pdp->p, j, &rkey, &rdata, pgno,flags);
2586 mdb_put(MDB_txn *txn, MDB_dbi dbi,
2587 MDB_val *key, MDB_val *data, unsigned int flags)
2589 int rc = MDB_SUCCESS, exact;
2594 assert(key != NULL);
2595 assert(data != NULL);
2600 if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
2604 if (txn->mt_env->me_txn != txn) {
2608 if (key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
2612 DPRINTF("==> put key %.*s, size %zu, data size %zu",
2613 (int)key->mv_size, (char *)key->mv_data, key->mv_size, data->mv_size);
2615 mpp.mp_parent = NULL;
2617 rc = mdb_search_page(txn, dbi, key, NULL, 1, &mpp);
2618 if (rc == MDB_SUCCESS) {
2619 leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, &ki);
2620 if (leaf && exact) {
2621 if (F_ISSET(flags, MDB_NOOVERWRITE)) {
2622 DPRINTF("duplicate key %.*s",
2623 (int)key->mv_size, (char *)key->mv_data);
2626 /* same size, just replace it */
2627 if (NODEDSZ(leaf) == data->mv_size) {
2628 memcpy(NODEDATA(leaf), data->mv_data, data->mv_size);
2631 mdb_del_node(mpp.mp_page, ki);
2633 if (leaf == NULL) { /* append if not found */
2634 ki = NUMKEYS(mpp.mp_page);
2635 DPRINTF("appending key at index %i", ki);
2637 } else if (rc == ENOENT) {
2639 /* new file, just write a root leaf page */
2640 DPRINTF("allocating new root leaf page");
2641 if ((dp = mdb_new_page(txn, dbi, P_LEAF, 1)) == NULL) {
2644 mpp.mp_page = &dp->p;
2645 txn->mt_dbs[dbi].md_root = mpp.mp_page->mp_pgno;
2646 txn->mt_dbs[dbi].md_depth++;
2652 assert(IS_LEAF(mpp.mp_page));
2653 DPRINTF("there are %u keys, should insert new key at index %i",
2654 NUMKEYS(mpp.mp_page), ki);
2656 if (SIZELEFT(mpp.mp_page) < mdb_leaf_size(txn->mt_env, key, data)) {
2657 rc = mdb_split(txn, dbi, &mpp.mp_page, &ki, key, data, P_INVALID);
2659 /* There is room already in this leaf page. */
2660 rc = mdb_add_node(txn, dbi, mpp.mp_page, ki, key, data, 0, 0);
2663 if (rc != MDB_SUCCESS)
2664 txn->mt_flags |= MDB_TXN_ERROR;
2666 txn->mt_dbs[dbi].md_entries++;
2673 mdbenv_get_flags(MDB_env *env, unsigned int *arg)
2678 *arg = env->me_flags;
2683 mdbenv_get_path(MDB_env *env, const char **arg)
2688 *arg = env->me_path;
2693 mdbenv_stat(MDB_env *env, MDB_stat *arg)
2695 if (env == NULL || arg == NULL)
2698 arg->ms_psize = env->me_psize;
2699 arg->ms_depth = env->me_meta->mm_dbs[MAIN_DBI].md_depth;
2700 arg->ms_branch_pages = env->me_meta->mm_dbs[MAIN_DBI].md_branch_pages;
2701 arg->ms_leaf_pages = env->me_meta->mm_dbs[MAIN_DBI].md_leaf_pages;
2702 arg->ms_overflow_pages = env->me_meta->mm_dbs[MAIN_DBI].md_overflow_pages;
2703 arg->ms_entries = env->me_meta->mm_dbs[MAIN_DBI].md_entries;
2708 int mdb_open(MDB_txn *txn, const char *name, unsigned int flags, MDB_dbi *dbi)
2721 /* Is the DB already open? */
2723 for (i=2; i<txn->mt_numdbs; i++) {
2724 if (len == txn->mt_dbxs[i].md_name.mv_size &&
2725 !strncmp(name, txn->mt_dbxs[i].md_name.mv_data, len)) {
2731 if (txn->mt_numdbs >= txn->mt_env->me_maxdbs - 1)
2734 /* Find the DB info */
2736 key.mv_data = (void *)name;
2737 rc = mdb_get(txn, MAIN_DBI, &key, &data);
2739 /* Create if requested */
2740 if (rc == ENOENT && (flags & MDB_CREATE)) {
2742 data.mv_size = sizeof(MDB_db);
2743 data.mv_data = &dummy;
2744 memset(&dummy, 0, sizeof(dummy));
2745 dummy.md_root = P_INVALID;
2746 dummy.md_flags = flags & 0xffff;
2747 rc = mdb_put(txn, 0, &key, &data, 0);
2750 /* OK, got info, add to table */
2751 if (rc == MDB_SUCCESS) {
2752 txn->mt_dbxs[txn->mt_numdbs].md_name.mv_data = strdup(name);
2753 txn->mt_dbxs[txn->mt_numdbs].md_name.mv_size = len;
2754 txn->mt_dbxs[txn->mt_numdbs].md_cmp = NULL;
2755 txn->mt_dbxs[txn->mt_numdbs].md_dcmp = NULL;
2756 txn->mt_dbxs[txn->mt_numdbs].md_rel = NULL;
2757 memcpy(&txn->mt_dbs[txn->mt_numdbs], data.mv_data, sizeof(MDB_db));
2758 *dbi = txn->mt_numdbs;
2765 int mdb_stat(MDB_txn *txn, MDB_dbi dbi, MDB_stat *arg)
2767 if (txn == NULL || arg == NULL)
2770 arg->ms_psize = txn->mt_env->me_psize;
2771 arg->ms_depth = txn->mt_dbs[dbi].md_depth;
2772 arg->ms_branch_pages = txn->mt_dbs[dbi].md_branch_pages;
2773 arg->ms_leaf_pages = txn->mt_dbs[dbi].md_leaf_pages;
2774 arg->ms_overflow_pages = txn->mt_dbs[dbi].md_overflow_pages;
2775 arg->ms_entries = txn->mt_dbs[dbi].md_entries;
2780 void mdb_close(MDB_txn *txn, MDB_dbi dbi)
2783 if (dbi <= MAIN_DBI || dbi >= txn->mt_numdbs)
2785 ptr = txn->mt_dbxs[dbi].md_name.mv_data;
2786 txn->mt_dbxs[dbi].md_name.mv_data = NULL;
2787 txn->mt_dbxs[dbi].md_name.mv_size = 0;