27 #define ULONG unsigned long
36 #if DEBUG && defined(__GNUC__)
37 # define DPRINTF(fmt, ...) \
38 fprintf(stderr, "%s:%d: " fmt "\n", __func__, __LINE__, ##__VA_ARGS__)
40 # define DPRINTF(...) ((void) 0)
45 #define MDB_MAGIC 0xBEEFC0DE
47 #define MAXKEYSIZE 255
49 #define P_INVALID (~0UL)
51 #define F_ISSET(w, f) (((w) & (f)) == (f))
53 typedef uint16_t indx_t;
55 #define DEFAULT_READERS 126
56 #define DEFAULT_MAPSIZE 1048576
58 /* Lock descriptor stuff */
63 typedef struct MDB_rxbody {
68 #define CACHELINE 64 /* most CPUs. Itanium uses 128 */
71 typedef struct MDB_reader {
73 /* cache line alignment */
74 char pad[CACHELINE-sizeof(MDB_rxbody)];
79 uint32_t mt_version; \
80 pthread_mutex_t mt_mutex; \
82 uint32_t mt_numreaders
83 typedef struct MDB_txbody {
87 typedef struct MDB_txninfo {
89 char pad[CACHELINE-sizeof(MDB_txbody)];
90 pthread_mutex_t mt_wmutex;
91 char pad2[CACHELINE-sizeof(pthread_mutex_t)];
92 MDB_reader mt_readers[1];
95 /* Common header for all page types. Overflow pages
96 * occupy a number of contiguous pages with no
97 * headers on any page after the first.
99 typedef struct MDB_page { /* represents a page of storage */
100 #define mp_pgno mp_p.p_pgno
102 pgno_t p_pgno; /* page number */
105 #define P_BRANCH 0x01 /* branch page */
106 #define P_LEAF 0x02 /* leaf page */
107 #define P_OVERFLOW 0x04 /* overflow page */
108 #define P_META 0x08 /* meta page */
109 #define P_DIRTY 0x10 /* dirty page */
111 #define mp_lower mp_pb.pb.pb_lower
112 #define mp_upper mp_pb.pb.pb_upper
113 #define mp_pages mp_pb.pb_pages
116 indx_t pb_lower; /* lower bound of free space */
117 indx_t pb_upper; /* upper bound of free space */
119 uint32_t pb_pages; /* number of overflow pages */
121 indx_t mp_ptrs[1]; /* dynamic size */
124 #define PAGEHDRSZ ((unsigned) offsetof(MDB_page, mp_ptrs))
126 #define NUMKEYS(p) (((p)->mp_lower - PAGEHDRSZ) >> 1)
127 #define SIZELEFT(p) (indx_t)((p)->mp_upper - (p)->mp_lower)
128 #define PAGEFILL(env, p) (1000L * ((env)->me_psize - PAGEHDRSZ - SIZELEFT(p)) / \
129 ((env)->me_psize - PAGEHDRSZ))
130 #define IS_LEAF(p) F_ISSET((p)->mp_flags, P_LEAF)
131 #define IS_BRANCH(p) F_ISSET((p)->mp_flags, P_BRANCH)
132 #define IS_OVERFLOW(p) F_ISSET((p)->mp_flags, P_OVERFLOW)
134 #define OVPAGES(size, psize) (PAGEHDRSZ + size + psize - 1) / psize;
136 typedef struct MDB_db {
140 ULONG md_branch_pages;
142 ULONG md_overflow_pages;
150 typedef struct MDB_meta { /* meta (footer) page content */
153 void *mm_address; /* address for fixed mapping */
154 size_t mm_mapsize; /* size of mmap region */
155 MDB_db mm_dbs[2]; /* first is free space, 2nd is main db */
156 #define mm_psize mm_dbs[0].md_pad
157 #define mm_flags mm_dbs[0].md_flags
158 pgno_t mm_last_pg; /* last used page in file */
159 ULONG mm_txnid; /* txnid that committed this page */
162 typedef struct MDB_dhead { /* a dirty page */
163 STAILQ_ENTRY(MDB_dpage) md_next; /* queue of dirty pages */
165 unsigned md_pi; /* parent index */
169 typedef struct MDB_dpage {
174 STAILQ_HEAD(dirty_queue, MDB_dpage);
176 typedef struct MDB_oldpages {
177 struct MDB_oldpages *mo_next;
179 pgno_t mo_pages[1]; /* dynamic */
182 typedef struct MDB_pageparent {
188 static MDB_dpage *mdb_alloc_page(MDB_txn *txn, MDB_page *parent, unsigned int parent_idx, int num);
189 static int mdb_touch(MDB_txn *txn, MDB_pageparent *mp);
191 typedef struct MDB_ppage { /* ordered list of pages */
192 SLIST_ENTRY(MDB_ppage) mp_entry;
194 unsigned int mp_ki; /* cursor index on page */
196 SLIST_HEAD(page_stack, MDB_ppage);
198 #define CURSOR_EMPTY(c) SLIST_EMPTY(&(c)->mc_stack)
199 #define CURSOR_TOP(c) SLIST_FIRST(&(c)->mc_stack)
200 #define CURSOR_POP(c) SLIST_REMOVE_HEAD(&(c)->mc_stack, mp_entry)
201 #define CURSOR_PUSH(c,p) SLIST_INSERT_HEAD(&(c)->mc_stack, p, mp_entry)
205 struct page_stack mc_stack; /* stack of parent pages */
207 short mc_initialized; /* 1 if initialized */
208 short mc_eof; /* 1 if end is reached */
211 #define METAHASHLEN offsetof(MDB_meta, mm_hash)
212 #define METADATA(p) ((void *)((char *)p + PAGEHDRSZ))
214 typedef struct MDB_node {
215 #define mn_pgno mn_p.np_pgno
216 #define mn_dsize mn_p.np_dsize
218 pgno_t np_pgno; /* child page number */
219 uint32_t np_dsize; /* leaf data size */
221 unsigned int mn_flags:4;
222 unsigned int mn_ksize:12; /* key size */
223 #define F_BIGDATA 0x01 /* data put on overflow page */
228 typedef struct MDB_dbx {
230 MDB_cmp_func *md_cmp; /* user compare function */
231 MDB_cmp_func *md_dcmp; /* user dupsort function */
232 MDB_rel_func *md_rel; /* user relocate function */
236 pgno_t mt_next_pgno; /* next unallocated page */
240 pgno_t *mt_free_pgs; /* this is an IDL */
242 struct dirty_queue *dirty_queue; /* modified pages */
245 MDB_dbx *mt_dbxs; /* array */
247 unsigned int mt_numdbs;
249 #define MDB_TXN_RDONLY 0x01 /* read-only transaction */
250 #define MDB_TXN_ERROR 0x02 /* an error has occurred */
251 #define MDB_TXN_METOGGLE 0x04 /* used meta page 1 */
252 unsigned int mt_flags;
259 unsigned int me_maxreaders;
260 unsigned int me_numdbs;
261 unsigned int me_maxdbs;
264 MDB_txninfo *me_txns;
265 MDB_meta *me_metas[2];
267 MDB_txn *me_txn; /* current write transaction */
269 off_t me_size; /* current file size */
270 unsigned int me_psize;
272 MDB_dbx *me_dbxs; /* array */
274 MDB_oldpages *me_pghead;
275 MDB_oldpages *me_pgtail;
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_read_data(MDB_env *env, MDB_node *leaf, MDB_val *data);
314 static int mdb_rebalance(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *mp);
315 static int mdb_update_key(MDB_page *mp, indx_t indx, MDB_val *key);
316 static int mdb_move_node(MDB_txn *txn, MDB_dbi dbi,
317 MDB_pageparent *src, indx_t srcindx,
318 MDB_pageparent *dst, indx_t dstindx);
319 static int mdb_merge(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *src,
320 MDB_pageparent *dst);
321 static int mdb_split(MDB_txn *txn, MDB_dbi dbi, MDB_page **mpp,
322 unsigned int *newindxp, MDB_val *newkey,
323 MDB_val *newdata, pgno_t newpgno);
324 static MDB_dpage *mdb_new_page(MDB_txn *txn, MDB_dbi dbi, uint32_t flags, int num);
326 static void cursor_pop_page(MDB_cursor *cursor);
327 static MDB_ppage *cursor_push_page(MDB_cursor *cursor,
330 static int mdb_set_key(MDB_node *node, MDB_val *key);
331 static int mdb_sibling(MDB_cursor *cursor, int move_right);
332 static int mdb_cursor_next(MDB_cursor *cursor,
333 MDB_val *key, MDB_val *data);
334 static int mdb_cursor_set(MDB_cursor *cursor,
335 MDB_val *key, MDB_val *data, int *exactp);
336 static int mdb_cursor_first(MDB_cursor *cursor,
337 MDB_val *key, MDB_val *data);
338 static int mdb_cursor_last(MDB_cursor *cursor,
339 MDB_val *key, MDB_val *data);
341 static size_t mdb_leaf_size(MDB_env *env, MDB_val *key,
343 static size_t mdb_branch_size(MDB_env *env, MDB_val *key);
345 static int memncmp(const void *s1, size_t n1,
346 const void *s2, size_t n2);
347 static int memnrcmp(const void *s1, size_t n1,
348 const void *s2, size_t n2);
351 memncmp(const void *s1, size_t n1, const void *s2, size_t n2)
353 int diff, len_diff = -1;
356 len_diff = (n1 > n2);
359 diff = memcmp(s1, s2, n1);
360 return diff ? diff : len_diff;
364 memnrcmp(const void *s1, size_t n1, const void *s2, size_t n2)
366 const unsigned char *p1, *p2, *p1_lim;
373 p1 = (const unsigned char *)s1 + n1 - 1;
374 p2 = (const unsigned char *)s2 + n2 - 1;
376 for (p1_lim = (n1 <= n2 ? s1 : s2); *p1 == *p2; p1--, p2--) {
378 return (p1 != s1) ? (p1 != p2) : (p2 != s2) ? -1 : 0;
384 mdb_cmp(MDB_txn *txn, MDB_dbi dbi, const MDB_val *a, const MDB_val *b)
386 return txn->mt_dbxs[dbi].md_cmp(a, b);
390 _mdb_cmp(MDB_txn *txn, MDB_dbi dbi, const MDB_val *key1, const MDB_val *key2)
392 if (F_ISSET(txn->mt_dbs[dbi].md_flags, MDB_REVERSEKEY))
393 return memnrcmp(key1->mv_data, key1->mv_size, key2->mv_data, key2->mv_size);
395 return memncmp((char *)key1->mv_data, key1->mv_size, key2->mv_data, key2->mv_size);
398 /* Allocate new page(s) for writing */
400 mdb_alloc_page(MDB_txn *txn, MDB_page *parent, unsigned int parent_idx, int num)
403 pgno_t pgno = P_INVALID;
405 if (txn->mt_env->me_pghead) {
406 ULONG oldest = txn->mt_txnid - 2;
408 for (i=0; i<txn->mt_env->me_txns->mt_numreaders; i++) {
409 ULONG mr = txn->mt_env->me_txns->mt_readers[i].mr_txnid;
412 oldest = txn->mt_env->me_txns->mt_readers[i].mr_txnid;
414 if (oldest > txn->mt_env->me_pghead->mo_txnid) {
415 MDB_oldpages *mop = txn->mt_env->me_pghead;
416 txn->mt_oldest = oldest;
418 /* FIXME: For now, always use fresh pages. We
419 * really ought to search the free list for a
424 /* peel pages off tail, so we only have to truncate the list */
425 pgno = MDB_IDL_LAST(mop->mo_pages);
426 if (MDB_IDL_IS_RANGE(mop->mo_pages)) {
428 if (mop->mo_pages[2] > mop->mo_pages[1])
429 mop->mo_pages[0] = 0;
433 if (MDB_IDL_IS_ZERO(mop->mo_pages)) {
434 txn->mt_env->me_pghead = mop->mo_next;
435 if (!txn->mt_env->me_pghead)
436 txn->mt_env->me_pgtail = NULL;
443 if ((dp = malloc(txn->mt_env->me_psize * num + sizeof(MDB_dhead))) == NULL)
446 dp->h.md_parent = parent;
447 dp->h.md_pi = parent_idx;
448 STAILQ_INSERT_TAIL(txn->mt_u.dirty_queue, dp, h.md_next);
449 if (pgno == P_INVALID) {
450 dp->p.mp_pgno = txn->mt_next_pgno;
451 txn->mt_next_pgno += num;
453 dp->p.mp_pgno = pgno;
459 /* Touch a page: make it dirty and re-insert into tree with updated pgno.
462 mdb_touch(MDB_txn *txn, MDB_pageparent *pp)
464 MDB_page *mp = pp->mp_page;
469 if (!F_ISSET(mp->mp_flags, P_DIRTY)) {
471 if ((dp = mdb_alloc_page(txn, pp->mp_parent, pp->mp_pi, 1)) == NULL)
473 DPRINTF("touched page %lu -> %lu", mp->mp_pgno, dp->p.mp_pgno);
474 mdb_idl_insert(txn->mt_free_pgs, mp->mp_pgno);
475 pgno = dp->p.mp_pgno;
476 memcpy(&dp->p, mp, txn->mt_env->me_psize);
479 mp->mp_flags |= P_DIRTY;
481 /* Update the page number to new touched page. */
482 if (pp->mp_parent != NULL)
483 NODEPGNO(NODEPTR(pp->mp_parent, pp->mp_pi)) = mp->mp_pgno;
490 mdbenv_sync(MDB_env *env)
493 if (!F_ISSET(env->me_flags, MDB_NOSYNC)) {
494 if (fsync(env->me_fd))
501 mdb_txn_begin(MDB_env *env, int rdonly, MDB_txn **ret)
506 if ((txn = calloc(1, sizeof(*txn))) == NULL) {
507 DPRINTF("calloc: %s", strerror(errno));
512 txn->mt_flags |= MDB_TXN_RDONLY;
514 txn->mt_u.dirty_queue = calloc(1, sizeof(*txn->mt_u.dirty_queue));
515 if (txn->mt_u.dirty_queue == NULL) {
519 STAILQ_INIT(txn->mt_u.dirty_queue);
521 pthread_mutex_lock(&env->me_txns->mt_wmutex);
522 env->me_txns->mt_txnid++;
523 txn->mt_free_pgs = malloc(MDB_IDL_UM_SIZEOF);
524 if (txn->mt_free_pgs == NULL) {
525 free(txn->mt_u.dirty_queue);
529 txn->mt_free_pgs[0] = 0;
532 txn->mt_txnid = env->me_txns->mt_txnid;
534 MDB_reader *r = pthread_getspecific(env->me_txkey);
537 pthread_mutex_lock(&env->me_txns->mt_mutex);
538 for (i=0; i<env->me_maxreaders; i++) {
539 if (env->me_txns->mt_readers[i].mr_pid == 0) {
540 env->me_txns->mt_readers[i].mr_pid = getpid();
541 env->me_txns->mt_readers[i].mr_tid = pthread_self();
542 r = &env->me_txns->mt_readers[i];
543 pthread_setspecific(env->me_txkey, r);
544 if (i >= env->me_txns->mt_numreaders)
545 env->me_txns->mt_numreaders = i+1;
549 pthread_mutex_unlock(&env->me_txns->mt_mutex);
550 if (i == env->me_maxreaders) {
554 r->mr_txnid = txn->mt_txnid;
555 txn->mt_u.reader = r;
562 if ((rc = mdbenv_read_meta(env, &toggle)) != MDB_SUCCESS) {
567 /* Copy the DB arrays */
568 txn->mt_numdbs = env->me_numdbs;
569 txn->mt_dbxs = env->me_dbxs; /* mostly static anyway */
570 txn->mt_dbs = malloc(env->me_maxdbs * sizeof(MDB_db));
571 memcpy(txn->mt_dbs, env->me_meta->mm_dbs, 2 * sizeof(MDB_db));
572 if (txn->mt_numdbs > 2)
573 memcpy(txn->mt_dbs+2, env->me_dbs[env->me_db_toggle]+2,
574 (txn->mt_numdbs - 2) * sizeof(MDB_db));
578 txn->mt_flags |= MDB_TXN_METOGGLE;
579 txn->mt_next_pgno = env->me_meta->mm_last_pg+1;
582 DPRINTF("begin transaction %lu on mdbenv %p, root page %lu",
583 txn->mt_txnid, (void *) env, txn->mt_dbs[MAIN_DBI].md_root);
590 mdb_txn_abort(MDB_txn *txn)
599 DPRINTF("abort transaction %lu on mdbenv %p, root page %lu",
600 txn->mt_txnid, (void *) env, txn->mt_dbs[MAIN_DBI].md_root);
604 if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
605 txn->mt_u.reader->mr_txnid = 0;
607 /* Discard all dirty pages. Return any re-used pages
610 MDB_IDL_ZERO(txn->mt_free_pgs);
611 while (!STAILQ_EMPTY(txn->mt_u.dirty_queue)) {
612 dp = STAILQ_FIRST(txn->mt_u.dirty_queue);
613 STAILQ_REMOVE_HEAD(txn->mt_u.dirty_queue, h.md_next);
614 if (dp->p.mp_pgno <= env->me_meta->mm_last_pg)
615 mdb_idl_insert(txn->mt_free_pgs, dp->p.mp_pgno);
618 /* put back to head of free list */
619 if (!MDB_IDL_IS_ZERO(txn->mt_free_pgs)) {
622 mop = malloc(sizeof(MDB_oldpages) + MDB_IDL_SIZEOF(txn->mt_free_pgs) - sizeof(pgno_t));
623 mop->mo_next = env->me_pghead;
624 mop->mo_txnid = txn->mt_oldest - 1;
625 if (!env->me_pghead) {
626 env->me_pgtail = mop;
628 env->me_pghead = mop;
629 memcpy(mop->mo_pages, txn->mt_free_pgs, MDB_IDL_SIZEOF(txn->mt_free_pgs));
632 free(txn->mt_free_pgs);
633 free(txn->mt_u.dirty_queue);
635 env->me_txns->mt_txnid--;
636 pthread_mutex_unlock(&env->me_txns->mt_wmutex);
643 mdb_txn_commit(MDB_txn *txn)
652 struct iovec iov[MDB_COMMIT_PAGES];
655 assert(txn->mt_env != NULL);
659 if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
660 DPRINTF("attempt to commit read-only transaction");
665 if (txn != env->me_txn) {
666 DPRINTF("attempt to commit unknown transaction");
671 if (F_ISSET(txn->mt_flags, MDB_TXN_ERROR)) {
672 DPRINTF("error flag is set, can't commit");
677 if (STAILQ_EMPTY(txn->mt_u.dirty_queue))
680 DPRINTF("committing transaction %lu on mdbenv %p, root page %lu",
681 txn->mt_txnid, (void *) env, txn->mt_dbs[MAIN_DBI].md_root);
683 /* Update DB root pointers. Their pages have already been
684 * touched so this is all in-place and cannot fail.
688 data.mv_size = sizeof(MDB_db);
690 for (i = 2; i < env->me_numdbs; i++) {
691 if (env->me_dbs[env->me_db_toggle][i].md_root != txn->mt_dbs[i].md_root) {
692 data.mv_data = &txn->mt_dbs[i];
693 mdb_put(txn, i, &txn->mt_dbxs[i].md_name, &data, 0);
696 for (i = env->me_numdbs; i < txn->mt_numdbs; i++) {
697 data.mv_data = &txn->mt_dbs[i];
698 mdb_put(txn, i, &txn->mt_dbxs[i].md_name, &data, 0);
702 /* Commit up to MDB_COMMIT_PAGES dirty pages to disk until done.
709 STAILQ_FOREACH(dp, txn->mt_u.dirty_queue, h.md_next) {
710 if (dp->p.mp_pgno != next) {
712 DPRINTF("committing %u dirty pages", n);
713 rc = writev(env->me_fd, iov, n);
717 DPRINTF("short write, filesystem full?");
719 DPRINTF("writev: %s", strerror(errno));
726 lseek(env->me_fd, dp->p.mp_pgno * env->me_psize, SEEK_SET);
727 next = dp->p.mp_pgno;
729 DPRINTF("committing page %lu", dp->p.mp_pgno);
730 iov[n].iov_len = env->me_psize * dp->h.md_num;
731 iov[n].iov_base = &dp->p;
732 size += iov[n].iov_len;
733 next = dp->p.mp_pgno + dp->h.md_num;
734 /* clear dirty flag */
735 dp->p.mp_flags &= ~P_DIRTY;
736 if (++n >= MDB_COMMIT_PAGES) {
745 DPRINTF("committing %u dirty pages", n);
746 rc = writev(env->me_fd, iov, n);
750 DPRINTF("short write, filesystem full?");
752 DPRINTF("writev: %s", strerror(errno));
759 /* Drop the dirty pages.
761 while (!STAILQ_EMPTY(txn->mt_u.dirty_queue)) {
762 dp = STAILQ_FIRST(txn->mt_u.dirty_queue);
763 STAILQ_REMOVE_HEAD(txn->mt_u.dirty_queue, h.md_next);
767 if ((n = mdbenv_sync(env)) != 0 ||
768 (n = mdbenv_write_meta(txn)) != MDB_SUCCESS ||
769 (n = mdbenv_sync(env)) != 0) {
775 /* update the DB tables */
777 int toggle = !env->me_db_toggle;
779 for (i = 0; i < env->me_numdbs; i++) {
780 if (env->me_dbs[toggle][i].md_root != txn->mt_dbs[i].md_root)
781 env->me_dbs[toggle][i] = txn->mt_dbs[i];
783 for (i = env->me_numdbs; i < txn->mt_numdbs; i++) {
784 env->me_dbxs[i] = txn->mt_dbxs[i];
785 env->me_dbs[toggle][i] = txn->mt_dbs[i];
787 env->me_db_toggle = toggle;
788 env->me_numdbs = txn->mt_numdbs;
793 /* add to tail of free list */
794 if (!MDB_IDL_IS_ZERO(txn->mt_free_pgs)) {
797 mop = malloc(sizeof(MDB_oldpages) + MDB_IDL_SIZEOF(txn->mt_free_pgs) - sizeof(pgno_t));
799 if (env->me_pghead) {
800 env->me_pgtail->mo_next = mop;
802 env->me_pghead = mop;
804 env->me_pgtail = mop;
805 memcpy(mop->mo_pages, txn->mt_free_pgs, MDB_IDL_SIZEOF(txn->mt_free_pgs));
806 mop->mo_txnid = txn->mt_txnid;
809 pthread_mutex_unlock(&env->me_txns->mt_wmutex);
810 free(txn->mt_free_pgs);
811 free(txn->mt_u.dirty_queue);
822 mdbenv_read_header(MDB_env *env, MDB_meta *meta)
831 /* We don't know the page size yet, so use a minimum value.
834 if ((rc = pread(env->me_fd, page, PAGESIZE, 0)) == 0) {
836 } else if (rc != PAGESIZE) {
839 DPRINTF("read: %s", strerror(errno));
843 p = (MDB_page *)page;
845 if (!F_ISSET(p->mp_flags, P_META)) {
846 DPRINTF("page %lu not a meta page", p->mp_pgno);
851 if (m->mm_magic != MDB_MAGIC) {
852 DPRINTF("meta has invalid magic");
856 if (m->mm_version != MDB_VERSION) {
857 DPRINTF("database is version %u, expected version %u",
858 m->mm_version, MDB_VERSION);
862 memcpy(meta, m, sizeof(*m));
867 mdbenv_init_meta(MDB_env *env, MDB_meta *meta)
874 DPRINTF("writing new meta page");
875 psize = sysconf(_SC_PAGE_SIZE);
877 meta->mm_magic = MDB_MAGIC;
878 meta->mm_version = MDB_VERSION;
879 meta->mm_psize = psize;
880 meta->mm_last_pg = 1;
881 meta->mm_flags = env->me_flags & 0xffff;
882 meta->mm_dbs[0].md_root = P_INVALID;
883 meta->mm_dbs[1].md_root = P_INVALID;
885 p = calloc(2, psize);
887 p->mp_flags = P_META;
890 memcpy(m, meta, sizeof(*meta));
892 q = (MDB_page *)((char *)p + psize);
895 q->mp_flags = P_META;
898 memcpy(m, meta, sizeof(*meta));
900 rc = write(env->me_fd, p, psize * 2);
902 return (rc == (int)psize * 2) ? MDB_SUCCESS : errno;
906 mdbenv_write_meta(MDB_txn *txn)
915 assert(txn->mt_env != NULL);
917 DPRINTF("writing meta page for root page %lu", txn->mt_dbs[MAIN_DBI].md_root);
922 off = offsetof(MDB_meta, mm_dbs[0].md_depth);
923 len = sizeof(MDB_meta) - off;
926 meta.mm_dbs[0] = txn->mt_dbs[0];
927 meta.mm_dbs[1] = txn->mt_dbs[1];
928 meta.mm_last_pg = txn->mt_next_pgno - 1;
929 meta.mm_txnid = txn->mt_txnid;
931 if (!F_ISSET(txn->mt_flags, MDB_TXN_METOGGLE))
932 off += env->me_psize;
935 lseek(env->me_fd, off, SEEK_SET);
936 rc = write(env->me_fd, ptr, len);
938 DPRINTF("write failed, disk error?");
946 mdbenv_read_meta(MDB_env *env, int *which)
952 if (env->me_metas[0]->mm_txnid < env->me_metas[1]->mm_txnid)
955 if (env->me_meta != env->me_metas[toggle])
956 env->me_meta = env->me_metas[toggle];
960 DPRINTF("Using meta page %d", toggle);
966 mdbenv_create(MDB_env **env)
970 e = calloc(1, sizeof(*e));
971 if (!e) return ENOMEM;
973 e->me_maxreaders = DEFAULT_READERS;
982 mdbenv_set_mapsize(MDB_env *env, size_t size)
986 env->me_mapsize = size;
991 mdbenv_set_maxdbs(MDB_env *env, int dbs)
995 env->me_maxdbs = dbs;
1000 mdbenv_set_maxreaders(MDB_env *env, int readers)
1002 env->me_maxreaders = readers;
1007 mdbenv_get_maxreaders(MDB_env *env, int *readers)
1009 if (!env || !readers)
1011 *readers = env->me_maxreaders;
1016 mdbenv_open2(MDB_env *env, unsigned int flags)
1022 env->me_flags = flags;
1024 memset(&meta, 0, sizeof(meta));
1026 if ((i = mdbenv_read_header(env, &meta)) != 0) {
1029 DPRINTF("new mdbenv");
1033 if (!env->me_mapsize) {
1034 env->me_mapsize = newenv ? DEFAULT_MAPSIZE : meta.mm_mapsize;
1038 if (meta.mm_address && (flags & MDB_FIXEDMAP))
1040 env->me_map = mmap(meta.mm_address, env->me_mapsize, PROT_READ, i,
1042 if (env->me_map == MAP_FAILED)
1046 meta.mm_mapsize = env->me_mapsize;
1047 if (flags & MDB_FIXEDMAP)
1048 meta.mm_address = env->me_map;
1049 i = mdbenv_init_meta(env, &meta);
1050 if (i != MDB_SUCCESS) {
1051 munmap(env->me_map, env->me_mapsize);
1055 env->me_psize = meta.mm_psize;
1057 p = (MDB_page *)(MDB_page *)(MDB_page *)(MDB_page *)(MDB_page *)(MDB_page *)(MDB_page *)(MDB_page *)(MDB_page *)env->me_map;
1058 env->me_metas[0] = METADATA(p);
1059 env->me_metas[1] = (MDB_meta *)((char *)env->me_metas[0] + meta.mm_psize);
1061 if ((i = mdbenv_read_meta(env, NULL)) != 0)
1064 DPRINTF("opened database version %u, pagesize %u",
1065 env->me_meta->mm_version, env->me_psize);
1066 DPRINTF("depth: %u", env->me_meta->mm_dbs[MAIN_DBI].md_depth);
1067 DPRINTF("entries: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_entries);
1068 DPRINTF("branch pages: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_branch_pages);
1069 DPRINTF("leaf pages: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_leaf_pages);
1070 DPRINTF("overflow pages: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_overflow_pages);
1071 DPRINTF("root: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_root);
1077 mdbenv_reader_dest(void *ptr)
1079 MDB_reader *reader = ptr;
1081 reader->mr_txnid = 0;
1086 /* downgrade the exclusive lock on the region back to shared */
1088 mdbenv_share_locks(MDB_env *env)
1090 struct flock lock_info;
1092 env->me_txns->mt_txnid = env->me_meta->mm_txnid;
1094 memset((void *)&lock_info, 0, sizeof(lock_info));
1095 lock_info.l_type = F_RDLCK;
1096 lock_info.l_whence = SEEK_SET;
1097 lock_info.l_start = 0;
1098 lock_info.l_len = 1;
1099 fcntl(env->me_lfd, F_SETLK, &lock_info);
1103 mdbenv_setup_locks(MDB_env *env, char *lpath, int mode, int *excl)
1107 struct flock lock_info;
1111 if ((env->me_lfd = open(lpath, O_RDWR|O_CREAT, mode)) == -1) {
1115 /* Try to get exclusive lock. If we succeed, then
1116 * nobody is using the lock region and we should initialize it.
1118 memset((void *)&lock_info, 0, sizeof(lock_info));
1119 lock_info.l_type = F_WRLCK;
1120 lock_info.l_whence = SEEK_SET;
1121 lock_info.l_start = 0;
1122 lock_info.l_len = 1;
1123 rc = fcntl(env->me_lfd, F_SETLK, &lock_info);
1127 lock_info.l_type = F_RDLCK;
1128 rc = fcntl(env->me_lfd, F_SETLK, &lock_info);
1134 size = lseek(env->me_lfd, 0, SEEK_END);
1135 rsize = (env->me_maxreaders-1) * sizeof(MDB_reader) + sizeof(MDB_txninfo);
1136 if (size < rsize && *excl) {
1137 if (ftruncate(env->me_lfd, rsize) != 0) {
1143 size = rsize - sizeof(MDB_txninfo);
1144 env->me_maxreaders = size/sizeof(MDB_reader) + 1;
1146 env->me_txns = mmap(0, rsize, PROT_READ|PROT_WRITE, MAP_SHARED,
1148 if (env->me_txns == MAP_FAILED) {
1153 pthread_mutexattr_t mattr;
1155 pthread_mutexattr_init(&mattr);
1156 pthread_mutexattr_setpshared(&mattr, PTHREAD_PROCESS_SHARED);
1157 pthread_mutex_init(&env->me_txns->mt_mutex, &mattr);
1158 pthread_mutex_init(&env->me_txns->mt_wmutex, &mattr);
1159 env->me_txns->mt_version = MDB_VERSION;
1160 env->me_txns->mt_magic = MDB_MAGIC;
1161 env->me_txns->mt_txnid = 0;
1162 env->me_txns->mt_numreaders = 0;
1165 if (env->me_txns->mt_magic != MDB_MAGIC) {
1166 DPRINTF("lock region has invalid magic");
1169 if (env->me_txns->mt_version != MDB_VERSION) {
1170 DPRINTF("lock region is version %u, expected version %u",
1171 env->me_txns->mt_version, MDB_VERSION);
1174 if (errno != EACCES && errno != EAGAIN) {
1188 mdbenv_open(MDB_env *env, const char *path, unsigned int flags, mode_t mode)
1190 int oflags, rc, len, excl;
1191 char *lpath, *dpath;
1194 lpath = malloc(len + sizeof("/lock.mdb") + len + sizeof("/data.db"));
1197 dpath = lpath + len + sizeof("/lock.mdb");
1198 sprintf(lpath, "%s/lock.mdb", path);
1199 sprintf(dpath, "%s/data.mdb", path);
1201 rc = mdbenv_setup_locks(env, lpath, mode, &excl);
1205 if (F_ISSET(flags, MDB_RDONLY))
1208 oflags = O_RDWR | O_CREAT;
1210 if ((env->me_fd = open(dpath, oflags, mode)) == -1)
1213 if ((rc = mdbenv_open2(env, flags)) != MDB_SUCCESS) {
1217 env->me_path = strdup(path);
1218 DPRINTF("opened dbenv %p", (void *) env);
1219 pthread_key_create(&env->me_txkey, mdbenv_reader_dest);
1221 mdbenv_share_locks(env);
1222 env->me_dbxs = calloc(env->me_maxdbs, sizeof(MDB_dbx));
1223 env->me_dbs[0] = calloc(env->me_maxdbs, sizeof(MDB_db));
1224 env->me_dbs[1] = calloc(env->me_maxdbs, sizeof(MDB_db));
1234 mdbenv_close(MDB_env *env)
1243 munmap(env->me_map, env->me_mapsize);
1247 size_t size = (env->me_maxreaders-1) * sizeof(MDB_reader) + sizeof(MDB_txninfo);
1248 munmap(env->me_txns, size);
1254 /* Search for key within a leaf page, using binary search.
1255 * Returns the smallest entry larger or equal to the key.
1256 * If exactp is non-null, stores whether the found entry was an exact match
1257 * in *exactp (1 or 0).
1258 * If kip is non-null, stores the index of the found entry in *kip.
1259 * If no entry larger or equal to the key is found, returns NULL.
1262 mdb_search_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp, MDB_val *key,
1263 int *exactp, unsigned int *kip)
1271 DPRINTF("searching %u keys in %s page %lu",
1273 IS_LEAF(mp) ? "leaf" : "branch",
1276 assert(NUMKEYS(mp) > 0);
1278 memset(&nodekey, 0, sizeof(nodekey));
1280 low = IS_LEAF(mp) ? 0 : 1;
1281 high = NUMKEYS(mp) - 1;
1282 while (low <= high) {
1283 i = (low + high) >> 1;
1284 node = NODEPTR(mp, i);
1286 nodekey.mv_size = node->mn_ksize;
1287 nodekey.mv_data = NODEKEY(node);
1289 if (txn->mt_dbxs[dbi].md_cmp)
1290 rc = txn->mt_dbxs[dbi].md_cmp(key, &nodekey);
1292 rc = _mdb_cmp(txn, dbi, key, &nodekey);
1295 DPRINTF("found leaf index %u [%.*s], rc = %i",
1296 i, (int)nodekey.mv_size, (char *)nodekey.mv_data, rc);
1298 DPRINTF("found branch index %u [%.*s -> %lu], rc = %i",
1299 i, (int)node->mn_ksize, (char *)NODEKEY(node),
1310 if (rc > 0) { /* Found entry is less than the key. */
1311 i++; /* Skip to get the smallest entry larger than key. */
1312 if (i >= NUMKEYS(mp))
1313 /* There is no entry larger or equal to the key. */
1317 *exactp = (rc == 0);
1318 if (kip) /* Store the key index if requested. */
1321 return NODEPTR(mp, i);
1325 cursor_pop_page(MDB_cursor *cursor)
1329 top = CURSOR_TOP(cursor);
1332 DPRINTF("popped page %lu off cursor %p", top->mp_page->mp_pgno, (void *) cursor);
1338 cursor_push_page(MDB_cursor *cursor, MDB_page *mp)
1342 DPRINTF("pushing page %lu on cursor %p", mp->mp_pgno, (void *) cursor);
1344 if ((ppage = calloc(1, sizeof(*ppage))) == NULL)
1346 ppage->mp_page = mp;
1347 CURSOR_PUSH(cursor, ppage);
1352 mdbenv_get_page(MDB_env *env, pgno_t pgno)
1355 MDB_txn *txn = env->me_txn;
1358 if (txn && !STAILQ_EMPTY(txn->mt_u.dirty_queue)) {
1360 STAILQ_FOREACH(dp, txn->mt_u.dirty_queue, h.md_next) {
1361 if (dp->p.mp_pgno == pgno) {
1369 p = (MDB_page *)(env->me_map + env->me_psize * pgno);
1375 mdb_search_page_root(MDB_txn *txn, MDB_dbi dbi, MDB_val *key,
1376 MDB_cursor *cursor, int modify, MDB_pageparent *mpp)
1378 MDB_page *mp = mpp->mp_page;
1381 if (cursor && cursor_push_page(cursor, mp) == NULL)
1384 while (IS_BRANCH(mp)) {
1388 DPRINTF("branch page %lu has %u keys", mp->mp_pgno, NUMKEYS(mp));
1389 assert(NUMKEYS(mp) > 1);
1390 DPRINTF("found index 0 to page %lu", NODEPGNO(NODEPTR(mp, 0)));
1392 if (key == NULL) /* Initialize cursor to first page. */
1394 else if (key->mv_size > MAXKEYSIZE && key->mv_data == NULL) {
1395 /* cursor to last page */
1399 node = mdb_search_node(txn, dbi, mp, key, &exact, &i);
1401 i = NUMKEYS(mp) - 1;
1409 DPRINTF("following index %u for key %.*s",
1410 i, (int)key->mv_size, (char *)key->mv_data);
1411 assert(i < NUMKEYS(mp));
1412 node = NODEPTR(mp, i);
1415 CURSOR_TOP(cursor)->mp_ki = i;
1417 mpp->mp_parent = mp;
1418 if ((mp = mdbenv_get_page(txn->mt_env, NODEPGNO(node))) == NULL)
1423 if (cursor && cursor_push_page(cursor, mp) == NULL)
1427 MDB_dhead *dh = ((MDB_dhead *)mp)-1;
1428 if ((rc = mdb_touch(txn, mpp)) != 0)
1430 dh = ((MDB_dhead *)mpp->mp_page)-1;
1431 dh->md_parent = mpp->mp_parent;
1432 dh->md_pi = mpp->mp_pi;
1439 DPRINTF("internal error, index points to a %02X page!?",
1444 DPRINTF("found leaf page %lu for key %.*s", mp->mp_pgno,
1445 key ? (int)key->mv_size : 0, key ? (char *)key->mv_data : NULL);
1450 /* Search for the page a given key should be in.
1451 * Stores a pointer to the found page in *mpp.
1452 * If key is NULL, search for the lowest page (used by mdb_cursor_first).
1453 * If cursor is non-null, pushes parent pages on the cursor stack.
1454 * If modify is true, visited pages are updated with new page numbers.
1457 mdb_search_page(MDB_txn *txn, MDB_dbi dbi, MDB_val *key,
1458 MDB_cursor *cursor, int modify, MDB_pageparent *mpp)
1463 /* Choose which root page to start with. If a transaction is given
1464 * use the root page from the transaction, otherwise read the last
1465 * committed root page.
1467 if (F_ISSET(txn->mt_flags, MDB_TXN_ERROR)) {
1468 DPRINTF("transaction has failed, must abort");
1471 root = txn->mt_dbs[dbi].md_root;
1473 if (root == P_INVALID) { /* Tree is empty. */
1474 DPRINTF("tree is empty");
1478 if ((mpp->mp_page = mdbenv_get_page(txn->mt_env, root)) == NULL)
1481 DPRINTF("root page has flags 0x%X", mpp->mp_page->mp_flags);
1484 /* For sub-databases, update main root first */
1485 if (dbi > MAIN_DBI && txn->mt_dbs[dbi].md_root ==
1486 txn->mt_env->me_dbs[txn->mt_env->me_db_toggle][dbi].md_root) {
1488 rc = mdb_search_page(txn, 0, &txn->mt_dbxs[dbi].md_name,
1493 if (!F_ISSET(mpp->mp_page->mp_flags, P_DIRTY)) {
1494 mpp->mp_parent = NULL;
1496 if ((rc = mdb_touch(txn, mpp)))
1498 txn->mt_dbs[dbi].md_root = mpp->mp_page->mp_pgno;
1502 return mdb_search_page_root(txn, dbi, key, cursor, modify, mpp);
1506 mdb_read_data(MDB_env *env, MDB_node *leaf, MDB_val *data)
1508 MDB_page *omp; /* overflow mpage */
1511 if (!F_ISSET(leaf->mn_flags, F_BIGDATA)) {
1512 data->mv_size = leaf->mn_dsize;
1513 data->mv_data = NODEDATA(leaf);
1517 /* Read overflow data.
1519 data->mv_size = leaf->mn_dsize;
1520 memcpy(&pgno, NODEDATA(leaf), sizeof(pgno));
1521 if ((omp = mdbenv_get_page(env, pgno)) == NULL) {
1522 DPRINTF("read overflow page %lu failed", pgno);
1525 data->mv_data = omp;
1531 mdb_get(MDB_txn *txn, MDB_dbi dbi,
1532 MDB_val *key, MDB_val *data)
1540 DPRINTF("===> get key [%.*s]", (int)key->mv_size, (char *)key->mv_data);
1542 if (key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
1546 if ((rc = mdb_search_page(txn, dbi, key, NULL, 0, &mpp)) != MDB_SUCCESS)
1549 leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, NULL);
1551 rc = mdb_read_data(txn->mt_env, leaf, data);
1560 mdb_sibling(MDB_cursor *cursor, int move_right)
1564 MDB_ppage *parent, *top;
1567 top = CURSOR_TOP(cursor);
1568 if ((parent = SLIST_NEXT(top, mp_entry)) == NULL) {
1569 return ENOENT; /* root has no siblings */
1572 DPRINTF("parent page is page %lu, index %u",
1573 parent->mp_page->mp_pgno, parent->mp_ki);
1575 cursor_pop_page(cursor);
1576 if (move_right ? (parent->mp_ki + 1 >= NUMKEYS(parent->mp_page))
1577 : (parent->mp_ki == 0)) {
1578 DPRINTF("no more keys left, moving to %s sibling",
1579 move_right ? "right" : "left");
1580 if ((rc = mdb_sibling(cursor, move_right)) != MDB_SUCCESS)
1582 parent = CURSOR_TOP(cursor);
1588 DPRINTF("just moving to %s index key %u",
1589 move_right ? "right" : "left", parent->mp_ki);
1591 assert(IS_BRANCH(parent->mp_page));
1593 indx = NODEPTR(parent->mp_page, parent->mp_ki);
1594 if ((mp = mdbenv_get_page(cursor->mc_txn->mt_env, indx->mn_pgno)) == NULL)
1597 mp->parent = parent->mp_page;
1598 mp->parent_index = parent->mp_ki;
1601 cursor_push_page(cursor, mp);
1607 mdb_set_key(MDB_node *node, MDB_val *key)
1612 key->mv_size = node->mn_ksize;
1613 key->mv_data = NODEKEY(node);
1619 mdb_cursor_next(MDB_cursor *cursor, MDB_val *key, MDB_val *data)
1625 if (cursor->mc_eof) {
1629 assert(cursor->mc_initialized);
1631 top = CURSOR_TOP(cursor);
1634 DPRINTF("cursor_next: top page is %lu in cursor %p", mp->mp_pgno, (void *) cursor);
1636 if (top->mp_ki + 1 >= NUMKEYS(mp)) {
1637 DPRINTF("=====> move to next sibling page");
1638 if (mdb_sibling(cursor, 1) != MDB_SUCCESS) {
1642 top = CURSOR_TOP(cursor);
1644 DPRINTF("next page is %lu, key index %u", mp->mp_pgno, top->mp_ki);
1648 DPRINTF("==> cursor points to page %lu with %u keys, key index %u",
1649 mp->mp_pgno, NUMKEYS(mp), top->mp_ki);
1651 assert(IS_LEAF(mp));
1652 leaf = NODEPTR(mp, top->mp_ki);
1654 if (data && mdb_read_data(cursor->mc_txn->mt_env, leaf, data) != MDB_SUCCESS)
1657 return mdb_set_key(leaf, key);
1661 mdb_cursor_set(MDB_cursor *cursor, MDB_val *key, MDB_val *data,
1671 assert(key->mv_size > 0);
1673 rc = mdb_search_page(cursor->mc_txn, cursor->mc_dbi, key, cursor, 0, &mpp);
1674 if (rc != MDB_SUCCESS)
1676 assert(IS_LEAF(mpp.mp_page));
1678 top = CURSOR_TOP(cursor);
1679 leaf = mdb_search_node(cursor->mc_txn, cursor->mc_dbi, mpp.mp_page, key, exactp, &top->mp_ki);
1680 if (exactp != NULL && !*exactp) {
1681 /* MDB_CURSOR_EXACT specified and not an exact match. */
1686 DPRINTF("===> inexact leaf not found, goto sibling");
1687 if ((rc = mdb_sibling(cursor, 1)) != MDB_SUCCESS)
1688 return rc; /* no entries matched */
1689 top = CURSOR_TOP(cursor);
1691 mpp.mp_page = top->mp_page;
1692 assert(IS_LEAF(mpp.mp_page));
1693 leaf = NODEPTR(mpp.mp_page, 0);
1696 cursor->mc_initialized = 1;
1699 if (data && (rc = mdb_read_data(cursor->mc_txn->mt_env, leaf, data)) != MDB_SUCCESS)
1702 rc = mdb_set_key(leaf, key);
1703 if (rc == MDB_SUCCESS) {
1704 DPRINTF("==> cursor placed on key %.*s",
1705 (int)key->mv_size, (char *)key->mv_data);
1713 mdb_cursor_first(MDB_cursor *cursor, MDB_val *key, MDB_val *data)
1719 rc = mdb_search_page(cursor->mc_txn, cursor->mc_dbi, NULL, cursor, 0, &mpp);
1720 if (rc != MDB_SUCCESS)
1722 assert(IS_LEAF(mpp.mp_page));
1724 leaf = NODEPTR(mpp.mp_page, 0);
1725 cursor->mc_initialized = 1;
1728 if (data && (rc = mdb_read_data(cursor->mc_txn->mt_env, leaf, data)) != MDB_SUCCESS)
1731 return mdb_set_key(leaf, key);
1735 mdb_cursor_last(MDB_cursor *cursor, MDB_val *key, MDB_val *data)
1742 lkey.mv_size = MAXKEYSIZE+1;
1743 lkey.mv_data = NULL;
1745 rc = mdb_search_page(cursor->mc_txn, cursor->mc_dbi, &lkey, cursor, 0, &mpp);
1746 if (rc != MDB_SUCCESS)
1748 assert(IS_LEAF(mpp.mp_page));
1750 leaf = NODEPTR(mpp.mp_page, NUMKEYS(mpp.mp_page)-1);
1751 cursor->mc_initialized = 1;
1754 if (data && (rc = mdb_read_data(cursor->mc_txn->mt_env, leaf, data)) != MDB_SUCCESS)
1757 return mdb_set_key(leaf, key);
1761 mdb_cursor_get(MDB_cursor *cursor, MDB_val *key, MDB_val *data,
1771 case MDB_CURSOR_EXACT:
1772 while (CURSOR_TOP(cursor) != NULL)
1773 cursor_pop_page(cursor);
1774 if (key == NULL || key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
1776 } else if (op == MDB_CURSOR_EXACT)
1777 rc = mdb_cursor_set(cursor, key, data, &exact);
1779 rc = mdb_cursor_set(cursor, key, data, NULL);
1782 if (!cursor->mc_initialized)
1783 rc = mdb_cursor_first(cursor, key, data);
1785 rc = mdb_cursor_next(cursor, key, data);
1788 while (CURSOR_TOP(cursor) != NULL)
1789 cursor_pop_page(cursor);
1790 rc = mdb_cursor_first(cursor, key, data);
1793 while (CURSOR_TOP(cursor) != NULL)
1794 cursor_pop_page(cursor);
1795 rc = mdb_cursor_last(cursor, key, data);
1798 DPRINTF("unhandled/unimplemented cursor operation %u", op);
1806 /* Allocate a page and initialize it
1809 mdb_new_page(MDB_txn *txn, MDB_dbi dbi, uint32_t flags, int num)
1813 if ((dp = mdb_alloc_page(txn, NULL, 0, num)) == NULL)
1815 DPRINTF("allocated new mpage %lu, page size %u",
1816 dp->p.mp_pgno, txn->mt_env->me_psize);
1817 dp->p.mp_flags = flags | P_DIRTY;
1818 dp->p.mp_lower = PAGEHDRSZ;
1819 dp->p.mp_upper = txn->mt_env->me_psize;
1821 if (IS_BRANCH(&dp->p))
1822 txn->mt_dbs[dbi].md_branch_pages++;
1823 else if (IS_LEAF(&dp->p))
1824 txn->mt_dbs[dbi].md_leaf_pages++;
1825 else if (IS_OVERFLOW(&dp->p)) {
1826 txn->mt_dbs[dbi].md_overflow_pages += num;
1827 dp->p.mp_pages = num;
1834 mdb_leaf_size(MDB_env *env, MDB_val *key, MDB_val *data)
1838 sz = LEAFSIZE(key, data);
1839 if (data->mv_size >= env->me_psize / MDB_MINKEYS) {
1840 /* put on overflow page */
1841 sz -= data->mv_size - sizeof(pgno_t);
1844 return sz + sizeof(indx_t);
1848 mdb_branch_size(MDB_env *env, MDB_val *key)
1853 if (sz >= env->me_psize / MDB_MINKEYS) {
1854 /* put on overflow page */
1855 /* not implemented */
1856 /* sz -= key->size - sizeof(pgno_t); */
1859 return sz + sizeof(indx_t);
1863 mdb_add_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp, indx_t indx,
1864 MDB_val *key, MDB_val *data, pgno_t pgno, uint8_t flags)
1867 size_t node_size = NODESIZE;
1870 MDB_dpage *ofp = NULL; /* overflow page */
1872 assert(mp->mp_upper >= mp->mp_lower);
1874 DPRINTF("add node [%.*s] to %s page %lu at index %i, key size %zu",
1875 key ? (int)key->mv_size : 0, key ? (char *)key->mv_data : NULL,
1876 IS_LEAF(mp) ? "leaf" : "branch",
1877 mp->mp_pgno, indx, key ? key->mv_size : 0);
1880 node_size += key->mv_size;
1884 if (F_ISSET(flags, F_BIGDATA)) {
1885 /* Data already on overflow page. */
1886 node_size += sizeof(pgno_t);
1887 } else if (data->mv_size >= txn->mt_env->me_psize / MDB_MINKEYS) {
1888 int ovpages = OVPAGES(data->mv_size, txn->mt_env->me_psize);
1889 /* Put data on overflow page. */
1890 DPRINTF("data size is %zu, put on overflow page",
1892 node_size += sizeof(pgno_t);
1893 if ((ofp = mdb_new_page(txn, dbi, P_OVERFLOW, ovpages)) == NULL)
1895 DPRINTF("allocated overflow page %lu", ofp->p.mp_pgno);
1898 node_size += data->mv_size;
1902 if (node_size + sizeof(indx_t) > SIZELEFT(mp)) {
1903 DPRINTF("not enough room in page %lu, got %u ptrs",
1904 mp->mp_pgno, NUMKEYS(mp));
1905 DPRINTF("upper - lower = %u - %u = %u", mp->mp_upper, mp->mp_lower,
1906 mp->mp_upper - mp->mp_lower);
1907 DPRINTF("node size = %zu", node_size);
1911 /* Move higher pointers up one slot. */
1912 for (i = NUMKEYS(mp); i > indx; i--)
1913 mp->mp_ptrs[i] = mp->mp_ptrs[i - 1];
1915 /* Adjust free space offsets. */
1916 ofs = mp->mp_upper - node_size;
1917 assert(ofs >= mp->mp_lower + sizeof(indx_t));
1918 mp->mp_ptrs[indx] = ofs;
1920 mp->mp_lower += sizeof(indx_t);
1922 /* Write the node data. */
1923 node = NODEPTR(mp, indx);
1924 node->mn_ksize = (key == NULL) ? 0 : key->mv_size;
1925 node->mn_flags = flags;
1927 node->mn_dsize = data->mv_size;
1929 node->mn_pgno = pgno;
1932 memcpy(NODEKEY(node), key->mv_data, key->mv_size);
1937 if (F_ISSET(flags, F_BIGDATA))
1938 memcpy(node->mn_data + key->mv_size, data->mv_data,
1941 memcpy(node->mn_data + key->mv_size, data->mv_data,
1944 memcpy(node->mn_data + key->mv_size, &ofp->p.mp_pgno,
1946 memcpy(METADATA(&ofp->p), data->mv_data, data->mv_size);
1954 mdb_del_node(MDB_page *mp, indx_t indx)
1957 indx_t i, j, numkeys, ptr;
1961 DPRINTF("delete node %u on %s page %lu", indx,
1962 IS_LEAF(mp) ? "leaf" : "branch", mp->mp_pgno);
1963 assert(indx < NUMKEYS(mp));
1965 node = NODEPTR(mp, indx);
1966 sz = NODESIZE + node->mn_ksize;
1968 if (F_ISSET(node->mn_flags, F_BIGDATA))
1969 sz += sizeof(pgno_t);
1971 sz += NODEDSZ(node);
1974 ptr = mp->mp_ptrs[indx];
1975 numkeys = NUMKEYS(mp);
1976 for (i = j = 0; i < numkeys; i++) {
1978 mp->mp_ptrs[j] = mp->mp_ptrs[i];
1979 if (mp->mp_ptrs[i] < ptr)
1980 mp->mp_ptrs[j] += sz;
1985 base = (char *)mp + mp->mp_upper;
1986 memmove(base + sz, base, ptr - mp->mp_upper);
1988 mp->mp_lower -= sizeof(indx_t);
1993 mdb_cursor_open(MDB_txn *txn, MDB_dbi dbi, MDB_cursor **ret)
1997 if (txn == NULL || ret == NULL)
2000 if ((cursor = calloc(1, sizeof(*cursor))) != NULL) {
2001 SLIST_INIT(&cursor->mc_stack);
2002 cursor->mc_dbi = dbi;
2003 cursor->mc_txn = txn;
2012 mdb_cursor_close(MDB_cursor *cursor)
2014 if (cursor != NULL) {
2015 while (!CURSOR_EMPTY(cursor))
2016 cursor_pop_page(cursor);
2018 /* btree_close(cursor->bt); */
2024 mdb_update_key(MDB_page *mp, indx_t indx, MDB_val *key)
2026 indx_t ptr, i, numkeys;
2032 node = NODEPTR(mp, indx);
2033 ptr = mp->mp_ptrs[indx];
2034 DPRINTF("update key %u (ofs %u) [%.*s] to [%.*s] on page %lu",
2036 (int)node->mn_ksize, (char *)NODEKEY(node),
2037 (int)key->mv_size, (char *)key->mv_data,
2040 delta = key->mv_size - node->mn_ksize;
2042 if (delta > 0 && SIZELEFT(mp) < delta) {
2043 DPRINTF("OUCH! Not enough room, delta = %d", delta);
2047 numkeys = NUMKEYS(mp);
2048 for (i = 0; i < numkeys; i++) {
2049 if (mp->mp_ptrs[i] <= ptr)
2050 mp->mp_ptrs[i] -= delta;
2053 base = (char *)mp + mp->mp_upper;
2054 len = ptr - mp->mp_upper + NODESIZE;
2055 memmove(base - delta, base, len);
2056 mp->mp_upper -= delta;
2058 node = NODEPTR(mp, indx);
2059 node->mn_ksize = key->mv_size;
2062 memcpy(NODEKEY(node), key->mv_data, key->mv_size);
2067 /* Move a node from src to dst.
2070 mdb_move_node(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *src, indx_t srcindx,
2071 MDB_pageparent *dst, indx_t dstindx)
2077 srcnode = NODEPTR(src->mp_page, srcindx);
2078 DPRINTF("moving %s node %u [%.*s] on page %lu to node %u on page %lu",
2079 IS_LEAF(src->mp_page) ? "leaf" : "branch",
2081 (int)srcnode->mn_ksize, (char *)NODEKEY(srcnode),
2082 src->mp_page->mp_pgno,
2083 dstindx, dst->mp_page->mp_pgno);
2085 /* Mark src and dst as dirty. */
2086 if ((rc = mdb_touch(txn, src)) ||
2087 (rc = mdb_touch(txn, dst)))
2090 /* Add the node to the destination page.
2092 key.mv_size = srcnode->mn_ksize;
2093 key.mv_data = NODEKEY(srcnode);
2094 data.mv_size = NODEDSZ(srcnode);
2095 data.mv_data = NODEDATA(srcnode);
2096 rc = mdb_add_node(txn, dbi, dst->mp_page, dstindx, &key, &data, NODEPGNO(srcnode),
2098 if (rc != MDB_SUCCESS)
2101 /* Delete the node from the source page.
2103 mdb_del_node(src->mp_page, srcindx);
2105 /* Update the parent separators.
2107 if (srcindx == 0 && src->mp_pi != 0) {
2108 DPRINTF("update separator for source page %lu to [%.*s]",
2109 src->mp_page->mp_pgno, (int)key.mv_size, (char *)key.mv_data);
2110 if ((rc = mdb_update_key(src->mp_parent, src->mp_pi,
2111 &key)) != MDB_SUCCESS)
2115 if (srcindx == 0 && IS_BRANCH(src->mp_page)) {
2117 nullkey.mv_size = 0;
2118 assert(mdb_update_key(src->mp_page, 0, &nullkey) == MDB_SUCCESS);
2121 if (dstindx == 0 && dst->mp_pi != 0) {
2122 DPRINTF("update separator for destination page %lu to [%.*s]",
2123 dst->mp_page->mp_pgno, (int)key.mv_size, (char *)key.mv_data);
2124 if ((rc = mdb_update_key(dst->mp_parent, dst->mp_pi,
2125 &key)) != MDB_SUCCESS)
2129 if (dstindx == 0 && IS_BRANCH(dst->mp_page)) {
2131 nullkey.mv_size = 0;
2132 assert(mdb_update_key(dst->mp_page, 0, &nullkey) == MDB_SUCCESS);
2139 mdb_merge(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *src, MDB_pageparent *dst)
2148 DPRINTF("merging page %lu and %lu", src->mp_page->mp_pgno, dst->mp_page->mp_pgno);
2150 assert(txn != NULL);
2151 assert(src->mp_parent); /* can't merge root page */
2152 assert(dst->mp_parent);
2154 /* Mark src and dst as dirty. */
2155 if ((rc = mdb_touch(txn, src)) ||
2156 (rc = mdb_touch(txn, dst)))
2159 /* Move all nodes from src to dst.
2161 for (i = 0; i < NUMKEYS(src->mp_page); i++) {
2162 srcnode = NODEPTR(src->mp_page, i);
2164 key.mv_size = srcnode->mn_ksize;
2165 key.mv_data = NODEKEY(srcnode);
2166 data.mv_size = NODEDSZ(srcnode);
2167 data.mv_data = NODEDATA(srcnode);
2168 rc = mdb_add_node(txn, dbi, dst->mp_page, NUMKEYS(dst->mp_page), &key,
2169 &data, NODEPGNO(srcnode), srcnode->mn_flags);
2170 if (rc != MDB_SUCCESS)
2174 DPRINTF("dst page %lu now has %u keys (%.1f%% filled)",
2175 dst->mp_page->mp_pgno, NUMKEYS(dst->mp_page), (float)PAGEFILL(txn->mt_env, dst->mp_page) / 10);
2177 /* Unlink the src page from parent.
2179 mdb_del_node(src->mp_parent, src->mp_pi);
2180 if (src->mp_pi == 0) {
2182 if ((rc = mdb_update_key(src->mp_parent, 0, &key)) != MDB_SUCCESS)
2186 if (IS_LEAF(src->mp_page))
2187 txn->mt_dbs[dbi].md_leaf_pages--;
2189 txn->mt_dbs[dbi].md_branch_pages--;
2191 mpp.mp_page = src->mp_parent;
2192 dh = (MDB_dhead *)src->mp_parent;
2194 mpp.mp_parent = dh->md_parent;
2195 mpp.mp_pi = dh->md_pi;
2197 return mdb_rebalance(txn, dbi, &mpp);
2200 #define FILL_THRESHOLD 250
2203 mdb_rebalance(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *mpp)
2208 indx_t si = 0, di = 0;
2210 assert(txn != NULL);
2211 assert(mpp != NULL);
2213 DPRINTF("rebalancing %s page %lu (has %u keys, %.1f%% full)",
2214 IS_LEAF(mpp->mp_page) ? "leaf" : "branch",
2215 mpp->mp_page->mp_pgno, NUMKEYS(mpp->mp_page), (float)PAGEFILL(txn->mt_env, mpp->mp_page) / 10);
2217 if (PAGEFILL(txn->mt_env, mpp->mp_page) >= FILL_THRESHOLD) {
2218 DPRINTF("no need to rebalance page %lu, above fill threshold",
2219 mpp->mp_page->mp_pgno);
2223 if (mpp->mp_parent == NULL) {
2224 if (NUMKEYS(mpp->mp_page) == 0) {
2225 DPRINTF("tree is completely empty");
2226 txn->mt_dbs[dbi].md_root = P_INVALID;
2227 txn->mt_dbs[dbi].md_depth--;
2228 txn->mt_dbs[dbi].md_leaf_pages--;
2229 } else if (IS_BRANCH(mpp->mp_page) && NUMKEYS(mpp->mp_page) == 1) {
2230 DPRINTF("collapsing root page!");
2231 txn->mt_dbs[dbi].md_root = NODEPGNO(NODEPTR(mpp->mp_page, 0));
2232 if ((root = mdbenv_get_page(txn->mt_env, txn->mt_dbs[dbi].md_root)) == NULL)
2234 txn->mt_dbs[dbi].md_depth--;
2235 txn->mt_dbs[dbi].md_branch_pages--;
2237 DPRINTF("root page doesn't need rebalancing");
2241 /* The parent (branch page) must have at least 2 pointers,
2242 * otherwise the tree is invalid.
2244 assert(NUMKEYS(mpp->mp_parent) > 1);
2246 /* Leaf page fill factor is below the threshold.
2247 * Try to move keys from left or right neighbor, or
2248 * merge with a neighbor page.
2253 if (mpp->mp_pi == 0) {
2254 /* We're the leftmost leaf in our parent.
2256 DPRINTF("reading right neighbor");
2257 node = NODEPTR(mpp->mp_parent, mpp->mp_pi + 1);
2258 if ((npp.mp_page = mdbenv_get_page(txn->mt_env, NODEPGNO(node))) == NULL)
2260 npp.mp_pi = mpp->mp_pi + 1;
2262 di = NUMKEYS(mpp->mp_page);
2264 /* There is at least one neighbor to the left.
2266 DPRINTF("reading left neighbor");
2267 node = NODEPTR(mpp->mp_parent, mpp->mp_pi - 1);
2268 if ((npp.mp_page = mdbenv_get_page(txn->mt_env, NODEPGNO(node))) == NULL)
2270 npp.mp_pi = mpp->mp_pi - 1;
2271 si = NUMKEYS(npp.mp_page) - 1;
2274 npp.mp_parent = mpp->mp_parent;
2276 DPRINTF("found neighbor page %lu (%u keys, %.1f%% full)",
2277 npp.mp_page->mp_pgno, NUMKEYS(npp.mp_page), (float)PAGEFILL(txn->mt_env, npp.mp_page) / 10);
2279 /* If the neighbor page is above threshold and has at least two
2280 * keys, move one key from it.
2282 * Otherwise we should try to merge them.
2284 if (PAGEFILL(txn->mt_env, npp.mp_page) >= FILL_THRESHOLD && NUMKEYS(npp.mp_page) >= 2)
2285 return mdb_move_node(txn, dbi, &npp, si, mpp, di);
2286 else { /* FIXME: if (has_enough_room()) */
2287 if (mpp->mp_pi == 0)
2288 return mdb_merge(txn, dbi, &npp, mpp);
2290 return mdb_merge(txn, dbi, mpp, &npp);
2295 mdb_del(MDB_txn *txn, MDB_dbi dbi,
2296 MDB_val *key, MDB_val *data)
2303 DPRINTF("========> delete key %.*s", (int)key->mv_size, (char *)key->mv_data);
2305 assert(key != NULL);
2307 if (txn == NULL || dbi >= txn->mt_numdbs)
2310 if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
2314 if (key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
2318 if ((rc = mdb_search_page(txn, dbi, key, NULL, 1, &mpp)) != MDB_SUCCESS)
2321 leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, &ki);
2322 if (leaf == NULL || !exact) {
2326 if (data && (rc = mdb_read_data(txn->mt_env, leaf, data)) != MDB_SUCCESS)
2329 mdb_del_node(mpp.mp_page, ki);
2330 /* add overflow pages to free list */
2331 if (F_ISSET(leaf->mn_flags, F_BIGDATA)) {
2335 memcpy(&pg, NODEDATA(leaf), sizeof(pg));
2336 ovpages = OVPAGES(NODEDSZ(leaf), txn->mt_env->me_psize);
2337 for (i=0; i<ovpages; i++) {
2338 mdb_idl_insert(txn->mt_free_pgs, pg);
2342 txn->mt_dbs[dbi].md_entries--;
2343 rc = mdb_rebalance(txn, dbi, &mpp);
2344 if (rc != MDB_SUCCESS)
2345 txn->mt_flags |= MDB_TXN_ERROR;
2350 /* Split page <*mpp>, and insert <key,(data|newpgno)> in either left or
2351 * right sibling, at index <*newindxp> (as if unsplit). Updates *mpp and
2352 * *newindxp with the actual values after split, ie if *mpp and *newindxp
2353 * refer to a node in the new right sibling page.
2356 mdb_split(MDB_txn *txn, MDB_dbi dbi, MDB_page **mpp, unsigned int *newindxp,
2357 MDB_val *newkey, MDB_val *newdata, pgno_t newpgno)
2360 int rc = MDB_SUCCESS, ins_new = 0;
2363 unsigned int i, j, split_indx;
2365 MDB_val sepkey, rkey, rdata;
2367 MDB_dpage *mdp, *rdp, *pdp;
2370 assert(txn != NULL);
2372 dh = ((MDB_dhead *)*mpp) - 1;
2373 mdp = (MDB_dpage *)dh;
2374 newindx = *newindxp;
2376 DPRINTF("-----> splitting %s page %lu and adding [%.*s] at index %i",
2377 IS_LEAF(&mdp->p) ? "leaf" : "branch", mdp->p.mp_pgno,
2378 (int)newkey->mv_size, (char *)newkey->mv_data, *newindxp);
2380 if (mdp->h.md_parent == NULL) {
2381 if ((pdp = mdb_new_page(txn, dbi, P_BRANCH, 1)) == NULL)
2384 mdp->h.md_parent = &pdp->p;
2385 txn->mt_dbs[dbi].md_root = pdp->p.mp_pgno;
2386 DPRINTF("root split! new root = %lu", pdp->p.mp_pgno);
2387 txn->mt_dbs[dbi].md_depth++;
2389 /* Add left (implicit) pointer. */
2390 if (mdb_add_node(txn, dbi, &pdp->p, 0, NULL, NULL,
2391 mdp->p.mp_pgno, 0) != MDB_SUCCESS)
2394 DPRINTF("parent branch page is %lu", mdp->h.md_parent->mp_pgno);
2397 /* Create a right sibling. */
2398 if ((rdp = mdb_new_page(txn, dbi, mdp->p.mp_flags, 1)) == NULL)
2400 rdp->h.md_parent = mdp->h.md_parent;
2401 rdp->h.md_pi = mdp->h.md_pi + 1;
2402 DPRINTF("new right sibling: page %lu", rdp->p.mp_pgno);
2404 /* Move half of the keys to the right sibling. */
2405 if ((copy = malloc(txn->mt_env->me_psize)) == NULL)
2407 memcpy(copy, &mdp->p, txn->mt_env->me_psize);
2408 memset(&mdp->p.mp_ptrs, 0, txn->mt_env->me_psize - PAGEHDRSZ);
2409 mdp->p.mp_lower = PAGEHDRSZ;
2410 mdp->p.mp_upper = txn->mt_env->me_psize;
2412 split_indx = NUMKEYS(copy) / 2 + 1;
2414 /* First find the separating key between the split pages.
2416 memset(&sepkey, 0, sizeof(sepkey));
2417 if (newindx == split_indx) {
2418 sepkey.mv_size = newkey->mv_size;
2419 sepkey.mv_data = newkey->mv_data;
2421 node = NODEPTR(copy, split_indx);
2422 sepkey.mv_size = node->mn_ksize;
2423 sepkey.mv_data = NODEKEY(node);
2426 DPRINTF("separator is [%.*s]", (int)sepkey.mv_size, (char *)sepkey.mv_data);
2428 /* Copy separator key to the parent.
2430 if (SIZELEFT(rdp->h.md_parent) < mdb_branch_size(txn->mt_env, &sepkey)) {
2431 rc = mdb_split(txn, dbi, &rdp->h.md_parent, &rdp->h.md_pi,
2432 &sepkey, NULL, rdp->p.mp_pgno);
2434 /* Right page might now have changed parent.
2435 * Check if left page also changed parent.
2437 if (rdp->h.md_parent != mdp->h.md_parent &&
2438 mdp->h.md_pi >= NUMKEYS(mdp->h.md_parent)) {
2439 mdp->h.md_parent = rdp->h.md_parent;
2440 mdp->h.md_pi = rdp->h.md_pi - 1;
2443 rc = mdb_add_node(txn, dbi, rdp->h.md_parent, rdp->h.md_pi,
2444 &sepkey, NULL, rdp->p.mp_pgno, 0);
2446 if (rc != MDB_SUCCESS) {
2451 for (i = j = 0; i <= NUMKEYS(copy); j++) {
2452 if (i < split_indx) {
2453 /* Re-insert in left sibling. */
2456 /* Insert in right sibling. */
2457 if (i == split_indx)
2458 /* Reset insert index for right sibling. */
2459 j = (i == newindx && ins_new);
2463 if (i == newindx && !ins_new) {
2464 /* Insert the original entry that caused the split. */
2465 rkey.mv_data = newkey->mv_data;
2466 rkey.mv_size = newkey->mv_size;
2467 if (IS_LEAF(&mdp->p)) {
2468 rdata.mv_data = newdata->mv_data;
2469 rdata.mv_size = newdata->mv_size;
2476 /* Update page and index for the new key. */
2479 } else if (i == NUMKEYS(copy)) {
2482 node = NODEPTR(copy, i);
2483 rkey.mv_data = NODEKEY(node);
2484 rkey.mv_size = node->mn_ksize;
2485 if (IS_LEAF(&mdp->p)) {
2486 rdata.mv_data = NODEDATA(node);
2487 rdata.mv_size = node->mn_dsize;
2489 pgno = node->mn_pgno;
2490 flags = node->mn_flags;
2495 if (!IS_LEAF(&mdp->p) && j == 0) {
2496 /* First branch index doesn't need key data. */
2500 rc = mdb_add_node(txn, dbi, &pdp->p, j, &rkey, &rdata, pgno,flags);
2508 mdb_put(MDB_txn *txn, MDB_dbi dbi,
2509 MDB_val *key, MDB_val *data, unsigned int flags)
2511 int rc = MDB_SUCCESS, exact;
2516 assert(key != NULL);
2517 assert(data != NULL);
2522 if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
2526 if (txn->mt_env->me_txn != txn) {
2530 if (key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
2534 DPRINTF("==> put key %.*s, size %zu, data size %zu",
2535 (int)key->mv_size, (char *)key->mv_data, key->mv_size, data->mv_size);
2537 rc = mdb_search_page(txn, dbi, key, NULL, 1, &mpp);
2538 if (rc == MDB_SUCCESS) {
2539 leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, &ki);
2540 if (leaf && exact) {
2541 if (F_ISSET(flags, MDB_NOOVERWRITE)) {
2542 DPRINTF("duplicate key %.*s",
2543 (int)key->mv_size, (char *)key->mv_data);
2546 /* same size, just replace it */
2547 if (NODEDSZ(leaf) == data->mv_size) {
2548 memcpy(NODEDATA(leaf), data->mv_data, data->mv_size);
2551 mdb_del_node(mpp.mp_page, ki);
2553 if (leaf == NULL) { /* append if not found */
2554 ki = NUMKEYS(mpp.mp_page);
2555 DPRINTF("appending key at index %i", ki);
2557 } else if (rc == ENOENT) {
2559 /* new file, just write a root leaf page */
2560 DPRINTF("allocating new root leaf page");
2561 if ((dp = mdb_new_page(txn, dbi, P_LEAF, 1)) == NULL) {
2564 mpp.mp_page = &dp->p;
2565 txn->mt_dbs[dbi].md_root = mpp.mp_page->mp_pgno;
2566 txn->mt_dbs[dbi].md_depth++;
2572 assert(IS_LEAF(mpp.mp_page));
2573 DPRINTF("there are %u keys, should insert new key at index %i",
2574 NUMKEYS(mpp.mp_page), ki);
2576 if (SIZELEFT(mpp.mp_page) < mdb_leaf_size(txn->mt_env, key, data)) {
2577 rc = mdb_split(txn, dbi, &mpp.mp_page, &ki, key, data, P_INVALID);
2579 /* There is room already in this leaf page. */
2580 rc = mdb_add_node(txn, dbi, mpp.mp_page, ki, key, data, 0, 0);
2583 if (rc != MDB_SUCCESS)
2584 txn->mt_flags |= MDB_TXN_ERROR;
2586 txn->mt_dbs[dbi].md_entries++;
2593 mdbenv_get_flags(MDB_env *env, unsigned int *arg)
2598 *arg = env->me_flags;
2603 mdbenv_get_path(MDB_env *env, const char **arg)
2608 *arg = env->me_path;
2613 mdbenv_stat(MDB_env *env, MDB_stat *arg)
2615 if (env == NULL || arg == NULL)
2618 arg->ms_psize = env->me_psize;
2619 arg->ms_depth = env->me_meta->mm_dbs[MAIN_DBI].md_depth;
2620 arg->ms_branch_pages = env->me_meta->mm_dbs[MAIN_DBI].md_branch_pages;
2621 arg->ms_leaf_pages = env->me_meta->mm_dbs[MAIN_DBI].md_leaf_pages;
2622 arg->ms_overflow_pages = env->me_meta->mm_dbs[MAIN_DBI].md_overflow_pages;
2623 arg->ms_entries = env->me_meta->mm_dbs[MAIN_DBI].md_entries;
2628 int mdb_open(MDB_txn *txn, const char *name, unsigned int flags, MDB_dbi *dbi)
2641 /* Is the DB already open? */
2643 for (i=2; i<txn->mt_numdbs; i++) {
2644 if (len == txn->mt_dbxs[i].md_name.mv_size &&
2645 !strncmp(name, txn->mt_dbxs[i].md_name.mv_data, len)) {
2651 if (txn->mt_numdbs >= txn->mt_env->me_maxdbs - 1)
2654 /* Find the DB info */
2656 key.mv_data = (void *)name;
2657 rc = mdb_get(txn, MAIN_DBI, &key, &data);
2659 /* Create if requested */
2660 if (rc == ENOENT && (flags & MDB_CREATE)) {
2662 data.mv_size = sizeof(MDB_db);
2663 data.mv_data = &dummy;
2664 memset(&dummy, 0, sizeof(dummy));
2665 dummy.md_root = P_INVALID;
2666 dummy.md_flags = flags & 0xffff;
2667 rc = mdb_put(txn, 0, &key, &data, 0);
2670 /* OK, got info, add to table */
2671 if (rc == MDB_SUCCESS) {
2672 txn->mt_dbxs[txn->mt_numdbs].md_name.mv_data = strdup(name);
2673 txn->mt_dbxs[txn->mt_numdbs].md_name.mv_size = len;
2674 txn->mt_dbxs[txn->mt_numdbs].md_cmp = NULL;
2675 txn->mt_dbxs[txn->mt_numdbs].md_dcmp = NULL;
2676 txn->mt_dbxs[txn->mt_numdbs].md_rel = NULL;
2677 memcpy(&txn->mt_dbs[txn->mt_numdbs], data.mv_data, sizeof(MDB_db));
2678 *dbi = txn->mt_numdbs;
2685 int mdb_stat(MDB_txn *txn, MDB_dbi dbi, MDB_stat *arg)
2687 if (txn == NULL || arg == NULL)
2690 arg->ms_psize = txn->mt_env->me_psize;
2691 arg->ms_depth = txn->mt_dbs[dbi].md_depth;
2692 arg->ms_branch_pages = txn->mt_dbs[dbi].md_branch_pages;
2693 arg->ms_leaf_pages = txn->mt_dbs[dbi].md_leaf_pages;
2694 arg->ms_overflow_pages = txn->mt_dbs[dbi].md_overflow_pages;
2695 arg->ms_entries = txn->mt_dbs[dbi].md_entries;
2700 void mdb_close(MDB_txn *txn, MDB_dbi dbi)
2703 if (dbi <= MAIN_DBI || dbi >= txn->mt_numdbs)
2705 ptr = txn->mt_dbxs[dbi].md_name.mv_data;
2706 txn->mt_dbxs[dbi].md_name.mv_data = NULL;
2707 txn->mt_dbxs[dbi].md_name.mv_size = 0;