27 #define ULONG unsigned long
36 #if (DEBUG +0) && 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 pgno_t mp_pgno; /* page number */
101 #define P_BRANCH 0x01 /* branch page */
102 #define P_LEAF 0x02 /* leaf page */
103 #define P_OVERFLOW 0x04 /* overflow page */
104 #define P_META 0x08 /* meta page */
105 #define P_DIRTY 0x10 /* dirty page */
107 #define mp_lower mp_pb.pb.pb_lower
108 #define mp_upper mp_pb.pb.pb_upper
109 #define mp_pages mp_pb.pb_pages
112 indx_t pb_lower; /* lower bound of free space */
113 indx_t pb_upper; /* upper bound of free space */
115 uint32_t pb_pages; /* number of overflow pages */
117 indx_t mp_ptrs[1]; /* dynamic size */
120 #define PAGEHDRSZ ((unsigned) offsetof(MDB_page, mp_ptrs))
122 #define NUMKEYS(p) (((p)->mp_lower - PAGEHDRSZ) >> 1)
123 #define SIZELEFT(p) (indx_t)((p)->mp_upper - (p)->mp_lower)
124 #define PAGEFILL(env, p) (1000L * ((env)->me_meta.mm_psize - PAGEHDRSZ - SIZELEFT(p)) / \
125 ((env)->me_meta.mm_psize - PAGEHDRSZ))
126 #define IS_LEAF(p) F_ISSET((p)->mp_flags, P_LEAF)
127 #define IS_BRANCH(p) F_ISSET((p)->mp_flags, P_BRANCH)
128 #define IS_OVERFLOW(p) F_ISSET((p)->mp_flags, P_OVERFLOW)
130 #define OVPAGES(size, psize) (PAGEHDRSZ + size + psize - 1) / psize;
132 typedef struct MDB_meta { /* meta (footer) page content */
135 void *mm_address; /* address for fixed mapping */
136 size_t mm_mapsize; /* size of mmap region */
137 pgno_t mm_last_pg; /* last used page in file */
138 ULONG mm_txnid; /* txnid that committed this page */
142 ULONG mm_branch_pages;
144 ULONG mm_overflow_pages;
149 typedef struct MDB_dhead { /* a dirty page */
150 STAILQ_ENTRY(MDB_dpage) md_next; /* queue of dirty pages */
152 unsigned md_pi; /* parent index */
156 typedef struct MDB_dpage {
161 STAILQ_HEAD(dirty_queue, MDB_dpage);
163 typedef struct MDB_oldpages {
164 struct MDB_oldpages *mo_next;
166 pgno_t mo_pages[1]; /* dynamic */
169 typedef struct MDB_pageparent {
175 static MDB_dpage *mdb_alloc_page(MDB_txn *txn, MDB_page *parent, unsigned int parent_idx, int num);
176 static int mdb_touch(MDB_txn *txn, MDB_pageparent *mp);
178 typedef struct MDB_ppage { /* ordered list of pages */
179 SLIST_ENTRY(MDB_ppage) mp_entry;
181 unsigned int mp_ki; /* cursor index on page */
183 SLIST_HEAD(page_stack, MDB_ppage);
185 #define CURSOR_EMPTY(c) SLIST_EMPTY(&(c)->mc_stack)
186 #define CURSOR_TOP(c) SLIST_FIRST(&(c)->mc_stack)
187 #define CURSOR_POP(c) SLIST_REMOVE_HEAD(&(c)->mc_stack, mp_entry)
188 #define CURSOR_PUSH(c,p) SLIST_INSERT_HEAD(&(c)->mc_stack, p, mp_entry)
192 struct page_stack mc_stack; /* stack of parent pages */
194 short mc_initialized; /* 1 if initialized */
195 short mc_eof; /* 1 if end is reached */
198 #define METAHASHLEN offsetof(MDB_meta, mm_hash)
199 #define METADATA(p) ((void *)((char *)p + PAGEHDRSZ))
201 typedef struct MDB_node {
202 #define mn_pgno mn_p.np_pgno
203 #define mn_dsize mn_p.np_dsize
205 pgno_t np_pgno; /* child page number */
206 uint32_t np_dsize; /* leaf data size */
208 unsigned int mn_flags:4;
209 unsigned int mn_ksize:12; /* key size */
210 #define F_BIGDATA 0x01 /* data put on overflow page */
214 typedef struct MDB_db {
218 ULONG md_branch_pages;
220 ULONG md_overflow_pages;
225 typedef struct MDB_dbx {
227 MDB_cmp_func *md_cmp; /* user compare function */
228 MDB_cmp_func *md_dcmp; /* user dupsort function */
229 MDB_rel_func *md_rel; /* user relocate function */
235 pgno_t mt_next_pgno; /* next unallocated page */
239 pgno_t *mt_free_pgs; /* this is an IDL */
241 struct dirty_queue *dirty_queue; /* modified pages */
244 MDB_dbx *mt_dbxs; /* array */
245 MDB_db mt_db0; /* just for write txns */
246 unsigned int mt_numdbs;
248 #define MDB_TXN_RDONLY 0x01 /* read-only transaction */
249 #define MDB_TXN_ERROR 0x02 /* an error has occurred */
250 #define MDB_TXN_METOGGLE 0x04 /* used meta page 1 */
251 unsigned int mt_flags;
258 unsigned int me_maxreaders;
261 MDB_txninfo *me_txns;
263 MDB_txn *me_txn; /* current write transaction */
265 off_t me_size; /* current file size */
266 pthread_key_t me_txkey; /* thread-key for readers */
267 MDB_oldpages *me_pghead;
268 MDB_oldpages *me_pgtail;
269 MDB_dbx *me_dbxs; /* array */
270 unsigned int me_numdbs;
273 #define NODESIZE offsetof(MDB_node, mn_data)
275 #define INDXSIZE(k) (NODESIZE + ((k) == NULL ? 0 : (k)->mv_size))
276 #define LEAFSIZE(k, d) (NODESIZE + (k)->mv_size + (d)->mv_size)
277 #define NODEPTR(p, i) ((MDB_node *)((char *)(p) + (p)->mp_ptrs[i]))
278 #define NODEKEY(node) (void *)((node)->mn_data)
279 #define NODEDATA(node) (void *)((char *)(node)->mn_data + (node)->mn_ksize)
280 #define NODEPGNO(node) ((node)->mn_pgno)
281 #define NODEDSZ(node) ((node)->mn_dsize)
283 #define MDB_COMMIT_PAGES 64 /* max number of pages to write in one commit */
284 #define MDB_MAXCACHE_DEF 1024 /* max number of pages to keep in cache */
286 static int mdb_search_page_root(MDB_txn *txn,
287 MDB_dbi dbi, MDB_val *key,
288 MDB_cursor *cursor, int modify,
289 MDB_pageparent *mpp);
290 static int mdb_search_page(MDB_txn *txn,
291 MDB_dbi dbi, MDB_val *key,
292 MDB_cursor *cursor, int modify,
293 MDB_pageparent *mpp);
295 static int mdbenv_read_header(MDB_env *env);
296 static int mdb_check_meta_page(MDB_page *p);
297 static int mdbenv_read_meta(MDB_env *env, int *which);
298 static int mdbenv_write_meta(MDB_txn *txn);
299 static MDB_page *mdbenv_get_page(MDB_env *env, pgno_t pgno);
301 static MDB_node *mdb_search_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp,
302 MDB_val *key, int *exactp, unsigned int *kip);
303 static int mdb_add_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp,
304 indx_t indx, MDB_val *key, MDB_val *data,
305 pgno_t pgno, uint8_t flags);
306 static void mdb_del_node(MDB_page *mp, indx_t indx);
307 static int mdb_read_data(MDB_env *env, MDB_node *leaf, MDB_val *data);
309 static int mdb_rebalance(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *mp);
310 static int mdb_update_key(MDB_page *mp, indx_t indx, MDB_val *key);
311 static int mdb_move_node(MDB_txn *txn, MDB_dbi dbi,
312 MDB_pageparent *src, indx_t srcindx,
313 MDB_pageparent *dst, indx_t dstindx);
314 static int mdb_merge(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *src,
315 MDB_pageparent *dst);
316 static int mdb_split(MDB_txn *txn, MDB_dbi dbi, MDB_page **mpp,
317 unsigned int *newindxp, MDB_val *newkey,
318 MDB_val *newdata, pgno_t newpgno);
319 static MDB_dpage *mdb_new_page(MDB_txn *txn, MDB_dbi dbi, uint32_t flags, int num);
321 static void cursor_pop_page(MDB_cursor *cursor);
322 static MDB_ppage *cursor_push_page(MDB_cursor *cursor,
325 static int mdb_set_key(MDB_node *node, MDB_val *key);
326 static int mdb_sibling(MDB_cursor *cursor, int move_right);
327 static int mdb_cursor_next(MDB_cursor *cursor,
328 MDB_val *key, MDB_val *data);
329 static int mdb_cursor_set(MDB_cursor *cursor,
330 MDB_val *key, MDB_val *data, int *exactp);
331 static int mdb_cursor_first(MDB_cursor *cursor,
332 MDB_val *key, MDB_val *data);
333 static int mdb_cursor_last(MDB_cursor *cursor,
334 MDB_val *key, MDB_val *data);
336 static size_t mdb_leaf_size(MDB_env *env, MDB_val *key,
338 static size_t mdb_branch_size(MDB_env *env, MDB_val *key);
340 static int memncmp(const void *s1, size_t n1,
341 const void *s2, size_t n2);
342 static int memnrcmp(const void *s1, size_t n1,
343 const void *s2, size_t n2);
346 memncmp(const void *s1, size_t n1, const void *s2, size_t n2)
348 int diff, len_diff = -1;
351 len_diff = (n1 > n2);
354 diff = memcmp(s1, s2, n1);
355 return diff ? diff : len_diff;
359 memnrcmp(const void *s1, size_t n1, const void *s2, size_t n2)
361 const unsigned char *p1, *p2, *p1_lim;
368 p1 = (const unsigned char *)s1 + n1 - 1;
369 p2 = (const unsigned char *)s2 + n2 - 1;
371 for (p1_lim = (n1 <= n2 ? s1 : s2); *p1 == *p2; p1--, p2--) {
373 return (p1 != s1) ? (p1 != p2) : (p2 != s2) ? -1 : 0;
379 mdb_cmp(MDB_txn *txn, MDB_dbi dbi, const MDB_val *a, const MDB_val *b)
381 return txn->mt_dbxs[dbi].md_cmp(a, b);
385 _mdb_cmp(MDB_txn *txn, MDB_dbi dbi, const MDB_val *key1, const MDB_val *key2)
387 if (F_ISSET(txn->mt_dbxs[dbi].md_db->md_flags, MDB_REVERSEKEY))
388 return memnrcmp(key1->mv_data, key1->mv_size, key2->mv_data, key2->mv_size);
390 return memncmp((char *)key1->mv_data, key1->mv_size, key2->mv_data, key2->mv_size);
393 /* Allocate new page(s) for writing */
395 mdb_alloc_page(MDB_txn *txn, MDB_page *parent, unsigned int parent_idx, int num)
398 pgno_t pgno = P_INVALID;
400 if (txn->mt_env->me_pghead) {
401 ULONG oldest = txn->mt_txnid - 2;
403 for (i=0; i<txn->mt_env->me_txns->mt_numreaders; i++) {
404 ULONG mr = txn->mt_env->me_txns->mt_readers[i].mr_txnid;
407 oldest = txn->mt_env->me_txns->mt_readers[i].mr_txnid;
409 if (oldest > txn->mt_env->me_pghead->mo_txnid) {
410 MDB_oldpages *mop = txn->mt_env->me_pghead;
411 txn->mt_oldest = oldest;
413 /* FIXME: For now, always use fresh pages. We
414 * really ought to search the free list for a
419 /* peel pages off tail, so we only have to truncate the list */
420 pgno = MDB_IDL_LAST(mop->mo_pages);
421 if (MDB_IDL_IS_RANGE(mop->mo_pages)) {
423 if (mop->mo_pages[2] > mop->mo_pages[1])
424 mop->mo_pages[0] = 0;
428 if (MDB_IDL_IS_ZERO(mop->mo_pages)) {
429 txn->mt_env->me_pghead = mop->mo_next;
430 if (!txn->mt_env->me_pghead)
431 txn->mt_env->me_pgtail = NULL;
438 if ((dp = malloc(txn->mt_env->me_meta.mm_psize * num + sizeof(MDB_dhead))) == NULL)
441 dp->h.md_parent = parent;
442 dp->h.md_pi = parent_idx;
443 STAILQ_INSERT_TAIL(txn->mt_u.dirty_queue, dp, h.md_next);
444 if (pgno == P_INVALID) {
445 dp->p.mp_pgno = txn->mt_next_pgno;
446 txn->mt_next_pgno += num;
448 dp->p.mp_pgno = pgno;
454 /* Touch a page: make it dirty and re-insert into tree with updated pgno.
457 mdb_touch(MDB_txn *txn, MDB_pageparent *pp)
459 MDB_page *mp = pp->mp_page;
464 if (!F_ISSET(mp->mp_flags, P_DIRTY)) {
466 if ((dp = mdb_alloc_page(txn, pp->mp_parent, pp->mp_pi, 1)) == NULL)
468 DPRINTF("touched page %lu -> %lu", mp->mp_pgno, dp->p.mp_pgno);
469 mdb_idl_insert(txn->mt_free_pgs, mp->mp_pgno);
470 pgno = dp->p.mp_pgno;
471 memcpy(&dp->p, mp, txn->mt_env->me_meta.mm_psize);
474 mp->mp_flags |= P_DIRTY;
476 /* Update the page number to new touched page. */
477 if (pp->mp_parent != NULL)
478 NODEPGNO(NODEPTR(pp->mp_parent, pp->mp_pi)) = mp->mp_pgno;
485 mdbenv_sync(MDB_env *env)
488 if (!F_ISSET(env->me_flags, MDB_NOSYNC)) {
489 if (fsync(env->me_fd))
495 #define DBX_CHUNK 16 /* space for 16 DBs at a time */
498 mdb_txn_begin(MDB_env *env, int rdonly, MDB_txn **ret)
503 if ((txn = calloc(1, sizeof(*txn))) == NULL) {
504 DPRINTF("calloc: %s", strerror(errno));
509 txn->mt_flags |= MDB_TXN_RDONLY;
511 txn->mt_u.dirty_queue = calloc(1, sizeof(*txn->mt_u.dirty_queue));
512 if (txn->mt_u.dirty_queue == NULL) {
516 STAILQ_INIT(txn->mt_u.dirty_queue);
518 pthread_mutex_lock(&env->me_txns->mt_wmutex);
519 env->me_txns->mt_txnid++;
520 txn->mt_free_pgs = malloc(MDB_IDL_UM_SIZEOF);
521 if (txn->mt_free_pgs == NULL) {
522 free(txn->mt_u.dirty_queue);
526 txn->mt_free_pgs[0] = 0;
529 txn->mt_txnid = env->me_txns->mt_txnid;
531 MDB_reader *r = pthread_getspecific(env->me_txkey);
534 pthread_mutex_lock(&env->me_txns->mt_mutex);
535 for (i=0; i<env->me_maxreaders; i++) {
536 if (env->me_txns->mt_readers[i].mr_pid == 0) {
537 env->me_txns->mt_readers[i].mr_pid = getpid();
538 env->me_txns->mt_readers[i].mr_tid = pthread_self();
539 r = &env->me_txns->mt_readers[i];
540 pthread_setspecific(env->me_txkey, r);
541 if (i >= env->me_txns->mt_numreaders)
542 env->me_txns->mt_numreaders = i+1;
546 pthread_mutex_unlock(&env->me_txns->mt_mutex);
547 if (i == env->me_maxreaders) {
551 r->mr_txnid = txn->mt_txnid;
552 txn->mt_u.reader = r;
559 if ((rc = mdbenv_read_meta(env, &toggle)) != MDB_SUCCESS) {
564 /* Copy the DB arrays */
565 txn->mt_numdbs = env->me_numdbs;
566 rc = (txn->mt_numdbs % DBX_CHUNK) + 1;
567 txn->mt_dbxs = malloc(rc * DBX_CHUNK * sizeof(MDB_dbx));
568 memcpy(txn->mt_dbxs, env->me_dbxs, txn->mt_numdbs * sizeof(MDB_dbx));
571 memcpy(&txn->mt_db0, txn->mt_dbxs[0].md_db, sizeof(txn->mt_db0));
572 txn->mt_dbxs[0].md_db = &txn->mt_db0;
574 txn->mt_flags |= MDB_TXN_METOGGLE;
575 txn->mt_next_pgno = env->me_meta.mm_last_pg+1;
578 DPRINTF("begin transaction %lu on mdbenv %p, root page %lu",
579 txn->mt_txnid, (void *) env, txn->mt_dbxs[0].md_db->md_root);
586 mdb_txn_abort(MDB_txn *txn)
595 DPRINTF("abort transaction %lu on mdbenv %p, root page %lu",
596 txn->mt_txnid, (void *) env, txn->mt_dbxs[0].md_db->md_root);
600 if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
601 txn->mt_u.reader->mr_txnid = 0;
603 /* Discard all dirty pages. Return any re-used pages
606 MDB_IDL_ZERO(txn->mt_free_pgs);
607 while (!STAILQ_EMPTY(txn->mt_u.dirty_queue)) {
608 dp = STAILQ_FIRST(txn->mt_u.dirty_queue);
609 STAILQ_REMOVE_HEAD(txn->mt_u.dirty_queue, h.md_next);
610 if (dp->p.mp_pgno <= env->me_meta.mm_last_pg)
611 mdb_idl_insert(txn->mt_free_pgs, dp->p.mp_pgno);
614 /* put back to head of free list */
615 if (!MDB_IDL_IS_ZERO(txn->mt_free_pgs)) {
618 mop = malloc(sizeof(MDB_oldpages) + MDB_IDL_SIZEOF(txn->mt_free_pgs) - sizeof(pgno_t));
619 mop->mo_next = env->me_pghead;
620 mop->mo_txnid = txn->mt_oldest - 1;
621 if (!env->me_pghead) {
622 env->me_pgtail = mop;
624 env->me_pghead = mop;
625 memcpy(mop->mo_pages, txn->mt_free_pgs, MDB_IDL_SIZEOF(txn->mt_free_pgs));
628 free(txn->mt_free_pgs);
629 free(txn->mt_u.dirty_queue);
631 env->me_txns->mt_txnid--;
632 pthread_mutex_unlock(&env->me_txns->mt_wmutex);
639 mdb_txn_commit(MDB_txn *txn)
647 struct iovec iov[MDB_COMMIT_PAGES];
650 assert(txn->mt_env != NULL);
654 if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
655 DPRINTF("attempt to commit read-only transaction");
660 if (txn != env->me_txn) {
661 DPRINTF("attempt to commit unknown transaction");
666 if (F_ISSET(txn->mt_flags, MDB_TXN_ERROR)) {
667 DPRINTF("error flag is set, can't commit");
672 if (STAILQ_EMPTY(txn->mt_u.dirty_queue))
675 DPRINTF("committing transaction %lu on mdbenv %p, root page %lu",
676 txn->mt_txnid, (void *) env, txn->mt_dbxs[0].md_db->md_root);
678 /* Commit up to MDB_COMMIT_PAGES dirty pages to disk until done.
685 STAILQ_FOREACH(dp, txn->mt_u.dirty_queue, h.md_next) {
686 if (dp->p.mp_pgno != next) {
688 DPRINTF("committing %u dirty pages", n);
689 rc = writev(env->me_fd, iov, n);
693 DPRINTF("short write, filesystem full?");
695 DPRINTF("writev: %s", strerror(errno));
702 lseek(env->me_fd, dp->p.mp_pgno * env->me_meta.mm_psize, SEEK_SET);
703 next = dp->p.mp_pgno;
705 DPRINTF("committing page %lu", dp->p.mp_pgno);
706 iov[n].iov_len = env->me_meta.mm_psize * dp->h.md_num;
707 iov[n].iov_base = &dp->p;
708 size += iov[n].iov_len;
709 next = dp->p.mp_pgno + dp->h.md_num;
710 /* clear dirty flag */
711 dp->p.mp_flags &= ~P_DIRTY;
712 if (++n >= MDB_COMMIT_PAGES) {
721 DPRINTF("committing %u dirty pages", n);
722 rc = writev(env->me_fd, iov, n);
726 DPRINTF("short write, filesystem full?");
728 DPRINTF("writev: %s", strerror(errno));
735 /* Drop the dirty pages.
737 while (!STAILQ_EMPTY(txn->mt_u.dirty_queue)) {
738 dp = STAILQ_FIRST(txn->mt_u.dirty_queue);
739 STAILQ_REMOVE_HEAD(txn->mt_u.dirty_queue, h.md_next);
743 if ((n = mdbenv_sync(env)) != 0 ||
744 (n = mdbenv_write_meta(txn)) != MDB_SUCCESS ||
745 (n = mdbenv_sync(env)) != 0) {
752 MDB_dbx *p1 = env->me_dbxs;
754 txn->mt_dbxs[0].md_db = env->me_dbxs[0].md_db;
755 env->me_dbxs = txn->mt_dbxs;
756 env->me_numdbs = txn->mt_numdbs;
761 /* add to tail of free list */
762 if (!MDB_IDL_IS_ZERO(txn->mt_free_pgs)) {
765 mop = malloc(sizeof(MDB_oldpages) + MDB_IDL_SIZEOF(txn->mt_free_pgs) - sizeof(pgno_t));
767 if (env->me_pghead) {
768 env->me_pgtail->mo_next = mop;
770 env->me_pghead = mop;
772 env->me_pgtail = mop;
773 memcpy(mop->mo_pages, txn->mt_free_pgs, MDB_IDL_SIZEOF(txn->mt_free_pgs));
774 mop->mo_txnid = txn->mt_txnid;
777 pthread_mutex_unlock(&env->me_txns->mt_wmutex);
778 free(txn->mt_free_pgs);
779 free(txn->mt_u.dirty_queue);
790 mdbenv_read_header(MDB_env *env)
799 /* We don't know the page size yet, so use a minimum value.
802 if ((rc = pread(env->me_fd, page, PAGESIZE, 0)) == 0) {
804 } else if (rc != PAGESIZE) {
807 DPRINTF("read: %s", strerror(errno));
811 p = (MDB_page *)page;
813 if (!F_ISSET(p->mp_flags, P_META)) {
814 DPRINTF("page %lu not a meta page", p->mp_pgno);
819 if (m->mm_magic != MDB_MAGIC) {
820 DPRINTF("meta has invalid magic");
824 if (m->mm_version != MDB_VERSION) {
825 DPRINTF("database is version %u, expected version %u",
826 m->mm_version, MDB_VERSION);
830 memcpy(&env->me_meta, m, sizeof(*m));
835 mdbenv_init_meta(MDB_env *env)
842 DPRINTF("writing new meta page");
843 psize = sysconf(_SC_PAGE_SIZE);
845 env->me_meta.mm_magic = MDB_MAGIC;
846 env->me_meta.mm_version = MDB_VERSION;
847 env->me_meta.mm_psize = psize;
848 env->me_meta.mm_flags = env->me_flags & 0xffff;
849 env->me_meta.mm_root = P_INVALID;
850 env->me_meta.mm_last_pg = 1;
852 p = calloc(2, psize);
854 p->mp_flags = P_META;
857 memcpy(meta, &env->me_meta, sizeof(*meta));
859 q = (MDB_page *)((char *)p + psize);
862 q->mp_flags = P_META;
865 memcpy(meta, &env->me_meta, sizeof(*meta));
867 rc = write(env->me_fd, p, psize * 2);
869 return (rc == (int)psize * 2) ? MDB_SUCCESS : errno;
873 mdbenv_write_meta(MDB_txn *txn)
882 assert(txn->mt_env != NULL);
884 DPRINTF("writing meta page for root page %lu", txn->mt_dbxs[0].md_db->md_root);
889 off = offsetof(MDB_meta, mm_last_pg);
890 len = sizeof(MDB_meta) - off;
893 meta.mm_last_pg = txn->mt_next_pgno - 1;
894 meta.mm_txnid = txn->mt_txnid;
895 meta.mm_psize = env->me_meta.mm_psize;
896 meta.mm_flags = env->me_meta.mm_flags;
897 meta.mm_depth = txn->mt_dbxs[0].md_db->md_depth;
898 meta.mm_branch_pages = txn->mt_dbxs[0].md_db->md_branch_pages;
899 meta.mm_leaf_pages = txn->mt_dbxs[0].md_db->md_leaf_pages;
900 meta.mm_overflow_pages = txn->mt_dbxs[0].md_db->md_overflow_pages;
901 meta.mm_entries = txn->mt_dbxs[0].md_db->md_entries;
902 meta.mm_root = txn->mt_dbxs[0].md_db->md_root;
904 if (!F_ISSET(txn->mt_flags, MDB_TXN_METOGGLE))
905 off += env->me_meta.mm_psize;
908 lseek(env->me_fd, off, SEEK_SET);
909 rc = write(env->me_fd, ptr, len);
911 DPRINTF("write failed, disk error?");
918 /* Returns true if page p is a valid meta page, false otherwise.
921 mdb_check_meta_page(MDB_page *p)
923 if (!F_ISSET(p->mp_flags, P_META)) {
924 DPRINTF("page %lu not a meta page", p->mp_pgno);
932 mdbenv_read_meta(MDB_env *env, int *which)
940 if ((mp0 = mdbenv_get_page(env, 0)) == NULL ||
941 (mp1 = mdbenv_get_page(env, 1)) == NULL)
944 rc = mdb_check_meta_page(mp0);
947 rc = mdb_check_meta_page(mp1);
950 meta[0] = METADATA(mp0);
951 meta[1] = METADATA(mp1);
953 if (meta[0]->mm_txnid < meta[1]->mm_txnid)
956 if (meta[toggle]->mm_txnid > env->me_meta.mm_txnid) {
957 memcpy(&env->me_meta, meta[toggle], sizeof(env->me_meta));
962 DPRINTF("Using meta page %d", toggle);
968 mdbenv_create(MDB_env **env)
972 e = calloc(1, sizeof(*e));
973 if (!e) return ENOMEM;
975 e->me_meta.mm_mapsize = DEFAULT_MAPSIZE;
976 e->me_maxreaders = DEFAULT_READERS;
984 mdbenv_set_mapsize(MDB_env *env, size_t size)
988 env->me_mapsize = env->me_meta.mm_mapsize = size;
993 mdbenv_set_maxreaders(MDB_env *env, int readers)
995 env->me_maxreaders = readers;
1000 mdbenv_get_maxreaders(MDB_env *env, int *readers)
1002 if (!env || !readers)
1004 *readers = env->me_maxreaders;
1009 mdbenv_open2(MDB_env *env, unsigned int flags)
1013 env->me_flags = flags;
1015 if ((i = mdbenv_read_header(env)) != 0) {
1018 DPRINTF("new mdbenv");
1022 if (!env->me_mapsize)
1023 env->me_mapsize = env->me_meta.mm_mapsize;
1026 if (env->me_meta.mm_address && (flags & MDB_FIXEDMAP))
1028 env->me_map = mmap(env->me_meta.mm_address, env->me_mapsize, PROT_READ, i,
1030 if (env->me_map == MAP_FAILED)
1034 env->me_meta.mm_mapsize = env->me_mapsize;
1035 if (flags & MDB_FIXEDMAP)
1036 env->me_meta.mm_address = env->me_map;
1037 i = mdbenv_init_meta(env);
1038 if (i != MDB_SUCCESS) {
1039 munmap(env->me_map, env->me_mapsize);
1044 if ((i = mdbenv_read_meta(env, NULL)) != 0)
1047 DPRINTF("opened database version %u, pagesize %u",
1048 env->me_meta.mm_version, env->me_meta.mm_psize);
1049 DPRINTF("depth: %u", env->me_meta.mm_depth);
1050 DPRINTF("entries: %lu", env->me_meta.mm_entries);
1051 DPRINTF("branch pages: %lu", env->me_meta.mm_branch_pages);
1052 DPRINTF("leaf pages: %lu", env->me_meta.mm_leaf_pages);
1053 DPRINTF("overflow pages: %lu", env->me_meta.mm_overflow_pages);
1054 DPRINTF("root: %lu", env->me_meta.mm_root);
1060 mdbenv_reader_dest(void *ptr)
1062 MDB_reader *reader = ptr;
1064 reader->mr_txnid = 0;
1070 mdbenv_share_locks(MDB_env *env)
1072 struct flock lock_info;
1074 env->me_txns->mt_txnid = env->me_meta.mm_txnid;
1076 memset((void *)&lock_info, 0, sizeof(lock_info));
1077 lock_info.l_type = F_RDLCK;
1078 lock_info.l_whence = SEEK_SET;
1079 lock_info.l_start = 0;
1080 lock_info.l_len = 1;
1081 fcntl(env->me_lfd, F_SETLK, &lock_info);
1085 mdbenv_setup_locks(MDB_env *env, char *lpath, int mode, int *excl)
1089 struct flock lock_info;
1093 if ((env->me_lfd = open(lpath, O_RDWR|O_CREAT, mode)) == -1) {
1097 /* Try to get exclusive lock. If we succeed, then
1098 * nobody is using the lock region and we should initialize it.
1100 memset((void *)&lock_info, 0, sizeof(lock_info));
1101 lock_info.l_type = F_WRLCK;
1102 lock_info.l_whence = SEEK_SET;
1103 lock_info.l_start = 0;
1104 lock_info.l_len = 1;
1105 rc = fcntl(env->me_lfd, F_SETLK, &lock_info);
1109 lock_info.l_type = F_RDLCK;
1110 rc = fcntl(env->me_lfd, F_SETLK, &lock_info);
1116 size = lseek(env->me_lfd, 0, SEEK_END);
1117 rsize = (env->me_maxreaders-1) * sizeof(MDB_reader) + sizeof(MDB_txninfo);
1118 if (size < rsize && *excl) {
1119 if (ftruncate(env->me_lfd, rsize) != 0) {
1125 size = rsize - sizeof(MDB_txninfo);
1126 env->me_maxreaders = size/sizeof(MDB_reader) + 1;
1128 env->me_txns = mmap(0, rsize, PROT_READ|PROT_WRITE, MAP_SHARED,
1130 if (env->me_txns == MAP_FAILED) {
1135 pthread_mutexattr_t mattr;
1137 pthread_mutexattr_init(&mattr);
1138 pthread_mutexattr_setpshared(&mattr, PTHREAD_PROCESS_SHARED);
1139 pthread_mutex_init(&env->me_txns->mt_mutex, &mattr);
1140 pthread_mutex_init(&env->me_txns->mt_wmutex, &mattr);
1141 env->me_txns->mt_version = MDB_VERSION;
1142 env->me_txns->mt_magic = MDB_MAGIC;
1143 env->me_txns->mt_txnid = 0;
1144 env->me_txns->mt_numreaders = 0;
1147 if (env->me_txns->mt_magic != MDB_MAGIC) {
1148 DPRINTF("lock region has invalid magic");
1151 if (env->me_txns->mt_version != MDB_VERSION) {
1152 DPRINTF("lock region is version %u, expected version %u",
1153 env->me_txns->mt_version, MDB_VERSION);
1156 if (errno != EACCES && errno != EAGAIN) {
1170 mdbenv_open(MDB_env *env, const char *path, unsigned int flags, mode_t mode)
1172 int oflags, rc, len, excl;
1173 char *lpath, *dpath;
1176 lpath = malloc(len + sizeof("/lock.mdb") + len + sizeof("/data.db"));
1179 dpath = lpath + len + sizeof("/lock.mdb");
1180 sprintf(lpath, "%s/lock.mdb", path);
1181 sprintf(dpath, "%s/data.mdb", path);
1183 rc = mdbenv_setup_locks(env, lpath, mode, &excl);
1187 if (F_ISSET(flags, MDB_RDONLY))
1190 oflags = O_RDWR | O_CREAT;
1192 if ((env->me_fd = open(dpath, oflags, mode)) == -1)
1195 if ((rc = mdbenv_open2(env, flags)) != MDB_SUCCESS) {
1199 env->me_path = strdup(path);
1200 DPRINTF("opened dbenv %p", (void *) env);
1201 pthread_key_create(&env->me_txkey, mdbenv_reader_dest);
1203 mdbenv_share_locks(env);
1204 env->me_dbxs = calloc(DBX_CHUNK, sizeof(MDB_dbx));
1206 env->me_dbxs[0].md_db = (MDB_db *)&env->me_meta.mm_psize;
1216 mdbenv_close(MDB_env *env)
1225 munmap(env->me_map, env->me_mapsize);
1229 size_t size = (env->me_maxreaders-1) * sizeof(MDB_reader) + sizeof(MDB_txninfo);
1230 munmap(env->me_txns, size);
1236 /* Search for key within a leaf page, using binary search.
1237 * Returns the smallest entry larger or equal to the key.
1238 * If exactp is non-null, stores whether the found entry was an exact match
1239 * in *exactp (1 or 0).
1240 * If kip is non-null, stores the index of the found entry in *kip.
1241 * If no entry larger or equal to the key is found, returns NULL.
1244 mdb_search_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp, MDB_val *key,
1245 int *exactp, unsigned int *kip)
1253 DPRINTF("searching %u keys in %s page %lu",
1255 IS_LEAF(mp) ? "leaf" : "branch",
1258 assert(NUMKEYS(mp) > 0);
1260 memset(&nodekey, 0, sizeof(nodekey));
1262 low = IS_LEAF(mp) ? 0 : 1;
1263 high = NUMKEYS(mp) - 1;
1264 while (low <= high) {
1265 i = (low + high) >> 1;
1266 node = NODEPTR(mp, i);
1268 nodekey.mv_size = node->mn_ksize;
1269 nodekey.mv_data = NODEKEY(node);
1271 if (txn->mt_dbxs[dbi].md_cmp)
1272 rc = txn->mt_dbxs[dbi].md_cmp(key, &nodekey);
1274 rc = _mdb_cmp(txn, dbi, key, &nodekey);
1277 DPRINTF("found leaf index %u [%.*s], rc = %i",
1278 i, (int)nodekey.mv_size, (char *)nodekey.mv_data, rc);
1280 DPRINTF("found branch index %u [%.*s -> %lu], rc = %i",
1281 i, (int)node->mn_ksize, (char *)NODEKEY(node),
1292 if (rc > 0) { /* Found entry is less than the key. */
1293 i++; /* Skip to get the smallest entry larger than key. */
1294 if (i >= NUMKEYS(mp))
1295 /* There is no entry larger or equal to the key. */
1299 *exactp = (rc == 0);
1300 if (kip) /* Store the key index if requested. */
1303 return NODEPTR(mp, i);
1307 cursor_pop_page(MDB_cursor *cursor)
1311 top = CURSOR_TOP(cursor);
1314 DPRINTF("popped page %lu off cursor %p", top->mp_page->mp_pgno, (void *) cursor);
1320 cursor_push_page(MDB_cursor *cursor, MDB_page *mp)
1324 DPRINTF("pushing page %lu on cursor %p", mp->mp_pgno, (void *) cursor);
1326 if ((ppage = calloc(1, sizeof(*ppage))) == NULL)
1328 ppage->mp_page = mp;
1329 CURSOR_PUSH(cursor, ppage);
1334 mdbenv_get_page(MDB_env *env, pgno_t pgno)
1337 MDB_txn *txn = env->me_txn;
1340 if (txn && !STAILQ_EMPTY(txn->mt_u.dirty_queue)) {
1342 STAILQ_FOREACH(dp, txn->mt_u.dirty_queue, h.md_next) {
1343 if (dp->p.mp_pgno == pgno) {
1351 p = (MDB_page *)(env->me_map + env->me_meta.mm_psize * pgno);
1357 mdb_search_page_root(MDB_txn *txn, MDB_dbi dbi, MDB_val *key,
1358 MDB_cursor *cursor, int modify, MDB_pageparent *mpp)
1360 MDB_page *mp = mpp->mp_page;
1363 if (cursor && cursor_push_page(cursor, mp) == NULL)
1366 while (IS_BRANCH(mp)) {
1370 DPRINTF("branch page %lu has %u keys", mp->mp_pgno, NUMKEYS(mp));
1371 assert(NUMKEYS(mp) > 1);
1372 DPRINTF("found index 0 to page %lu", NODEPGNO(NODEPTR(mp, 0)));
1374 if (key == NULL) /* Initialize cursor to first page. */
1376 else if (key->mv_size > MAXKEYSIZE && key->mv_data == NULL) {
1377 /* cursor to last page */
1381 node = mdb_search_node(txn, dbi, mp, key, &exact, &i);
1383 i = NUMKEYS(mp) - 1;
1391 DPRINTF("following index %u for key %.*s",
1392 i, (int)key->mv_size, (char *)key->mv_data);
1393 assert(i < NUMKEYS(mp));
1394 node = NODEPTR(mp, i);
1397 CURSOR_TOP(cursor)->mp_ki = i;
1399 mpp->mp_parent = mp;
1400 if ((mp = mdbenv_get_page(txn->mt_env, NODEPGNO(node))) == NULL)
1405 if (cursor && cursor_push_page(cursor, mp) == NULL)
1409 MDB_dhead *dh = ((MDB_dhead *)mp)-1;
1410 if ((rc = mdb_touch(txn, mpp)) != 0)
1412 dh = ((MDB_dhead *)mpp->mp_page)-1;
1413 dh->md_parent = mpp->mp_parent;
1414 dh->md_pi = mpp->mp_pi;
1421 DPRINTF("internal error, index points to a %02X page!?",
1426 DPRINTF("found leaf page %lu for key %.*s", mp->mp_pgno,
1427 key ? (int)key->mv_size : 0, key ? (char *)key->mv_data : NULL);
1432 /* Search for the page a given key should be in.
1433 * Stores a pointer to the found page in *mpp.
1434 * If key is NULL, search for the lowest page (used by mdb_cursor_first).
1435 * If cursor is non-null, pushes parent pages on the cursor stack.
1436 * If modify is true, visited pages are updated with new page numbers.
1439 mdb_search_page(MDB_txn *txn, MDB_dbi dbi, MDB_val *key,
1440 MDB_cursor *cursor, int modify, MDB_pageparent *mpp)
1445 /* Choose which root page to start with. If a transaction is given
1446 * use the root page from the transaction, otherwise read the last
1447 * committed root page.
1449 if (F_ISSET(txn->mt_flags, MDB_TXN_ERROR)) {
1450 DPRINTF("transaction has failed, must abort");
1453 root = txn->mt_dbxs[dbi].md_db->md_root;
1455 if (root == P_INVALID) { /* Tree is empty. */
1456 DPRINTF("tree is empty");
1460 if ((mpp->mp_page = mdbenv_get_page(txn->mt_env, root)) == NULL)
1463 DPRINTF("root page has flags 0x%X", mpp->mp_page->mp_flags);
1466 /* For sub-databases, update main root first */
1467 if (dbi && !F_ISSET(txn->mt_dbxs[dbi].md_page->mp_flags, P_DIRTY)) {
1469 rc = mdb_search_page(txn, 0, &txn->mt_dbxs[dbi].md_name,
1474 if (!F_ISSET(mpp->mp_page->mp_flags, P_DIRTY)) {
1475 mpp->mp_parent = NULL;
1477 if ((rc = mdb_touch(txn, mpp)))
1479 txn->mt_dbxs[dbi].md_db->md_root = mpp->mp_page->mp_pgno;
1483 return mdb_search_page_root(txn, dbi, key, cursor, modify, mpp);
1487 mdb_read_data(MDB_env *env, MDB_node *leaf, MDB_val *data)
1489 MDB_page *omp; /* overflow mpage */
1492 if (!F_ISSET(leaf->mn_flags, F_BIGDATA)) {
1493 data->mv_size = leaf->mn_dsize;
1494 data->mv_data = NODEDATA(leaf);
1498 /* Read overflow data.
1500 data->mv_size = leaf->mn_dsize;
1501 memcpy(&pgno, NODEDATA(leaf), sizeof(pgno));
1502 if ((omp = mdbenv_get_page(env, pgno)) == NULL) {
1503 DPRINTF("read overflow page %lu failed", pgno);
1506 data->mv_data = omp;
1512 mdb_get(MDB_txn *txn, MDB_dbi dbi,
1513 MDB_val *key, MDB_val *data)
1521 DPRINTF("===> get key [%.*s]", (int)key->mv_size, (char *)key->mv_data);
1523 if (key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
1527 if ((rc = mdb_search_page(txn, dbi, key, NULL, 0, &mpp)) != MDB_SUCCESS)
1530 leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, NULL);
1532 rc = mdb_read_data(txn->mt_env, leaf, data);
1541 mdb_sibling(MDB_cursor *cursor, int move_right)
1545 MDB_ppage *parent, *top;
1548 top = CURSOR_TOP(cursor);
1549 if ((parent = SLIST_NEXT(top, mp_entry)) == NULL) {
1550 return ENOENT; /* root has no siblings */
1553 DPRINTF("parent page is page %lu, index %u",
1554 parent->mp_page->mp_pgno, parent->mp_ki);
1556 cursor_pop_page(cursor);
1557 if (move_right ? (parent->mp_ki + 1 >= NUMKEYS(parent->mp_page))
1558 : (parent->mp_ki == 0)) {
1559 DPRINTF("no more keys left, moving to %s sibling",
1560 move_right ? "right" : "left");
1561 if ((rc = mdb_sibling(cursor, move_right)) != MDB_SUCCESS)
1563 parent = CURSOR_TOP(cursor);
1569 DPRINTF("just moving to %s index key %u",
1570 move_right ? "right" : "left", parent->mp_ki);
1572 assert(IS_BRANCH(parent->mp_page));
1574 indx = NODEPTR(parent->mp_page, parent->mp_ki);
1575 if ((mp = mdbenv_get_page(cursor->mc_txn->mt_env, indx->mn_pgno)) == NULL)
1578 mp->parent = parent->mp_page;
1579 mp->parent_index = parent->mp_ki;
1582 cursor_push_page(cursor, mp);
1588 mdb_set_key(MDB_node *node, MDB_val *key)
1593 key->mv_size = node->mn_ksize;
1594 key->mv_data = NODEKEY(node);
1600 mdb_cursor_next(MDB_cursor *cursor, MDB_val *key, MDB_val *data)
1606 if (cursor->mc_eof) {
1610 assert(cursor->mc_initialized);
1612 top = CURSOR_TOP(cursor);
1615 DPRINTF("cursor_next: top page is %lu in cursor %p", mp->mp_pgno, (void *) cursor);
1617 if (top->mp_ki + 1 >= NUMKEYS(mp)) {
1618 DPRINTF("=====> move to next sibling page");
1619 if (mdb_sibling(cursor, 1) != MDB_SUCCESS) {
1623 top = CURSOR_TOP(cursor);
1625 DPRINTF("next page is %lu, key index %u", mp->mp_pgno, top->mp_ki);
1629 DPRINTF("==> cursor points to page %lu with %u keys, key index %u",
1630 mp->mp_pgno, NUMKEYS(mp), top->mp_ki);
1632 assert(IS_LEAF(mp));
1633 leaf = NODEPTR(mp, top->mp_ki);
1635 if (data && mdb_read_data(cursor->mc_txn->mt_env, leaf, data) != MDB_SUCCESS)
1638 return mdb_set_key(leaf, key);
1642 mdb_cursor_set(MDB_cursor *cursor, MDB_val *key, MDB_val *data,
1652 assert(key->mv_size > 0);
1654 rc = mdb_search_page(cursor->mc_txn, cursor->mc_dbi, key, cursor, 0, &mpp);
1655 if (rc != MDB_SUCCESS)
1657 assert(IS_LEAF(mpp.mp_page));
1659 top = CURSOR_TOP(cursor);
1660 leaf = mdb_search_node(cursor->mc_txn, cursor->mc_dbi, mpp.mp_page, key, exactp, &top->mp_ki);
1661 if (exactp != NULL && !*exactp) {
1662 /* MDB_CURSOR_EXACT specified and not an exact match. */
1667 DPRINTF("===> inexact leaf not found, goto sibling");
1668 if ((rc = mdb_sibling(cursor, 1)) != MDB_SUCCESS)
1669 return rc; /* no entries matched */
1670 top = CURSOR_TOP(cursor);
1672 mpp.mp_page = top->mp_page;
1673 assert(IS_LEAF(mpp.mp_page));
1674 leaf = NODEPTR(mpp.mp_page, 0);
1677 cursor->mc_initialized = 1;
1680 if (data && (rc = mdb_read_data(cursor->mc_txn->mt_env, leaf, data)) != MDB_SUCCESS)
1683 rc = mdb_set_key(leaf, key);
1684 if (rc == MDB_SUCCESS) {
1685 DPRINTF("==> cursor placed on key %.*s",
1686 (int)key->mv_size, (char *)key->mv_data);
1694 mdb_cursor_first(MDB_cursor *cursor, MDB_val *key, MDB_val *data)
1700 rc = mdb_search_page(cursor->mc_txn, cursor->mc_dbi, NULL, cursor, 0, &mpp);
1701 if (rc != MDB_SUCCESS)
1703 assert(IS_LEAF(mpp.mp_page));
1705 leaf = NODEPTR(mpp.mp_page, 0);
1706 cursor->mc_initialized = 1;
1709 if (data && (rc = mdb_read_data(cursor->mc_txn->mt_env, leaf, data)) != MDB_SUCCESS)
1712 return mdb_set_key(leaf, key);
1716 mdb_cursor_last(MDB_cursor *cursor, MDB_val *key, MDB_val *data)
1723 lkey.mv_size = MAXKEYSIZE+1;
1724 lkey.mv_data = NULL;
1726 rc = mdb_search_page(cursor->mc_txn, cursor->mc_dbi, &lkey, cursor, 0, &mpp);
1727 if (rc != MDB_SUCCESS)
1729 assert(IS_LEAF(mpp.mp_page));
1731 leaf = NODEPTR(mpp.mp_page, NUMKEYS(mpp.mp_page)-1);
1732 cursor->mc_initialized = 1;
1735 if (data && (rc = mdb_read_data(cursor->mc_txn->mt_env, leaf, data)) != MDB_SUCCESS)
1738 return mdb_set_key(leaf, key);
1742 mdb_cursor_get(MDB_cursor *cursor, MDB_val *key, MDB_val *data,
1752 case MDB_CURSOR_EXACT:
1753 while (CURSOR_TOP(cursor) != NULL)
1754 cursor_pop_page(cursor);
1755 if (key == NULL || key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
1757 } else if (op == MDB_CURSOR_EXACT)
1758 rc = mdb_cursor_set(cursor, key, data, &exact);
1760 rc = mdb_cursor_set(cursor, key, data, NULL);
1763 if (!cursor->mc_initialized)
1764 rc = mdb_cursor_first(cursor, key, data);
1766 rc = mdb_cursor_next(cursor, key, data);
1769 while (CURSOR_TOP(cursor) != NULL)
1770 cursor_pop_page(cursor);
1771 rc = mdb_cursor_first(cursor, key, data);
1774 while (CURSOR_TOP(cursor) != NULL)
1775 cursor_pop_page(cursor);
1776 rc = mdb_cursor_last(cursor, key, data);
1779 DPRINTF("unhandled/unimplemented cursor operation %u", op);
1787 /* Allocate a page and initialize it
1790 mdb_new_page(MDB_txn *txn, MDB_dbi dbi, uint32_t flags, int num)
1794 if ((dp = mdb_alloc_page(txn, NULL, 0, num)) == NULL)
1796 DPRINTF("allocated new mpage %lu, page size %u",
1797 dp->p.mp_pgno, txn->mt_env->me_meta.mm_psize);
1798 dp->p.mp_flags = flags | P_DIRTY;
1799 dp->p.mp_lower = PAGEHDRSZ;
1800 dp->p.mp_upper = txn->mt_env->me_meta.mm_psize;
1802 if (IS_BRANCH(&dp->p))
1803 txn->mt_dbxs[dbi].md_db->md_branch_pages++;
1804 else if (IS_LEAF(&dp->p))
1805 txn->mt_dbxs[dbi].md_db->md_leaf_pages++;
1806 else if (IS_OVERFLOW(&dp->p)) {
1807 txn->mt_dbxs[dbi].md_db->md_overflow_pages += num;
1808 dp->p.mp_pages = num;
1815 mdb_leaf_size(MDB_env *env, MDB_val *key, MDB_val *data)
1819 sz = LEAFSIZE(key, data);
1820 if (data->mv_size >= env->me_meta.mm_psize / MDB_MINKEYS) {
1821 /* put on overflow page */
1822 sz -= data->mv_size - sizeof(pgno_t);
1825 return sz + sizeof(indx_t);
1829 mdb_branch_size(MDB_env *env, MDB_val *key)
1834 if (sz >= env->me_meta.mm_psize / MDB_MINKEYS) {
1835 /* put on overflow page */
1836 /* not implemented */
1837 /* sz -= key->size - sizeof(pgno_t); */
1840 return sz + sizeof(indx_t);
1844 mdb_add_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp, indx_t indx,
1845 MDB_val *key, MDB_val *data, pgno_t pgno, uint8_t flags)
1848 size_t node_size = NODESIZE;
1851 MDB_dpage *ofp = NULL; /* overflow page */
1853 assert(mp->mp_upper >= mp->mp_lower);
1855 DPRINTF("add node [%.*s] to %s page %lu at index %i, key size %zu",
1856 key ? (int)key->mv_size : 0, key ? (char *)key->mv_data : NULL,
1857 IS_LEAF(mp) ? "leaf" : "branch",
1858 mp->mp_pgno, indx, key ? key->mv_size : 0);
1861 node_size += key->mv_size;
1865 if (F_ISSET(flags, F_BIGDATA)) {
1866 /* Data already on overflow page. */
1867 node_size += sizeof(pgno_t);
1868 } else if (data->mv_size >= txn->mt_env->me_meta.mm_psize / MDB_MINKEYS) {
1869 int ovpages = OVPAGES(data->mv_size, txn->mt_env->me_meta.mm_psize);
1870 /* Put data on overflow page. */
1871 DPRINTF("data size is %zu, put on overflow page",
1873 node_size += sizeof(pgno_t);
1874 if ((ofp = mdb_new_page(txn, dbi, P_OVERFLOW, ovpages)) == NULL)
1876 DPRINTF("allocated overflow page %lu", ofp->p.mp_pgno);
1879 node_size += data->mv_size;
1883 if (node_size + sizeof(indx_t) > SIZELEFT(mp)) {
1884 DPRINTF("not enough room in page %lu, got %u ptrs",
1885 mp->mp_pgno, NUMKEYS(mp));
1886 DPRINTF("upper - lower = %u - %u = %u", mp->mp_upper, mp->mp_lower,
1887 mp->mp_upper - mp->mp_lower);
1888 DPRINTF("node size = %zu", node_size);
1892 /* Move higher pointers up one slot. */
1893 for (i = NUMKEYS(mp); i > indx; i--)
1894 mp->mp_ptrs[i] = mp->mp_ptrs[i - 1];
1896 /* Adjust free space offsets. */
1897 ofs = mp->mp_upper - node_size;
1898 assert(ofs >= mp->mp_lower + sizeof(indx_t));
1899 mp->mp_ptrs[indx] = ofs;
1901 mp->mp_lower += sizeof(indx_t);
1903 /* Write the node data. */
1904 node = NODEPTR(mp, indx);
1905 node->mn_ksize = (key == NULL) ? 0 : key->mv_size;
1906 node->mn_flags = flags;
1908 node->mn_dsize = data->mv_size;
1910 node->mn_pgno = pgno;
1913 memcpy(NODEKEY(node), key->mv_data, key->mv_size);
1918 if (F_ISSET(flags, F_BIGDATA))
1919 memcpy(node->mn_data + key->mv_size, data->mv_data,
1922 memcpy(node->mn_data + key->mv_size, data->mv_data,
1925 memcpy(node->mn_data + key->mv_size, &ofp->p.mp_pgno,
1927 memcpy(METADATA(&ofp->p), data->mv_data, data->mv_size);
1935 mdb_del_node(MDB_page *mp, indx_t indx)
1938 indx_t i, j, numkeys, ptr;
1942 DPRINTF("delete node %u on %s page %lu", indx,
1943 IS_LEAF(mp) ? "leaf" : "branch", mp->mp_pgno);
1944 assert(indx < NUMKEYS(mp));
1946 node = NODEPTR(mp, indx);
1947 sz = NODESIZE + node->mn_ksize;
1949 if (F_ISSET(node->mn_flags, F_BIGDATA))
1950 sz += sizeof(pgno_t);
1952 sz += NODEDSZ(node);
1955 ptr = mp->mp_ptrs[indx];
1956 numkeys = NUMKEYS(mp);
1957 for (i = j = 0; i < numkeys; i++) {
1959 mp->mp_ptrs[j] = mp->mp_ptrs[i];
1960 if (mp->mp_ptrs[i] < ptr)
1961 mp->mp_ptrs[j] += sz;
1966 base = (char *)mp + mp->mp_upper;
1967 memmove(base + sz, base, ptr - mp->mp_upper);
1969 mp->mp_lower -= sizeof(indx_t);
1974 mdb_cursor_open(MDB_txn *txn, MDB_dbi dbi, MDB_cursor **ret)
1978 if (txn == NULL || ret == NULL)
1981 if ((cursor = calloc(1, sizeof(*cursor))) != NULL) {
1982 SLIST_INIT(&cursor->mc_stack);
1983 cursor->mc_dbi = dbi;
1984 cursor->mc_txn = txn;
1993 mdb_cursor_close(MDB_cursor *cursor)
1995 if (cursor != NULL) {
1996 while (!CURSOR_EMPTY(cursor))
1997 cursor_pop_page(cursor);
1999 /* btree_close(cursor->bt); */
2005 mdb_update_key(MDB_page *mp, indx_t indx, MDB_val *key)
2007 indx_t ptr, i, numkeys;
2013 node = NODEPTR(mp, indx);
2014 ptr = mp->mp_ptrs[indx];
2015 DPRINTF("update key %u (ofs %u) [%.*s] to [%.*s] on page %lu",
2017 (int)node->mn_ksize, (char *)NODEKEY(node),
2018 (int)key->mv_size, (char *)key->mv_data,
2021 delta = key->mv_size - node->mn_ksize;
2023 if (delta > 0 && SIZELEFT(mp) < delta) {
2024 DPRINTF("OUCH! Not enough room, delta = %d", delta);
2028 numkeys = NUMKEYS(mp);
2029 for (i = 0; i < numkeys; i++) {
2030 if (mp->mp_ptrs[i] <= ptr)
2031 mp->mp_ptrs[i] -= delta;
2034 base = (char *)mp + mp->mp_upper;
2035 len = ptr - mp->mp_upper + NODESIZE;
2036 memmove(base - delta, base, len);
2037 mp->mp_upper -= delta;
2039 node = NODEPTR(mp, indx);
2040 node->mn_ksize = key->mv_size;
2043 memcpy(NODEKEY(node), key->mv_data, key->mv_size);
2048 /* Move a node from src to dst.
2051 mdb_move_node(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *src, indx_t srcindx,
2052 MDB_pageparent *dst, indx_t dstindx)
2058 srcnode = NODEPTR(src->mp_page, srcindx);
2059 DPRINTF("moving %s node %u [%.*s] on page %lu to node %u on page %lu",
2060 IS_LEAF(src->mp_page) ? "leaf" : "branch",
2062 (int)srcnode->mn_ksize, (char *)NODEKEY(srcnode),
2063 src->mp_page->mp_pgno,
2064 dstindx, dst->mp_page->mp_pgno);
2066 /* Mark src and dst as dirty. */
2067 if ((rc = mdb_touch(txn, src)) ||
2068 (rc = mdb_touch(txn, dst)))
2071 /* Add the node to the destination page.
2073 key.mv_size = srcnode->mn_ksize;
2074 key.mv_data = NODEKEY(srcnode);
2075 data.mv_size = NODEDSZ(srcnode);
2076 data.mv_data = NODEDATA(srcnode);
2077 rc = mdb_add_node(txn, dbi, dst->mp_page, dstindx, &key, &data, NODEPGNO(srcnode),
2079 if (rc != MDB_SUCCESS)
2082 /* Delete the node from the source page.
2084 mdb_del_node(src->mp_page, srcindx);
2086 /* Update the parent separators.
2088 if (srcindx == 0 && src->mp_pi != 0) {
2089 DPRINTF("update separator for source page %lu to [%.*s]",
2090 src->mp_page->mp_pgno, (int)key.mv_size, (char *)key.mv_data);
2091 if ((rc = mdb_update_key(src->mp_parent, src->mp_pi,
2092 &key)) != MDB_SUCCESS)
2096 if (srcindx == 0 && IS_BRANCH(src->mp_page)) {
2098 nullkey.mv_size = 0;
2099 assert(mdb_update_key(src->mp_page, 0, &nullkey) == MDB_SUCCESS);
2102 if (dstindx == 0 && dst->mp_pi != 0) {
2103 DPRINTF("update separator for destination page %lu to [%.*s]",
2104 dst->mp_page->mp_pgno, (int)key.mv_size, (char *)key.mv_data);
2105 if ((rc = mdb_update_key(dst->mp_parent, dst->mp_pi,
2106 &key)) != MDB_SUCCESS)
2110 if (dstindx == 0 && IS_BRANCH(dst->mp_page)) {
2112 nullkey.mv_size = 0;
2113 assert(mdb_update_key(dst->mp_page, 0, &nullkey) == MDB_SUCCESS);
2120 mdb_merge(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *src, MDB_pageparent *dst)
2129 DPRINTF("merging page %lu and %lu", src->mp_page->mp_pgno, dst->mp_page->mp_pgno);
2131 assert(txn != NULL);
2132 assert(src->mp_parent); /* can't merge root page */
2133 assert(dst->mp_parent);
2135 /* Mark src and dst as dirty. */
2136 if ((rc = mdb_touch(txn, src)) ||
2137 (rc = mdb_touch(txn, dst)))
2140 /* Move all nodes from src to dst.
2142 for (i = 0; i < NUMKEYS(src->mp_page); i++) {
2143 srcnode = NODEPTR(src->mp_page, i);
2145 key.mv_size = srcnode->mn_ksize;
2146 key.mv_data = NODEKEY(srcnode);
2147 data.mv_size = NODEDSZ(srcnode);
2148 data.mv_data = NODEDATA(srcnode);
2149 rc = mdb_add_node(txn, dbi, dst->mp_page, NUMKEYS(dst->mp_page), &key,
2150 &data, NODEPGNO(srcnode), srcnode->mn_flags);
2151 if (rc != MDB_SUCCESS)
2155 DPRINTF("dst page %lu now has %u keys (%.1f%% filled)",
2156 dst->mp_page->mp_pgno, NUMKEYS(dst->mp_page), (float)PAGEFILL(txn->mt_env, dst->mp_page) / 10);
2158 /* Unlink the src page from parent.
2160 mdb_del_node(src->mp_parent, src->mp_pi);
2161 if (src->mp_pi == 0) {
2163 if ((rc = mdb_update_key(src->mp_parent, 0, &key)) != MDB_SUCCESS)
2167 if (IS_LEAF(src->mp_page))
2168 txn->mt_dbxs[dbi].md_db->md_leaf_pages--;
2170 txn->mt_dbxs[dbi].md_db->md_branch_pages--;
2172 mpp.mp_page = src->mp_parent;
2173 dh = (MDB_dhead *)src->mp_parent;
2175 mpp.mp_parent = dh->md_parent;
2176 mpp.mp_pi = dh->md_pi;
2178 return mdb_rebalance(txn, dbi, &mpp);
2181 #define FILL_THRESHOLD 250
2184 mdb_rebalance(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *mpp)
2189 indx_t si = 0, di = 0;
2191 assert(txn != NULL);
2192 assert(mpp != NULL);
2194 DPRINTF("rebalancing %s page %lu (has %u keys, %.1f%% full)",
2195 IS_LEAF(mpp->mp_page) ? "leaf" : "branch",
2196 mpp->mp_page->mp_pgno, NUMKEYS(mpp->mp_page), (float)PAGEFILL(txn->mt_env, mpp->mp_page) / 10);
2198 if (PAGEFILL(txn->mt_env, mpp->mp_page) >= FILL_THRESHOLD) {
2199 DPRINTF("no need to rebalance page %lu, above fill threshold",
2200 mpp->mp_page->mp_pgno);
2204 if (mpp->mp_parent == NULL) {
2205 if (NUMKEYS(mpp->mp_page) == 0) {
2206 DPRINTF("tree is completely empty");
2207 txn->mt_dbxs[dbi].md_db->md_root = P_INVALID;
2208 txn->mt_dbxs[dbi].md_db->md_depth--;
2209 txn->mt_dbxs[dbi].md_db->md_leaf_pages--;
2210 } else if (IS_BRANCH(mpp->mp_page) && NUMKEYS(mpp->mp_page) == 1) {
2211 DPRINTF("collapsing root page!");
2212 txn->mt_dbxs[dbi].md_db->md_root = NODEPGNO(NODEPTR(mpp->mp_page, 0));
2213 if ((root = mdbenv_get_page(txn->mt_env, txn->mt_dbxs[dbi].md_db->md_root)) == NULL)
2215 txn->mt_dbxs[dbi].md_db->md_depth--;
2216 txn->mt_dbxs[dbi].md_db->md_branch_pages--;
2218 DPRINTF("root page doesn't need rebalancing");
2222 /* The parent (branch page) must have at least 2 pointers,
2223 * otherwise the tree is invalid.
2225 assert(NUMKEYS(mpp->mp_parent) > 1);
2227 /* Leaf page fill factor is below the threshold.
2228 * Try to move keys from left or right neighbor, or
2229 * merge with a neighbor page.
2234 if (mpp->mp_pi == 0) {
2235 /* We're the leftmost leaf in our parent.
2237 DPRINTF("reading right neighbor");
2238 node = NODEPTR(mpp->mp_parent, mpp->mp_pi + 1);
2239 if ((npp.mp_page = mdbenv_get_page(txn->mt_env, NODEPGNO(node))) == NULL)
2241 npp.mp_pi = mpp->mp_pi + 1;
2243 di = NUMKEYS(mpp->mp_page);
2245 /* There is at least one neighbor to the left.
2247 DPRINTF("reading left neighbor");
2248 node = NODEPTR(mpp->mp_parent, mpp->mp_pi - 1);
2249 if ((npp.mp_page = mdbenv_get_page(txn->mt_env, NODEPGNO(node))) == NULL)
2251 npp.mp_pi = mpp->mp_pi - 1;
2252 si = NUMKEYS(npp.mp_page) - 1;
2255 npp.mp_parent = mpp->mp_parent;
2257 DPRINTF("found neighbor page %lu (%u keys, %.1f%% full)",
2258 npp.mp_page->mp_pgno, NUMKEYS(npp.mp_page), (float)PAGEFILL(txn->mt_env, npp.mp_page) / 10);
2260 /* If the neighbor page is above threshold and has at least two
2261 * keys, move one key from it.
2263 * Otherwise we should try to merge them.
2265 if (PAGEFILL(txn->mt_env, npp.mp_page) >= FILL_THRESHOLD && NUMKEYS(npp.mp_page) >= 2)
2266 return mdb_move_node(txn, dbi, &npp, si, mpp, di);
2267 else { /* FIXME: if (has_enough_room()) */
2268 if (mpp->mp_pi == 0)
2269 return mdb_merge(txn, dbi, &npp, mpp);
2271 return mdb_merge(txn, dbi, mpp, &npp);
2276 mdb_del(MDB_txn *txn, MDB_dbi dbi,
2277 MDB_val *key, MDB_val *data)
2284 DPRINTF("========> delete key %.*s", (int)key->mv_size, (char *)key->mv_data);
2286 assert(key != NULL);
2288 if (txn == NULL || dbi >= txn->mt_numdbs)
2291 if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
2295 if (key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
2299 if ((rc = mdb_search_page(txn, dbi, key, NULL, 1, &mpp)) != MDB_SUCCESS)
2302 leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, &ki);
2303 if (leaf == NULL || !exact) {
2307 if (data && (rc = mdb_read_data(txn->mt_env, leaf, data)) != MDB_SUCCESS)
2310 mdb_del_node(mpp.mp_page, ki);
2311 /* add overflow pages to free list */
2312 if (F_ISSET(leaf->mn_flags, F_BIGDATA)) {
2316 memcpy(&pg, NODEDATA(leaf), sizeof(pg));
2317 ovpages = OVPAGES(NODEDSZ(leaf), txn->mt_env->me_meta.mm_psize);
2318 for (i=0; i<ovpages; i++) {
2319 mdb_idl_insert(txn->mt_free_pgs, pg);
2323 txn->mt_dbxs[dbi].md_db->md_entries--;
2324 rc = mdb_rebalance(txn, dbi, &mpp);
2325 if (rc != MDB_SUCCESS)
2326 txn->mt_flags |= MDB_TXN_ERROR;
2331 /* Split page <*mpp>, and insert <key,(data|newpgno)> in either left or
2332 * right sibling, at index <*newindxp> (as if unsplit). Updates *mpp and
2333 * *newindxp with the actual values after split, ie if *mpp and *newindxp
2334 * refer to a node in the new right sibling page.
2337 mdb_split(MDB_txn *txn, MDB_dbi dbi, MDB_page **mpp, unsigned int *newindxp,
2338 MDB_val *newkey, MDB_val *newdata, pgno_t newpgno)
2341 int rc = MDB_SUCCESS, ins_new = 0;
2344 unsigned int i, j, split_indx;
2346 MDB_val sepkey, rkey, rdata;
2348 MDB_dpage *mdp, *rdp, *pdp;
2351 assert(txn != NULL);
2353 dh = ((MDB_dhead *)*mpp) - 1;
2354 mdp = (MDB_dpage *)dh;
2355 newindx = *newindxp;
2357 DPRINTF("-----> splitting %s page %lu and adding [%.*s] at index %i",
2358 IS_LEAF(&mdp->p) ? "leaf" : "branch", mdp->p.mp_pgno,
2359 (int)newkey->mv_size, (char *)newkey->mv_data, *newindxp);
2361 if (mdp->h.md_parent == NULL) {
2362 if ((pdp = mdb_new_page(txn, dbi, P_BRANCH, 1)) == NULL)
2365 mdp->h.md_parent = &pdp->p;
2366 txn->mt_dbxs[dbi].md_db->md_root = pdp->p.mp_pgno;
2367 DPRINTF("root split! new root = %lu", pdp->p.mp_pgno);
2368 txn->mt_dbxs[dbi].md_db->md_depth++;
2370 /* Add left (implicit) pointer. */
2371 if (mdb_add_node(txn, dbi, &pdp->p, 0, NULL, NULL,
2372 mdp->p.mp_pgno, 0) != MDB_SUCCESS)
2375 DPRINTF("parent branch page is %lu", mdp->h.md_parent->mp_pgno);
2378 /* Create a right sibling. */
2379 if ((rdp = mdb_new_page(txn, dbi, mdp->p.mp_flags, 1)) == NULL)
2381 rdp->h.md_parent = mdp->h.md_parent;
2382 rdp->h.md_pi = mdp->h.md_pi + 1;
2383 DPRINTF("new right sibling: page %lu", rdp->p.mp_pgno);
2385 /* Move half of the keys to the right sibling. */
2386 if ((copy = malloc(txn->mt_env->me_meta.mm_psize)) == NULL)
2388 memcpy(copy, &mdp->p, txn->mt_env->me_meta.mm_psize);
2389 memset(&mdp->p.mp_ptrs, 0, txn->mt_env->me_meta.mm_psize - PAGEHDRSZ);
2390 mdp->p.mp_lower = PAGEHDRSZ;
2391 mdp->p.mp_upper = txn->mt_env->me_meta.mm_psize;
2393 split_indx = NUMKEYS(copy) / 2 + 1;
2395 /* First find the separating key between the split pages.
2397 memset(&sepkey, 0, sizeof(sepkey));
2398 if (newindx == split_indx) {
2399 sepkey.mv_size = newkey->mv_size;
2400 sepkey.mv_data = newkey->mv_data;
2402 node = NODEPTR(copy, split_indx);
2403 sepkey.mv_size = node->mn_ksize;
2404 sepkey.mv_data = NODEKEY(node);
2407 DPRINTF("separator is [%.*s]", (int)sepkey.mv_size, (char *)sepkey.mv_data);
2409 /* Copy separator key to the parent.
2411 if (SIZELEFT(rdp->h.md_parent) < mdb_branch_size(txn->mt_env, &sepkey)) {
2412 rc = mdb_split(txn, dbi, &rdp->h.md_parent, &rdp->h.md_pi,
2413 &sepkey, NULL, rdp->p.mp_pgno);
2415 /* Right page might now have changed parent.
2416 * Check if left page also changed parent.
2418 if (rdp->h.md_parent != mdp->h.md_parent &&
2419 mdp->h.md_pi >= NUMKEYS(mdp->h.md_parent)) {
2420 mdp->h.md_parent = rdp->h.md_parent;
2421 mdp->h.md_pi = rdp->h.md_pi - 1;
2424 rc = mdb_add_node(txn, dbi, rdp->h.md_parent, rdp->h.md_pi,
2425 &sepkey, NULL, rdp->p.mp_pgno, 0);
2427 if (rc != MDB_SUCCESS) {
2432 for (i = j = 0; i <= NUMKEYS(copy); j++) {
2433 if (i < split_indx) {
2434 /* Re-insert in left sibling. */
2437 /* Insert in right sibling. */
2438 if (i == split_indx)
2439 /* Reset insert index for right sibling. */
2440 j = (i == newindx && ins_new);
2444 if (i == newindx && !ins_new) {
2445 /* Insert the original entry that caused the split. */
2446 rkey.mv_data = newkey->mv_data;
2447 rkey.mv_size = newkey->mv_size;
2448 if (IS_LEAF(&mdp->p)) {
2449 rdata.mv_data = newdata->mv_data;
2450 rdata.mv_size = newdata->mv_size;
2457 /* Update page and index for the new key. */
2460 } else if (i == NUMKEYS(copy)) {
2463 node = NODEPTR(copy, i);
2464 rkey.mv_data = NODEKEY(node);
2465 rkey.mv_size = node->mn_ksize;
2466 if (IS_LEAF(&mdp->p)) {
2467 rdata.mv_data = NODEDATA(node);
2468 rdata.mv_size = node->mn_dsize;
2470 pgno = node->mn_pgno;
2471 flags = node->mn_flags;
2476 if (!IS_LEAF(&mdp->p) && j == 0) {
2477 /* First branch index doesn't need key data. */
2481 rc = mdb_add_node(txn, dbi, &pdp->p, j, &rkey, &rdata, pgno,flags);
2489 mdb_put(MDB_txn *txn, MDB_dbi dbi,
2490 MDB_val *key, MDB_val *data, unsigned int flags)
2492 int rc = MDB_SUCCESS, exact;
2497 assert(key != NULL);
2498 assert(data != NULL);
2503 if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
2507 if (txn->mt_env->me_txn != txn) {
2511 if (key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
2515 DPRINTF("==> put key %.*s, size %zu, data size %zu",
2516 (int)key->mv_size, (char *)key->mv_data, key->mv_size, data->mv_size);
2518 rc = mdb_search_page(txn, dbi, key, NULL, 1, &mpp);
2519 if (rc == MDB_SUCCESS) {
2520 leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, &ki);
2521 if (leaf && exact) {
2522 if (F_ISSET(flags, MDB_NOOVERWRITE)) {
2523 DPRINTF("duplicate key %.*s",
2524 (int)key->mv_size, (char *)key->mv_data);
2527 mdb_del_node(mpp.mp_page, ki);
2529 if (leaf == NULL) { /* append if not found */
2530 ki = NUMKEYS(mpp.mp_page);
2531 DPRINTF("appending key at index %i", ki);
2533 } else if (rc == ENOENT) {
2535 /* new file, just write a root leaf page */
2536 DPRINTF("allocating new root leaf page");
2537 if ((dp = mdb_new_page(txn, dbi, P_LEAF, 1)) == NULL) {
2540 mpp.mp_page = &dp->p;
2541 txn->mt_dbxs[dbi].md_db->md_root = mpp.mp_page->mp_pgno;
2542 txn->mt_dbxs[dbi].md_db->md_depth++;
2548 assert(IS_LEAF(mpp.mp_page));
2549 DPRINTF("there are %u keys, should insert new key at index %i",
2550 NUMKEYS(mpp.mp_page), ki);
2552 if (SIZELEFT(mpp.mp_page) < mdb_leaf_size(txn->mt_env, key, data)) {
2553 rc = mdb_split(txn, dbi, &mpp.mp_page, &ki, key, data, P_INVALID);
2555 /* There is room already in this leaf page. */
2556 rc = mdb_add_node(txn, dbi, mpp.mp_page, ki, key, data, 0, 0);
2559 if (rc != MDB_SUCCESS)
2560 txn->mt_flags |= MDB_TXN_ERROR;
2562 txn->mt_dbxs[dbi].md_db->md_entries++;
2569 mdbenv_get_flags(MDB_env *env, unsigned int *arg)
2574 *arg = env->me_flags;
2579 mdbenv_get_path(MDB_env *env, const char **arg)
2584 *arg = env->me_path;
2589 mdbenv_stat(MDB_env *env, MDB_stat *arg)
2591 if (env == NULL || arg == NULL)
2594 arg->ms_psize = env->me_meta.mm_psize;
2595 arg->ms_depth = env->me_meta.mm_depth;
2596 arg->ms_branch_pages = env->me_meta.mm_branch_pages;
2597 arg->ms_leaf_pages = env->me_meta.mm_leaf_pages;
2598 arg->ms_overflow_pages = env->me_meta.mm_overflow_pages;
2599 arg->ms_entries = env->me_meta.mm_entries;
2604 int mdb_open(MDB_txn *txn, const char *name, unsigned int flags, MDB_dbi *dbi)
2617 /* Is the DB already open? */
2619 for (i=0; i<txn->mt_numdbs; i++) {
2620 if (len == txn->mt_dbxs[i].md_name.mv_size &&
2621 !strcmp(name, txn->mt_dbxs[i].md_name.mv_data)) {
2627 /* Find the DB info */
2629 key.mv_data = (void *)name;
2630 rc = mdb_get(txn, 0, &key, &data);
2632 /* Create if requested */
2633 if (rc == ENOENT && (flags & MDB_CREATE)) {
2635 data.mv_size = sizeof(MDB_db);
2636 data.mv_data = &dummy;
2637 memset(&dummy, 0, sizeof(dummy));
2638 dummy.md_root = P_INVALID;
2639 dummy.md_flags = flags & 0xffff;
2640 rc = mdb_put(txn, 0, &key, &data, 0);
2641 if (rc == MDB_SUCCESS)
2642 rc = mdb_get(txn, 0, &key, &data);
2645 /* OK, got info, add to table */
2646 if (rc == MDB_SUCCESS) {
2647 /* Is there a free slot? */
2648 if ((txn->mt_numdbs & (DBX_CHUNK-1)) == 0) {
2651 i = txn->mt_numdbs + DBX_CHUNK;
2652 p1 = realloc(txn->mt_dbxs, i * sizeof(MDB_dbx));
2657 txn->mt_dbxs[txn->mt_numdbs].md_name.mv_data = strdup(name);
2658 txn->mt_dbxs[txn->mt_numdbs].md_name.mv_size = len;
2659 txn->mt_dbxs[txn->mt_numdbs].md_cmp = NULL;
2660 txn->mt_dbxs[txn->mt_numdbs].md_dcmp = NULL;
2661 txn->mt_dbxs[txn->mt_numdbs].md_rel = NULL;
2662 txn->mt_dbxs[txn->mt_numdbs].md_db = data.mv_data;
2663 *dbi = txn->mt_numdbs;
2670 int mdb_stat(MDB_txn *txn, MDB_dbi dbi, MDB_stat *arg)
2672 if (txn == NULL || arg == NULL)
2675 arg->ms_psize = txn->mt_env->me_meta.mm_psize;
2676 arg->ms_depth = txn->mt_dbxs[dbi].md_db->md_depth;
2677 arg->ms_branch_pages = txn->mt_dbxs[dbi].md_db->md_branch_pages;
2678 arg->ms_leaf_pages = txn->mt_dbxs[dbi].md_db->md_leaf_pages;
2679 arg->ms_overflow_pages = txn->mt_dbxs[dbi].md_db->md_overflow_pages;
2680 arg->ms_entries = txn->mt_dbxs[dbi].md_db->md_entries;
2685 void mdb_close(MDB_txn *txn, MDB_dbi dbi)
2687 if (dbi >= txn->mt_numdbs)
2689 free(txn->mt_dbxs[dbi].md_name.mv_data);
2690 txn->mt_dbxs[dbi].md_name.mv_data = NULL;
2691 txn->mt_dbxs[dbi].md_name.mv_size = 0;