X-Git-Url: https://git.sur5r.net/?a=blobdiff_plain;f=libraries%2Flibmdb%2Fmdb.c;h=b814169b7100b28507ee562573019ac60ab6bb55;hb=1f8c3369b90f54d7fd58c5e327343ca1c708e813;hp=5785227f64e4acb633ea7f547d0d0a501096f5a1;hpb=e91dae3ed52ea2ced5d747b3ab95534c957c8e6d;p=openldap diff --git a/libraries/libmdb/mdb.c b/libraries/libmdb/mdb.c index 5785227f64..b814169b71 100644 --- a/libraries/libmdb/mdb.c +++ b/libraries/libmdb/mdb.c @@ -113,7 +113,6 @@ #endif #endif -#define ULONG unsigned long typedef ULONG pgno_t; #ifndef DEBUG @@ -351,29 +350,32 @@ typedef struct MDB_ppage { /* ordered list of pages */ struct MDB_xcursor; struct MDB_cursor { + struct MDB_xcursor *mc_xcursor; MDB_txn *mc_txn; - MDB_ppage mc_stack[CURSOR_STACK]; /* stack of parent pages */ - unsigned int mc_snum; /* number of pushed pages */ MDB_dbi mc_dbi; - short mc_initialized; /* 1 if initialized */ - short mc_eof; /* 1 if end is reached */ - struct MDB_xcursor *mc_xcursor; + unsigned int mc_snum; /* number of pushed pages */ + unsigned int mc_flags; +#define C_INITIALIZED 0x01 +#define C_EOF 0x02 +#define C_XDIRTY 0x04 + MDB_ppage mc_stack[CURSOR_STACK]; /* stack of parent pages */ }; #define METADATA(p) ((void *)((char *)(p) + PAGEHDRSZ)) +/* We guarantee 2-byte alignment for nodes */ typedef struct MDB_node { -#define mn_pgno mn_p.np_pgno -#define mn_dsize mn_p.np_dsize - union { - pgno_t np_pgno; /* child page number */ - uint32_t np_dsize; /* leaf data size */ - } mn_p; - unsigned int mn_flags:4; - unsigned int mn_ksize:12; /* key size */ + /* lo and hi are used for data size on leaf nodes and for + * child pgno on branch nodes. On 64 bit platforms, flags + * is also used for pgno. (branch nodes ignore flags) + */ + unsigned short mn_lo; + unsigned short mn_hi; + unsigned short mn_flags; #define F_BIGDATA 0x01 /* data put on overflow page */ #define F_SUBDATA 0x02 /* data is a sub-database */ #define F_DUPDATA 0x04 /* data has duplicates */ + unsigned short mn_ksize; /* key size */ char mn_data[1]; } MDB_node; @@ -439,7 +441,7 @@ struct MDB_env { pthread_key_t me_txkey; /* thread-key for readers */ MDB_dpage *me_dpages; pgno_t me_free_pgs[MDB_IDL_UM_SIZE]; - ID2 me_dirty_list[MDB_IDL_DB_SIZE]; + ID2 me_dirty_list[MDB_IDL_UM_SIZE]; LAZY_RWLOCK_DEF(me_dblock); #ifdef _WIN32 HANDLE me_rmutex; /* Windows mutexes don't reside in shared mem */ @@ -454,8 +456,19 @@ struct MDB_env { #define NODEPTR(p, i) ((MDB_node *)((char *)(p) + (p)->mp_ptrs[i])) #define NODEKEY(node) (void *)((node)->mn_data) #define NODEDATA(node) (void *)((char *)(node)->mn_data + (node)->mn_ksize) -#define NODEPGNO(node) ((node)->mn_pgno) -#define NODEDSZ(node) ((node)->mn_dsize) +#if LONG_MAX == 0x7fffffff +#define NODEPGNO(node) ((node)->mn_lo | ((node)->mn_hi << 16)) +#define SETPGNO(node,pgno) do { \ + (node)->mn_lo = (pgno) & 0xffff; (node)->mn_hi = (pgno) >> 16;} while(0) +#else +#define NODEPGNO(node) ((node)->mn_lo | ((node)->mn_hi << 16) | ((unsigned long)(node)->mn_flags << 32)) +#define SETPGNO(node,pgno) do { \ + (node)->mn_lo = (pgno) & 0xffff; (node)->mn_hi = (pgno) >> 16; \ + (node)->mn_flags = (pgno) >> 32; } while(0) +#endif +#define NODEDSZ(node) ((node)->mn_lo | ((unsigned)(node)->mn_hi << 16)) +#define SETDSZ(node,size) do { \ + (node)->mn_lo = (size) & 0xffff; (node)->mn_hi = (size) >> 16;} while(0) #define NODEKSZ(node) ((node)->mn_ksize) #define LEAF2KEY(p, i, ks) ((char *)(p) + PAGEHDRSZ + ((i)*(ks))) @@ -483,53 +496,50 @@ static int mdb_add_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp, indx_t indx, MDB_val *key, MDB_val *data, pgno_t pgno, uint8_t flags); static void mdb_del_node(MDB_page *mp, indx_t indx, int ksize); -static int mdb_del0(MDB_txn *txn, MDB_dbi dbi, unsigned int ki, +static int mdb_del0(MDB_cursor *mc, unsigned int ki, MDB_pageparent *mpp, MDB_node *leaf); -static int mdb_put0(MDB_txn *txn, MDB_dbi dbi, - MDB_val *key, MDB_val *data, unsigned int flags); static int mdb_read_data(MDB_txn *txn, MDB_node *leaf, MDB_val *data); -static int mdb_rebalance(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *mp); -static int mdb_update_key(MDB_page *mp, indx_t indx, MDB_val *key); -static int mdb_move_node(MDB_txn *txn, MDB_dbi dbi, +static int mdb_rebalance(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *mp); +static int mdb_update_key(MDB_page *mp, indx_t indx, MDB_val *key); +static int mdb_move_node(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *src, indx_t srcindx, MDB_pageparent *dst, indx_t dstindx); -static int mdb_merge(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *src, +static int mdb_merge(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *src, MDB_pageparent *dst); -static int mdb_split(MDB_txn *txn, MDB_dbi dbi, MDB_page **mpp, +static int mdb_split(MDB_txn *txn, MDB_dbi dbi, MDB_page **mpp, unsigned int *newindxp, MDB_val *newkey, MDB_val *newdata, pgno_t newpgno); static MDB_dpage *mdb_new_page(MDB_txn *txn, MDB_dbi dbi, uint32_t flags, int num); -static void cursor_pop_page(MDB_cursor *cursor); +static void cursor_pop_page(MDB_cursor *cursor); static MDB_ppage *cursor_push_page(MDB_cursor *cursor, MDB_page *mp); -static int mdb_sibling(MDB_cursor *cursor, int move_right); -static int mdb_cursor_next(MDB_cursor *cursor, +static int mdb_sibling(MDB_cursor *cursor, int move_right); +static int mdb_cursor_next(MDB_cursor *cursor, MDB_val *key, MDB_val *data, MDB_cursor_op op); -static int mdb_cursor_prev(MDB_cursor *cursor, +static int mdb_cursor_prev(MDB_cursor *cursor, MDB_val *key, MDB_val *data, MDB_cursor_op op); -static int mdb_cursor_set(MDB_cursor *cursor, +static int mdb_cursor_set(MDB_cursor *cursor, MDB_val *key, MDB_val *data, MDB_cursor_op op, int *exactp); -static int mdb_cursor_first(MDB_cursor *cursor, +static int mdb_cursor_first(MDB_cursor *cursor, MDB_val *key, MDB_val *data); -static int mdb_cursor_last(MDB_cursor *cursor, +static int mdb_cursor_last(MDB_cursor *cursor, MDB_val *key, MDB_val *data); -static void mdb_xcursor_init0(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx); -static void mdb_xcursor_init1(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx, - MDB_page *mp, MDB_node *node); -static void mdb_xcursor_fini(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx); +static void mdb_xcursor_init0(MDB_cursor *mc); +static void mdb_xcursor_init1(MDB_cursor *mc, MDB_page *mp, MDB_node *node); +static void mdb_xcursor_init2(MDB_cursor *mc); +static void mdb_xcursor_fini(MDB_cursor *mc); -static size_t mdb_leaf_size(MDB_env *env, MDB_val *key, +static size_t mdb_leaf_size(MDB_env *env, MDB_val *key, MDB_val *data); -static size_t mdb_branch_size(MDB_env *env, MDB_val *key); +static size_t mdb_branch_size(MDB_env *env, MDB_val *key); + +static void mdb_default_cmp(MDB_txn *txn, MDB_dbi dbi); -static int memncmp(const void *s1, size_t n1, - const void *s2, size_t n2); -static int memnrcmp(const void *s1, size_t n1, - const void *s2, size_t n2); +static MDB_cmp_func memncmp, memnrcmp, intcmp, cintcmp; #ifdef _WIN32 static SECURITY_DESCRIPTOR mdb_null_sd; @@ -537,39 +547,6 @@ static SECURITY_ATTRIBUTES mdb_all_sa; static int mdb_sec_inited; #endif -static int -memncmp(const void *s1, size_t n1, const void *s2, size_t n2) -{ - int diff, len_diff = -1; - - if (n1 >= n2) { - len_diff = (n1 > n2); - n1 = n2; - } - diff = memcmp(s1, s2, n1); - return diff ? diff : len_diff; -} - -static int -memnrcmp(const void *s1, size_t n1, const void *s2, size_t n2) -{ - const unsigned char *p1, *p2, *p1_lim; - - if (n2 == 0) - return n1 != 0; - if (n1 == 0) - return -1; - - p1 = (const unsigned char *)s1 + n1 - 1; - p2 = (const unsigned char *)s2 + n2 - 1; - - for (p1_lim = (n1 <= n2 ? s1 : s2); *p1 == *p2; p1--, p2--) { - if (p1 == p1_lim) - return (p1 != s1) ? (p1 != p2) : (p2 != s2) ? -1 : 0; - } - return *p1 - *p2; -} - char * mdb_version(int *maj, int *min, int *pat) { @@ -609,8 +586,12 @@ mdb_dkey(MDB_val *key, char *buf) unsigned int i; if (key->mv_size > MAXKEYSIZE) return "MAXKEYSIZE"; +#if 1 for (i=0; imv_size; i++) ptr += sprintf(ptr, "%02x", *c++); +#else + sprintf(buf, "%.*s", key->mv_size, key->mv_data); +#endif return buf; } #endif @@ -618,17 +599,7 @@ mdb_dkey(MDB_val *key, char *buf) int mdb_cmp(MDB_txn *txn, MDB_dbi dbi, const MDB_val *a, const MDB_val *b) { - if (txn->mt_dbxs[dbi].md_cmp) - return txn->mt_dbxs[dbi].md_cmp(a, b); - - if (txn->mt_dbs[dbi].md_flags & (MDB_REVERSEKEY -#if __BYTE_ORDER == __LITTLE_ENDIAN - |MDB_INTEGERKEY -#endif - )) - return memnrcmp(a->mv_data, a->mv_size, b->mv_data, b->mv_size); - else - return memncmp((char *)a->mv_data, a->mv_size, b->mv_data, b->mv_size); + return txn->mt_dbxs[dbi].md_cmp(a, b); } int @@ -636,15 +607,8 @@ mdb_dcmp(MDB_txn *txn, MDB_dbi dbi, const MDB_val *a, const MDB_val *b) { if (txn->mt_dbxs[dbi].md_dcmp) return txn->mt_dbxs[dbi].md_dcmp(a, b); - - if (txn->mt_dbs[dbi].md_flags & (0 -#if __BYTE_ORDER == __LITTLE_ENDIAN - |MDB_INTEGERDUP -#endif - )) - return memnrcmp(a->mv_data, a->mv_size, b->mv_data, b->mv_size); else - return memncmp((char *)a->mv_data, a->mv_size, b->mv_data, b->mv_size); + return EINVAL; /* too bad you can't distinguish this from a valid result */ } /* Allocate new page(s) for writing */ @@ -660,13 +624,18 @@ mdb_alloc_page(MDB_txn *txn, MDB_dbi dbi, MDB_page *parent, unsigned int parent_ if (!txn->mt_env->me_pghead && dbi != FREE_DBI && txn->mt_dbs[FREE_DBI].md_root != P_INVALID) { /* See if there's anything in the free DB */ + MDB_cursor mc; MDB_pageparent mpp; MDB_node *leaf; ULONG *kptr, oldest; mpp.mp_parent = NULL; mpp.mp_pi = 0; - mdb_search_page(txn, FREE_DBI, NULL, NULL, 0, &mpp); + mc.mc_txn = txn; + mc.mc_dbi = FREE_DBI; + mc.mc_snum = 0; + mc.mc_flags = 0; + mdb_search_page(txn, FREE_DBI, NULL, &mc, 0, &mpp); leaf = NODEPTR(mpp.mp_page, 0); kptr = (ULONG *)NODEKEY(leaf); @@ -684,6 +653,7 @@ mdb_alloc_page(MDB_txn *txn, MDB_dbi dbi, MDB_page *parent, unsigned int parent_ /* It's usable, grab it. */ MDB_oldpages *mop; + MDB_ppage *top; MDB_val data; pgno_t *idl; @@ -706,11 +676,10 @@ mdb_alloc_page(MDB_txn *txn, MDB_dbi dbi, MDB_page *parent, unsigned int parent_ } #endif /* drop this IDL from the DB */ - mpp.mp_parent = NULL; - mpp.mp_pi = 0; - mdb_search_page(txn, FREE_DBI, NULL, NULL, 1, &mpp); - leaf = NODEPTR(mpp.mp_page, 0); - mdb_del0(txn, FREE_DBI, 0, &mpp, leaf); + top = CURSOR_TOP(&mc); + top->mp_ki = 0; + mc.mc_flags = C_INITIALIZED; + mdb_cursor_del(&mc, 0); } } if (txn->mt_env->me_pghead) { @@ -783,7 +752,7 @@ mdb_touch(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *pp) return ENOMEM; DPRINTF("touched db %u page %lu -> %lu", dbi, mp->mp_pgno, dp->p.mp_pgno); assert(mp->mp_pgno != dp->p.mp_pgno); - mdb_midl_insert(txn->mt_free_pgs, mp->mp_pgno); + mdb_midl_append(txn->mt_free_pgs, mp->mp_pgno); pgno = dp->p.mp_pgno; memcpy(&dp->p, mp, txn->mt_env->me_psize); mp = &dp->p; @@ -792,7 +761,7 @@ mdb_touch(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *pp) /* Update the page number to new touched page. */ if (pp->mp_parent != NULL) - NODEPGNO(NODEPTR(pp->mp_parent, pp->mp_pi)) = mp->mp_pgno; + SETPGNO(NODEPTR(pp->mp_parent, pp->mp_pi), mp->mp_pgno); pp->mp_page = mp; } return 0; @@ -998,6 +967,7 @@ mdb_txn_commit(MDB_txn *txn) MDB_dpage *dp; MDB_env *env; pgno_t next; + MDB_cursor mc; assert(txn != NULL); assert(txn->mt_env != NULL); @@ -1027,6 +997,10 @@ mdb_txn_commit(MDB_txn *txn) DPRINTF("committing txn %lu %p on mdbenv %p, root page %lu", txn->mt_txnid, txn, (void *)env, txn->mt_dbs[MAIN_DBI].md_root); + mc.mc_txn = txn; + mc.mc_dbi = FREE_DBI; + mc.mc_flags = 0; + /* should only be one record now */ if (env->me_pghead) { MDB_pageparent mpp; @@ -1034,7 +1008,8 @@ mdb_txn_commit(MDB_txn *txn) /* make sure first page of freeDB is touched and on freelist */ mpp.mp_parent = NULL; mpp.mp_pi = 0; - mdb_search_page(txn, FREE_DBI, NULL, NULL, 1, &mpp); + mc.mc_snum = 0; + mdb_search_page(txn, FREE_DBI, NULL, &mc, 1, &mpp); } /* save to free list */ if (!MDB_IDL_IS_ZERO(txn->mt_free_pgs)) { @@ -1047,8 +1022,10 @@ mdb_txn_commit(MDB_txn *txn) key.mv_data = NULL; mpp.mp_parent = NULL; mpp.mp_pi = 0; - mdb_search_page(txn, FREE_DBI, &key, NULL, 1, &mpp); + mc.mc_snum = 0; + mdb_search_page(txn, FREE_DBI, &key, &mc, 1, &mpp); + mdb_midl_sort(txn->mt_free_pgs); #if DEBUG > 1 { unsigned int i; @@ -1071,7 +1048,7 @@ mdb_txn_commit(MDB_txn *txn) do { i = txn->mt_free_pgs[0]; data.mv_size = MDB_IDL_SIZEOF(txn->mt_free_pgs); - rc = mdb_put0(txn, FREE_DBI, &key, &data, 0); + rc = mdb_cursor_put(&mc, &key, &data, 0); if (rc) { mdb_txn_abort(txn); return rc; @@ -1088,7 +1065,7 @@ mdb_txn_commit(MDB_txn *txn) key.mv_data = (char *)&mop->mo_txnid; data.mv_size = MDB_IDL_SIZEOF(mop->mo_pages); data.mv_data = mop->mo_pages; - mdb_put0(txn, FREE_DBI, &key, &data, 0); + mdb_cursor_put(&mc, &key, &data, 0); free(env->me_pghead); env->me_pghead = NULL; } @@ -1100,10 +1077,12 @@ mdb_txn_commit(MDB_txn *txn) MDB_val data; data.mv_size = sizeof(MDB_db); + mc.mc_dbi = MAIN_DBI; + mc.mc_flags = 0; for (i = 2; i < txn->mt_numdbs; i++) { if (txn->mt_dbxs[i].md_dirty) { data.mv_data = &txn->mt_dbs[i]; - mdb_put0(txn, MAIN_DBI, &txn->mt_dbxs[i].md_name, &data, 0); + mdb_cursor_put(&mc, &txn->mt_dbxs[i].md_name, &data, 0); } } } @@ -1958,6 +1937,88 @@ mdb_env_close(MDB_env *env) free(env); } +/* only for aligned ints */ +static int +intcmp(const MDB_val *a, const MDB_val *b) +{ + if (a->mv_size == sizeof(long)) + { + unsigned long *la, *lb; + la = a->mv_data; + lb = b->mv_data; + return *la - *lb; + } else { + unsigned int *ia, *ib; + ia = a->mv_data; + ib = b->mv_data; + return *ia - *ib; + } +} + +/* ints must always be the same size */ +static int +cintcmp(const MDB_val *a, const MDB_val *b) +{ +#if __BYTE_ORDER == __LITTLE_ENDIAN + unsigned char *u, *c; + int x; + + u = a->mv_data + a->mv_size; + c = b->mv_data + a->mv_size; + while(u > (unsigned char *)a->mv_data) { + x = *--u - *--c; + if (x) break; + } + return x; +#else + return memcmp(a->mv_data, b->mv_data, a->mv_size); +#endif +} + +static int +memncmp(const MDB_val *a, const MDB_val *b) +{ + int diff, len_diff; + unsigned int len; + + len = a->mv_size; + len_diff = a->mv_size - b->mv_size; + if (len_diff > 0) + len = b->mv_size; + diff = memcmp(a->mv_data, b->mv_data, len); + return diff ? diff : len_diff; +} + +static int +memnrcmp(const MDB_val *a, const MDB_val *b) +{ + const unsigned char *p1, *p2, *p1_lim; + int diff, len_diff; + + if (b->mv_size == 0) + return a->mv_size != 0; + if (a->mv_size == 0) + return -1; + + p1 = (const unsigned char *)a->mv_data + a->mv_size - 1; + p2 = (const unsigned char *)b->mv_data + b->mv_size - 1; + + len_diff = a->mv_size - b->mv_size; + if (len_diff < 0) + p1_lim = p1 - a->mv_size; + else + p1_lim = p1 - b->mv_size; + + while (p1 > p1_lim) { + diff = *p1 - *p2; + if (diff) + return diff; + p1--; + p2--; + } + return len_diff; +} + /* Search for key within a leaf page, using binary search. * Returns the smallest entry larger or equal to the key. * If exactp is non-null, stores whether the found entry was an exact match @@ -1969,29 +2030,33 @@ static MDB_node * mdb_search_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp, MDB_val *key, int *exactp, unsigned int *kip) { - unsigned int i = 0; + unsigned int i = 0, nkeys; int low, high; int rc = 0; MDB_node *node = NULL; MDB_val nodekey; + MDB_cmp_func *cmp; DKBUF; + nkeys = NUMKEYS(mp); + DPRINTF("searching %u keys in %s page %lu", - NUMKEYS(mp), - IS_LEAF(mp) ? "leaf" : "branch", + nkeys, IS_LEAF(mp) ? "leaf" : "branch", mp->mp_pgno); - assert(NUMKEYS(mp) > 0); - - memset(&nodekey, 0, sizeof(nodekey)); + assert(nkeys > 0); low = IS_LEAF(mp) ? 0 : 1; - high = NUMKEYS(mp) - 1; + high = nkeys - 1; + cmp = txn->mt_dbxs[dbi].md_cmp; + if (IS_LEAF2(mp)) { + nodekey.mv_size = txn->mt_dbs[dbi].md_pad; + node = NODEPTR(mp, 0); /* fake */ + } while (low <= high) { i = (low + high) >> 1; if (IS_LEAF2(mp)) { - nodekey.mv_size = txn->mt_dbs[dbi].md_pad; nodekey.mv_data = LEAF2KEY(mp, i, nodekey.mv_size); } else { node = NODEPTR(mp, i); @@ -2000,14 +2065,16 @@ mdb_search_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp, MDB_val *key, nodekey.mv_data = NODEKEY(node); } - rc = mdb_cmp(txn, dbi, key, &nodekey); + rc = cmp(key, &nodekey); +#if DEBUG if (IS_LEAF(mp)) DPRINTF("found leaf index %u [%s], rc = %i", i, DKEY(&nodekey), rc); else DPRINTF("found branch index %u [%s -> %lu], rc = %i", i, DKEY(&nodekey), NODEPGNO(node), rc); +#endif if (rc == 0) break; @@ -2019,17 +2086,19 @@ mdb_search_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp, MDB_val *key, if (rc > 0) { /* Found entry is less than the key. */ i++; /* Skip to get the smallest entry larger than key. */ - if (i >= NUMKEYS(mp)) - /* There is no entry larger or equal to the key. */ - return NULL; + if (!IS_LEAF2(mp)) + node = NODEPTR(mp, i); } if (exactp) *exactp = (rc == 0); if (kip) /* Store the key index if requested. */ *kip = i; + if (i >= nkeys) + /* There is no entry larger or equal to the key. */ + return NULL; /* nodeptr is fake for LEAF2 */ - return IS_LEAF2(mp) ? NODEPTR(mp, 0) : NODEPTR(mp, i); + return node; } static void @@ -2142,7 +2211,7 @@ mdb_search_page_root(MDB_txn *txn, MDB_dbi dbi, MDB_val *key, return ENOMEM; if (modify) { - MDB_dhead *dh = ((MDB_dhead *)mp)-1; + MDB_dhead *dh; if ((rc = mdb_touch(txn, dbi, mpp)) != 0) return rc; dh = ((MDB_dhead *)mpp->mp_page)-1; @@ -2229,14 +2298,14 @@ mdb_read_data(MDB_txn *txn, MDB_node *leaf, MDB_val *data) int rc; if (!F_ISSET(leaf->mn_flags, F_BIGDATA)) { - data->mv_size = leaf->mn_dsize; + data->mv_size = NODEDSZ(leaf); data->mv_data = NODEDATA(leaf); return MDB_SUCCESS; } /* Read overflow data. */ - data->mv_size = leaf->mn_dsize; + data->mv_size = NODEDSZ(leaf); memcpy(&pgno, NODEDATA(leaf), sizeof(pgno)); if ((rc = mdb_get_page(txn, pgno, &omp))) { DPRINTF("read overflow page %lu failed", pgno); @@ -2251,9 +2320,9 @@ int mdb_get(MDB_txn *txn, MDB_dbi dbi, MDB_val *key, MDB_val *data) { - int rc, exact; - MDB_node *leaf; - MDB_pageparent mpp; + MDB_cursor mc; + MDB_xcursor mx; + int exact = 0; DKBUF; assert(key); @@ -2267,36 +2336,16 @@ mdb_get(MDB_txn *txn, MDB_dbi dbi, return EINVAL; } - if ((rc = mdb_search_page(txn, dbi, key, NULL, 0, &mpp)) != MDB_SUCCESS) - return rc; - - leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, NULL); - if (leaf && exact) { - /* Return first duplicate data item */ - if (F_ISSET(leaf->mn_flags, F_DUPDATA)) { - MDB_xcursor mx; - - mdb_xcursor_init0(txn, dbi, &mx); - mdb_xcursor_init1(txn, dbi, &mx, mpp.mp_page, leaf); - rc = mdb_search_page(&mx.mx_txn, mx.mx_cursor.mc_dbi, NULL, NULL, 0, &mpp); - if (rc != MDB_SUCCESS) - return rc; - if (IS_LEAF2(mpp.mp_page)) { - data->mv_size = txn->mt_dbs[dbi].md_pad; - data->mv_data = LEAF2KEY(mpp.mp_page, 0, data->mv_size); - } else { - leaf = NODEPTR(mpp.mp_page, 0); - data->mv_size = NODEKSZ(leaf); - data->mv_data = NODEKEY(leaf); - } - } else { - rc = mdb_read_data(txn, leaf, data); - } + mc.mc_txn = txn; + mc.mc_dbi = dbi; + mc.mc_flags = 0; + if (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT) { + mc.mc_xcursor = &mx; + mdb_xcursor_init0(&mc); } else { - rc = MDB_NOTFOUND; + mc.mc_xcursor = NULL; } - - return rc; + return mdb_cursor_set(&mc, key, data, MDB_SET, &exact); } static int @@ -2354,11 +2403,11 @@ mdb_cursor_next(MDB_cursor *cursor, MDB_val *key, MDB_val *data, MDB_cursor_op o MDB_node *leaf; int rc; - if (cursor->mc_eof) { + if (cursor->mc_flags & C_EOF) { return MDB_NOTFOUND; } - assert(cursor->mc_initialized); + assert(cursor->mc_flags & C_INITIALIZED); top = CURSOR_TOP(cursor); mp = top->mp_page; @@ -2372,7 +2421,7 @@ mdb_cursor_next(MDB_cursor *cursor, MDB_val *key, MDB_val *data, MDB_cursor_op o return rc; } } else { - cursor->mc_xcursor->mx_cursor.mc_initialized = 0; + cursor->mc_xcursor->mx_cursor.mc_flags = 0; if (op == MDB_NEXT_DUP) return MDB_NOTFOUND; } @@ -2383,7 +2432,7 @@ mdb_cursor_next(MDB_cursor *cursor, MDB_val *key, MDB_val *data, MDB_cursor_op o if (top->mp_ki + 1 >= NUMKEYS(mp)) { DPUTS("=====> move to next sibling page"); if (mdb_sibling(cursor, 1) != MDB_SUCCESS) { - cursor->mc_eof = 1; + cursor->mc_flags |= C_EOF; return MDB_NOTFOUND; } top = CURSOR_TOP(cursor); @@ -2405,7 +2454,7 @@ mdb_cursor_next(MDB_cursor *cursor, MDB_val *key, MDB_val *data, MDB_cursor_op o leaf = NODEPTR(mp, top->mp_ki); if (F_ISSET(leaf->mn_flags, F_DUPDATA)) { - mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, mp, leaf); + mdb_xcursor_init1(cursor, mp, leaf); } if (data) { if ((rc = mdb_read_data(cursor->mc_txn, leaf, data) != MDB_SUCCESS)) @@ -2430,7 +2479,7 @@ mdb_cursor_prev(MDB_cursor *cursor, MDB_val *key, MDB_val *data, MDB_cursor_op o MDB_node *leaf; int rc; - assert(cursor->mc_initialized); + assert(cursor->mc_flags & C_INITIALIZED); top = CURSOR_TOP(cursor); mp = top->mp_page; @@ -2443,7 +2492,7 @@ mdb_cursor_prev(MDB_cursor *cursor, MDB_val *key, MDB_val *data, MDB_cursor_op o if (op != MDB_PREV || rc == MDB_SUCCESS) return rc; } else { - cursor->mc_xcursor->mx_cursor.mc_initialized = 0; + cursor->mc_xcursor->mx_cursor.mc_flags = 0; if (op == MDB_PREV_DUP) return MDB_NOTFOUND; } @@ -2455,7 +2504,7 @@ mdb_cursor_prev(MDB_cursor *cursor, MDB_val *key, MDB_val *data, MDB_cursor_op o if (top->mp_ki == 0) { DPUTS("=====> move to prev sibling page"); if (mdb_sibling(cursor, 0) != MDB_SUCCESS) { - cursor->mc_initialized = 0; + cursor->mc_flags &= ~C_INITIALIZED; return MDB_NOTFOUND; } top = CURSOR_TOP(cursor); @@ -2465,7 +2514,7 @@ mdb_cursor_prev(MDB_cursor *cursor, MDB_val *key, MDB_val *data, MDB_cursor_op o } else top->mp_ki--; - cursor->mc_eof = 0; + cursor->mc_flags &= ~C_EOF; DPRINTF("==> cursor points to page %lu with %u keys, key index %u", mp->mp_pgno, NUMKEYS(mp), top->mp_ki); @@ -2480,7 +2529,7 @@ mdb_cursor_prev(MDB_cursor *cursor, MDB_val *key, MDB_val *data, MDB_cursor_op o leaf = NODEPTR(mp, top->mp_ki); if (F_ISSET(leaf->mn_flags, F_DUPDATA)) { - mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, mp, leaf); + mdb_xcursor_init1(cursor, mp, leaf); } if (data) { if ((rc = mdb_read_data(cursor->mc_txn, leaf, data) != MDB_SUCCESS)) @@ -2512,27 +2561,69 @@ mdb_cursor_set(MDB_cursor *cursor, MDB_val *key, MDB_val *data, assert(key->mv_size > 0); /* See if we're already on the right page */ - if (cursor->mc_initialized) { + if (cursor->mc_flags & C_INITIALIZED) { MDB_val nodekey; + top = CURSOR_TOP(cursor); - /* Don't try this for LEAF2 pages. Maybe support that later. */ - if ((top->mp_page->mp_flags & (P_LEAF|P_LEAF2)) == P_LEAF) { + if (top->mp_page->mp_flags & P_LEAF2) { + nodekey.mv_size = cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_pad; + nodekey.mv_data = LEAF2KEY(top->mp_page, 0, nodekey.mv_size); + } else { leaf = NODEPTR(top->mp_page, 0); MDB_SET_KEY(leaf, &nodekey); - rc = mdb_cmp(cursor->mc_txn, cursor->mc_dbi, key, &nodekey); - if (rc >= 0) { - leaf = NODEPTR(top->mp_page, NUMKEYS(top->mp_page)-1); - MDB_SET_KEY(leaf, &nodekey); - rc = mdb_cmp(cursor->mc_txn, cursor->mc_dbi, key, &nodekey); - if (rc <= 0) { - /* we're already on the right page */ + } + rc = cursor->mc_txn->mt_dbxs[cursor->mc_dbi].md_cmp(key, &nodekey); + if (rc == 0) { + /* Probably happens rarely, but first node on the page + * was the one we wanted. + */ + top->mp_ki = 0; +set1: + mpp.mp_page = top->mp_page; + if (exactp) + *exactp = 1; + rc = 0; + leaf = NODEPTR(top->mp_page, top->mp_ki); + goto set3; + } + if (rc > 0) { + unsigned int i; + if (NUMKEYS(top->mp_page) > 1) { + if (top->mp_page->mp_flags & P_LEAF2) { + nodekey.mv_data = LEAF2KEY(top->mp_page, + NUMKEYS(top->mp_page)-1, nodekey.mv_size); + } else { + leaf = NODEPTR(top->mp_page, NUMKEYS(top->mp_page)-1); + MDB_SET_KEY(leaf, &nodekey); + } + rc = cursor->mc_txn->mt_dbxs[cursor->mc_dbi].md_cmp(key, &nodekey); + if (rc == 0) { + /* last node was the one we wanted */ + top->mp_ki = NUMKEYS(top->mp_page)-1; + goto set1; + } + if (rc < 0) { + /* This is definitely the right page, skip search_page */ mpp.mp_page = top->mp_page; rc = 0; goto set2; } } + /* If any parents have right-sibs, search. + * Otherwise, there's nothing further. + */ + for (i=0; imc_snum-1; i++) + if (cursor->mc_stack[i].mp_ki < + NUMKEYS(cursor->mc_stack[i].mp_page)-1) + break; + if (i == cursor->mc_snum - 1) { + /* There are no other pages */ + top->mp_ki = NUMKEYS(top->mp_page); + return MDB_NOTFOUND; + } } } + cursor->mc_snum = 0; rc = mdb_search_page(cursor->mc_txn, cursor->mc_dbi, key, cursor, 0, &mpp); @@ -2560,8 +2651,9 @@ set2: leaf = NODEPTR(mpp.mp_page, 0); } - cursor->mc_initialized = 1; - cursor->mc_eof = 0; +set3: + cursor->mc_flags |= C_INITIALIZED; + cursor->mc_flags &= ~C_EOF; if (IS_LEAF2(mpp.mp_page)) { key->mv_size = cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_pad; @@ -2570,7 +2662,7 @@ set2: } if (F_ISSET(leaf->mn_flags, F_DUPDATA)) { - mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, mpp.mp_page, leaf); + mdb_xcursor_init1(cursor, mpp.mp_page, leaf); } if (data) { if (F_ISSET(leaf->mn_flags, F_DUPDATA)) { @@ -2580,6 +2672,7 @@ set2: int ex2, *ex2p; if (op == MDB_GET_BOTH) { ex2p = &ex2; + ex2 = 0; } else { ex2p = NULL; } @@ -2591,7 +2684,7 @@ set2: MDB_val d2; if ((rc = mdb_read_data(cursor->mc_txn, leaf, &d2)) != MDB_SUCCESS) return rc; - rc = mdb_dcmp(cursor->mc_txn, cursor->mc_dbi, data, &d2); + rc = cursor->mc_txn->mt_dbxs[cursor->mc_dbi].md_dcmp(data, &d2); if (rc) { if (op == MDB_GET_BOTH || rc > 0) return MDB_NOTFOUND; @@ -2626,8 +2719,8 @@ mdb_cursor_first(MDB_cursor *cursor, MDB_val *key, MDB_val *data) assert(IS_LEAF(mpp.mp_page)); leaf = NODEPTR(mpp.mp_page, 0); - cursor->mc_initialized = 1; - cursor->mc_eof = 0; + cursor->mc_flags |= C_INITIALIZED; + cursor->mc_flags &= ~C_EOF; if (IS_LEAF2(mpp.mp_page)) { key->mv_size = cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_pad; @@ -2637,13 +2730,13 @@ mdb_cursor_first(MDB_cursor *cursor, MDB_val *key, MDB_val *data) if (data) { if (F_ISSET(leaf->mn_flags, F_DUPDATA)) { - mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, mpp.mp_page, leaf); + mdb_xcursor_init1(cursor, mpp.mp_page, leaf); rc = mdb_cursor_first(&cursor->mc_xcursor->mx_cursor, data, NULL); if (rc) return rc; } else { if (cursor->mc_xcursor) - cursor->mc_xcursor->mx_cursor.mc_initialized = 0; + cursor->mc_xcursor->mx_cursor.mc_flags = 0; if ((rc = mdb_read_data(cursor->mc_txn, leaf, data)) != MDB_SUCCESS) return rc; } @@ -2672,8 +2765,8 @@ mdb_cursor_last(MDB_cursor *cursor, MDB_val *key, MDB_val *data) assert(IS_LEAF(mpp.mp_page)); leaf = NODEPTR(mpp.mp_page, NUMKEYS(mpp.mp_page)-1); - cursor->mc_initialized = 1; - cursor->mc_eof = 0; + cursor->mc_flags |= C_INITIALIZED; + cursor->mc_flags &= ~C_EOF; top = CURSOR_TOP(cursor); top->mp_ki = NUMKEYS(top->mp_page) - 1; @@ -2686,7 +2779,7 @@ mdb_cursor_last(MDB_cursor *cursor, MDB_val *key, MDB_val *data) if (data) { if (F_ISSET(leaf->mn_flags, F_DUPDATA)) { - mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, mpp.mp_page, leaf); + mdb_xcursor_init1(cursor, mpp.mp_page, leaf); rc = mdb_cursor_last(&cursor->mc_xcursor->mx_cursor, data, NULL); if (rc) return rc; @@ -2729,12 +2822,13 @@ mdb_cursor_get(MDB_cursor *cursor, MDB_val *key, MDB_val *data, case MDB_GET_MULTIPLE: if (data == NULL || !(cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPFIXED) || - !cursor->mc_initialized) { + !(cursor->mc_flags & C_INITIALIZED)) { rc = EINVAL; break; } rc = MDB_SUCCESS; - if (!cursor->mc_xcursor->mx_cursor.mc_initialized || cursor->mc_xcursor->mx_cursor.mc_eof) + if (!(cursor->mc_xcursor->mx_cursor.mc_flags & C_INITIALIZED) || + (cursor->mc_xcursor->mx_cursor.mc_flags & C_EOF)) break; goto fetchm; case MDB_NEXT_MULTIPLE: @@ -2743,12 +2837,12 @@ mdb_cursor_get(MDB_cursor *cursor, MDB_val *key, MDB_val *data, rc = EINVAL; break; } - if (!cursor->mc_initialized) + if (!(cursor->mc_flags & C_INITIALIZED)) rc = mdb_cursor_first(cursor, key, data); else rc = mdb_cursor_next(cursor, key, data, MDB_NEXT_DUP); if (rc == MDB_SUCCESS) { - if (cursor->mc_xcursor->mx_cursor.mc_initialized) { + if (cursor->mc_xcursor->mx_cursor.mc_flags & C_INITIALIZED) { MDB_ppage *top; fetchm: top = CURSOR_TOP(&cursor->mc_xcursor->mx_cursor); @@ -2764,7 +2858,7 @@ fetchm: case MDB_NEXT: case MDB_NEXT_DUP: case MDB_NEXT_NODUP: - if (!cursor->mc_initialized) + if (!(cursor->mc_flags & C_INITIALIZED)) rc = mdb_cursor_first(cursor, key, data); else rc = mdb_cursor_next(cursor, key, data, op); @@ -2772,7 +2866,7 @@ fetchm: case MDB_PREV: case MDB_PREV_DUP: case MDB_PREV_NODUP: - if (!cursor->mc_initialized || cursor->mc_eof) + if (!(cursor->mc_flags & C_INITIALIZED) || (cursor->mc_flags & C_EOF)) rc = mdb_cursor_last(cursor, key, data); else rc = mdb_cursor_prev(cursor, key, data, op); @@ -2780,9 +2874,29 @@ fetchm: case MDB_FIRST: rc = mdb_cursor_first(cursor, key, data); break; + case MDB_FIRST_DUP: + if (data == NULL || + !(cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) || + !(cursor->mc_flags & C_INITIALIZED) || + !(cursor->mc_xcursor->mx_cursor.mc_flags & C_INITIALIZED)) { + rc = EINVAL; + break; + } + rc = mdb_cursor_first(&cursor->mc_xcursor->mx_cursor, data, NULL); + break; case MDB_LAST: rc = mdb_cursor_last(cursor, key, data); break; + case MDB_LAST_DUP: + if (data == NULL || + !(cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) || + !(cursor->mc_flags & C_INITIALIZED) || + !(cursor->mc_xcursor->mx_cursor.mc_flags & C_INITIALIZED)) { + rc = EINVAL; + break; + } + rc = mdb_cursor_last(&cursor->mc_xcursor->mx_cursor, data, NULL); + break; default: DPRINTF("unhandled/unimplemented cursor operation %u", op); rc = EINVAL; @@ -2792,6 +2906,337 @@ fetchm: return rc; } +static int +mdb_cursor_touch(MDB_cursor *mc) +{ + MDB_pageparent mpp; + MDB_dhead *dh; + unsigned int i; + int rc; + + mpp.mp_parent = NULL; + mpp.mp_pi = 0; + if (mc->mc_dbi > MAIN_DBI && !mc->mc_txn->mt_dbxs[mc->mc_dbi].md_dirty) { + rc = mdb_search_page(mc->mc_txn, MAIN_DBI, &mc->mc_txn->mt_dbxs[mc->mc_dbi].md_name, + NULL, 1, &mpp); + if (rc) return rc; + mc->mc_txn->mt_dbxs[mc->mc_dbi].md_dirty = 1; + } + for(i=0; imc_snum; i++) { + mpp.mp_page = mc->mc_stack[i].mp_page; + if (!F_ISSET(mpp.mp_page->mp_flags, P_DIRTY)) { + rc = mdb_touch(mc->mc_txn, mc->mc_dbi, &mpp); + if (rc) return rc; + mc->mc_stack[i].mp_page = mpp.mp_page; + if (!i) { + mc->mc_txn->mt_dbs[mc->mc_dbi].md_root = mpp.mp_page->mp_pgno; + } + } + dh = ((MDB_dhead *)mpp.mp_page)-1; + dh->md_parent = mpp.mp_parent; + dh->md_pi = mpp.mp_pi; + mpp.mp_parent = mpp.mp_page; + mpp.mp_pi = mc->mc_stack[i].mp_ki; + } + return MDB_SUCCESS; +} + +int +mdb_cursor_put(MDB_cursor *mc, MDB_val *key, MDB_val *data, + unsigned int flags) +{ + MDB_ppage *top; + MDB_node *leaf; + MDB_val xdata, *rdata, dkey; + MDB_db dummy; + char dbuf[PAGESIZE]; + int do_sub = 0; + size_t nsize; + DKBUF; + int rc, rc2; + + if (F_ISSET(mc->mc_txn->mt_flags, MDB_TXN_RDONLY)) + return EACCES; + + DPRINTF("==> put db %u key [%s], size %zu, data size %zu", + mc->mc_dbi, DKEY(key), key->mv_size, data->mv_size); + + dkey.mv_size = 0; + + if (flags == MDB_CURRENT) { + if (!(mc->mc_flags & C_INITIALIZED)) + return EINVAL; + rc = MDB_SUCCESS; + } else if (mc->mc_txn->mt_dbs[mc->mc_dbi].md_root == P_INVALID) { + MDB_dpage *dp; + /* new database, write a root leaf page */ + DPUTS("allocating new root leaf page"); + if ((dp = mdb_new_page(mc->mc_txn, mc->mc_dbi, P_LEAF, 1)) == NULL) { + return ENOMEM; + } + mc->mc_snum = 0; + cursor_push_page(mc, &dp->p); + mc->mc_txn->mt_dbs[mc->mc_dbi].md_root = dp->p.mp_pgno; + mc->mc_txn->mt_dbs[mc->mc_dbi].md_depth++; + mc->mc_txn->mt_dbxs[mc->mc_dbi].md_dirty = 1; + if ((mc->mc_txn->mt_dbs[mc->mc_dbi].md_flags & (MDB_DUPSORT|MDB_DUPFIXED)) + == MDB_DUPFIXED) + dp->p.mp_flags |= P_LEAF2; + mc->mc_flags |= C_INITIALIZED; + rc = MDB_NOTFOUND; + goto top; + } else { + int exact = 0; + rc = mdb_cursor_set(mc, key, NULL, MDB_SET, &exact); + if (flags == MDB_NOOVERWRITE && rc == 0) { + DPRINTF("duplicate key [%s]", DKEY(key)); + return MDB_KEYEXIST; + } + if (rc && rc != MDB_NOTFOUND) + return rc; + } + + /* Cursor is positioned, now make sure all pages are writable */ + rc2 = mdb_cursor_touch(mc); + if (rc2) return rc2; + +top: + top = CURSOR_TOP(mc); + + /* The key already exists */ + if (rc == MDB_SUCCESS) { + /* there's only a key anyway, so this is a no-op */ + if (IS_LEAF2(top->mp_page)) { + unsigned int ksize = mc->mc_txn->mt_dbs[mc->mc_dbi].md_pad; + if (key->mv_size != ksize) + return EINVAL; + if (flags == MDB_CURRENT) { + char *ptr = LEAF2KEY(top->mp_page, top->mp_ki, ksize); + memcpy(ptr, key->mv_data, ksize); + } + return MDB_SUCCESS; + } + + leaf = NODEPTR(top->mp_page, top->mp_ki); + + /* DB has dups? */ + if (F_ISSET(mc->mc_txn->mt_dbs[mc->mc_dbi].md_flags, MDB_DUPSORT)) { + /* Was a single item before, must convert now */ + if (!F_ISSET(leaf->mn_flags, F_DUPDATA)) { + dkey.mv_size = NODEDSZ(leaf); + dkey.mv_data = dbuf; + memcpy(dbuf, NODEDATA(leaf), dkey.mv_size); + /* data matches, ignore it */ + if (!mdb_dcmp(mc->mc_txn, mc->mc_dbi, data, &dkey)) + return (flags == MDB_NODUPDATA) ? MDB_KEYEXIST : MDB_SUCCESS; + memset(&dummy, 0, sizeof(dummy)); + if (mc->mc_txn->mt_dbs[mc->mc_dbi].md_flags & MDB_DUPFIXED) { + dummy.md_pad = data->mv_size; + dummy.md_flags = MDB_DUPFIXED; + if (mc->mc_txn->mt_dbs[mc->mc_dbi].md_flags & MDB_INTEGERDUP) + dummy.md_flags |= MDB_INTEGERKEY; + } + dummy.md_root = P_INVALID; + if (dkey.mv_size == sizeof(MDB_db)) { + memcpy(NODEDATA(leaf), &dummy, sizeof(dummy)); + goto put_sub; + } + mdb_del_node(top->mp_page, top->mp_ki, 0); + do_sub = 1; + rdata = &xdata; + xdata.mv_size = sizeof(MDB_db); + xdata.mv_data = &dummy; + goto new_sub; + } + goto put_sub; + } + /* same size, just replace it */ + if (!F_ISSET(leaf->mn_flags, F_BIGDATA) && + NODEDSZ(leaf) == data->mv_size) { + memcpy(NODEDATA(leaf), data->mv_data, data->mv_size); + goto done; + } + mdb_del_node(top->mp_page, top->mp_ki, 0); + } else { + DPRINTF("inserting key at index %i", top->mp_ki); + } + + rdata = data; + +new_sub: + nsize = IS_LEAF2(top->mp_page) ? key->mv_size : mdb_leaf_size(mc->mc_txn->mt_env, key, rdata); + if (SIZELEFT(top->mp_page) < nsize) { + rc = mdb_split(mc->mc_txn, mc->mc_dbi, &top->mp_page, &top->mp_ki, key, rdata, P_INVALID); + } else { + /* There is room already in this leaf page. */ + rc = mdb_add_node(mc->mc_txn, mc->mc_dbi, top->mp_page, top->mp_ki, key, rdata, 0, 0); + } + + if (rc != MDB_SUCCESS) + mc->mc_txn->mt_flags |= MDB_TXN_ERROR; + else { + /* Remember if we just added a subdatabase */ + if (flags & F_SUBDATA) { + leaf = NODEPTR(top->mp_page, top->mp_ki); + leaf->mn_flags |= F_SUBDATA; + } + + /* Now store the actual data in the child DB. Note that we're + * storing the user data in the keys field, so there are strict + * size limits on dupdata. The actual data fields of the child + * DB are all zero size. + */ + if (do_sub) { + leaf = NODEPTR(top->mp_page, top->mp_ki); +put_sub: + if (flags == MDB_CURRENT) + mdb_xcursor_init2(mc); + else + mdb_xcursor_init1(mc, top->mp_page, leaf); + xdata.mv_size = 0; + xdata.mv_data = ""; + if (flags == MDB_NODUPDATA) + flags = MDB_NOOVERWRITE; + /* converted, write the original data first */ + if (dkey.mv_size) { + rc = mdb_cursor_put(&mc->mc_xcursor->mx_cursor, &dkey, &xdata, flags); + if (rc) return rc; + leaf->mn_flags |= F_DUPDATA; + } + rc = mdb_cursor_put(&mc->mc_xcursor->mx_cursor, data, &xdata, flags); + mdb_xcursor_fini(mc); + memcpy(NODEDATA(leaf), + &mc->mc_xcursor->mx_txn.mt_dbs[mc->mc_xcursor->mx_cursor.mc_dbi], + sizeof(MDB_db)); + } + mc->mc_txn->mt_dbs[mc->mc_dbi].md_entries++; + } +done: + return rc; +} + +int +mdb_cursor_del(MDB_cursor *mc, unsigned int flags) +{ + MDB_pageparent mpp; + MDB_ppage *top; + MDB_node *leaf; + int rc; + + if (F_ISSET(mc->mc_txn->mt_flags, MDB_TXN_RDONLY)) + return EACCES; + + if (!mc->mc_flags & C_INITIALIZED) + return EINVAL; + + rc = mdb_cursor_touch(mc); + if (rc) return rc; + + top = CURSOR_TOP(mc); + leaf = NODEPTR(top->mp_page, top->mp_ki); + mpp.mp_page = top->mp_page; + if (mc->mc_snum > 1) { + MDB_ppage *parent = CURSOR_PARENT(mc); + mpp.mp_parent = parent->mp_page; + mpp.mp_pi = parent->mp_ki; + } else { + mpp.mp_parent = NULL; + mpp.mp_pi = 0; + } + + if (!IS_LEAF2(top->mp_page) && F_ISSET(leaf->mn_flags, F_DUPDATA)) { + MDB_pageparent mp2; + + if (flags != MDB_NODUPDATA) { + mdb_xcursor_init2(mc); + rc = mdb_cursor_del(&mc->mc_xcursor->mx_cursor, 0); + mdb_xcursor_fini(mc); + /* If sub-DB still has entries, we're done */ + if (mc->mc_xcursor->mx_txn.mt_dbs[mc->mc_xcursor->mx_cursor.mc_dbi].md_root + != P_INVALID) { + memcpy(NODEDATA(leaf), + &mc->mc_xcursor->mx_txn.mt_dbs[mc->mc_xcursor->mx_cursor.mc_dbi], + sizeof(MDB_db)); + mc->mc_txn->mt_dbs[mc->mc_dbi].md_entries--; + return rc; + } + /* otherwise fall thru and delete the sub-DB */ + } + + /* add all the child DB's pages to the free list */ + mc->mc_xcursor->mx_cursor.mc_snum = 0; + rc = mdb_search_page(&mc->mc_xcursor->mx_txn, mc->mc_xcursor->mx_cursor.mc_dbi, + NULL, &mc->mc_xcursor->mx_cursor, 0, &mp2); + if (rc == MDB_SUCCESS) { + MDB_ppage *top, *parent; + MDB_node *ni; + unsigned int i; +#if 0 + MDB_dpage *dp; + ID2 mid; + int dirty_root = 0; +#endif + + mc->mc_txn->mt_dbs[mc->mc_dbi].md_entries -= + mc->mc_xcursor->mx_txn.mt_dbs[mc->mc_xcursor->mx_cursor.mc_dbi].md_entries; + + cursor_pop_page(&mc->mc_xcursor->mx_cursor); + if (mc->mc_xcursor->mx_cursor.mc_snum) { +#if 0 + if (mc->mc_xcursor->mx_cursor.mc_stack[0].mp_page->mp_flags & P_DIRTY) { + dirty_root = 1; + } +#endif + while (mc->mc_xcursor->mx_cursor.mc_snum > 1) { + top = CURSOR_TOP(&mc->mc_xcursor->mx_cursor); + parent = CURSOR_PARENT(&mc->mc_xcursor->mx_cursor); + for (i=0; imp_page); i++) { + MDB_page *mp; + pgno_t pg; + ni = NODEPTR(top->mp_page, i); + pg = NODEPGNO(ni); + if ((rc = mdb_get_page(mc->mc_txn, pg, &mp))) + return rc; +#if 0 + if (mp->mp_flags & P_DIRTY) { + /* drop it */ + mid.mid = pg; + mdb_mid2l_delete(mc->mc_txn->mt_u.dirty_list, &mid); + dp = mid.mptr; + dp->h.md_parent = (MDB_page *)mc->mc_txn->mt_env->me_dpages; + mc->mc_txn->mt_env->me_dpages = dp; + } else +#endif + { + /* free it */ + mdb_midl_append(mc->mc_txn->mt_free_pgs, pg); + } + } + rc = mdb_sibling(&mc->mc_xcursor->mx_cursor, 1); + if (rc) break; + } + } +#if 0 + if (dirty_root) { + /* drop it */ + mid.mid = mc->mc_xcursor->mx_txn.mt_dbs[mc->mc_xcursor->mx_cursor.mc_dbi].md_root; + mdb_mid2l_delete(mc->mc_txn->mt_u.dirty_list, &mid); + dp = mid.mptr; + dp->h.md_parent = (MDB_page *)mc->mc_txn->mt_env->me_dpages; + mc->mc_txn->mt_env->me_dpages = dp; + } else +#endif + { + /* free it */ + mdb_midl_append(mc->mc_txn->mt_free_pgs, + mc->mc_xcursor->mx_txn.mt_dbs[mc->mc_xcursor->mx_cursor.mc_dbi].md_root); + } + } + } + + return mdb_del0(mc, top->mp_ki, &mpp, leaf); +} + /* Allocate a page and initialize it */ static MDB_dpage * @@ -2829,6 +3274,7 @@ mdb_leaf_size(MDB_env *env, MDB_val *key, MDB_val *data) /* put on overflow page */ sz -= data->mv_size - sizeof(pgno_t); } + sz += sz & 1; return sz + sizeof(indx_t); } @@ -2904,6 +3350,7 @@ mdb_add_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp, indx_t indx, node_size += data->mv_size; } } + node_size += node_size & 1; if (node_size + sizeof(indx_t) > SIZELEFT(mp)) { DPRINTF("not enough room in page %lu, got %u ptrs", @@ -2930,9 +3377,9 @@ mdb_add_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp, indx_t indx, node->mn_ksize = (key == NULL) ? 0 : key->mv_size; node->mn_flags = flags; if (IS_LEAF(mp)) - node->mn_dsize = data->mv_size; + SETDSZ(node,data->mv_size); else - NODEPGNO(node) = pgno; + SETPGNO(node,pgno); if (key) memcpy(NODEKEY(node), key->mv_data, key->mv_size); @@ -2986,6 +3433,7 @@ mdb_del_node(MDB_page *mp, indx_t indx, int ksize) else sz += NODEDSZ(node); } + sz += sz & 1; ptr = mp->mp_ptrs[indx]; numkeys = NUMKEYS(mp); @@ -3006,17 +3454,18 @@ mdb_del_node(MDB_page *mp, indx_t indx, int ksize) } static void -mdb_xcursor_init0(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx) +mdb_xcursor_init0(MDB_cursor *mc) { + MDB_xcursor *mx = mc->mc_xcursor; MDB_dbi dbn; - mx->mx_txn = *txn; + mx->mx_txn = *mc->mc_txn; mx->mx_txn.mt_dbxs = mx->mx_dbxs; mx->mx_txn.mt_dbs = mx->mx_dbs; - mx->mx_dbxs[0] = txn->mt_dbxs[0]; - mx->mx_dbxs[1] = txn->mt_dbxs[1]; - if (dbi > 1) { - mx->mx_dbxs[2] = txn->mt_dbxs[dbi]; + mx->mx_dbxs[0] = mc->mc_txn->mt_dbxs[0]; + mx->mx_dbxs[1] = mc->mc_txn->mt_dbxs[1]; + if (mc->mc_dbi > 1) { + mx->mx_dbxs[2] = mc->mc_txn->mt_dbxs[mc->mc_dbi]; dbn = 2; } else { dbn = 1; @@ -3026,49 +3475,68 @@ mdb_xcursor_init0(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx) mx->mx_dbxs[dbn+1].md_rel = mx->mx_dbxs[dbn].md_rel; mx->mx_dbxs[dbn+1].md_dirty = 0; mx->mx_txn.mt_numdbs = dbn+2; + mx->mx_txn.mt_u = mc->mc_txn->mt_u; - mx->mx_cursor.mc_snum = 0; + mx->mx_cursor.mc_xcursor = NULL; mx->mx_cursor.mc_txn = &mx->mx_txn; mx->mx_cursor.mc_dbi = dbn+1; } static void -mdb_xcursor_init1(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx, MDB_page *mp, MDB_node *node) +mdb_xcursor_init1(MDB_cursor *mc, MDB_page *mp, MDB_node *node) { MDB_db *db = NODEDATA(node); + MDB_xcursor *mx = mc->mc_xcursor; MDB_dbi dbn; - mx->mx_dbs[0] = txn->mt_dbs[0]; - mx->mx_dbs[1] = txn->mt_dbs[1]; - if (dbi > 1) { - mx->mx_dbs[2] = txn->mt_dbs[dbi]; + mx->mx_dbs[0] = mc->mc_txn->mt_dbs[0]; + mx->mx_dbs[1] = mc->mc_txn->mt_dbs[1]; + if (mc->mc_dbi > 1) { + mx->mx_dbs[2] = mc->mc_txn->mt_dbs[mc->mc_dbi]; + mx->mx_dbxs[2].md_dirty = mc->mc_txn->mt_dbxs[mc->mc_dbi].md_dirty; dbn = 3; } else { dbn = 2; } - DPRINTF("Sub-db %u for db %u root page %lu", dbn, dbi, db->md_root); + DPRINTF("Sub-db %u for db %u root page %lu", dbn, mc->mc_dbi, db->md_root); mx->mx_dbs[dbn] = *db; if (F_ISSET(mp->mp_flags, P_DIRTY)) mx->mx_dbxs[dbn].md_dirty = 1; mx->mx_dbxs[dbn].md_name.mv_data = NODEKEY(node); mx->mx_dbxs[dbn].md_name.mv_size = node->mn_ksize; - mx->mx_txn.mt_next_pgno = txn->mt_next_pgno; - mx->mx_txn.mt_u = txn->mt_u; - mx->mx_cursor.mc_initialized = 0; - mx->mx_cursor.mc_eof = 0; + mx->mx_txn.mt_next_pgno = mc->mc_txn->mt_next_pgno; + mx->mx_cursor.mc_snum = 0; + mx->mx_cursor.mc_flags = 0; +} + +static void +mdb_xcursor_init2(MDB_cursor *mc) +{ + MDB_xcursor *mx = mc->mc_xcursor; + MDB_dbi dbn; + mx->mx_dbs[0] = mc->mc_txn->mt_dbs[0]; + mx->mx_dbs[1] = mc->mc_txn->mt_dbs[1]; + if (mc->mc_dbi > 1) { + mx->mx_dbs[2] = mc->mc_txn->mt_dbs[mc->mc_dbi]; + mx->mx_dbxs[2].md_dirty = mc->mc_txn->mt_dbxs[mc->mc_dbi].md_dirty; + dbn = 3; + } else { + dbn = 2; + } + DPRINTF("Sub-db %u for db %u root page %lu", dbn, mc->mc_dbi, + mx->mx_dbs[dbn].md_root); + mx->mx_txn.mt_next_pgno = mc->mc_txn->mt_next_pgno; } static void -mdb_xcursor_fini(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx) +mdb_xcursor_fini(MDB_cursor *mc) { - txn->mt_next_pgno = mx->mx_txn.mt_next_pgno; - txn->mt_u = mx->mx_txn.mt_u; - txn->mt_dbs[0] = mx->mx_dbs[0]; - txn->mt_dbs[1] = mx->mx_dbs[1]; - txn->mt_dbxs[0].md_dirty = mx->mx_dbxs[0].md_dirty; - txn->mt_dbxs[1].md_dirty = mx->mx_dbxs[1].md_dirty; - if (dbi > 1) { - txn->mt_dbs[dbi] = mx->mx_dbs[2]; - txn->mt_dbxs[dbi].md_dirty = mx->mx_dbxs[2].md_dirty; + MDB_xcursor *mx = mc->mc_xcursor; + mc->mc_txn->mt_next_pgno = mx->mx_txn.mt_next_pgno; + mc->mc_txn->mt_dbs[0] = mx->mx_dbs[0]; + mc->mc_txn->mt_dbs[1] = mx->mx_dbs[1]; + if (mc->mc_dbi > 1) { + mc->mc_txn->mt_dbs[mc->mc_dbi] = mx->mx_dbs[2]; + mc->mc_txn->mt_dbxs[mc->mc_dbi].md_dirty = mx->mx_dbxs[2].md_dirty; } } @@ -3090,7 +3558,7 @@ mdb_cursor_open(MDB_txn *txn, MDB_dbi dbi, MDB_cursor **ret) if (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT) { MDB_xcursor *mx = (MDB_xcursor *)(cursor + 1); cursor->mc_xcursor = mx; - mdb_xcursor_init0(txn, dbi, mx); + mdb_xcursor_init0(cursor); } } else { return ENOMEM; @@ -3119,7 +3587,7 @@ mdb_cursor_count(MDB_cursor *mc, unsigned long *countp) if (!F_ISSET(leaf->mn_flags, F_DUPDATA)) { *countp = 1; } else { - if (!mc->mc_xcursor->mx_cursor.mc_initialized) + if (!(mc->mc_xcursor->mx_cursor.mc_flags & C_INITIALIZED)) return EINVAL; *countp = mc->mc_xcursor->mx_txn.mt_dbs[mc->mc_xcursor->mx_cursor.mc_dbi].md_entries; @@ -3127,6 +3595,32 @@ mdb_cursor_count(MDB_cursor *mc, unsigned long *countp) return MDB_SUCCESS; } +#if 0 +static void +mdb_cursor_reset(MDB_cursor *mc) +{ + if (mc != NULL) { + if (mc->mc_flags & C_XDIRTY) { + + mdb_xcursor_fini(mc); + + /* If sub-DB still has entries, update root record */ + if (mc->mc_xcursor->mx_txn.mt_dbs[mc->mc_xcursor->mx_cursor.mc_dbi].md_root + != P_INVALID) { + MDB_ppage *top; + MDB_node *leaf; + top = CURSOR_TOP(mc); + leaf = NODEPTR(top->mp_page, top->mp_ki); + memcpy(NODEDATA(leaf), + &mc->mc_xcursor->mx_txn.mt_dbs[mc->mc_xcursor->mx_cursor.mc_dbi], + sizeof(MDB_db)); + } + mc->mc_flags ^= C_XDIRTY; + } + } +} +#endif + void mdb_cursor_close(MDB_cursor *cursor) { @@ -3203,7 +3697,16 @@ mdb_move_node(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *src, indx_t srcindx, data.mv_size = 0; data.mv_data = NULL; } else { - srcnode = NODEPTR(src->mp_page, srcindx); + if (srcindx == 0 && IS_BRANCH(src->mp_page)) { + /* must find the lowest key below src */ + MDB_pageparent mpp; + mpp.mp_page = src->mp_page; + mpp.mp_pi = 0; + mdb_search_page_root(txn, dbi, NULL, NULL, 0, &mpp); + srcnode = NODEPTR(mpp.mp_page, 0); + } else { + srcnode = NODEPTR(src->mp_page, srcindx); + } key.mv_size = NODEKSZ(srcnode); key.mv_data = NODEKEY(srcnode); data.mv_size = NODEDSZ(srcnode); @@ -3227,17 +3730,17 @@ mdb_move_node(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *src, indx_t srcindx, */ mdb_del_node(src->mp_page, srcindx, key.mv_size); - /* The key value just changed due to del_node, find it again. - */ - if (!IS_LEAF2(src->mp_page)) { - srcnode = NODEPTR(src->mp_page, srcindx); - key.mv_data = NODEKEY(srcnode); - } - /* Update the parent separators. */ if (srcindx == 0) { if (src->mp_pi != 0) { + if (IS_LEAF2(src->mp_page)) { + key.mv_data = LEAF2KEY(src->mp_page, srcindx, key.mv_size); + } else { + srcnode = NODEPTR(src->mp_page, srcindx); + key.mv_size = NODEKSZ(srcnode); + key.mv_data = NODEKEY(srcnode); + } DPRINTF("update separator for source page %lu to [%s]", src->mp_page->mp_pgno, DKEY(&key)); if ((rc = mdb_update_key(src->mp_parent, src->mp_pi, @@ -3253,6 +3756,13 @@ mdb_move_node(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *src, indx_t srcindx, if (dstindx == 0) { if (dst->mp_pi != 0) { + if (IS_LEAF2(src->mp_page)) { + key.mv_data = LEAF2KEY(dst->mp_page, 0, key.mv_size); + } else { + srcnode = NODEPTR(dst->mp_page, 0); + key.mv_size = NODEKSZ(srcnode); + key.mv_data = NODEKEY(srcnode); + } DPRINTF("update separator for destination page %lu to [%s]", dst->mp_page->mp_pgno, DKEY(&key)); if ((rc = mdb_update_key(dst->mp_parent, dst->mp_pi, @@ -3373,8 +3883,10 @@ mdb_rebalance(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *mpp) txn->mt_dbs[dbi].md_root = P_INVALID; txn->mt_dbs[dbi].md_depth = 0; txn->mt_dbs[dbi].md_leaf_pages = 0; + mdb_midl_append(txn->mt_free_pgs, mpp->mp_page->mp_pgno); } else if (IS_BRANCH(mpp->mp_page) && NUMKEYS(mpp->mp_page) == 1) { DPUTS("collapsing root page!"); + mdb_midl_append(txn->mt_free_pgs, mpp->mp_page->mp_pgno); txn->mt_dbs[dbi].md_root = NODEPGNO(NODEPTR(mpp->mp_page, 0)); if ((rc = mdb_get_page(txn, txn->mt_dbs[dbi].md_root, &root))) return rc; @@ -3439,7 +3951,7 @@ mdb_rebalance(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *mpp) } static int -mdb_del0(MDB_txn *txn, MDB_dbi dbi, unsigned int ki, MDB_pageparent *mpp, MDB_node *leaf) +mdb_del0(MDB_cursor *mc, unsigned int ki, MDB_pageparent *mpp, MDB_node *leaf) { int rc; @@ -3449,18 +3961,18 @@ mdb_del0(MDB_txn *txn, MDB_dbi dbi, unsigned int ki, MDB_pageparent *mpp, MDB_no pgno_t pg; memcpy(&pg, NODEDATA(leaf), sizeof(pg)); - ovpages = OVPAGES(NODEDSZ(leaf), txn->mt_env->me_psize); + ovpages = OVPAGES(NODEDSZ(leaf), mc->mc_txn->mt_env->me_psize); for (i=0; imt_free_pgs, pg); + mdb_midl_append(mc->mc_txn->mt_free_pgs, pg); pg++; } } - mdb_del_node(mpp->mp_page, ki, txn->mt_dbs[dbi].md_pad); - txn->mt_dbs[dbi].md_entries--; - rc = mdb_rebalance(txn, dbi, mpp); + mdb_del_node(mpp->mp_page, ki, mc->mc_txn->mt_dbs[mc->mc_dbi].md_pad); + mc->mc_txn->mt_dbs[mc->mc_dbi].md_entries--; + rc = mdb_rebalance(mc->mc_txn, mc->mc_dbi, mpp); if (rc != MDB_SUCCESS) - txn->mt_flags |= MDB_TXN_ERROR; + mc->mc_txn->mt_flags |= MDB_TXN_ERROR; return rc; } @@ -3469,10 +3981,11 @@ int mdb_del(MDB_txn *txn, MDB_dbi dbi, MDB_val *key, MDB_val *data) { + MDB_cursor mc; + MDB_xcursor mx; + MDB_cursor_op op; + MDB_val rdata, *xdata; int rc, exact; - unsigned int ki; - MDB_node *leaf; - MDB_pageparent mpp; DKBUF; assert(key != NULL); @@ -3490,67 +4003,29 @@ mdb_del(MDB_txn *txn, MDB_dbi dbi, return EINVAL; } - mpp.mp_parent = NULL; - mpp.mp_pi = 0; - if ((rc = mdb_search_page(txn, dbi, key, NULL, 1, &mpp)) != MDB_SUCCESS) - return rc; - - leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, &ki); - if (leaf == NULL || !exact) { - return MDB_NOTFOUND; + mc.mc_txn = txn; + mc.mc_dbi = dbi; + mc.mc_flags = 0; + if (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT) { + mc.mc_xcursor = &mx; + mdb_xcursor_init0(&mc); + } else { + mc.mc_xcursor = NULL; } - if (!IS_LEAF2(mpp.mp_page) && F_ISSET(leaf->mn_flags, F_DUPDATA)) { - MDB_xcursor mx; - MDB_pageparent mp2; - - mdb_xcursor_init0(txn, dbi, &mx); - mdb_xcursor_init1(txn, dbi, &mx, mpp.mp_page, leaf); - if (data) { - rc = mdb_del(&mx.mx_txn, mx.mx_cursor.mc_dbi, data, NULL); - mdb_xcursor_fini(txn, dbi, &mx); - /* If sub-DB still has entries, we're done */ - if (mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi].md_root != P_INVALID) { - memcpy(NODEDATA(leaf), &mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi], - sizeof(MDB_db)); - txn->mt_dbs[dbi].md_entries--; - return rc; - } - /* otherwise fall thru and delete the sub-DB */ - } else { - /* add all the child DB's pages to the free list */ - rc = mdb_search_page(&mx.mx_txn, mx.mx_cursor.mc_dbi, - NULL, &mx.mx_cursor, 0, &mp2); - if (rc == MDB_SUCCESS) { - MDB_ppage *top, *parent; - MDB_node *ni; - unsigned int i; - - cursor_pop_page(&mx.mx_cursor); - if (mx.mx_cursor.mc_snum) { - top = CURSOR_TOP(&mx.mx_cursor); - while (mx.mx_cursor.mc_snum > 1) { - parent = CURSOR_PARENT(&mx.mx_cursor); - for (i=0; imp_page); i++) { - ni = NODEPTR(top->mp_page, i); - mdb_midl_insert(txn->mt_free_pgs, NODEPGNO(ni)); - } - parent->mp_ki++; - if (parent->mp_ki >= NUMKEYS(parent->mp_page)) { - cursor_pop_page(&mx.mx_cursor); - top = parent; - } else { - ni = NODEPTR(parent->mp_page, parent->mp_ki); - rc = mdb_get_page(&mx.mx_txn, NODEPGNO(ni), &top->mp_page); - } - } - } - mdb_midl_insert(txn->mt_free_pgs, mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi].md_root); - } - } + exact = 0; + if (data) { + op = MDB_GET_BOTH; + rdata = *data; + xdata = &rdata; + } else { + op = MDB_SET; + xdata = NULL; } - - return mdb_del0(txn, dbi, ki, &mpp, leaf); + rc = mdb_cursor_set(&mc, key, xdata, op, &exact); + if (rc == 0) + rc = mdb_cursor_del(&mc, data ? 0 : MDB_NODUPDATA); + return rc; } /* Split page <*mpp>, and insert in either left or @@ -3662,34 +4137,34 @@ mdb_split(MDB_txn *txn, MDB_dbi dbi, MDB_page **mpp, unsigned int *newindxp, /* Maximum free space in an empty page */ pmax = txn->mt_env->me_psize - PAGEHDRSZ; nsize = mdb_leaf_size(txn->mt_env, newkey, newdata); - if (newindx <= split_indx) { -split1: + if (newindx < split_indx) { psize = nsize; for (i=0; ip, i); - psize += NODESIZE + NODEKSZ(node); + psize += NODESIZE + NODEKSZ(node) + sizeof(indx_t); if (F_ISSET(node->mn_flags, F_BIGDATA)) psize += sizeof(pgno_t); else psize += NODEDSZ(node); + psize += psize & 1; if (psize > pmax) { - split_indx--; - goto split1; + split_indx = i; + break; } } } else { -split2: psize = nsize; - for (i=split_indx; i=split_indx; i--) { node = NODEPTR(&mdp->p, i); - psize += NODESIZE + NODEKSZ(node); + psize += NODESIZE + NODEKSZ(node) + sizeof(indx_t); if (F_ISSET(node->mn_flags, F_BIGDATA)) psize += sizeof(pgno_t); else psize += NODEDSZ(node); + psize += psize & 1; if (psize > pmax) { - split_indx++; - goto split2; + split_indx = i+1; + break; } } } @@ -3697,7 +4172,6 @@ split2: /* First find the separating key between the split pages. */ - memset(&sepkey, 0, sizeof(sepkey)); if (newindx == split_indx) { sepkey.mv_size = newkey->mv_size; sepkey.mv_data = newkey->mv_data; @@ -3777,7 +4251,7 @@ newsep: rkey.mv_size = node->mn_ksize; if (IS_LEAF(&mdp->p)) { rdata.mv_data = NODEDATA(node); - rdata.mv_size = node->mn_dsize; + rdata.mv_size = NODEDSZ(node); } else pgno = NODEPGNO(node); flags = node->mn_flags; @@ -3804,161 +4278,13 @@ newsep: return rc; } -static int -mdb_put0(MDB_txn *txn, MDB_dbi dbi, - MDB_val *key, MDB_val *data, unsigned int flags) -{ - int rc = MDB_SUCCESS, exact; - unsigned int ki; - MDB_node *leaf; - MDB_pageparent mpp; - MDB_val xdata, *rdata, dkey; - MDB_db dummy; - char dbuf[PAGESIZE]; - int do_sub = 0; - size_t nsize; - DKBUF; - - DPRINTF("==> put db %u key [%s], size %zu, data size %zu", - dbi, DKEY(key), key->mv_size, data->mv_size); - - dkey.mv_size = 0; - mpp.mp_parent = NULL; - mpp.mp_pi = 0; - rc = mdb_search_page(txn, dbi, key, NULL, 1, &mpp); - if (rc == MDB_SUCCESS) { - leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, &ki); - if (leaf && exact) { - if (flags == MDB_NOOVERWRITE) { - DPRINTF("duplicate key [%s]", DKEY(key)); - return MDB_KEYEXIST; - } - /* there's only a key anyway, so this is a no-op */ - if (IS_LEAF2(mpp.mp_page)) - return MDB_SUCCESS; - - if (F_ISSET(txn->mt_dbs[dbi].md_flags, MDB_DUPSORT)) { - /* Was a single item before, must convert now */ - if (!F_ISSET(leaf->mn_flags, F_DUPDATA)) { - dkey.mv_size = NODEDSZ(leaf); - dkey.mv_data = dbuf; - memcpy(dbuf, NODEDATA(leaf), dkey.mv_size); - /* data matches, ignore it */ - if (!mdb_dcmp(txn, dbi, data, &dkey)) - return (flags == MDB_NODUPDATA) ? MDB_KEYEXIST : MDB_SUCCESS; - memset(&dummy, 0, sizeof(dummy)); - if (txn->mt_dbs[dbi].md_flags & MDB_DUPFIXED) { - dummy.md_pad = data->mv_size; - dummy.md_flags = MDB_DUPFIXED; - if (txn->mt_dbs[dbi].md_flags & MDB_INTEGERDUP) - dummy.md_flags |= MDB_INTEGERKEY; - } - dummy.md_root = P_INVALID; - if (dkey.mv_size == sizeof(MDB_db)) { - memcpy(NODEDATA(leaf), &dummy, sizeof(dummy)); - goto put_sub; - } - mdb_del_node(mpp.mp_page, ki, 0); - do_sub = 1; - rdata = &xdata; - xdata.mv_size = sizeof(MDB_db); - xdata.mv_data = &dummy; - goto new_sub; - } - goto put_sub; - } - /* same size, just replace it */ - if (!F_ISSET(leaf->mn_flags, F_BIGDATA) && - NODEDSZ(leaf) == data->mv_size) { - memcpy(NODEDATA(leaf), data->mv_data, data->mv_size); - goto done; - } - mdb_del_node(mpp.mp_page, ki, 0); - } - if (leaf == NULL) { /* append if not found */ - ki = NUMKEYS(mpp.mp_page); - DPRINTF("appending key at index %i", ki); - } - } else if (rc == MDB_NOTFOUND) { - MDB_dpage *dp; - /* new file, just write a root leaf page */ - DPUTS("allocating new root leaf page"); - if ((dp = mdb_new_page(txn, dbi, P_LEAF, 1)) == NULL) { - return ENOMEM; - } - mpp.mp_page = &dp->p; - txn->mt_dbs[dbi].md_root = mpp.mp_page->mp_pgno; - txn->mt_dbs[dbi].md_depth++; - txn->mt_dbxs[dbi].md_dirty = 1; - if ((txn->mt_dbs[dbi].md_flags & (MDB_DUPSORT|MDB_DUPFIXED)) == MDB_DUPFIXED) - mpp.mp_page->mp_flags |= P_LEAF2; - ki = 0; - } - else - goto done; - - assert(IS_LEAF(mpp.mp_page)); - DPRINTF("there are %u keys, should insert new key at index %i", - NUMKEYS(mpp.mp_page), ki); - - rdata = data; - -new_sub: - nsize = IS_LEAF2(mpp.mp_page) ? key->mv_size : mdb_leaf_size(txn->mt_env, key, rdata); - if (SIZELEFT(mpp.mp_page) < nsize) { - rc = mdb_split(txn, dbi, &mpp.mp_page, &ki, key, rdata, P_INVALID); - } else { - /* There is room already in this leaf page. */ - rc = mdb_add_node(txn, dbi, mpp.mp_page, ki, key, rdata, 0, 0); - } - - if (rc != MDB_SUCCESS) - txn->mt_flags |= MDB_TXN_ERROR; - else { - /* Remember if we just added a subdatabase */ - if (flags & F_SUBDATA) { - leaf = NODEPTR(mpp.mp_page, ki); - leaf->mn_flags |= F_SUBDATA; - } - - /* Now store the actual data in the child DB. Note that we're - * storing the user data in the keys field, so there are strict - * size limits on dupdata. The actual data fields of the child - * DB are all zero size. - */ - if (do_sub) { - MDB_xcursor mx; - - leaf = NODEPTR(mpp.mp_page, ki); -put_sub: - mdb_xcursor_init0(txn, dbi, &mx); - mdb_xcursor_init1(txn, dbi, &mx, mpp.mp_page, leaf); - xdata.mv_size = 0; - xdata.mv_data = ""; - if (flags == MDB_NODUPDATA) - flags = MDB_NOOVERWRITE; - /* converted, write the original data first */ - if (dkey.mv_size) { - rc = mdb_put0(&mx.mx_txn, mx.mx_cursor.mc_dbi, &dkey, &xdata, flags); - if (rc) return rc; - leaf->mn_flags |= F_DUPDATA; - } - rc = mdb_put0(&mx.mx_txn, mx.mx_cursor.mc_dbi, data, &xdata, flags); - mdb_xcursor_fini(txn, dbi, &mx); - memcpy(NODEDATA(leaf), &mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi], - sizeof(MDB_db)); - } - txn->mt_dbs[dbi].md_entries++; - } - -done: - return rc; -} - int mdb_put(MDB_txn *txn, MDB_dbi dbi, MDB_val *key, MDB_val *data, unsigned int flags) { + MDB_cursor mc; + MDB_xcursor mx; + assert(key != NULL); assert(data != NULL); @@ -3976,7 +4302,17 @@ mdb_put(MDB_txn *txn, MDB_dbi dbi, if ((flags & (MDB_NOOVERWRITE|MDB_NODUPDATA)) != flags) return EINVAL; - return mdb_put0(txn, dbi, key, data, flags); + mc.mc_txn = txn; + mc.mc_dbi = dbi; + mc.mc_snum = 0; + mc.mc_flags = 0; + if (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT) { + mc.mc_xcursor = &mx; + mdb_xcursor_init0(&mc); + } else { + mc.mc_xcursor = NULL; + } + return mdb_cursor_put(&mc, key, data, flags); } int @@ -4037,6 +4373,32 @@ mdb_env_stat(MDB_env *env, MDB_stat *arg) return mdb_stat0(env, &env->me_metas[toggle]->mm_dbs[MAIN_DBI], arg); } +static void +mdb_default_cmp(MDB_txn *txn, MDB_dbi dbi) +{ + if (txn->mt_dbs[dbi].md_flags & MDB_REVERSEKEY) + txn->mt_dbxs[dbi].md_cmp = memnrcmp; + else if (txn->mt_dbs[dbi].md_flags & MDB_INTEGERKEY) + txn->mt_dbxs[dbi].md_cmp = cintcmp; + else + txn->mt_dbxs[dbi].md_cmp = memncmp; + + if (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT) { + if (txn->mt_dbs[dbi].md_flags & MDB_INTEGERDUP) { + if (txn->mt_dbs[dbi].md_flags & MDB_DUPFIXED) + txn->mt_dbxs[dbi].md_dcmp = intcmp; + else + txn->mt_dbxs[dbi].md_dcmp = cintcmp; + } else if (txn->mt_dbs[dbi].md_flags & MDB_REVERSEDUP) { + txn->mt_dbxs[dbi].md_dcmp = memnrcmp; + } else { + txn->mt_dbxs[dbi].md_dcmp = memncmp; + } + } else { + txn->mt_dbxs[dbi].md_dcmp = NULL; + } +} + int mdb_open(MDB_txn *txn, const char *name, unsigned int flags, MDB_dbi *dbi) { MDB_val key, data; @@ -4044,14 +4406,23 @@ int mdb_open(MDB_txn *txn, const char *name, unsigned int flags, MDB_dbi *dbi) int rc, dirty = 0; size_t len; + if (txn->mt_dbxs[FREE_DBI].md_cmp == NULL) { + mdb_default_cmp(txn, FREE_DBI); + } + /* main DB? */ if (!name) { *dbi = MAIN_DBI; if (flags & (MDB_DUPSORT|MDB_REVERSEKEY|MDB_INTEGERKEY)) txn->mt_dbs[MAIN_DBI].md_flags |= (flags & (MDB_DUPSORT|MDB_REVERSEKEY|MDB_INTEGERKEY)); + mdb_default_cmp(txn, MAIN_DBI); return MDB_SUCCESS; } + if (txn->mt_dbxs[MAIN_DBI].md_cmp == NULL) { + mdb_default_cmp(txn, MAIN_DBI); + } + /* Is the DB already open? */ len = strlen(name); for (i=2; imt_numdbs; i++) { @@ -4072,13 +4443,17 @@ int mdb_open(MDB_txn *txn, const char *name, unsigned int flags, MDB_dbi *dbi) /* Create if requested */ if (rc == MDB_NOTFOUND && (flags & MDB_CREATE)) { + MDB_cursor mc; MDB_db dummy; data.mv_size = sizeof(MDB_db); data.mv_data = &dummy; memset(&dummy, 0, sizeof(dummy)); dummy.md_root = P_INVALID; dummy.md_flags = flags & 0xffff; - rc = mdb_put0(txn, MAIN_DBI, &key, &data, F_SUBDATA); + mc.mc_txn = txn; + mc.mc_dbi = MAIN_DBI; + mc.mc_flags = 0; + rc = mdb_cursor_put(&mc, &key, &data, F_SUBDATA); dirty = 1; } @@ -4086,8 +4461,6 @@ int mdb_open(MDB_txn *txn, const char *name, unsigned int flags, MDB_dbi *dbi) if (rc == MDB_SUCCESS) { txn->mt_dbxs[txn->mt_numdbs].md_name.mv_data = strdup(name); txn->mt_dbxs[txn->mt_numdbs].md_name.mv_size = len; - txn->mt_dbxs[txn->mt_numdbs].md_cmp = NULL; - txn->mt_dbxs[txn->mt_numdbs].md_dcmp = NULL; txn->mt_dbxs[txn->mt_numdbs].md_rel = NULL; txn->mt_dbxs[txn->mt_numdbs].md_parent = MAIN_DBI; txn->mt_dbxs[txn->mt_numdbs].md_dirty = dirty; @@ -4095,6 +4468,7 @@ int mdb_open(MDB_txn *txn, const char *name, unsigned int flags, MDB_dbi *dbi) *dbi = txn->mt_numdbs; txn->mt_env->me_dbs[0][txn->mt_numdbs] = txn->mt_dbs[txn->mt_numdbs]; txn->mt_env->me_dbs[1][txn->mt_numdbs] = txn->mt_dbs[txn->mt_numdbs]; + mdb_default_cmp(txn, txn->mt_numdbs); txn->mt_numdbs++; }