mdb_drop optimization

[openldap] / libraries / liblmdb / mdb.c

diff --git a/libraries/liblmdb/mdb.c b/libraries/liblmdb/mdb.c
index a624cba3c4f0418c9ab142b1d7032e78b8985ea6..6c0ee3f0fc1d8a556d344c9090d8263956b1ab92 100644 (file)
--- a/libraries/liblmdb/mdb.c
+++ b/libraries/liblmdb/mdb.c
@@ -5,7 +5,7 @@
  *     BerkeleyDB API, but much simplified.
  */
 /*
- * Copyright 2011-2015 Howard Chu, Symas Corp.
+ * Copyright 2011-2016 Howard Chu, Symas Corp.
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -198,7 +198,7 @@ typedef SSIZE_T     ssize_t;
 #define ESECT
 #endif
 
-#ifdef _MSC_VER
+#ifdef _WIN32
 #define CALL_CONV WINAPI
 #else
 #define CALL_CONV
@@ -1271,7 +1271,7 @@ typedef struct MDB_ntxn {
 #endif
 
        /** max bytes to write in one call */
-#define MAX_WRITE              (0x80000000U >> (sizeof(ssize_t) == 4))
+#define MAX_WRITE              (0x40000000U >> (sizeof(ssize_t) == 4))
 
        /** Check \b txn and \b dbi arguments to a function */
 #define TXN_DBI_EXIST(txn, dbi, validity) \
@@ -1417,8 +1417,9 @@ mdb_strerror(int err)
         *      This works as long as no function between the call to mdb_strerror
         *      and the actual use of the message uses more than 4K of stack.
         */
-       char pad[4096];
-       char buf[1024], *ptr = buf;
+#define MSGSIZE        1024
+#define PADSIZE 4096
+       char buf[MSGSIZE+PADSIZE], *ptr = buf;
 #endif
        int i;
        if (!err)
@@ -1450,7 +1451,7 @@ mdb_strerror(int err)
        buf[0] = 0;
        FormatMessageA(FORMAT_MESSAGE_FROM_SYSTEM |
                FORMAT_MESSAGE_IGNORE_INSERTS,
-               NULL, err, 0, ptr, sizeof(buf), (va_list *)pad);
+               NULL, err, 0, ptr, MSGSIZE, (va_list *)buf+MSGSIZE);
        return ptr;
 #else
        return strerror(err);
@@ -5668,8 +5669,10 @@ mdb_cursor_next(MDB_cursor *mc, MDB_val *key, MDB_val *data, MDB_cursor_op op)
 
        DPRINTF(("cursor_next: top page is %"Z"u in cursor %p",
                mdb_dbg_pgno(mp), (void *) mc));
-       if (mc->mc_flags & C_DEL)
+       if (mc->mc_flags & C_DEL) {
+               mc->mc_flags ^= C_DEL;
                goto skip;
+       }
 
        if (mc->mc_ki[mc->mc_top] + 1u >= NUMKEYS(mp)) {
                DPUTS("=====> move to next sibling page");
@@ -5748,6 +5751,8 @@ mdb_cursor_prev(MDB_cursor *mc, MDB_val *key, MDB_val *data, MDB_cursor_op op)
        DPRINTF(("cursor_prev: top page is %"Z"u in cursor %p",
                mdb_dbg_pgno(mp), (void *) mc));
 
+       mc->mc_flags &= ~(C_EOF|C_DEL);
+
        if (mc->mc_ki[mc->mc_top] == 0)  {
                DPUTS("=====> move to prev sibling page");
                if ((rc = mdb_cursor_sibling(mc, 0)) != MDB_SUCCESS) {
@@ -5759,8 +5764,6 @@ mdb_cursor_prev(MDB_cursor *mc, MDB_val *key, MDB_val *data, MDB_cursor_op op)
        } else
                mc->mc_ki[mc->mc_top]--;
 
-       mc->mc_flags &= ~C_EOF;
-
        DPRINTF(("==> cursor points to page %"Z"u with %u keys, key index %u",
            mdb_dbg_pgno(mp), NUMKEYS(mp), mc->mc_ki[mc->mc_top]));
 
@@ -5971,8 +5974,8 @@ set1:
                                if (op == MDB_GET_BOTH || rc > 0)
                                        return MDB_NOTFOUND;
                                rc = 0;
-                               *data = olddata;
                        }
+                       *data = olddata;
 
                } else {
                        if (mc->mc_xcursor)
@@ -6491,7 +6494,7 @@ more:
 #endif
                                /* does data match? */
                                if (!dcmp(data, &olddata)) {
-                                       if (flags & MDB_NODUPDATA)
+                                       if (flags & (MDB_NODUPDATA|MDB_APPENDDUP))
                                                return MDB_KEYEXIST;
                                        /* overwrite it */
                                        goto current;
@@ -9157,7 +9160,7 @@ mdb_env_copyfd0(MDB_env *env, HANDLE fd)
        MDB_txn *txn = NULL;
        mdb_mutexref_t wmutex = NULL;
        int rc;
-       size_t wsize;
+       size_t wsize, w3;
        char *ptr;
 #ifdef _WIN32
        DWORD len, w2;
@@ -9216,15 +9219,15 @@ mdb_env_copyfd0(MDB_env *env, HANDLE fd)
        if (rc)
                goto leave;
 
-       w2 = txn->mt_next_pgno * env->me_psize;
+       w3 = txn->mt_next_pgno * env->me_psize;
        {
                size_t fsize = 0;
                if ((rc = mdb_fsize(env->me_fd, &fsize)))
                        goto leave;
-               if (w2 > fsize)
-                       w2 = fsize;
+               if (w3 > fsize)
+                       w3 = fsize;
        }
-       wsize = w2 - wsize;
+       wsize = w3 - wsize;
        while (wsize > 0) {
                if (wsize > MAX_WRITE)
                        w2 = MAX_WRITE;
@@ -9293,7 +9296,7 @@ mdb_env_copy2(MDB_env *env, const char *path, unsigned int flags)
 #ifdef _WIN32
        rc = utf8_to_utf16(lpath, -1, &wpath, NULL);
        if (rc)
-               return rc;
+               goto leave;
        newfd = CreateFileW(wpath, GENERIC_WRITE, 0, NULL, CREATE_NEW,
                                FILE_FLAG_NO_BUFFERING|FILE_FLAG_WRITE_THROUGH, NULL);
        free(wpath);
@@ -9657,8 +9660,11 @@ mdb_drop0(MDB_cursor *mc, int subs)
 
                /* DUPSORT sub-DBs have no ovpages/DBs. Omit scanning leaves.
                 * This also avoids any P_LEAF2 pages, which have no nodes.
+                * Also if the DB doesn't have sub-DBs and has no overflow
+                * pages, omit scanning leaves.
                 */
-               if (mc->mc_flags & C_SUB)
+               if ((mc->mc_flags & C_SUB) ||
+                       (!subs && !mc->mc_db->md_overflow_pages))
                        mdb_cursor_pop(mc);
 
                mdb_cursor_copy(mc, &mx);
@@ -9680,6 +9686,9 @@ mdb_drop0(MDB_cursor *mc, int subs)
                                                        pg, omp->mp_pages);
                                                if (rc)
                                                        goto done;
+                                               mc->mc_db->md_overflow_pages -= omp->mp_pages;
+                                               if (!mc->mc_db->md_overflow_pages && !subs)
+                                                       break;
                                        } else if (subs && (ni->mn_flags & F_SUBDATA)) {
                                                mdb_xcursor_init1(mc, ni);
                                                rc = mdb_drop0(&mc->mc_xcursor->mx_cursor, 0);
@@ -9687,6 +9696,8 @@ mdb_drop0(MDB_cursor *mc, int subs)
                                                        goto done;
                                        }
                                }
+                               if (!subs && !mc->mc_db->md_overflow_pages)
+                                       goto pop;
                        } else {
                                if ((rc = mdb_midl_need(&txn->mt_free_pgs, n)) != 0)
                                        goto done;
@@ -9708,6 +9719,7 @@ mdb_drop0(MDB_cursor *mc, int subs)
                                /* no more siblings, go back to beginning
                                 * of previous level.
                                 */
+pop:
                                mdb_cursor_pop(mc);
                                mc->mc_ki[0] = 0;
                                for (i=1; i<mc->mc_snum; i++) {