ITS#7682 mdb_env_copy(): Avoid Linux O_DIRECT bug.

[openldap] / libraries / liblmdb / mdb.c

diff --git a/libraries/liblmdb/mdb.c b/libraries/liblmdb/mdb.c
index f59fe51bc4eaae1621d9ad8dee76b52b6aeb3f76..46dea53c877d2f0a645d34aaa89345cd801323eb 100644 (file)
--- a/libraries/liblmdb/mdb.c
+++ b/libraries/liblmdb/mdb.c
@@ -933,6 +933,7 @@ struct MDB_cursor {
 #define C_INITIALIZED  0x01    /**< cursor has been initialized and is valid */
 #define C_EOF  0x02                    /**< No more data */
 #define C_SUB  0x04                    /**< Cursor is a sub-cursor */
+#define C_DEL  0x08                    /**< last op was a cursor_del */
 #define C_SPLITTING    0x20            /**< Cursor is in page_split */
 #define C_UNTRACK      0x40            /**< Un-track cursor when closing */
 /** @} */
@@ -4245,17 +4246,18 @@ mdb_env_copy(MDB_env *env, const char *path)
        newfd = CreateFile(lpath, GENERIC_WRITE, 0, NULL, CREATE_NEW,
                                FILE_FLAG_NO_BUFFERING|FILE_FLAG_WRITE_THROUGH, NULL);
 #else
-       newfd = open(lpath, O_WRONLY|O_CREAT|O_EXCL
-#ifdef O_DIRECT
-               |O_DIRECT
-#endif
-               , 0666);
+       newfd = open(lpath, O_WRONLY|O_CREAT|O_EXCL, 0666);
 #endif
        if (newfd == INVALID_HANDLE_VALUE) {
                rc = ErrCode();
                goto leave;
        }
 
+#ifdef O_DIRECT
+       /* Set O_DIRECT if the file system supports it */
+       if ((rc = fcntl(newfd, F_GETFL)) != -1)
+               (void) fcntl(newfd, F_SETFL, rc | O_DIRECT);
+#endif
 #ifdef F_NOCACHE       /* __APPLE__ */
        rc = fcntl(newfd, F_NOCACHE, 1);
        if (rc) {
@@ -5006,6 +5008,8 @@ 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", mp->mp_pgno, (void *) mc));
+       if (mc->mc_flags & C_DEL)
+               goto skip;
 
        if (mc->mc_ki[mc->mc_top] + 1u >= NUMKEYS(mp)) {
                DPUTS("=====> move to next sibling page");
@@ -5018,6 +5022,7 @@ mdb_cursor_next(MDB_cursor *mc, MDB_val *key, MDB_val *data, MDB_cursor_op op)
        } else
                mc->mc_ki[mc->mc_top]++;
 
+skip:
        DPRINTF(("==> cursor points to page %"Z"u with %u keys, key index %u",
            mp->mp_pgno, NUMKEYS(mp), mc->mc_ki[mc->mc_top]));
 
@@ -5223,7 +5228,11 @@ mdb_cursor_set(MDB_cursor *mc, MDB_val *key, MDB_val *data,
                if (!mc->mc_top) {
                        /* There are no other pages */
                        mc->mc_ki[mc->mc_top] = 0;
-                       return MDB_NOTFOUND;
+                       if (op == MDB_SET_RANGE) {
+                               rc = 0;
+                               goto set1;
+                       } else
+                               return MDB_NOTFOUND;
                }
        }
 
@@ -5287,6 +5296,7 @@ set1:
                        if (rc) {
                                if (op == MDB_GET_BOTH || rc > 0)
                                        return MDB_NOTFOUND;
+                               rc = 0;
                        }
 
                } else {
@@ -5432,6 +5442,8 @@ mdb_cursor_get(MDB_cursor *mc, MDB_val *key, MDB_val *data,
                                MDB_GET_KEY(leaf, key);
                                if (data) {
                                        if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
+                                               if (mc->mc_flags & C_DEL)
+                                                       mdb_xcursor_init1(mc, leaf);
                                                rc = mdb_cursor_get(&mc->mc_xcursor->mx_cursor, data, NULL, MDB_GET_CURRENT);
                                        } else {
                                                rc = mdb_node_read(mc->mc_txn, leaf, data);
@@ -5556,6 +5568,9 @@ fetchm:
                break;
        }
 
+       if (mc->mc_flags & C_DEL)
+               mc->mc_flags ^= C_DEL;
+
        return rc;
 }
 
@@ -5676,6 +5691,9 @@ mdb_cursor_put(MDB_cursor *mc, MDB_val *key, MDB_val *data,
                        return rc;
        }
 
+       if (mc->mc_flags & C_DEL)
+               mc->mc_flags ^= C_DEL;
+
        /* Cursor is positioned, check for room in the dirty list */
        if (!nospill) {
                if (flags & MDB_MULTIPLE) {
@@ -6105,6 +6123,7 @@ mdb_cursor_del(MDB_cursor *mc, unsigned int flags)
                                        }
                                }
                                mc->mc_db->md_entries--;
+                               mc->mc_flags |= C_DEL;
                                return rc;
                        }
                        /* otherwise fall thru and delete the sub-DB */
@@ -7228,8 +7247,11 @@ mdb_rebalance(MDB_cursor *mc)
        else {
                if (mc->mc_ki[ptop] == 0)
                        rc = mdb_page_merge(&mn, mc);
-               else
+               else {
+                       mn.mc_ki[mn.mc_top] += mc->mc_ki[mn.mc_top] + 1;
                        rc = mdb_page_merge(mc, &mn);
+                       mdb_cursor_copy(&mn, mc);
+               }
                mc->mc_flags &= ~(C_INITIALIZED|C_EOF);
        }
        return rc;
@@ -7280,12 +7302,16 @@ mdb_cursor_del0(MDB_cursor *mc, MDB_node *leaf)
                        if (!(m2->mc_flags & C_INITIALIZED))
                                continue;
                        if (m2->mc_pg[mc->mc_top] == mp) {
-                               if (m2->mc_ki[mc->mc_top] > ki)
-                                       m2->mc_ki[mc->mc_top]--;
+                               if (m2->mc_ki[mc->mc_top] >= ki) {
+                                       m2->mc_flags |= C_DEL;
+                                       if (m2->mc_ki[mc->mc_top] > ki)
+                                               m2->mc_ki[mc->mc_top]--;
+                               }
                                if (m2->mc_ki[mc->mc_top] >= nkeys)
                                        mdb_cursor_sibling(m2, 1);
                        }
                }
+               mc->mc_flags |= C_DEL;
        }
 
        return rc;