Daniel Shahaf <[email protected]> writes: > [email protected] wrote on Fri, Jul 08, 2011 at 14:02:42 -0000: >> Author: philip >> Date: Fri Jul 8 14:02:42 2011 >> New Revision: 1144316 >> >> URL: http://svn.apache.org/viewvc?rev=1144316&view=rev >> Log: >> Fix issue 3953, mod_dav_svn failing to reject bogus mergeinfo on commit. >> Move server-side mergeinfo validation so that all RA layers use it. >> > ... >> @@ -222,6 +255,9 @@ svn_repos_fs_change_node_prop(svn_fs_roo >> const svn_string_t *value, >> apr_pool_t *pool) >> { >> + if (value && strcmp(name, SVN_PROP_MERGEINFO) == 0) >> + SVN_ERR(verify_mergeinfo(value, path, pool)); >> + >> /* Validate the property, then call the wrapped function. */ >> SVN_ERR(svn_repos__validate_prop(name, value, pool)); >> return svn_fs_change_node_prop(root, path, name, value, pool); >> > > Shouldn't the call to verify_mergeinfo() be made by > svn_repos__validate_prop(), rather than here?
Perhaps. svn_repos__validate_prop is also used by svn_repos_fs_change_txn_props, is it appropriate to validate mergeinfo there? We should probably split the validate function into two, one for node props and one for revprops. -- uberSVN: Apache Subversion Made Easy http://www.uberSVN.com
