From: Pierangelo Masarati Date: Mon, 12 Nov 2001 11:29:40 +0000 (+0000) Subject: trims space after comma in dn.regex acls (prelude to normalization) X-Git-Tag: LDBM_PRE_GIANT_RWLOCK~881 X-Git-Url: https://git.sur5r.net/?a=commitdiff_plain;h=b4e7f0873aed39fab727120c4d53491b8dafbfbd;p=openldap trims space after comma in dn.regex acls (prelude to normalization) --- diff --git a/servers/slapd/aclparse.c b/servers/slapd/aclparse.c index 0a2a573184..00c93c23b1 100644 --- a/servers/slapd/aclparse.c +++ b/servers/slapd/aclparse.c @@ -21,6 +21,8 @@ static void split(char *line, int splitchar, char **left, char **right); static void access_append(Access **l, Access *a); static void acl_usage(void) LDAP_GCCATTR((noreturn)); +static char *acl_regex_normalized_dn(const char *pattern); + #ifdef LDAP_DEBUG static void print_acl(Backend *be, AccessControl *a); static void print_access(Access *b); @@ -167,7 +169,7 @@ parse_acl( a->acl_dn_pat = ch_strdup( "*" ); } else { - a->acl_dn_pat = ch_strdup( right ); + a->acl_dn_pat = acl_regex_normalized_dn( right ); } } else if ( strcasecmp( style, "base" ) == 0 ) { a->acl_dn_style = ACL_STYLE_BASE; @@ -342,8 +344,8 @@ parse_acl( pat = ch_strdup( "*" ); } else { - regtest(fname, lineno, right); - pat = ch_strdup( right ); + pat = acl_regex_normalized_dn( right ); + regtest(fname, lineno, pat); } } else if ( right == NULL || *right == '\0' ) { fprintf( stderr, @@ -452,8 +454,9 @@ parse_acl( b->a_group_style = sty; if (sty == ACL_STYLE_REGEX) { - regtest(fname, lineno, right); - b->a_group_pat = ch_strdup( right ); + char *tmp = acl_regex_normalized_dn( right ); + regtest(fname, lineno, tmp); + b->a_group_pat = tmp; } else { b->a_group_pat = ch_strdup( right ); dn_normalize(b->a_group_pat); @@ -578,9 +581,12 @@ parse_acl( b->a_peername_style = sty; if (sty == ACL_STYLE_REGEX) { - regtest(fname, lineno, right); + char *tmp = acl_regex_normalized_dn( right ); + regtest(fname, lineno, tmp); + b->a_peername_pat = tmp; + } else { + b->a_peername_pat = ch_strdup( right ); } - b->a_peername_pat = ch_strdup( right ); continue; } @@ -601,9 +607,12 @@ parse_acl( b->a_sockname_style = sty; if (sty == ACL_STYLE_REGEX) { - regtest(fname, lineno, right); + char *tmp = acl_regex_normalized_dn( right ); + regtest(fname, lineno, tmp); + b->a_sockname_pat = tmp; + } else { + b->a_sockname_pat = ch_strdup( right ); } - b->a_sockname_pat = ch_strdup( right ); continue; } @@ -624,9 +633,12 @@ parse_acl( b->a_domain_style = sty; if (sty == ACL_STYLE_REGEX) { - regtest(fname, lineno, right); + char *tmp = acl_regex_normalized_dn( right ); + regtest(fname, lineno, tmp); + b->a_domain_pat = tmp; + } else { + b->a_domain_pat = ch_strdup( right ); } - b->a_domain_pat = ch_strdup( right ); continue; } @@ -647,9 +659,12 @@ parse_acl( b->a_sockurl_style = sty; if (sty == ACL_STYLE_REGEX) { - regtest(fname, lineno, right); + char *tmp = acl_regex_normalized_dn( right ); + regtest(fname, lineno, tmp); + b->a_sockurl_pat = tmp; + } else { + b->a_sockurl_pat = ch_strdup( right ); } - b->a_sockurl_pat = ch_strdup( right ); continue; } @@ -1131,6 +1146,45 @@ acl_usage( void ) exit( EXIT_FAILURE ); } +/* + * At present it simply eats the (optional) space after + * a RDN separator (,) + * Eventually will evolve in a more complete normalization + */ +static char * +acl_regex_normalized_dn( + const char *pattern +) +{ + char *str, *p; + + str = ch_strdup( pattern ); + + if ( str == NULL ) { + return( NULL ); + } + + for ( p = str; p[ 0 ]; p++ ) { + /* escape */ + if ( p[ 0 ] == '\\' ) { + p++; + } + + if ( p[ 0 ] == ',' ) { + if ( p[ 1 ] == ' ' ) { + char *q; + + for ( q = &p[ 2 ]; q[ 0 ] == ' '; q++ ) { + /* DO NOTHING */ ; + } + AC_MEMCPY( &p[ 1 ], &q[ 0 ], strlen( q ) + 1 ); + } + } + } + + return( str ); +} + static void split( char *line,