directives: correctly handle quoted strings in directives

Quoted strings should be valid inside directives, including ones that
contain a ] character. Also check for garbage on the line after the
directive.

Signed-off-by: H. Peter Anvin (Intel) <hpa@zytor.com>

Author: H. Peter Anvin (Intel)

asm/directiv.c

@@ -22,6 +22,7 @@
 #include "listing.h"
 #include "labels.h"
 #include "iflag.h"
+#include "quote.h"
 
 struct cpunames {
     const char *name;
@@ -178,34 +179,51 @@ static int get_bits(const char *value)
 
 static enum directive parse_directive_line(char **directive, char **value)
 {
-    char *p, *q, *buf;
+    char *p, *q, *eol, *buf;
+    char c;
 
     buf = nasm_skip_spaces(*directive);
 
     /*
      * It should be enclosed in [ ].
-     * XXX: we don't check there is nothing else on the remainder of the
-     * line, except a possible comment.
+     *
+     * Strip off the comments.  We should really strip the comments in
+     * generic code, not here.  While we're at it, it would be better
+     * to pass the backend a series of tokens instead of a raw string,
+     * and actually process quoted strings for it, like of like argv
+     * is handled in C.
      */
     if (*buf != '[')
         return D_none;
-    q = strchr(buf, ']');
-    if (!q)
-        return D_corrupt;
+
+    q = buf;
+    while ((c = *q) != ']') {
+        switch (c) {
+        case '\0':
+        case ';':
+            return D_corrupt;   /* No ] in directive */
+        case '\'':
+        case '\"':
+        case '`':
+            q = nasm_skip_string(q);
+            if (!*q++)
+                return D_corrupt;
+            break;
+        default:
+            q++;
+            break;
+        }
+    }
 
     /*
-     * Strip off the comments.  XXX: this doesn't account for quoted
-     * strings inside a directive.  We should really strip the
-     * comments in generic code, not here.  While we're at it, it
-     * would be better to pass the backend a series of tokens instead
-     * of a raw string, and actually process quoted strings for it,
-     * like of like argv is handled in C.
+     * Found the ] at the end of the directive. Make sure there isn't
+     * anything else at the end of the line, except a possible
+     * comment.
      */
-    p = strchr(buf, ';');
-    if (p) {
-        if (p < q) /* ouch! somewhere inside */
-            return D_corrupt;
-        *p = '\0';
+    eol = nasm_skip_spaces(q+1);
+    if (*eol != '\0' && *eol != ';') {
+        nasm_warn(WARN_DIRECTIVE_GARBAGE_EOL,
+                  "garbage found on line after directive");
     }
 
     /* no brace, no trailing spaces */
@@ -264,7 +282,7 @@ bool process_directives(char *directive)
 
     switch (d) {
     case D_none:
-        return D_none;      /* Not a directive */
+        return false;
 
     case D_corrupt:
 	nasm_nonfatal("invalid directive line");
@@ -285,6 +303,12 @@ bool process_directives(char *directive)
             default:
                 panic();
             }
+        } else if (d < D_pseudo_ops) {
+            nasm_nonfatal("internal error: unimplemented directive [%s]",
+                          directive);
+            break;
+        } else {
+            goto unknown;
         }
         break;
 
@@ -617,11 +641,10 @@ bool process_directives(char *directive)
         break;
     }
 
-
     /* A common error message */
     if (bad_param) {
         nasm_nonfatal("invalid parameter to [%s] directive", directive);
     }
 
-    return d != D_none;
+    return true;
 }

asm/directiv.dat

@@ -17,6 +17,12 @@
 ;; In the future, this will be turned into a general list of keywords
 ;; to be parsed in special contexts.
 ;;
+;; #special tokens are used to separate different classes of tokens.
+;; Make sure to add new tokens to the correct place. If a token is used
+;; in both a directive and in pragmas, they should be in the directive
+;; part of the list. Tokens used in pragmas *only* go at the end of the
+;; list.
+;;
 
 ; --- General configuration
 #name directive
@@ -25,10 +31,15 @@
 #header directiv.h
 
 ; --- Special enum values
+;     These must be first in the list.
 #special none = 0			; Must be zero
 #special unknown
 #special corrupt
 
+; --- True directives
+;     These tokens are accepted as directives by the global code or
+;     ignored if not implemented.
+
 ; --- Global directives
 absolute
 bits
@@ -40,16 +51,29 @@ dollarhex
 extern
 float
 global
-static
 list
+pragma
+required
+sectalign
 section
 segment
+static
 warning
-sectalign
-pragma
-required
+
+; --- Common output directives/pragmas
+prefix
+suffix
+postfix
+gprefix
+gsuffix
+gpostfix
+lprefix
+lsuffix
+lpostfix
 
 ; --- Pseudo-op list, for the benefit of %isdirective
+;     Tokens put in this part will be rejected as either directives or
+;     pragmas
 #special pseudo_ops
 db
 dw
@@ -71,6 +95,8 @@ incbin
 equ
 
 ; --- Format-specific directives
+;     Tokens put in this part of the list will be forwarded to the
+;     backend ofmt->directive() method.
 #special ofmt
 export				; outcoff, outobj
 group				; outobj
@@ -83,7 +109,8 @@ osabi				; outelf
 safeseh				; outcoff
 uppercase			; outieee, outobj
 
-; --- The following are tokens used in pragmas, not actual directives
+; --- The following are tokens used in pragmas, not actual directives.
+;     They will not be accepted as directives.
 #special pragma_tokens
 
 ; --- Assembler pragmas
@@ -92,17 +119,6 @@ limit
 ; --- Listing pragmas
 options
 
-; --- Common output pragmas
-prefix
-suffix
-postfix
-gprefix
-gsuffix
-gpostfix
-lprefix
-lsuffix
-lpostfix
-
 ; --- Backend-specific pragmas
 subsections_via_symbols		; macho
 no_dead_strip			; macho

asm/warnings.dat

@@ -9,6 +9,11 @@ db-empty [on] no operand for data declaration
     This is permitted, but often indicative of an error.
     See \k{db}.
 
+directive-garbage-eol [err] garbage after directive
+    Text was found after a directive. This is a warning so it can be
+    suppressed, because previous versions of NASM did not check for
+    this condition.
+
 ea-absolute [on] absolute address cannot be RIP-relative
     Warns that an address that is inherently absolute cannot
     be generated with RIP-relative encoding using \c{REL},