Change the way options are parsed on the `#!'-line of a shell-script. Instead

of having the kernel parse that line and add an entry to the argument list for
each 'separate word' it finds, have it add only one entry which holds all
the words found on that line.  The old behavior is useful in some situations,
but it does not match the way any other operating system will parse that line.

This has been discussed in the thread "Bug in #! processing - One More Time"
on the freebsd-arch mailing list (starting back on Feb 24, 2005).  The first
few messages in that thread provide the background in much detail.

PR:		16393
Reviewed by:	freebsd-arch

1 files changed, 174 insertions, 7 deletions

diff --git a/sys/kern/imgact_shell.c b/sys/kern/imgact_shell.c
index 04df003..3955855 100644
--- a/sys/kern/imgact_shell.c
+++ b/sys/kern/imgact_shell.c
@@ -36,6 +36,15 @@ __FBSDID("$FreeBSD$");
 #include <sys/imgact.h>
 #include <sys/kernel.h>
 
+#define	KEEP_OLDCODE	1
+#if BYTE_ORDER == LITTLE_ENDIAN		/* temp for OLD_CODE kludge */
+#define	DBG_MAGIC	0x2B23		/* #+ in "little-endian" */
+#define	OLD_MAGIC	0x3C23		/* #< */
+#else
+#define	DBG_MAGIC	0x232B		/* #+ in big-endian */
+#define	OLD_MAGIC	0x233C		/* #< */
+#endif
+
 #if BYTE_ORDER == LITTLE_ENDIAN
 #define SHELLMAGIC	0x2123 /* #! */
 #else
@@ -43,15 +52,65 @@ __FBSDID("$FreeBSD$");
 #endif
 
 /*
- * Shell interpreter image activator. An interpreter name beginning
- *	at imgp->args->begin_argv is the minimal successful exit requirement.
+ * At the time of this writing, MAXSHELLCMDLEN == PAGE_SIZE.  This is
+ * significant because the caller has only mapped in one page of the
+ * file we're reading.  This code should be changed to know how to
+ * read in the second page, but I'm not doing that just yet...
+ */
+#if MAXSHELLCMDLEN > PAGE_SIZE
+#error "MAXSHELLCMDLEN is larger than a single page!"
+#endif
+
+/**
+ * Shell interpreter image activator. An interpreter name beginning at
+ * imgp->args->begin_argv is the minimal successful exit requirement.
+ *
+ * If the given file is a shell-script, then the first line will start
+ * with the two characters `#!' (aka SHELLMAGIC), followed by the name
+ * of the shell-interpreter to run, followed by zero or more tokens.
+ *
+ * The interpreter is then started up such that it will see:
+ *    arg[0] -> The name of interpreter as specified after `#!' in the
+ *		first line of the script.  The interpreter name must
+ *		not be longer than MAXSHELLCMDLEN bytes.
+ *    arg[1] -> *If* there are any additional tokens on the first line,
+ *		then we add a new arg[1], which is a copy of the rest of
+ *		that line.  The copy starts at the first token after the
+ *		interpreter name.  We leave it to the interpreter to
+ *		parse the tokens in that value.
+ *    arg[x] -> the full pathname of the script.  This will either be
+ *		arg[2] or arg[1], depending on whether or not tokens
+ *		were found after the interpreter name.
+ *  arg[x+1] -> all the arguments that were specified on the original
+ *		command line.
+ *
+ * This processing is described in the execve(2) man page.
+ */
+
+/*
+ * HISTORICAL NOTE: From 1993 to mid-2005, FreeBSD parsed out the tokens as
+ * found on the first line of the script, and setup each token as a separate
+ * value in arg[].  This extra processing did not match the behavior of other
+ * OS's, and caused a few subtle problems.  For one, it meant the kernel was
+ * deciding how those values should be parsed (wrt characters for quoting or
+ * comments, etc), while the interpreter might have other rules for parsing.
+ * It also meant the interpreter had no way of knowing which arguments came
+ * from the first line of the shell script, and which arguments were specified
+ * by the user on the command line.
+ *
+ * Only few things in the base system might depend on that non-standard
+ * processing (mainly /bin/sh and /usr/bin/env).  And for programs which are
+ * not in the base system, the "newer" behavior matches how NetBSD, OpenBSD,
+ * Linux, Solaris, AIX, IRIX, and many other Unixes have set up the arg-list
+ * for the interpreter.  So if a program can handle this behavior on those
+ * other OS's, it should be able to handle it for FreeBSD too.
  */
 int
 exec_shell_imgact(imgp)
 	struct image_params *imgp;
 {
 	const char *image_header = imgp->image_header;
-	const char *ihp;
+	const char *ihp, *interpb, *interpe, *maxp, *optb, *opte;
 	int error, offset;
 	size_t length, clength;
 	struct vattr vattr;
@@ -79,14 +138,34 @@ exec_shell_imgact(imgp)
 	if (error)
 		return (error);
 
-	clength = (vattr.va_size > MAXSHELLCMDLEN) ?
-	    MAXSHELLCMDLEN : vattr.va_size;
+	/*
+	 * Copy shell name and arguments from image_header into a string
+	 *	buffer.  Remember that the caller has mapped only the
+	 *	first page of the file into memory.
+	 */
+	clength = (vattr.va_size > PAGE_SIZE) ? PAGE_SIZE : vattr.va_size;
+
+	maxp = &image_header[clength];
+	ihp = &image_header[2];
+#if KEEP_OLDCODE
+	/*
+	 * XXX - Temporarily provide a quick-and-dirty way to get the
+	 * older, non-standard option-parsing behavior, just in case
+	 * someone finds themselves in an emergency where they need it.
+	 * This will not be documented.  It is only for initial testing.
+	 */
+	if (*(const short *)ihp == OLD_MAGIC)
+		ihp += 2;
+	else
+		goto new_code;
+	interpb = ihp;
+
 	/*
 	 * Figure out the number of bytes that need to be reserved in the
 	 * argument string to copy the contents of the interpreter's command
 	 * line into the argument string.
 	 */
-	ihp = &image_header[2];
+	ihp = interpb;
 	offset = 0;
 	while (ihp < &image_header[clength]) {
 		/* Skip any whitespace */
@@ -152,7 +231,7 @@ exec_shell_imgact(imgp)
 	 * Loop through the interpreter name yet again, copying as
 	 * we go.
 	 */
-	ihp = &image_header[2];
+	ihp = interpb;
 	offset = 0;
 	while (ihp < &image_header[clength]) {
 		/* Skip whitespace */
@@ -174,8 +253,96 @@ exec_shell_imgact(imgp)
 		imgp->args->begin_argv[offset++] = '\0';
 		imgp->args->argc++;
 	}
+	goto common_end;
+new_code:
+#endif
+	/*
+	 * Find the beginning and end of the interpreter_name.  If the
+	 * line does not include any interpreter, or if the name which
+	 * was found is too long, we bail out.
+	 */
+	while (ihp < maxp && ((*ihp == ' ') || (*ihp == '\t')))
+		ihp++;
+	interpb = ihp;
+	while (ihp < maxp && ((*ihp != ' ') && (*ihp != '\t') && (*ihp != '\n')
+	    && (*ihp != '\0')))
+		ihp++;
+	interpe = ihp;
+	if (interpb == interpe)
+		return (ENOEXEC);
+	if ((interpe - interpb) >= MAXSHELLCMDLEN)
+		return (ENAMETOOLONG);
 
 	/*
+	 * Find the beginning of the options (if any), and the end-of-line.
+	 * Then trim the trailing blanks off the value.  Note that some
+	 * other operating systems do *not* trim the trailing whitespace...
+	 */
+	while (ihp < maxp && ((*ihp == ' ') || (*ihp == '\t')))
+		ihp++;
+	optb = ihp;
+	while (ihp < maxp && ((*ihp != '\n') && (*ihp != '\0')))
+		ihp++;
+	opte = ihp;
+	while (--ihp > interpe && ((*ihp == ' ') || (*ihp == '\t')))
+		opte = ihp;
+
+	/*
+	 * We need to "pop" (remove) the present value of arg[0], and "push"
+	 * either two or three new values in the arg[] list.  To do this,
+	 * we first shift all the other values in the `begin_argv' area to
+	 * provide the exact amount of room for the values added.  Set up
+	 * `offset' as the number of bytes to be added to the `begin_argv'
+	 * area, and 'length' as the number of bytes being removed.
+	 */
+	offset = interpe - interpb + 1;			/* interpreter */
+	if (opte != optb)				/* options (if any) */
+		offset += opte - optb + 1;
+	offset += strlen(imgp->args->fname) + 1;	/* fname of script */
+	length = (imgp->args->argc == 0) ? 0 :
+	    strlen(imgp->args->begin_argv) + 1;		/* bytes to delete */
+
+	if (offset - length > imgp->args->stringspace)
+		return (E2BIG);
+
+	bcopy(imgp->args->begin_argv + length, imgp->args->begin_argv + offset,
+	    imgp->args->endp - (imgp->args->begin_argv + length));
+
+	offset -= length;		/* calculate actual adjustment */
+	imgp->args->begin_envv += offset;
+	imgp->args->endp += offset;
+	imgp->args->stringspace -= offset;
+
+	/*
+	 * If there was no arg[0] when we started, then the interpreter_name
+	 * is adding an argument (instead of replacing the arg[0] we started
+	 * with).  And we're always adding an argument when we include the
+	 * full pathname of the original script.
+	 */
+	if (imgp->args->argc == 0)
+		imgp->args->argc = 1;
+	imgp->args->argc++;
+
+	/*
+	 * The original arg[] list has been shifted appropriately.  Copy in
+	 * the interpreter name and options-string.
+	 */
+	length = interpe - interpb;
+	bcopy(interpb, imgp->args->buf, length);
+	*(imgp->args->buf + length) = '\0';
+	offset = length + 1;
+	if (opte != optb) {
+		length = opte - optb;
+		bcopy(optb, imgp->args->buf + offset, length);
+		*(imgp->args->buf + offset + length) = '\0';
+		offset += length + 1;
+		imgp->args->argc++;
+	}
+
+#if KEEP_OLDCODE
+common_end:
+#endif
+	/*
 	 * Finally, add the filename onto the end for the interpreter to
 	 * use and copy the interpreter's name to imgp->interpreter_name
 	 * for exec to use.