4.3bsd/VAX stdio: buffer overruns

mouse at mcgill-vision.UUCP mouse at mcgill-vision.UUCP
Wed Jan 28 19:44:55 AEST 1987


Index:	src/lib/libc/vax/stdio/doprnt.c

Description:
	4.3 stdio on a VAX will, sometimes, grossly overrun its buffer,
	scribbling on whatever happens to follow said buffer.

Repeat-By:
	This is difficult to demonstrate easily, because it depends on
	the output being line buffered and on what happens to follow
	the output buffer in memory.  But....

	#include <stdio.h>
	char buf[BUFSIZ+1];
	main()
	{
	 /* no \n - don't want to flush the buffer */
	 printf("Hello world....");
	}

	Run this under adb.  Examine __iob+14 immediately after the
	printf and notice that the _cnt field is zero even though the
	_ptr has been advanced properly.

	Provided you always use printf, this problem doesn't show
	itself, because _doprnt checks for _ptr > _base+_bufsiz.  Where
	it bites is when you mix printf() and putc() (or fputc()),
	because putc() checks _cnt and assumes _ptr and _base agree
	with it.  Thus, it is possible to overrun _sobuf by up to
	BUFSIZ-1 characters (use printf to nearly fill _sobuf, then you
	can do BUFSIZ more putc()s before it will flush).

Fix:
	I suspect some of the code in _doprnt (see Index: line):

	strout2:		# enter here to force out r2; r0,r1 must be set
		# do some tricks with line buffering (_IOLBF) first
		movl	fdesc,r3
	<six lines omitted here>
		movb	r2,*4(r3)		# line buffered and not buffer full
		incl	4(r3)			#  and not newline
		clrl	(r3)			#  just stuff it and fix _cnt
		incl	nchar			# count the char

	I don't like the way it "fixes" _cnt.  But I don't understand
	this code entirely (not the assembly, I know VAX assembly, I
	mean _doprnt in general), so I'm not sure what to do about it.
	Perhaps the clrl should be decl?

					der Mouse

USA: {ihnp4,decvax,akgua,utzoo,etc}!utcsri!mcgill-vision!mouse
     think!mosart!mcgill-vision!mouse
Europe: mcvax!decvax!utcsri!mcgill-vision!mouse
ARPAnet: think!mosart!mcgill-vision!mouse at harvard.harvard.edu



More information about the Comp.unix.wizards mailing list