Re: Symlinks and Cryogenic Sleep

[ant () NOTATLA DEMON CO UK (Antonomasia)]

Olaf Kirch asked about checking files when you reopen them and
questioned the usefulness of

>       if (lstat(fname, &stb1) >= 0 && S_ISREG(stb1.st_mode)) {
>               fd = open(fname, O_RDWR);
>               if (fd < 0 || fstat(fd, &stb2) < 0
>                || ino_or_dev_mismatch(&stb1, &stb2))
>                       raise_big_stink()
>       } else {
>               /* do the O_EXCL thing */
>       }

Mark A. Heilpern" <heilpern () MINDSPRING COM> and
der Mouse <mouse () RODENTS MONTREAL QC CA> maintain that mortals can only
send a SIGSTOP to their own processes.

When I send a SIGSTOP to a passwd process (uid=me,euid=0) I get:

linux 2.0.36:  stopped
linux 2.2.12:  stopped
OpenBSD 2.5:   stopped

No doubt Olaf selected SIGSTOP in his example because it cannot be handled.

Goetz Babin-Ebell <babinebell () TRUSTCENTER DE> provided some code which
I've not tested but looks as if it will leak open files and will
call fopen(cpFile,"a"); first and lstat() afterwards.  This could lead
to the creation of unintended files at the symlink target.  Only comparison
to S_IFLNK is done, leaving named pipes in the running.
It might also be raced either side of the lstat() call.  fstat() is not used.

While I'm on this I'll mention a code scanner I wrote last year for checking
file races.  It follows a description of an unpublished scanner by Matt Bishop
and Michael Dilger and is demonstrated on sendmail-8.6.10.  In Perl.
http://www.notatla.demon.co.uk/SOFTWARE/SCANNER/scanner-1.0b.tar.gz

Olaf's suggested function ino_or_dev_mismatch(&stb1, &stb2) could be
extended to check the file's owner and group remain unchanged.  This
means even if a file is switched the attacker gains nothing - he has
to replace it with an equivalent file.

--
##############################################################
# Antonomasia   ant () notatla demon co uk                      #
# See http://www.notatla.demon.co.uk/                        #
##############################################################