On August 15, 2014 11:51:53 PM CEST, Paul de Weerd <we...@weirdnet.nl> wrote:
>On Fri, Aug 15, 2014 at 06:55:49PM +0200, Alexander Hall wrote:
>| On 08/15/14 16:22, Paul de Weerd wrote:
>| >On Fri, Aug 15, 2014 at 04:07:21PM +0200, Alexander Hall wrote:
>| >| On August 15, 2014 2:04:56 PM CEST, Theo de Raadt
><dera...@cvs.openbsd.org> wrote:
>| >| >> Is it safe to generate some randomness in
>/tftpboot/etc/random.seed
>| >| >for
>| >| >> clients that PXE boot?
>| >| >
>| >| >I do not even know if that file will be read... is it?
>| >|
>| >| IIRC, it is tried but deemed unsafe (0555) and therefore isn't
>used, but causes a warning. Maybe it had changed since.
>| >
>| >What do you mean? You don't get permissions with a TFTP transfer.
>If
>| >the tftpd can read the file, it'll send it to you. If it can't, it
>| >won't (surprise, surprise ;).
>
>Ah, I see what you mean. All files transferred via tftp are
>considered to have 0444 permissions by the requesting process (quite a
>sane approach) and therefore not accepted (not so sure about that part
>in this case).
>
>I'm guessing the reasoning not to trust world writable files is to
>prevent attackers from putting with known contents into the mix. I'm
>not sure if that's any worse than not having anything at all (and the
>contents of that buffer thereby being predictable, perhaps? not even
>sure whether that's true...)
>
>At any rate, this changes that to allow world readable files (still
>not taking world writable files). We can't check S_IWOTH over tftp,
>we should probably assume 0777 for files transferred that way. But,
>if you're trusting the kernel you're getting over tftp, then why the
>hell are you not trusting random.seed? That attacker that could maybe
>influence your randomness would NEVER touch your kernel to ensure it
>only produces well known (to them) randomness. That would be way too
>easy...
>
>Note that this diff doesn't fix the repeated attempts at loading the
>seed file. Still don't grok what's going on there.
I
Please remember that this change isn't limited to tftp, but affects all
potential ways we fetch the seed. Either way, I think it's safe enough, and I'm
all for this change, so ok halex@ unless someone finds an obvious flaw.
/Alexander
>
>Index: boot.c
>===================================================================
>RCS file: /cvs/src/sys/stand/boot/boot.c,v
>retrieving revision 1.43
>diff -u -p -r1.43 boot.c
>--- boot.c 19 Feb 2014 22:02:15 -0000 1.43
>+++ boot.c 15 Aug 2014 21:41:01 -0000
>@@ -153,7 +153,7 @@ loadrandom(char *name, char *buf, size_t
> }
> if (fstat(fd, &sb) == -1 ||
> sb.st_uid != 0 ||
>- (sb.st_mode & (S_IWOTH|S_IROTH)))
>+ (sb.st_mode & (S_IWOTH)))
> goto fail;
> (void) read(fd, buf, buflen);
> fail:
>
>| From sys/lib/libsa/tftp.c:
>|
>| ...
>| int
>| tftp_stat(struct open_file *f, struct stat *sb)
>| {
>| struct tftp_handle *tftpfile;
>| tftpfile = (struct tftp_handle *) f->f_fsdata;
>|
>| sb->st_mode = 0444;
>| ...
>|
>| Ok, it wasn't 0555, but 0444. Close enough.
>|
>| Then, in sys/stand/boot/boot.c:
>|
>| ...
>| void
>| loadrandom(char *name, char *buf, size_t buflen)
>| {
>| ...
>| if (fstat(fd, &sb) == -1 ||
>| sb.st_uid != 0 ||
>| (sb.st_mode & (S_IWOTH|S_IROTH)))
>| goto fail;
>| ...
>|
>| So, actually, if the file exists, it will not emit a warning (like it
>does
>| if it's non-existent), but silently ignore it. :-)
>|
>| /Alexander
>|
