2007-02-20 13:27:24

by Petr Tesařík

[permalink] [raw]

Subject: [PATCH] free pages in remove_arg_zero()

---

Hello,

I've found a bug when executing scripts:

When a script is loaded, load_script() replaces argv[0] with the
name of the interpreter and the filename passed to the exec syscall.
However, there is no guarantee that the length of the interpreter
name plus the length of the filename is greater than the length of
the original argv[0]. If the difference happens to cross a page boundary,
setup_arg_pages() will call install_arg_page() with an address outside
the VMA.

Therefore, remove_arg_zero() must free all pages which would be unused
after the argument is removed.

Signed-off-by: Petr Tesarik <[email protected]>

--- linux-sles9.orig/fs/exec.c
+++ linux-sles9/fs/exec.c
@@ -1000,6 +1000,8 @@ void remove_arg_zero(struct linux_binprm
continue;
offset = 0;
kunmap_atomic(kaddr, KM_USER0);
+ __free_page(page);
+ bprm->page[bprm->p/PAGE_SIZE - 1] = NULL;
inside:
page = bprm->page[bprm->p/PAGE_SIZE];
kaddr = kmap_atomic(page, KM_USER0);

---

Attachments:

signature.asc (189.00 B)
Toto je digit?ln? podepsan? ??st zpr?vy

2007-02-21 23:42:09

by Andrew Morton

[permalink] [raw]

Subject: Re: [PATCH] free pages in remove_arg_zero()

---

On Tue, 20 Feb 2007 14:29:04 +0100
Petr Tesa____k <[email protected]> wrote:

> I've found a bug when executing scripts:

You've found more than that.

> When a script is loaded, load_script() replaces argv[0] with the
> name of the interpreter and the filename passed to the exec syscall.
> However, there is no guarantee that the length of the interpreter
> name plus the length of the filename is greater than the length of
> the original argv[0]. If the difference happens to cross a page boundary,
> setup_arg_pages() will call install_arg_page() with an address outside
> the VMA.
>
> Therefore, remove_arg_zero() must free all pages which would be unused
> after the argument is removed.
>
> Signed-off-by: Petr Tesarik <[email protected]>
>
> --- linux-sles9.orig/fs/exec.c
> +++ linux-sles9/fs/exec.c
> @@ -1000,6 +1000,8 @@ void remove_arg_zero(struct linux_binprm
> continue;
> offset = 0;
> kunmap_atomic(kaddr, KM_USER0);
> + __free_page(page);
> + bprm->page[bprm->p/PAGE_SIZE - 1] = NULL;
> inside:
> page = bprm->page[bprm->p/PAGE_SIZE];
> kaddr = kmap_atomic(page, KM_USER0);

I am not surprised that remove_arg_zero() is buggy. Let us look at the
stupid thing:

void remove_arg_zero(struct linux_binprm *bprm)
{
if (bprm->argc) {
unsigned long offset;
char * kaddr;
struct page *page;

offset = bprm->p % PAGE_SIZE;
goto inside;

while (bprm->p++, *(kaddr+offset++)) {
if (offset != PAGE_SIZE)
continue;
offset = 0;
kunmap_atomic(kaddr, KM_USER0);
inside:
page = bprm->page[bprm->p/PAGE_SIZE];
kaddr = kmap_atomic(page, KM_USER0);
}
kunmap_atomic(kaddr, KM_USER0);
bprm->argc--;
}
}

I mean.... what the hell?

As you appear to have managed to work out what the sorry thing is trying to
do, would you have time to simply rip it out and completely rewrite it,
including a nice comment telling the world what this function's function is?

Because what we have there is beyond repairing.

Thanks.