Thank you for your contribution! Sashiko AI review found 2 potential issue(s) 
to consider:

Pre-existing issues:
- [High] The integer overflow vulnerability fixed in user-space still exists in 
the kernel's bootconfig loader (`init/main.c`), allowing a crafted initrd to 
bypass bounds checks and cause an out-of-bounds read and kernel panic.
- [High] Truncation of `off_t` to `int` causes a heap buffer overflow and 
underflow when processing large standalone bootconfig files.
--

commit 1cf2181ef71cd516068d73bd6993bef0811e1b9d
Author: Masami Hiramatsu (Google) <[email protected]>
Subject: tools/bootconfig: Fix integer overflow in size check

This patch fixes an integer overflow vulnerability in load_xbc_from_initrd() 
within the bootconfig tool. On 32-bit systems, a maliciously large bootconfig 
size could wrap around, bypassing bounds checks and causing memory corruption. 
The fix rejects overly large sizes upfront and verifies against stat.st_size 
in a way that avoids underflow.

Closes: https://lore.kernel.org/all/[email protected]/

> diff --git a/tools/bootconfig/main.c b/tools/bootconfig/main.c
> index 7dc9fff9b6370..bf6a4e2163154 100644
> --- a/tools/bootconfig/main.c
> +++ b/tools/bootconfig/main.c
> @@ -218,7 +218,8 @@ static int load_xbc_from_initrd(int fd, char **buf)
>       csum = le32toh(csum);
>  
>       /* Wrong size error  */
> -     if (stat.st_size < size + BOOTCONFIG_FOOTER_SIZE) {
> +     if (size > XBC_DATA_MAX ||
> +         size > stat.st_size - BOOTCONFIG_FOOTER_SIZE) {

[Severity: High]
This isn't a bug introduced by this patch, but the exact same integer
overflow vulnerability appears to exist in the kernel's bootconfig loader.

On 32-bit systems, a crafted initrd with a huge bootconfig size
(e.g., 0xFFFFFFFF) can cause the pointer arithmetic to wrap around:

init/main.c:get_boot_config_from_initrd() {
    ...
        data = ((void *)hdr) - size;
        if ((unsigned long)data < initrd_start) {
                pr_err("bootconfig size %d is greater than initrd size %ld\n",
    ...
}

Because data wraps around, the bounds check
(unsigned long)data < initrd_start evaluates to false.

The kernel then calls xbc_calc_checksum(data, size), attempting to read
4GB of memory, which hits unmapped pages and triggers a fatal page fault.

Could we apply a similar bounds check to get_boot_config_from_initrd() in
init/main.c?


[Severity: High]
This is a pre-existing issue, but does load_xbc_file() have a truncation
issue that can cause a heap buffer overflow?

When processing large standalone bootconfig files, load_xbc_file() passes
the 64-bit off_t file size into load_xbc_fd(), which expects a 32-bit int:

tools/bootconfig/main.c:load_xbc_file() {
    ...
        ret = load_xbc_fd(fd, buf, stat.st_size);
    ...
}

tools/bootconfig/main.c:load_xbc_fd(int fd, char **buf, int size) {
    ...
        *buf = malloc(size + 1);
        if (!*buf)
                return -ENOMEM;

        ret = read(fd, *buf, size);
        if (ret < 0)
                return -errno;
        (*buf)[size] = '\0';
    ...
}

If a user or script runs bootconfig -a on an extremely large file
(exactly 4GB - 1 byte in size), stat.st_size truncates to -1 when
passed as the int size parameter.

malloc(-1 + 1) correctly evaluates to malloc(0) and succeeds.
read(fd, *buf, -1) is then called, where -1 implicitly casts to SIZE_MAX,
causing the kernel to write up to ~2GB of file data into the 0-byte buffer.

Finally, (*buf)[size] = '\0' writes a null byte at index -1, corrupting
heap metadata.

Should load_xbc_fd() take a size_t instead of an int?

>               pr_err("bootconfig size is too big\n");
>               return -E2BIG;
>       }

-- 
Sashiko AI review ยท 
https://sashiko.dev/#/patchset/178900114498.187028.9558747732041210116.stgit@devnote2?part=1

Reply via email to