[PATCHv6 12/13] tools/kexec: Introduce a bpf-prog to parse zboot image format

[bot+bpf-ci at kernel.org]

> diff --git a/tools/kexec/zboot_parser_bpf.c b/tools/kexec/zboot_parser_bpf.c
> new file mode 100644
> index 000000000000..54c4b762b332
> --- /dev/null
> +++ b/tools/kexec/zboot_parser_bpf.c

[ ... ]

> +/* see drivers/firmware/efi/libstub/zboot-header.S */
> +struct linux_pe_zboot_header {
> +	unsigned int mz_magic;
> +	char image_type[4];
> +	unsigned int payload_offset;
> +	unsigned int payload_size;
> +	unsigned int reserved[2];
> +	char comp_type[4];
> +	unsigned int linux_pe_magic;
> +	unsigned int pe_header_offset;
> +} __attribute__((packed));

[ ... ]

> +	unsigned int payload_offset = zboot_header->payload_offset;
> +	unsigned int payload_size = zboot_header->payload_size;
> +	bpf_printk("zboot image payload offset=0x%x, size=0x%x\n", payload_offset, payload_size);
> +	/* sane check */
> +	if (payload_size > image_sz) {
> +		bpf_ringbuf_discard(buf, BPF_RB_NO_WAKEUP);
> +		bpf_printk("Invalid zboot image payload offset and size\n");
> +		return 0;
> +	}

Should this bounds check also verify that payload_offset + payload_size
does not exceed image_sz? The current check only validates that payload_size
fits, but if payload_offset is large (e.g., near image_sz), the read below
could extend past the end of the kernel buffer.

For example, with payload_offset = image_sz - 10 and payload_size = 100,
the check "payload_size > image_sz" passes, but the bpf_probe_read below
would read 90 bytes past the valid buffer boundary.

> +	unsigned int max_payload = MAX_RECORD_SIZE - sizeof(struct cmd_hdr);
> +	if (payload_size >= max_payload) {
> +		bpf_ringbuf_discard(buf, BPF_RB_NO_WAKEUP);
> +		bpf_printk("Err: payload_size > MAX_RECORD_SIZE\n");
> +		return 0;
> +	}
> +	void *dst = (void *)buf + sizeof(struct cmd_hdr);
> +	/* Overwrite buf */
> +	struct cmd_hdr *cmd = (struct cmd_hdr *)buf;
> +	cmd->cmd = KEXEC_BPF_CMD_DECOMPRESS;
> +	cmd->subcmd = KEXEC_BPF_SUBCMD_KERNEL;
> +	/* 4 bytes original size is appended after vmlinuz.bin */
> +	cmd->payload_len = payload_size - 4;
> +	bpf_probe_read(dst, payload_size, context->kernel + payload_offset);
                                                           ^^^^^^^^^^^^^^^

This is where the out-of-bounds read could occur if payload_offset +
payload_size > image_sz. While bpf_probe_read is safe and will not crash
(it uses copy_from_kernel_nofault internally), it could read from adjacent
kernel memory if the zboot header contains invalid offset/size values.


---
AI reviewed your patch. Please fix the bug or email reply why it's not a bug.
See: https://github.com/kernel-patches/vmtest/blob/master/ci/claude/README.md

CI run summary: https://github.com/kernel-patches/bpf/actions/runs/21147860407