Re: [PATCH] ima: fix kexec measurement buffer overflow after failed loads
From: Jérémy Jean
Date: Thu Nov 12 2026 - 07:56:29 EST
On 2026-11-12 03:07, Mimi Zohar wrote:
On Sun, 2026-11-08 at 19:04 +0000, Jérémy Jean wrote:
When kexec_file_load() fails to install a new image, it can leave the
global IMA scratch buffer larger than the IMA segment of the current
image. When that image is executed, ima_update_kexec_buffer() copies
the larger log past the end of its reserved segment. In a test, KASAN
reports it as:
BUG: KASAN: vmalloc-out-of-bounds in ima_update_kexec_buffer+0x265/0x280
Write of size 868426 at addr ffffc9000004e000 by task init/1
...
Call Trace:
...
__asan_memcpy+0x3b/0x60
ima_update_kexec_buffer+0x265/0x280
...
blocking_notifier_call_chain+0x88/0xb0
kernel_restart_prepare+0x18/0x40
kernel_kexec+0xe4/0x130
Limit the dump to the smaller value between the scratch size and the
installed image segment size. Update the IMA mapping only after
machine_kexec_post_load() succeeds.
Fixes: 9f0ec4b16f2b ("ima: kexec: move IMA log copy from kexec load to execute")
Cc: stable@xxxxxxxxxxxxxxx
Signed-off-by: Jérémy Jean <Jeremy.Jean@xxxxxxxxxxxxxxxxx>
Thank you for the report. I agree with Mike that the machine_kexec_post_load()
reordering should be a separate patch.
For the overflow, rather than clamping at execute time, how about not changing
ima_kexec_file.size for an image that isn't installed?
1. Set ima_kexec_file.size only in ima_kexec_post_load(), from
image->ima_buffer_size, and stop ima_free_kexec_file_buf() from
resetting it.
2. Reuse the buffer if size >= segment_size, instead of only if they
are equal, so the buffer is never smaller than size. The buffer no
longer tracks size, and with staging the list can shrink.
Please don't remove the "segment size can't change between kexec load and
execute" comments. They are still correct.
Dear Mimi and Mike,
Thank you both for your feedback. I tried to took them into consideration
for the v2 I have just sent. As requested, I split the fixes into different
commits, and try to address the bigger problem in this new version.
Best regards,
Jérémy