[PATCH v15 08/23] arm64: kexec_file: Fix image->elf_headers memory leak during retry loop

From: Jinjie Ruan

Date: Mon Jun 01 2026 - 06:04:19 EST


Sashiko AI code review pointed out a potential memory leak of
image->elf_headers when load_other_segments() fails on error paths.

In the arm64 kexec_file file-load path, kexec_image.c runs a retry loop
calling kexec_add_buffer() to find a suitable location for the kernel
segment. On each iteration, load_other_segments() is invoked to allocate
and populate alternative segments such as initrd, DTB, and ELF headers.

However, if a placement or allocation failure occurs later in
load_other_segments() (e.g., when adding initrd or dtb), the execution
jumps to the out_err label. While this path restores image->nr_segments
via orig_segments, it returns an error back to the caller without freeing
the previously allocated image->elf_headers vmalloc buffer.

As a result, the retry loop in image_load() unconditionally allocates
new ELF headers on the next iteration and overwrites image->elf_headers,
permanently leaking the memory blocks allocated in previous iterations.

To fix this, decouple the ELF header allocation from the target-seeking
retry loop. Since the contents and size of ELF headers only depend on
the host memory layout and do not change with the kernel's physical
placement, move prepare_elf_headers() completely outside and prior to
the while retry loop in image_load().

And if kexec_add_buffer() for elf headers fails, not need to vfree
headers, because the err path will vfree `image->elf_headers` by calling
arch_kimage_file_post_load_cleanup().

This optimization eliminates redundant memory allocation/deallocation
overhead during kexec placement retries and eradicates the Use-After-Free
and memory leak risk.

Concurrently, remove the prepare_elf_headers() call from inside
load_other_segments() and have it directly reuse the single, pre-allocated
image->elf_headers.

Cc: Catalin Marinas <catalin.marinas@xxxxxxx>
Cc: Will Deacon <will@xxxxxxxxxx>
Cc: Thomas Huth <thuth@xxxxxxxxxx>
Cc: Breno Leitao <leitao@xxxxxxxxxx>
Cc: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
Cc: Yeoreum Yun <yeoreum.yun@xxxxxxx>
Cc: Coiby Xu <coxu@xxxxxxxxxx>
Cc: Baoquan He <bhe@xxxxxxxxxx>
Cc: Kees Cook <kees@xxxxxxxxxx>
Cc: Benjamin Gwin <bgwin@xxxxxxxxxx>
Cc: stable@xxxxxxxxxxxxxxx
Fixes: 108aa503657e ("arm64: kexec_file: try more regions if loading segments fails")
Signed-off-by: Jinjie Ruan <ruanjinjie@xxxxxxxxxx>
---
v15:
- Use image->elf_headers and image->elf_headers_sz instead of adding function
parameters for load_other_segments() to simplify the fix.
---
arch/arm64/include/asm/kexec.h | 1 +
arch/arm64/kernel/kexec_image.c | 16 ++++++++++++++++
arch/arm64/kernel/machine_kexec_file.c | 23 +++++------------------
3 files changed, 22 insertions(+), 18 deletions(-)

diff --git a/arch/arm64/include/asm/kexec.h b/arch/arm64/include/asm/kexec.h
index 892e5bebda95..7ffa2ff5fcfd 100644
--- a/arch/arm64/include/asm/kexec.h
+++ b/arch/arm64/include/asm/kexec.h
@@ -128,6 +128,7 @@ extern int load_other_segments(struct kimage *image,
unsigned long kernel_load_addr, unsigned long kernel_size,
char *initrd, unsigned long initrd_len,
char *cmdline);
+extern int prepare_elf_headers(void **addr, unsigned long *sz);
#endif

#endif /* __ASSEMBLER__ */
diff --git a/arch/arm64/kernel/kexec_image.c b/arch/arm64/kernel/kexec_image.c
index ffcb7f9075e6..424b9527db09 100644
--- a/arch/arm64/kernel/kexec_image.c
+++ b/arch/arm64/kernel/kexec_image.c
@@ -89,6 +89,22 @@ static void *image_load(struct kimage *image,

kernel_segment_number = image->nr_segments;

+#ifdef CONFIG_CRASH_DUMP
+ if (image->type == KEXEC_TYPE_CRASH) {
+ /* load elf core header */
+ unsigned long headers_sz;
+ void *headers;
+
+ ret = prepare_elf_headers(&headers, &headers_sz);
+ if (ret) {
+ pr_err("Preparing elf core header failed\n");
+ return ERR_PTR(ret);
+ }
+ image->elf_headers = headers;
+ image->elf_headers_sz = headers_sz;
+ }
+#endif
+
/*
* The location of the kernel segment may make it impossible to satisfy
* the other segment requirements, so we try repeatedly to find a
diff --git a/arch/arm64/kernel/machine_kexec_file.c b/arch/arm64/kernel/machine_kexec_file.c
index 13c247c28866..4cbb71e1f8ed 100644
--- a/arch/arm64/kernel/machine_kexec_file.c
+++ b/arch/arm64/kernel/machine_kexec_file.c
@@ -40,7 +40,7 @@ int arch_kimage_file_post_load_cleanup(struct kimage *image)
}

#ifdef CONFIG_CRASH_DUMP
-static int prepare_elf_headers(void **addr, unsigned long *sz)
+int prepare_elf_headers(void **addr, unsigned long *sz)
{
struct crash_mem *cmem;
unsigned int nr_ranges;
@@ -105,32 +105,19 @@ int load_other_segments(struct kimage *image,
kbuf.buf_min = kernel_load_addr + kernel_size;

#ifdef CONFIG_CRASH_DUMP
- /* load elf core header */
- void *headers;
- unsigned long headers_sz;
if (image->type == KEXEC_TYPE_CRASH) {
- ret = prepare_elf_headers(&headers, &headers_sz);
- if (ret) {
- pr_err("Preparing elf core header failed\n");
- goto out_err;
- }
-
- kbuf.buffer = headers;
- kbuf.bufsz = headers_sz;
+ kbuf.buffer = image->elf_headers;
+ kbuf.bufsz = image->elf_headers_sz;
kbuf.mem = KEXEC_BUF_MEM_UNKNOWN;
- kbuf.memsz = headers_sz;
+ kbuf.memsz = image->elf_headers_sz;
kbuf.buf_align = SZ_64K; /* largest supported page size */
kbuf.buf_max = ULONG_MAX;
kbuf.top_down = true;

ret = kexec_add_buffer(&kbuf);
- if (ret) {
- vfree(headers);
+ if (ret)
goto out_err;
- }
- image->elf_headers = headers;
image->elf_load_addr = kbuf.mem;
- image->elf_headers_sz = headers_sz;

kexec_dprintk("Loaded elf core header at 0x%lx bufsz=0x%lx memsz=0x%lx\n",
image->elf_load_addr, kbuf.bufsz, kbuf.memsz);
--
2.34.1