write_comp_data() reads the comparison record count from area[0] and uses
it to index the coverage buffer:

	area = (u64 *)t->kcov_area;
	max_pos = t->kcov_size * sizeof(unsigned long);

	count = READ_ONCE(area[0]);

	/* Every record is KCOV_WORDS_PER_CMP 64-bit words. */
	start_index = 1 + count * KCOV_WORDS_PER_CMP;
	end_pos = (start_index + KCOV_WORDS_PER_CMP) * sizeof(u64);
	if (likely(end_pos <= max_pos)) {

The buffer is mmap'd writable into the collecting process, so count is
under its control and end_pos <= max_pos is its only bound. A count that
wraps the u64 multiply leaves end_pos below max_pos, so the check passes
while the record store lands 24 bytes before the buffer, in the unmapped
vmalloc guard page, and faults:

  BUG: unable to handle page fault for address: ffa0000000b60fe8
  #PF: supervisor write access in kernel mode
  #PF: error_code(0x0002) - not-present page
  Oops: 0002 [#1] SMP KASAN NOPTI
  RIP: 0010:write_comp_data+0x7e/0xa0
  ...
  Kernel panic - not syncing: Fatal exception

Bound count first: only max_pos / (sizeof(u64) * KCOV_WORDS_PER_CMP)
records fit, so a larger count is not a valid index and is dropped.
kcov_move_area() bounds the same untrusted count this way, and no count
the end_pos <= max_pos check accepts reaches that limit, so no valid
record is lost.

Fixes: ded97d2c2b2c ("kcov: support comparison operands collection")
Cc: stable@vger.kernel.org
Assisted-by: Hawkeye:GLM-5.3-flash
Assisted-by: Qoder:Qwen3.8-Max
Signed-off-by: Fang Xieyan <fangxy@xiaopeng.com>
---

Found by reading write_comp_data() in kernel/kcov.c. kcov_mmap() only sets
VM_DONTEXPAND, so a process that maps the coverage buffer PROT_WRITE controls
area[0] and its record count, which write_comp_data() trusts. kcov is a
root-only debugfs file (debugfs_create_file_unsafe("kcov", 0600, ...)) and
write_comp_data() exists only under CONFIG_KCOV_ENABLE_COMPARISONS, so this is
a local, debug-kernel robustness fix: the process corrupts its own buffer and
the kernel oopses. It crosses no privilege boundary.

Reproducer: open /sys/kernel/debug/kcov, KCOV_ENABLE with KCOV_TRACE_CMP, mmap
the buffer PROT_WRITE, store 0x1fffffffffffffff into area[0], then execute a
comparison so the callback fires. Unpatched, the wrapped end_pos passes the
end_pos <= max_pos check and the store faults; patched, the record is dropped
and collection continues with rc=0. area[0] keeps its corrupt value, so later
records are dropped too until userspace resets it, as with a full buffer.

Both cases ran on 704340f1cd0d (9 commits past v7.3-rc3): x86_64 defconfig
plus CONFIG_KCOV=y, CONFIG_KCOV_ENABLE_COMPARISONS=y and CONFIG_KASAN_GENERIC
(with CONFIG_KASAN_VMALLOC=y), gcc 13.2.0, QEMU under TCG; the unpatched and
patched kernels use byte-identical .config and differ only by this patch.
write_comp_data() is notrace and kcov.c is not KASAN-instrumented, so the
out-of-bounds store is a bare #PF on the vmalloc guard page, not a KASAN
report; KASAN shows up only as a build flag in the Oops line.

The fault address is not reproducible byte for byte: the kcov area is
vmalloc'd, so its page moves between boots (the quoted splat faulted at
ffa0000000b60fe8, a later re-run at ffa0000000b10fe8). The invariant is the
low 12 bits ...fe8 - the store always lands 24 bytes below the page-aligned
buffer, into the guard page.

 kernel/kcov.c | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/kernel/kcov.c b/kernel/kcov.c
index 35420f0..54eaae9 100644
--- a/kernel/kcov.c
+++ b/kernel/kcov.c
@@ -253,6 +253,14 @@ static void notrace write_comp_data(u64 type, u64 arg1, u64 arg2, u64 ip)
 
 	count = READ_ONCE(area[0]);
 
+	/*
+	 * area[0] is writable by the collecting process, so count cannot be
+	 * trusted. Bound it to the records that fit, as kcov_move_area()
+	 * does, so the end_pos multiply below cannot wrap past its check.
+	 */
+	if (count >= max_pos / (sizeof(u64) * KCOV_WORDS_PER_CMP))
+		return;
+
 	/* Every record is KCOV_WORDS_PER_CMP 64-bit words. */
 	start_index = 1 + count * KCOV_WORDS_PER_CMP;
 	end_pos = (start_index + KCOV_WORDS_PER_CMP) * sizeof(u64);
-- 
2.50.1 (Apple Git-155)

On Thu, Sep 17, 2026 at 12:43 PM Fang Xieyan <fangxy@xiaopeng.com> wrote:
>
> write_comp_data() reads the comparison record count from area[0] and uses
> it to index the coverage buffer:
>
>         area = (u64 *)t->kcov_area;
>         max_pos = t->kcov_size * sizeof(unsigned long);
>
>         count = READ_ONCE(area[0]);
>
>         /* Every record is KCOV_WORDS_PER_CMP 64-bit words. */
>         start_index = 1 + count * KCOV_WORDS_PER_CMP;
>         end_pos = (start_index + KCOV_WORDS_PER_CMP) * sizeof(u64);
>         if (likely(end_pos <= max_pos)) {
>
> The buffer is mmap'd writable into the collecting process, so count is
> under its control and end_pos <= max_pos is its only bound. A count that
> wraps the u64 multiply leaves end_pos below max_pos, so the check passes
> while the record store lands 24 bytes before the buffer, in the unmapped
> vmalloc guard page, and faults:
>
>   BUG: unable to handle page fault for address: ffa0000000b60fe8
>   #PF: supervisor write access in kernel mode
>   #PF: error_code(0x0002) - not-present page
>   Oops: 0002 [#1] SMP KASAN NOPTI
>   RIP: 0010:write_comp_data+0x7e/0xa0
>   ...
>   Kernel panic - not syncing: Fatal exception
>
> Bound count first: only max_pos / (sizeof(u64) * KCOV_WORDS_PER_CMP)
> records fit, so a larger count is not a valid index and is dropped.
> kcov_move_area() bounds the same untrusted count this way, and no count
> the end_pos <= max_pos check accepts reaches that limit, so no valid
> record is lost.
>
> Fixes: ded97d2c2b2c ("kcov: support comparison operands collection")
> Cc: stable@vger.kernel.org
> Assisted-by: Hawkeye:GLM-5.3-flash
> Assisted-by: Qoder:Qwen3.8-Max
> Signed-off-by: Fang Xieyan <fangxy@xiaopeng.com>

Reviewed-by: Alexander Potapenko <glider@google.com>

I was reviewing
https://lore.kernel.org/all/20260915033345.904567-1-eadavis@sina.com/
yesterday, but this patch supersedes Edward's.