Move mmu notification mechanism inside mm lock to prevent race condition
in other components which depend on it. The notifier will invalidate
memory range. Depending upon the number of iterations, different memory
ranges would be invalidated.
The following warning would be removed by this patch:
WARNING: CPU: 0 PID: 5067 at arch/x86/kvm/../../../virt/kvm/kvm_main.c:734 kvm_mmu_notifier_change_pte+0x860/0x960 arch/x86/kvm/../../../virt/kvm/kvm_main.c:734
There is no behavioural and performance change with this patch when
there is no component registered with the mmu notifier.
Fixes: 52526ca7fdb9 ("fs/proc/task_mmu: implement IOCTL to get and optionally clear info about PTEs")
Reported-by: [email protected]
Link: https://lore.kernel.org/all/[email protected]/
Cc: Sean Christopherson <[email protected]>
Cc: [email protected]
Signed-off-by: Muhammad Usama Anjum <[email protected]>
---
fs/proc/task_mmu.c | 22 ++++++++++++----------
1 file changed, 12 insertions(+), 10 deletions(-)
diff --git a/fs/proc/task_mmu.c b/fs/proc/task_mmu.c
index 62b16f42d5d2..56c2e7357494 100644
--- a/fs/proc/task_mmu.c
+++ b/fs/proc/task_mmu.c
@@ -2448,13 +2448,6 @@ static long do_pagemap_scan(struct mm_struct *mm, unsigned long uarg)
if (ret)
return ret;
- /* Protection change for the range is going to happen. */
- if (p.arg.flags & PM_SCAN_WP_MATCHING) {
- mmu_notifier_range_init(&range, MMU_NOTIFY_PROTECTION_VMA, 0,
- mm, p.arg.start, p.arg.end);
- mmu_notifier_invalidate_range_start(&range);
- }
-
for (walk_start = p.arg.start; walk_start < p.arg.end;
walk_start = p.arg.walk_end) {
long n_out;
@@ -2467,8 +2460,20 @@ static long do_pagemap_scan(struct mm_struct *mm, unsigned long uarg)
ret = mmap_read_lock_killable(mm);
if (ret)
break;
+
+ /* Protection change for the range is going to happen. */
+ if (p.arg.flags & PM_SCAN_WP_MATCHING) {
+ mmu_notifier_range_init(&range, MMU_NOTIFY_PROTECTION_VMA, 0,
+ mm, walk_start, p.arg.end);
+ mmu_notifier_invalidate_range_start(&range);
+ }
+
ret = walk_page_range(mm, walk_start, p.arg.end,
&pagemap_scan_ops, &p);
+
+ if (p.arg.flags & PM_SCAN_WP_MATCHING)
+ mmu_notifier_invalidate_range_end(&range);
+
mmap_read_unlock(mm);
n_out = pagemap_scan_flush_buffer(&p);
@@ -2494,9 +2499,6 @@ static long do_pagemap_scan(struct mm_struct *mm, unsigned long uarg)
if (pagemap_scan_writeback_args(&p.arg, uarg))
ret = -EFAULT;
- if (p.arg.flags & PM_SCAN_WP_MATCHING)
- mmu_notifier_invalidate_range_end(&range);
-
kfree(p.vec_buf);
return ret;
}
--
2.42.0
On Tue, Jan 09, 2024, Muhammad Usama Anjum wrote:
> Move mmu notification mechanism inside mm lock to prevent race condition
> in other components which depend on it. The notifier will invalidate
> memory range. Depending upon the number of iterations, different memory
> ranges would be invalidated.
>
> The following warning would be removed by this patch:
> WARNING: CPU: 0 PID: 5067 at arch/x86/kvm/../../../virt/kvm/kvm_main.c:734 kvm_mmu_notifier_change_pte+0x860/0x960 arch/x86/kvm/../../../virt/kvm/kvm_main.c:734
>
> There is no behavioural and performance change with this patch when
> there is no component registered with the mmu notifier.
>
> Fixes: 52526ca7fdb9 ("fs/proc/task_mmu: implement IOCTL to get and optionally clear info about PTEs")
> Reported-by: [email protected]
> Link: https://lore.kernel.org/all/[email protected]/
> Cc: Sean Christopherson <[email protected]>
> Cc: [email protected]
> Signed-off-by: Muhammad Usama Anjum <[email protected]>
> ---
> fs/proc/task_mmu.c | 22 ++++++++++++----------
> 1 file changed, 12 insertions(+), 10 deletions(-)
>
> diff --git a/fs/proc/task_mmu.c b/fs/proc/task_mmu.c
> index 62b16f42d5d2..56c2e7357494 100644
> --- a/fs/proc/task_mmu.c
> +++ b/fs/proc/task_mmu.c
> @@ -2448,13 +2448,6 @@ static long do_pagemap_scan(struct mm_struct *mm, unsigned long uarg)
> if (ret)
> return ret;
>
> - /* Protection change for the range is going to happen. */
> - if (p.arg.flags & PM_SCAN_WP_MATCHING) {
> - mmu_notifier_range_init(&range, MMU_NOTIFY_PROTECTION_VMA, 0,
> - mm, p.arg.start, p.arg.end);
> - mmu_notifier_invalidate_range_start(&range);
> - }
> -
> for (walk_start = p.arg.start; walk_start < p.arg.end;
> walk_start = p.arg.walk_end) {
> long n_out;
Nit, might be worth moving
struct mmu_notifier_range range;
inside the loop to guard against stale usage, but that's definitely optional.
Reviewed-by: Sean Christopherson <[email protected]>
On Tue, 9 Jan 2024 08:28:06 -0800 Sean Christopherson <[email protected]> wrote:
> > - /* Protection change for the range is going to happen. */
> > - if (p.arg.flags & PM_SCAN_WP_MATCHING) {
> > - mmu_notifier_range_init(&range, MMU_NOTIFY_PROTECTION_VMA, 0,
> > - mm, p.arg.start, p.arg.end);
> > - mmu_notifier_invalidate_range_start(&range);
> > - }
> > -
> > for (walk_start = p.arg.start; walk_start < p.arg.end;
> > walk_start = p.arg.walk_end) {
> > long n_out;
>
> Nit, might be worth moving
>
> struct mmu_notifier_range range;
>
> inside the loop to guard against stale usage, but that's definitely optional.
Yes, I think that's nicer.
--- a/fs/proc/task_mmu.c~fs-proc-task_mmu-move-mmu-notification-mechanism-inside-mm-lock-fix
+++ a/fs/proc/task_mmu.c
@@ -2432,7 +2432,6 @@ static long pagemap_scan_flush_buffer(st
static long do_pagemap_scan(struct mm_struct *mm, unsigned long uarg)
{
- struct mmu_notifier_range range;
struct pagemap_scan_private p = {0};
unsigned long walk_start;
size_t n_ranges_out = 0;
@@ -2450,6 +2449,7 @@ static long do_pagemap_scan(struct mm_st
for (walk_start = p.arg.start; walk_start < p.arg.end;
walk_start = p.arg.walk_end) {
+ struct mmu_notifier_range range;
long n_out;
if (fatal_signal_pending(current)) {
_
I'm surprised this code doesn't generate a might-be-used-uninitialized
warning. I guess gcc got smarter.