The patch solves the following problem: file system specific routines
involved in ordinary routine writeback process BUG_ON page_buffers()
because a page goes to writeback without buffer-heads attached.
The way how kvm_set_pfn_dirty calls SetPageDirty works only for anon
mappings. For file mappings it is obviously incorrect - there page_mkwrite
must be called. It's not easy to add page_mkwrite call to kvm_set_pfn_dirty
because there is no universal way to find vma by pfn. But actually
SetPageDirty may be simply skipped in those cases. Below is a
justification.
When guest modifies the content of a page with file mapping, kernel kvm
makes the page dirty by the following call-path:
vmx_handle_exit ->
handle_ept_violation ->
__get_user_pages ->
page_mkwrite ->
SetPageDirty
Since then, the page is dirty from both guest and host point of view. Then
the host makes writeback and marks the page as write-protected. So any
further write from the guest triggers call-path above again.
So, for file mappings, it's not possible to have new data written to a page
inside the guest w/o corresponding SetPageDirty on the host.
This makes explicit SetPageDirty from kvm_set_pfn_dirty redundant.
Signed-off-by: Maxim Patlasov <mpatlasov@xxxxxxxxxxxxx>
---
virt/kvm/kvm_main.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
index a11cfd2..5a7d3fa 100644
--- a/virt/kvm/kvm_main.c
+++ b/virt/kvm/kvm_main.c
@@ -1582,7 +1582,8 @@ void kvm_set_pfn_dirty(kvm_pfn_t pfn)
if (!kvm_is_reserved_pfn(pfn)) {
struct page *page = pfn_to_page(pfn);
- if (!PageReserved(page))
+ if (!PageReserved(page) &&
+ (!page->mapping || PageAnon(page)))
SetPageDirty(page);
}
}
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
