lkml.org 
[lkml]   [2019]   [Dec]   [18]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH v4 07/19] KVM: Explicitly free allocated-but-unused dirty bitmap
On Tue, Dec 17, 2019 at 02:51:18PM -0800, Sean Christopherson wrote:
> On Tue, Dec 17, 2019 at 05:24:46PM -0500, Peter Xu wrote:
> > On Tue, Dec 17, 2019 at 12:40:29PM -0800, Sean Christopherson wrote:
> > > Explicitly free an allocated-but-unused dirty bitmap instead of relying
> > > on kvm_free_memslot() if an error occurs in __kvm_set_memory_region().
> > > There is no longer a need to abuse kvm_free_memslot() to free arch
> > > specific resources as arch specific code is now called only after the
> > > common flow is guaranteed to succeed. Arch code can still fail, but
> > > it's responsible for its own cleanup in that case.
> > >
> > > Eliminating the error path's abuse of kvm_free_memslot() paves the way
> > > for simplifying kvm_free_memslot(), i.e. dropping its @dont param.
> > >
> > > Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
> > > ---
> > > virt/kvm/kvm_main.c | 7 ++++---
> > > 1 file changed, 4 insertions(+), 3 deletions(-)
> > >
> > > diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
> > > index d403e93e3028..6b2261a9e139 100644
> > > --- a/virt/kvm/kvm_main.c
> > > +++ b/virt/kvm/kvm_main.c
> > > @@ -1096,7 +1096,7 @@ int __kvm_set_memory_region(struct kvm *kvm,
> > >
> > > slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT);
> > > if (!slots)
> > > - goto out_free;
> > > + goto out_bitmap;
> > > memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
> > >
> > > if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
> > > @@ -1144,8 +1144,9 @@ int __kvm_set_memory_region(struct kvm *kvm,
> > > if (change == KVM_MR_DELETE || change == KVM_MR_MOVE)
> > > slots = install_new_memslots(kvm, as_id, slots);
> > > kvfree(slots);
> > > -out_free:
> > > - kvm_free_memslot(kvm, &new, &old);
> > > +out_bitmap:
> > > + if (new.dirty_bitmap && !old.dirty_bitmap)
> > > + kvm_destroy_dirty_bitmap(&new);
> >
> > What if both the old and new have KVM_MEM_LOG_DIRTY_PAGES set?
> > kvm_free_memslot() did cover that but I see that you explicitly
> > dropped it. Could I ask why? Thanks,
>
> In that case, old.dirty_bitmap == new.dirty_bitmap, i.e. shouldn't be freed
> by this error path since doing so would result in a use-after-free via the
> old memslot.
>
> The kvm_free_memslot() logic is the same, albeit in a very twisted way.

Yes it is. :)

>
> In __kvm_set_memory_region(), @old and @new start with the same dirty_bitmap.
>
> new = old = *slot;
>
> And @new is modified based on KVM_MEM_LOG_DIRTY_PAGES. If LOG_DIRTY_PAGES
> is set in both @new and @old, then both the "if" and "else if" evaluate
> false, i.e. new.dirty_bitmap == old.dirty_bitmap.
>
> /* Allocate/free page dirty bitmap as needed */
> if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
> new.dirty_bitmap = NULL;
> else if (!new.dirty_bitmap) {
> r = kvm_create_dirty_bitmap(&new);
> if (r)
> return r;
> }
>
> Subbing "@free <= @new" and "@dont <= @old" in kvm_free_memslot()
>
> static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
> struct kvm_memory_slot *dont)
> {
> if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
> kvm_destroy_dirty_bitmap(free);
>
>
> yeids this, since @old is obviously non-NULL
>
> if (new.dirty_bitmap != old.dirty_bitmap)
> kvm_destroy_dirty_bitmap(&new);
>
> The dirty_bitmap allocation logic guarantees that new.dirty_bitmap is
> a) NULL (the "if" case")
> b) != old.dirty_bitmap iff old.dirty_bitmap == NULL (the "else if" case)
> c) == old.dirty_bitmap (the implicit "else" case).
>
> kvm_free_memslot() frees @new.dirty_bitmap iff its != @old.dirty_bitmap,
> thus the explicit destroy only needs to check for (b).

Thanks for explaining with such a detail.

Reviewed-by: Peter Xu <peterx@redhat.com>

--
Peter Xu

\
 
 \ /
  Last update: 2019-12-18 17:18    [W:0.050 / U:24.916 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site