On Mon, Jul 29, 2024 at 2:05 PM Barry Song <21cnbao@xxxxxxxxx> wrote:
>
> On Mon, Jul 29, 2024 at 3:42 PM Jason Wang <jasowang@xxxxxxxxxx> wrote:
> >
> > On Thu, Jul 25, 2024 at 3:00 PM Barry Song <21cnbao@xxxxxxxxx> wrote:
> > >
> > > On Thu, Jul 25, 2024 at 6:08 PM Michal Hocko <mhocko@xxxxxxxx> wrote:
> > > >
> > > > On Thu 25-07-24 10:50:45, Barry Song wrote:
> > > > > On Thu, Jul 25, 2024 at 12:27 AM Michal Hocko <mhocko@xxxxxxxx> wrote:
> > > > > >
> > > > > > On Wed 24-07-24 20:55:40, Barry Song wrote:
> > > > [...]
> > > > > > > diff --git a/drivers/vdpa/vdpa_user/iova_domain.c b/drivers/vdpa/vdpa_user/iova_domain.c
> > > > > > > index 791d38d6284c..eff700e5f7a2 100644
> > > > > > > --- a/drivers/vdpa/vdpa_user/iova_domain.c
> > > > > > > +++ b/drivers/vdpa/vdpa_user/iova_domain.c
> > > > > > > @@ -287,28 +287,44 @@ void vduse_domain_remove_user_bounce_pages(struct vduse_iova_domain *domain)
> > > > > > > {
> > > > > > > struct vduse_bounce_map *map;
> > > > > > > unsigned long i, count;
> > > > > > > + struct page **pages = NULL;
> > > > > > >
> > > > > > > write_lock(&domain->bounce_lock);
> > > > > > > if (!domain->user_bounce_pages)
> > > > > > > goto out;
> > > > > > > -
> > > > > > > count = domain->bounce_size >> PAGE_SHIFT;
> > > > > > > + write_unlock(&domain->bounce_lock);
> > > > > > > +
> > > > > > > + pages = kmalloc_array(count, sizeof(*pages), GFP_KERNEL | __GFP_NOFAIL);
> > > > > > > + for (i = 0; i < count; i++)
> > > > > > > + pages[i] = alloc_page(GFP_KERNEL | __GFP_NOFAIL);
> > > > > >
> > > > > > AFAICS vduse_domain_release calls this function with
> > > > > > spin_lock(&domain->iotlb_lock) so dropping &domain->bounce_lock is not
> > > > > > sufficient.
> > > > >
> > > > > yes. this is true:
> > > > >
> > > > > static int vduse_domain_release(struct inode *inode, struct file *file)
> > > > > {
> > > > > struct vduse_iova_domain *domain = file->private_data;
> > > > >
> > > > > spin_lock(&domain->iotlb_lock);
> > > > > vduse_iotlb_del_range(domain, 0, ULLONG_MAX);
> > > > > vduse_domain_remove_user_bounce_pages(domain);
> > > > > vduse_domain_free_kernel_bounce_pages(domain);
> > > > > spin_unlock(&domain->iotlb_lock);
> > > > > put_iova_domain(&domain->stream_iovad);
> > > > > put_iova_domain(&domain->consistent_iovad);
> > > > > vhost_iotlb_free(domain->iotlb);
> > > > > vfree(domain->bounce_maps);
> > > > > kfree(domain);
> > > > >
> > > > > return 0;
> > > > > }
> > > > >
> > > > > This is quite a pain. I admit I don't have knowledge of this driver, and I don't
> > > > > think it's safe to release two locks and then reacquire them. The situation is
> > > > > rather complex. Therefore, I would prefer if the VDPA maintainers could
> > > > > take the lead in implementing a proper fix.
> > > >
> > > > Would it be possible to move all that work to a deferred context?
> > >
> > > My understanding is that we need to be aware of both the iotlb_lock and
> > > bounce_lock to implement the correct changes. As long as we still need
> > > to acquire these two locks in a deferred context, there doesn't seem to
> > > be any difference.
> > >
> > > I can do the memory pre-allocation before spin_lock(&domain->iotlb_lock),
> > > but I have no knowledge whether the "count" will change after I make
> > > the preallocation.
> > >
> > > diff --git a/drivers/vdpa/vdpa_user/iova_domain.c
> > > b/drivers/vdpa/vdpa_user/iova_domain.c
> > > index 791d38d6284c..7ec87ef33d42 100644
> > > --- a/drivers/vdpa/vdpa_user/iova_domain.c
> > > +++ b/drivers/vdpa/vdpa_user/iova_domain.c
> > > @@ -544,9 +544,12 @@ static int vduse_domain_release(struct inode
> > > *inode, struct file *file)
> > > {
> > > struct vduse_iova_domain *domain = file->private_data;
> > >
> > > + struct page **pages;
> > > + spin_lock(&domain->iotlb_lock); maybe also + bounce_lock?
> > > + count = domain->bounce_size >> PAGE_SHIFT;
> > > + spin_unlock(&domain->iotlb_lock);
> >
> > We probably don't need any lock here as bounce_size won't be changed .
> >
> > > +
> > > + preallocate_count_pages(pages, count);
> > > +
> > > ....
> > > spin_lock(&domain->iotlb_lock);
> > > vduse_iotlb_del_range(domain, 0, ULLONG_MAX);
> > > - vduse_domain_remove_user_bounce_pages(domain);
> > > + vduse_domain_remove_user_bounce_pages(domain, pages);
> > > vduse_domain_free_kernel_bounce_pages(domain);
> > > spin_unlock(&domain->iotlb_lock);
> > > put_iova_domain(&domain->stream_iovad);
> >
> > This seems to work.
>
> Thanks, Jason. I personally have no knowledge of vDPA. Could you please help
> review and test the patch below?
>
> From 1f3cae091159bfcaffdb4a999a4a8e37db2eacf1 Mon Sep 17 00:00:00 2001
> From: Barry Song <v-songbaohua@xxxxxxxx>
> Date: Wed, 24 Jul 2024 20:55:40 +1200
> Subject: [PATCH RFC v2] vpda: try to fix the potential crash due to misusing
> __GFP_NOFAIL
> MIME-Version: 1.0
> Content-Type: text/plain; charset=UTF-8
> Content-Transfer-Encoding: 8bit
>
> mm doesn't support non-blockable __GFP_NOFAIL allocation. Because
> __GFP_NOFAIL without direct reclamation may just result in a busy
> loop within non-sleepable contexts.
>
> static inline struct page *
> __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
> struct alloc_context *ac)
> {
> ...
> /*
> * Make sure that __GFP_NOFAIL request doesn't leak out and make sure
> * we always retry
> */
> if (gfp_mask & __GFP_NOFAIL) {
> /*
> * All existing users of the __GFP_NOFAIL are blockable, so warn
> * of any new users that actually require GFP_NOWAIT
> */
> if (WARN_ON_ONCE_GFP(!can_direct_reclaim, gfp_mask))
> goto fail;
> ...
> }
> ...
> fail:
> warn_alloc(gfp_mask, ac->nodemask,
> "page allocation failure: order:%u", order);
> got_pg:
> return page;
> }
>
> Let's move the memory allocation out of the atomic context and use
> the normal sleepable context to get pages.
>
> [RFC]: This has only been compile-tested; I'd prefer if the VDPA maintainers
> handles it.
>
> Cc: "Michael S. Tsirkin" <mst@xxxxxxxxxx>
> Cc: Jason Wang <jasowang@xxxxxxxxxx>
> Cc: Xuan Zhuo <xuanzhuo@xxxxxxxxxxxxxxxxx>
> Cc: "Eugenio Pérez" <eperezma@xxxxxxxxxx>
> Cc: Maxime Coquelin <maxime.coquelin@xxxxxxxxxx>
> Signed-off-by: Barry Song <v-songbaohua@xxxxxxxx>
> ---
> drivers/vdpa/vdpa_user/iova_domain.c | 21 ++++++++++++++++-----
> drivers/vdpa/vdpa_user/iova_domain.h | 3 ++-
> drivers/vdpa/vdpa_user/vduse_dev.c | 13 ++++++++++++-
> 3 files changed, 30 insertions(+), 7 deletions(-)
>
> diff --git a/drivers/vdpa/vdpa_user/iova_domain.c b/drivers/vdpa/vdpa_user/iova_domain.c
> index 791d38d6284c..014809ac2b7c 100644
> --- a/drivers/vdpa/vdpa_user/iova_domain.c
> +++ b/drivers/vdpa/vdpa_user/iova_domain.c
> @@ -283,7 +283,7 @@ int vduse_domain_add_user_bounce_pages(struct vduse_iova_domain *domain,
> return ret;
> }
>
> -void vduse_domain_remove_user_bounce_pages(struct vduse_iova_domain *domain)
> +void vduse_domain_remove_user_bounce_pages(struct vduse_iova_domain *domain, struct page **pages)
> {
> struct vduse_bounce_map *map;
> unsigned long i, count;
> @@ -294,15 +294,16 @@ void vduse_domain_remove_user_bounce_pages(struct vduse_iova_domain *domain)
>
> count = domain->bounce_size >> PAGE_SHIFT;
> for (i = 0; i < count; i++) {
> - struct page *page = NULL;
> + struct page *page = pages[i];
>
> map = &domain->bounce_maps[i];
> - if (WARN_ON(!map->bounce_page))
> + if (WARN_ON(!map->bounce_page)) {
> + put_page(page);
> continue;
> + }
>
> /* Copy user page to kernel page if it's in use */
> if (map->orig_phys != INVALID_PHYS_ADDR) {
> - page = alloc_page(GFP_ATOMIC | __GFP_NOFAIL);
> memcpy_from_page(page_address(page),
> map->bounce_page, 0, PAGE_SIZE);
> }
> @@ -543,10 +544,19 @@ static int vduse_domain_mmap(struct file *file, struct vm_area_struct *vma)
> static int vduse_domain_release(struct inode *inode, struct file *file)
> {
> struct vduse_iova_domain *domain = file->private_data;
> + struct page **pages = NULL;
> + unsigned long count, i;
> +
> + if (domain->user_bounce_pages) {
> + count = domain->bounce_size >> PAGE_SHIFT;
> + pages = kmalloc_array(count, sizeof(*pages), GFP_KERNEL | __GFP_NOFAIL);
> + for (i = 0; i < count; i++)
> + pages[i] = alloc_page(GFP_KERNEL | __GFP_NOFAIL);
> + }
>
> spin_lock(&domain->iotlb_lock);
> vduse_iotlb_del_range(domain, 0, ULLONG_MAX);
> - vduse_domain_remove_user_bounce_pages(domain);
> + vduse_domain_remove_user_bounce_pages(domain, pages);
> vduse_domain_free_kernel_bounce_pages(domain);
> spin_unlock(&domain->iotlb_lock);
> put_iova_domain(&domain->stream_iovad);
> @@ -554,6 +564,7 @@ static int vduse_domain_release(struct inode *inode, struct file *file)
> vhost_iotlb_free(domain->iotlb);
> vfree(domain->bounce_maps);
> kfree(domain);
> + kfree(pages);
>
> return 0;
> }
> diff --git a/drivers/vdpa/vdpa_user/iova_domain.h b/drivers/vdpa/vdpa_user/iova_domain.h
> index f92f22a7267d..db0b793d86db 100644
> --- a/drivers/vdpa/vdpa_user/iova_domain.h
> +++ b/drivers/vdpa/vdpa_user/iova_domain.h
> @@ -74,7 +74,8 @@ void vduse_domain_reset_bounce_map(struct vduse_iova_domain *domain);
> int vduse_domain_add_user_bounce_pages(struct vduse_iova_domain *domain,
> struct page **pages, int count);
>
> -void vduse_domain_remove_user_bounce_pages(struct vduse_iova_domain *domain);
> +void vduse_domain_remove_user_bounce_pages(struct vduse_iova_domain *domain,
> + struct page **pages);
>
> void vduse_domain_destroy(struct vduse_iova_domain *domain);
>
> diff --git a/drivers/vdpa/vdpa_user/vduse_dev.c b/drivers/vdpa/vdpa_user/vduse_dev.c
> index 7ae99691efdf..df7c1b6f1350 100644
> --- a/drivers/vdpa/vdpa_user/vduse_dev.c
> +++ b/drivers/vdpa/vdpa_user/vduse_dev.c
> @@ -1030,6 +1030,8 @@ static int vduse_dev_queue_irq_work(struct vduse_dev *dev,
> static int vduse_dev_dereg_umem(struct vduse_dev *dev,
> u64 iova, u64 size)
> {
> + struct page **pages = NULL;
> + unsigned long count, i;
> int ret;
>
> mutex_lock(&dev->mem_lock);
> @@ -1044,13 +1046,22 @@ static int vduse_dev_dereg_umem(struct vduse_dev *dev,
> if (dev->umem->iova != iova || size != dev->domain->bounce_size)
> goto unlock;
>
> - vduse_domain_remove_user_bounce_pages(dev->domain);
> + if (dev->domain->user_bounce_pages) {
> + count = dev->domain->bounce_size >> PAGE_SHIFT;
> + pages = kmalloc_array(count, sizeof(*pages),
> + GFP_KERNEL | __GFP_NOFAIL);
> + for (i = 0; i < count; i++)
> + pages[i] = alloc_page(GFP_KERNEL | __GFP_NOFAIL);
> + }
Nit: there's some code duplication with vduse_domain_release().
Others look good to me.
Would you like to post a formal patch?
Thanks
> +
> + vduse_domain_remove_user_bounce_pages(dev->domain, pages);
> unpin_user_pages_dirty_lock(dev->umem->pages,
> dev->umem->npages, true);
> atomic64_sub(dev->umem->npages, &dev->umem->mm->pinned_vm);
> mmdrop(dev->umem->mm);
> vfree(dev->umem->pages);
> kfree(dev->umem);
> + kfree(pages);
> dev->umem = NULL;
> ret = 0;
> unlock:
> --
> 2.34.1
>
> >
> > Thanks
> >
> > >
> > >
> > > > --
> > > > Michal Hocko
> > > > SUSE Labs
> > >
>
> Thanks
> Barry
>
