Re: [PATCH v3 14/28] userfaultfd: wp: handle COW properly for uffd-wp

From: Jerome Glisse
Date: Thu Apr 18 2019 - 16:51:26 EST


On Wed, Mar 20, 2019 at 10:06:28AM +0800, Peter Xu wrote:
> This allows uffd-wp to support write-protected pages for COW.
>
> For example, the uffd write-protected PTE could also be write-protected
> by other usages like COW or zero pages. When that happens, we can't
> simply set the write bit in the PTE since otherwise it'll change the
> content of every single reference to the page. Instead, we should do
> the COW first if necessary, then handle the uffd-wp fault.
>
> To correctly copy the page, we'll also need to carry over the
> _PAGE_UFFD_WP bit if it was set in the original PTE.
>
> For huge PMDs, we just simply split the huge PMDs where we want to
> resolve an uffd-wp page fault always. That matches what we do with
> general huge PMD write protections. In that way, we resolved the huge
> PMD copy-on-write issue into PTE copy-on-write.
>
> Signed-off-by: Peter Xu <peterx@xxxxxxxxxx>

This one has a bug see below.


> ---
> mm/memory.c | 5 +++-
> mm/mprotect.c | 64 ++++++++++++++++++++++++++++++++++++++++++++++++---
> 2 files changed, 65 insertions(+), 4 deletions(-)
>
> diff --git a/mm/memory.c b/mm/memory.c
> index e7a4b9650225..b8a4c0bab461 100644
> --- a/mm/memory.c
> +++ b/mm/memory.c
> @@ -2291,7 +2291,10 @@ vm_fault_t wp_page_copy(struct vm_fault *vmf)
> }
> flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte));
> entry = mk_pte(new_page, vma->vm_page_prot);
> - entry = maybe_mkwrite(pte_mkdirty(entry), vma);
> + if (pte_uffd_wp(vmf->orig_pte))
> + entry = pte_mkuffd_wp(entry);
> + else
> + entry = maybe_mkwrite(pte_mkdirty(entry), vma);
> /*
> * Clear the pte entry and flush it first, before updating the
> * pte with the new entry. This will avoid a race condition
> diff --git a/mm/mprotect.c b/mm/mprotect.c
> index 9d4433044c21..855dddb07ff2 100644
> --- a/mm/mprotect.c
> +++ b/mm/mprotect.c
> @@ -73,18 +73,18 @@ static unsigned long change_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
> flush_tlb_batched_pending(vma->vm_mm);
> arch_enter_lazy_mmu_mode();
> do {
> +retry_pte:
> oldpte = *pte;
> if (pte_present(oldpte)) {
> pte_t ptent;
> bool preserve_write = prot_numa && pte_write(oldpte);
> + struct page *page;
>
> /*
> * Avoid trapping faults against the zero or KSM
> * pages. See similar comment in change_huge_pmd.
> */
> if (prot_numa) {
> - struct page *page;
> -
> page = vm_normal_page(vma, addr, oldpte);
> if (!page || PageKsm(page))
> continue;
> @@ -114,6 +114,54 @@ static unsigned long change_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
> continue;
> }
>
> + /*
> + * Detect whether we'll need to COW before
> + * resolving an uffd-wp fault. Note that this
> + * includes detection of the zero page (where
> + * page==NULL)
> + */
> + if (uffd_wp_resolve) {
> + /* If the fault is resolved already, skip */
> + if (!pte_uffd_wp(*pte))
> + continue;
> + page = vm_normal_page(vma, addr, oldpte);
> + if (!page || page_mapcount(page) > 1) {
> + struct vm_fault vmf = {
> + .vma = vma,
> + .address = addr & PAGE_MASK,
> + .page = page,
> + .orig_pte = oldpte,
> + .pmd = pmd,
> + /* pte and ptl not needed */
> + };
> + vm_fault_t ret;
> +
> + if (page)
> + get_page(page);
> + arch_leave_lazy_mmu_mode();
> + pte_unmap_unlock(pte, ptl);
> + ret = wp_page_copy(&vmf);
> + /* PTE is changed, or OOM */
> + if (ret == 0)
> + /* It's done by others */
> + continue;

This is wrong if ret == 0 you still need to remap the pte before
continuing as otherwise you will go to next pte without the page
table lock for the directory. So 0 case must be handled after
arch_enter_lazy_mmu_mode() below.

Sorry i should have catch that in previous review.


> + else if (WARN_ON(ret != VM_FAULT_WRITE))
> + return pages;
> + pte = pte_offset_map_lock(vma->vm_mm,
> + pmd, addr,
> + &ptl);
> + arch_enter_lazy_mmu_mode();
> + if (!pte_present(*pte))
> + /*
> + * This PTE could have been
> + * modified after COW
> + * before we have taken the
> + * lock; retry this PTE
> + */
> + goto retry_pte;
> + }
> + }
> +
> ptent = ptep_modify_prot_start(mm, addr, pte);
> ptent = pte_modify(ptent, newprot);
> if (preserve_write)

> unsigned long pages = 0;
> unsigned long nr_huge_updates = 0;
> struct mmu_notifier_range range;
> + bool uffd_wp_resolve = cp_flags & MM_CP_UFFD_WP_RESOLVE;
>
> range.start = 0;
>
> @@ -202,7 +251,16 @@ static inline unsigned long change_pmd_range(struct vm_area_struct *vma,
> }
>
> if (is_swap_pmd(*pmd) || pmd_trans_huge(*pmd) || pmd_devmap(*pmd)) {
> - if (next - addr != HPAGE_PMD_SIZE) {
> + /*
> + * When resolving an userfaultfd write
> + * protection fault, it's not easy to identify
> + * whether a THP is shared with others and
> + * whether we'll need to do copy-on-write, so
> + * just split it always for now to simply the
> + * procedure. And that's the policy too for
> + * general THP write-protect in af9e4d5f2de2.
> + */
> + if (next - addr != HPAGE_PMD_SIZE || uffd_wp_resolve) {

Just a nit pick can you please add () to next - addr ie:
if ((next - addr) != HPAGE_PMD_SIZE || uffd_wp_resolve) {

I know it is not needed but each time i bump into this i
have to scratch my head for second to remember the operator
rules :)

> __split_huge_pmd(vma, pmd, addr, false, NULL);
> } else {
> int nr_ptes = change_huge_pmd(vma, pmd, addr,
> --
> 2.17.1
>