Commit f45ec5ff16a75 ("userfaultfd: wp: support swap and page migration") introduced support for tracking the uffd wp bit during page migration. However the non-swap PTE variant was used to set the flag for zone device private pages which are a type of swap page.
This leads to corruption of the swap offset if the original PTE has the uffd_wp flag set.
Fixes: f45ec5ff16a75 ("userfaultfd: wp: support swap and page migration") Signed-off-by: Alistair Popple alistair@popple.id.au Reviewed-by: Peter Xu peterx@redhat.com Cc: stable@vger.kernel.org --- mm/migrate.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/mm/migrate.c b/mm/migrate.c index 34a842a8eb6a..ddb64253fe3e 100644 --- a/mm/migrate.c +++ b/mm/migrate.c @@ -251,7 +251,7 @@ static bool remove_migration_pte(struct page *page, struct vm_area_struct *vma, entry = make_device_private_entry(new, pte_write(pte)); pte = swp_entry_to_pte(entry); if (pte_swp_uffd_wp(*pvmw.pte)) - pte = pte_mkuffd_wp(pte); + pte = pte_swp_mkuffd_wp(pte); } }
During memory migration a pte is temporarily replaced with a migration swap pte. Some pte bits from the existing mapping such as the soft-dirty and uffd write-protect bits are preserved by copying these to the temporary migration swap pte.
However these bits are not stored at the same location for swap and non-swap ptes. Therefore testing these bits requires using the appropriate helper function for the given pte type.
Unfortunately several code locations were found where the wrong helper function is being used to test soft_dirty and uffd_wp bits which leads to them getting incorrectly set or cleared during page-migration.
Fix these by using the correct tests based on pte type.
Fixes: a5430dda8a3a ("mm/migrate: support un-addressable ZONE_DEVICE page in migration") Fixes: 8c3328f1f36a ("mm/migrate: migrate_vma() unmap page from vma while collecting pages") Fixes: f45ec5ff16a7 ("userfaultfd: wp: support swap and page migration") Signed-off-by: Alistair Popple alistair@popple.id.au Cc: stable@vger.kernel.org --- mm/migrate.c | 15 +++++++++++---- mm/rmap.c | 9 +++++++-- 2 files changed, 18 insertions(+), 6 deletions(-)
diff --git a/mm/migrate.c b/mm/migrate.c index ddb64253fe3e..12f63806d0ac 100644 --- a/mm/migrate.c +++ b/mm/migrate.c @@ -2427,10 +2427,17 @@ static int migrate_vma_collect_pmd(pmd_t *pmdp, entry = make_migration_entry(page, mpfn & MIGRATE_PFN_WRITE); swp_pte = swp_entry_to_pte(entry); - if (pte_soft_dirty(pte)) - swp_pte = pte_swp_mksoft_dirty(swp_pte); - if (pte_uffd_wp(pte)) - swp_pte = pte_swp_mkuffd_wp(swp_pte); + if (pte_present(pte)) { + if (pte_soft_dirty(pte)) + swp_pte = pte_swp_mksoft_dirty(swp_pte); + if (pte_uffd_wp(pte)) + swp_pte = pte_swp_mkuffd_wp(swp_pte); + } else { + if (pte_swp_soft_dirty(pte)) + swp_pte = pte_swp_mksoft_dirty(swp_pte); + if (pte_swp_uffd_wp(pte)) + swp_pte = pte_swp_mkuffd_wp(swp_pte); + } set_pte_at(mm, addr, ptep, swp_pte);
/* diff --git a/mm/rmap.c b/mm/rmap.c index 83cc459edc40..9425260774a1 100644 --- a/mm/rmap.c +++ b/mm/rmap.c @@ -1511,9 +1511,14 @@ static bool try_to_unmap_one(struct page *page, struct vm_area_struct *vma, */ entry = make_migration_entry(page, 0); swp_pte = swp_entry_to_pte(entry); - if (pte_soft_dirty(pteval)) + + /* + * pteval maps a zone device page and is therefore + * a swap pte. + */ + if (pte_swp_soft_dirty(pteval)) swp_pte = pte_swp_mksoft_dirty(swp_pte); - if (pte_uffd_wp(pteval)) + if (pte_swp_uffd_wp(pteval)) swp_pte = pte_swp_mkuffd_wp(swp_pte); set_pte_at(mm, pvmw.address, pvmw.pte, swp_pte); /*
On Tue, Aug 25, 2020 at 04:42:32PM +1000, Alistair Popple wrote:
During memory migration a pte is temporarily replaced with a migration swap pte. Some pte bits from the existing mapping such as the soft-dirty and uffd write-protect bits are preserved by copying these to the temporary migration swap pte.
However these bits are not stored at the same location for swap and non-swap ptes. Therefore testing these bits requires using the appropriate helper function for the given pte type.
Unfortunately several code locations were found where the wrong helper function is being used to test soft_dirty and uffd_wp bits which leads to them getting incorrectly set or cleared during page-migration.
Fix these by using the correct tests based on pte type.
Fixes: a5430dda8a3a ("mm/migrate: support un-addressable ZONE_DEVICE page in migration") Fixes: 8c3328f1f36a ("mm/migrate: migrate_vma() unmap page from vma while collecting pages") Fixes: f45ec5ff16a7 ("userfaultfd: wp: support swap and page migration") Signed-off-by: Alistair Popple alistair@popple.id.au Cc: stable@vger.kernel.org
Reviewed-by: Peter Xu peterx@redhat.com
linux-stable-mirror@lists.linaro.org