From: Jan Beulich Subject: IOMMU/x86: fix type ref-counting race upon IOMMU page table construction When arch_iommu_populate_page_table() gets invoked for an already running guest, simply looking at page types once isn't enough, as they may change at any time. Add logic to re-check the type after having mapped the page, unmapping it again if needed. This is XSA-285. Signed-off-by: Jan Beulich Tentatively-Acked-by: Andrew Cooper --- a/xen/drivers/passthrough/x86/iommu.c +++ b/xen/drivers/passthrough/x86/iommu.c @@ -68,6 +68,27 @@ int arch_iommu_populate_page_table(struct domain *d) rc = hd->platform_ops->map_page(d, gfn, mfn, IOMMUF_readable | IOMMUF_writable); + + /* + * We may be working behind the back of a running guest, which + * may change the type of a page at any time. We can't prevent + * this (for instance, by bumping the type count while mapping + * the page) without causing legitimate guest type-change + * operations to fail. So after adding the page to the IOMMU, + * check again to make sure this is still valid. NB that the + * writable entry in the iommu is harmless until later, when + * the actual device gets assigned. + */ + if ( !rc && !is_hvm_domain(d) && + ((page->u.inuse.type_info & PGT_type_mask) != + PGT_writable_page) ) + { + rc = hd->platform_ops->unmap_page(d, gfn); + /* If the type changed yet again, simply force a retry. */ + if ( !rc && ((page->u.inuse.type_info & PGT_type_mask) == + PGT_writable_page) ) + rc = -ERESTART; + } } if ( rc ) {