IOMMU: hold page ref until after deferred TLB flush

When moving around a page via XENMAPSPACE_gmfn_range, deferring the TLB
flush for the "from" GFN range requires that the page remains allocated
to the guest until the TLB flush has actually occurred. Otherwise a
parallel hypercall to remove the page would only flush the TLB for the
GFN it has been moved to, but not the one is was mapped at originally.

This is part of XSA-346.

Fixes: cf95b2a9fd5a ("iommu: Introduce per cpu flag (iommu_dont_flush_iotlb) to avoid unnecessary iotlb... ")
Reported-by: Julien Grall <jgrall@amazon.com>
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Acked-by: Julien Grall <jgrall@amazon.com>
master commit: 5777a3742d88ff1c0ebc626ceb4fd47f9b3dc6d5
master date: 2020-10-20 14:21:32 +0200

diff --git a/xen/arch/arm/mm.c b/xen/arch/arm/mm.c
index ffe940fbede7ab325db20e773e5b118f27b05142..de6009e54bd18186a21be32b469a495cf670e2d8 100644 (file)
--- a/xen/arch/arm/mm.c
+++ b/xen/arch/arm/mm.c
@@ -1211,7 +1211,7 @@ void share_xen_page_with_guest(struct page_info *page, struct domain *d,
 int xenmem_add_to_physmap_one(
     struct domain *d,
     unsigned int space,
-    union xen_add_to_physmap_batch_extra extra,
+    union add_to_physmap_extra extra,
     unsigned long idx,
     gfn_t gfn)
 {
@@ -1284,10 +1284,6 @@ int xenmem_add_to_physmap_one(
         break;
     }
     case XENMAPSPACE_dev_mmio:
-        /* extra should be 0. Reserved for future use. */
-        if ( extra.res0 )
-            return -EOPNOTSUPP;
-
         rc = map_dev_mmio_region(d, gfn, 1, _mfn(idx));
         return rc;
 

diff --git a/xen/arch/x86/mm.c b/xen/arch/x86/mm.c
index d6ba8c4bb419e4738d568ef7b150209824c6161c..e7b8f4ee4b08541e8037e26d473fbb2a82852cdd 100644 (file)
--- a/xen/arch/x86/mm.c
+++ b/xen/arch/x86/mm.c
@@ -4704,7 +4704,7 @@ static int handle_iomem_range(unsigned long s, unsigned long e, void *p)
 int xenmem_add_to_physmap_one(
     struct domain *d,
     unsigned int space,
-    union xen_add_to_physmap_batch_extra extra,
+    union add_to_physmap_extra extra,
     unsigned long idx,
     gfn_t gpfn)
 {
@@ -4791,9 +4791,20 @@ int xenmem_add_to_physmap_one(
         rc = guest_physmap_add_page(d, gpfn, mfn, PAGE_ORDER_4K);
 
  put_both:
-    /* In the XENMAPSPACE_gmfn case, we took a ref of the gfn at the top. */
+    /*
+     * In the XENMAPSPACE_gmfn case, we took a ref of the gfn at the top.
+     * We also may need to transfer ownership of the page reference to our
+     * caller.
+     */
     if ( space == XENMAPSPACE_gmfn )
+    {
         put_gfn(d, gfn);
+        if ( !rc && extra.ppage )
+        {
+            *extra.ppage = page;
+            page = NULL;
+        }
+    }
 
     if ( page )
         put_page(page);

diff --git a/xen/common/memory.c b/xen/common/memory.c
index 212b293bd87af2fbab1e1e0914dd8d905a87516f..b352a30af6458ad298caf6487bea1b5951a81207 100644 (file)
--- a/xen/common/memory.c
+++ b/xen/common/memory.c
@@ -824,11 +824,10 @@ int xenmem_add_to_physmap(struct domain *d, struct xen_add_to_physmap *xatp,
 {
     unsigned int done = 0;
     long rc = 0;
-    union xen_add_to_physmap_batch_extra extra;
+    union add_to_physmap_extra extra = {};
+    struct page_info *pages[16];
 
-    if ( xatp->space != XENMAPSPACE_gmfn_foreign )
-        extra.res0 = 0;
-    else
+    if ( xatp->space == XENMAPSPACE_gmfn_foreign )
         extra.foreign_domid = DOMID_INVALID;
 
     if ( xatp->space != XENMAPSPACE_gmfn_range )
@@ -843,7 +842,10 @@ int xenmem_add_to_physmap(struct domain *d, struct xen_add_to_physmap *xatp,
     xatp->size -= start;
 
     if ( has_iommu_pt(d) )
+    {
        this_cpu(iommu_dont_flush_iotlb) = 1;
+       extra.ppage = &pages[0];
+    }
 
     while ( xatp->size > done )
     {
@@ -855,8 +857,12 @@ int xenmem_add_to_physmap(struct domain *d, struct xen_add_to_physmap *xatp,
         xatp->idx++;
         xatp->gpfn++;
 
+        if ( extra.ppage )
+            ++extra.ppage;
+
         /* Check for continuation if it's not the last iteration. */
-        if ( xatp->size > ++done && hypercall_preempt_check() )
+        if ( (++done > ARRAY_SIZE(pages) && extra.ppage) ||
+             (xatp->size > done && hypercall_preempt_check()) )
         {
             rc = start + done;
             break;
@@ -866,6 +872,7 @@ int xenmem_add_to_physmap(struct domain *d, struct xen_add_to_physmap *xatp,
     if ( has_iommu_pt(d) )
     {
         int ret;
+        unsigned int i;
 
         this_cpu(iommu_dont_flush_iotlb) = 0;
 
@@ -874,6 +881,15 @@ int xenmem_add_to_physmap(struct domain *d, struct xen_add_to_physmap *xatp,
         if ( unlikely(ret) && rc >= 0 )
             rc = ret;
 
+        /*
+         * Now that the IOMMU TLB flush was done for the original GFN, drop
+         * the page references. The 2nd flush below is fine to make later, as
+         * whoever removes the page again from its new GFN will have to do
+         * another flush anyway.
+         */
+        for ( i = 0; i < done; ++i )
+            put_page(pages[i]);
+
         ret = iommu_iotlb_flush(d, _dfn(xatp->gpfn - done), done,
                                 IOMMU_FLUSHF_added | IOMMU_FLUSHF_modified);
         if ( unlikely(ret) && rc >= 0 )
@@ -887,6 +903,8 @@ static int xenmem_add_to_physmap_batch(struct domain *d,
                                        struct xen_add_to_physmap_batch *xatpb,
                                        unsigned int extent)
 {
+    union add_to_physmap_extra extra = {};
+
     if ( xatpb->size < extent )
         return -EILSEQ;
 
@@ -895,6 +913,19 @@ static int xenmem_add_to_physmap_batch(struct domain *d,
          !guest_handle_subrange_okay(xatpb->errs, extent, xatpb->size - 1) )
         return -EFAULT;
 
+    switch ( xatpb->space )
+    {
+    case XENMAPSPACE_dev_mmio:
+        /* res0 is reserved for future use. */
+        if ( xatpb->u.res0 )
+            return -EOPNOTSUPP;
+        break;
+
+    case XENMAPSPACE_gmfn_foreign:
+        extra.foreign_domid = xatpb->u.foreign_domid;
+        break;
+    }
+
     while ( xatpb->size > extent )
     {
         xen_ulong_t idx;
@@ -907,8 +938,7 @@ static int xenmem_add_to_physmap_batch(struct domain *d,
                                                extent, 1)) )
             return -EFAULT;
 
-        rc = xenmem_add_to_physmap_one(d, xatpb->space,
-                                       xatpb->u,
+        rc = xenmem_add_to_physmap_one(d, xatpb->space, extra,
                                        idx, _gfn(gpfn));
 
         if ( unlikely(__copy_to_guest_offset(xatpb->errs, extent, &rc, 1)) )

diff --git a/xen/include/xen/mm.h b/xen/include/xen/mm.h
index e9711472343c8452824fcff8066a3e88379691f1..d60f17db7a18ffbc34935a0965539a28c56544e3 100644 (file)
--- a/xen/include/xen/mm.h
+++ b/xen/include/xen/mm.h
@@ -583,8 +583,22 @@ void scrub_one_page(struct page_info *);
                       &(d)->xenpage_list : &(d)->page_list)
 #endif
 
+union add_to_physmap_extra {
+    /*
+     * XENMAPSPACE_gmfn: When deferring TLB flushes, a page reference needs
+     * to be kept until after the flush, so the page can't get removed from
+     * the domain (and re-used for another purpose) beforehand. By passing
+     * non-NULL, the caller of xenmem_add_to_physmap_one() indicates it wants
+     * to have ownership of such a reference transferred in the success case.
+     */
+    struct page_info **ppage;
+
+    /* XENMAPSPACE_gmfn_foreign */
+    domid_t foreign_domid;
+};
+
 int xenmem_add_to_physmap_one(struct domain *d, unsigned int space,
-                              union xen_add_to_physmap_batch_extra extra,
+                              union add_to_physmap_extra extra,
                               unsigned long idx, gfn_t gfn);
 
 int xenmem_add_to_physmap(struct domain *d, struct xen_add_to_physmap *xatp,