x86/mm: Ensure maps used by nested hvm code cannot be paged out

The nested hvm code maps pages of the guest hvm. These maps live beyond
a hypervisor entry/exit pair, and thus their liveness cannot be ensured
with get_gfn/put_gfn critical sections. Ensure their liveness by
increasing the page ref count, instead.

Signed-off-by: Andres Lagar-Cavilla <andres@lagarcavilla.org>
Acked-by: Tim Deegan <tim@xen.org>
Committed-by: Tim Deegan <tim@xen.org>

diff --git a/xen/arch/x86/hvm/hvm.c b/xen/arch/x86/hvm/hvm.c
index 9ac42b52b6aac0cf509cb33c4a642e9b52cf1481..a38bd7b1dbdcd6c8779ab085057f6c35bad84d2e 100644 (file)
--- a/xen/arch/x86/hvm/hvm.c
+++ b/xen/arch/x86/hvm/hvm.c
@@ -1823,12 +1823,16 @@ int hvm_virtual_to_linear_addr(
     return 0;
 }
 
-/* We leave this function holding a lock on the p2m entry */
+/* On non-NULL return, we leave this function holding an additional 
+ * ref on the underlying mfn, if any */
 static void *__hvm_map_guest_frame(unsigned long gfn, bool_t writable)
 {
+    void *map;
     unsigned long mfn;
     p2m_type_t p2mt;
+    struct page_info *pg;
     struct domain *d = current->domain;
+    int rc;
 
     mfn = mfn_x(writable
                 ? get_gfn_unshare(d, gfn, &p2mt)
@@ -1850,7 +1854,21 @@ static void *__hvm_map_guest_frame(unsigned long gfn, bool_t writable)
     if ( writable )
         paging_mark_dirty(d, mfn);
 
-    return map_domain_page(mfn);
+    /* Get a ref on the page, considering that it could be shared */
+    pg = mfn_to_page(mfn);
+    rc = get_page(pg, d);
+    if ( !rc && !writable )
+        /* Page could be shared */
+        rc = get_page(pg, dom_cow);
+    if ( !rc )
+    {
+        put_gfn(d, gfn);
+        return NULL;
+    }
+
+    map = map_domain_page(mfn);
+    put_gfn(d, gfn);
+    return map;
 }
 
 void *hvm_map_guest_frame_rw(unsigned long gfn)
@@ -1866,11 +1884,16 @@ void *hvm_map_guest_frame_ro(unsigned long gfn)
 void hvm_unmap_guest_frame(void *p)
 {
     if ( p )
+    {
+        unsigned long mfn = domain_page_map_to_mfn(p);
         unmap_domain_page(p);
+        put_page(mfn_to_page(mfn));
+    }
 }
 
-static void *hvm_map_entry(unsigned long va, unsigned long *gfn)
+static void *hvm_map_entry(unsigned long va)
 {
+    unsigned long gfn;
     uint32_t pfec;
     char *v;
 
@@ -1887,11 +1910,11 @@ static void *hvm_map_entry(unsigned long va, unsigned long *gfn)
      * treat it as a kernel-mode read (i.e. no access checks).
      */
     pfec = PFEC_page_present;
-    *gfn = paging_gva_to_gfn(current, va, &pfec);
+    gfn = paging_gva_to_gfn(current, va, &pfec);
     if ( (pfec == PFEC_page_paged) || (pfec == PFEC_page_shared) )
         goto fail;
 
-    v = hvm_map_guest_frame_rw(*gfn);
+    v = hvm_map_guest_frame_rw(gfn);
     if ( v == NULL )
         goto fail;
 
@@ -1902,11 +1925,9 @@ static void *hvm_map_entry(unsigned long va, unsigned long *gfn)
     return NULL;
 }
 
-static void hvm_unmap_entry(void *p, unsigned long gfn)
+static void hvm_unmap_entry(void *p)
 {
     hvm_unmap_guest_frame(p);
-    if ( p && (gfn != INVALID_GFN) )
-        put_gfn(current->domain, gfn);
 }
 
 static int hvm_load_segment_selector(
@@ -1918,7 +1939,6 @@ static int hvm_load_segment_selector(
     int fault_type = TRAP_invalid_tss;
     struct cpu_user_regs *regs = guest_cpu_user_regs();
     struct vcpu *v = current;
-    unsigned long pdesc_gfn = INVALID_GFN;
 
     if ( regs->eflags & X86_EFLAGS_VM )
     {
@@ -1952,7 +1972,7 @@ static int hvm_load_segment_selector(
     if ( ((sel & 0xfff8) + 7) > desctab.limit )
         goto fail;
 
-    pdesc = hvm_map_entry(desctab.base + (sel & 0xfff8), &pdesc_gfn);
+    pdesc = hvm_map_entry(desctab.base + (sel & 0xfff8));
     if ( pdesc == NULL )
         goto hvm_map_fail;
 
@@ -2012,7 +2032,7 @@ static int hvm_load_segment_selector(
     desc.b |= 0x100;
 
  skip_accessed_flag:
-    hvm_unmap_entry(pdesc, pdesc_gfn);
+    hvm_unmap_entry(pdesc);
 
     segr.base = (((desc.b <<  0) & 0xff000000u) |
                  ((desc.b << 16) & 0x00ff0000u) |
@@ -2028,7 +2048,7 @@ static int hvm_load_segment_selector(
     return 0;
 
  unmap_and_fail:
-    hvm_unmap_entry(pdesc, pdesc_gfn);
+    hvm_unmap_entry(pdesc);
  fail:
     hvm_inject_exception(fault_type, sel & 0xfffc, 0);
  hvm_map_fail:
@@ -2043,7 +2063,7 @@ void hvm_task_switch(
     struct cpu_user_regs *regs = guest_cpu_user_regs();
     struct segment_register gdt, tr, prev_tr, segr;
     struct desc_struct *optss_desc = NULL, *nptss_desc = NULL, tss_desc;
-    unsigned long eflags, optss_gfn = INVALID_GFN, nptss_gfn = INVALID_GFN;
+    unsigned long eflags;
     int exn_raised, rc;
     struct {
         u16 back_link,__blh;
@@ -2069,11 +2089,11 @@ void hvm_task_switch(
         goto out;
     }
 
-    optss_desc = hvm_map_entry(gdt.base + (prev_tr.sel & 0xfff8), &optss_gfn);
+    optss_desc = hvm_map_entry(gdt.base + (prev_tr.sel & 0xfff8)); 
     if ( optss_desc == NULL )
         goto out;
 
-    nptss_desc = hvm_map_entry(gdt.base + (tss_sel & 0xfff8), &nptss_gfn);
+    nptss_desc = hvm_map_entry(gdt.base + (tss_sel & 0xfff8)); 
     if ( nptss_desc == NULL )
         goto out;
 
@@ -2238,8 +2258,8 @@ void hvm_task_switch(
     }
 
  out:
-    hvm_unmap_entry(optss_desc, optss_gfn);
-    hvm_unmap_entry(nptss_desc, nptss_gfn);
+    hvm_unmap_entry(optss_desc);
+    hvm_unmap_entry(nptss_desc);
 }
 
 #define HVMCOPY_from_guest (0u<<0)

diff --git a/xen/arch/x86/hvm/svm/nestedsvm.c b/xen/arch/x86/hvm/svm/nestedsvm.c
index c3912a58138608092d5ad956c4f847892c1681cc..40a68bf898c1383a9727b898bc7aa4f3c143a269 100644 (file)
--- a/xen/arch/x86/hvm/svm/nestedsvm.c
+++ b/xen/arch/x86/hvm/svm/nestedsvm.c
@@ -81,10 +81,6 @@ int nestedsvm_vmcb_map(struct vcpu *v, uint64_t vmcbaddr)
         if (nv->nv_vvmcx == NULL)
             return 0;
         nv->nv_vvmcxaddr = vmcbaddr;
-        /* put_gfn here even though the map survives beyond this caller.
-         * The map can likely survive beyond a hypervisor exit, thus we
-         * need to put the gfn */
-        put_gfn(current->domain, vmcbaddr >> PAGE_SHIFT);
     }
 
     return 1;
@@ -358,7 +354,6 @@ static int nsvm_vmrun_permissionmap(struct vcpu *v, bool_t viopm)
     ioport_80 = test_bit(0x80, ns_viomap);
     ioport_ed = test_bit(0xed, ns_viomap);
     hvm_unmap_guest_frame(ns_viomap);
-    put_gfn(current->domain, svm->ns_iomap_pa >> PAGE_SHIFT);
 
     svm->ns_iomap = nestedhvm_vcpu_iomap_get(ioport_80, ioport_ed);
 
@@ -889,7 +884,6 @@ nsvm_vmcb_guest_intercepts_ioio(paddr_t iopm_pa, uint64_t exitinfo1)
 
     enabled = test_bit(port, io_bitmap);
     hvm_unmap_guest_frame(io_bitmap);
-    put_gfn(current->domain, gfn);
 
     if (!enabled)
         return NESTEDHVM_VMEXIT_HOST;

diff --git a/xen/arch/x86/hvm/vmx/vvmx.c b/xen/arch/x86/hvm/vmx/vvmx.c
index 3d7092bbef12f5e2bb98df87e8662e73de9f304a..31682edd791e1aefe191910de3236f8ae78f4e42 100644 (file)
--- a/xen/arch/x86/hvm/vmx/vvmx.c
+++ b/xen/arch/x86/hvm/vmx/vvmx.c
@@ -560,10 +560,7 @@ static void __map_io_bitmap(struct vcpu *v, u64 vmcs_reg)
     if (nvmx->iobitmap[index])
         hvm_unmap_guest_frame (nvmx->iobitmap[index]); 
     gpa = __get_vvmcs(vcpu_nestedhvm(v).nv_vvmcx, vmcs_reg);
-    nvmx->iobitmap[index] = hvm_map_guest_frame_ro (gpa >> PAGE_SHIFT);
-    /* See comment in nestedsvm_vmcb_map re putting this gfn and 
-     * liveness of the map it backs */
-    put_gfn(current->domain, gpa >> PAGE_SHIFT);
+    nvmx->iobitmap[index] = hvm_map_guest_frame_ro(gpa >> PAGE_SHIFT);
 }
 
 static inline void map_io_bitmap_all(struct vcpu *v)
@@ -1138,12 +1135,9 @@ int nvmx_handle_vmptrld(struct cpu_user_regs *regs)
 
     if ( nvcpu->nv_vvmcxaddr == VMCX_EADDR )
     {
-        nvcpu->nv_vvmcx = hvm_map_guest_frame_rw (gpa >> PAGE_SHIFT);
+        nvcpu->nv_vvmcx = hvm_map_guest_frame_rw(gpa >> PAGE_SHIFT);
         nvcpu->nv_vvmcxaddr = gpa;
         map_io_bitmap_all (v);
-        /* See comment in nestedsvm_vmcb_map regarding putting this 
-         * gfn and liveness of the map that uses it */
-        put_gfn(current->domain, gpa >> PAGE_SHIFT);
     }
 
     vmreturn(regs, VMSUCCEED);
@@ -1205,7 +1199,6 @@ int nvmx_handle_vmclear(struct cpu_user_regs *regs)
         if ( vvmcs ) 
             __set_vvmcs(vvmcs, NVMX_LAUNCH_STATE, 0);
         hvm_unmap_guest_frame(vvmcs);
-        put_gfn(current->domain, gpa >> PAGE_SHIFT);
     }
 
     vmreturn(regs, VMSUCCEED);