path: root/main/xen/xsa97-hap-4_2.patch

x86/paging: make log-dirty operations preemptible

Both the freeing and the inspection of the bitmap get done in (nested)
loops which - besides having a rather high iteration count in general,
albeit that would be covered by XSA-77 - have the number of non-trivial
iterations they need to perform (indirectly) controllable by both the
guest they are for and any domain controlling the guest (including the
one running qemu for it).

This is XSA-97.

Signed-off-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Tim Deegan <tim@xen.org>

--- a/xen/arch/x86/domain.c
+++ b/xen/arch/x86/domain.c
@@ -2136,7 +2136,9 @@ int domain_relinquish_resources(struct d
         pci_release_devices(d);
 
         /* Tear down paging-assistance stuff. */
-        paging_teardown(d);
+        ret = paging_teardown(d);
+        if ( ret )
+            return ret;
 
         /* Drop the in-use references to page-table bases. */
         for_each_vcpu ( d, v )
--- a/xen/arch/x86/domctl.c
+++ b/xen/arch/x86/domctl.c
@@ -66,6 +66,9 @@ long arch_do_domctl(
                                 &domctl->u.shadow_op,
                                 guest_handle_cast(u_domctl, void));
             rcu_unlock_domain(d);
+            if ( ret == -EAGAIN )
+                return hypercall_create_continuation(__HYPERVISOR_domctl,
+                                                     "h", u_domctl);
             copy_to_guest(u_domctl, domctl, 1);
         } 
     }
--- a/xen/arch/x86/mm/hap/hap.c
+++ b/xen/arch/x86/mm/hap/hap.c
@@ -678,8 +678,7 @@ int hap_domctl(struct domain *d, xen_dom
         paging_unlock(d);
         if ( preempted )
             /* Not finished.  Set up to re-run the call. */
-            rc = hypercall_create_continuation(__HYPERVISOR_domctl, "h",
-                                               u_domctl);
+            rc = -EAGAIN;
         else
             /* Finished.  Return the new allocation */
             sc->mb = hap_get_allocation(d);
--- a/xen/arch/x86/mm/paging.c
+++ b/xen/arch/x86/mm/paging.c
@@ -26,6 +26,7 @@
 #include <asm/shadow.h>
 #include <asm/p2m.h>
 #include <asm/hap.h>
+#include <asm/event.h>
 #include <asm/hvm/nestedhvm.h>
 #include <xen/numa.h>
 #include <xsm/xsm.h>
@@ -116,26 +117,46 @@ static void paging_free_log_dirty_page(s
     d->arch.paging.free_page(d, mfn_to_page(mfn));
 }
 
-void paging_free_log_dirty_bitmap(struct domain *d)
+static int paging_free_log_dirty_bitmap(struct domain *d, int rc)
 {
     mfn_t *l4, *l3, *l2;
     int i4, i3, i2;
 
+    paging_lock(d);
+
     if ( !mfn_valid(d->arch.paging.log_dirty.top) )
-        return;
+    {
+        paging_unlock(d);
+        return 0;
+    }
 
-    paging_lock(d);
+    if ( !d->arch.paging.preempt.vcpu )
+    {
+        memset(&d->arch.paging.preempt.log_dirty, 0,
+               sizeof(d->arch.paging.preempt.log_dirty));
+        ASSERT(rc <= 0);
+        d->arch.paging.preempt.log_dirty.done = -rc;
+    }
+    else if ( d->arch.paging.preempt.vcpu != current ||
+              d->arch.paging.preempt.op != XEN_DOMCTL_SHADOW_OP_OFF )
+    {
+        paging_unlock(d);
+        return -EBUSY;
+    }
 
     l4 = map_domain_page(mfn_x(d->arch.paging.log_dirty.top));
+    i4 = d->arch.paging.preempt.log_dirty.i4;
+    i3 = d->arch.paging.preempt.log_dirty.i3;
+    rc = 0;
 
-    for ( i4 = 0; i4 < LOGDIRTY_NODE_ENTRIES; i4++ )
+    for ( ; i4 < LOGDIRTY_NODE_ENTRIES; i4++, i3 = 0 )
     {
         if ( !mfn_valid(l4[i4]) )
             continue;
 
         l3 = map_domain_page(mfn_x(l4[i4]));
 
-        for ( i3 = 0; i3 < LOGDIRTY_NODE_ENTRIES; i3++ )
+        for ( ; i3 < LOGDIRTY_NODE_ENTRIES; i3++ )
         {
             if ( !mfn_valid(l3[i3]) )
                 continue;
@@ -148,20 +169,54 @@ void paging_free_log_dirty_bitmap(struct
 
             unmap_domain_page(l2);
             paging_free_log_dirty_page(d, l3[i3]);
+            l3[i3] = _mfn(INVALID_MFN);
+
+            if ( i3 < LOGDIRTY_NODE_ENTRIES - 1 && hypercall_preempt_check() )
+            {
+                d->arch.paging.preempt.log_dirty.i3 = i3 + 1;
+                d->arch.paging.preempt.log_dirty.i4 = i4;
+                rc = -EAGAIN;
+                break;
+            }
         }
 
         unmap_domain_page(l3);
+        if ( rc )
+            break;
         paging_free_log_dirty_page(d, l4[i4]);
+        l4[i4] = _mfn(INVALID_MFN);
+
+        if ( i4 < LOGDIRTY_NODE_ENTRIES - 1 && hypercall_preempt_check() )
+        {
+            d->arch.paging.preempt.log_dirty.i3 = 0;
+            d->arch.paging.preempt.log_dirty.i4 = i4 + 1;
+            rc = -EAGAIN;
+            break;
+        }
     }
 
     unmap_domain_page(l4);
-    paging_free_log_dirty_page(d, d->arch.paging.log_dirty.top);
-    d->arch.paging.log_dirty.top = _mfn(INVALID_MFN);
 
-    ASSERT(d->arch.paging.log_dirty.allocs == 0);
-    d->arch.paging.log_dirty.failed_allocs = 0;
+    if ( !rc )
+    {
+        paging_free_log_dirty_page(d, d->arch.paging.log_dirty.top);
+        d->arch.paging.log_dirty.top = _mfn(INVALID_MFN);
+
+        ASSERT(d->arch.paging.log_dirty.allocs == 0);
+        d->arch.paging.log_dirty.failed_allocs = 0;
+
+        rc = -d->arch.paging.preempt.log_dirty.done;
+        d->arch.paging.preempt.vcpu = NULL;
+    }
+    else
+    {
+        d->arch.paging.preempt.vcpu = current;
+        d->arch.paging.preempt.op = XEN_DOMCTL_SHADOW_OP_OFF;
+    }
 
     paging_unlock(d);
+
+    return rc;
 }
 
 int paging_log_dirty_enable(struct domain *d)
@@ -178,15 +233,25 @@ int paging_log_dirty_enable(struct domai
     return ret;
 }
 
-int paging_log_dirty_disable(struct domain *d)
+static int paging_log_dirty_disable(struct domain *d, bool_t resuming)
 {
-    int ret;
+    int ret = 1;
+
+    if ( !resuming )
+    {
+        domain_pause(d);
+        /* Safe because the domain is paused. */
+        ret = d->arch.paging.log_dirty.disable_log_dirty(d);
+        ASSERT(ret <= 0);
+    }
 
-    domain_pause(d);
-    /* Safe because the domain is paused. */
-    ret = d->arch.paging.log_dirty.disable_log_dirty(d);
     if ( !paging_mode_log_dirty(d) )
-        paging_free_log_dirty_bitmap(d);
+    {
+        ret = paging_free_log_dirty_bitmap(d, ret);
+        if ( ret == -EAGAIN )
+            return ret;
+    }
+
     domain_unpause(d);
 
     return ret;
@@ -326,7 +391,9 @@ int paging_mfn_is_dirty(struct domain *d
 
 /* Read a domain's log-dirty bitmap and stats.  If the operation is a CLEAN,
  * clear the bitmap and stats as well. */
-int paging_log_dirty_op(struct domain *d, struct xen_domctl_shadow_op *sc)
+static int paging_log_dirty_op(struct domain *d,
+                               struct xen_domctl_shadow_op *sc,
+                               bool_t resuming)
 {
     int rv = 0, clean = 0, peek = 1;
     unsigned long pages = 0;
@@ -334,9 +401,22 @@ int paging_log_dirty_op(struct domain *d
     unsigned long *l1 = NULL;
     int i4, i3, i2;
 
-    domain_pause(d);
+    if ( !resuming )
+        domain_pause(d);
     paging_lock(d);
 
+    if ( !d->arch.paging.preempt.vcpu )
+        memset(&d->arch.paging.preempt.log_dirty, 0,
+               sizeof(d->arch.paging.preempt.log_dirty));
+    else if ( d->arch.paging.preempt.vcpu != current ||
+              d->arch.paging.preempt.op != sc->op )
+    {
+        paging_unlock(d);
+        ASSERT(!resuming);
+        domain_unpause(d);
+        return -EBUSY;
+    }
+
     clean = (sc->op == XEN_DOMCTL_SHADOW_OP_CLEAN);
 
     PAGING_DEBUG(LOGDIRTY, "log-dirty %s: dom %u faults=%u dirty=%u\n",
@@ -365,17 +445,15 @@ int paging_log_dirty_op(struct domain *d
         goto out;
     }
 
-    pages = 0;
     l4 = paging_map_log_dirty_bitmap(d);
+    i4 = d->arch.paging.preempt.log_dirty.i4;
+    i3 = d->arch.paging.preempt.log_dirty.i3;
+    pages = d->arch.paging.preempt.log_dirty.done;
 
-    for ( i4 = 0;
-          (pages < sc->pages) && (i4 < LOGDIRTY_NODE_ENTRIES);
-          i4++ )
+    for ( ; (pages < sc->pages) && (i4 < LOGDIRTY_NODE_ENTRIES); i4++, i3 = 0 )
     {
         l3 = (l4 && mfn_valid(l4[i4])) ? map_domain_page(mfn_x(l4[i4])) : NULL;
-        for ( i3 = 0;
-              (pages < sc->pages) && (i3 < LOGDIRTY_NODE_ENTRIES);
-              i3++ )
+        for ( ; (pages < sc->pages) && (i3 < LOGDIRTY_NODE_ENTRIES); i3++ )
         {
             l2 = ((l3 && mfn_valid(l3[i3])) ?
                   map_domain_page(mfn_x(l3[i3])) : NULL);
@@ -410,18 +488,51 @@ int paging_log_dirty_op(struct domain *d
             }
             if ( l2 )
                 unmap_domain_page(l2);
+
+            if ( i3 < LOGDIRTY_NODE_ENTRIES - 1 && hypercall_preempt_check() )
+            {
+                d->arch.paging.preempt.log_dirty.i4 = i4;
+                d->arch.paging.preempt.log_dirty.i3 = i3 + 1;
+                rv = -EAGAIN;
+                break;
+            }
         }
         if ( l3 )
             unmap_domain_page(l3);
+
+        if ( !rv && i4 < LOGDIRTY_NODE_ENTRIES - 1 &&
+             hypercall_preempt_check() )
+        {
+            d->arch.paging.preempt.log_dirty.i4 = i4 + 1;
+            d->arch.paging.preempt.log_dirty.i3 = 0;
+            rv = -EAGAIN;
+        }
+        if ( rv )
+            break;
     }
     if ( l4 )
         unmap_domain_page(l4);
 
-    if ( pages < sc->pages )
-        sc->pages = pages;
+    if ( !rv )
+        d->arch.paging.preempt.vcpu = NULL;
+    else
+    {
+        d->arch.paging.preempt.vcpu = current;
+        d->arch.paging.preempt.op = sc->op;
+        d->arch.paging.preempt.log_dirty.done = pages;
+    }
 
     paging_unlock(d);
 
+    if ( rv )
+    {
+        /* Never leave the domain paused for other errors. */
+        ASSERT(rv == -EAGAIN);
+        return rv;
+    }
+
+    if ( pages < sc->pages )
+        sc->pages = pages;
     if ( clean )
     {
         /* We need to further call clean_dirty_bitmap() functions of specific
@@ -432,6 +543,7 @@ int paging_log_dirty_op(struct domain *d
     return rv;
 
  out:
+    d->arch.paging.preempt.vcpu = NULL;
     paging_unlock(d);
     domain_unpause(d);
 
@@ -498,12 +610,6 @@ void paging_log_dirty_init(struct domain
     d->arch.paging.log_dirty.clean_dirty_bitmap = clean_dirty_bitmap;
 }
 
-/* This function fress log dirty bitmap resources. */
-static void paging_log_dirty_teardown(struct domain*d)
-{
-    paging_free_log_dirty_bitmap(d);
-}
-
 /************************************************/
 /*           CODE FOR PAGING SUPPORT            */
 /************************************************/
@@ -547,6 +653,7 @@ void paging_vcpu_init(struct vcpu *v)
 int paging_domctl(struct domain *d, xen_domctl_shadow_op_t *sc,
                   XEN_GUEST_HANDLE(void) u_domctl)
 {
+    bool_t resuming = 0;
     int rc;
 
     if ( unlikely(d == current->domain) )
@@ -569,6 +676,20 @@ int paging_domctl(struct domain *d, xen_
         return -EINVAL;
     }
 
+    if ( d->arch.paging.preempt.vcpu )
+    {
+        if ( d->arch.paging.preempt.vcpu != current ||
+             d->arch.paging.preempt.op != sc->op )
+        {
+            printk(XENLOG_G_DEBUG
+                   "d%d:v%d: Paging op %#x on Dom%u with unfinished prior op %#x\n",
+                   current->domain->domain_id, current->vcpu_id,
+                   sc->op, d->domain_id, d->arch.paging.preempt.op);
+            return -EBUSY;
+        }
+        resuming = 1;
+    }
+
     rc = xsm_shadow_control(d, sc->op);
     if ( rc )
         return rc;
@@ -594,13 +714,13 @@ int paging_domctl(struct domain *d, xen_
 
     case XEN_DOMCTL_SHADOW_OP_OFF:
         if ( paging_mode_log_dirty(d) )
-            if ( (rc = paging_log_dirty_disable(d)) != 0 )
+            if ( (rc = paging_log_dirty_disable(d, resuming)) != 0 )
                 return rc;
         break;
 
     case XEN_DOMCTL_SHADOW_OP_CLEAN:
     case XEN_DOMCTL_SHADOW_OP_PEEK:
-        return paging_log_dirty_op(d, sc);
+        return paging_log_dirty_op(d, sc, resuming);
     }
 
     /* Here, dispatch domctl to the appropriate paging code */
@@ -611,18 +731,24 @@ int paging_domctl(struct domain *d, xen_
 }
 
 /* Call when destroying a domain */
-void paging_teardown(struct domain *d)
+int paging_teardown(struct domain *d)
 {
+    int rc;
+
     if ( hap_enabled(d) )
         hap_teardown(d);
     else
         shadow_teardown(d);
 
     /* clean up log dirty resources. */
-    paging_log_dirty_teardown(d);
+    rc = paging_free_log_dirty_bitmap(d, 0);
+    if ( rc == -EAGAIN )
+        return rc;
 
     /* Move populate-on-demand cache back to domain_list for destruction */
     p2m_pod_empty_cache(d);
+
+    return rc;
 }
 
 /* Call once all of the references to the domain have gone away */
--- a/xen/arch/x86/mm/shadow/common.c
+++ b/xen/arch/x86/mm/shadow/common.c
@@ -3829,8 +3829,7 @@ int shadow_domctl(struct domain *d, 
         paging_unlock(d);
         if ( preempted )
             /* Not finished.  Set up to re-run the call. */
-            rc = hypercall_create_continuation(
-                __HYPERVISOR_domctl, "h", u_domctl);
+            rc = -EAGAIN;
         else 
             /* Finished.  Return the new allocation */
             sc->mb = shadow_get_allocation(d);
--- a/xen/common/domain.c
+++ b/xen/common/domain.c
@@ -479,7 +479,6 @@ int domain_kill(struct domain *d)
         rc = domain_relinquish_resources(d);
         if ( rc != 0 )
         {
-            BUG_ON(rc != -EAGAIN);
             break;
         }
         if ( sched_move_domain(d, cpupool0) )
--- a/xen/include/asm-x86/domain.h
+++ b/xen/include/asm-x86/domain.h
@@ -193,6 +193,20 @@ struct paging_domain {
     struct hap_domain       hap;
     /* log dirty support */
     struct log_dirty_domain log_dirty;
+
+    /* preemption handling */
+    struct {
+        struct vcpu *vcpu;
+        unsigned int op;
+        union {
+            struct {
+                unsigned long done:PADDR_BITS - PAGE_SHIFT;
+                unsigned long i4:PAGETABLE_ORDER;
+                unsigned long i3:PAGETABLE_ORDER;
+            } log_dirty;
+        };
+    } preempt;
+
     /* alloc/free pages from the pool for paging-assistance structures
      * (used by p2m and log-dirty code for their tries) */
     struct page_info * (*alloc_page)(struct domain *d);
--- a/xen/include/asm-x86/paging.h
+++ b/xen/include/asm-x86/paging.h
@@ -141,9 +141,6 @@ struct paging_mode {
 /*****************************************************************************
  * Log dirty code */
 
-/* free log dirty bitmap resource */
-void paging_free_log_dirty_bitmap(struct domain *d);
-
 /* get the dirty bitmap for a specific range of pfns */
 void paging_log_dirty_range(struct domain *d,
                             unsigned long begin_pfn,
@@ -153,9 +150,6 @@ void paging_log_dirty_range(struct domai
 /* enable log dirty */
 int paging_log_dirty_enable(struct domain *d);
 
-/* disable log dirty */
-int paging_log_dirty_disable(struct domain *d);
-
 /* log dirty initialization */
 void paging_log_dirty_init(struct domain *d,
                            int  (*enable_log_dirty)(struct domain *d),
@@ -218,7 +212,7 @@ int paging_domctl(struct domain *d, xen_
                   XEN_GUEST_HANDLE(void) u_domctl);
 
 /* Call when destroying a domain */
-void paging_teardown(struct domain *d);
+int paging_teardown(struct domain *d);
 
 /* Call once all of the references to the domain have gone away */
 void paging_final_teardown(struct domain *d);