path: root/sysutils/xenkernel413/patches/patch-XSA401

$NetBSD: patch-XSA401,v 1.1 2022/06/24 13:47:37 bouyer Exp $

From: Andrew Cooper <andrew.cooper3@citrix.com>
Subject: x86/pv: Clean up _get_page_type()

Various fixes for clarity, ahead of making complicated changes.

 * Split the overflow check out of the if/else chain for type handling, as
   it's somewhat unrelated.
 * Comment the main if/else chain to explain what is going on.  Adjust one
   ASSERT() and state the bit layout for validate-locked and partial states.
 * Correct the comment about TLB flushing, as it's backwards.  The problem
   case is when writeable mappings are retained to a page becoming read-only,
   as it allows the guest to bypass Xen's safety checks for updates.
 * Reduce the scope of 'y'.  It is an artefact of the cmpxchg loop and not
   valid for use by subsequent logic.  Switch to using ACCESS_ONCE() to treat
   all reads as explicitly volatile.  The only thing preventing the validated
   wait-loop being infinite is the compiler barrier hidden in cpu_relax().
 * Replace one page_get_owner(page) with the already-calculated 'd' already in
   scope.

No functional change.

This is part of XSA-401 / CVE-2022-26362.

Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
Signed-off-by: George Dunlap <george.dunlap@eu.citrix.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: George Dunlap <george.dunlap@citrix.com>

diff --git a/xen/arch/x86/mm.c b/xen/arch/x86/mm.c
index ad89bfb45fff..96738b027827 100644
--- xen/arch/x86/mm.c.orig
+++ xen/arch/x86/mm.c
@@ -2978,16 +2978,17 @@ static int _put_page_type(struct page_info *page, unsigned int flags,
 static int _get_page_type(struct page_info *page, unsigned long type,
                           bool preemptible)
 {
-    unsigned long nx, x, y = page->u.inuse.type_info;
+    unsigned long nx, x;
     int rc = 0;
 
     ASSERT(!(type & ~(PGT_type_mask | PGT_pae_xen_l2)));
     ASSERT(!in_irq());
 
-    for ( ; ; )
+    for ( unsigned long y = ACCESS_ONCE(page->u.inuse.type_info); ; )
     {
         x  = y;
         nx = x + 1;
+
         if ( unlikely((nx & PGT_count_mask) == 0) )
         {
             gdprintk(XENLOG_WARNING,
@@ -2995,8 +2996,15 @@ static int _get_page_type(struct page_info *page, unsigned long type,
                      mfn_x(page_to_mfn(page)));
             return -EINVAL;
         }
-        else if ( unlikely((x & PGT_count_mask) == 0) )
+
+        if ( unlikely((x & PGT_count_mask) == 0) )
         {
+            /*
+             * Typeref 0 -> 1.
+             *
+             * Type changes are permitted when the typeref is 0.  If the type
+             * actually changes, the page needs re-validating.
+             */
             struct domain *d = page_get_owner(page);
 
             if ( d && shadow_mode_enabled(d) )
@@ -3007,8 +3015,8 @@ static int _get_page_type(struct page_info *page, unsigned long type,
             {
                 /*
                  * On type change we check to flush stale TLB entries. It is
-                 * vital that no other CPUs are left with mappings of a frame
-                 * which is about to become writeable to the guest.
+                 * vital that no other CPUs are left with writeable mappings
+                 * to a frame which is intending to become pgtable/segdesc.
                  */
                 cpumask_t *mask = this_cpu(scratch_cpumask);
 
@@ -3020,7 +3028,7 @@ static int _get_page_type(struct page_info *page, unsigned long type,
 
                 if ( unlikely(!cpumask_empty(mask)) &&
                      /* Shadow mode: track only writable pages. */
-                     (!shadow_mode_enabled(page_get_owner(page)) ||
+                     (!shadow_mode_enabled(d) ||
                       ((nx & PGT_type_mask) == PGT_writable_page)) )
                 {
                     perfc_incr(need_flush_tlb_flush);
@@ -3041,7 +3049,14 @@ static int _get_page_type(struct page_info *page, unsigned long type,
         }
         else if ( unlikely((x & (PGT_type_mask|PGT_pae_xen_l2)) != type) )
         {
-            /* Don't log failure if it could be a recursive-mapping attempt. */
+            /*
+             * else, we're trying to take a new reference, of the wrong type.
+             *
+             * This (being able to prohibit use of the wrong type) is what the
+             * typeref system exists for, but skip printing the failure if it
+             * looks like a recursive mapping, as subsequent logic might
+             * ultimately permit the attempt.
+             */
             if ( ((x & PGT_type_mask) == PGT_l2_page_table) &&
                  (type == PGT_l1_page_table) )
                 return -EINVAL;
@@ -3060,18 +3075,46 @@ static int _get_page_type(struct page_info *page, unsigned long type,
         }
         else if ( unlikely(!(x & PGT_validated)) )
         {
+            /*
+             * else, the count is non-zero, and we're grabbing the right type;
+             * but the page hasn't been validated yet.
+             *
+             * The page is in one of two states (depending on PGT_partial),
+             * and should have exactly one reference.
+             */
+            ASSERT((x & (PGT_type_mask | PGT_count_mask)) == (type | 1));
+
             if ( !(x & PGT_partial) )
             {
-                /* Someone else is updating validation of this page. Wait... */
+                /*
+                 * The page has been left in the "validate locked" state
+                 * (i.e. PGT_[type] | 1) which means that a concurrent caller
+                 * of _get_page_type() is in the middle of validation.
+                 *
+                 * Spin waiting for the concurrent user to complete (partial
+                 * or fully validated), then restart our attempt to acquire a
+                 * type reference.
+                 */
                 do {
                     if ( preemptible && hypercall_preempt_check() )
                         return -EINTR;
                     cpu_relax();
-                } while ( (y = page->u.inuse.type_info) == x );
+                } while ( (y = ACCESS_ONCE(page->u.inuse.type_info)) == x );
                 continue;
             }
-            /* Type ref count was left at 1 when PGT_partial got set. */
-            ASSERT((x & PGT_count_mask) == 1);
+
+            /*
+             * The page has been left in the "partial" state
+             * (i.e., PGT_[type] | PGT_partial | 1).
+             *
+             * Rather than bumping the type count, we need to try to grab the
+             * validation lock; if we succeed, we need to validate the page,
+             * then drop the general ref associated with the PGT_partial bit.
+             *
+             * We grab the validation lock by setting nx to (PGT_[type] | 1)
+             * (i.e., non-zero type count, neither PGT_validated nor
+             * PGT_partial set).
+             */
             nx = x & ~PGT_partial;
         }
 
@@ -3116,6 +3159,13 @@ static int _get_page_type(struct page_info *page, unsigned long type,
     }
 
  out:
+    /*
+     * Did we drop the PGT_partial bit when acquiring the typeref?  If so,
+     * drop the general reference that went along with it.
+     *
+     * N.B. validate_page() may have have re-set PGT_partial, not reflected in
+     * nx, but will have taken an extra ref when doing so.
+     */
     if ( (x & PGT_partial) && !(nx & PGT_partial) )
         put_page(page);
 
From: Andrew Cooper <andrew.cooper3@citrix.com>
Subject: x86/pv: Fix ABAC cmpxchg() race in _get_page_type()

_get_page_type() suffers from a race condition where it incorrectly assumes
that because 'x' was read and a subsequent a cmpxchg() succeeds, the type
cannot have changed in-between.  Consider:

CPU A:
  1. Creates an L2e referencing pg
     `-> _get_page_type(pg, PGT_l1_page_table), sees count 0, type PGT_writable_page
  2.     Issues flush_tlb_mask()
CPU B:
  3. Creates a writeable mapping of pg
     `-> _get_page_type(pg, PGT_writable_page), count increases to 1
  4. Writes into new mapping, creating a TLB entry for pg
  5. Removes the writeable mapping of pg
     `-> _put_page_type(pg), count goes back down to 0
CPU A:
  7.     Issues cmpxchg(), setting count 1, type PGT_l1_page_table

CPU B now has a writeable mapping to pg, which Xen believes is a pagetable and
suitably protected (i.e. read-only).  The TLB flush in step 2 must be deferred
until after the guest is prohibited from creating new writeable mappings,
which is after step 7.

Defer all safety actions until after the cmpxchg() has successfully taken the
intended typeref, because that is what prevents concurrent users from using
the old type.

Also remove the early validation for writeable and shared pages.  This removes
race conditions where one half of a parallel mapping attempt can return
successfully before:
 * The IOMMU pagetables are in sync with the new page type
 * Writeable mappings to shared pages have been torn down

This is part of XSA-401 / CVE-2022-26362.

Reported-by: Jann Horn <jannh@google.com>
Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: George Dunlap <george.dunlap@citrix.com>

diff --git a/xen/arch/x86/mm.c b/xen/arch/x86/mm.c
index 96738b027827..ee91c7fe5f69 100644
--- xen/arch/x86/mm.c.orig
+++ xen/arch/x86/mm.c
@@ -3005,46 +3005,12 @@ static int _get_page_type(struct page_info *page, unsigned long type,
              * Type changes are permitted when the typeref is 0.  If the type
              * actually changes, the page needs re-validating.
              */
-            struct domain *d = page_get_owner(page);
-
-            if ( d && shadow_mode_enabled(d) )
-               shadow_prepare_page_type_change(d, page, type);
 
             ASSERT(!(x & PGT_pae_xen_l2));
             if ( (x & PGT_type_mask) != type )
             {
-                /*
-                 * On type change we check to flush stale TLB entries. It is
-                 * vital that no other CPUs are left with writeable mappings
-                 * to a frame which is intending to become pgtable/segdesc.
-                 */
-                cpumask_t *mask = this_cpu(scratch_cpumask);
-
-                BUG_ON(in_irq());
-                cpumask_copy(mask, d->dirty_cpumask);
-
-                /* Don't flush if the timestamp is old enough */
-                tlbflush_filter(mask, page->tlbflush_timestamp);
-
-                if ( unlikely(!cpumask_empty(mask)) &&
-                     /* Shadow mode: track only writable pages. */
-                     (!shadow_mode_enabled(d) ||
-                      ((nx & PGT_type_mask) == PGT_writable_page)) )
-                {
-                    perfc_incr(need_flush_tlb_flush);
-                    flush_tlb_mask(mask);
-                }
-
-                /* We lose existing type and validity. */
                 nx &= ~(PGT_type_mask | PGT_validated);
                 nx |= type;
-
-                /*
-                 * No special validation needed for writable pages.
-                 * Page tables and GDT/LDT need to be scanned for validity.
-                 */
-                if ( type == PGT_writable_page || type == PGT_shared_page )
-                    nx |= PGT_validated;
             }
         }
         else if ( unlikely((x & (PGT_type_mask|PGT_pae_xen_l2)) != type) )
@@ -3125,6 +3091,46 @@ static int _get_page_type(struct page_info *page, unsigned long type,
             return -EINTR;
     }
 
+    /*
+     * One typeref has been taken and is now globally visible.
+     *
+     * The page is either in the "validate locked" state (PGT_[type] | 1) or
+     * fully validated (PGT_[type] | PGT_validated | >0).
+     */
+
+    if ( unlikely((x & PGT_count_mask) == 0) )
+    {
+        struct domain *d = page_get_owner(page);
+
+        if ( d && shadow_mode_enabled(d) )
+            shadow_prepare_page_type_change(d, page, type);
+
+        if ( (x & PGT_type_mask) != type )
+        {
+            /*
+             * On type change we check to flush stale TLB entries. It is
+             * vital that no other CPUs are left with writeable mappings
+             * to a frame which is intending to become pgtable/segdesc.
+             */
+            cpumask_t *mask = this_cpu(scratch_cpumask);
+
+            BUG_ON(in_irq());
+            cpumask_copy(mask, d->dirty_cpumask);
+
+            /* Don't flush if the timestamp is old enough */
+            tlbflush_filter(mask, page->tlbflush_timestamp);
+
+            if ( unlikely(!cpumask_empty(mask)) &&
+                 /* Shadow mode: track only writable pages. */
+                 (!shadow_mode_enabled(d) ||
+                  ((nx & PGT_type_mask) == PGT_writable_page)) )
+            {
+                perfc_incr(need_flush_tlb_flush);
+                flush_tlb_mask(mask);
+            }
+        }
+    }
+
     if ( unlikely((x & PGT_type_mask) != type) )
     {
         /* Special pages should not be accessible from devices. */
@@ -3149,13 +3155,25 @@ static int _get_page_type(struct page_info *page, unsigned long type,
 
     if ( unlikely(!(nx & PGT_validated)) )
     {
-        if ( !(x & PGT_partial) )
+        /*
+         * No special validation needed for writable or shared pages.  Page
+         * tables and GDT/LDT need to have their contents audited.
+         *
+         * per validate_page(), non-atomic updates are fine here.
+         */
+        if ( type == PGT_writable_page || type == PGT_shared_page )
+            page->u.inuse.type_info |= PGT_validated;
+        else
         {
-            page->nr_validated_ptes = 0;
-            page->partial_flags = 0;
-            page->linear_pt_count = 0;
+            if ( !(x & PGT_partial) )
+            {
+                page->nr_validated_ptes = 0;
+                page->partial_flags = 0;
+                page->linear_pt_count = 0;
+            }
+
+            rc = alloc_page_type(page, type, preemptible);
         }
-        rc = alloc_page_type(page, type, preemptible);
     }
 
  out: