x86: account for shadow stack in exception-from-stub recovery

Dealing with exceptions raised from within emulation stubs involves
discarding return address (replaced by exception related information).
Such discarding of course also requires removing the corresponding entry
from the shadow stack.

Also amend the comment in fixup_exception_return(), to further clarify
why use of ptr[1] can't be an out-of-bounds access.

This is CVE-2023-46841 / XSA-451.

Fixes: 209fb9919b50 ("x86/extable: Adjust extable handling to be shadow stack compatible")
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>
master commit: 91f5f7a9154919a765c3933521760acffeddbf28
master date: 2024-02-27 13:49:22 +0100

diff --git a/xen/arch/x86/extable.c b/xen/arch/x86/extable.c
index c18cf97875643df1aea78a6aed081d3bb5f05ef5..f25f56b71430df9663b412a28592a5c0c7883e4a 100644 (file)
--- a/xen/arch/x86/extable.c
+++ b/xen/arch/x86/extable.c
@@ -86,26 +86,29 @@ search_one_extable(const struct exception_table_entry *first,
 }
 
 unsigned long
-search_exception_table(const struct cpu_user_regs *regs)
+search_exception_table(const struct cpu_user_regs *regs, unsigned long *stub_ra)
 {
     const struct virtual_region *region = find_text_region(regs->rip);
     unsigned long stub = this_cpu(stubs.addr);
 
     if ( region && region->ex )
+    {
+        *stub_ra = 0;
         return search_one_extable(region->ex, region->ex_end, regs->rip);
+    }
 
     if ( regs->rip >= stub + STUB_BUF_SIZE / 2 &&
          regs->rip < stub + STUB_BUF_SIZE &&
          regs->rsp > (unsigned long)regs &&
          regs->rsp < (unsigned long)get_cpu_info() )
     {
-        unsigned long retptr = *(unsigned long *)regs->rsp;
+        unsigned long retaddr = *(unsigned long *)regs->rsp, fixup;
 
-        region = find_text_region(retptr);
-        retptr = region && region->ex
-                 ? search_one_extable(region->ex, region->ex_end, retptr)
-                 : 0;
-        if ( retptr )
+        region = find_text_region(retaddr);
+        fixup = region && region->ex
+                ? search_one_extable(region->ex, region->ex_end, retaddr)
+                : 0;
+        if ( fixup )
         {
             /*
              * Put trap number and error code on the stack (in place of the
@@ -117,7 +120,8 @@ search_exception_table(const struct cpu_user_regs *regs)
             };
 
             *(unsigned long *)regs->rsp = token.raw;
-            return retptr;
+            *stub_ra = retaddr;
+            return fixup;
         }
     }
 

diff --git a/xen/arch/x86/traps.c b/xen/arch/x86/traps.c
index a142a63dd8691f95e77280f3733510aeb3845bd9..d74dcc343ba0c8dfceb470042beab936c2154d2c 100644 (file)
--- a/xen/arch/x86/traps.c
+++ b/xen/arch/x86/traps.c
@@ -895,7 +895,7 @@ static void do_reserved_trap(struct cpu_user_regs *regs)
 }
 
 static void fixup_exception_return(struct cpu_user_regs *regs,
-                                   unsigned long fixup)
+                                   unsigned long fixup, unsigned long stub_ra)
 {
     if ( IS_ENABLED(CONFIG_XEN_SHSTK) )
     {
@@ -912,7 +912,8 @@ static void fixup_exception_return(struct cpu_user_regs *regs,
             /*
              * Search for %rip.  The shstk currently looks like this:
              *
-             *   ...  [Likely pointed to by SSP]
+             *   tok  [Supervisor token, == &tok | BUSY, only with FRED inactive]
+             *   ...  [Pointed to by SSP for most exceptions, empty in IST cases]
              *   %cs  [== regs->cs]
              *   %rip [== regs->rip]
              *   SSP  [Likely points to 3 slots higher, above %cs]
@@ -930,7 +931,56 @@ static void fixup_exception_return(struct cpu_user_regs *regs,
              */
             if ( ptr[0] == regs->rip && ptr[1] == regs->cs )
             {
+                unsigned long primary_shstk =
+                    (ssp & ~(STACK_SIZE - 1)) +
+                    (PRIMARY_SHSTK_SLOT + 1) * PAGE_SIZE - 8;
+
                 wrss(fixup, ptr);
+
+                if ( !stub_ra )
+                    goto shstk_done;
+
+                /*
+                 * Stub recovery ought to happen only when the outer context
+                 * was on the main shadow stack.  We need to also "pop" the
+                 * stub's return address from the interrupted context's shadow
+                 * stack.  That is,
+                 * - if we're still on the main stack, we need to move the
+                 *   entire stack (up to and including the exception frame)
+                 *   up by one slot, incrementing the original SSP in the
+                 *   exception frame,
+                 * - if we're on an IST stack, we need to increment the
+                 *   original SSP.
+                 */
+                BUG_ON((ptr[-1] ^ primary_shstk) >> PAGE_SHIFT);
+
+                if ( (ssp ^ primary_shstk) >> PAGE_SHIFT )
+                {
+                    /*
+                     * We're on an IST stack.  First make sure the two return
+                     * addresses actually match.  Then increment the interrupted
+                     * context's SSP.
+                     */
+                    BUG_ON(stub_ra != *(unsigned long*)ptr[-1]);
+                    wrss(ptr[-1] + 8, &ptr[-1]);
+                    goto shstk_done;
+                }
+
+                /* Make sure the two return addresses actually match. */
+                BUG_ON(stub_ra != ptr[2]);
+
+                /* Move exception frame, updating SSP there. */
+                wrss(ptr[1], &ptr[2]); /* %cs */
+                wrss(ptr[0], &ptr[1]); /* %rip */
+                wrss(ptr[-1] + 8, &ptr[0]); /* SSP */
+
+                /* Move all newer entries. */
+                while ( --ptr != _p(ssp) )
+                    wrss(ptr[-1], &ptr[0]);
+
+                /* Finally account for our own stack having shifted up. */
+                asm volatile ( "incsspd %0" :: "r" (2) );
+
                 goto shstk_done;
             }
         }
@@ -951,7 +1001,8 @@ static void fixup_exception_return(struct cpu_user_regs *regs,
 
 static bool extable_fixup(struct cpu_user_regs *regs, bool print)
 {
-    unsigned long fixup = search_exception_table(regs);
+    unsigned long stub_ra = 0;
+    unsigned long fixup = search_exception_table(regs, &stub_ra);
 
     if ( unlikely(fixup == 0) )
         return false;
@@ -965,7 +1016,7 @@ static bool extable_fixup(struct cpu_user_regs *regs, bool print)
                vec_name(regs->entry_vector), regs->error_code,
                _p(regs->rip), _p(regs->rip), _p(fixup));
 
-    fixup_exception_return(regs, fixup);
+    fixup_exception_return(regs, fixup, stub_ra);
     this_cpu(last_extable_addr) = regs->rip;
 
     return true;
@@ -1256,7 +1307,7 @@ void do_invalid_op(struct cpu_user_regs *regs)
         void (*fn)(struct cpu_user_regs *) = bug_ptr(bug);
 
         fn(regs);
-        fixup_exception_return(regs, (unsigned long)eip);
+        fixup_exception_return(regs, (unsigned long)eip, 0);
         return;
     }
 
@@ -1277,7 +1328,7 @@ void do_invalid_op(struct cpu_user_regs *regs)
     case BUGFRAME_warn:
         printk("Xen WARN at %s%s:%d\n", prefix, filename, lineno);
         show_execution_state(regs);
-        fixup_exception_return(regs, (unsigned long)eip);
+        fixup_exception_return(regs, (unsigned long)eip, 0);
         return;
 
     case BUGFRAME_bug:

diff --git a/xen/include/asm-x86/uaccess.h b/xen/include/asm-x86/uaccess.h
index 684fccd95c1a69f4b4f3b94ca2041bfb2c19bcbf..74bb222c031098dee02a0b23cc288d235466b55c 100644 (file)
--- a/xen/include/asm-x86/uaccess.h
+++ b/xen/include/asm-x86/uaccess.h
@@ -421,7 +421,8 @@ union stub_exception_token {
     unsigned long raw;
 };
 
-extern unsigned long search_exception_table(const struct cpu_user_regs *regs);
+extern unsigned long search_exception_table(const struct cpu_user_regs *regs,
+                                            unsigned long *stub_ra);
 extern void sort_exception_tables(void);
 extern void sort_exception_table(struct exception_table_entry *start,
                                  const struct exception_table_entry *stop);