Enable shstk2

build/Jamfile.v2

@@ -797,7 +797,7 @@ alias asm_sources
      asm/jump_x86_64_sysv_elf_gas.S
      asm/ontop_x86_64_sysv_elf_gas.S
    : <abi>x32
-     <address-model>64
+     <address-model>32
      <architecture>x86
      <binary-format>elf
      <toolset>clang
@@ -808,7 +808,7 @@ alias asm_sources
      asm/jump_x86_64_sysv_elf_gas.S
      asm/ontop_x86_64_sysv_elf_gas.S
    : <abi>x32
-     <address-model>64
+     <address-model>32
      <architecture>x86
      <binary-format>elf
      <toolset>gcc
@@ -819,7 +819,7 @@ alias asm_sources
      asm/jump_x86_64_sysv_elf_gas.S
      asm/ontop_x86_64_sysv_elf_gas.S
    : <abi>x32
-     <address-model>64
+     <address-model>32
      <architecture>x86
      <binary-format>elf
      <toolset>intel

doc/stack.qbk

@@ -364,4 +364,20 @@ when linking against Boost binaries.
 [endsect]
 
 
+[section:stack_protect Support for stack protection]
+
+Compiler switch `-fstack-protector` changes the default context switching logic.
+Users must define `BOOST_CONTEXT_TLS_STACK_PROTECTOR` before including any
+Boost.Context headers if stack protection is enabled.
+
+[endsect]
+
+[section:shadow_stack Support for shadow stack protection]
+
+Shadow stack is part of Intel's Control-Flow Enforcement Technology. Users must
+define `BOOST_CONTEXT_SHADOW_STACK` before including any Boost.Context headers
+if shadow stack protection is enabled.
+
+[endsect]
+
 [endsect]

include/boost/context/continuation_fcontext.hpp

@@ -44,6 +44,17 @@
 # include BOOST_ABI_PREFIX
 #endif
 
+#if defined __CET__
+#  include <cet.h>
+#  include <sys/mman.h>
+#  define SHSTK_ENABLED (__CET__ & 0x2)
+#  define BOOST_CONTEXT_SHADOW_STACK (SHSTK_ENABLED && SHADOW_STACK_SYSCALL)
+#  define __NR_map_shadow_stack 451
+#ifndef SHADOW_STACK_SET_TOKEN
+#  define SHADOW_STACK_SET_TOKEN 0x1
+#endif
+#endif
+
 #if defined(BOOST_MSVC)
 # pragma warning(push)
 # pragma warning(disable: 4702)
@@ -62,6 +73,12 @@ transfer_t context_unwind( transfer_t t) {
 template< typename Rec >
 transfer_t context_exit( transfer_t t) noexcept {
     Rec * rec = static_cast< Rec * >( t.data);
+#if BOOST_CONTEXT_SHADOW_STACK
+    // destory shadow stack
+    std::size_t ss_size = *((unsigned long*)(reinterpret_cast< uintptr_t >( rec)- 16));
+    long unsigned int ss_base = *((unsigned long*)(reinterpret_cast< uintptr_t >( rec)- 8));
+    munmap((void *)ss_base, ss_size);
+#endif
     // destroy context stack
     rec->deallocate();
     return { nullptr, nullptr };
@@ -168,6 +185,25 @@ fcontext_t create_context1( StackAlloc && salloc, Fn && fn) {
             reinterpret_cast< uintptr_t >( sctx.sp) - static_cast< uintptr_t >( sctx.size) );
     // create fast-context
     const std::size_t size = reinterpret_cast< uintptr_t >( stack_top) - reinterpret_cast< uintptr_t >( stack_bottom);
+
+#if BOOST_CONTEXT_SHADOW_STACK
+    std::size_t ss_size = size >> 5;
+    // align shadow stack to 8 bytes.
+	ss_size = (ss_size + 7) & ~7;
+    // Todo: shadow stack occupies at least 4KB
+    ss_size = (ss_size > 4096) ? size : 4096;
+    // create shadow stack
+    void *ss_base = (void *)syscall(__NR_map_shadow_stack, 0, ss_size, SHADOW_STACK_SET_TOKEN);
+    BOOST_ASSERT(ss_base != -1);
+    unsigned long ss_sp = (unsigned long)ss_base + ss_size;
+    /* pass the shadow stack pointer to make_fcontext
+	 i.e., link the new shadow stack with the new fcontext
+	 TODO should be a better way? */
+    *((unsigned long*)(reinterpret_cast< uintptr_t >( stack_top)- 8)) = ss_sp;
+    /* Todo: place shadow stack info in 64byte gap */
+    *((unsigned long*)(reinterpret_cast< uintptr_t >( storage)- 8)) = (unsigned long) ss_base;
+    *((unsigned long*)(reinterpret_cast< uintptr_t >( storage)- 16)) = ss_size;
+#endif
     const fcontext_t fctx = make_fcontext( stack_top, size, & context_entry< Record >);
     BOOST_ASSERT( nullptr != fctx);
     // transfer control structure to context-stack
@@ -190,6 +226,25 @@ fcontext_t create_context2( preallocated palloc, StackAlloc && salloc, Fn && fn)
             reinterpret_cast< uintptr_t >( palloc.sctx.sp) - static_cast< uintptr_t >( palloc.sctx.size) );
     // create fast-context
     const std::size_t size = reinterpret_cast< uintptr_t >( stack_top) - reinterpret_cast< uintptr_t >( stack_bottom);
+
+#if BOOST_CONTEXT_SHADOW_STACK
+    std::size_t ss_size = size >> 5;
+    // align shadow stack to 8 bytes.
+	ss_size = (ss_size + 7) & ~7;
+    // Todo: shadow stack occupies at least 4KB
+    ss_size = (ss_size > 4096) ? size : 4096;
+    // create shadow stack
+    void *ss_base = (void *)syscall(__NR_map_shadow_stack, 0, ss_size, SHADOW_STACK_SET_TOKEN);
+    BOOST_ASSERT(ss_base != -1);
+    unsigned long ss_sp = (unsigned long)ss_base + ss_size;
+    /* pass the shadow stack pointer to make_fcontext
+	 i.e., link the new shadow stack with the new fcontext
+	 TODO should be a better way? */
+    *((unsigned long*)(reinterpret_cast< uintptr_t >( stack_top)- 8)) = ss_sp;
+    /* Todo: place shadow stack info in 64byte gap */
+    *((unsigned long*)(reinterpret_cast< uintptr_t >( storage)- 8)) = (unsigned long) ss_base;
+    *((unsigned long*)(reinterpret_cast< uintptr_t >( storage)- 16)) = ss_size;
+#endif
     const fcontext_t fctx = make_fcontext( stack_top, size, & context_entry< Record >);
     BOOST_ASSERT( nullptr != fctx);
     // transfer control structure to context-stack

include/boost/context/fiber_fcontext.hpp

@@ -44,6 +44,17 @@
 # include BOOST_ABI_PREFIX
 #endif
 
+#if defined __CET__
+#  include <cet.h>
+#  include <sys/mman.h>
+#  define SHSTK_ENABLED (__CET__ & 0x2)
+#  define BOOST_CONTEXT_SHADOW_STACK (SHSTK_ENABLED && SHADOW_STACK_SYSCALL)
+#  define __NR_map_shadow_stack 451
+#ifndef SHADOW_STACK_SET_TOKEN
+#  define SHADOW_STACK_SET_TOKEN 0x1
+#endif
+#endif
+
 #if defined(BOOST_MSVC)
 # pragma warning(push)
 # pragma warning(disable: 4702)
@@ -62,6 +73,12 @@ transfer_t fiber_unwind( transfer_t t) {
 template< typename Rec >
 transfer_t fiber_exit( transfer_t t) noexcept {
     Rec * rec = static_cast< Rec * >( t.data);
+#if BOOST_CONTEXT_SHADOW_STACK
+    // destory shadow stack
+    std::size_t ss_size = *((unsigned long*)(reinterpret_cast< uintptr_t >( rec)- 16));
+    long unsigned int ss_base = *((unsigned long*)(reinterpret_cast< uintptr_t >( rec)- 8));
+    munmap((void *)ss_base, ss_size);
+#endif
     // destroy context stack
     rec->deallocate();
     return { nullptr, nullptr };
@@ -165,6 +182,25 @@ fcontext_t create_fiber1( StackAlloc && salloc, Fn && fn) {
             reinterpret_cast< uintptr_t >( sctx.sp) - static_cast< uintptr_t >( sctx.size) );
     // create fast-context
     const std::size_t size = reinterpret_cast< uintptr_t >( stack_top) - reinterpret_cast< uintptr_t >( stack_bottom);
+
+#if BOOST_CONTEXT_SHADOW_STACK
+    std::size_t ss_size = size >> 5;
+    // align shadow stack to 8 bytes.
+	ss_size = (ss_size + 7) & ~7;
+    // Todo: shadow stack occupies at least 4KB
+    ss_size = (ss_size > 4096) ? size : 4096;
+    // create shadow stack
+    void *ss_base = (void *)syscall(__NR_map_shadow_stack, 0, ss_size, SHADOW_STACK_SET_TOKEN);
+    BOOST_ASSERT(ss_base != -1);
+    unsigned long ss_sp = (unsigned long)ss_base + ss_size;
+    /* pass the shadow stack pointer to make_fcontext
+	 i.e., link the new shadow stack with the new fcontext
+	 TODO should be a better way? */
+    *((unsigned long*)(reinterpret_cast< uintptr_t >( stack_top)- 8)) = ss_sp;
+    /* Todo: place shadow stack info in 64byte gap */
+    *((unsigned long*)(reinterpret_cast< uintptr_t >( storage)- 8)) = (unsigned long) ss_base;
+    *((unsigned long*)(reinterpret_cast< uintptr_t >( storage)- 16)) = ss_size;
+#endif
     const fcontext_t fctx = make_fcontext( stack_top, size, & fiber_entry< Record >);
     BOOST_ASSERT( nullptr != fctx);
     // transfer control structure to context-stack
@@ -187,6 +223,25 @@ fcontext_t create_fiber2( preallocated palloc, StackAlloc && salloc, Fn && fn) {
             reinterpret_cast< uintptr_t >( palloc.sctx.sp) - static_cast< uintptr_t >( palloc.sctx.size) );
     // create fast-context
     const std::size_t size = reinterpret_cast< uintptr_t >( stack_top) - reinterpret_cast< uintptr_t >( stack_bottom);
+
+#if BOOST_CONTEXT_SHADOW_STACK
+    std::size_t ss_size = size >> 5;
+    // align shadow stack to 8 bytes.
+	ss_size = (ss_size + 7) & ~7;
+    // Todo: shadow stack occupies at least 4KB
+    ss_size = (ss_size > 4096) ? size : 4096;
+    // create shadow stack
+    void *ss_base = (void *)syscall(__NR_map_shadow_stack, 0, ss_size, SHADOW_STACK_SET_TOKEN);
+    BOOST_ASSERT(ss_base != -1);
+    unsigned long ss_sp = (unsigned long)ss_base + ss_size;
+    /* pass the shadow stack pointer to make_fcontext
+	 i.e., link the new shadow stack with the new fcontext
+	 TODO should be a better way? */
+    *((unsigned long*)(reinterpret_cast< uintptr_t >( stack_top)- 8)) = ss_sp;
+    /* Todo: place shadow stack info in 64byte gap */
+    *((unsigned long*)(reinterpret_cast< uintptr_t >( storage)- 8)) = (unsigned long) ss_base;
+    *((unsigned long*)(reinterpret_cast< uintptr_t >( storage)- 16)) = ss_size;
+#endif
     const fcontext_t fctx = make_fcontext( stack_top, size, & fiber_entry< Record >);
     BOOST_ASSERT( nullptr != fctx);
     // transfer control structure to context-stack
@@ -204,14 +259,6 @@ class fiber {
     friend detail::transfer_t
     detail::fiber_ontop( detail::transfer_t);
 
-    template< typename StackAlloc, typename Fn >
-    friend fiber
-    callcc( std::allocator_arg_t, StackAlloc &&, Fn &&);
-
-    template< typename StackAlloc, typename Fn >
-    friend fiber
-    callcc( std::allocator_arg_t, preallocated, StackAlloc &&, Fn &&);
-
     detail::fcontext_t  fctx_{ nullptr };
 
     fiber( detail::fcontext_t fctx) noexcept :

include/boost/context/fiber_ucontext.hpp

@@ -397,14 +397,6 @@ class BOOST_CONTEXT_DECL fiber {
 	template< typename Ctx, typename StackAlloc, typename Fn >
 	friend detail::fiber_activation_record * detail::create_fiber2( preallocated, StackAlloc &&, Fn &&);
 
-    template< typename StackAlloc, typename Fn >
-    friend fiber
-    callcc( std::allocator_arg_t, StackAlloc &&, Fn &&);
-
-    template< typename StackAlloc, typename Fn >
-    friend fiber
-    callcc( std::allocator_arg_t, preallocated, StackAlloc &&, Fn &&);
-
     detail::fiber_activation_record   *   ptr_{ nullptr };
 
     fiber( detail::fiber_activation_record * ptr) noexcept :

include/boost/context/fiber_winfib.hpp

@@ -290,14 +290,6 @@ class BOOST_CONTEXT_DECL fiber {
     template< typename Ctx, typename StackAlloc, typename Fn >
     friend detail::fiber_activation_record * detail::create_fiber2( preallocated, StackAlloc &&, Fn &&);
 
-    template< typename StackAlloc, typename Fn >
-    friend fiber
-    callcc( std::allocator_arg_t, StackAlloc &&, Fn &&);
-
-    template< typename StackAlloc, typename Fn >
-    friend fiber
-    callcc( std::allocator_arg_t, preallocated, StackAlloc &&, Fn &&);
-
     detail::fiber_activation_record   *   ptr_{ nullptr };
 
     fiber( detail::fiber_activation_record * ptr) noexcept :

src/asm/jump_i386_sysv_elf_gas.S

@@ -12,14 +12,14 @@
  *  ----------------------------------------------------------------------------------  *
  *  |   0x0   |   0x4   |   0x8   |   0xc   |   0x10   |   0x14  |   0x18  |   0x1c  |  *
  *  ----------------------------------------------------------------------------------  *
- *  | fc_mxcsr|fc_x87_cw|   EDI   |   ESI   |   EBX    |   EBP   |   EIP   |  hidden |  *
+ *  | fc_mxcsr|fc_x87_cw|  guard  |   EDI   |   ESI    |   EBX   |   EBP   |   EIP   |  *
  *  ----------------------------------------------------------------------------------  *
  *  ----------------------------------------------------------------------------------  *
  *  |    8    |    9    |    10   |    11   |    12    |    13   |    14   |    15   |  *
  *  ----------------------------------------------------------------------------------  *
- *  |   0x20  |   0x24  |                                                            |  *
+ *  |   0x20  |   0x24  |   0x28  |                                                  |  *
  *  ----------------------------------------------------------------------------------  *
- *  |    to   |   data  |                                                            |  *
+ *  |  hidden |    to   |   data  |                                                  |  *
  *  ----------------------------------------------------------------------------------  *
  *                                                                                      *
  ****************************************************************************************/
@@ -30,50 +30,60 @@
 .align 2
 .type jump_fcontext,@function
 jump_fcontext:
-    leal  -0x18(%esp), %esp  /* prepare stack */
+    leal  -0x1c(%esp), %esp  /* prepare stack */
 
 #if !defined(BOOST_USE_TSX)
     stmxcsr  (%esp)     /* save MMX control- and status-word */
     fnstcw   0x4(%esp)  /* save x87 control-word */
 #endif
 
-    movl  %edi, 0x8(%esp)  /* save EDI */
-    movl  %esi, 0xc(%esp)  /* save ESI */
-    movl  %ebx, 0x10(%esp)  /* save EBX */
-    movl  %ebp, 0x14(%esp)  /* save EBP */
+#if defined(BOOST_CONTEXT_TLS_STACK_PROTECTOR)
+    movl  %gs:0x14, %ecx    /* read stack guard from TLS record */
+    movl  %ecx, 0x8(%esp)   /* save stack guard */
+#endif
+
+    movl  %edi, 0xc(%esp)   /* save EDI */
+    movl  %esi, 0x10(%esp)  /* save ESI */
+    movl  %ebx, 0x14(%esp)  /* save EBX */
+    movl  %ebp, 0x18(%esp)  /* save EBP */
 
     /* store ESP (pointing to context-data) in ECX */
     movl  %esp, %ecx
 
     /* first arg of jump_fcontext() == fcontext to jump to */
-    movl  0x20(%esp), %eax
+    movl  0x24(%esp), %eax
 
     /* second arg of jump_fcontext() == data to be transferred */
-    movl  0x24(%esp), %edx
+    movl  0x28(%esp), %edx
 
     /* restore ESP (pointing to context-data) from EAX */
     movl  %eax, %esp
 
     /* address of returned transport_t */
-    movl 0x1c(%esp), %eax
+    movl 0x20(%esp), %eax
     /* return parent fcontext_t */
     movl  %ecx, (%eax)
     /* return data */
     movl %edx, 0x4(%eax)
 
-    movl  0x18(%esp), %ecx  /* restore EIP */
+    movl  0x1c(%esp), %ecx  /* restore EIP */
 
 #if !defined(BOOST_USE_TSX)
     ldmxcsr  (%esp)     /* restore MMX control- and status-word */
     fldcw    0x4(%esp)  /* restore x87 control-word */
 #endif
 
-    movl  0x8(%esp), %edi  /* restore EDI */
-    movl  0xc(%esp), %esi  /* restore ESI */
-    movl  0x10(%esp), %ebx  /* restore EBX */
-    movl  0x14(%esp), %ebp  /* restore EBP */
+#if defined(BOOST_CONTEXT_TLS_STACK_PROTECTOR)
+    movl  0x8(%esp), %edx  /* load stack guard */
+    movl  %edx, %gs:0x14   /* restore stack guard to TLS record */
+#endif
+
+    movl  0xc(%esp), %edi  /* restore EDI */
+    movl  0x10(%esp), %esi  /* restore ESI */
+    movl  0x14(%esp), %ebx  /* restore EBX */
+    movl  0x18(%esp), %ebp  /* restore EBP */
 
-    leal  0x20(%esp), %esp  /* prepare stack */
+    leal  0x24(%esp), %esp  /* prepare stack */
 
     /* jump to context */
     jmp *%ecx