CONTRACTS: support bounded user defined memory-predicates

Author: Remi Delmas

regression/contracts-dfcc/user-defined-memory-predicates/main.c

@@ -0,0 +1,77 @@
+#include <stdbool.h>
+#include <stdlib.h>
+
+typedef struct buffer_t
+{
+  size_t size;
+  char *arr;
+  char *cursor;
+} buffer_t;
+
+typedef struct double_buffer_t
+{
+  buffer_t *first;
+  buffer_t *second;
+} double_buffer_t;
+
+bool is_sized_array(char *arr, size_t size)
+{
+  return __CPROVER_is_fresh(arr, size);
+}
+
+bool is_ranged_cursor(char *cursor, char *arr, size_t size)
+{
+  void *lb = arr;
+  void *ub = arr + size;
+  __CPROVER_pointer_in_range_dfcc(lb, cursor, ub);
+}
+
+bool is_buffer(buffer_t *b)
+{
+  return __CPROVER_is_fresh(b, sizeof(*b)) && (0 < b->size && b->size <= 10) &&
+         is_sized_array(b->arr, b->size) &&
+         is_ranged_cursor(b->cursor, b->arr, b->size);
+}
+
+bool is_double_buffer(double_buffer_t *b)
+{
+  return __CPROVER_is_fresh(b, sizeof(*b)) && is_buffer(b->first) &&
+         is_buffer(b->second);
+}
+
+typedef struct list_t
+{
+  int value;
+  struct list_t *next;
+} list_t;
+
+// true iff list of len max_len with values in [-10,10]
+bool is_list(size_t max_len, list_t *l)
+{
+  if(max_len == 0)
+    return l == NULL;
+  else
+    return __CPROVER_is_fresh(l, sizeof(*l)) && -10 <= l->value &&
+           l->value <= 10 && is_list(max_len - 1, l->next);
+}
+
+int foo(list_t *l, double_buffer_t *b)
+  // clang-format off
+  __CPROVER_requires(is_list(3, l))
+  __CPROVER_requires(is_double_buffer(b))
+  __CPROVER_ensures(-28 <= __CPROVER_return_value &&
+                    __CPROVER_return_value <= 50)
+// clang-format on
+{
+  // access the assumed data structure
+  return l->value + l->next->value + l->next->next->value + b->first->size +
+         b->second->size;
+}
+
+int main()
+{
+  list_t *l;
+  double_buffer_t *b;
+  int return_value = foo(l, b);
+  assert(-28 <= return_value && return_value <= 50);
+}

regression/contracts-dfcc/user-defined-memory-predicates/test.desc

@@ -0,0 +1,9 @@
+CORE
+main.c
+--dfcc main --enforce-contract foo _ --pointer-check
+^EXIT=0$
+^SIGNAL=0$
+^VERIFICATION SUCCESSFUL$
+--
+--
+Checks if user-defined predicates work;

src/ansi-c/c_typecheck_expr.cpp

@@ -2342,6 +2342,17 @@ exprt c_typecheck_baset::do_special_functions(
     // returning nil leaves the call expression in place
     return nil_exprt();
   }
+  else if(identifier == CPROVER_PREFIX "pointer_in_range_dfcc")
+  {
+    // experimental feature for DFCC contracts encodings -- do not use
+    if(expr.arguments().size() != 3)
+    {
+      error().source_location = f_op.source_location();
+      error() << "pointer_in_range expects three arguments" << eom;
+      throw 0;
+    }
+    return nil_exprt();
+  }
   else if(identifier == CPROVER_PREFIX "same_object")
   {
     if(expr.arguments().size()!=2)

src/ansi-c/cprover_builtin_headers.h

@@ -50,6 +50,8 @@ __CPROVER_bool __CPROVER_is_freeable(const void *mem);
 __CPROVER_bool __CPROVER_was_freed(const void *mem);
 __CPROVER_bool __CPROVER_is_fresh(const void *mem, __CPROVER_size_t size);
 __CPROVER_bool __CPROVER_obeys_contract(void (*)(void), void (*)(void));
+// same as pointer_in_range with experimental support in contracts
+__CPROVER_bool __CPROVER_pointer_in_range_dfcc(void *lb, void *ptr, void *ub);
 void __CPROVER_old(const void *);
 void __CPROVER_loop_entry(const void *);
 

src/ansi-c/library/cprover_contracts.c

@@ -1283,6 +1283,50 @@ __CPROVER_HIDE:;
   }
 }
 
+__CPROVER_size_t __VERIFIER_nondet_size();
+
+__CPROVER_bool __CPROVER_contracts_pointer_in_range_dfcc(
+  void *lb,
+  void **ptr,
+  void *ub,
+  __CPROVER_contracts_write_set_ptr_t write_set)
+{
+__CPROVER_HIDE:;
+  __CPROVER_assert(
+    (write_set != 0) & ((write_set->assume_requires_ctx == 1) |
+                        (write_set->assert_requires_ctx == 1) |
+                        (write_set->assume_ensures_ctx == 1) |
+                        (write_set->assert_ensures_ctx == 1)),
+    "__CPROVER_pointer_in_range_dfcc is used only in requires or ensures "
+    "clauses");
+  __CPROVER_assert(__CPROVER_r_ok(lb, 0), "lb pointer must be valid");
+  __CPROVER_assert(__CPROVER_r_ok(ub, 0), "ub pointer must be valid");
+  __CPROVER_assert(
+    __CPROVER_same_object(lb, ub),
+    "lb and ub pointers must have the same object");
+  __CPROVER_ssize_t lb_offset = __CPROVER_POINTER_OFFSET(lb);
+  __CPROVER_ssize_t ub_offset = __CPROVER_POINTER_OFFSET(ub);
+  __CPROVER_assert(
+    lb_offset <= ub_offset, "lb and ub pointers must be ordered");
+  if(write_set->assume_requires_ctx | write_set->assume_ensures_ctx)
+  {
+    // add nondet offset
+    __CPROVER_size_t offset = __VERIFIER_nondet_size();
+
+    // this cast is safe because we prove that ub and lb are ordered
+    __CPROVER_size_t max_offset = (__CPROVER_size_t)(ub_offset - lb_offset);
+    __CPROVER_assume(offset <= max_offset);
+    *ptr = (char *)lb + offset;
+    return 1;
+  }
+  else /* write_set->assert_requires_ctx | write_set->assert_ensures_ctx */
+  {
+    __CPROVER_ssize_t offset = __CPROVER_POINTER_OFFSET(ptr);
+    return __CPROVER_same_object(lb, ptr) && lb_offset <= offset &&
+           offset <= ub_offset;
+  }
+}
+
 /// \brief Returns the start address of the conditional address range found at
 /// index \p idx in  \p set->contract_assigns.
 void *__CPROVER_contracts_write_set_havoc_get_assignable_target(

src/goto-instrument/Makefile

@@ -22,12 +22,14 @@ SRC = accelerate/accelerate.cpp \
       contracts/dynamic-frames/dfcc_is_fresh.cpp \
       contracts/dynamic-frames/dfcc_is_freeable.cpp \
       contracts/dynamic-frames/dfcc_obeys_contract.cpp \
+      contracts/dynamic-frames/dfcc_lift_memory_predicates.cpp \
       contracts/dynamic-frames/dfcc_instrument.cpp \
       contracts/dynamic-frames/dfcc_spec_functions.cpp \
       contracts/dynamic-frames/dfcc_contract_functions.cpp \
       contracts/dynamic-frames/dfcc_wrapper_program.cpp \
       contracts/dynamic-frames/dfcc_contract_handler.cpp \
       contracts/dynamic-frames/dfcc_swap_and_wrap.cpp \
+      contracts/dynamic-frames/dfcc_pointer_in_range.cpp \
       contracts/dynamic-frames/dfcc.cpp \
       contracts/havoc_assigns_clause_targets.cpp \
       contracts/instrument_spec_assigns.cpp \

src/goto-instrument/contracts/doc/developer/contracts-dev-arch.md

@@ -28,6 +28,7 @@ Each of these translation passes is implemented in a specific class:
  :-------------------------------|:---------------------------------------
  @ref dfcc_instrumentt           | Implements @ref contracts-dev-spec-dfcc for @ref goto_functiont, @ref goto_programt, or subsequences of instructions of @ref goto_programt
  @ref dfcc_is_fresht             | Implements @ref contracts-dev-spec-is-fresh
+ @ref dfcc_pointer_in_ranget     | Implements @ref contracts-dev-spec-pointer-in-range
  @ref dfcc_is_freeablet          | Implements @ref contracts-dev-spec-is-freeable
  @ref dfcc_obeys_contractt       | Implements @ref contracts-dev-spec-obeys-contract
  @ref dfcc_spec_functionst       | Implements @ref contracts-dev-spec-spec-rewriting

src/goto-instrument/contracts/doc/developer/contracts-dev-spec-memory-predicates-rewriting.md

@@ -0,0 +1,175 @@
+# Rewriting Declarative Assign and Frees Specification Functions {#contracts-dev-spec-memory-predicates-rewriting}
+
+Back to top @ref contracts-dev-spec
+
+@tableofcontents
+
+The C extensions for contract specification provide three pointer-related memory
+predicates:
+
+```c
+// Holds iff ptr is pointing to an object distinct to all objects pointed to by
+// other __CPROVER_is_fresh occurrences in the contract's pre and post conditions
+__CPROVER_bool __CPROVER_is_fresh(void *ptr, size_t size);
+
+// Holds iff ptr is a valid pointer pointing between lb and ub pointers.
+// \pre lb and ub must be valid pointers pointing in the same object.
+__CPROVER_bool __CPROVER_in_range(void *ptr, void *lb, void *ub);
+
+// Holds iff the function pointer \p fptr points to a function satisfying
+// \p contract.
+__CPROVER_bool __CPROVER_obeys_contract(void (*fptr)(void), void (*contract)(void));
+```
+
+Users are free to express pre and post conditions using these predicates,
+and we let users define their own functions in terms of these predicates.
+
+
+For instance, one could write a predicate defining linked lists of at most `len`
+elements as follows:
+
+```c
+struct list_t {
+  int value;
+  struct list_t *next;
+};
+
+bool is_list(struct list_t *ptr, size_t len)
+{
+  if(ptr)
+    return len > 0 &&
+    __CPROVER_is_fresh(ptr, sizeof(list_t)) && is_list(ptr->next, len-1));
+  else
+    return 0;
+}
+```
+
+One can also simply describe finite structures:
+
+```c
+struct buffer_t {
+  size_t size;
+  char *arr;
+  char *cursor;
+};
+
+struct two_buffers_t {
+  struct buffer_t *left;
+  struct buffer_t *right;
+};
+
+bool is_buffer(struct buffer_t *ptr)
+{
+  return
+    __CPROVER_is_fresh(ptr, sizeof(struct buffer_t)) &&
+    __CPROVER_is_fresh(ptr->arr, ptr->size) &&
+    __CPROVER_in_range(ptr->cursor, ptr->arr, ptr->arr + size-1);
+}
+
+bool is_two_buffers(struct two_buffers_t *ptr)
+{
+  return
+    __CPROVER_is_fresh(ptr, sizeof(struct two_buffers_t)) &&
+    is_buffer(ptr->left) &&
+    is_buffer(ptr->right);
+}
+```
+
+By rewriting such user-defined predicate we achieve two things:
+1. in assumption contexts, evaluating the predicate allocates the data stucture
+  specified by the predicate;
+1. in assertion contexts, evaluating the predicate checks that a given pointer
+  satisfies the predicate definiton;
+
+To be able to achieve point 1., the user-defined functions need to be
+transformed into functions that take their pointer arguments by reference
+instead of by value, so that enforcing the assumption described by the
+predicate can be done by updating the pointer in place using a side effect.
+
+## Collecting user-defined memory predicates {#contracts-dev-spec-memory-predicate-collect}
+
+We first run a pass that collects all user-defined functions that are defined
+in terms of one of the three memory predicates using this fixpoint algorithm:
+
+```
+functions_to_lift =
+ {`__CPROVER_is_fresh`, `__CPROVER_in_range`, `__CPROVER_obeys_contract`};
+
+updated = true;
+while(updated)
+{
+  updated = false;
+  for(function : goto_model)
+  {
+    if(!contains(function_to_lift, function) &&
+        calls_one_of(function, functions_to_lift) &&)
+    {
+      rewrite(function);
+      functions_to_lift.insert(function);
+      updated = true;
+    }
+  }
+}
+```
+
+## Rewriting user-defined memory predicates {#contracts-dev-spec-memory-predicate-collect}
+
+Rewriting a user-defined memory predicate consists in:
+- Adding an extra write set parameter to the predicate signature;
+- identifying the subset of its pointer-typed arguments which eventually gets
+  passed to a built-in or user-defined memory predicate in a lifted parameter position; Let's call this subset `Lifted`;
+- in the signature of the function, lift the types of all parameters of `Lifted` into pointer-to-pointer types;
+- rewrite all occurences of a lifted parameter `p` into `*p`;
+- add an `address_of` operator to all arguments passed to a lifted parameter of a memory predicate;
+- pass the write set as ghost parameter to function calls;
+- rewrite calls to built-in predicates to call their library implementation;
+
+
+This yields the following result
+```c
+bool is_list(
+  struct list_t **ptr,
+  size_t len,
+  __CPROVER_contracts_write_set_ptr_t __write_set)
+{
+  if(*ptr)
+  {
+    return (len > 0 &&
+      __CPROVER_contracts_is_fresh(ptr, sizeof(list_t), __write_set) &&
+      is_list(&((*ptr)->next), len-1, __write_set));
+  }
+  else
+  }
+    return 0;
+  }
+}
+```
+
+```c
+bool is_buffer(struct buffer_t **ptr,
+__CPROVER_contracts_write_set_ptr_t __write_set)
+{
+  return
+    __CPROVER_contracts_is_fresh(
+      ptr, sizeof(struct buffer_t), __write_set) &&
+    __CPROVER_contracts_is_fresh(
+      &((*ptr)->arr), (*ptr)->size, __write_set) &&
+    __CPROVER_contracts_in_range(
+      &((*ptr)->cursor), (*ptr)->arr, (*ptr)->arr + size-1, __write_set);
+}
+
+bool is_two_buffers(struct two_buffers_t **ptr,
+__CPROVER_contracts_write_set_ptr_t __write_set)
+{
+  return
+    __CPROVER_contracts_is_fresh(
+      ptr, sizeof(struct two_buffers_t), __write_set) &&
+    is_buffer(&((*ptr)->left), __write_set) &&
+    is_buffer(&((*ptr)->right), __write_set);
+}
+```
+
+---
+ Prev | Next
+:-----|:------
+ @ref contracts-dev-spec-spec-rewriting | @ref contracts-dev-spec-dfcc

src/goto-instrument/contracts/doc/developer/contracts-dev-spec-pointer-in-range.md

@@ -0,0 +1,26 @@
+# Rewriting Calls to the __CPROVER_pointer_in_range_dfcc Predicate {#contracts-dev-spec-pointer-in-range}
+
+Back to top @ref contracts-dev-spec
+
+@tableofcontents
+
+In goto programs encoding pre or post conditions (generated from the contract
+clauses) and in all user-defined functions, we simply replace calls to
+`__CPROVER_pointer_in_range_dfcc` with calls to the library implementation
+
+```c
+__CPROVER_contracts_pointer_in_range_dfcc(
+    void *lb,
+    void **ptr,
+    void *ub,
+    __CPROVER_contracts_write_set_ptr_t write_set);
+```
+
+This function implements the `__CPROVER_pointer_in_range_dfcc` behaviour in all
+possible contexts (contract checking vs replacement, requires vs ensures clause
+context, as described by the flags carried by the write set parameter).
+
+---
+ Prev | Next
+:-----|:------
+ @ref contracts-dev | @ref contracts-dev-spec-reminder