cprover
symex_function_call.cpp
Go to the documentation of this file.
1 /*******************************************************************\
2 
3 Module: Symbolic Execution of ANSI-C
4 
5 Author: Daniel Kroening, kroening@kroening.com
6 
7 \*******************************************************************/
8 
11 
12 #include "goto_symex.h"
13 
14 #include <analyses/guard_expr.h>
15 #include <util/arith_tools.h>
16 #include <util/byte_operators.h>
17 #include <util/c_types.h>
18 #include <util/exception_utils.h>
19 #include <util/fresh_symbol.h>
20 #include <util/invariant.h>
21 #include <util/prefix.h>
22 #include <util/range.h>
23 
24 #include "expr_skeleton.h"
25 #include "symex_assign.h"
26 
27 static void locality(
28  const irep_idt &function_identifier,
29  goto_symext::statet &state,
30  path_storaget &path_storage,
31  const goto_functionst::goto_functiont &goto_function,
32  const namespacet &ns);
33 
34 bool goto_symext::get_unwind_recursion(const irep_idt &, unsigned, unsigned)
35 {
36  return false;
37 }
38 
40  const irep_idt &function_identifier,
41  const goto_functionst::goto_functiont &goto_function,
42  statet &state,
43  const exprt::operandst &arguments)
44 {
45  // iterates over the arguments
46  exprt::operandst::const_iterator it1=arguments.begin();
47 
48  // iterates over the types of the parameters
49  for(const auto &identifier : goto_function.parameter_identifiers)
50  {
51  INVARIANT(
52  !identifier.empty(), "function parameter must have an identifier");
53  state.call_stack().top().parameter_names.push_back(identifier);
54 
55  const symbolt &symbol=ns.lookup(identifier);
56  symbol_exprt lhs=symbol.symbol_expr();
57 
58  // this is the type that the n-th argument should have
59  const typet &parameter_type = symbol.type;
60 
61  exprt rhs;
62 
63  // if you run out of actual arguments there was a mismatch
64  if(it1==arguments.end())
65  {
66  log.warning() << state.source.pc->source_location.as_string()
67  << ": "
68  "call to '"
69  << id2string(function_identifier)
70  << "': "
71  "not enough arguments, inserting non-deterministic value"
72  << log.eom;
73 
75  parameter_type, state.source.pc->source_location);
76  }
77  else
78  rhs=*it1;
79 
80  if(rhs.is_nil())
81  {
82  // 'nil' argument doesn't get assigned
83  }
84  else
85  {
86  // It should be the same exact type.
87  if(parameter_type != rhs.type())
88  {
89  const typet &rhs_type = rhs.type();
90 
91  // But we are willing to do some limited conversion.
92  // This is highly dubious, obviously.
93  // clang-format off
94  if(
95  (parameter_type.id() == ID_signedbv ||
96  parameter_type.id() == ID_unsignedbv ||
97  parameter_type.id() == ID_c_enum_tag ||
98  parameter_type.id() == ID_bool ||
99  parameter_type.id() == ID_pointer ||
100  parameter_type.id() == ID_union ||
101  parameter_type.id() == ID_union_tag) &&
102  (rhs_type.id() == ID_signedbv ||
103  rhs_type.id() == ID_unsignedbv ||
104  rhs_type.id() == ID_c_bit_field ||
105  rhs_type.id() == ID_c_enum_tag ||
106  rhs_type.id() == ID_bool ||
107  rhs_type.id() == ID_pointer ||
108  rhs_type.id() == ID_union ||
109  rhs_type.id() == ID_union_tag))
110  {
111  rhs=
113  byte_extract_id(),
114  rhs,
115  from_integer(0, index_type()),
116  parameter_type);
117  }
118  // clang-format on
119  else
120  {
121  std::ostringstream error;
122  error << state.source.pc->source_location.as_string() << ": "
123  << "function call: parameter \"" << identifier
124  << "\" type mismatch:\ngot " << rhs.type().pretty()
125  << "\nexpected " << parameter_type.pretty();
126  throw unsupported_operation_exceptiont(error.str());
127  }
128  }
129 
130  assignment_typet assignment_type;
131 
132  // We hide if we are in a hidden function.
133  if(state.call_stack().top().hidden_function)
134  assignment_type =
136  else
137  assignment_type =
139 
140  lhs = to_symbol_expr(clean_expr(std::move(lhs), state, true));
141  rhs = clean_expr(std::move(rhs), state, false);
142 
143  exprt::operandst lhs_conditions;
144  symex_assignt{state, assignment_type, ns, symex_config, target}
145  .assign_rec(lhs, expr_skeletont{}, rhs, lhs_conditions);
146  }
147 
148  if(it1!=arguments.end())
149  it1++;
150  }
151 
152  if(to_code_type(ns.lookup(function_identifier).type).has_ellipsis())
153  {
154  // These are va_arg arguments; their types may differ from call to call
155  for(; it1 != arguments.end(); it1++)
156  {
157  symbolt &va_arg = get_fresh_aux_symbol(
158  it1->type(),
159  id2string(function_identifier),
160  "va_arg",
161  state.source.pc->source_location,
162  ns.lookup(function_identifier).mode,
163  state.symbol_table);
164  va_arg.is_parameter = true;
165 
166  state.call_stack().top().parameter_names.push_back(va_arg.name);
167 
168  symex_assign(state, va_arg.symbol_expr(), *it1);
169  }
170  }
171  else if(it1!=arguments.end())
172  {
173  // we got too many arguments, but we will just ignore them
174  }
175 }
176 
178  const get_goto_functiont &get_goto_function,
179  statet &state,
180  const code_function_callt &code)
181 {
182  const exprt &function=code.function();
183 
184  // If at some point symex_function_call can support more
185  // expression ids(), like ID_Dereference, please expand the
186  // precondition appropriately.
187  PRECONDITION(function.id() == ID_symbol);
189 }
190 
192  const get_goto_functiont &get_goto_function,
193  statet &state,
194  const code_function_callt &original_code)
195 {
196  code_function_callt code = original_code;
197 
198  if(code.lhs().is_not_nil())
199  code.lhs() = clean_expr(std::move(code.lhs()), state, true);
200 
201  code.function() = clean_expr(std::move(code.function()), state, false);
202 
203  for(auto &argument : code.arguments())
204  argument = clean_expr(std::move(argument), state, false);
205 
206  target.location(state.guard.as_expr(), state.source);
207 
208  PRECONDITION(code.function().id() == ID_symbol);
209 
210  const irep_idt &identifier=
212 
213  if(has_prefix(id2string(identifier), CPROVER_FKT_PREFIX))
214  {
215  symex_fkt(state, code);
216  }
217  else
219 }
220 
222  const get_goto_functiont &get_goto_function,
223  statet &state,
224  const code_function_callt &call)
225 {
226  const irep_idt &identifier=
228 
229  const goto_functionst::goto_functiont &goto_function =
230  get_goto_function(identifier);
231 
232  path_storage.dirty.populate_dirty_for_function(identifier, goto_function);
233 
234  auto emplace_safe_pointers_result =
235  path_storage.safe_pointers.emplace(identifier, local_safe_pointerst{});
236  if(emplace_safe_pointers_result.second)
237  emplace_safe_pointers_result.first->second(goto_function.body);
238 
239  const bool stop_recursing = get_unwind_recursion(
240  identifier,
241  state.source.thread_nr,
242  state.call_stack().top().loop_iterations[identifier].count);
243 
244  // see if it's too much
245  if(stop_recursing)
246  {
248  {
249  // it's ok, ignore
250  }
251  else
252  {
254  vcc(false_exprt(), "recursion unwinding assertion", state);
255 
256  // Rule out this path:
257  symex_assume_l2(state, false_exprt());
258  }
259 
260  symex_transition(state);
261  return;
262  }
263 
264  // read the arguments -- before the locality renaming
265  const exprt::operandst &arguments = call.arguments();
266  const std::vector<renamedt<exprt, L2>> renamed_arguments =
267  make_range(arguments).map(
268  [&](const exprt &a) { return state.rename(a, ns); });
269 
270  // we hide the call if the caller and callee are both hidden
271  const bool hidden =
272  state.call_stack().top().hidden_function && goto_function.is_hidden();
273 
274  // record the call
276  state.guard.as_expr(), identifier, renamed_arguments, state.source, hidden);
277 
278  if(!goto_function.body_available())
279  {
280  no_body(identifier);
281 
282  // record the return
284  state.guard.as_expr(), identifier, state.source, hidden);
285 
286  if(call.lhs().is_not_nil())
287  {
288  const auto rhs =
290  symex_assign(state, call.lhs(), rhs);
291  }
292 
294  {
295  // assign non det to function arguments if pointers
296  // are not const
297  for(const auto &arg : call.arguments())
298  {
299  if(
300  arg.type().id() == ID_pointer &&
301  !arg.type().subtype().get_bool(ID_C_constant) &&
302  arg.type().subtype().id() != ID_code)
303  {
304  exprt object = dereference_exprt(arg, arg.type().subtype());
305  exprt cleaned_object = clean_expr(object, state, true);
306  const guardt guard(true_exprt(), state.guard_manager);
307  havoc_rec(state, guard, cleaned_object);
308  }
309  }
310  }
311 
312  symex_transition(state);
313  return;
314  }
315 
316  // produce a new frame
317  PRECONDITION(!state.call_stack().empty());
318  framet &frame = state.call_stack().new_frame(state.source, state.guard);
319 
320  // Only enable loop analysis when complexity is enabled.
322  {
323  // Analyzes loops if required.
324  path_storage.add_function_loops(identifier, goto_function.body);
325  frame.loops_info = path_storage.get_loop_analysis(identifier);
326  }
327 
328  // preserve locality of local variables
329  locality(identifier, state, path_storage, goto_function, ns);
330 
331  // assign actuals to formal parameters
332  parameter_assignments(identifier, goto_function, state, arguments);
333 
334  frame.end_of_function=--goto_function.body.instructions.end();
335  frame.return_value=call.lhs();
336  frame.function_identifier=identifier;
337  frame.hidden_function = goto_function.is_hidden();
338 
339  const framet &p_frame = state.call_stack().previous_frame();
340  for(const auto &pair : p_frame.loop_iterations)
341  {
342  if(pair.second.is_recursion)
343  frame.loop_iterations.insert(pair);
344  }
345 
346  // increase unwinding counter
347  frame.loop_iterations[identifier].is_recursion=true;
348  frame.loop_iterations[identifier].count++;
349 
350  state.source.function_id = identifier;
351  symex_transition(state, goto_function.body.instructions.begin(), false);
352 }
353 
355 static void pop_frame(
356  goto_symext::statet &state,
357  const path_storaget &path_storage,
358  bool doing_path_exploration)
359 {
360  PRECONDITION(!state.call_stack().empty());
361 
362  const framet &frame = state.call_stack().top();
363 
364  // restore program counter
365  symex_transition(state, frame.calling_location.pc, false);
367 
368  // restore L1 renaming
369  state.level1.restore_from(frame.old_level1);
370 
371  // If the program is multi-threaded then the state guard is used to
372  // accumulate assumptions (in symex_assume_l2) and must be left alone.
373  // If however it is single-threaded then we should restore the guard, as the
374  // guard coming out of the function may be more complex (e.g. if the callee
375  // was { if(x) while(true) { } } then the guard may still be `!x`),
376  // but at this point all control-flow paths have either converged or been
377  // proven unviable, so we can stop specifying the callee's constraints when
378  // we generate an assumption or VCC.
379 
380  // If we're doing path exploration then we do tail-duplication, and we
381  // actually *are* in a more-restricted context than we were when the
382  // function began.
383  if(state.threads.size() == 1 && !doing_path_exploration)
384  {
385  state.guard = frame.guard_at_function_start;
386  }
387 
388  for(const irep_idt &l1_o_id : frame.local_objects)
389  {
390  const auto l2_entry_opt = state.get_level2().current_names.find(l1_o_id);
391 
392  if(
393  l2_entry_opt.has_value() &&
394  (state.threads.size() == 1 ||
395  !path_storage.dirty(l2_entry_opt->get().first.get_object_name())))
396  {
397  state.drop_existing_l1_name(l1_o_id);
398  }
399  }
400 
401  state.call_stack().pop();
402 }
403 
406 {
407  const bool hidden = state.call_stack().top().hidden_function;
408 
409  // first record the return
411  state.guard.as_expr(), state.source.function_id, state.source, hidden);
412 
413  // then get rid of the frame
415 }
416 
419 static void locality(
420  const irep_idt &function_identifier,
421  goto_symext::statet &state,
422  path_storaget &path_storage,
423  const goto_functionst::goto_functiont &goto_function,
424  const namespacet &ns)
425 {
426  unsigned &frame_nr=
427  state.threads[state.source.thread_nr].function_frame[function_identifier];
428  frame_nr++;
429 
430  for(const auto &param : goto_function.parameter_identifiers)
431  {
432  (void)state.add_object(
433  ns.lookup(param).symbol_expr(),
434  [&path_storage, &frame_nr](const irep_idt &l0_name) {
435  return path_storage.get_unique_l1_index(l0_name, frame_nr);
436  },
437  ns);
438  }
439 }
constant_exprt from_integer(const mp_integer &int_value, const typet &type)
Definition: arith_tools.cpp:99
Expression classes for byte-level operators.
bitvector_typet index_type()
Definition: c_types.cpp:16
Expression of type type extracted from some object op starting at position offset (given in number of...
void pop()
Definition: call_stack.h:36
framet & new_frame(symex_targett::sourcet calling_location, const guardt &guard)
Definition: call_stack.h:30
const framet & previous_frame()
Definition: call_stack.h:42
framet & top()
Definition: call_stack.h:17
codet representation of a function call statement.
Definition: std_code.h:1215
exprt & function()
Definition: std_code.h:1250
argumentst & arguments()
Definition: std_code.h:1260
bool has_ellipsis() const
Definition: std_types.h:816
Operator to dereference a pointer.
Definition: pointer_expr.h:256
dstringt has one field, an unsigned integer no which is an index into a static table of strings.
Definition: dstring.h:37
Expression in which some part is missing and can be substituted for another expression.
Definition: expr_skeleton.h:26
Base class for all expressions.
Definition: expr.h:54
std::vector< exprt > operandst
Definition: expr.h:56
const source_locationt & source_location() const
Definition: expr.h:234
typet & type()
Return the type of the expression.
Definition: expr.h:82
The Boolean constant false.
Definition: std_expr.h:2726
::goto_functiont goto_functiont
guardt guard
Definition: goto_state.h:54
const symex_level2t & get_level2() const
Definition: goto_state.h:41
Central data structure: state.
ssa_exprt add_object(const symbol_exprt &expr, std::function< std::size_t(const irep_idt &)> index_generator, const namespacet &ns)
Instantiate the object expr.
NODISCARD renamedt< exprt, level > rename(exprt expr, const namespacet &ns)
Rewrites symbol expressions in exprt, applying a suffix to each symbol reflecting its most recent ver...
call_stackt & call_stack()
symex_level1t level1
guard_managert & guard_manager
symbol_tablet symbol_table
contains symbols that are minted during symbolic execution, such as dynamically created objects etc.
symex_targett::sourcet source
std::vector< threadt > threads
void drop_existing_l1_name(const irep_idt &l1_identifier)
Drops an L1 name from the local L2 map.
virtual void no_body(const irep_idt &identifier)
Log a warning that a function has no body.
Definition: goto_symex.h:432
virtual void vcc(const exprt &, const std::string &msg, statet &)
Definition: symex_main.cpp:185
static get_goto_functiont get_goto_function(abstract_goto_modelt &goto_model)
Return a function to get/load a goto function from the given goto model Create a default delegate to ...
Definition: symex_main.cpp:493
virtual void symex_function_call_symbol(const get_goto_functiont &get_goto_function, statet &state, const code_function_callt &code)
Symbolic execution of a call to a function call.
path_storaget & path_storage
Symbolic execution paths to be resumed later.
Definition: goto_symex.h:797
void havoc_rec(statet &state, const guardt &guard, const exprt &dest)
Definition: symex_other.cpp:19
symex_target_equationt & target
The equation that this execution is building up.
Definition: goto_symex.h:264
virtual bool get_unwind_recursion(const irep_idt &identifier, unsigned thread_nr, unsigned unwind)
exprt clean_expr(exprt expr, statet &state, bool write)
Clean up an expression.
void parameter_assignments(const irep_idt &function_identifier, const goto_functionst::goto_functiont &goto_function, statet &state, const exprt::operandst &arguments)
Iterates over arguments and assigns them to the parameters, which are symbols whose name and type are...
namespacet ns
Initialized just before symbolic execution begins, to point to both outer_symbol_table and the symbol...
Definition: goto_symex.h:256
virtual void symex_end_of_function(statet &)
Symbolically execute a END_FUNCTION instruction.
virtual void symex_fkt(statet &state, const code_function_callt &code)
Symbolically execute a FUNCTION_CALL instruction for a function whose name starts with CPROVER_FKT_PR...
void symex_assign(statet &state, const exprt &lhs, const exprt &rhs)
Symbolically execute an ASSIGN instruction or simulate such an execution for a synthetic assignment.
Definition: goto_symex.cpp:39
virtual void symex_function_call(const get_goto_functiont &get_goto_function, statet &state, const code_function_callt &code)
Symbolically execute a FUNCTION_CALL instruction.
std::function< const goto_functionst::goto_functiont &(const irep_idt &)> get_goto_functiont
The type of delegate functions that retrieve a goto_functiont for a particular function identifier.
Definition: goto_symex.h:95
messaget log
The messaget to write log messages to.
Definition: goto_symex.h:276
const symex_configt symex_config
The configuration to use for this symbolic execution.
Definition: goto_symex.h:183
void symex_assume_l2(statet &, const exprt &cond)
Definition: symex_main.cpp:222
virtual void symex_function_call_code(const get_goto_functiont &get_goto_function, statet &state, const code_function_callt &call)
Symbolic execution of a function call by inlining.
exprt as_expr() const
Definition: guard_expr.h:49
void populate_dirty_for_function(const irep_idt &id, const goto_functionst::goto_functiont &function)
Analyse the given function with dirtyt if it hasn't been seen before.
Definition: dirty.cpp:78
std::string pretty(unsigned indent=0, unsigned max_indent=0) const
Definition: irep.cpp:492
bool is_not_nil() const
Definition: irep.h:391
const irep_idt & id() const
Definition: irep.h:407
bool is_nil() const
Definition: irep.h:387
A very simple, cheap analysis to determine when dereference operations are trivially guarded by a che...
mstreamt & warning() const
Definition: message.h:404
static eomt eom
Definition: message.h:297
A namespacet is essentially one or two symbol tables bound together, to allow for symbol lookups in t...
Definition: namespace.h:92
bool lookup(const irep_idt &name, const symbolt *&symbol) const override
See documentation for namespace_baset::lookup().
Definition: namespace.cpp:140
Storage for symbolic execution paths to be resumed later.
Definition: path_storage.h:38
incremental_dirtyt dirty
Local variables are considered 'dirty' if they've had an address taken and therefore may be referred ...
Definition: path_storage.h:116
std::size_t get_unique_l1_index(const irep_idt &id, std::size_t minimum_index)
Provide a unique L1 index for a given id, starting from minimum_index.
Definition: path_storage.h:104
std::shared_ptr< lexical_loopst > get_loop_analysis(const irep_idt &function_id)
Definition: path_storage.h:131
std::unordered_map< irep_idt, local_safe_pointerst > safe_pointers
Map function identifiers to local_safe_pointerst instances.
Definition: path_storage.h:100
void add_function_loops(const irep_idt &identifier, const goto_programt &body)
Generates a loop analysis for the instructions in goto_programt and keys it against function ID.
Definition: path_storage.h:120
optionalt< std::reference_wrapper< const mapped_type > > find(const key_type &k) const
Find element.
Definition: sharing_map.h:1449
A side_effect_exprt that returns a non-deterministically chosen value.
Definition: std_code.h:1968
Expression to hold a symbol (variable)
Definition: std_expr.h:81
const irep_idt & get_identifier() const
Definition: std_expr.h:110
Symbol table entry.
Definition: symbol.h:28
bool is_parameter
Definition: symbol.h:67
class symbol_exprt symbol_expr() const
Produces a symbol_exprt for a symbol.
Definition: symbol.cpp:122
typet type
Type of symbol.
Definition: symbol.h:31
irep_idt name
The unique identifier.
Definition: symbol.h:40
Functor for symex assignment.
Definition: symex_assign.h:26
virtual void function_return(const exprt &guard, const irep_idt &function_id, const sourcet &source, bool hidden)
Record return from a function.
virtual void location(const exprt &guard, const sourcet &source)
Record a location.
virtual void function_call(const exprt &guard, const irep_idt &function_id, const std::vector< renamedt< exprt, L2 >> &ssa_function_arguments, const sourcet &source, bool hidden)
Record a function call.
The Boolean constant true.
Definition: std_expr.h:2717
The type of an expression, extends irept.
Definition: type.h:28
Thrown when we encounter an instruction, parameters to an instruction etc.
bool has_prefix(const std::string &s, const std::string &prefix)
Definition: converter.cpp:13
#define CPROVER_FKT_PREFIX
Expression skeleton.
symbolt & get_fresh_aux_symbol(const typet &type, const std::string &name_prefix, const std::string &basename_prefix, const source_locationt &source_location, const irep_idt &symbol_mode, const namespacet &ns, symbol_table_baset &symbol_table)
Installs a fresh-named symbol with respect to the given namespace ns with the requested name pattern ...
Fresh auxiliary symbol creation.
Symbolic Execution.
void symex_transition(goto_symext::statet &state)
Transition to the next instruction, which increments the internal program counter and initializes the...
Definition: symex_main.cpp:150
Guard Data Structure.
const std::string & id2string(const irep_idt &d)
Definition: irep.h:49
Ranges: pair of begin and end iterators, which can be initialized from containers,...
ranget< iteratort > make_range(iteratort begin, iteratort end)
Definition: range.h:524
#define PRECONDITION(CONDITION)
Definition: invariant.h:464
const symbol_exprt & to_symbol_expr(const exprt &expr)
Cast an exprt to a symbol_exprt.
Definition: std_expr.h:190
const code_typet & to_code_type(const typet &type)
Cast a typet to a code_typet.
Definition: std_types.h:949
Stack frames – these are used for function calls and for exceptions.
Definition: frame.h:21
guardt guard_at_function_start
Definition: frame.h:31
exprt return_value
Definition: frame.h:33
symex_targett::sourcet calling_location
Definition: frame.h:29
std::unordered_map< irep_idt, loop_infot > loop_iterations
Definition: frame.h:71
bool hidden_function
Definition: frame.h:34
std::vector< irep_idt > parameter_names
Definition: frame.h:30
symex_level1t old_level1
Definition: frame.h:36
goto_programt::const_targett end_of_function
Definition: frame.h:32
std::set< irep_idt > local_objects
Definition: frame.h:38
irep_idt function_identifier
Definition: frame.h:27
std::shared_ptr< lexical_loopst > loops_info
Definition: frame.h:68
bool partial_loops
Definition: symex_config.h:35
bool complexity_limits_active
Whether this run of symex is under complexity limits.
Definition: symex_config.h:56
bool unwinding_assertions
Definition: symex_config.h:33
bool havoc_undefined_functions
Definition: symex_config.h:37
bool doing_path_exploration
Definition: symex_config.h:23
void restore_from(const symex_level1t &other)
Insert the content of other into this renaming.
symex_renaming_levelt current_names
goto_programt::const_targett pc
Definition: symex_target.h:43
Symbolic Execution of assignments.
static void locality(const irep_idt &function_identifier, goto_symext::statet &state, path_storaget &path_storage, const goto_functionst::goto_functiont &goto_function, const namespacet &ns)
Preserves locality of parameters of a given function by applying L1 renaming to them.
static void pop_frame(goto_symext::statet &state, const path_storaget &path_storage, bool doing_path_exploration)
pop one call frame
irep_idt byte_extract_id()