cprover
Loading...
Searching...
No Matches
symex_function_call.cpp
Go to the documentation of this file.
1/*******************************************************************\
2
3Module: Symbolic Execution of ANSI-C
4
5Author: Daniel Kroening, kroening@kroening.com
6
7\*******************************************************************/
8
11
12#include "goto_symex.h"
13
14#include <util/arith_tools.h>
15#include <util/byte_operators.h>
16#include <util/c_types.h>
18#include <util/fresh_symbol.h>
19#include <util/invariant.h>
20#include <util/prefix.h>
21#include <util/range.h>
22#include <util/std_code.h>
23
24#include "expr_skeleton.h"
25#include "path_storage.h"
26#include "symex_assign.h"
27
28static void locality(
29 const irep_idt &function_identifier,
31 path_storaget &path_storage,
32 const goto_functionst::goto_functiont &goto_function,
33 const namespacet &ns);
34
35bool goto_symext::get_unwind_recursion(const irep_idt &, unsigned, unsigned)
36{
37 return false;
38}
39
41 const irep_idt &function_identifier,
42 const goto_functionst::goto_functiont &goto_function,
43 statet &state,
44 const exprt::operandst &arguments)
45{
46 // iterates over the arguments
47 exprt::operandst::const_iterator it1=arguments.begin();
48
49 // iterates over the types of the parameters
50 for(const auto &identifier : goto_function.parameter_identifiers)
51 {
53 !identifier.empty(), "function parameter must have an identifier");
54 state.call_stack().top().parameter_names.push_back(identifier);
55
56 const symbolt &symbol=ns.lookup(identifier);
57 symbol_exprt lhs=symbol.symbol_expr();
58
59 // this is the type that the n-th argument should have
60 const typet &parameter_type = symbol.type;
61
62 exprt rhs;
63
64 // if you run out of actual arguments there was a mismatch
65 if(it1==arguments.end())
66 {
67 log.warning() << state.source.pc->source_location().as_string()
68 << ": "
69 "call to '"
70 << id2string(function_identifier)
71 << "': "
72 "not enough arguments, inserting non-deterministic value"
73 << log.eom;
74
76 parameter_type, state.source.pc->source_location());
77 }
78 else
79 rhs=*it1;
80
81 if(rhs.is_nil())
82 {
83 // 'nil' argument doesn't get assigned
84 }
85 else
86 {
87 // It should be the same exact type.
88 if(parameter_type != rhs.type())
89 {
90 const typet &rhs_type = rhs.type();
91
92 // But we are willing to do some limited conversion.
93 // This is highly dubious, obviously.
94 // clang-format off
95 if(
96 (parameter_type.id() == ID_signedbv ||
99 parameter_type.id() == ID_bool ||
100 parameter_type.id() == ID_pointer ||
101 parameter_type.id() == ID_union ||
102 parameter_type.id() == ID_union_tag) &&
103 (rhs_type.id() == ID_signedbv ||
104 rhs_type.id() == ID_unsignedbv ||
105 rhs_type.id() == ID_c_bit_field ||
106 rhs_type.id() == ID_c_enum_tag ||
107 rhs_type.id() == ID_bool ||
108 rhs_type.id() == ID_pointer ||
109 rhs_type.id() == ID_union ||
110 rhs_type.id() == ID_union_tag))
111 // clang-format on
112 {
113 rhs = make_byte_extract(
115 }
116 else
117 {
118 std::ostringstream error;
119 error << state.source.pc->source_location().as_string() << ": "
120 << "function call: parameter \"" << identifier
121 << "\" type mismatch:\ngot " << rhs.type().pretty()
122 << "\nexpected " << parameter_type.pretty();
123 throw unsupported_operation_exceptiont(error.str());
124 }
125 }
126
127 assignment_typet assignment_type;
128
129 // We hide if we are in a hidden function.
130 if(state.call_stack().top().hidden_function)
131 assignment_type =
133 else
134 assignment_type =
136
137 lhs = to_symbol_expr(clean_expr(std::move(lhs), state, true));
138 rhs = clean_expr(std::move(rhs), state, false);
139
141 symex_assignt{state, assignment_type, ns, symex_config, target}
142 .assign_rec(lhs, expr_skeletont{}, rhs, lhs_conditions);
143 }
144
145 if(it1!=arguments.end())
146 it1++;
147 }
148
149 if(to_code_type(ns.lookup(function_identifier).type).has_ellipsis())
150 {
151 // These are va_arg arguments; their types may differ from call to call
152 for(; it1 != arguments.end(); it1++)
153 {
155 it1->type(),
156 id2string(function_identifier),
157 "va_arg",
158 state.source.pc->source_location(),
159 ns.lookup(function_identifier).mode,
160 state.symbol_table);
161 va_arg.is_parameter = true;
162
163 state.call_stack().top().parameter_names.push_back(va_arg.name);
164
165 symex_assign(state, va_arg.symbol_expr(), *it1);
166 }
167 }
168 else if(it1!=arguments.end())
169 {
170 // we got too many arguments, but we will just ignore them
171 }
172}
173
175 const get_goto_functiont &get_goto_function,
176 statet &state,
177 const goto_programt::instructiont &instruction)
178{
179 const exprt &function = instruction.call_function();
180
181 // If at some point symex_function_call can support more
182 // expression ids(), like ID_Dereference, please expand the
183 // precondition appropriately.
184 PRECONDITION(function.id() == ID_symbol);
185
188 state,
189 instruction.call_lhs(),
190 to_symbol_expr(instruction.call_function()),
191 instruction.call_arguments());
192}
193
195 const get_goto_functiont &get_goto_function,
196 statet &state,
197 const exprt &lhs,
198 const symbol_exprt &function,
199 const exprt::operandst &arguments)
200{
202
203 if(lhs.is_nil())
204 cleaned_lhs = lhs;
205 else
206 cleaned_lhs = clean_expr(lhs, state, true);
207
208 // no need to clean the function, which is a symbol only
209
211
212 for(auto &argument : arguments)
213 cleaned_arguments.push_back(clean_expr(argument, state, false));
214
215 target.location(state.guard.as_expr(), state.source);
216
219}
220
222 const get_goto_functiont &get_goto_function,
223 statet &state,
224 const exprt &cleaned_lhs,
225 const symbol_exprt &function,
227{
228 const irep_idt &identifier = function.get_identifier();
229
230 const goto_functionst::goto_functiont &goto_function =
231 get_goto_function(identifier);
232
233 path_storage.dirty.populate_dirty_for_function(identifier, goto_function);
234
236 path_storage.safe_pointers.emplace(identifier, local_safe_pointerst{});
238 emplace_safe_pointers_result.first->second(goto_function.body);
239
241 identifier,
242 state.source.thread_nr,
243 state.call_stack().top().loop_iterations[identifier].count);
244
245 // see if it's too much
247 {
249 {
250 // it's ok, ignore
251 }
252 else
253 {
255 vcc(false_exprt(), "recursion unwinding assertion", state);
256
257 // Rule out this path:
259 }
260
261 symex_transition(state);
262 return;
263 }
264
265 // read the arguments -- before the locality renaming
266 const std::vector<renamedt<exprt, L2>> renamed_arguments =
267 make_range(cleaned_arguments).map([&](const exprt &a) {
268 return state.rename(a, ns);
269 });
270
271 // we hide the call if the caller and callee are both hidden
272 const bool hidden =
273 state.call_stack().top().hidden_function && goto_function.is_hidden();
274
275 // record the call
277 state.guard.as_expr(), identifier, renamed_arguments, state.source, hidden);
278
279 if(!goto_function.body_available())
280 {
281 no_body(identifier);
282
283 // record the return
285 state.guard.as_expr(), identifier, state.source, hidden);
286
287 if(cleaned_lhs.is_not_nil())
288 {
289 const auto rhs = side_effect_expr_nondett(
290 cleaned_lhs.type(), state.source.pc->source_location());
291 symex_assign(state, cleaned_lhs, rhs);
292 }
293
295 {
296 // assign non det to function arguments if pointers
297 // are not const
298 for(const auto &arg : cleaned_arguments)
299 {
300 if(
301 arg.type().id() == ID_pointer &&
303 to_pointer_type(arg.type()).base_type().id() != ID_code)
304 {
305 exprt object =
306 dereference_exprt(arg, to_pointer_type(arg.type()).base_type());
307 exprt cleaned_object = clean_expr(object, state, true);
308 const guardt guard(true_exprt(), state.guard_manager);
309 havoc_rec(state, guard, cleaned_object);
310 }
311 }
312 }
313
314 symex_transition(state);
315 return;
316 }
317
318 // produce a new frame
319 PRECONDITION(!state.call_stack().empty());
320 framet &frame = state.call_stack().new_frame(state.source, state.guard);
321
322 // Only enable loop analysis when complexity is enabled.
324 {
325 // Analyzes loops if required.
326 path_storage.add_function_loops(identifier, goto_function.body);
327 frame.loops_info = path_storage.get_loop_analysis(identifier);
328 }
329
330 // preserve locality of local variables
331 locality(identifier, state, path_storage, goto_function, ns);
332
333 // assign actuals to formal parameters
334 parameter_assignments(identifier, goto_function, state, cleaned_arguments);
335
336 frame.call_lhs = cleaned_lhs;
337 frame.end_of_function = --goto_function.body.instructions.end();
338 frame.function_identifier=identifier;
339 frame.hidden_function = goto_function.is_hidden();
340
341 // set up the 'return value symbol' when needed
342 if(frame.call_lhs.is_not_nil())
343 {
345 "goto_symex::return_value::" + id2string(identifier);
346
348 {
349 const symbolt &function_symbol = ns.lookup(identifier);
351 new_symbol; // these are thread-local and have dynamic lifetime
352 new_symbol.base_name = "return_value";
354 new_symbol.type = to_code_type(function_symbol.type).return_type();
355 new_symbol.mode = function_symbol.mode;
356 state.symbol_table.add(new_symbol);
357 }
358
359 frame.return_value_symbol =
361 }
362
363 const framet &p_frame = state.call_stack().previous_frame();
364 for(const auto &pair : p_frame.loop_iterations)
365 {
366 if(pair.second.is_recursion)
367 frame.loop_iterations.insert(pair);
368 }
369
370 // increase unwinding counter
371 frame.loop_iterations[identifier].is_recursion=true;
372 frame.loop_iterations[identifier].count++;
373
374 state.source.function_id = identifier;
375 symex_transition(state, goto_function.body.instructions.begin(), false);
376}
377
379static void pop_frame(
380 goto_symext::statet &state,
381 const path_storaget &path_storage,
382 bool doing_path_exploration)
383{
384 PRECONDITION(!state.call_stack().empty());
385
386 const framet &frame = state.call_stack().top();
387
388 // restore program counter
389 symex_transition(state, frame.calling_location.pc, false);
390 state.source.function_id = frame.calling_location.function_id;
391
392 // restore L1 renaming
393 state.level1.restore_from(frame.old_level1);
394
395 // If the program is multi-threaded then the state guard is used to
396 // accumulate assumptions (in symex_assume_l2) and must be left alone.
397 // If however it is single-threaded then we should restore the guard, as the
398 // guard coming out of the function may be more complex (e.g. if the callee
399 // was { if(x) while(true) { } } then the guard may still be `!x`),
400 // but at this point all control-flow paths have either converged or been
401 // proven unviable, so we can stop specifying the callee's constraints when
402 // we generate an assumption or VCC.
403
404 // If we're doing path exploration then we do tail-duplication, and we
405 // actually *are* in a more-restricted context than we were when the
406 // function began.
407 if(state.threads.size() == 1 && !doing_path_exploration)
408 {
409 state.guard = frame.guard_at_function_start;
410 }
411
412 for(const irep_idt &l1_o_id : frame.local_objects)
413 {
414 const auto l2_entry_opt = state.get_level2().current_names.find(l1_o_id);
415
416 if(
417 l2_entry_opt.has_value() &&
418 (state.threads.size() == 1 ||
419 !path_storage.dirty(l2_entry_opt->get().first.get_object_name())))
420 {
422 }
423 }
424
425 state.call_stack().pop();
426}
427
430{
431 PRECONDITION(!state.call_stack().empty());
432
433 const bool hidden = state.call_stack().top().hidden_function;
434
435 // first record the return
437 state.guard.as_expr(), state.source.function_id, state.source, hidden);
438
439 // before we drop the frame, remember the call LHS
440 // and the return value symbol, if any
441 auto call_lhs = state.call_stack().top().call_lhs;
443
444 // now get rid of the frame
446
447 // after dropping the frame, assign the return value, if any
448 if(state.reachable && call_lhs.is_not_nil())
449 {
451 return_value_symbol.has_value(),
452 "must have return value symbol when assigning call lhs");
453 // the type of the call lhs and the return type might not match
455 return_value_symbol.value(), call_lhs.type());
456 symex_assign(state, call_lhs, casted_return_value);
457 }
458}
459
462static void locality(
463 const irep_idt &function_identifier,
464 goto_symext::statet &state,
465 path_storaget &path_storage,
466 const goto_functionst::goto_functiont &goto_function,
467 const namespacet &ns)
468{
469 unsigned &frame_nr=
470 state.threads[state.source.thread_nr].function_frame[function_identifier];
471 frame_nr++;
472
473 for(const auto &param : goto_function.parameter_identifiers)
474 {
475 (void)state.add_object(
476 ns.lookup(param).symbol_expr(),
477 [&path_storage, &frame_nr](const irep_idt &l0_name) {
478 return path_storage.get_unique_l1_index(l0_name, frame_nr);
479 },
480 ns);
481 }
482}
constant_exprt from_integer(const mp_integer &int_value, const typet &type)
Expression classes for byte-level operators.
bitvector_typet c_index_type()
Definition c_types.cpp:16
ait supplies three of the four components needed: an abstract interpreter (in this case handling func...
Definition ai.h:564
Internally generated symbol table entryThis is a symbol generated as part of translation to or modifi...
Definition symbol.h:147
void pop()
Definition call_stack.h:36
framet & top()
Definition call_stack.h:17
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
const typet & return_type() const
Definition std_types.h:645
bool has_ellipsis() const
Definition std_types.h:611
Operator to dereference a pointer.
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.
Base class for all expressions.
Definition expr.h:54
std::vector< exprt > operandst
Definition expr.h:56
typet & type()
Return the type of the expression.
Definition expr.h:82
The Boolean constant false.
Definition std_expr.h:2865
::goto_functiont goto_functiont
This class represents an instruction in the GOTO intermediate representation.
const exprt & call_lhs() const
Get the lhs of a FUNCTION_CALL (may be nil)
const exprt::operandst & call_arguments() const
Get the arguments of a FUNCTION_CALL.
const exprt & call_function() const
Get the function that is called for FUNCTION_CALL.
guardt guard
Definition goto_state.h:58
bool reachable
Is this code reachable? If not we can take shortcuts such as not entering function calls,...
Definition goto_state.h:62
const symex_level2t & get_level2() const
Definition goto_state.h:45
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.
call_stackt & call_stack()
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...
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 symex_function_call_symbol(const get_goto_functiont &get_goto_function, statet &state, const exprt &lhs, const symbol_exprt &function, const exprt::operandst &arguments)
Symbolic execution of a call to a function call.
virtual void no_body(const irep_idt &identifier)
Log a warning that a function has no body.
Definition goto_symex.h:428
virtual void vcc(const exprt &, const std::string &msg, statet &)
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 ...
path_storaget & path_storage
Symbolic execution paths to be resumed later.
Definition goto_symex.h:788
void havoc_rec(statet &state, const guardt &guard, const exprt &dest)
symex_target_equationt & target
The equation that this execution is building up.
Definition goto_symex.h:251
virtual bool get_unwind_recursion(const irep_idt &identifier, unsigned thread_nr, unsigned unwind)
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:243
virtual void symex_end_of_function(statet &)
Symbolically execute a END_FUNCTION instruction.
exprt clean_expr(exprt expr, statet &state, bool write)
Clean up an expression.
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.
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:82
virtual void symex_function_call_post_clean(const get_goto_functiont &get_goto_function, statet &state, const exprt &cleaned_lhs, const symbol_exprt &function, const exprt::operandst &cleaned_arguments)
Symbolic execution of a function call by inlining.
messaget log
The messaget to write log messages to.
Definition goto_symex.h:263
const symex_configt symex_config
The configuration to use for this symbolic execution.
Definition goto_symex.h:170
void symex_assume_l2(statet &, const exprt &cond)
virtual void symex_function_call(const get_goto_functiont &get_goto_function, statet &state, const goto_programt::instructiont &instruction)
Symbolically execute a FUNCTION_CALL instruction.
exprt as_expr() const
Definition guard_expr.h:46
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
bool get_bool(const irep_idt &name) const
Definition irep.cpp:58
std::string pretty(unsigned indent=0, unsigned max_indent=0) const
Definition irep.cpp:495
const irep_idt & id() const
Definition irep.h:396
bool is_nil() const
Definition irep.h:376
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:91
bool lookup(const irep_idt &name, const symbolt *&symbol) const override
See documentation for namespace_baset::lookup().
Storage for symbolic execution paths to be resumed later.
incremental_dirtyt dirty
Local variables are considered 'dirty' if they've had an address taken and therefore may be referred ...
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.
std::shared_ptr< lexical_loopst > get_loop_analysis(const irep_idt &function_id)
std::unordered_map< irep_idt, local_safe_pointerst > safe_pointers
Map function identifiers to local_safe_pointerst instances.
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.
const typet & base_type() const
The type of the data what we point to.
optionalt< std::reference_wrapper< const mapped_type > > find(const key_type &k) const
Find element.
A side_effect_exprt that returns a non-deterministically chosen value.
Definition std_code.h:1520
Expression to hold a symbol (variable)
Definition std_expr.h:80
const irep_idt & get_identifier() const
Definition std_expr.h:109
bool has_symbol(const irep_idt &name) const
Check whether a symbol exists in the symbol table.
bool add(const symbolt &symbol)
Add a new symbol to the symbol table.
Symbol table entry.
Definition symbol.h:28
class symbol_exprt symbol_expr() const
Produces a symbol_exprt for a symbol.
Definition symbol.cpp:121
typet type
Type of symbol.
Definition symbol.h:31
irep_idt name
The unique identifier.
Definition symbol.h:40
irep_idt irep_idt base_name
Name of module the symbol belongs to.
Definition symbol.h:46
irep_idt mode
Language mode.
Definition symbol.h:49
Functor for symex assignment.
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:2856
static exprt conditional_cast(const exprt &expr, const typet &type)
Definition std_expr.h:1928
The type of an expression, extends irept.
Definition type.h:29
Thrown when we encounter an instruction, parameters to an instruction etc.
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...
const std::string & id2string(const irep_idt &d)
Definition irep.h:47
Storage of symbolic execution paths to resume.
const pointer_typet & to_pointer_type(const typet &type)
Cast a typet to a pointer_typet.
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
symbol_exprt return_value_symbol(const irep_idt &identifier, const namespacet &ns)
produces the symbol that is used to store the return value of the function with the given identifier
#define DATA_INVARIANT(CONDITION, REASON)
This condition should be used to document that assumptions that are made on goto_functions,...
Definition invariant.h:510
#define PRECONDITION(CONDITION)
Definition invariant.h:463
#define INVARIANT(CONDITION, REASON)
This macro uses the wrapper function 'invariant_violated_string'.
Definition invariant.h:423
const symbol_exprt & to_symbol_expr(const exprt &expr)
Cast an exprt to a symbol_exprt.
Definition std_expr.h:189
const code_typet & to_code_type(const typet &type)
Cast a typet to a code_typet.
Definition std_types.h:744
Stack frames – these are used for function calls and for exceptions.
Definition frame.h:22
exprt call_lhs
Definition frame.h:34
std::unordered_map< irep_idt, loop_infot > loop_iterations
Definition frame.h:73
bool hidden_function
Definition frame.h:36
std::vector< irep_idt > parameter_names
Definition frame.h:31
optionalt< symbol_exprt > return_value_symbol
Definition frame.h:35
bool complexity_limits_active
Whether this run of symex is under complexity limits.
bool unwinding_assertions
bool havoc_undefined_functions
bool doing_path_exploration
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
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
byte_extract_exprt make_byte_extract(const exprt &_op, const exprt &_offset, const typet &_type)
Construct a byte_extract_exprt with endianness and byte width matching the current configuration.