]>
Commit | Line | Data |
---|---|---|
2402dc6b DM |
1 | /* Implementation of <stdarg.h> within analyzer. |
2 | Copyright (C) 2022 Free Software Foundation, Inc. | |
3 | Contributed by David Malcolm <dmalcolm@redhat.com>. | |
4 | ||
5 | This file is part of GCC. | |
6 | ||
7 | GCC is free software; you can redistribute it and/or modify it | |
8 | under the terms of the GNU General Public License as published by | |
9 | the Free Software Foundation; either version 3, or (at your option) | |
10 | any later version. | |
11 | ||
12 | GCC is distributed in the hope that it will be useful, but | |
13 | WITHOUT ANY WARRANTY; without even the implied warranty of | |
14 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
15 | General Public License for more details. | |
16 | ||
17 | You should have received a copy of the GNU General Public License | |
18 | along with GCC; see the file COPYING3. If not see | |
19 | <http://www.gnu.org/licenses/>. */ | |
20 | ||
21 | #include "config.h" | |
22 | #include "system.h" | |
23 | #include "coretypes.h" | |
24 | #include "tree.h" | |
25 | #include "function.h" | |
26 | #include "basic-block.h" | |
27 | #include "gimple.h" | |
28 | #include "diagnostic-path.h" | |
29 | #include "json.h" | |
30 | #include "analyzer/analyzer.h" | |
31 | #include "analyzer/analyzer-logging.h" | |
32 | #include "analyzer/sm.h" | |
33 | #include "analyzer/pending-diagnostic.h" | |
34 | #include "tristate.h" | |
35 | #include "selftest.h" | |
36 | #include "analyzer/call-string.h" | |
37 | #include "analyzer/program-point.h" | |
38 | #include "analyzer/store.h" | |
39 | #include "analyzer/region-model.h" | |
40 | #include "analyzer/program-state.h" | |
41 | #include "analyzer/checker-path.h" | |
42 | #include "digraph.h" | |
43 | #include "ordered-hash-map.h" | |
44 | #include "cfg.h" | |
45 | #include "gimple-iterator.h" | |
46 | #include "analyzer/supergraph.h" | |
47 | #include "alloc-pool.h" | |
48 | #include "fibonacci_heap.h" | |
49 | #include "shortest-paths.h" | |
50 | #include "sbitmap.h" | |
51 | #include "analyzer/diagnostic-manager.h" | |
52 | #include "analyzer/exploded-graph.h" | |
f443024b | 53 | #include "diagnostic-metadata.h" |
2402dc6b DM |
54 | |
55 | #if ENABLE_ANALYZER | |
56 | ||
57 | namespace ana { | |
58 | ||
59 | /* Implementation of <stdarg.h> within analyzer. | |
60 | ||
61 | Objectives: | |
62 | - detection of interprocedural type errors involving va_arg | |
63 | - tracking of symbolic values interprocedurally from variadic call | |
64 | through to va_arg unpacking | |
65 | - detection of missing va_end | |
66 | - detection of va_arg outside of a va_start/va_end pair | |
67 | - detection of uses of a va_list after the frame in containing the | |
68 | va_start has returned | |
69 | ||
70 | The analyzer runs *before* the "stdarg" and "lower_vaarg" gimple | |
71 | passes, which have target-dependent effects. | |
72 | ||
73 | This file implements a state machine on svalues for tracking when | |
74 | va_start has been called, so that we can detect missing va_end, | |
75 | and misplaced va_arg, etc. | |
76 | To do this requires an svalue that can have state, so we implement va_start | |
77 | by creating a stack-allocated region, and use a pointer to that region | |
78 | as the svalue that has state. | |
79 | ||
80 | We call this stack-allocated region the "impl_reg". Allocating it on | |
81 | the stack ensures that it is invalidated when the frame containing | |
82 | the va_start returns, leading to | |
83 | -Wanalyzer-use-of-pointer-in-stale-stack-frame on attempts to use such | |
84 | a va_list. | |
85 | ||
86 | To track svalues from variadic calls interprocedurally, we implement | |
87 | variadic arguments via new child regions of the callee's frame_region, | |
88 | var_arg_region, each one representing a storage slot for one of the | |
89 | variadic arguments, accessed by index. | |
90 | ||
91 | We have: | |
92 | ||
93 | stack frame: | |
94 | va_list: &impl_reg | |
95 | 'impl_reg': pointer to next var_arg_region | |
96 | var_arg_region for arg 0 | |
97 | ... | |
98 | var_arg_region for arg N-1 | |
99 | ||
100 | Hence given test_1 in stdarg-1.c, at the call to: | |
101 | ||
102 | __analyzer_called_by_test_1 (int placeholder, ...); | |
103 | ||
104 | here: | |
105 | ||
106 | __analyzer_called_by_test_1 (42, "foo", 1066, '@'); | |
107 | ||
108 | we push this frame for the called function: | |
109 | clusters within frame: ‘__analyzer_called_by_test_1’@2 | |
110 | cluster for: placeholder: (int)42 | |
111 | cluster for: VAR_ARG_REG(frame: ‘__analyzer_called_by_test_1’@2, arg_idx: 0): &"foo" (TOUCHED) | |
112 | cluster for: VAR_ARG_REG(frame: ‘__analyzer_called_by_test_1’@2, arg_idx: 1): (int)1066 (TOUCHED) | |
113 | cluster for: VAR_ARG_REG(frame: ‘__analyzer_called_by_test_1’@2, arg_idx: 2): (int)64 (TOUCHED) | |
114 | where the called function's frame has been populated with both the value | |
115 | of the regular argument "placeholder", and with values for 3 variadic | |
116 | arguments. | |
117 | ||
118 | At the call to | |
119 | va_start (ap, placeholder); | |
120 | we allocate a region ALLOCA_REGION for ap to point to, populate that | |
121 | region with the address of variadic argument 0, and set sm-state of | |
122 | &ALLOCA_REGION to "started": | |
123 | clusters within frame: ‘__analyzer_called_by_test_1’@2 | |
124 | cluster for: placeholder: (int)42 | |
125 | cluster for: VAR_ARG_REG(frame: ‘__analyzer_called_by_test_1’@2, arg_idx: 0): &"foo" (TOUCHED) | |
126 | cluster for: VAR_ARG_REG(frame: ‘__analyzer_called_by_test_1’@2, arg_idx: 1): (int)1066 (TOUCHED) | |
127 | cluster for: VAR_ARG_REG(frame: ‘__analyzer_called_by_test_1’@2, arg_idx: 2): (int)64 (TOUCHED) | |
128 | cluster for: ap: &ALLOCA_REGION | |
129 | cluster for: ALLOCA_REGION: &VAR_ARG_REG(frame: ‘__analyzer_called_by_test_1’@2, arg_idx: 0) (TOUCHED) | |
130 | va_list: | |
131 | 0x4c83700: &ALLOCA_REGION: started | |
132 | ||
133 | At each call to | |
134 | va_arg (ap, TYPE); | |
135 | we can look within *ap, locate the region holding the next variadic | |
136 | argument to be extracted, extract the svalue, and advance the index | |
137 | by effectively updating *ap. | |
138 | ||
139 | At the va_end, we can set &ALLOCA_REGION's state to "ended". | |
140 | ||
141 | The various __builtin_va_* accept ap by pointer, so we have e.g.: | |
142 | ||
143 | __builtin_va_start (&ap, [...]); | |
144 | ||
145 | except for the 2nd param of __builtin_va_copy, where the type | |
146 | is already target-dependent (see the discussion of BT_VALIST_ARG | |
147 | below). */ | |
148 | ||
149 | /* Get a tree for diagnostics. | |
150 | Typically we have "&ap", but it will make more sense to | |
151 | the user as just "ap", so strip off the ADDR_EXPR. */ | |
152 | ||
153 | static tree | |
154 | get_va_list_diag_arg (tree va_list_tree) | |
155 | { | |
156 | if (TREE_CODE (va_list_tree) == ADDR_EXPR) | |
157 | va_list_tree = TREE_OPERAND (va_list_tree, 0); | |
158 | return va_list_tree; | |
159 | } | |
160 | ||
161 | /* Get argument ARG_IDX of type BT_VALIST_ARG (for use by va_copy). | |
162 | ||
163 | builtin-types.def has: | |
164 | DEF_PRIMITIVE_TYPE (BT_VALIST_ARG, va_list_arg_type_node) | |
165 | ||
166 | and c_common_nodes_and_builtins initializes va_list_arg_type_node | |
167 | based on whether TREE_CODE (va_list_type_node) is of ARRAY_TYPE or | |
168 | not, giving either one or zero levels of indirection. */ | |
169 | ||
170 | static const svalue * | |
171 | get_BT_VALIST_ARG (const region_model *model, | |
172 | region_model_context *ctxt, | |
173 | const gcall *call, | |
174 | unsigned arg_idx) | |
175 | { | |
176 | tree arg = gimple_call_arg (call, arg_idx); | |
177 | const svalue *arg_sval = model->get_rvalue (arg, ctxt); | |
178 | if (const svalue *cast = arg_sval->maybe_undo_cast ()) | |
179 | arg_sval = cast; | |
180 | if (TREE_CODE (va_list_type_node) == ARRAY_TYPE) | |
181 | { | |
182 | /* va_list_arg_type_node is a pointer to a va_list; | |
183 | return *ARG_SVAL. */ | |
184 | const region *src_reg = model->deref_rvalue (arg_sval, arg, ctxt); | |
185 | const svalue *src_reg_sval = model->get_store_value (src_reg, ctxt); | |
186 | if (const svalue *cast = src_reg_sval->maybe_undo_cast ()) | |
187 | src_reg_sval = cast; | |
188 | return src_reg_sval; | |
189 | } | |
190 | else | |
191 | { | |
192 | /* va_list_arg_type_node is a va_list; return ARG_SVAL. */ | |
193 | return arg_sval; | |
194 | } | |
195 | } | |
196 | ||
197 | namespace { | |
198 | ||
199 | /* A state machine for tracking the state of a va_list, so that | |
200 | we can enforce that each va_start is paired with a va_end, | |
201 | and va_arg only happens within a va_start/va_end pair. | |
202 | Specifically, this tracks the state of the &ALLOCA_BUFFER | |
203 | that va_start/va_copy allocate. */ | |
204 | ||
205 | class va_list_state_machine : public state_machine | |
206 | { | |
207 | public: | |
208 | va_list_state_machine (logger *logger); | |
209 | ||
ff171cb1 | 210 | bool inherited_state_p () const final override { return false; } |
2402dc6b DM |
211 | |
212 | bool on_stmt (sm_context *sm_ctxt, | |
213 | const supernode *node, | |
ff171cb1 | 214 | const gimple *stmt) const final override; |
2402dc6b | 215 | |
ff171cb1 | 216 | bool can_purge_p (state_t s) const final override |
2402dc6b DM |
217 | { |
218 | return s != m_started; | |
219 | } | |
ff171cb1 | 220 | pending_diagnostic *on_leak (tree var) const final override; |
2402dc6b DM |
221 | |
222 | /* State for a va_list that the result of a va_start or va_copy. */ | |
223 | state_t m_started; | |
224 | ||
225 | /* State for a va_list that has had va_end called on it. */ | |
226 | state_t m_ended; | |
227 | ||
228 | private: | |
229 | void on_va_start (sm_context *sm_ctxt, const supernode *node, | |
230 | const gcall *call) const; | |
231 | void on_va_copy (sm_context *sm_ctxt, const supernode *node, | |
232 | const gcall *call) const; | |
233 | void on_va_arg (sm_context *sm_ctxt, const supernode *node, | |
234 | const gcall *call) const; | |
235 | void on_va_end (sm_context *sm_ctxt, const supernode *node, | |
236 | const gcall *call) const; | |
237 | void check_for_ended_va_list (sm_context *sm_ctxt, | |
238 | const supernode *node, | |
239 | const gcall *call, | |
240 | const svalue *arg, | |
241 | const char *usage_fnname) const; | |
242 | }; | |
243 | ||
244 | /* va_list_state_machine's ctor. */ | |
245 | ||
246 | va_list_state_machine::va_list_state_machine (logger *logger) | |
247 | : state_machine ("va_list", logger) | |
248 | { | |
249 | m_started = add_state ("started"); | |
250 | m_ended = add_state ("ended"); | |
251 | } | |
252 | ||
253 | /* Implementation of the various "va_*" functions for | |
254 | va_list_state_machine. */ | |
255 | ||
256 | bool | |
257 | va_list_state_machine::on_stmt (sm_context *sm_ctxt, | |
258 | const supernode *node, | |
259 | const gimple *stmt) const | |
260 | { | |
261 | if (const gcall *call = dyn_cast <const gcall *> (stmt)) | |
262 | { | |
263 | if (gimple_call_internal_p (call) | |
264 | && gimple_call_internal_fn (call) == IFN_VA_ARG) | |
265 | { | |
266 | on_va_arg (sm_ctxt, node, call); | |
267 | return false; | |
268 | } | |
269 | ||
270 | if (tree callee_fndecl = sm_ctxt->get_fndecl_for_call (call)) | |
271 | if (fndecl_built_in_p (callee_fndecl, BUILT_IN_NORMAL) | |
272 | && gimple_builtin_call_types_compatible_p (call, callee_fndecl)) | |
273 | switch (DECL_UNCHECKED_FUNCTION_CODE (callee_fndecl)) | |
274 | { | |
275 | default: | |
276 | break; | |
277 | ||
278 | case BUILT_IN_VA_START: | |
279 | on_va_start (sm_ctxt, node, call); | |
280 | break; | |
281 | ||
282 | case BUILT_IN_VA_COPY: | |
283 | on_va_copy (sm_ctxt, node, call); | |
284 | break; | |
285 | ||
286 | case BUILT_IN_VA_END: | |
287 | on_va_end (sm_ctxt, node, call); | |
288 | break; | |
289 | } | |
290 | } | |
291 | return false; | |
292 | } | |
293 | ||
294 | /* Get the svalue for which va_list_state_machine holds state on argument ARG_ | |
295 | IDX to CALL. */ | |
296 | ||
297 | static const svalue * | |
298 | get_stateful_arg (sm_context *sm_ctxt, const gcall *call, unsigned arg_idx) | |
299 | { | |
300 | tree ap = gimple_call_arg (call, arg_idx); | |
301 | if (ap | |
302 | && POINTER_TYPE_P (TREE_TYPE (ap))) | |
303 | { | |
304 | if (const program_state *new_state = sm_ctxt->get_new_program_state ()) | |
305 | { | |
306 | const region_model *new_model = new_state->m_region_model; | |
307 | const svalue *ptr_sval = new_model->get_rvalue (ap, NULL); | |
308 | const region *reg = new_model->deref_rvalue (ptr_sval, ap, NULL); | |
309 | const svalue *impl_sval = new_model->get_store_value (reg, NULL); | |
310 | if (const svalue *cast = impl_sval->maybe_undo_cast ()) | |
311 | impl_sval = cast; | |
312 | return impl_sval; | |
313 | } | |
314 | } | |
315 | return NULL; | |
316 | } | |
317 | ||
318 | /* Abstract class for diagnostics relating to va_list_state_machine. */ | |
319 | ||
320 | class va_list_sm_diagnostic : public pending_diagnostic | |
321 | { | |
322 | public: | |
ff171cb1 | 323 | bool subclass_equal_p (const pending_diagnostic &base_other) const override |
2402dc6b DM |
324 | { |
325 | const va_list_sm_diagnostic &other | |
326 | = (const va_list_sm_diagnostic &)base_other; | |
327 | return (m_ap_sval == other.m_ap_sval | |
328 | && same_tree_p (m_ap_tree, other.m_ap_tree)); | |
329 | } | |
330 | ||
331 | label_text describe_state_change (const evdesc::state_change &change) | |
ff171cb1 | 332 | override |
2402dc6b DM |
333 | { |
334 | if (const char *fnname = maybe_get_fnname (change)) | |
335 | return change.formatted_print ("%qs called here", fnname); | |
336 | return label_text (); | |
337 | } | |
338 | ||
6cf276dd DM |
339 | diagnostic_event::meaning |
340 | get_meaning_for_state_change (const evdesc::state_change &change) | |
341 | const final override | |
342 | { | |
343 | if (change.m_new_state == m_sm.m_started) | |
344 | return diagnostic_event::meaning (diagnostic_event::VERB_acquire, | |
345 | diagnostic_event::NOUN_resource); | |
346 | if (change.m_new_state == m_sm.m_ended) | |
347 | return diagnostic_event::meaning (diagnostic_event::VERB_release, | |
348 | diagnostic_event::NOUN_resource); | |
349 | return diagnostic_event::meaning (); | |
350 | } | |
351 | ||
2402dc6b DM |
352 | protected: |
353 | va_list_sm_diagnostic (const va_list_state_machine &sm, | |
354 | const svalue *ap_sval, tree ap_tree) | |
355 | : m_sm (sm), m_ap_sval (ap_sval), m_ap_tree (ap_tree) | |
356 | {} | |
357 | ||
358 | static const char *maybe_get_fnname (const evdesc::state_change &change) | |
359 | { | |
360 | if (change.m_event.m_stmt) | |
361 | if (const gcall *call = as_a <const gcall *> (change.m_event.m_stmt)) | |
362 | if (tree callee_fndecl = gimple_call_fndecl (call)) | |
363 | { | |
364 | if (fndecl_built_in_p (callee_fndecl, BUILT_IN_NORMAL)) | |
365 | switch (DECL_UNCHECKED_FUNCTION_CODE (callee_fndecl)) | |
366 | { | |
367 | case BUILT_IN_VA_START: | |
368 | return "va_start"; | |
369 | case BUILT_IN_VA_COPY: | |
370 | return "va_copy"; | |
371 | case BUILT_IN_VA_END: | |
372 | return "va_end"; | |
373 | } | |
374 | } | |
375 | return NULL; | |
376 | } | |
377 | ||
378 | const va_list_state_machine &m_sm; | |
379 | const svalue *m_ap_sval; | |
380 | tree m_ap_tree; | |
381 | }; | |
382 | ||
383 | /* Concrete class for -Wanalyzer-va-list-use-after-va-end: | |
384 | complain about use of a va_list after va_end has been called on it. */ | |
385 | ||
386 | class va_list_use_after_va_end : public va_list_sm_diagnostic | |
387 | { | |
388 | public: | |
389 | va_list_use_after_va_end (const va_list_state_machine &sm, | |
390 | const svalue *ap_sval, tree ap_tree, | |
391 | const char *usage_fnname) | |
392 | : va_list_sm_diagnostic (sm, ap_sval, ap_tree), | |
393 | m_usage_fnname (usage_fnname) | |
394 | { | |
395 | } | |
396 | ||
ff171cb1 | 397 | int get_controlling_option () const final override |
2402dc6b DM |
398 | { |
399 | return OPT_Wanalyzer_va_list_use_after_va_end; | |
400 | } | |
401 | ||
402 | bool operator== (const va_list_use_after_va_end &other) const | |
403 | { | |
404 | return (va_list_sm_diagnostic::subclass_equal_p (other) | |
405 | && 0 == strcmp (m_usage_fnname, other.m_usage_fnname)); | |
406 | } | |
407 | ||
ff171cb1 | 408 | bool emit (rich_location *rich_loc) final override |
2402dc6b DM |
409 | { |
410 | auto_diagnostic_group d; | |
411 | return warning_at (rich_loc, get_controlling_option (), | |
412 | "%qs after %qs", m_usage_fnname, "va_end"); | |
413 | } | |
414 | ||
ff171cb1 | 415 | const char *get_kind () const final override |
2402dc6b DM |
416 | { |
417 | return "va_list_use_after_va_end"; | |
418 | } | |
419 | ||
420 | label_text describe_state_change (const evdesc::state_change &change) | |
ff171cb1 | 421 | final override |
2402dc6b DM |
422 | { |
423 | if (change.m_new_state == m_sm.m_ended) | |
424 | m_va_end_event = change.m_event_id; | |
425 | return va_list_sm_diagnostic::describe_state_change (change); | |
426 | } | |
427 | ||
ff171cb1 | 428 | label_text describe_final_event (const evdesc::final_event &ev) final override |
2402dc6b DM |
429 | { |
430 | if (ev.m_expr) | |
431 | { | |
432 | if (m_va_end_event.known_p ()) | |
433 | return ev.formatted_print | |
434 | ("%qs on %qE after %qs at %@", | |
435 | m_usage_fnname, ev.m_expr, "va_end", &m_va_end_event); | |
436 | else | |
437 | return ev.formatted_print | |
438 | ("%qs on %qE after %qs", | |
439 | m_usage_fnname, ev.m_expr, "va_end"); | |
440 | } | |
441 | else | |
442 | { | |
443 | if (m_va_end_event.known_p ()) | |
444 | return ev.formatted_print | |
445 | ("%qs after %qs at %@", | |
446 | m_usage_fnname, "va_end", &m_va_end_event); | |
447 | else | |
448 | return ev.formatted_print | |
449 | ("%qs after %qs", | |
450 | m_usage_fnname, "va_end"); | |
451 | } | |
452 | } | |
453 | ||
454 | private: | |
455 | diagnostic_event_id_t m_va_end_event; | |
456 | const char *m_usage_fnname; | |
457 | }; | |
458 | ||
459 | /* Concrete class for -Wanalyzer-va-list-leak: | |
460 | complain about a va_list in the "started" state that doesn't get after | |
461 | va_end called on it. */ | |
462 | ||
463 | class va_list_leak : public va_list_sm_diagnostic | |
464 | { | |
465 | public: | |
466 | va_list_leak (const va_list_state_machine &sm, | |
467 | const svalue *ap_sval, tree ap_tree) | |
468 | : va_list_sm_diagnostic (sm, ap_sval, ap_tree), | |
469 | m_start_event_fnname (NULL) | |
470 | { | |
471 | } | |
472 | ||
ff171cb1 | 473 | int get_controlling_option () const final override |
2402dc6b DM |
474 | { |
475 | return OPT_Wanalyzer_va_list_leak; | |
476 | } | |
477 | ||
478 | bool operator== (const va_list_leak &other) const | |
479 | { | |
480 | return va_list_sm_diagnostic::subclass_equal_p (other); | |
481 | } | |
482 | ||
2ac1459f | 483 | bool emit (rich_location *rich_loc) final override |
2402dc6b DM |
484 | { |
485 | auto_diagnostic_group d; | |
486 | return warning_at (rich_loc, get_controlling_option (), | |
487 | "missing call to %qs", "va_end"); | |
488 | } | |
489 | ||
ff171cb1 | 490 | const char *get_kind () const final override { return "va_list_leak"; } |
2402dc6b DM |
491 | |
492 | label_text describe_state_change (const evdesc::state_change &change) | |
ff171cb1 | 493 | final override |
2402dc6b DM |
494 | { |
495 | if (change.m_new_state == m_sm.m_started) | |
496 | { | |
497 | m_start_event = change.m_event_id; | |
498 | m_start_event_fnname = maybe_get_fnname (change); | |
499 | } | |
500 | return va_list_sm_diagnostic::describe_state_change (change); | |
501 | } | |
502 | ||
ff171cb1 | 503 | label_text describe_final_event (const evdesc::final_event &ev) final override |
2402dc6b DM |
504 | { |
505 | if (ev.m_expr) | |
506 | { | |
507 | if (m_start_event.known_p () && m_start_event_fnname) | |
508 | return ev.formatted_print | |
509 | ("missing call to %qs on %qE to match %qs at %@", | |
510 | "va_end", ev.m_expr, m_start_event_fnname, &m_start_event); | |
511 | else | |
512 | return ev.formatted_print | |
513 | ("missing call to %qs on %qE", | |
514 | "va_end", ev.m_expr); | |
515 | } | |
516 | else | |
517 | { | |
518 | if (m_start_event.known_p () && m_start_event_fnname) | |
519 | return ev.formatted_print | |
520 | ("missing call to %qs to match %qs at %@", | |
521 | "va_end", m_start_event_fnname, &m_start_event); | |
522 | else | |
523 | return ev.formatted_print | |
524 | ("missing call to %qs", | |
525 | "va_end"); | |
526 | } | |
527 | } | |
528 | ||
529 | private: | |
530 | diagnostic_event_id_t m_start_event; | |
531 | const char *m_start_event_fnname; | |
532 | }; | |
533 | ||
534 | /* Update state machine for a "va_start" call. */ | |
535 | ||
536 | void | |
537 | va_list_state_machine::on_va_start (sm_context *sm_ctxt, | |
538 | const supernode *, | |
539 | const gcall *call) const | |
540 | { | |
541 | const svalue *arg = get_stateful_arg (sm_ctxt, call, 0); | |
542 | if (arg) | |
543 | { | |
544 | /* Transition from start state to "started". */ | |
545 | if (sm_ctxt->get_state (call, arg) == m_start) | |
546 | sm_ctxt->set_next_state (call, arg, m_started); | |
547 | } | |
548 | } | |
549 | ||
550 | /* Complain if ARG is in the "ended" state. */ | |
551 | ||
552 | void | |
553 | va_list_state_machine::check_for_ended_va_list (sm_context *sm_ctxt, | |
554 | const supernode *node, | |
555 | const gcall *call, | |
556 | const svalue *arg, | |
557 | const char *usage_fnname) const | |
558 | { | |
559 | if (sm_ctxt->get_state (call, arg) == m_ended) | |
560 | sm_ctxt->warn (node, call, arg, | |
561 | new va_list_use_after_va_end (*this, arg, NULL_TREE, | |
562 | usage_fnname)); | |
563 | } | |
564 | ||
565 | /* Get the svalue with associated va_list_state_machine state for a | |
566 | BT_VALIST_ARG for ARG_IDX of CALL, if SM_CTXT supports this, | |
567 | or NULL otherwise. */ | |
568 | ||
569 | static const svalue * | |
570 | get_stateful_BT_VALIST_ARG (sm_context *sm_ctxt, | |
571 | const gcall *call, | |
572 | unsigned arg_idx) | |
573 | { | |
574 | if (const program_state *new_state = sm_ctxt->get_new_program_state ()) | |
575 | { | |
576 | const region_model *new_model = new_state->m_region_model; | |
577 | const svalue *arg = get_BT_VALIST_ARG (new_model, NULL, call, arg_idx); | |
578 | return arg; | |
579 | } | |
580 | return NULL; | |
581 | } | |
582 | ||
583 | /* Update state machine for a "va_copy" call. */ | |
584 | ||
585 | void | |
586 | va_list_state_machine::on_va_copy (sm_context *sm_ctxt, | |
587 | const supernode *node, | |
588 | const gcall *call) const | |
589 | { | |
590 | const svalue *src_arg = get_stateful_BT_VALIST_ARG (sm_ctxt, call, 1); | |
591 | if (src_arg) | |
592 | check_for_ended_va_list (sm_ctxt, node, call, src_arg, "va_copy"); | |
593 | ||
594 | const svalue *dst_arg = get_stateful_arg (sm_ctxt, call, 0); | |
595 | if (dst_arg) | |
596 | { | |
597 | /* Transition from start state to "started". */ | |
598 | if (sm_ctxt->get_state (call, dst_arg) == m_start) | |
599 | sm_ctxt->set_next_state (call, dst_arg, m_started); | |
600 | } | |
601 | } | |
602 | ||
603 | /* Update state machine for a "va_arg" call. */ | |
604 | ||
605 | void | |
606 | va_list_state_machine::on_va_arg (sm_context *sm_ctxt, | |
607 | const supernode *node, | |
608 | const gcall *call) const | |
609 | { | |
610 | const svalue *arg = get_stateful_arg (sm_ctxt, call, 0); | |
611 | if (arg) | |
612 | check_for_ended_va_list (sm_ctxt, node, call, arg, "va_arg"); | |
613 | } | |
614 | ||
615 | /* Update state machine for a "va_end" call. */ | |
616 | ||
617 | void | |
618 | va_list_state_machine::on_va_end (sm_context *sm_ctxt, | |
619 | const supernode *node, | |
620 | const gcall *call) const | |
621 | { | |
622 | const svalue *arg = get_stateful_arg (sm_ctxt, call, 0); | |
623 | if (arg) | |
624 | { | |
625 | state_t s = sm_ctxt->get_state (call, arg); | |
626 | /* Transition from "started" to "ended". */ | |
627 | if (s == m_started) | |
628 | sm_ctxt->set_next_state (call, arg, m_ended); | |
629 | else if (s == m_ended) | |
630 | check_for_ended_va_list (sm_ctxt, node, call, arg, "va_end"); | |
631 | } | |
632 | } | |
633 | ||
634 | /* Implementation of state_machine::on_leak vfunc for va_list_state_machine | |
635 | (for complaining about leaks of values in state 'started'). */ | |
636 | ||
637 | pending_diagnostic * | |
638 | va_list_state_machine::on_leak (tree var) const | |
639 | { | |
640 | return new va_list_leak (*this, NULL, var); | |
641 | } | |
642 | ||
643 | } // anonymous namespace | |
644 | ||
645 | /* Internal interface to this file. */ | |
646 | ||
647 | state_machine * | |
648 | make_va_list_state_machine (logger *logger) | |
649 | { | |
650 | return new va_list_state_machine (logger); | |
651 | } | |
652 | ||
653 | /* Handle the on_call_pre part of "__builtin_va_start". */ | |
654 | ||
655 | void | |
656 | region_model::impl_call_va_start (const call_details &cd) | |
657 | { | |
658 | const svalue *out_ptr = cd.get_arg_svalue (0); | |
659 | const region *out_reg | |
660 | = deref_rvalue (out_ptr, cd.get_arg_tree (0), cd.get_ctxt ()); | |
661 | ||
662 | /* "*out_ptr = &IMPL_REGION;". */ | |
663 | const region *impl_reg = m_mgr->create_region_for_alloca (m_current_frame); | |
664 | ||
665 | /* We abuse the types here, since va_list_type isn't | |
666 | necessarily anything to do with a pointer. */ | |
667 | const svalue *ptr_to_impl_reg = m_mgr->get_ptr_svalue (NULL_TREE, impl_reg); | |
668 | set_value (out_reg, ptr_to_impl_reg, cd.get_ctxt ()); | |
669 | ||
670 | /* "*(&IMPL_REGION) = VA_LIST_VAL (0);". */ | |
671 | const region *init_var_arg_reg | |
672 | = m_mgr->get_var_arg_region (get_current_frame (), 0); | |
673 | const svalue *ap_sval = m_mgr->get_ptr_svalue (NULL_TREE, init_var_arg_reg); | |
674 | set_value (impl_reg, ap_sval, cd.get_ctxt ()); | |
675 | } | |
676 | ||
677 | /* Handle the on_call_pre part of "__builtin_va_copy". */ | |
678 | ||
679 | void | |
680 | region_model::impl_call_va_copy (const call_details &cd) | |
681 | { | |
682 | const svalue *out_dst_ptr = cd.get_arg_svalue (0); | |
683 | const svalue *in_va_list | |
684 | = get_BT_VALIST_ARG (this, cd.get_ctxt (), cd.get_call_stmt (), 1); | |
685 | in_va_list = check_for_poison (in_va_list, | |
686 | get_va_list_diag_arg (cd.get_arg_tree (1)), | |
687 | cd.get_ctxt ()); | |
688 | ||
689 | const region *out_dst_reg | |
690 | = deref_rvalue (out_dst_ptr, cd.get_arg_tree (0), cd.get_ctxt ()); | |
691 | ||
692 | /* "*out_dst_ptr = &NEW_IMPL_REGION;". */ | |
693 | const region *new_impl_reg | |
694 | = m_mgr->create_region_for_alloca (m_current_frame); | |
695 | const svalue *ptr_to_new_impl_reg | |
696 | = m_mgr->get_ptr_svalue (NULL_TREE, new_impl_reg); | |
697 | set_value (out_dst_reg, ptr_to_new_impl_reg, cd.get_ctxt ()); | |
698 | ||
699 | if (const region *old_impl_reg = in_va_list->maybe_get_region ()) | |
700 | { | |
701 | ||
702 | /* "(NEW_IMPL_REGION) = (OLD_IMPL_REGION);". */ | |
703 | const svalue *existing_sval | |
704 | = get_store_value (old_impl_reg, cd.get_ctxt ()); | |
705 | set_value (new_impl_reg, existing_sval, cd.get_ctxt ()); | |
706 | } | |
707 | } | |
708 | ||
709 | /* Get the number of variadic arguments to CALLEE_FNDECL at CALL_STMT. */ | |
710 | ||
711 | static int | |
712 | get_num_variadic_arguments (tree callee_fndecl, | |
713 | const gcall *call_stmt) | |
714 | { | |
715 | int num_positional = 0; | |
716 | for (tree iter_parm = DECL_ARGUMENTS (callee_fndecl); iter_parm; | |
717 | iter_parm = DECL_CHAIN (iter_parm)) | |
718 | num_positional++; | |
719 | return gimple_call_num_args (call_stmt) - num_positional; | |
720 | } | |
721 | ||
722 | /* An abstract subclass of pending_diagnostic for diagnostics relating | |
723 | to bad va_arg invocations. | |
724 | ||
725 | This shows the number of variadic arguments at the call of interest. | |
726 | Ideally we'd also be able to highlight individual arguments, but | |
727 | that location information isn't generally available from the middle end. */ | |
728 | ||
729 | class va_arg_diagnostic : public pending_diagnostic | |
730 | { | |
731 | public: | |
732 | /* Override of pending_diagnostic::add_call_event, | |
733 | adding a custom call_event subclass. */ | |
734 | void add_call_event (const exploded_edge &eedge, | |
ff171cb1 | 735 | checker_path *emission_path) override |
2402dc6b DM |
736 | { |
737 | /* As per call_event, but show the number of variadic arguments | |
738 | in the call. */ | |
739 | class va_arg_call_event : public call_event | |
740 | { | |
741 | public: | |
742 | va_arg_call_event (const exploded_edge &eedge, | |
743 | location_t loc, tree fndecl, int depth, | |
744 | int num_variadic_arguments) | |
745 | : call_event (eedge, loc, fndecl, depth), | |
746 | m_num_variadic_arguments (num_variadic_arguments) | |
747 | { | |
748 | } | |
749 | ||
ff171cb1 | 750 | label_text get_desc (bool can_colorize) const override |
2402dc6b DM |
751 | { |
752 | return make_label_text_n | |
753 | (can_colorize, m_num_variadic_arguments, | |
754 | "calling %qE from %qE with %i variadic argument", | |
755 | "calling %qE from %qE with %i variadic arguments", | |
756 | get_callee_fndecl (), | |
757 | get_caller_fndecl (), | |
758 | m_num_variadic_arguments); | |
759 | } | |
760 | private: | |
761 | int m_num_variadic_arguments; | |
762 | }; | |
763 | ||
764 | const frame_region *frame_reg = m_var_arg_reg->get_frame_region (); | |
765 | const exploded_node *dst_node = eedge.m_dest; | |
766 | if (dst_node->get_state ().m_region_model->get_current_frame () | |
767 | == frame_reg) | |
768 | { | |
769 | const exploded_node *src_node = eedge.m_src; | |
770 | const program_point &src_point = src_node->get_point (); | |
771 | const int src_stack_depth = src_point.get_stack_depth (); | |
772 | const gimple *last_stmt = src_point.get_supernode ()->get_last_stmt (); | |
773 | const gcall *call_stmt = as_a <const gcall *> (last_stmt); | |
774 | int num_variadic_arguments | |
775 | = get_num_variadic_arguments (dst_node->get_function ()->decl, | |
776 | call_stmt); | |
777 | emission_path->add_event | |
778 | (new va_arg_call_event (eedge, | |
779 | (last_stmt | |
780 | ? last_stmt->location | |
781 | : UNKNOWN_LOCATION), | |
782 | src_point.get_fndecl (), | |
783 | src_stack_depth, | |
784 | num_variadic_arguments)); | |
785 | } | |
786 | else | |
787 | pending_diagnostic::add_call_event (eedge, emission_path); | |
788 | } | |
789 | ||
790 | protected: | |
791 | va_arg_diagnostic (tree va_list_tree, const var_arg_region *var_arg_reg) | |
792 | : m_va_list_tree (va_list_tree), m_var_arg_reg (var_arg_reg) | |
793 | {} | |
794 | ||
ff171cb1 | 795 | bool subclass_equal_p (const pending_diagnostic &base_other) const override |
2402dc6b DM |
796 | { |
797 | const va_arg_diagnostic &other = (const va_arg_diagnostic &)base_other; | |
798 | return (same_tree_p (m_va_list_tree, other.m_va_list_tree) | |
799 | && m_var_arg_reg == other.m_var_arg_reg); | |
800 | } | |
801 | ||
802 | /* Get the number of arguments consumed so far from the va_list | |
803 | (*before* this va_arg call). */ | |
804 | unsigned get_num_consumed () const | |
805 | { | |
806 | return m_var_arg_reg->get_index (); | |
807 | } | |
808 | ||
809 | /* Get a 1-based index of which variadic argument is being consumed. */ | |
810 | unsigned get_variadic_index_for_diagnostic () const | |
811 | { | |
812 | return get_num_consumed () + 1; | |
813 | } | |
814 | ||
815 | /* User-readable expr for the va_list argument to va_arg. */ | |
816 | tree m_va_list_tree; | |
817 | ||
818 | /* The region that the va_arg attempted to access. */ | |
819 | const var_arg_region *m_var_arg_reg; | |
820 | }; | |
821 | ||
822 | /* A subclass of pending_diagnostic for complaining about a type mismatch | |
823 | between the result of: | |
824 | va_arg (AP); | |
825 | and the type of the argument that was passed to the variadic call. */ | |
826 | ||
827 | class va_arg_type_mismatch : public va_arg_diagnostic | |
828 | { | |
829 | public: | |
830 | va_arg_type_mismatch (tree va_list_tree, const var_arg_region *var_arg_reg, | |
831 | tree expected_type, tree actual_type) | |
832 | : va_arg_diagnostic (va_list_tree, var_arg_reg), | |
833 | m_expected_type (expected_type), m_actual_type (actual_type) | |
834 | {} | |
835 | ||
ff171cb1 | 836 | const char *get_kind () const final override |
2402dc6b DM |
837 | { |
838 | return "va_arg_type_mismatch"; | |
839 | } | |
840 | ||
841 | bool subclass_equal_p (const pending_diagnostic &base_other) | |
ff171cb1 | 842 | const final override |
2402dc6b DM |
843 | { |
844 | if (!va_arg_diagnostic::subclass_equal_p (base_other)) | |
845 | return false; | |
846 | const va_arg_type_mismatch &other | |
847 | = (const va_arg_type_mismatch &)base_other; | |
848 | return (same_tree_p (m_expected_type, other.m_expected_type) | |
849 | && same_tree_p (m_actual_type, other.m_actual_type)); | |
850 | } | |
851 | ||
ff171cb1 | 852 | int get_controlling_option () const final override |
2402dc6b DM |
853 | { |
854 | return OPT_Wanalyzer_va_arg_type_mismatch; | |
855 | } | |
856 | ||
ff171cb1 | 857 | bool emit (rich_location *rich_loc) final override |
2402dc6b DM |
858 | { |
859 | auto_diagnostic_group d; | |
f0da5f0a DM |
860 | diagnostic_metadata m; |
861 | /* "CWE-686: Function Call With Incorrect Argument Type". */ | |
862 | m.add_cwe (686); | |
2402dc6b | 863 | bool warned |
f0da5f0a DM |
864 | = warning_meta (rich_loc, m, get_controlling_option (), |
865 | "%<va_arg%> expected %qT but received %qT" | |
866 | " for variadic argument %i of %qE", | |
867 | m_expected_type, m_actual_type, | |
868 | get_variadic_index_for_diagnostic (), m_va_list_tree); | |
2402dc6b DM |
869 | return warned; |
870 | } | |
871 | ||
ff171cb1 | 872 | label_text describe_final_event (const evdesc::final_event &ev) final override |
2402dc6b DM |
873 | { |
874 | return ev.formatted_print ("%<va_arg%> expected %qT but received %qT" | |
875 | " for variadic argument %i of %qE", | |
876 | m_expected_type, m_actual_type, | |
877 | get_variadic_index_for_diagnostic (), | |
878 | m_va_list_tree); | |
879 | } | |
880 | ||
881 | private: | |
882 | tree m_expected_type; | |
883 | tree m_actual_type; | |
884 | }; | |
885 | ||
886 | /* A subclass of pending_diagnostic for complaining about a | |
887 | va_arg (AP); | |
888 | after all of the args in AP have been consumed. */ | |
889 | ||
890 | class va_list_exhausted : public va_arg_diagnostic | |
891 | { | |
892 | public: | |
893 | va_list_exhausted (tree va_list_tree, const var_arg_region *var_arg_reg) | |
894 | : va_arg_diagnostic (va_list_tree, var_arg_reg) | |
895 | {} | |
896 | ||
ff171cb1 | 897 | const char *get_kind () const final override |
2402dc6b DM |
898 | { |
899 | return "va_list_exhausted"; | |
900 | } | |
901 | ||
ff171cb1 | 902 | int get_controlling_option () const final override |
2402dc6b DM |
903 | { |
904 | return OPT_Wanalyzer_va_list_exhausted; | |
905 | } | |
906 | ||
ff171cb1 | 907 | bool emit (rich_location *rich_loc) final override |
2402dc6b DM |
908 | { |
909 | auto_diagnostic_group d; | |
f443024b DM |
910 | diagnostic_metadata m; |
911 | /* CWE-685: Function Call With Incorrect Number of Arguments. */ | |
912 | m.add_cwe (685); | |
913 | bool warned = warning_meta (rich_loc, m, get_controlling_option (), | |
914 | "%qE has no more arguments (%i consumed)", | |
915 | m_va_list_tree, get_num_consumed ()); | |
2402dc6b DM |
916 | return warned; |
917 | } | |
918 | ||
ff171cb1 | 919 | label_text describe_final_event (const evdesc::final_event &ev) final override |
2402dc6b DM |
920 | { |
921 | return ev.formatted_print ("%qE has no more arguments (%i consumed)", | |
922 | m_va_list_tree, get_num_consumed ()); | |
923 | } | |
924 | }; | |
925 | ||
926 | /* Return true if it's OK to copy a value from ARG_TYPE to LHS_TYPE via | |
927 | va_arg (where argument promotion has already happened). */ | |
928 | ||
929 | static bool | |
930 | va_arg_compatible_types_p (tree lhs_type, tree arg_type) | |
931 | { | |
932 | return compat_types_p (arg_type, lhs_type); | |
933 | } | |
934 | ||
935 | /* If AP_SVAL is a pointer to a var_arg_region, return that var_arg_region. | |
936 | Otherwise return NULL. */ | |
937 | ||
938 | static const var_arg_region * | |
939 | maybe_get_var_arg_region (const svalue *ap_sval) | |
940 | { | |
941 | if (const region *reg = ap_sval->maybe_get_region ()) | |
942 | return reg->dyn_cast_var_arg_region (); | |
943 | return NULL; | |
944 | } | |
945 | ||
946 | /* Handle the on_call_pre part of "__builtin_va_arg". */ | |
947 | ||
948 | void | |
949 | region_model::impl_call_va_arg (const call_details &cd) | |
950 | { | |
951 | region_model_context *ctxt = cd.get_ctxt (); | |
952 | ||
953 | const svalue *in_ptr = cd.get_arg_svalue (0); | |
954 | const region *ap_reg = deref_rvalue (in_ptr, cd.get_arg_tree (0), ctxt); | |
955 | ||
956 | const svalue *ap_sval = get_store_value (ap_reg, ctxt); | |
957 | if (const svalue *cast = ap_sval->maybe_undo_cast ()) | |
958 | ap_sval = cast; | |
959 | ||
960 | tree va_list_tree = get_va_list_diag_arg (cd.get_arg_tree (0)); | |
961 | ap_sval = check_for_poison (ap_sval, va_list_tree, ctxt); | |
962 | ||
963 | if (const region *impl_reg = ap_sval->maybe_get_region ()) | |
964 | { | |
965 | const svalue *old_impl_sval = get_store_value (impl_reg, ctxt); | |
966 | if (const var_arg_region *arg_reg | |
967 | = maybe_get_var_arg_region (old_impl_sval)) | |
968 | { | |
969 | bool saw_problem = false; | |
970 | ||
971 | const frame_region *frame_reg = arg_reg->get_frame_region (); | |
972 | unsigned next_arg_idx = arg_reg->get_index (); | |
973 | ||
b852aa7f | 974 | if (frame_reg->get_stack_depth () > 1) |
2402dc6b DM |
975 | { |
976 | /* The interprocedural case: the called frame will have been | |
977 | populated with any variadic aruguments. | |
978 | Attempt to extract arg_reg to cd's return region (which already | |
979 | has a conjured_svalue), or warn if there's a problem | |
980 | (incompatible types, or if we've run out of args). */ | |
981 | if (const svalue *arg_sval | |
982 | = m_store.get_any_binding (m_mgr->get_store_manager (), | |
983 | arg_reg)) | |
984 | { | |
985 | tree lhs_type = cd.get_lhs_type (); | |
986 | tree arg_type = arg_sval->get_type (); | |
987 | if (va_arg_compatible_types_p (lhs_type, arg_type)) | |
988 | cd.maybe_set_lhs (arg_sval); | |
989 | else | |
990 | { | |
991 | if (ctxt) | |
992 | ctxt->warn (new va_arg_type_mismatch (va_list_tree, | |
993 | arg_reg, | |
994 | lhs_type, | |
995 | arg_type)); | |
996 | saw_problem = true; | |
997 | } | |
998 | } | |
999 | else | |
1000 | { | |
1001 | if (ctxt) | |
1002 | ctxt->warn (new va_list_exhausted (va_list_tree, arg_reg)); | |
1003 | saw_problem = true; | |
1004 | } | |
1005 | } | |
1006 | else | |
1007 | { | |
1008 | /* This frame is an entry-point to the analysis, so there won't be | |
1009 | any specific var_arg_regions populated within it. | |
1010 | We already have a conjured_svalue for the result, so leave | |
1011 | it untouched. */ | |
b852aa7f | 1012 | gcc_assert (frame_reg->get_stack_depth () == 1); |
2402dc6b DM |
1013 | } |
1014 | ||
1015 | if (saw_problem) | |
1016 | { | |
1017 | /* Set impl_reg to UNKNOWN to suppress further warnings. */ | |
1018 | const svalue *new_ap_sval | |
1019 | = m_mgr->get_or_create_unknown_svalue (impl_reg->get_type ()); | |
1020 | set_value (impl_reg, new_ap_sval, ctxt); | |
1021 | } | |
1022 | else | |
1023 | { | |
1024 | /* Update impl_reg to advance to the next arg. */ | |
1025 | const region *next_var_arg_region | |
1026 | = m_mgr->get_var_arg_region (frame_reg, next_arg_idx + 1); | |
1027 | const svalue *new_ap_sval | |
1028 | = m_mgr->get_ptr_svalue (NULL_TREE, next_var_arg_region); | |
1029 | set_value (impl_reg, new_ap_sval, ctxt); | |
1030 | } | |
1031 | } | |
1032 | } | |
1033 | } | |
1034 | ||
1035 | /* Handle the on_call_post part of "__builtin_va_end". */ | |
1036 | ||
1037 | void | |
1038 | region_model::impl_call_va_end (const call_details &) | |
1039 | { | |
1040 | /* No-op. */ | |
1041 | } | |
1042 | ||
1043 | } // namespace ana | |
1044 | ||
1045 | #endif /* #if ENABLE_ANALYZER */ |