]>
Commit | Line | Data |
---|---|---|
85ec4feb | 1 | /* Copyright (C) 2016-2018 Free Software Foundation, Inc. |
88d0c3f0 MS |
2 | Contributed by Martin Sebor <msebor@redhat.com>. |
3 | ||
4 | This file is part of GCC. | |
5 | ||
6 | GCC is free software; you can redistribute it and/or modify it under | |
7 | the terms of the GNU General Public License as published by the Free | |
8 | Software Foundation; either version 3, or (at your option) any later | |
9 | version. | |
10 | ||
11 | GCC is distributed in the hope that it will be useful, but WITHOUT ANY | |
12 | WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
13 | FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License | |
14 | for more details. | |
15 | ||
16 | You should have received a copy of the GNU General Public License | |
17 | along with GCC; see the file COPYING3. If not see | |
18 | <http://www.gnu.org/licenses/>. */ | |
19 | ||
20 | /* This file implements the printf-return-value pass. The pass does | |
21 | two things: 1) it analyzes calls to formatted output functions like | |
22 | sprintf looking for possible buffer overflows and calls to bounded | |
23 | functions like snprintf for early truncation (and under the control | |
24 | of the -Wformat-length option issues warnings), and 2) under the | |
25 | control of the -fprintf-return-value option it folds the return | |
26 | value of safe calls into constants, making it possible to eliminate | |
27 | code that depends on the value of those constants. | |
28 | ||
29 | For all functions (bounded or not) the pass uses the size of the | |
30 | destination object. That means that it will diagnose calls to | |
31 | snprintf not on the basis of the size specified by the function's | |
32 | second argument but rathger on the basis of the size the first | |
33 | argument points to (if possible). For bound-checking built-ins | |
34 | like __builtin___snprintf_chk the pass uses the size typically | |
35 | determined by __builtin_object_size and passed to the built-in | |
36 | by the Glibc inline wrapper. | |
37 | ||
38 | The pass handles all forms standard sprintf format directives, | |
39 | including character, integer, floating point, pointer, and strings, | |
5fe39ec0 | 40 | with the standard C flags, widths, and precisions. For integers |
88d0c3f0 MS |
41 | and strings it computes the length of output itself. For floating |
42 | point it uses MPFR to fornmat known constants with up and down | |
43 | rounding and uses the resulting range of output lengths. For | |
44 | strings it uses the length of string literals and the sizes of | |
45 | character arrays that a character pointer may point to as a bound | |
46 | on the longest string. */ | |
47 | ||
48 | #include "config.h" | |
49 | #include "system.h" | |
50 | #include "coretypes.h" | |
51 | #include "backend.h" | |
52 | #include "tree.h" | |
53 | #include "gimple.h" | |
54 | #include "tree-pass.h" | |
55 | #include "ssa.h" | |
56 | #include "gimple-fold.h" | |
57 | #include "gimple-pretty-print.h" | |
58 | #include "diagnostic-core.h" | |
59 | #include "fold-const.h" | |
60 | #include "gimple-iterator.h" | |
61 | #include "tree-ssa.h" | |
62 | #include "tree-object-size.h" | |
63 | #include "params.h" | |
64 | #include "tree-cfg.h" | |
d7b0fcaa | 65 | #include "tree-ssa-propagate.h" |
88d0c3f0 MS |
66 | #include "calls.h" |
67 | #include "cfgloop.h" | |
68 | #include "intl.h" | |
d127ae9f | 69 | #include "langhooks.h" |
88d0c3f0 | 70 | |
91e3ec29 | 71 | #include "attribs.h" |
88d0c3f0 MS |
72 | #include "builtins.h" |
73 | #include "stor-layout.h" | |
74 | ||
75 | #include "realmpfr.h" | |
76 | #include "target.h" | |
88d0c3f0 MS |
77 | |
78 | #include "cpplib.h" | |
79 | #include "input.h" | |
80 | #include "toplev.h" | |
81 | #include "substring-locations.h" | |
82 | #include "diagnostic.h" | |
a09ca93a | 83 | #include "domwalk.h" |
aa67d03c JL |
84 | #include "alloc-pool.h" |
85 | #include "vr-values.h" | |
86 | #include "gimple-ssa-evrp-analyze.h" | |
88d0c3f0 | 87 | |
1eb4547b MS |
88 | /* The likely worst case value of MB_LEN_MAX for the target, large enough |
89 | for UTF-8. Ideally, this would be obtained by a target hook if it were | |
90 | to be used for optimization but it's good enough as is for warnings. */ | |
954b452a | 91 | #define target_mb_len_max() 6 |
1eb4547b | 92 | |
cfce1a4a MS |
93 | /* The maximum number of bytes a single non-string directive can result |
94 | in. This is the result of printf("%.*Lf", INT_MAX, -LDBL_MAX) for | |
95 | LDBL_MAX_10_EXP of 4932. */ | |
96 | #define IEEE_MAX_10_EXP 4932 | |
97 | #define target_dir_max() (target_int_max () + IEEE_MAX_10_EXP + 2) | |
98 | ||
88d0c3f0 MS |
99 | namespace { |
100 | ||
101 | const pass_data pass_data_sprintf_length = { | |
102 | GIMPLE_PASS, // pass type | |
103 | "printf-return-value", // pass name | |
104 | OPTGROUP_NONE, // optinfo_flags | |
105 | TV_NONE, // tv_id | |
106 | PROP_cfg, // properties_required | |
107 | 0, // properties_provided | |
108 | 0, // properties_destroyed | |
109 | 0, // properties_start | |
110 | 0, // properties_finish | |
111 | }; | |
112 | ||
02f1ef40 MS |
113 | /* Set to the warning level for the current function which is equal |
114 | either to warn_format_trunc for bounded functions or to | |
115 | warn_format_overflow otherwise. */ | |
116 | ||
117 | static int warn_level; | |
118 | ||
88d0c3f0 MS |
119 | struct format_result; |
120 | ||
a09ca93a JL |
121 | class sprintf_dom_walker : public dom_walker |
122 | { | |
123 | public: | |
124 | sprintf_dom_walker () : dom_walker (CDI_DOMINATORS) {} | |
125 | ~sprintf_dom_walker () {} | |
126 | ||
b95f2911 | 127 | edge before_dom_children (basic_block) FINAL OVERRIDE; |
aa67d03c | 128 | void after_dom_children (basic_block) FINAL OVERRIDE; |
a09ca93a JL |
129 | bool handle_gimple_call (gimple_stmt_iterator *); |
130 | ||
131 | struct call_info; | |
132 | bool compute_format_length (call_info &, format_result *); | |
aa67d03c | 133 | class evrp_range_analyzer evrp_range_analyzer; |
a09ca93a JL |
134 | }; |
135 | ||
88d0c3f0 MS |
136 | class pass_sprintf_length : public gimple_opt_pass |
137 | { | |
138 | bool fold_return_value; | |
139 | ||
140 | public: | |
141 | pass_sprintf_length (gcc::context *ctxt) | |
142 | : gimple_opt_pass (pass_data_sprintf_length, ctxt), | |
143 | fold_return_value (false) | |
144 | { } | |
145 | ||
146 | opt_pass * clone () { return new pass_sprintf_length (m_ctxt); } | |
147 | ||
148 | virtual bool gate (function *); | |
149 | ||
150 | virtual unsigned int execute (function *); | |
151 | ||
152 | void set_pass_param (unsigned int n, bool param) | |
153 | { | |
154 | gcc_assert (n == 0); | |
155 | fold_return_value = param; | |
156 | } | |
157 | ||
88d0c3f0 MS |
158 | }; |
159 | ||
160 | bool | |
161 | pass_sprintf_length::gate (function *) | |
162 | { | |
02f1ef40 MS |
163 | /* Run the pass iff -Warn-format-overflow or -Warn-format-truncation |
164 | is specified and either not optimizing and the pass is being invoked | |
165 | early, or when optimizing and the pass is being invoked during | |
166 | optimization (i.e., "late"). */ | |
954b452a MS |
167 | return ((warn_format_overflow > 0 |
168 | || warn_format_trunc > 0 | |
169 | || flag_printf_return_value) | |
91eaca5e | 170 | && (optimize > 0) == fold_return_value); |
88d0c3f0 MS |
171 | } |
172 | ||
5d93da1d MS |
173 | /* The minimum, maximum, likely, and unlikely maximum number of bytes |
174 | of output either a formatting function or an individual directive | |
175 | can result in. */ | |
176 | ||
177 | struct result_range | |
178 | { | |
179 | /* The absolute minimum number of bytes. The result of a successful | |
180 | conversion is guaranteed to be no less than this. (An erroneous | |
181 | conversion can be indicated by MIN > HOST_WIDE_INT_MAX.) */ | |
182 | unsigned HOST_WIDE_INT min; | |
183 | /* The likely maximum result that is used in diagnostics. In most | |
184 | cases MAX is the same as the worst case UNLIKELY result. */ | |
185 | unsigned HOST_WIDE_INT max; | |
186 | /* The likely result used to trigger diagnostics. For conversions | |
187 | that result in a range of bytes [MIN, MAX], LIKELY is somewhere | |
188 | in that range. */ | |
189 | unsigned HOST_WIDE_INT likely; | |
190 | /* In rare cases (e.g., for nultibyte characters) UNLIKELY gives | |
191 | the worst cases maximum result of a directive. In most cases | |
192 | UNLIKELY == MAX. UNLIKELY is used to control the return value | |
193 | optimization but not in diagnostics. */ | |
194 | unsigned HOST_WIDE_INT unlikely; | |
195 | }; | |
196 | ||
88d0c3f0 MS |
197 | /* The result of a call to a formatted function. */ |
198 | ||
199 | struct format_result | |
200 | { | |
5d93da1d MS |
201 | /* Range of characters written by the formatted function. |
202 | Setting the minimum to HOST_WIDE_INT_MAX disables all | |
203 | length tracking for the remainder of the format string. */ | |
204 | result_range range; | |
88d0c3f0 | 205 | |
1eb4547b | 206 | /* True when the range above is obtained from known values of |
5d93da1d MS |
207 | directive arguments, or bounds on the amount of output such |
208 | as width and precision, and not the result of heuristics that | |
209 | depend on warning levels. It's used to issue stricter diagnostics | |
210 | in cases where strings of unknown lengths are bounded by the arrays | |
211 | they are determined to refer to. KNOWNRANGE must not be used for | |
212 | the return value optimization. */ | |
1eb4547b MS |
213 | bool knownrange; |
214 | ||
bbcbd744 MS |
215 | /* True if no individual directive could fail or result in more than |
216 | 4095 bytes of output (the total NUMBER_CHARS_{MIN,MAX} might be | |
217 | greater). Implementations are not required to handle directives | |
218 | that produce more than 4K bytes (leading to undefined behavior) | |
219 | and so when one is found it disables the return value optimization. | |
220 | Similarly, directives that can fail (such as wide character | |
221 | directives) disable the optimization. */ | |
222 | bool posunder4k; | |
88d0c3f0 MS |
223 | |
224 | /* True when a floating point directive has been seen in the format | |
225 | string. */ | |
226 | bool floating; | |
227 | ||
228 | /* True when an intermediate result has caused a warning. Used to | |
229 | avoid issuing duplicate warnings while finishing the processing | |
5d93da1d | 230 | of a call. WARNED also disables the return value optimization. */ |
88d0c3f0 MS |
231 | bool warned; |
232 | ||
233 | /* Preincrement the number of output characters by 1. */ | |
234 | format_result& operator++ () | |
235 | { | |
236 | return *this += 1; | |
237 | } | |
238 | ||
239 | /* Postincrement the number of output characters by 1. */ | |
240 | format_result operator++ (int) | |
241 | { | |
242 | format_result prev (*this); | |
243 | *this += 1; | |
244 | return prev; | |
245 | } | |
246 | ||
247 | /* Increment the number of output characters by N. */ | |
954b452a | 248 | format_result& operator+= (unsigned HOST_WIDE_INT); |
88d0c3f0 MS |
249 | }; |
250 | ||
954b452a MS |
251 | format_result& |
252 | format_result::operator+= (unsigned HOST_WIDE_INT n) | |
253 | { | |
254 | gcc_assert (n < HOST_WIDE_INT_MAX); | |
255 | ||
5d93da1d MS |
256 | if (range.min < HOST_WIDE_INT_MAX) |
257 | range.min += n; | |
258 | ||
259 | if (range.max < HOST_WIDE_INT_MAX) | |
260 | range.max += n; | |
954b452a | 261 | |
5d93da1d MS |
262 | if (range.likely < HOST_WIDE_INT_MAX) |
263 | range.likely += n; | |
954b452a | 264 | |
5d93da1d MS |
265 | if (range.unlikely < HOST_WIDE_INT_MAX) |
266 | range.unlikely += n; | |
954b452a MS |
267 | |
268 | return *this; | |
269 | } | |
270 | ||
88d0c3f0 MS |
271 | /* Return the value of INT_MIN for the target. */ |
272 | ||
de6aa933 | 273 | static inline HOST_WIDE_INT |
88d0c3f0 MS |
274 | target_int_min () |
275 | { | |
de6aa933 | 276 | return tree_to_shwi (TYPE_MIN_VALUE (integer_type_node)); |
1eb4547b MS |
277 | } |
278 | ||
279 | /* Return the value of INT_MAX for the target. */ | |
280 | ||
281 | static inline unsigned HOST_WIDE_INT | |
282 | target_int_max () | |
283 | { | |
de6aa933 | 284 | return tree_to_uhwi (TYPE_MAX_VALUE (integer_type_node)); |
1eb4547b MS |
285 | } |
286 | ||
287 | /* Return the value of SIZE_MAX for the target. */ | |
288 | ||
289 | static inline unsigned HOST_WIDE_INT | |
290 | target_size_max () | |
291 | { | |
de6aa933 | 292 | return tree_to_uhwi (TYPE_MAX_VALUE (size_type_node)); |
88d0c3f0 MS |
293 | } |
294 | ||
d127ae9f MS |
295 | /* A straightforward mapping from the execution character set to the host |
296 | character set indexed by execution character. */ | |
297 | ||
298 | static char target_to_host_charmap[256]; | |
299 | ||
300 | /* Initialize a mapping from the execution character set to the host | |
301 | character set. */ | |
302 | ||
303 | static bool | |
304 | init_target_to_host_charmap () | |
305 | { | |
306 | /* If the percent sign is non-zero the mapping has already been | |
307 | initialized. */ | |
308 | if (target_to_host_charmap['%']) | |
309 | return true; | |
310 | ||
311 | /* Initialize the target_percent character (done elsewhere). */ | |
312 | if (!init_target_chars ()) | |
313 | return false; | |
314 | ||
315 | /* The subset of the source character set used by printf conversion | |
316 | specifications (strictly speaking, not all letters are used but | |
317 | they are included here for the sake of simplicity). The dollar | |
318 | sign must be included even though it's not in the basic source | |
319 | character set. */ | |
320 | const char srcset[] = " 0123456789!\"#%&'()*+,-./:;<=>?[\\]^_{|}~$" | |
321 | "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz"; | |
322 | ||
323 | /* Set the mapping for all characters to some ordinary value (i,e., | |
324 | not none used in printf conversion specifications) and overwrite | |
325 | those that are used by conversion specifications with their | |
326 | corresponding values. */ | |
327 | memset (target_to_host_charmap + 1, '?', sizeof target_to_host_charmap - 1); | |
328 | ||
329 | /* Are the two sets of characters the same? */ | |
330 | bool all_same_p = true; | |
331 | ||
332 | for (const char *pc = srcset; *pc; ++pc) | |
333 | { | |
334 | /* Slice off the high end bits in case target characters are | |
335 | signed. All values are expected to be non-nul, otherwise | |
336 | there's a problem. */ | |
337 | if (unsigned char tc = lang_hooks.to_target_charset (*pc)) | |
338 | { | |
339 | target_to_host_charmap[tc] = *pc; | |
340 | if (tc != *pc) | |
341 | all_same_p = false; | |
342 | } | |
343 | else | |
344 | return false; | |
345 | ||
346 | } | |
347 | ||
348 | /* Set the first element to a non-zero value if the mapping | |
349 | is 1-to-1, otherwise leave it clear (NUL is assumed to be | |
350 | the same in both character sets). */ | |
351 | target_to_host_charmap[0] = all_same_p; | |
352 | ||
353 | return true; | |
354 | } | |
355 | ||
356 | /* Return the host source character corresponding to the character | |
357 | CH in the execution character set if one exists, or some innocuous | |
358 | (non-special, non-nul) source character otherwise. */ | |
359 | ||
360 | static inline unsigned char | |
361 | target_to_host (unsigned char ch) | |
362 | { | |
363 | return target_to_host_charmap[ch]; | |
364 | } | |
365 | ||
366 | /* Convert an initial substring of the string TARGSTR consisting of | |
367 | characters in the execution character set into a string in the | |
368 | source character set on the host and store up to HOSTSZ characters | |
369 | in the buffer pointed to by HOSTR. Return HOSTR. */ | |
370 | ||
371 | static const char* | |
372 | target_to_host (char *hostr, size_t hostsz, const char *targstr) | |
373 | { | |
374 | /* Make sure the buffer is reasonably big. */ | |
375 | gcc_assert (hostsz > 4); | |
376 | ||
377 | /* The interesting subset of source and execution characters are | |
378 | the same so no conversion is necessary. However, truncate | |
379 | overlong strings just like the translated strings are. */ | |
380 | if (target_to_host_charmap['\0'] == 1) | |
381 | { | |
382 | strncpy (hostr, targstr, hostsz - 4); | |
383 | if (strlen (targstr) >= hostsz) | |
384 | strcpy (hostr + hostsz - 4, "..."); | |
385 | return hostr; | |
386 | } | |
387 | ||
388 | /* Convert the initial substring of TARGSTR to the corresponding | |
389 | characters in the host set, appending "..." if TARGSTR is too | |
390 | long to fit. Using the static buffer assumes the function is | |
391 | not called in between sequence points (which it isn't). */ | |
392 | for (char *ph = hostr; ; ++targstr) | |
393 | { | |
394 | *ph++ = target_to_host (*targstr); | |
395 | if (!*targstr) | |
396 | break; | |
397 | ||
398 | if (size_t (ph - hostr) == hostsz - 4) | |
399 | { | |
400 | *ph = '\0'; | |
401 | strcat (ph, "..."); | |
402 | break; | |
403 | } | |
404 | } | |
405 | ||
406 | return hostr; | |
407 | } | |
408 | ||
409 | /* Convert the sequence of decimal digits in the execution character | |
410 | starting at S to a long, just like strtol does. Return the result | |
411 | and set *END to one past the last converted character. On range | |
412 | error set ERANGE to the digit that caused it. */ | |
413 | ||
414 | static inline long | |
415 | target_strtol10 (const char **ps, const char **erange) | |
416 | { | |
417 | unsigned HOST_WIDE_INT val = 0; | |
418 | for ( ; ; ++*ps) | |
419 | { | |
420 | unsigned char c = target_to_host (**ps); | |
421 | if (ISDIGIT (c)) | |
422 | { | |
423 | c -= '0'; | |
424 | ||
425 | /* Check for overflow. */ | |
426 | if (val > (LONG_MAX - c) / 10LU) | |
427 | { | |
428 | val = LONG_MAX; | |
429 | *erange = *ps; | |
430 | ||
431 | /* Skip the remaining digits. */ | |
432 | do | |
433 | c = target_to_host (*++*ps); | |
434 | while (ISDIGIT (c)); | |
435 | break; | |
436 | } | |
437 | else | |
438 | val = val * 10 + c; | |
439 | } | |
440 | else | |
441 | break; | |
442 | } | |
443 | ||
444 | return val; | |
445 | } | |
446 | ||
88d0c3f0 MS |
447 | /* Given FORMAT, set *PLOC to the source location of the format string |
448 | and return the format string if it is known or null otherwise. */ | |
449 | ||
450 | static const char* | |
451 | get_format_string (tree format, location_t *ploc) | |
452 | { | |
88d0c3f0 MS |
453 | *ploc = EXPR_LOC_OR_LOC (format, input_location); |
454 | ||
6b2ffa6a | 455 | return c_getstr (format); |
88d0c3f0 MS |
456 | } |
457 | ||
346114b5 | 458 | /* For convenience and brevity, shorter named entrypoints of |
c896ecfe DM |
459 | format_string_diagnostic_t::emit_warning_va and |
460 | format_string_diagnostic_t::emit_warning_n_va. | |
346114b5 JJ |
461 | These have to be functions with the attribute so that exgettext |
462 | works properly. */ | |
88d0c3f0 | 463 | |
346114b5 JJ |
464 | static bool |
465 | ATTRIBUTE_GCC_DIAG (5, 6) | |
466 | fmtwarn (const substring_loc &fmt_loc, location_t param_loc, | |
467 | const char *corrected_substring, int opt, const char *gmsgid, ...) | |
468 | { | |
c896ecfe DM |
469 | format_string_diagnostic_t diag (fmt_loc, NULL, param_loc, NULL, |
470 | corrected_substring); | |
346114b5 JJ |
471 | va_list ap; |
472 | va_start (ap, gmsgid); | |
c896ecfe | 473 | bool warned = diag.emit_warning_va (opt, gmsgid, &ap); |
346114b5 | 474 | va_end (ap); |
88d0c3f0 | 475 | |
346114b5 JJ |
476 | return warned; |
477 | } | |
88d0c3f0 MS |
478 | |
479 | static bool | |
346114b5 JJ |
480 | ATTRIBUTE_GCC_DIAG (6, 8) ATTRIBUTE_GCC_DIAG (7, 8) |
481 | fmtwarn_n (const substring_loc &fmt_loc, location_t param_loc, | |
482 | const char *corrected_substring, int opt, unsigned HOST_WIDE_INT n, | |
483 | const char *singular_gmsgid, const char *plural_gmsgid, ...) | |
484 | { | |
c896ecfe DM |
485 | format_string_diagnostic_t diag (fmt_loc, NULL, param_loc, NULL, |
486 | corrected_substring); | |
346114b5 JJ |
487 | va_list ap; |
488 | va_start (ap, plural_gmsgid); | |
c896ecfe DM |
489 | bool warned = diag.emit_warning_n_va (opt, n, singular_gmsgid, plural_gmsgid, |
490 | &ap); | |
346114b5 JJ |
491 | va_end (ap); |
492 | ||
493 | return warned; | |
494 | } | |
88d0c3f0 MS |
495 | |
496 | /* Format length modifiers. */ | |
497 | ||
498 | enum format_lengths | |
499 | { | |
500 | FMT_LEN_none, | |
501 | FMT_LEN_hh, // char argument | |
502 | FMT_LEN_h, // short | |
503 | FMT_LEN_l, // long | |
504 | FMT_LEN_ll, // long long | |
505 | FMT_LEN_L, // long double (and GNU long long) | |
506 | FMT_LEN_z, // size_t | |
507 | FMT_LEN_t, // ptrdiff_t | |
508 | FMT_LEN_j // intmax_t | |
509 | }; | |
510 | ||
511 | ||
88d0c3f0 MS |
512 | /* Description of the result of conversion either of a single directive |
513 | or the whole format string. */ | |
514 | ||
515 | struct fmtresult | |
516 | { | |
954b452a MS |
517 | /* Construct a FMTRESULT object with all counters initialized |
518 | to MIN. KNOWNRANGE is set when MIN is valid. */ | |
519 | fmtresult (unsigned HOST_WIDE_INT min = HOST_WIDE_INT_MAX) | |
a41cc59d | 520 | : argmin (), argmax (), nonstr (), |
954b452a | 521 | knownrange (min < HOST_WIDE_INT_MAX), |
bbcbd744 | 522 | mayfail (), nullp () |
1eb4547b | 523 | { |
954b452a MS |
524 | range.min = min; |
525 | range.max = min; | |
5d93da1d MS |
526 | range.likely = min; |
527 | range.unlikely = min; | |
954b452a MS |
528 | } |
529 | ||
5d93da1d MS |
530 | /* Construct a FMTRESULT object with MIN, MAX, and LIKELY counters. |
531 | KNOWNRANGE is set when both MIN and MAX are valid. */ | |
532 | fmtresult (unsigned HOST_WIDE_INT min, unsigned HOST_WIDE_INT max, | |
533 | unsigned HOST_WIDE_INT likely = HOST_WIDE_INT_MAX) | |
a41cc59d | 534 | : argmin (), argmax (), nonstr (), |
954b452a | 535 | knownrange (min < HOST_WIDE_INT_MAX && max < HOST_WIDE_INT_MAX), |
bbcbd744 | 536 | mayfail (), nullp () |
954b452a MS |
537 | { |
538 | range.min = min; | |
a151e93b | 539 | range.max = max; |
5d93da1d MS |
540 | range.likely = max < likely ? min : likely; |
541 | range.unlikely = max; | |
1eb4547b MS |
542 | } |
543 | ||
31c87a43 MS |
544 | /* Adjust result upward to reflect the RANGE of values the specified |
545 | width or precision is known to be in. */ | |
546 | fmtresult& adjust_for_width_or_precision (const HOST_WIDE_INT[2], | |
5d93da1d MS |
547 | tree = NULL_TREE, |
548 | unsigned = 0, unsigned = 0); | |
549 | ||
550 | /* Return the maximum number of decimal digits a value of TYPE | |
551 | formats as on output. */ | |
31c87a43 | 552 | static unsigned type_max_digits (tree, int); |
5d93da1d | 553 | |
88d0c3f0 MS |
554 | /* The range a directive's argument is in. */ |
555 | tree argmin, argmax; | |
556 | ||
557 | /* The minimum and maximum number of bytes that a directive | |
558 | results in on output for an argument in the range above. */ | |
559 | result_range range; | |
560 | ||
a41cc59d MS |
561 | /* Non-nul when the argument of a string directive is not a nul |
562 | terminated string. */ | |
563 | tree nonstr; | |
564 | ||
1eb4547b MS |
565 | /* True when the range above is obtained from a known value of |
566 | a directive's argument or its bounds and not the result of | |
567 | heuristics that depend on warning levels. */ | |
568 | bool knownrange; | |
569 | ||
bbcbd744 MS |
570 | /* True for a directive that may fail (such as wide character |
571 | directives). */ | |
572 | bool mayfail; | |
573 | ||
2c8c83ac MS |
574 | /* True when the argument is a null pointer. */ |
575 | bool nullp; | |
88d0c3f0 MS |
576 | }; |
577 | ||
31c87a43 MS |
578 | /* Adjust result upward to reflect the range ADJUST of values the |
579 | specified width or precision is known to be in. When non-null, | |
580 | TYPE denotes the type of the directive whose result is being | |
581 | adjusted, BASE gives the base of the directive (octal, decimal, | |
582 | or hex), and ADJ denotes the additional adjustment to the LIKELY | |
583 | counter that may need to be added when ADJUST is a range. */ | |
5d93da1d MS |
584 | |
585 | fmtresult& | |
31c87a43 | 586 | fmtresult::adjust_for_width_or_precision (const HOST_WIDE_INT adjust[2], |
5d93da1d MS |
587 | tree type /* = NULL_TREE */, |
588 | unsigned base /* = 0 */, | |
589 | unsigned adj /* = 0 */) | |
590 | { | |
591 | bool minadjusted = false; | |
592 | ||
5d93da1d | 593 | /* Adjust the minimum and likely counters. */ |
8657c838 | 594 | if (adjust[0] >= 0) |
5d93da1d MS |
595 | { |
596 | if (range.min < (unsigned HOST_WIDE_INT)adjust[0]) | |
597 | { | |
598 | range.min = adjust[0]; | |
599 | minadjusted = true; | |
600 | } | |
601 | ||
602 | /* Adjust the likely counter. */ | |
603 | if (range.likely < range.min) | |
604 | range.likely = range.min; | |
605 | } | |
606 | else if (adjust[0] == target_int_min () | |
607 | && (unsigned HOST_WIDE_INT)adjust[1] == target_int_max ()) | |
608 | knownrange = false; | |
609 | ||
610 | /* Adjust the maximum counter. */ | |
8657c838 | 611 | if (adjust[1] > 0) |
5d93da1d MS |
612 | { |
613 | if (range.max < (unsigned HOST_WIDE_INT)adjust[1]) | |
614 | { | |
615 | range.max = adjust[1]; | |
616 | ||
617 | /* Set KNOWNRANGE if both the minimum and maximum have been | |
618 | adjusted. Otherwise leave it at what it was before. */ | |
619 | knownrange = minadjusted; | |
620 | } | |
621 | } | |
622 | ||
623 | if (warn_level > 1 && type) | |
624 | { | |
625 | /* For large non-constant width or precision whose range spans | |
626 | the maximum number of digits produced by the directive for | |
627 | any argument, set the likely number of bytes to be at most | |
628 | the number digits plus other adjustment determined by the | |
629 | caller (one for sign or two for the hexadecimal "0x" | |
630 | prefix). */ | |
631 | unsigned dirdigs = type_max_digits (type, base); | |
632 | if (adjust[0] < dirdigs && dirdigs < adjust[1] | |
633 | && range.likely < dirdigs) | |
634 | range.likely = dirdigs + adj; | |
635 | } | |
636 | else if (range.likely < (range.min ? range.min : 1)) | |
637 | { | |
638 | /* Conservatively, set LIKELY to at least MIN but no less than | |
639 | 1 unless MAX is zero. */ | |
640 | range.likely = (range.min | |
641 | ? range.min | |
642 | : range.max && (range.max < HOST_WIDE_INT_MAX | |
643 | || warn_level > 1) ? 1 : 0); | |
644 | } | |
645 | ||
646 | /* Finally adjust the unlikely counter to be at least as large as | |
647 | the maximum. */ | |
648 | if (range.unlikely < range.max) | |
649 | range.unlikely = range.max; | |
650 | ||
651 | return *this; | |
652 | } | |
653 | ||
654 | /* Return the maximum number of digits a value of TYPE formats in | |
655 | BASE on output, not counting base prefix . */ | |
656 | ||
657 | unsigned | |
658 | fmtresult::type_max_digits (tree type, int base) | |
659 | { | |
660 | unsigned prec = TYPE_PRECISION (type); | |
7af7cb51 MS |
661 | switch (base) |
662 | { | |
663 | case 8: | |
664 | return (prec + 2) / 3; | |
665 | case 10: | |
666 | /* Decimal approximation: yields 3, 5, 10, and 20 for precision | |
667 | of 8, 16, 32, and 64 bits. */ | |
668 | return prec * 301 / 1000 + 1; | |
669 | case 16: | |
670 | return prec / 4; | |
671 | } | |
5d93da1d | 672 | |
7af7cb51 | 673 | gcc_unreachable (); |
5d93da1d MS |
674 | } |
675 | ||
31c87a43 | 676 | static bool |
bbe2e114 JL |
677 | get_int_range (tree, HOST_WIDE_INT *, HOST_WIDE_INT *, bool, HOST_WIDE_INT, |
678 | class vr_values *vr_values); | |
31c87a43 MS |
679 | |
680 | /* Description of a format directive. A directive is either a plain | |
681 | string or a conversion specification that starts with '%'. */ | |
88d0c3f0 | 682 | |
954b452a | 683 | struct directive |
88d0c3f0 | 684 | { |
954b452a MS |
685 | /* The 1-based directive number (for debugging). */ |
686 | unsigned dirno; | |
687 | ||
688 | /* The first character of the directive and its length. */ | |
689 | const char *beg; | |
690 | size_t len; | |
691 | ||
88d0c3f0 MS |
692 | /* A bitmap of flags, one for each character. */ |
693 | unsigned flags[256 / sizeof (int)]; | |
88d0c3f0 | 694 | |
31c87a43 MS |
695 | /* The range of values of the specified width, or -1 if not specified. */ |
696 | HOST_WIDE_INT width[2]; | |
697 | /* The range of values of the specified precision, or -1 if not | |
698 | specified. */ | |
699 | HOST_WIDE_INT prec[2]; | |
88d0c3f0 MS |
700 | |
701 | /* Length modifier. */ | |
702 | format_lengths modifier; | |
703 | ||
704 | /* Format specifier character. */ | |
705 | char specifier; | |
706 | ||
954b452a MS |
707 | /* The argument of the directive or null when the directive doesn't |
708 | take one or when none is available (such as for vararg functions). */ | |
709 | tree arg; | |
88d0c3f0 | 710 | |
5d93da1d MS |
711 | /* Format conversion function that given a directive and an argument |
712 | returns the formatting result. */ | |
bbe2e114 | 713 | fmtresult (*fmtfunc) (const directive &, tree, vr_values *); |
88d0c3f0 MS |
714 | |
715 | /* Return True when a the format flag CHR has been used. */ | |
716 | bool get_flag (char chr) const | |
717 | { | |
718 | unsigned char c = chr & 0xff; | |
719 | return (flags[c / (CHAR_BIT * sizeof *flags)] | |
720 | & (1U << (c % (CHAR_BIT * sizeof *flags)))); | |
721 | } | |
722 | ||
723 | /* Make a record of the format flag CHR having been used. */ | |
724 | void set_flag (char chr) | |
725 | { | |
726 | unsigned char c = chr & 0xff; | |
727 | flags[c / (CHAR_BIT * sizeof *flags)] | |
728 | |= (1U << (c % (CHAR_BIT * sizeof *flags))); | |
729 | } | |
730 | ||
731 | /* Reset the format flag CHR. */ | |
732 | void clear_flag (char chr) | |
733 | { | |
734 | unsigned char c = chr & 0xff; | |
735 | flags[c / (CHAR_BIT * sizeof *flags)] | |
736 | &= ~(1U << (c % (CHAR_BIT * sizeof *flags))); | |
737 | } | |
954b452a | 738 | |
31c87a43 | 739 | /* Set both bounds of the width range to VAL. */ |
954b452a MS |
740 | void set_width (HOST_WIDE_INT val) |
741 | { | |
31c87a43 | 742 | width[0] = width[1] = val; |
954b452a MS |
743 | } |
744 | ||
31c87a43 MS |
745 | /* Set the width range according to ARG, with both bounds being |
746 | no less than 0. For a constant ARG set both bounds to its value | |
747 | or 0, whichever is greater. For a non-constant ARG in some range | |
748 | set width to its range adjusting each bound to -1 if it's less. | |
749 | For an indeterminate ARG set width to [0, INT_MAX]. */ | |
bbe2e114 | 750 | void set_width (tree arg, vr_values *vr_values) |
954b452a | 751 | { |
bbe2e114 | 752 | get_int_range (arg, width, width + 1, true, 0, vr_values); |
954b452a MS |
753 | } |
754 | ||
31c87a43 | 755 | /* Set both bounds of the precision range to VAL. */ |
954b452a MS |
756 | void set_precision (HOST_WIDE_INT val) |
757 | { | |
31c87a43 | 758 | prec[0] = prec[1] = val; |
954b452a MS |
759 | } |
760 | ||
31c87a43 MS |
761 | /* Set the precision range according to ARG, with both bounds being |
762 | no less than -1. For a constant ARG set both bounds to its value | |
763 | or -1 whichever is greater. For a non-constant ARG in some range | |
764 | set precision to its range adjusting each bound to -1 if it's less. | |
765 | For an indeterminate ARG set precision to [-1, INT_MAX]. */ | |
bbe2e114 | 766 | void set_precision (tree arg, vr_values *vr_values) |
954b452a | 767 | { |
bbe2e114 | 768 | get_int_range (arg, prec, prec + 1, false, -1, vr_values); |
954b452a | 769 | } |
71dedb33 MS |
770 | |
771 | /* Return true if both width and precision are known to be | |
772 | either constant or in some range, false otherwise. */ | |
773 | bool known_width_and_precision () const | |
774 | { | |
775 | return ((width[1] < 0 | |
776 | || (unsigned HOST_WIDE_INT)width[1] <= target_int_max ()) | |
777 | && (prec[1] < 0 | |
778 | || (unsigned HOST_WIDE_INT)prec[1] < target_int_max ())); | |
779 | } | |
88d0c3f0 MS |
780 | }; |
781 | ||
782 | /* Return the logarithm of X in BASE. */ | |
783 | ||
784 | static int | |
785 | ilog (unsigned HOST_WIDE_INT x, int base) | |
786 | { | |
787 | int res = 0; | |
788 | do | |
789 | { | |
790 | ++res; | |
791 | x /= base; | |
792 | } while (x); | |
793 | return res; | |
794 | } | |
795 | ||
796 | /* Return the number of bytes resulting from converting into a string | |
5b8999d6 MS |
797 | the INTEGER_CST tree node X in BASE with a minimum of PREC digits. |
798 | PLUS indicates whether 1 for a plus sign should be added for positive | |
799 | numbers, and PREFIX whether the length of an octal ('O') or hexadecimal | |
800 | ('0x') prefix should be added for nonzero numbers. Return -1 if X cannot | |
801 | be represented. */ | |
802 | ||
803 | static HOST_WIDE_INT | |
804 | tree_digits (tree x, int base, HOST_WIDE_INT prec, bool plus, bool prefix) | |
88d0c3f0 MS |
805 | { |
806 | unsigned HOST_WIDE_INT absval; | |
807 | ||
5b8999d6 | 808 | HOST_WIDE_INT res; |
88d0c3f0 MS |
809 | |
810 | if (TYPE_UNSIGNED (TREE_TYPE (x))) | |
811 | { | |
812 | if (tree_fits_uhwi_p (x)) | |
813 | { | |
814 | absval = tree_to_uhwi (x); | |
815 | res = plus; | |
816 | } | |
817 | else | |
818 | return -1; | |
819 | } | |
820 | else | |
821 | { | |
822 | if (tree_fits_shwi_p (x)) | |
823 | { | |
824 | HOST_WIDE_INT i = tree_to_shwi (x); | |
209042e6 MS |
825 | if (HOST_WIDE_INT_MIN == i) |
826 | { | |
827 | /* Avoid undefined behavior due to negating a minimum. */ | |
828 | absval = HOST_WIDE_INT_MAX; | |
829 | res = 1; | |
830 | } | |
831 | else if (i < 0) | |
832 | { | |
833 | absval = -i; | |
834 | res = 1; | |
835 | } | |
836 | else | |
837 | { | |
838 | absval = i; | |
839 | res = plus; | |
840 | } | |
88d0c3f0 MS |
841 | } |
842 | else | |
843 | return -1; | |
844 | } | |
845 | ||
5b8999d6 MS |
846 | int ndigs = ilog (absval, base); |
847 | ||
848 | res += prec < ndigs ? ndigs : prec; | |
88d0c3f0 | 849 | |
f589a1ce MS |
850 | /* Adjust a non-zero value for the base prefix, either hexadecimal, |
851 | or, unless precision has resulted in a leading zero, also octal. */ | |
852 | if (prefix && absval && (base == 16 || prec <= ndigs)) | |
88d0c3f0 MS |
853 | { |
854 | if (base == 8) | |
855 | res += 1; | |
856 | else if (base == 16) | |
857 | res += 2; | |
858 | } | |
859 | ||
860 | return res; | |
861 | } | |
862 | ||
863 | /* Given the formatting result described by RES and NAVAIL, the number | |
31c87a43 | 864 | of available in the destination, return the range of bytes remaining |
88d0c3f0 MS |
865 | in the destination. */ |
866 | ||
867 | static inline result_range | |
868 | bytes_remaining (unsigned HOST_WIDE_INT navail, const format_result &res) | |
869 | { | |
870 | result_range range; | |
871 | ||
872 | if (HOST_WIDE_INT_MAX <= navail) | |
873 | { | |
5d93da1d | 874 | range.min = range.max = range.likely = range.unlikely = navail; |
88d0c3f0 MS |
875 | return range; |
876 | } | |
877 | ||
5d93da1d MS |
878 | /* The lower bound of the available range is the available size |
879 | minus the maximum output size, and the upper bound is the size | |
880 | minus the minimum. */ | |
881 | range.max = res.range.min < navail ? navail - res.range.min : 0; | |
88d0c3f0 | 882 | |
5d93da1d | 883 | range.likely = res.range.likely < navail ? navail - res.range.likely : 0; |
88d0c3f0 | 884 | |
5d93da1d MS |
885 | if (res.range.max < HOST_WIDE_INT_MAX) |
886 | range.min = res.range.max < navail ? navail - res.range.max : 0; | |
88d0c3f0 | 887 | else |
5d93da1d | 888 | range.min = range.likely; |
88d0c3f0 | 889 | |
5d93da1d MS |
890 | range.unlikely = (res.range.unlikely < navail |
891 | ? navail - res.range.unlikely : 0); | |
88d0c3f0 | 892 | |
5d93da1d | 893 | return range; |
88d0c3f0 MS |
894 | } |
895 | ||
896 | /* Description of a call to a formatted function. */ | |
897 | ||
a09ca93a | 898 | struct sprintf_dom_walker::call_info |
88d0c3f0 MS |
899 | { |
900 | /* Function call statement. */ | |
901 | gimple *callstmt; | |
902 | ||
903 | /* Function called. */ | |
904 | tree func; | |
905 | ||
906 | /* Called built-in function code. */ | |
907 | built_in_function fncode; | |
908 | ||
909 | /* Format argument and format string extracted from it. */ | |
910 | tree format; | |
911 | const char *fmtstr; | |
912 | ||
913 | /* The location of the format argument. */ | |
914 | location_t fmtloc; | |
915 | ||
916 | /* The destination object size for __builtin___xxx_chk functions | |
917 | typically determined by __builtin_object_size, or -1 if unknown. */ | |
918 | unsigned HOST_WIDE_INT objsize; | |
919 | ||
920 | /* Number of the first variable argument. */ | |
921 | unsigned HOST_WIDE_INT argidx; | |
922 | ||
923 | /* True for functions like snprintf that specify the size of | |
924 | the destination, false for others like sprintf that don't. */ | |
925 | bool bounded; | |
d7b0fcaa MS |
926 | |
927 | /* True for bounded functions like snprintf that specify a zero-size | |
928 | buffer as a request to compute the size of output without actually | |
51ad5d5b JJ |
929 | writing any. NOWRITE is cleared in response to the %n directive |
930 | which has side-effects similar to writing output. */ | |
d7b0fcaa | 931 | bool nowrite; |
efcc8d38 MS |
932 | |
933 | /* Return true if the called function's return value is used. */ | |
934 | bool retval_used () const | |
935 | { | |
936 | return gimple_get_lhs (callstmt); | |
937 | } | |
938 | ||
939 | /* Return the warning option corresponding to the called function. */ | |
940 | int warnopt () const | |
941 | { | |
bce6f760 | 942 | return bounded ? OPT_Wformat_truncation_ : OPT_Wformat_overflow_; |
efcc8d38 | 943 | } |
88d0c3f0 MS |
944 | }; |
945 | ||
954b452a MS |
946 | /* Return the result of formatting a no-op directive (such as '%n'). */ |
947 | ||
948 | static fmtresult | |
bbe2e114 | 949 | format_none (const directive &, tree, vr_values *) |
954b452a MS |
950 | { |
951 | fmtresult res (0); | |
954b452a MS |
952 | return res; |
953 | } | |
954 | ||
88d0c3f0 MS |
955 | /* Return the result of formatting the '%%' directive. */ |
956 | ||
957 | static fmtresult | |
bbe2e114 | 958 | format_percent (const directive &, tree, vr_values *) |
88d0c3f0 | 959 | { |
954b452a | 960 | fmtresult res (1); |
88d0c3f0 MS |
961 | return res; |
962 | } | |
963 | ||
964 | ||
42aa1173 JJ |
965 | /* Compute intmax_type_node and uintmax_type_node similarly to how |
966 | tree.c builds size_type_node. */ | |
88d0c3f0 MS |
967 | |
968 | static void | |
969 | build_intmax_type_nodes (tree *pintmax, tree *puintmax) | |
970 | { | |
42aa1173 | 971 | if (strcmp (UINTMAX_TYPE, "unsigned int") == 0) |
88d0c3f0 MS |
972 | { |
973 | *pintmax = integer_type_node; | |
974 | *puintmax = unsigned_type_node; | |
975 | } | |
42aa1173 | 976 | else if (strcmp (UINTMAX_TYPE, "long unsigned int") == 0) |
88d0c3f0 MS |
977 | { |
978 | *pintmax = long_integer_type_node; | |
979 | *puintmax = long_unsigned_type_node; | |
980 | } | |
42aa1173 | 981 | else if (strcmp (UINTMAX_TYPE, "long long unsigned int") == 0) |
88d0c3f0 MS |
982 | { |
983 | *pintmax = long_long_integer_type_node; | |
984 | *puintmax = long_long_unsigned_type_node; | |
985 | } | |
986 | else | |
987 | { | |
988 | for (int i = 0; i < NUM_INT_N_ENTS; i++) | |
989 | if (int_n_enabled_p[i]) | |
990 | { | |
991 | char name[50]; | |
992 | sprintf (name, "__int%d unsigned", int_n_data[i].bitsize); | |
993 | ||
42aa1173 | 994 | if (strcmp (name, UINTMAX_TYPE) == 0) |
88d0c3f0 MS |
995 | { |
996 | *pintmax = int_n_trees[i].signed_type; | |
997 | *puintmax = int_n_trees[i].unsigned_type; | |
42aa1173 | 998 | return; |
88d0c3f0 MS |
999 | } |
1000 | } | |
42aa1173 | 1001 | gcc_unreachable (); |
88d0c3f0 MS |
1002 | } |
1003 | } | |
1004 | ||
3812f460 MS |
1005 | /* Determine the range [*PMIN, *PMAX] that the expression ARG is |
1006 | in and that is representable in type int. | |
1007 | Return true when the range is a subrange of that of int. | |
1008 | When ARG is null it is as if it had the full range of int. | |
31c87a43 MS |
1009 | When ABSOLUTE is true the range reflects the absolute value of |
1010 | the argument. When ABSOLUTE is false, negative bounds of | |
1011 | the determined range are replaced with NEGBOUND. */ | |
1012 | ||
1013 | static bool | |
3812f460 | 1014 | get_int_range (tree arg, HOST_WIDE_INT *pmin, HOST_WIDE_INT *pmax, |
bbe2e114 JL |
1015 | bool absolute, HOST_WIDE_INT negbound, |
1016 | class vr_values *vr_values) | |
31c87a43 | 1017 | { |
3812f460 MS |
1018 | /* The type of the result. */ |
1019 | const_tree type = integer_type_node; | |
1020 | ||
31c87a43 MS |
1021 | bool knownrange = false; |
1022 | ||
1023 | if (!arg) | |
1024 | { | |
3812f460 MS |
1025 | *pmin = tree_to_shwi (TYPE_MIN_VALUE (type)); |
1026 | *pmax = tree_to_shwi (TYPE_MAX_VALUE (type)); | |
31c87a43 | 1027 | } |
e602bbfc MS |
1028 | else if (TREE_CODE (arg) == INTEGER_CST |
1029 | && TYPE_PRECISION (TREE_TYPE (arg)) <= TYPE_PRECISION (type)) | |
31c87a43 MS |
1030 | { |
1031 | /* For a constant argument return its value adjusted as specified | |
1032 | by NEGATIVE and NEGBOUND and return true to indicate that the | |
1033 | result is known. */ | |
1034 | *pmin = tree_fits_shwi_p (arg) ? tree_to_shwi (arg) : tree_to_uhwi (arg); | |
1035 | *pmax = *pmin; | |
1036 | knownrange = true; | |
1037 | } | |
1038 | else | |
1039 | { | |
1040 | /* True if the argument's range cannot be determined. */ | |
1041 | bool unknown = true; | |
1042 | ||
3812f460 | 1043 | tree argtype = TREE_TYPE (arg); |
31c87a43 | 1044 | |
3812f460 MS |
1045 | /* Ignore invalid arguments with greater precision that that |
1046 | of the expected type (e.g., in sprintf("%*i", 12LL, i)). | |
1047 | They will have been detected and diagnosed by -Wformat and | |
1048 | so it's not important to complicate this code to try to deal | |
1049 | with them again. */ | |
31c87a43 | 1050 | if (TREE_CODE (arg) == SSA_NAME |
3812f460 MS |
1051 | && INTEGRAL_TYPE_P (argtype) |
1052 | && TYPE_PRECISION (argtype) <= TYPE_PRECISION (type)) | |
31c87a43 MS |
1053 | { |
1054 | /* Try to determine the range of values of the integer argument. */ | |
5a54a15e | 1055 | value_range *vr = vr_values->get_value_range (arg); |
54994253 | 1056 | if (range_int_cst_p (vr)) |
31c87a43 MS |
1057 | { |
1058 | HOST_WIDE_INT type_min | |
3812f460 MS |
1059 | = (TYPE_UNSIGNED (argtype) |
1060 | ? tree_to_uhwi (TYPE_MIN_VALUE (argtype)) | |
1061 | : tree_to_shwi (TYPE_MIN_VALUE (argtype))); | |
31c87a43 | 1062 | |
3812f460 | 1063 | HOST_WIDE_INT type_max = tree_to_uhwi (TYPE_MAX_VALUE (argtype)); |
31c87a43 | 1064 | |
54994253 AH |
1065 | *pmin = TREE_INT_CST_LOW (vr->min ()); |
1066 | *pmax = TREE_INT_CST_LOW (vr->max ()); | |
31c87a43 | 1067 | |
3812f460 MS |
1068 | if (*pmin < *pmax) |
1069 | { | |
1070 | /* Return true if the adjusted range is a subrange of | |
1071 | the full range of the argument's type. *PMAX may | |
1072 | be less than *PMIN when the argument is unsigned | |
1073 | and its upper bound is in excess of TYPE_MAX. In | |
1074 | that (invalid) case disregard the range and use that | |
1075 | of the expected type instead. */ | |
1076 | knownrange = type_min < *pmin || *pmax < type_max; | |
1077 | ||
1078 | unknown = false; | |
1079 | } | |
31c87a43 MS |
1080 | } |
1081 | } | |
1082 | ||
1083 | /* Handle an argument with an unknown range as if none had been | |
1084 | provided. */ | |
1085 | if (unknown) | |
bbe2e114 JL |
1086 | return get_int_range (NULL_TREE, pmin, pmax, absolute, |
1087 | negbound, vr_values); | |
31c87a43 MS |
1088 | } |
1089 | ||
1090 | /* Adjust each bound as specified by ABSOLUTE and NEGBOUND. */ | |
1091 | if (absolute) | |
1092 | { | |
1093 | if (*pmin < 0) | |
1094 | { | |
1095 | if (*pmin == *pmax) | |
1096 | *pmin = *pmax = -*pmin; | |
1097 | else | |
1098 | { | |
3812f460 MS |
1099 | /* Make sure signed overlow is avoided. */ |
1100 | gcc_assert (*pmin != HOST_WIDE_INT_MIN); | |
1101 | ||
31c87a43 MS |
1102 | HOST_WIDE_INT tmp = -*pmin; |
1103 | *pmin = 0; | |
1104 | if (*pmax < tmp) | |
1105 | *pmax = tmp; | |
1106 | } | |
1107 | } | |
1108 | } | |
1109 | else if (*pmin < negbound) | |
1110 | *pmin = negbound; | |
1111 | ||
1112 | return knownrange; | |
1113 | } | |
1114 | ||
573aa7d4 MS |
1115 | /* With the range [*ARGMIN, *ARGMAX] of an integer directive's actual |
1116 | argument, due to the conversion from either *ARGMIN or *ARGMAX to | |
1117 | the type of the directive's formal argument it's possible for both | |
1118 | to result in the same number of bytes or a range of bytes that's | |
1119 | less than the number of bytes that would result from formatting | |
1120 | some other value in the range [*ARGMIN, *ARGMAX]. This can be | |
1121 | determined by checking for the actual argument being in the range | |
1122 | of the type of the directive. If it isn't it must be assumed to | |
1123 | take on the full range of the directive's type. | |
5b00f9d2 JJ |
1124 | Return true when the range has been adjusted to the full range |
1125 | of DIRTYPE, and false otherwise. */ | |
573aa7d4 MS |
1126 | |
1127 | static bool | |
1128 | adjust_range_for_overflow (tree dirtype, tree *argmin, tree *argmax) | |
1129 | { | |
1130 | tree argtype = TREE_TYPE (*argmin); | |
1131 | unsigned argprec = TYPE_PRECISION (argtype); | |
1132 | unsigned dirprec = TYPE_PRECISION (dirtype); | |
1133 | ||
1134 | /* If the actual argument and the directive's argument have the same | |
1135 | precision and sign there can be no overflow and so there is nothing | |
1136 | to adjust. */ | |
1137 | if (argprec == dirprec && TYPE_SIGN (argtype) == TYPE_SIGN (dirtype)) | |
1138 | return false; | |
1139 | ||
1140 | /* The logic below was inspired/lifted from the CONVERT_EXPR_CODE_P | |
1141 | branch in the extract_range_from_unary_expr function in tree-vrp.c. */ | |
1142 | ||
1143 | if (TREE_CODE (*argmin) == INTEGER_CST | |
1144 | && TREE_CODE (*argmax) == INTEGER_CST | |
1145 | && (dirprec >= argprec | |
1146 | || integer_zerop (int_const_binop (RSHIFT_EXPR, | |
1147 | int_const_binop (MINUS_EXPR, | |
1148 | *argmax, | |
1149 | *argmin), | |
1150 | size_int (dirprec))))) | |
1151 | { | |
1152 | *argmin = force_fit_type (dirtype, wi::to_widest (*argmin), 0, false); | |
1153 | *argmax = force_fit_type (dirtype, wi::to_widest (*argmax), 0, false); | |
1154 | ||
1155 | /* If *ARGMIN is still less than *ARGMAX the conversion above | |
1156 | is safe. Otherwise, it has overflowed and would be unsafe. */ | |
1157 | if (tree_int_cst_le (*argmin, *argmax)) | |
1158 | return false; | |
1159 | } | |
1160 | ||
5b00f9d2 JJ |
1161 | *argmin = TYPE_MIN_VALUE (dirtype); |
1162 | *argmax = TYPE_MAX_VALUE (dirtype); | |
573aa7d4 MS |
1163 | return true; |
1164 | } | |
1165 | ||
88d0c3f0 | 1166 | /* Return a range representing the minimum and maximum number of bytes |
31c87a43 MS |
1167 | that the format directive DIR will output for any argument given |
1168 | the WIDTH and PRECISION (extracted from DIR). This function is | |
1169 | used when the directive argument or its value isn't known. */ | |
88d0c3f0 MS |
1170 | |
1171 | static fmtresult | |
bbe2e114 | 1172 | format_integer (const directive &dir, tree arg, vr_values *vr_values) |
88d0c3f0 | 1173 | { |
42aa1173 JJ |
1174 | tree intmax_type_node; |
1175 | tree uintmax_type_node; | |
88d0c3f0 | 1176 | |
02f1ef40 MS |
1177 | /* Base to format the number in. */ |
1178 | int base; | |
1179 | ||
5d93da1d MS |
1180 | /* True when a conversion is preceded by a prefix indicating the base |
1181 | of the argument (octal or hexadecimal). */ | |
1182 | bool maybebase = dir.get_flag ('#'); | |
1183 | ||
02f1ef40 MS |
1184 | /* True when a signed conversion is preceded by a sign or space. */ |
1185 | bool maybesign = false; | |
1186 | ||
1187 | /* True for signed conversions (i.e., 'd' and 'i'). */ | |
1188 | bool sign = false; | |
1189 | ||
1190 | switch (dir.specifier) | |
1191 | { | |
1192 | case 'd': | |
1193 | case 'i': | |
1194 | /* Space and '+' are only meaningful for signed conversions. */ | |
1195 | maybesign = dir.get_flag (' ') | dir.get_flag ('+'); | |
1196 | sign = true; | |
1197 | base = 10; | |
1198 | break; | |
1199 | case 'u': | |
1200 | base = 10; | |
1201 | break; | |
1202 | case 'o': | |
1203 | base = 8; | |
1204 | break; | |
1205 | case 'X': | |
1206 | case 'x': | |
1207 | base = 16; | |
1208 | break; | |
1209 | default: | |
1210 | gcc_unreachable (); | |
1211 | } | |
88d0c3f0 MS |
1212 | |
1213 | /* The type of the "formal" argument expected by the directive. */ | |
1214 | tree dirtype = NULL_TREE; | |
1215 | ||
1216 | /* Determine the expected type of the argument from the length | |
1217 | modifier. */ | |
954b452a | 1218 | switch (dir.modifier) |
88d0c3f0 MS |
1219 | { |
1220 | case FMT_LEN_none: | |
954b452a | 1221 | if (dir.specifier == 'p') |
88d0c3f0 MS |
1222 | dirtype = ptr_type_node; |
1223 | else | |
1224 | dirtype = sign ? integer_type_node : unsigned_type_node; | |
1225 | break; | |
1226 | ||
1227 | case FMT_LEN_h: | |
1228 | dirtype = sign ? short_integer_type_node : short_unsigned_type_node; | |
1229 | break; | |
1230 | ||
1231 | case FMT_LEN_hh: | |
1232 | dirtype = sign ? signed_char_type_node : unsigned_char_type_node; | |
1233 | break; | |
1234 | ||
1235 | case FMT_LEN_l: | |
1236 | dirtype = sign ? long_integer_type_node : long_unsigned_type_node; | |
1237 | break; | |
1238 | ||
1239 | case FMT_LEN_L: | |
1240 | case FMT_LEN_ll: | |
ed30d2a6 MS |
1241 | dirtype = (sign |
1242 | ? long_long_integer_type_node | |
1243 | : long_long_unsigned_type_node); | |
88d0c3f0 MS |
1244 | break; |
1245 | ||
1246 | case FMT_LEN_z: | |
42aa1173 | 1247 | dirtype = signed_or_unsigned_type_for (!sign, size_type_node); |
88d0c3f0 MS |
1248 | break; |
1249 | ||
1250 | case FMT_LEN_t: | |
42aa1173 | 1251 | dirtype = signed_or_unsigned_type_for (!sign, ptrdiff_type_node); |
88d0c3f0 MS |
1252 | break; |
1253 | ||
1254 | case FMT_LEN_j: | |
42aa1173 | 1255 | build_intmax_type_nodes (&intmax_type_node, &uintmax_type_node); |
88d0c3f0 MS |
1256 | dirtype = sign ? intmax_type_node : uintmax_type_node; |
1257 | break; | |
1258 | ||
1259 | default: | |
42aa1173 | 1260 | return fmtresult (); |
88d0c3f0 MS |
1261 | } |
1262 | ||
1263 | /* The type of the argument to the directive, either deduced from | |
1264 | the actual non-constant argument if one is known, or from | |
1265 | the directive itself when none has been provided because it's | |
1266 | a va_list. */ | |
1267 | tree argtype = NULL_TREE; | |
1268 | ||
1269 | if (!arg) | |
1270 | { | |
1271 | /* When the argument has not been provided, use the type of | |
1272 | the directive's argument as an approximation. This will | |
1273 | result in false positives for directives like %i with | |
1274 | arguments with smaller precision (such as short or char). */ | |
1275 | argtype = dirtype; | |
1276 | } | |
1277 | else if (TREE_CODE (arg) == INTEGER_CST) | |
1278 | { | |
88d0c3f0 MS |
1279 | /* When a constant argument has been provided use its value |
1280 | rather than type to determine the length of the output. */ | |
5d93da1d | 1281 | fmtresult res; |
de6aa933 | 1282 | |
31c87a43 | 1283 | if ((dir.prec[0] <= 0 && dir.prec[1] >= 0) && integer_zerop (arg)) |
de6aa933 | 1284 | { |
573aa7d4 MS |
1285 | /* As a special case, a precision of zero with a zero argument |
1286 | results in zero bytes except in base 8 when the '#' flag is | |
1287 | specified, and for signed conversions in base 8 and 10 when | |
71dedb33 MS |
1288 | either the space or '+' flag has been specified and it results |
1289 | in just one byte (with width having the normal effect). This | |
1290 | must extend to the case of a specified precision with | |
1291 | an unknown value because it can be zero. */ | |
5d93da1d | 1292 | res.range.min = ((base == 8 && dir.get_flag ('#')) || maybesign); |
31c87a43 | 1293 | if (res.range.min == 0 && dir.prec[0] != dir.prec[1]) |
5d93da1d MS |
1294 | { |
1295 | res.range.max = 1; | |
1296 | res.range.likely = 1; | |
1297 | } | |
1298 | else | |
1299 | { | |
1300 | res.range.max = res.range.min; | |
1301 | res.range.likely = res.range.min; | |
1302 | } | |
de6aa933 MS |
1303 | } |
1304 | else | |
1305 | { | |
1306 | /* Convert the argument to the type of the directive. */ | |
1307 | arg = fold_convert (dirtype, arg); | |
88d0c3f0 | 1308 | |
31c87a43 | 1309 | res.range.min = tree_digits (arg, base, dir.prec[0], |
5d93da1d | 1310 | maybesign, maybebase); |
31c87a43 | 1311 | if (dir.prec[0] == dir.prec[1]) |
5d93da1d | 1312 | res.range.max = res.range.min; |
31c87a43 MS |
1313 | else |
1314 | res.range.max = tree_digits (arg, base, dir.prec[1], | |
1315 | maybesign, maybebase); | |
5d93da1d | 1316 | res.range.likely = res.range.min; |
4d82d0bc | 1317 | res.knownrange = true; |
de6aa933 | 1318 | } |
88d0c3f0 | 1319 | |
5d93da1d MS |
1320 | res.range.unlikely = res.range.max; |
1321 | ||
1322 | /* Bump up the counters if WIDTH is greater than LEN. */ | |
1323 | res.adjust_for_width_or_precision (dir.width, dirtype, base, | |
1324 | (sign | maybebase) + (base == 16)); | |
1325 | /* Bump up the counters again if PRECision is greater still. */ | |
1326 | res.adjust_for_width_or_precision (dir.prec, dirtype, base, | |
1327 | (sign | maybebase) + (base == 16)); | |
31c87a43 | 1328 | |
88d0c3f0 MS |
1329 | return res; |
1330 | } | |
9c582dbb | 1331 | else if (INTEGRAL_TYPE_P (TREE_TYPE (arg)) |
88d0c3f0 | 1332 | || TREE_CODE (TREE_TYPE (arg)) == POINTER_TYPE) |
0e817197 JJ |
1333 | /* Determine the type of the provided non-constant argument. */ |
1334 | argtype = TREE_TYPE (arg); | |
88d0c3f0 | 1335 | else |
0e817197 JJ |
1336 | /* Don't bother with invalid arguments since they likely would |
1337 | have already been diagnosed, and disable any further checking | |
1338 | of the format string by returning [-1, -1]. */ | |
1339 | return fmtresult (); | |
88d0c3f0 | 1340 | |
1eb4547b | 1341 | fmtresult res; |
88d0c3f0 MS |
1342 | |
1343 | /* Using either the range the non-constant argument is in, or its | |
1344 | type (either "formal" or actual), create a range of values that | |
1345 | constrain the length of output given the warning level. */ | |
1346 | tree argmin = NULL_TREE; | |
1347 | tree argmax = NULL_TREE; | |
1348 | ||
053d5e0c JJ |
1349 | if (arg |
1350 | && TREE_CODE (arg) == SSA_NAME | |
9c582dbb | 1351 | && INTEGRAL_TYPE_P (argtype)) |
88d0c3f0 MS |
1352 | { |
1353 | /* Try to determine the range of values of the integer argument | |
1354 | (range information is not available for pointers). */ | |
e506dc87 | 1355 | value_range *vr = vr_values->get_value_range (arg); |
54994253 | 1356 | if (range_int_cst_p (vr)) |
88d0c3f0 | 1357 | { |
54994253 AH |
1358 | argmin = vr->min (); |
1359 | argmax = vr->max (); | |
573aa7d4 MS |
1360 | |
1361 | /* Set KNOWNRANGE if the argument is in a known subrange | |
71dedb33 MS |
1362 | of the directive's type and neither width nor precision |
1363 | is unknown. (KNOWNRANGE may be reset below). */ | |
573aa7d4 | 1364 | res.knownrange |
71dedb33 MS |
1365 | = ((!tree_int_cst_equal (TYPE_MIN_VALUE (dirtype), argmin) |
1366 | || !tree_int_cst_equal (TYPE_MAX_VALUE (dirtype), argmax)) | |
1367 | && dir.known_width_and_precision ()); | |
573aa7d4 MS |
1368 | |
1369 | res.argmin = argmin; | |
1370 | res.argmax = argmax; | |
88d0c3f0 | 1371 | } |
54994253 | 1372 | else if (vr->kind () == VR_ANTI_RANGE) |
88d0c3f0 MS |
1373 | { |
1374 | /* Handle anti-ranges if/when bug 71690 is resolved. */ | |
1375 | } | |
54994253 | 1376 | else if (vr->varying_p () || vr->undefined_p ()) |
88d0c3f0 MS |
1377 | { |
1378 | /* The argument here may be the result of promoting the actual | |
1379 | argument to int. Try to determine the type of the actual | |
5fe39ec0 | 1380 | argument before promotion and narrow down its range that |
88d0c3f0 MS |
1381 | way. */ |
1382 | gimple *def = SSA_NAME_DEF_STMT (arg); | |
5fe39ec0 | 1383 | if (is_gimple_assign (def)) |
88d0c3f0 MS |
1384 | { |
1385 | tree_code code = gimple_assign_rhs_code (def); | |
1eb4547b MS |
1386 | if (code == INTEGER_CST) |
1387 | { | |
1388 | arg = gimple_assign_rhs1 (def); | |
bbe2e114 | 1389 | return format_integer (dir, arg, vr_values); |
1eb4547b MS |
1390 | } |
1391 | ||
88d0c3f0 | 1392 | if (code == NOP_EXPR) |
0e817197 JJ |
1393 | { |
1394 | tree type = TREE_TYPE (gimple_assign_rhs1 (def)); | |
9c582dbb | 1395 | if (INTEGRAL_TYPE_P (type) |
0e817197 JJ |
1396 | || TREE_CODE (type) == POINTER_TYPE) |
1397 | argtype = type; | |
1398 | } | |
88d0c3f0 MS |
1399 | } |
1400 | } | |
1401 | } | |
1402 | ||
1403 | if (!argmin) | |
1404 | { | |
5b00f9d2 | 1405 | if (TREE_CODE (argtype) == POINTER_TYPE) |
88d0c3f0 | 1406 | { |
5b00f9d2 JJ |
1407 | argmin = build_int_cst (pointer_sized_int_node, 0); |
1408 | argmax = build_all_ones_cst (pointer_sized_int_node); | |
88d0c3f0 MS |
1409 | } |
1410 | else | |
1411 | { | |
5b00f9d2 JJ |
1412 | argmin = TYPE_MIN_VALUE (argtype); |
1413 | argmax = TYPE_MAX_VALUE (argtype); | |
88d0c3f0 | 1414 | } |
573aa7d4 MS |
1415 | } |
1416 | ||
1417 | /* Clear KNOWNRANGE if the range has been adjusted to the maximum | |
1418 | of the directive. If it has been cleared then since ARGMIN and/or | |
1419 | ARGMAX have been adjusted also adjust the corresponding ARGMIN and | |
1420 | ARGMAX in the result to include in diagnostics. */ | |
1421 | if (adjust_range_for_overflow (dirtype, &argmin, &argmax)) | |
1422 | { | |
1423 | res.knownrange = false; | |
1424 | res.argmin = argmin; | |
1425 | res.argmax = argmax; | |
1426 | } | |
1427 | ||
5b00f9d2 JJ |
1428 | /* Recursively compute the minimum and maximum from the known range. */ |
1429 | if (TYPE_UNSIGNED (dirtype) || tree_int_cst_sgn (argmin) >= 0) | |
573aa7d4 | 1430 | { |
5b00f9d2 JJ |
1431 | /* For unsigned conversions/directives or signed when |
1432 | the minimum is positive, use the minimum and maximum to compute | |
1433 | the shortest and longest output, respectively. */ | |
bbe2e114 JL |
1434 | res.range.min = format_integer (dir, argmin, vr_values).range.min; |
1435 | res.range.max = format_integer (dir, argmax, vr_values).range.max; | |
573aa7d4 | 1436 | } |
5b00f9d2 | 1437 | else if (tree_int_cst_sgn (argmax) < 0) |
573aa7d4 | 1438 | { |
5b00f9d2 JJ |
1439 | /* For signed conversions/directives if maximum is negative, |
1440 | use the minimum as the longest output and maximum as the | |
1441 | shortest output. */ | |
bbe2e114 JL |
1442 | res.range.min = format_integer (dir, argmax, vr_values).range.min; |
1443 | res.range.max = format_integer (dir, argmin, vr_values).range.max; | |
573aa7d4 | 1444 | } |
5b00f9d2 | 1445 | else |
88d0c3f0 | 1446 | { |
5b00f9d2 JJ |
1447 | /* Otherwise, 0 is inside of the range and minimum negative. Use 0 |
1448 | as the shortest output and for the longest output compute the | |
1449 | length of the output of both minimum and maximum and pick the | |
1450 | longer. */ | |
bbe2e114 JL |
1451 | unsigned HOST_WIDE_INT max1 |
1452 | = format_integer (dir, argmin, vr_values).range.max; | |
1453 | unsigned HOST_WIDE_INT max2 | |
1454 | = format_integer (dir, argmax, vr_values).range.max; | |
1455 | res.range.min | |
1456 | = format_integer (dir, integer_zero_node, vr_values).range.min; | |
5b00f9d2 | 1457 | res.range.max = MAX (max1, max2); |
88d0c3f0 MS |
1458 | } |
1459 | ||
34a64e47 | 1460 | /* If the range is known, use the maximum as the likely length. */ |
f589a1ce MS |
1461 | if (res.knownrange) |
1462 | res.range.likely = res.range.max; | |
1463 | else | |
1464 | { | |
34a64e47 JJ |
1465 | /* Otherwise, use the minimum. Except for the case where for %#x or |
1466 | %#o the minimum is just for a single value in the range (0) and | |
1467 | for all other values it is something longer, like 0x1 or 01. | |
1468 | Use the length for value 1 in that case instead as the likely | |
1469 | length. */ | |
f589a1ce | 1470 | res.range.likely = res.range.min; |
34a64e47 JJ |
1471 | if (maybebase |
1472 | && base != 10 | |
1473 | && (tree_int_cst_sgn (argmin) < 0 || tree_int_cst_sgn (argmax) > 0)) | |
f589a1ce MS |
1474 | { |
1475 | if (res.range.min == 1) | |
1476 | res.range.likely += base == 8 ? 1 : 2; | |
1477 | else if (res.range.min == 2 | |
1478 | && base == 16 | |
1479 | && (dir.width[0] == 2 || dir.prec[0] == 2)) | |
1480 | ++res.range.likely; | |
1481 | } | |
1482 | } | |
1483 | ||
5d93da1d MS |
1484 | res.range.unlikely = res.range.max; |
1485 | res.adjust_for_width_or_precision (dir.width, dirtype, base, | |
1486 | (sign | maybebase) + (base == 16)); | |
1487 | res.adjust_for_width_or_precision (dir.prec, dirtype, base, | |
1488 | (sign | maybebase) + (base == 16)); | |
1489 | ||
88d0c3f0 MS |
1490 | return res; |
1491 | } | |
1492 | ||
cfce1a4a MS |
1493 | /* Return the number of bytes that a format directive consisting of FLAGS, |
1494 | PRECision, format SPECification, and MPFR rounding specifier RNDSPEC, | |
1495 | would result for argument X under ideal conditions (i.e., if PREC | |
1496 | weren't excessive). MPFR 3.1 allocates large amounts of memory for | |
1497 | values of PREC with large magnitude and can fail (see MPFR bug #21056). | |
1498 | This function works around those problems. */ | |
1499 | ||
1500 | static unsigned HOST_WIDE_INT | |
1501 | get_mpfr_format_length (mpfr_ptr x, const char *flags, HOST_WIDE_INT prec, | |
1502 | char spec, char rndspec) | |
1503 | { | |
1504 | char fmtstr[40]; | |
1505 | ||
1506 | HOST_WIDE_INT len = strlen (flags); | |
1507 | ||
1508 | fmtstr[0] = '%'; | |
1509 | memcpy (fmtstr + 1, flags, len); | |
1510 | memcpy (fmtstr + 1 + len, ".*R", 3); | |
1511 | fmtstr[len + 4] = rndspec; | |
1512 | fmtstr[len + 5] = spec; | |
1513 | fmtstr[len + 6] = '\0'; | |
1514 | ||
5b4f088d MS |
1515 | spec = TOUPPER (spec); |
1516 | if (spec == 'E' || spec == 'F') | |
1517 | { | |
1518 | /* For %e, specify the precision explicitly since mpfr_sprintf | |
1519 | does its own thing just to be different (see MPFR bug 21088). */ | |
1520 | if (prec < 0) | |
1521 | prec = 6; | |
1522 | } | |
1523 | else | |
1524 | { | |
1525 | /* Avoid passing negative precisions with larger magnitude to MPFR | |
1526 | to avoid exposing its bugs. (A negative precision is supposed | |
1527 | to be ignored.) */ | |
1528 | if (prec < 0) | |
1529 | prec = -1; | |
1530 | } | |
cfce1a4a MS |
1531 | |
1532 | HOST_WIDE_INT p = prec; | |
1533 | ||
71dedb33 | 1534 | if (spec == 'G' && !strchr (flags, '#')) |
cfce1a4a | 1535 | { |
71dedb33 MS |
1536 | /* For G/g without the pound flag, precision gives the maximum number |
1537 | of significant digits which is bounded by LDBL_MAX_10_EXP, or, for | |
1538 | a 128 bit IEEE extended precision, 4932. Using twice as much here | |
1539 | should be more than sufficient for any real format. */ | |
cfce1a4a MS |
1540 | if ((IEEE_MAX_10_EXP * 2) < prec) |
1541 | prec = IEEE_MAX_10_EXP * 2; | |
1542 | p = prec; | |
1543 | } | |
1544 | else | |
1545 | { | |
1546 | /* Cap precision arbitrarily at 1KB and add the difference | |
1547 | (if any) to the MPFR result. */ | |
8657c838 | 1548 | if (prec > 1024) |
cfce1a4a MS |
1549 | p = 1024; |
1550 | } | |
1551 | ||
1552 | len = mpfr_snprintf (NULL, 0, fmtstr, (int)p, x); | |
1553 | ||
1554 | /* Handle the unlikely (impossible?) error by returning more than | |
1555 | the maximum dictated by the function's return type. */ | |
1556 | if (len < 0) | |
1557 | return target_dir_max () + 1; | |
1558 | ||
1559 | /* Adjust the return value by the difference. */ | |
1560 | if (p < prec) | |
1561 | len += prec - p; | |
1562 | ||
1563 | return len; | |
1564 | } | |
1565 | ||
88d0c3f0 | 1566 | /* Return the number of bytes to format using the format specifier |
5b4f088d MS |
1567 | SPEC and the precision PREC the largest value in the real floating |
1568 | TYPE. */ | |
88d0c3f0 | 1569 | |
cfce1a4a MS |
1570 | static unsigned HOST_WIDE_INT |
1571 | format_floating_max (tree type, char spec, HOST_WIDE_INT prec) | |
88d0c3f0 MS |
1572 | { |
1573 | machine_mode mode = TYPE_MODE (type); | |
1574 | ||
1575 | /* IBM Extended mode. */ | |
1576 | if (MODE_COMPOSITE_P (mode)) | |
1577 | mode = DFmode; | |
1578 | ||
1579 | /* Get the real type format desription for the target. */ | |
1580 | const real_format *rfmt = REAL_MODE_FORMAT (mode); | |
1581 | REAL_VALUE_TYPE rv; | |
1582 | ||
954b452a | 1583 | real_maxval (&rv, 0, mode); |
88d0c3f0 MS |
1584 | |
1585 | /* Convert the GCC real value representation with the precision | |
1586 | of the real type to the mpfr_t format with the GCC default | |
1587 | round-to-nearest mode. */ | |
1588 | mpfr_t x; | |
1589 | mpfr_init2 (x, rfmt->p); | |
89dc4c94 | 1590 | mpfr_from_real (x, &rv, GMP_RNDN); |
88d0c3f0 | 1591 | |
1eb4547b | 1592 | /* Return a value one greater to account for the leading minus sign. */ |
0a95c7e2 ML |
1593 | unsigned HOST_WIDE_INT r |
1594 | = 1 + get_mpfr_format_length (x, "", prec, spec, 'D'); | |
1595 | mpfr_clear (x); | |
1596 | return r; | |
88d0c3f0 MS |
1597 | } |
1598 | ||
1599 | /* Return a range representing the minimum and maximum number of bytes | |
a983542b MS |
1600 | that the directive DIR will output for any argument. PREC gives |
1601 | the adjusted precision range to account for negative precisions | |
1602 | meaning the default 6. This function is used when the directive | |
1603 | argument or its value isn't known. */ | |
88d0c3f0 MS |
1604 | |
1605 | static fmtresult | |
a983542b | 1606 | format_floating (const directive &dir, const HOST_WIDE_INT prec[2]) |
88d0c3f0 MS |
1607 | { |
1608 | tree type; | |
88d0c3f0 | 1609 | |
954b452a | 1610 | switch (dir.modifier) |
88d0c3f0 | 1611 | { |
30dc6ee9 | 1612 | case FMT_LEN_l: |
88d0c3f0 MS |
1613 | case FMT_LEN_none: |
1614 | type = double_type_node; | |
1615 | break; | |
1616 | ||
1617 | case FMT_LEN_L: | |
1618 | type = long_double_type_node; | |
88d0c3f0 MS |
1619 | break; |
1620 | ||
1621 | case FMT_LEN_ll: | |
1622 | type = long_double_type_node; | |
88d0c3f0 MS |
1623 | break; |
1624 | ||
1625 | default: | |
1eb4547b | 1626 | return fmtresult (); |
88d0c3f0 MS |
1627 | } |
1628 | ||
1629 | /* The minimum and maximum number of bytes produced by the directive. */ | |
1eb4547b | 1630 | fmtresult res; |
88d0c3f0 | 1631 | |
5d93da1d MS |
1632 | /* The minimum output as determined by flags. It's always at least 1. |
1633 | When plus or space are set the output is preceded by either a sign | |
1634 | or a space. */ | |
a983542b MS |
1635 | unsigned flagmin = (1 /* for the first digit */ |
1636 | + (dir.get_flag ('+') | dir.get_flag (' '))); | |
88d0c3f0 | 1637 | |
7af7cb51 MS |
1638 | /* The minimum is 3 for "inf" and "nan" for all specifiers, plus 1 |
1639 | for the plus sign/space with the '+' and ' ' flags, respectively, | |
1640 | unless reduced below. */ | |
1641 | res.range.min = 2 + flagmin; | |
1642 | ||
5d93da1d MS |
1643 | /* When the pound flag is set the decimal point is included in output |
1644 | regardless of precision. Whether or not a decimal point is included | |
1645 | otherwise depends on the specification and precision. */ | |
1646 | bool radix = dir.get_flag ('#'); | |
88d0c3f0 | 1647 | |
954b452a | 1648 | switch (dir.specifier) |
88d0c3f0 MS |
1649 | { |
1650 | case 'A': | |
1651 | case 'a': | |
1652 | { | |
5d93da1d | 1653 | HOST_WIDE_INT minprec = 6 + !radix /* decimal point */; |
31c87a43 | 1654 | if (dir.prec[0] <= 0) |
5d93da1d | 1655 | minprec = 0; |
31c87a43 MS |
1656 | else if (dir.prec[0] > 0) |
1657 | minprec = dir.prec[0] + !radix /* decimal point */; | |
5d93da1d | 1658 | |
7af7cb51 MS |
1659 | res.range.likely = (2 /* 0x */ |
1660 | + flagmin | |
1661 | + radix | |
1662 | + minprec | |
1663 | + 3 /* p+0 */); | |
5d93da1d | 1664 | |
a983542b | 1665 | res.range.max = format_floating_max (type, 'a', prec[1]); |
5d93da1d MS |
1666 | |
1667 | /* The unlikely maximum accounts for the longest multibyte | |
1668 | decimal point character. */ | |
31c87a43 | 1669 | res.range.unlikely = res.range.max; |
a983542b | 1670 | if (dir.prec[1] > 0) |
31c87a43 | 1671 | res.range.unlikely += target_mb_len_max () - 1; |
5b4f088d | 1672 | |
88d0c3f0 MS |
1673 | break; |
1674 | } | |
1675 | ||
1676 | case 'E': | |
1677 | case 'e': | |
1678 | { | |
a983542b MS |
1679 | /* Minimum output attributable to precision and, when it's |
1680 | non-zero, decimal point. */ | |
1681 | HOST_WIDE_INT minprec = prec[0] ? prec[0] + !radix : 0; | |
1682 | ||
7af7cb51 | 1683 | /* The likely minimum output is "[-+]1.234567e+00" regardless |
88d0c3f0 | 1684 | of the value of the actual argument. */ |
7af7cb51 MS |
1685 | res.range.likely = (flagmin |
1686 | + radix | |
1687 | + minprec | |
1688 | + 2 /* e+ */ + 2); | |
5d93da1d | 1689 | |
a983542b | 1690 | res.range.max = format_floating_max (type, 'e', prec[1]); |
5d93da1d MS |
1691 | |
1692 | /* The unlikely maximum accounts for the longest multibyte | |
1693 | decimal point character. */ | |
31c87a43 MS |
1694 | if (dir.prec[0] != dir.prec[1] |
1695 | || dir.prec[0] == -1 || dir.prec[0] > 0) | |
5d93da1d MS |
1696 | res.range.unlikely = res.range.max + target_mb_len_max () -1; |
1697 | else | |
1698 | res.range.unlikely = res.range.max; | |
88d0c3f0 MS |
1699 | break; |
1700 | } | |
1701 | ||
1702 | case 'F': | |
1703 | case 'f': | |
1704 | { | |
a983542b MS |
1705 | /* Minimum output attributable to precision and, when it's non-zero, |
1706 | decimal point. */ | |
1707 | HOST_WIDE_INT minprec = prec[0] ? prec[0] + !radix : 0; | |
1708 | ||
7af7cb51 MS |
1709 | /* For finite numbers (i.e., not infinity or NaN) the lower bound |
1710 | when precision isn't specified is 8 bytes ("1.23456" since | |
1711 | precision is taken to be 6). When precision is zero, the lower | |
1712 | bound is 1 byte (e.g., "1"). Otherwise, when precision is greater | |
1713 | than zero, then the lower bound is 2 plus precision (plus flags). | |
1714 | But in all cases, the lower bound is no greater than 3. */ | |
1715 | unsigned HOST_WIDE_INT min = flagmin + radix + minprec; | |
1716 | if (min < res.range.min) | |
1717 | res.range.min = min; | |
5d93da1d MS |
1718 | |
1719 | /* Compute the upper bound for -TYPE_MAX. */ | |
a983542b | 1720 | res.range.max = format_floating_max (type, 'f', prec[1]); |
5d93da1d | 1721 | |
71dedb33 MS |
1722 | /* The minimum output with unknown precision is a single byte |
1723 | (e.g., "0") but the more likely output is 3 bytes ("0.0"). */ | |
1724 | if (dir.prec[0] < 0 && dir.prec[1] > 0) | |
1725 | res.range.likely = 3; | |
1726 | else | |
7af7cb51 | 1727 | res.range.likely = min; |
5d93da1d MS |
1728 | |
1729 | /* The unlikely maximum accounts for the longest multibyte | |
1730 | decimal point character. */ | |
31c87a43 MS |
1731 | if (dir.prec[0] != dir.prec[1] |
1732 | || dir.prec[0] == -1 || dir.prec[0] > 0) | |
5d93da1d | 1733 | res.range.unlikely = res.range.max + target_mb_len_max () - 1; |
88d0c3f0 MS |
1734 | break; |
1735 | } | |
5b4f088d | 1736 | |
88d0c3f0 MS |
1737 | case 'G': |
1738 | case 'g': | |
1739 | { | |
5b4f088d MS |
1740 | /* The %g output depends on precision and the exponent of |
1741 | the argument. Since the value of the argument isn't known | |
1742 | the lower bound on the range of bytes (not counting flags | |
71dedb33 MS |
1743 | or width) is 1 plus radix (i.e., either "0" or "0." for |
1744 | "%g" and "%#g", respectively, with a zero argument). */ | |
7af7cb51 MS |
1745 | unsigned HOST_WIDE_INT min = flagmin + radix; |
1746 | if (min < res.range.min) | |
1747 | res.range.min = min; | |
71dedb33 MS |
1748 | |
1749 | char spec = 'g'; | |
1750 | HOST_WIDE_INT maxprec = dir.prec[1]; | |
1751 | if (radix && maxprec) | |
1752 | { | |
1753 | /* When the pound flag (radix) is set, trailing zeros aren't | |
1754 | trimmed and so the longest output is the same as for %e, | |
1755 | except with precision minus 1 (as specified in C11). */ | |
1756 | spec = 'e'; | |
1757 | if (maxprec > 0) | |
1758 | --maxprec; | |
1759 | else if (maxprec < 0) | |
1760 | maxprec = 5; | |
1761 | } | |
a983542b MS |
1762 | else |
1763 | maxprec = prec[1]; | |
71dedb33 MS |
1764 | |
1765 | res.range.max = format_floating_max (type, spec, maxprec); | |
1766 | ||
1767 | /* The likely output is either the maximum computed above | |
1768 | minus 1 (assuming the maximum is positive) when precision | |
1769 | is known (or unspecified), or the same minimum as for %e | |
1770 | (which is computed for a non-negative argument). Unlike | |
1771 | for the other specifiers above the likely output isn't | |
1772 | the minimum because for %g that's 1 which is unlikely. */ | |
1773 | if (dir.prec[1] < 0 | |
1774 | || (unsigned HOST_WIDE_INT)dir.prec[1] < target_int_max ()) | |
1775 | res.range.likely = res.range.max - 1; | |
1776 | else | |
1777 | { | |
1778 | HOST_WIDE_INT minprec = 6 + !radix /* decimal point */; | |
1779 | res.range.likely = (flagmin | |
1780 | + radix | |
1781 | + minprec | |
1782 | + 2 /* e+ */ + 2); | |
1783 | } | |
5d93da1d MS |
1784 | |
1785 | /* The unlikely maximum accounts for the longest multibyte | |
1786 | decimal point character. */ | |
1787 | res.range.unlikely = res.range.max + target_mb_len_max () - 1; | |
88d0c3f0 MS |
1788 | break; |
1789 | } | |
1790 | ||
1791 | default: | |
1eb4547b | 1792 | return fmtresult (); |
88d0c3f0 MS |
1793 | } |
1794 | ||
5d93da1d MS |
1795 | /* Bump up the byte counters if WIDTH is greater. */ |
1796 | res.adjust_for_width_or_precision (dir.width); | |
88d0c3f0 MS |
1797 | return res; |
1798 | } | |
1799 | ||
1800 | /* Return a range representing the minimum and maximum number of bytes | |
31c87a43 MS |
1801 | that the directive DIR will write on output for the floating argument |
1802 | ARG. */ | |
88d0c3f0 MS |
1803 | |
1804 | static fmtresult | |
bbe2e114 | 1805 | format_floating (const directive &dir, tree arg, vr_values *) |
88d0c3f0 | 1806 | { |
31c87a43 | 1807 | HOST_WIDE_INT prec[] = { dir.prec[0], dir.prec[1] }; |
aa215959 JJ |
1808 | tree type = (dir.modifier == FMT_LEN_L || dir.modifier == FMT_LEN_ll |
1809 | ? long_double_type_node : double_type_node); | |
88d0c3f0 | 1810 | |
71dedb33 MS |
1811 | /* For an indeterminate precision the lower bound must be assumed |
1812 | to be zero. */ | |
5d93da1d | 1813 | if (TOUPPER (dir.specifier) == 'A') |
88d0c3f0 | 1814 | { |
71dedb33 MS |
1815 | /* Get the number of fractional decimal digits needed to represent |
1816 | the argument without a loss of accuracy. */ | |
71dedb33 MS |
1817 | unsigned fmtprec |
1818 | = REAL_MODE_FORMAT (TYPE_MODE (type))->p; | |
1819 | ||
1820 | /* The precision of the IEEE 754 double format is 53. | |
1821 | The precision of all other GCC binary double formats | |
1822 | is 56 or less. */ | |
1823 | unsigned maxprec = fmtprec <= 56 ? 13 : 15; | |
1824 | ||
5d93da1d MS |
1825 | /* For %a, leave the minimum precision unspecified to let |
1826 | MFPR trim trailing zeros (as it and many other systems | |
1827 | including Glibc happen to do) and set the maximum | |
1828 | precision to reflect what it would be with trailing zeros | |
1829 | present (as Solaris and derived systems do). */ | |
71dedb33 | 1830 | if (dir.prec[1] < 0) |
5d93da1d | 1831 | { |
71dedb33 MS |
1832 | /* Both bounds are negative implies that precision has |
1833 | not been specified. */ | |
1834 | prec[0] = maxprec; | |
1835 | prec[1] = -1; | |
1836 | } | |
1837 | else if (dir.prec[0] < 0) | |
1838 | { | |
1839 | /* With a negative lower bound and a non-negative upper | |
1840 | bound set the minimum precision to zero and the maximum | |
1841 | to the greater of the maximum precision (i.e., with | |
1842 | trailing zeros present) and the specified upper bound. */ | |
1843 | prec[0] = 0; | |
1844 | prec[1] = dir.prec[1] < maxprec ? maxprec : dir.prec[1]; | |
1845 | } | |
1846 | } | |
1847 | else if (dir.prec[0] < 0) | |
1848 | { | |
1849 | if (dir.prec[1] < 0) | |
1850 | { | |
1851 | /* A precision in a strictly negative range is ignored and | |
1852 | the default of 6 is used instead. */ | |
1853 | prec[0] = prec[1] = 6; | |
1854 | } | |
1855 | else | |
1856 | { | |
1857 | /* For a precision in a partly negative range, the lower bound | |
1858 | must be assumed to be zero and the new upper bound is the | |
1859 | greater of 6 (the default precision used when the specified | |
1860 | precision is negative) and the upper bound of the specified | |
1861 | range. */ | |
1862 | prec[0] = 0; | |
1863 | prec[1] = dir.prec[1] < 6 ? 6 : dir.prec[1]; | |
5d93da1d | 1864 | } |
88d0c3f0 MS |
1865 | } |
1866 | ||
aa215959 JJ |
1867 | if (!arg |
1868 | || TREE_CODE (arg) != REAL_CST | |
1869 | || !useless_type_conversion_p (type, TREE_TYPE (arg))) | |
a983542b MS |
1870 | return format_floating (dir, prec); |
1871 | ||
954b452a MS |
1872 | /* The minimum and maximum number of bytes produced by the directive. */ |
1873 | fmtresult res; | |
88d0c3f0 | 1874 | |
954b452a MS |
1875 | /* Get the real type format desription for the target. */ |
1876 | const REAL_VALUE_TYPE *rvp = TREE_REAL_CST_PTR (arg); | |
1877 | const real_format *rfmt = REAL_MODE_FORMAT (TYPE_MODE (TREE_TYPE (arg))); | |
1878 | ||
7af7cb51 MS |
1879 | if (!real_isfinite (rvp)) |
1880 | { | |
1881 | /* The format for Infinity and NaN is "[-]inf"/"[-]infinity" | |
1882 | and "[-]nan" with the choice being implementation-defined | |
1883 | but not locale dependent. */ | |
1884 | bool sign = dir.get_flag ('+') || real_isneg (rvp); | |
1885 | res.range.min = 3 + sign; | |
1886 | ||
1887 | res.range.likely = res.range.min; | |
1888 | res.range.max = res.range.min; | |
bd1f3566 MS |
1889 | /* The unlikely maximum is "[-/+]infinity" or "[-/+][qs]nan". |
1890 | For NaN, the C/POSIX standards specify two formats: | |
1891 | "[-/+]nan" | |
1892 | and | |
1893 | "[-/+]nan(n-char-sequence)" | |
1894 | No known printf implementation outputs the latter format but AIX | |
1895 | outputs QNaN and SNaN for quiet and signalling NaN, respectively, | |
1896 | so the unlikely maximum reflects that. */ | |
1897 | res.range.unlikely = sign + (real_isinf (rvp) ? 8 : 4); | |
7af7cb51 MS |
1898 | |
1899 | /* The range for infinity and NaN is known unless either width | |
1900 | or precision is unknown. Width has the same effect regardless | |
1901 | of whether the argument is finite. Precision is either ignored | |
1902 | (e.g., Glibc) or can have an effect on the short vs long format | |
1903 | such as inf/infinity (e.g., Solaris). */ | |
1904 | res.knownrange = dir.known_width_and_precision (); | |
1905 | ||
1906 | /* Adjust the range for width but ignore precision. */ | |
1907 | res.adjust_for_width_or_precision (dir.width); | |
1908 | ||
1909 | return res; | |
1910 | } | |
1911 | ||
954b452a MS |
1912 | char fmtstr [40]; |
1913 | char *pfmt = fmtstr; | |
88d0c3f0 | 1914 | |
954b452a MS |
1915 | /* Append flags. */ |
1916 | for (const char *pf = "-+ #0"; *pf; ++pf) | |
1917 | if (dir.get_flag (*pf)) | |
1918 | *pfmt++ = *pf; | |
88d0c3f0 | 1919 | |
954b452a | 1920 | *pfmt = '\0'; |
88d0c3f0 | 1921 | |
954b452a MS |
1922 | { |
1923 | /* Set up an array to easily iterate over. */ | |
1924 | unsigned HOST_WIDE_INT* const minmax[] = { | |
1925 | &res.range.min, &res.range.max | |
1926 | }; | |
88d0c3f0 | 1927 | |
954b452a | 1928 | for (int i = 0; i != sizeof minmax / sizeof *minmax; ++i) |
5b4f088d | 1929 | { |
954b452a MS |
1930 | /* Convert the GCC real value representation with the precision |
1931 | of the real type to the mpfr_t format rounding down in the | |
1932 | first iteration that computes the minimm and up in the second | |
1933 | that computes the maximum. This order is arbibtrary because | |
1934 | rounding in either direction can result in longer output. */ | |
1935 | mpfr_t mpfrval; | |
1936 | mpfr_init2 (mpfrval, rfmt->p); | |
cf8e20bf | 1937 | mpfr_from_real (mpfrval, rvp, i ? GMP_RNDU : GMP_RNDD); |
954b452a MS |
1938 | |
1939 | /* Use the MPFR rounding specifier to round down in the first | |
1940 | iteration and then up. In most but not all cases this will | |
1941 | result in the same number of bytes. */ | |
1942 | char rndspec = "DU"[i]; | |
1943 | ||
1944 | /* Format it and store the result in the corresponding member | |
1945 | of the result struct. */ | |
5d93da1d MS |
1946 | *minmax[i] = get_mpfr_format_length (mpfrval, fmtstr, prec[i], |
1947 | dir.specifier, rndspec); | |
0a95c7e2 | 1948 | mpfr_clear (mpfrval); |
5b4f088d | 1949 | } |
954b452a | 1950 | } |
5b4f088d | 1951 | |
954b452a MS |
1952 | /* Make sure the minimum is less than the maximum (MPFR rounding |
1953 | in the call to mpfr_snprintf can result in the reverse. */ | |
1954 | if (res.range.max < res.range.min) | |
1955 | { | |
1956 | unsigned HOST_WIDE_INT tmp = res.range.min; | |
1957 | res.range.min = res.range.max; | |
1958 | res.range.max = tmp; | |
1959 | } | |
1eb4547b | 1960 | |
71dedb33 MS |
1961 | /* The range is known unless either width or precision is unknown. */ |
1962 | res.knownrange = dir.known_width_and_precision (); | |
1963 | ||
1964 | /* For the same floating point constant, unless width or precision | |
1965 | is unknown, use the longer output as the likely maximum since | |
1966 | with round to nearest either is equally likely. Otheriwse, when | |
1967 | precision is unknown, use the greater of the minimum and 3 as | |
1968 | the likely output (for "0.0" since zero precision is unlikely). */ | |
1969 | if (res.knownrange) | |
1970 | res.range.likely = res.range.max; | |
1971 | else if (res.range.min < 3 | |
1972 | && dir.prec[0] < 0 | |
1973 | && (unsigned HOST_WIDE_INT)dir.prec[1] == target_int_max ()) | |
1974 | res.range.likely = 3; | |
1975 | else | |
1976 | res.range.likely = res.range.min; | |
5d93da1d | 1977 | |
5d93da1d MS |
1978 | res.range.unlikely = res.range.max; |
1979 | ||
1980 | if (res.range.max > 2 && (prec[0] != 0 || prec[1] != 0)) | |
954b452a | 1981 | { |
5d93da1d MS |
1982 | /* Unless the precision is zero output longer than 2 bytes may |
1983 | include the decimal point which must be a single character | |
1984 | up to MB_LEN_MAX in length. This is overly conservative | |
1985 | since in some conversions some constants result in no decimal | |
1986 | point (e.g., in %g). */ | |
1987 | res.range.unlikely += target_mb_len_max () - 1; | |
88d0c3f0 MS |
1988 | } |
1989 | ||
5d93da1d | 1990 | res.adjust_for_width_or_precision (dir.width); |
954b452a | 1991 | return res; |
88d0c3f0 MS |
1992 | } |
1993 | ||
1994 | /* Return a FMTRESULT struct set to the lengths of the shortest and longest | |
1995 | strings referenced by the expression STR, or (-1, -1) when not known. | |
1996 | Used by the format_string function below. */ | |
1997 | ||
1998 | static fmtresult | |
4148b00d | 1999 | get_string_length (tree str, unsigned eltsize) |
88d0c3f0 MS |
2000 | { |
2001 | if (!str) | |
1eb4547b | 2002 | return fmtresult (); |
88d0c3f0 | 2003 | |
a41cc59d MS |
2004 | c_strlen_data data; |
2005 | memset (&data, 0, sizeof (c_strlen_data)); | |
2006 | tree slen = c_strlen (str, 1, &data, eltsize); | |
2007 | if (slen && TREE_CODE (slen) == INTEGER_CST) | |
2008 | { | |
2009 | /* The string is properly terminated and | |
2010 | we know its length. */ | |
2011 | fmtresult res (tree_to_shwi (slen)); | |
2012 | res.nonstr = NULL_TREE; | |
2013 | return res; | |
2014 | } | |
2015 | else if (!slen | |
2016 | && data.decl | |
2017 | && data.len | |
2018 | && TREE_CODE (data.len) == INTEGER_CST) | |
2019 | { | |
2020 | /* STR was not properly NUL terminated, but we have | |
2021 | length information about the unterminated string. */ | |
2022 | fmtresult res (tree_to_shwi (data.len)); | |
2023 | res.nonstr = data.decl; | |
2024 | return res; | |
2025 | } | |
2026 | ||
88d0c3f0 MS |
2027 | /* Determine the length of the shortest and longest string referenced |
2028 | by STR. Strings of unknown lengths are bounded by the sizes of | |
2029 | arrays that subexpressions of STR may refer to. Pointers that | |
2030 | aren't known to point any such arrays result in LENRANGE[1] set | |
a41cc59d MS |
2031 | to SIZE_MAX. NONSTR is set to the declaration of the constant |
2032 | array that is known not to be nul-terminated. */ | |
88d0c3f0 | 2033 | tree lenrange[2]; |
a41cc59d MS |
2034 | tree nonstr; |
2035 | bool flexarray = get_range_strlen (str, lenrange, eltsize, false, &nonstr); | |
88d0c3f0 MS |
2036 | |
2037 | if (lenrange [0] || lenrange [1]) | |
2038 | { | |
5d93da1d MS |
2039 | HOST_WIDE_INT min |
2040 | = (tree_fits_uhwi_p (lenrange[0]) | |
2041 | ? tree_to_uhwi (lenrange[0]) | |
2042 | : 0); | |
2043 | ||
2044 | HOST_WIDE_INT max | |
2045 | = (tree_fits_uhwi_p (lenrange[1]) | |
2046 | ? tree_to_uhwi (lenrange[1]) | |
2047 | : HOST_WIDE_INT_M1U); | |
88d0c3f0 | 2048 | |
5d93da1d MS |
2049 | /* get_range_strlen() returns the target value of SIZE_MAX for |
2050 | strings of unknown length. Bump it up to HOST_WIDE_INT_M1U | |
2051 | which may be bigger. */ | |
2052 | if ((unsigned HOST_WIDE_INT)min == target_size_max ()) | |
2053 | min = HOST_WIDE_INT_M1U; | |
2054 | if ((unsigned HOST_WIDE_INT)max == target_size_max ()) | |
2055 | max = HOST_WIDE_INT_M1U; | |
88d0c3f0 | 2056 | |
5d93da1d | 2057 | fmtresult res (min, max); |
a41cc59d | 2058 | res.nonstr = nonstr; |
5d93da1d MS |
2059 | |
2060 | /* Set RES.KNOWNRANGE to true if and only if all strings referenced | |
88d0c3f0 MS |
2061 | by STR are known to be bounded (though not necessarily by their |
2062 | actual length but perhaps by their maximum possible length). */ | |
5d93da1d MS |
2063 | if (res.range.max < target_int_max ()) |
2064 | { | |
2065 | res.knownrange = true; | |
2066 | /* When the the length of the longest string is known and not | |
2067 | excessive use it as the likely length of the string(s). */ | |
2068 | res.range.likely = res.range.max; | |
2069 | } | |
2070 | else | |
2071 | { | |
ebee1eb9 | 2072 | /* When the upper bound is unknown (it can be zero or excessive) |
5d93da1d | 2073 | set the likely length to the greater of 1 and the length of |
ebee1eb9 | 2074 | the shortest string and reset the lower bound to zero. */ |
5d93da1d | 2075 | res.range.likely = res.range.min ? res.range.min : warn_level > 1; |
ebee1eb9 | 2076 | res.range.min = 0; |
5d93da1d MS |
2077 | } |
2078 | ||
3f343040 MS |
2079 | /* If the range of string length has been estimated from the size |
2080 | of an array at the end of a struct assume that it's longer than | |
2081 | the array bound says it is in case it's used as a poor man's | |
2082 | flexible array member, such as in struct S { char a[4]; }; */ | |
2083 | res.range.unlikely = flexarray ? HOST_WIDE_INT_MAX : res.range.max; | |
88d0c3f0 | 2084 | |
88d0c3f0 MS |
2085 | return res; |
2086 | } | |
2087 | ||
4148b00d | 2088 | return fmtresult (); |
88d0c3f0 MS |
2089 | } |
2090 | ||
2091 | /* Return the minimum and maximum number of characters formatted | |
954b452a MS |
2092 | by the '%c' format directives and its wide character form for |
2093 | the argument ARG. ARG can be null (for functions such as | |
2094 | vsprinf). */ | |
88d0c3f0 MS |
2095 | |
2096 | static fmtresult | |
bbe2e114 | 2097 | format_character (const directive &dir, tree arg, vr_values *vr_values) |
88d0c3f0 | 2098 | { |
1eb4547b | 2099 | fmtresult res; |
88d0c3f0 | 2100 | |
5d93da1d MS |
2101 | res.knownrange = true; |
2102 | ||
bbcbd744 MS |
2103 | if (dir.specifier == 'C' |
2104 | || dir.modifier == FMT_LEN_l) | |
5d93da1d | 2105 | { |
5d93da1d MS |
2106 | /* A wide character can result in as few as zero bytes. */ |
2107 | res.range.min = 0; | |
2108 | ||
31c87a43 | 2109 | HOST_WIDE_INT min, max; |
bbe2e114 | 2110 | if (get_int_range (arg, &min, &max, false, 0, vr_values)) |
5d93da1d | 2111 | { |
31c87a43 MS |
2112 | if (min == 0 && max == 0) |
2113 | { | |
2114 | /* The NUL wide character results in no bytes. */ | |
2115 | res.range.max = 0; | |
2116 | res.range.likely = 0; | |
2117 | res.range.unlikely = 0; | |
2118 | } | |
bbcbd744 | 2119 | else if (min >= 0 && min < 128) |
31c87a43 | 2120 | { |
bbcbd744 MS |
2121 | /* Be conservative if the target execution character set |
2122 | is not a 1-to-1 mapping to the source character set or | |
2123 | if the source set is not ASCII. */ | |
2124 | bool one_2_one_ascii | |
2125 | = (target_to_host_charmap[0] == 1 && target_to_host ('a') == 97); | |
2126 | ||
31c87a43 MS |
2127 | /* A wide character in the ASCII range most likely results |
2128 | in a single byte, and only unlikely in up to MB_LEN_MAX. */ | |
bbcbd744 | 2129 | res.range.max = one_2_one_ascii ? 1 : target_mb_len_max ();; |
31c87a43 MS |
2130 | res.range.likely = 1; |
2131 | res.range.unlikely = target_mb_len_max (); | |
bbcbd744 | 2132 | res.mayfail = !one_2_one_ascii; |
31c87a43 MS |
2133 | } |
2134 | else | |
2135 | { | |
2136 | /* A wide character outside the ASCII range likely results | |
2137 | in up to two bytes, and only unlikely in up to MB_LEN_MAX. */ | |
2138 | res.range.max = target_mb_len_max (); | |
2139 | res.range.likely = 2; | |
2140 | res.range.unlikely = res.range.max; | |
bbcbd744 MS |
2141 | /* Converting such a character may fail. */ |
2142 | res.mayfail = true; | |
31c87a43 | 2143 | } |
5d93da1d MS |
2144 | } |
2145 | else | |
2146 | { | |
31c87a43 MS |
2147 | /* An unknown wide character is treated the same as a wide |
2148 | character outside the ASCII range. */ | |
5d93da1d MS |
2149 | res.range.max = target_mb_len_max (); |
2150 | res.range.likely = 2; | |
2151 | res.range.unlikely = res.range.max; | |
bbcbd744 | 2152 | res.mayfail = true; |
5d93da1d | 2153 | } |
954b452a MS |
2154 | } |
2155 | else | |
2156 | { | |
2157 | /* A plain '%c' directive. Its ouput is exactly 1. */ | |
2158 | res.range.min = res.range.max = 1; | |
31c87a43 | 2159 | res.range.likely = res.range.unlikely = 1; |
954b452a | 2160 | res.knownrange = true; |
954b452a MS |
2161 | } |
2162 | ||
5d93da1d MS |
2163 | /* Bump up the byte counters if WIDTH is greater. */ |
2164 | return res.adjust_for_width_or_precision (dir.width); | |
954b452a MS |
2165 | } |
2166 | ||
2167 | /* Return the minimum and maximum number of characters formatted | |
31c87a43 MS |
2168 | by the '%s' format directive and its wide character form for |
2169 | the argument ARG. ARG can be null (for functions such as | |
2170 | vsprinf). */ | |
954b452a MS |
2171 | |
2172 | static fmtresult | |
bbe2e114 | 2173 | format_string (const directive &dir, tree arg, vr_values *) |
954b452a MS |
2174 | { |
2175 | fmtresult res; | |
88d0c3f0 | 2176 | |
954b452a | 2177 | /* Compute the range the argument's length can be in. */ |
0863decd JL |
2178 | int count_by = 1; |
2179 | if (dir.specifier == 'S' || dir.modifier == FMT_LEN_l) | |
2180 | { | |
2181 | /* Get a node for a C type that will be the same size | |
2182 | as a wchar_t on the target. */ | |
2183 | tree node = get_typenode_from_name (MODIFIED_WCHAR_TYPE); | |
2184 | ||
2185 | /* Now that we have a suitable node, get the number of | |
2186 | bytes it occupies. */ | |
2187 | count_by = int_size_in_bytes (node); | |
2188 | gcc_checking_assert (count_by == 2 || count_by == 4); | |
2189 | } | |
2190 | ||
4148b00d | 2191 | fmtresult slen = get_string_length (arg, count_by); |
a151e93b MS |
2192 | if (slen.range.min == slen.range.max |
2193 | && slen.range.min < HOST_WIDE_INT_MAX) | |
88d0c3f0 | 2194 | { |
5d93da1d MS |
2195 | /* The argument is either a string constant or it refers |
2196 | to one of a number of strings of the same length. */ | |
88d0c3f0 | 2197 | |
954b452a MS |
2198 | /* A '%s' directive with a string argument with constant length. */ |
2199 | res.range = slen.range; | |
88d0c3f0 | 2200 | |
bbcbd744 MS |
2201 | if (dir.specifier == 'S' |
2202 | || dir.modifier == FMT_LEN_l) | |
954b452a | 2203 | { |
5d93da1d MS |
2204 | /* In the worst case the length of output of a wide string S |
2205 | is bounded by MB_LEN_MAX * wcslen (S). */ | |
2206 | res.range.max *= target_mb_len_max (); | |
2207 | res.range.unlikely = res.range.max; | |
2208 | /* It's likely that the the total length is not more that | |
2209 | 2 * wcslen (S).*/ | |
2210 | res.range.likely = res.range.min * 2; | |
88d0c3f0 | 2211 | |
8657c838 | 2212 | if (dir.prec[1] >= 0 |
31c87a43 | 2213 | && (unsigned HOST_WIDE_INT)dir.prec[1] < res.range.max) |
5d93da1d | 2214 | { |
31c87a43 MS |
2215 | res.range.max = dir.prec[1]; |
2216 | res.range.likely = dir.prec[1]; | |
2217 | res.range.unlikely = dir.prec[1]; | |
5d93da1d | 2218 | } |
88d0c3f0 | 2219 | |
31c87a43 MS |
2220 | if (dir.prec[0] < 0 && dir.prec[1] > -1) |
2221 | res.range.min = 0; | |
8657c838 | 2222 | else if (dir.prec[0] >= 0) |
31c87a43 MS |
2223 | res.range.likely = dir.prec[0]; |
2224 | ||
5d93da1d MS |
2225 | /* Even a non-empty wide character string need not convert into |
2226 | any bytes. */ | |
2227 | res.range.min = 0; | |
bbcbd744 MS |
2228 | |
2229 | /* A non-empty wide character conversion may fail. */ | |
2230 | if (slen.range.max > 0) | |
2231 | res.mayfail = true; | |
954b452a | 2232 | } |
5d93da1d | 2233 | else |
954b452a | 2234 | { |
5d93da1d MS |
2235 | res.knownrange = true; |
2236 | ||
31c87a43 | 2237 | if (dir.prec[0] < 0 && dir.prec[1] > -1) |
5d93da1d | 2238 | res.range.min = 0; |
31c87a43 MS |
2239 | else if ((unsigned HOST_WIDE_INT)dir.prec[0] < res.range.min) |
2240 | res.range.min = dir.prec[0]; | |
2241 | ||
2242 | if ((unsigned HOST_WIDE_INT)dir.prec[1] < res.range.max) | |
5d93da1d | 2243 | { |
31c87a43 MS |
2244 | res.range.max = dir.prec[1]; |
2245 | res.range.likely = dir.prec[1]; | |
2246 | res.range.unlikely = dir.prec[1]; | |
5d93da1d | 2247 | } |
954b452a MS |
2248 | } |
2249 | } | |
2250 | else if (arg && integer_zerop (arg)) | |
2251 | { | |
2252 | /* Handle null pointer argument. */ | |
88d0c3f0 | 2253 | |
954b452a MS |
2254 | fmtresult res (0); |
2255 | res.nullp = true; | |
2256 | return res; | |
2257 | } | |
2258 | else | |
2259 | { | |
ebee1eb9 MS |
2260 | /* For a '%s' and '%ls' directive with a non-constant string (either |
2261 | one of a number of strings of known length or an unknown string) | |
2262 | the minimum number of characters is lesser of PRECISION[0] and | |
2263 | the length of the shortest known string or zero, and the maximum | |
2264 | is the lessser of the length of the longest known string or | |
2265 | PTRDIFF_MAX and PRECISION[1]. The likely length is either | |
2266 | the minimum at level 1 and the greater of the minimum and 1 | |
2267 | at level 2. This result is adjust upward for width (if it's | |
2268 | specified). */ | |
2269 | ||
bbcbd744 MS |
2270 | if (dir.specifier == 'S' |
2271 | || dir.modifier == FMT_LEN_l) | |
ebee1eb9 MS |
2272 | { |
2273 | /* A wide character converts to as few as zero bytes. */ | |
2274 | slen.range.min = 0; | |
2275 | if (slen.range.max < target_int_max ()) | |
2276 | slen.range.max *= target_mb_len_max (); | |
2277 | ||
2278 | if (slen.range.likely < target_int_max ()) | |
2279 | slen.range.likely *= 2; | |
2280 | ||
2281 | if (slen.range.likely < target_int_max ()) | |
2282 | slen.range.unlikely *= target_mb_len_max (); | |
bbcbd744 MS |
2283 | |
2284 | /* A non-empty wide character conversion may fail. */ | |
2285 | if (slen.range.max > 0) | |
2286 | res.mayfail = true; | |
ebee1eb9 MS |
2287 | } |
2288 | ||
2289 | res.range = slen.range; | |
88d0c3f0 | 2290 | |
8657c838 | 2291 | if (dir.prec[0] >= 0) |
954b452a | 2292 | { |
ebee1eb9 MS |
2293 | /* Adjust the minimum to zero if the string length is unknown, |
2294 | or at most the lower bound of the precision otherwise. */ | |
954b452a | 2295 | if (slen.range.min >= target_int_max ()) |
ebee1eb9 | 2296 | res.range.min = 0; |
31c87a43 | 2297 | else if ((unsigned HOST_WIDE_INT)dir.prec[0] < slen.range.min) |
ebee1eb9 | 2298 | res.range.min = dir.prec[0]; |
954b452a | 2299 | |
ebee1eb9 | 2300 | /* Make both maxima no greater than the upper bound of precision. */ |
31c87a43 | 2301 | if ((unsigned HOST_WIDE_INT)dir.prec[1] < slen.range.max |
954b452a | 2302 | || slen.range.max >= target_int_max ()) |
5d93da1d | 2303 | { |
ebee1eb9 MS |
2304 | res.range.max = dir.prec[1]; |
2305 | res.range.unlikely = dir.prec[1]; | |
5d93da1d | 2306 | } |
ebee1eb9 MS |
2307 | |
2308 | /* If precision is constant, set the likely counter to the lesser | |
2309 | of it and the maximum string length. Otherwise, if the lower | |
2310 | bound of precision is greater than zero, set the likely counter | |
2311 | to the minimum. Otherwise set it to zero or one based on | |
2312 | the warning level. */ | |
2313 | if (dir.prec[0] == dir.prec[1]) | |
2314 | res.range.likely | |
2315 | = ((unsigned HOST_WIDE_INT)dir.prec[0] < slen.range.max | |
2316 | ? dir.prec[0] : slen.range.max); | |
2317 | else if (dir.prec[0] > 0) | |
2318 | res.range.likely = res.range.min; | |
2319 | else | |
2320 | res.range.likely = warn_level > 1; | |
2321 | } | |
2322 | else if (dir.prec[1] >= 0) | |
2323 | { | |
2324 | res.range.min = 0; | |
2325 | if ((unsigned HOST_WIDE_INT)dir.prec[1] < slen.range.max) | |
2326 | res.range.max = dir.prec[1]; | |
2327 | res.range.likely = dir.prec[1] ? warn_level > 1 : 0; | |
954b452a MS |
2328 | } |
2329 | else if (slen.range.min >= target_int_max ()) | |
2330 | { | |
ebee1eb9 MS |
2331 | res.range.min = 0; |
2332 | res.range.max = HOST_WIDE_INT_MAX; | |
2333 | /* At level 1 strings of unknown length are assumed to be | |
5d93da1d MS |
2334 | empty, while at level 1 they are assumed to be one byte |
2335 | long. */ | |
ebee1eb9 MS |
2336 | res.range.likely = warn_level > 1; |
2337 | } | |
2338 | else | |
2339 | { | |
2340 | /* A string of unknown length unconstrained by precision is | |
2341 | assumed to be empty at level 1 and just one character long | |
2342 | at higher levels. */ | |
2343 | if (res.range.likely >= target_int_max ()) | |
2344 | res.range.likely = warn_level > 1; | |
88d0c3f0 | 2345 | } |
954b452a | 2346 | |
ebee1eb9 | 2347 | res.range.unlikely = res.range.max; |
88d0c3f0 MS |
2348 | } |
2349 | ||
a41cc59d MS |
2350 | /* If the argument isn't a nul-terminated string and the number |
2351 | of bytes on output isn't bounded by precision, set NONSTR. */ | |
2352 | if (slen.nonstr && slen.range.min < (unsigned HOST_WIDE_INT)dir.prec[0]) | |
2353 | res.nonstr = slen.nonstr; | |
2354 | ||
5d93da1d MS |
2355 | /* Bump up the byte counters if WIDTH is greater. */ |
2356 | return res.adjust_for_width_or_precision (dir.width); | |
2357 | } | |
2358 | ||
2359 | /* Format plain string (part of the format string itself). */ | |
2360 | ||
2361 | static fmtresult | |
bbe2e114 | 2362 | format_plain (const directive &dir, tree, vr_values *) |
5d93da1d MS |
2363 | { |
2364 | fmtresult res (dir.len); | |
2365 | return res; | |
2366 | } | |
2367 | ||
2368 | /* Return true if the RESULT of a directive in a call describe by INFO | |
2369 | should be diagnosed given the AVAILable space in the destination. */ | |
2370 | ||
2371 | static bool | |
a09ca93a | 2372 | should_warn_p (const sprintf_dom_walker::call_info &info, |
5d93da1d MS |
2373 | const result_range &avail, const result_range &result) |
2374 | { | |
2375 | if (result.max <= avail.min) | |
2376 | { | |
2377 | /* The least amount of space remaining in the destination is big | |
2378 | enough for the longest output. */ | |
2379 | return false; | |
2380 | } | |
2381 | ||
2382 | if (info.bounded) | |
de6aa933 | 2383 | { |
8657c838 | 2384 | if (warn_format_trunc == 1 && result.min <= avail.max |
5d93da1d MS |
2385 | && info.retval_used ()) |
2386 | { | |
2387 | /* The likely amount of space remaining in the destination is big | |
2388 | enough for the least output and the return value is used. */ | |
2389 | return false; | |
2390 | } | |
88d0c3f0 | 2391 | |
8657c838 | 2392 | if (warn_format_trunc == 1 && result.likely <= avail.likely |
5d93da1d MS |
2393 | && !info.retval_used ()) |
2394 | { | |
2395 | /* The likely amount of space remaining in the destination is big | |
2396 | enough for the likely output and the return value is unused. */ | |
2397 | return false; | |
2398 | } | |
88d0c3f0 | 2399 | |
5d93da1d MS |
2400 | if (warn_format_trunc == 2 |
2401 | && result.likely <= avail.min | |
2402 | && (result.max <= avail.min | |
2403 | || result.max > HOST_WIDE_INT_MAX)) | |
2404 | { | |
2405 | /* The minimum amount of space remaining in the destination is big | |
2406 | enough for the longest output. */ | |
2407 | return false; | |
2408 | } | |
de6aa933 | 2409 | } |
5d93da1d MS |
2410 | else |
2411 | { | |
8657c838 | 2412 | if (warn_level == 1 && result.likely <= avail.likely) |
5d93da1d MS |
2413 | { |
2414 | /* The likely amount of space remaining in the destination is big | |
2415 | enough for the likely output. */ | |
2416 | return false; | |
2417 | } | |
88d0c3f0 | 2418 | |
5d93da1d MS |
2419 | if (warn_level == 2 |
2420 | && result.likely <= avail.min | |
2421 | && (result.max <= avail.min | |
2422 | || result.max > HOST_WIDE_INT_MAX)) | |
2423 | { | |
2424 | /* The minimum amount of space remaining in the destination is big | |
2425 | enough for the longest output. */ | |
2426 | return false; | |
2427 | } | |
2428 | } | |
1eb4547b | 2429 | |
5d93da1d | 2430 | return true; |
88d0c3f0 MS |
2431 | } |
2432 | ||
02f1ef40 MS |
2433 | /* At format string location describe by DIRLOC in a call described |
2434 | by INFO, issue a warning for a directive DIR whose output may be | |
2435 | in excess of the available space AVAIL_RANGE in the destination | |
2436 | given the formatting result FMTRES. This function does nothing | |
2437 | except decide whether to issue a warning for a possible write | |
2438 | past the end or truncation and, if so, format the warning. | |
2439 | Return true if a warning has been issued. */ | |
2440 | ||
2441 | static bool | |
89b6abbb | 2442 | maybe_warn (substring_loc &dirloc, location_t argloc, |
a09ca93a | 2443 | const sprintf_dom_walker::call_info &info, |
5d93da1d | 2444 | const result_range &avail_range, const result_range &res, |
02f1ef40 MS |
2445 | const directive &dir) |
2446 | { | |
5d93da1d MS |
2447 | if (!should_warn_p (info, avail_range, res)) |
2448 | return false; | |
2449 | ||
2450 | /* A warning will definitely be issued below. */ | |
2451 | ||
2452 | /* The maximum byte count to reference in the warning. Larger counts | |
2453 | imply that the upper bound is unknown (and could be anywhere between | |
2454 | RES.MIN + 1 and SIZE_MAX / 2) are printed as "N or more bytes" rather | |
2455 | than "between N and X" where X is some huge number. */ | |
2456 | unsigned HOST_WIDE_INT maxbytes = target_dir_max (); | |
02f1ef40 | 2457 | |
5d93da1d MS |
2458 | /* True when there is enough room in the destination for the least |
2459 | amount of a directive's output but not enough for its likely or | |
2460 | maximum output. */ | |
2461 | bool maybe = (res.min <= avail_range.max | |
2462 | && (avail_range.min < res.likely | |
2463 | || (res.max < HOST_WIDE_INT_MAX | |
2464 | && avail_range.min < res.max))); | |
2465 | ||
d127ae9f MS |
2466 | /* Buffer for the directive in the host character set (used when |
2467 | the source character set is different). */ | |
2468 | char hostdir[32]; | |
2469 | ||
5d93da1d | 2470 | if (avail_range.min == avail_range.max) |
02f1ef40 | 2471 | { |
5d93da1d MS |
2472 | /* The size of the destination region is exact. */ |
2473 | unsigned HOST_WIDE_INT navail = avail_range.max; | |
2474 | ||
d127ae9f | 2475 | if (target_to_host (*dir.beg) != '%') |
02f1ef40 | 2476 | { |
5d93da1d MS |
2477 | /* For plain character directives (i.e., the format string itself) |
2478 | but not others, point the caret at the first character that's | |
2479 | past the end of the destination. */ | |
36367d1f JJ |
2480 | if (navail < dir.len) |
2481 | dirloc.set_caret_index (dirloc.get_caret_idx () + navail); | |
02f1ef40 | 2482 | } |
5d93da1d MS |
2483 | |
2484 | if (*dir.beg == '\0') | |
02f1ef40 | 2485 | { |
5d93da1d MS |
2486 | /* This is the terminating nul. */ |
2487 | gcc_assert (res.min == 1 && res.min == res.max); | |
2488 | ||
89b6abbb | 2489 | return fmtwarn (dirloc, UNKNOWN_LOCATION, NULL, info.warnopt (), |
346114b5 JJ |
2490 | info.bounded |
2491 | ? (maybe | |
2492 | ? G_("%qE output may be truncated before the " | |
2493 | "last format character") | |
2494 | : G_("%qE output truncated before the last " | |
2495 | "format character")) | |
2496 | : (maybe | |
2497 | ? G_("%qE may write a terminating nul past the " | |
2498 | "end of the destination") | |
2499 | : G_("%qE writing a terminating nul past the " | |
2500 | "end of the destination")), | |
2501 | info.func); | |
5d93da1d MS |
2502 | } |
2503 | ||
2504 | if (res.min == res.max) | |
2505 | { | |
346114b5 JJ |
2506 | const char *d = target_to_host (hostdir, sizeof hostdir, dir.beg); |
2507 | if (!info.bounded) | |
2508 | return fmtwarn_n (dirloc, argloc, NULL, info.warnopt (), res.min, | |
2509 | "%<%.*s%> directive writing %wu byte into a " | |
2510 | "region of size %wu", | |
2511 | "%<%.*s%> directive writing %wu bytes into a " | |
2512 | "region of size %wu", | |
2513 | (int) dir.len, d, res.min, navail); | |
2514 | else if (maybe) | |
2515 | return fmtwarn_n (dirloc, argloc, NULL, info.warnopt (), res.min, | |
2516 | "%<%.*s%> directive output may be truncated " | |
2517 | "writing %wu byte into a region of size %wu", | |
2518 | "%<%.*s%> directive output may be truncated " | |
2519 | "writing %wu bytes into a region of size %wu", | |
2520 | (int) dir.len, d, res.min, navail); | |
2521 | else | |
2522 | return fmtwarn_n (dirloc, argloc, NULL, info.warnopt (), res.min, | |
2523 | "%<%.*s%> directive output truncated writing " | |
2524 | "%wu byte into a region of size %wu", | |
2525 | "%<%.*s%> directive output truncated writing " | |
2526 | "%wu bytes into a region of size %wu", | |
2527 | (int) dir.len, d, res.min, navail); | |
5d93da1d | 2528 | } |
8657c838 | 2529 | if (res.min == 0 && res.max < maxbytes) |
346114b5 JJ |
2530 | return fmtwarn (dirloc, argloc, NULL, |
2531 | info.warnopt (), | |
2532 | info.bounded | |
2533 | ? (maybe | |
2534 | ? G_("%<%.*s%> directive output may be truncated " | |
2535 | "writing up to %wu bytes into a region of " | |
2536 | "size %wu") | |
2537 | : G_("%<%.*s%> directive output truncated writing " | |
2538 | "up to %wu bytes into a region of size %wu")) | |
2539 | : G_("%<%.*s%> directive writing up to %wu bytes " | |
2540 | "into a region of size %wu"), (int) dir.len, | |
2541 | target_to_host (hostdir, sizeof hostdir, dir.beg), | |
2542 | res.max, navail); | |
5d93da1d | 2543 | |
8657c838 | 2544 | if (res.min == 0 && maxbytes <= res.max) |
346114b5 JJ |
2545 | /* This is a special case to avoid issuing the potentially |
2546 | confusing warning: | |
2547 | writing 0 or more bytes into a region of size 0. */ | |
2548 | return fmtwarn (dirloc, argloc, NULL, info.warnopt (), | |
2549 | info.bounded | |
2550 | ? (maybe | |
2551 | ? G_("%<%.*s%> directive output may be truncated " | |
2552 | "writing likely %wu or more bytes into a " | |
2553 | "region of size %wu") | |
2554 | : G_("%<%.*s%> directive output truncated writing " | |
2555 | "likely %wu or more bytes into a region of " | |
2556 | "size %wu")) | |
2557 | : G_("%<%.*s%> directive writing likely %wu or more " | |
2558 | "bytes into a region of size %wu"), (int) dir.len, | |
2559 | target_to_host (hostdir, sizeof hostdir, dir.beg), | |
2560 | res.likely, navail); | |
5d93da1d MS |
2561 | |
2562 | if (res.max < maxbytes) | |
346114b5 JJ |
2563 | return fmtwarn (dirloc, argloc, NULL, info.warnopt (), |
2564 | info.bounded | |
2565 | ? (maybe | |
2566 | ? G_("%<%.*s%> directive output may be truncated " | |
2567 | "writing between %wu and %wu bytes into a " | |
2568 | "region of size %wu") | |
2569 | : G_("%<%.*s%> directive output truncated " | |
2570 | "writing between %wu and %wu bytes into a " | |
2571 | "region of size %wu")) | |
2572 | : G_("%<%.*s%> directive writing between %wu and " | |
2573 | "%wu bytes into a region of size %wu"), | |
2574 | (int) dir.len, | |
2575 | target_to_host (hostdir, sizeof hostdir, dir.beg), | |
2576 | res.min, res.max, navail); | |
2577 | ||
2578 | return fmtwarn (dirloc, argloc, NULL, info.warnopt (), | |
2579 | info.bounded | |
2580 | ? (maybe | |
2581 | ? G_("%<%.*s%> directive output may be truncated " | |
2582 | "writing %wu or more bytes into a region of " | |
2583 | "size %wu") | |
2584 | : G_("%<%.*s%> directive output truncated writing " | |
2585 | "%wu or more bytes into a region of size %wu")) | |
2586 | : G_("%<%.*s%> directive writing %wu or more bytes " | |
2587 | "into a region of size %wu"), (int) dir.len, | |
d127ae9f | 2588 | target_to_host (hostdir, sizeof hostdir, dir.beg), |
5d93da1d | 2589 | res.min, navail); |
02f1ef40 MS |
2590 | } |
2591 | ||
5d93da1d MS |
2592 | /* The size of the destination region is a range. */ |
2593 | ||
d127ae9f | 2594 | if (target_to_host (*dir.beg) != '%') |
5d93da1d MS |
2595 | { |
2596 | unsigned HOST_WIDE_INT navail = avail_range.max; | |
2597 | ||
2598 | /* For plain character directives (i.e., the format string itself) | |
2599 | but not others, point the caret at the first character that's | |
2600 | past the end of the destination. */ | |
36367d1f JJ |
2601 | if (navail < dir.len) |
2602 | dirloc.set_caret_index (dirloc.get_caret_idx () + navail); | |
5d93da1d MS |
2603 | } |
2604 | ||
2605 | if (*dir.beg == '\0') | |
2606 | { | |
2607 | gcc_assert (res.min == 1 && res.min == res.max); | |
2608 | ||
346114b5 JJ |
2609 | return fmtwarn (dirloc, UNKNOWN_LOCATION, NULL, info.warnopt (), |
2610 | info.bounded | |
2611 | ? (maybe | |
2612 | ? G_("%qE output may be truncated before the last " | |
2613 | "format character") | |
2614 | : G_("%qE output truncated before the last format " | |
2615 | "character")) | |
2616 | : (maybe | |
2617 | ? G_("%qE may write a terminating nul past the end " | |
2618 | "of the destination") | |
2619 | : G_("%qE writing a terminating nul past the end " | |
2620 | "of the destination")), info.func); | |
5d93da1d MS |
2621 | } |
2622 | ||
2623 | if (res.min == res.max) | |
2624 | { | |
346114b5 JJ |
2625 | const char *d = target_to_host (hostdir, sizeof hostdir, dir.beg); |
2626 | if (!info.bounded) | |
2627 | return fmtwarn_n (dirloc, argloc, NULL, info.warnopt (), res.min, | |
2628 | "%<%.*s%> directive writing %wu byte into a region " | |
2629 | "of size between %wu and %wu", | |
2630 | "%<%.*s%> directive writing %wu bytes into a region " | |
2631 | "of size between %wu and %wu", (int) dir.len, d, | |
2632 | res.min, avail_range.min, avail_range.max); | |
2633 | else if (maybe) | |
2634 | return fmtwarn_n (dirloc, argloc, NULL, info.warnopt (), res.min, | |
2635 | "%<%.*s%> directive output may be truncated writing " | |
2636 | "%wu byte into a region of size between %wu and %wu", | |
2637 | "%<%.*s%> directive output may be truncated writing " | |
2638 | "%wu bytes into a region of size between %wu and " | |
2639 | "%wu", (int) dir.len, d, res.min, avail_range.min, | |
2640 | avail_range.max); | |
2641 | else | |
2642 | return fmtwarn_n (dirloc, argloc, NULL, info.warnopt (), res.min, | |
2643 | "%<%.*s%> directive output truncated writing %wu " | |
2644 | "byte into a region of size between %wu and %wu", | |
2645 | "%<%.*s%> directive output truncated writing %wu " | |
2646 | "bytes into a region of size between %wu and %wu", | |
2647 | (int) dir.len, d, res.min, avail_range.min, | |
2648 | avail_range.max); | |
5d93da1d MS |
2649 | } |
2650 | ||
8657c838 | 2651 | if (res.min == 0 && res.max < maxbytes) |
346114b5 JJ |
2652 | return fmtwarn (dirloc, argloc, NULL, info.warnopt (), |
2653 | info.bounded | |
2654 | ? (maybe | |
2655 | ? G_("%<%.*s%> directive output may be truncated " | |
2656 | "writing up to %wu bytes into a region of size " | |
2657 | "between %wu and %wu") | |
2658 | : G_("%<%.*s%> directive output truncated writing " | |
2659 | "up to %wu bytes into a region of size between " | |
2660 | "%wu and %wu")) | |
2661 | : G_("%<%.*s%> directive writing up to %wu bytes " | |
2662 | "into a region of size between %wu and %wu"), | |
2663 | (int) dir.len, | |
2664 | target_to_host (hostdir, sizeof hostdir, dir.beg), | |
2665 | res.max, avail_range.min, avail_range.max); | |
5d93da1d | 2666 | |
8657c838 | 2667 | if (res.min == 0 && maxbytes <= res.max) |
346114b5 JJ |
2668 | /* This is a special case to avoid issuing the potentially confusing |
2669 | warning: | |
2670 | writing 0 or more bytes into a region of size between 0 and N. */ | |
2671 | return fmtwarn (dirloc, argloc, NULL, info.warnopt (), | |
2672 | info.bounded | |
2673 | ? (maybe | |
2674 | ? G_("%<%.*s%> directive output may be truncated " | |
2675 | "writing likely %wu or more bytes into a region " | |
2676 | "of size between %wu and %wu") | |
2677 | : G_("%<%.*s%> directive output truncated writing " | |
2678 | "likely %wu or more bytes into a region of size " | |
2679 | "between %wu and %wu")) | |
2680 | : G_("%<%.*s%> directive writing likely %wu or more bytes " | |
2681 | "into a region of size between %wu and %wu"), | |
2682 | (int) dir.len, | |
2683 | target_to_host (hostdir, sizeof hostdir, dir.beg), | |
2684 | res.likely, avail_range.min, avail_range.max); | |
5d93da1d MS |
2685 | |
2686 | if (res.max < maxbytes) | |
346114b5 JJ |
2687 | return fmtwarn (dirloc, argloc, NULL, info.warnopt (), |
2688 | info.bounded | |
2689 | ? (maybe | |
2690 | ? G_("%<%.*s%> directive output may be truncated " | |
2691 | "writing between %wu and %wu bytes into a region " | |
2692 | "of size between %wu and %wu") | |
2693 | : G_("%<%.*s%> directive output truncated writing " | |
2694 | "between %wu and %wu bytes into a region of size " | |
2695 | "between %wu and %wu")) | |
2696 | : G_("%<%.*s%> directive writing between %wu and " | |
2697 | "%wu bytes into a region of size between %wu and " | |
2698 | "%wu"), (int) dir.len, | |
2699 | target_to_host (hostdir, sizeof hostdir, dir.beg), | |
2700 | res.min, res.max, avail_range.min, avail_range.max); | |
2701 | ||
2702 | return fmtwarn (dirloc, argloc, NULL, info.warnopt (), | |
2703 | info.bounded | |
2704 | ? (maybe | |
2705 | ? G_("%<%.*s%> directive output may be truncated writing " | |
2706 | "%wu or more bytes into a region of size between " | |
2707 | "%wu and %wu") | |
2708 | : G_("%<%.*s%> directive output truncated writing " | |
2709 | "%wu or more bytes into a region of size between " | |
2710 | "%wu and %wu")) | |
2711 | : G_("%<%.*s%> directive writing %wu or more bytes " | |
2712 | "into a region of size between %wu and %wu"), | |
2713 | (int) dir.len, | |
d127ae9f MS |
2714 | target_to_host (hostdir, sizeof hostdir, dir.beg), |
2715 | res.min, avail_range.min, avail_range.max); | |
02f1ef40 MS |
2716 | } |
2717 | ||
31c87a43 MS |
2718 | /* Compute the length of the output resulting from the directive DIR |
2719 | in a call described by INFO and update the overall result of the call | |
2720 | in *RES. Return true if the directive has been handled. */ | |
88d0c3f0 | 2721 | |
954b452a | 2722 | static bool |
a09ca93a | 2723 | format_directive (const sprintf_dom_walker::call_info &info, |
bbe2e114 JL |
2724 | format_result *res, const directive &dir, |
2725 | class vr_values *vr_values) | |
88d0c3f0 MS |
2726 | { |
2727 | /* Offset of the beginning of the directive from the beginning | |
2728 | of the format string. */ | |
31c87a43 MS |
2729 | size_t offset = dir.beg - info.fmtstr; |
2730 | size_t start = offset; | |
2731 | size_t length = offset + dir.len - !!dir.len; | |
88d0c3f0 MS |
2732 | |
2733 | /* Create a location for the whole directive from the % to the format | |
2734 | specifier. */ | |
2735 | substring_loc dirloc (info.fmtloc, TREE_TYPE (info.format), | |
31c87a43 | 2736 | offset, start, length); |
88d0c3f0 | 2737 | |
89b6abbb | 2738 | /* Also get the location of the argument if possible. |
88d0c3f0 | 2739 | This doesn't work for integer literals or function calls. */ |
89b6abbb DM |
2740 | location_t argloc = UNKNOWN_LOCATION; |
2741 | if (dir.arg) | |
2742 | argloc = EXPR_LOCATION (dir.arg); | |
88d0c3f0 MS |
2743 | |
2744 | /* Bail when there is no function to compute the output length, | |
2745 | or when minimum length checking has been disabled. */ | |
5d93da1d | 2746 | if (!dir.fmtfunc || res->range.min >= HOST_WIDE_INT_MAX) |
954b452a | 2747 | return false; |
88d0c3f0 | 2748 | |
31c87a43 | 2749 | /* Compute the range of lengths of the formatted output. */ |
bbe2e114 | 2750 | fmtresult fmtres = dir.fmtfunc (dir, dir.arg, vr_values); |
88d0c3f0 | 2751 | |
1eb4547b MS |
2752 | /* Record whether the output of all directives is known to be |
2753 | bounded by some maximum, implying that their arguments are | |
2754 | either known exactly or determined to be in a known range | |
2755 | or, for strings, limited by the upper bounds of the arrays | |
2756 | they refer to. */ | |
2757 | res->knownrange &= fmtres.knownrange; | |
88d0c3f0 | 2758 | |
1eb4547b | 2759 | if (!fmtres.knownrange) |
88d0c3f0 | 2760 | { |
1eb4547b | 2761 | /* Only when the range is known, check it against the host value |
cfce1a4a MS |
2762 | of INT_MAX + (the number of bytes of the "%.*Lf" directive with |
2763 | INT_MAX precision, which is the longest possible output of any | |
2764 | single directive). That's the largest valid byte count (though | |
2765 | not valid call to a printf-like function because it can never | |
2766 | return such a count). Otherwise, the range doesn't correspond | |
2767 | to known values of the argument. */ | |
2768 | if (fmtres.range.max > target_dir_max ()) | |
1eb4547b MS |
2769 | { |
2770 | /* Normalize the MAX counter to avoid having to deal with it | |
2771 | later. The counter can be less than HOST_WIDE_INT_M1U | |
2772 | when compiling for an ILP32 target on an LP64 host. */ | |
2773 | fmtres.range.max = HOST_WIDE_INT_M1U; | |
2774 | /* Disable exact and maximum length checking after a failure | |
2775 | to determine the maximum number of characters (for example | |
2776 | for wide characters or wide character strings) but continue | |
2777 | tracking the minimum number of characters. */ | |
5d93da1d | 2778 | res->range.max = HOST_WIDE_INT_M1U; |
1eb4547b MS |
2779 | } |
2780 | ||
cfce1a4a | 2781 | if (fmtres.range.min > target_dir_max ()) |
1eb4547b MS |
2782 | { |
2783 | /* Disable exact length checking after a failure to determine | |
2784 | even the minimum number of characters (it shouldn't happen | |
2785 | except in an error) but keep tracking the minimum and maximum | |
2786 | number of characters. */ | |
954b452a | 2787 | return true; |
1eb4547b | 2788 | } |
88d0c3f0 MS |
2789 | } |
2790 | ||
d127ae9f MS |
2791 | /* Buffer for the directive in the host character set (used when |
2792 | the source character set is different). */ | |
2793 | char hostdir[32]; | |
2794 | ||
5d93da1d MS |
2795 | int dirlen = dir.len; |
2796 | ||
2c8c83ac MS |
2797 | if (fmtres.nullp) |
2798 | { | |
89b6abbb | 2799 | fmtwarn (dirloc, argloc, NULL, info.warnopt (), |
91e3ec29 MS |
2800 | "%G%<%.*s%> directive argument is null", |
2801 | info.callstmt, dirlen, | |
2802 | target_to_host (hostdir, sizeof hostdir, dir.beg)); | |
2c8c83ac MS |
2803 | |
2804 | /* Don't bother processing the rest of the format string. */ | |
2805 | res->warned = true; | |
5d93da1d MS |
2806 | res->range.min = HOST_WIDE_INT_M1U; |
2807 | res->range.max = HOST_WIDE_INT_M1U; | |
954b452a | 2808 | return false; |
2c8c83ac MS |
2809 | } |
2810 | ||
88d0c3f0 MS |
2811 | /* Compute the number of available bytes in the destination. There |
2812 | must always be at least one byte of space for the terminating | |
2813 | NUL that's appended after the format string has been processed. */ | |
5d93da1d | 2814 | result_range avail_range = bytes_remaining (info.objsize, *res); |
88d0c3f0 | 2815 | |
02f1ef40 MS |
2816 | bool warned = res->warned; |
2817 | ||
2818 | if (!warned) | |
89b6abbb | 2819 | warned = maybe_warn (dirloc, argloc, info, avail_range, |
02f1ef40 MS |
2820 | fmtres.range, dir); |
2821 | ||
5d93da1d MS |
2822 | /* Bump up the total maximum if it isn't too big. */ |
2823 | if (res->range.max < HOST_WIDE_INT_MAX | |
2824 | && fmtres.range.max < HOST_WIDE_INT_MAX) | |
2825 | res->range.max += fmtres.range.max; | |
88d0c3f0 | 2826 | |
5d93da1d | 2827 | /* Raise the total unlikely maximum by the larger of the maximum |
bf00c9e0 MS |
2828 | and the unlikely maximum. */ |
2829 | unsigned HOST_WIDE_INT save = res->range.unlikely; | |
5d93da1d MS |
2830 | if (fmtres.range.max < fmtres.range.unlikely) |
2831 | res->range.unlikely += fmtres.range.unlikely; | |
88d0c3f0 | 2832 | else |
5d93da1d MS |
2833 | res->range.unlikely += fmtres.range.max; |
2834 | ||
bf00c9e0 MS |
2835 | if (res->range.unlikely < save) |
2836 | res->range.unlikely = HOST_WIDE_INT_M1U; | |
2837 | ||
5d93da1d MS |
2838 | res->range.min += fmtres.range.min; |
2839 | res->range.likely += fmtres.range.likely; | |
88d0c3f0 MS |
2840 | |
2841 | /* Has the minimum directive output length exceeded the maximum | |
2842 | of 4095 bytes required to be supported? */ | |
2843 | bool minunder4k = fmtres.range.min < 4096; | |
5d93da1d | 2844 | bool maxunder4k = fmtres.range.max < 4096; |
bbcbd744 MS |
2845 | /* Clear POSUNDER4K in the overall result if the maximum has exceeded |
2846 | the 4k (this is necessary to avoid the return value optimization | |
5d93da1d MS |
2847 | that may not be safe in the maximum case). */ |
2848 | if (!maxunder4k) | |
bbcbd744 MS |
2849 | res->posunder4k = false; |
2850 | /* Also clear POSUNDER4K if the directive may fail. */ | |
2851 | if (fmtres.mayfail) | |
2852 | res->posunder4k = false; | |
88d0c3f0 | 2853 | |
5d93da1d MS |
2854 | if (!warned |
2855 | /* Only warn at level 2. */ | |
01512446 | 2856 | && warn_level > 1 |
5d93da1d MS |
2857 | && (!minunder4k |
2858 | || (!maxunder4k && fmtres.range.max < HOST_WIDE_INT_MAX))) | |
88d0c3f0 MS |
2859 | { |
2860 | /* The directive output may be longer than the maximum required | |
2861 | to be handled by an implementation according to 7.21.6.1, p15 | |
2862 | of C11. Warn on this only at level 2 but remember this and | |
2863 | prevent folding the return value when done. This allows for | |
2864 | the possibility of the actual libc call failing due to ENOMEM | |
2865 | (like Glibc does under some conditions). */ | |
2866 | ||
2867 | if (fmtres.range.min == fmtres.range.max) | |
346114b5 | 2868 | warned = fmtwarn (dirloc, argloc, NULL, info.warnopt (), |
573aa7d4 | 2869 | "%<%.*s%> directive output of %wu bytes exceeds " |
346114b5 | 2870 | "minimum required size of 4095", dirlen, |
d127ae9f MS |
2871 | target_to_host (hostdir, sizeof hostdir, dir.beg), |
2872 | fmtres.range.min); | |
88d0c3f0 | 2873 | else |
346114b5 JJ |
2874 | warned = fmtwarn (dirloc, argloc, NULL, info.warnopt (), |
2875 | minunder4k | |
2876 | ? G_("%<%.*s%> directive output between %wu and %wu " | |
2877 | "bytes may exceed minimum required size of " | |
2878 | "4095") | |
2879 | : G_("%<%.*s%> directive output between %wu and %wu " | |
2880 | "bytes exceeds minimum required size of 4095"), | |
2881 | dirlen, | |
2882 | target_to_host (hostdir, sizeof hostdir, dir.beg), | |
2883 | fmtres.range.min, fmtres.range.max); | |
88d0c3f0 MS |
2884 | } |
2885 | ||
5d93da1d MS |
2886 | /* Has the likely and maximum directive output exceeded INT_MAX? */ |
2887 | bool likelyximax = *dir.beg && res->range.likely > target_int_max (); | |
bf00c9e0 MS |
2888 | /* Don't consider the maximum to be in excess when it's the result |
2889 | of a string of unknown length (i.e., whose maximum has been set | |
2890 | to be greater than or equal to HOST_WIDE_INT_MAX. */ | |
2891 | bool maxximax = (*dir.beg | |
2892 | && res->range.max > target_int_max () | |
2893 | && res->range.max < HOST_WIDE_INT_MAX); | |
88d0c3f0 | 2894 | |
573aa7d4 | 2895 | if (!warned |
5d93da1d MS |
2896 | /* Warn for the likely output size at level 1. */ |
2897 | && (likelyximax | |
2898 | /* But only warn for the maximum at level 2. */ | |
01512446 | 2899 | || (warn_level > 1 |
5d93da1d MS |
2900 | && maxximax |
2901 | && fmtres.range.max < HOST_WIDE_INT_MAX))) | |
88d0c3f0 MS |
2902 | { |
2903 | /* The directive output causes the total length of output | |
2904 | to exceed INT_MAX bytes. */ | |
2905 | ||
2906 | if (fmtres.range.min == fmtres.range.max) | |
89b6abbb | 2907 | warned = fmtwarn (dirloc, argloc, NULL, info.warnopt (), |
573aa7d4 | 2908 | "%<%.*s%> directive output of %wu bytes causes " |
346114b5 | 2909 | "result to exceed %<INT_MAX%>", dirlen, |
d127ae9f MS |
2910 | target_to_host (hostdir, sizeof hostdir, dir.beg), |
2911 | fmtres.range.min); | |
88d0c3f0 | 2912 | else |
346114b5 JJ |
2913 | warned = fmtwarn (dirloc, argloc, NULL, info.warnopt (), |
2914 | fmtres.range.min > target_int_max () | |
40659769 ML |
2915 | ? G_("%<%.*s%> directive output between %wu and " |
2916 | "%wu bytes causes result to exceed " | |
2917 | "%<INT_MAX%>") | |
2918 | : G_("%<%.*s%> directive output between %wu and " | |
2919 | "%wu bytes may cause result to exceed " | |
2920 | "%<INT_MAX%>"), dirlen, | |
346114b5 JJ |
2921 | target_to_host (hostdir, sizeof hostdir, dir.beg), |
2922 | fmtres.range.min, fmtres.range.max); | |
88d0c3f0 | 2923 | } |
573aa7d4 | 2924 | |
a41cc59d MS |
2925 | if (!warned && fmtres.nonstr) |
2926 | { | |
2927 | warned = fmtwarn (dirloc, argloc, NULL, info.warnopt (), | |
2928 | "%<%.*s%> directive argument is not a nul-terminated " | |
2929 | "string", | |
2930 | dirlen, | |
2931 | target_to_host (hostdir, sizeof hostdir, dir.beg)); | |
2932 | if (warned && DECL_P (fmtres.nonstr)) | |
2933 | inform (DECL_SOURCE_LOCATION (fmtres.nonstr), | |
2934 | "referenced argument declared here"); | |
2935 | return false; | |
2936 | } | |
2937 | ||
5d93da1d MS |
2938 | if (warned && fmtres.range.min < fmtres.range.likely |
2939 | && fmtres.range.likely < fmtres.range.max) | |
1c89478a | 2940 | inform_n (info.fmtloc, fmtres.range.likely, |
281cbf31 JJ |
2941 | "assuming directive output of %wu byte", |
2942 | "assuming directive output of %wu bytes", | |
5d93da1d | 2943 | fmtres.range.likely); |
5d93da1d | 2944 | |
573aa7d4 MS |
2945 | if (warned && fmtres.argmin) |
2946 | { | |
2947 | if (fmtres.argmin == fmtres.argmax) | |
2948 | inform (info.fmtloc, "directive argument %qE", fmtres.argmin); | |
2949 | else if (fmtres.knownrange) | |
2950 | inform (info.fmtloc, "directive argument in the range [%E, %E]", | |
2951 | fmtres.argmin, fmtres.argmax); | |
2952 | else | |
2953 | inform (info.fmtloc, | |
2954 | "using the range [%E, %E] for directive argument", | |
2955 | fmtres.argmin, fmtres.argmax); | |
2956 | } | |
2957 | ||
2958 | res->warned |= warned; | |
02f1ef40 | 2959 | |
5d93da1d | 2960 | if (!dir.beg[0] && res->warned && info.objsize < HOST_WIDE_INT_MAX) |
88d0c3f0 MS |
2961 | { |
2962 | /* If a warning has been issued for buffer overflow or truncation | |
2963 | (but not otherwise) help the user figure out how big a buffer | |
2964 | they need. */ | |
2965 | ||
2966 | location_t callloc = gimple_location (info.callstmt); | |
2967 | ||
5d93da1d MS |
2968 | unsigned HOST_WIDE_INT min = res->range.min; |
2969 | unsigned HOST_WIDE_INT max = res->range.max; | |
88d0c3f0 | 2970 | |
5d93da1d MS |
2971 | if (min == max) |
2972 | inform (callloc, | |
2973 | (min == 1 | |
2974 | ? G_("%qE output %wu byte into a destination of size %wu") | |
2975 | : G_("%qE output %wu bytes into a destination of size %wu")), | |
2976 | info.func, min, info.objsize); | |
2977 | else if (max < HOST_WIDE_INT_MAX) | |
88d0c3f0 | 2978 | inform (callloc, |
5d93da1d | 2979 | "%qE output between %wu and %wu bytes into " |
88d0c3f0 | 2980 | "a destination of size %wu", |
5d93da1d MS |
2981 | info.func, min, max, info.objsize); |
2982 | else if (min < res->range.likely && res->range.likely < max) | |
2983 | inform (callloc, | |
2984 | "%qE output %wu or more bytes (assuming %wu) into " | |
2985 | "a destination of size %wu", | |
2986 | info.func, min, res->range.likely, info.objsize); | |
88d0c3f0 MS |
2987 | else |
2988 | inform (callloc, | |
5d93da1d MS |
2989 | "%qE output %wu or more bytes into a destination of size %wu", |
2990 | info.func, min, info.objsize); | |
88d0c3f0 MS |
2991 | } |
2992 | ||
5d93da1d | 2993 | if (dump_file && *dir.beg) |
88d0c3f0 | 2994 | { |
a6fd2cd7 MS |
2995 | fprintf (dump_file, |
2996 | " Result: " | |
2997 | HOST_WIDE_INT_PRINT_DEC ", " HOST_WIDE_INT_PRINT_DEC ", " | |
2998 | HOST_WIDE_INT_PRINT_DEC ", " HOST_WIDE_INT_PRINT_DEC " (" | |
2999 | HOST_WIDE_INT_PRINT_DEC ", " HOST_WIDE_INT_PRINT_DEC ", " | |
3000 | HOST_WIDE_INT_PRINT_DEC ", " HOST_WIDE_INT_PRINT_DEC ")\n", | |
3001 | fmtres.range.min, fmtres.range.likely, | |
3002 | fmtres.range.max, fmtres.range.unlikely, | |
3003 | res->range.min, res->range.likely, | |
3004 | res->range.max, res->range.unlikely); | |
88d0c3f0 | 3005 | } |
5d93da1d MS |
3006 | |
3007 | return true; | |
88d0c3f0 MS |
3008 | } |
3009 | ||
954b452a MS |
3010 | /* Parse a format directive in function call described by INFO starting |
3011 | at STR and populate DIR structure. Bump up *ARGNO by the number of | |
3012 | arguments extracted for the directive. Return the length of | |
3013 | the directive. */ | |
88d0c3f0 | 3014 | |
954b452a | 3015 | static size_t |
a09ca93a | 3016 | parse_directive (sprintf_dom_walker::call_info &info, |
954b452a | 3017 | directive &dir, format_result *res, |
bbe2e114 JL |
3018 | const char *str, unsigned *argno, |
3019 | vr_values *vr_values) | |
88d0c3f0 | 3020 | { |
d127ae9f | 3021 | const char *pcnt = strchr (str, target_percent); |
954b452a | 3022 | dir.beg = str; |
88d0c3f0 | 3023 | |
954b452a MS |
3024 | if (size_t len = pcnt ? pcnt - str : *str ? strlen (str) : 1) |
3025 | { | |
3026 | /* This directive is either a plain string or the terminating nul | |
3027 | (which isn't really a directive but it simplifies things to | |
3028 | handle it as if it were). */ | |
3029 | dir.len = len; | |
5d93da1d | 3030 | dir.fmtfunc = format_plain; |
88d0c3f0 | 3031 | |
954b452a MS |
3032 | if (dump_file) |
3033 | { | |
a6fd2cd7 MS |
3034 | fprintf (dump_file, " Directive %u at offset " |
3035 | HOST_WIDE_INT_PRINT_UNSIGNED ": \"%.*s\", " | |
3036 | "length = " HOST_WIDE_INT_PRINT_UNSIGNED "\n", | |
9976a810 | 3037 | dir.dirno, |
a6fd2cd7 | 3038 | (unsigned HOST_WIDE_INT)(size_t)(dir.beg - info.fmtstr), |
a5a4a9df | 3039 | (int)dir.len, dir.beg, (unsigned HOST_WIDE_INT) dir.len); |
954b452a MS |
3040 | } |
3041 | ||
3042 | return len - !*str; | |
3043 | } | |
88d0c3f0 | 3044 | |
954b452a MS |
3045 | const char *pf = pcnt + 1; |
3046 | ||
3047 | /* POSIX numbered argument index or zero when none. */ | |
d127ae9f | 3048 | HOST_WIDE_INT dollar = 0; |
954b452a MS |
3049 | |
3050 | /* With and precision. -1 when not specified, HOST_WIDE_INT_MIN | |
3051 | when given by a va_list argument, and a non-negative value | |
3052 | when specified in the format string itself. */ | |
3053 | HOST_WIDE_INT width = -1; | |
3054 | HOST_WIDE_INT precision = -1; | |
88d0c3f0 | 3055 | |
d127ae9f MS |
3056 | /* Pointers to the beginning of the width and precision decimal |
3057 | string (if any) within the directive. */ | |
3058 | const char *pwidth = 0; | |
3059 | const char *pprec = 0; | |
3060 | ||
3061 | /* When the value of the decimal string that specifies width or | |
3062 | precision is out of range, points to the digit that causes | |
3063 | the value to exceed the limit. */ | |
3064 | const char *werange = NULL; | |
3065 | const char *perange = NULL; | |
3066 | ||
954b452a MS |
3067 | /* Width specified via the asterisk. Need not be INTEGER_CST. |
3068 | For vararg functions set to void_node. */ | |
3069 | tree star_width = NULL_TREE; | |
3070 | ||
3071 | /* Width specified via the asterisk. Need not be INTEGER_CST. | |
3072 | For vararg functions set to void_node. */ | |
3073 | tree star_precision = NULL_TREE; | |
3074 | ||
d127ae9f | 3075 | if (ISDIGIT (target_to_host (*pf))) |
88d0c3f0 | 3076 | { |
954b452a MS |
3077 | /* This could be either a POSIX positional argument, the '0' |
3078 | flag, or a width, depending on what follows. Store it as | |
3079 | width and sort it out later after the next character has | |
3080 | been seen. */ | |
d127ae9f MS |
3081 | pwidth = pf; |
3082 | width = target_strtol10 (&pf, &werange); | |
954b452a | 3083 | } |
d127ae9f | 3084 | else if (target_to_host (*pf) == '*') |
954b452a MS |
3085 | { |
3086 | /* Similarly to the block above, this could be either a POSIX | |
3087 | positional argument or a width, depending on what follows. */ | |
3088 | if (*argno < gimple_call_num_args (info.callstmt)) | |
3089 | star_width = gimple_call_arg (info.callstmt, (*argno)++); | |
3090 | else | |
3091 | star_width = void_node; | |
3092 | ++pf; | |
3093 | } | |
88d0c3f0 | 3094 | |
d127ae9f | 3095 | if (target_to_host (*pf) == '$') |
954b452a MS |
3096 | { |
3097 | /* Handle the POSIX dollar sign which references the 1-based | |
3098 | positional argument number. */ | |
3099 | if (width != -1) | |
3100 | dollar = width + info.argidx; | |
3101 | else if (star_width | |
e602bbfc MS |
3102 | && TREE_CODE (star_width) == INTEGER_CST |
3103 | && (TYPE_PRECISION (TREE_TYPE (star_width)) | |
3104 | <= TYPE_PRECISION (integer_type_node))) | |
954b452a | 3105 | dollar = width + tree_to_shwi (star_width); |
88d0c3f0 | 3106 | |
954b452a MS |
3107 | /* Bail when the numbered argument is out of range (it will |
3108 | have already been diagnosed by -Wformat). */ | |
3109 | if (dollar == 0 | |
d127ae9f | 3110 | || dollar == (int)info.argidx |
954b452a MS |
3111 | || dollar > gimple_call_num_args (info.callstmt)) |
3112 | return false; | |
3113 | ||
3114 | --dollar; | |
88d0c3f0 | 3115 | |
954b452a MS |
3116 | star_width = NULL_TREE; |
3117 | width = -1; | |
3118 | ++pf; | |
3119 | } | |
88d0c3f0 | 3120 | |
954b452a MS |
3121 | if (dollar || !star_width) |
3122 | { | |
3123 | if (width != -1) | |
88d0c3f0 | 3124 | { |
954b452a MS |
3125 | if (width == 0) |
3126 | { | |
3127 | /* The '0' that has been interpreted as a width above is | |
3128 | actually a flag. Reset HAVE_WIDTH, set the '0' flag, | |
3129 | and continue processing other flags. */ | |
3130 | width = -1; | |
3131 | dir.set_flag ('0'); | |
3132 | } | |
3133 | else if (!dollar) | |
3134 | { | |
3135 | /* (Non-zero) width has been seen. The next character | |
3136 | is either a period or a digit. */ | |
3137 | goto start_precision; | |
3138 | } | |
88d0c3f0 | 3139 | } |
954b452a MS |
3140 | /* When either '$' has been seen, or width has not been seen, |
3141 | the next field is the optional flags followed by an optional | |
3142 | width. */ | |
3143 | for ( ; ; ) { | |
d127ae9f | 3144 | switch (target_to_host (*pf)) |
954b452a MS |
3145 | { |
3146 | case ' ': | |
3147 | case '0': | |
3148 | case '+': | |
3149 | case '-': | |
3150 | case '#': | |
d127ae9f | 3151 | dir.set_flag (target_to_host (*pf++)); |
954b452a MS |
3152 | break; |
3153 | ||
3154 | default: | |
3155 | goto start_width; | |
3156 | } | |
3157 | } | |
88d0c3f0 | 3158 | |
954b452a | 3159 | start_width: |
d127ae9f | 3160 | if (ISDIGIT (target_to_host (*pf))) |
88d0c3f0 | 3161 | { |
d127ae9f MS |
3162 | werange = 0; |
3163 | pwidth = pf; | |
3164 | width = target_strtol10 (&pf, &werange); | |
88d0c3f0 | 3165 | } |
d127ae9f | 3166 | else if (target_to_host (*pf) == '*') |
88d0c3f0 | 3167 | { |
954b452a MS |
3168 | if (*argno < gimple_call_num_args (info.callstmt)) |
3169 | star_width = gimple_call_arg (info.callstmt, (*argno)++); | |
5b4f088d | 3170 | else |
954b452a MS |
3171 | { |
3172 | /* This is (likely) a va_list. It could also be an invalid | |
3173 | call with insufficient arguments. */ | |
3174 | star_width = void_node; | |
3175 | } | |
88d0c3f0 MS |
3176 | ++pf; |
3177 | } | |
d127ae9f | 3178 | else if (target_to_host (*pf) == '\'') |
88d0c3f0 | 3179 | { |
954b452a MS |
3180 | /* The POSIX apostrophe indicating a numeric grouping |
3181 | in the current locale. Even though it's possible to | |
3182 | estimate the upper bound on the size of the output | |
3183 | based on the number of digits it probably isn't worth | |
3184 | continuing. */ | |
3185 | return 0; | |
3186 | } | |
3187 | } | |
88d0c3f0 | 3188 | |
954b452a | 3189 | start_precision: |
d127ae9f | 3190 | if (target_to_host (*pf) == '.') |
954b452a MS |
3191 | { |
3192 | ++pf; | |
88d0c3f0 | 3193 | |
d127ae9f | 3194 | if (ISDIGIT (target_to_host (*pf))) |
954b452a | 3195 | { |
d127ae9f MS |
3196 | pprec = pf; |
3197 | precision = target_strtol10 (&pf, &perange); | |
88d0c3f0 | 3198 | } |
d127ae9f | 3199 | else if (target_to_host (*pf) == '*') |
88d0c3f0 | 3200 | { |
954b452a MS |
3201 | if (*argno < gimple_call_num_args (info.callstmt)) |
3202 | star_precision = gimple_call_arg (info.callstmt, (*argno)++); | |
3203 | else | |
88d0c3f0 | 3204 | { |
954b452a MS |
3205 | /* This is (likely) a va_list. It could also be an invalid |
3206 | call with insufficient arguments. */ | |
3207 | star_precision = void_node; | |
88d0c3f0 | 3208 | } |
954b452a | 3209 | ++pf; |
88d0c3f0 | 3210 | } |
954b452a MS |
3211 | else |
3212 | { | |
3213 | /* The decimal precision or the asterisk are optional. | |
3214 | When neither is dirified it's taken to be zero. */ | |
3215 | precision = 0; | |
3216 | } | |
3217 | } | |
88d0c3f0 | 3218 | |
d127ae9f | 3219 | switch (target_to_host (*pf)) |
954b452a MS |
3220 | { |
3221 | case 'h': | |
d127ae9f | 3222 | if (target_to_host (pf[1]) == 'h') |
88d0c3f0 MS |
3223 | { |
3224 | ++pf; | |
954b452a | 3225 | dir.modifier = FMT_LEN_hh; |
88d0c3f0 | 3226 | } |
954b452a MS |
3227 | else |
3228 | dir.modifier = FMT_LEN_h; | |
3229 | ++pf; | |
3230 | break; | |
3231 | ||
3232 | case 'j': | |
3233 | dir.modifier = FMT_LEN_j; | |
3234 | ++pf; | |
3235 | break; | |
88d0c3f0 | 3236 | |
954b452a MS |
3237 | case 'L': |
3238 | dir.modifier = FMT_LEN_L; | |
3239 | ++pf; | |
3240 | break; | |
3241 | ||
3242 | case 'l': | |
d127ae9f | 3243 | if (target_to_host (pf[1]) == 'l') |
88d0c3f0 | 3244 | { |
88d0c3f0 | 3245 | ++pf; |
954b452a MS |
3246 | dir.modifier = FMT_LEN_ll; |
3247 | } | |
3248 | else | |
3249 | dir.modifier = FMT_LEN_l; | |
3250 | ++pf; | |
3251 | break; | |
88d0c3f0 | 3252 | |
954b452a MS |
3253 | case 't': |
3254 | dir.modifier = FMT_LEN_t; | |
3255 | ++pf; | |
3256 | break; | |
88d0c3f0 | 3257 | |
954b452a MS |
3258 | case 'z': |
3259 | dir.modifier = FMT_LEN_z; | |
3260 | ++pf; | |
3261 | break; | |
3262 | } | |
88d0c3f0 | 3263 | |
d127ae9f | 3264 | switch (target_to_host (*pf)) |
954b452a MS |
3265 | { |
3266 | /* Handle a sole '%' character the same as "%%" but since it's | |
3267 | undefined prevent the result from being folded. */ | |
3268 | case '\0': | |
3269 | --pf; | |
5d93da1d | 3270 | res->range.min = res->range.max = HOST_WIDE_INT_M1U; |
954b452a MS |
3271 | /* FALLTHRU */ |
3272 | case '%': | |
3273 | dir.fmtfunc = format_percent; | |
3274 | break; | |
88d0c3f0 | 3275 | |
954b452a MS |
3276 | case 'a': |
3277 | case 'A': | |
3278 | case 'e': | |
3279 | case 'E': | |
3280 | case 'f': | |
3281 | case 'F': | |
3282 | case 'g': | |
3283 | case 'G': | |
3284 | res->floating = true; | |
3285 | dir.fmtfunc = format_floating; | |
3286 | break; | |
88d0c3f0 | 3287 | |
954b452a MS |
3288 | case 'd': |
3289 | case 'i': | |
3290 | case 'o': | |
3291 | case 'u': | |
3292 | case 'x': | |
3293 | case 'X': | |
3294 | dir.fmtfunc = format_integer; | |
3295 | break; | |
3296 | ||
3297 | case 'p': | |
3298 | /* The %p output is implementation-defined. It's possible | |
3299 | to determine this format but due to extensions (edirially | |
3300 | those of the Linux kernel -- see bug 78512) the first %p | |
3301 | in the format string disables any further processing. */ | |
3302 | return false; | |
3303 | ||
3304 | case 'n': | |
3305 | /* %n has side-effects even when nothing is actually printed to | |
3306 | any buffer. */ | |
3307 | info.nowrite = false; | |
3308 | dir.fmtfunc = format_none; | |
3309 | break; | |
3310 | ||
bbcbd744 | 3311 | case 'C': |
954b452a | 3312 | case 'c': |
bbcbd744 | 3313 | /* POSIX wide character and C/POSIX narrow character. */ |
954b452a MS |
3314 | dir.fmtfunc = format_character; |
3315 | break; | |
3316 | ||
3317 | case 'S': | |
3318 | case 's': | |
bbcbd744 | 3319 | /* POSIX wide string and C/POSIX narrow character string. */ |
954b452a MS |
3320 | dir.fmtfunc = format_string; |
3321 | break; | |
3322 | ||
3323 | default: | |
3324 | /* Unknown conversion specification. */ | |
3325 | return 0; | |
3326 | } | |
3327 | ||
d127ae9f MS |
3328 | dir.specifier = target_to_host (*pf++); |
3329 | ||
3330 | /* Store the length of the format directive. */ | |
3331 | dir.len = pf - pcnt; | |
3332 | ||
3333 | /* Buffer for the directive in the host character set (used when | |
3334 | the source character set is different). */ | |
3335 | char hostdir[32]; | |
954b452a MS |
3336 | |
3337 | if (star_width) | |
3338 | { | |
3812f460 | 3339 | if (INTEGRAL_TYPE_P (TREE_TYPE (star_width))) |
bbe2e114 | 3340 | dir.set_width (star_width, vr_values); |
954b452a MS |
3341 | else |
3342 | { | |
3343 | /* Width specified by a va_list takes on the range [0, -INT_MIN] | |
3344 | (width is the absolute value of that specified). */ | |
31c87a43 MS |
3345 | dir.width[0] = 0; |
3346 | dir.width[1] = target_int_max () + 1; | |
88d0c3f0 | 3347 | } |
954b452a MS |
3348 | } |
3349 | else | |
d127ae9f MS |
3350 | { |
3351 | if (width == LONG_MAX && werange) | |
3352 | { | |
3353 | size_t begin = dir.beg - info.fmtstr + (pwidth - pcnt); | |
3354 | size_t caret = begin + (werange - pcnt); | |
3355 | size_t end = pf - info.fmtstr - 1; | |
3356 | ||
3357 | /* Create a location for the width part of the directive, | |
3358 | pointing the caret at the first out-of-range digit. */ | |
3359 | substring_loc dirloc (info.fmtloc, TREE_TYPE (info.format), | |
3360 | caret, begin, end); | |
3361 | ||
346114b5 JJ |
3362 | fmtwarn (dirloc, UNKNOWN_LOCATION, NULL, info.warnopt (), |
3363 | "%<%.*s%> directive width out of range", (int) dir.len, | |
3364 | target_to_host (hostdir, sizeof hostdir, dir.beg)); | |
d127ae9f MS |
3365 | } |
3366 | ||
3367 | dir.set_width (width); | |
3368 | } | |
88d0c3f0 | 3369 | |
954b452a MS |
3370 | if (star_precision) |
3371 | { | |
3812f460 | 3372 | if (INTEGRAL_TYPE_P (TREE_TYPE (star_precision))) |
bbe2e114 | 3373 | dir.set_precision (star_precision, vr_values); |
954b452a | 3374 | else |
88d0c3f0 | 3375 | { |
954b452a MS |
3376 | /* Precision specified by a va_list takes on the range [-1, INT_MAX] |
3377 | (unlike width, negative precision is ignored). */ | |
31c87a43 MS |
3378 | dir.prec[0] = -1; |
3379 | dir.prec[1] = target_int_max (); | |
954b452a MS |
3380 | } |
3381 | } | |
3382 | else | |
d127ae9f MS |
3383 | { |
3384 | if (precision == LONG_MAX && perange) | |
3385 | { | |
3386 | size_t begin = dir.beg - info.fmtstr + (pprec - pcnt) - 1; | |
3387 | size_t caret = dir.beg - info.fmtstr + (perange - pcnt) - 1; | |
3388 | size_t end = pf - info.fmtstr - 2; | |
3389 | ||
3390 | /* Create a location for the precision part of the directive, | |
3391 | including the leading period, pointing the caret at the first | |
3392 | out-of-range digit . */ | |
3393 | substring_loc dirloc (info.fmtloc, TREE_TYPE (info.format), | |
3394 | caret, begin, end); | |
3395 | ||
346114b5 JJ |
3396 | fmtwarn (dirloc, UNKNOWN_LOCATION, NULL, info.warnopt (), |
3397 | "%<%.*s%> directive precision out of range", (int) dir.len, | |
3398 | target_to_host (hostdir, sizeof hostdir, dir.beg)); | |
d127ae9f MS |
3399 | } |
3400 | ||
3401 | dir.set_precision (precision); | |
3402 | } | |
88d0c3f0 | 3403 | |
954b452a MS |
3404 | /* Extract the argument if the directive takes one and if it's |
3405 | available (e.g., the function doesn't take a va_list). Treat | |
3406 | missing arguments the same as va_list, even though they will | |
3407 | have likely already been diagnosed by -Wformat. */ | |
3408 | if (dir.specifier != '%' | |
3409 | && *argno < gimple_call_num_args (info.callstmt)) | |
3410 | dir.arg = gimple_call_arg (info.callstmt, dollar ? dollar : (*argno)++); | |
88d0c3f0 | 3411 | |
954b452a MS |
3412 | if (dump_file) |
3413 | { | |
a6fd2cd7 MS |
3414 | fprintf (dump_file, |
3415 | " Directive %u at offset " HOST_WIDE_INT_PRINT_UNSIGNED | |
3416 | ": \"%.*s\"", | |
3417 | dir.dirno, | |
3418 | (unsigned HOST_WIDE_INT)(size_t)(dir.beg - info.fmtstr), | |
954b452a MS |
3419 | (int)dir.len, dir.beg); |
3420 | if (star_width) | |
31c87a43 MS |
3421 | { |
3422 | if (dir.width[0] == dir.width[1]) | |
a6fd2cd7 MS |
3423 | fprintf (dump_file, ", width = " HOST_WIDE_INT_PRINT_DEC, |
3424 | dir.width[0]); | |
31c87a43 | 3425 | else |
a6fd2cd7 MS |
3426 | fprintf (dump_file, |
3427 | ", width in range [" HOST_WIDE_INT_PRINT_DEC | |
3428 | ", " HOST_WIDE_INT_PRINT_DEC "]", | |
3429 | dir.width[0], dir.width[1]); | |
31c87a43 | 3430 | } |
88d0c3f0 | 3431 | |
954b452a | 3432 | if (star_precision) |
31c87a43 MS |
3433 | { |
3434 | if (dir.prec[0] == dir.prec[1]) | |
a6fd2cd7 MS |
3435 | fprintf (dump_file, ", precision = " HOST_WIDE_INT_PRINT_DEC, |
3436 | dir.prec[0]); | |
31c87a43 | 3437 | else |
a6fd2cd7 MS |
3438 | fprintf (dump_file, |
3439 | ", precision in range [" HOST_WIDE_INT_PRINT_DEC | |
3440 | HOST_WIDE_INT_PRINT_DEC "]", | |
3441 | dir.prec[0], dir.prec[1]); | |
31c87a43 | 3442 | } |
954b452a MS |
3443 | fputc ('\n', dump_file); |
3444 | } | |
3445 | ||
3446 | return dir.len; | |
3447 | } | |
3448 | ||
3449 | /* Compute the length of the output resulting from the call to a formatted | |
3450 | output function described by INFO and store the result of the call in | |
3451 | *RES. Issue warnings for detected past the end writes. Return true | |
3452 | if the complete format string has been processed and *RES can be relied | |
3453 | on, false otherwise (e.g., when a unknown or unhandled directive was seen | |
3454 | that caused the processing to be terminated early). */ | |
3455 | ||
3456 | bool | |
a09ca93a | 3457 | sprintf_dom_walker::compute_format_length (call_info &info, |
aa67d03c | 3458 | format_result *res) |
954b452a | 3459 | { |
02f1ef40 MS |
3460 | if (dump_file) |
3461 | { | |
3462 | location_t callloc = gimple_location (info.callstmt); | |
3463 | fprintf (dump_file, "%s:%i: ", | |
3464 | LOCATION_FILE (callloc), LOCATION_LINE (callloc)); | |
3465 | print_generic_expr (dump_file, info.func, dump_flags); | |
3466 | ||
a6fd2cd7 MS |
3467 | fprintf (dump_file, |
3468 | ": objsize = " HOST_WIDE_INT_PRINT_UNSIGNED | |
3469 | ", fmtstr = \"%s\"\n", | |
3470 | info.objsize, info.fmtstr); | |
02f1ef40 MS |
3471 | } |
3472 | ||
31c87a43 | 3473 | /* Reset the minimum and maximum byte counters. */ |
5d93da1d | 3474 | res->range.min = res->range.max = 0; |
954b452a MS |
3475 | |
3476 | /* No directive has been seen yet so the length of output is bounded | |
bbcbd744 MS |
3477 | by the known range [0, 0] (with no conversion resulting in a failure |
3478 | or producing more than 4K bytes) until determined otherwise. */ | |
954b452a | 3479 | res->knownrange = true; |
954b452a MS |
3480 | res->floating = false; |
3481 | res->warned = false; | |
3482 | ||
3483 | /* 1-based directive counter. */ | |
3484 | unsigned dirno = 1; | |
3485 | ||
3486 | /* The variadic argument counter. */ | |
3487 | unsigned argno = info.argidx; | |
3488 | ||
3489 | for (const char *pf = info.fmtstr; ; ++dirno) | |
3490 | { | |
3491 | directive dir = directive (); | |
3492 | dir.dirno = dirno; | |
88d0c3f0 | 3493 | |
bbe2e114 JL |
3494 | size_t n = parse_directive (info, dir, res, pf, &argno, |
3495 | evrp_range_analyzer.get_vr_values ()); | |
88d0c3f0 | 3496 | |
5d93da1d | 3497 | /* Return failure if the format function fails. */ |
bbe2e114 JL |
3498 | if (!format_directive (info, res, dir, |
3499 | evrp_range_analyzer.get_vr_values ())) | |
5d93da1d | 3500 | return false; |
88d0c3f0 | 3501 | |
954b452a MS |
3502 | /* Return success the directive is zero bytes long and it's |
3503 | the last think in the format string (i.e., it's the terminating | |
3504 | nul, which isn't really a directive but handling it as one makes | |
3505 | things simpler). */ | |
3506 | if (!n) | |
3507 | return *pf == '\0'; | |
88d0c3f0 | 3508 | |
954b452a | 3509 | pf += n; |
88d0c3f0 | 3510 | } |
01a4551c | 3511 | |
5d93da1d | 3512 | /* The complete format string was processed (with or without warnings). */ |
01a4551c | 3513 | return true; |
88d0c3f0 MS |
3514 | } |
3515 | ||
3516 | /* Return the size of the object referenced by the expression DEST if | |
3517 | available, or -1 otherwise. */ | |
3518 | ||
3519 | static unsigned HOST_WIDE_INT | |
3520 | get_destination_size (tree dest) | |
3521 | { | |
91e3ec29 MS |
3522 | /* When there is no destination return -1. */ |
3523 | if (!dest) | |
3524 | return HOST_WIDE_INT_M1U; | |
3525 | ||
02f1ef40 MS |
3526 | /* Initialize object size info before trying to compute it. */ |
3527 | init_object_sizes (); | |
3528 | ||
88d0c3f0 MS |
3529 | /* Use __builtin_object_size to determine the size of the destination |
3530 | object. When optimizing, determine the smallest object (such as | |
3531 | a member array as opposed to the whole enclosing object), otherwise | |
3532 | use type-zero object size to determine the size of the enclosing | |
3533 | object (the function fails without optimization in this type). */ | |
91eaca5e | 3534 | int ost = optimize > 0; |
88d0c3f0 MS |
3535 | unsigned HOST_WIDE_INT size; |
3536 | if (compute_builtin_object_size (dest, ost, &size)) | |
3537 | return size; | |
3538 | ||
3539 | return HOST_WIDE_INT_M1U; | |
3540 | } | |
3541 | ||
a104bd88 MS |
3542 | /* Return true if the call described by INFO with result RES safe to |
3543 | optimize (i.e., no undefined behavior), and set RETVAL to the range | |
3544 | of its return values. */ | |
88d0c3f0 | 3545 | |
3eefa646 | 3546 | static bool |
a09ca93a | 3547 | is_call_safe (const sprintf_dom_walker::call_info &info, |
a104bd88 MS |
3548 | const format_result &res, bool under4k, |
3549 | unsigned HOST_WIDE_INT retval[2]) | |
88d0c3f0 | 3550 | { |
bbcbd744 | 3551 | if (under4k && !res.posunder4k) |
a104bd88 | 3552 | return false; |
a151e93b | 3553 | |
5d93da1d | 3554 | /* The minimum return value. */ |
a104bd88 | 3555 | retval[0] = res.range.min; |
a151e93b | 3556 | |
5d93da1d MS |
3557 | /* The maximum return value is in most cases bounded by RES.RANGE.MAX |
3558 | but in cases involving multibyte characters could be as large as | |
3559 | RES.RANGE.UNLIKELY. */ | |
a104bd88 | 3560 | retval[1] |
5d93da1d | 3561 | = res.range.unlikely < res.range.max ? res.range.max : res.range.unlikely; |
a151e93b MS |
3562 | |
3563 | /* Adjust the number of bytes which includes the terminating nul | |
3564 | to reflect the return value of the function which does not. | |
3565 | Because the valid range of the function is [INT_MIN, INT_MAX], | |
3566 | a valid range before the adjustment below is [0, INT_MAX + 1] | |
3567 | (the functions only return negative values on error or undefined | |
3568 | behavior). */ | |
a104bd88 MS |
3569 | if (retval[0] <= target_int_max () + 1) |
3570 | --retval[0]; | |
3571 | if (retval[1] <= target_int_max () + 1) | |
3572 | --retval[1]; | |
a151e93b | 3573 | |
88d0c3f0 MS |
3574 | /* Avoid the return value optimization when the behavior of the call |
3575 | is undefined either because any directive may have produced 4K or | |
3576 | more of output, or the return value exceeds INT_MAX, or because | |
3577 | the output overflows the destination object (but leave it enabled | |
3578 | when the function is bounded because then the behavior is well- | |
3579 | defined). */ | |
a104bd88 MS |
3580 | if (retval[0] == retval[1] |
3581 | && (info.bounded || retval[0] < info.objsize) | |
3582 | && retval[0] <= target_int_max ()) | |
3583 | return true; | |
3584 | ||
3585 | if ((info.bounded || retval[1] < info.objsize) | |
3586 | && (retval[0] < target_int_max () | |
3587 | && retval[1] < target_int_max ())) | |
3588 | return true; | |
3589 | ||
3590 | if (!under4k && (info.bounded || retval[0] < info.objsize)) | |
3591 | return true; | |
3592 | ||
3593 | return false; | |
3594 | } | |
3595 | ||
3596 | /* Given a suitable result RES of a call to a formatted output function | |
3597 | described by INFO, substitute the result for the return value of | |
3598 | the call. The result is suitable if the number of bytes it represents | |
3599 | is known and exact. A result that isn't suitable for substitution may | |
3600 | have its range set to the range of return values, if that is known. | |
3601 | Return true if the call is removed and gsi_next should not be performed | |
3602 | in the caller. */ | |
3603 | ||
3604 | static bool | |
3605 | try_substitute_return_value (gimple_stmt_iterator *gsi, | |
a09ca93a | 3606 | const sprintf_dom_walker::call_info &info, |
a104bd88 MS |
3607 | const format_result &res) |
3608 | { | |
3609 | tree lhs = gimple_get_lhs (info.callstmt); | |
3610 | ||
3611 | /* Set to true when the entire call has been removed. */ | |
3612 | bool removed = false; | |
3613 | ||
3614 | /* The minimum and maximum return value. */ | |
3615 | unsigned HOST_WIDE_INT retval[2]; | |
3616 | bool safe = is_call_safe (info, res, true, retval); | |
3617 | ||
3618 | if (safe | |
3619 | && retval[0] == retval[1] | |
cd123354 JJ |
3620 | /* Not prepared to handle possibly throwing calls here; they shouldn't |
3621 | appear in non-artificial testcases, except when the __*_chk routines | |
3622 | are badly declared. */ | |
3623 | && !stmt_ends_bb_p (info.callstmt)) | |
88d0c3f0 | 3624 | { |
a104bd88 | 3625 | tree cst = build_int_cst (integer_type_node, retval[0]); |
d7b0fcaa | 3626 | |
a151e93b MS |
3627 | if (lhs == NULL_TREE |
3628 | && info.nowrite) | |
3629 | { | |
3630 | /* Remove the call to the bounded function with a zero size | |
3631 | (e.g., snprintf(0, 0, "%i", 123)) if there is no lhs. */ | |
3632 | unlink_stmt_vdef (info.callstmt); | |
3633 | gsi_remove (gsi, true); | |
3634 | removed = true; | |
3635 | } | |
3636 | else if (info.nowrite) | |
d7b0fcaa MS |
3637 | { |
3638 | /* Replace the call to the bounded function with a zero size | |
3639 | (e.g., snprintf(0, 0, "%i", 123) with the constant result | |
a151e93b | 3640 | of the function. */ |
d7b0fcaa MS |
3641 | if (!update_call_from_tree (gsi, cst)) |
3642 | gimplify_and_update_call_from_tree (gsi, cst); | |
3643 | gimple *callstmt = gsi_stmt (*gsi); | |
3644 | update_stmt (callstmt); | |
3645 | } | |
a151e93b | 3646 | else if (lhs) |
d7b0fcaa MS |
3647 | { |
3648 | /* Replace the left-hand side of the call with the constant | |
a151e93b | 3649 | result of the formatted function. */ |
d7b0fcaa MS |
3650 | gimple_call_set_lhs (info.callstmt, NULL_TREE); |
3651 | gimple *g = gimple_build_assign (lhs, cst); | |
3652 | gsi_insert_after (gsi, g, GSI_NEW_STMT); | |
3653 | update_stmt (info.callstmt); | |
3654 | } | |
88d0c3f0 MS |
3655 | |
3656 | if (dump_file) | |
3657 | { | |
a151e93b MS |
3658 | if (removed) |
3659 | fprintf (dump_file, " Removing call statement."); | |
3660 | else | |
3661 | { | |
3662 | fprintf (dump_file, " Substituting "); | |
3663 | print_generic_expr (dump_file, cst, dump_flags); | |
3664 | fprintf (dump_file, " for %s.\n", | |
3665 | info.nowrite ? "statement" : "return value"); | |
3666 | } | |
88d0c3f0 MS |
3667 | } |
3668 | } | |
a151e93b | 3669 | else if (lhs) |
3eefa646 | 3670 | { |
a151e93b | 3671 | bool setrange = false; |
88d0c3f0 | 3672 | |
a104bd88 MS |
3673 | if (safe |
3674 | && (info.bounded || retval[1] < info.objsize) | |
3675 | && (retval[0] < target_int_max () | |
3676 | && retval[1] < target_int_max ())) | |
88d0c3f0 MS |
3677 | { |
3678 | /* If the result is in a valid range bounded by the size of | |
3679 | the destination set it so that it can be used for subsequent | |
3680 | optimizations. */ | |
3681 | int prec = TYPE_PRECISION (integer_type_node); | |
3682 | ||
a104bd88 MS |
3683 | wide_int min = wi::shwi (retval[0], prec); |
3684 | wide_int max = wi::shwi (retval[1], prec); | |
a151e93b MS |
3685 | set_range_info (lhs, VR_RANGE, min, max); |
3686 | ||
3687 | setrange = true; | |
88d0c3f0 MS |
3688 | } |
3689 | ||
3690 | if (dump_file) | |
3691 | { | |
3692 | const char *inbounds | |
a104bd88 MS |
3693 | = (retval[0] < info.objsize |
3694 | ? (retval[1] < info.objsize | |
88d0c3f0 MS |
3695 | ? "in" : "potentially out-of") |
3696 | : "out-of"); | |
3697 | ||
a151e93b | 3698 | const char *what = setrange ? "Setting" : "Discarding"; |
a104bd88 | 3699 | if (retval[0] != retval[1]) |
88d0c3f0 | 3700 | fprintf (dump_file, |
a6fd2cd7 MS |
3701 | " %s %s-bounds return value range [" |
3702 | HOST_WIDE_INT_PRINT_UNSIGNED ", " | |
3703 | HOST_WIDE_INT_PRINT_UNSIGNED "].\n", | |
3704 | what, inbounds, retval[0], retval[1]); | |
88d0c3f0 | 3705 | else |
a6fd2cd7 MS |
3706 | fprintf (dump_file, " %s %s-bounds return value " |
3707 | HOST_WIDE_INT_PRINT_UNSIGNED ".\n", | |
3708 | what, inbounds, retval[0]); | |
88d0c3f0 MS |
3709 | } |
3710 | } | |
3eefa646 | 3711 | |
02f1ef40 MS |
3712 | if (dump_file) |
3713 | fputc ('\n', dump_file); | |
3714 | ||
3715 | return removed; | |
88d0c3f0 MS |
3716 | } |
3717 | ||
a104bd88 MS |
3718 | /* Try to simplify a s{,n}printf call described by INFO with result |
3719 | RES by replacing it with a simpler and presumably more efficient | |
3720 | call (such as strcpy). */ | |
3721 | ||
3722 | static bool | |
3723 | try_simplify_call (gimple_stmt_iterator *gsi, | |
a09ca93a | 3724 | const sprintf_dom_walker::call_info &info, |
a104bd88 MS |
3725 | const format_result &res) |
3726 | { | |
3727 | unsigned HOST_WIDE_INT dummy[2]; | |
3728 | if (!is_call_safe (info, res, info.retval_used (), dummy)) | |
3729 | return false; | |
3730 | ||
3731 | switch (info.fncode) | |
3732 | { | |
3733 | case BUILT_IN_SNPRINTF: | |
3734 | return gimple_fold_builtin_snprintf (gsi); | |
3735 | ||
3736 | case BUILT_IN_SPRINTF: | |
3737 | return gimple_fold_builtin_sprintf (gsi); | |
3738 | ||
3739 | default: | |
3740 | ; | |
3741 | } | |
3742 | ||
3743 | return false; | |
3744 | } | |
3745 | ||
91e3ec29 MS |
3746 | /* Return the zero-based index of the format string argument of a printf |
3747 | like function and set *IDX_ARGS to the first format argument. When | |
3748 | no such index exists return UINT_MAX. */ | |
3749 | ||
3750 | static unsigned | |
3751 | get_user_idx_format (tree fndecl, unsigned *idx_args) | |
3752 | { | |
3753 | tree attrs = lookup_attribute ("format", DECL_ATTRIBUTES (fndecl)); | |
3754 | if (!attrs) | |
3755 | attrs = lookup_attribute ("format", TYPE_ATTRIBUTES (TREE_TYPE (fndecl))); | |
3756 | ||
3757 | if (!attrs) | |
3758 | return UINT_MAX; | |
3759 | ||
3760 | attrs = TREE_VALUE (attrs); | |
3761 | ||
3762 | tree archetype = TREE_VALUE (attrs); | |
3763 | if (strcmp ("printf", IDENTIFIER_POINTER (archetype))) | |
3764 | return UINT_MAX; | |
3765 | ||
3766 | attrs = TREE_CHAIN (attrs); | |
3767 | tree fmtarg = TREE_VALUE (attrs); | |
3768 | ||
3769 | attrs = TREE_CHAIN (attrs); | |
3770 | tree elliparg = TREE_VALUE (attrs); | |
3771 | ||
3772 | /* Attribute argument indices are 1-based but we use zero-based. */ | |
3773 | *idx_args = tree_to_uhwi (elliparg) - 1; | |
3774 | return tree_to_uhwi (fmtarg) - 1; | |
3775 | } | |
3776 | ||
88d0c3f0 | 3777 | /* Determine if a GIMPLE CALL is to one of the sprintf-like built-in |
3eefa646 JJ |
3778 | functions and if so, handle it. Return true if the call is removed |
3779 | and gsi_next should not be performed in the caller. */ | |
88d0c3f0 | 3780 | |
3eefa646 | 3781 | bool |
a09ca93a | 3782 | sprintf_dom_walker::handle_gimple_call (gimple_stmt_iterator *gsi) |
88d0c3f0 MS |
3783 | { |
3784 | call_info info = call_info (); | |
3785 | ||
d7b0fcaa | 3786 | info.callstmt = gsi_stmt (*gsi); |
91e3ec29 MS |
3787 | info.func = gimple_call_fndecl (info.callstmt); |
3788 | if (!info.func) | |
3eefa646 | 3789 | return false; |
88d0c3f0 MS |
3790 | |
3791 | info.fncode = DECL_FUNCTION_CODE (info.func); | |
3792 | ||
91e3ec29 MS |
3793 | /* Format string argument number (valid for all functions). */ |
3794 | unsigned idx_format = UINT_MAX; | |
3795 | if (!gimple_call_builtin_p (info.callstmt, BUILT_IN_NORMAL)) | |
3796 | { | |
3797 | unsigned idx_args; | |
3798 | idx_format = get_user_idx_format (info.func, &idx_args); | |
3799 | if (idx_format == UINT_MAX) | |
3800 | return false; | |
3801 | info.argidx = idx_args; | |
3802 | } | |
3803 | ||
88d0c3f0 MS |
3804 | /* The size of the destination as in snprintf(dest, size, ...). */ |
3805 | unsigned HOST_WIDE_INT dstsize = HOST_WIDE_INT_M1U; | |
3806 | ||
3807 | /* The size of the destination determined by __builtin_object_size. */ | |
3808 | unsigned HOST_WIDE_INT objsize = HOST_WIDE_INT_M1U; | |
3809 | ||
91e3ec29 MS |
3810 | /* Zero-based buffer size argument number (snprintf and vsnprintf). */ |
3811 | unsigned idx_dstsize = UINT_MAX; | |
88d0c3f0 MS |
3812 | |
3813 | /* Object size argument number (snprintf_chk and vsnprintf_chk). */ | |
91e3ec29 | 3814 | unsigned idx_objsize = UINT_MAX; |
88d0c3f0 | 3815 | |
91e3ec29 MS |
3816 | /* Destinaton argument number (valid for sprintf functions only). */ |
3817 | unsigned idx_dstptr = 0; | |
88d0c3f0 MS |
3818 | |
3819 | switch (info.fncode) | |
3820 | { | |
91e3ec29 MS |
3821 | case BUILT_IN_NONE: |
3822 | // User-defined function with attribute format (printf). | |
3823 | idx_dstptr = -1; | |
3824 | break; | |
3825 | ||
3826 | case BUILT_IN_FPRINTF: | |
3827 | // Signature: | |
3828 | // __builtin_fprintf (FILE*, format, ...) | |
3829 | idx_format = 1; | |
3830 | info.argidx = 2; | |
3831 | idx_dstptr = -1; | |
3832 | break; | |
3833 | ||
3834 | case BUILT_IN_FPRINTF_CHK: | |
3835 | // Signature: | |
3836 | // __builtin_fprintf_chk (FILE*, ost, format, ...) | |
3837 | idx_format = 2; | |
3838 | info.argidx = 3; | |
3839 | idx_dstptr = -1; | |
3840 | break; | |
3841 | ||
3842 | case BUILT_IN_FPRINTF_UNLOCKED: | |
3843 | // Signature: | |
3844 | // __builtin_fprintf_unnlocked (FILE*, format, ...) | |
3845 | idx_format = 1; | |
3846 | info.argidx = 2; | |
3847 | idx_dstptr = -1; | |
3848 | break; | |
3849 | ||
3850 | case BUILT_IN_PRINTF: | |
3851 | // Signature: | |
3852 | // __builtin_printf (format, ...) | |
3853 | idx_format = 0; | |
3854 | info.argidx = 1; | |
3855 | idx_dstptr = -1; | |
3856 | break; | |
3857 | ||
3858 | case BUILT_IN_PRINTF_CHK: | |
3859 | // Signature: | |
3860 | // __builtin_printf_chk (it, format, ...) | |
3861 | idx_format = 1; | |
3862 | info.argidx = 2; | |
3863 | idx_dstptr = -1; | |
3864 | break; | |
3865 | ||
3866 | case BUILT_IN_PRINTF_UNLOCKED: | |
3867 | // Signature: | |
3868 | // __builtin_printf (format, ...) | |
3869 | idx_format = 0; | |
3870 | info.argidx = 1; | |
3871 | idx_dstptr = -1; | |
3872 | break; | |
3873 | ||
88d0c3f0 MS |
3874 | case BUILT_IN_SPRINTF: |
3875 | // Signature: | |
3876 | // __builtin_sprintf (dst, format, ...) | |
3877 | idx_format = 1; | |
3878 | info.argidx = 2; | |
3879 | break; | |
3880 | ||
5fe39ec0 JJ |
3881 | case BUILT_IN_SPRINTF_CHK: |
3882 | // Signature: | |
3883 | // __builtin___sprintf_chk (dst, ost, objsize, format, ...) | |
3884 | idx_objsize = 2; | |
3885 | idx_format = 3; | |
3886 | info.argidx = 4; | |
3887 | break; | |
3888 | ||
88d0c3f0 MS |
3889 | case BUILT_IN_SNPRINTF: |
3890 | // Signature: | |
3891 | // __builtin_snprintf (dst, size, format, ...) | |
3892 | idx_dstsize = 1; | |
3893 | idx_format = 2; | |
3894 | info.argidx = 3; | |
3895 | info.bounded = true; | |
3896 | break; | |
3897 | ||
3898 | case BUILT_IN_SNPRINTF_CHK: | |
3899 | // Signature: | |
5fe39ec0 | 3900 | // __builtin___snprintf_chk (dst, size, ost, objsize, format, ...) |
88d0c3f0 MS |
3901 | idx_dstsize = 1; |
3902 | idx_objsize = 3; | |
3903 | idx_format = 4; | |
3904 | info.argidx = 5; | |
3905 | info.bounded = true; | |
3906 | break; | |
3907 | ||
91e3ec29 MS |
3908 | case BUILT_IN_VFPRINTF: |
3909 | // Signature: | |
3910 | // __builtin_vprintf (FILE*, format, va_list) | |
3911 | idx_format = 1; | |
3912 | info.argidx = -1; | |
3913 | idx_dstptr = -1; | |
3914 | break; | |
3915 | ||
3916 | case BUILT_IN_VFPRINTF_CHK: | |
3917 | // Signature: | |
3918 | // __builtin___vfprintf_chk (FILE*, ost, format, va_list) | |
3919 | idx_format = 2; | |
3920 | info.argidx = -1; | |
3921 | idx_dstptr = -1; | |
3922 | break; | |
3923 | ||
3924 | case BUILT_IN_VPRINTF: | |
3925 | // Signature: | |
3926 | // __builtin_vprintf (format, va_list) | |
3927 | idx_format = 0; | |
3928 | info.argidx = -1; | |
3929 | idx_dstptr = -1; | |
3930 | break; | |
3931 | ||
3932 | case BUILT_IN_VPRINTF_CHK: | |
3933 | // Signature: | |
3934 | // __builtin___vprintf_chk (ost, format, va_list) | |
3935 | idx_format = 1; | |
3936 | info.argidx = -1; | |
3937 | idx_dstptr = -1; | |
3938 | break; | |
3939 | ||
88d0c3f0 MS |
3940 | case BUILT_IN_VSNPRINTF: |
3941 | // Signature: | |
3942 | // __builtin_vsprintf (dst, size, format, va) | |
3943 | idx_dstsize = 1; | |
3944 | idx_format = 2; | |
3945 | info.argidx = -1; | |
3946 | info.bounded = true; | |
3947 | break; | |
3948 | ||
3949 | case BUILT_IN_VSNPRINTF_CHK: | |
3950 | // Signature: | |
3951 | // __builtin___vsnprintf_chk (dst, size, ost, objsize, format, va) | |
3952 | idx_dstsize = 1; | |
2b1b5938 MS |
3953 | idx_objsize = 3; |
3954 | idx_format = 4; | |
88d0c3f0 MS |
3955 | info.argidx = -1; |
3956 | info.bounded = true; | |
3957 | break; | |
3958 | ||
3959 | case BUILT_IN_VSPRINTF: | |
3960 | // Signature: | |
3961 | // __builtin_vsprintf (dst, format, va) | |
3962 | idx_format = 1; | |
3963 | info.argidx = -1; | |
3964 | break; | |
3965 | ||
3966 | case BUILT_IN_VSPRINTF_CHK: | |
3967 | // Signature: | |
3968 | // __builtin___vsprintf_chk (dst, ost, objsize, format, va) | |
3969 | idx_format = 3; | |
3970 | idx_objsize = 2; | |
3971 | info.argidx = -1; | |
3972 | break; | |
3973 | ||
3974 | default: | |
3eefa646 | 3975 | return false; |
88d0c3f0 MS |
3976 | } |
3977 | ||
02f1ef40 MS |
3978 | /* Set the global warning level for this function. */ |
3979 | warn_level = info.bounded ? warn_format_trunc : warn_format_overflow; | |
3980 | ||
91e3ec29 MS |
3981 | /* For all string functions the first argument is a pointer to |
3982 | the destination. */ | |
3983 | tree dstptr = (idx_dstptr < gimple_call_num_args (info.callstmt) | |
3984 | ? gimple_call_arg (info.callstmt, 0) : NULL_TREE); | |
2c8c83ac | 3985 | |
88d0c3f0 MS |
3986 | info.format = gimple_call_arg (info.callstmt, idx_format); |
3987 | ||
ee75687b MS |
3988 | /* True when the destination size is constant as opposed to the lower |
3989 | or upper bound of a range. */ | |
3990 | bool dstsize_cst_p = true; | |
3991 | ||
91e3ec29 | 3992 | if (idx_dstsize == UINT_MAX) |
88d0c3f0 | 3993 | { |
de6aa933 MS |
3994 | /* For non-bounded functions like sprintf, determine the size |
3995 | of the destination from the object or pointer passed to it | |
3996 | as the first argument. */ | |
2c8c83ac | 3997 | dstsize = get_destination_size (dstptr); |
88d0c3f0 MS |
3998 | } |
3999 | else if (tree size = gimple_call_arg (info.callstmt, idx_dstsize)) | |
4000 | { | |
4001 | /* For bounded functions try to get the size argument. */ | |
4002 | ||
4003 | if (TREE_CODE (size) == INTEGER_CST) | |
4004 | { | |
4005 | dstsize = tree_to_uhwi (size); | |
1eb4547b | 4006 | /* No object can be larger than SIZE_MAX bytes (half the address |
ee92e7ba | 4007 | space) on the target. |
de6aa933 MS |
4008 | The functions are defined only for output of at most INT_MAX |
4009 | bytes. Specifying a bound in excess of that limit effectively | |
4010 | defeats the bounds checking (and on some implementations such | |
4011 | as Solaris cause the function to fail with EINVAL). */ | |
ee92e7ba MS |
4012 | if (dstsize > target_size_max () / 2) |
4013 | { | |
4014 | /* Avoid warning if -Wstringop-overflow is specified since | |
4015 | it also warns for the same thing though only for the | |
4016 | checking built-ins. */ | |
91e3ec29 | 4017 | if ((idx_objsize == UINT_MAX |
ee92e7ba | 4018 | || !warn_stringop_overflow)) |
efcc8d38 | 4019 | warning_at (gimple_location (info.callstmt), info.warnopt (), |
ee92e7ba MS |
4020 | "specified bound %wu exceeds maximum object size " |
4021 | "%wu", | |
4022 | dstsize, target_size_max () / 2); | |
4023 | } | |
de6aa933 | 4024 | else if (dstsize > target_int_max ()) |
efcc8d38 | 4025 | warning_at (gimple_location (info.callstmt), info.warnopt (), |
3a66f913 | 4026 | "specified bound %wu exceeds %<INT_MAX%>", |
88d0c3f0 MS |
4027 | dstsize); |
4028 | } | |
4029 | else if (TREE_CODE (size) == SSA_NAME) | |
4030 | { | |
4031 | /* Try to determine the range of values of the argument | |
ee75687b MS |
4032 | and use the greater of the two at level 1 and the smaller |
4033 | of them at level 2. */ | |
c3684506 | 4034 | value_range *vr = evrp_range_analyzer.get_value_range (size); |
54994253 | 4035 | if (range_int_cst_p (vr)) |
c3684506 | 4036 | dstsize = (warn_level < 2 |
54994253 AH |
4037 | ? TREE_INT_CST_LOW (vr->max ()) |
4038 | : TREE_INT_CST_LOW (vr->min ())); | |
ee75687b MS |
4039 | |
4040 | /* The destination size is not constant. If the function is | |
4041 | bounded (e.g., snprintf) a lower bound of zero doesn't | |
4042 | necessarily imply it can be eliminated. */ | |
4043 | dstsize_cst_p = false; | |
88d0c3f0 MS |
4044 | } |
4045 | } | |
4046 | ||
91e3ec29 | 4047 | if (idx_objsize != UINT_MAX) |
3eefa646 JJ |
4048 | if (tree size = gimple_call_arg (info.callstmt, idx_objsize)) |
4049 | if (tree_fits_uhwi_p (size)) | |
4050 | objsize = tree_to_uhwi (size); | |
88d0c3f0 MS |
4051 | |
4052 | if (info.bounded && !dstsize) | |
4053 | { | |
4054 | /* As a special case, when the explicitly specified destination | |
4055 | size argument (to a bounded function like snprintf) is zero | |
4056 | it is a request to determine the number of bytes on output | |
4057 | without actually producing any. Pretend the size is | |
4058 | unlimited in this case. */ | |
4059 | info.objsize = HOST_WIDE_INT_MAX; | |
ee75687b | 4060 | info.nowrite = dstsize_cst_p; |
88d0c3f0 MS |
4061 | } |
4062 | else | |
4063 | { | |
2c8c83ac MS |
4064 | /* For calls to non-bounded functions or to those of bounded |
4065 | functions with a non-zero size, warn if the destination | |
4066 | pointer is null. */ | |
91e3ec29 | 4067 | if (dstptr && integer_zerop (dstptr)) |
2c8c83ac MS |
4068 | { |
4069 | /* This is diagnosed with -Wformat only when the null is a constant | |
4070 | pointer. The warning here diagnoses instances where the pointer | |
4071 | is not constant. */ | |
4072 | location_t loc = gimple_location (info.callstmt); | |
4073 | warning_at (EXPR_LOC_OR_LOC (dstptr, loc), | |
91e3ec29 MS |
4074 | info.warnopt (), "%Gnull destination pointer", |
4075 | info.callstmt); | |
3eefa646 | 4076 | return false; |
2c8c83ac MS |
4077 | } |
4078 | ||
88d0c3f0 MS |
4079 | /* Set the object size to the smaller of the two arguments |
4080 | of both have been specified and they're not equal. */ | |
4081 | info.objsize = dstsize < objsize ? dstsize : objsize; | |
4082 | ||
4083 | if (info.bounded | |
ee92e7ba MS |
4084 | && dstsize < target_size_max () / 2 && objsize < dstsize |
4085 | /* Avoid warning if -Wstringop-overflow is specified since | |
4086 | it also warns for the same thing though only for the | |
4087 | checking built-ins. */ | |
91e3ec29 | 4088 | && (idx_objsize == UINT_MAX |
ee92e7ba | 4089 | || !warn_stringop_overflow)) |
88d0c3f0 | 4090 | { |
efcc8d38 | 4091 | warning_at (gimple_location (info.callstmt), info.warnopt (), |
ee92e7ba | 4092 | "specified bound %wu exceeds the size %wu " |
88d0c3f0 MS |
4093 | "of the destination object", dstsize, objsize); |
4094 | } | |
4095 | } | |
4096 | ||
91e3ec29 MS |
4097 | /* Determine if the format argument may be null and warn if not |
4098 | and if the argument is null. */ | |
4099 | if (integer_zerop (info.format) | |
4100 | && gimple_call_builtin_p (info.callstmt, BUILT_IN_NORMAL)) | |
88d0c3f0 | 4101 | { |
2c8c83ac MS |
4102 | location_t loc = gimple_location (info.callstmt); |
4103 | warning_at (EXPR_LOC_OR_LOC (info.format, loc), | |
91e3ec29 MS |
4104 | info.warnopt (), "%Gnull format string", |
4105 | info.callstmt); | |
3eefa646 | 4106 | return false; |
88d0c3f0 MS |
4107 | } |
4108 | ||
4109 | info.fmtstr = get_format_string (info.format, &info.fmtloc); | |
4110 | if (!info.fmtstr) | |
3eefa646 | 4111 | return false; |
88d0c3f0 MS |
4112 | |
4113 | /* The result is the number of bytes output by the formatted function, | |
4114 | including the terminating NUL. */ | |
4115 | format_result res = format_result (); | |
01a4551c | 4116 | |
91e3ec29 MS |
4117 | /* I/O functions with no destination argument (i.e., all forms of fprintf |
4118 | and printf) may fail under any conditions. Others (i.e., all forms of | |
4119 | sprintf) may only fail under specific conditions determined for each | |
4120 | directive. Clear POSUNDER4K for the former set of functions and set | |
4121 | it to true for the latter (it can only be cleared later, but it is | |
4122 | never set to true again). */ | |
4123 | res.posunder4k = dstptr; | |
4124 | ||
01a4551c | 4125 | bool success = compute_format_length (info, &res); |
a41cc59d MS |
4126 | if (res.warned) |
4127 | gimple_set_no_warning (info.callstmt, true); | |
88d0c3f0 MS |
4128 | |
4129 | /* When optimizing and the printf return value optimization is enabled, | |
4130 | attempt to substitute the computed result for the return value of | |
4131 | the call. Avoid this optimization when -frounding-math is in effect | |
4132 | and the format string contains a floating point directive. */ | |
a104bd88 MS |
4133 | bool call_removed = false; |
4134 | if (success && optimize > 0) | |
4135 | { | |
4136 | /* Save a copy of the iterator pointing at the call. The iterator | |
4137 | may change to point past the call in try_substitute_return_value | |
4138 | but the original value is needed in try_simplify_call. */ | |
4139 | gimple_stmt_iterator gsi_call = *gsi; | |
02f1ef40 | 4140 | |
a104bd88 MS |
4141 | if (flag_printf_return_value |
4142 | && (!flag_rounding_math || !res.floating)) | |
4143 | call_removed = try_substitute_return_value (gsi, info, res); | |
4144 | ||
4145 | if (!call_removed) | |
4146 | try_simplify_call (&gsi_call, info, res); | |
4147 | } | |
4148 | ||
4149 | return call_removed; | |
88d0c3f0 MS |
4150 | } |
4151 | ||
a09ca93a JL |
4152 | edge |
4153 | sprintf_dom_walker::before_dom_children (basic_block bb) | |
4154 | { | |
aa67d03c | 4155 | evrp_range_analyzer.enter (bb); |
a09ca93a JL |
4156 | for (gimple_stmt_iterator si = gsi_start_bb (bb); !gsi_end_p (si); ) |
4157 | { | |
4158 | /* Iterate over statements, looking for function calls. */ | |
4159 | gimple *stmt = gsi_stmt (si); | |
4160 | ||
aa67d03c JL |
4161 | /* First record ranges generated by this statement. */ |
4162 | evrp_range_analyzer.record_ranges_from_stmt (stmt, false); | |
4163 | ||
a09ca93a JL |
4164 | if (is_gimple_call (stmt) && handle_gimple_call (&si)) |
4165 | /* If handle_gimple_call returns true, the iterator is | |
4166 | already pointing to the next statement. */ | |
4167 | continue; | |
4168 | ||
4169 | gsi_next (&si); | |
4170 | } | |
4171 | return NULL; | |
4172 | } | |
4173 | ||
aa67d03c JL |
4174 | void |
4175 | sprintf_dom_walker::after_dom_children (basic_block bb) | |
4176 | { | |
4177 | evrp_range_analyzer.leave (bb); | |
4178 | } | |
4179 | ||
88d0c3f0 MS |
4180 | /* Execute the pass for function FUN. */ |
4181 | ||
4182 | unsigned int | |
4183 | pass_sprintf_length::execute (function *fun) | |
4184 | { | |
d127ae9f MS |
4185 | init_target_to_host_charmap (); |
4186 | ||
a09ca93a | 4187 | calculate_dominance_info (CDI_DOMINATORS); |
3eefa646 | 4188 | |
a09ca93a JL |
4189 | sprintf_dom_walker sprintf_dom_walker; |
4190 | sprintf_dom_walker.walk (ENTRY_BLOCK_PTR_FOR_FN (fun)); | |
88d0c3f0 | 4191 | |
02f1ef40 | 4192 | /* Clean up object size info. */ |
eb07c7cf | 4193 | fini_object_sizes (); |
88d0c3f0 MS |
4194 | return 0; |
4195 | } | |
4196 | ||
4197 | } /* Unnamed namespace. */ | |
4198 | ||
4199 | /* Return a pointer to a pass object newly constructed from the context | |
4200 | CTXT. */ | |
4201 | ||
4202 | gimple_opt_pass * | |
4203 | make_pass_sprintf_length (gcc::context *ctxt) | |
4204 | { | |
4205 | return new pass_sprintf_length (ctxt); | |
4206 | } |