OpenMP: Update documentation of metadirective implementation status.
[gcc.git] / gcc / gimple-ssa-sprintf.cc
blobbcec540524222394df6421d436777eb1c912303d
1 /* Copyright (C) 2016-2025 Free Software Foundation, Inc.
2 Contributed by Martin Sebor <msebor@redhat.com>.
4 This file is part of GCC.
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.
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.
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/>. */
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.
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 rather 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.
38 The pass handles all forms standard sprintf format directives,
39 including character, integer, floating point, pointer, and strings,
40 with the standard C flags, widths, and precisions. For integers
41 and strings it computes the length of output itself. For floating
42 point it uses MPFR to format 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. */
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-iterator.h"
57 #include "gimple-fold.h"
58 #include "gimple-pretty-print.h"
59 #include "diagnostic-core.h"
60 #include "fold-const.h"
61 #include "tree-ssa.h"
62 #include "tree-object-size.h"
63 #include "tree-cfg.h"
64 #include "tree-ssa-propagate.h"
65 #include "calls.h"
66 #include "cfgloop.h"
67 #include "tree-scalar-evolution.h"
68 #include "tree-ssa-loop.h"
69 #include "intl.h"
70 #include "langhooks.h"
72 #include "attribs.h"
73 #include "builtins.h"
74 #include "pointer-query.h"
75 #include "stor-layout.h"
77 #include "realmpfr.h"
78 #include "target.h"
80 #include "cpplib.h"
81 #include "input.h"
82 #include "toplev.h"
83 #include "substring-locations.h"
84 #include "diagnostic.h"
85 #include "domwalk.h"
86 #include "alloc-pool.h"
87 #include "vr-values.h"
88 #include "tree-ssa-strlen.h"
89 #include "tree-dfa.h"
91 /* The likely worst case value of MB_LEN_MAX for the target, large enough
92 for UTF-8. Ideally, this would be obtained by a target hook if it were
93 to be used for optimization but it's good enough as is for warnings. */
94 #define target_mb_len_max() 6
96 /* The maximum number of bytes a single non-string directive can result
97 in. This is the result of printf("%.*Lf", INT_MAX, -LDBL_MAX) for
98 LDBL_MAX_10_EXP of 4932. */
99 #define IEEE_MAX_10_EXP 4932
100 #define target_dir_max() (target_int_max () + IEEE_MAX_10_EXP + 2)
102 namespace {
104 /* Set to the warning level for the current function which is equal
105 either to warn_format_trunc for bounded functions or to
106 warn_format_overflow otherwise. */
108 static int warn_level;
110 /* The minimum, maximum, likely, and unlikely maximum number of bytes
111 of output either a formatting function or an individual directive
112 can result in. */
114 struct result_range
116 /* The absolute minimum number of bytes. The result of a successful
117 conversion is guaranteed to be no less than this. (An erroneous
118 conversion can be indicated by MIN > HOST_WIDE_INT_MAX.) */
119 unsigned HOST_WIDE_INT min;
120 /* The likely maximum result that is used in diagnostics. In most
121 cases MAX is the same as the worst case UNLIKELY result. */
122 unsigned HOST_WIDE_INT max;
123 /* The likely result used to trigger diagnostics. For conversions
124 that result in a range of bytes [MIN, MAX], LIKELY is somewhere
125 in that range. */
126 unsigned HOST_WIDE_INT likely;
127 /* In rare cases (e.g., for multibyte characters) UNLIKELY gives
128 the worst cases maximum result of a directive. In most cases
129 UNLIKELY == MAX. UNLIKELY is used to control the return value
130 optimization but not in diagnostics. */
131 unsigned HOST_WIDE_INT unlikely;
134 /* Return the value of INT_MIN for the target. */
136 static inline HOST_WIDE_INT
137 target_int_min ()
139 return tree_to_shwi (TYPE_MIN_VALUE (integer_type_node));
142 /* Return the value of INT_MAX for the target. */
144 static inline unsigned HOST_WIDE_INT
145 target_int_max ()
147 return tree_to_uhwi (TYPE_MAX_VALUE (integer_type_node));
150 /* Return the value of SIZE_MAX for the target. */
152 static inline unsigned HOST_WIDE_INT
153 target_size_max ()
155 return tree_to_uhwi (TYPE_MAX_VALUE (size_type_node));
158 /* A straightforward mapping from the execution character set to the host
159 character set indexed by execution character. */
161 static char target_to_host_charmap[256];
163 /* Initialize a mapping from the execution character set to the host
164 character set. */
166 static bool
167 init_target_to_host_charmap ()
169 /* If the percent sign is non-zero the mapping has already been
170 initialized. */
171 if (target_to_host_charmap['%'])
172 return true;
174 /* Initialize the target_percent character (done elsewhere). */
175 if (!init_target_chars ())
176 return false;
178 /* The subset of the source character set used by printf conversion
179 specifications (strictly speaking, not all letters are used but
180 they are included here for the sake of simplicity). The dollar
181 sign must be included even though it's not in the basic source
182 character set. */
183 const char srcset[] = " 0123456789!\"#%&'()*+,-./:;<=>?[\\]^_{|}~$"
184 "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz";
186 /* Set the mapping for all characters to some ordinary value (i,e.,
187 not none used in printf conversion specifications) and overwrite
188 those that are used by conversion specifications with their
189 corresponding values. */
190 memset (target_to_host_charmap + 1, '?', sizeof target_to_host_charmap - 1);
192 /* Are the two sets of characters the same? */
193 bool all_same_p = true;
195 for (const char *pc = srcset; *pc; ++pc)
197 /* Slice off the high end bits in case target characters are
198 signed. All values are expected to be non-nul, otherwise
199 there's a problem. */
200 if (unsigned char tc = lang_hooks.to_target_charset (*pc))
202 target_to_host_charmap[tc] = *pc;
203 if (tc != *pc)
204 all_same_p = false;
206 else
207 return false;
211 /* Set the first element to a non-zero value if the mapping
212 is 1-to-1, otherwise leave it clear (NUL is assumed to be
213 the same in both character sets). */
214 target_to_host_charmap[0] = all_same_p;
216 return true;
219 /* Return the host source character corresponding to the character
220 CH in the execution character set if one exists, or some innocuous
221 (non-special, non-nul) source character otherwise. */
223 static inline unsigned char
224 target_to_host (unsigned char ch)
226 return target_to_host_charmap[ch];
229 /* Convert an initial substring of the string TARGSTR consisting of
230 characters in the execution character set into a string in the
231 source character set on the host and store up to HOSTSZ characters
232 in the buffer pointed to by HOSTR. Return HOSTR. */
234 static const char*
235 target_to_host (char *hostr, size_t hostsz, const char *targstr)
237 /* Make sure the buffer is reasonably big. */
238 gcc_assert (hostsz > 4);
240 /* The interesting subset of source and execution characters are
241 the same so no conversion is necessary. However, truncate
242 overlong strings just like the translated strings are. */
243 if (target_to_host_charmap['\0'] == 1)
245 size_t len = strlen (targstr);
246 if (len >= hostsz)
248 memcpy (hostr, targstr, hostsz - 4);
249 strcpy (hostr + hostsz - 4, "...");
251 else
252 memcpy (hostr, targstr, len + 1);
253 return hostr;
256 /* Convert the initial substring of TARGSTR to the corresponding
257 characters in the host set, appending "..." if TARGSTR is too
258 long to fit. Using the static buffer assumes the function is
259 not called in between sequence points (which it isn't). */
260 for (char *ph = hostr; ; ++targstr)
262 *ph++ = target_to_host (*targstr);
263 if (!*targstr)
264 break;
266 if (size_t (ph - hostr) == hostsz)
268 strcpy (ph - 4, "...");
269 break;
273 return hostr;
276 /* Convert the sequence of decimal digits in the execution character
277 starting at *PS to a HOST_WIDE_INT, analogously to strtol. Return
278 the result and set *PS to one past the last converted character.
279 On range error set ERANGE to the digit that caused it. */
281 static inline HOST_WIDE_INT
282 target_strtowi (const char **ps, const char **erange)
284 unsigned HOST_WIDE_INT val = 0;
285 for ( ; ; ++*ps)
287 unsigned char c = target_to_host (**ps);
288 if (ISDIGIT (c))
290 c -= '0';
292 /* Check for overflow. */
293 if (val > ((unsigned HOST_WIDE_INT) HOST_WIDE_INT_MAX - c) / 10LU)
295 val = HOST_WIDE_INT_MAX;
296 *erange = *ps;
298 /* Skip the remaining digits. */
300 c = target_to_host (*++*ps);
301 while (ISDIGIT (c));
302 break;
304 else
305 val = val * 10 + c;
307 else
308 break;
311 return val;
314 /* Given FORMAT, set *PLOC to the source location of the format string
315 and return the format string if it is known or null otherwise. */
317 static const char*
318 get_format_string (tree format, location_t *ploc)
320 *ploc = EXPR_LOC_OR_LOC (format, input_location);
322 return c_getstr (format);
325 /* For convenience and brevity, shorter named entrypoints of
326 format_string_diagnostic_t::emit_warning_va and
327 format_string_diagnostic_t::emit_warning_n_va.
328 These have to be functions with the attribute so that exgettext
329 works properly. */
331 static bool
332 ATTRIBUTE_GCC_DIAG (5, 6)
333 fmtwarn (const substring_loc &fmt_loc, location_t param_loc,
334 const char *corrected_substring, opt_code opt,
335 const char *gmsgid, ...)
337 format_string_diagnostic_t diag (fmt_loc, NULL, param_loc, NULL,
338 corrected_substring);
339 va_list ap;
340 va_start (ap, gmsgid);
341 bool warned = diag.emit_warning_va (opt, gmsgid, &ap);
342 va_end (ap);
344 return warned;
347 static bool
348 ATTRIBUTE_GCC_DIAG (6, 8) ATTRIBUTE_GCC_DIAG (7, 8)
349 fmtwarn_n (const substring_loc &fmt_loc, location_t param_loc,
350 const char *corrected_substring, opt_code opt,
351 unsigned HOST_WIDE_INT n,
352 const char *singular_gmsgid, const char *plural_gmsgid, ...)
354 format_string_diagnostic_t diag (fmt_loc, NULL, param_loc, NULL,
355 corrected_substring);
356 va_list ap;
357 va_start (ap, plural_gmsgid);
358 bool warned = diag.emit_warning_n_va (opt, n, singular_gmsgid, plural_gmsgid,
359 &ap);
360 va_end (ap);
362 return warned;
365 /* Format length modifiers. */
367 enum format_lengths
369 FMT_LEN_none,
370 FMT_LEN_hh, // char argument
371 FMT_LEN_h, // short
372 FMT_LEN_l, // long
373 FMT_LEN_ll, // long long
374 FMT_LEN_L, // long double (and GNU long long)
375 FMT_LEN_z, // size_t
376 FMT_LEN_t, // ptrdiff_t
377 FMT_LEN_j // intmax_t
381 /* Description of the result of conversion either of a single directive
382 or the whole format string. */
384 class fmtresult
386 public:
387 /* Construct a FMTRESULT object with all counters initialized
388 to MIN. KNOWNRANGE is set when MIN is valid. */
389 fmtresult (unsigned HOST_WIDE_INT min = HOST_WIDE_INT_MAX)
390 : argmin (), argmax (), dst_offset (HOST_WIDE_INT_MIN), nonstr (),
391 knownrange (min < HOST_WIDE_INT_MAX),
392 mayfail (), nullp ()
394 range.min = min;
395 range.max = min;
396 range.likely = min;
397 range.unlikely = min;
400 /* Construct a FMTRESULT object with MIN, MAX, and LIKELY counters.
401 KNOWNRANGE is set when both MIN and MAX are valid. */
402 fmtresult (unsigned HOST_WIDE_INT min, unsigned HOST_WIDE_INT max,
403 unsigned HOST_WIDE_INT likely = HOST_WIDE_INT_MAX)
404 : argmin (), argmax (), dst_offset (HOST_WIDE_INT_MIN), nonstr (),
405 knownrange (min < HOST_WIDE_INT_MAX && max < HOST_WIDE_INT_MAX),
406 mayfail (), nullp ()
408 range.min = min;
409 range.max = max;
410 range.likely = max < likely ? min : likely;
411 range.unlikely = max;
414 /* Adjust result upward to reflect the RANGE of values the specified
415 width or precision is known to be in. */
416 fmtresult& adjust_for_width_or_precision (const HOST_WIDE_INT[2],
417 tree = NULL_TREE,
418 unsigned = 0, unsigned = 0);
420 /* Return the maximum number of decimal digits a value of TYPE
421 formats as on output. */
422 static unsigned type_max_digits (tree, int);
424 /* The range a directive's argument is in. */
425 tree argmin, argmax;
427 /* The starting offset into the destination of the formatted function
428 call of the %s argument that points into (aliases with) the same
429 destination array. */
430 HOST_WIDE_INT dst_offset;
432 /* The minimum and maximum number of bytes that a directive
433 results in on output for an argument in the range above. */
434 result_range range;
436 /* Non-nul when the argument of a string directive is not a nul
437 terminated string. */
438 tree nonstr;
440 /* True when the range above is obtained from a known value of
441 a directive's argument or its bounds and not the result of
442 heuristics that depend on warning levels. */
443 bool knownrange;
445 /* True for a directive that may fail (such as wide character
446 directives). */
447 bool mayfail;
449 /* True when the argument is a null pointer. */
450 bool nullp;
453 /* Adjust result upward to reflect the range ADJUST of values the
454 specified width or precision is known to be in. When non-null,
455 TYPE denotes the type of the directive whose result is being
456 adjusted, BASE gives the base of the directive (octal, decimal,
457 or hex), and ADJ denotes the additional adjustment to the LIKELY
458 counter that may need to be added when ADJUST is a range. */
460 fmtresult&
461 fmtresult::adjust_for_width_or_precision (const HOST_WIDE_INT adjust[2],
462 tree type /* = NULL_TREE */,
463 unsigned base /* = 0 */,
464 unsigned adj /* = 0 */)
466 bool minadjusted = false;
468 /* Adjust the minimum and likely counters. */
469 if (adjust[0] >= 0)
471 if (range.min < (unsigned HOST_WIDE_INT)adjust[0])
473 range.min = adjust[0];
474 minadjusted = true;
477 /* Adjust the likely counter. */
478 if (range.likely < range.min)
479 range.likely = range.min;
481 else if (adjust[0] == target_int_min ()
482 && (unsigned HOST_WIDE_INT)adjust[1] == target_int_max ())
483 knownrange = false;
485 /* Adjust the maximum counter. */
486 if (adjust[1] > 0)
488 if (range.max < (unsigned HOST_WIDE_INT)adjust[1])
490 range.max = adjust[1];
492 /* Set KNOWNRANGE if both the minimum and maximum have been
493 adjusted. Otherwise leave it at what it was before. */
494 knownrange = minadjusted;
498 if (warn_level > 1 && type)
500 /* For large non-constant width or precision whose range spans
501 the maximum number of digits produced by the directive for
502 any argument, set the likely number of bytes to be at most
503 the number digits plus other adjustment determined by the
504 caller (one for sign or two for the hexadecimal "0x"
505 prefix). */
506 unsigned dirdigs = type_max_digits (type, base);
507 if (adjust[0] < dirdigs && dirdigs < adjust[1]
508 && range.likely < dirdigs)
509 range.likely = dirdigs + adj;
511 else if (range.likely < (range.min ? range.min : 1))
513 /* Conservatively, set LIKELY to at least MIN but no less than
514 1 unless MAX is zero. */
515 range.likely = (range.min
516 ? range.min
517 : range.max && (range.max < HOST_WIDE_INT_MAX
518 || warn_level > 1) ? 1 : 0);
521 /* Finally adjust the unlikely counter to be at least as large as
522 the maximum. */
523 if (range.unlikely < range.max)
524 range.unlikely = range.max;
526 return *this;
529 /* Return the maximum number of digits a value of TYPE formats in
530 BASE on output, not counting base prefix . */
532 unsigned
533 fmtresult::type_max_digits (tree type, int base)
535 unsigned prec = TYPE_PRECISION (type);
536 switch (base)
538 case 2:
539 return prec;
540 case 8:
541 return (prec + 2) / 3;
542 case 10:
543 /* Decimal approximation: yields 3, 5, 10, and 20 for precision
544 of 8, 16, 32, and 64 bits. */
545 return prec * 301 / 1000 + 1;
546 case 16:
547 return prec / 4;
550 gcc_unreachable ();
553 static bool
554 get_int_range (tree, gimple *, HOST_WIDE_INT *, HOST_WIDE_INT *,
555 bool, HOST_WIDE_INT, range_query *);
557 struct call_info;
559 /* Description of a format directive. A directive is either a plain
560 string or a conversion specification that starts with '%'. */
562 struct directive
564 directive (const call_info *inf, unsigned dno)
565 : info (inf), dirno (dno), argno (), beg (), len (), flags (),
566 width (), prec (), modifier (), specifier (), arg (), fmtfunc ()
569 /* Reference to the info structure describing the call that this
570 directive is a part of. */
571 const call_info *info;
573 /* The 1-based directive number (for debugging). */
574 unsigned dirno;
576 /* The zero-based argument number of the directive's argument ARG in
577 the function's argument list. */
578 unsigned argno;
580 /* The first character of the directive and its length. */
581 const char *beg;
582 size_t len;
584 /* A bitmap of flags, one for each character. */
585 unsigned flags[256 / sizeof (int)];
587 /* The range of values of the specified width, or -1 if not specified. */
588 HOST_WIDE_INT width[2];
589 /* The range of values of the specified precision, or -1 if not
590 specified. */
591 HOST_WIDE_INT prec[2];
593 /* Length modifier. */
594 format_lengths modifier;
596 /* Format specifier character. */
597 char specifier;
599 /* The argument of the directive or null when the directive doesn't
600 take one or when none is available (such as for vararg functions). */
601 tree arg;
603 /* Format conversion function that given a directive and an argument
604 returns the formatting result. */
605 fmtresult (*fmtfunc) (const directive &, tree, pointer_query &);
607 /* Return True when the format flag CHR has been used. */
608 bool get_flag (char chr) const
610 unsigned char c = chr & 0xff;
611 return (flags[c / (CHAR_BIT * sizeof *flags)]
612 & (1U << (c % (CHAR_BIT * sizeof *flags))));
615 /* Make a record of the format flag CHR having been used. */
616 void set_flag (char chr)
618 unsigned char c = chr & 0xff;
619 flags[c / (CHAR_BIT * sizeof *flags)]
620 |= (1U << (c % (CHAR_BIT * sizeof *flags)));
623 /* Reset the format flag CHR. */
624 void clear_flag (char chr)
626 unsigned char c = chr & 0xff;
627 flags[c / (CHAR_BIT * sizeof *flags)]
628 &= ~(1U << (c % (CHAR_BIT * sizeof *flags)));
631 /* Set both bounds of the width range to VAL. */
632 void set_width (HOST_WIDE_INT val)
634 width[0] = width[1] = val;
637 /* Set the width range according to ARG, with both bounds being
638 no less than 0. For a constant ARG set both bounds to its value
639 or 0, whichever is greater. For a non-constant ARG in some range
640 set width to its range adjusting each bound to -1 if it's less.
641 For an indeterminate ARG set width to [0, INT_MAX]. */
642 void set_width (tree arg, range_query *);
644 /* Set both bounds of the precision range to VAL. */
645 void set_precision (HOST_WIDE_INT val)
647 prec[0] = prec[1] = val;
650 /* Set the precision range according to ARG, with both bounds being
651 no less than -1. For a constant ARG set both bounds to its value
652 or -1 whichever is greater. For a non-constant ARG in some range
653 set precision to its range adjusting each bound to -1 if it's less.
654 For an indeterminate ARG set precision to [-1, INT_MAX]. */
655 void set_precision (tree arg, range_query *query);
657 /* Return true if both width and precision are known to be
658 either constant or in some range, false otherwise. */
659 bool known_width_and_precision () const
661 return ((width[1] < 0
662 || (unsigned HOST_WIDE_INT)width[1] <= target_int_max ())
663 && (prec[1] < 0
664 || (unsigned HOST_WIDE_INT)prec[1] < target_int_max ()));
668 /* The result of a call to a formatted function. */
670 struct format_result
672 format_result ()
673 : range (), aliases (), alias_count (), knownrange (), posunder4k (),
674 floating (), warned () { /* No-op. */ }
676 ~format_result ()
678 XDELETEVEC (aliases);
681 /* Range of characters written by the formatted function.
682 Setting the minimum to HOST_WIDE_INT_MAX disables all
683 length tracking for the remainder of the format string. */
684 result_range range;
686 struct alias_info
688 directive dir; /* The directive that aliases the destination. */
689 HOST_WIDE_INT offset; /* The offset at which it aliases it. */
690 result_range range; /* The raw result of the directive. */
693 /* An array of directives whose pointer argument aliases a part
694 of the destination object of the formatted function. */
695 alias_info *aliases;
696 unsigned alias_count;
698 /* True when the range above is obtained from known values of
699 directive arguments, or bounds on the amount of output such
700 as width and precision, and not the result of heuristics that
701 depend on warning levels. It's used to issue stricter diagnostics
702 in cases where strings of unknown lengths are bounded by the arrays
703 they are determined to refer to. KNOWNRANGE must not be used for
704 the return value optimization. */
705 bool knownrange;
707 /* True if no individual directive could fail or result in more than
708 4095 bytes of output (the total NUMBER_CHARS_{MIN,MAX} might be
709 greater). Implementations are not required to handle directives
710 that produce more than 4K bytes (leading to undefined behavior)
711 and so when one is found it disables the return value optimization.
712 Similarly, directives that can fail (such as wide character
713 directives) disable the optimization. */
714 bool posunder4k;
716 /* True when a floating point directive has been seen in the format
717 string. */
718 bool floating;
720 /* True when an intermediate result has caused a warning. Used to
721 avoid issuing duplicate warnings while finishing the processing
722 of a call. WARNED also disables the return value optimization. */
723 bool warned;
725 /* Preincrement the number of output characters by 1. */
726 format_result& operator++ ()
728 return *this += 1;
731 /* Postincrement the number of output characters by 1. */
732 format_result operator++ (int)
734 format_result prev (*this);
735 *this += 1;
736 return prev;
739 /* Increment the number of output characters by N. */
740 format_result& operator+= (unsigned HOST_WIDE_INT);
742 /* Add a directive to the sequence of those with potentially aliasing
743 arguments. */
744 void append_alias (const directive &, HOST_WIDE_INT, const result_range &);
746 private:
747 /* Not copyable or assignable. */
748 format_result (format_result&);
749 void operator= (format_result&);
752 format_result&
753 format_result::operator+= (unsigned HOST_WIDE_INT n)
755 gcc_assert (n < HOST_WIDE_INT_MAX);
757 if (range.min < HOST_WIDE_INT_MAX)
758 range.min += n;
760 if (range.max < HOST_WIDE_INT_MAX)
761 range.max += n;
763 if (range.likely < HOST_WIDE_INT_MAX)
764 range.likely += n;
766 if (range.unlikely < HOST_WIDE_INT_MAX)
767 range.unlikely += n;
769 return *this;
772 void
773 format_result::append_alias (const directive &d, HOST_WIDE_INT off,
774 const result_range &resrng)
776 unsigned cnt = alias_count + 1;
777 alias_info *ar = XNEWVEC (alias_info, cnt);
779 for (unsigned i = 0; i != alias_count; ++i)
780 ar[i] = aliases[i];
782 ar[alias_count].dir = d;
783 ar[alias_count].offset = off;
784 ar[alias_count].range = resrng;
786 XDELETEVEC (aliases);
788 alias_count = cnt;
789 aliases = ar;
792 /* Return the logarithm of X in BASE. */
794 static int
795 ilog (unsigned HOST_WIDE_INT x, int base)
797 int res = 0;
800 ++res;
801 x /= base;
802 } while (x);
803 return res;
806 /* Return the number of bytes resulting from converting into a string
807 the INTEGER_CST tree node X in BASE with a minimum of PREC digits.
808 PLUS indicates whether 1 for a plus sign should be added for positive
809 numbers, and PREFIX whether the length of an octal ('0') or hexadecimal
810 ('0x') or binary ('0b') prefix should be added for nonzero numbers.
811 Return -1 if X cannot be represented. */
813 static HOST_WIDE_INT
814 tree_digits (tree x, int base, HOST_WIDE_INT prec, bool plus, bool prefix)
816 unsigned HOST_WIDE_INT absval;
818 HOST_WIDE_INT res;
820 if (TYPE_UNSIGNED (TREE_TYPE (x)))
822 if (tree_fits_uhwi_p (x))
824 absval = tree_to_uhwi (x);
825 res = plus;
827 else
828 return -1;
830 else
832 if (tree_fits_shwi_p (x))
834 HOST_WIDE_INT i = tree_to_shwi (x);
835 if (HOST_WIDE_INT_MIN == i)
837 /* Avoid undefined behavior due to negating a minimum. */
838 absval = HOST_WIDE_INT_MAX;
839 res = 1;
841 else if (i < 0)
843 absval = -i;
844 res = 1;
846 else
848 absval = i;
849 res = plus;
852 else
853 return -1;
856 int ndigs = ilog (absval, base);
858 res += prec < ndigs ? ndigs : prec;
860 /* Adjust a non-zero value for the base prefix, either hexadecimal,
861 or, unless precision has resulted in a leading zero, also octal. */
862 if (prefix && absval)
864 if (base == 8 && prec <= ndigs)
865 res += 1;
866 else if (base == 16 || base == 2) /* 0x...(0X...) or 0b...(0B...). */
867 res += 2;
870 return res;
873 /* Description of a call to a formatted function. */
875 struct call_info
877 /* Function call statement. */
878 gimple *callstmt;
880 /* Function called. */
881 tree func;
883 /* Called built-in function code. */
884 built_in_function fncode;
886 /* The "origin" of the destination pointer argument, which is either
887 the DECL of the destination buffer being written into or a pointer
888 that points to it, plus some offset. */
889 tree dst_origin;
891 /* For a destination pointing to a struct array member, the offset of
892 the member. */
893 HOST_WIDE_INT dst_field;
895 /* The offset into the destination buffer. */
896 HOST_WIDE_INT dst_offset;
898 /* Format argument and format string extracted from it. */
899 tree format;
900 const char *fmtstr;
902 /* The location of the format argument. */
903 location_t fmtloc;
905 /* The destination object size for __builtin___xxx_chk functions
906 typically determined by __builtin_object_size, or -1 if unknown. */
907 unsigned HOST_WIDE_INT objsize;
909 /* Number of the first variable argument. */
910 unsigned HOST_WIDE_INT argidx;
912 /* True for functions like snprintf that specify the size of
913 the destination, false for others like sprintf that don't. */
914 bool bounded;
916 /* True for bounded functions like snprintf that specify a zero-size
917 buffer as a request to compute the size of output without actually
918 writing any. NOWRITE is cleared in response to the %n directive
919 which has side-effects similar to writing output. */
920 bool nowrite;
922 /* Return true if the called function's return value is used. */
923 bool retval_used () const
925 return gimple_get_lhs (callstmt);
928 /* Return the warning option corresponding to the called function. */
929 opt_code warnopt () const
931 return bounded ? OPT_Wformat_truncation_ : OPT_Wformat_overflow_;
934 /* Return true for calls to file formatted functions. */
935 bool is_file_func () const
937 return (fncode == BUILT_IN_FPRINTF
938 || fncode == BUILT_IN_FPRINTF_CHK
939 || fncode == BUILT_IN_FPRINTF_UNLOCKED
940 || fncode == BUILT_IN_VFPRINTF
941 || fncode == BUILT_IN_VFPRINTF_CHK);
944 /* Return true for calls to string formatted functions. */
945 bool is_string_func () const
947 return (fncode == BUILT_IN_SPRINTF
948 || fncode == BUILT_IN_SPRINTF_CHK
949 || fncode == BUILT_IN_SNPRINTF
950 || fncode == BUILT_IN_SNPRINTF_CHK
951 || fncode == BUILT_IN_VSPRINTF
952 || fncode == BUILT_IN_VSPRINTF_CHK
953 || fncode == BUILT_IN_VSNPRINTF
954 || fncode == BUILT_IN_VSNPRINTF_CHK);
958 void
959 directive::set_width (tree arg, range_query *query)
961 get_int_range (arg, info->callstmt, width, width + 1, true, 0, query);
964 void
965 directive::set_precision (tree arg, range_query *query)
967 get_int_range (arg, info->callstmt, prec, prec + 1, false, -1, query);
970 /* Return the result of formatting a no-op directive (such as '%n'). */
972 static fmtresult
973 format_none (const directive &, tree, pointer_query &)
975 fmtresult res (0);
976 return res;
979 /* Return the result of formatting the '%%' directive. */
981 static fmtresult
982 format_percent (const directive &, tree, pointer_query &)
984 fmtresult res (1);
985 return res;
989 /* Compute intmax_type_node and uintmax_type_node similarly to how
990 tree.cc builds size_type_node. */
992 static void
993 build_intmax_type_nodes (tree *pintmax, tree *puintmax)
995 if (strcmp (UINTMAX_TYPE, "unsigned int") == 0)
997 *pintmax = integer_type_node;
998 *puintmax = unsigned_type_node;
1000 else if (strcmp (UINTMAX_TYPE, "long unsigned int") == 0)
1002 *pintmax = long_integer_type_node;
1003 *puintmax = long_unsigned_type_node;
1005 else if (strcmp (UINTMAX_TYPE, "long long unsigned int") == 0)
1007 *pintmax = long_long_integer_type_node;
1008 *puintmax = long_long_unsigned_type_node;
1010 else
1012 for (int i = 0; i < NUM_INT_N_ENTS; i++)
1013 if (int_n_enabled_p[i])
1015 char name[50], altname[50];
1016 sprintf (name, "__int%d unsigned", int_n_data[i].bitsize);
1017 sprintf (altname, "__int%d__ unsigned", int_n_data[i].bitsize);
1019 if (strcmp (name, UINTMAX_TYPE) == 0
1020 || strcmp (altname, UINTMAX_TYPE) == 0)
1022 *pintmax = int_n_trees[i].signed_type;
1023 *puintmax = int_n_trees[i].unsigned_type;
1024 return;
1027 gcc_unreachable ();
1031 /* Determine the range [*PMIN, *PMAX] that the expression ARG is
1032 in and that is representable in type int.
1033 Return true when the range is a subrange of that of int.
1034 When ARG is null it is as if it had the full range of int.
1035 When ABSOLUTE is true the range reflects the absolute value of
1036 the argument. When ABSOLUTE is false, negative bounds of
1037 the determined range are replaced with NEGBOUND. */
1039 static bool
1040 get_int_range (tree arg, gimple *stmt,
1041 HOST_WIDE_INT *pmin, HOST_WIDE_INT *pmax,
1042 bool absolute, HOST_WIDE_INT negbound,
1043 range_query *query)
1045 /* The type of the result. */
1046 const_tree type = integer_type_node;
1048 bool knownrange = false;
1050 if (!arg)
1052 *pmin = tree_to_shwi (TYPE_MIN_VALUE (type));
1053 *pmax = tree_to_shwi (TYPE_MAX_VALUE (type));
1055 else if (TREE_CODE (arg) == INTEGER_CST
1056 && TYPE_PRECISION (TREE_TYPE (arg)) <= TYPE_PRECISION (type))
1058 /* For a constant argument return its value adjusted as specified
1059 by NEGATIVE and NEGBOUND and return true to indicate that the
1060 result is known. */
1061 *pmin = tree_fits_shwi_p (arg) ? tree_to_shwi (arg) : tree_to_uhwi (arg);
1062 *pmax = *pmin;
1063 knownrange = true;
1065 else
1067 /* True if the argument's range cannot be determined. */
1068 bool unknown = true;
1070 tree argtype = TREE_TYPE (arg);
1072 /* Ignore invalid arguments with greater precision that that
1073 of the expected type (e.g., in sprintf("%*i", 12LL, i)).
1074 They will have been detected and diagnosed by -Wformat and
1075 so it's not important to complicate this code to try to deal
1076 with them again. */
1077 if (TREE_CODE (arg) == SSA_NAME
1078 && INTEGRAL_TYPE_P (argtype)
1079 && TYPE_PRECISION (argtype) <= TYPE_PRECISION (type))
1081 /* Try to determine the range of values of the integer argument. */
1082 int_range_max vr;
1083 query->range_of_expr (vr, arg, stmt);
1085 if (!vr.undefined_p () && !vr.varying_p ())
1087 HOST_WIDE_INT type_min
1088 = (TYPE_UNSIGNED (argtype)
1089 ? tree_to_uhwi (TYPE_MIN_VALUE (argtype))
1090 : tree_to_shwi (TYPE_MIN_VALUE (argtype)));
1092 HOST_WIDE_INT type_max = tree_to_uhwi (TYPE_MAX_VALUE (argtype));
1094 tree type = TREE_TYPE (arg);
1095 tree tmin = wide_int_to_tree (type, vr.lower_bound ());
1096 tree tmax = wide_int_to_tree (type, vr.upper_bound ());
1097 *pmin = TREE_INT_CST_LOW (tmin);
1098 *pmax = TREE_INT_CST_LOW (tmax);
1100 if (*pmin < *pmax)
1102 /* Return true if the adjusted range is a subrange of
1103 the full range of the argument's type. *PMAX may
1104 be less than *PMIN when the argument is unsigned
1105 and its upper bound is in excess of TYPE_MAX. In
1106 that (invalid) case disregard the range and use that
1107 of the expected type instead. */
1108 knownrange = type_min < *pmin || *pmax < type_max;
1110 unknown = false;
1115 /* Handle an argument with an unknown range as if none had been
1116 provided. */
1117 if (unknown)
1118 return get_int_range (NULL_TREE, NULL, pmin, pmax, absolute,
1119 negbound, query);
1122 /* Adjust each bound as specified by ABSOLUTE and NEGBOUND. */
1123 if (absolute)
1125 if (*pmin < 0)
1127 if (*pmin == *pmax)
1128 *pmin = *pmax = -*pmin;
1129 else
1131 /* Make sure signed overlow is avoided. */
1132 gcc_assert (*pmin != HOST_WIDE_INT_MIN);
1134 HOST_WIDE_INT tmp = -*pmin;
1135 *pmin = 0;
1136 if (*pmax < tmp)
1137 *pmax = tmp;
1141 else if (*pmin < negbound)
1142 *pmin = negbound;
1144 return knownrange;
1147 /* With the range [*ARGMIN, *ARGMAX] of an integer directive's actual
1148 argument, due to the conversion from either *ARGMIN or *ARGMAX to
1149 the type of the directive's formal argument it's possible for both
1150 to result in the same number of bytes or a range of bytes that's
1151 less than the number of bytes that would result from formatting
1152 some other value in the range [*ARGMIN, *ARGMAX]. This can be
1153 determined by checking for the actual argument being in the range
1154 of the type of the directive. If it isn't it must be assumed to
1155 take on the full range of the directive's type.
1156 Return true when the range has been adjusted to the full range
1157 of DIRTYPE, and false otherwise. */
1159 static bool
1160 adjust_range_for_overflow (tree dirtype, tree *argmin, tree *argmax)
1162 tree argtype = TREE_TYPE (*argmin);
1163 unsigned argprec = TYPE_PRECISION (argtype);
1164 unsigned dirprec = TYPE_PRECISION (dirtype);
1166 /* If the actual argument and the directive's argument have the same
1167 precision and sign there can be no overflow and so there is nothing
1168 to adjust. */
1169 if (argprec == dirprec && TYPE_SIGN (argtype) == TYPE_SIGN (dirtype))
1170 return false;
1172 /* The logic below was inspired/lifted from the CONVERT_EXPR_CODE_P
1173 branch in the extract_range_from_unary_expr function in tree-vrp.cc. */
1175 if (TREE_CODE (*argmin) == INTEGER_CST
1176 && TREE_CODE (*argmax) == INTEGER_CST
1177 && (dirprec >= argprec
1178 || integer_zerop (int_const_binop (RSHIFT_EXPR,
1179 int_const_binop (MINUS_EXPR,
1180 *argmax,
1181 *argmin),
1182 size_int (dirprec)))))
1184 unsigned int maxprec = MAX (argprec, dirprec);
1185 *argmin = force_fit_type (dirtype,
1186 wide_int::from (wi::to_wide (*argmin), maxprec,
1187 TYPE_SIGN (argtype)),
1188 0, false);
1189 *argmax = force_fit_type (dirtype,
1190 wide_int::from (wi::to_wide (*argmax), maxprec,
1191 TYPE_SIGN (argtype)),
1192 0, false);
1194 /* If *ARGMIN is still less than *ARGMAX the conversion above
1195 is safe. Otherwise, it has overflowed and would be unsafe. */
1196 if (tree_int_cst_le (*argmin, *argmax))
1197 return false;
1200 *argmin = TYPE_MIN_VALUE (dirtype);
1201 *argmax = TYPE_MAX_VALUE (dirtype);
1202 return true;
1205 /* Return a range representing the minimum and maximum number of bytes
1206 that the format directive DIR will output for any argument given
1207 the WIDTH and PRECISION (extracted from DIR). This function is
1208 used when the directive argument or its value isn't known. */
1210 static fmtresult
1211 format_integer (const directive &dir, tree arg, pointer_query &ptr_qry)
1213 tree intmax_type_node;
1214 tree uintmax_type_node;
1216 /* Base to format the number in. */
1217 int base;
1219 /* True when a conversion is preceded by a prefix indicating the base
1220 of the argument (octal or hexadecimal). */
1221 const bool maybebase = dir.get_flag ('#');
1223 /* True when a signed conversion is preceded by a sign or space. */
1224 bool maybesign = false;
1226 /* True for signed conversions (i.e., 'd' and 'i'). */
1227 bool sign = false;
1229 switch (dir.specifier)
1231 case 'd':
1232 case 'i':
1233 /* Space and '+' are only meaningful for signed conversions. */
1234 maybesign = dir.get_flag (' ') | dir.get_flag ('+');
1235 sign = true;
1236 base = 10;
1237 break;
1238 case 'u':
1239 base = 10;
1240 break;
1241 case 'b':
1242 case 'B':
1243 base = 2;
1244 break;
1245 case 'o':
1246 base = 8;
1247 break;
1248 case 'X':
1249 case 'x':
1250 base = 16;
1251 break;
1252 default:
1253 gcc_unreachable ();
1256 const unsigned adj = (sign | maybebase) + (base == 2 || base == 16);
1258 /* The type of the "formal" argument expected by the directive. */
1259 tree dirtype = NULL_TREE;
1261 /* Determine the expected type of the argument from the length
1262 modifier. */
1263 switch (dir.modifier)
1265 case FMT_LEN_none:
1266 if (dir.specifier == 'p')
1267 dirtype = ptr_type_node;
1268 else
1269 dirtype = sign ? integer_type_node : unsigned_type_node;
1270 break;
1272 case FMT_LEN_h:
1273 dirtype = sign ? short_integer_type_node : short_unsigned_type_node;
1274 break;
1276 case FMT_LEN_hh:
1277 dirtype = sign ? signed_char_type_node : unsigned_char_type_node;
1278 break;
1280 case FMT_LEN_l:
1281 dirtype = sign ? long_integer_type_node : long_unsigned_type_node;
1282 break;
1284 case FMT_LEN_L:
1285 case FMT_LEN_ll:
1286 dirtype = (sign
1287 ? long_long_integer_type_node
1288 : long_long_unsigned_type_node);
1289 break;
1291 case FMT_LEN_z:
1292 dirtype = signed_or_unsigned_type_for (!sign, size_type_node);
1293 break;
1295 case FMT_LEN_t:
1296 dirtype = signed_or_unsigned_type_for (!sign, ptrdiff_type_node);
1297 break;
1299 case FMT_LEN_j:
1300 build_intmax_type_nodes (&intmax_type_node, &uintmax_type_node);
1301 dirtype = sign ? intmax_type_node : uintmax_type_node;
1302 break;
1304 default:
1305 return fmtresult ();
1308 /* The type of the argument to the directive, either deduced from
1309 the actual non-constant argument if one is known, or from
1310 the directive itself when none has been provided because it's
1311 a va_list. */
1312 tree argtype = NULL_TREE;
1314 if (!arg)
1316 /* When the argument has not been provided, use the type of
1317 the directive's argument as an approximation. This will
1318 result in false positives for directives like %i with
1319 arguments with smaller precision (such as short or char). */
1320 argtype = dirtype;
1322 else if (TREE_CODE (arg) == INTEGER_CST)
1324 /* When a constant argument has been provided use its value
1325 rather than type to determine the length of the output. */
1326 fmtresult res;
1328 if ((dir.prec[0] <= 0 && dir.prec[1] >= 0) && integer_zerop (arg))
1330 /* As a special case, a precision of zero with a zero argument
1331 results in zero bytes except in base 8 when the '#' flag is
1332 specified, and for signed conversions in base 8 and 10 when
1333 either the space or '+' flag has been specified and it results
1334 in just one byte (with width having the normal effect). This
1335 must extend to the case of a specified precision with
1336 an unknown value because it can be zero. */
1337 res.range.min = ((base == 8 && dir.get_flag ('#')) || maybesign);
1338 if (res.range.min == 0 && dir.prec[0] != dir.prec[1])
1340 res.range.max = 1;
1341 res.range.likely = 1;
1343 else
1345 res.range.max = res.range.min;
1346 res.range.likely = res.range.min;
1349 else
1351 /* Convert the argument to the type of the directive. */
1352 arg = fold_convert (dirtype, arg);
1354 res.range.min = tree_digits (arg, base, dir.prec[0],
1355 maybesign, maybebase);
1356 if (dir.prec[0] == dir.prec[1])
1357 res.range.max = res.range.min;
1358 else
1359 res.range.max = tree_digits (arg, base, dir.prec[1],
1360 maybesign, maybebase);
1361 res.range.likely = res.range.min;
1362 res.knownrange = true;
1365 res.range.unlikely = res.range.max;
1367 /* Bump up the counters if WIDTH is greater than LEN. */
1368 res.adjust_for_width_or_precision (dir.width, dirtype, base, adj);
1369 /* Bump up the counters again if PRECision is greater still. */
1370 res.adjust_for_width_or_precision (dir.prec, dirtype, base, adj);
1372 return res;
1374 else if (INTEGRAL_TYPE_P (TREE_TYPE (arg))
1375 || TREE_CODE (TREE_TYPE (arg)) == POINTER_TYPE)
1376 /* Determine the type of the provided non-constant argument. */
1377 argtype = TREE_TYPE (arg);
1378 else
1379 /* Don't bother with invalid arguments since they likely would
1380 have already been diagnosed, and disable any further checking
1381 of the format string by returning [-1, -1]. */
1382 return fmtresult ();
1384 fmtresult res;
1386 /* Using either the range the non-constant argument is in, or its
1387 type (either "formal" or actual), create a range of values that
1388 constrain the length of output given the warning level. */
1389 tree argmin = NULL_TREE;
1390 tree argmax = NULL_TREE;
1392 if (arg
1393 && TREE_CODE (arg) == SSA_NAME
1394 && INTEGRAL_TYPE_P (argtype))
1396 /* Try to determine the range of values of the integer argument
1397 (range information is not available for pointers). */
1398 int_range_max vr;
1399 ptr_qry.rvals->range_of_expr (vr, arg, dir.info->callstmt);
1401 if (!vr.varying_p () && !vr.undefined_p ())
1403 argmin = wide_int_to_tree (TREE_TYPE (arg), vr.lower_bound ());
1404 argmax = wide_int_to_tree (TREE_TYPE (arg), vr.upper_bound ());
1406 /* Set KNOWNRANGE if the argument is in a known subrange
1407 of the directive's type and neither width nor precision
1408 is unknown. (KNOWNRANGE may be reset below). */
1409 res.knownrange
1410 = ((!tree_int_cst_equal (TYPE_MIN_VALUE (dirtype), argmin)
1411 || !tree_int_cst_equal (TYPE_MAX_VALUE (dirtype), argmax))
1412 && dir.known_width_and_precision ());
1414 res.argmin = argmin;
1415 res.argmax = argmax;
1417 else
1419 /* The argument here may be the result of promoting the actual
1420 argument to int. Try to determine the type of the actual
1421 argument before promotion and narrow down its range that
1422 way. */
1423 gimple *def = SSA_NAME_DEF_STMT (arg);
1424 if (is_gimple_assign (def))
1426 tree_code code = gimple_assign_rhs_code (def);
1427 if (code == INTEGER_CST)
1429 arg = gimple_assign_rhs1 (def);
1430 return format_integer (dir, arg, ptr_qry);
1433 if (code == NOP_EXPR)
1435 tree type = TREE_TYPE (gimple_assign_rhs1 (def));
1436 if (INTEGRAL_TYPE_P (type)
1437 || TREE_CODE (type) == POINTER_TYPE)
1438 argtype = type;
1444 if (!argmin)
1446 if (TREE_CODE (argtype) == POINTER_TYPE)
1448 argmin = build_int_cst (pointer_sized_int_node, 0);
1449 argmax = build_all_ones_cst (pointer_sized_int_node);
1451 else
1453 argmin = TYPE_MIN_VALUE (argtype);
1454 argmax = TYPE_MAX_VALUE (argtype);
1458 /* Clear KNOWNRANGE if the range has been adjusted to the maximum
1459 of the directive. If it has been cleared then since ARGMIN and/or
1460 ARGMAX have been adjusted also adjust the corresponding ARGMIN and
1461 ARGMAX in the result to include in diagnostics. */
1462 if (adjust_range_for_overflow (dirtype, &argmin, &argmax))
1464 res.knownrange = false;
1465 res.argmin = argmin;
1466 res.argmax = argmax;
1469 /* Recursively compute the minimum and maximum from the known range. */
1470 if (TYPE_UNSIGNED (dirtype) || tree_int_cst_sgn (argmin) >= 0)
1472 /* For unsigned conversions/directives or signed when
1473 the minimum is positive, use the minimum and maximum to compute
1474 the shortest and longest output, respectively. */
1475 res.range.min = format_integer (dir, argmin, ptr_qry).range.min;
1476 res.range.max = format_integer (dir, argmax, ptr_qry).range.max;
1478 else if (tree_int_cst_sgn (argmax) < 0)
1480 /* For signed conversions/directives if maximum is negative,
1481 use the minimum as the longest output and maximum as the
1482 shortest output. */
1483 res.range.min = format_integer (dir, argmax, ptr_qry).range.min;
1484 res.range.max = format_integer (dir, argmin, ptr_qry).range.max;
1486 else
1488 /* Otherwise, 0 is inside of the range and minimum negative. Use 0
1489 as the shortest output and for the longest output compute the
1490 length of the output of both minimum and maximum and pick the
1491 longer. */
1492 unsigned HOST_WIDE_INT max1
1493 = format_integer (dir, argmin, ptr_qry).range.max;
1494 unsigned HOST_WIDE_INT max2
1495 = format_integer (dir, argmax, ptr_qry).range.max;
1496 res.range.min
1497 = format_integer (dir, integer_zero_node, ptr_qry).range.min;
1498 res.range.max = MAX (max1, max2);
1501 /* If the range is known, use the maximum as the likely length. */
1502 if (res.knownrange)
1503 res.range.likely = res.range.max;
1504 else
1506 /* Otherwise, use the minimum. Except for the case where for %#x or
1507 %#o the minimum is just for a single value in the range (0) and
1508 for all other values it is something longer, like 0x1 or 01.
1509 Use the length for value 1 in that case instead as the likely
1510 length. */
1511 res.range.likely = res.range.min;
1512 if (maybebase
1513 && base != 10
1514 && (tree_int_cst_sgn (argmin) < 0 || tree_int_cst_sgn (argmax) > 0))
1516 if (res.range.min == 1)
1517 res.range.likely += base == 8 ? 1 : 2;
1518 else if (res.range.min == 2
1519 && (base == 16 || base == 2)
1520 && (dir.width[0] == 2 || dir.prec[0] == 2))
1521 ++res.range.likely;
1525 res.range.unlikely = res.range.max;
1526 res.adjust_for_width_or_precision (dir.width, dirtype, base, adj);
1527 res.adjust_for_width_or_precision (dir.prec, dirtype, base, adj);
1529 return res;
1532 /* Return the number of bytes that a format directive consisting of FLAGS,
1533 PRECision, format SPECification, and MPFR rounding specifier RNDSPEC,
1534 would result for argument X under ideal conditions (i.e., if PREC
1535 weren't excessive). MPFR 3.1 allocates large amounts of memory for
1536 values of PREC with large magnitude and can fail (see MPFR bug #21056).
1537 This function works around those problems. */
1539 static unsigned HOST_WIDE_INT
1540 get_mpfr_format_length (mpfr_ptr x, const char *flags, HOST_WIDE_INT prec,
1541 char spec, char rndspec)
1543 char fmtstr[40];
1545 HOST_WIDE_INT len = strlen (flags);
1547 fmtstr[0] = '%';
1548 memcpy (fmtstr + 1, flags, len);
1549 memcpy (fmtstr + 1 + len, ".*R", 3);
1550 fmtstr[len + 4] = rndspec;
1551 fmtstr[len + 5] = spec;
1552 fmtstr[len + 6] = '\0';
1554 spec = TOUPPER (spec);
1555 if (spec == 'E' || spec == 'F')
1557 /* For %e, specify the precision explicitly since mpfr_sprintf
1558 does its own thing just to be different (see MPFR bug 21088). */
1559 if (prec < 0)
1560 prec = 6;
1562 else
1564 /* Avoid passing negative precisions with larger magnitude to MPFR
1565 to avoid exposing its bugs. (A negative precision is supposed
1566 to be ignored.) */
1567 if (prec < 0)
1568 prec = -1;
1571 HOST_WIDE_INT p = prec;
1573 if (spec == 'G' && !strchr (flags, '#'))
1575 /* For G/g without the pound flag, precision gives the maximum number
1576 of significant digits which is bounded by LDBL_MAX_10_EXP, or, for
1577 a 128 bit IEEE extended precision, 4932. Using twice as much here
1578 should be more than sufficient for any real format. */
1579 if ((IEEE_MAX_10_EXP * 2) < prec)
1580 prec = IEEE_MAX_10_EXP * 2;
1581 p = prec;
1583 else
1585 /* Cap precision arbitrarily at 1KB and add the difference
1586 (if any) to the MPFR result. */
1587 if (prec > 1024)
1588 p = 1024;
1591 len = mpfr_snprintf (NULL, 0, fmtstr, (int)p, x);
1593 /* Handle the unlikely (impossible?) error by returning more than
1594 the maximum dictated by the function's return type. */
1595 if (len < 0)
1596 return target_dir_max () + 1;
1598 /* Adjust the return value by the difference. */
1599 if (p < prec)
1600 len += prec - p;
1602 return len;
1605 /* Return the number of bytes to format using the format specifier
1606 SPEC and the precision PREC the largest value in the real floating
1607 TYPE. */
1609 static unsigned HOST_WIDE_INT
1610 format_floating_max (tree type, char spec, HOST_WIDE_INT prec)
1612 machine_mode mode = TYPE_MODE (type);
1614 /* IBM Extended mode. */
1615 if (MODE_COMPOSITE_P (mode))
1616 mode = DFmode;
1618 /* Get the real type format description for the target. */
1619 const real_format *rfmt = REAL_MODE_FORMAT (mode);
1620 REAL_VALUE_TYPE rv;
1622 real_maxval (&rv, 0, mode);
1624 /* Convert the GCC real value representation with the precision
1625 of the real type to the mpfr_t format with the GCC default
1626 round-to-nearest mode. */
1627 mpfr_t x;
1628 mpfr_init2 (x, rfmt->p);
1629 mpfr_from_real (x, &rv, MPFR_RNDN);
1631 /* Return a value one greater to account for the leading minus sign. */
1632 unsigned HOST_WIDE_INT r
1633 = 1 + get_mpfr_format_length (x, "", prec, spec, 'D');
1634 mpfr_clear (x);
1635 return r;
1638 /* Return a range representing the minimum and maximum number of bytes
1639 that the directive DIR will output for any argument. PREC gives
1640 the adjusted precision range to account for negative precisions
1641 meaning the default 6. This function is used when the directive
1642 argument or its value isn't known. */
1644 static fmtresult
1645 format_floating (const directive &dir, const HOST_WIDE_INT prec[2])
1647 tree type;
1649 switch (dir.modifier)
1651 case FMT_LEN_l:
1652 case FMT_LEN_none:
1653 type = double_type_node;
1654 break;
1656 case FMT_LEN_L:
1657 type = long_double_type_node;
1658 break;
1660 case FMT_LEN_ll:
1661 type = long_double_type_node;
1662 break;
1664 default:
1665 return fmtresult ();
1668 /* The minimum and maximum number of bytes produced by the directive. */
1669 fmtresult res;
1671 /* The minimum output as determined by flags. It's always at least 1.
1672 When plus or space are set the output is preceded by either a sign
1673 or a space. */
1674 unsigned flagmin = (1 /* for the first digit */
1675 + (dir.get_flag ('+') | dir.get_flag (' ')));
1677 /* The minimum is 3 for "inf" and "nan" for all specifiers, plus 1
1678 for the plus sign/space with the '+' and ' ' flags, respectively,
1679 unless reduced below. */
1680 res.range.min = 2 + flagmin;
1682 /* When the pound flag is set the decimal point is included in output
1683 regardless of precision. Whether or not a decimal point is included
1684 otherwise depends on the specification and precision. */
1685 bool radix = dir.get_flag ('#');
1687 switch (dir.specifier)
1689 case 'A':
1690 case 'a':
1692 HOST_WIDE_INT minprec = 6 + !radix /* decimal point */;
1693 if (dir.prec[0] <= 0)
1694 minprec = 0;
1695 else if (dir.prec[0] > 0)
1696 minprec = dir.prec[0] + !radix /* decimal point */;
1698 res.range.likely = (2 /* 0x */
1699 + flagmin
1700 + radix
1701 + minprec
1702 + 3 /* p+0 */);
1704 res.range.max = format_floating_max (type, 'a', prec[1]);
1706 /* The unlikely maximum accounts for the longest multibyte
1707 decimal point character. */
1708 res.range.unlikely = res.range.max;
1709 if (dir.prec[1] > 0)
1710 res.range.unlikely += target_mb_len_max () - 1;
1712 break;
1715 case 'E':
1716 case 'e':
1718 /* Minimum output attributable to precision and, when it's
1719 non-zero, decimal point. */
1720 HOST_WIDE_INT minprec = prec[0] ? prec[0] + !radix : 0;
1722 /* The likely minimum output is "[-+]1.234567e+00" regardless
1723 of the value of the actual argument. */
1724 res.range.likely = (flagmin
1725 + radix
1726 + minprec
1727 + 2 /* e+ */ + 2);
1729 res.range.max = format_floating_max (type, 'e', prec[1]);
1731 /* The unlikely maximum accounts for the longest multibyte
1732 decimal point character. */
1733 if (dir.prec[0] != dir.prec[1]
1734 || dir.prec[0] == -1 || dir.prec[0] > 0)
1735 res.range.unlikely = res.range.max + target_mb_len_max () -1;
1736 else
1737 res.range.unlikely = res.range.max;
1738 break;
1741 case 'F':
1742 case 'f':
1744 /* Minimum output attributable to precision and, when it's non-zero,
1745 decimal point. */
1746 HOST_WIDE_INT minprec = prec[0] ? prec[0] + !radix : 0;
1748 /* For finite numbers (i.e., not infinity or NaN) the lower bound
1749 when precision isn't specified is 8 bytes ("1.23456" since
1750 precision is taken to be 6). When precision is zero, the lower
1751 bound is 1 byte (e.g., "1"). Otherwise, when precision is greater
1752 than zero, then the lower bound is 2 plus precision (plus flags).
1753 But in all cases, the lower bound is no greater than 3. */
1754 unsigned HOST_WIDE_INT min = flagmin + radix + minprec;
1755 if (min < res.range.min)
1756 res.range.min = min;
1758 /* Compute the upper bound for -TYPE_MAX. */
1759 res.range.max = format_floating_max (type, 'f', prec[1]);
1761 /* The minimum output with unknown precision is a single byte
1762 (e.g., "0") but the more likely output is 3 bytes ("0.0"). */
1763 if (dir.prec[0] < 0 && dir.prec[1] > 0)
1764 res.range.likely = 3;
1765 else
1766 res.range.likely = min;
1768 /* The unlikely maximum accounts for the longest multibyte
1769 decimal point character. */
1770 if (dir.prec[0] != dir.prec[1]
1771 || dir.prec[0] == -1 || dir.prec[0] > 0)
1772 res.range.unlikely = res.range.max + target_mb_len_max () - 1;
1773 break;
1776 case 'G':
1777 case 'g':
1779 /* The %g output depends on precision and the exponent of
1780 the argument. Since the value of the argument isn't known
1781 the lower bound on the range of bytes (not counting flags
1782 or width) is 1 plus radix (i.e., either "0" or "0." for
1783 "%g" and "%#g", respectively, with a zero argument). */
1784 unsigned HOST_WIDE_INT min = flagmin + radix;
1785 if (min < res.range.min)
1786 res.range.min = min;
1788 char spec = 'g';
1789 HOST_WIDE_INT maxprec = dir.prec[1];
1790 if (radix && maxprec)
1792 /* When the pound flag (radix) is set, trailing zeros aren't
1793 trimmed and so the longest output is the same as for %e,
1794 except with precision minus 1 (as specified in C11). */
1795 spec = 'e';
1796 if (maxprec > 0)
1797 --maxprec;
1798 else if (maxprec < 0)
1799 maxprec = 5;
1801 else
1802 maxprec = prec[1];
1804 res.range.max = format_floating_max (type, spec, maxprec);
1806 /* The likely output is either the maximum computed above
1807 minus 1 (assuming the maximum is positive) when precision
1808 is known (or unspecified), or the same minimum as for %e
1809 (which is computed for a non-negative argument). Unlike
1810 for the other specifiers above the likely output isn't
1811 the minimum because for %g that's 1 which is unlikely. */
1812 if (dir.prec[1] < 0
1813 || (unsigned HOST_WIDE_INT)dir.prec[1] < target_int_max ())
1814 res.range.likely = res.range.max - 1;
1815 else
1817 HOST_WIDE_INT minprec = 6 + !radix /* decimal point */;
1818 res.range.likely = (flagmin
1819 + radix
1820 + minprec
1821 + 2 /* e+ */ + 2);
1824 /* The unlikely maximum accounts for the longest multibyte
1825 decimal point character. */
1826 res.range.unlikely = res.range.max + target_mb_len_max () - 1;
1827 break;
1830 default:
1831 return fmtresult ();
1834 /* Bump up the byte counters if WIDTH is greater. */
1835 res.adjust_for_width_or_precision (dir.width);
1836 return res;
1839 /* Return a range representing the minimum and maximum number of bytes
1840 that the directive DIR will write on output for the floating argument
1841 ARG. */
1843 static fmtresult
1844 format_floating (const directive &dir, tree arg, pointer_query &)
1846 HOST_WIDE_INT prec[] = { dir.prec[0], dir.prec[1] };
1847 tree type = (dir.modifier == FMT_LEN_L || dir.modifier == FMT_LEN_ll
1848 ? long_double_type_node : double_type_node);
1850 /* For an indeterminate precision the lower bound must be assumed
1851 to be zero. */
1852 if (TOUPPER (dir.specifier) == 'A')
1854 /* Get the number of fractional decimal digits needed to represent
1855 the argument without a loss of accuracy. */
1856 unsigned fmtprec
1857 = REAL_MODE_FORMAT (TYPE_MODE (type))->p;
1859 /* The precision of the IEEE 754 double format is 53.
1860 The precision of all other GCC binary double formats
1861 is 56 or less. */
1862 unsigned maxprec = fmtprec <= 56 ? 13 : 15;
1864 /* For %a, leave the minimum precision unspecified to let
1865 MFPR trim trailing zeros (as it and many other systems
1866 including Glibc happen to do) and set the maximum
1867 precision to reflect what it would be with trailing zeros
1868 present (as Solaris and derived systems do). */
1869 if (dir.prec[1] < 0)
1871 /* Both bounds are negative implies that precision has
1872 not been specified. */
1873 prec[0] = maxprec;
1874 prec[1] = -1;
1876 else if (dir.prec[0] < 0)
1878 /* With a negative lower bound and a non-negative upper
1879 bound set the minimum precision to zero and the maximum
1880 to the greater of the maximum precision (i.e., with
1881 trailing zeros present) and the specified upper bound. */
1882 prec[0] = 0;
1883 prec[1] = dir.prec[1] < maxprec ? maxprec : dir.prec[1];
1886 else if (dir.prec[0] < 0)
1888 if (dir.prec[1] < 0)
1890 /* A precision in a strictly negative range is ignored and
1891 the default of 6 is used instead. */
1892 prec[0] = prec[1] = 6;
1894 else
1896 /* For a precision in a partly negative range, the lower bound
1897 must be assumed to be zero and the new upper bound is the
1898 greater of 6 (the default precision used when the specified
1899 precision is negative) and the upper bound of the specified
1900 range. */
1901 prec[0] = 0;
1902 prec[1] = dir.prec[1] < 6 ? 6 : dir.prec[1];
1906 if (!arg
1907 || TREE_CODE (arg) != REAL_CST
1908 || !useless_type_conversion_p (type, TREE_TYPE (arg)))
1909 return format_floating (dir, prec);
1911 /* The minimum and maximum number of bytes produced by the directive. */
1912 fmtresult res;
1914 /* Get the real type format description for the target. */
1915 const REAL_VALUE_TYPE *rvp = TREE_REAL_CST_PTR (arg);
1916 const real_format *rfmt = REAL_MODE_FORMAT (TYPE_MODE (TREE_TYPE (arg)));
1918 if (!real_isfinite (rvp))
1920 /* The format for Infinity and NaN is "[-]inf"/"[-]infinity"
1921 and "[-]nan" with the choice being implementation-defined
1922 but not locale dependent. */
1923 bool sign = dir.get_flag ('+') || real_isneg (rvp);
1924 res.range.min = 3 + sign;
1926 res.range.likely = res.range.min;
1927 res.range.max = res.range.min;
1928 /* The unlikely maximum is "[-/+]infinity" or "[-/+][qs]nan".
1929 For NaN, the C/POSIX standards specify two formats:
1930 "[-/+]nan"
1932 "[-/+]nan(n-char-sequence)"
1933 No known printf implementation outputs the latter format but AIX
1934 outputs QNaN and SNaN for quiet and signalling NaN, respectively,
1935 so the unlikely maximum reflects that. */
1936 res.range.unlikely = sign + (real_isinf (rvp) ? 8 : 4);
1938 /* The range for infinity and NaN is known unless either width
1939 or precision is unknown. Width has the same effect regardless
1940 of whether the argument is finite. Precision is either ignored
1941 (e.g., Glibc) or can have an effect on the short vs long format
1942 such as inf/infinity (e.g., Solaris). */
1943 res.knownrange = dir.known_width_and_precision ();
1945 /* Adjust the range for width but ignore precision. */
1946 res.adjust_for_width_or_precision (dir.width);
1948 return res;
1951 char fmtstr [40];
1952 char *pfmt = fmtstr;
1954 /* Append flags. */
1955 for (const char *pf = "-+ #0"; *pf; ++pf)
1956 if (dir.get_flag (*pf))
1957 *pfmt++ = *pf;
1959 *pfmt = '\0';
1962 /* Set up an array to easily iterate over. */
1963 unsigned HOST_WIDE_INT* const minmax[] = {
1964 &res.range.min, &res.range.max
1967 for (int i = 0; i != ARRAY_SIZE (minmax); ++i)
1969 /* Convert the GCC real value representation with the precision
1970 of the real type to the mpfr_t format rounding down in the
1971 first iteration that computes the minimum and up in the second
1972 that computes the maximum. This order is arbitrary because
1973 rounding in either direction can result in longer output. */
1974 mpfr_t mpfrval;
1975 mpfr_init2 (mpfrval, rfmt->p);
1976 mpfr_from_real (mpfrval, rvp, i ? MPFR_RNDU : MPFR_RNDD);
1978 /* Use the MPFR rounding specifier to round down in the first
1979 iteration and then up. In most but not all cases this will
1980 result in the same number of bytes. */
1981 char rndspec = "DU"[i];
1983 /* Format it and store the result in the corresponding member
1984 of the result struct. */
1985 *minmax[i] = get_mpfr_format_length (mpfrval, fmtstr, prec[i],
1986 dir.specifier, rndspec);
1987 mpfr_clear (mpfrval);
1991 /* Make sure the minimum is less than the maximum (MPFR rounding
1992 in the call to mpfr_snprintf can result in the reverse. */
1993 if (res.range.max < res.range.min)
1995 unsigned HOST_WIDE_INT tmp = res.range.min;
1996 res.range.min = res.range.max;
1997 res.range.max = tmp;
2000 /* The range is known unless either width or precision is unknown. */
2001 res.knownrange = dir.known_width_and_precision ();
2003 /* For the same floating point constant, unless width or precision
2004 is unknown, use the longer output as the likely maximum since
2005 with round to nearest either is equally likely. Otherwise, when
2006 precision is unknown, use the greater of the minimum and 3 as
2007 the likely output (for "0.0" since zero precision is unlikely). */
2008 if (res.knownrange)
2009 res.range.likely = res.range.max;
2010 else if (res.range.min < 3
2011 && dir.prec[0] < 0
2012 && (unsigned HOST_WIDE_INT)dir.prec[1] == target_int_max ())
2013 res.range.likely = 3;
2014 else
2015 res.range.likely = res.range.min;
2017 res.range.unlikely = res.range.max;
2019 if (res.range.max > 2 && (prec[0] != 0 || prec[1] != 0))
2021 /* Unless the precision is zero output longer than 2 bytes may
2022 include the decimal point which must be a single character
2023 up to MB_LEN_MAX in length. This is overly conservative
2024 since in some conversions some constants result in no decimal
2025 point (e.g., in %g). */
2026 res.range.unlikely += target_mb_len_max () - 1;
2029 res.adjust_for_width_or_precision (dir.width);
2030 return res;
2033 /* Return a FMTRESULT struct set to the lengths of the shortest and longest
2034 strings referenced by the expression STR, or (-1, -1) when not known.
2035 Used by the format_string function below. */
2037 static fmtresult
2038 get_string_length (tree str, gimple *stmt, unsigned HOST_WIDE_INT max_size,
2039 unsigned eltsize, pointer_query &ptr_qry)
2041 if (!str)
2042 return fmtresult ();
2044 /* Try to determine the dynamic string length first.
2045 Set MAXBOUND to an arbitrary non-null non-integer node as a request
2046 to have it set to the length of the longest string in a PHI. */
2047 c_strlen_data lendata = { };
2048 lendata.maxbound = str;
2049 if (eltsize == 1)
2050 get_range_strlen_dynamic (str, stmt, &lendata, ptr_qry);
2051 else
2053 /* Determine the length of the shortest and longest string referenced
2054 by STR. Strings of unknown lengths are bounded by the sizes of
2055 arrays that subexpressions of STR may refer to. Pointers that
2056 aren't known to point any such arrays result in LENDATA.MAXLEN
2057 set to SIZE_MAX. */
2058 get_range_strlen (str, &lendata, eltsize);
2061 /* If LENDATA.MAXBOUND is not equal to .MINLEN it corresponds to the bound
2062 of the largest array STR refers to, if known, or it's set to SIZE_MAX
2063 otherwise. */
2065 /* Return the default result when nothing is known about the string. */
2066 if ((lendata.maxbound && !tree_fits_uhwi_p (lendata.maxbound))
2067 || !tree_fits_uhwi_p (lendata.maxlen))
2069 fmtresult res;
2070 res.nonstr = lendata.decl;
2071 return res;
2074 unsigned HOST_WIDE_INT lenmax = tree_to_uhwi (max_object_size ()) - 2;
2075 if (integer_zerop (lendata.minlen)
2076 && (!lendata.maxbound || lenmax <= tree_to_uhwi (lendata.maxbound))
2077 && lenmax <= tree_to_uhwi (lendata.maxlen))
2079 if (max_size > 0 && max_size < HOST_WIDE_INT_MAX)
2081 /* Adjust the conservative unknown/unbounded result if MAX_SIZE
2082 is valid. Set UNLIKELY to maximum in case MAX_SIZE refers
2083 to a subobject.
2084 TODO: This is overly conservative. Set UNLIKELY to the size
2085 of the outermost enclosing declared object. */
2086 fmtresult res (0, max_size - 1);
2087 res.nonstr = lendata.decl;
2088 res.range.likely = res.range.max;
2089 res.range.unlikely = HOST_WIDE_INT_MAX;
2090 return res;
2093 fmtresult res;
2094 res.nonstr = lendata.decl;
2095 return res;
2098 /* The minimum length of the string. */
2099 HOST_WIDE_INT min
2100 = (tree_fits_uhwi_p (lendata.minlen)
2101 ? tree_to_uhwi (lendata.minlen)
2102 : 0);
2104 /* The maximum length of the string; initially set to MAXBOUND which
2105 may be less than MAXLEN, but may be adjusted up below. */
2106 HOST_WIDE_INT max
2107 = (lendata.maxbound && tree_fits_uhwi_p (lendata.maxbound)
2108 ? tree_to_uhwi (lendata.maxbound)
2109 : HOST_WIDE_INT_M1U);
2111 /* True if either the maximum length is unknown or (conservatively)
2112 the array bound is less than the maximum length. That can happen
2113 when the length of the string is unknown but the array in which
2114 the string is stored is a member of a struct. The warning uses
2115 the size of the member as the upper bound but the optimization
2116 doesn't. The optimization could still use the size of
2117 enclosing object as the upper bound but that's not done here. */
2118 const bool unbounded
2119 = (integer_all_onesp (lendata.maxlen)
2120 || (lendata.maxbound
2121 && tree_int_cst_lt (lendata.maxbound, lendata.maxlen)));
2123 /* Set the max/likely counters to unbounded when a minimum is known
2124 but the maximum length isn't bounded. This implies that STR is
2125 a conditional expression involving a string of known length and
2126 an expression of unknown/unbounded length. */
2127 if (min
2128 && (unsigned HOST_WIDE_INT)min < HOST_WIDE_INT_M1U
2129 && unbounded)
2130 max = HOST_WIDE_INT_M1U;
2132 /* get_range_strlen() returns the target value of SIZE_MAX for
2133 strings of unknown length. Bump it up to HOST_WIDE_INT_M1U
2134 which may be bigger. */
2135 if ((unsigned HOST_WIDE_INT)min == target_size_max ())
2136 min = HOST_WIDE_INT_M1U;
2137 if ((unsigned HOST_WIDE_INT)max == target_size_max ())
2138 max = HOST_WIDE_INT_M1U;
2140 fmtresult res (min, max);
2141 res.nonstr = lendata.decl;
2143 /* Set RES.KNOWNRANGE to true if and only if all strings referenced
2144 by STR are known to be bounded (though not necessarily by their
2145 actual length but perhaps by their maximum possible length). */
2146 if (res.range.max < target_int_max ())
2148 res.knownrange = true;
2149 /* When the length of the longest string is known and not
2150 excessive use it as the likely length of the string(s). */
2151 res.range.likely = res.range.max;
2153 else
2155 /* When the upper bound is unknown (it can be zero or excessive)
2156 set the likely length to the greater of 1. If MAXBOUND is
2157 known, also reset the length of the lower bound to zero. */
2158 res.range.likely = res.range.min ? res.range.min : warn_level > 1;
2159 if (lendata.maxbound && !integer_all_onesp (lendata.maxbound))
2160 res.range.min = 0;
2163 res.range.unlikely = unbounded ? HOST_WIDE_INT_MAX : res.range.max;
2165 return res;
2168 /* Return the minimum and maximum number of characters formatted
2169 by the '%c' format directives and its wide character form for
2170 the argument ARG. ARG can be null (for functions such as
2171 vsprinf). */
2173 static fmtresult
2174 format_character (const directive &dir, tree arg, pointer_query &ptr_qry)
2176 fmtresult res;
2178 res.knownrange = true;
2180 if (dir.specifier == 'C' || dir.modifier == FMT_LEN_l)
2182 /* A wide character can result in as few as zero bytes. */
2183 res.range.min = 0;
2185 HOST_WIDE_INT min, max;
2186 if (get_int_range (arg, dir.info->callstmt, &min, &max, false, 0,
2187 ptr_qry.rvals))
2189 if (min == 0 && max == 0)
2191 /* In strict reading of older ISO C or POSIX, this required
2192 no characters to be emitted. ISO C23 changes that, so
2193 does POSIX, to match what has been implemented in most of the
2194 implementations, namely emitting a single NUL character.
2195 Let's use 0 for minimum and 1 for all the other values. */
2196 res.range.max = 1;
2197 res.range.likely = res.range.unlikely = 1;
2199 else if (min >= 0 && min < 128)
2201 /* Be conservative if the target execution character set
2202 is not a 1-to-1 mapping to the source character set or
2203 if the source set is not ASCII. */
2204 bool one_2_one_ascii
2205 = (target_to_host_charmap[0] == 1
2206 && target_to_host ('a') == 97);
2208 /* A wide character in the ASCII range most likely results
2209 in a single byte, and only unlikely in up to MB_LEN_MAX. */
2210 res.range.max = one_2_one_ascii ? 1 : target_mb_len_max ();
2211 res.range.likely = 1;
2212 res.range.unlikely = target_mb_len_max ();
2213 res.mayfail = !one_2_one_ascii;
2215 else
2217 /* A wide character outside the ASCII range likely results
2218 in up to two bytes, and only unlikely in up to MB_LEN_MAX. */
2219 res.range.max = target_mb_len_max ();
2220 res.range.likely = 2;
2221 res.range.unlikely = res.range.max;
2222 /* Converting such a character may fail. */
2223 res.mayfail = true;
2226 else
2228 /* An unknown wide character is treated the same as a wide
2229 character outside the ASCII range. */
2230 res.range.max = target_mb_len_max ();
2231 res.range.likely = 2;
2232 res.range.unlikely = res.range.max;
2233 res.mayfail = true;
2236 else
2238 /* A plain '%c' directive. Its output is exactly 1. */
2239 res.range.min = res.range.max = 1;
2240 res.range.likely = res.range.unlikely = 1;
2243 /* Bump up the byte counters if WIDTH is greater. */
2244 return res.adjust_for_width_or_precision (dir.width);
2247 /* If TYPE is an array or struct or union, increment *FLDOFF by the starting
2248 offset of the member that *OFF points into if one can be determined and
2249 set *FLDSIZE to its size in bytes and decrement *OFF by the same.
2250 Otherwise do nothing. */
2252 static void
2253 set_aggregate_size_and_offset (tree type, HOST_WIDE_INT *fldoff,
2254 HOST_WIDE_INT *fldsize, HOST_WIDE_INT *off)
2256 /* The byte offset of the most basic struct member the byte
2257 offset *OFF corresponds to, or for a (multidimensional)
2258 array member, the byte offset of the array element. */
2259 if (TREE_CODE (type) == ARRAY_TYPE
2260 && TREE_CODE (TREE_TYPE (type)) == ARRAY_TYPE)
2262 HOST_WIDE_INT index = 0, arrsize = 0;
2263 if (array_elt_at_offset (type, *off, &index, &arrsize))
2265 *fldoff += index;
2266 *fldsize = arrsize;
2268 /* Otherwise leave *FLDOFF et al. unchanged. */
2270 else if (RECORD_OR_UNION_TYPE_P (type))
2272 HOST_WIDE_INT index = 0;
2273 tree sub = field_at_offset (type, NULL_TREE, *off, &index);
2274 if (sub)
2276 tree subsize = DECL_SIZE_UNIT (sub);
2277 if (*fldsize < HOST_WIDE_INT_MAX
2278 && subsize
2279 && tree_fits_uhwi_p (subsize))
2280 *fldsize = tree_to_uhwi (subsize);
2281 else
2282 *fldsize = HOST_WIDE_INT_MAX;
2283 *fldoff += index;
2284 *off -= index;
2286 /* Otherwise leave *FLDOFF et al. unchanged. */
2290 /* For an expression X of pointer type, recursively try to find its origin
2291 (either object DECL or pointer such as PARM_DECL) Y and return such a Y.
2292 When X refers to an array element or struct member, set *FLDOFF to
2293 the offset of the element or member from the beginning of the "most
2294 derived" object and *FLDSIZE to its size. When nonnull, set *OFF to
2295 the overall offset from the beginning of the object so that
2296 *FLDOFF <= *OFF. */
2298 static tree
2299 get_origin_and_offset_r (tree x, HOST_WIDE_INT *fldoff, HOST_WIDE_INT *fldsize,
2300 HOST_WIDE_INT *off)
2302 HOST_WIDE_INT sizebuf = -1;
2303 if (!fldsize)
2304 fldsize = &sizebuf;
2306 if (DECL_P (x))
2308 /* Set the size if it hasn't been set yet. */
2309 if (tree size = DECL_SIZE_UNIT (x))
2310 if (*fldsize < 0 && tree_fits_shwi_p (size))
2311 *fldsize = tree_to_shwi (size);
2312 return x;
2315 switch (TREE_CODE (x))
2317 case ADDR_EXPR:
2318 x = TREE_OPERAND (x, 0);
2319 return get_origin_and_offset_r (x, fldoff, fldsize, off);
2321 case ARRAY_REF:
2323 tree sub = TREE_OPERAND (x, 1);
2324 unsigned HOST_WIDE_INT idx =
2325 tree_fits_uhwi_p (sub) ? tree_to_uhwi (sub) : HOST_WIDE_INT_MAX;
2327 tree elsz = array_ref_element_size (x);
2328 unsigned HOST_WIDE_INT elbytes =
2329 tree_fits_shwi_p (elsz) ? tree_to_shwi (elsz) : HOST_WIDE_INT_MAX;
2331 unsigned HOST_WIDE_INT byteoff = idx * elbytes;
2333 if (byteoff < HOST_WIDE_INT_MAX
2334 && elbytes < HOST_WIDE_INT_MAX
2335 && (elbytes == 0 || byteoff / elbytes == idx))
2337 /* For in-bounds constant offsets into constant-sized arrays
2338 bump up *OFF, and for what's likely arrays or structs of
2339 arrays, also *FLDOFF, as necessary. */
2340 if (off)
2341 *off += byteoff;
2342 if (elbytes > 1)
2343 *fldoff += byteoff;
2345 else
2346 *fldoff = HOST_WIDE_INT_MAX;
2348 x = TREE_OPERAND (x, 0);
2349 return get_origin_and_offset_r (x, fldoff, fldsize, off);
2352 case MEM_REF:
2353 if (off)
2355 tree offset = TREE_OPERAND (x, 1);
2356 *off = (tree_fits_uhwi_p (offset)
2357 ? tree_to_uhwi (offset) : HOST_WIDE_INT_MAX);
2360 x = TREE_OPERAND (x, 0);
2362 if (off)
2364 tree xtype
2365 = (TREE_CODE (x) == ADDR_EXPR
2366 ? TREE_TYPE (TREE_OPERAND (x, 0)) : TREE_TYPE (TREE_TYPE (x)));
2368 set_aggregate_size_and_offset (xtype, fldoff, fldsize, off);
2371 return get_origin_and_offset_r (x, fldoff, fldsize, nullptr);
2373 case COMPONENT_REF:
2375 tree foff = component_ref_field_offset (x);
2376 tree fld = TREE_OPERAND (x, 1);
2377 if (!tree_fits_shwi_p (foff)
2378 || !tree_fits_shwi_p (DECL_FIELD_BIT_OFFSET (fld)))
2379 return x;
2380 *fldoff += (tree_to_shwi (foff)
2381 + (tree_to_shwi (DECL_FIELD_BIT_OFFSET (fld))
2382 / BITS_PER_UNIT));
2384 get_origin_and_offset_r (fld, fldoff, fldsize, off);
2385 x = TREE_OPERAND (x, 0);
2386 return get_origin_and_offset_r (x, fldoff, nullptr, off);
2389 case SSA_NAME:
2391 gimple *def = SSA_NAME_DEF_STMT (x);
2392 if (is_gimple_assign (def))
2394 tree_code code = gimple_assign_rhs_code (def);
2395 if (code == ADDR_EXPR)
2397 x = gimple_assign_rhs1 (def);
2398 return get_origin_and_offset_r (x, fldoff, fldsize, off);
2401 if (code == POINTER_PLUS_EXPR)
2403 tree offset = gimple_assign_rhs2 (def);
2404 if (off && tree_fits_uhwi_p (offset))
2405 *off = tree_to_uhwi (offset);
2407 x = gimple_assign_rhs1 (def);
2408 x = get_origin_and_offset_r (x, fldoff, fldsize, off);
2409 if (off && !tree_fits_uhwi_p (offset))
2410 *off = HOST_WIDE_INT_MAX;
2411 if (off)
2413 tree xtype = TREE_TYPE (x);
2414 set_aggregate_size_and_offset (xtype, fldoff, fldsize, off);
2416 return x;
2418 else if (code == VAR_DECL)
2420 x = gimple_assign_rhs1 (def);
2421 return get_origin_and_offset_r (x, fldoff, fldsize, off);
2424 else if (gimple_nop_p (def) && SSA_NAME_VAR (x))
2425 x = SSA_NAME_VAR (x);
2427 tree xtype = TREE_TYPE (x);
2428 if (POINTER_TYPE_P (xtype))
2429 xtype = TREE_TYPE (xtype);
2431 if (off)
2432 set_aggregate_size_and_offset (xtype, fldoff, fldsize, off);
2435 default:
2436 break;
2439 return x;
2442 /* Nonrecursive version of the above.
2443 The function never returns null unless X is null to begin with. */
2445 static tree
2446 get_origin_and_offset (tree x, HOST_WIDE_INT *fldoff, HOST_WIDE_INT *off,
2447 HOST_WIDE_INT *fldsize = nullptr)
2449 if (!x)
2450 return NULL_TREE;
2452 HOST_WIDE_INT sizebuf;
2453 if (!fldsize)
2454 fldsize = &sizebuf;
2456 /* Invalidate *FLDSIZE. */
2457 *fldsize = -1;
2458 *fldoff = *off = 0;
2460 return get_origin_and_offset_r (x, fldoff, fldsize, off);
2463 /* If ARG refers to the same (sub)object or array element as described
2464 by DST and DST_FLD, return the byte offset into the struct member or
2465 array element referenced by ARG and set *ARG_SIZE to the size of
2466 the (sub)object. Otherwise return HOST_WIDE_INT_MIN to indicate
2467 that ARG and DST do not refer to the same object. */
2469 static HOST_WIDE_INT
2470 alias_offset (tree arg, HOST_WIDE_INT *arg_size,
2471 tree dst, HOST_WIDE_INT dst_fld)
2473 /* See if the argument refers to the same base object as the destination
2474 of the formatted function call, and if so, try to determine if they
2475 can alias. */
2476 if (!arg || !dst || !ptr_derefs_may_alias_p (arg, dst))
2477 return HOST_WIDE_INT_MIN;
2479 /* The two arguments may refer to the same object. If they both refer
2480 to a struct member, see if the members are one and the same. If so,
2481 return the offset into the member. */
2482 HOST_WIDE_INT arg_off = 0, arg_fld = 0;
2484 tree arg_orig = get_origin_and_offset (arg, &arg_fld, &arg_off, arg_size);
2486 if (arg_orig == dst && arg_fld == dst_fld)
2487 return arg_off;
2489 return HOST_WIDE_INT_MIN;
2492 /* Return the minimum and maximum number of characters formatted
2493 by the '%s' format directive and its wide character form for
2494 the argument ARG. ARG can be null (for functions such as
2495 vsprinf). */
2497 static fmtresult
2498 format_string (const directive &dir, tree arg, pointer_query &ptr_qry)
2500 fmtresult res;
2502 /* The size of the (sub)object ARG refers to. Used to adjust
2503 the conservative get_string_length() result. */
2504 HOST_WIDE_INT arg_size = 0;
2506 if (warn_restrict)
2508 /* See if ARG might alias the destination of the call with
2509 DST_ORIGIN and DST_FIELD. If so, store the starting offset
2510 so that the overlap can be determined for certain later,
2511 when the amount of output of the call (including subsequent
2512 directives) has been computed. Otherwise, store HWI_MIN. */
2513 res.dst_offset = alias_offset (arg, &arg_size, dir.info->dst_origin,
2514 dir.info->dst_field);
2515 if (res.dst_offset >= 0 && res.dst_offset <= arg_size)
2516 arg_size -= res.dst_offset;
2517 else
2518 arg_size = 0;
2521 /* Compute the range the argument's length can be in. */
2522 int count_by = 1;
2523 if (dir.specifier == 'S' || dir.modifier == FMT_LEN_l)
2525 /* Get a node for a C type that will be the same size
2526 as a wchar_t on the target. */
2527 tree node = get_typenode_from_name (MODIFIED_WCHAR_TYPE);
2529 /* Now that we have a suitable node, get the number of
2530 bytes it occupies. */
2531 count_by = int_size_in_bytes (node);
2532 gcc_checking_assert (count_by == 2 || count_by == 4);
2535 fmtresult slen =
2536 get_string_length (arg, dir.info->callstmt, arg_size, count_by, ptr_qry);
2537 if (slen.range.min == slen.range.max
2538 && slen.range.min < HOST_WIDE_INT_MAX)
2540 /* The argument is either a string constant or it refers
2541 to one of a number of strings of the same length. */
2543 /* A '%s' directive with a string argument with constant length. */
2544 res.range = slen.range;
2546 if (dir.specifier == 'S'
2547 || dir.modifier == FMT_LEN_l)
2549 /* In the worst case the length of output of a wide string S
2550 is bounded by MB_LEN_MAX * wcslen (S). */
2551 res.range.max *= target_mb_len_max ();
2552 res.range.unlikely = res.range.max;
2553 /* It's likely that the total length is not more that
2554 2 * wcslen (S).*/
2555 res.range.likely = res.range.min * 2;
2557 if (dir.prec[1] >= 0
2558 && (unsigned HOST_WIDE_INT)dir.prec[1] < res.range.max)
2560 res.range.max = dir.prec[1];
2561 res.range.likely = dir.prec[1];
2562 res.range.unlikely = dir.prec[1];
2565 if (dir.prec[0] < 0 && dir.prec[1] > -1)
2566 res.range.min = 0;
2567 else if (dir.prec[0] >= 0)
2568 res.range.likely = dir.prec[0];
2570 /* Even a non-empty wide character string need not convert into
2571 any bytes. */
2572 res.range.min = 0;
2574 /* A non-empty wide character conversion may fail. */
2575 if (slen.range.max > 0)
2576 res.mayfail = true;
2578 else
2580 res.knownrange = true;
2582 if (dir.prec[0] < 0 && dir.prec[1] > -1)
2583 res.range.min = 0;
2584 else if ((unsigned HOST_WIDE_INT)dir.prec[0] < res.range.min)
2585 res.range.min = dir.prec[0];
2587 if ((unsigned HOST_WIDE_INT)dir.prec[1] < res.range.max)
2589 res.range.max = dir.prec[1];
2590 res.range.likely = dir.prec[1];
2591 res.range.unlikely = dir.prec[1];
2595 else if (arg && integer_zerop (arg))
2597 /* Handle null pointer argument. */
2599 fmtresult res (0);
2600 res.nullp = true;
2601 return res;
2603 else
2605 /* For a '%s' and '%ls' directive with a non-constant string (either
2606 one of a number of strings of known length or an unknown string)
2607 the minimum number of characters is lesser of PRECISION[0] and
2608 the length of the shortest known string or zero, and the maximum
2609 is the lesser of the length of the longest known string or
2610 PTRDIFF_MAX and PRECISION[1]. The likely length is either
2611 the minimum at level 1 and the greater of the minimum and 1
2612 at level 2. This result is adjust upward for width (if it's
2613 specified). */
2615 if (dir.specifier == 'S'
2616 || dir.modifier == FMT_LEN_l)
2618 /* A wide character converts to as few as zero bytes. */
2619 slen.range.min = 0;
2620 if (slen.range.max < target_int_max ())
2621 slen.range.max *= target_mb_len_max ();
2623 if (slen.range.likely < target_int_max ())
2624 slen.range.likely *= 2;
2626 if (slen.range.unlikely < target_int_max ())
2627 slen.range.unlikely *= target_mb_len_max ();
2629 /* A non-empty wide character conversion may fail. */
2630 if (slen.range.max > 0)
2631 res.mayfail = true;
2634 res.range = slen.range;
2636 if (dir.prec[0] >= 0)
2638 /* Adjust the minimum to zero if the string length is unknown,
2639 or at most the lower bound of the precision otherwise. */
2640 if (slen.range.min >= target_int_max ())
2641 res.range.min = 0;
2642 else if ((unsigned HOST_WIDE_INT)dir.prec[0] < slen.range.min)
2643 res.range.min = dir.prec[0];
2645 /* Make both maxima no greater than the upper bound of precision. */
2646 if ((unsigned HOST_WIDE_INT)dir.prec[1] < slen.range.max
2647 || slen.range.max >= target_int_max ())
2649 res.range.max = dir.prec[1];
2650 res.range.unlikely = dir.prec[1];
2653 /* If precision is constant, set the likely counter to the lesser
2654 of it and the maximum string length. Otherwise, if the lower
2655 bound of precision is greater than zero, set the likely counter
2656 to the minimum. Otherwise set it to zero or one based on
2657 the warning level. */
2658 if (dir.prec[0] == dir.prec[1])
2659 res.range.likely
2660 = ((unsigned HOST_WIDE_INT)dir.prec[0] < slen.range.max
2661 ? dir.prec[0] : slen.range.max);
2662 else if (dir.prec[0] > 0)
2663 res.range.likely = res.range.min;
2664 else
2665 res.range.likely = warn_level > 1;
2667 else if (dir.prec[1] >= 0)
2669 res.range.min = 0;
2670 if ((unsigned HOST_WIDE_INT)dir.prec[1] < slen.range.max)
2671 res.range.max = dir.prec[1];
2672 res.range.likely = dir.prec[1] ? warn_level > 1 : 0;
2673 if ((unsigned HOST_WIDE_INT)dir.prec[1] < slen.range.unlikely)
2674 res.range.unlikely = dir.prec[1];
2676 else if (slen.range.min >= target_int_max ())
2678 res.range.min = 0;
2679 res.range.max = HOST_WIDE_INT_MAX;
2680 /* At level 1 strings of unknown length are assumed to be
2681 empty, while at level 1 they are assumed to be one byte
2682 long. */
2683 res.range.likely = warn_level > 1;
2684 res.range.unlikely = HOST_WIDE_INT_MAX;
2686 else
2688 /* A string of unknown length unconstrained by precision is
2689 assumed to be empty at level 1 and just one character long
2690 at higher levels. */
2691 if (res.range.likely >= target_int_max ())
2692 res.range.likely = warn_level > 1;
2696 /* If the argument isn't a nul-terminated string and the number
2697 of bytes on output isn't bounded by precision, set NONSTR. */
2698 if (slen.nonstr && slen.range.min < (unsigned HOST_WIDE_INT)dir.prec[0])
2699 res.nonstr = slen.nonstr;
2701 /* Bump up the byte counters if WIDTH is greater. */
2702 return res.adjust_for_width_or_precision (dir.width);
2705 /* Format plain string (part of the format string itself). */
2707 static fmtresult
2708 format_plain (const directive &dir, tree, pointer_query &)
2710 fmtresult res (dir.len);
2711 return res;
2714 /* Return true if the RESULT of a directive in a call describe by INFO
2715 should be diagnosed given the AVAILable space in the destination. */
2717 static bool
2718 should_warn_p (const call_info &info,
2719 const result_range &avail, const result_range &result)
2721 if (result.max <= avail.min)
2723 /* The least amount of space remaining in the destination is big
2724 enough for the longest output. */
2725 return false;
2728 if (info.bounded)
2730 if (warn_format_trunc == 1 && result.min <= avail.max
2731 && info.retval_used ())
2733 /* The likely amount of space remaining in the destination is big
2734 enough for the least output and the return value is used. */
2735 return false;
2738 if (warn_format_trunc == 1 && result.likely <= avail.likely
2739 && !info.retval_used ())
2741 /* The likely amount of space remaining in the destination is big
2742 enough for the likely output and the return value is unused. */
2743 return false;
2746 if (warn_format_trunc == 2
2747 && result.likely <= avail.min
2748 && (result.max <= avail.min
2749 || result.max > HOST_WIDE_INT_MAX))
2751 /* The minimum amount of space remaining in the destination is big
2752 enough for the longest output. */
2753 return false;
2756 else
2758 if (warn_level == 1 && result.likely <= avail.likely)
2760 /* The likely amount of space remaining in the destination is big
2761 enough for the likely output. */
2762 return false;
2765 if (warn_level == 2
2766 && result.likely <= avail.min
2767 && (result.max <= avail.min
2768 || result.max > HOST_WIDE_INT_MAX))
2770 /* The minimum amount of space remaining in the destination is big
2771 enough for the longest output. */
2772 return false;
2776 return true;
2779 /* At format string location describe by DIRLOC in a call described
2780 by INFO, issue a warning for a directive DIR whose output may be
2781 in excess of the available space AVAIL_RANGE in the destination
2782 given the formatting result FMTRES. This function does nothing
2783 except decide whether to issue a warning for a possible write
2784 past the end or truncation and, if so, format the warning.
2785 Return true if a warning has been issued. */
2787 static bool
2788 maybe_warn (substring_loc &dirloc, location_t argloc,
2789 const call_info &info,
2790 const result_range &avail_range, const result_range &res,
2791 const directive &dir)
2793 if (!should_warn_p (info, avail_range, res))
2794 return false;
2796 /* A warning will definitely be issued below. */
2798 /* The maximum byte count to reference in the warning. Larger counts
2799 imply that the upper bound is unknown (and could be anywhere between
2800 RES.MIN + 1 and SIZE_MAX / 2) are printed as "N or more bytes" rather
2801 than "between N and X" where X is some huge number. */
2802 unsigned HOST_WIDE_INT maxbytes = target_dir_max ();
2804 /* True when there is enough room in the destination for the least
2805 amount of a directive's output but not enough for its likely or
2806 maximum output. */
2807 bool maybe = (res.min <= avail_range.max
2808 && (avail_range.min < res.likely
2809 || (res.max < HOST_WIDE_INT_MAX
2810 && avail_range.min < res.max)));
2812 /* Buffer for the directive in the host character set (used when
2813 the source character set is different). */
2814 char hostdir[32];
2816 if (avail_range.min == avail_range.max)
2818 /* The size of the destination region is exact. */
2819 unsigned HOST_WIDE_INT navail = avail_range.max;
2821 if (target_to_host (*dir.beg) != '%')
2823 /* For plain character directives (i.e., the format string itself)
2824 but not others, point the caret at the first character that's
2825 past the end of the destination. */
2826 if (navail < dir.len)
2827 dirloc.set_caret_index (dirloc.get_caret_idx () + navail);
2830 if (*dir.beg == '\0')
2832 /* This is the terminating nul. */
2833 gcc_assert (res.min == 1 && res.min == res.max);
2835 return fmtwarn (dirloc, UNKNOWN_LOCATION, NULL, info.warnopt (),
2836 info.bounded
2837 ? (maybe
2838 ? G_("%qE output may be truncated before the "
2839 "last format character")
2840 : G_("%qE output truncated before the last "
2841 "format character"))
2842 : (maybe
2843 ? G_("%qE may write a terminating nul past the "
2844 "end of the destination")
2845 : G_("%qE writing a terminating nul past the "
2846 "end of the destination")),
2847 info.func);
2850 if (res.min == res.max)
2852 const char *d = target_to_host (hostdir, sizeof hostdir, dir.beg);
2853 if (!info.bounded)
2854 return fmtwarn_n (dirloc, argloc, NULL, info.warnopt (), res.min,
2855 "%<%.*s%> directive writing %wu byte into a "
2856 "region of size %wu",
2857 "%<%.*s%> directive writing %wu bytes into a "
2858 "region of size %wu",
2859 (int) dir.len, d, res.min, navail);
2860 else if (maybe)
2861 return fmtwarn_n (dirloc, argloc, NULL, info.warnopt (), res.min,
2862 "%<%.*s%> directive output may be truncated "
2863 "writing %wu byte into a region of size %wu",
2864 "%<%.*s%> directive output may be truncated "
2865 "writing %wu bytes into a region of size %wu",
2866 (int) dir.len, d, res.min, navail);
2867 else
2868 return fmtwarn_n (dirloc, argloc, NULL, info.warnopt (), res.min,
2869 "%<%.*s%> directive output truncated writing "
2870 "%wu byte into a region of size %wu",
2871 "%<%.*s%> directive output truncated writing "
2872 "%wu bytes into a region of size %wu",
2873 (int) dir.len, d, res.min, navail);
2875 if (res.min == 0 && res.max < maxbytes)
2876 return fmtwarn (dirloc, argloc, NULL,
2877 info.warnopt (),
2878 info.bounded
2879 ? (maybe
2880 ? G_("%<%.*s%> directive output may be truncated "
2881 "writing up to %wu bytes into a region of "
2882 "size %wu")
2883 : G_("%<%.*s%> directive output truncated writing "
2884 "up to %wu bytes into a region of size %wu"))
2885 : G_("%<%.*s%> directive writing up to %wu bytes "
2886 "into a region of size %wu"), (int) dir.len,
2887 target_to_host (hostdir, sizeof hostdir, dir.beg),
2888 res.max, navail);
2890 if (res.min == 0 && maxbytes <= res.max)
2891 /* This is a special case to avoid issuing the potentially
2892 confusing warning:
2893 writing 0 or more bytes into a region of size 0. */
2894 return fmtwarn (dirloc, argloc, NULL, info.warnopt (),
2895 info.bounded
2896 ? (maybe
2897 ? G_("%<%.*s%> directive output may be truncated "
2898 "writing likely %wu or more bytes into a "
2899 "region of size %wu")
2900 : G_("%<%.*s%> directive output truncated writing "
2901 "likely %wu or more bytes into a region of "
2902 "size %wu"))
2903 : G_("%<%.*s%> directive writing likely %wu or more "
2904 "bytes into a region of size %wu"), (int) dir.len,
2905 target_to_host (hostdir, sizeof hostdir, dir.beg),
2906 res.likely, navail);
2908 if (res.max < maxbytes)
2909 return fmtwarn (dirloc, argloc, NULL, info.warnopt (),
2910 info.bounded
2911 ? (maybe
2912 ? G_("%<%.*s%> directive output may be truncated "
2913 "writing between %wu and %wu bytes into a "
2914 "region of size %wu")
2915 : G_("%<%.*s%> directive output truncated "
2916 "writing between %wu and %wu bytes into a "
2917 "region of size %wu"))
2918 : G_("%<%.*s%> directive writing between %wu and "
2919 "%wu bytes into a region of size %wu"),
2920 (int) dir.len,
2921 target_to_host (hostdir, sizeof hostdir, dir.beg),
2922 res.min, res.max, navail);
2924 return fmtwarn (dirloc, argloc, NULL, info.warnopt (),
2925 info.bounded
2926 ? (maybe
2927 ? G_("%<%.*s%> directive output may be truncated "
2928 "writing %wu or more bytes into a region of "
2929 "size %wu")
2930 : G_("%<%.*s%> directive output truncated writing "
2931 "%wu or more bytes into a region of size %wu"))
2932 : G_("%<%.*s%> directive writing %wu or more bytes "
2933 "into a region of size %wu"), (int) dir.len,
2934 target_to_host (hostdir, sizeof hostdir, dir.beg),
2935 res.min, navail);
2938 /* The size of the destination region is a range. */
2940 if (target_to_host (*dir.beg) != '%')
2942 unsigned HOST_WIDE_INT navail = avail_range.max;
2944 /* For plain character directives (i.e., the format string itself)
2945 but not others, point the caret at the first character that's
2946 past the end of the destination. */
2947 if (navail < dir.len)
2948 dirloc.set_caret_index (dirloc.get_caret_idx () + navail);
2951 if (*dir.beg == '\0')
2953 gcc_assert (res.min == 1 && res.min == res.max);
2955 return fmtwarn (dirloc, UNKNOWN_LOCATION, NULL, info.warnopt (),
2956 info.bounded
2957 ? (maybe
2958 ? G_("%qE output may be truncated before the last "
2959 "format character")
2960 : G_("%qE output truncated before the last format "
2961 "character"))
2962 : (maybe
2963 ? G_("%qE may write a terminating nul past the end "
2964 "of the destination")
2965 : G_("%qE writing a terminating nul past the end "
2966 "of the destination")), info.func);
2969 if (res.min == res.max)
2971 const char *d = target_to_host (hostdir, sizeof hostdir, dir.beg);
2972 if (!info.bounded)
2973 return fmtwarn_n (dirloc, argloc, NULL, info.warnopt (), res.min,
2974 "%<%.*s%> directive writing %wu byte into a region "
2975 "of size between %wu and %wu",
2976 "%<%.*s%> directive writing %wu bytes into a region "
2977 "of size between %wu and %wu", (int) dir.len, d,
2978 res.min, avail_range.min, avail_range.max);
2979 else if (maybe)
2980 return fmtwarn_n (dirloc, argloc, NULL, info.warnopt (), res.min,
2981 "%<%.*s%> directive output may be truncated writing "
2982 "%wu byte into a region of size between %wu and %wu",
2983 "%<%.*s%> directive output may be truncated writing "
2984 "%wu bytes into a region of size between %wu and "
2985 "%wu", (int) dir.len, d, res.min, avail_range.min,
2986 avail_range.max);
2987 else
2988 return fmtwarn_n (dirloc, argloc, NULL, info.warnopt (), res.min,
2989 "%<%.*s%> directive output truncated writing %wu "
2990 "byte into a region of size between %wu and %wu",
2991 "%<%.*s%> directive output truncated writing %wu "
2992 "bytes into a region of size between %wu and %wu",
2993 (int) dir.len, d, res.min, avail_range.min,
2994 avail_range.max);
2997 if (res.min == 0 && res.max < maxbytes)
2998 return fmtwarn (dirloc, argloc, NULL, info.warnopt (),
2999 info.bounded
3000 ? (maybe
3001 ? G_("%<%.*s%> directive output may be truncated "
3002 "writing up to %wu bytes into a region of size "
3003 "between %wu and %wu")
3004 : G_("%<%.*s%> directive output truncated writing "
3005 "up to %wu bytes into a region of size between "
3006 "%wu and %wu"))
3007 : G_("%<%.*s%> directive writing up to %wu bytes "
3008 "into a region of size between %wu and %wu"),
3009 (int) dir.len,
3010 target_to_host (hostdir, sizeof hostdir, dir.beg),
3011 res.max, avail_range.min, avail_range.max);
3013 if (res.min == 0 && maxbytes <= res.max)
3014 /* This is a special case to avoid issuing the potentially confusing
3015 warning:
3016 writing 0 or more bytes into a region of size between 0 and N. */
3017 return fmtwarn (dirloc, argloc, NULL, info.warnopt (),
3018 info.bounded
3019 ? (maybe
3020 ? G_("%<%.*s%> directive output may be truncated "
3021 "writing likely %wu or more bytes into a region "
3022 "of size between %wu and %wu")
3023 : G_("%<%.*s%> directive output truncated writing "
3024 "likely %wu or more bytes into a region of size "
3025 "between %wu and %wu"))
3026 : G_("%<%.*s%> directive writing likely %wu or more bytes "
3027 "into a region of size between %wu and %wu"),
3028 (int) dir.len,
3029 target_to_host (hostdir, sizeof hostdir, dir.beg),
3030 res.likely, avail_range.min, avail_range.max);
3032 if (res.max < maxbytes)
3033 return fmtwarn (dirloc, argloc, NULL, info.warnopt (),
3034 info.bounded
3035 ? (maybe
3036 ? G_("%<%.*s%> directive output may be truncated "
3037 "writing between %wu and %wu bytes into a region "
3038 "of size between %wu and %wu")
3039 : G_("%<%.*s%> directive output truncated writing "
3040 "between %wu and %wu bytes into a region of size "
3041 "between %wu and %wu"))
3042 : G_("%<%.*s%> directive writing between %wu and "
3043 "%wu bytes into a region of size between %wu and "
3044 "%wu"), (int) dir.len,
3045 target_to_host (hostdir, sizeof hostdir, dir.beg),
3046 res.min, res.max, avail_range.min, avail_range.max);
3048 return fmtwarn (dirloc, argloc, NULL, info.warnopt (),
3049 info.bounded
3050 ? (maybe
3051 ? G_("%<%.*s%> directive output may be truncated writing "
3052 "%wu or more bytes into a region of size between "
3053 "%wu and %wu")
3054 : G_("%<%.*s%> directive output truncated writing "
3055 "%wu or more bytes into a region of size between "
3056 "%wu and %wu"))
3057 : G_("%<%.*s%> directive writing %wu or more bytes "
3058 "into a region of size between %wu and %wu"),
3059 (int) dir.len,
3060 target_to_host (hostdir, sizeof hostdir, dir.beg),
3061 res.min, avail_range.min, avail_range.max);
3064 /* Given the formatting result described by RES and NAVAIL, the number
3065 of available bytes in the destination, return the range of bytes
3066 remaining in the destination. */
3068 static inline result_range
3069 bytes_remaining (unsigned HOST_WIDE_INT navail, const format_result &res)
3071 result_range range;
3073 if (HOST_WIDE_INT_MAX <= navail)
3075 range.min = range.max = range.likely = range.unlikely = navail;
3076 return range;
3079 /* The lower bound of the available range is the available size
3080 minus the maximum output size, and the upper bound is the size
3081 minus the minimum. */
3082 range.max = res.range.min < navail ? navail - res.range.min : 0;
3084 range.likely = res.range.likely < navail ? navail - res.range.likely : 0;
3086 if (res.range.max < HOST_WIDE_INT_MAX)
3087 range.min = res.range.max < navail ? navail - res.range.max : 0;
3088 else
3089 range.min = range.likely;
3091 range.unlikely = (res.range.unlikely < navail
3092 ? navail - res.range.unlikely : 0);
3094 return range;
3097 /* Compute the length of the output resulting from the directive DIR
3098 in a call described by INFO and update the overall result of the call
3099 in *RES. Return true if the directive has been handled. */
3101 static bool
3102 format_directive (const call_info &info,
3103 format_result *res, const directive &dir,
3104 pointer_query &ptr_qry)
3106 /* Offset of the beginning of the directive from the beginning
3107 of the format string. */
3108 size_t offset = dir.beg - info.fmtstr;
3109 size_t start = offset;
3110 size_t length = offset + dir.len - !!dir.len;
3112 /* Create a location for the whole directive from the % to the format
3113 specifier. */
3114 substring_loc dirloc (info.fmtloc, TREE_TYPE (info.format),
3115 offset, start, length);
3117 /* Also get the location of the argument if possible.
3118 This doesn't work for integer literals or function calls. */
3119 location_t argloc = UNKNOWN_LOCATION;
3120 if (dir.arg)
3121 argloc = EXPR_LOCATION (dir.arg);
3123 /* Bail when there is no function to compute the output length,
3124 or when minimum length checking has been disabled. */
3125 if (!dir.fmtfunc || res->range.min >= HOST_WIDE_INT_MAX)
3126 return false;
3128 /* Compute the range of lengths of the formatted output. */
3129 fmtresult fmtres = dir.fmtfunc (dir, dir.arg, ptr_qry);
3131 /* Record whether the output of all directives is known to be
3132 bounded by some maximum, implying that their arguments are
3133 either known exactly or determined to be in a known range
3134 or, for strings, limited by the upper bounds of the arrays
3135 they refer to. */
3136 res->knownrange &= fmtres.knownrange;
3138 if (!fmtres.knownrange)
3140 /* Only when the range is known, check it against the host value
3141 of INT_MAX + (the number of bytes of the "%.*Lf" directive with
3142 INT_MAX precision, which is the longest possible output of any
3143 single directive). That's the largest valid byte count (though
3144 not valid call to a printf-like function because it can never
3145 return such a count). Otherwise, the range doesn't correspond
3146 to known values of the argument. */
3147 if (fmtres.range.max > target_dir_max ())
3149 /* Normalize the MAX counter to avoid having to deal with it
3150 later. The counter can be less than HOST_WIDE_INT_M1U
3151 when compiling for an ILP32 target on an LP64 host. */
3152 fmtres.range.max = HOST_WIDE_INT_M1U;
3153 /* Disable exact and maximum length checking after a failure
3154 to determine the maximum number of characters (for example
3155 for wide characters or wide character strings) but continue
3156 tracking the minimum number of characters. */
3157 res->range.max = HOST_WIDE_INT_M1U;
3160 if (fmtres.range.min > target_dir_max ())
3162 /* Disable exact length checking after a failure to determine
3163 even the minimum number of characters (it shouldn't happen
3164 except in an error) but keep tracking the minimum and maximum
3165 number of characters. */
3166 return true;
3170 /* Buffer for the directive in the host character set (used when
3171 the source character set is different). */
3172 char hostdir[32];
3174 int dirlen = dir.len;
3176 if (fmtres.nullp)
3178 fmtwarn (dirloc, argloc, NULL, info.warnopt (),
3179 "%<%.*s%> directive argument is null",
3180 dirlen, target_to_host (hostdir, sizeof hostdir, dir.beg));
3182 /* Don't bother processing the rest of the format string. */
3183 res->warned = true;
3184 res->range.min = HOST_WIDE_INT_M1U;
3185 res->range.max = HOST_WIDE_INT_M1U;
3186 return false;
3189 /* Compute the number of available bytes in the destination. There
3190 must always be at least one byte of space for the terminating
3191 NUL that's appended after the format string has been processed. */
3192 result_range avail_range = bytes_remaining (info.objsize, *res);
3194 /* If the argument aliases a part of the destination of the formatted
3195 call at offset FMTRES.DST_OFFSET append the directive and its result
3196 to the set of aliases for later processing. */
3197 if (fmtres.dst_offset != HOST_WIDE_INT_MIN)
3198 res->append_alias (dir, fmtres.dst_offset, fmtres.range);
3200 bool warned = res->warned;
3202 if (!warned)
3203 warned = maybe_warn (dirloc, argloc, info, avail_range,
3204 fmtres.range, dir);
3206 /* Bump up the total maximum if it isn't too big. */
3207 if (res->range.max < HOST_WIDE_INT_MAX
3208 && fmtres.range.max < HOST_WIDE_INT_MAX)
3209 res->range.max += fmtres.range.max;
3211 /* Raise the total unlikely maximum by the larger of the maximum
3212 and the unlikely maximum. */
3213 unsigned HOST_WIDE_INT save = res->range.unlikely;
3214 if (fmtres.range.max < fmtres.range.unlikely)
3215 res->range.unlikely += fmtres.range.unlikely;
3216 else
3217 res->range.unlikely += fmtres.range.max;
3219 if (res->range.unlikely < save)
3220 res->range.unlikely = HOST_WIDE_INT_M1U;
3222 res->range.min += fmtres.range.min;
3223 res->range.likely += fmtres.range.likely;
3225 /* Has the minimum directive output length exceeded the maximum
3226 of 4095 bytes required to be supported? */
3227 bool minunder4k = fmtres.range.min < 4096;
3228 bool maxunder4k = fmtres.range.max < 4096;
3229 /* Clear POSUNDER4K in the overall result if the maximum has exceeded
3230 the 4k (this is necessary to avoid the return value optimization
3231 that may not be safe in the maximum case). */
3232 if (!maxunder4k)
3233 res->posunder4k = false;
3234 /* Also clear POSUNDER4K if the directive may fail. */
3235 if (fmtres.mayfail)
3236 res->posunder4k = false;
3238 if (!warned
3239 /* Only warn at level 2. */
3240 && warn_level > 1
3241 /* Only warn for string functions. */
3242 && info.is_string_func ()
3243 && (!minunder4k
3244 || (!maxunder4k && fmtres.range.max < HOST_WIDE_INT_MAX)))
3246 /* The directive output may be longer than the maximum required
3247 to be handled by an implementation according to 7.21.6.1, p15
3248 of C11. Warn on this only at level 2 but remember this and
3249 prevent folding the return value when done. This allows for
3250 the possibility of the actual libc call failing due to ENOMEM
3251 (like Glibc does with very large precision or width).
3252 Issue the "may exceed" warning only for string functions and
3253 not for fprintf or printf. */
3255 if (fmtres.range.min == fmtres.range.max)
3256 warned = fmtwarn (dirloc, argloc, NULL, info.warnopt (),
3257 "%<%.*s%> directive output of %wu bytes exceeds "
3258 "minimum required size of 4095", dirlen,
3259 target_to_host (hostdir, sizeof hostdir, dir.beg),
3260 fmtres.range.min);
3261 else if (!minunder4k)
3262 warned = fmtwarn (dirloc, argloc, NULL, info.warnopt (),
3263 "%<%.*s%> directive output between %wu and %wu "
3264 "bytes exceeds minimum required size of 4095",
3265 dirlen,
3266 target_to_host (hostdir, sizeof hostdir, dir.beg),
3267 fmtres.range.min, fmtres.range.max);
3268 else if (!info.retval_used () && info.is_string_func ())
3269 warned = fmtwarn (dirloc, argloc, NULL, info.warnopt (),
3270 "%<%.*s%> directive output between %wu and %wu "
3271 "bytes may exceed minimum required size of "
3272 "4095",
3273 dirlen,
3274 target_to_host (hostdir, sizeof hostdir, dir.beg),
3275 fmtres.range.min, fmtres.range.max);
3278 /* Has the likely and maximum directive output exceeded INT_MAX? */
3279 bool likelyximax = *dir.beg && res->range.likely > target_int_max ();
3280 /* Don't consider the maximum to be in excess when it's the result
3281 of a string of unknown length (i.e., whose maximum has been set
3282 to be greater than or equal to HOST_WIDE_INT_MAX. */
3283 bool maxximax = (*dir.beg
3284 && res->range.max > target_int_max ()
3285 && res->range.max < HOST_WIDE_INT_MAX);
3287 if (!warned
3288 /* Warn for the likely output size at level 1. */
3289 && (likelyximax
3290 /* But only warn for the maximum at level 2. */
3291 || (warn_level > 1
3292 && maxximax
3293 && fmtres.range.max < HOST_WIDE_INT_MAX)))
3295 if (fmtres.range.min > target_int_max ())
3297 /* The directive output exceeds INT_MAX bytes. */
3298 if (fmtres.range.min == fmtres.range.max)
3299 warned = fmtwarn (dirloc, argloc, NULL, info.warnopt (),
3300 "%<%.*s%> directive output of %wu bytes exceeds "
3301 "%<INT_MAX%>", dirlen,
3302 target_to_host (hostdir, sizeof hostdir, dir.beg),
3303 fmtres.range.min);
3304 else
3305 warned = fmtwarn (dirloc, argloc, NULL, info.warnopt (),
3306 "%<%.*s%> directive output between %wu and "
3307 "%wu bytes exceeds %<INT_MAX%>", dirlen,
3308 target_to_host (hostdir, sizeof hostdir, dir.beg),
3309 fmtres.range.min, fmtres.range.max);
3311 else if (res->range.min > target_int_max ())
3313 /* The directive output is under INT_MAX but causes the result
3314 to exceed INT_MAX bytes. */
3315 if (fmtres.range.min == fmtres.range.max)
3316 warned = fmtwarn (dirloc, argloc, NULL, info.warnopt (),
3317 "%<%.*s%> directive output of %wu bytes causes "
3318 "result to exceed %<INT_MAX%>", dirlen,
3319 target_to_host (hostdir, sizeof hostdir, dir.beg),
3320 fmtres.range.min);
3321 else
3322 warned = fmtwarn (dirloc, argloc, NULL, info.warnopt (),
3323 "%<%.*s%> directive output between %wu and "
3324 "%wu bytes causes result to exceed %<INT_MAX%>",
3325 dirlen,
3326 target_to_host (hostdir, sizeof hostdir, dir.beg),
3327 fmtres.range.min, fmtres.range.max);
3329 else if ((!info.retval_used () || !info.bounded)
3330 && (info.is_string_func ()))
3331 /* Warn for calls to string functions that either aren't bounded
3332 (sprintf) or whose return value isn't used. */
3333 warned = fmtwarn (dirloc, argloc, NULL, info.warnopt (),
3334 "%<%.*s%> directive output between %wu and "
3335 "%wu bytes may cause result to exceed "
3336 "%<INT_MAX%>", dirlen,
3337 target_to_host (hostdir, sizeof hostdir, dir.beg),
3338 fmtres.range.min, fmtres.range.max);
3341 if (!warned && fmtres.nonstr)
3343 warned = fmtwarn (dirloc, argloc, NULL, info.warnopt (),
3344 "%<%.*s%> directive argument is not a nul-terminated "
3345 "string",
3346 dirlen,
3347 target_to_host (hostdir, sizeof hostdir, dir.beg));
3348 if (warned && DECL_P (fmtres.nonstr))
3349 inform (DECL_SOURCE_LOCATION (fmtres.nonstr),
3350 "referenced argument declared here");
3351 return false;
3354 if (warned && fmtres.range.min < fmtres.range.likely
3355 && fmtres.range.likely < fmtres.range.max)
3356 inform_n (info.fmtloc, fmtres.range.likely,
3357 "assuming directive output of %wu byte",
3358 "assuming directive output of %wu bytes",
3359 fmtres.range.likely);
3361 if (warned && fmtres.argmin)
3363 if (fmtres.argmin == fmtres.argmax)
3364 inform (info.fmtloc, "directive argument %qE", fmtres.argmin);
3365 else if (fmtres.knownrange)
3366 inform (info.fmtloc, "directive argument in the range [%E, %E]",
3367 fmtres.argmin, fmtres.argmax);
3368 else
3369 inform (info.fmtloc,
3370 "using the range [%E, %E] for directive argument",
3371 fmtres.argmin, fmtres.argmax);
3374 res->warned |= warned;
3376 if (!dir.beg[0] && res->warned)
3378 location_t callloc = gimple_location (info.callstmt);
3380 unsigned HOST_WIDE_INT min = res->range.min;
3381 unsigned HOST_WIDE_INT max = res->range.max;
3383 if (info.objsize < HOST_WIDE_INT_MAX)
3385 /* If a warning has been issued for buffer overflow or truncation
3386 help the user figure out how big a buffer they need. */
3388 if (min == max)
3389 inform_n (callloc, min,
3390 "%qE output %wu byte into a destination of size %wu",
3391 "%qE output %wu bytes into a destination of size %wu",
3392 info.func, min, info.objsize);
3393 else if (max < HOST_WIDE_INT_MAX)
3394 inform (callloc,
3395 "%qE output between %wu and %wu bytes into "
3396 "a destination of size %wu",
3397 info.func, min, max, info.objsize);
3398 else if (min < res->range.likely && res->range.likely < max)
3399 inform (callloc,
3400 "%qE output %wu or more bytes (assuming %wu) into "
3401 "a destination of size %wu",
3402 info.func, min, res->range.likely, info.objsize);
3403 else
3404 inform (callloc,
3405 "%qE output %wu or more bytes into a destination of size "
3406 "%wu",
3407 info.func, min, info.objsize);
3409 else if (!info.is_string_func ())
3411 /* If the warning is for a file function like fprintf
3412 of printf with no destination size just print the computed
3413 result. */
3414 if (min == max)
3415 inform_n (callloc, min,
3416 "%qE output %wu byte", "%qE output %wu bytes",
3417 info.func, min);
3418 else if (max < HOST_WIDE_INT_MAX)
3419 inform (callloc,
3420 "%qE output between %wu and %wu bytes",
3421 info.func, min, max);
3422 else if (min < res->range.likely && res->range.likely < max)
3423 inform (callloc,
3424 "%qE output %wu or more bytes (assuming %wu)",
3425 info.func, min, res->range.likely);
3426 else
3427 inform (callloc,
3428 "%qE output %wu or more bytes",
3429 info.func, min);
3433 if (dump_file && *dir.beg)
3435 fprintf (dump_file,
3436 " Result: "
3437 HOST_WIDE_INT_PRINT_DEC ", " HOST_WIDE_INT_PRINT_DEC ", "
3438 HOST_WIDE_INT_PRINT_DEC ", " HOST_WIDE_INT_PRINT_DEC " ("
3439 HOST_WIDE_INT_PRINT_DEC ", " HOST_WIDE_INT_PRINT_DEC ", "
3440 HOST_WIDE_INT_PRINT_DEC ", " HOST_WIDE_INT_PRINT_DEC ")\n",
3441 fmtres.range.min, fmtres.range.likely,
3442 fmtres.range.max, fmtres.range.unlikely,
3443 res->range.min, res->range.likely,
3444 res->range.max, res->range.unlikely);
3447 return true;
3450 /* Parse a format directive in function call described by INFO starting
3451 at STR and populate DIR structure. Bump up *ARGNO by the number of
3452 arguments extracted for the directive. Return the length of
3453 the directive. */
3455 static size_t
3456 parse_directive (call_info &info,
3457 directive &dir, format_result *res,
3458 const char *str, unsigned *argno,
3459 range_query *query)
3461 const char *pcnt = strchr (str, target_percent);
3462 dir.beg = str;
3464 if (size_t len = pcnt ? pcnt - str : *str ? strlen (str) : 1)
3466 /* This directive is either a plain string or the terminating nul
3467 (which isn't really a directive but it simplifies things to
3468 handle it as if it were). */
3469 dir.len = len;
3470 dir.fmtfunc = format_plain;
3472 if (dump_file)
3474 fprintf (dump_file, " Directive %u at offset "
3475 HOST_WIDE_INT_PRINT_UNSIGNED ": \"%.*s\", "
3476 "length = " HOST_WIDE_INT_PRINT_UNSIGNED "\n",
3477 dir.dirno,
3478 (unsigned HOST_WIDE_INT)(size_t)(dir.beg - info.fmtstr),
3479 (int)dir.len, dir.beg, (unsigned HOST_WIDE_INT) dir.len);
3482 return len - !*str;
3485 /* Set the directive argument's number to correspond to its position
3486 in the formatted function call's argument list. */
3487 dir.argno = *argno;
3489 const char *pf = pcnt + 1;
3491 /* POSIX numbered argument index or zero when none. */
3492 HOST_WIDE_INT dollar = 0;
3494 /* With and precision. -1 when not specified, HOST_WIDE_INT_MIN
3495 when given by a va_list argument, and a non-negative value
3496 when specified in the format string itself. */
3497 HOST_WIDE_INT width = -1;
3498 HOST_WIDE_INT precision = -1;
3500 /* Pointers to the beginning of the width and precision decimal
3501 string (if any) within the directive. */
3502 const char *pwidth = 0;
3503 const char *pprec = 0;
3505 /* When the value of the decimal string that specifies width or
3506 precision is out of range, points to the digit that causes
3507 the value to exceed the limit. */
3508 const char *werange = NULL;
3509 const char *perange = NULL;
3511 /* Width specified via the asterisk. Need not be INTEGER_CST.
3512 For vararg functions set to void_node. */
3513 tree star_width = NULL_TREE;
3515 /* Width specified via the asterisk. Need not be INTEGER_CST.
3516 For vararg functions set to void_node. */
3517 tree star_precision = NULL_TREE;
3519 if (ISDIGIT (target_to_host (*pf)))
3521 /* This could be either a POSIX positional argument, the '0'
3522 flag, or a width, depending on what follows. Store it as
3523 width and sort it out later after the next character has
3524 been seen. */
3525 pwidth = pf;
3526 width = target_strtowi (&pf, &werange);
3528 else if (target_to_host (*pf) == '*')
3530 /* Similarly to the block above, this could be either a POSIX
3531 positional argument or a width, depending on what follows. */
3532 if (*argno < gimple_call_num_args (info.callstmt))
3533 star_width = gimple_call_arg (info.callstmt, (*argno)++);
3534 else
3535 star_width = void_node;
3536 ++pf;
3539 if (target_to_host (*pf) == '$')
3541 /* Handle the POSIX dollar sign which references the 1-based
3542 positional argument number. */
3543 if (width != -1)
3544 dollar = width + info.argidx;
3545 else if (star_width
3546 && TREE_CODE (star_width) == INTEGER_CST
3547 && (TYPE_PRECISION (TREE_TYPE (star_width))
3548 <= TYPE_PRECISION (integer_type_node)))
3549 dollar = width + tree_to_shwi (star_width);
3551 /* Bail when the numbered argument is out of range (it will
3552 have already been diagnosed by -Wformat). */
3553 if (dollar == 0
3554 || dollar == (int)info.argidx
3555 || dollar > gimple_call_num_args (info.callstmt))
3556 return false;
3558 --dollar;
3560 star_width = NULL_TREE;
3561 width = -1;
3562 ++pf;
3565 if (dollar || !star_width)
3567 if (width != -1)
3569 if (width == 0)
3571 /* The '0' that has been interpreted as a width above is
3572 actually a flag. Reset HAVE_WIDTH, set the '0' flag,
3573 and continue processing other flags. */
3574 width = -1;
3575 dir.set_flag ('0');
3577 else if (!dollar)
3579 /* (Non-zero) width has been seen. The next character
3580 is either a period or a digit. */
3581 goto start_precision;
3584 /* When either '$' has been seen, or width has not been seen,
3585 the next field is the optional flags followed by an optional
3586 width. */
3587 for ( ; ; ) {
3588 switch (target_to_host (*pf))
3590 case ' ':
3591 case '0':
3592 case '+':
3593 case '-':
3594 case '#':
3595 dir.set_flag (target_to_host (*pf++));
3596 break;
3598 default:
3599 goto start_width;
3603 start_width:
3604 if (ISDIGIT (target_to_host (*pf)))
3606 werange = 0;
3607 pwidth = pf;
3608 width = target_strtowi (&pf, &werange);
3610 else if (target_to_host (*pf) == '*')
3612 if (*argno < gimple_call_num_args (info.callstmt))
3613 star_width = gimple_call_arg (info.callstmt, (*argno)++);
3614 else
3616 /* This is (likely) a va_list. It could also be an invalid
3617 call with insufficient arguments. */
3618 star_width = void_node;
3620 ++pf;
3622 else if (target_to_host (*pf) == '\'')
3624 /* The POSIX apostrophe indicating a numeric grouping
3625 in the current locale. Even though it's possible to
3626 estimate the upper bound on the size of the output
3627 based on the number of digits it probably isn't worth
3628 continuing. */
3629 return 0;
3633 start_precision:
3634 if (target_to_host (*pf) == '.')
3636 ++pf;
3638 if (ISDIGIT (target_to_host (*pf)))
3640 pprec = pf;
3641 precision = target_strtowi (&pf, &perange);
3643 else if (target_to_host (*pf) == '*')
3645 if (*argno < gimple_call_num_args (info.callstmt))
3646 star_precision = gimple_call_arg (info.callstmt, (*argno)++);
3647 else
3649 /* This is (likely) a va_list. It could also be an invalid
3650 call with insufficient arguments. */
3651 star_precision = void_node;
3653 ++pf;
3655 else
3657 /* The decimal precision or the asterisk are optional.
3658 When neither is specified it's taken to be zero. */
3659 precision = 0;
3663 switch (target_to_host (*pf))
3665 case 'h':
3666 if (target_to_host (pf[1]) == 'h')
3668 ++pf;
3669 dir.modifier = FMT_LEN_hh;
3671 else
3672 dir.modifier = FMT_LEN_h;
3673 ++pf;
3674 break;
3676 case 'j':
3677 dir.modifier = FMT_LEN_j;
3678 ++pf;
3679 break;
3681 case 'L':
3682 dir.modifier = FMT_LEN_L;
3683 ++pf;
3684 break;
3686 case 'l':
3687 if (target_to_host (pf[1]) == 'l')
3689 ++pf;
3690 dir.modifier = FMT_LEN_ll;
3692 else
3693 dir.modifier = FMT_LEN_l;
3694 ++pf;
3695 break;
3697 case 't':
3698 dir.modifier = FMT_LEN_t;
3699 ++pf;
3700 break;
3702 case 'z':
3703 dir.modifier = FMT_LEN_z;
3704 ++pf;
3705 break;
3708 switch (target_to_host (*pf))
3710 /* Handle a sole '%' character the same as "%%" but since it's
3711 undefined prevent the result from being folded. */
3712 case '\0':
3713 --pf;
3714 res->range.min = res->range.max = HOST_WIDE_INT_M1U;
3715 /* FALLTHRU */
3716 case '%':
3717 dir.fmtfunc = format_percent;
3718 break;
3720 case 'a':
3721 case 'A':
3722 case 'e':
3723 case 'E':
3724 case 'f':
3725 case 'F':
3726 case 'g':
3727 case 'G':
3728 res->floating = true;
3729 dir.fmtfunc = format_floating;
3730 break;
3732 case 'd':
3733 case 'i':
3734 case 'o':
3735 case 'u':
3736 case 'x':
3737 case 'X':
3738 dir.fmtfunc = format_integer;
3739 break;
3741 case 'b':
3742 case 'B':
3743 dir.fmtfunc = format_integer;
3744 break;
3746 case 'p':
3747 /* The %p output is implementation-defined. It's possible
3748 to determine this format but due to extensions (especially
3749 those of the Linux kernel -- see bug 78512) the first %p
3750 in the format string disables any further processing. */
3751 return false;
3753 case 'n':
3754 /* %n has side-effects even when nothing is actually printed to
3755 any buffer. */
3756 info.nowrite = false;
3757 dir.fmtfunc = format_none;
3758 break;
3760 case 'C':
3761 case 'c':
3762 /* POSIX wide character and C/POSIX narrow character. */
3763 dir.fmtfunc = format_character;
3764 break;
3766 case 'S':
3767 case 's':
3768 /* POSIX wide string and C/POSIX narrow character string. */
3769 dir.fmtfunc = format_string;
3770 break;
3772 default:
3773 /* Unknown conversion specification. */
3774 return 0;
3777 dir.specifier = target_to_host (*pf++);
3779 /* Store the length of the format directive. */
3780 dir.len = pf - pcnt;
3782 /* Buffer for the directive in the host character set (used when
3783 the source character set is different). */
3784 char hostdir[32];
3786 if (star_width)
3788 if (INTEGRAL_TYPE_P (TREE_TYPE (star_width)))
3789 dir.set_width (star_width, query);
3790 else
3792 /* Width specified by a va_list takes on the range [0, -INT_MIN]
3793 (width is the absolute value of that specified). */
3794 dir.width[0] = 0;
3795 dir.width[1] = target_int_max () + 1;
3798 else
3800 if (width == HOST_WIDE_INT_MAX && werange)
3802 size_t begin = dir.beg - info.fmtstr + (pwidth - pcnt);
3803 size_t caret = begin + (werange - pcnt);
3804 size_t end = pf - info.fmtstr - 1;
3806 /* Create a location for the width part of the directive,
3807 pointing the caret at the first out-of-range digit. */
3808 substring_loc dirloc (info.fmtloc, TREE_TYPE (info.format),
3809 caret, begin, end);
3811 fmtwarn (dirloc, UNKNOWN_LOCATION, NULL, info.warnopt (),
3812 "%<%.*s%> directive width out of range", (int) dir.len,
3813 target_to_host (hostdir, sizeof hostdir, dir.beg));
3816 dir.set_width (width);
3819 if (star_precision)
3821 if (INTEGRAL_TYPE_P (TREE_TYPE (star_precision)))
3822 dir.set_precision (star_precision, query);
3823 else
3825 /* Precision specified by a va_list takes on the range [-1, INT_MAX]
3826 (unlike width, negative precision is ignored). */
3827 dir.prec[0] = -1;
3828 dir.prec[1] = target_int_max ();
3831 else
3833 if (precision == HOST_WIDE_INT_MAX && perange)
3835 size_t begin = dir.beg - info.fmtstr + (pprec - pcnt) - 1;
3836 size_t caret = dir.beg - info.fmtstr + (perange - pcnt) - 1;
3837 size_t end = pf - info.fmtstr - 2;
3839 /* Create a location for the precision part of the directive,
3840 including the leading period, pointing the caret at the first
3841 out-of-range digit . */
3842 substring_loc dirloc (info.fmtloc, TREE_TYPE (info.format),
3843 caret, begin, end);
3845 fmtwarn (dirloc, UNKNOWN_LOCATION, NULL, info.warnopt (),
3846 "%<%.*s%> directive precision out of range", (int) dir.len,
3847 target_to_host (hostdir, sizeof hostdir, dir.beg));
3850 dir.set_precision (precision);
3853 /* Extract the argument if the directive takes one and if it's
3854 available (e.g., the function doesn't take a va_list). Treat
3855 missing arguments the same as va_list, even though they will
3856 have likely already been diagnosed by -Wformat. */
3857 if (dir.specifier != '%'
3858 && *argno < gimple_call_num_args (info.callstmt))
3859 dir.arg = gimple_call_arg (info.callstmt, dollar ? dollar : (*argno)++);
3861 if (dump_file)
3863 fprintf (dump_file,
3864 " Directive %u at offset " HOST_WIDE_INT_PRINT_UNSIGNED
3865 ": \"%.*s\"",
3866 dir.dirno,
3867 (unsigned HOST_WIDE_INT)(size_t)(dir.beg - info.fmtstr),
3868 (int)dir.len, dir.beg);
3869 if (star_width)
3871 if (dir.width[0] == dir.width[1])
3872 fprintf (dump_file, ", width = " HOST_WIDE_INT_PRINT_DEC,
3873 dir.width[0]);
3874 else
3875 fprintf (dump_file,
3876 ", width in range [" HOST_WIDE_INT_PRINT_DEC
3877 ", " HOST_WIDE_INT_PRINT_DEC "]",
3878 dir.width[0], dir.width[1]);
3881 if (star_precision)
3883 if (dir.prec[0] == dir.prec[1])
3884 fprintf (dump_file, ", precision = " HOST_WIDE_INT_PRINT_DEC,
3885 dir.prec[0]);
3886 else
3887 fprintf (dump_file,
3888 ", precision in range [" HOST_WIDE_INT_PRINT_DEC
3889 HOST_WIDE_INT_PRINT_DEC "]",
3890 dir.prec[0], dir.prec[1]);
3892 fputc ('\n', dump_file);
3895 return dir.len;
3898 /* Diagnose overlap between destination and %s directive arguments. */
3900 static void
3901 maybe_warn_overlap (call_info &info, format_result *res)
3903 /* Two vectors of 1-based indices corresponding to either certainly
3904 or possibly aliasing arguments. */
3905 auto_vec<int, 16> aliasarg[2];
3907 /* Go through the array of potentially aliasing directives and collect
3908 argument numbers of those that do or may overlap the destination
3909 object given the full result. */
3910 for (unsigned i = 0; i != res->alias_count; ++i)
3912 const format_result::alias_info &alias = res->aliases[i];
3914 enum { possible = -1, none = 0, certain = 1 } overlap = none;
3916 /* If the precision is zero there is no overlap. (This only
3917 considers %s directives and ignores %n.) */
3918 if (alias.dir.prec[0] == 0 && alias.dir.prec[1] == 0)
3919 continue;
3921 if (alias.offset == HOST_WIDE_INT_MAX
3922 || info.dst_offset == HOST_WIDE_INT_MAX)
3923 overlap = possible;
3924 else if (alias.offset == info.dst_offset)
3925 overlap = alias.dir.prec[0] == 0 ? possible : certain;
3926 else
3928 /* Determine overlap from the range of output and offsets
3929 into the same destination as the source, and rule out
3930 impossible overlap. */
3931 unsigned HOST_WIDE_INT albeg = alias.offset;
3932 unsigned HOST_WIDE_INT dstbeg = info.dst_offset;
3934 unsigned HOST_WIDE_INT alend = albeg + alias.range.min;
3935 unsigned HOST_WIDE_INT dstend = dstbeg + res->range.min - 1;
3937 if ((albeg <= dstbeg && alend > dstbeg)
3938 || (albeg >= dstbeg && albeg < dstend))
3939 overlap = certain;
3940 else
3942 alend = albeg + alias.range.max;
3943 if (alend < albeg)
3944 alend = HOST_WIDE_INT_M1U;
3946 dstend = dstbeg + res->range.max - 1;
3947 if (dstend < dstbeg)
3948 dstend = HOST_WIDE_INT_M1U;
3950 if ((albeg >= dstbeg && albeg <= dstend)
3951 || (alend >= dstbeg && alend <= dstend))
3952 overlap = possible;
3956 if (overlap == none)
3957 continue;
3959 /* Append the 1-based argument number. */
3960 aliasarg[overlap != certain].safe_push (alias.dir.argno + 1);
3962 /* Disable any kind of optimization. */
3963 res->range.unlikely = HOST_WIDE_INT_M1U;
3966 tree arg0 = gimple_call_arg (info.callstmt, 0);
3967 location_t loc = gimple_location (info.callstmt);
3969 bool aliaswarn = false;
3971 unsigned ncertain = aliasarg[0].length ();
3972 unsigned npossible = aliasarg[1].length ();
3973 if (ncertain && npossible)
3975 /* If there are multiple arguments that overlap, some certainly
3976 and some possibly, handle both sets in a single diagnostic. */
3977 aliaswarn
3978 = warning_at (loc, OPT_Wrestrict,
3979 "%qE arguments %Z and maybe %Z overlap destination "
3980 "object %qE",
3981 info.func, aliasarg[0].address (), ncertain,
3982 aliasarg[1].address (), npossible,
3983 info.dst_origin);
3985 else if (ncertain)
3987 /* There is only one set of two or more arguments and they all
3988 certainly overlap the destination. */
3989 aliaswarn
3990 = warning_n (loc, OPT_Wrestrict, ncertain,
3991 "%qE argument %Z overlaps destination object %qE",
3992 "%qE arguments %Z overlap destination object %qE",
3993 info.func, aliasarg[0].address (), ncertain,
3994 info.dst_origin);
3996 else if (npossible)
3998 /* There is only one set of two or more arguments and they all
3999 may overlap (but need not). */
4000 aliaswarn
4001 = warning_n (loc, OPT_Wrestrict, npossible,
4002 "%qE argument %Z may overlap destination object %qE",
4003 "%qE arguments %Z may overlap destination object %qE",
4004 info.func, aliasarg[1].address (), npossible,
4005 info.dst_origin);
4008 if (aliaswarn)
4010 res->warned = true;
4012 if (info.dst_origin != arg0)
4014 /* If its location is different from the first argument of the call
4015 point either at the destination object itself or at the expression
4016 that was used to determine the overlap. */
4017 loc = (DECL_P (info.dst_origin)
4018 ? DECL_SOURCE_LOCATION (info.dst_origin)
4019 : EXPR_LOCATION (info.dst_origin));
4020 if (loc != UNKNOWN_LOCATION)
4021 inform (loc,
4022 "destination object referenced by %<restrict%>-qualified "
4023 "argument 1 was declared here");
4028 /* Compute the length of the output resulting from the call to a formatted
4029 output function described by INFO and store the result of the call in
4030 *RES. Issue warnings for detected past the end writes. Return true
4031 if the complete format string has been processed and *RES can be relied
4032 on, false otherwise (e.g., when a unknown or unhandled directive was seen
4033 that caused the processing to be terminated early). */
4035 static bool
4036 compute_format_length (call_info &info, format_result *res,
4037 pointer_query &ptr_qry)
4039 if (dump_file)
4041 location_t callloc = gimple_location (info.callstmt);
4042 fprintf (dump_file, "%s:%i: ",
4043 LOCATION_FILE (callloc), LOCATION_LINE (callloc));
4044 print_generic_expr (dump_file, info.func, dump_flags);
4046 fprintf (dump_file,
4047 ": objsize = " HOST_WIDE_INT_PRINT_UNSIGNED
4048 ", fmtstr = \"%s\"\n",
4049 info.objsize, info.fmtstr);
4052 /* Reset the minimum and maximum byte counters. */
4053 res->range.min = res->range.max = 0;
4055 /* No directive has been seen yet so the length of output is bounded
4056 by the known range [0, 0] (with no conversion resulting in a failure
4057 or producing more than 4K bytes) until determined otherwise. */
4058 res->knownrange = true;
4059 res->floating = false;
4060 res->warned = false;
4062 /* 1-based directive counter. */
4063 unsigned dirno = 1;
4065 /* The variadic argument counter. */
4066 unsigned argno = info.argidx;
4068 bool success = true;
4070 for (const char *pf = info.fmtstr; ; ++dirno)
4072 directive dir (&info, dirno);
4074 size_t n = parse_directive (info, dir, res, pf, &argno, ptr_qry.rvals);
4076 /* Return failure if the format function fails. */
4077 if (!format_directive (info, res, dir, ptr_qry))
4078 return false;
4080 /* Return success when the directive is zero bytes long and it's
4081 the last thing in the format string (i.e., it's the terminating
4082 nul, which isn't really a directive but handling it as one makes
4083 things simpler). */
4084 if (!n)
4086 success = *pf == '\0';
4087 break;
4090 pf += n;
4093 maybe_warn_overlap (info, res);
4095 /* The complete format string was processed (with or without warnings). */
4096 return success;
4099 /* Return the size of the object referenced by the expression DEST in
4100 statement STMT, if available, or the maximum possible size otherwise. */
4102 static unsigned HOST_WIDE_INT
4103 get_destination_size (tree dest, gimple *stmt, pointer_query &ptr_qry)
4105 /* When there is no destination return the maximum. */
4106 if (!dest)
4107 return HOST_WIDE_INT_MAX;
4109 /* Use compute_objsize to determine the size of the destination object. */
4110 access_ref aref;
4111 if (!ptr_qry.get_ref (dest, stmt, &aref))
4112 return HOST_WIDE_INT_MAX;
4114 offset_int remsize = aref.size_remaining ();
4115 if (!wi::fits_uhwi_p (remsize))
4116 return HOST_WIDE_INT_MAX;
4118 return remsize.to_uhwi ();
4121 /* Return true if the call described by INFO with result RES safe to
4122 optimize (i.e., no undefined behavior), and set RETVAL to the range
4123 of its return values. */
4125 static bool
4126 is_call_safe (const call_info &info,
4127 const format_result &res, bool under4k,
4128 unsigned HOST_WIDE_INT retval[2])
4130 if (under4k && !res.posunder4k)
4131 return false;
4133 /* The minimum return value. */
4134 retval[0] = res.range.min;
4136 /* The maximum return value is in most cases bounded by RES.RANGE.MAX
4137 but in cases involving multibyte characters could be as large as
4138 RES.RANGE.UNLIKELY. */
4139 retval[1]
4140 = res.range.unlikely < res.range.max ? res.range.max : res.range.unlikely;
4142 /* Adjust the number of bytes which includes the terminating nul
4143 to reflect the return value of the function which does not.
4144 Because the valid range of the function is [INT_MIN, INT_MAX],
4145 a valid range before the adjustment below is [0, INT_MAX + 1]
4146 (the functions only return negative values on error or undefined
4147 behavior). */
4148 if (retval[0] <= target_int_max () + 1)
4149 --retval[0];
4150 if (retval[1] <= target_int_max () + 1)
4151 --retval[1];
4153 /* Avoid the return value optimization when the behavior of the call
4154 is undefined either because any directive may have produced 4K or
4155 more of output, or the return value exceeds INT_MAX, or because
4156 the output overflows the destination object (but leave it enabled
4157 when the function is bounded because then the behavior is well-
4158 defined). */
4159 if (retval[0] == retval[1]
4160 && (info.bounded || retval[0] < info.objsize)
4161 && retval[0] <= target_int_max ())
4162 return true;
4164 if ((info.bounded || retval[1] < info.objsize)
4165 && (retval[0] < target_int_max ()
4166 && retval[1] < target_int_max ()))
4167 return true;
4169 if (!under4k && (info.bounded || retval[0] < info.objsize))
4170 return true;
4172 return false;
4175 /* Given a suitable result RES of a call to a formatted output function
4176 described by INFO, substitute the result for the return value of
4177 the call. The result is suitable if the number of bytes it represents
4178 is known and exact. A result that isn't suitable for substitution may
4179 have its range set to the range of return values, if that is known.
4180 Return true if the call is removed and gsi_next should not be performed
4181 in the caller. */
4183 static bool
4184 try_substitute_return_value (gimple_stmt_iterator *gsi,
4185 const call_info &info,
4186 const format_result &res)
4188 tree lhs = gimple_get_lhs (info.callstmt);
4190 /* Set to true when the entire call has been removed. */
4191 bool removed = false;
4193 /* The minimum and maximum return value. */
4194 unsigned HOST_WIDE_INT retval[2] = {0};
4195 bool safe = is_call_safe (info, res, true, retval);
4197 if (safe
4198 && retval[0] == retval[1]
4199 /* Not prepared to handle possibly throwing calls here; they shouldn't
4200 appear in non-artificial testcases, except when the __*_chk routines
4201 are badly declared. */
4202 && !stmt_ends_bb_p (info.callstmt))
4204 tree cst = build_int_cst (lhs ? TREE_TYPE (lhs) : integer_type_node,
4205 retval[0]);
4207 if (lhs == NULL_TREE && info.nowrite)
4209 /* Remove the call to the bounded function with a zero size
4210 (e.g., snprintf(0, 0, "%i", 123)) if there is no lhs. */
4211 unlink_stmt_vdef (info.callstmt);
4212 gsi_remove (gsi, true);
4213 removed = true;
4215 else if (info.nowrite)
4217 /* Replace the call to the bounded function with a zero size
4218 (e.g., snprintf(0, 0, "%i", 123) with the constant result
4219 of the function. */
4220 gimplify_and_update_call_from_tree (gsi, cst);
4221 gimple *callstmt = gsi_stmt (*gsi);
4222 update_stmt (callstmt);
4224 else if (lhs)
4226 /* Replace the left-hand side of the call with the constant
4227 result of the formatted function. */
4228 gimple_call_set_lhs (info.callstmt, NULL_TREE);
4229 gimple *g = gimple_build_assign (lhs, cst);
4230 gsi_insert_after (gsi, g, GSI_NEW_STMT);
4231 update_stmt (info.callstmt);
4234 if (dump_file)
4236 if (removed)
4237 fprintf (dump_file, " Removing call statement.");
4238 else
4240 fprintf (dump_file, " Substituting ");
4241 print_generic_expr (dump_file, cst, dump_flags);
4242 fprintf (dump_file, " for %s.\n",
4243 info.nowrite ? "statement" : "return value");
4247 else if (lhs && types_compatible_p (TREE_TYPE (lhs), integer_type_node))
4249 bool setrange = false;
4251 if (safe
4252 && (info.bounded || retval[1] < info.objsize)
4253 && (retval[0] < target_int_max ()
4254 && retval[1] < target_int_max ()))
4256 /* If the result is in a valid range bounded by the size of
4257 the destination set it so that it can be used for subsequent
4258 optimizations. */
4259 int prec = TYPE_PRECISION (integer_type_node);
4261 wide_int min = wi::shwi (retval[0], prec);
4262 wide_int max = wi::shwi (retval[1], prec);
4263 int_range_max r (TREE_TYPE (lhs), min, max);
4264 set_range_info (lhs, r);
4266 setrange = true;
4269 if (dump_file)
4271 const char *inbounds
4272 = (retval[0] < info.objsize
4273 ? (retval[1] < info.objsize
4274 ? "in" : "potentially out-of")
4275 : "out-of");
4277 const char *what = setrange ? "Setting" : "Discarding";
4278 if (retval[0] != retval[1])
4279 fprintf (dump_file,
4280 " %s %s-bounds return value range ["
4281 HOST_WIDE_INT_PRINT_UNSIGNED ", "
4282 HOST_WIDE_INT_PRINT_UNSIGNED "].\n",
4283 what, inbounds, retval[0], retval[1]);
4284 else
4285 fprintf (dump_file, " %s %s-bounds return value "
4286 HOST_WIDE_INT_PRINT_UNSIGNED ".\n",
4287 what, inbounds, retval[0]);
4291 if (dump_file)
4292 fputc ('\n', dump_file);
4294 return removed;
4297 /* Try to simplify a s{,n}printf call described by INFO with result
4298 RES by replacing it with a simpler and presumably more efficient
4299 call (such as strcpy). */
4301 static bool
4302 try_simplify_call (gimple_stmt_iterator *gsi,
4303 const call_info &info,
4304 const format_result &res)
4306 unsigned HOST_WIDE_INT dummy[2];
4307 if (!is_call_safe (info, res, info.retval_used (), dummy))
4308 return false;
4310 switch (info.fncode)
4312 case BUILT_IN_SNPRINTF:
4313 return gimple_fold_builtin_snprintf (gsi);
4315 case BUILT_IN_SPRINTF:
4316 return gimple_fold_builtin_sprintf (gsi);
4318 default:
4322 return false;
4325 /* Return the zero-based index of the format string argument of a printf
4326 like function and set *IDX_ARGS to the first format argument. When
4327 no such index exists return UINT_MAX. */
4329 static unsigned
4330 get_user_idx_format (tree fndecl, unsigned *idx_args)
4332 tree attrs = lookup_attribute ("format", DECL_ATTRIBUTES (fndecl));
4333 if (!attrs)
4334 attrs = lookup_attribute ("format", TYPE_ATTRIBUTES (TREE_TYPE (fndecl)));
4336 if (!attrs)
4337 return UINT_MAX;
4339 attrs = TREE_VALUE (attrs);
4341 tree archetype = TREE_VALUE (attrs);
4342 if (strcmp ("printf", IDENTIFIER_POINTER (archetype)))
4343 return UINT_MAX;
4345 attrs = TREE_CHAIN (attrs);
4346 tree fmtarg = TREE_VALUE (attrs);
4348 attrs = TREE_CHAIN (attrs);
4349 tree elliparg = TREE_VALUE (attrs);
4351 /* Attribute argument indices are 1-based but we use zero-based. */
4352 *idx_args = tree_to_uhwi (elliparg) - 1;
4353 return tree_to_uhwi (fmtarg) - 1;
4356 } /* Unnamed namespace. */
4358 /* Determine if a GIMPLE call at *GSI is to one of the sprintf-like built-in
4359 functions and if so, handle it. Return true if the call is removed and
4360 gsi_next should not be performed in the caller. */
4362 bool
4363 handle_printf_call (gimple_stmt_iterator *gsi, pointer_query &ptr_qry)
4365 init_target_to_host_charmap ();
4367 call_info info = call_info ();
4369 info.callstmt = gsi_stmt (*gsi);
4370 info.func = gimple_call_fndecl (info.callstmt);
4371 if (!info.func)
4372 return false;
4374 /* Format string argument number (valid for all functions). */
4375 unsigned idx_format = UINT_MAX;
4376 if (gimple_call_builtin_p (info.callstmt, BUILT_IN_NORMAL))
4377 info.fncode = DECL_FUNCTION_CODE (info.func);
4378 else
4380 unsigned idx_args;
4381 idx_format = get_user_idx_format (info.func, &idx_args);
4382 if (idx_format == UINT_MAX
4383 || idx_format >= gimple_call_num_args (info.callstmt)
4384 || idx_args > gimple_call_num_args (info.callstmt)
4385 || !POINTER_TYPE_P (TREE_TYPE (gimple_call_arg (info.callstmt,
4386 idx_format))))
4387 return false;
4388 info.fncode = BUILT_IN_NONE;
4389 info.argidx = idx_args;
4392 /* The size of the destination as in snprintf(dest, size, ...). */
4393 unsigned HOST_WIDE_INT dstsize = HOST_WIDE_INT_M1U;
4395 /* The size of the destination determined by __builtin_object_size. */
4396 unsigned HOST_WIDE_INT objsize = HOST_WIDE_INT_M1U;
4398 /* Zero-based buffer size argument number (snprintf and vsnprintf). */
4399 unsigned idx_dstsize = UINT_MAX;
4401 /* Object size argument number (snprintf_chk and vsnprintf_chk). */
4402 unsigned idx_objsize = UINT_MAX;
4404 /* Destinaton argument number (valid for sprintf functions only). */
4405 unsigned idx_dstptr = 0;
4407 switch (info.fncode)
4409 case BUILT_IN_NONE:
4410 // User-defined function with attribute format (printf).
4411 idx_dstptr = -1;
4412 break;
4414 case BUILT_IN_FPRINTF:
4415 // Signature:
4416 // __builtin_fprintf (FILE*, format, ...)
4417 idx_format = 1;
4418 info.argidx = 2;
4419 idx_dstptr = -1;
4420 break;
4422 case BUILT_IN_FPRINTF_CHK:
4423 // Signature:
4424 // __builtin_fprintf_chk (FILE*, ost, format, ...)
4425 idx_format = 2;
4426 info.argidx = 3;
4427 idx_dstptr = -1;
4428 break;
4430 case BUILT_IN_FPRINTF_UNLOCKED:
4431 // Signature:
4432 // __builtin_fprintf_unnlocked (FILE*, format, ...)
4433 idx_format = 1;
4434 info.argidx = 2;
4435 idx_dstptr = -1;
4436 break;
4438 case BUILT_IN_PRINTF:
4439 // Signature:
4440 // __builtin_printf (format, ...)
4441 idx_format = 0;
4442 info.argidx = 1;
4443 idx_dstptr = -1;
4444 break;
4446 case BUILT_IN_PRINTF_CHK:
4447 // Signature:
4448 // __builtin_printf_chk (ost, format, ...)
4449 idx_format = 1;
4450 info.argidx = 2;
4451 idx_dstptr = -1;
4452 break;
4454 case BUILT_IN_PRINTF_UNLOCKED:
4455 // Signature:
4456 // __builtin_printf (format, ...)
4457 idx_format = 0;
4458 info.argidx = 1;
4459 idx_dstptr = -1;
4460 break;
4462 case BUILT_IN_SPRINTF:
4463 // Signature:
4464 // __builtin_sprintf (dst, format, ...)
4465 idx_format = 1;
4466 info.argidx = 2;
4467 break;
4469 case BUILT_IN_SPRINTF_CHK:
4470 // Signature:
4471 // __builtin___sprintf_chk (dst, ost, objsize, format, ...)
4472 idx_objsize = 2;
4473 idx_format = 3;
4474 info.argidx = 4;
4475 break;
4477 case BUILT_IN_SNPRINTF:
4478 // Signature:
4479 // __builtin_snprintf (dst, size, format, ...)
4480 idx_dstsize = 1;
4481 idx_format = 2;
4482 info.argidx = 3;
4483 info.bounded = true;
4484 break;
4486 case BUILT_IN_SNPRINTF_CHK:
4487 // Signature:
4488 // __builtin___snprintf_chk (dst, size, ost, objsize, format, ...)
4489 idx_dstsize = 1;
4490 idx_objsize = 3;
4491 idx_format = 4;
4492 info.argidx = 5;
4493 info.bounded = true;
4494 break;
4496 case BUILT_IN_VFPRINTF:
4497 // Signature:
4498 // __builtin_vprintf (FILE*, format, va_list)
4499 idx_format = 1;
4500 info.argidx = -1;
4501 idx_dstptr = -1;
4502 break;
4504 case BUILT_IN_VFPRINTF_CHK:
4505 // Signature:
4506 // __builtin___vfprintf_chk (FILE*, ost, format, va_list)
4507 idx_format = 2;
4508 info.argidx = -1;
4509 idx_dstptr = -1;
4510 break;
4512 case BUILT_IN_VPRINTF:
4513 // Signature:
4514 // __builtin_vprintf (format, va_list)
4515 idx_format = 0;
4516 info.argidx = -1;
4517 idx_dstptr = -1;
4518 break;
4520 case BUILT_IN_VPRINTF_CHK:
4521 // Signature:
4522 // __builtin___vprintf_chk (ost, format, va_list)
4523 idx_format = 1;
4524 info.argidx = -1;
4525 idx_dstptr = -1;
4526 break;
4528 case BUILT_IN_VSNPRINTF:
4529 // Signature:
4530 // __builtin_vsprintf (dst, size, format, va)
4531 idx_dstsize = 1;
4532 idx_format = 2;
4533 info.argidx = -1;
4534 info.bounded = true;
4535 break;
4537 case BUILT_IN_VSNPRINTF_CHK:
4538 // Signature:
4539 // __builtin___vsnprintf_chk (dst, size, ost, objsize, format, va)
4540 idx_dstsize = 1;
4541 idx_objsize = 3;
4542 idx_format = 4;
4543 info.argidx = -1;
4544 info.bounded = true;
4545 break;
4547 case BUILT_IN_VSPRINTF:
4548 // Signature:
4549 // __builtin_vsprintf (dst, format, va)
4550 idx_format = 1;
4551 info.argidx = -1;
4552 break;
4554 case BUILT_IN_VSPRINTF_CHK:
4555 // Signature:
4556 // __builtin___vsprintf_chk (dst, ost, objsize, format, va)
4557 idx_format = 3;
4558 idx_objsize = 2;
4559 info.argidx = -1;
4560 break;
4562 default:
4563 return false;
4566 /* Set the global warning level for this function. */
4567 warn_level = info.bounded ? warn_format_trunc : warn_format_overflow;
4569 /* For all string functions the first argument is a pointer to
4570 the destination. */
4571 tree dstptr = (idx_dstptr < gimple_call_num_args (info.callstmt)
4572 ? gimple_call_arg (info.callstmt, 0) : NULL_TREE);
4574 info.format = gimple_call_arg (info.callstmt, idx_format);
4576 /* True when the destination size is constant as opposed to the lower
4577 or upper bound of a range. */
4578 bool dstsize_cst_p = true;
4579 bool posunder4k = true;
4581 if (idx_dstsize == UINT_MAX)
4583 /* For non-bounded functions like sprintf, determine the size
4584 of the destination from the object or pointer passed to it
4585 as the first argument. */
4586 dstsize = get_destination_size (dstptr, info.callstmt, ptr_qry);
4588 else if (tree size = gimple_call_arg (info.callstmt, idx_dstsize))
4590 /* For bounded functions try to get the size argument. */
4592 if (TREE_CODE (size) == INTEGER_CST)
4594 dstsize = tree_to_uhwi (size);
4595 /* No object can be larger than SIZE_MAX bytes (half the address
4596 space) on the target.
4597 The functions are defined only for output of at most INT_MAX
4598 bytes. Specifying a bound in excess of that limit effectively
4599 defeats the bounds checking (and on some implementations such
4600 as Solaris cause the function to fail with EINVAL). */
4601 if (dstsize > target_size_max () / 2)
4603 /* Avoid warning if -Wstringop-overflow is specified since
4604 it also warns for the same thing though only for the
4605 checking built-ins. */
4606 if ((idx_objsize == UINT_MAX
4607 || !warn_stringop_overflow))
4608 warning_at (gimple_location (info.callstmt), info.warnopt (),
4609 "specified bound %wu exceeds maximum object size "
4610 "%wu",
4611 dstsize, target_size_max () / 2);
4612 /* POSIX requires snprintf to fail if DSTSIZE is greater
4613 than INT_MAX. Even though not all POSIX implementations
4614 conform to the requirement, avoid folding in this case. */
4615 posunder4k = false;
4617 else if (dstsize > target_int_max ())
4619 warning_at (gimple_location (info.callstmt), info.warnopt (),
4620 "specified bound %wu exceeds %<INT_MAX%>",
4621 dstsize);
4622 /* POSIX requires snprintf to fail if DSTSIZE is greater
4623 than INT_MAX. Avoid folding in that case. */
4624 posunder4k = false;
4627 else if (TREE_CODE (size) == SSA_NAME)
4629 /* Try to determine the range of values of the argument
4630 and use the greater of the two at level 1 and the smaller
4631 of them at level 2. */
4632 int_range_max vr;
4633 ptr_qry.rvals->range_of_expr (vr, size, info.callstmt);
4635 if (!vr.undefined_p ())
4637 tree type = TREE_TYPE (size);
4638 tree tmin = wide_int_to_tree (type, vr.lower_bound ());
4639 tree tmax = wide_int_to_tree (type, vr.upper_bound ());
4640 unsigned HOST_WIDE_INT minsize = TREE_INT_CST_LOW (tmin);
4641 unsigned HOST_WIDE_INT maxsize = TREE_INT_CST_LOW (tmax);
4642 dstsize = warn_level < 2 ? maxsize : minsize;
4644 if (minsize > target_int_max ())
4645 warning_at (gimple_location (info.callstmt), info.warnopt (),
4646 "specified bound range [%wu, %wu] exceeds "
4647 "%<INT_MAX%>",
4648 minsize, maxsize);
4650 /* POSIX requires snprintf to fail if DSTSIZE is greater
4651 than INT_MAX. Avoid folding if that's possible. */
4652 if (maxsize > target_int_max ())
4653 posunder4k = false;
4656 /* The destination size is not constant. If the function is
4657 bounded (e.g., snprintf) a lower bound of zero doesn't
4658 necessarily imply it can be eliminated. */
4659 dstsize_cst_p = false;
4663 if (idx_objsize != UINT_MAX)
4664 if (tree size = gimple_call_arg (info.callstmt, idx_objsize))
4665 if (tree_fits_uhwi_p (size))
4666 objsize = tree_to_uhwi (size);
4668 if (info.bounded && !dstsize)
4670 /* As a special case, when the explicitly specified destination
4671 size argument (to a bounded function like snprintf) is zero
4672 it is a request to determine the number of bytes on output
4673 without actually producing any. Pretend the size is
4674 unlimited in this case. */
4675 info.objsize = HOST_WIDE_INT_MAX;
4676 info.nowrite = dstsize_cst_p;
4678 else
4680 /* For calls to non-bounded functions or to those of bounded
4681 functions with a non-zero size, warn if the destination
4682 pointer is null. */
4683 if (dstptr && integer_zerop (dstptr))
4685 /* This is diagnosed with -Wformat only when the null is a constant
4686 pointer. The warning here diagnoses instances where the pointer
4687 is not constant. */
4688 location_t loc = gimple_location (info.callstmt);
4689 warning_at (EXPR_LOC_OR_LOC (dstptr, loc),
4690 info.warnopt (), "null destination pointer");
4691 return false;
4694 /* Set the object size to the smaller of the two arguments
4695 of both have been specified and they're not equal. */
4696 info.objsize = dstsize < objsize ? dstsize : objsize;
4698 if (info.bounded
4699 && dstsize < target_size_max () / 2 && objsize < dstsize
4700 /* Avoid warning if -Wstringop-overflow is specified since
4701 it also warns for the same thing though only for the
4702 checking built-ins. */
4703 && (idx_objsize == UINT_MAX
4704 || !warn_stringop_overflow))
4706 warning_at (gimple_location (info.callstmt), info.warnopt (),
4707 "specified bound %wu exceeds the size %wu "
4708 "of the destination object", dstsize, objsize);
4712 /* Determine if the format argument may be null and warn if not
4713 and if the argument is null. */
4714 if (integer_zerop (info.format)
4715 && gimple_call_builtin_p (info.callstmt, BUILT_IN_NORMAL))
4717 location_t loc = gimple_location (info.callstmt);
4718 warning_at (EXPR_LOC_OR_LOC (info.format, loc),
4719 info.warnopt (), "null format string");
4720 return false;
4723 info.fmtstr = get_format_string (info.format, &info.fmtloc);
4724 if (!info.fmtstr)
4725 return false;
4727 if (warn_restrict)
4729 /* Compute the origin of the destination pointer and its offset
4730 from the base object/pointer if possible. */
4731 info.dst_offset = 0;
4732 info.dst_origin = get_origin_and_offset (dstptr, &info.dst_field,
4733 &info.dst_offset);
4736 /* The result is the number of bytes output by the formatted function,
4737 including the terminating NUL. */
4738 format_result res;
4740 /* I/O functions with no destination argument (i.e., all forms of fprintf
4741 and printf) may fail under any conditions. Others (i.e., all forms of
4742 sprintf) may only fail under specific conditions determined for each
4743 directive. Clear POSUNDER4K for the former set of functions and set
4744 it to true for the latter (it can only be cleared later, but it is
4745 never set to true again). */
4746 res.posunder4k = posunder4k && dstptr;
4748 bool success = compute_format_length (info, &res, ptr_qry);
4749 if (res.warned)
4750 suppress_warning (info.callstmt, info.warnopt ());
4752 /* When optimizing and the printf return value optimization is enabled,
4753 attempt to substitute the computed result for the return value of
4754 the call. Avoid this optimization when -frounding-math is in effect
4755 and the format string contains a floating point directive. */
4756 bool call_removed = false;
4757 if (success && optimize > 0)
4759 /* Save a copy of the iterator pointing at the call. The iterator
4760 may change to point past the call in try_substitute_return_value
4761 but the original value is needed in try_simplify_call. */
4762 gimple_stmt_iterator gsi_call = *gsi;
4764 if (flag_printf_return_value
4765 && (!flag_rounding_math || !res.floating))
4766 call_removed = try_substitute_return_value (gsi, info, res);
4768 if (!call_removed)
4769 try_simplify_call (&gsi_call, info, res);
4772 return call_removed;