1 /* Copyright (C) 2016-2020 Free Software Foundation, Inc.
2    Contributed by Martin Sebor <msebor@redhat.com>.
3 
4 This file is part of GCC.
5 
6 GCC is free software; you can redistribute it and/or modify it under
7 the terms of the GNU General Public License as published by the Free
8 Software Foundation; either version 3, or (at your option) any later
9 version.
10 
11 GCC is distributed in the hope that it will be useful, but WITHOUT ANY
12 WARRANTY; without even the implied warranty of MERCHANTABILITY or
13 FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
14 for more details.
15 
16 You should have received a copy of the GNU General Public License
17 along with GCC; see the file COPYING3.  If not see
18 <http://www.gnu.org/licenses/>.  */
19 
20 /* This file implements the printf-return-value pass.  The pass does
21    two things: 1) it analyzes calls to formatted output functions like
22    sprintf looking for possible buffer overflows and calls to bounded
23    functions like snprintf for early truncation (and under the control
24    of the -Wformat-length option issues warnings), and 2) under the
25    control of the -fprintf-return-value option it folds the return
26    value of safe calls into constants, making it possible to eliminate
27    code that depends on the value of those constants.
28 
29    For all functions (bounded or not) the pass uses the size of the
30    destination object.  That means that it will diagnose calls to
31    snprintf not on the basis of the size specified by the function's
32    second argument but rathger on the basis of the size the first
33    argument points to (if possible).  For bound-checking built-ins
34    like __builtin___snprintf_chk the pass uses the size typically
35    determined by __builtin_object_size and passed to the built-in
36    by the Glibc inline wrapper.
37 
38    The pass handles all forms standard sprintf format directives,
39    including character, integer, floating point, pointer, and strings,
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 fornmat known constants with up and down
43    rounding and uses the resulting range of output lengths.  For
44    strings it uses the length of string literals and the sizes of
45    character arrays that a character pointer may point to as a bound
46    on the longest string.  */
47 
48 #include "config.h"
49 #include "system.h"
50 #include "coretypes.h"
51 #include "backend.h"
52 #include "tree.h"
53 #include "gimple.h"
54 #include "tree-pass.h"
55 #include "ssa.h"
56 #include "gimple-fold.h"
57 #include "gimple-pretty-print.h"
58 #include "diagnostic-core.h"
59 #include "fold-const.h"
60 #include "gimple-iterator.h"
61 #include "tree-ssa.h"
62 #include "tree-object-size.h"
63 #include "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"
71 
72 #include "attribs.h"
73 #include "builtins.h"
74 #include "stor-layout.h"
75 
76 #include "realmpfr.h"
77 #include "target.h"
78 
79 #include "cpplib.h"
80 #include "input.h"
81 #include "toplev.h"
82 #include "substring-locations.h"
83 #include "diagnostic.h"
84 #include "domwalk.h"
85 #include "alloc-pool.h"
86 #include "vr-values.h"
87 #include "tree-ssa-strlen.h"
88 #include "tree-dfa.h"
89 
90 /* The likely worst case value of MB_LEN_MAX for the target, large enough
91    for UTF-8.  Ideally, this would be obtained by a target hook if it were
92    to be used for optimization but it's good enough as is for warnings.  */
93 #define target_mb_len_max()   6
94 
95 /* The maximum number of bytes a single non-string directive can result
96    in.  This is the result of printf("%.*Lf", INT_MAX, -LDBL_MAX) for
97    LDBL_MAX_10_EXP of 4932.  */
98 #define IEEE_MAX_10_EXP    4932
99 #define target_dir_max()   (target_int_max () + IEEE_MAX_10_EXP + 2)
100 
101 namespace {
102 
103 /* Set to the warning level for the current function which is equal
104    either to warn_format_trunc for bounded functions or to
105    warn_format_overflow otherwise.  */
106 
107 static int warn_level;
108 
109 /* The minimum, maximum, likely, and unlikely maximum number of bytes
110    of output either a formatting function or an individual directive
111    can result in.  */
112 
113 struct result_range
114 {
115   /* The absolute minimum number of bytes.  The result of a successful
116      conversion is guaranteed to be no less than this.  (An erroneous
117      conversion can be indicated by MIN > HOST_WIDE_INT_MAX.)  */
118   unsigned HOST_WIDE_INT min;
119   /* The likely maximum result that is used in diagnostics.  In most
120      cases MAX is the same as the worst case UNLIKELY result.  */
121   unsigned HOST_WIDE_INT max;
122   /* The likely result used to trigger diagnostics.  For conversions
123      that result in a range of bytes [MIN, MAX], LIKELY is somewhere
124      in that range.  */
125   unsigned HOST_WIDE_INT likely;
126   /* In rare cases (e.g., for nultibyte characters) UNLIKELY gives
127      the worst cases maximum result of a directive.  In most cases
128      UNLIKELY == MAX.  UNLIKELY is used to control the return value
129      optimization but not in diagnostics.  */
130   unsigned HOST_WIDE_INT unlikely;
131 };
132 
133 /* Return the value of INT_MIN for the target.  */
134 
135 static inline HOST_WIDE_INT
target_int_min()136 target_int_min ()
137 {
138   return tree_to_shwi (TYPE_MIN_VALUE (integer_type_node));
139 }
140 
141 /* Return the value of INT_MAX for the target.  */
142 
143 static inline unsigned HOST_WIDE_INT
target_int_max()144 target_int_max ()
145 {
146   return tree_to_uhwi (TYPE_MAX_VALUE (integer_type_node));
147 }
148 
149 /* Return the value of SIZE_MAX for the target.  */
150 
151 static inline unsigned HOST_WIDE_INT
target_size_max()152 target_size_max ()
153 {
154   return tree_to_uhwi (TYPE_MAX_VALUE (size_type_node));
155 }
156 
157 /* A straightforward mapping from the execution character set to the host
158    character set indexed by execution character.  */
159 
160 static char target_to_host_charmap[256];
161 
162 /* Initialize a mapping from the execution character set to the host
163    character set.  */
164 
165 static bool
init_target_to_host_charmap()166 init_target_to_host_charmap ()
167 {
168   /* If the percent sign is non-zero the mapping has already been
169      initialized.  */
170   if (target_to_host_charmap['%'])
171     return true;
172 
173   /* Initialize the target_percent character (done elsewhere).  */
174   if (!init_target_chars ())
175     return false;
176 
177   /* The subset of the source character set used by printf conversion
178      specifications (strictly speaking, not all letters are used but
179      they are included here for the sake of simplicity).  The dollar
180      sign must be included even though it's not in the basic source
181      character set.  */
182   const char srcset[] = " 0123456789!\"#%&'()*+,-./:;<=>?[\\]^_{|}~$"
183     "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz";
184 
185   /* Set the mapping for all characters to some ordinary value (i,e.,
186      not none used in printf conversion specifications) and overwrite
187      those that are used by conversion specifications with their
188      corresponding values.  */
189   memset (target_to_host_charmap + 1, '?', sizeof target_to_host_charmap - 1);
190 
191   /* Are the two sets of characters the same?  */
192   bool all_same_p = true;
193 
194   for (const char *pc = srcset; *pc; ++pc)
195     {
196       /* Slice off the high end bits in case target characters are
197 	 signed.  All values are expected to be non-nul, otherwise
198 	 there's a problem.  */
199       if (unsigned char tc = lang_hooks.to_target_charset (*pc))
200 	{
201 	  target_to_host_charmap[tc] = *pc;
202 	  if (tc != *pc)
203 	    all_same_p = false;
204 	}
205       else
206 	return false;
207 
208     }
209 
210   /* Set the first element to a non-zero value if the mapping
211      is 1-to-1, otherwise leave it clear (NUL is assumed to be
212      the same in both character sets).  */
213   target_to_host_charmap[0] = all_same_p;
214 
215   return true;
216 }
217 
218 /* Return the host source character corresponding to the character
219    CH in the execution character set if one exists, or some innocuous
220    (non-special, non-nul) source character otherwise.  */
221 
222 static inline unsigned char
target_to_host(unsigned char ch)223 target_to_host (unsigned char ch)
224 {
225   return target_to_host_charmap[ch];
226 }
227 
228 /* Convert an initial substring of the string TARGSTR consisting of
229    characters in the execution character set into a string in the
230    source character set on the host and store up to HOSTSZ characters
231    in the buffer pointed to by HOSTR.  Return HOSTR.  */
232 
233 static const char*
target_to_host(char * hostr,size_t hostsz,const char * targstr)234 target_to_host (char *hostr, size_t hostsz, const char *targstr)
235 {
236   /* Make sure the buffer is reasonably big.  */
237   gcc_assert (hostsz > 4);
238 
239   /* The interesting subset of source and execution characters are
240      the same so no conversion is necessary.  However, truncate
241      overlong strings just like the translated strings are.  */
242   if (target_to_host_charmap['\0'] == 1)
243     {
244       size_t len = strlen (targstr);
245       if (len >= hostsz)
246 	{
247 	  memcpy (hostr, targstr, hostsz - 4);
248 	  strcpy (hostr + hostsz - 4, "...");
249 	}
250       else
251 	memcpy (hostr, targstr, len + 1);
252       return hostr;
253     }
254 
255   /* Convert the initial substring of TARGSTR to the corresponding
256      characters in the host set, appending "..." if TARGSTR is too
257      long to fit.  Using the static buffer assumes the function is
258      not called in between sequence points (which it isn't).  */
259   for (char *ph = hostr; ; ++targstr)
260     {
261       *ph++ = target_to_host (*targstr);
262       if (!*targstr)
263 	break;
264 
265       if (size_t (ph - hostr) == hostsz)
266 	{
267 	  strcpy (ph - 4, "...");
268 	  break;
269 	}
270     }
271 
272   return hostr;
273 }
274 
275 /* Convert the sequence of decimal digits in the execution character
276    starting at *PS to a HOST_WIDE_INT, analogously to strtol.  Return
277    the result and set *PS to one past the last converted character.
278    On range error set ERANGE to the digit that caused it.  */
279 
280 static inline HOST_WIDE_INT
target_strtowi(const char ** ps,const char ** erange)281 target_strtowi (const char **ps, const char **erange)
282 {
283   unsigned HOST_WIDE_INT val = 0;
284   for ( ; ; ++*ps)
285     {
286       unsigned char c = target_to_host (**ps);
287       if (ISDIGIT (c))
288 	{
289 	  c -= '0';
290 
291 	  /* Check for overflow.  */
292 	  if (val > ((unsigned HOST_WIDE_INT) HOST_WIDE_INT_MAX - c) / 10LU)
293 	    {
294 	      val = HOST_WIDE_INT_MAX;
295 	      *erange = *ps;
296 
297 	      /* Skip the remaining digits.  */
298 	      do
299 		c = target_to_host (*++*ps);
300 	      while (ISDIGIT (c));
301 	      break;
302 	    }
303 	  else
304 	    val = val * 10 + c;
305 	}
306       else
307 	break;
308     }
309 
310   return val;
311 }
312 
313 /* Given FORMAT, set *PLOC to the source location of the format string
314    and return the format string if it is known or null otherwise.  */
315 
316 static const char*
get_format_string(tree format,location_t * ploc)317 get_format_string (tree format, location_t *ploc)
318 {
319   *ploc = EXPR_LOC_OR_LOC (format, input_location);
320 
321   return c_getstr (format);
322 }
323 
324 /* For convenience and brevity, shorter named entrypoints of
325    format_string_diagnostic_t::emit_warning_va and
326    format_string_diagnostic_t::emit_warning_n_va.
327    These have to be functions with the attribute so that exgettext
328    works properly.  */
329 
330 static bool
331 ATTRIBUTE_GCC_DIAG (5, 6)
fmtwarn(const substring_loc & fmt_loc,location_t param_loc,const char * corrected_substring,int opt,const char * gmsgid,...)332 fmtwarn (const substring_loc &fmt_loc, location_t param_loc,
333 	 const char *corrected_substring, int opt, const char *gmsgid, ...)
334 {
335   format_string_diagnostic_t diag (fmt_loc, NULL, param_loc, NULL,
336 				   corrected_substring);
337   va_list ap;
338   va_start (ap, gmsgid);
339   bool warned = diag.emit_warning_va (opt, gmsgid, &ap);
340   va_end (ap);
341 
342   return warned;
343 }
344 
345 static bool
346 ATTRIBUTE_GCC_DIAG (6, 8) ATTRIBUTE_GCC_DIAG (7, 8)
fmtwarn_n(const substring_loc & fmt_loc,location_t param_loc,const char * corrected_substring,int opt,unsigned HOST_WIDE_INT n,const char * singular_gmsgid,const char * plural_gmsgid,...)347 fmtwarn_n (const substring_loc &fmt_loc, location_t param_loc,
348 	   const char *corrected_substring, int opt, unsigned HOST_WIDE_INT n,
349 	   const char *singular_gmsgid, const char *plural_gmsgid, ...)
350 {
351   format_string_diagnostic_t diag (fmt_loc, NULL, param_loc, NULL,
352 				   corrected_substring);
353   va_list ap;
354   va_start (ap, plural_gmsgid);
355   bool warned = diag.emit_warning_n_va (opt, n, singular_gmsgid, plural_gmsgid,
356 					&ap);
357   va_end (ap);
358 
359   return warned;
360 }
361 
362 /* Format length modifiers.  */
363 
364 enum format_lengths
365 {
366   FMT_LEN_none,
367   FMT_LEN_hh,    // char argument
368   FMT_LEN_h,     // short
369   FMT_LEN_l,     // long
370   FMT_LEN_ll,    // long long
371   FMT_LEN_L,     // long double (and GNU long long)
372   FMT_LEN_z,     // size_t
373   FMT_LEN_t,     // ptrdiff_t
374   FMT_LEN_j      // intmax_t
375 };
376 
377 
378 /* Description of the result of conversion either of a single directive
379    or the whole format string.  */
380 
381 class fmtresult
382 {
383 public:
384   /* Construct a FMTRESULT object with all counters initialized
385      to MIN.  KNOWNRANGE is set when MIN is valid.  */
386   fmtresult (unsigned HOST_WIDE_INT min = HOST_WIDE_INT_MAX)
argmin()387   : argmin (), argmax (), dst_offset (HOST_WIDE_INT_MIN), nonstr (),
388     knownrange (min < HOST_WIDE_INT_MAX),
389     mayfail (), nullp ()
390   {
391     range.min = min;
392     range.max = min;
393     range.likely = min;
394     range.unlikely = min;
395   }
396 
397   /* Construct a FMTRESULT object with MIN, MAX, and LIKELY counters.
398      KNOWNRANGE is set when both MIN and MAX are valid.   */
399   fmtresult (unsigned HOST_WIDE_INT min, unsigned HOST_WIDE_INT max,
400 	     unsigned HOST_WIDE_INT likely = HOST_WIDE_INT_MAX)
argmin()401   : argmin (), argmax (), dst_offset (HOST_WIDE_INT_MIN), nonstr (),
402     knownrange (min < HOST_WIDE_INT_MAX && max < HOST_WIDE_INT_MAX),
403     mayfail (), nullp ()
404   {
405     range.min = min;
406     range.max = max;
407     range.likely = max < likely ? min : likely;
408     range.unlikely = max;
409   }
410 
411   /* Adjust result upward to reflect the RANGE of values the specified
412      width or precision is known to be in.  */
413   fmtresult& adjust_for_width_or_precision (const HOST_WIDE_INT[2],
414 					    tree = NULL_TREE,
415 					    unsigned = 0, unsigned = 0);
416 
417   /* Return the maximum number of decimal digits a value of TYPE
418      formats as on output.  */
419   static unsigned type_max_digits (tree, int);
420 
421   /* The range a directive's argument is in.  */
422   tree argmin, argmax;
423 
424   /* The starting offset into the destination of the formatted function
425      call of the %s argument that points into (aliases with) the same
426      destination array.  */
427   HOST_WIDE_INT dst_offset;
428 
429   /* The minimum and maximum number of bytes that a directive
430      results in on output for an argument in the range above.  */
431   result_range range;
432 
433   /* Non-nul when the argument of a string directive is not a nul
434      terminated string.  */
435   tree nonstr;
436 
437   /* True when the range above is obtained from a known value of
438      a directive's argument or its bounds and not the result of
439      heuristics that depend on warning levels.  */
440   bool knownrange;
441 
442   /* True for a directive that may fail (such as wide character
443      directives).  */
444   bool mayfail;
445 
446   /* True when the argument is a null pointer.  */
447   bool nullp;
448 };
449 
450 /* Adjust result upward to reflect the range ADJUST of values the
451    specified width or precision is known to be in.  When non-null,
452    TYPE denotes the type of the directive whose result is being
453    adjusted, BASE gives the base of the directive (octal, decimal,
454    or hex), and ADJ denotes the additional adjustment to the LIKELY
455    counter that may need to be added when ADJUST is a range.  */
456 
457 fmtresult&
adjust_for_width_or_precision(const HOST_WIDE_INT adjust[2],tree type,unsigned base,unsigned adj)458 fmtresult::adjust_for_width_or_precision (const HOST_WIDE_INT adjust[2],
459 					  tree type /* = NULL_TREE */,
460 					  unsigned base /* = 0 */,
461 					  unsigned adj /* = 0 */)
462 {
463   bool minadjusted = false;
464 
465   /* Adjust the minimum and likely counters.  */
466   if (adjust[0] >= 0)
467     {
468       if (range.min < (unsigned HOST_WIDE_INT)adjust[0])
469 	{
470 	  range.min = adjust[0];
471 	  minadjusted = true;
472 	}
473 
474       /* Adjust the likely counter.  */
475       if (range.likely < range.min)
476 	range.likely = range.min;
477     }
478   else if (adjust[0] == target_int_min ()
479 	   && (unsigned HOST_WIDE_INT)adjust[1] == target_int_max ())
480     knownrange = false;
481 
482   /* Adjust the maximum counter.  */
483   if (adjust[1] > 0)
484     {
485       if (range.max < (unsigned HOST_WIDE_INT)adjust[1])
486 	{
487 	  range.max = adjust[1];
488 
489 	  /* Set KNOWNRANGE if both the minimum and maximum have been
490 	     adjusted.  Otherwise leave it at what it was before.  */
491 	  knownrange = minadjusted;
492 	}
493     }
494 
495   if (warn_level > 1 && type)
496     {
497       /* For large non-constant width or precision whose range spans
498 	 the maximum number of digits produced by the directive for
499 	 any argument, set the likely number of bytes to be at most
500 	 the number digits plus other adjustment determined by the
501 	 caller (one for sign or two for the hexadecimal "0x"
502 	 prefix).  */
503       unsigned dirdigs = type_max_digits (type, base);
504       if (adjust[0] < dirdigs && dirdigs < adjust[1]
505 	  && range.likely < dirdigs)
506 	range.likely = dirdigs + adj;
507     }
508   else if (range.likely < (range.min ? range.min : 1))
509     {
510       /* Conservatively, set LIKELY to at least MIN but no less than
511 	 1 unless MAX is zero.  */
512       range.likely = (range.min
513 		      ? range.min
514 		      : range.max && (range.max < HOST_WIDE_INT_MAX
515 				      || warn_level > 1) ? 1 : 0);
516     }
517 
518   /* Finally adjust the unlikely counter to be at least as large as
519      the maximum.  */
520   if (range.unlikely < range.max)
521     range.unlikely = range.max;
522 
523   return *this;
524 }
525 
526 /* Return the maximum number of digits a value of TYPE formats in
527    BASE on output, not counting base prefix .  */
528 
529 unsigned
type_max_digits(tree type,int base)530 fmtresult::type_max_digits (tree type, int base)
531 {
532   unsigned prec = TYPE_PRECISION (type);
533   switch (base)
534     {
535     case 8:
536       return (prec + 2) / 3;
537     case 10:
538       /* Decimal approximation: yields 3, 5, 10, and 20 for precision
539 	 of 8, 16, 32, and 64 bits.  */
540       return prec * 301 / 1000 + 1;
541     case 16:
542       return prec / 4;
543     }
544 
545   gcc_unreachable ();
546 }
547 
548 static bool
549 get_int_range (tree, HOST_WIDE_INT *, HOST_WIDE_INT *, bool, HOST_WIDE_INT,
550 	       const vr_values *);
551 
552 struct call_info;
553 
554 /* Description of a format directive.  A directive is either a plain
555    string or a conversion specification that starts with '%'.  */
556 
557 struct directive
558 {
directivedirective559   directive (const call_info *inf, unsigned dno)
560     : info (inf), dirno (dno), argno (), beg (), len (), flags (),
561     width (), prec (),  modifier (), specifier (), arg (), fmtfunc ()
562   { }
563 
564   /* Reference to the info structure describing the call that this
565      directive is a part of.  */
566   const call_info *info;
567 
568   /* The 1-based directive number (for debugging).  */
569   unsigned dirno;
570 
571   /* The zero-based argument number of the directive's argument ARG in
572      the function's argument list.  */
573   unsigned argno;
574 
575   /* The first character of the directive and its length.  */
576   const char *beg;
577   size_t len;
578 
579   /* A bitmap of flags, one for each character.  */
580   unsigned flags[256 / sizeof (int)];
581 
582   /* The range of values of the specified width, or -1 if not specified.  */
583   HOST_WIDE_INT width[2];
584   /* The range of values of the specified precision, or -1 if not
585      specified.  */
586   HOST_WIDE_INT prec[2];
587 
588   /* Length modifier.  */
589   format_lengths modifier;
590 
591   /* Format specifier character.  */
592   char specifier;
593 
594   /* The argument of the directive or null when the directive doesn't
595      take one or when none is available (such as for vararg functions).  */
596   tree arg;
597 
598   /* Format conversion function that given a directive and an argument
599      returns the formatting result.  */
600   fmtresult (*fmtfunc) (const directive &, tree, const vr_values *);
601 
602   /* Return True when a the format flag CHR has been used.  */
get_flagdirective603   bool get_flag (char chr) const
604   {
605     unsigned char c = chr & 0xff;
606     return (flags[c / (CHAR_BIT * sizeof *flags)]
607 	    & (1U << (c % (CHAR_BIT * sizeof *flags))));
608   }
609 
610   /* Make a record of the format flag CHR having been used.  */
set_flagdirective611   void set_flag (char chr)
612   {
613     unsigned char c = chr & 0xff;
614     flags[c / (CHAR_BIT * sizeof *flags)]
615       |= (1U << (c % (CHAR_BIT * sizeof *flags)));
616   }
617 
618   /* Reset the format flag CHR.  */
clear_flagdirective619   void clear_flag (char chr)
620   {
621     unsigned char c = chr & 0xff;
622     flags[c / (CHAR_BIT * sizeof *flags)]
623       &= ~(1U << (c % (CHAR_BIT * sizeof *flags)));
624   }
625 
626   /* Set both bounds of the width range to VAL.  */
set_widthdirective627   void set_width (HOST_WIDE_INT val)
628   {
629     width[0] = width[1] = val;
630   }
631 
632   /* Set the width range according to ARG, with both bounds being
633      no less than 0.  For a constant ARG set both bounds to its value
634      or 0, whichever is greater.  For a non-constant ARG in some range
635      set width to its range adjusting each bound to -1 if it's less.
636      For an indeterminate ARG set width to [0, INT_MAX].  */
set_widthdirective637   void set_width (tree arg, const vr_values *vr)
638   {
639     get_int_range (arg, width, width + 1, true, 0, vr);
640   }
641 
642   /* Set both bounds of the precision range to VAL.  */
set_precisiondirective643   void set_precision (HOST_WIDE_INT val)
644   {
645     prec[0] = prec[1] = val;
646   }
647 
648   /* Set the precision range according to ARG, with both bounds being
649      no less than -1.  For a constant ARG set both bounds to its value
650      or -1 whichever is greater.  For a non-constant ARG in some range
651      set precision to its range adjusting each bound to -1 if it's less.
652      For an indeterminate ARG set precision to [-1, INT_MAX].  */
set_precisiondirective653   void set_precision (tree arg, const vr_values *vr)
654   {
655     get_int_range (arg, prec, prec + 1, false, -1, vr);
656   }
657 
658   /* Return true if both width and precision are known to be
659      either constant or in some range, false otherwise.  */
known_width_and_precisiondirective660   bool known_width_and_precision () const
661   {
662     return ((width[1] < 0
663 	     || (unsigned HOST_WIDE_INT)width[1] <= target_int_max ())
664 	    && (prec[1] < 0
665 		|| (unsigned HOST_WIDE_INT)prec[1] < target_int_max ()));
666   }
667 };
668 
669 /* The result of a call to a formatted function.  */
670 
671 struct format_result
672 {
format_resultformat_result673   format_result ()
674     : range (), aliases (), alias_count (), knownrange (), posunder4k (),
675     floating (), warned () { /* No-op.  */ }
676 
~format_resultformat_result677   ~format_result ()
678   {
679     XDELETEVEC (aliases);
680   }
681 
682   /* Range of characters written by the formatted function.
683      Setting the minimum to HOST_WIDE_INT_MAX disables all
684      length tracking for the remainder of the format string.  */
685   result_range range;
686 
687   struct alias_info
688   {
689     directive dir;          /* The directive that aliases the destination.  */
690     HOST_WIDE_INT offset;   /* The offset at which it aliases it.  */
691     result_range range;     /* The raw result of the directive.  */
692   };
693 
694   /* An array of directives whose pointer argument aliases a part
695      of the destination object of the formatted function.  */
696   alias_info *aliases;
697   unsigned alias_count;
698 
699   /* True when the range above is obtained from known values of
700      directive arguments, or bounds on the amount of output such
701      as width and precision, and not the result of  heuristics that
702      depend on warning levels.  It's used to issue stricter diagnostics
703      in cases where strings of unknown lengths are bounded by the arrays
704      they are determined to refer to.  KNOWNRANGE must not be used for
705      the return value optimization.  */
706   bool knownrange;
707 
708   /* True if no individual directive could fail or result in more than
709      4095 bytes of output (the total NUMBER_CHARS_{MIN,MAX} might be
710      greater).  Implementations are not required to handle directives
711      that produce more than 4K bytes (leading to undefined behavior)
712      and so when one is found it disables the return value optimization.
713      Similarly, directives that can fail (such as wide character
714      directives) disable the optimization.  */
715   bool posunder4k;
716 
717   /* True when a floating point directive has been seen in the format
718      string.  */
719   bool floating;
720 
721   /* True when an intermediate result has caused a warning.  Used to
722      avoid issuing duplicate warnings while finishing the processing
723      of a call.  WARNED also disables the return value optimization.  */
724   bool warned;
725 
726   /* Preincrement the number of output characters by 1.  */
727   format_result& operator++ ()
728   {
729     return *this += 1;
730   }
731 
732   /* Postincrement the number of output characters by 1.  */
733   format_result operator++ (int)
734   {
735     format_result prev (*this);
736     *this += 1;
737     return prev;
738   }
739 
740   /* Increment the number of output characters by N.  */
741   format_result& operator+= (unsigned HOST_WIDE_INT);
742 
743   /* Add a directive to the sequence of those with potentially aliasing
744      arguments.  */
745   void append_alias (const directive &, HOST_WIDE_INT, const result_range &);
746 
747 private:
748   /* Not copyable or assignable.  */
749   format_result (format_result&);
750   void operator= (format_result&);
751 };
752 
753 format_result&
754 format_result::operator+= (unsigned HOST_WIDE_INT n)
755 {
756   gcc_assert (n < HOST_WIDE_INT_MAX);
757 
758   if (range.min < HOST_WIDE_INT_MAX)
759     range.min += n;
760 
761   if (range.max < HOST_WIDE_INT_MAX)
762     range.max += n;
763 
764   if (range.likely < HOST_WIDE_INT_MAX)
765     range.likely += n;
766 
767   if (range.unlikely < HOST_WIDE_INT_MAX)
768     range.unlikely += n;
769 
770   return *this;
771 }
772 
773 void
append_alias(const directive & d,HOST_WIDE_INT off,const result_range & resrng)774 format_result::append_alias (const directive &d, HOST_WIDE_INT off,
775 			     const result_range &resrng)
776 {
777   unsigned cnt = alias_count + 1;
778   alias_info *ar = XNEWVEC (alias_info, cnt);
779 
780   for (unsigned i = 0; i != alias_count; ++i)
781     ar[i] = aliases[i];
782 
783   ar[alias_count].dir = d;
784   ar[alias_count].offset = off;
785   ar[alias_count].range = resrng;
786 
787   XDELETEVEC (aliases);
788 
789   alias_count = cnt;
790   aliases = ar;
791 }
792 
793 /* Return the logarithm of X in BASE.  */
794 
795 static int
ilog(unsigned HOST_WIDE_INT x,int base)796 ilog (unsigned HOST_WIDE_INT x, int base)
797 {
798   int res = 0;
799   do
800     {
801       ++res;
802       x /= base;
803     } while (x);
804   return res;
805 }
806 
807 /* Return the number of bytes resulting from converting into a string
808    the INTEGER_CST tree node X in BASE with a minimum of PREC digits.
809    PLUS indicates whether 1 for a plus sign should be added for positive
810    numbers, and PREFIX whether the length of an octal ('O') or hexadecimal
811    ('0x') prefix should be added for nonzero numbers.  Return -1 if X cannot
812    be represented.  */
813 
814 static HOST_WIDE_INT
tree_digits(tree x,int base,HOST_WIDE_INT prec,bool plus,bool prefix)815 tree_digits (tree x, int base, HOST_WIDE_INT prec, bool plus, bool prefix)
816 {
817   unsigned HOST_WIDE_INT absval;
818 
819   HOST_WIDE_INT res;
820 
821   if (TYPE_UNSIGNED (TREE_TYPE (x)))
822     {
823       if (tree_fits_uhwi_p (x))
824 	{
825 	  absval = tree_to_uhwi (x);
826 	  res = plus;
827 	}
828       else
829 	return -1;
830     }
831   else
832     {
833       if (tree_fits_shwi_p (x))
834 	{
835 	  HOST_WIDE_INT i = tree_to_shwi (x);
836          if (HOST_WIDE_INT_MIN == i)
837            {
838              /* Avoid undefined behavior due to negating a minimum.  */
839              absval = HOST_WIDE_INT_MAX;
840              res = 1;
841            }
842          else if (i < 0)
843 	   {
844 	     absval = -i;
845 	     res = 1;
846 	   }
847 	 else
848 	   {
849 	     absval = i;
850 	     res = plus;
851 	   }
852 	}
853       else
854 	return -1;
855     }
856 
857   int ndigs = ilog (absval, base);
858 
859   res += prec < ndigs ? ndigs : prec;
860 
861   /* Adjust a non-zero value for the base prefix, either hexadecimal,
862      or, unless precision has resulted in a leading zero, also octal.  */
863   if (prefix && absval && (base == 16 || prec <= ndigs))
864     {
865       if (base == 8)
866 	res += 1;
867       else if (base == 16)
868 	res += 2;
869     }
870 
871   return res;
872 }
873 
874 /* Description of a call to a formatted function.  */
875 
876 struct call_info
877 {
878   /* Function call statement.  */
879   gimple *callstmt;
880 
881   /* Function called.  */
882   tree func;
883 
884   /* Called built-in function code.  */
885   built_in_function fncode;
886 
887   /* The "origin" of the destination pointer argument, which is either
888      the DECL of the destination buffer being written into or a pointer
889      that points to it, plus some offset.  */
890   tree dst_origin;
891 
892   /* For a destination pointing to a struct array member, the offset of
893      the member.  */
894   HOST_WIDE_INT dst_field;
895 
896   /* The offset into the destination buffer.  */
897   HOST_WIDE_INT dst_offset;
898 
899   /* Format argument and format string extracted from it.  */
900   tree format;
901   const char *fmtstr;
902 
903   /* The location of the format argument.  */
904   location_t fmtloc;
905 
906   /* The destination object size for __builtin___xxx_chk functions
907      typically determined by __builtin_object_size, or -1 if unknown.  */
908   unsigned HOST_WIDE_INT objsize;
909 
910   /* Number of the first variable argument.  */
911   unsigned HOST_WIDE_INT argidx;
912 
913   /* True for functions like snprintf that specify the size of
914      the destination, false for others like sprintf that don't.  */
915   bool bounded;
916 
917   /* True for bounded functions like snprintf that specify a zero-size
918      buffer as a request to compute the size of output without actually
919      writing any.  NOWRITE is cleared in response to the %n directive
920      which has side-effects similar to writing output.  */
921   bool nowrite;
922 
923   /* Return true if the called function's return value is used.  */
retval_usedcall_info924   bool retval_used () const
925   {
926     return gimple_get_lhs (callstmt);
927   }
928 
929   /* Return the warning option corresponding to the called function.  */
warnoptcall_info930   int warnopt () const
931   {
932     return bounded ? OPT_Wformat_truncation_ : OPT_Wformat_overflow_;
933   }
934 
935   /* Return true for calls to file formatted functions.  */
is_file_funccall_info936   bool is_file_func () const
937   {
938     return (fncode == BUILT_IN_FPRINTF
939 	    || fncode == BUILT_IN_FPRINTF_CHK
940 	    || fncode == BUILT_IN_FPRINTF_UNLOCKED
941 	    || fncode == BUILT_IN_VFPRINTF
942 	    || fncode == BUILT_IN_VFPRINTF_CHK);
943   }
944 
945   /* Return true for calls to string formatted functions.  */
is_string_funccall_info946   bool is_string_func () const
947   {
948     return (fncode == BUILT_IN_SPRINTF
949 	    || fncode == BUILT_IN_SPRINTF_CHK
950 	    || fncode == BUILT_IN_SNPRINTF
951 	    || fncode == BUILT_IN_SNPRINTF_CHK
952 	    || fncode == BUILT_IN_VSPRINTF
953 	    || fncode == BUILT_IN_VSPRINTF_CHK
954 	    || fncode == BUILT_IN_VSNPRINTF
955 	    || fncode == BUILT_IN_VSNPRINTF_CHK);
956   }
957 };
958 
959 /* Return the result of formatting a no-op directive (such as '%n').  */
960 
961 static fmtresult
format_none(const directive &,tree,const vr_values *)962 format_none (const directive &, tree, const vr_values *)
963 {
964   fmtresult res (0);
965   return res;
966 }
967 
968 /* Return the result of formatting the '%%' directive.  */
969 
970 static fmtresult
format_percent(const directive &,tree,const vr_values *)971 format_percent (const directive &, tree, const vr_values *)
972 {
973   fmtresult res (1);
974   return res;
975 }
976 
977 
978 /* Compute intmax_type_node and uintmax_type_node similarly to how
979    tree.c builds size_type_node.  */
980 
981 static void
build_intmax_type_nodes(tree * pintmax,tree * puintmax)982 build_intmax_type_nodes (tree *pintmax, tree *puintmax)
983 {
984   if (strcmp (UINTMAX_TYPE, "unsigned int") == 0)
985     {
986       *pintmax = integer_type_node;
987       *puintmax = unsigned_type_node;
988     }
989   else if (strcmp (UINTMAX_TYPE, "long unsigned int") == 0)
990     {
991       *pintmax = long_integer_type_node;
992       *puintmax = long_unsigned_type_node;
993     }
994   else if (strcmp (UINTMAX_TYPE, "long long unsigned int") == 0)
995     {
996       *pintmax = long_long_integer_type_node;
997       *puintmax = long_long_unsigned_type_node;
998     }
999   else
1000     {
1001       for (int i = 0; i < NUM_INT_N_ENTS; i++)
1002 	if (int_n_enabled_p[i])
1003 	  {
1004 	    char name[50], altname[50];
1005 	    sprintf (name, "__int%d unsigned", int_n_data[i].bitsize);
1006 	    sprintf (altname, "__int%d__ unsigned", int_n_data[i].bitsize);
1007 
1008 	    if (strcmp (name, UINTMAX_TYPE) == 0
1009 		|| strcmp (altname, UINTMAX_TYPE) == 0)
1010 	      {
1011 	        *pintmax = int_n_trees[i].signed_type;
1012 	        *puintmax = int_n_trees[i].unsigned_type;
1013 		return;
1014 	      }
1015 	  }
1016       gcc_unreachable ();
1017     }
1018 }
1019 
1020 /* Determine the range [*PMIN, *PMAX] that the expression ARG is
1021    in and that is representable in type int.
1022    Return true when the range is a subrange of that of int.
1023    When ARG is null it is as if it had the full range of int.
1024    When ABSOLUTE is true the range reflects the absolute value of
1025    the argument.  When ABSOLUTE is false, negative bounds of
1026    the determined range are replaced with NEGBOUND.  */
1027 
1028 static bool
get_int_range(tree arg,HOST_WIDE_INT * pmin,HOST_WIDE_INT * pmax,bool absolute,HOST_WIDE_INT negbound,const class vr_values * vr_values)1029 get_int_range (tree arg, HOST_WIDE_INT *pmin, HOST_WIDE_INT *pmax,
1030 	       bool absolute, HOST_WIDE_INT negbound,
1031 	       const class vr_values *vr_values)
1032 {
1033   /* The type of the result.  */
1034   const_tree type = integer_type_node;
1035 
1036   bool knownrange = false;
1037 
1038   if (!arg)
1039     {
1040       *pmin = tree_to_shwi (TYPE_MIN_VALUE (type));
1041       *pmax = tree_to_shwi (TYPE_MAX_VALUE (type));
1042     }
1043   else if (TREE_CODE (arg) == INTEGER_CST
1044 	   && TYPE_PRECISION (TREE_TYPE (arg)) <= TYPE_PRECISION (type))
1045     {
1046       /* For a constant argument return its value adjusted as specified
1047 	 by NEGATIVE and NEGBOUND and return true to indicate that the
1048 	 result is known.  */
1049       *pmin = tree_fits_shwi_p (arg) ? tree_to_shwi (arg) : tree_to_uhwi (arg);
1050       *pmax = *pmin;
1051       knownrange = true;
1052     }
1053   else
1054     {
1055       /* True if the argument's range cannot be determined.  */
1056       bool unknown = true;
1057 
1058       tree argtype = TREE_TYPE (arg);
1059 
1060       /* Ignore invalid arguments with greater precision that that
1061 	 of the expected type (e.g., in sprintf("%*i", 12LL, i)).
1062 	 They will have been detected and diagnosed by -Wformat and
1063 	 so it's not important to complicate this code to try to deal
1064 	 with them again.  */
1065       if (TREE_CODE (arg) == SSA_NAME
1066 	  && INTEGRAL_TYPE_P (argtype)
1067 	  && TYPE_PRECISION (argtype) <= TYPE_PRECISION (type))
1068 	{
1069 	  /* Try to determine the range of values of the integer argument.  */
1070 	  const value_range_equiv *vr
1071 	    = CONST_CAST (class vr_values *, vr_values)->get_value_range (arg);
1072 
1073 	  if (range_int_cst_p (vr))
1074 	    {
1075 	      HOST_WIDE_INT type_min
1076 		= (TYPE_UNSIGNED (argtype)
1077 		   ? tree_to_uhwi (TYPE_MIN_VALUE (argtype))
1078 		   : tree_to_shwi (TYPE_MIN_VALUE (argtype)));
1079 
1080 	      HOST_WIDE_INT type_max = tree_to_uhwi (TYPE_MAX_VALUE (argtype));
1081 
1082 	      *pmin = TREE_INT_CST_LOW (vr->min ());
1083 	      *pmax = TREE_INT_CST_LOW (vr->max ());
1084 
1085 	      if (*pmin < *pmax)
1086 		{
1087 		  /* Return true if the adjusted range is a subrange of
1088 		     the full range of the argument's type.  *PMAX may
1089 		     be less than *PMIN when the argument is unsigned
1090 		     and its upper bound is in excess of TYPE_MAX.  In
1091 		     that (invalid) case disregard the range and use that
1092 		     of the expected type instead.  */
1093 		  knownrange = type_min < *pmin || *pmax < type_max;
1094 
1095 		  unknown = false;
1096 		}
1097 	    }
1098 	}
1099 
1100       /* Handle an argument with an unknown range as if none had been
1101 	 provided.  */
1102       if (unknown)
1103 	return get_int_range (NULL_TREE, pmin, pmax, absolute,
1104 			      negbound, vr_values);
1105     }
1106 
1107   /* Adjust each bound as specified by ABSOLUTE and NEGBOUND.  */
1108   if (absolute)
1109     {
1110       if (*pmin < 0)
1111 	{
1112 	  if (*pmin == *pmax)
1113 	    *pmin = *pmax = -*pmin;
1114 	  else
1115 	    {
1116 	      /* Make sure signed overlow is avoided.  */
1117 	      gcc_assert (*pmin != HOST_WIDE_INT_MIN);
1118 
1119 	      HOST_WIDE_INT tmp = -*pmin;
1120 	      *pmin = 0;
1121 	      if (*pmax < tmp)
1122 		*pmax = tmp;
1123 	    }
1124 	}
1125     }
1126   else if (*pmin < negbound)
1127     *pmin = negbound;
1128 
1129   return knownrange;
1130 }
1131 
1132 /* With the range [*ARGMIN, *ARGMAX] of an integer directive's actual
1133    argument, due to the conversion from either *ARGMIN or *ARGMAX to
1134    the type of the directive's formal argument it's possible for both
1135    to result in the same number of bytes or a range of bytes that's
1136    less than the number of bytes that would result from formatting
1137    some other value in the range [*ARGMIN, *ARGMAX].  This can be
1138    determined by checking for the actual argument being in the range
1139    of the type of the directive.  If it isn't it must be assumed to
1140    take on the full range of the directive's type.
1141    Return true when the range has been adjusted to the full range
1142    of DIRTYPE, and false otherwise.  */
1143 
1144 static bool
adjust_range_for_overflow(tree dirtype,tree * argmin,tree * argmax)1145 adjust_range_for_overflow (tree dirtype, tree *argmin, tree *argmax)
1146 {
1147   tree argtype = TREE_TYPE (*argmin);
1148   unsigned argprec = TYPE_PRECISION (argtype);
1149   unsigned dirprec = TYPE_PRECISION (dirtype);
1150 
1151   /* If the actual argument and the directive's argument have the same
1152      precision and sign there can be no overflow and so there is nothing
1153      to adjust.  */
1154   if (argprec == dirprec && TYPE_SIGN (argtype) == TYPE_SIGN (dirtype))
1155     return false;
1156 
1157   /* The logic below was inspired/lifted from the CONVERT_EXPR_CODE_P
1158      branch in the extract_range_from_unary_expr function in tree-vrp.c.  */
1159 
1160   if (TREE_CODE (*argmin) == INTEGER_CST
1161       && TREE_CODE (*argmax) == INTEGER_CST
1162       && (dirprec >= argprec
1163 	  || integer_zerop (int_const_binop (RSHIFT_EXPR,
1164 					     int_const_binop (MINUS_EXPR,
1165 							      *argmax,
1166 							      *argmin),
1167 					     size_int (dirprec)))))
1168     {
1169       *argmin = force_fit_type (dirtype, wi::to_widest (*argmin), 0, false);
1170       *argmax = force_fit_type (dirtype, wi::to_widest (*argmax), 0, false);
1171 
1172       /* If *ARGMIN is still less than *ARGMAX the conversion above
1173 	 is safe.  Otherwise, it has overflowed and would be unsafe.  */
1174       if (tree_int_cst_le (*argmin, *argmax))
1175 	return false;
1176     }
1177 
1178   *argmin = TYPE_MIN_VALUE (dirtype);
1179   *argmax = TYPE_MAX_VALUE (dirtype);
1180   return true;
1181 }
1182 
1183 /* Return a range representing the minimum and maximum number of bytes
1184    that the format directive DIR will output for any argument given
1185    the WIDTH and PRECISION (extracted from DIR).  This function is
1186    used when the directive argument or its value isn't known.  */
1187 
1188 static fmtresult
format_integer(const directive & dir,tree arg,const vr_values * vr_values)1189 format_integer (const directive &dir, tree arg, const vr_values *vr_values)
1190 {
1191   tree intmax_type_node;
1192   tree uintmax_type_node;
1193 
1194   /* Base to format the number in.  */
1195   int base;
1196 
1197   /* True when a conversion is preceded by a prefix indicating the base
1198      of the argument (octal or hexadecimal).  */
1199   bool maybebase = dir.get_flag ('#');
1200 
1201   /* True when a signed conversion is preceded by a sign or space.  */
1202   bool maybesign = false;
1203 
1204   /* True for signed conversions (i.e., 'd' and 'i').  */
1205   bool sign = false;
1206 
1207   switch (dir.specifier)
1208     {
1209     case 'd':
1210     case 'i':
1211       /* Space and '+' are  only meaningful for signed conversions.  */
1212       maybesign = dir.get_flag (' ') | dir.get_flag ('+');
1213       sign = true;
1214       base = 10;
1215       break;
1216     case 'u':
1217       base = 10;
1218       break;
1219     case 'o':
1220       base = 8;
1221       break;
1222     case 'X':
1223     case 'x':
1224       base = 16;
1225       break;
1226     default:
1227       gcc_unreachable ();
1228     }
1229 
1230   /* The type of the "formal" argument expected by the directive.  */
1231   tree dirtype = NULL_TREE;
1232 
1233   /* Determine the expected type of the argument from the length
1234      modifier.  */
1235   switch (dir.modifier)
1236     {
1237     case FMT_LEN_none:
1238       if (dir.specifier == 'p')
1239 	dirtype = ptr_type_node;
1240       else
1241 	dirtype = sign ? integer_type_node : unsigned_type_node;
1242       break;
1243 
1244     case FMT_LEN_h:
1245       dirtype = sign ? short_integer_type_node : short_unsigned_type_node;
1246       break;
1247 
1248     case FMT_LEN_hh:
1249       dirtype = sign ? signed_char_type_node : unsigned_char_type_node;
1250       break;
1251 
1252     case FMT_LEN_l:
1253       dirtype = sign ? long_integer_type_node : long_unsigned_type_node;
1254       break;
1255 
1256     case FMT_LEN_L:
1257     case FMT_LEN_ll:
1258       dirtype = (sign
1259 		 ? long_long_integer_type_node
1260 		 : long_long_unsigned_type_node);
1261       break;
1262 
1263     case FMT_LEN_z:
1264       dirtype = signed_or_unsigned_type_for (!sign, size_type_node);
1265       break;
1266 
1267     case FMT_LEN_t:
1268       dirtype = signed_or_unsigned_type_for (!sign, ptrdiff_type_node);
1269       break;
1270 
1271     case FMT_LEN_j:
1272       build_intmax_type_nodes (&intmax_type_node, &uintmax_type_node);
1273       dirtype = sign ? intmax_type_node : uintmax_type_node;
1274       break;
1275 
1276     default:
1277       return fmtresult ();
1278     }
1279 
1280   /* The type of the argument to the directive, either deduced from
1281      the actual non-constant argument if one is known, or from
1282      the directive itself when none has been provided because it's
1283      a va_list.  */
1284   tree argtype = NULL_TREE;
1285 
1286   if (!arg)
1287     {
1288       /* When the argument has not been provided, use the type of
1289 	 the directive's argument as an approximation.  This will
1290 	 result in false positives for directives like %i with
1291 	 arguments with smaller precision (such as short or char).  */
1292       argtype = dirtype;
1293     }
1294   else if (TREE_CODE (arg) == INTEGER_CST)
1295     {
1296       /* When a constant argument has been provided use its value
1297 	 rather than type to determine the length of the output.  */
1298       fmtresult res;
1299 
1300       if ((dir.prec[0] <= 0 && dir.prec[1] >= 0) && integer_zerop (arg))
1301 	{
1302 	  /* As a special case, a precision of zero with a zero argument
1303 	     results in zero bytes except in base 8 when the '#' flag is
1304 	     specified, and for signed conversions in base 8 and 10 when
1305 	     either the space or '+' flag has been specified and it results
1306 	     in just one byte (with width having the normal effect).  This
1307 	     must extend to the case of a specified precision with
1308 	     an unknown value because it can be zero.  */
1309 	  res.range.min = ((base == 8 && dir.get_flag ('#')) || maybesign);
1310 	  if (res.range.min == 0 && dir.prec[0] != dir.prec[1])
1311 	    {
1312 	      res.range.max = 1;
1313 	      res.range.likely = 1;
1314 	    }
1315 	  else
1316 	    {
1317 	      res.range.max = res.range.min;
1318 	      res.range.likely = res.range.min;
1319 	    }
1320 	}
1321       else
1322 	{
1323 	  /* Convert the argument to the type of the directive.  */
1324 	  arg = fold_convert (dirtype, arg);
1325 
1326 	  res.range.min = tree_digits (arg, base, dir.prec[0],
1327 				       maybesign, maybebase);
1328 	  if (dir.prec[0] == dir.prec[1])
1329 	    res.range.max = res.range.min;
1330 	  else
1331 	    res.range.max = tree_digits (arg, base, dir.prec[1],
1332 					 maybesign, maybebase);
1333 	  res.range.likely = res.range.min;
1334 	  res.knownrange = true;
1335 	}
1336 
1337       res.range.unlikely = res.range.max;
1338 
1339       /* Bump up the counters if WIDTH is greater than LEN.  */
1340       res.adjust_for_width_or_precision (dir.width, dirtype, base,
1341 					 (sign | maybebase) + (base == 16));
1342       /* Bump up the counters again if PRECision is greater still.  */
1343       res.adjust_for_width_or_precision (dir.prec, dirtype, base,
1344 					 (sign | maybebase) + (base == 16));
1345 
1346       return res;
1347     }
1348   else if (INTEGRAL_TYPE_P (TREE_TYPE (arg))
1349 	   || TREE_CODE (TREE_TYPE (arg)) == POINTER_TYPE)
1350     /* Determine the type of the provided non-constant argument.  */
1351     argtype = TREE_TYPE (arg);
1352   else
1353     /* Don't bother with invalid arguments since they likely would
1354        have already been diagnosed, and disable any further checking
1355        of the format string by returning [-1, -1].  */
1356     return fmtresult ();
1357 
1358   fmtresult res;
1359 
1360   /* Using either the range the non-constant argument is in, or its
1361      type (either "formal" or actual), create a range of values that
1362      constrain the length of output given the warning level.  */
1363   tree argmin = NULL_TREE;
1364   tree argmax = NULL_TREE;
1365 
1366   if (arg
1367       && TREE_CODE (arg) == SSA_NAME
1368       && INTEGRAL_TYPE_P (argtype))
1369     {
1370       /* Try to determine the range of values of the integer argument
1371 	 (range information is not available for pointers).  */
1372       const value_range_equiv *vr
1373 	= CONST_CAST (class vr_values *, vr_values)->get_value_range (arg);
1374 
1375       if (range_int_cst_p (vr))
1376 	{
1377 	  argmin = vr->min ();
1378 	  argmax = vr->max ();
1379 
1380 	  /* Set KNOWNRANGE if the argument is in a known subrange
1381 	     of the directive's type and neither width nor precision
1382 	     is unknown.  (KNOWNRANGE may be reset below).  */
1383 	  res.knownrange
1384 	    = ((!tree_int_cst_equal (TYPE_MIN_VALUE (dirtype), argmin)
1385 		|| !tree_int_cst_equal (TYPE_MAX_VALUE (dirtype), argmax))
1386 	       && dir.known_width_and_precision ());
1387 
1388 	  res.argmin = argmin;
1389 	  res.argmax = argmax;
1390 	}
1391       else if (vr->kind () == VR_ANTI_RANGE)
1392 	{
1393 	  /* Handle anti-ranges if/when bug 71690 is resolved.  */
1394 	}
1395       else if (vr->varying_p () || vr->undefined_p ())
1396 	{
1397 	  /* The argument here may be the result of promoting the actual
1398 	     argument to int.  Try to determine the type of the actual
1399 	     argument before promotion and narrow down its range that
1400 	     way.  */
1401 	  gimple *def = SSA_NAME_DEF_STMT (arg);
1402 	  if (is_gimple_assign (def))
1403 	    {
1404 	      tree_code code = gimple_assign_rhs_code (def);
1405 	      if (code == INTEGER_CST)
1406 		{
1407 		  arg = gimple_assign_rhs1 (def);
1408 		  return format_integer (dir, arg, vr_values);
1409 		}
1410 
1411 	      if (code == NOP_EXPR)
1412 		{
1413 		  tree type = TREE_TYPE (gimple_assign_rhs1 (def));
1414 		  if (INTEGRAL_TYPE_P (type)
1415 		      || TREE_CODE (type) == POINTER_TYPE)
1416 		    argtype = type;
1417 		}
1418 	    }
1419 	}
1420     }
1421 
1422   if (!argmin)
1423     {
1424       if (TREE_CODE (argtype) == POINTER_TYPE)
1425 	{
1426 	  argmin = build_int_cst (pointer_sized_int_node, 0);
1427 	  argmax = build_all_ones_cst (pointer_sized_int_node);
1428 	}
1429       else
1430 	{
1431 	  argmin = TYPE_MIN_VALUE (argtype);
1432 	  argmax = TYPE_MAX_VALUE (argtype);
1433 	}
1434     }
1435 
1436   /* Clear KNOWNRANGE if the range has been adjusted to the maximum
1437      of the directive.  If it has been cleared then since ARGMIN and/or
1438      ARGMAX have been adjusted also adjust the corresponding ARGMIN and
1439      ARGMAX in the result to include in diagnostics.  */
1440   if (adjust_range_for_overflow (dirtype, &argmin, &argmax))
1441     {
1442       res.knownrange = false;
1443       res.argmin = argmin;
1444       res.argmax = argmax;
1445     }
1446 
1447   /* Recursively compute the minimum and maximum from the known range.  */
1448   if (TYPE_UNSIGNED (dirtype) || tree_int_cst_sgn (argmin) >= 0)
1449     {
1450       /* For unsigned conversions/directives or signed when
1451 	 the minimum is positive, use the minimum and maximum to compute
1452 	 the shortest and longest output, respectively.  */
1453       res.range.min = format_integer (dir, argmin, vr_values).range.min;
1454       res.range.max = format_integer (dir, argmax, vr_values).range.max;
1455     }
1456   else if (tree_int_cst_sgn (argmax) < 0)
1457     {
1458       /* For signed conversions/directives if maximum is negative,
1459 	 use the minimum as the longest output and maximum as the
1460 	 shortest output.  */
1461       res.range.min = format_integer (dir, argmax, vr_values).range.min;
1462       res.range.max = format_integer (dir, argmin, vr_values).range.max;
1463     }
1464   else
1465     {
1466       /* Otherwise, 0 is inside of the range and minimum negative.  Use 0
1467 	 as the shortest output and for the longest output compute the
1468 	 length of the output of both minimum and maximum and pick the
1469 	 longer.  */
1470       unsigned HOST_WIDE_INT max1
1471 	= format_integer (dir, argmin, vr_values).range.max;
1472       unsigned HOST_WIDE_INT max2
1473 	= format_integer (dir, argmax, vr_values).range.max;
1474       res.range.min
1475 	= format_integer (dir, integer_zero_node, vr_values).range.min;
1476       res.range.max = MAX (max1, max2);
1477     }
1478 
1479   /* If the range is known, use the maximum as the likely length.  */
1480   if (res.knownrange)
1481     res.range.likely = res.range.max;
1482   else
1483     {
1484       /* Otherwise, use the minimum.  Except for the case where for %#x or
1485          %#o the minimum is just for a single value in the range (0) and
1486          for all other values it is something longer, like 0x1 or 01.
1487 	  Use the length for value 1 in that case instead as the likely
1488 	  length.  */
1489       res.range.likely = res.range.min;
1490       if (maybebase
1491 	  && base != 10
1492 	  && (tree_int_cst_sgn (argmin) < 0 || tree_int_cst_sgn (argmax) > 0))
1493 	{
1494 	  if (res.range.min == 1)
1495 	    res.range.likely += base == 8 ? 1 : 2;
1496 	  else if (res.range.min == 2
1497 		   && base == 16
1498 		   && (dir.width[0] == 2 || dir.prec[0] == 2))
1499 	    ++res.range.likely;
1500 	}
1501     }
1502 
1503   res.range.unlikely = res.range.max;
1504   res.adjust_for_width_or_precision (dir.width, dirtype, base,
1505 				     (sign | maybebase) + (base == 16));
1506   res.adjust_for_width_or_precision (dir.prec, dirtype, base,
1507 				     (sign | maybebase) + (base == 16));
1508 
1509   return res;
1510 }
1511 
1512 /* Return the number of bytes that a format directive consisting of FLAGS,
1513    PRECision, format SPECification, and MPFR rounding specifier RNDSPEC,
1514    would result for argument X under ideal conditions (i.e., if PREC
1515    weren't excessive).  MPFR 3.1 allocates large amounts of memory for
1516    values of PREC with large magnitude and can fail (see MPFR bug #21056).
1517    This function works around those problems.  */
1518 
1519 static unsigned HOST_WIDE_INT
get_mpfr_format_length(mpfr_ptr x,const char * flags,HOST_WIDE_INT prec,char spec,char rndspec)1520 get_mpfr_format_length (mpfr_ptr x, const char *flags, HOST_WIDE_INT prec,
1521 			char spec, char rndspec)
1522 {
1523   char fmtstr[40];
1524 
1525   HOST_WIDE_INT len = strlen (flags);
1526 
1527   fmtstr[0] = '%';
1528   memcpy (fmtstr + 1, flags, len);
1529   memcpy (fmtstr + 1 + len, ".*R", 3);
1530   fmtstr[len + 4] = rndspec;
1531   fmtstr[len + 5] = spec;
1532   fmtstr[len + 6] = '\0';
1533 
1534   spec = TOUPPER (spec);
1535   if (spec == 'E' || spec == 'F')
1536     {
1537       /* For %e, specify the precision explicitly since mpfr_sprintf
1538 	 does its own thing just to be different (see MPFR bug 21088).  */
1539       if (prec < 0)
1540 	prec = 6;
1541     }
1542   else
1543     {
1544       /* Avoid passing negative precisions with larger magnitude to MPFR
1545 	 to avoid exposing its bugs.  (A negative precision is supposed
1546 	 to be ignored.)  */
1547       if (prec < 0)
1548 	prec = -1;
1549     }
1550 
1551   HOST_WIDE_INT p = prec;
1552 
1553   if (spec == 'G' && !strchr (flags, '#'))
1554     {
1555       /* For G/g without the pound flag, precision gives the maximum number
1556 	 of significant digits which is bounded by LDBL_MAX_10_EXP, or, for
1557 	 a 128 bit IEEE extended precision, 4932.  Using twice as much here
1558 	 should be more than sufficient for any real format.  */
1559       if ((IEEE_MAX_10_EXP * 2) < prec)
1560 	prec = IEEE_MAX_10_EXP * 2;
1561       p = prec;
1562     }
1563   else
1564     {
1565       /* Cap precision arbitrarily at 1KB and add the difference
1566 	 (if any) to the MPFR result.  */
1567       if (prec > 1024)
1568 	p = 1024;
1569     }
1570 
1571   len = mpfr_snprintf (NULL, 0, fmtstr, (int)p, x);
1572 
1573   /* Handle the unlikely (impossible?) error by returning more than
1574      the maximum dictated by the function's return type.  */
1575   if (len < 0)
1576     return target_dir_max () + 1;
1577 
1578   /* Adjust the return value by the difference.  */
1579   if (p < prec)
1580     len += prec - p;
1581 
1582   return len;
1583 }
1584 
1585 /* Return the number of bytes to format using the format specifier
1586    SPEC and the precision PREC the largest value in the real floating
1587    TYPE.  */
1588 
1589 static unsigned HOST_WIDE_INT
format_floating_max(tree type,char spec,HOST_WIDE_INT prec)1590 format_floating_max (tree type, char spec, HOST_WIDE_INT prec)
1591 {
1592   machine_mode mode = TYPE_MODE (type);
1593 
1594   /* IBM Extended mode.  */
1595   if (MODE_COMPOSITE_P (mode))
1596     mode = DFmode;
1597 
1598   /* Get the real type format desription for the target.  */
1599   const real_format *rfmt = REAL_MODE_FORMAT (mode);
1600   REAL_VALUE_TYPE rv;
1601 
1602   real_maxval (&rv, 0, mode);
1603 
1604   /* Convert the GCC real value representation with the precision
1605      of the real type to the mpfr_t format with the GCC default
1606      round-to-nearest mode.  */
1607   mpfr_t x;
1608   mpfr_init2 (x, rfmt->p);
1609   mpfr_from_real (x, &rv, MPFR_RNDN);
1610 
1611   /* Return a value one greater to account for the leading minus sign.  */
1612   unsigned HOST_WIDE_INT r
1613     = 1 + get_mpfr_format_length (x, "", prec, spec, 'D');
1614   mpfr_clear (x);
1615   return r;
1616 }
1617 
1618 /* Return a range representing the minimum and maximum number of bytes
1619    that the directive DIR will output for any argument.  PREC gives
1620    the adjusted precision range to account for negative precisions
1621    meaning the default 6.  This function is used when the directive
1622    argument or its value isn't known.  */
1623 
1624 static fmtresult
format_floating(const directive & dir,const HOST_WIDE_INT prec[2])1625 format_floating (const directive &dir, const HOST_WIDE_INT prec[2])
1626 {
1627   tree type;
1628 
1629   switch (dir.modifier)
1630     {
1631     case FMT_LEN_l:
1632     case FMT_LEN_none:
1633       type = double_type_node;
1634       break;
1635 
1636     case FMT_LEN_L:
1637       type = long_double_type_node;
1638       break;
1639 
1640     case FMT_LEN_ll:
1641       type = long_double_type_node;
1642       break;
1643 
1644     default:
1645       return fmtresult ();
1646     }
1647 
1648   /* The minimum and maximum number of bytes produced by the directive.  */
1649   fmtresult res;
1650 
1651   /* The minimum output as determined by flags.  It's always at least 1.
1652      When plus or space are set the output is preceded by either a sign
1653      or a space.  */
1654   unsigned flagmin = (1 /* for the first digit */
1655 		      + (dir.get_flag ('+') | dir.get_flag (' ')));
1656 
1657   /* The minimum is 3 for "inf" and "nan" for all specifiers, plus 1
1658      for the plus sign/space with the '+' and ' ' flags, respectively,
1659      unless reduced below.  */
1660   res.range.min = 2 + flagmin;
1661 
1662   /* When the pound flag is set the decimal point is included in output
1663      regardless of precision.  Whether or not a decimal point is included
1664      otherwise depends on the specification and precision.  */
1665   bool radix = dir.get_flag ('#');
1666 
1667   switch (dir.specifier)
1668     {
1669     case 'A':
1670     case 'a':
1671       {
1672 	HOST_WIDE_INT minprec = 6 + !radix /* decimal point */;
1673 	if (dir.prec[0] <= 0)
1674 	  minprec = 0;
1675 	else if (dir.prec[0] > 0)
1676 	  minprec = dir.prec[0] + !radix /* decimal point */;
1677 
1678 	res.range.likely = (2 /* 0x */
1679 			    + flagmin
1680 			    + radix
1681 			    + minprec
1682 			    + 3 /* p+0 */);
1683 
1684 	res.range.max = format_floating_max (type, 'a', prec[1]);
1685 
1686 	/* The unlikely maximum accounts for the longest multibyte
1687 	   decimal point character.  */
1688 	res.range.unlikely = res.range.max;
1689 	if (dir.prec[1] > 0)
1690 	  res.range.unlikely += target_mb_len_max () - 1;
1691 
1692 	break;
1693       }
1694 
1695     case 'E':
1696     case 'e':
1697       {
1698 	/* Minimum output attributable to precision and, when it's
1699 	   non-zero, decimal point.  */
1700 	HOST_WIDE_INT minprec = prec[0] ? prec[0] + !radix : 0;
1701 
1702 	/* The likely minimum output is "[-+]1.234567e+00" regardless
1703 	   of the value of the actual argument.  */
1704 	res.range.likely = (flagmin
1705 			    + radix
1706 			    + minprec
1707 			    + 2 /* e+ */ + 2);
1708 
1709 	res.range.max = format_floating_max (type, 'e', prec[1]);
1710 
1711 	/* The unlikely maximum accounts for the longest multibyte
1712 	   decimal point character.  */
1713 	if (dir.prec[0] != dir.prec[1]
1714 	    || dir.prec[0] == -1 || dir.prec[0] > 0)
1715 	  res.range.unlikely = res.range.max + target_mb_len_max () -1;
1716 	else
1717 	  res.range.unlikely = res.range.max;
1718 	break;
1719       }
1720 
1721     case 'F':
1722     case 'f':
1723       {
1724 	/* Minimum output attributable to precision and, when it's non-zero,
1725 	   decimal point.  */
1726 	HOST_WIDE_INT minprec = prec[0] ? prec[0] + !radix : 0;
1727 
1728 	/* For finite numbers (i.e., not infinity or NaN) the lower bound
1729 	   when precision isn't specified is 8 bytes ("1.23456" since
1730 	   precision is taken to be 6).  When precision is zero, the lower
1731 	   bound is 1 byte (e.g., "1").  Otherwise, when precision is greater
1732 	   than zero, then the lower bound is 2 plus precision (plus flags).
1733 	   But in all cases, the lower bound is no greater than 3.  */
1734 	unsigned HOST_WIDE_INT min = flagmin + radix + minprec;
1735 	if (min < res.range.min)
1736 	  res.range.min = min;
1737 
1738 	/* Compute the upper bound for -TYPE_MAX.  */
1739 	res.range.max = format_floating_max (type, 'f', prec[1]);
1740 
1741 	/* The minimum output with unknown precision is a single byte
1742 	   (e.g., "0") but the more likely output is 3 bytes ("0.0").  */
1743 	if (dir.prec[0] < 0 && dir.prec[1] > 0)
1744 	  res.range.likely = 3;
1745 	else
1746 	  res.range.likely = min;
1747 
1748 	/* The unlikely maximum accounts for the longest multibyte
1749 	   decimal point character.  */
1750 	if (dir.prec[0] != dir.prec[1]
1751 	    || dir.prec[0] == -1 || dir.prec[0] > 0)
1752 	  res.range.unlikely = res.range.max + target_mb_len_max () - 1;
1753 	break;
1754       }
1755 
1756     case 'G':
1757     case 'g':
1758       {
1759 	/* The %g output depends on precision and the exponent of
1760 	   the argument.  Since the value of the argument isn't known
1761 	   the lower bound on the range of bytes (not counting flags
1762 	   or width) is 1 plus radix (i.e., either "0" or "0." for
1763 	   "%g" and "%#g", respectively, with a zero argument).  */
1764 	unsigned HOST_WIDE_INT min = flagmin + radix;
1765 	if (min < res.range.min)
1766 	  res.range.min = min;
1767 
1768 	char spec = 'g';
1769 	HOST_WIDE_INT maxprec = dir.prec[1];
1770 	if (radix && maxprec)
1771 	  {
1772 	    /* When the pound flag (radix) is set, trailing zeros aren't
1773 	       trimmed and so the longest output is the same as for %e,
1774 	       except with precision minus 1 (as specified in C11).  */
1775 	    spec = 'e';
1776 	    if (maxprec > 0)
1777 	      --maxprec;
1778 	    else if (maxprec < 0)
1779 	      maxprec = 5;
1780 	  }
1781 	else
1782 	  maxprec = prec[1];
1783 
1784 	res.range.max = format_floating_max (type, spec, maxprec);
1785 
1786 	/* The likely output is either the maximum computed above
1787 	   minus 1 (assuming the maximum is positive) when precision
1788 	   is known (or unspecified), or the same minimum as for %e
1789 	   (which is computed for a non-negative argument).  Unlike
1790 	   for the other specifiers above the likely output isn't
1791 	   the minimum because for %g that's 1 which is unlikely.  */
1792 	if (dir.prec[1] < 0
1793 	    || (unsigned HOST_WIDE_INT)dir.prec[1] < target_int_max ())
1794 	  res.range.likely = res.range.max - 1;
1795 	else
1796 	  {
1797 	    HOST_WIDE_INT minprec = 6 + !radix /* decimal point */;
1798 	    res.range.likely = (flagmin
1799 				+ radix
1800 				+ minprec
1801 				+ 2 /* e+ */ + 2);
1802 	  }
1803 
1804 	/* The unlikely maximum accounts for the longest multibyte
1805 	   decimal point character.  */
1806 	res.range.unlikely = res.range.max + target_mb_len_max () - 1;
1807 	break;
1808       }
1809 
1810     default:
1811       return fmtresult ();
1812     }
1813 
1814   /* Bump up the byte counters if WIDTH is greater.  */
1815   res.adjust_for_width_or_precision (dir.width);
1816   return res;
1817 }
1818 
1819 /* Return a range representing the minimum and maximum number of bytes
1820    that the directive DIR will write on output for the floating argument
1821    ARG.  */
1822 
1823 static fmtresult
format_floating(const directive & dir,tree arg,const vr_values *)1824 format_floating (const directive &dir, tree arg, const vr_values *)
1825 {
1826   HOST_WIDE_INT prec[] = { dir.prec[0], dir.prec[1] };
1827   tree type = (dir.modifier == FMT_LEN_L || dir.modifier == FMT_LEN_ll
1828 	       ? long_double_type_node : double_type_node);
1829 
1830   /* For an indeterminate precision the lower bound must be assumed
1831      to be zero.  */
1832   if (TOUPPER (dir.specifier) == 'A')
1833     {
1834       /* Get the number of fractional decimal digits needed to represent
1835 	 the argument without a loss of accuracy.  */
1836       unsigned fmtprec
1837 	= REAL_MODE_FORMAT (TYPE_MODE (type))->p;
1838 
1839       /* The precision of the IEEE 754 double format is 53.
1840 	 The precision of all other GCC binary double formats
1841 	 is 56 or less.  */
1842       unsigned maxprec = fmtprec <= 56 ? 13 : 15;
1843 
1844       /* For %a, leave the minimum precision unspecified to let
1845 	 MFPR trim trailing zeros (as it and many other systems
1846 	 including Glibc happen to do) and set the maximum
1847 	 precision to reflect what it would be with trailing zeros
1848 	 present (as Solaris and derived systems do).  */
1849       if (dir.prec[1] < 0)
1850 	{
1851 	  /* Both bounds are negative implies that precision has
1852 	     not been specified.  */
1853 	  prec[0] = maxprec;
1854 	  prec[1] = -1;
1855 	}
1856       else if (dir.prec[0] < 0)
1857 	{
1858 	  /* With a negative lower bound and a non-negative upper
1859 	     bound set the minimum precision to zero and the maximum
1860 	     to the greater of the maximum precision (i.e., with
1861 	     trailing zeros present) and the specified upper bound.  */
1862 	  prec[0] = 0;
1863 	  prec[1] = dir.prec[1] < maxprec ? maxprec : dir.prec[1];
1864 	}
1865     }
1866   else if (dir.prec[0] < 0)
1867     {
1868       if (dir.prec[1] < 0)
1869 	{
1870 	  /* A precision in a strictly negative range is ignored and
1871 	     the default of 6 is used instead.  */
1872 	  prec[0] = prec[1] = 6;
1873 	}
1874       else
1875 	{
1876 	  /* For a precision in a partly negative range, the lower bound
1877 	     must be assumed to be zero and the new upper bound is the
1878 	     greater of 6 (the default precision used when the specified
1879 	     precision is negative) and the upper bound of the specified
1880 	     range.  */
1881 	  prec[0] = 0;
1882 	  prec[1] = dir.prec[1] < 6 ? 6 : dir.prec[1];
1883 	}
1884     }
1885 
1886   if (!arg
1887       || TREE_CODE (arg) != REAL_CST
1888       || !useless_type_conversion_p (type, TREE_TYPE (arg)))
1889     return format_floating (dir, prec);
1890 
1891   /* The minimum and maximum number of bytes produced by the directive.  */
1892   fmtresult res;
1893 
1894   /* Get the real type format desription for the target.  */
1895   const REAL_VALUE_TYPE *rvp = TREE_REAL_CST_PTR (arg);
1896   const real_format *rfmt = REAL_MODE_FORMAT (TYPE_MODE (TREE_TYPE (arg)));
1897 
1898   if (!real_isfinite (rvp))
1899     {
1900       /* The format for Infinity and NaN is "[-]inf"/"[-]infinity"
1901 	 and "[-]nan" with the choice being implementation-defined
1902 	 but not locale dependent.  */
1903       bool sign = dir.get_flag ('+') || real_isneg (rvp);
1904       res.range.min = 3 + sign;
1905 
1906       res.range.likely = res.range.min;
1907       res.range.max = res.range.min;
1908       /* The unlikely maximum is "[-/+]infinity" or "[-/+][qs]nan".
1909 	 For NaN, the C/POSIX standards specify two formats:
1910 	   "[-/+]nan"
1911 	 and
1912 	   "[-/+]nan(n-char-sequence)"
1913 	 No known printf implementation outputs the latter format but AIX
1914 	 outputs QNaN and SNaN for quiet and signalling NaN, respectively,
1915 	 so the unlikely maximum reflects that.  */
1916       res.range.unlikely = sign + (real_isinf (rvp) ? 8 : 4);
1917 
1918       /* The range for infinity and NaN is known unless either width
1919 	 or precision is unknown.  Width has the same effect regardless
1920 	 of whether the argument is finite.  Precision is either ignored
1921 	 (e.g., Glibc) or can have an effect on the short vs long format
1922 	 such as inf/infinity (e.g., Solaris).  */
1923       res.knownrange = dir.known_width_and_precision ();
1924 
1925       /* Adjust the range for width but ignore precision.  */
1926       res.adjust_for_width_or_precision (dir.width);
1927 
1928       return res;
1929     }
1930 
1931   char fmtstr [40];
1932   char *pfmt = fmtstr;
1933 
1934   /* Append flags.  */
1935   for (const char *pf = "-+ #0"; *pf; ++pf)
1936     if (dir.get_flag (*pf))
1937       *pfmt++ = *pf;
1938 
1939   *pfmt = '\0';
1940 
1941   {
1942     /* Set up an array to easily iterate over.  */
1943     unsigned HOST_WIDE_INT* const minmax[] = {
1944       &res.range.min, &res.range.max
1945     };
1946 
1947     for (int i = 0; i != sizeof minmax / sizeof *minmax; ++i)
1948       {
1949 	/* Convert the GCC real value representation with the precision
1950 	   of the real type to the mpfr_t format rounding down in the
1951 	   first iteration that computes the minimm and up in the second
1952 	   that computes the maximum.  This order is arbibtrary because
1953 	   rounding in either direction can result in longer output.  */
1954 	mpfr_t mpfrval;
1955 	mpfr_init2 (mpfrval, rfmt->p);
1956 	mpfr_from_real (mpfrval, rvp, i ? MPFR_RNDU : MPFR_RNDD);
1957 
1958 	/* Use the MPFR rounding specifier to round down in the first
1959 	   iteration and then up.  In most but not all cases this will
1960 	   result in the same number of bytes.  */
1961 	char rndspec = "DU"[i];
1962 
1963 	/* Format it and store the result in the corresponding member
1964 	   of the result struct.  */
1965 	*minmax[i] = get_mpfr_format_length (mpfrval, fmtstr, prec[i],
1966 					     dir.specifier, rndspec);
1967 	mpfr_clear (mpfrval);
1968       }
1969   }
1970 
1971   /* Make sure the minimum is less than the maximum (MPFR rounding
1972      in the call to mpfr_snprintf can result in the reverse.  */
1973   if (res.range.max < res.range.min)
1974     {
1975       unsigned HOST_WIDE_INT tmp = res.range.min;
1976       res.range.min = res.range.max;
1977       res.range.max = tmp;
1978     }
1979 
1980   /* The range is known unless either width or precision is unknown.  */
1981   res.knownrange = dir.known_width_and_precision ();
1982 
1983   /* For the same floating point constant, unless width or precision
1984      is unknown, use the longer output as the likely maximum since
1985      with round to nearest either is equally likely.  Otheriwse, when
1986      precision is unknown, use the greater of the minimum and 3 as
1987      the likely output (for "0.0" since zero precision is unlikely).  */
1988   if (res.knownrange)
1989     res.range.likely = res.range.max;
1990   else if (res.range.min < 3
1991 	   && dir.prec[0] < 0
1992 	   && (unsigned HOST_WIDE_INT)dir.prec[1] == target_int_max ())
1993     res.range.likely = 3;
1994   else
1995     res.range.likely = res.range.min;
1996 
1997   res.range.unlikely = res.range.max;
1998 
1999   if (res.range.max > 2 && (prec[0] != 0 || prec[1] != 0))
2000     {
2001       /* Unless the precision is zero output longer than 2 bytes may
2002 	 include the decimal point which must be a single character
2003 	 up to MB_LEN_MAX in length.  This is overly conservative
2004 	 since in some conversions some constants result in no decimal
2005 	 point (e.g., in %g).  */
2006       res.range.unlikely += target_mb_len_max () - 1;
2007     }
2008 
2009   res.adjust_for_width_or_precision (dir.width);
2010   return res;
2011 }
2012 
2013 /* Return a FMTRESULT struct set to the lengths of the shortest and longest
2014    strings referenced by the expression STR, or (-1, -1) when not known.
2015    Used by the format_string function below.  */
2016 
2017 static fmtresult
get_string_length(tree str,unsigned eltsize,const vr_values * vr)2018 get_string_length (tree str, unsigned eltsize, const vr_values *vr)
2019 {
2020   if (!str)
2021     return fmtresult ();
2022 
2023   /* Try to determine the dynamic string length first.
2024      Set MAXBOUND to an arbitrary non-null non-integer node as a request
2025      to have it set to the length of the longest string in a PHI.  */
2026   c_strlen_data lendata = { };
2027   lendata.maxbound = str;
2028   if (eltsize == 1)
2029     get_range_strlen_dynamic (str, &lendata, vr);
2030   else
2031     {
2032       /* Determine the length of the shortest and longest string referenced
2033 	 by STR.  Strings of unknown lengths are bounded by the sizes of
2034 	 arrays that subexpressions of STR may refer to.  Pointers that
2035 	 aren't known to point any such arrays result in LENDATA.MAXLEN
2036 	 set to SIZE_MAX.  */
2037       get_range_strlen (str, &lendata, eltsize);
2038     }
2039 
2040   /* If LENDATA.MAXBOUND is not equal to .MINLEN it corresponds to the bound
2041      of the largest array STR refers to, if known, or it's set to SIZE_MAX
2042      otherwise.  */
2043 
2044   /* Return the default result when nothing is known about the string.  */
2045   if ((lendata.maxbound && !tree_fits_uhwi_p (lendata.maxbound))
2046       || !tree_fits_uhwi_p (lendata.maxlen))
2047     {
2048       fmtresult res;
2049       res.nonstr = lendata.decl;
2050       return res;
2051     }
2052 
2053   unsigned HOST_WIDE_INT lenmax = tree_to_uhwi (max_object_size ()) - 2;
2054   if (integer_zerop (lendata.minlen)
2055       && (!lendata.maxbound || lenmax <= tree_to_uhwi (lendata.maxbound))
2056       && lenmax <= tree_to_uhwi (lendata.maxlen))
2057     {
2058       fmtresult res;
2059       res.nonstr = lendata.decl;
2060       return res;
2061     }
2062 
2063   HOST_WIDE_INT min
2064     = (tree_fits_uhwi_p (lendata.minlen)
2065        ? tree_to_uhwi (lendata.minlen)
2066        : 0);
2067 
2068   HOST_WIDE_INT max
2069     = (lendata.maxbound && tree_fits_uhwi_p (lendata.maxbound)
2070        ? tree_to_uhwi (lendata.maxbound)
2071        : HOST_WIDE_INT_M1U);
2072 
2073   const bool unbounded = integer_all_onesp (lendata.maxlen);
2074 
2075   /* Set the max/likely counters to unbounded when a minimum is known
2076      but the maximum length isn't bounded.  This implies that STR is
2077      a conditional expression involving a string of known length and
2078      and an expression of unknown/unbounded length.  */
2079   if (min
2080       && (unsigned HOST_WIDE_INT)min < HOST_WIDE_INT_M1U
2081       && unbounded)
2082     max = HOST_WIDE_INT_M1U;
2083 
2084   /* get_range_strlen() returns the target value of SIZE_MAX for
2085      strings of unknown length.  Bump it up to HOST_WIDE_INT_M1U
2086      which may be bigger.  */
2087   if ((unsigned HOST_WIDE_INT)min == target_size_max ())
2088     min = HOST_WIDE_INT_M1U;
2089   if ((unsigned HOST_WIDE_INT)max == target_size_max ())
2090     max = HOST_WIDE_INT_M1U;
2091 
2092   fmtresult res (min, max);
2093   res.nonstr = lendata.decl;
2094 
2095   /* Set RES.KNOWNRANGE to true if and only if all strings referenced
2096      by STR are known to be bounded (though not necessarily by their
2097      actual length but perhaps by their maximum possible length).  */
2098   if (res.range.max < target_int_max ())
2099     {
2100       res.knownrange = true;
2101       /* When the length of the longest string is known and not
2102 	 excessive use it as the likely length of the string(s).  */
2103       res.range.likely = res.range.max;
2104     }
2105   else
2106     {
2107       /* When the upper bound is unknown (it can be zero or excessive)
2108 	 set the likely length to the greater of 1.  If MAXBOUND is
2109 	 known, also reset the length of the lower bound to zero.  */
2110       res.range.likely = res.range.min ? res.range.min : warn_level > 1;
2111       if (lendata.maxbound && !integer_all_onesp (lendata.maxbound))
2112 	res.range.min = 0;
2113     }
2114 
2115   res.range.unlikely = unbounded ? HOST_WIDE_INT_MAX : res.range.max;
2116 
2117   return res;
2118 }
2119 
2120 /* Return the minimum and maximum number of characters formatted
2121    by the '%c' format directives and its wide character form for
2122    the argument ARG.  ARG can be null (for functions such as
2123    vsprinf).  */
2124 
2125 static fmtresult
format_character(const directive & dir,tree arg,const vr_values * vr_values)2126 format_character (const directive &dir, tree arg, const vr_values *vr_values)
2127 {
2128   fmtresult res;
2129 
2130   res.knownrange = true;
2131 
2132   if (dir.specifier == 'C'
2133       || dir.modifier == FMT_LEN_l)
2134     {
2135       /* A wide character can result in as few as zero bytes.  */
2136       res.range.min = 0;
2137 
2138       HOST_WIDE_INT min, max;
2139       if (get_int_range (arg, &min, &max, false, 0, vr_values))
2140 	{
2141 	  if (min == 0 && max == 0)
2142 	    {
2143 	      /* The NUL wide character results in no bytes.  */
2144 	      res.range.max = 0;
2145 	      res.range.likely = 0;
2146 	      res.range.unlikely = 0;
2147 	    }
2148 	  else if (min >= 0 && min < 128)
2149 	    {
2150 	      /* Be conservative if the target execution character set
2151 		 is not a 1-to-1 mapping to the source character set or
2152 		 if the source set is not ASCII.  */
2153 	      bool one_2_one_ascii
2154 		= (target_to_host_charmap[0] == 1 && target_to_host ('a') == 97);
2155 
2156 	      /* A wide character in the ASCII range most likely results
2157 		 in a single byte, and only unlikely in up to MB_LEN_MAX.  */
2158 	      res.range.max = one_2_one_ascii ? 1 : target_mb_len_max ();;
2159 	      res.range.likely = 1;
2160 	      res.range.unlikely = target_mb_len_max ();
2161 	      res.mayfail = !one_2_one_ascii;
2162 	    }
2163 	  else
2164 	    {
2165 	      /* A wide character outside the ASCII range likely results
2166 		 in up to two bytes, and only unlikely in up to MB_LEN_MAX.  */
2167 	      res.range.max = target_mb_len_max ();
2168 	      res.range.likely = 2;
2169 	      res.range.unlikely = res.range.max;
2170 	      /* Converting such a character may fail.  */
2171 	      res.mayfail = true;
2172 	    }
2173 	}
2174       else
2175 	{
2176 	  /* An unknown wide character is treated the same as a wide
2177 	     character outside the ASCII range.  */
2178 	  res.range.max = target_mb_len_max ();
2179 	  res.range.likely = 2;
2180 	  res.range.unlikely = res.range.max;
2181 	  res.mayfail = true;
2182 	}
2183     }
2184   else
2185     {
2186       /* A plain '%c' directive.  Its ouput is exactly 1.  */
2187       res.range.min = res.range.max = 1;
2188       res.range.likely = res.range.unlikely = 1;
2189       res.knownrange = true;
2190     }
2191 
2192   /* Bump up the byte counters if WIDTH is greater.  */
2193   return res.adjust_for_width_or_precision (dir.width);
2194 }
2195 
2196 /* Determine the offset *INDEX of the first byte of an array element of
2197    TYPE (possibly recursively) into which the byte offset OFF points.
2198    On success set *INDEX to the offset of the first byte and return type.
2199    Otherwise, if no such element can be found, return null.  */
2200 
2201 static tree
array_elt_at_offset(tree type,HOST_WIDE_INT off,HOST_WIDE_INT * index)2202 array_elt_at_offset (tree type, HOST_WIDE_INT off, HOST_WIDE_INT *index)
2203 {
2204   gcc_assert (TREE_CODE (type) == ARRAY_TYPE);
2205 
2206   tree eltype = type;
2207   while (TREE_CODE (TREE_TYPE (eltype)) == ARRAY_TYPE)
2208     eltype = TREE_TYPE (eltype);
2209 
2210   if (TYPE_MODE (TREE_TYPE (eltype)) != TYPE_MODE (char_type_node))
2211     eltype = TREE_TYPE (eltype);
2212 
2213   if (eltype == type)
2214     {
2215       *index = 0;
2216       return type;
2217     }
2218 
2219   HOST_WIDE_INT typsz = int_size_in_bytes (type);
2220   HOST_WIDE_INT eltsz = int_size_in_bytes (eltype);
2221   if (off < typsz * eltsz)
2222     {
2223       *index = (off / eltsz) * eltsz;
2224       return TREE_CODE (eltype) == ARRAY_TYPE ? TREE_TYPE (eltype) : eltype;
2225     }
2226 
2227   return NULL_TREE;
2228 }
2229 
2230 /* Determine the offset *INDEX of the first byte of a struct member of TYPE
2231    (possibly recursively) into which the byte offset OFF points.  On success
2232    set *INDEX to the offset of the first byte and return true.  Otherwise,
2233    if no such member can be found, return false.  */
2234 
2235 static bool
field_at_offset(tree type,HOST_WIDE_INT off,HOST_WIDE_INT * index)2236 field_at_offset (tree type, HOST_WIDE_INT off, HOST_WIDE_INT *index)
2237 {
2238   gcc_assert (RECORD_OR_UNION_TYPE_P (type));
2239 
2240   for (tree fld = TYPE_FIELDS (type); fld; fld = TREE_CHAIN (fld))
2241     {
2242       if (TREE_CODE (fld) != FIELD_DECL || DECL_ARTIFICIAL (fld))
2243 	continue;
2244 
2245       tree fldtype = TREE_TYPE (fld);
2246       HOST_WIDE_INT fldoff = int_byte_position (fld);
2247 
2248       /* If the size is not available the field is a flexible array
2249 	 member.  Treat this case as success.  */
2250       tree typesize = TYPE_SIZE_UNIT (fldtype);
2251       HOST_WIDE_INT fldsize = (tree_fits_uhwi_p (typesize)
2252 			       ? tree_to_uhwi (typesize)
2253 			       : off);
2254 
2255       if (fldoff + fldsize < off)
2256 	continue;
2257 
2258       if (TREE_CODE (fldtype) == ARRAY_TYPE)
2259 	{
2260 	  HOST_WIDE_INT idx = 0;
2261 	  if (tree ft = array_elt_at_offset (fldtype, off, &idx))
2262 	    fldtype = ft;
2263 	  else
2264 	    break;
2265 
2266 	  *index += idx;
2267 	  fldoff -= idx;
2268 	  off -= idx;
2269 	}
2270 
2271       if (RECORD_OR_UNION_TYPE_P (fldtype))
2272 	{
2273 	  *index += fldoff;
2274 	  return field_at_offset (fldtype, off - fldoff, index);
2275 	}
2276 
2277       *index += fldoff;
2278       return true;
2279     }
2280 
2281   return false;
2282 }
2283 
2284 /* For an expression X of pointer type, recursively try to find the same
2285    origin (object or pointer) as Y it references and return such an X.
2286    When X refers to a struct member, set *FLDOFF to the offset of the
2287    member from the beginning of the "most derived" object.  */
2288 
2289 static tree
get_origin_and_offset(tree x,HOST_WIDE_INT * fldoff,HOST_WIDE_INT * off)2290 get_origin_and_offset (tree x, HOST_WIDE_INT *fldoff, HOST_WIDE_INT *off)
2291 {
2292   if (!x)
2293     return NULL_TREE;
2294 
2295   switch (TREE_CODE (x))
2296     {
2297     case ADDR_EXPR:
2298       x = TREE_OPERAND (x, 0);
2299       return get_origin_and_offset (x, fldoff, off);
2300 
2301     case ARRAY_REF:
2302       {
2303 	tree offset = TREE_OPERAND (x, 1);
2304 	HOST_WIDE_INT idx = (tree_fits_uhwi_p (offset)
2305 			     ? tree_to_uhwi (offset) : HOST_WIDE_INT_MAX);
2306 
2307 	tree eltype = TREE_TYPE (x);
2308 	if (TREE_CODE (eltype) == INTEGER_TYPE)
2309 	  {
2310 	    if (off)
2311 	      *off = idx;
2312 	  }
2313 	else if (idx < HOST_WIDE_INT_MAX)
2314 	  *fldoff += idx * int_size_in_bytes (eltype);
2315 	else
2316 	  *fldoff = idx;
2317 
2318 	x = TREE_OPERAND (x, 0);
2319 	return get_origin_and_offset (x, fldoff, NULL);
2320       }
2321 
2322     case MEM_REF:
2323       if (off)
2324 	{
2325 	  tree offset = TREE_OPERAND (x, 1);
2326 	  *off = (tree_fits_uhwi_p (offset)
2327 		  ? tree_to_uhwi (offset) : HOST_WIDE_INT_MAX);
2328 	}
2329 
2330       x = TREE_OPERAND (x, 0);
2331 
2332       if (off)
2333 	{
2334 	  tree xtype
2335 	    = (TREE_CODE (x) == ADDR_EXPR
2336 	       ? TREE_TYPE (TREE_OPERAND (x, 0)) : TREE_TYPE (TREE_TYPE (x)));
2337 
2338 	  /* The byte offset of the most basic struct member the byte
2339 	     offset *OFF corresponds to, or for a (multidimensional)
2340 	     array member, the byte offset of the array element.  */
2341 	  HOST_WIDE_INT index = 0;
2342 
2343 	  if ((RECORD_OR_UNION_TYPE_P (xtype)
2344 	       && field_at_offset (xtype, *off, &index))
2345 	      || (TREE_CODE (xtype) == ARRAY_TYPE
2346 		  && TREE_CODE (TREE_TYPE (xtype)) == ARRAY_TYPE
2347 		  && array_elt_at_offset (xtype, *off, &index)))
2348 	    {
2349 	      *fldoff += index;
2350 	      *off -= index;
2351 	    }
2352 	}
2353 
2354       return get_origin_and_offset (x, fldoff, NULL);
2355 
2356     case COMPONENT_REF:
2357       {
2358 	tree fld = TREE_OPERAND (x, 1);
2359 	*fldoff += int_byte_position (fld);
2360 
2361 	get_origin_and_offset (fld, fldoff, off);
2362 	x = TREE_OPERAND (x, 0);
2363 	return get_origin_and_offset (x, fldoff, off);
2364       }
2365 
2366     case SSA_NAME:
2367       {
2368 	gimple *def = SSA_NAME_DEF_STMT (x);
2369 	if (is_gimple_assign (def))
2370 	  {
2371 	    tree_code code = gimple_assign_rhs_code (def);
2372 	    if (code == ADDR_EXPR)
2373 	      {
2374 		x = gimple_assign_rhs1 (def);
2375 		return get_origin_and_offset (x, fldoff, off);
2376 	      }
2377 
2378 	    if (code == POINTER_PLUS_EXPR)
2379 	      {
2380 		tree offset = gimple_assign_rhs2 (def);
2381 		if (off)
2382 		  *off = (tree_fits_uhwi_p (offset)
2383 			  ? tree_to_uhwi (offset) : HOST_WIDE_INT_MAX);
2384 
2385 		x = gimple_assign_rhs1 (def);
2386 		return get_origin_and_offset (x, fldoff, NULL);
2387 	      }
2388 	    else if (code == VAR_DECL)
2389 	      {
2390 		x = gimple_assign_rhs1 (def);
2391 		return get_origin_and_offset (x, fldoff, off);
2392 	      }
2393 	  }
2394 	else if (gimple_nop_p (def) && SSA_NAME_VAR (x))
2395 	  x = SSA_NAME_VAR (x);
2396       }
2397 
2398     default:
2399       break;
2400     }
2401 
2402   return x;
2403 }
2404 
2405 /* If ARG refers to the same (sub)object or array element as described
2406    by DST and DST_FLD, return the byte offset into the struct member or
2407    array element referenced by ARG.  Otherwise return HOST_WIDE_INT_MIN
2408    to indicate that ARG and DST do not refer to the same object.  */
2409 
2410 static HOST_WIDE_INT
alias_offset(tree arg,tree dst,HOST_WIDE_INT dst_fld)2411 alias_offset (tree arg, tree dst, HOST_WIDE_INT dst_fld)
2412 {
2413   /* See if the argument refers to the same base object as the destination
2414      of the formatted function call, and if so, try to determine if they
2415      can alias.  */
2416   if (!arg || !dst || !ptr_derefs_may_alias_p (arg, dst))
2417     return HOST_WIDE_INT_MIN;
2418 
2419   /* The two arguments may refer to the same object.  If they both refer
2420      to a struct member, see if the members are one and the same.  */
2421   HOST_WIDE_INT arg_off = 0, arg_fld = 0;
2422 
2423   tree arg_orig = get_origin_and_offset (arg, &arg_fld, &arg_off);
2424 
2425   if (arg_orig == dst && arg_fld == dst_fld)
2426     return arg_off;
2427 
2428   return HOST_WIDE_INT_MIN;
2429 }
2430 
2431 /* Return the minimum and maximum number of characters formatted
2432    by the '%s' format directive and its wide character form for
2433    the argument ARG.  ARG can be null (for functions such as
2434    vsprinf).  */
2435 
2436 static fmtresult
format_string(const directive & dir,tree arg,const vr_values * vr_values)2437 format_string (const directive &dir, tree arg, const vr_values *vr_values)
2438 {
2439   fmtresult res;
2440 
2441   if (warn_restrict)
2442     {
2443       /* See if ARG might alias the destination of the call with
2444 	 DST_ORIGIN and DST_FIELD.  If so, store the starting offset
2445 	 so that the overlap can be determined for certain later,
2446 	 when the amount of output of the call (including subsequent
2447 	 directives) has been computed.  Otherwise, store HWI_MIN.  */
2448       res.dst_offset = alias_offset (arg, dir.info->dst_origin,
2449 				     dir.info->dst_field);
2450     }
2451 
2452   /* Compute the range the argument's length can be in.  */
2453   int count_by = 1;
2454   if (dir.specifier == 'S' || dir.modifier == FMT_LEN_l)
2455     {
2456       /* Get a node for a C type that will be the same size
2457 	 as a wchar_t on the target.  */
2458       tree node = get_typenode_from_name (MODIFIED_WCHAR_TYPE);
2459 
2460       /* Now that we have a suitable node, get the number of
2461 	 bytes it occupies.  */
2462       count_by = int_size_in_bytes (node);
2463       gcc_checking_assert (count_by == 2 || count_by == 4);
2464     }
2465 
2466   fmtresult slen = get_string_length (arg, count_by, vr_values);
2467   if (slen.range.min == slen.range.max
2468       && slen.range.min < HOST_WIDE_INT_MAX)
2469     {
2470       /* The argument is either a string constant or it refers
2471 	 to one of a number of strings of the same length.  */
2472 
2473       /* A '%s' directive with a string argument with constant length.  */
2474       res.range = slen.range;
2475 
2476       if (dir.specifier == 'S'
2477 	  || dir.modifier == FMT_LEN_l)
2478 	{
2479 	  /* In the worst case the length of output of a wide string S
2480 	     is bounded by MB_LEN_MAX * wcslen (S).  */
2481 	  res.range.max *= target_mb_len_max ();
2482 	  res.range.unlikely = res.range.max;
2483 	  /* It's likely that the total length is not more that
2484 	     2 * wcslen (S).*/
2485 	  res.range.likely = res.range.min * 2;
2486 
2487 	  if (dir.prec[1] >= 0
2488 	      && (unsigned HOST_WIDE_INT)dir.prec[1] < res.range.max)
2489 	    {
2490 	      res.range.max = dir.prec[1];
2491 	      res.range.likely = dir.prec[1];
2492 	      res.range.unlikely = dir.prec[1];
2493 	    }
2494 
2495 	  if (dir.prec[0] < 0 && dir.prec[1] > -1)
2496 	    res.range.min = 0;
2497 	  else if (dir.prec[0] >= 0)
2498 	    res.range.likely = dir.prec[0];
2499 
2500 	  /* Even a non-empty wide character string need not convert into
2501 	     any bytes.  */
2502 	  res.range.min = 0;
2503 
2504 	  /* A non-empty wide character conversion may fail.  */
2505 	  if (slen.range.max > 0)
2506 	    res.mayfail = true;
2507 	}
2508       else
2509 	{
2510 	  res.knownrange = true;
2511 
2512 	  if (dir.prec[0] < 0 && dir.prec[1] > -1)
2513 	    res.range.min = 0;
2514 	  else if ((unsigned HOST_WIDE_INT)dir.prec[0] < res.range.min)
2515 	    res.range.min = dir.prec[0];
2516 
2517 	  if ((unsigned HOST_WIDE_INT)dir.prec[1] < res.range.max)
2518 	    {
2519 	      res.range.max = dir.prec[1];
2520 	      res.range.likely = dir.prec[1];
2521 	      res.range.unlikely = dir.prec[1];
2522 	    }
2523 	}
2524     }
2525   else if (arg && integer_zerop (arg))
2526     {
2527       /* Handle null pointer argument.  */
2528 
2529       fmtresult res (0);
2530       res.nullp = true;
2531       return res;
2532     }
2533   else
2534     {
2535       /* For a '%s' and '%ls' directive with a non-constant string (either
2536 	 one of a number of strings of known length or an unknown string)
2537 	 the minimum number of characters is lesser of PRECISION[0] and
2538 	 the length of the shortest known string or zero, and the maximum
2539 	 is the lessser of the length of the longest known string or
2540 	 PTRDIFF_MAX and PRECISION[1].  The likely length is either
2541 	 the minimum at level 1 and the greater of the minimum and 1
2542 	 at level 2.  This result is adjust upward for width (if it's
2543 	 specified).  */
2544 
2545       if (dir.specifier == 'S'
2546 	  || dir.modifier == FMT_LEN_l)
2547 	{
2548 	  /* A wide character converts to as few as zero bytes.  */
2549 	  slen.range.min = 0;
2550 	  if (slen.range.max < target_int_max ())
2551 	    slen.range.max *= target_mb_len_max ();
2552 
2553 	  if (slen.range.likely < target_int_max ())
2554 	    slen.range.likely *= 2;
2555 
2556 	  if (slen.range.likely < target_int_max ())
2557 	    slen.range.unlikely *= target_mb_len_max ();
2558 
2559 	  /* A non-empty wide character conversion may fail.  */
2560 	  if (slen.range.max > 0)
2561 	    res.mayfail = true;
2562 	}
2563 
2564       res.range = slen.range;
2565 
2566       if (dir.prec[0] >= 0)
2567 	{
2568 	  /* Adjust the minimum to zero if the string length is unknown,
2569 	     or at most the lower bound of the precision otherwise.  */
2570 	  if (slen.range.min >= target_int_max ())
2571 	    res.range.min = 0;
2572 	  else if ((unsigned HOST_WIDE_INT)dir.prec[0] < slen.range.min)
2573 	    res.range.min = dir.prec[0];
2574 
2575 	  /* Make both maxima no greater than the upper bound of precision.  */
2576 	  if ((unsigned HOST_WIDE_INT)dir.prec[1] < slen.range.max
2577 	      || slen.range.max >= target_int_max ())
2578 	    {
2579 	      res.range.max = dir.prec[1];
2580 	      res.range.unlikely = dir.prec[1];
2581 	    }
2582 
2583 	  /* If precision is constant, set the likely counter to the lesser
2584 	     of it and the maximum string length.  Otherwise, if the lower
2585 	     bound of precision is greater than zero, set the likely counter
2586 	     to the minimum.  Otherwise set it to zero or one based on
2587 	     the warning level.  */
2588 	  if (dir.prec[0] == dir.prec[1])
2589 	    res.range.likely
2590 	      = ((unsigned HOST_WIDE_INT)dir.prec[0] < slen.range.max
2591 		 ? dir.prec[0] : slen.range.max);
2592 	  else if (dir.prec[0] > 0)
2593 	    res.range.likely = res.range.min;
2594 	  else
2595 	    res.range.likely = warn_level > 1;
2596 	}
2597       else if (dir.prec[1] >= 0)
2598 	{
2599 	  res.range.min = 0;
2600 	  if ((unsigned HOST_WIDE_INT)dir.prec[1] < slen.range.max)
2601 	    res.range.max = dir.prec[1];
2602 	  res.range.likely = dir.prec[1] ? warn_level > 1 : 0;
2603 	  if ((unsigned HOST_WIDE_INT)dir.prec[1] < slen.range.unlikely)
2604 	    res.range.unlikely = dir.prec[1];
2605 	}
2606       else if (slen.range.min >= target_int_max ())
2607 	{
2608 	  res.range.min = 0;
2609 	  res.range.max = HOST_WIDE_INT_MAX;
2610 	  /* At level 1 strings of unknown length are assumed to be
2611 	     empty, while at level 1 they are assumed to be one byte
2612 	     long.  */
2613 	  res.range.likely = warn_level > 1;
2614 	  res.range.unlikely = HOST_WIDE_INT_MAX;
2615 	}
2616       else
2617 	{
2618 	  /* A string of unknown length unconstrained by precision is
2619 	     assumed to be empty at level 1 and just one character long
2620 	     at higher levels.  */
2621 	  if (res.range.likely >= target_int_max ())
2622 	    res.range.likely = warn_level > 1;
2623 	}
2624     }
2625 
2626   /* If the argument isn't a nul-terminated string and the number
2627      of bytes on output isn't bounded by precision, set NONSTR.  */
2628   if (slen.nonstr && slen.range.min < (unsigned HOST_WIDE_INT)dir.prec[0])
2629     res.nonstr = slen.nonstr;
2630 
2631   /* Bump up the byte counters if WIDTH is greater.  */
2632   return res.adjust_for_width_or_precision (dir.width);
2633 }
2634 
2635 /* Format plain string (part of the format string itself).  */
2636 
2637 static fmtresult
format_plain(const directive & dir,tree,const vr_values *)2638 format_plain (const directive &dir, tree, const vr_values *)
2639 {
2640   fmtresult res (dir.len);
2641   return res;
2642 }
2643 
2644 /* Return true if the RESULT of a directive in a call describe by INFO
2645    should be diagnosed given the AVAILable space in the destination.  */
2646 
2647 static bool
should_warn_p(const call_info & info,const result_range & avail,const result_range & result)2648 should_warn_p (const call_info &info,
2649 	       const result_range &avail, const result_range &result)
2650 {
2651   if (result.max <= avail.min)
2652     {
2653       /* The least amount of space remaining in the destination is big
2654 	 enough for the longest output.  */
2655       return false;
2656     }
2657 
2658   if (info.bounded)
2659     {
2660       if (warn_format_trunc == 1 && result.min <= avail.max
2661 	  && info.retval_used ())
2662 	{
2663 	  /* The likely amount of space remaining in the destination is big
2664 	     enough for the least output and the return value is used.  */
2665 	  return false;
2666 	}
2667 
2668       if (warn_format_trunc == 1 && result.likely <= avail.likely
2669 	  && !info.retval_used ())
2670 	{
2671 	  /* The likely amount of space remaining in the destination is big
2672 	     enough for the likely output and the return value is unused.  */
2673 	  return false;
2674 	}
2675 
2676       if (warn_format_trunc == 2
2677 	  && result.likely <= avail.min
2678 	  && (result.max <= avail.min
2679 	      || result.max > HOST_WIDE_INT_MAX))
2680 	{
2681 	  /* The minimum amount of space remaining in the destination is big
2682 	     enough for the longest output.  */
2683 	  return false;
2684 	}
2685     }
2686   else
2687     {
2688       if (warn_level == 1 && result.likely <= avail.likely)
2689 	{
2690 	  /* The likely amount of space remaining in the destination is big
2691 	     enough for the likely output.  */
2692 	  return false;
2693 	}
2694 
2695       if (warn_level == 2
2696 	  && result.likely <= avail.min
2697 	  && (result.max <= avail.min
2698 	      || result.max > HOST_WIDE_INT_MAX))
2699 	{
2700 	  /* The minimum amount of space remaining in the destination is big
2701 	     enough for the longest output.  */
2702 	  return false;
2703 	}
2704     }
2705 
2706   return true;
2707 }
2708 
2709 /* At format string location describe by DIRLOC in a call described
2710    by INFO, issue a warning for a directive DIR whose output may be
2711    in excess of the available space AVAIL_RANGE in the destination
2712    given the formatting result FMTRES.  This function does nothing
2713    except decide whether to issue a warning for a possible write
2714    past the end or truncation and, if so, format the warning.
2715    Return true if a warning has been issued.  */
2716 
2717 static bool
maybe_warn(substring_loc & dirloc,location_t argloc,const call_info & info,const result_range & avail_range,const result_range & res,const directive & dir)2718 maybe_warn (substring_loc &dirloc, location_t argloc,
2719 	    const call_info &info,
2720 	    const result_range &avail_range, const result_range &res,
2721 	    const directive &dir)
2722 {
2723   if (!should_warn_p (info, avail_range, res))
2724     return false;
2725 
2726   /* A warning will definitely be issued below.  */
2727 
2728   /* The maximum byte count to reference in the warning.  Larger counts
2729      imply that the upper bound is unknown (and could be anywhere between
2730      RES.MIN + 1 and SIZE_MAX / 2) are printed as "N or more bytes" rather
2731      than "between N and X" where X is some huge number.  */
2732   unsigned HOST_WIDE_INT maxbytes = target_dir_max ();
2733 
2734   /* True when there is enough room in the destination for the least
2735      amount of a directive's output but not enough for its likely or
2736      maximum output.  */
2737   bool maybe = (res.min <= avail_range.max
2738 		&& (avail_range.min < res.likely
2739 		    || (res.max < HOST_WIDE_INT_MAX
2740 			&& avail_range.min < res.max)));
2741 
2742   /* Buffer for the directive in the host character set (used when
2743      the source character set is different).  */
2744   char hostdir[32];
2745 
2746   if (avail_range.min == avail_range.max)
2747     {
2748       /* The size of the destination region is exact.  */
2749       unsigned HOST_WIDE_INT navail = avail_range.max;
2750 
2751       if (target_to_host (*dir.beg) != '%')
2752 	{
2753 	  /* For plain character directives (i.e., the format string itself)
2754 	     but not others, point the caret at the first character that's
2755 	     past the end of the destination.  */
2756 	  if (navail < dir.len)
2757 	    dirloc.set_caret_index (dirloc.get_caret_idx () + navail);
2758 	}
2759 
2760       if (*dir.beg == '\0')
2761 	{
2762 	  /* This is the terminating nul.  */
2763 	  gcc_assert (res.min == 1 && res.min == res.max);
2764 
2765 	  return fmtwarn (dirloc, UNKNOWN_LOCATION, NULL, info.warnopt (),
2766 			  info.bounded
2767 			  ? (maybe
2768 			     ? G_("%qE output may be truncated before the "
2769 				  "last format character")
2770 			     : G_("%qE output truncated before the last "
2771 				  "format character"))
2772 			  : (maybe
2773 			     ? G_("%qE may write a terminating nul past the "
2774 				  "end of the destination")
2775 			     : G_("%qE writing a terminating nul past the "
2776 				  "end of the destination")),
2777 			  info.func);
2778 	}
2779 
2780       if (res.min == res.max)
2781 	{
2782 	  const char *d = target_to_host (hostdir, sizeof hostdir, dir.beg);
2783 	  if (!info.bounded)
2784 	    return fmtwarn_n (dirloc, argloc, NULL, info.warnopt (), res.min,
2785 			      "%<%.*s%> directive writing %wu byte into a "
2786 			      "region of size %wu",
2787 			      "%<%.*s%> directive writing %wu bytes into a "
2788 			      "region of size %wu",
2789 			      (int) dir.len, d, res.min, navail);
2790 	  else if (maybe)
2791 	    return fmtwarn_n (dirloc, argloc, NULL, info.warnopt (), res.min,
2792 			      "%<%.*s%> directive output may be truncated "
2793 			      "writing %wu byte into a region of size %wu",
2794 			      "%<%.*s%> directive output may be truncated "
2795 			      "writing %wu bytes into a region of size %wu",
2796 			      (int) dir.len, d, res.min, navail);
2797 	  else
2798 	    return fmtwarn_n (dirloc, argloc, NULL, info.warnopt (), res.min,
2799 			      "%<%.*s%> directive output truncated writing "
2800 			      "%wu byte into a region of size %wu",
2801 			      "%<%.*s%> directive output truncated writing "
2802 			      "%wu bytes into a region of size %wu",
2803 			      (int) dir.len, d, res.min, navail);
2804 	}
2805       if (res.min == 0 && res.max < maxbytes)
2806 	return fmtwarn (dirloc, argloc, NULL,
2807 			info.warnopt (),
2808 			info.bounded
2809 			? (maybe
2810 			   ? G_("%<%.*s%> directive output may be truncated "
2811 				"writing up to %wu bytes into a region of "
2812 				"size %wu")
2813 			   : G_("%<%.*s%> directive output truncated writing "
2814 				"up to %wu bytes into a region of size %wu"))
2815 			: G_("%<%.*s%> directive writing up to %wu bytes "
2816 			     "into a region of size %wu"), (int) dir.len,
2817 			target_to_host (hostdir, sizeof hostdir, dir.beg),
2818 			res.max, navail);
2819 
2820       if (res.min == 0 && maxbytes <= res.max)
2821 	/* This is a special case to avoid issuing the potentially
2822 	   confusing warning:
2823 	     writing 0 or more bytes into a region of size 0.  */
2824 	return fmtwarn (dirloc, argloc, NULL, info.warnopt (),
2825 			info.bounded
2826 			? (maybe
2827 			   ? G_("%<%.*s%> directive output may be truncated "
2828 				"writing likely %wu or more bytes into a "
2829 				"region of size %wu")
2830 			   : G_("%<%.*s%> directive output truncated writing "
2831 				"likely %wu or more bytes into a region of "
2832 				"size %wu"))
2833 			: G_("%<%.*s%> directive writing likely %wu or more "
2834 			     "bytes into a region of size %wu"), (int) dir.len,
2835 			target_to_host (hostdir, sizeof hostdir, dir.beg),
2836 			res.likely, navail);
2837 
2838       if (res.max < maxbytes)
2839 	return fmtwarn (dirloc, argloc, NULL, info.warnopt (),
2840 			info.bounded
2841 			? (maybe
2842 			   ? G_("%<%.*s%> directive output may be truncated "
2843 				"writing between %wu and %wu bytes into a "
2844 				"region of size %wu")
2845 			   : G_("%<%.*s%> directive output truncated "
2846 				"writing between %wu and %wu bytes into a "
2847 				"region of size %wu"))
2848 			: G_("%<%.*s%> directive writing between %wu and "
2849 			     "%wu bytes into a region of size %wu"),
2850 			(int) dir.len,
2851 			target_to_host (hostdir, sizeof hostdir, dir.beg),
2852 			res.min, res.max, navail);
2853 
2854       return fmtwarn (dirloc, argloc, NULL, info.warnopt (),
2855 		      info.bounded
2856 		      ? (maybe
2857 			 ? G_("%<%.*s%> directive output may be truncated "
2858 			      "writing %wu or more bytes into a region of "
2859 			      "size %wu")
2860 			 : G_("%<%.*s%> directive output truncated writing "
2861 			      "%wu or more bytes into a region of size %wu"))
2862 		      : G_("%<%.*s%> directive writing %wu or more bytes "
2863 			   "into a region of size %wu"), (int) dir.len,
2864 		      target_to_host (hostdir, sizeof hostdir, dir.beg),
2865 		      res.min, navail);
2866     }
2867 
2868   /* The size of the destination region is a range.  */
2869 
2870   if (target_to_host (*dir.beg) != '%')
2871     {
2872       unsigned HOST_WIDE_INT navail = avail_range.max;
2873 
2874       /* For plain character directives (i.e., the format string itself)
2875 	 but not others, point the caret at the first character that's
2876 	 past the end of the destination.  */
2877       if (navail < dir.len)
2878 	dirloc.set_caret_index (dirloc.get_caret_idx () + navail);
2879     }
2880 
2881   if (*dir.beg == '\0')
2882     {
2883       gcc_assert (res.min == 1 && res.min == res.max);
2884 
2885       return fmtwarn (dirloc, UNKNOWN_LOCATION, NULL, info.warnopt (),
2886 		      info.bounded
2887 		      ? (maybe
2888 			 ? G_("%qE output may be truncated before the last "
2889 			      "format character")
2890 			 : G_("%qE output truncated before the last format "
2891 			      "character"))
2892 		      : (maybe
2893 			 ? G_("%qE may write a terminating nul past the end "
2894 			      "of the destination")
2895 			 : G_("%qE writing a terminating nul past the end "
2896 			      "of the destination")), info.func);
2897     }
2898 
2899   if (res.min == res.max)
2900     {
2901       const char *d = target_to_host (hostdir, sizeof hostdir, dir.beg);
2902       if (!info.bounded)
2903 	return fmtwarn_n (dirloc, argloc, NULL, info.warnopt (), res.min,
2904 			  "%<%.*s%> directive writing %wu byte into a region "
2905 			  "of size between %wu and %wu",
2906 			  "%<%.*s%> directive writing %wu bytes into a region "
2907 			  "of size between %wu and %wu", (int) dir.len, d,
2908 			  res.min, avail_range.min, avail_range.max);
2909       else if (maybe)
2910 	return fmtwarn_n (dirloc, argloc, NULL, info.warnopt (), res.min,
2911 			  "%<%.*s%> directive output may be truncated writing "
2912 			  "%wu byte into a region of size between %wu and %wu",
2913 			  "%<%.*s%> directive output may be truncated writing "
2914 			  "%wu bytes into a region of size between %wu and "
2915 			  "%wu", (int) dir.len, d, res.min, avail_range.min,
2916 			  avail_range.max);
2917       else
2918 	return fmtwarn_n (dirloc, argloc, NULL, info.warnopt (), res.min,
2919 			  "%<%.*s%> directive output truncated writing %wu "
2920 			  "byte into a region of size between %wu and %wu",
2921 			  "%<%.*s%> directive output truncated writing %wu "
2922 			  "bytes into a region of size between %wu and %wu",
2923 			  (int) dir.len, d, res.min, avail_range.min,
2924 			  avail_range.max);
2925     }
2926 
2927   if (res.min == 0 && res.max < maxbytes)
2928     return fmtwarn (dirloc, argloc, NULL, info.warnopt (),
2929 		    info.bounded
2930 		    ? (maybe
2931 		       ? G_("%<%.*s%> directive output may be truncated "
2932 			    "writing up to %wu bytes into a region of size "
2933 			    "between %wu and %wu")
2934 		       : G_("%<%.*s%> directive output truncated writing "
2935 			    "up to %wu bytes into a region of size between "
2936 			    "%wu and %wu"))
2937 		    : G_("%<%.*s%> directive writing up to %wu bytes "
2938 			 "into a region of size between %wu and %wu"),
2939 		    (int) dir.len,
2940 		    target_to_host (hostdir, sizeof hostdir, dir.beg),
2941 		    res.max, avail_range.min, avail_range.max);
2942 
2943   if (res.min == 0 && maxbytes <= res.max)
2944     /* This is a special case to avoid issuing the potentially confusing
2945        warning:
2946 	 writing 0 or more bytes into a region of size between 0 and N.  */
2947     return fmtwarn (dirloc, argloc, NULL, info.warnopt (),
2948 		    info.bounded
2949 		    ? (maybe
2950 		       ? G_("%<%.*s%> directive output may be truncated "
2951 			    "writing likely %wu or more bytes into a region "
2952 			    "of size between %wu and %wu")
2953 		       : G_("%<%.*s%> directive output truncated writing "
2954 			    "likely %wu or more bytes into a region of size "
2955 			    "between %wu and %wu"))
2956 		    : G_("%<%.*s%> directive writing likely %wu or more bytes "
2957 			 "into a region of size between %wu and %wu"),
2958 		    (int) dir.len,
2959 		    target_to_host (hostdir, sizeof hostdir, dir.beg),
2960 		    res.likely, avail_range.min, avail_range.max);
2961 
2962   if (res.max < maxbytes)
2963     return fmtwarn (dirloc, argloc, NULL, info.warnopt (),
2964 		    info.bounded
2965 		    ? (maybe
2966 		       ? G_("%<%.*s%> directive output may be truncated "
2967 			    "writing between %wu and %wu bytes into a region "
2968 			    "of size between %wu and %wu")
2969 		       : G_("%<%.*s%> directive output truncated writing "
2970 			    "between %wu and %wu bytes into a region of size "
2971 			    "between %wu and %wu"))
2972 		    : G_("%<%.*s%> directive writing between %wu and "
2973 			 "%wu bytes into a region of size between %wu and "
2974 			 "%wu"), (int) dir.len,
2975 		    target_to_host (hostdir, sizeof hostdir, dir.beg),
2976 		    res.min, res.max, avail_range.min, avail_range.max);
2977 
2978   return fmtwarn (dirloc, argloc, NULL, info.warnopt (),
2979 		  info.bounded
2980 		  ? (maybe
2981 		     ? G_("%<%.*s%> directive output may be truncated writing "
2982 			  "%wu or more bytes into a region of size between "
2983 			  "%wu and %wu")
2984 		     : G_("%<%.*s%> directive output truncated writing "
2985 			  "%wu or more bytes into a region of size between "
2986 			  "%wu and %wu"))
2987 		  : G_("%<%.*s%> directive writing %wu or more bytes "
2988 		       "into a region of size between %wu and %wu"),
2989 		  (int) dir.len,
2990 		  target_to_host (hostdir, sizeof hostdir, dir.beg),
2991 		  res.min, avail_range.min, avail_range.max);
2992 }
2993 
2994 /* Given the formatting result described by RES and NAVAIL, the number
2995    of available in the destination, return the range of bytes remaining
2996    in the destination.  */
2997 
2998 static inline result_range
bytes_remaining(unsigned HOST_WIDE_INT navail,const format_result & res)2999 bytes_remaining (unsigned HOST_WIDE_INT navail, const format_result &res)
3000 {
3001   result_range range;
3002 
3003   if (HOST_WIDE_INT_MAX <= navail)
3004     {
3005       range.min = range.max = range.likely = range.unlikely = navail;
3006       return range;
3007     }
3008 
3009   /* The lower bound of the available range is the available size
3010      minus the maximum output size, and the upper bound is the size
3011      minus the minimum.  */
3012   range.max = res.range.min < navail ? navail - res.range.min : 0;
3013 
3014   range.likely = res.range.likely < navail ? navail - res.range.likely : 0;
3015 
3016   if (res.range.max < HOST_WIDE_INT_MAX)
3017     range.min = res.range.max < navail ? navail - res.range.max : 0;
3018   else
3019     range.min = range.likely;
3020 
3021   range.unlikely = (res.range.unlikely < navail
3022 		    ? navail - res.range.unlikely : 0);
3023 
3024   return range;
3025 }
3026 
3027 /* Compute the length of the output resulting from the directive DIR
3028    in a call described by INFO and update the overall result of the call
3029    in *RES.  Return true if the directive has been handled.  */
3030 
3031 static bool
format_directive(const call_info & info,format_result * res,const directive & dir,const class vr_values * vr_values)3032 format_directive (const call_info &info,
3033 		  format_result *res, const directive &dir,
3034 		  const class vr_values *vr_values)
3035 {
3036   /* Offset of the beginning of the directive from the beginning
3037      of the format string.  */
3038   size_t offset = dir.beg - info.fmtstr;
3039   size_t start = offset;
3040   size_t length = offset + dir.len - !!dir.len;
3041 
3042   /* Create a location for the whole directive from the % to the format
3043      specifier.  */
3044   substring_loc dirloc (info.fmtloc, TREE_TYPE (info.format),
3045 			offset, start, length);
3046 
3047   /* Also get the location of the argument if possible.
3048      This doesn't work for integer literals or function calls.  */
3049   location_t argloc = UNKNOWN_LOCATION;
3050   if (dir.arg)
3051     argloc = EXPR_LOCATION (dir.arg);
3052 
3053   /* Bail when there is no function to compute the output length,
3054      or when minimum length checking has been disabled.   */
3055   if (!dir.fmtfunc || res->range.min >= HOST_WIDE_INT_MAX)
3056     return false;
3057 
3058   /* Compute the range of lengths of the formatted output.  */
3059   fmtresult fmtres = dir.fmtfunc (dir, dir.arg, vr_values);
3060 
3061   /* Record whether the output of all directives is known to be
3062      bounded by some maximum, implying that their arguments are
3063      either known exactly or determined to be in a known range
3064      or, for strings, limited by the upper bounds of the arrays
3065      they refer to.  */
3066   res->knownrange &= fmtres.knownrange;
3067 
3068   if (!fmtres.knownrange)
3069     {
3070       /* Only when the range is known, check it against the host value
3071 	 of INT_MAX + (the number of bytes of the "%.*Lf" directive with
3072 	 INT_MAX precision, which is the longest possible output of any
3073 	 single directive).  That's the largest valid byte count (though
3074 	 not valid call to a printf-like function because it can never
3075 	 return such a count).  Otherwise, the range doesn't correspond
3076 	 to known values of the argument.  */
3077       if (fmtres.range.max > target_dir_max ())
3078 	{
3079 	  /* Normalize the MAX counter to avoid having to deal with it
3080 	     later.  The counter can be less than HOST_WIDE_INT_M1U
3081 	     when compiling for an ILP32 target on an LP64 host.  */
3082 	  fmtres.range.max = HOST_WIDE_INT_M1U;
3083 	  /* Disable exact and maximum length checking after a failure
3084 	     to determine the maximum number of characters (for example
3085 	     for wide characters or wide character strings) but continue
3086 	     tracking the minimum number of characters.  */
3087 	  res->range.max = HOST_WIDE_INT_M1U;
3088 	}
3089 
3090       if (fmtres.range.min > target_dir_max ())
3091 	{
3092 	  /* Disable exact length checking after a failure to determine
3093 	     even the minimum number of characters (it shouldn't happen
3094 	     except in an error) but keep tracking the minimum and maximum
3095 	     number of characters.  */
3096 	  return true;
3097 	}
3098     }
3099 
3100   /* Buffer for the directive in the host character set (used when
3101      the source character set is different).  */
3102   char hostdir[32];
3103 
3104   int dirlen = dir.len;
3105 
3106   if (fmtres.nullp)
3107     {
3108       fmtwarn (dirloc, argloc, NULL, info.warnopt (),
3109 	       "%G%<%.*s%> directive argument is null",
3110 	       info.callstmt, dirlen,
3111 	       target_to_host (hostdir, sizeof hostdir, dir.beg));
3112 
3113       /* Don't bother processing the rest of the format string.  */
3114       res->warned = true;
3115       res->range.min = HOST_WIDE_INT_M1U;
3116       res->range.max = HOST_WIDE_INT_M1U;
3117       return false;
3118     }
3119 
3120   /* Compute the number of available bytes in the destination.  There
3121      must always be at least one byte of space for the terminating
3122      NUL that's appended after the format string has been processed.  */
3123   result_range avail_range = bytes_remaining (info.objsize, *res);
3124 
3125   /* If the argument aliases a part of the destination of the formatted
3126      call at offset FMTRES.DST_OFFSET append the directive and its result
3127      to the set of aliases for later processing.  */
3128   if (fmtres.dst_offset != HOST_WIDE_INT_MIN)
3129     res->append_alias (dir, fmtres.dst_offset, fmtres.range);
3130 
3131   bool warned = res->warned;
3132 
3133   if (!warned)
3134     warned = maybe_warn (dirloc, argloc, info, avail_range,
3135 			 fmtres.range, dir);
3136 
3137   /* Bump up the total maximum if it isn't too big.  */
3138   if (res->range.max < HOST_WIDE_INT_MAX
3139       && fmtres.range.max < HOST_WIDE_INT_MAX)
3140     res->range.max += fmtres.range.max;
3141 
3142   /* Raise the total unlikely maximum by the larger of the maximum
3143      and the unlikely maximum.  */
3144   unsigned HOST_WIDE_INT save = res->range.unlikely;
3145   if (fmtres.range.max < fmtres.range.unlikely)
3146     res->range.unlikely += fmtres.range.unlikely;
3147   else
3148     res->range.unlikely += fmtres.range.max;
3149 
3150   if (res->range.unlikely < save)
3151     res->range.unlikely = HOST_WIDE_INT_M1U;
3152 
3153   res->range.min += fmtres.range.min;
3154   res->range.likely += fmtres.range.likely;
3155 
3156   /* Has the minimum directive output length exceeded the maximum
3157      of 4095 bytes required to be supported?  */
3158   bool minunder4k = fmtres.range.min < 4096;
3159   bool maxunder4k = fmtres.range.max < 4096;
3160   /* Clear POSUNDER4K in the overall result if the maximum has exceeded
3161      the 4k (this is necessary to avoid the return value optimization
3162      that may not be safe in the maximum case).  */
3163   if (!maxunder4k)
3164     res->posunder4k = false;
3165   /* Also clear POSUNDER4K if the directive may fail.  */
3166   if (fmtres.mayfail)
3167     res->posunder4k = false;
3168 
3169   if (!warned
3170       /* Only warn at level 2.  */
3171       && warn_level > 1
3172       /* Only warn for string functions.  */
3173       && info.is_string_func ()
3174       && (!minunder4k
3175 	  || (!maxunder4k && fmtres.range.max < HOST_WIDE_INT_MAX)))
3176     {
3177       /* The directive output may be longer than the maximum required
3178 	 to be handled by an implementation according to 7.21.6.1, p15
3179 	 of C11.  Warn on this only at level 2 but remember this and
3180 	 prevent folding the return value when done.  This allows for
3181 	 the possibility of the actual libc call failing due to ENOMEM
3182 	 (like Glibc does with very large precision or width).
3183 	 Issue the "may exceed" warning only for string functions and
3184 	 not for fprintf or printf.  */
3185 
3186       if (fmtres.range.min == fmtres.range.max)
3187 	warned = fmtwarn (dirloc, argloc, NULL, info.warnopt (),
3188 			  "%<%.*s%> directive output of %wu bytes exceeds "
3189 			  "minimum required size of 4095", dirlen,
3190 			  target_to_host (hostdir, sizeof hostdir, dir.beg),
3191 			  fmtres.range.min);
3192       else if (!minunder4k)
3193 	warned = fmtwarn (dirloc, argloc, NULL, info.warnopt (),
3194 			  "%<%.*s%> directive output between %wu and %wu "
3195 			  "bytes exceeds minimum required size of 4095",
3196 			  dirlen,
3197 			  target_to_host (hostdir, sizeof hostdir, dir.beg),
3198 			  fmtres.range.min, fmtres.range.max);
3199       else if (!info.retval_used () && info.is_string_func ())
3200 	warned = fmtwarn (dirloc, argloc, NULL, info.warnopt (),
3201 			  "%<%.*s%> directive output between %wu and %wu "
3202 			  "bytes may exceed minimum required size of "
3203 			  "4095",
3204 			  dirlen,
3205 			  target_to_host (hostdir, sizeof hostdir, dir.beg),
3206 			  fmtres.range.min, fmtres.range.max);
3207     }
3208 
3209   /* Has the likely and maximum directive output exceeded INT_MAX?  */
3210   bool likelyximax = *dir.beg && res->range.likely > target_int_max ();
3211   /* Don't consider the maximum to be in excess when it's the result
3212      of a string of unknown length (i.e., whose maximum has been set
3213      to be greater than or equal to HOST_WIDE_INT_MAX.  */
3214   bool maxximax = (*dir.beg
3215 		   && res->range.max > target_int_max ()
3216 		   && res->range.max < HOST_WIDE_INT_MAX);
3217 
3218   if (!warned
3219       /* Warn for the likely output size at level 1.  */
3220       && (likelyximax
3221 	  /* But only warn for the maximum at level 2.  */
3222 	  || (warn_level > 1
3223 	      && maxximax
3224 	      && fmtres.range.max < HOST_WIDE_INT_MAX)))
3225     {
3226       if (fmtres.range.min > target_int_max ())
3227 	{
3228 	  /* The directive output exceeds INT_MAX bytes.  */
3229 	  if (fmtres.range.min == fmtres.range.max)
3230 	    warned = fmtwarn (dirloc, argloc, NULL, info.warnopt (),
3231 			      "%<%.*s%> directive output of %wu bytes exceeds "
3232 			      "%<INT_MAX%>", dirlen,
3233 			      target_to_host (hostdir, sizeof hostdir, dir.beg),
3234 			      fmtres.range.min);
3235 	  else
3236 	    warned = fmtwarn (dirloc, argloc, NULL, info.warnopt (),
3237 			      "%<%.*s%> directive output between %wu and "
3238 			      "%wu bytes exceeds %<INT_MAX%>", dirlen,
3239 			      target_to_host (hostdir, sizeof hostdir, dir.beg),
3240 			      fmtres.range.min, fmtres.range.max);
3241 	}
3242       else if (res->range.min > target_int_max ())
3243 	{
3244 	  /* The directive output is under INT_MAX but causes the result
3245 	     to exceed INT_MAX bytes.  */
3246 	  if (fmtres.range.min == fmtres.range.max)
3247 	    warned = fmtwarn (dirloc, argloc, NULL, info.warnopt (),
3248 			      "%<%.*s%> directive output of %wu bytes causes "
3249 			      "result to exceed %<INT_MAX%>", dirlen,
3250 			      target_to_host (hostdir, sizeof hostdir, dir.beg),
3251 			      fmtres.range.min);
3252 	  else
3253 	    warned = fmtwarn (dirloc, argloc, NULL, info.warnopt (),
3254 			      "%<%.*s%> directive output between %wu and "
3255 			      "%wu bytes causes result to exceed %<INT_MAX%>",
3256 			      dirlen,
3257 			      target_to_host (hostdir, sizeof hostdir, dir.beg),
3258 			      fmtres.range.min, fmtres.range.max);
3259 	}
3260       else if ((!info.retval_used () || !info.bounded)
3261 	       && (info.is_string_func ()))
3262 	/* Warn for calls to string functions that either aren't bounded
3263 	   (sprintf) or whose return value isn't used.  */
3264 	warned = fmtwarn (dirloc, argloc, NULL, info.warnopt (),
3265 			  "%<%.*s%> directive output between %wu and "
3266 			  "%wu bytes may cause result to exceed "
3267 			  "%<INT_MAX%>", dirlen,
3268 			  target_to_host (hostdir, sizeof hostdir, dir.beg),
3269 			  fmtres.range.min, fmtres.range.max);
3270     }
3271 
3272   if (!warned && fmtres.nonstr)
3273     {
3274       warned = fmtwarn (dirloc, argloc, NULL, info.warnopt (),
3275 			"%<%.*s%> directive argument is not a nul-terminated "
3276 			"string",
3277 			dirlen,
3278 			target_to_host (hostdir, sizeof hostdir, dir.beg));
3279       if (warned && DECL_P (fmtres.nonstr))
3280 	inform (DECL_SOURCE_LOCATION (fmtres.nonstr),
3281 		"referenced argument declared here");
3282       return false;
3283     }
3284 
3285   if (warned && fmtres.range.min < fmtres.range.likely
3286       && fmtres.range.likely < fmtres.range.max)
3287     inform_n (info.fmtloc, fmtres.range.likely,
3288 	      "assuming directive output of %wu byte",
3289 	      "assuming directive output of %wu bytes",
3290 	      fmtres.range.likely);
3291 
3292   if (warned && fmtres.argmin)
3293     {
3294       if (fmtres.argmin == fmtres.argmax)
3295 	inform (info.fmtloc, "directive argument %qE", fmtres.argmin);
3296       else if (fmtres.knownrange)
3297 	inform (info.fmtloc, "directive argument in the range [%E, %E]",
3298 		fmtres.argmin, fmtres.argmax);
3299       else
3300 	inform (info.fmtloc,
3301 		"using the range [%E, %E] for directive argument",
3302 		fmtres.argmin, fmtres.argmax);
3303     }
3304 
3305   res->warned |= warned;
3306 
3307   if (!dir.beg[0] && res->warned)
3308     {
3309       location_t callloc = gimple_location (info.callstmt);
3310 
3311       unsigned HOST_WIDE_INT min = res->range.min;
3312       unsigned HOST_WIDE_INT max = res->range.max;
3313 
3314       if (info.objsize < HOST_WIDE_INT_MAX)
3315 	{
3316 	  /* If a warning has been issued for buffer overflow or truncation
3317 	     help the user figure out how big a buffer they need.  */
3318 
3319 	  if (min == max)
3320 	    inform_n (callloc, min,
3321 		      "%qE output %wu byte into a destination of size %wu",
3322 		      "%qE output %wu bytes into a destination of size %wu",
3323 		      info.func, min, info.objsize);
3324 	  else if (max < HOST_WIDE_INT_MAX)
3325 	    inform (callloc,
3326 		    "%qE output between %wu and %wu bytes into "
3327 		    "a destination of size %wu",
3328 		    info.func, min, max, info.objsize);
3329 	  else if (min < res->range.likely && res->range.likely < max)
3330 	    inform (callloc,
3331 		    "%qE output %wu or more bytes (assuming %wu) into "
3332 		    "a destination of size %wu",
3333 		    info.func, min, res->range.likely, info.objsize);
3334 	  else
3335 	    inform (callloc,
3336 		    "%qE output %wu or more bytes into a destination of size "
3337 		    "%wu",
3338 		    info.func, min, info.objsize);
3339 	}
3340       else if (!info.is_string_func ())
3341 	{
3342 	  /* If the warning is for a file function like fprintf
3343 	     of printf with no destination size just print the computed
3344 	     result.  */
3345 	  if (min == max)
3346 	    inform_n (callloc, min,
3347 		      "%qE output %wu byte", "%qE output %wu bytes",
3348 		      info.func, min);
3349 	  else if (max < HOST_WIDE_INT_MAX)
3350 	    inform (callloc,
3351 		    "%qE output between %wu and %wu bytes",
3352 		    info.func, min, max);
3353 	  else if (min < res->range.likely && res->range.likely < max)
3354 	    inform (callloc,
3355 		    "%qE output %wu or more bytes (assuming %wu)",
3356 		    info.func, min, res->range.likely);
3357 	  else
3358 	    inform (callloc,
3359 		    "%qE output %wu or more bytes",
3360 		    info.func, min);
3361 	}
3362     }
3363 
3364   if (dump_file && *dir.beg)
3365     {
3366       fprintf (dump_file,
3367 	       "    Result: "
3368 	       HOST_WIDE_INT_PRINT_DEC ", " HOST_WIDE_INT_PRINT_DEC ", "
3369 	       HOST_WIDE_INT_PRINT_DEC ", " HOST_WIDE_INT_PRINT_DEC " ("
3370 	       HOST_WIDE_INT_PRINT_DEC ", " HOST_WIDE_INT_PRINT_DEC ", "
3371 	       HOST_WIDE_INT_PRINT_DEC ", " HOST_WIDE_INT_PRINT_DEC ")\n",
3372 	       fmtres.range.min, fmtres.range.likely,
3373 	       fmtres.range.max, fmtres.range.unlikely,
3374 	       res->range.min, res->range.likely,
3375 	       res->range.max, res->range.unlikely);
3376     }
3377 
3378   return true;
3379 }
3380 
3381 /* Parse a format directive in function call described by INFO starting
3382    at STR and populate DIR structure.  Bump up *ARGNO by the number of
3383    arguments extracted for the directive.  Return the length of
3384    the directive.  */
3385 
3386 static size_t
parse_directive(call_info & info,directive & dir,format_result * res,const char * str,unsigned * argno,const vr_values * vr_values)3387 parse_directive (call_info &info,
3388 		 directive &dir, format_result *res,
3389 		 const char *str, unsigned *argno,
3390 		 const vr_values *vr_values)
3391 {
3392   const char *pcnt = strchr (str, target_percent);
3393   dir.beg = str;
3394 
3395   if (size_t len = pcnt ? pcnt - str : *str ? strlen (str) : 1)
3396     {
3397       /* This directive is either a plain string or the terminating nul
3398 	 (which isn't really a directive but it simplifies things to
3399 	 handle it as if it were).  */
3400       dir.len = len;
3401       dir.fmtfunc = format_plain;
3402 
3403       if (dump_file)
3404 	{
3405 	  fprintf (dump_file, "  Directive %u at offset "
3406 		   HOST_WIDE_INT_PRINT_UNSIGNED ": \"%.*s\", "
3407 		   "length = " HOST_WIDE_INT_PRINT_UNSIGNED "\n",
3408 		   dir.dirno,
3409 		   (unsigned HOST_WIDE_INT)(size_t)(dir.beg - info.fmtstr),
3410 		   (int)dir.len, dir.beg, (unsigned HOST_WIDE_INT) dir.len);
3411 	}
3412 
3413       return len - !*str;
3414     }
3415 
3416   /* Set the directive argument's number to correspond to its position
3417      in the formatted function call's argument list.  */
3418   dir.argno = *argno;
3419 
3420   const char *pf = pcnt + 1;
3421 
3422     /* POSIX numbered argument index or zero when none.  */
3423   HOST_WIDE_INT dollar = 0;
3424 
3425   /* With and precision.  -1 when not specified, HOST_WIDE_INT_MIN
3426      when given by a va_list argument, and a non-negative value
3427      when specified in the format string itself.  */
3428   HOST_WIDE_INT width = -1;
3429   HOST_WIDE_INT precision = -1;
3430 
3431   /* Pointers to the beginning of the width and precision decimal
3432      string (if any) within the directive.  */
3433   const char *pwidth = 0;
3434   const char *pprec = 0;
3435 
3436   /* When the value of the decimal string that specifies width or
3437      precision is out of range, points to the digit that causes
3438      the value to exceed the limit.  */
3439   const char *werange = NULL;
3440   const char *perange = NULL;
3441 
3442   /* Width specified via the asterisk.  Need not be INTEGER_CST.
3443      For vararg functions set to void_node.  */
3444   tree star_width = NULL_TREE;
3445 
3446   /* Width specified via the asterisk.  Need not be INTEGER_CST.
3447      For vararg functions set to void_node.  */
3448   tree star_precision = NULL_TREE;
3449 
3450   if (ISDIGIT (target_to_host (*pf)))
3451     {
3452       /* This could be either a POSIX positional argument, the '0'
3453 	 flag, or a width, depending on what follows.  Store it as
3454 	 width and sort it out later after the next character has
3455 	 been seen.  */
3456       pwidth = pf;
3457       width = target_strtowi (&pf, &werange);
3458     }
3459   else if (target_to_host (*pf) == '*')
3460     {
3461       /* Similarly to the block above, this could be either a POSIX
3462 	 positional argument or a width, depending on what follows.  */
3463       if (*argno < gimple_call_num_args (info.callstmt))
3464 	star_width = gimple_call_arg (info.callstmt, (*argno)++);
3465       else
3466 	star_width = void_node;
3467       ++pf;
3468     }
3469 
3470   if (target_to_host (*pf) == '$')
3471     {
3472       /* Handle the POSIX dollar sign which references the 1-based
3473 	 positional argument number.  */
3474       if (width != -1)
3475 	dollar = width + info.argidx;
3476       else if (star_width
3477 	       && TREE_CODE (star_width) == INTEGER_CST
3478 	       && (TYPE_PRECISION (TREE_TYPE (star_width))
3479 		   <= TYPE_PRECISION (integer_type_node)))
3480 	dollar = width + tree_to_shwi (star_width);
3481 
3482       /* Bail when the numbered argument is out of range (it will
3483 	 have already been diagnosed by -Wformat).  */
3484       if (dollar == 0
3485 	  || dollar == (int)info.argidx
3486 	  || dollar > gimple_call_num_args (info.callstmt))
3487 	return false;
3488 
3489       --dollar;
3490 
3491       star_width = NULL_TREE;
3492       width = -1;
3493       ++pf;
3494     }
3495 
3496   if (dollar || !star_width)
3497     {
3498       if (width != -1)
3499 	{
3500 	  if (width == 0)
3501 	    {
3502 	      /* The '0' that has been interpreted as a width above is
3503 		 actually a flag.  Reset HAVE_WIDTH, set the '0' flag,
3504 		 and continue processing other flags.  */
3505 	      width = -1;
3506 	      dir.set_flag ('0');
3507 	    }
3508 	  else if (!dollar)
3509 	    {
3510 	      /* (Non-zero) width has been seen.  The next character
3511 		 is either a period or a digit.  */
3512 	      goto start_precision;
3513 	    }
3514 	}
3515       /* When either '$' has been seen, or width has not been seen,
3516 	 the next field is the optional flags followed by an optional
3517 	 width.  */
3518       for ( ; ; ) {
3519 	switch (target_to_host (*pf))
3520 	  {
3521 	  case ' ':
3522 	  case '0':
3523 	  case '+':
3524 	  case '-':
3525 	  case '#':
3526 	    dir.set_flag (target_to_host (*pf++));
3527 	    break;
3528 
3529 	  default:
3530 	    goto start_width;
3531 	  }
3532       }
3533 
3534     start_width:
3535       if (ISDIGIT (target_to_host (*pf)))
3536 	{
3537 	  werange = 0;
3538 	  pwidth = pf;
3539 	  width = target_strtowi (&pf, &werange);
3540 	}
3541       else if (target_to_host (*pf) == '*')
3542 	{
3543 	  if (*argno < gimple_call_num_args (info.callstmt))
3544 	    star_width = gimple_call_arg (info.callstmt, (*argno)++);
3545 	  else
3546 	    {
3547 	      /* This is (likely) a va_list.  It could also be an invalid
3548 		 call with insufficient arguments.  */
3549 	      star_width = void_node;
3550 	    }
3551 	  ++pf;
3552 	}
3553       else if (target_to_host (*pf) == '\'')
3554 	{
3555 	  /* The POSIX apostrophe indicating a numeric grouping
3556 	     in the current locale.  Even though it's possible to
3557 	     estimate the upper bound on the size of the output
3558 	     based on the number of digits it probably isn't worth
3559 	     continuing.  */
3560 	  return 0;
3561 	}
3562     }
3563 
3564  start_precision:
3565   if (target_to_host (*pf) == '.')
3566     {
3567       ++pf;
3568 
3569       if (ISDIGIT (target_to_host (*pf)))
3570 	{
3571 	  pprec = pf;
3572 	  precision = target_strtowi (&pf, &perange);
3573 	}
3574       else if (target_to_host (*pf) == '*')
3575 	{
3576 	  if (*argno < gimple_call_num_args (info.callstmt))
3577 	    star_precision = gimple_call_arg (info.callstmt, (*argno)++);
3578 	  else
3579 	    {
3580 	      /* This is (likely) a va_list.  It could also be an invalid
3581 		 call with insufficient arguments.  */
3582 	      star_precision = void_node;
3583 	    }
3584 	  ++pf;
3585 	}
3586       else
3587 	{
3588 	  /* The decimal precision or the asterisk are optional.
3589 	     When neither is dirified it's taken to be zero.  */
3590 	  precision = 0;
3591 	}
3592     }
3593 
3594   switch (target_to_host (*pf))
3595     {
3596     case 'h':
3597       if (target_to_host (pf[1]) == 'h')
3598 	{
3599 	  ++pf;
3600 	  dir.modifier = FMT_LEN_hh;
3601 	}
3602       else
3603 	dir.modifier = FMT_LEN_h;
3604       ++pf;
3605       break;
3606 
3607     case 'j':
3608       dir.modifier = FMT_LEN_j;
3609       ++pf;
3610       break;
3611 
3612     case 'L':
3613       dir.modifier = FMT_LEN_L;
3614       ++pf;
3615       break;
3616 
3617     case 'l':
3618       if (target_to_host (pf[1]) == 'l')
3619 	{
3620 	  ++pf;
3621 	  dir.modifier = FMT_LEN_ll;
3622 	}
3623       else
3624 	dir.modifier = FMT_LEN_l;
3625       ++pf;
3626       break;
3627 
3628     case 't':
3629       dir.modifier = FMT_LEN_t;
3630       ++pf;
3631       break;
3632 
3633     case 'z':
3634       dir.modifier = FMT_LEN_z;
3635       ++pf;
3636       break;
3637     }
3638 
3639   switch (target_to_host (*pf))
3640     {
3641       /* Handle a sole '%' character the same as "%%" but since it's
3642 	 undefined prevent the result from being folded.  */
3643     case '\0':
3644       --pf;
3645       res->range.min = res->range.max = HOST_WIDE_INT_M1U;
3646       /* FALLTHRU */
3647     case '%':
3648       dir.fmtfunc = format_percent;
3649       break;
3650 
3651     case 'a':
3652     case 'A':
3653     case 'e':
3654     case 'E':
3655     case 'f':
3656     case 'F':
3657     case 'g':
3658     case 'G':
3659       res->floating = true;
3660       dir.fmtfunc = format_floating;
3661       break;
3662 
3663     case 'd':
3664     case 'i':
3665     case 'o':
3666     case 'u':
3667     case 'x':
3668     case 'X':
3669       dir.fmtfunc = format_integer;
3670       break;
3671 
3672     case 'p':
3673       /* The %p output is implementation-defined.  It's possible
3674 	 to determine this format but due to extensions (edirially
3675 	 those of the Linux kernel -- see bug 78512) the first %p
3676 	 in the format string disables any further processing.  */
3677       return false;
3678 
3679     case 'n':
3680       /* %n has side-effects even when nothing is actually printed to
3681 	 any buffer.  */
3682       info.nowrite = false;
3683       dir.fmtfunc = format_none;
3684       break;
3685 
3686     case 'C':
3687     case 'c':
3688       /* POSIX wide character and C/POSIX narrow character.  */
3689       dir.fmtfunc = format_character;
3690       break;
3691 
3692     case 'S':
3693     case 's':
3694       /* POSIX wide string and C/POSIX narrow character string.  */
3695       dir.fmtfunc = format_string;
3696       break;
3697 
3698     default:
3699       /* Unknown conversion specification.  */
3700       return 0;
3701     }
3702 
3703   dir.specifier = target_to_host (*pf++);
3704 
3705   /* Store the length of the format directive.  */
3706   dir.len = pf - pcnt;
3707 
3708   /* Buffer for the directive in the host character set (used when
3709      the source character set is different).  */
3710   char hostdir[32];
3711 
3712   if (star_width)
3713     {
3714       if (INTEGRAL_TYPE_P (TREE_TYPE (star_width)))
3715 	dir.set_width (star_width, vr_values);
3716       else
3717 	{
3718 	  /* Width specified by a va_list takes on the range [0, -INT_MIN]
3719 	     (width is the absolute value of that specified).  */
3720 	  dir.width[0] = 0;
3721 	  dir.width[1] = target_int_max () + 1;
3722 	}
3723     }
3724   else
3725     {
3726       if (width == HOST_WIDE_INT_MAX && werange)
3727 	{
3728 	  size_t begin = dir.beg - info.fmtstr + (pwidth - pcnt);
3729 	  size_t caret = begin + (werange - pcnt);
3730 	  size_t end = pf - info.fmtstr - 1;
3731 
3732 	  /* Create a location for the width part of the directive,
3733 	     pointing the caret at the first out-of-range digit.  */
3734 	  substring_loc dirloc (info.fmtloc, TREE_TYPE (info.format),
3735 				caret, begin, end);
3736 
3737 	  fmtwarn (dirloc, UNKNOWN_LOCATION, NULL, info.warnopt (),
3738 		   "%<%.*s%> directive width out of range", (int) dir.len,
3739 		   target_to_host (hostdir, sizeof hostdir, dir.beg));
3740 	}
3741 
3742       dir.set_width (width);
3743     }
3744 
3745   if (star_precision)
3746     {
3747       if (INTEGRAL_TYPE_P (TREE_TYPE (star_precision)))
3748 	dir.set_precision (star_precision, vr_values);
3749       else
3750 	{
3751 	  /* Precision specified by a va_list takes on the range [-1, INT_MAX]
3752 	     (unlike width, negative precision is ignored).  */
3753 	  dir.prec[0] = -1;
3754 	  dir.prec[1] = target_int_max ();
3755 	}
3756     }
3757   else
3758     {
3759       if (precision == HOST_WIDE_INT_MAX && perange)
3760 	{
3761 	  size_t begin = dir.beg - info.fmtstr + (pprec - pcnt) - 1;
3762 	  size_t caret = dir.beg - info.fmtstr + (perange - pcnt) - 1;
3763 	  size_t end = pf - info.fmtstr - 2;
3764 
3765 	  /* Create a location for the precision part of the directive,
3766 	     including the leading period, pointing the caret at the first
3767 	     out-of-range digit .  */
3768 	  substring_loc dirloc (info.fmtloc, TREE_TYPE (info.format),
3769 				caret, begin, end);
3770 
3771 	  fmtwarn (dirloc, UNKNOWN_LOCATION, NULL, info.warnopt (),
3772 		   "%<%.*s%> directive precision out of range", (int) dir.len,
3773 		   target_to_host (hostdir, sizeof hostdir, dir.beg));
3774 	}
3775 
3776       dir.set_precision (precision);
3777     }
3778 
3779   /* Extract the argument if the directive takes one and if it's
3780      available (e.g., the function doesn't take a va_list).  Treat
3781      missing arguments the same as va_list, even though they will
3782      have likely already been diagnosed by -Wformat.  */
3783   if (dir.specifier != '%'
3784       && *argno < gimple_call_num_args (info.callstmt))
3785     dir.arg = gimple_call_arg (info.callstmt, dollar ? dollar : (*argno)++);
3786 
3787   if (dump_file)
3788     {
3789       fprintf (dump_file,
3790 	       "  Directive %u at offset " HOST_WIDE_INT_PRINT_UNSIGNED
3791 	       ": \"%.*s\"",
3792 	       dir.dirno,
3793 	       (unsigned HOST_WIDE_INT)(size_t)(dir.beg - info.fmtstr),
3794 	       (int)dir.len, dir.beg);
3795       if (star_width)
3796 	{
3797 	  if (dir.width[0] == dir.width[1])
3798 	    fprintf (dump_file, ", width = " HOST_WIDE_INT_PRINT_DEC,
3799 		     dir.width[0]);
3800 	  else
3801 	    fprintf (dump_file,
3802 		     ", width in range [" HOST_WIDE_INT_PRINT_DEC
3803 		     ", " HOST_WIDE_INT_PRINT_DEC "]",
3804 		     dir.width[0], dir.width[1]);
3805 	}
3806 
3807       if (star_precision)
3808 	{
3809 	  if (dir.prec[0] == dir.prec[1])
3810 	    fprintf (dump_file, ", precision = " HOST_WIDE_INT_PRINT_DEC,
3811 		     dir.prec[0]);
3812 	  else
3813 	    fprintf (dump_file,
3814 		     ", precision in range [" HOST_WIDE_INT_PRINT_DEC
3815 		     HOST_WIDE_INT_PRINT_DEC "]",
3816 		     dir.prec[0], dir.prec[1]);
3817 	}
3818       fputc ('\n', dump_file);
3819     }
3820 
3821   return dir.len;
3822 }
3823 
3824 /* Diagnose overlap between destination and %s directive arguments.  */
3825 
3826 static void
maybe_warn_overlap(call_info & info,format_result * res)3827 maybe_warn_overlap (call_info &info, format_result *res)
3828 {
3829   /* Two vectors of 1-based indices corresponding to either certainly
3830      or possibly aliasing arguments.  */
3831   auto_vec<int, 16> aliasarg[2];
3832 
3833   /* Go through the array of potentially aliasing directives and collect
3834      argument numbers of those that do or may overlap the destination
3835      object given the full result.  */
3836   for (unsigned i = 0; i != res->alias_count; ++i)
3837     {
3838       const format_result::alias_info &alias = res->aliases[i];
3839 
3840       enum { possible = -1, none = 0, certain = 1 } overlap = none;
3841 
3842       /* If the precision is zero there is no overlap.  (This only
3843 	 considers %s directives and ignores %n.)  */
3844       if (alias.dir.prec[0] == 0 && alias.dir.prec[1] == 0)
3845 	continue;
3846 
3847       if (alias.offset == HOST_WIDE_INT_MAX
3848 	  || info.dst_offset == HOST_WIDE_INT_MAX)
3849 	overlap = possible;
3850       else if (alias.offset == info.dst_offset)
3851 	overlap = alias.dir.prec[0] == 0 ? possible : certain;
3852       else
3853 	{
3854 	  /* Determine overlap from the range of output and offsets
3855 	     into the same destination as the source, and rule out
3856 	     impossible overlap.  */
3857 	  unsigned HOST_WIDE_INT albeg = alias.offset;
3858 	  unsigned HOST_WIDE_INT dstbeg = info.dst_offset;
3859 
3860 	  unsigned HOST_WIDE_INT alend = albeg + alias.range.min;
3861 	  unsigned HOST_WIDE_INT dstend = dstbeg + res->range.min - 1;
3862 
3863 	  if ((albeg <= dstbeg && alend > dstbeg)
3864 	      || (albeg >= dstbeg && albeg < dstend))
3865 	    overlap = certain;
3866 	  else
3867 	    {
3868 	      alend = albeg + alias.range.max;
3869 	      if (alend < albeg)
3870 		alend = HOST_WIDE_INT_M1U;
3871 
3872 	      dstend = dstbeg + res->range.max - 1;
3873 	      if (dstend < dstbeg)
3874 		dstend = HOST_WIDE_INT_M1U;
3875 
3876 	      if ((albeg >= dstbeg && albeg <= dstend)
3877 		  || (alend >= dstbeg && alend <= dstend))
3878 		overlap = possible;
3879 	    }
3880 	}
3881 
3882       if (overlap == none)
3883 	continue;
3884 
3885       /* Append the 1-based argument number.  */
3886       aliasarg[overlap != certain].safe_push (alias.dir.argno + 1);
3887 
3888       /* Disable any kind of optimization.  */
3889       res->range.unlikely = HOST_WIDE_INT_M1U;
3890     }
3891 
3892   tree arg0 = gimple_call_arg (info.callstmt, 0);
3893   location_t loc = gimple_location (info.callstmt);
3894 
3895   bool aliaswarn = false;
3896 
3897   unsigned ncertain = aliasarg[0].length ();
3898   unsigned npossible = aliasarg[1].length ();
3899   if (ncertain && npossible)
3900     {
3901       /* If there are multiple arguments that overlap, some certainly
3902 	 and some possibly, handle both sets in a single diagnostic.  */
3903       aliaswarn
3904 	= warning_at (loc, OPT_Wrestrict,
3905 		      "%qE arguments %Z and maybe %Z overlap destination "
3906 		      "object %qE",
3907 		      info.func, aliasarg[0].address (), ncertain,
3908 		      aliasarg[1].address (), npossible,
3909 		      info.dst_origin);
3910     }
3911   else if (ncertain)
3912     {
3913       /* There is only one set of two or more arguments and they all
3914 	 certainly overlap the destination.  */
3915       aliaswarn
3916 	= warning_n (loc, OPT_Wrestrict, ncertain,
3917 		     "%qE argument %Z overlaps destination object %qE",
3918 		     "%qE arguments %Z overlap destination object %qE",
3919 		     info.func, aliasarg[0].address (), ncertain,
3920 		     info.dst_origin);
3921     }
3922   else if (npossible)
3923     {
3924       /* There is only one set of two or more arguments and they all
3925 	 may overlap (but need not).  */
3926       aliaswarn
3927 	= warning_n (loc, OPT_Wrestrict, npossible,
3928 		     "%qE argument %Z may overlap destination object %qE",
3929 		     "%qE arguments %Z may overlap destination object %qE",
3930 		     info.func, aliasarg[1].address (), npossible,
3931 		     info.dst_origin);
3932     }
3933 
3934   if (aliaswarn)
3935     {
3936       res->warned = true;
3937 
3938       if (info.dst_origin != arg0)
3939 	{
3940 	  /* If its location is different from the first argument of the call
3941 	     point either at the destination object itself or at the expression
3942 	     that was used to determine the overlap.  */
3943 	  loc = (DECL_P (info.dst_origin)
3944 		 ? DECL_SOURCE_LOCATION (info.dst_origin)
3945 		 : EXPR_LOCATION (info.dst_origin));
3946 	  if (loc != UNKNOWN_LOCATION)
3947 	    inform (loc,
3948 		    "destination object referenced by %<restrict%>-qualified "
3949 		    "argument 1 was declared here");
3950 	}
3951     }
3952 }
3953 
3954 /* Compute the length of the output resulting from the call to a formatted
3955    output function described by INFO and store the result of the call in
3956    *RES.  Issue warnings for detected past the end writes.  Return true
3957    if the complete format string has been processed and *RES can be relied
3958    on, false otherwise (e.g., when a unknown or unhandled directive was seen
3959    that caused the processing to be terminated early).  */
3960 
3961 static bool
compute_format_length(call_info & info,format_result * res,const vr_values * vr)3962 compute_format_length (call_info &info, format_result *res, const vr_values *vr)
3963 {
3964   if (dump_file)
3965     {
3966       location_t callloc = gimple_location (info.callstmt);
3967       fprintf (dump_file, "%s:%i: ",
3968 	       LOCATION_FILE (callloc), LOCATION_LINE (callloc));
3969       print_generic_expr (dump_file, info.func, dump_flags);
3970 
3971       fprintf (dump_file,
3972 	       ": objsize = " HOST_WIDE_INT_PRINT_UNSIGNED
3973 	       ", fmtstr = \"%s\"\n",
3974 	       info.objsize, info.fmtstr);
3975     }
3976 
3977   /* Reset the minimum and maximum byte counters.  */
3978   res->range.min = res->range.max = 0;
3979 
3980   /* No directive has been seen yet so the length of output is bounded
3981      by the known range [0, 0] (with no conversion resulting in a failure
3982      or producing more than 4K bytes) until determined otherwise.  */
3983   res->knownrange = true;
3984   res->floating = false;
3985   res->warned = false;
3986 
3987   /* 1-based directive counter.  */
3988   unsigned dirno = 1;
3989 
3990   /* The variadic argument counter.  */
3991   unsigned argno = info.argidx;
3992 
3993   bool success = true;
3994 
3995   for (const char *pf = info.fmtstr; ; ++dirno)
3996     {
3997       directive dir (&info, dirno);
3998 
3999       size_t n = parse_directive (info, dir, res, pf, &argno, vr);
4000 
4001       /* Return failure if the format function fails.  */
4002       if (!format_directive (info, res, dir, vr))
4003 	return false;
4004 
4005       /* Return success when the directive is zero bytes long and it's
4006 	 the last thing in the format string (i.e., it's the terminating
4007 	 nul, which isn't really a directive but handling it as one makes
4008 	 things simpler).  */
4009       if (!n)
4010 	{
4011 	  success = *pf == '\0';
4012 	  break;
4013 	}
4014 
4015       pf += n;
4016     }
4017 
4018   maybe_warn_overlap (info, res);
4019 
4020   /* The complete format string was processed (with or without warnings).  */
4021   return success;
4022 }
4023 
4024 /* Return the size of the object referenced by the expression DEST if
4025    available, or the maximum possible size otherwise.  */
4026 
4027 static unsigned HOST_WIDE_INT
get_destination_size(tree dest)4028 get_destination_size (tree dest)
4029 {
4030   /* When there is no destination return the maximum.  */
4031   if (!dest)
4032     return HOST_WIDE_INT_MAX;
4033 
4034   /* Initialize object size info before trying to compute it.  */
4035   init_object_sizes ();
4036 
4037   /* Use __builtin_object_size to determine the size of the destination
4038      object.  When optimizing, determine the smallest object (such as
4039      a member array as opposed to the whole enclosing object), otherwise
4040      use type-zero object size to determine the size of the enclosing
4041      object (the function fails without optimization in this type).  */
4042   int ost = optimize > 0;
4043   unsigned HOST_WIDE_INT size;
4044   if (compute_builtin_object_size (dest, ost, &size))
4045     return size;
4046 
4047   return HOST_WIDE_INT_MAX;
4048 }
4049 
4050 /* Return true if the call described by INFO with result RES safe to
4051    optimize (i.e., no undefined behavior), and set RETVAL to the range
4052    of its return values.  */
4053 
4054 static bool
is_call_safe(const call_info & info,const format_result & res,bool under4k,unsigned HOST_WIDE_INT retval[2])4055 is_call_safe (const call_info &info,
4056 	      const format_result &res, bool under4k,
4057 	      unsigned HOST_WIDE_INT retval[2])
4058 {
4059   if (under4k && !res.posunder4k)
4060     return false;
4061 
4062   /* The minimum return value.  */
4063   retval[0] = res.range.min;
4064 
4065   /* The maximum return value is in most cases bounded by RES.RANGE.MAX
4066      but in cases involving multibyte characters could be as large as
4067      RES.RANGE.UNLIKELY.  */
4068   retval[1]
4069     = res.range.unlikely < res.range.max ? res.range.max : res.range.unlikely;
4070 
4071   /* Adjust the number of bytes which includes the terminating nul
4072      to reflect the return value of the function which does not.
4073      Because the valid range of the function is [INT_MIN, INT_MAX],
4074      a valid range before the adjustment below is [0, INT_MAX + 1]
4075      (the functions only return negative values on error or undefined
4076      behavior).  */
4077   if (retval[0] <= target_int_max () + 1)
4078     --retval[0];
4079   if (retval[1] <= target_int_max () + 1)
4080     --retval[1];
4081 
4082   /* Avoid the return value optimization when the behavior of the call
4083      is undefined either because any directive may have produced 4K or
4084      more of output, or the return value exceeds INT_MAX, or because
4085      the output overflows the destination object (but leave it enabled
4086      when the function is bounded because then the behavior is well-
4087      defined).  */
4088   if (retval[0] == retval[1]
4089       && (info.bounded || retval[0] < info.objsize)
4090       && retval[0] <= target_int_max ())
4091     return true;
4092 
4093   if ((info.bounded || retval[1] < info.objsize)
4094       && (retval[0] < target_int_max ()
4095 	  && retval[1] < target_int_max ()))
4096     return true;
4097 
4098   if (!under4k && (info.bounded || retval[0] < info.objsize))
4099     return true;
4100 
4101   return false;
4102 }
4103 
4104 /* Given a suitable result RES of a call to a formatted output function
4105    described by INFO, substitute the result for the return value of
4106    the call.  The result is suitable if the number of bytes it represents
4107    is known and exact.  A result that isn't suitable for substitution may
4108    have its range set to the range of return values, if that is known.
4109    Return true if the call is removed and gsi_next should not be performed
4110    in the caller.  */
4111 
4112 static bool
try_substitute_return_value(gimple_stmt_iterator * gsi,const call_info & info,const format_result & res)4113 try_substitute_return_value (gimple_stmt_iterator *gsi,
4114 			     const call_info &info,
4115 			     const format_result &res)
4116 {
4117   tree lhs = gimple_get_lhs (info.callstmt);
4118 
4119   /* Set to true when the entire call has been removed.  */
4120   bool removed = false;
4121 
4122   /* The minimum and maximum return value.  */
4123   unsigned HOST_WIDE_INT retval[2] = {0};
4124   bool safe = is_call_safe (info, res, true, retval);
4125 
4126   if (safe
4127       && retval[0] == retval[1]
4128       /* Not prepared to handle possibly throwing calls here; they shouldn't
4129 	 appear in non-artificial testcases, except when the __*_chk routines
4130 	 are badly declared.  */
4131       && !stmt_ends_bb_p (info.callstmt))
4132     {
4133       tree cst = build_int_cst (lhs ? TREE_TYPE (lhs) : integer_type_node,
4134 				retval[0]);
4135 
4136       if (lhs == NULL_TREE && info.nowrite)
4137 	{
4138 	  /* Remove the call to the bounded function with a zero size
4139 	     (e.g., snprintf(0, 0, "%i", 123)) if there is no lhs.  */
4140 	  unlink_stmt_vdef (info.callstmt);
4141 	  gsi_remove (gsi, true);
4142 	  removed = true;
4143 	}
4144       else if (info.nowrite)
4145 	{
4146 	  /* Replace the call to the bounded function with a zero size
4147 	     (e.g., snprintf(0, 0, "%i", 123) with the constant result
4148 	     of the function.  */
4149 	  if (!update_call_from_tree (gsi, cst))
4150 	    gimplify_and_update_call_from_tree (gsi, cst);
4151 	  gimple *callstmt = gsi_stmt (*gsi);
4152 	  update_stmt (callstmt);
4153 	}
4154       else if (lhs)
4155 	{
4156 	  /* Replace the left-hand side of the call with the constant
4157 	     result of the formatted function.  */
4158 	  gimple_call_set_lhs (info.callstmt, NULL_TREE);
4159 	  gimple *g = gimple_build_assign (lhs, cst);
4160 	  gsi_insert_after (gsi, g, GSI_NEW_STMT);
4161 	  update_stmt (info.callstmt);
4162 	}
4163 
4164       if (dump_file)
4165 	{
4166 	  if (removed)
4167 	    fprintf (dump_file, "  Removing call statement.");
4168 	  else
4169 	    {
4170 	      fprintf (dump_file, "  Substituting ");
4171 	      print_generic_expr (dump_file, cst, dump_flags);
4172 	      fprintf (dump_file, " for %s.\n",
4173 		       info.nowrite ? "statement" : "return value");
4174 	    }
4175 	}
4176     }
4177   else if (lhs && types_compatible_p (TREE_TYPE (lhs), integer_type_node))
4178     {
4179       bool setrange = false;
4180 
4181       if (safe
4182 	  && (info.bounded || retval[1] < info.objsize)
4183 	  && (retval[0] < target_int_max ()
4184 	      && retval[1] < target_int_max ()))
4185 	{
4186 	  /* If the result is in a valid range bounded by the size of
4187 	     the destination set it so that it can be used for subsequent
4188 	     optimizations.  */
4189 	  int prec = TYPE_PRECISION (integer_type_node);
4190 
4191 	  wide_int min = wi::shwi (retval[0], prec);
4192 	  wide_int max = wi::shwi (retval[1], prec);
4193 	  set_range_info (lhs, VR_RANGE, min, max);
4194 
4195 	  setrange = true;
4196 	}
4197 
4198       if (dump_file)
4199 	{
4200 	  const char *inbounds
4201 	    = (retval[0] < info.objsize
4202 	       ? (retval[1] < info.objsize
4203 		  ? "in" : "potentially out-of")
4204 	       : "out-of");
4205 
4206 	  const char *what = setrange ? "Setting" : "Discarding";
4207 	  if (retval[0] != retval[1])
4208 	    fprintf (dump_file,
4209 		     "  %s %s-bounds return value range ["
4210 		     HOST_WIDE_INT_PRINT_UNSIGNED ", "
4211 		     HOST_WIDE_INT_PRINT_UNSIGNED "].\n",
4212 		     what, inbounds, retval[0], retval[1]);
4213 	  else
4214 	    fprintf (dump_file, "  %s %s-bounds return value "
4215 		     HOST_WIDE_INT_PRINT_UNSIGNED ".\n",
4216 		     what, inbounds, retval[0]);
4217 	}
4218     }
4219 
4220   if (dump_file)
4221     fputc ('\n', dump_file);
4222 
4223   return removed;
4224 }
4225 
4226 /* Try to simplify a s{,n}printf call described by INFO with result
4227    RES by replacing it with a simpler and presumably more efficient
4228    call (such as strcpy).  */
4229 
4230 static bool
try_simplify_call(gimple_stmt_iterator * gsi,const call_info & info,const format_result & res)4231 try_simplify_call (gimple_stmt_iterator *gsi,
4232 		   const call_info &info,
4233 		   const format_result &res)
4234 {
4235   unsigned HOST_WIDE_INT dummy[2];
4236   if (!is_call_safe (info, res, info.retval_used (), dummy))
4237     return false;
4238 
4239   switch (info.fncode)
4240     {
4241     case BUILT_IN_SNPRINTF:
4242       return gimple_fold_builtin_snprintf (gsi);
4243 
4244     case BUILT_IN_SPRINTF:
4245       return gimple_fold_builtin_sprintf (gsi);
4246 
4247     default:
4248       ;
4249     }
4250 
4251   return false;
4252 }
4253 
4254 /* Return the zero-based index of the format string argument of a printf
4255    like function and set *IDX_ARGS to the first format argument.  When
4256    no such index exists return UINT_MAX.  */
4257 
4258 static unsigned
get_user_idx_format(tree fndecl,unsigned * idx_args)4259 get_user_idx_format (tree fndecl, unsigned *idx_args)
4260 {
4261   tree attrs = lookup_attribute ("format", DECL_ATTRIBUTES (fndecl));
4262   if (!attrs)
4263     attrs = lookup_attribute ("format", TYPE_ATTRIBUTES (TREE_TYPE (fndecl)));
4264 
4265   if (!attrs)
4266     return UINT_MAX;
4267 
4268   attrs = TREE_VALUE (attrs);
4269 
4270   tree archetype = TREE_VALUE (attrs);
4271   if (strcmp ("printf", IDENTIFIER_POINTER (archetype)))
4272     return UINT_MAX;
4273 
4274   attrs = TREE_CHAIN (attrs);
4275   tree fmtarg = TREE_VALUE (attrs);
4276 
4277   attrs = TREE_CHAIN (attrs);
4278   tree elliparg = TREE_VALUE (attrs);
4279 
4280   /* Attribute argument indices are 1-based but we use zero-based.  */
4281   *idx_args = tree_to_uhwi (elliparg) - 1;
4282   return tree_to_uhwi (fmtarg) - 1;
4283 }
4284 
4285 }   /* Unnamed namespace.  */
4286 
4287 /* Determine if a GIMPLE call at *GSI is to one of the sprintf-like built-in
4288    functions and if so, handle it.  Return true if the call is removed and
4289    gsi_next should not be performed in the caller.  */
4290 
4291 bool
handle_printf_call(gimple_stmt_iterator * gsi,const vr_values * vr_values)4292 handle_printf_call (gimple_stmt_iterator *gsi, const vr_values *vr_values)
4293 {
4294   init_target_to_host_charmap ();
4295 
4296   call_info info = call_info ();
4297 
4298   info.callstmt = gsi_stmt (*gsi);
4299   info.func = gimple_call_fndecl (info.callstmt);
4300   if (!info.func)
4301     return false;
4302 
4303   /* Format string argument number (valid for all functions).  */
4304   unsigned idx_format = UINT_MAX;
4305   if (gimple_call_builtin_p (info.callstmt, BUILT_IN_NORMAL))
4306     info.fncode = DECL_FUNCTION_CODE (info.func);
4307   else
4308     {
4309       unsigned idx_args;
4310       idx_format = get_user_idx_format (info.func, &idx_args);
4311       if (idx_format == UINT_MAX
4312 	  || idx_format >= gimple_call_num_args (info.callstmt)
4313 	  || idx_args > gimple_call_num_args (info.callstmt)
4314 	  || !POINTER_TYPE_P (TREE_TYPE (gimple_call_arg (info.callstmt,
4315 							  idx_format))))
4316 	return false;
4317       info.fncode = BUILT_IN_NONE;
4318       info.argidx = idx_args;
4319     }
4320 
4321   /* The size of the destination as in snprintf(dest, size, ...).  */
4322   unsigned HOST_WIDE_INT dstsize = HOST_WIDE_INT_M1U;
4323 
4324   /* The size of the destination determined by __builtin_object_size.  */
4325   unsigned HOST_WIDE_INT objsize = HOST_WIDE_INT_M1U;
4326 
4327   /* Zero-based buffer size argument number (snprintf and vsnprintf).  */
4328   unsigned idx_dstsize = UINT_MAX;
4329 
4330   /* Object size argument number (snprintf_chk and vsnprintf_chk).  */
4331   unsigned idx_objsize = UINT_MAX;
4332 
4333   /* Destinaton argument number (valid for sprintf functions only).  */
4334   unsigned idx_dstptr = 0;
4335 
4336   switch (info.fncode)
4337     {
4338     case BUILT_IN_NONE:
4339       // User-defined function with attribute format (printf).
4340       idx_dstptr = -1;
4341       break;
4342 
4343     case BUILT_IN_FPRINTF:
4344       // Signature:
4345       //   __builtin_fprintf (FILE*, format, ...)
4346       idx_format = 1;
4347       info.argidx = 2;
4348       idx_dstptr = -1;
4349       break;
4350 
4351     case BUILT_IN_FPRINTF_CHK:
4352       // Signature:
4353       //   __builtin_fprintf_chk (FILE*, ost, format, ...)
4354       idx_format = 2;
4355       info.argidx = 3;
4356       idx_dstptr = -1;
4357       break;
4358 
4359     case BUILT_IN_FPRINTF_UNLOCKED:
4360       // Signature:
4361       //   __builtin_fprintf_unnlocked (FILE*, format, ...)
4362       idx_format = 1;
4363       info.argidx = 2;
4364       idx_dstptr = -1;
4365       break;
4366 
4367     case BUILT_IN_PRINTF:
4368       // Signature:
4369       //   __builtin_printf (format, ...)
4370       idx_format = 0;
4371       info.argidx = 1;
4372       idx_dstptr = -1;
4373       break;
4374 
4375     case BUILT_IN_PRINTF_CHK:
4376       // Signature:
4377       //   __builtin_printf_chk (ost, format, ...)
4378       idx_format = 1;
4379       info.argidx = 2;
4380       idx_dstptr = -1;
4381       break;
4382 
4383     case BUILT_IN_PRINTF_UNLOCKED:
4384       // Signature:
4385       //   __builtin_printf (format, ...)
4386       idx_format = 0;
4387       info.argidx = 1;
4388       idx_dstptr = -1;
4389       break;
4390 
4391     case BUILT_IN_SPRINTF:
4392       // Signature:
4393       //   __builtin_sprintf (dst, format, ...)
4394       idx_format = 1;
4395       info.argidx = 2;
4396       break;
4397 
4398     case BUILT_IN_SPRINTF_CHK:
4399       // Signature:
4400       //   __builtin___sprintf_chk (dst, ost, objsize, format, ...)
4401       idx_objsize = 2;
4402       idx_format = 3;
4403       info.argidx = 4;
4404       break;
4405 
4406     case BUILT_IN_SNPRINTF:
4407       // Signature:
4408       //   __builtin_snprintf (dst, size, format, ...)
4409       idx_dstsize = 1;
4410       idx_format = 2;
4411       info.argidx = 3;
4412       info.bounded = true;
4413       break;
4414 
4415     case BUILT_IN_SNPRINTF_CHK:
4416       // Signature:
4417       //   __builtin___snprintf_chk (dst, size, ost, objsize, format, ...)
4418       idx_dstsize = 1;
4419       idx_objsize = 3;
4420       idx_format = 4;
4421       info.argidx = 5;
4422       info.bounded = true;
4423       break;
4424 
4425     case BUILT_IN_VFPRINTF:
4426       // Signature:
4427       //   __builtin_vprintf (FILE*, format, va_list)
4428       idx_format = 1;
4429       info.argidx = -1;
4430       idx_dstptr = -1;
4431       break;
4432 
4433     case BUILT_IN_VFPRINTF_CHK:
4434       // Signature:
4435       //   __builtin___vfprintf_chk (FILE*, ost, format, va_list)
4436       idx_format = 2;
4437       info.argidx = -1;
4438       idx_dstptr = -1;
4439       break;
4440 
4441     case BUILT_IN_VPRINTF:
4442       // Signature:
4443       //   __builtin_vprintf (format, va_list)
4444       idx_format = 0;
4445       info.argidx = -1;
4446       idx_dstptr = -1;
4447       break;
4448 
4449     case BUILT_IN_VPRINTF_CHK:
4450       // Signature:
4451       //   __builtin___vprintf_chk (ost, format, va_list)
4452       idx_format = 1;
4453       info.argidx = -1;
4454       idx_dstptr = -1;
4455       break;
4456 
4457     case BUILT_IN_VSNPRINTF:
4458       // Signature:
4459       //   __builtin_vsprintf (dst, size, format, va)
4460       idx_dstsize = 1;
4461       idx_format = 2;
4462       info.argidx = -1;
4463       info.bounded = true;
4464       break;
4465 
4466     case BUILT_IN_VSNPRINTF_CHK:
4467       // Signature:
4468       //   __builtin___vsnprintf_chk (dst, size, ost, objsize, format, va)
4469       idx_dstsize = 1;
4470       idx_objsize = 3;
4471       idx_format = 4;
4472       info.argidx = -1;
4473       info.bounded = true;
4474       break;
4475 
4476     case BUILT_IN_VSPRINTF:
4477       // Signature:
4478       //   __builtin_vsprintf (dst, format, va)
4479       idx_format = 1;
4480       info.argidx = -1;
4481       break;
4482 
4483     case BUILT_IN_VSPRINTF_CHK:
4484       // Signature:
4485       //   __builtin___vsprintf_chk (dst, ost, objsize, format, va)
4486       idx_format = 3;
4487       idx_objsize = 2;
4488       info.argidx = -1;
4489       break;
4490 
4491     default:
4492       return false;
4493     }
4494 
4495   /* Set the global warning level for this function.  */
4496   warn_level = info.bounded ? warn_format_trunc : warn_format_overflow;
4497 
4498   /* For all string functions the first argument is a pointer to
4499      the destination.  */
4500   tree dstptr = (idx_dstptr < gimple_call_num_args (info.callstmt)
4501 		 ? gimple_call_arg (info.callstmt, 0) : NULL_TREE);
4502 
4503   info.format = gimple_call_arg (info.callstmt, idx_format);
4504 
4505   /* True when the destination size is constant as opposed to the lower
4506      or upper bound of a range.  */
4507   bool dstsize_cst_p = true;
4508   bool posunder4k = true;
4509 
4510   if (idx_dstsize == UINT_MAX)
4511     {
4512       /* For non-bounded functions like sprintf, determine the size
4513 	 of the destination from the object or pointer passed to it
4514 	 as the first argument.  */
4515       dstsize = get_destination_size (dstptr);
4516     }
4517   else if (tree size = gimple_call_arg (info.callstmt, idx_dstsize))
4518     {
4519       /* For bounded functions try to get the size argument.  */
4520 
4521       if (TREE_CODE (size) == INTEGER_CST)
4522 	{
4523 	  dstsize = tree_to_uhwi (size);
4524 	  /* No object can be larger than SIZE_MAX bytes (half the address
4525 	     space) on the target.
4526 	     The functions are defined only for output of at most INT_MAX
4527 	     bytes.  Specifying a bound in excess of that limit effectively
4528 	     defeats the bounds checking (and on some implementations such
4529 	     as Solaris cause the function to fail with EINVAL).  */
4530 	  if (dstsize > target_size_max () / 2)
4531 	    {
4532 	      /* Avoid warning if -Wstringop-overflow is specified since
4533 		 it also warns for the same thing though only for the
4534 		 checking built-ins.  */
4535 	      if ((idx_objsize == UINT_MAX
4536 		   || !warn_stringop_overflow))
4537 		warning_at (gimple_location (info.callstmt), info.warnopt (),
4538 			    "specified bound %wu exceeds maximum object size "
4539 			    "%wu",
4540 			    dstsize, target_size_max () / 2);
4541 	      /* POSIX requires snprintf to fail if DSTSIZE is greater
4542 		 than INT_MAX.  Even though not all POSIX implementations
4543 		 conform to the requirement, avoid folding in this case.  */
4544 	      posunder4k = false;
4545 	    }
4546 	  else if (dstsize > target_int_max ())
4547 	    {
4548 	      warning_at (gimple_location (info.callstmt), info.warnopt (),
4549 			  "specified bound %wu exceeds %<INT_MAX%>",
4550 			  dstsize);
4551 	      /* POSIX requires snprintf to fail if DSTSIZE is greater
4552 		 than INT_MAX.  Avoid folding in that case.  */
4553 	      posunder4k = false;
4554 	    }
4555 	}
4556       else if (TREE_CODE (size) == SSA_NAME)
4557 	{
4558 	  /* Try to determine the range of values of the argument
4559 	     and use the greater of the two at level 1 and the smaller
4560 	     of them at level 2.  */
4561 	  const value_range_equiv *vr
4562 	    = CONST_CAST (class vr_values *, vr_values)->get_value_range (size);
4563 
4564 	  if (range_int_cst_p (vr))
4565 	    {
4566 	      unsigned HOST_WIDE_INT minsize = TREE_INT_CST_LOW (vr->min ());
4567 	      unsigned HOST_WIDE_INT maxsize = TREE_INT_CST_LOW (vr->max ());
4568 	      dstsize = warn_level < 2 ? maxsize : minsize;
4569 
4570 	      if (minsize > target_int_max ())
4571 		warning_at (gimple_location (info.callstmt), info.warnopt (),
4572 			    "specified bound range [%wu, %wu] exceeds "
4573 			    "%<INT_MAX%>",
4574 			    minsize, maxsize);
4575 
4576 	      /* POSIX requires snprintf to fail if DSTSIZE is greater
4577 		 than INT_MAX.  Avoid folding if that's possible.  */
4578 	      if (maxsize > target_int_max ())
4579 		posunder4k = false;
4580 	    }
4581 	  else if (vr->varying_p ())
4582 	    {
4583 	      /* POSIX requires snprintf to fail if DSTSIZE is greater
4584 		 than INT_MAX.  Since SIZE's range is unknown, avoid
4585 		 folding.  */
4586 	      posunder4k = false;
4587 	    }
4588 
4589 	  /* The destination size is not constant.  If the function is
4590 	     bounded (e.g., snprintf) a lower bound of zero doesn't
4591 	     necessarily imply it can be eliminated.  */
4592 	  dstsize_cst_p = false;
4593 	}
4594     }
4595 
4596   if (idx_objsize != UINT_MAX)
4597     if (tree size = gimple_call_arg (info.callstmt, idx_objsize))
4598       if (tree_fits_uhwi_p (size))
4599 	objsize = tree_to_uhwi (size);
4600 
4601   if (info.bounded && !dstsize)
4602     {
4603       /* As a special case, when the explicitly specified destination
4604 	 size argument (to a bounded function like snprintf) is zero
4605 	 it is a request to determine the number of bytes on output
4606 	 without actually producing any.  Pretend the size is
4607 	 unlimited in this case.  */
4608       info.objsize = HOST_WIDE_INT_MAX;
4609       info.nowrite = dstsize_cst_p;
4610     }
4611   else
4612     {
4613       /* For calls to non-bounded functions or to those of bounded
4614 	 functions with a non-zero size, warn if the destination
4615 	 pointer is null.  */
4616       if (dstptr && integer_zerop (dstptr))
4617 	{
4618 	  /* This is diagnosed with -Wformat only when the null is a constant
4619 	     pointer.  The warning here diagnoses instances where the pointer
4620 	     is not constant.  */
4621 	  location_t loc = gimple_location (info.callstmt);
4622 	  warning_at (EXPR_LOC_OR_LOC (dstptr, loc),
4623 		      info.warnopt (), "%Gnull destination pointer",
4624 		      info.callstmt);
4625 	  return false;
4626 	}
4627 
4628       /* Set the object size to the smaller of the two arguments
4629 	 of both have been specified and they're not equal.  */
4630       info.objsize = dstsize < objsize ? dstsize : objsize;
4631 
4632       if (info.bounded
4633 	  && dstsize < target_size_max () / 2 && objsize < dstsize
4634 	  /* Avoid warning if -Wstringop-overflow is specified since
4635 	     it also warns for the same thing though only for the
4636 	     checking built-ins.  */
4637 	  && (idx_objsize == UINT_MAX
4638 	      || !warn_stringop_overflow))
4639 	{
4640 	  warning_at (gimple_location (info.callstmt), info.warnopt (),
4641 		      "specified bound %wu exceeds the size %wu "
4642 		      "of the destination object", dstsize, objsize);
4643 	}
4644     }
4645 
4646   /* Determine if the format argument may be null and warn if not
4647      and if the argument is null.  */
4648   if (integer_zerop (info.format)
4649       && gimple_call_builtin_p (info.callstmt, BUILT_IN_NORMAL))
4650     {
4651       location_t loc = gimple_location (info.callstmt);
4652       warning_at (EXPR_LOC_OR_LOC (info.format, loc),
4653 		  info.warnopt (), "%Gnull format string",
4654 		  info.callstmt);
4655       return false;
4656     }
4657 
4658   info.fmtstr = get_format_string (info.format, &info.fmtloc);
4659   if (!info.fmtstr)
4660     return false;
4661 
4662   if (warn_restrict)
4663     {
4664       /* Compute the origin of the destination pointer and its offset
4665 	 from the base object/pointer if possible.  */
4666       info.dst_offset = 0;
4667       info.dst_origin = get_origin_and_offset (dstptr, &info.dst_field,
4668 					       &info.dst_offset);
4669     }
4670 
4671   /* The result is the number of bytes output by the formatted function,
4672      including the terminating NUL.  */
4673   format_result res;
4674 
4675   /* I/O functions with no destination argument (i.e., all forms of fprintf
4676      and printf) may fail under any conditions.  Others (i.e., all forms of
4677      sprintf) may only fail under specific conditions determined for each
4678      directive.  Clear POSUNDER4K for the former set of functions and set
4679      it to true for the latter (it can only be cleared later, but it is
4680      never set to true again).  */
4681   res.posunder4k = posunder4k && dstptr;
4682 
4683   bool success = compute_format_length (info, &res, vr_values);
4684   if (res.warned)
4685     gimple_set_no_warning (info.callstmt, true);
4686 
4687   /* When optimizing and the printf return value optimization is enabled,
4688      attempt to substitute the computed result for the return value of
4689      the call.  Avoid this optimization when -frounding-math is in effect
4690      and the format string contains a floating point directive.  */
4691   bool call_removed = false;
4692   if (success && optimize > 0)
4693     {
4694       /* Save a copy of the iterator pointing at the call.  The iterator
4695 	 may change to point past the call in try_substitute_return_value
4696 	 but the original value is needed in try_simplify_call.  */
4697       gimple_stmt_iterator gsi_call = *gsi;
4698 
4699       if (flag_printf_return_value
4700 	  && (!flag_rounding_math || !res.floating))
4701 	call_removed = try_substitute_return_value (gsi, info, res);
4702 
4703       if (!call_removed)
4704 	try_simplify_call (&gsi_call, info, res);
4705     }
4706 
4707   return call_removed;
4708 }
4709