* tree-outof-ssa.h (ssaexpand): Add partitions_for_undefined_values.
[official-gcc.git] / gcc / gimple-ssa-sprintf.c
blob7899e09195fe200088d95d54d94019e5c013b48a
1 /* Copyright (C) 2016-2017 Free Software Foundation, Inc.
2 Contributed by Martin Sebor <msebor@redhat.com>.
4 This file is part of GCC.
6 GCC is free software; you can redistribute it and/or modify it under
7 the terms of the GNU General Public License as published by the Free
8 Software Foundation; either version 3, or (at your option) any later
9 version.
11 GCC is distributed in the hope that it will be useful, but WITHOUT ANY
12 WARRANTY; without even the implied warranty of MERCHANTABILITY or
13 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
14 for more details.
16 You should have received a copy of the GNU General Public License
17 along with GCC; see the file COPYING3. If not see
18 <http://www.gnu.org/licenses/>. */
20 /* This file implements the printf-return-value pass. The pass does
21 two things: 1) it analyzes calls to formatted output functions like
22 sprintf looking for possible buffer overflows and calls to bounded
23 functions like snprintf for early truncation (and under the control
24 of the -Wformat-length option issues warnings), and 2) under the
25 control of the -fprintf-return-value option it folds the return
26 value of safe calls into constants, making it possible to eliminate
27 code that depends on the value of those constants.
29 For all functions (bounded or not) the pass uses the size of the
30 destination object. That means that it will diagnose calls to
31 snprintf not on the basis of the size specified by the function's
32 second argument but 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.
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. */
48 #include "config.h"
49 #include "system.h"
50 #include "coretypes.h"
51 #include "backend.h"
52 #include "tree.h"
53 #include "gimple.h"
54 #include "tree-pass.h"
55 #include "ssa.h"
56 #include "gimple-fold.h"
57 #include "gimple-pretty-print.h"
58 #include "diagnostic-core.h"
59 #include "fold-const.h"
60 #include "gimple-iterator.h"
61 #include "tree-ssa.h"
62 #include "tree-object-size.h"
63 #include "params.h"
64 #include "tree-cfg.h"
65 #include "tree-ssa-propagate.h"
66 #include "calls.h"
67 #include "cfgloop.h"
68 #include "intl.h"
69 #include "langhooks.h"
71 #include "builtins.h"
72 #include "stor-layout.h"
74 #include "realmpfr.h"
75 #include "target.h"
77 #include "cpplib.h"
78 #include "input.h"
79 #include "toplev.h"
80 #include "substring-locations.h"
81 #include "diagnostic.h"
83 /* The likely worst case value of MB_LEN_MAX for the target, large enough
84 for UTF-8. Ideally, this would be obtained by a target hook if it were
85 to be used for optimization but it's good enough as is for warnings. */
86 #define target_mb_len_max() 6
88 /* The maximum number of bytes a single non-string directive can result
89 in. This is the result of printf("%.*Lf", INT_MAX, -LDBL_MAX) for
90 LDBL_MAX_10_EXP of 4932. */
91 #define IEEE_MAX_10_EXP 4932
92 #define target_dir_max() (target_int_max () + IEEE_MAX_10_EXP + 2)
94 namespace {
96 const pass_data pass_data_sprintf_length = {
97 GIMPLE_PASS, // pass type
98 "printf-return-value", // pass name
99 OPTGROUP_NONE, // optinfo_flags
100 TV_NONE, // tv_id
101 PROP_cfg, // properties_required
102 0, // properties_provided
103 0, // properties_destroyed
104 0, // properties_start
105 0, // properties_finish
108 /* Set to the warning level for the current function which is equal
109 either to warn_format_trunc for bounded functions or to
110 warn_format_overflow otherwise. */
112 static int warn_level;
114 struct format_result;
116 class pass_sprintf_length : public gimple_opt_pass
118 bool fold_return_value;
120 public:
121 pass_sprintf_length (gcc::context *ctxt)
122 : gimple_opt_pass (pass_data_sprintf_length, ctxt),
123 fold_return_value (false)
126 opt_pass * clone () { return new pass_sprintf_length (m_ctxt); }
128 virtual bool gate (function *);
130 virtual unsigned int execute (function *);
132 void set_pass_param (unsigned int n, bool param)
134 gcc_assert (n == 0);
135 fold_return_value = param;
138 bool handle_gimple_call (gimple_stmt_iterator *);
140 struct call_info;
141 bool compute_format_length (call_info &, format_result *);
144 bool
145 pass_sprintf_length::gate (function *)
147 /* Run the pass iff -Warn-format-overflow or -Warn-format-truncation
148 is specified and either not optimizing and the pass is being invoked
149 early, or when optimizing and the pass is being invoked during
150 optimization (i.e., "late"). */
151 return ((warn_format_overflow > 0
152 || warn_format_trunc > 0
153 || flag_printf_return_value)
154 && (optimize > 0) == fold_return_value);
157 /* The minimum, maximum, likely, and unlikely maximum number of bytes
158 of output either a formatting function or an individual directive
159 can result in. */
161 struct result_range
163 /* The absolute minimum number of bytes. The result of a successful
164 conversion is guaranteed to be no less than this. (An erroneous
165 conversion can be indicated by MIN > HOST_WIDE_INT_MAX.) */
166 unsigned HOST_WIDE_INT min;
167 /* The likely maximum result that is used in diagnostics. In most
168 cases MAX is the same as the worst case UNLIKELY result. */
169 unsigned HOST_WIDE_INT max;
170 /* The likely result used to trigger diagnostics. For conversions
171 that result in a range of bytes [MIN, MAX], LIKELY is somewhere
172 in that range. */
173 unsigned HOST_WIDE_INT likely;
174 /* In rare cases (e.g., for nultibyte characters) UNLIKELY gives
175 the worst cases maximum result of a directive. In most cases
176 UNLIKELY == MAX. UNLIKELY is used to control the return value
177 optimization but not in diagnostics. */
178 unsigned HOST_WIDE_INT unlikely;
181 /* The result of a call to a formatted function. */
183 struct format_result
185 /* Range of characters written by the formatted function.
186 Setting the minimum to HOST_WIDE_INT_MAX disables all
187 length tracking for the remainder of the format string. */
188 result_range range;
190 /* True when the range above is obtained from known values of
191 directive arguments, or bounds on the amount of output such
192 as width and precision, and not the result of heuristics that
193 depend on warning levels. It's used to issue stricter diagnostics
194 in cases where strings of unknown lengths are bounded by the arrays
195 they are determined to refer to. KNOWNRANGE must not be used for
196 the return value optimization. */
197 bool knownrange;
199 /* True if no individual directive resulted in more than 4095 bytes
200 of output (the total NUMBER_CHARS_{MIN,MAX} might be greater).
201 Implementations are not required to handle directives that produce
202 more than 4K bytes (leading to undefined behavior) and so when one
203 is found it disables the return value optimization. */
204 bool under4k;
206 /* True when a floating point directive has been seen in the format
207 string. */
208 bool floating;
210 /* True when an intermediate result has caused a warning. Used to
211 avoid issuing duplicate warnings while finishing the processing
212 of a call. WARNED also disables the return value optimization. */
213 bool warned;
215 /* Preincrement the number of output characters by 1. */
216 format_result& operator++ ()
218 return *this += 1;
221 /* Postincrement the number of output characters by 1. */
222 format_result operator++ (int)
224 format_result prev (*this);
225 *this += 1;
226 return prev;
229 /* Increment the number of output characters by N. */
230 format_result& operator+= (unsigned HOST_WIDE_INT);
233 format_result&
234 format_result::operator+= (unsigned HOST_WIDE_INT n)
236 gcc_assert (n < HOST_WIDE_INT_MAX);
238 if (range.min < HOST_WIDE_INT_MAX)
239 range.min += n;
241 if (range.max < HOST_WIDE_INT_MAX)
242 range.max += n;
244 if (range.likely < HOST_WIDE_INT_MAX)
245 range.likely += n;
247 if (range.unlikely < HOST_WIDE_INT_MAX)
248 range.unlikely += n;
250 return *this;
253 /* Return the value of INT_MIN for the target. */
255 static inline HOST_WIDE_INT
256 target_int_min ()
258 return tree_to_shwi (TYPE_MIN_VALUE (integer_type_node));
261 /* Return the value of INT_MAX for the target. */
263 static inline unsigned HOST_WIDE_INT
264 target_int_max ()
266 return tree_to_uhwi (TYPE_MAX_VALUE (integer_type_node));
269 /* Return the value of SIZE_MAX for the target. */
271 static inline unsigned HOST_WIDE_INT
272 target_size_max ()
274 return tree_to_uhwi (TYPE_MAX_VALUE (size_type_node));
277 /* A straightforward mapping from the execution character set to the host
278 character set indexed by execution character. */
280 static char target_to_host_charmap[256];
282 /* Initialize a mapping from the execution character set to the host
283 character set. */
285 static bool
286 init_target_to_host_charmap ()
288 /* If the percent sign is non-zero the mapping has already been
289 initialized. */
290 if (target_to_host_charmap['%'])
291 return true;
293 /* Initialize the target_percent character (done elsewhere). */
294 if (!init_target_chars ())
295 return false;
297 /* The subset of the source character set used by printf conversion
298 specifications (strictly speaking, not all letters are used but
299 they are included here for the sake of simplicity). The dollar
300 sign must be included even though it's not in the basic source
301 character set. */
302 const char srcset[] = " 0123456789!\"#%&'()*+,-./:;<=>?[\\]^_{|}~$"
303 "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz";
305 /* Set the mapping for all characters to some ordinary value (i,e.,
306 not none used in printf conversion specifications) and overwrite
307 those that are used by conversion specifications with their
308 corresponding values. */
309 memset (target_to_host_charmap + 1, '?', sizeof target_to_host_charmap - 1);
311 /* Are the two sets of characters the same? */
312 bool all_same_p = true;
314 for (const char *pc = srcset; *pc; ++pc)
316 /* Slice off the high end bits in case target characters are
317 signed. All values are expected to be non-nul, otherwise
318 there's a problem. */
319 if (unsigned char tc = lang_hooks.to_target_charset (*pc))
321 target_to_host_charmap[tc] = *pc;
322 if (tc != *pc)
323 all_same_p = false;
325 else
326 return false;
330 /* Set the first element to a non-zero value if the mapping
331 is 1-to-1, otherwise leave it clear (NUL is assumed to be
332 the same in both character sets). */
333 target_to_host_charmap[0] = all_same_p;
335 return true;
338 /* Return the host source character corresponding to the character
339 CH in the execution character set if one exists, or some innocuous
340 (non-special, non-nul) source character otherwise. */
342 static inline unsigned char
343 target_to_host (unsigned char ch)
345 return target_to_host_charmap[ch];
348 /* Convert an initial substring of the string TARGSTR consisting of
349 characters in the execution character set into a string in the
350 source character set on the host and store up to HOSTSZ characters
351 in the buffer pointed to by HOSTR. Return HOSTR. */
353 static const char*
354 target_to_host (char *hostr, size_t hostsz, const char *targstr)
356 /* Make sure the buffer is reasonably big. */
357 gcc_assert (hostsz > 4);
359 /* The interesting subset of source and execution characters are
360 the same so no conversion is necessary. However, truncate
361 overlong strings just like the translated strings are. */
362 if (target_to_host_charmap['\0'] == 1)
364 strncpy (hostr, targstr, hostsz - 4);
365 if (strlen (targstr) >= hostsz)
366 strcpy (hostr + hostsz - 4, "...");
367 return hostr;
370 /* Convert the initial substring of TARGSTR to the corresponding
371 characters in the host set, appending "..." if TARGSTR is too
372 long to fit. Using the static buffer assumes the function is
373 not called in between sequence points (which it isn't). */
374 for (char *ph = hostr; ; ++targstr)
376 *ph++ = target_to_host (*targstr);
377 if (!*targstr)
378 break;
380 if (size_t (ph - hostr) == hostsz - 4)
382 *ph = '\0';
383 strcat (ph, "...");
384 break;
388 return hostr;
391 /* Convert the sequence of decimal digits in the execution character
392 starting at S to a long, just like strtol does. Return the result
393 and set *END to one past the last converted character. On range
394 error set ERANGE to the digit that caused it. */
396 static inline long
397 target_strtol10 (const char **ps, const char **erange)
399 unsigned HOST_WIDE_INT val = 0;
400 for ( ; ; ++*ps)
402 unsigned char c = target_to_host (**ps);
403 if (ISDIGIT (c))
405 c -= '0';
407 /* Check for overflow. */
408 if (val > (LONG_MAX - c) / 10LU)
410 val = LONG_MAX;
411 *erange = *ps;
413 /* Skip the remaining digits. */
415 c = target_to_host (*++*ps);
416 while (ISDIGIT (c));
417 break;
419 else
420 val = val * 10 + c;
422 else
423 break;
426 return val;
429 /* Return the constant initial value of DECL if available or DECL
430 otherwise. Same as the synonymous function in c/c-typeck.c. */
432 static tree
433 decl_constant_value (tree decl)
435 if (/* Don't change a variable array bound or initial value to a constant
436 in a place where a variable is invalid. Note that DECL_INITIAL
437 isn't valid for a PARM_DECL. */
438 current_function_decl != 0
439 && TREE_CODE (decl) != PARM_DECL
440 && !TREE_THIS_VOLATILE (decl)
441 && TREE_READONLY (decl)
442 && DECL_INITIAL (decl) != 0
443 && TREE_CODE (DECL_INITIAL (decl)) != ERROR_MARK
444 /* This is invalid if initial value is not constant.
445 If it has either a function call, a memory reference,
446 or a variable, then re-evaluating it could give different results. */
447 && TREE_CONSTANT (DECL_INITIAL (decl))
448 /* Check for cases where this is sub-optimal, even though valid. */
449 && TREE_CODE (DECL_INITIAL (decl)) != CONSTRUCTOR)
450 return DECL_INITIAL (decl);
451 return decl;
454 /* Given FORMAT, set *PLOC to the source location of the format string
455 and return the format string if it is known or null otherwise. */
457 static const char*
458 get_format_string (tree format, location_t *ploc)
460 if (VAR_P (format))
462 /* Pull out a constant value if the front end didn't. */
463 format = decl_constant_value (format);
464 STRIP_NOPS (format);
467 if (integer_zerop (format))
469 /* FIXME: Diagnose null format string if it hasn't been diagnosed
470 by -Wformat (the latter diagnoses only nul pointer constants,
471 this pass can do better). */
472 return NULL;
475 HOST_WIDE_INT offset = 0;
477 if (TREE_CODE (format) == POINTER_PLUS_EXPR)
479 tree arg0 = TREE_OPERAND (format, 0);
480 tree arg1 = TREE_OPERAND (format, 1);
481 STRIP_NOPS (arg0);
482 STRIP_NOPS (arg1);
484 if (TREE_CODE (arg1) != INTEGER_CST)
485 return NULL;
487 format = arg0;
489 /* POINTER_PLUS_EXPR offsets are to be interpreted signed. */
490 if (!cst_and_fits_in_hwi (arg1))
491 return NULL;
493 offset = int_cst_value (arg1);
496 if (TREE_CODE (format) != ADDR_EXPR)
497 return NULL;
499 *ploc = EXPR_LOC_OR_LOC (format, input_location);
501 format = TREE_OPERAND (format, 0);
503 if (TREE_CODE (format) == ARRAY_REF
504 && tree_fits_shwi_p (TREE_OPERAND (format, 1))
505 && (offset += tree_to_shwi (TREE_OPERAND (format, 1))) >= 0)
506 format = TREE_OPERAND (format, 0);
508 if (offset < 0)
509 return NULL;
511 tree array_init;
512 tree array_size = NULL_TREE;
514 if (VAR_P (format)
515 && TREE_CODE (TREE_TYPE (format)) == ARRAY_TYPE
516 && (array_init = decl_constant_value (format)) != format
517 && TREE_CODE (array_init) == STRING_CST)
519 /* Extract the string constant initializer. Note that this may
520 include a trailing NUL character that is not in the array (e.g.
521 const char a[3] = "foo";). */
522 array_size = DECL_SIZE_UNIT (format);
523 format = array_init;
526 if (TREE_CODE (format) != STRING_CST)
527 return NULL;
529 tree type = TREE_TYPE (format);
531 scalar_int_mode char_mode;
532 if (!is_int_mode (TYPE_MODE (TREE_TYPE (type)), &char_mode)
533 || GET_MODE_SIZE (char_mode) != 1)
535 /* Wide format string. */
536 return NULL;
539 const char *fmtstr = TREE_STRING_POINTER (format);
540 unsigned fmtlen = TREE_STRING_LENGTH (format);
542 if (array_size)
544 /* Variable length arrays can't be initialized. */
545 gcc_assert (TREE_CODE (array_size) == INTEGER_CST);
547 if (tree_fits_shwi_p (array_size))
549 HOST_WIDE_INT array_size_value = tree_to_shwi (array_size);
550 if (array_size_value > 0
551 && array_size_value == (int) array_size_value
552 && fmtlen > array_size_value)
553 fmtlen = array_size_value;
556 if (offset)
558 if (offset >= fmtlen)
559 return NULL;
561 fmtstr += offset;
562 fmtlen -= offset;
565 if (fmtlen < 1 || fmtstr[--fmtlen] != 0)
567 /* FIXME: Diagnose an unterminated format string if it hasn't been
568 diagnosed by -Wformat. Similarly to a null format pointer,
569 -Wformay diagnoses only nul pointer constants, this pass can
570 do better). */
571 return NULL;
574 return fmtstr;
577 /* The format_warning_at_substring function is not used here in a way
578 that makes using attribute format viable. Suppress the warning. */
580 #pragma GCC diagnostic push
581 #pragma GCC diagnostic ignored "-Wsuggest-attribute=format"
583 /* For convenience and brevity. */
585 static bool
586 (* const fmtwarn) (const substring_loc &, const source_range *,
587 const char *, int, const char *, ...)
588 = format_warning_at_substring;
590 /* Format length modifiers. */
592 enum format_lengths
594 FMT_LEN_none,
595 FMT_LEN_hh, // char argument
596 FMT_LEN_h, // short
597 FMT_LEN_l, // long
598 FMT_LEN_ll, // long long
599 FMT_LEN_L, // long double (and GNU long long)
600 FMT_LEN_z, // size_t
601 FMT_LEN_t, // ptrdiff_t
602 FMT_LEN_j // intmax_t
606 /* Description of the result of conversion either of a single directive
607 or the whole format string. */
609 struct fmtresult
611 /* Construct a FMTRESULT object with all counters initialized
612 to MIN. KNOWNRANGE is set when MIN is valid. */
613 fmtresult (unsigned HOST_WIDE_INT min = HOST_WIDE_INT_MAX)
614 : argmin (), argmax (),
615 knownrange (min < HOST_WIDE_INT_MAX),
616 nullp ()
618 range.min = min;
619 range.max = min;
620 range.likely = min;
621 range.unlikely = min;
624 /* Construct a FMTRESULT object with MIN, MAX, and LIKELY counters.
625 KNOWNRANGE is set when both MIN and MAX are valid. */
626 fmtresult (unsigned HOST_WIDE_INT min, unsigned HOST_WIDE_INT max,
627 unsigned HOST_WIDE_INT likely = HOST_WIDE_INT_MAX)
628 : argmin (), argmax (),
629 knownrange (min < HOST_WIDE_INT_MAX && max < HOST_WIDE_INT_MAX),
630 nullp ()
632 range.min = min;
633 range.max = max;
634 range.likely = max < likely ? min : likely;
635 range.unlikely = max;
638 /* Adjust result upward to reflect the RANGE of values the specified
639 width or precision is known to be in. */
640 fmtresult& adjust_for_width_or_precision (const HOST_WIDE_INT[2],
641 tree = NULL_TREE,
642 unsigned = 0, unsigned = 0);
644 /* Return the maximum number of decimal digits a value of TYPE
645 formats as on output. */
646 static unsigned type_max_digits (tree, int);
648 /* The range a directive's argument is in. */
649 tree argmin, argmax;
651 /* The minimum and maximum number of bytes that a directive
652 results in on output for an argument in the range above. */
653 result_range range;
655 /* True when the range above is obtained from a known value of
656 a directive's argument or its bounds and not the result of
657 heuristics that depend on warning levels. */
658 bool knownrange;
660 /* True when the argument is a null pointer. */
661 bool nullp;
664 /* Adjust result upward to reflect the range ADJUST of values the
665 specified width or precision is known to be in. When non-null,
666 TYPE denotes the type of the directive whose result is being
667 adjusted, BASE gives the base of the directive (octal, decimal,
668 or hex), and ADJ denotes the additional adjustment to the LIKELY
669 counter that may need to be added when ADJUST is a range. */
671 fmtresult&
672 fmtresult::adjust_for_width_or_precision (const HOST_WIDE_INT adjust[2],
673 tree type /* = NULL_TREE */,
674 unsigned base /* = 0 */,
675 unsigned adj /* = 0 */)
677 bool minadjusted = false;
679 /* Adjust the minimum and likely counters. */
680 if (adjust[0] >= 0)
682 if (range.min < (unsigned HOST_WIDE_INT)adjust[0])
684 range.min = adjust[0];
685 minadjusted = true;
688 /* Adjust the likely counter. */
689 if (range.likely < range.min)
690 range.likely = range.min;
692 else if (adjust[0] == target_int_min ()
693 && (unsigned HOST_WIDE_INT)adjust[1] == target_int_max ())
694 knownrange = false;
696 /* Adjust the maximum counter. */
697 if (adjust[1] > 0)
699 if (range.max < (unsigned HOST_WIDE_INT)adjust[1])
701 range.max = adjust[1];
703 /* Set KNOWNRANGE if both the minimum and maximum have been
704 adjusted. Otherwise leave it at what it was before. */
705 knownrange = minadjusted;
709 if (warn_level > 1 && type)
711 /* For large non-constant width or precision whose range spans
712 the maximum number of digits produced by the directive for
713 any argument, set the likely number of bytes to be at most
714 the number digits plus other adjustment determined by the
715 caller (one for sign or two for the hexadecimal "0x"
716 prefix). */
717 unsigned dirdigs = type_max_digits (type, base);
718 if (adjust[0] < dirdigs && dirdigs < adjust[1]
719 && range.likely < dirdigs)
720 range.likely = dirdigs + adj;
722 else if (range.likely < (range.min ? range.min : 1))
724 /* Conservatively, set LIKELY to at least MIN but no less than
725 1 unless MAX is zero. */
726 range.likely = (range.min
727 ? range.min
728 : range.max && (range.max < HOST_WIDE_INT_MAX
729 || warn_level > 1) ? 1 : 0);
732 /* Finally adjust the unlikely counter to be at least as large as
733 the maximum. */
734 if (range.unlikely < range.max)
735 range.unlikely = range.max;
737 return *this;
740 /* Return the maximum number of digits a value of TYPE formats in
741 BASE on output, not counting base prefix . */
743 unsigned
744 fmtresult::type_max_digits (tree type, int base)
746 unsigned prec = TYPE_PRECISION (type);
747 if (base == 8)
748 return (prec + 2) / 3;
750 if (base == 16)
751 return prec / 4;
753 /* Decimal approximation: yields 3, 5, 10, and 20 for precision
754 of 8, 16, 32, and 64 bits. */
755 return prec * 301 / 1000 + 1;
758 static bool
759 get_int_range (tree, HOST_WIDE_INT *, HOST_WIDE_INT *, bool, HOST_WIDE_INT);
761 /* Description of a format directive. A directive is either a plain
762 string or a conversion specification that starts with '%'. */
764 struct directive
766 /* The 1-based directive number (for debugging). */
767 unsigned dirno;
769 /* The first character of the directive and its length. */
770 const char *beg;
771 size_t len;
773 /* A bitmap of flags, one for each character. */
774 unsigned flags[256 / sizeof (int)];
776 /* The range of values of the specified width, or -1 if not specified. */
777 HOST_WIDE_INT width[2];
778 /* The range of values of the specified precision, or -1 if not
779 specified. */
780 HOST_WIDE_INT prec[2];
782 /* Length modifier. */
783 format_lengths modifier;
785 /* Format specifier character. */
786 char specifier;
788 /* The argument of the directive or null when the directive doesn't
789 take one or when none is available (such as for vararg functions). */
790 tree arg;
792 /* Format conversion function that given a directive and an argument
793 returns the formatting result. */
794 fmtresult (*fmtfunc) (const directive &, tree);
796 /* Return True when a the format flag CHR has been used. */
797 bool get_flag (char chr) const
799 unsigned char c = chr & 0xff;
800 return (flags[c / (CHAR_BIT * sizeof *flags)]
801 & (1U << (c % (CHAR_BIT * sizeof *flags))));
804 /* Make a record of the format flag CHR having been used. */
805 void set_flag (char chr)
807 unsigned char c = chr & 0xff;
808 flags[c / (CHAR_BIT * sizeof *flags)]
809 |= (1U << (c % (CHAR_BIT * sizeof *flags)));
812 /* Reset the format flag CHR. */
813 void clear_flag (char chr)
815 unsigned char c = chr & 0xff;
816 flags[c / (CHAR_BIT * sizeof *flags)]
817 &= ~(1U << (c % (CHAR_BIT * sizeof *flags)));
820 /* Set both bounds of the width range to VAL. */
821 void set_width (HOST_WIDE_INT val)
823 width[0] = width[1] = val;
826 /* Set the width range according to ARG, with both bounds being
827 no less than 0. For a constant ARG set both bounds to its value
828 or 0, whichever is greater. For a non-constant ARG in some range
829 set width to its range adjusting each bound to -1 if it's less.
830 For an indeterminate ARG set width to [0, INT_MAX]. */
831 void set_width (tree arg)
833 get_int_range (arg, width, width + 1, true, 0);
836 /* Set both bounds of the precision range to VAL. */
837 void set_precision (HOST_WIDE_INT val)
839 prec[0] = prec[1] = val;
842 /* Set the precision range according to ARG, with both bounds being
843 no less than -1. For a constant ARG set both bounds to its value
844 or -1 whichever is greater. For a non-constant ARG in some range
845 set precision to its range adjusting each bound to -1 if it's less.
846 For an indeterminate ARG set precision to [-1, INT_MAX]. */
847 void set_precision (tree arg)
849 get_int_range (arg, prec, prec + 1, false, -1);
852 /* Return true if both width and precision are known to be
853 either constant or in some range, false otherwise. */
854 bool known_width_and_precision () const
856 return ((width[1] < 0
857 || (unsigned HOST_WIDE_INT)width[1] <= target_int_max ())
858 && (prec[1] < 0
859 || (unsigned HOST_WIDE_INT)prec[1] < target_int_max ()));
863 /* Return the logarithm of X in BASE. */
865 static int
866 ilog (unsigned HOST_WIDE_INT x, int base)
868 int res = 0;
871 ++res;
872 x /= base;
873 } while (x);
874 return res;
877 /* Return the number of bytes resulting from converting into a string
878 the INTEGER_CST tree node X in BASE with a minimum of PREC digits.
879 PLUS indicates whether 1 for a plus sign should be added for positive
880 numbers, and PREFIX whether the length of an octal ('O') or hexadecimal
881 ('0x') prefix should be added for nonzero numbers. Return -1 if X cannot
882 be represented. */
884 static HOST_WIDE_INT
885 tree_digits (tree x, int base, HOST_WIDE_INT prec, bool plus, bool prefix)
887 unsigned HOST_WIDE_INT absval;
889 HOST_WIDE_INT res;
891 if (TYPE_UNSIGNED (TREE_TYPE (x)))
893 if (tree_fits_uhwi_p (x))
895 absval = tree_to_uhwi (x);
896 res = plus;
898 else
899 return -1;
901 else
903 if (tree_fits_shwi_p (x))
905 HOST_WIDE_INT i = tree_to_shwi (x);
906 if (HOST_WIDE_INT_MIN == i)
908 /* Avoid undefined behavior due to negating a minimum. */
909 absval = HOST_WIDE_INT_MAX;
910 res = 1;
912 else if (i < 0)
914 absval = -i;
915 res = 1;
917 else
919 absval = i;
920 res = plus;
923 else
924 return -1;
927 int ndigs = ilog (absval, base);
929 res += prec < ndigs ? ndigs : prec;
931 /* Adjust a non-zero value for the base prefix, either hexadecimal,
932 or, unless precision has resulted in a leading zero, also octal. */
933 if (prefix && absval && (base == 16 || prec <= ndigs))
935 if (base == 8)
936 res += 1;
937 else if (base == 16)
938 res += 2;
941 return res;
944 /* Given the formatting result described by RES and NAVAIL, the number
945 of available in the destination, return the range of bytes remaining
946 in the destination. */
948 static inline result_range
949 bytes_remaining (unsigned HOST_WIDE_INT navail, const format_result &res)
951 result_range range;
953 if (HOST_WIDE_INT_MAX <= navail)
955 range.min = range.max = range.likely = range.unlikely = navail;
956 return range;
959 /* The lower bound of the available range is the available size
960 minus the maximum output size, and the upper bound is the size
961 minus the minimum. */
962 range.max = res.range.min < navail ? navail - res.range.min : 0;
964 range.likely = res.range.likely < navail ? navail - res.range.likely : 0;
966 if (res.range.max < HOST_WIDE_INT_MAX)
967 range.min = res.range.max < navail ? navail - res.range.max : 0;
968 else
969 range.min = range.likely;
971 range.unlikely = (res.range.unlikely < navail
972 ? navail - res.range.unlikely : 0);
974 return range;
977 /* Description of a call to a formatted function. */
979 struct pass_sprintf_length::call_info
981 /* Function call statement. */
982 gimple *callstmt;
984 /* Function called. */
985 tree func;
987 /* Called built-in function code. */
988 built_in_function fncode;
990 /* Format argument and format string extracted from it. */
991 tree format;
992 const char *fmtstr;
994 /* The location of the format argument. */
995 location_t fmtloc;
997 /* The destination object size for __builtin___xxx_chk functions
998 typically determined by __builtin_object_size, or -1 if unknown. */
999 unsigned HOST_WIDE_INT objsize;
1001 /* Number of the first variable argument. */
1002 unsigned HOST_WIDE_INT argidx;
1004 /* True for functions like snprintf that specify the size of
1005 the destination, false for others like sprintf that don't. */
1006 bool bounded;
1008 /* True for bounded functions like snprintf that specify a zero-size
1009 buffer as a request to compute the size of output without actually
1010 writing any. NOWRITE is cleared in response to the %n directive
1011 which has side-effects similar to writing output. */
1012 bool nowrite;
1014 /* Return true if the called function's return value is used. */
1015 bool retval_used () const
1017 return gimple_get_lhs (callstmt);
1020 /* Return the warning option corresponding to the called function. */
1021 int warnopt () const
1023 return bounded ? OPT_Wformat_truncation_ : OPT_Wformat_overflow_;
1027 /* Return the result of formatting a no-op directive (such as '%n'). */
1029 static fmtresult
1030 format_none (const directive &, tree)
1032 fmtresult res (0);
1033 return res;
1036 /* Return the result of formatting the '%%' directive. */
1038 static fmtresult
1039 format_percent (const directive &, tree)
1041 fmtresult res (1);
1042 return res;
1046 /* Compute intmax_type_node and uintmax_type_node similarly to how
1047 tree.c builds size_type_node. */
1049 static void
1050 build_intmax_type_nodes (tree *pintmax, tree *puintmax)
1052 if (strcmp (UINTMAX_TYPE, "unsigned int") == 0)
1054 *pintmax = integer_type_node;
1055 *puintmax = unsigned_type_node;
1057 else if (strcmp (UINTMAX_TYPE, "long unsigned int") == 0)
1059 *pintmax = long_integer_type_node;
1060 *puintmax = long_unsigned_type_node;
1062 else if (strcmp (UINTMAX_TYPE, "long long unsigned int") == 0)
1064 *pintmax = long_long_integer_type_node;
1065 *puintmax = long_long_unsigned_type_node;
1067 else
1069 for (int i = 0; i < NUM_INT_N_ENTS; i++)
1070 if (int_n_enabled_p[i])
1072 char name[50];
1073 sprintf (name, "__int%d unsigned", int_n_data[i].bitsize);
1075 if (strcmp (name, UINTMAX_TYPE) == 0)
1077 *pintmax = int_n_trees[i].signed_type;
1078 *puintmax = int_n_trees[i].unsigned_type;
1079 return;
1082 gcc_unreachable ();
1086 /* Determine the range [*PMIN, *PMAX] that the expression ARG is
1087 in and that is representable in type int.
1088 Return true when the range is a subrange of that of int.
1089 When ARG is null it is as if it had the full range of int.
1090 When ABSOLUTE is true the range reflects the absolute value of
1091 the argument. When ABSOLUTE is false, negative bounds of
1092 the determined range are replaced with NEGBOUND. */
1094 static bool
1095 get_int_range (tree arg, HOST_WIDE_INT *pmin, HOST_WIDE_INT *pmax,
1096 bool absolute, HOST_WIDE_INT negbound)
1098 /* The type of the result. */
1099 const_tree type = integer_type_node;
1101 bool knownrange = false;
1103 if (!arg)
1105 *pmin = tree_to_shwi (TYPE_MIN_VALUE (type));
1106 *pmax = tree_to_shwi (TYPE_MAX_VALUE (type));
1108 else if (TREE_CODE (arg) == INTEGER_CST
1109 && TYPE_PRECISION (TREE_TYPE (arg)) <= TYPE_PRECISION (type))
1111 /* For a constant argument return its value adjusted as specified
1112 by NEGATIVE and NEGBOUND and return true to indicate that the
1113 result is known. */
1114 *pmin = tree_fits_shwi_p (arg) ? tree_to_shwi (arg) : tree_to_uhwi (arg);
1115 *pmax = *pmin;
1116 knownrange = true;
1118 else
1120 /* True if the argument's range cannot be determined. */
1121 bool unknown = true;
1123 tree argtype = TREE_TYPE (arg);
1125 /* Ignore invalid arguments with greater precision that that
1126 of the expected type (e.g., in sprintf("%*i", 12LL, i)).
1127 They will have been detected and diagnosed by -Wformat and
1128 so it's not important to complicate this code to try to deal
1129 with them again. */
1130 if (TREE_CODE (arg) == SSA_NAME
1131 && INTEGRAL_TYPE_P (argtype)
1132 && TYPE_PRECISION (argtype) <= TYPE_PRECISION (type))
1134 /* Try to determine the range of values of the integer argument. */
1135 wide_int min, max;
1136 enum value_range_type range_type = get_range_info (arg, &min, &max);
1137 if (range_type == VR_RANGE)
1139 HOST_WIDE_INT type_min
1140 = (TYPE_UNSIGNED (argtype)
1141 ? tree_to_uhwi (TYPE_MIN_VALUE (argtype))
1142 : tree_to_shwi (TYPE_MIN_VALUE (argtype)));
1144 HOST_WIDE_INT type_max = tree_to_uhwi (TYPE_MAX_VALUE (argtype));
1146 *pmin = min.to_shwi ();
1147 *pmax = max.to_shwi ();
1149 if (*pmin < *pmax)
1151 /* Return true if the adjusted range is a subrange of
1152 the full range of the argument's type. *PMAX may
1153 be less than *PMIN when the argument is unsigned
1154 and its upper bound is in excess of TYPE_MAX. In
1155 that (invalid) case disregard the range and use that
1156 of the expected type instead. */
1157 knownrange = type_min < *pmin || *pmax < type_max;
1159 unknown = false;
1164 /* Handle an argument with an unknown range as if none had been
1165 provided. */
1166 if (unknown)
1167 return get_int_range (NULL_TREE, pmin, pmax, absolute, negbound);
1170 /* Adjust each bound as specified by ABSOLUTE and NEGBOUND. */
1171 if (absolute)
1173 if (*pmin < 0)
1175 if (*pmin == *pmax)
1176 *pmin = *pmax = -*pmin;
1177 else
1179 /* Make sure signed overlow is avoided. */
1180 gcc_assert (*pmin != HOST_WIDE_INT_MIN);
1182 HOST_WIDE_INT tmp = -*pmin;
1183 *pmin = 0;
1184 if (*pmax < tmp)
1185 *pmax = tmp;
1189 else if (*pmin < negbound)
1190 *pmin = negbound;
1192 return knownrange;
1195 /* With the range [*ARGMIN, *ARGMAX] of an integer directive's actual
1196 argument, due to the conversion from either *ARGMIN or *ARGMAX to
1197 the type of the directive's formal argument it's possible for both
1198 to result in the same number of bytes or a range of bytes that's
1199 less than the number of bytes that would result from formatting
1200 some other value in the range [*ARGMIN, *ARGMAX]. This can be
1201 determined by checking for the actual argument being in the range
1202 of the type of the directive. If it isn't it must be assumed to
1203 take on the full range of the directive's type.
1204 Return true when the range has been adjusted to the full range
1205 of DIRTYPE, and false otherwise. */
1207 static bool
1208 adjust_range_for_overflow (tree dirtype, tree *argmin, tree *argmax)
1210 tree argtype = TREE_TYPE (*argmin);
1211 unsigned argprec = TYPE_PRECISION (argtype);
1212 unsigned dirprec = TYPE_PRECISION (dirtype);
1214 /* If the actual argument and the directive's argument have the same
1215 precision and sign there can be no overflow and so there is nothing
1216 to adjust. */
1217 if (argprec == dirprec && TYPE_SIGN (argtype) == TYPE_SIGN (dirtype))
1218 return false;
1220 /* The logic below was inspired/lifted from the CONVERT_EXPR_CODE_P
1221 branch in the extract_range_from_unary_expr function in tree-vrp.c. */
1223 if (TREE_CODE (*argmin) == INTEGER_CST
1224 && TREE_CODE (*argmax) == INTEGER_CST
1225 && (dirprec >= argprec
1226 || integer_zerop (int_const_binop (RSHIFT_EXPR,
1227 int_const_binop (MINUS_EXPR,
1228 *argmax,
1229 *argmin),
1230 size_int (dirprec)))))
1232 *argmin = force_fit_type (dirtype, wi::to_widest (*argmin), 0, false);
1233 *argmax = force_fit_type (dirtype, wi::to_widest (*argmax), 0, false);
1235 /* If *ARGMIN is still less than *ARGMAX the conversion above
1236 is safe. Otherwise, it has overflowed and would be unsafe. */
1237 if (tree_int_cst_le (*argmin, *argmax))
1238 return false;
1241 *argmin = TYPE_MIN_VALUE (dirtype);
1242 *argmax = TYPE_MAX_VALUE (dirtype);
1243 return true;
1246 /* Return a range representing the minimum and maximum number of bytes
1247 that the format directive DIR will output for any argument given
1248 the WIDTH and PRECISION (extracted from DIR). This function is
1249 used when the directive argument or its value isn't known. */
1251 static fmtresult
1252 format_integer (const directive &dir, tree arg)
1254 tree intmax_type_node;
1255 tree uintmax_type_node;
1257 /* Base to format the number in. */
1258 int base;
1260 /* True when a conversion is preceded by a prefix indicating the base
1261 of the argument (octal or hexadecimal). */
1262 bool maybebase = dir.get_flag ('#');
1264 /* True when a signed conversion is preceded by a sign or space. */
1265 bool maybesign = false;
1267 /* True for signed conversions (i.e., 'd' and 'i'). */
1268 bool sign = false;
1270 switch (dir.specifier)
1272 case 'd':
1273 case 'i':
1274 /* Space and '+' are only meaningful for signed conversions. */
1275 maybesign = dir.get_flag (' ') | dir.get_flag ('+');
1276 sign = true;
1277 base = 10;
1278 break;
1279 case 'u':
1280 base = 10;
1281 break;
1282 case 'o':
1283 base = 8;
1284 break;
1285 case 'X':
1286 case 'x':
1287 base = 16;
1288 break;
1289 default:
1290 gcc_unreachable ();
1293 /* The type of the "formal" argument expected by the directive. */
1294 tree dirtype = NULL_TREE;
1296 /* Determine the expected type of the argument from the length
1297 modifier. */
1298 switch (dir.modifier)
1300 case FMT_LEN_none:
1301 if (dir.specifier == 'p')
1302 dirtype = ptr_type_node;
1303 else
1304 dirtype = sign ? integer_type_node : unsigned_type_node;
1305 break;
1307 case FMT_LEN_h:
1308 dirtype = sign ? short_integer_type_node : short_unsigned_type_node;
1309 break;
1311 case FMT_LEN_hh:
1312 dirtype = sign ? signed_char_type_node : unsigned_char_type_node;
1313 break;
1315 case FMT_LEN_l:
1316 dirtype = sign ? long_integer_type_node : long_unsigned_type_node;
1317 break;
1319 case FMT_LEN_L:
1320 case FMT_LEN_ll:
1321 dirtype = (sign
1322 ? long_long_integer_type_node
1323 : long_long_unsigned_type_node);
1324 break;
1326 case FMT_LEN_z:
1327 dirtype = signed_or_unsigned_type_for (!sign, size_type_node);
1328 break;
1330 case FMT_LEN_t:
1331 dirtype = signed_or_unsigned_type_for (!sign, ptrdiff_type_node);
1332 break;
1334 case FMT_LEN_j:
1335 build_intmax_type_nodes (&intmax_type_node, &uintmax_type_node);
1336 dirtype = sign ? intmax_type_node : uintmax_type_node;
1337 break;
1339 default:
1340 return fmtresult ();
1343 /* The type of the argument to the directive, either deduced from
1344 the actual non-constant argument if one is known, or from
1345 the directive itself when none has been provided because it's
1346 a va_list. */
1347 tree argtype = NULL_TREE;
1349 if (!arg)
1351 /* When the argument has not been provided, use the type of
1352 the directive's argument as an approximation. This will
1353 result in false positives for directives like %i with
1354 arguments with smaller precision (such as short or char). */
1355 argtype = dirtype;
1357 else if (TREE_CODE (arg) == INTEGER_CST)
1359 /* When a constant argument has been provided use its value
1360 rather than type to determine the length of the output. */
1361 fmtresult res;
1363 if ((dir.prec[0] <= 0 && dir.prec[1] >= 0) && integer_zerop (arg))
1365 /* As a special case, a precision of zero with a zero argument
1366 results in zero bytes except in base 8 when the '#' flag is
1367 specified, and for signed conversions in base 8 and 10 when
1368 either the space or '+' flag has been specified and it results
1369 in just one byte (with width having the normal effect). This
1370 must extend to the case of a specified precision with
1371 an unknown value because it can be zero. */
1372 res.range.min = ((base == 8 && dir.get_flag ('#')) || maybesign);
1373 if (res.range.min == 0 && dir.prec[0] != dir.prec[1])
1375 res.range.max = 1;
1376 res.range.likely = 1;
1378 else
1380 res.range.max = res.range.min;
1381 res.range.likely = res.range.min;
1384 else
1386 /* Convert the argument to the type of the directive. */
1387 arg = fold_convert (dirtype, arg);
1389 res.range.min = tree_digits (arg, base, dir.prec[0],
1390 maybesign, maybebase);
1391 if (dir.prec[0] == dir.prec[1])
1392 res.range.max = res.range.min;
1393 else
1394 res.range.max = tree_digits (arg, base, dir.prec[1],
1395 maybesign, maybebase);
1396 res.range.likely = res.range.min;
1397 res.knownrange = true;
1400 res.range.unlikely = res.range.max;
1402 /* Bump up the counters if WIDTH is greater than LEN. */
1403 res.adjust_for_width_or_precision (dir.width, dirtype, base,
1404 (sign | maybebase) + (base == 16));
1405 /* Bump up the counters again if PRECision is greater still. */
1406 res.adjust_for_width_or_precision (dir.prec, dirtype, base,
1407 (sign | maybebase) + (base == 16));
1409 return res;
1411 else if (INTEGRAL_TYPE_P (TREE_TYPE (arg))
1412 || TREE_CODE (TREE_TYPE (arg)) == POINTER_TYPE)
1413 /* Determine the type of the provided non-constant argument. */
1414 argtype = TREE_TYPE (arg);
1415 else
1416 /* Don't bother with invalid arguments since they likely would
1417 have already been diagnosed, and disable any further checking
1418 of the format string by returning [-1, -1]. */
1419 return fmtresult ();
1421 fmtresult res;
1423 /* Using either the range the non-constant argument is in, or its
1424 type (either "formal" or actual), create a range of values that
1425 constrain the length of output given the warning level. */
1426 tree argmin = NULL_TREE;
1427 tree argmax = NULL_TREE;
1429 if (arg
1430 && TREE_CODE (arg) == SSA_NAME
1431 && INTEGRAL_TYPE_P (argtype))
1433 /* Try to determine the range of values of the integer argument
1434 (range information is not available for pointers). */
1435 wide_int min, max;
1436 enum value_range_type range_type = get_range_info (arg, &min, &max);
1437 if (range_type == VR_RANGE)
1439 argmin = wide_int_to_tree (argtype, min);
1440 argmax = wide_int_to_tree (argtype, max);
1442 /* Set KNOWNRANGE if the argument is in a known subrange
1443 of the directive's type and neither width nor precision
1444 is unknown. (KNOWNRANGE may be reset below). */
1445 res.knownrange
1446 = ((!tree_int_cst_equal (TYPE_MIN_VALUE (dirtype), argmin)
1447 || !tree_int_cst_equal (TYPE_MAX_VALUE (dirtype), argmax))
1448 && dir.known_width_and_precision ());
1450 res.argmin = argmin;
1451 res.argmax = argmax;
1453 else if (range_type == VR_ANTI_RANGE)
1455 /* Handle anti-ranges if/when bug 71690 is resolved. */
1457 else if (range_type == VR_VARYING)
1459 /* The argument here may be the result of promoting the actual
1460 argument to int. Try to determine the type of the actual
1461 argument before promotion and narrow down its range that
1462 way. */
1463 gimple *def = SSA_NAME_DEF_STMT (arg);
1464 if (is_gimple_assign (def))
1466 tree_code code = gimple_assign_rhs_code (def);
1467 if (code == INTEGER_CST)
1469 arg = gimple_assign_rhs1 (def);
1470 return format_integer (dir, arg);
1473 if (code == NOP_EXPR)
1475 tree type = TREE_TYPE (gimple_assign_rhs1 (def));
1476 if (INTEGRAL_TYPE_P (type)
1477 || TREE_CODE (type) == POINTER_TYPE)
1478 argtype = type;
1484 if (!argmin)
1486 if (TREE_CODE (argtype) == POINTER_TYPE)
1488 argmin = build_int_cst (pointer_sized_int_node, 0);
1489 argmax = build_all_ones_cst (pointer_sized_int_node);
1491 else
1493 argmin = TYPE_MIN_VALUE (argtype);
1494 argmax = TYPE_MAX_VALUE (argtype);
1498 /* Clear KNOWNRANGE if the range has been adjusted to the maximum
1499 of the directive. If it has been cleared then since ARGMIN and/or
1500 ARGMAX have been adjusted also adjust the corresponding ARGMIN and
1501 ARGMAX in the result to include in diagnostics. */
1502 if (adjust_range_for_overflow (dirtype, &argmin, &argmax))
1504 res.knownrange = false;
1505 res.argmin = argmin;
1506 res.argmax = argmax;
1509 /* Recursively compute the minimum and maximum from the known range. */
1510 if (TYPE_UNSIGNED (dirtype) || tree_int_cst_sgn (argmin) >= 0)
1512 /* For unsigned conversions/directives or signed when
1513 the minimum is positive, use the minimum and maximum to compute
1514 the shortest and longest output, respectively. */
1515 res.range.min = format_integer (dir, argmin).range.min;
1516 res.range.max = format_integer (dir, argmax).range.max;
1518 else if (tree_int_cst_sgn (argmax) < 0)
1520 /* For signed conversions/directives if maximum is negative,
1521 use the minimum as the longest output and maximum as the
1522 shortest output. */
1523 res.range.min = format_integer (dir, argmax).range.min;
1524 res.range.max = format_integer (dir, argmin).range.max;
1526 else
1528 /* Otherwise, 0 is inside of the range and minimum negative. Use 0
1529 as the shortest output and for the longest output compute the
1530 length of the output of both minimum and maximum and pick the
1531 longer. */
1532 unsigned HOST_WIDE_INT max1 = format_integer (dir, argmin).range.max;
1533 unsigned HOST_WIDE_INT max2 = format_integer (dir, argmax).range.max;
1534 res.range.min = format_integer (dir, integer_zero_node).range.min;
1535 res.range.max = MAX (max1, max2);
1538 /* If the range is known, use the maximum as the likely length. */
1539 if (res.knownrange)
1540 res.range.likely = res.range.max;
1541 else
1543 /* Otherwise, use the minimum. Except for the case where for %#x or
1544 %#o the minimum is just for a single value in the range (0) and
1545 for all other values it is something longer, like 0x1 or 01.
1546 Use the length for value 1 in that case instead as the likely
1547 length. */
1548 res.range.likely = res.range.min;
1549 if (maybebase
1550 && base != 10
1551 && (tree_int_cst_sgn (argmin) < 0 || tree_int_cst_sgn (argmax) > 0))
1553 if (res.range.min == 1)
1554 res.range.likely += base == 8 ? 1 : 2;
1555 else if (res.range.min == 2
1556 && base == 16
1557 && (dir.width[0] == 2 || dir.prec[0] == 2))
1558 ++res.range.likely;
1562 res.range.unlikely = res.range.max;
1563 res.adjust_for_width_or_precision (dir.width, dirtype, base,
1564 (sign | maybebase) + (base == 16));
1565 res.adjust_for_width_or_precision (dir.prec, dirtype, base,
1566 (sign | maybebase) + (base == 16));
1568 return res;
1571 /* Return the number of bytes that a format directive consisting of FLAGS,
1572 PRECision, format SPECification, and MPFR rounding specifier RNDSPEC,
1573 would result for argument X under ideal conditions (i.e., if PREC
1574 weren't excessive). MPFR 3.1 allocates large amounts of memory for
1575 values of PREC with large magnitude and can fail (see MPFR bug #21056).
1576 This function works around those problems. */
1578 static unsigned HOST_WIDE_INT
1579 get_mpfr_format_length (mpfr_ptr x, const char *flags, HOST_WIDE_INT prec,
1580 char spec, char rndspec)
1582 char fmtstr[40];
1584 HOST_WIDE_INT len = strlen (flags);
1586 fmtstr[0] = '%';
1587 memcpy (fmtstr + 1, flags, len);
1588 memcpy (fmtstr + 1 + len, ".*R", 3);
1589 fmtstr[len + 4] = rndspec;
1590 fmtstr[len + 5] = spec;
1591 fmtstr[len + 6] = '\0';
1593 spec = TOUPPER (spec);
1594 if (spec == 'E' || spec == 'F')
1596 /* For %e, specify the precision explicitly since mpfr_sprintf
1597 does its own thing just to be different (see MPFR bug 21088). */
1598 if (prec < 0)
1599 prec = 6;
1601 else
1603 /* Avoid passing negative precisions with larger magnitude to MPFR
1604 to avoid exposing its bugs. (A negative precision is supposed
1605 to be ignored.) */
1606 if (prec < 0)
1607 prec = -1;
1610 HOST_WIDE_INT p = prec;
1612 if (spec == 'G' && !strchr (flags, '#'))
1614 /* For G/g without the pound flag, precision gives the maximum number
1615 of significant digits which is bounded by LDBL_MAX_10_EXP, or, for
1616 a 128 bit IEEE extended precision, 4932. Using twice as much here
1617 should be more than sufficient for any real format. */
1618 if ((IEEE_MAX_10_EXP * 2) < prec)
1619 prec = IEEE_MAX_10_EXP * 2;
1620 p = prec;
1622 else
1624 /* Cap precision arbitrarily at 1KB and add the difference
1625 (if any) to the MPFR result. */
1626 if (prec > 1024)
1627 p = 1024;
1630 len = mpfr_snprintf (NULL, 0, fmtstr, (int)p, x);
1632 /* Handle the unlikely (impossible?) error by returning more than
1633 the maximum dictated by the function's return type. */
1634 if (len < 0)
1635 return target_dir_max () + 1;
1637 /* Adjust the return value by the difference. */
1638 if (p < prec)
1639 len += prec - p;
1641 return len;
1644 /* Return the number of bytes to format using the format specifier
1645 SPEC and the precision PREC the largest value in the real floating
1646 TYPE. */
1648 static unsigned HOST_WIDE_INT
1649 format_floating_max (tree type, char spec, HOST_WIDE_INT prec)
1651 machine_mode mode = TYPE_MODE (type);
1653 /* IBM Extended mode. */
1654 if (MODE_COMPOSITE_P (mode))
1655 mode = DFmode;
1657 /* Get the real type format desription for the target. */
1658 const real_format *rfmt = REAL_MODE_FORMAT (mode);
1659 REAL_VALUE_TYPE rv;
1661 real_maxval (&rv, 0, mode);
1663 /* Convert the GCC real value representation with the precision
1664 of the real type to the mpfr_t format with the GCC default
1665 round-to-nearest mode. */
1666 mpfr_t x;
1667 mpfr_init2 (x, rfmt->p);
1668 mpfr_from_real (x, &rv, GMP_RNDN);
1670 /* Return a value one greater to account for the leading minus sign. */
1671 unsigned HOST_WIDE_INT r
1672 = 1 + get_mpfr_format_length (x, "", prec, spec, 'D');
1673 mpfr_clear (x);
1674 return r;
1677 /* Return a range representing the minimum and maximum number of bytes
1678 that the directive DIR will output for any argument. PREC gives
1679 the adjusted precision range to account for negative precisions
1680 meaning the default 6. This function is used when the directive
1681 argument or its value isn't known. */
1683 static fmtresult
1684 format_floating (const directive &dir, const HOST_WIDE_INT prec[2])
1686 tree type;
1688 switch (dir.modifier)
1690 case FMT_LEN_l:
1691 case FMT_LEN_none:
1692 type = double_type_node;
1693 break;
1695 case FMT_LEN_L:
1696 type = long_double_type_node;
1697 break;
1699 case FMT_LEN_ll:
1700 type = long_double_type_node;
1701 break;
1703 default:
1704 return fmtresult ();
1707 /* The minimum and maximum number of bytes produced by the directive. */
1708 fmtresult res;
1710 /* The minimum output as determined by flags. It's always at least 1.
1711 When plus or space are set the output is preceded by either a sign
1712 or a space. */
1713 unsigned flagmin = (1 /* for the first digit */
1714 + (dir.get_flag ('+') | dir.get_flag (' ')));
1716 /* When the pound flag is set the decimal point is included in output
1717 regardless of precision. Whether or not a decimal point is included
1718 otherwise depends on the specification and precision. */
1719 bool radix = dir.get_flag ('#');
1721 switch (dir.specifier)
1723 case 'A':
1724 case 'a':
1726 HOST_WIDE_INT minprec = 6 + !radix /* decimal point */;
1727 if (dir.prec[0] <= 0)
1728 minprec = 0;
1729 else if (dir.prec[0] > 0)
1730 minprec = dir.prec[0] + !radix /* decimal point */;
1732 res.range.min = (2 /* 0x */
1733 + flagmin
1734 + radix
1735 + minprec
1736 + 3 /* p+0 */);
1738 res.range.max = format_floating_max (type, 'a', prec[1]);
1739 res.range.likely = res.range.min;
1741 /* The unlikely maximum accounts for the longest multibyte
1742 decimal point character. */
1743 res.range.unlikely = res.range.max;
1744 if (dir.prec[1] > 0)
1745 res.range.unlikely += target_mb_len_max () - 1;
1747 break;
1750 case 'E':
1751 case 'e':
1753 /* Minimum output attributable to precision and, when it's
1754 non-zero, decimal point. */
1755 HOST_WIDE_INT minprec = prec[0] ? prec[0] + !radix : 0;
1757 /* The minimum output is "[-+]1.234567e+00" regardless
1758 of the value of the actual argument. */
1759 res.range.min = (flagmin
1760 + radix
1761 + minprec
1762 + 2 /* e+ */ + 2);
1764 res.range.max = format_floating_max (type, 'e', prec[1]);
1765 res.range.likely = res.range.min;
1767 /* The unlikely maximum accounts for the longest multibyte
1768 decimal point character. */
1769 if (dir.prec[0] != dir.prec[1]
1770 || dir.prec[0] == -1 || dir.prec[0] > 0)
1771 res.range.unlikely = res.range.max + target_mb_len_max () -1;
1772 else
1773 res.range.unlikely = res.range.max;
1774 break;
1777 case 'F':
1778 case 'f':
1780 /* Minimum output attributable to precision and, when it's non-zero,
1781 decimal point. */
1782 HOST_WIDE_INT minprec = prec[0] ? prec[0] + !radix : 0;
1784 /* The lower bound when precision isn't specified is 8 bytes
1785 ("1.23456" since precision is taken to be 6). When precision
1786 is zero, the lower bound is 1 byte (e.g., "1"). Otherwise,
1787 when precision is greater than zero, then the lower bound
1788 is 2 plus precision (plus flags). */
1789 res.range.min = flagmin + radix + minprec;
1791 /* Compute the upper bound for -TYPE_MAX. */
1792 res.range.max = format_floating_max (type, 'f', prec[1]);
1794 /* The minimum output with unknown precision is a single byte
1795 (e.g., "0") but the more likely output is 3 bytes ("0.0"). */
1796 if (dir.prec[0] < 0 && dir.prec[1] > 0)
1797 res.range.likely = 3;
1798 else
1799 res.range.likely = res.range.min;
1801 /* The unlikely maximum accounts for the longest multibyte
1802 decimal point character. */
1803 if (dir.prec[0] != dir.prec[1]
1804 || dir.prec[0] == -1 || dir.prec[0] > 0)
1805 res.range.unlikely = res.range.max + target_mb_len_max () - 1;
1806 break;
1809 case 'G':
1810 case 'g':
1812 /* The %g output depends on precision and the exponent of
1813 the argument. Since the value of the argument isn't known
1814 the lower bound on the range of bytes (not counting flags
1815 or width) is 1 plus radix (i.e., either "0" or "0." for
1816 "%g" and "%#g", respectively, with a zero argument). */
1817 res.range.min = flagmin + radix;
1819 char spec = 'g';
1820 HOST_WIDE_INT maxprec = dir.prec[1];
1821 if (radix && maxprec)
1823 /* When the pound flag (radix) is set, trailing zeros aren't
1824 trimmed and so the longest output is the same as for %e,
1825 except with precision minus 1 (as specified in C11). */
1826 spec = 'e';
1827 if (maxprec > 0)
1828 --maxprec;
1829 else if (maxprec < 0)
1830 maxprec = 5;
1832 else
1833 maxprec = prec[1];
1835 res.range.max = format_floating_max (type, spec, maxprec);
1837 /* The likely output is either the maximum computed above
1838 minus 1 (assuming the maximum is positive) when precision
1839 is known (or unspecified), or the same minimum as for %e
1840 (which is computed for a non-negative argument). Unlike
1841 for the other specifiers above the likely output isn't
1842 the minimum because for %g that's 1 which is unlikely. */
1843 if (dir.prec[1] < 0
1844 || (unsigned HOST_WIDE_INT)dir.prec[1] < target_int_max ())
1845 res.range.likely = res.range.max - 1;
1846 else
1848 HOST_WIDE_INT minprec = 6 + !radix /* decimal point */;
1849 res.range.likely = (flagmin
1850 + radix
1851 + minprec
1852 + 2 /* e+ */ + 2);
1855 /* The unlikely maximum accounts for the longest multibyte
1856 decimal point character. */
1857 res.range.unlikely = res.range.max + target_mb_len_max () - 1;
1858 break;
1861 default:
1862 return fmtresult ();
1865 /* Bump up the byte counters if WIDTH is greater. */
1866 res.adjust_for_width_or_precision (dir.width);
1867 return res;
1870 /* Return a range representing the minimum and maximum number of bytes
1871 that the directive DIR will write on output for the floating argument
1872 ARG. */
1874 static fmtresult
1875 format_floating (const directive &dir, tree arg)
1877 HOST_WIDE_INT prec[] = { dir.prec[0], dir.prec[1] };
1879 /* For an indeterminate precision the lower bound must be assumed
1880 to be zero. */
1881 if (TOUPPER (dir.specifier) == 'A')
1883 /* Get the number of fractional decimal digits needed to represent
1884 the argument without a loss of accuracy. */
1885 tree type = arg ? TREE_TYPE (arg) :
1886 (dir.modifier == FMT_LEN_L || dir.modifier == FMT_LEN_ll
1887 ? long_double_type_node : double_type_node);
1889 unsigned fmtprec
1890 = REAL_MODE_FORMAT (TYPE_MODE (type))->p;
1892 /* The precision of the IEEE 754 double format is 53.
1893 The precision of all other GCC binary double formats
1894 is 56 or less. */
1895 unsigned maxprec = fmtprec <= 56 ? 13 : 15;
1897 /* For %a, leave the minimum precision unspecified to let
1898 MFPR trim trailing zeros (as it and many other systems
1899 including Glibc happen to do) and set the maximum
1900 precision to reflect what it would be with trailing zeros
1901 present (as Solaris and derived systems do). */
1902 if (dir.prec[1] < 0)
1904 /* Both bounds are negative implies that precision has
1905 not been specified. */
1906 prec[0] = maxprec;
1907 prec[1] = -1;
1909 else if (dir.prec[0] < 0)
1911 /* With a negative lower bound and a non-negative upper
1912 bound set the minimum precision to zero and the maximum
1913 to the greater of the maximum precision (i.e., with
1914 trailing zeros present) and the specified upper bound. */
1915 prec[0] = 0;
1916 prec[1] = dir.prec[1] < maxprec ? maxprec : dir.prec[1];
1919 else if (dir.prec[0] < 0)
1921 if (dir.prec[1] < 0)
1923 /* A precision in a strictly negative range is ignored and
1924 the default of 6 is used instead. */
1925 prec[0] = prec[1] = 6;
1927 else
1929 /* For a precision in a partly negative range, the lower bound
1930 must be assumed to be zero and the new upper bound is the
1931 greater of 6 (the default precision used when the specified
1932 precision is negative) and the upper bound of the specified
1933 range. */
1934 prec[0] = 0;
1935 prec[1] = dir.prec[1] < 6 ? 6 : dir.prec[1];
1939 if (!arg || TREE_CODE (arg) != REAL_CST)
1940 return format_floating (dir, prec);
1942 /* The minimum and maximum number of bytes produced by the directive. */
1943 fmtresult res;
1945 /* Get the real type format desription for the target. */
1946 const REAL_VALUE_TYPE *rvp = TREE_REAL_CST_PTR (arg);
1947 const real_format *rfmt = REAL_MODE_FORMAT (TYPE_MODE (TREE_TYPE (arg)));
1949 char fmtstr [40];
1950 char *pfmt = fmtstr;
1952 /* Append flags. */
1953 for (const char *pf = "-+ #0"; *pf; ++pf)
1954 if (dir.get_flag (*pf))
1955 *pfmt++ = *pf;
1957 *pfmt = '\0';
1960 /* Set up an array to easily iterate over. */
1961 unsigned HOST_WIDE_INT* const minmax[] = {
1962 &res.range.min, &res.range.max
1965 for (int i = 0; i != sizeof minmax / sizeof *minmax; ++i)
1967 /* Convert the GCC real value representation with the precision
1968 of the real type to the mpfr_t format rounding down in the
1969 first iteration that computes the minimm and up in the second
1970 that computes the maximum. This order is arbibtrary because
1971 rounding in either direction can result in longer output. */
1972 mpfr_t mpfrval;
1973 mpfr_init2 (mpfrval, rfmt->p);
1974 mpfr_from_real (mpfrval, rvp, i ? GMP_RNDU : GMP_RNDD);
1976 /* Use the MPFR rounding specifier to round down in the first
1977 iteration and then up. In most but not all cases this will
1978 result in the same number of bytes. */
1979 char rndspec = "DU"[i];
1981 /* Format it and store the result in the corresponding member
1982 of the result struct. */
1983 *minmax[i] = get_mpfr_format_length (mpfrval, fmtstr, prec[i],
1984 dir.specifier, rndspec);
1985 mpfr_clear (mpfrval);
1989 /* Make sure the minimum is less than the maximum (MPFR rounding
1990 in the call to mpfr_snprintf can result in the reverse. */
1991 if (res.range.max < res.range.min)
1993 unsigned HOST_WIDE_INT tmp = res.range.min;
1994 res.range.min = res.range.max;
1995 res.range.max = tmp;
1998 /* The range is known unless either width or precision is unknown. */
1999 res.knownrange = dir.known_width_and_precision ();
2001 /* For the same floating point constant, unless width or precision
2002 is unknown, use the longer output as the likely maximum since
2003 with round to nearest either is equally likely. Otheriwse, when
2004 precision is unknown, use the greater of the minimum and 3 as
2005 the likely output (for "0.0" since zero precision is unlikely). */
2006 if (res.knownrange)
2007 res.range.likely = res.range.max;
2008 else if (res.range.min < 3
2009 && dir.prec[0] < 0
2010 && (unsigned HOST_WIDE_INT)dir.prec[1] == target_int_max ())
2011 res.range.likely = 3;
2012 else
2013 res.range.likely = res.range.min;
2015 res.range.unlikely = res.range.max;
2017 if (res.range.max > 2 && (prec[0] != 0 || prec[1] != 0))
2019 /* Unless the precision is zero output longer than 2 bytes may
2020 include the decimal point which must be a single character
2021 up to MB_LEN_MAX in length. This is overly conservative
2022 since in some conversions some constants result in no decimal
2023 point (e.g., in %g). */
2024 res.range.unlikely += target_mb_len_max () - 1;
2027 res.adjust_for_width_or_precision (dir.width);
2028 return res;
2031 /* Return a FMTRESULT struct set to the lengths of the shortest and longest
2032 strings referenced by the expression STR, or (-1, -1) when not known.
2033 Used by the format_string function below. */
2035 static fmtresult
2036 get_string_length (tree str)
2038 if (!str)
2039 return fmtresult ();
2041 if (tree slen = c_strlen (str, 1))
2043 /* Simply return the length of the string. */
2044 fmtresult res (tree_to_shwi (slen));
2045 return res;
2048 /* Determine the length of the shortest and longest string referenced
2049 by STR. Strings of unknown lengths are bounded by the sizes of
2050 arrays that subexpressions of STR may refer to. Pointers that
2051 aren't known to point any such arrays result in LENRANGE[1] set
2052 to SIZE_MAX. */
2053 tree lenrange[2];
2054 bool flexarray = get_range_strlen (str, lenrange);
2056 if (lenrange [0] || lenrange [1])
2058 HOST_WIDE_INT min
2059 = (tree_fits_uhwi_p (lenrange[0])
2060 ? tree_to_uhwi (lenrange[0])
2061 : 0);
2063 HOST_WIDE_INT max
2064 = (tree_fits_uhwi_p (lenrange[1])
2065 ? tree_to_uhwi (lenrange[1])
2066 : HOST_WIDE_INT_M1U);
2068 /* get_range_strlen() returns the target value of SIZE_MAX for
2069 strings of unknown length. Bump it up to HOST_WIDE_INT_M1U
2070 which may be bigger. */
2071 if ((unsigned HOST_WIDE_INT)min == target_size_max ())
2072 min = HOST_WIDE_INT_M1U;
2073 if ((unsigned HOST_WIDE_INT)max == target_size_max ())
2074 max = HOST_WIDE_INT_M1U;
2076 fmtresult res (min, max);
2078 /* Set RES.KNOWNRANGE to true if and only if all strings referenced
2079 by STR are known to be bounded (though not necessarily by their
2080 actual length but perhaps by their maximum possible length). */
2081 if (res.range.max < target_int_max ())
2083 res.knownrange = true;
2084 /* When the the length of the longest string is known and not
2085 excessive use it as the likely length of the string(s). */
2086 res.range.likely = res.range.max;
2088 else
2090 /* When the upper bound is unknown (it can be zero or excessive)
2091 set the likely length to the greater of 1 and the length of
2092 the shortest string and reset the lower bound to zero. */
2093 res.range.likely = res.range.min ? res.range.min : warn_level > 1;
2094 res.range.min = 0;
2097 /* If the range of string length has been estimated from the size
2098 of an array at the end of a struct assume that it's longer than
2099 the array bound says it is in case it's used as a poor man's
2100 flexible array member, such as in struct S { char a[4]; }; */
2101 res.range.unlikely = flexarray ? HOST_WIDE_INT_MAX : res.range.max;
2103 return res;
2106 return get_string_length (NULL_TREE);
2109 /* Return the minimum and maximum number of characters formatted
2110 by the '%c' format directives and its wide character form for
2111 the argument ARG. ARG can be null (for functions such as
2112 vsprinf). */
2114 static fmtresult
2115 format_character (const directive &dir, tree arg)
2117 fmtresult res;
2119 res.knownrange = true;
2121 if (dir.modifier == FMT_LEN_l)
2123 /* A wide character can result in as few as zero bytes. */
2124 res.range.min = 0;
2126 HOST_WIDE_INT min, max;
2127 if (get_int_range (arg, &min, &max, false, 0))
2129 if (min == 0 && max == 0)
2131 /* The NUL wide character results in no bytes. */
2132 res.range.max = 0;
2133 res.range.likely = 0;
2134 res.range.unlikely = 0;
2136 else if (min > 0 && min < 128)
2138 /* A wide character in the ASCII range most likely results
2139 in a single byte, and only unlikely in up to MB_LEN_MAX. */
2140 res.range.max = 1;
2141 res.range.likely = 1;
2142 res.range.unlikely = target_mb_len_max ();
2144 else
2146 /* A wide character outside the ASCII range likely results
2147 in up to two bytes, and only unlikely in up to MB_LEN_MAX. */
2148 res.range.max = target_mb_len_max ();
2149 res.range.likely = 2;
2150 res.range.unlikely = res.range.max;
2153 else
2155 /* An unknown wide character is treated the same as a wide
2156 character outside the ASCII range. */
2157 res.range.max = target_mb_len_max ();
2158 res.range.likely = 2;
2159 res.range.unlikely = res.range.max;
2162 else
2164 /* A plain '%c' directive. Its ouput is exactly 1. */
2165 res.range.min = res.range.max = 1;
2166 res.range.likely = res.range.unlikely = 1;
2167 res.knownrange = true;
2170 /* Bump up the byte counters if WIDTH is greater. */
2171 return res.adjust_for_width_or_precision (dir.width);
2174 /* Return the minimum and maximum number of characters formatted
2175 by the '%s' format directive and its wide character form for
2176 the argument ARG. ARG can be null (for functions such as
2177 vsprinf). */
2179 static fmtresult
2180 format_string (const directive &dir, tree arg)
2182 fmtresult res;
2184 /* Compute the range the argument's length can be in. */
2185 fmtresult slen = get_string_length (arg);
2186 if (slen.range.min == slen.range.max
2187 && slen.range.min < HOST_WIDE_INT_MAX)
2189 /* The argument is either a string constant or it refers
2190 to one of a number of strings of the same length. */
2192 /* A '%s' directive with a string argument with constant length. */
2193 res.range = slen.range;
2195 if (dir.modifier == FMT_LEN_l)
2197 /* In the worst case the length of output of a wide string S
2198 is bounded by MB_LEN_MAX * wcslen (S). */
2199 res.range.max *= target_mb_len_max ();
2200 res.range.unlikely = res.range.max;
2201 /* It's likely that the the total length is not more that
2202 2 * wcslen (S).*/
2203 res.range.likely = res.range.min * 2;
2205 if (dir.prec[1] >= 0
2206 && (unsigned HOST_WIDE_INT)dir.prec[1] < res.range.max)
2208 res.range.max = dir.prec[1];
2209 res.range.likely = dir.prec[1];
2210 res.range.unlikely = dir.prec[1];
2213 if (dir.prec[0] < 0 && dir.prec[1] > -1)
2214 res.range.min = 0;
2215 else if (dir.prec[0] >= 0)
2216 res.range.likely = dir.prec[0];
2218 /* Even a non-empty wide character string need not convert into
2219 any bytes. */
2220 res.range.min = 0;
2222 else
2224 res.knownrange = true;
2226 if (dir.prec[0] < 0 && dir.prec[1] > -1)
2227 res.range.min = 0;
2228 else if ((unsigned HOST_WIDE_INT)dir.prec[0] < res.range.min)
2229 res.range.min = dir.prec[0];
2231 if ((unsigned HOST_WIDE_INT)dir.prec[1] < res.range.max)
2233 res.range.max = dir.prec[1];
2234 res.range.likely = dir.prec[1];
2235 res.range.unlikely = dir.prec[1];
2239 else if (arg && integer_zerop (arg))
2241 /* Handle null pointer argument. */
2243 fmtresult res (0);
2244 res.nullp = true;
2245 return res;
2247 else
2249 /* For a '%s' and '%ls' directive with a non-constant string (either
2250 one of a number of strings of known length or an unknown string)
2251 the minimum number of characters is lesser of PRECISION[0] and
2252 the length of the shortest known string or zero, and the maximum
2253 is the lessser of the length of the longest known string or
2254 PTRDIFF_MAX and PRECISION[1]. The likely length is either
2255 the minimum at level 1 and the greater of the minimum and 1
2256 at level 2. This result is adjust upward for width (if it's
2257 specified). */
2259 if (dir.modifier == FMT_LEN_l)
2261 /* A wide character converts to as few as zero bytes. */
2262 slen.range.min = 0;
2263 if (slen.range.max < target_int_max ())
2264 slen.range.max *= target_mb_len_max ();
2266 if (slen.range.likely < target_int_max ())
2267 slen.range.likely *= 2;
2269 if (slen.range.likely < target_int_max ())
2270 slen.range.unlikely *= target_mb_len_max ();
2273 res.range = slen.range;
2275 if (dir.prec[0] >= 0)
2277 /* Adjust the minimum to zero if the string length is unknown,
2278 or at most the lower bound of the precision otherwise. */
2279 if (slen.range.min >= target_int_max ())
2280 res.range.min = 0;
2281 else if ((unsigned HOST_WIDE_INT)dir.prec[0] < slen.range.min)
2282 res.range.min = dir.prec[0];
2284 /* Make both maxima no greater than the upper bound of precision. */
2285 if ((unsigned HOST_WIDE_INT)dir.prec[1] < slen.range.max
2286 || slen.range.max >= target_int_max ())
2288 res.range.max = dir.prec[1];
2289 res.range.unlikely = dir.prec[1];
2292 /* If precision is constant, set the likely counter to the lesser
2293 of it and the maximum string length. Otherwise, if the lower
2294 bound of precision is greater than zero, set the likely counter
2295 to the minimum. Otherwise set it to zero or one based on
2296 the warning level. */
2297 if (dir.prec[0] == dir.prec[1])
2298 res.range.likely
2299 = ((unsigned HOST_WIDE_INT)dir.prec[0] < slen.range.max
2300 ? dir.prec[0] : slen.range.max);
2301 else if (dir.prec[0] > 0)
2302 res.range.likely = res.range.min;
2303 else
2304 res.range.likely = warn_level > 1;
2306 else if (dir.prec[1] >= 0)
2308 res.range.min = 0;
2309 if ((unsigned HOST_WIDE_INT)dir.prec[1] < slen.range.max)
2310 res.range.max = dir.prec[1];
2311 res.range.likely = dir.prec[1] ? warn_level > 1 : 0;
2313 else if (slen.range.min >= target_int_max ())
2315 res.range.min = 0;
2316 res.range.max = HOST_WIDE_INT_MAX;
2317 /* At level 1 strings of unknown length are assumed to be
2318 empty, while at level 1 they are assumed to be one byte
2319 long. */
2320 res.range.likely = warn_level > 1;
2322 else
2324 /* A string of unknown length unconstrained by precision is
2325 assumed to be empty at level 1 and just one character long
2326 at higher levels. */
2327 if (res.range.likely >= target_int_max ())
2328 res.range.likely = warn_level > 1;
2331 res.range.unlikely = res.range.max;
2334 /* Bump up the byte counters if WIDTH is greater. */
2335 return res.adjust_for_width_or_precision (dir.width);
2338 /* Format plain string (part of the format string itself). */
2340 static fmtresult
2341 format_plain (const directive &dir, tree)
2343 fmtresult res (dir.len);
2344 return res;
2347 /* Return true if the RESULT of a directive in a call describe by INFO
2348 should be diagnosed given the AVAILable space in the destination. */
2350 static bool
2351 should_warn_p (const pass_sprintf_length::call_info &info,
2352 const result_range &avail, const result_range &result)
2354 if (result.max <= avail.min)
2356 /* The least amount of space remaining in the destination is big
2357 enough for the longest output. */
2358 return false;
2361 if (info.bounded)
2363 if (warn_format_trunc == 1 && result.min <= avail.max
2364 && info.retval_used ())
2366 /* The likely amount of space remaining in the destination is big
2367 enough for the least output and the return value is used. */
2368 return false;
2371 if (warn_format_trunc == 1 && result.likely <= avail.likely
2372 && !info.retval_used ())
2374 /* The likely amount of space remaining in the destination is big
2375 enough for the likely output and the return value is unused. */
2376 return false;
2379 if (warn_format_trunc == 2
2380 && result.likely <= avail.min
2381 && (result.max <= avail.min
2382 || result.max > HOST_WIDE_INT_MAX))
2384 /* The minimum amount of space remaining in the destination is big
2385 enough for the longest output. */
2386 return false;
2389 else
2391 if (warn_level == 1 && result.likely <= avail.likely)
2393 /* The likely amount of space remaining in the destination is big
2394 enough for the likely output. */
2395 return false;
2398 if (warn_level == 2
2399 && result.likely <= avail.min
2400 && (result.max <= avail.min
2401 || result.max > HOST_WIDE_INT_MAX))
2403 /* The minimum amount of space remaining in the destination is big
2404 enough for the longest output. */
2405 return false;
2409 return true;
2412 /* At format string location describe by DIRLOC in a call described
2413 by INFO, issue a warning for a directive DIR whose output may be
2414 in excess of the available space AVAIL_RANGE in the destination
2415 given the formatting result FMTRES. This function does nothing
2416 except decide whether to issue a warning for a possible write
2417 past the end or truncation and, if so, format the warning.
2418 Return true if a warning has been issued. */
2420 static bool
2421 maybe_warn (substring_loc &dirloc, source_range *pargrange,
2422 const pass_sprintf_length::call_info &info,
2423 const result_range &avail_range, const result_range &res,
2424 const directive &dir)
2426 if (!should_warn_p (info, avail_range, res))
2427 return false;
2429 /* A warning will definitely be issued below. */
2431 /* The maximum byte count to reference in the warning. Larger counts
2432 imply that the upper bound is unknown (and could be anywhere between
2433 RES.MIN + 1 and SIZE_MAX / 2) are printed as "N or more bytes" rather
2434 than "between N and X" where X is some huge number. */
2435 unsigned HOST_WIDE_INT maxbytes = target_dir_max ();
2437 /* True when there is enough room in the destination for the least
2438 amount of a directive's output but not enough for its likely or
2439 maximum output. */
2440 bool maybe = (res.min <= avail_range.max
2441 && (avail_range.min < res.likely
2442 || (res.max < HOST_WIDE_INT_MAX
2443 && avail_range.min < res.max)));
2445 /* Buffer for the directive in the host character set (used when
2446 the source character set is different). */
2447 char hostdir[32];
2449 if (avail_range.min == avail_range.max)
2451 /* The size of the destination region is exact. */
2452 unsigned HOST_WIDE_INT navail = avail_range.max;
2454 if (target_to_host (*dir.beg) != '%')
2456 /* For plain character directives (i.e., the format string itself)
2457 but not others, point the caret at the first character that's
2458 past the end of the destination. */
2459 dirloc.set_caret_index (dirloc.get_caret_idx () + navail);
2462 if (*dir.beg == '\0')
2464 /* This is the terminating nul. */
2465 gcc_assert (res.min == 1 && res.min == res.max);
2467 const char *fmtstr
2468 = (info.bounded
2469 ? (maybe
2470 ? G_("%qE output may be truncated before the last format "
2471 "character")
2472 : G_("%qE output truncated before the last format character"))
2473 : (maybe
2474 ? G_("%qE may write a terminating nul past the end "
2475 "of the destination")
2476 : G_("%qE writing a terminating nul past the end "
2477 "of the destination")));
2479 return fmtwarn (dirloc, NULL, NULL, info.warnopt (), fmtstr,
2480 info.func);
2483 if (res.min == res.max)
2485 const char* fmtstr
2486 = (res.min == 1
2487 ? (info.bounded
2488 ? (maybe
2489 ? G_("%<%.*s%> directive output may be truncated writing "
2490 "%wu byte into a region of size %wu")
2491 : G_("%<%.*s%> directive output truncated writing "
2492 "%wu byte into a region of size %wu"))
2493 : G_("%<%.*s%> directive writing %wu byte "
2494 "into a region of size %wu"))
2495 : (info.bounded
2496 ? (maybe
2497 ? G_("%<%.*s%> directive output may be truncated writing "
2498 "%wu bytes into a region of size %wu")
2499 : G_("%<%.*s%> directive output truncated writing "
2500 "%wu bytes into a region of size %wu"))
2501 : G_("%<%.*s%> directive writing %wu bytes "
2502 "into a region of size %wu")));
2503 return fmtwarn (dirloc, pargrange, NULL,
2504 info.warnopt (), fmtstr, dir.len,
2505 target_to_host (hostdir, sizeof hostdir, dir.beg),
2506 res.min, navail);
2509 if (res.min == 0 && res.max < maxbytes)
2511 const char* fmtstr
2512 = (info.bounded
2513 ? (maybe
2514 ? G_("%<%.*s%> directive output may be truncated writing "
2515 "up to %wu bytes into a region of size %wu")
2516 : G_("%<%.*s%> directive output truncated writing "
2517 "up to %wu bytes into a region of size %wu"))
2518 : G_("%<%.*s%> directive writing up to %wu bytes "
2519 "into a region of size %wu"));
2520 return fmtwarn (dirloc, pargrange, NULL,
2521 info.warnopt (), fmtstr, dir.len,
2522 target_to_host (hostdir, sizeof hostdir, dir.beg),
2523 res.max, navail);
2526 if (res.min == 0 && maxbytes <= res.max)
2528 /* This is a special case to avoid issuing the potentially
2529 confusing warning:
2530 writing 0 or more bytes into a region of size 0. */
2531 const char* fmtstr
2532 = (info.bounded
2533 ? (maybe
2534 ? G_("%<%.*s%> directive output may be truncated writing "
2535 "likely %wu or more bytes into a region of size %wu")
2536 : G_("%<%.*s%> directive output truncated writing "
2537 "likely %wu or more bytes into a region of size %wu"))
2538 : G_("%<%.*s%> directive writing likely %wu or more bytes "
2539 "into a region of size %wu"));
2540 return fmtwarn (dirloc, pargrange, NULL,
2541 info.warnopt (), fmtstr, dir.len,
2542 target_to_host (hostdir, sizeof hostdir, dir.beg),
2543 res.likely, navail);
2546 if (res.max < maxbytes)
2548 const char* fmtstr
2549 = (info.bounded
2550 ? (maybe
2551 ? G_("%<%.*s%> directive output may be truncated writing "
2552 "between %wu and %wu bytes into a region of size %wu")
2553 : G_("%<%.*s%> directive output truncated writing "
2554 "between %wu and %wu bytes into a region of size %wu"))
2555 : G_("%<%.*s%> directive writing between %wu and "
2556 "%wu bytes into a region of size %wu"));
2557 return fmtwarn (dirloc, pargrange, NULL,
2558 info.warnopt (), fmtstr, dir.len,
2559 target_to_host (hostdir, sizeof hostdir, dir.beg),
2560 res.min, res.max, navail);
2563 const char* fmtstr
2564 = (info.bounded
2565 ? (maybe
2566 ? G_("%<%.*s%> directive output may be truncated writing "
2567 "%wu or more bytes into a region of size %wu")
2568 : G_("%<%.*s%> directive output truncated writing "
2569 "%wu or more bytes into a region of size %wu"))
2570 : G_("%<%.*s%> directive writing %wu or more bytes "
2571 "into a region of size %wu"));
2572 return fmtwarn (dirloc, pargrange, NULL,
2573 info.warnopt (), fmtstr, dir.len,
2574 target_to_host (hostdir, sizeof hostdir, dir.beg),
2575 res.min, navail);
2578 /* The size of the destination region is a range. */
2580 if (target_to_host (*dir.beg) != '%')
2582 unsigned HOST_WIDE_INT navail = avail_range.max;
2584 /* For plain character directives (i.e., the format string itself)
2585 but not others, point the caret at the first character that's
2586 past the end of the destination. */
2587 dirloc.set_caret_index (dirloc.get_caret_idx () + navail);
2590 if (*dir.beg == '\0')
2592 gcc_assert (res.min == 1 && res.min == res.max);
2594 const char *fmtstr
2595 = (info.bounded
2596 ? (maybe
2597 ? G_("%qE output may be truncated before the last format "
2598 "character")
2599 : G_("%qE output truncated before the last format character"))
2600 : (maybe
2601 ? G_("%qE may write a terminating nul past the end "
2602 "of the destination")
2603 : G_("%qE writing a terminating nul past the end "
2604 "of the destination")));
2606 return fmtwarn (dirloc, NULL, NULL, info.warnopt (), fmtstr,
2607 info.func);
2610 if (res.min == res.max)
2612 const char* fmtstr
2613 = (res.min == 1
2614 ? (info.bounded
2615 ? (maybe
2616 ? G_("%<%.*s%> directive output may be truncated writing "
2617 "%wu byte into a region of size between %wu and %wu")
2618 : G_("%<%.*s%> directive output truncated writing "
2619 "%wu byte into a region of size between %wu and %wu"))
2620 : G_("%<%.*s%> directive writing %wu byte "
2621 "into a region of size between %wu and %wu"))
2622 : (info.bounded
2623 ? (maybe
2624 ? G_("%<%.*s%> directive output may be truncated writing "
2625 "%wu bytes into a region of size between %wu and %wu")
2626 : G_("%<%.*s%> directive output truncated writing "
2627 "%wu bytes into a region of size between %wu and %wu"))
2628 : G_("%<%.*s%> directive writing %wu bytes "
2629 "into a region of size between %wu and %wu")));
2631 return fmtwarn (dirloc, pargrange, NULL,
2632 info.warnopt (), fmtstr, dir.len,
2633 target_to_host (hostdir, sizeof hostdir, dir.beg),
2634 res.min, avail_range.min, avail_range.max);
2637 if (res.min == 0 && res.max < maxbytes)
2639 const char* fmtstr
2640 = (info.bounded
2641 ? (maybe
2642 ? G_("%<%.*s%> directive output may be truncated writing "
2643 "up to %wu bytes into a region of size between "
2644 "%wu and %wu")
2645 : G_("%<%.*s%> directive output truncated writing "
2646 "up to %wu bytes into a region of size between "
2647 "%wu and %wu"))
2648 : G_("%<%.*s%> directive writing up to %wu bytes "
2649 "into a region of size between %wu and %wu"));
2650 return fmtwarn (dirloc, pargrange, NULL,
2651 info.warnopt (), fmtstr, dir.len,
2652 target_to_host (hostdir, sizeof hostdir, dir.beg),
2653 res.max, avail_range.min, avail_range.max);
2656 if (res.min == 0 && maxbytes <= res.max)
2658 /* This is a special case to avoid issuing the potentially confusing
2659 warning:
2660 writing 0 or more bytes into a region of size between 0 and N. */
2661 const char* fmtstr
2662 = (info.bounded
2663 ? (maybe
2664 ? G_("%<%.*s%> directive output may be truncated writing "
2665 "likely %wu or more bytes into a region of size between "
2666 "%wu and %wu")
2667 : G_("%<%.*s%> directive output truncated writing likely "
2668 "%wu or more bytes into a region of size between "
2669 "%wu and %wu"))
2670 : G_("%<%.*s%> directive writing likely %wu or more bytes "
2671 "into a region of size between %wu and %wu"));
2672 return fmtwarn (dirloc, pargrange, NULL,
2673 info.warnopt (), fmtstr, dir.len,
2674 target_to_host (hostdir, sizeof hostdir, dir.beg),
2675 res.likely, avail_range.min, avail_range.max);
2678 if (res.max < maxbytes)
2680 const char* fmtstr
2681 = (info.bounded
2682 ? (maybe
2683 ? G_("%<%.*s%> directive output may be truncated writing "
2684 "between %wu and %wu bytes into a region of size "
2685 "between %wu and %wu")
2686 : G_("%<%.*s%> directive output truncated writing "
2687 "between %wu and %wu bytes into a region of size "
2688 "between %wu and %wu"))
2689 : G_("%<%.*s%> directive writing between %wu and "
2690 "%wu bytes into a region of size between %wu and %wu"));
2691 return fmtwarn (dirloc, pargrange, NULL,
2692 info.warnopt (), fmtstr, dir.len,
2693 target_to_host (hostdir, sizeof hostdir, dir.beg),
2694 res.min, res.max, avail_range.min, avail_range.max);
2697 const char* fmtstr
2698 = (info.bounded
2699 ? (maybe
2700 ? G_("%<%.*s%> directive output may be truncated writing "
2701 "%wu or more bytes into a region of size between "
2702 "%wu and %wu")
2703 : G_("%<%.*s%> directive output truncated writing "
2704 "%wu or more bytes into a region of size between "
2705 "%wu and %wu"))
2706 : G_("%<%.*s%> directive writing %wu or more bytes "
2707 "into a region of size between %wu and %wu"));
2708 return fmtwarn (dirloc, pargrange, NULL,
2709 info.warnopt (), fmtstr, dir.len,
2710 target_to_host (hostdir, sizeof hostdir, dir.beg),
2711 res.min, avail_range.min, avail_range.max);
2714 /* Compute the length of the output resulting from the directive DIR
2715 in a call described by INFO and update the overall result of the call
2716 in *RES. Return true if the directive has been handled. */
2718 static bool
2719 format_directive (const pass_sprintf_length::call_info &info,
2720 format_result *res, const directive &dir)
2722 /* Offset of the beginning of the directive from the beginning
2723 of the format string. */
2724 size_t offset = dir.beg - info.fmtstr;
2725 size_t start = offset;
2726 size_t length = offset + dir.len - !!dir.len;
2728 /* Create a location for the whole directive from the % to the format
2729 specifier. */
2730 substring_loc dirloc (info.fmtloc, TREE_TYPE (info.format),
2731 offset, start, length);
2733 /* Also create a location range for the argument if possible.
2734 This doesn't work for integer literals or function calls. */
2735 source_range argrange;
2736 source_range *pargrange;
2737 if (dir.arg && CAN_HAVE_LOCATION_P (dir.arg))
2739 argrange = EXPR_LOCATION_RANGE (dir.arg);
2740 pargrange = &argrange;
2742 else
2743 pargrange = NULL;
2745 /* Bail when there is no function to compute the output length,
2746 or when minimum length checking has been disabled. */
2747 if (!dir.fmtfunc || res->range.min >= HOST_WIDE_INT_MAX)
2748 return false;
2750 /* Compute the range of lengths of the formatted output. */
2751 fmtresult fmtres = dir.fmtfunc (dir, dir.arg);
2753 /* Record whether the output of all directives is known to be
2754 bounded by some maximum, implying that their arguments are
2755 either known exactly or determined to be in a known range
2756 or, for strings, limited by the upper bounds of the arrays
2757 they refer to. */
2758 res->knownrange &= fmtres.knownrange;
2760 if (!fmtres.knownrange)
2762 /* Only when the range is known, check it against the host value
2763 of INT_MAX + (the number of bytes of the "%.*Lf" directive with
2764 INT_MAX precision, which is the longest possible output of any
2765 single directive). That's the largest valid byte count (though
2766 not valid call to a printf-like function because it can never
2767 return such a count). Otherwise, the range doesn't correspond
2768 to known values of the argument. */
2769 if (fmtres.range.max > target_dir_max ())
2771 /* Normalize the MAX counter to avoid having to deal with it
2772 later. The counter can be less than HOST_WIDE_INT_M1U
2773 when compiling for an ILP32 target on an LP64 host. */
2774 fmtres.range.max = HOST_WIDE_INT_M1U;
2775 /* Disable exact and maximum length checking after a failure
2776 to determine the maximum number of characters (for example
2777 for wide characters or wide character strings) but continue
2778 tracking the minimum number of characters. */
2779 res->range.max = HOST_WIDE_INT_M1U;
2782 if (fmtres.range.min > target_dir_max ())
2784 /* Disable exact length checking after a failure to determine
2785 even the minimum number of characters (it shouldn't happen
2786 except in an error) but keep tracking the minimum and maximum
2787 number of characters. */
2788 return true;
2792 /* Buffer for the directive in the host character set (used when
2793 the source character set is different). */
2794 char hostdir[32];
2796 int dirlen = dir.len;
2798 if (fmtres.nullp)
2800 fmtwarn (dirloc, pargrange, NULL, info.warnopt (),
2801 "%<%.*s%> directive argument is null",
2802 dirlen, target_to_host (hostdir, sizeof hostdir, dir.beg));
2804 /* Don't bother processing the rest of the format string. */
2805 res->warned = true;
2806 res->range.min = HOST_WIDE_INT_M1U;
2807 res->range.max = HOST_WIDE_INT_M1U;
2808 return false;
2811 /* Compute the number of available bytes in the destination. There
2812 must always be at least one byte of space for the terminating
2813 NUL that's appended after the format string has been processed. */
2814 result_range avail_range = bytes_remaining (info.objsize, *res);
2816 bool warned = res->warned;
2818 if (!warned)
2819 warned = maybe_warn (dirloc, pargrange, info, avail_range,
2820 fmtres.range, dir);
2822 /* Bump up the total maximum if it isn't too big. */
2823 if (res->range.max < HOST_WIDE_INT_MAX
2824 && fmtres.range.max < HOST_WIDE_INT_MAX)
2825 res->range.max += fmtres.range.max;
2827 /* Raise the total unlikely maximum by the larger of the maximum
2828 and the unlikely maximum. */
2829 unsigned HOST_WIDE_INT save = res->range.unlikely;
2830 if (fmtres.range.max < fmtres.range.unlikely)
2831 res->range.unlikely += fmtres.range.unlikely;
2832 else
2833 res->range.unlikely += fmtres.range.max;
2835 if (res->range.unlikely < save)
2836 res->range.unlikely = HOST_WIDE_INT_M1U;
2838 res->range.min += fmtres.range.min;
2839 res->range.likely += fmtres.range.likely;
2841 /* Has the minimum directive output length exceeded the maximum
2842 of 4095 bytes required to be supported? */
2843 bool minunder4k = fmtres.range.min < 4096;
2844 bool maxunder4k = fmtres.range.max < 4096;
2845 /* Clear UNDER4K in the overall result if the maximum has exceeded
2846 the 4k (this is necessary to avoid the return valuye optimization
2847 that may not be safe in the maximum case). */
2848 if (!maxunder4k)
2849 res->under4k = false;
2851 if (!warned
2852 /* Only warn at level 2. */
2853 && 1 < warn_level
2854 && (!minunder4k
2855 || (!maxunder4k && fmtres.range.max < HOST_WIDE_INT_MAX)))
2857 /* The directive output may be longer than the maximum required
2858 to be handled by an implementation according to 7.21.6.1, p15
2859 of C11. Warn on this only at level 2 but remember this and
2860 prevent folding the return value when done. This allows for
2861 the possibility of the actual libc call failing due to ENOMEM
2862 (like Glibc does under some conditions). */
2864 if (fmtres.range.min == fmtres.range.max)
2865 warned = fmtwarn (dirloc, pargrange, NULL,
2866 info.warnopt (),
2867 "%<%.*s%> directive output of %wu bytes exceeds "
2868 "minimum required size of 4095",
2869 dirlen,
2870 target_to_host (hostdir, sizeof hostdir, dir.beg),
2871 fmtres.range.min);
2872 else
2874 const char *fmtstr
2875 = (minunder4k
2876 ? G_("%<%.*s%> directive output between %wu and %wu "
2877 "bytes may exceed minimum required size of 4095")
2878 : G_("%<%.*s%> directive output between %wu and %wu "
2879 "bytes exceeds minimum required size of 4095"));
2881 warned = fmtwarn (dirloc, pargrange, NULL,
2882 info.warnopt (), fmtstr, dirlen,
2883 target_to_host (hostdir, sizeof hostdir, dir.beg),
2884 fmtres.range.min, fmtres.range.max);
2888 /* Has the likely and maximum directive output exceeded INT_MAX? */
2889 bool likelyximax = *dir.beg && res->range.likely > target_int_max ();
2890 /* Don't consider the maximum to be in excess when it's the result
2891 of a string of unknown length (i.e., whose maximum has been set
2892 to be greater than or equal to HOST_WIDE_INT_MAX. */
2893 bool maxximax = (*dir.beg
2894 && res->range.max > target_int_max ()
2895 && res->range.max < HOST_WIDE_INT_MAX);
2897 if (!warned
2898 /* Warn for the likely output size at level 1. */
2899 && (likelyximax
2900 /* But only warn for the maximum at level 2. */
2901 || (1 < warn_level
2902 && maxximax
2903 && fmtres.range.max < HOST_WIDE_INT_MAX)))
2905 /* The directive output causes the total length of output
2906 to exceed INT_MAX bytes. */
2908 if (fmtres.range.min == fmtres.range.max)
2909 warned = fmtwarn (dirloc, pargrange, NULL, info.warnopt (),
2910 "%<%.*s%> directive output of %wu bytes causes "
2911 "result to exceed %<INT_MAX%>",
2912 dirlen,
2913 target_to_host (hostdir, sizeof hostdir, dir.beg),
2914 fmtres.range.min);
2915 else
2917 const char *fmtstr
2918 = (fmtres.range.min > target_int_max ()
2919 ? G_ ("%<%.*s%> directive output between %wu and %wu "
2920 "bytes causes result to exceed %<INT_MAX%>")
2921 : G_ ("%<%.*s%> directive output between %wu and %wu "
2922 "bytes may cause result to exceed %<INT_MAX%>"));
2923 warned = fmtwarn (dirloc, pargrange, NULL,
2924 info.warnopt (), fmtstr, dirlen,
2925 target_to_host (hostdir, sizeof hostdir, dir.beg),
2926 fmtres.range.min, fmtres.range.max);
2930 if (warned && fmtres.range.min < fmtres.range.likely
2931 && fmtres.range.likely < fmtres.range.max)
2933 inform (info.fmtloc,
2934 (1 == fmtres.range.likely
2935 ? G_("assuming directive output of %wu byte")
2936 : G_("assuming directive output of %wu bytes")),
2937 fmtres.range.likely);
2940 if (warned && fmtres.argmin)
2942 if (fmtres.argmin == fmtres.argmax)
2943 inform (info.fmtloc, "directive argument %qE", fmtres.argmin);
2944 else if (fmtres.knownrange)
2945 inform (info.fmtloc, "directive argument in the range [%E, %E]",
2946 fmtres.argmin, fmtres.argmax);
2947 else
2948 inform (info.fmtloc,
2949 "using the range [%E, %E] for directive argument",
2950 fmtres.argmin, fmtres.argmax);
2953 res->warned |= warned;
2955 if (!dir.beg[0] && res->warned && info.objsize < HOST_WIDE_INT_MAX)
2957 /* If a warning has been issued for buffer overflow or truncation
2958 (but not otherwise) help the user figure out how big a buffer
2959 they need. */
2961 location_t callloc = gimple_location (info.callstmt);
2963 unsigned HOST_WIDE_INT min = res->range.min;
2964 unsigned HOST_WIDE_INT max = res->range.max;
2966 if (min == max)
2967 inform (callloc,
2968 (min == 1
2969 ? G_("%qE output %wu byte into a destination of size %wu")
2970 : G_("%qE output %wu bytes into a destination of size %wu")),
2971 info.func, min, info.objsize);
2972 else if (max < HOST_WIDE_INT_MAX)
2973 inform (callloc,
2974 "%qE output between %wu and %wu bytes into "
2975 "a destination of size %wu",
2976 info.func, min, max, info.objsize);
2977 else if (min < res->range.likely && res->range.likely < max)
2978 inform (callloc,
2979 "%qE output %wu or more bytes (assuming %wu) into "
2980 "a destination of size %wu",
2981 info.func, min, res->range.likely, info.objsize);
2982 else
2983 inform (callloc,
2984 "%qE output %wu or more bytes into a destination of size %wu",
2985 info.func, min, info.objsize);
2988 if (dump_file && *dir.beg)
2990 fprintf (dump_file, " Result: %lli, %lli, %lli, %lli "
2991 "(%lli, %lli, %lli, %lli)\n",
2992 (long long)fmtres.range.min,
2993 (long long)fmtres.range.likely,
2994 (long long)fmtres.range.max,
2995 (long long)fmtres.range.unlikely,
2996 (long long)res->range.min,
2997 (long long)res->range.likely,
2998 (long long)res->range.max,
2999 (long long)res->range.unlikely);
3002 return true;
3005 #pragma GCC diagnostic pop
3007 /* Parse a format directive in function call described by INFO starting
3008 at STR and populate DIR structure. Bump up *ARGNO by the number of
3009 arguments extracted for the directive. Return the length of
3010 the directive. */
3012 static size_t
3013 parse_directive (pass_sprintf_length::call_info &info,
3014 directive &dir, format_result *res,
3015 const char *str, unsigned *argno)
3017 const char *pcnt = strchr (str, target_percent);
3018 dir.beg = str;
3020 if (size_t len = pcnt ? pcnt - str : *str ? strlen (str) : 1)
3022 /* This directive is either a plain string or the terminating nul
3023 (which isn't really a directive but it simplifies things to
3024 handle it as if it were). */
3025 dir.len = len;
3026 dir.fmtfunc = format_plain;
3028 if (dump_file)
3030 fprintf (dump_file, " Directive %u at offset %llu: \"%.*s\", "
3031 "length = %llu\n",
3032 dir.dirno,
3033 (unsigned long long)(size_t)(dir.beg - info.fmtstr),
3034 (int)dir.len, dir.beg, (unsigned long long)dir.len);
3037 return len - !*str;
3040 const char *pf = pcnt + 1;
3042 /* POSIX numbered argument index or zero when none. */
3043 HOST_WIDE_INT dollar = 0;
3045 /* With and precision. -1 when not specified, HOST_WIDE_INT_MIN
3046 when given by a va_list argument, and a non-negative value
3047 when specified in the format string itself. */
3048 HOST_WIDE_INT width = -1;
3049 HOST_WIDE_INT precision = -1;
3051 /* Pointers to the beginning of the width and precision decimal
3052 string (if any) within the directive. */
3053 const char *pwidth = 0;
3054 const char *pprec = 0;
3056 /* When the value of the decimal string that specifies width or
3057 precision is out of range, points to the digit that causes
3058 the value to exceed the limit. */
3059 const char *werange = NULL;
3060 const char *perange = NULL;
3062 /* Width specified via the asterisk. Need not be INTEGER_CST.
3063 For vararg functions set to void_node. */
3064 tree star_width = NULL_TREE;
3066 /* Width specified via the asterisk. Need not be INTEGER_CST.
3067 For vararg functions set to void_node. */
3068 tree star_precision = NULL_TREE;
3070 if (ISDIGIT (target_to_host (*pf)))
3072 /* This could be either a POSIX positional argument, the '0'
3073 flag, or a width, depending on what follows. Store it as
3074 width and sort it out later after the next character has
3075 been seen. */
3076 pwidth = pf;
3077 width = target_strtol10 (&pf, &werange);
3079 else if (target_to_host (*pf) == '*')
3081 /* Similarly to the block above, this could be either a POSIX
3082 positional argument or a width, depending on what follows. */
3083 if (*argno < gimple_call_num_args (info.callstmt))
3084 star_width = gimple_call_arg (info.callstmt, (*argno)++);
3085 else
3086 star_width = void_node;
3087 ++pf;
3090 if (target_to_host (*pf) == '$')
3092 /* Handle the POSIX dollar sign which references the 1-based
3093 positional argument number. */
3094 if (width != -1)
3095 dollar = width + info.argidx;
3096 else if (star_width
3097 && TREE_CODE (star_width) == INTEGER_CST
3098 && (TYPE_PRECISION (TREE_TYPE (star_width))
3099 <= TYPE_PRECISION (integer_type_node)))
3100 dollar = width + tree_to_shwi (star_width);
3102 /* Bail when the numbered argument is out of range (it will
3103 have already been diagnosed by -Wformat). */
3104 if (dollar == 0
3105 || dollar == (int)info.argidx
3106 || dollar > gimple_call_num_args (info.callstmt))
3107 return false;
3109 --dollar;
3111 star_width = NULL_TREE;
3112 width = -1;
3113 ++pf;
3116 if (dollar || !star_width)
3118 if (width != -1)
3120 if (width == 0)
3122 /* The '0' that has been interpreted as a width above is
3123 actually a flag. Reset HAVE_WIDTH, set the '0' flag,
3124 and continue processing other flags. */
3125 width = -1;
3126 dir.set_flag ('0');
3128 else if (!dollar)
3130 /* (Non-zero) width has been seen. The next character
3131 is either a period or a digit. */
3132 goto start_precision;
3135 /* When either '$' has been seen, or width has not been seen,
3136 the next field is the optional flags followed by an optional
3137 width. */
3138 for ( ; ; ) {
3139 switch (target_to_host (*pf))
3141 case ' ':
3142 case '0':
3143 case '+':
3144 case '-':
3145 case '#':
3146 dir.set_flag (target_to_host (*pf++));
3147 break;
3149 default:
3150 goto start_width;
3154 start_width:
3155 if (ISDIGIT (target_to_host (*pf)))
3157 werange = 0;
3158 pwidth = pf;
3159 width = target_strtol10 (&pf, &werange);
3161 else if (target_to_host (*pf) == '*')
3163 if (*argno < gimple_call_num_args (info.callstmt))
3164 star_width = gimple_call_arg (info.callstmt, (*argno)++);
3165 else
3167 /* This is (likely) a va_list. It could also be an invalid
3168 call with insufficient arguments. */
3169 star_width = void_node;
3171 ++pf;
3173 else if (target_to_host (*pf) == '\'')
3175 /* The POSIX apostrophe indicating a numeric grouping
3176 in the current locale. Even though it's possible to
3177 estimate the upper bound on the size of the output
3178 based on the number of digits it probably isn't worth
3179 continuing. */
3180 return 0;
3184 start_precision:
3185 if (target_to_host (*pf) == '.')
3187 ++pf;
3189 if (ISDIGIT (target_to_host (*pf)))
3191 pprec = pf;
3192 precision = target_strtol10 (&pf, &perange);
3194 else if (target_to_host (*pf) == '*')
3196 if (*argno < gimple_call_num_args (info.callstmt))
3197 star_precision = gimple_call_arg (info.callstmt, (*argno)++);
3198 else
3200 /* This is (likely) a va_list. It could also be an invalid
3201 call with insufficient arguments. */
3202 star_precision = void_node;
3204 ++pf;
3206 else
3208 /* The decimal precision or the asterisk are optional.
3209 When neither is dirified it's taken to be zero. */
3210 precision = 0;
3214 switch (target_to_host (*pf))
3216 case 'h':
3217 if (target_to_host (pf[1]) == 'h')
3219 ++pf;
3220 dir.modifier = FMT_LEN_hh;
3222 else
3223 dir.modifier = FMT_LEN_h;
3224 ++pf;
3225 break;
3227 case 'j':
3228 dir.modifier = FMT_LEN_j;
3229 ++pf;
3230 break;
3232 case 'L':
3233 dir.modifier = FMT_LEN_L;
3234 ++pf;
3235 break;
3237 case 'l':
3238 if (target_to_host (pf[1]) == 'l')
3240 ++pf;
3241 dir.modifier = FMT_LEN_ll;
3243 else
3244 dir.modifier = FMT_LEN_l;
3245 ++pf;
3246 break;
3248 case 't':
3249 dir.modifier = FMT_LEN_t;
3250 ++pf;
3251 break;
3253 case 'z':
3254 dir.modifier = FMT_LEN_z;
3255 ++pf;
3256 break;
3259 switch (target_to_host (*pf))
3261 /* Handle a sole '%' character the same as "%%" but since it's
3262 undefined prevent the result from being folded. */
3263 case '\0':
3264 --pf;
3265 res->range.min = res->range.max = HOST_WIDE_INT_M1U;
3266 /* FALLTHRU */
3267 case '%':
3268 dir.fmtfunc = format_percent;
3269 break;
3271 case 'a':
3272 case 'A':
3273 case 'e':
3274 case 'E':
3275 case 'f':
3276 case 'F':
3277 case 'g':
3278 case 'G':
3279 res->floating = true;
3280 dir.fmtfunc = format_floating;
3281 break;
3283 case 'd':
3284 case 'i':
3285 case 'o':
3286 case 'u':
3287 case 'x':
3288 case 'X':
3289 dir.fmtfunc = format_integer;
3290 break;
3292 case 'p':
3293 /* The %p output is implementation-defined. It's possible
3294 to determine this format but due to extensions (edirially
3295 those of the Linux kernel -- see bug 78512) the first %p
3296 in the format string disables any further processing. */
3297 return false;
3299 case 'n':
3300 /* %n has side-effects even when nothing is actually printed to
3301 any buffer. */
3302 info.nowrite = false;
3303 dir.fmtfunc = format_none;
3304 break;
3306 case 'c':
3307 dir.fmtfunc = format_character;
3308 break;
3310 case 'S':
3311 case 's':
3312 dir.fmtfunc = format_string;
3313 break;
3315 default:
3316 /* Unknown conversion specification. */
3317 return 0;
3320 dir.specifier = target_to_host (*pf++);
3322 /* Store the length of the format directive. */
3323 dir.len = pf - pcnt;
3325 /* Buffer for the directive in the host character set (used when
3326 the source character set is different). */
3327 char hostdir[32];
3329 if (star_width)
3331 if (INTEGRAL_TYPE_P (TREE_TYPE (star_width)))
3332 dir.set_width (star_width);
3333 else
3335 /* Width specified by a va_list takes on the range [0, -INT_MIN]
3336 (width is the absolute value of that specified). */
3337 dir.width[0] = 0;
3338 dir.width[1] = target_int_max () + 1;
3341 else
3343 if (width == LONG_MAX && werange)
3345 size_t begin = dir.beg - info.fmtstr + (pwidth - pcnt);
3346 size_t caret = begin + (werange - pcnt);
3347 size_t end = pf - info.fmtstr - 1;
3349 /* Create a location for the width part of the directive,
3350 pointing the caret at the first out-of-range digit. */
3351 substring_loc dirloc (info.fmtloc, TREE_TYPE (info.format),
3352 caret, begin, end);
3354 fmtwarn (dirloc, NULL, NULL,
3355 info.warnopt (), "%<%.*s%> directive width out of range",
3356 dir.len, target_to_host (hostdir, sizeof hostdir, dir.beg));
3359 dir.set_width (width);
3362 if (star_precision)
3364 if (INTEGRAL_TYPE_P (TREE_TYPE (star_precision)))
3365 dir.set_precision (star_precision);
3366 else
3368 /* Precision specified by a va_list takes on the range [-1, INT_MAX]
3369 (unlike width, negative precision is ignored). */
3370 dir.prec[0] = -1;
3371 dir.prec[1] = target_int_max ();
3374 else
3376 if (precision == LONG_MAX && perange)
3378 size_t begin = dir.beg - info.fmtstr + (pprec - pcnt) - 1;
3379 size_t caret = dir.beg - info.fmtstr + (perange - pcnt) - 1;
3380 size_t end = pf - info.fmtstr - 2;
3382 /* Create a location for the precision part of the directive,
3383 including the leading period, pointing the caret at the first
3384 out-of-range digit . */
3385 substring_loc dirloc (info.fmtloc, TREE_TYPE (info.format),
3386 caret, begin, end);
3388 fmtwarn (dirloc, NULL, NULL,
3389 info.warnopt (), "%<%.*s%> directive precision out of range",
3390 dir.len, target_to_host (hostdir, sizeof hostdir, dir.beg));
3393 dir.set_precision (precision);
3396 /* Extract the argument if the directive takes one and if it's
3397 available (e.g., the function doesn't take a va_list). Treat
3398 missing arguments the same as va_list, even though they will
3399 have likely already been diagnosed by -Wformat. */
3400 if (dir.specifier != '%'
3401 && *argno < gimple_call_num_args (info.callstmt))
3402 dir.arg = gimple_call_arg (info.callstmt, dollar ? dollar : (*argno)++);
3404 if (dump_file)
3406 fprintf (dump_file, " Directive %u at offset %llu: \"%.*s\"",
3407 dir.dirno, (unsigned long long)(size_t)(dir.beg - info.fmtstr),
3408 (int)dir.len, dir.beg);
3409 if (star_width)
3411 if (dir.width[0] == dir.width[1])
3412 fprintf (dump_file, ", width = %lli", (long long)dir.width[0]);
3413 else
3414 fprintf (dump_file, ", width in range [%lli, %lli]",
3415 (long long)dir.width[0], (long long)dir.width[1]);
3418 if (star_precision)
3420 if (dir.prec[0] == dir.prec[1])
3421 fprintf (dump_file, ", precision = %lli", (long long)dir.prec[0]);
3422 else
3423 fprintf (dump_file, ", precision in range [%lli, %lli]",
3424 (long long)dir.prec[0], (long long)dir.prec[1]);
3426 fputc ('\n', dump_file);
3429 return dir.len;
3432 /* Compute the length of the output resulting from the call to a formatted
3433 output function described by INFO and store the result of the call in
3434 *RES. Issue warnings for detected past the end writes. Return true
3435 if the complete format string has been processed and *RES can be relied
3436 on, false otherwise (e.g., when a unknown or unhandled directive was seen
3437 that caused the processing to be terminated early). */
3439 bool
3440 pass_sprintf_length::compute_format_length (call_info &info,
3441 format_result *res)
3443 if (dump_file)
3445 location_t callloc = gimple_location (info.callstmt);
3446 fprintf (dump_file, "%s:%i: ",
3447 LOCATION_FILE (callloc), LOCATION_LINE (callloc));
3448 print_generic_expr (dump_file, info.func, dump_flags);
3450 fprintf (dump_file, ": objsize = %llu, fmtstr = \"%s\"\n",
3451 (unsigned long long)info.objsize, info.fmtstr);
3454 /* Reset the minimum and maximum byte counters. */
3455 res->range.min = res->range.max = 0;
3457 /* No directive has been seen yet so the length of output is bounded
3458 by the known range [0, 0] (with no conversion producing more than
3459 4K bytes) until determined otherwise. */
3460 res->knownrange = true;
3461 res->under4k = true;
3462 res->floating = false;
3463 res->warned = false;
3465 /* 1-based directive counter. */
3466 unsigned dirno = 1;
3468 /* The variadic argument counter. */
3469 unsigned argno = info.argidx;
3471 for (const char *pf = info.fmtstr; ; ++dirno)
3473 directive dir = directive ();
3474 dir.dirno = dirno;
3476 size_t n = parse_directive (info, dir, res, pf, &argno);
3478 /* Return failure if the format function fails. */
3479 if (!format_directive (info, res, dir))
3480 return false;
3482 /* Return success the directive is zero bytes long and it's
3483 the last think in the format string (i.e., it's the terminating
3484 nul, which isn't really a directive but handling it as one makes
3485 things simpler). */
3486 if (!n)
3487 return *pf == '\0';
3489 pf += n;
3492 /* The complete format string was processed (with or without warnings). */
3493 return true;
3496 /* Return the size of the object referenced by the expression DEST if
3497 available, or -1 otherwise. */
3499 static unsigned HOST_WIDE_INT
3500 get_destination_size (tree dest)
3502 /* Initialize object size info before trying to compute it. */
3503 init_object_sizes ();
3505 /* Use __builtin_object_size to determine the size of the destination
3506 object. When optimizing, determine the smallest object (such as
3507 a member array as opposed to the whole enclosing object), otherwise
3508 use type-zero object size to determine the size of the enclosing
3509 object (the function fails without optimization in this type). */
3510 int ost = optimize > 0;
3511 unsigned HOST_WIDE_INT size;
3512 if (compute_builtin_object_size (dest, ost, &size))
3513 return size;
3515 return HOST_WIDE_INT_M1U;
3518 /* Return true if the call described by INFO with result RES safe to
3519 optimize (i.e., no undefined behavior), and set RETVAL to the range
3520 of its return values. */
3522 static bool
3523 is_call_safe (const pass_sprintf_length::call_info &info,
3524 const format_result &res, bool under4k,
3525 unsigned HOST_WIDE_INT retval[2])
3527 if (under4k && !res.under4k)
3528 return false;
3530 /* The minimum return value. */
3531 retval[0] = res.range.min;
3533 /* The maximum return value is in most cases bounded by RES.RANGE.MAX
3534 but in cases involving multibyte characters could be as large as
3535 RES.RANGE.UNLIKELY. */
3536 retval[1]
3537 = res.range.unlikely < res.range.max ? res.range.max : res.range.unlikely;
3539 /* Adjust the number of bytes which includes the terminating nul
3540 to reflect the return value of the function which does not.
3541 Because the valid range of the function is [INT_MIN, INT_MAX],
3542 a valid range before the adjustment below is [0, INT_MAX + 1]
3543 (the functions only return negative values on error or undefined
3544 behavior). */
3545 if (retval[0] <= target_int_max () + 1)
3546 --retval[0];
3547 if (retval[1] <= target_int_max () + 1)
3548 --retval[1];
3550 /* Avoid the return value optimization when the behavior of the call
3551 is undefined either because any directive may have produced 4K or
3552 more of output, or the return value exceeds INT_MAX, or because
3553 the output overflows the destination object (but leave it enabled
3554 when the function is bounded because then the behavior is well-
3555 defined). */
3556 if (retval[0] == retval[1]
3557 && (info.bounded || retval[0] < info.objsize)
3558 && retval[0] <= target_int_max ())
3559 return true;
3561 if ((info.bounded || retval[1] < info.objsize)
3562 && (retval[0] < target_int_max ()
3563 && retval[1] < target_int_max ()))
3564 return true;
3566 if (!under4k && (info.bounded || retval[0] < info.objsize))
3567 return true;
3569 return false;
3572 /* Given a suitable result RES of a call to a formatted output function
3573 described by INFO, substitute the result for the return value of
3574 the call. The result is suitable if the number of bytes it represents
3575 is known and exact. A result that isn't suitable for substitution may
3576 have its range set to the range of return values, if that is known.
3577 Return true if the call is removed and gsi_next should not be performed
3578 in the caller. */
3580 static bool
3581 try_substitute_return_value (gimple_stmt_iterator *gsi,
3582 const pass_sprintf_length::call_info &info,
3583 const format_result &res)
3585 tree lhs = gimple_get_lhs (info.callstmt);
3587 /* Set to true when the entire call has been removed. */
3588 bool removed = false;
3590 /* The minimum and maximum return value. */
3591 unsigned HOST_WIDE_INT retval[2];
3592 bool safe = is_call_safe (info, res, true, retval);
3594 if (safe
3595 && retval[0] == retval[1]
3596 /* Not prepared to handle possibly throwing calls here; they shouldn't
3597 appear in non-artificial testcases, except when the __*_chk routines
3598 are badly declared. */
3599 && !stmt_ends_bb_p (info.callstmt))
3601 tree cst = build_int_cst (integer_type_node, retval[0]);
3603 if (lhs == NULL_TREE
3604 && info.nowrite)
3606 /* Remove the call to the bounded function with a zero size
3607 (e.g., snprintf(0, 0, "%i", 123)) if there is no lhs. */
3608 unlink_stmt_vdef (info.callstmt);
3609 gsi_remove (gsi, true);
3610 removed = true;
3612 else if (info.nowrite)
3614 /* Replace the call to the bounded function with a zero size
3615 (e.g., snprintf(0, 0, "%i", 123) with the constant result
3616 of the function. */
3617 if (!update_call_from_tree (gsi, cst))
3618 gimplify_and_update_call_from_tree (gsi, cst);
3619 gimple *callstmt = gsi_stmt (*gsi);
3620 update_stmt (callstmt);
3622 else if (lhs)
3624 /* Replace the left-hand side of the call with the constant
3625 result of the formatted function. */
3626 gimple_call_set_lhs (info.callstmt, NULL_TREE);
3627 gimple *g = gimple_build_assign (lhs, cst);
3628 gsi_insert_after (gsi, g, GSI_NEW_STMT);
3629 update_stmt (info.callstmt);
3632 if (dump_file)
3634 if (removed)
3635 fprintf (dump_file, " Removing call statement.");
3636 else
3638 fprintf (dump_file, " Substituting ");
3639 print_generic_expr (dump_file, cst, dump_flags);
3640 fprintf (dump_file, " for %s.\n",
3641 info.nowrite ? "statement" : "return value");
3645 else if (lhs)
3647 bool setrange = false;
3649 if (safe
3650 && (info.bounded || retval[1] < info.objsize)
3651 && (retval[0] < target_int_max ()
3652 && retval[1] < target_int_max ()))
3654 /* If the result is in a valid range bounded by the size of
3655 the destination set it so that it can be used for subsequent
3656 optimizations. */
3657 int prec = TYPE_PRECISION (integer_type_node);
3659 wide_int min = wi::shwi (retval[0], prec);
3660 wide_int max = wi::shwi (retval[1], prec);
3661 set_range_info (lhs, VR_RANGE, min, max);
3663 setrange = true;
3666 if (dump_file)
3668 const char *inbounds
3669 = (retval[0] < info.objsize
3670 ? (retval[1] < info.objsize
3671 ? "in" : "potentially out-of")
3672 : "out-of");
3674 const char *what = setrange ? "Setting" : "Discarding";
3675 if (retval[0] != retval[1])
3676 fprintf (dump_file,
3677 " %s %s-bounds return value range [%llu, %llu].\n",
3678 what, inbounds,
3679 (unsigned long long)retval[0],
3680 (unsigned long long)retval[1]);
3681 else
3682 fprintf (dump_file, " %s %s-bounds return value %llu.\n",
3683 what, inbounds, (unsigned long long)retval[0]);
3687 if (dump_file)
3688 fputc ('\n', dump_file);
3690 return removed;
3693 /* Try to simplify a s{,n}printf call described by INFO with result
3694 RES by replacing it with a simpler and presumably more efficient
3695 call (such as strcpy). */
3697 static bool
3698 try_simplify_call (gimple_stmt_iterator *gsi,
3699 const pass_sprintf_length::call_info &info,
3700 const format_result &res)
3702 unsigned HOST_WIDE_INT dummy[2];
3703 if (!is_call_safe (info, res, info.retval_used (), dummy))
3704 return false;
3706 switch (info.fncode)
3708 case BUILT_IN_SNPRINTF:
3709 return gimple_fold_builtin_snprintf (gsi);
3711 case BUILT_IN_SPRINTF:
3712 return gimple_fold_builtin_sprintf (gsi);
3714 default:
3718 return false;
3721 /* Determine if a GIMPLE CALL is to one of the sprintf-like built-in
3722 functions and if so, handle it. Return true if the call is removed
3723 and gsi_next should not be performed in the caller. */
3725 bool
3726 pass_sprintf_length::handle_gimple_call (gimple_stmt_iterator *gsi)
3728 call_info info = call_info ();
3730 info.callstmt = gsi_stmt (*gsi);
3731 if (!gimple_call_builtin_p (info.callstmt, BUILT_IN_NORMAL))
3732 return false;
3734 info.func = gimple_call_fndecl (info.callstmt);
3735 info.fncode = DECL_FUNCTION_CODE (info.func);
3737 /* The size of the destination as in snprintf(dest, size, ...). */
3738 unsigned HOST_WIDE_INT dstsize = HOST_WIDE_INT_M1U;
3740 /* The size of the destination determined by __builtin_object_size. */
3741 unsigned HOST_WIDE_INT objsize = HOST_WIDE_INT_M1U;
3743 /* Buffer size argument number (snprintf and vsnprintf). */
3744 unsigned HOST_WIDE_INT idx_dstsize = HOST_WIDE_INT_M1U;
3746 /* Object size argument number (snprintf_chk and vsnprintf_chk). */
3747 unsigned HOST_WIDE_INT idx_objsize = HOST_WIDE_INT_M1U;
3749 /* Format string argument number (valid for all functions). */
3750 unsigned idx_format;
3752 switch (info.fncode)
3754 case BUILT_IN_SPRINTF:
3755 // Signature:
3756 // __builtin_sprintf (dst, format, ...)
3757 idx_format = 1;
3758 info.argidx = 2;
3759 break;
3761 case BUILT_IN_SPRINTF_CHK:
3762 // Signature:
3763 // __builtin___sprintf_chk (dst, ost, objsize, format, ...)
3764 idx_objsize = 2;
3765 idx_format = 3;
3766 info.argidx = 4;
3767 break;
3769 case BUILT_IN_SNPRINTF:
3770 // Signature:
3771 // __builtin_snprintf (dst, size, format, ...)
3772 idx_dstsize = 1;
3773 idx_format = 2;
3774 info.argidx = 3;
3775 info.bounded = true;
3776 break;
3778 case BUILT_IN_SNPRINTF_CHK:
3779 // Signature:
3780 // __builtin___snprintf_chk (dst, size, ost, objsize, format, ...)
3781 idx_dstsize = 1;
3782 idx_objsize = 3;
3783 idx_format = 4;
3784 info.argidx = 5;
3785 info.bounded = true;
3786 break;
3788 case BUILT_IN_VSNPRINTF:
3789 // Signature:
3790 // __builtin_vsprintf (dst, size, format, va)
3791 idx_dstsize = 1;
3792 idx_format = 2;
3793 info.argidx = -1;
3794 info.bounded = true;
3795 break;
3797 case BUILT_IN_VSNPRINTF_CHK:
3798 // Signature:
3799 // __builtin___vsnprintf_chk (dst, size, ost, objsize, format, va)
3800 idx_dstsize = 1;
3801 idx_objsize = 3;
3802 idx_format = 4;
3803 info.argidx = -1;
3804 info.bounded = true;
3805 break;
3807 case BUILT_IN_VSPRINTF:
3808 // Signature:
3809 // __builtin_vsprintf (dst, format, va)
3810 idx_format = 1;
3811 info.argidx = -1;
3812 break;
3814 case BUILT_IN_VSPRINTF_CHK:
3815 // Signature:
3816 // __builtin___vsprintf_chk (dst, ost, objsize, format, va)
3817 idx_format = 3;
3818 idx_objsize = 2;
3819 info.argidx = -1;
3820 break;
3822 default:
3823 return false;
3826 /* Set the global warning level for this function. */
3827 warn_level = info.bounded ? warn_format_trunc : warn_format_overflow;
3829 /* The first argument is a pointer to the destination. */
3830 tree dstptr = gimple_call_arg (info.callstmt, 0);
3832 info.format = gimple_call_arg (info.callstmt, idx_format);
3834 /* True when the destination size is constant as opposed to the lower
3835 or upper bound of a range. */
3836 bool dstsize_cst_p = true;
3838 if (idx_dstsize == HOST_WIDE_INT_M1U)
3840 /* For non-bounded functions like sprintf, determine the size
3841 of the destination from the object or pointer passed to it
3842 as the first argument. */
3843 dstsize = get_destination_size (dstptr);
3845 else if (tree size = gimple_call_arg (info.callstmt, idx_dstsize))
3847 /* For bounded functions try to get the size argument. */
3849 if (TREE_CODE (size) == INTEGER_CST)
3851 dstsize = tree_to_uhwi (size);
3852 /* No object can be larger than SIZE_MAX bytes (half the address
3853 space) on the target.
3854 The functions are defined only for output of at most INT_MAX
3855 bytes. Specifying a bound in excess of that limit effectively
3856 defeats the bounds checking (and on some implementations such
3857 as Solaris cause the function to fail with EINVAL). */
3858 if (dstsize > target_size_max () / 2)
3860 /* Avoid warning if -Wstringop-overflow is specified since
3861 it also warns for the same thing though only for the
3862 checking built-ins. */
3863 if ((idx_objsize == HOST_WIDE_INT_M1U
3864 || !warn_stringop_overflow))
3865 warning_at (gimple_location (info.callstmt), info.warnopt (),
3866 "specified bound %wu exceeds maximum object size "
3867 "%wu",
3868 dstsize, target_size_max () / 2);
3870 else if (dstsize > target_int_max ())
3871 warning_at (gimple_location (info.callstmt), info.warnopt (),
3872 "specified bound %wu exceeds %<INT_MAX%>",
3873 dstsize);
3875 else if (TREE_CODE (size) == SSA_NAME)
3877 /* Try to determine the range of values of the argument
3878 and use the greater of the two at level 1 and the smaller
3879 of them at level 2. */
3880 wide_int min, max;
3881 enum value_range_type range_type
3882 = get_range_info (size, &min, &max);
3883 if (range_type == VR_RANGE)
3885 dstsize
3886 = (warn_level < 2
3887 ? wi::fits_uhwi_p (max) ? max.to_uhwi () : max.to_shwi ()
3888 : wi::fits_uhwi_p (min) ? min.to_uhwi () : min.to_shwi ());
3891 /* The destination size is not constant. If the function is
3892 bounded (e.g., snprintf) a lower bound of zero doesn't
3893 necessarily imply it can be eliminated. */
3894 dstsize_cst_p = false;
3898 if (idx_objsize != HOST_WIDE_INT_M1U)
3899 if (tree size = gimple_call_arg (info.callstmt, idx_objsize))
3900 if (tree_fits_uhwi_p (size))
3901 objsize = tree_to_uhwi (size);
3903 if (info.bounded && !dstsize)
3905 /* As a special case, when the explicitly specified destination
3906 size argument (to a bounded function like snprintf) is zero
3907 it is a request to determine the number of bytes on output
3908 without actually producing any. Pretend the size is
3909 unlimited in this case. */
3910 info.objsize = HOST_WIDE_INT_MAX;
3911 info.nowrite = dstsize_cst_p;
3913 else
3915 /* For calls to non-bounded functions or to those of bounded
3916 functions with a non-zero size, warn if the destination
3917 pointer is null. */
3918 if (integer_zerop (dstptr))
3920 /* This is diagnosed with -Wformat only when the null is a constant
3921 pointer. The warning here diagnoses instances where the pointer
3922 is not constant. */
3923 location_t loc = gimple_location (info.callstmt);
3924 warning_at (EXPR_LOC_OR_LOC (dstptr, loc),
3925 info.warnopt (), "null destination pointer");
3926 return false;
3929 /* Set the object size to the smaller of the two arguments
3930 of both have been specified and they're not equal. */
3931 info.objsize = dstsize < objsize ? dstsize : objsize;
3933 if (info.bounded
3934 && dstsize < target_size_max () / 2 && objsize < dstsize
3935 /* Avoid warning if -Wstringop-overflow is specified since
3936 it also warns for the same thing though only for the
3937 checking built-ins. */
3938 && (idx_objsize == HOST_WIDE_INT_M1U
3939 || !warn_stringop_overflow))
3941 warning_at (gimple_location (info.callstmt), info.warnopt (),
3942 "specified bound %wu exceeds the size %wu "
3943 "of the destination object", dstsize, objsize);
3947 if (integer_zerop (info.format))
3949 /* This is diagnosed with -Wformat only when the null is a constant
3950 pointer. The warning here diagnoses instances where the pointer
3951 is not constant. */
3952 location_t loc = gimple_location (info.callstmt);
3953 warning_at (EXPR_LOC_OR_LOC (info.format, loc),
3954 info.warnopt (), "null format string");
3955 return false;
3958 info.fmtstr = get_format_string (info.format, &info.fmtloc);
3959 if (!info.fmtstr)
3960 return false;
3962 /* The result is the number of bytes output by the formatted function,
3963 including the terminating NUL. */
3964 format_result res = format_result ();
3966 bool success = compute_format_length (info, &res);
3968 /* When optimizing and the printf return value optimization is enabled,
3969 attempt to substitute the computed result for the return value of
3970 the call. Avoid this optimization when -frounding-math is in effect
3971 and the format string contains a floating point directive. */
3972 bool call_removed = false;
3973 if (success && optimize > 0)
3975 /* Save a copy of the iterator pointing at the call. The iterator
3976 may change to point past the call in try_substitute_return_value
3977 but the original value is needed in try_simplify_call. */
3978 gimple_stmt_iterator gsi_call = *gsi;
3980 if (flag_printf_return_value
3981 && (!flag_rounding_math || !res.floating))
3982 call_removed = try_substitute_return_value (gsi, info, res);
3984 if (!call_removed)
3985 try_simplify_call (&gsi_call, info, res);
3988 return call_removed;
3991 /* Execute the pass for function FUN. */
3993 unsigned int
3994 pass_sprintf_length::execute (function *fun)
3996 init_target_to_host_charmap ();
3998 basic_block bb;
3999 FOR_EACH_BB_FN (bb, fun)
4001 for (gimple_stmt_iterator si = gsi_start_bb (bb); !gsi_end_p (si); )
4003 /* Iterate over statements, looking for function calls. */
4004 gimple *stmt = gsi_stmt (si);
4006 if (is_gimple_call (stmt) && handle_gimple_call (&si))
4007 /* If handle_gimple_call returns true, the iterator is
4008 already pointing to the next statement. */
4009 continue;
4011 gsi_next (&si);
4015 /* Clean up object size info. */
4016 fini_object_sizes ();
4018 return 0;
4021 } /* Unnamed namespace. */
4023 /* Return a pointer to a pass object newly constructed from the context
4024 CTXT. */
4026 gimple_opt_pass *
4027 make_pass_sprintf_length (gcc::context *ctxt)
4029 return new pass_sprintf_length (ctxt);