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
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
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. */
50 #include "coretypes.h"
54 #include "tree-pass.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"
62 #include "tree-object-size.h"
65 #include "tree-ssa-propagate.h"
71 #include "stor-layout.h"
79 #include "substring-locations.h"
80 #include "diagnostic.h"
82 /* The likely worst case value of MB_LEN_MAX for the target, large enough
83 for UTF-8. Ideally, this would be obtained by a target hook if it were
84 to be used for optimization but it's good enough as is for warnings. */
85 #define target_mb_len_max() 6
87 /* The maximum number of bytes a single non-string directive can result
88 in. This is the result of printf("%.*Lf", INT_MAX, -LDBL_MAX) for
89 LDBL_MAX_10_EXP of 4932. */
90 #define IEEE_MAX_10_EXP 4932
91 #define target_dir_max() (target_int_max () + IEEE_MAX_10_EXP + 2)
95 const pass_data pass_data_sprintf_length
= {
96 GIMPLE_PASS
, // pass type
97 "printf-return-value", // pass name
98 OPTGROUP_NONE
, // optinfo_flags
100 PROP_cfg
, // properties_required
101 0, // properties_provided
102 0, // properties_destroyed
103 0, // properties_start
104 0, // properties_finish
107 /* Set to the warning level for the current function which is equal
108 either to warn_format_trunc for bounded functions or to
109 warn_format_overflow otherwise. */
111 static int warn_level
;
113 struct format_result
;
115 class pass_sprintf_length
: public gimple_opt_pass
117 bool fold_return_value
;
120 pass_sprintf_length (gcc::context
*ctxt
)
121 : gimple_opt_pass (pass_data_sprintf_length
, ctxt
),
122 fold_return_value (false)
125 opt_pass
* clone () { return new pass_sprintf_length (m_ctxt
); }
127 virtual bool gate (function
*);
129 virtual unsigned int execute (function
*);
131 void set_pass_param (unsigned int n
, bool param
)
134 fold_return_value
= param
;
137 bool handle_gimple_call (gimple_stmt_iterator
*);
140 bool compute_format_length (call_info
&, format_result
*);
144 pass_sprintf_length::gate (function
*)
146 /* Run the pass iff -Warn-format-overflow or -Warn-format-truncation
147 is specified and either not optimizing and the pass is being invoked
148 early, or when optimizing and the pass is being invoked during
149 optimization (i.e., "late"). */
150 return ((warn_format_overflow
> 0
151 || warn_format_trunc
> 0
152 || flag_printf_return_value
)
153 && (optimize
> 0) == fold_return_value
);
156 /* The minimum, maximum, likely, and unlikely maximum number of bytes
157 of output either a formatting function or an individual directive
162 /* The absolute minimum number of bytes. The result of a successful
163 conversion is guaranteed to be no less than this. (An erroneous
164 conversion can be indicated by MIN > HOST_WIDE_INT_MAX.) */
165 unsigned HOST_WIDE_INT min
;
166 /* The likely maximum result that is used in diagnostics. In most
167 cases MAX is the same as the worst case UNLIKELY result. */
168 unsigned HOST_WIDE_INT max
;
169 /* The likely result used to trigger diagnostics. For conversions
170 that result in a range of bytes [MIN, MAX], LIKELY is somewhere
172 unsigned HOST_WIDE_INT likely
;
173 /* In rare cases (e.g., for nultibyte characters) UNLIKELY gives
174 the worst cases maximum result of a directive. In most cases
175 UNLIKELY == MAX. UNLIKELY is used to control the return value
176 optimization but not in diagnostics. */
177 unsigned HOST_WIDE_INT unlikely
;
180 /* The result of a call to a formatted function. */
184 /* Range of characters written by the formatted function.
185 Setting the minimum to HOST_WIDE_INT_MAX disables all
186 length tracking for the remainder of the format string. */
189 /* True when the range above is obtained from known values of
190 directive arguments, or bounds on the amount of output such
191 as width and precision, and not the result of heuristics that
192 depend on warning levels. It's used to issue stricter diagnostics
193 in cases where strings of unknown lengths are bounded by the arrays
194 they are determined to refer to. KNOWNRANGE must not be used for
195 the return value optimization. */
198 /* True if no individual directive resulted in more than 4095 bytes
199 of output (the total NUMBER_CHARS_{MIN,MAX} might be greater).
200 Implementations are not required to handle directives that produce
201 more than 4K bytes (leading to undefined behavior) and so when one
202 is found it disables the return value optimization. */
205 /* True when a floating point directive has been seen in the format
209 /* True when an intermediate result has caused a warning. Used to
210 avoid issuing duplicate warnings while finishing the processing
211 of a call. WARNED also disables the return value optimization. */
214 /* Preincrement the number of output characters by 1. */
215 format_result
& operator++ ()
220 /* Postincrement the number of output characters by 1. */
221 format_result
operator++ (int)
223 format_result
prev (*this);
228 /* Increment the number of output characters by N. */
229 format_result
& operator+= (unsigned HOST_WIDE_INT
);
233 format_result::operator+= (unsigned HOST_WIDE_INT n
)
235 gcc_assert (n
< HOST_WIDE_INT_MAX
);
237 if (range
.min
< HOST_WIDE_INT_MAX
)
240 if (range
.max
< HOST_WIDE_INT_MAX
)
243 if (range
.likely
< HOST_WIDE_INT_MAX
)
246 if (range
.unlikely
< HOST_WIDE_INT_MAX
)
252 /* Return the value of INT_MIN for the target. */
254 static inline HOST_WIDE_INT
257 return tree_to_shwi (TYPE_MIN_VALUE (integer_type_node
));
260 /* Return the value of INT_MAX for the target. */
262 static inline unsigned HOST_WIDE_INT
265 return tree_to_uhwi (TYPE_MAX_VALUE (integer_type_node
));
268 /* Return the value of SIZE_MAX for the target. */
270 static inline unsigned HOST_WIDE_INT
273 return tree_to_uhwi (TYPE_MAX_VALUE (size_type_node
));
276 /* Return the constant initial value of DECL if available or DECL
277 otherwise. Same as the synonymous function in c/c-typeck.c. */
280 decl_constant_value (tree decl
)
282 if (/* Don't change a variable array bound or initial value to a constant
283 in a place where a variable is invalid. Note that DECL_INITIAL
284 isn't valid for a PARM_DECL. */
285 current_function_decl
!= 0
286 && TREE_CODE (decl
) != PARM_DECL
287 && !TREE_THIS_VOLATILE (decl
)
288 && TREE_READONLY (decl
)
289 && DECL_INITIAL (decl
) != 0
290 && TREE_CODE (DECL_INITIAL (decl
)) != ERROR_MARK
291 /* This is invalid if initial value is not constant.
292 If it has either a function call, a memory reference,
293 or a variable, then re-evaluating it could give different results. */
294 && TREE_CONSTANT (DECL_INITIAL (decl
))
295 /* Check for cases where this is sub-optimal, even though valid. */
296 && TREE_CODE (DECL_INITIAL (decl
)) != CONSTRUCTOR
)
297 return DECL_INITIAL (decl
);
301 /* Given FORMAT, set *PLOC to the source location of the format string
302 and return the format string if it is known or null otherwise. */
305 get_format_string (tree format
, location_t
*ploc
)
309 /* Pull out a constant value if the front end didn't. */
310 format
= decl_constant_value (format
);
314 if (integer_zerop (format
))
316 /* FIXME: Diagnose null format string if it hasn't been diagnosed
317 by -Wformat (the latter diagnoses only nul pointer constants,
318 this pass can do better). */
322 HOST_WIDE_INT offset
= 0;
324 if (TREE_CODE (format
) == POINTER_PLUS_EXPR
)
326 tree arg0
= TREE_OPERAND (format
, 0);
327 tree arg1
= TREE_OPERAND (format
, 1);
331 if (TREE_CODE (arg1
) != INTEGER_CST
)
336 /* POINTER_PLUS_EXPR offsets are to be interpreted signed. */
337 if (!cst_and_fits_in_hwi (arg1
))
340 offset
= int_cst_value (arg1
);
343 if (TREE_CODE (format
) != ADDR_EXPR
)
346 *ploc
= EXPR_LOC_OR_LOC (format
, input_location
);
348 format
= TREE_OPERAND (format
, 0);
350 if (TREE_CODE (format
) == ARRAY_REF
351 && tree_fits_shwi_p (TREE_OPERAND (format
, 1))
352 && (offset
+= tree_to_shwi (TREE_OPERAND (format
, 1))) >= 0)
353 format
= TREE_OPERAND (format
, 0);
359 tree array_size
= NULL_TREE
;
362 && TREE_CODE (TREE_TYPE (format
)) == ARRAY_TYPE
363 && (array_init
= decl_constant_value (format
)) != format
364 && TREE_CODE (array_init
) == STRING_CST
)
366 /* Extract the string constant initializer. Note that this may
367 include a trailing NUL character that is not in the array (e.g.
368 const char a[3] = "foo";). */
369 array_size
= DECL_SIZE_UNIT (format
);
373 if (TREE_CODE (format
) != STRING_CST
)
376 if (TYPE_MAIN_VARIANT (TREE_TYPE (TREE_TYPE (format
))) != char_type_node
)
378 /* Wide format string. */
382 const char *fmtstr
= TREE_STRING_POINTER (format
);
383 unsigned fmtlen
= TREE_STRING_LENGTH (format
);
387 /* Variable length arrays can't be initialized. */
388 gcc_assert (TREE_CODE (array_size
) == INTEGER_CST
);
390 if (tree_fits_shwi_p (array_size
))
392 HOST_WIDE_INT array_size_value
= tree_to_shwi (array_size
);
393 if (array_size_value
> 0
394 && array_size_value
== (int) array_size_value
395 && fmtlen
> array_size_value
)
396 fmtlen
= array_size_value
;
401 if (offset
>= fmtlen
)
408 if (fmtlen
< 1 || fmtstr
[--fmtlen
] != 0)
410 /* FIXME: Diagnose an unterminated format string if it hasn't been
411 diagnosed by -Wformat. Similarly to a null format pointer,
412 -Wformay diagnoses only nul pointer constants, this pass can
420 /* The format_warning_at_substring function is not used here in a way
421 that makes using attribute format viable. Suppress the warning. */
423 #pragma GCC diagnostic push
424 #pragma GCC diagnostic ignored "-Wsuggest-attribute=format"
426 /* For convenience and brevity. */
429 (* const fmtwarn
) (const substring_loc
&, const source_range
*,
430 const char *, int, const char *, ...)
431 = format_warning_at_substring
;
433 /* Format length modifiers. */
438 FMT_LEN_hh
, // char argument
441 FMT_LEN_ll
, // long long
442 FMT_LEN_L
, // long double (and GNU long long)
444 FMT_LEN_t
, // ptrdiff_t
445 FMT_LEN_j
// intmax_t
449 /* Description of the result of conversion either of a single directive
450 or the whole format string. */
454 /* Construct a FMTRESULT object with all counters initialized
455 to MIN. KNOWNRANGE is set when MIN is valid. */
456 fmtresult (unsigned HOST_WIDE_INT min
= HOST_WIDE_INT_MAX
)
457 : argmin (), argmax (),
458 knownrange (min
< HOST_WIDE_INT_MAX
),
464 range
.unlikely
= min
;
467 /* Construct a FMTRESULT object with MIN, MAX, and LIKELY counters.
468 KNOWNRANGE is set when both MIN and MAX are valid. */
469 fmtresult (unsigned HOST_WIDE_INT min
, unsigned HOST_WIDE_INT max
,
470 unsigned HOST_WIDE_INT likely
= HOST_WIDE_INT_MAX
)
471 : argmin (), argmax (),
472 knownrange (min
< HOST_WIDE_INT_MAX
&& max
< HOST_WIDE_INT_MAX
),
477 range
.likely
= max
< likely
? min
: likely
;
478 range
.unlikely
= max
;
481 /* Adjust result upward to reflect the RANGE of values the specified
482 width or precision is known to be in. */
483 fmtresult
& adjust_for_width_or_precision (const HOST_WIDE_INT
[2],
485 unsigned = 0, unsigned = 0);
487 /* Return the maximum number of decimal digits a value of TYPE
488 formats as on output. */
489 static unsigned type_max_digits (tree
, int);
491 /* The range a directive's argument is in. */
494 /* The minimum and maximum number of bytes that a directive
495 results in on output for an argument in the range above. */
498 /* True when the range above is obtained from a known value of
499 a directive's argument or its bounds and not the result of
500 heuristics that depend on warning levels. */
503 /* True when the argument is a null pointer. */
507 /* Adjust result upward to reflect the range ADJUST of values the
508 specified width or precision is known to be in. When non-null,
509 TYPE denotes the type of the directive whose result is being
510 adjusted, BASE gives the base of the directive (octal, decimal,
511 or hex), and ADJ denotes the additional adjustment to the LIKELY
512 counter that may need to be added when ADJUST is a range. */
515 fmtresult::adjust_for_width_or_precision (const HOST_WIDE_INT adjust
[2],
516 tree type
/* = NULL_TREE */,
517 unsigned base
/* = 0 */,
518 unsigned adj
/* = 0 */)
520 bool minadjusted
= false;
522 /* Adjust the minimum and likely counters. */
525 if (range
.min
< (unsigned HOST_WIDE_INT
)adjust
[0])
527 range
.min
= adjust
[0];
531 /* Adjust the likely counter. */
532 if (range
.likely
< range
.min
)
533 range
.likely
= range
.min
;
535 else if (adjust
[0] == target_int_min ()
536 && (unsigned HOST_WIDE_INT
)adjust
[1] == target_int_max ())
539 /* Adjust the maximum counter. */
542 if (range
.max
< (unsigned HOST_WIDE_INT
)adjust
[1])
544 range
.max
= adjust
[1];
546 /* Set KNOWNRANGE if both the minimum and maximum have been
547 adjusted. Otherwise leave it at what it was before. */
548 knownrange
= minadjusted
;
552 if (warn_level
> 1 && type
)
554 /* For large non-constant width or precision whose range spans
555 the maximum number of digits produced by the directive for
556 any argument, set the likely number of bytes to be at most
557 the number digits plus other adjustment determined by the
558 caller (one for sign or two for the hexadecimal "0x"
560 unsigned dirdigs
= type_max_digits (type
, base
);
561 if (adjust
[0] < dirdigs
&& dirdigs
< adjust
[1]
562 && range
.likely
< dirdigs
)
563 range
.likely
= dirdigs
+ adj
;
565 else if (range
.likely
< (range
.min
? range
.min
: 1))
567 /* Conservatively, set LIKELY to at least MIN but no less than
568 1 unless MAX is zero. */
569 range
.likely
= (range
.min
571 : range
.max
&& (range
.max
< HOST_WIDE_INT_MAX
572 || warn_level
> 1) ? 1 : 0);
575 /* Finally adjust the unlikely counter to be at least as large as
577 if (range
.unlikely
< range
.max
)
578 range
.unlikely
= range
.max
;
583 /* Return the maximum number of digits a value of TYPE formats in
584 BASE on output, not counting base prefix . */
587 fmtresult::type_max_digits (tree type
, int base
)
589 unsigned prec
= TYPE_PRECISION (type
);
591 return (prec
+ 2) / 3;
596 /* Decimal approximation: yields 3, 5, 10, and 20 for precision
597 of 8, 16, 32, and 64 bits. */
598 return prec
* 301 / 1000 + 1;
602 get_int_range (tree
, tree
, HOST_WIDE_INT
*, HOST_WIDE_INT
*,
603 bool, HOST_WIDE_INT
);
605 /* Description of a format directive. A directive is either a plain
606 string or a conversion specification that starts with '%'. */
610 /* The 1-based directive number (for debugging). */
613 /* The first character of the directive and its length. */
617 /* A bitmap of flags, one for each character. */
618 unsigned flags
[256 / sizeof (int)];
620 /* The range of values of the specified width, or -1 if not specified. */
621 HOST_WIDE_INT width
[2];
622 /* The range of values of the specified precision, or -1 if not
624 HOST_WIDE_INT prec
[2];
626 /* Length modifier. */
627 format_lengths modifier
;
629 /* Format specifier character. */
632 /* The argument of the directive or null when the directive doesn't
633 take one or when none is available (such as for vararg functions). */
636 /* Format conversion function that given a directive and an argument
637 returns the formatting result. */
638 fmtresult (*fmtfunc
) (const directive
&, tree
);
640 /* Return True when a the format flag CHR has been used. */
641 bool get_flag (char chr
) const
643 unsigned char c
= chr
& 0xff;
644 return (flags
[c
/ (CHAR_BIT
* sizeof *flags
)]
645 & (1U << (c
% (CHAR_BIT
* sizeof *flags
))));
648 /* Make a record of the format flag CHR having been used. */
649 void set_flag (char chr
)
651 unsigned char c
= chr
& 0xff;
652 flags
[c
/ (CHAR_BIT
* sizeof *flags
)]
653 |= (1U << (c
% (CHAR_BIT
* sizeof *flags
)));
656 /* Reset the format flag CHR. */
657 void clear_flag (char chr
)
659 unsigned char c
= chr
& 0xff;
660 flags
[c
/ (CHAR_BIT
* sizeof *flags
)]
661 &= ~(1U << (c
% (CHAR_BIT
* sizeof *flags
)));
664 /* Set both bounds of the width range to VAL. */
665 void set_width (HOST_WIDE_INT val
)
667 width
[0] = width
[1] = val
;
670 /* Set the width range according to ARG, with both bounds being
671 no less than 0. For a constant ARG set both bounds to its value
672 or 0, whichever is greater. For a non-constant ARG in some range
673 set width to its range adjusting each bound to -1 if it's less.
674 For an indeterminate ARG set width to [0, INT_MAX]. */
675 void set_width (tree arg
)
677 get_int_range (arg
, integer_type_node
, width
, width
+ 1, true, 0);
680 /* Set both bounds of the precision range to VAL. */
681 void set_precision (HOST_WIDE_INT val
)
683 prec
[0] = prec
[1] = val
;
686 /* Set the precision range according to ARG, with both bounds being
687 no less than -1. For a constant ARG set both bounds to its value
688 or -1 whichever is greater. For a non-constant ARG in some range
689 set precision to its range adjusting each bound to -1 if it's less.
690 For an indeterminate ARG set precision to [-1, INT_MAX]. */
691 void set_precision (tree arg
)
693 get_int_range (arg
, integer_type_node
, prec
, prec
+ 1, false, -1);
696 /* Return true if both width and precision are known to be
697 either constant or in some range, false otherwise. */
698 bool known_width_and_precision () const
700 return ((width
[1] < 0
701 || (unsigned HOST_WIDE_INT
)width
[1] <= target_int_max ())
703 || (unsigned HOST_WIDE_INT
)prec
[1] < target_int_max ()));
707 /* Return the logarithm of X in BASE. */
710 ilog (unsigned HOST_WIDE_INT x
, int base
)
721 /* Return the number of bytes resulting from converting into a string
722 the INTEGER_CST tree node X in BASE with a minimum of PREC digits.
723 PLUS indicates whether 1 for a plus sign should be added for positive
724 numbers, and PREFIX whether the length of an octal ('O') or hexadecimal
725 ('0x') prefix should be added for nonzero numbers. Return -1 if X cannot
729 tree_digits (tree x
, int base
, HOST_WIDE_INT prec
, bool plus
, bool prefix
)
731 unsigned HOST_WIDE_INT absval
;
735 if (TYPE_UNSIGNED (TREE_TYPE (x
)))
737 if (tree_fits_uhwi_p (x
))
739 absval
= tree_to_uhwi (x
);
747 if (tree_fits_shwi_p (x
))
749 HOST_WIDE_INT i
= tree_to_shwi (x
);
750 if (HOST_WIDE_INT_MIN
== i
)
752 /* Avoid undefined behavior due to negating a minimum. */
753 absval
= HOST_WIDE_INT_MAX
;
771 int ndigs
= ilog (absval
, base
);
773 res
+= prec
< ndigs
? ndigs
: prec
;
775 /* Adjust a non-zero value for the base prefix, either hexadecimal,
776 or, unless precision has resulted in a leading zero, also octal. */
777 if (prefix
&& absval
&& (base
== 16 || prec
<= ndigs
))
788 /* Given the formatting result described by RES and NAVAIL, the number
789 of available in the destination, return the range of bytes remaining
790 in the destination. */
792 static inline result_range
793 bytes_remaining (unsigned HOST_WIDE_INT navail
, const format_result
&res
)
797 if (HOST_WIDE_INT_MAX
<= navail
)
799 range
.min
= range
.max
= range
.likely
= range
.unlikely
= navail
;
803 /* The lower bound of the available range is the available size
804 minus the maximum output size, and the upper bound is the size
805 minus the minimum. */
806 range
.max
= res
.range
.min
< navail
? navail
- res
.range
.min
: 0;
808 range
.likely
= res
.range
.likely
< navail
? navail
- res
.range
.likely
: 0;
810 if (res
.range
.max
< HOST_WIDE_INT_MAX
)
811 range
.min
= res
.range
.max
< navail
? navail
- res
.range
.max
: 0;
813 range
.min
= range
.likely
;
815 range
.unlikely
= (res
.range
.unlikely
< navail
816 ? navail
- res
.range
.unlikely
: 0);
821 /* Description of a call to a formatted function. */
823 struct pass_sprintf_length::call_info
825 /* Function call statement. */
828 /* Function called. */
831 /* Called built-in function code. */
832 built_in_function fncode
;
834 /* Format argument and format string extracted from it. */
838 /* The location of the format argument. */
841 /* The destination object size for __builtin___xxx_chk functions
842 typically determined by __builtin_object_size, or -1 if unknown. */
843 unsigned HOST_WIDE_INT objsize
;
845 /* Number of the first variable argument. */
846 unsigned HOST_WIDE_INT argidx
;
848 /* True for functions like snprintf that specify the size of
849 the destination, false for others like sprintf that don't. */
852 /* True for bounded functions like snprintf that specify a zero-size
853 buffer as a request to compute the size of output without actually
854 writing any. NOWRITE is cleared in response to the %n directive
855 which has side-effects similar to writing output. */
858 /* Return true if the called function's return value is used. */
859 bool retval_used () const
861 return gimple_get_lhs (callstmt
);
864 /* Return the warning option corresponding to the called function. */
867 return bounded
? OPT_Wformat_truncation_
: OPT_Wformat_overflow_
;
871 /* Return the result of formatting a no-op directive (such as '%n'). */
874 format_none (const directive
&, tree
)
880 /* Return the result of formatting the '%%' directive. */
883 format_percent (const directive
&, tree
)
890 /* Compute intmax_type_node and uintmax_type_node similarly to how
891 tree.c builds size_type_node. */
894 build_intmax_type_nodes (tree
*pintmax
, tree
*puintmax
)
896 if (strcmp (UINTMAX_TYPE
, "unsigned int") == 0)
898 *pintmax
= integer_type_node
;
899 *puintmax
= unsigned_type_node
;
901 else if (strcmp (UINTMAX_TYPE
, "long unsigned int") == 0)
903 *pintmax
= long_integer_type_node
;
904 *puintmax
= long_unsigned_type_node
;
906 else if (strcmp (UINTMAX_TYPE
, "long long unsigned int") == 0)
908 *pintmax
= long_long_integer_type_node
;
909 *puintmax
= long_long_unsigned_type_node
;
913 for (int i
= 0; i
< NUM_INT_N_ENTS
; i
++)
914 if (int_n_enabled_p
[i
])
917 sprintf (name
, "__int%d unsigned", int_n_data
[i
].bitsize
);
919 if (strcmp (name
, UINTMAX_TYPE
) == 0)
921 *pintmax
= int_n_trees
[i
].signed_type
;
922 *puintmax
= int_n_trees
[i
].unsigned_type
;
930 /* Determine the range [*PMIN, *PMAX] that the expression ARG of TYPE
931 is in. Return true when the range is a subrange of that of TYPE.
932 Whn ARG is null it is as if it had the full range of TYPE.
933 When ABSOLUTE is true the range reflects the absolute value of
934 the argument. When ABSOLUTE is false, negative bounds of
935 the determined range are replaced with NEGBOUND. */
938 get_int_range (tree arg
, tree type
, HOST_WIDE_INT
*pmin
, HOST_WIDE_INT
*pmax
,
939 bool absolute
, HOST_WIDE_INT negbound
)
941 bool knownrange
= false;
945 *pmin
= (TYPE_UNSIGNED (type
)
946 ? tree_to_uhwi (TYPE_MIN_VALUE (type
))
947 : tree_to_shwi (TYPE_MIN_VALUE (type
)));
948 *pmax
= tree_to_uhwi (TYPE_MAX_VALUE (type
));
950 else if (TREE_CODE (arg
) == INTEGER_CST
)
952 /* For a constant argument return its value adjusted as specified
953 by NEGATIVE and NEGBOUND and return true to indicate that the
955 *pmin
= tree_fits_shwi_p (arg
) ? tree_to_shwi (arg
) : tree_to_uhwi (arg
);
961 /* True if the argument's range cannot be determined. */
964 type
= TREE_TYPE (arg
);
966 if (TREE_CODE (arg
) == SSA_NAME
967 && TREE_CODE (type
) == INTEGER_TYPE
)
969 /* Try to determine the range of values of the integer argument. */
971 enum value_range_type range_type
= get_range_info (arg
, &min
, &max
);
972 if (range_type
== VR_RANGE
)
974 HOST_WIDE_INT type_min
975 = (TYPE_UNSIGNED (type
)
976 ? tree_to_uhwi (TYPE_MIN_VALUE (type
))
977 : tree_to_shwi (TYPE_MIN_VALUE (type
)));
979 HOST_WIDE_INT type_max
= tree_to_uhwi (TYPE_MAX_VALUE (type
));
981 *pmin
= min
.to_shwi ();
982 *pmax
= max
.to_shwi ();
984 /* Return true if the adjusted range is a subrange of
985 the full range of the argument's type. */
986 knownrange
= type_min
< *pmin
|| *pmax
< type_max
;
992 /* Handle an argument with an unknown range as if none had been
995 return get_int_range (NULL_TREE
, type
, pmin
, pmax
, absolute
, negbound
);
998 /* Adjust each bound as specified by ABSOLUTE and NEGBOUND. */
1004 *pmin
= *pmax
= -*pmin
;
1007 HOST_WIDE_INT tmp
= -*pmin
;
1014 else if (*pmin
< negbound
)
1020 /* With the range [*ARGMIN, *ARGMAX] of an integer directive's actual
1021 argument, due to the conversion from either *ARGMIN or *ARGMAX to
1022 the type of the directive's formal argument it's possible for both
1023 to result in the same number of bytes or a range of bytes that's
1024 less than the number of bytes that would result from formatting
1025 some other value in the range [*ARGMIN, *ARGMAX]. This can be
1026 determined by checking for the actual argument being in the range
1027 of the type of the directive. If it isn't it must be assumed to
1028 take on the full range of the directive's type.
1029 Return true when the range has been adjusted to the full range
1030 of DIRTYPE, and false otherwise. */
1033 adjust_range_for_overflow (tree dirtype
, tree
*argmin
, tree
*argmax
)
1035 tree argtype
= TREE_TYPE (*argmin
);
1036 unsigned argprec
= TYPE_PRECISION (argtype
);
1037 unsigned dirprec
= TYPE_PRECISION (dirtype
);
1039 /* If the actual argument and the directive's argument have the same
1040 precision and sign there can be no overflow and so there is nothing
1042 if (argprec
== dirprec
&& TYPE_SIGN (argtype
) == TYPE_SIGN (dirtype
))
1045 /* The logic below was inspired/lifted from the CONVERT_EXPR_CODE_P
1046 branch in the extract_range_from_unary_expr function in tree-vrp.c. */
1048 if (TREE_CODE (*argmin
) == INTEGER_CST
1049 && TREE_CODE (*argmax
) == INTEGER_CST
1050 && (dirprec
>= argprec
1051 || integer_zerop (int_const_binop (RSHIFT_EXPR
,
1052 int_const_binop (MINUS_EXPR
,
1055 size_int (dirprec
)))))
1057 *argmin
= force_fit_type (dirtype
, wi::to_widest (*argmin
), 0, false);
1058 *argmax
= force_fit_type (dirtype
, wi::to_widest (*argmax
), 0, false);
1060 /* If *ARGMIN is still less than *ARGMAX the conversion above
1061 is safe. Otherwise, it has overflowed and would be unsafe. */
1062 if (tree_int_cst_le (*argmin
, *argmax
))
1066 *argmin
= TYPE_MIN_VALUE (dirtype
);
1067 *argmax
= TYPE_MAX_VALUE (dirtype
);
1071 /* Return a range representing the minimum and maximum number of bytes
1072 that the format directive DIR will output for any argument given
1073 the WIDTH and PRECISION (extracted from DIR). This function is
1074 used when the directive argument or its value isn't known. */
1077 format_integer (const directive
&dir
, tree arg
)
1079 tree intmax_type_node
;
1080 tree uintmax_type_node
;
1082 /* Base to format the number in. */
1085 /* True when a conversion is preceded by a prefix indicating the base
1086 of the argument (octal or hexadecimal). */
1087 bool maybebase
= dir
.get_flag ('#');
1089 /* True when a signed conversion is preceded by a sign or space. */
1090 bool maybesign
= false;
1092 /* True for signed conversions (i.e., 'd' and 'i'). */
1095 switch (dir
.specifier
)
1099 /* Space and '+' are only meaningful for signed conversions. */
1100 maybesign
= dir
.get_flag (' ') | dir
.get_flag ('+');
1118 /* The type of the "formal" argument expected by the directive. */
1119 tree dirtype
= NULL_TREE
;
1121 /* Determine the expected type of the argument from the length
1123 switch (dir
.modifier
)
1126 if (dir
.specifier
== 'p')
1127 dirtype
= ptr_type_node
;
1129 dirtype
= sign
? integer_type_node
: unsigned_type_node
;
1133 dirtype
= sign
? short_integer_type_node
: short_unsigned_type_node
;
1137 dirtype
= sign
? signed_char_type_node
: unsigned_char_type_node
;
1141 dirtype
= sign
? long_integer_type_node
: long_unsigned_type_node
;
1147 ? long_long_integer_type_node
1148 : long_long_unsigned_type_node
);
1152 dirtype
= signed_or_unsigned_type_for (!sign
, size_type_node
);
1156 dirtype
= signed_or_unsigned_type_for (!sign
, ptrdiff_type_node
);
1160 build_intmax_type_nodes (&intmax_type_node
, &uintmax_type_node
);
1161 dirtype
= sign
? intmax_type_node
: uintmax_type_node
;
1165 return fmtresult ();
1168 /* The type of the argument to the directive, either deduced from
1169 the actual non-constant argument if one is known, or from
1170 the directive itself when none has been provided because it's
1172 tree argtype
= NULL_TREE
;
1176 /* When the argument has not been provided, use the type of
1177 the directive's argument as an approximation. This will
1178 result in false positives for directives like %i with
1179 arguments with smaller precision (such as short or char). */
1182 else if (TREE_CODE (arg
) == INTEGER_CST
)
1184 /* When a constant argument has been provided use its value
1185 rather than type to determine the length of the output. */
1188 if ((dir
.prec
[0] <= 0 && dir
.prec
[1] >= 0) && integer_zerop (arg
))
1190 /* As a special case, a precision of zero with a zero argument
1191 results in zero bytes except in base 8 when the '#' flag is
1192 specified, and for signed conversions in base 8 and 10 when
1193 either the space or '+' flag has been specified and it results
1194 in just one byte (with width having the normal effect). This
1195 must extend to the case of a specified precision with
1196 an unknown value because it can be zero. */
1197 res
.range
.min
= ((base
== 8 && dir
.get_flag ('#')) || maybesign
);
1198 if (res
.range
.min
== 0 && dir
.prec
[0] != dir
.prec
[1])
1201 res
.range
.likely
= 1;
1205 res
.range
.max
= res
.range
.min
;
1206 res
.range
.likely
= res
.range
.min
;
1211 /* Convert the argument to the type of the directive. */
1212 arg
= fold_convert (dirtype
, arg
);
1214 res
.range
.min
= tree_digits (arg
, base
, dir
.prec
[0],
1215 maybesign
, maybebase
);
1216 if (dir
.prec
[0] == dir
.prec
[1])
1217 res
.range
.max
= res
.range
.min
;
1219 res
.range
.max
= tree_digits (arg
, base
, dir
.prec
[1],
1220 maybesign
, maybebase
);
1221 res
.range
.likely
= res
.range
.min
;
1224 res
.range
.unlikely
= res
.range
.max
;
1226 /* Bump up the counters if WIDTH is greater than LEN. */
1227 res
.adjust_for_width_or_precision (dir
.width
, dirtype
, base
,
1228 (sign
| maybebase
) + (base
== 16));
1229 /* Bump up the counters again if PRECision is greater still. */
1230 res
.adjust_for_width_or_precision (dir
.prec
, dirtype
, base
,
1231 (sign
| maybebase
) + (base
== 16));
1235 else if (TREE_CODE (TREE_TYPE (arg
)) == INTEGER_TYPE
1236 || TREE_CODE (TREE_TYPE (arg
)) == POINTER_TYPE
)
1237 /* Determine the type of the provided non-constant argument. */
1238 argtype
= TREE_TYPE (arg
);
1240 /* Don't bother with invalid arguments since they likely would
1241 have already been diagnosed, and disable any further checking
1242 of the format string by returning [-1, -1]. */
1243 return fmtresult ();
1247 /* Using either the range the non-constant argument is in, or its
1248 type (either "formal" or actual), create a range of values that
1249 constrain the length of output given the warning level. */
1250 tree argmin
= NULL_TREE
;
1251 tree argmax
= NULL_TREE
;
1254 && TREE_CODE (arg
) == SSA_NAME
1255 && TREE_CODE (argtype
) == INTEGER_TYPE
)
1257 /* Try to determine the range of values of the integer argument
1258 (range information is not available for pointers). */
1260 enum value_range_type range_type
= get_range_info (arg
, &min
, &max
);
1261 if (range_type
== VR_RANGE
)
1263 argmin
= wide_int_to_tree (argtype
, min
);
1264 argmax
= wide_int_to_tree (argtype
, max
);
1266 /* Set KNOWNRANGE if the argument is in a known subrange
1267 of the directive's type and neither width nor precision
1268 is unknown. (KNOWNRANGE may be reset below). */
1270 = ((!tree_int_cst_equal (TYPE_MIN_VALUE (dirtype
), argmin
)
1271 || !tree_int_cst_equal (TYPE_MAX_VALUE (dirtype
), argmax
))
1272 && dir
.known_width_and_precision ());
1274 res
.argmin
= argmin
;
1275 res
.argmax
= argmax
;
1277 else if (range_type
== VR_ANTI_RANGE
)
1279 /* Handle anti-ranges if/when bug 71690 is resolved. */
1281 else if (range_type
== VR_VARYING
)
1283 /* The argument here may be the result of promoting the actual
1284 argument to int. Try to determine the type of the actual
1285 argument before promotion and narrow down its range that
1287 gimple
*def
= SSA_NAME_DEF_STMT (arg
);
1288 if (is_gimple_assign (def
))
1290 tree_code code
= gimple_assign_rhs_code (def
);
1291 if (code
== INTEGER_CST
)
1293 arg
= gimple_assign_rhs1 (def
);
1294 return format_integer (dir
, arg
);
1297 if (code
== NOP_EXPR
)
1299 tree type
= TREE_TYPE (gimple_assign_rhs1 (def
));
1300 if (TREE_CODE (type
) == INTEGER_TYPE
1301 || TREE_CODE (type
) == POINTER_TYPE
)
1310 if (TREE_CODE (argtype
) == POINTER_TYPE
)
1312 argmin
= build_int_cst (pointer_sized_int_node
, 0);
1313 argmax
= build_all_ones_cst (pointer_sized_int_node
);
1317 argmin
= TYPE_MIN_VALUE (argtype
);
1318 argmax
= TYPE_MAX_VALUE (argtype
);
1322 /* Clear KNOWNRANGE if the range has been adjusted to the maximum
1323 of the directive. If it has been cleared then since ARGMIN and/or
1324 ARGMAX have been adjusted also adjust the corresponding ARGMIN and
1325 ARGMAX in the result to include in diagnostics. */
1326 if (adjust_range_for_overflow (dirtype
, &argmin
, &argmax
))
1328 res
.knownrange
= false;
1329 res
.argmin
= argmin
;
1330 res
.argmax
= argmax
;
1333 /* Recursively compute the minimum and maximum from the known range. */
1334 if (TYPE_UNSIGNED (dirtype
) || tree_int_cst_sgn (argmin
) >= 0)
1336 /* For unsigned conversions/directives or signed when
1337 the minimum is positive, use the minimum and maximum to compute
1338 the shortest and longest output, respectively. */
1339 res
.range
.min
= format_integer (dir
, argmin
).range
.min
;
1340 res
.range
.max
= format_integer (dir
, argmax
).range
.max
;
1342 else if (tree_int_cst_sgn (argmax
) < 0)
1344 /* For signed conversions/directives if maximum is negative,
1345 use the minimum as the longest output and maximum as the
1347 res
.range
.min
= format_integer (dir
, argmax
).range
.min
;
1348 res
.range
.max
= format_integer (dir
, argmin
).range
.max
;
1352 /* Otherwise, 0 is inside of the range and minimum negative. Use 0
1353 as the shortest output and for the longest output compute the
1354 length of the output of both minimum and maximum and pick the
1356 unsigned HOST_WIDE_INT max1
= format_integer (dir
, argmin
).range
.max
;
1357 unsigned HOST_WIDE_INT max2
= format_integer (dir
, argmax
).range
.max
;
1358 res
.range
.min
= format_integer (dir
, integer_zero_node
).range
.min
;
1359 res
.range
.max
= MAX (max1
, max2
);
1362 /* If the range is known, use the maximum as the likely length. */
1364 res
.range
.likely
= res
.range
.max
;
1367 /* Otherwise, use the minimum. Except for the case where for %#x or
1368 %#o the minimum is just for a single value in the range (0) and
1369 for all other values it is something longer, like 0x1 or 01.
1370 Use the length for value 1 in that case instead as the likely
1372 res
.range
.likely
= res
.range
.min
;
1375 && (tree_int_cst_sgn (argmin
) < 0 || tree_int_cst_sgn (argmax
) > 0))
1377 if (res
.range
.min
== 1)
1378 res
.range
.likely
+= base
== 8 ? 1 : 2;
1379 else if (res
.range
.min
== 2
1381 && (dir
.width
[0] == 2 || dir
.prec
[0] == 2))
1386 res
.range
.unlikely
= res
.range
.max
;
1387 res
.adjust_for_width_or_precision (dir
.width
, dirtype
, base
,
1388 (sign
| maybebase
) + (base
== 16));
1389 res
.adjust_for_width_or_precision (dir
.prec
, dirtype
, base
,
1390 (sign
| maybebase
) + (base
== 16));
1395 /* Return the number of bytes that a format directive consisting of FLAGS,
1396 PRECision, format SPECification, and MPFR rounding specifier RNDSPEC,
1397 would result for argument X under ideal conditions (i.e., if PREC
1398 weren't excessive). MPFR 3.1 allocates large amounts of memory for
1399 values of PREC with large magnitude and can fail (see MPFR bug #21056).
1400 This function works around those problems. */
1402 static unsigned HOST_WIDE_INT
1403 get_mpfr_format_length (mpfr_ptr x
, const char *flags
, HOST_WIDE_INT prec
,
1404 char spec
, char rndspec
)
1408 HOST_WIDE_INT len
= strlen (flags
);
1411 memcpy (fmtstr
+ 1, flags
, len
);
1412 memcpy (fmtstr
+ 1 + len
, ".*R", 3);
1413 fmtstr
[len
+ 4] = rndspec
;
1414 fmtstr
[len
+ 5] = spec
;
1415 fmtstr
[len
+ 6] = '\0';
1417 spec
= TOUPPER (spec
);
1418 if (spec
== 'E' || spec
== 'F')
1420 /* For %e, specify the precision explicitly since mpfr_sprintf
1421 does its own thing just to be different (see MPFR bug 21088). */
1427 /* Avoid passing negative precisions with larger magnitude to MPFR
1428 to avoid exposing its bugs. (A negative precision is supposed
1434 HOST_WIDE_INT p
= prec
;
1436 if (spec
== 'G' && !strchr (flags
, '#'))
1438 /* For G/g without the pound flag, precision gives the maximum number
1439 of significant digits which is bounded by LDBL_MAX_10_EXP, or, for
1440 a 128 bit IEEE extended precision, 4932. Using twice as much here
1441 should be more than sufficient for any real format. */
1442 if ((IEEE_MAX_10_EXP
* 2) < prec
)
1443 prec
= IEEE_MAX_10_EXP
* 2;
1448 /* Cap precision arbitrarily at 1KB and add the difference
1449 (if any) to the MPFR result. */
1454 len
= mpfr_snprintf (NULL
, 0, fmtstr
, (int)p
, x
);
1456 /* Handle the unlikely (impossible?) error by returning more than
1457 the maximum dictated by the function's return type. */
1459 return target_dir_max () + 1;
1461 /* Adjust the return value by the difference. */
1468 /* Return the number of bytes to format using the format specifier
1469 SPEC and the precision PREC the largest value in the real floating
1472 static unsigned HOST_WIDE_INT
1473 format_floating_max (tree type
, char spec
, HOST_WIDE_INT prec
)
1475 machine_mode mode
= TYPE_MODE (type
);
1477 /* IBM Extended mode. */
1478 if (MODE_COMPOSITE_P (mode
))
1481 /* Get the real type format desription for the target. */
1482 const real_format
*rfmt
= REAL_MODE_FORMAT (mode
);
1485 real_maxval (&rv
, 0, mode
);
1487 /* Convert the GCC real value representation with the precision
1488 of the real type to the mpfr_t format with the GCC default
1489 round-to-nearest mode. */
1491 mpfr_init2 (x
, rfmt
->p
);
1492 mpfr_from_real (x
, &rv
, GMP_RNDN
);
1494 /* Return a value one greater to account for the leading minus sign. */
1495 unsigned HOST_WIDE_INT r
1496 = 1 + get_mpfr_format_length (x
, "", prec
, spec
, 'D');
1501 /* Return a range representing the minimum and maximum number of bytes
1502 that the directive DIR will output for any argument. This function
1503 is used when the directive argument or its value isn't known. */
1506 format_floating (const directive
&dir
)
1510 switch (dir
.modifier
)
1514 type
= double_type_node
;
1518 type
= long_double_type_node
;
1522 type
= long_double_type_node
;
1526 return fmtresult ();
1529 /* The minimum and maximum number of bytes produced by the directive. */
1532 /* The minimum output as determined by flags. It's always at least 1.
1533 When plus or space are set the output is preceded by either a sign
1535 int flagmin
= (1 /* for the first digit */
1536 + (dir
.get_flag ('+') | dir
.get_flag (' ')));
1538 /* When the pound flag is set the decimal point is included in output
1539 regardless of precision. Whether or not a decimal point is included
1540 otherwise depends on the specification and precision. */
1541 bool radix
= dir
.get_flag ('#');
1543 switch (dir
.specifier
)
1548 HOST_WIDE_INT minprec
= 6 + !radix
/* decimal point */;
1549 if (dir
.prec
[0] <= 0)
1551 else if (dir
.prec
[0] > 0)
1552 minprec
= dir
.prec
[0] + !radix
/* decimal point */;
1554 res
.range
.min
= (2 /* 0x */
1560 res
.range
.max
= format_floating_max (type
, 'a', dir
.prec
[1]);
1561 res
.range
.likely
= res
.range
.min
;
1563 /* The unlikely maximum accounts for the longest multibyte
1564 decimal point character. */
1565 res
.range
.unlikely
= res
.range
.max
;
1566 if (dir
.prec
[0] != dir
.prec
[1]
1567 || dir
.prec
[0] == -1 || dir
.prec
[0] > 0)
1568 res
.range
.unlikely
+= target_mb_len_max () - 1;
1576 /* The minimum output is "[-+]1.234567e+00" regardless
1577 of the value of the actual argument. */
1578 HOST_WIDE_INT minprec
= 6 + !radix
/* decimal point */;
1579 if ((dir
.prec
[0] < 0 && dir
.prec
[1] > -1) || dir
.prec
[0] == 0)
1581 else if (dir
.prec
[0] > 0)
1582 minprec
= dir
.prec
[0] + !radix
/* decimal point */;
1584 res
.range
.min
= (flagmin
1588 /* MPFR uses a precision of 16 by default for some reason.
1589 Set it to the C default of 6. */
1590 int maxprec
= dir
.prec
[1] < 0 ? 6 : dir
.prec
[1];
1591 res
.range
.max
= format_floating_max (type
, 'e', maxprec
);
1593 res
.range
.likely
= res
.range
.min
;
1595 /* The unlikely maximum accounts for the longest multibyte
1596 decimal point character. */
1597 if (dir
.prec
[0] != dir
.prec
[1]
1598 || dir
.prec
[0] == -1 || dir
.prec
[0] > 0)
1599 res
.range
.unlikely
= res
.range
.max
+ target_mb_len_max () -1;
1601 res
.range
.unlikely
= res
.range
.max
;
1608 /* The lower bound when precision isn't specified is 8 bytes
1609 ("1.23456" since precision is taken to be 6). When precision
1610 is zero, the lower bound is 1 byte (e.g., "1"). Otherwise,
1611 when precision is greater than zero, then the lower bound
1612 is 2 plus precision (plus flags). */
1613 HOST_WIDE_INT minprec
= 0;
1614 if (dir
.prec
[0] < 0)
1615 minprec
= dir
.prec
[1] < 0 ? 6 + !radix
/* decimal point */ : 0;
1616 else if (dir
.prec
[0])
1617 minprec
= dir
.prec
[0] + !radix
/* decimal point */;
1619 res
.range
.min
= flagmin
+ radix
+ minprec
;
1621 /* Compute the upper bound for -TYPE_MAX. */
1622 res
.range
.max
= format_floating_max (type
, 'f', dir
.prec
[1]);
1624 /* The minimum output with unknown precision is a single byte
1625 (e.g., "0") but the more likely output is 3 bytes ("0.0"). */
1626 if (dir
.prec
[0] < 0 && dir
.prec
[1] > 0)
1627 res
.range
.likely
= 3;
1629 res
.range
.likely
= res
.range
.min
;
1631 /* The unlikely maximum accounts for the longest multibyte
1632 decimal point character. */
1633 if (dir
.prec
[0] != dir
.prec
[1]
1634 || dir
.prec
[0] == -1 || dir
.prec
[0] > 0)
1635 res
.range
.unlikely
= res
.range
.max
+ target_mb_len_max () - 1;
1642 /* The %g output depends on precision and the exponent of
1643 the argument. Since the value of the argument isn't known
1644 the lower bound on the range of bytes (not counting flags
1645 or width) is 1 plus radix (i.e., either "0" or "0." for
1646 "%g" and "%#g", respectively, with a zero argument). */
1647 res
.range
.min
= flagmin
+ radix
;
1650 HOST_WIDE_INT maxprec
= dir
.prec
[1];
1651 if (radix
&& maxprec
)
1653 /* When the pound flag (radix) is set, trailing zeros aren't
1654 trimmed and so the longest output is the same as for %e,
1655 except with precision minus 1 (as specified in C11). */
1659 else if (maxprec
< 0)
1663 res
.range
.max
= format_floating_max (type
, spec
, maxprec
);
1665 /* The likely output is either the maximum computed above
1666 minus 1 (assuming the maximum is positive) when precision
1667 is known (or unspecified), or the same minimum as for %e
1668 (which is computed for a non-negative argument). Unlike
1669 for the other specifiers above the likely output isn't
1670 the minimum because for %g that's 1 which is unlikely. */
1672 || (unsigned HOST_WIDE_INT
)dir
.prec
[1] < target_int_max ())
1673 res
.range
.likely
= res
.range
.max
- 1;
1676 HOST_WIDE_INT minprec
= 6 + !radix
/* decimal point */;
1677 res
.range
.likely
= (flagmin
1683 /* The unlikely maximum accounts for the longest multibyte
1684 decimal point character. */
1685 res
.range
.unlikely
= res
.range
.max
+ target_mb_len_max () - 1;
1690 return fmtresult ();
1693 /* Bump up the byte counters if WIDTH is greater. */
1694 res
.adjust_for_width_or_precision (dir
.width
);
1698 /* Return a range representing the minimum and maximum number of bytes
1699 that the directive DIR will write on output for the floating argument
1703 format_floating (const directive
&dir
, tree arg
)
1705 if (!arg
|| TREE_CODE (arg
) != REAL_CST
)
1706 return format_floating (dir
);
1708 HOST_WIDE_INT prec
[] = { dir
.prec
[0], dir
.prec
[1] };
1710 /* For an indeterminate precision the lower bound must be assumed
1712 if (TOUPPER (dir
.specifier
) == 'A')
1714 /* Get the number of fractional decimal digits needed to represent
1715 the argument without a loss of accuracy. */
1716 tree type
= arg
? TREE_TYPE (arg
) :
1717 (dir
.modifier
== FMT_LEN_L
|| dir
.modifier
== FMT_LEN_ll
1718 ? long_double_type_node
: double_type_node
);
1721 = REAL_MODE_FORMAT (TYPE_MODE (type
))->p
;
1723 /* The precision of the IEEE 754 double format is 53.
1724 The precision of all other GCC binary double formats
1726 unsigned maxprec
= fmtprec
<= 56 ? 13 : 15;
1728 /* For %a, leave the minimum precision unspecified to let
1729 MFPR trim trailing zeros (as it and many other systems
1730 including Glibc happen to do) and set the maximum
1731 precision to reflect what it would be with trailing zeros
1732 present (as Solaris and derived systems do). */
1733 if (dir
.prec
[1] < 0)
1735 /* Both bounds are negative implies that precision has
1736 not been specified. */
1740 else if (dir
.prec
[0] < 0)
1742 /* With a negative lower bound and a non-negative upper
1743 bound set the minimum precision to zero and the maximum
1744 to the greater of the maximum precision (i.e., with
1745 trailing zeros present) and the specified upper bound. */
1747 prec
[1] = dir
.prec
[1] < maxprec
? maxprec
: dir
.prec
[1];
1750 else if (dir
.prec
[0] < 0)
1752 if (dir
.prec
[1] < 0)
1754 /* A precision in a strictly negative range is ignored and
1755 the default of 6 is used instead. */
1756 prec
[0] = prec
[1] = 6;
1760 /* For a precision in a partly negative range, the lower bound
1761 must be assumed to be zero and the new upper bound is the
1762 greater of 6 (the default precision used when the specified
1763 precision is negative) and the upper bound of the specified
1766 prec
[1] = dir
.prec
[1] < 6 ? 6 : dir
.prec
[1];
1770 /* The minimum and maximum number of bytes produced by the directive. */
1773 /* Get the real type format desription for the target. */
1774 const REAL_VALUE_TYPE
*rvp
= TREE_REAL_CST_PTR (arg
);
1775 const real_format
*rfmt
= REAL_MODE_FORMAT (TYPE_MODE (TREE_TYPE (arg
)));
1778 char *pfmt
= fmtstr
;
1781 for (const char *pf
= "-+ #0"; *pf
; ++pf
)
1782 if (dir
.get_flag (*pf
))
1788 /* Set up an array to easily iterate over. */
1789 unsigned HOST_WIDE_INT
* const minmax
[] = {
1790 &res
.range
.min
, &res
.range
.max
1793 for (int i
= 0; i
!= sizeof minmax
/ sizeof *minmax
; ++i
)
1795 /* Convert the GCC real value representation with the precision
1796 of the real type to the mpfr_t format rounding down in the
1797 first iteration that computes the minimm and up in the second
1798 that computes the maximum. This order is arbibtrary because
1799 rounding in either direction can result in longer output. */
1801 mpfr_init2 (mpfrval
, rfmt
->p
);
1802 mpfr_from_real (mpfrval
, rvp
, i
? GMP_RNDU
: GMP_RNDD
);
1804 /* Use the MPFR rounding specifier to round down in the first
1805 iteration and then up. In most but not all cases this will
1806 result in the same number of bytes. */
1807 char rndspec
= "DU"[i
];
1809 /* Format it and store the result in the corresponding member
1810 of the result struct. */
1811 *minmax
[i
] = get_mpfr_format_length (mpfrval
, fmtstr
, prec
[i
],
1812 dir
.specifier
, rndspec
);
1813 mpfr_clear (mpfrval
);
1817 /* Make sure the minimum is less than the maximum (MPFR rounding
1818 in the call to mpfr_snprintf can result in the reverse. */
1819 if (res
.range
.max
< res
.range
.min
)
1821 unsigned HOST_WIDE_INT tmp
= res
.range
.min
;
1822 res
.range
.min
= res
.range
.max
;
1823 res
.range
.max
= tmp
;
1826 /* The range is known unless either width or precision is unknown. */
1827 res
.knownrange
= dir
.known_width_and_precision ();
1829 /* For the same floating point constant, unless width or precision
1830 is unknown, use the longer output as the likely maximum since
1831 with round to nearest either is equally likely. Otheriwse, when
1832 precision is unknown, use the greater of the minimum and 3 as
1833 the likely output (for "0.0" since zero precision is unlikely). */
1835 res
.range
.likely
= res
.range
.max
;
1836 else if (res
.range
.min
< 3
1838 && (unsigned HOST_WIDE_INT
)dir
.prec
[1] == target_int_max ())
1839 res
.range
.likely
= 3;
1841 res
.range
.likely
= res
.range
.min
;
1843 res
.range
.unlikely
= res
.range
.max
;
1845 if (res
.range
.max
> 2 && (prec
[0] != 0 || prec
[1] != 0))
1847 /* Unless the precision is zero output longer than 2 bytes may
1848 include the decimal point which must be a single character
1849 up to MB_LEN_MAX in length. This is overly conservative
1850 since in some conversions some constants result in no decimal
1851 point (e.g., in %g). */
1852 res
.range
.unlikely
+= target_mb_len_max () - 1;
1855 res
.adjust_for_width_or_precision (dir
.width
);
1859 /* Return a FMTRESULT struct set to the lengths of the shortest and longest
1860 strings referenced by the expression STR, or (-1, -1) when not known.
1861 Used by the format_string function below. */
1864 get_string_length (tree str
)
1867 return fmtresult ();
1869 if (tree slen
= c_strlen (str
, 1))
1871 /* Simply return the length of the string. */
1872 fmtresult
res (tree_to_shwi (slen
));
1876 /* Determine the length of the shortest and longest string referenced
1877 by STR. Strings of unknown lengths are bounded by the sizes of
1878 arrays that subexpressions of STR may refer to. Pointers that
1879 aren't known to point any such arrays result in LENRANGE[1] set
1882 bool flexarray
= get_range_strlen (str
, lenrange
);
1884 if (lenrange
[0] || lenrange
[1])
1887 = (tree_fits_uhwi_p (lenrange
[0])
1888 ? tree_to_uhwi (lenrange
[0])
1892 = (tree_fits_uhwi_p (lenrange
[1])
1893 ? tree_to_uhwi (lenrange
[1])
1894 : HOST_WIDE_INT_M1U
);
1896 /* get_range_strlen() returns the target value of SIZE_MAX for
1897 strings of unknown length. Bump it up to HOST_WIDE_INT_M1U
1898 which may be bigger. */
1899 if ((unsigned HOST_WIDE_INT
)min
== target_size_max ())
1900 min
= HOST_WIDE_INT_M1U
;
1901 if ((unsigned HOST_WIDE_INT
)max
== target_size_max ())
1902 max
= HOST_WIDE_INT_M1U
;
1904 fmtresult
res (min
, max
);
1906 /* Set RES.KNOWNRANGE to true if and only if all strings referenced
1907 by STR are known to be bounded (though not necessarily by their
1908 actual length but perhaps by their maximum possible length). */
1909 if (res
.range
.max
< target_int_max ())
1911 res
.knownrange
= true;
1912 /* When the the length of the longest string is known and not
1913 excessive use it as the likely length of the string(s). */
1914 res
.range
.likely
= res
.range
.max
;
1918 /* When the upper bound is unknown (it can be zero or excessive)
1919 set the likely length to the greater of 1 and the length of
1920 the shortest string and reset the lower bound to zero. */
1921 res
.range
.likely
= res
.range
.min
? res
.range
.min
: warn_level
> 1;
1925 /* If the range of string length has been estimated from the size
1926 of an array at the end of a struct assume that it's longer than
1927 the array bound says it is in case it's used as a poor man's
1928 flexible array member, such as in struct S { char a[4]; }; */
1929 res
.range
.unlikely
= flexarray
? HOST_WIDE_INT_MAX
: res
.range
.max
;
1934 return get_string_length (NULL_TREE
);
1937 /* Return the minimum and maximum number of characters formatted
1938 by the '%c' format directives and its wide character form for
1939 the argument ARG. ARG can be null (for functions such as
1943 format_character (const directive
&dir
, tree arg
)
1947 res
.knownrange
= true;
1949 if (dir
.modifier
== FMT_LEN_l
)
1951 /* A wide character can result in as few as zero bytes. */
1954 HOST_WIDE_INT min
, max
;
1955 if (get_int_range (arg
, integer_type_node
, &min
, &max
, false, 0))
1957 if (min
== 0 && max
== 0)
1959 /* The NUL wide character results in no bytes. */
1961 res
.range
.likely
= 0;
1962 res
.range
.unlikely
= 0;
1964 else if (min
> 0 && min
< 128)
1966 /* A wide character in the ASCII range most likely results
1967 in a single byte, and only unlikely in up to MB_LEN_MAX. */
1969 res
.range
.likely
= 1;
1970 res
.range
.unlikely
= target_mb_len_max ();
1974 /* A wide character outside the ASCII range likely results
1975 in up to two bytes, and only unlikely in up to MB_LEN_MAX. */
1976 res
.range
.max
= target_mb_len_max ();
1977 res
.range
.likely
= 2;
1978 res
.range
.unlikely
= res
.range
.max
;
1983 /* An unknown wide character is treated the same as a wide
1984 character outside the ASCII range. */
1985 res
.range
.max
= target_mb_len_max ();
1986 res
.range
.likely
= 2;
1987 res
.range
.unlikely
= res
.range
.max
;
1992 /* A plain '%c' directive. Its ouput is exactly 1. */
1993 res
.range
.min
= res
.range
.max
= 1;
1994 res
.range
.likely
= res
.range
.unlikely
= 1;
1995 res
.knownrange
= true;
1998 /* Bump up the byte counters if WIDTH is greater. */
1999 return res
.adjust_for_width_or_precision (dir
.width
);
2002 /* Return the minimum and maximum number of characters formatted
2003 by the '%s' format directive and its wide character form for
2004 the argument ARG. ARG can be null (for functions such as
2008 format_string (const directive
&dir
, tree arg
)
2012 /* Compute the range the argument's length can be in. */
2013 fmtresult slen
= get_string_length (arg
);
2014 if (slen
.range
.min
== slen
.range
.max
2015 && slen
.range
.min
< HOST_WIDE_INT_MAX
)
2017 /* The argument is either a string constant or it refers
2018 to one of a number of strings of the same length. */
2020 /* A '%s' directive with a string argument with constant length. */
2021 res
.range
= slen
.range
;
2023 if (dir
.modifier
== FMT_LEN_l
)
2025 /* In the worst case the length of output of a wide string S
2026 is bounded by MB_LEN_MAX * wcslen (S). */
2027 res
.range
.max
*= target_mb_len_max ();
2028 res
.range
.unlikely
= res
.range
.max
;
2029 /* It's likely that the the total length is not more that
2031 res
.range
.likely
= res
.range
.min
* 2;
2033 if (dir
.prec
[1] >= 0
2034 && (unsigned HOST_WIDE_INT
)dir
.prec
[1] < res
.range
.max
)
2036 res
.range
.max
= dir
.prec
[1];
2037 res
.range
.likely
= dir
.prec
[1];
2038 res
.range
.unlikely
= dir
.prec
[1];
2041 if (dir
.prec
[0] < 0 && dir
.prec
[1] > -1)
2043 else if (dir
.prec
[0] >= 0)
2044 res
.range
.likely
= dir
.prec
[0];
2046 /* Even a non-empty wide character string need not convert into
2052 res
.knownrange
= true;
2054 if (dir
.prec
[0] < 0 && dir
.prec
[1] > -1)
2056 else if ((unsigned HOST_WIDE_INT
)dir
.prec
[0] < res
.range
.min
)
2057 res
.range
.min
= dir
.prec
[0];
2059 if ((unsigned HOST_WIDE_INT
)dir
.prec
[1] < res
.range
.max
)
2061 res
.range
.max
= dir
.prec
[1];
2062 res
.range
.likely
= dir
.prec
[1];
2063 res
.range
.unlikely
= dir
.prec
[1];
2067 else if (arg
&& integer_zerop (arg
))
2069 /* Handle null pointer argument. */
2077 /* For a '%s' and '%ls' directive with a non-constant string (either
2078 one of a number of strings of known length or an unknown string)
2079 the minimum number of characters is lesser of PRECISION[0] and
2080 the length of the shortest known string or zero, and the maximum
2081 is the lessser of the length of the longest known string or
2082 PTRDIFF_MAX and PRECISION[1]. The likely length is either
2083 the minimum at level 1 and the greater of the minimum and 1
2084 at level 2. This result is adjust upward for width (if it's
2087 if (dir
.modifier
== FMT_LEN_l
)
2089 /* A wide character converts to as few as zero bytes. */
2091 if (slen
.range
.max
< target_int_max ())
2092 slen
.range
.max
*= target_mb_len_max ();
2094 if (slen
.range
.likely
< target_int_max ())
2095 slen
.range
.likely
*= 2;
2097 if (slen
.range
.likely
< target_int_max ())
2098 slen
.range
.unlikely
*= target_mb_len_max ();
2101 res
.range
= slen
.range
;
2103 if (dir
.prec
[0] >= 0)
2105 /* Adjust the minimum to zero if the string length is unknown,
2106 or at most the lower bound of the precision otherwise. */
2107 if (slen
.range
.min
>= target_int_max ())
2109 else if ((unsigned HOST_WIDE_INT
)dir
.prec
[0] < slen
.range
.min
)
2110 res
.range
.min
= dir
.prec
[0];
2112 /* Make both maxima no greater than the upper bound of precision. */
2113 if ((unsigned HOST_WIDE_INT
)dir
.prec
[1] < slen
.range
.max
2114 || slen
.range
.max
>= target_int_max ())
2116 res
.range
.max
= dir
.prec
[1];
2117 res
.range
.unlikely
= dir
.prec
[1];
2120 /* If precision is constant, set the likely counter to the lesser
2121 of it and the maximum string length. Otherwise, if the lower
2122 bound of precision is greater than zero, set the likely counter
2123 to the minimum. Otherwise set it to zero or one based on
2124 the warning level. */
2125 if (dir
.prec
[0] == dir
.prec
[1])
2127 = ((unsigned HOST_WIDE_INT
)dir
.prec
[0] < slen
.range
.max
2128 ? dir
.prec
[0] : slen
.range
.max
);
2129 else if (dir
.prec
[0] > 0)
2130 res
.range
.likely
= res
.range
.min
;
2132 res
.range
.likely
= warn_level
> 1;
2134 else if (dir
.prec
[1] >= 0)
2137 if ((unsigned HOST_WIDE_INT
)dir
.prec
[1] < slen
.range
.max
)
2138 res
.range
.max
= dir
.prec
[1];
2139 res
.range
.likely
= dir
.prec
[1] ? warn_level
> 1 : 0;
2141 else if (slen
.range
.min
>= target_int_max ())
2144 res
.range
.max
= HOST_WIDE_INT_MAX
;
2145 /* At level 1 strings of unknown length are assumed to be
2146 empty, while at level 1 they are assumed to be one byte
2148 res
.range
.likely
= warn_level
> 1;
2152 /* A string of unknown length unconstrained by precision is
2153 assumed to be empty at level 1 and just one character long
2154 at higher levels. */
2155 if (res
.range
.likely
>= target_int_max ())
2156 res
.range
.likely
= warn_level
> 1;
2159 res
.range
.unlikely
= res
.range
.max
;
2162 /* Bump up the byte counters if WIDTH is greater. */
2163 return res
.adjust_for_width_or_precision (dir
.width
);
2166 /* Format plain string (part of the format string itself). */
2169 format_plain (const directive
&dir
, tree
)
2171 fmtresult
res (dir
.len
);
2175 /* Return true if the RESULT of a directive in a call describe by INFO
2176 should be diagnosed given the AVAILable space in the destination. */
2179 should_warn_p (const pass_sprintf_length::call_info
&info
,
2180 const result_range
&avail
, const result_range
&result
)
2182 if (result
.max
<= avail
.min
)
2184 /* The least amount of space remaining in the destination is big
2185 enough for the longest output. */
2191 if (warn_format_trunc
== 1 && result
.min
<= avail
.max
2192 && info
.retval_used ())
2194 /* The likely amount of space remaining in the destination is big
2195 enough for the least output and the return value is used. */
2199 if (warn_format_trunc
== 1 && result
.likely
<= avail
.likely
2200 && !info
.retval_used ())
2202 /* The likely amount of space remaining in the destination is big
2203 enough for the likely output and the return value is unused. */
2207 if (warn_format_trunc
== 2
2208 && result
.likely
<= avail
.min
2209 && (result
.max
<= avail
.min
2210 || result
.max
> HOST_WIDE_INT_MAX
))
2212 /* The minimum amount of space remaining in the destination is big
2213 enough for the longest output. */
2219 if (warn_level
== 1 && result
.likely
<= avail
.likely
)
2221 /* The likely amount of space remaining in the destination is big
2222 enough for the likely output. */
2227 && result
.likely
<= avail
.min
2228 && (result
.max
<= avail
.min
2229 || result
.max
> HOST_WIDE_INT_MAX
))
2231 /* The minimum amount of space remaining in the destination is big
2232 enough for the longest output. */
2240 /* At format string location describe by DIRLOC in a call described
2241 by INFO, issue a warning for a directive DIR whose output may be
2242 in excess of the available space AVAIL_RANGE in the destination
2243 given the formatting result FMTRES. This function does nothing
2244 except decide whether to issue a warning for a possible write
2245 past the end or truncation and, if so, format the warning.
2246 Return true if a warning has been issued. */
2249 maybe_warn (substring_loc
&dirloc
, source_range
*pargrange
,
2250 const pass_sprintf_length::call_info
&info
,
2251 const result_range
&avail_range
, const result_range
&res
,
2252 const directive
&dir
)
2254 if (!should_warn_p (info
, avail_range
, res
))
2257 /* A warning will definitely be issued below. */
2259 /* The maximum byte count to reference in the warning. Larger counts
2260 imply that the upper bound is unknown (and could be anywhere between
2261 RES.MIN + 1 and SIZE_MAX / 2) are printed as "N or more bytes" rather
2262 than "between N and X" where X is some huge number. */
2263 unsigned HOST_WIDE_INT maxbytes
= target_dir_max ();
2265 /* True when there is enough room in the destination for the least
2266 amount of a directive's output but not enough for its likely or
2268 bool maybe
= (res
.min
<= avail_range
.max
2269 && (avail_range
.min
< res
.likely
2270 || (res
.max
< HOST_WIDE_INT_MAX
2271 && avail_range
.min
< res
.max
)));
2273 if (avail_range
.min
== avail_range
.max
)
2275 /* The size of the destination region is exact. */
2276 unsigned HOST_WIDE_INT navail
= avail_range
.max
;
2278 if (*dir
.beg
!= '%')
2280 /* For plain character directives (i.e., the format string itself)
2281 but not others, point the caret at the first character that's
2282 past the end of the destination. */
2283 dirloc
.set_caret_index (dirloc
.get_caret_idx () + navail
);
2286 if (*dir
.beg
== '\0')
2288 /* This is the terminating nul. */
2289 gcc_assert (res
.min
== 1 && res
.min
== res
.max
);
2294 ? G_("%qE output may be truncated before the last format "
2296 : G_("%qE output truncated before the last format character"))
2298 ? G_("%qE may write a terminating nul past the end "
2299 "of the destination")
2300 : G_("%qE writing a terminating nul past the end "
2301 "of the destination")));
2303 return fmtwarn (dirloc
, NULL
, NULL
, info
.warnopt (), fmtstr
,
2307 if (res
.min
== res
.max
)
2313 ? G_("%<%.*s%> directive output may be truncated writing "
2314 "%wu byte into a region of size %wu")
2315 : G_("%<%.*s%> directive output truncated writing "
2316 "%wu byte into a region of size %wu"))
2317 : G_("%<%.*s%> directive writing %wu byte "
2318 "into a region of size %wu"))
2321 ? G_("%<%.*s%> directive output may be truncated writing "
2322 "%wu bytes into a region of size %wu")
2323 : G_("%<%.*s%> directive output truncated writing "
2324 "%wu bytes into a region of size %wu"))
2325 : G_("%<%.*s%> directive writing %wu bytes "
2326 "into a region of size %wu")));
2327 return fmtwarn (dirloc
, pargrange
, NULL
,
2328 info
.warnopt (), fmtstr
,
2329 dir
.len
, dir
.beg
, res
.min
,
2333 if (res
.min
== 0 && res
.max
< maxbytes
)
2338 ? G_("%<%.*s%> directive output may be truncated writing "
2339 "up to %wu bytes into a region of size %wu")
2340 : G_("%<%.*s%> directive output truncated writing "
2341 "up to %wu bytes into a region of size %wu"))
2342 : G_("%<%.*s%> directive writing up to %wu bytes "
2343 "into a region of size %wu"));
2344 return fmtwarn (dirloc
, pargrange
, NULL
,
2345 info
.warnopt (), fmtstr
,
2350 if (res
.min
== 0 && maxbytes
<= res
.max
)
2352 /* This is a special case to avoid issuing the potentially
2354 writing 0 or more bytes into a region of size 0. */
2358 ? G_("%<%.*s%> directive output may be truncated writing "
2359 "likely %wu or more bytes into a region of size %wu")
2360 : G_("%<%.*s%> directive output truncated writing "
2361 "likely %wu or more bytes into a region of size %wu"))
2362 : G_("%<%.*s%> directive writing likely %wu or more bytes "
2363 "into a region of size %wu"));
2364 return fmtwarn (dirloc
, pargrange
, NULL
,
2365 info
.warnopt (), fmtstr
,
2367 res
.likely
, navail
);
2370 if (res
.max
< maxbytes
)
2375 ? G_("%<%.*s%> directive output may be truncated writing "
2376 "between %wu and %wu bytes into a region of size %wu")
2377 : G_("%<%.*s%> directive output truncated writing "
2378 "between %wu and %wu bytes into a region of size %wu"))
2379 : G_("%<%.*s%> directive writing between %wu and "
2380 "%wu bytes into a region of size %wu"));
2381 return fmtwarn (dirloc
, pargrange
, NULL
,
2382 info
.warnopt (), fmtstr
,
2391 ? G_("%<%.*s%> directive output may be truncated writing "
2392 "%wu or more bytes into a region of size %wu")
2393 : G_("%<%.*s%> directive output truncated writing "
2394 "%wu or more bytes into a region of size %wu"))
2395 : G_("%<%.*s%> directive writing %wu or more bytes "
2396 "into a region of size %wu"));
2397 return fmtwarn (dirloc
, pargrange
, NULL
,
2398 info
.warnopt (), fmtstr
,
2403 /* The size of the destination region is a range. */
2405 if (*dir
.beg
!= '%')
2407 unsigned HOST_WIDE_INT navail
= avail_range
.max
;
2409 /* For plain character directives (i.e., the format string itself)
2410 but not others, point the caret at the first character that's
2411 past the end of the destination. */
2412 dirloc
.set_caret_index (dirloc
.get_caret_idx () + navail
);
2415 if (*dir
.beg
== '\0')
2417 gcc_assert (res
.min
== 1 && res
.min
== res
.max
);
2422 ? G_("%qE output may be truncated before the last format "
2424 : G_("%qE output truncated before the last format character"))
2426 ? G_("%qE may write a terminating nul past the end "
2427 "of the destination")
2428 : G_("%qE writing a terminating nul past the end "
2429 "of the destination")));
2431 return fmtwarn (dirloc
, NULL
, NULL
, info
.warnopt (), fmtstr
,
2435 if (res
.min
== res
.max
)
2441 ? G_("%<%.*s%> directive output may be truncated writing "
2442 "%wu byte into a region of size between %wu and %wu")
2443 : G_("%<%.*s%> directive output truncated writing "
2444 "%wu byte into a region of size between %wu and %wu"))
2445 : G_("%<%.*s%> directive writing %wu byte "
2446 "into a region of size between %wu and %wu"))
2449 ? G_("%<%.*s%> directive output may be truncated writing "
2450 "%wu bytes into a region of size between %wu and %wu")
2451 : G_("%<%.*s%> directive output truncated writing "
2452 "%wu bytes into a region of size between %wu and %wu"))
2453 : G_("%<%.*s%> directive writing %wu bytes "
2454 "into a region of size between %wu and %wu")));
2456 return fmtwarn (dirloc
, pargrange
, NULL
,
2457 info
.warnopt (), fmtstr
,
2458 dir
.len
, dir
.beg
, res
.min
,
2459 avail_range
.min
, avail_range
.max
);
2462 if (res
.min
== 0 && res
.max
< maxbytes
)
2467 ? G_("%<%.*s%> directive output may be truncated writing "
2468 "up to %wu bytes into a region of size between "
2470 : G_("%<%.*s%> directive output truncated writing "
2471 "up to %wu bytes into a region of size between "
2473 : G_("%<%.*s%> directive writing up to %wu bytes "
2474 "into a region of size between %wu and %wu"));
2475 return fmtwarn (dirloc
, pargrange
, NULL
,
2476 info
.warnopt (), fmtstr
,
2477 dir
.len
, dir
.beg
, res
.max
,
2478 avail_range
.min
, avail_range
.max
);
2481 if (res
.min
== 0 && maxbytes
<= res
.max
)
2483 /* This is a special case to avoid issuing the potentially confusing
2485 writing 0 or more bytes into a region of size between 0 and N. */
2489 ? G_("%<%.*s%> directive output may be truncated writing "
2490 "likely %wu or more bytes into a region of size between "
2492 : G_("%<%.*s%> directive output truncated writing likely "
2493 "%wu or more bytes into a region of size between "
2495 : G_("%<%.*s%> directive writing likely %wu or more bytes "
2496 "into a region of size between %wu and %wu"));
2497 return fmtwarn (dirloc
, pargrange
, NULL
,
2498 info
.warnopt (), fmtstr
,
2499 dir
.len
, dir
.beg
, res
.likely
,
2500 avail_range
.min
, avail_range
.max
);
2503 if (res
.max
< maxbytes
)
2508 ? G_("%<%.*s%> directive output may be truncated writing "
2509 "between %wu and %wu bytes into a region of size "
2510 "between %wu and %wu")
2511 : G_("%<%.*s%> directive output truncated writing "
2512 "between %wu and %wu bytes into a region of size "
2513 "between %wu and %wu"))
2514 : G_("%<%.*s%> directive writing between %wu and "
2515 "%wu bytes into a region of size between %wu and %wu"));
2516 return fmtwarn (dirloc
, pargrange
, NULL
,
2517 info
.warnopt (), fmtstr
,
2520 avail_range
.min
, avail_range
.max
);
2526 ? G_("%<%.*s%> directive output may be truncated writing "
2527 "%wu or more bytes into a region of size between "
2529 : G_("%<%.*s%> directive output truncated writing "
2530 "%wu or more bytes into a region of size between "
2532 : G_("%<%.*s%> directive writing %wu or more bytes "
2533 "into a region of size between %wu and %wu"));
2534 return fmtwarn (dirloc
, pargrange
, NULL
,
2535 info
.warnopt (), fmtstr
,
2538 avail_range
.min
, avail_range
.max
);
2541 /* Compute the length of the output resulting from the directive DIR
2542 in a call described by INFO and update the overall result of the call
2543 in *RES. Return true if the directive has been handled. */
2546 format_directive (const pass_sprintf_length::call_info
&info
,
2547 format_result
*res
, const directive
&dir
)
2549 /* Offset of the beginning of the directive from the beginning
2550 of the format string. */
2551 size_t offset
= dir
.beg
- info
.fmtstr
;
2552 size_t start
= offset
;
2553 size_t length
= offset
+ dir
.len
- !!dir
.len
;
2555 /* Create a location for the whole directive from the % to the format
2557 substring_loc
dirloc (info
.fmtloc
, TREE_TYPE (info
.format
),
2558 offset
, start
, length
);
2560 /* Also create a location range for the argument if possible.
2561 This doesn't work for integer literals or function calls. */
2562 source_range argrange
;
2563 source_range
*pargrange
;
2564 if (dir
.arg
&& CAN_HAVE_LOCATION_P (dir
.arg
))
2566 argrange
= EXPR_LOCATION_RANGE (dir
.arg
);
2567 pargrange
= &argrange
;
2572 /* Bail when there is no function to compute the output length,
2573 or when minimum length checking has been disabled. */
2574 if (!dir
.fmtfunc
|| res
->range
.min
>= HOST_WIDE_INT_MAX
)
2577 /* Compute the range of lengths of the formatted output. */
2578 fmtresult fmtres
= dir
.fmtfunc (dir
, dir
.arg
);
2580 /* Record whether the output of all directives is known to be
2581 bounded by some maximum, implying that their arguments are
2582 either known exactly or determined to be in a known range
2583 or, for strings, limited by the upper bounds of the arrays
2585 res
->knownrange
&= fmtres
.knownrange
;
2587 if (!fmtres
.knownrange
)
2589 /* Only when the range is known, check it against the host value
2590 of INT_MAX + (the number of bytes of the "%.*Lf" directive with
2591 INT_MAX precision, which is the longest possible output of any
2592 single directive). That's the largest valid byte count (though
2593 not valid call to a printf-like function because it can never
2594 return such a count). Otherwise, the range doesn't correspond
2595 to known values of the argument. */
2596 if (fmtres
.range
.max
> target_dir_max ())
2598 /* Normalize the MAX counter to avoid having to deal with it
2599 later. The counter can be less than HOST_WIDE_INT_M1U
2600 when compiling for an ILP32 target on an LP64 host. */
2601 fmtres
.range
.max
= HOST_WIDE_INT_M1U
;
2602 /* Disable exact and maximum length checking after a failure
2603 to determine the maximum number of characters (for example
2604 for wide characters or wide character strings) but continue
2605 tracking the minimum number of characters. */
2606 res
->range
.max
= HOST_WIDE_INT_M1U
;
2609 if (fmtres
.range
.min
> target_dir_max ())
2611 /* Disable exact length checking after a failure to determine
2612 even the minimum number of characters (it shouldn't happen
2613 except in an error) but keep tracking the minimum and maximum
2614 number of characters. */
2619 int dirlen
= dir
.len
;
2623 fmtwarn (dirloc
, pargrange
, NULL
, info
.warnopt (),
2624 "%<%.*s%> directive argument is null",
2627 /* Don't bother processing the rest of the format string. */
2629 res
->range
.min
= HOST_WIDE_INT_M1U
;
2630 res
->range
.max
= HOST_WIDE_INT_M1U
;
2634 /* Compute the number of available bytes in the destination. There
2635 must always be at least one byte of space for the terminating
2636 NUL that's appended after the format string has been processed. */
2637 result_range avail_range
= bytes_remaining (info
.objsize
, *res
);
2639 bool warned
= res
->warned
;
2642 warned
= maybe_warn (dirloc
, pargrange
, info
, avail_range
,
2645 /* Bump up the total maximum if it isn't too big. */
2646 if (res
->range
.max
< HOST_WIDE_INT_MAX
2647 && fmtres
.range
.max
< HOST_WIDE_INT_MAX
)
2648 res
->range
.max
+= fmtres
.range
.max
;
2650 /* Raise the total unlikely maximum by the larger of the maximum
2651 and the unlikely maximum. */
2652 unsigned HOST_WIDE_INT save
= res
->range
.unlikely
;
2653 if (fmtres
.range
.max
< fmtres
.range
.unlikely
)
2654 res
->range
.unlikely
+= fmtres
.range
.unlikely
;
2656 res
->range
.unlikely
+= fmtres
.range
.max
;
2658 if (res
->range
.unlikely
< save
)
2659 res
->range
.unlikely
= HOST_WIDE_INT_M1U
;
2661 res
->range
.min
+= fmtres
.range
.min
;
2662 res
->range
.likely
+= fmtres
.range
.likely
;
2664 /* Has the minimum directive output length exceeded the maximum
2665 of 4095 bytes required to be supported? */
2666 bool minunder4k
= fmtres
.range
.min
< 4096;
2667 bool maxunder4k
= fmtres
.range
.max
< 4096;
2668 /* Clear UNDER4K in the overall result if the maximum has exceeded
2669 the 4k (this is necessary to avoid the return valuye optimization
2670 that may not be safe in the maximum case). */
2672 res
->under4k
= false;
2675 /* Only warn at level 2. */
2678 || (!maxunder4k
&& fmtres
.range
.max
< HOST_WIDE_INT_MAX
)))
2680 /* The directive output may be longer than the maximum required
2681 to be handled by an implementation according to 7.21.6.1, p15
2682 of C11. Warn on this only at level 2 but remember this and
2683 prevent folding the return value when done. This allows for
2684 the possibility of the actual libc call failing due to ENOMEM
2685 (like Glibc does under some conditions). */
2687 if (fmtres
.range
.min
== fmtres
.range
.max
)
2688 warned
= fmtwarn (dirloc
, pargrange
, NULL
,
2690 "%<%.*s%> directive output of %wu bytes exceeds "
2691 "minimum required size of 4095",
2692 dirlen
, dir
.beg
, fmtres
.range
.min
);
2697 ? G_("%<%.*s%> directive output between %wu and %wu "
2698 "bytes may exceed minimum required size of 4095")
2699 : G_("%<%.*s%> directive output between %wu and %wu "
2700 "bytes exceeds minimum required size of 4095"));
2702 warned
= fmtwarn (dirloc
, pargrange
, NULL
,
2703 info
.warnopt (), fmtstr
,
2705 fmtres
.range
.min
, fmtres
.range
.max
);
2709 /* Has the likely and maximum directive output exceeded INT_MAX? */
2710 bool likelyximax
= *dir
.beg
&& res
->range
.likely
> target_int_max ();
2711 /* Don't consider the maximum to be in excess when it's the result
2712 of a string of unknown length (i.e., whose maximum has been set
2713 to be greater than or equal to HOST_WIDE_INT_MAX. */
2714 bool maxximax
= (*dir
.beg
2715 && res
->range
.max
> target_int_max ()
2716 && res
->range
.max
< HOST_WIDE_INT_MAX
);
2719 /* Warn for the likely output size at level 1. */
2721 /* But only warn for the maximum at level 2. */
2724 && fmtres
.range
.max
< HOST_WIDE_INT_MAX
)))
2726 /* The directive output causes the total length of output
2727 to exceed INT_MAX bytes. */
2729 if (fmtres
.range
.min
== fmtres
.range
.max
)
2730 warned
= fmtwarn (dirloc
, pargrange
, NULL
, info
.warnopt (),
2731 "%<%.*s%> directive output of %wu bytes causes "
2732 "result to exceed %<INT_MAX%>",
2733 dirlen
, dir
.beg
, fmtres
.range
.min
);
2737 = (fmtres
.range
.min
> target_int_max ()
2738 ? G_ ("%<%.*s%> directive output between %wu and %wu "
2739 "bytes causes result to exceed %<INT_MAX%>")
2740 : G_ ("%<%.*s%> directive output between %wu and %wu "
2741 "bytes may cause result to exceed %<INT_MAX%>"));
2742 warned
= fmtwarn (dirloc
, pargrange
, NULL
,
2743 info
.warnopt (), fmtstr
,
2745 fmtres
.range
.min
, fmtres
.range
.max
);
2749 if (warned
&& fmtres
.range
.min
< fmtres
.range
.likely
2750 && fmtres
.range
.likely
< fmtres
.range
.max
)
2752 inform (info
.fmtloc
,
2753 (1 == fmtres
.range
.likely
2754 ? G_("assuming directive output of %wu byte")
2755 : G_("assuming directive output of %wu bytes")),
2756 fmtres
.range
.likely
);
2759 if (warned
&& fmtres
.argmin
)
2761 if (fmtres
.argmin
== fmtres
.argmax
)
2762 inform (info
.fmtloc
, "directive argument %qE", fmtres
.argmin
);
2763 else if (fmtres
.knownrange
)
2764 inform (info
.fmtloc
, "directive argument in the range [%E, %E]",
2765 fmtres
.argmin
, fmtres
.argmax
);
2767 inform (info
.fmtloc
,
2768 "using the range [%E, %E] for directive argument",
2769 fmtres
.argmin
, fmtres
.argmax
);
2772 res
->warned
|= warned
;
2774 if (!dir
.beg
[0] && res
->warned
&& info
.objsize
< HOST_WIDE_INT_MAX
)
2776 /* If a warning has been issued for buffer overflow or truncation
2777 (but not otherwise) help the user figure out how big a buffer
2780 location_t callloc
= gimple_location (info
.callstmt
);
2782 unsigned HOST_WIDE_INT min
= res
->range
.min
;
2783 unsigned HOST_WIDE_INT max
= res
->range
.max
;
2788 ? G_("%qE output %wu byte into a destination of size %wu")
2789 : G_("%qE output %wu bytes into a destination of size %wu")),
2790 info
.func
, min
, info
.objsize
);
2791 else if (max
< HOST_WIDE_INT_MAX
)
2793 "%qE output between %wu and %wu bytes into "
2794 "a destination of size %wu",
2795 info
.func
, min
, max
, info
.objsize
);
2796 else if (min
< res
->range
.likely
&& res
->range
.likely
< max
)
2798 "%qE output %wu or more bytes (assuming %wu) into "
2799 "a destination of size %wu",
2800 info
.func
, min
, res
->range
.likely
, info
.objsize
);
2803 "%qE output %wu or more bytes into a destination of size %wu",
2804 info
.func
, min
, info
.objsize
);
2807 if (dump_file
&& *dir
.beg
)
2809 fprintf (dump_file
, " Result: %lli, %lli, %lli, %lli "
2810 "(%lli, %lli, %lli, %lli)\n",
2811 (long long)fmtres
.range
.min
,
2812 (long long)fmtres
.range
.likely
,
2813 (long long)fmtres
.range
.max
,
2814 (long long)fmtres
.range
.unlikely
,
2815 (long long)res
->range
.min
,
2816 (long long)res
->range
.likely
,
2817 (long long)res
->range
.max
,
2818 (long long)res
->range
.unlikely
);
2824 #pragma GCC diagnostic pop
2826 /* Parse a format directive in function call described by INFO starting
2827 at STR and populate DIR structure. Bump up *ARGNO by the number of
2828 arguments extracted for the directive. Return the length of
2832 parse_directive (pass_sprintf_length::call_info
&info
,
2833 directive
&dir
, format_result
*res
,
2834 const char *str
, unsigned *argno
)
2836 const char *pcnt
= strchr (str
, '%');
2839 if (size_t len
= pcnt
? pcnt
- str
: *str
? strlen (str
) : 1)
2841 /* This directive is either a plain string or the terminating nul
2842 (which isn't really a directive but it simplifies things to
2843 handle it as if it were). */
2845 dir
.fmtfunc
= format_plain
;
2849 fprintf (dump_file
, " Directive %u at offset %llu: \"%.*s\", "
2852 (unsigned long long)(size_t)(dir
.beg
- info
.fmtstr
),
2853 (int)dir
.len
, dir
.beg
, (unsigned long long)dir
.len
);
2859 const char *pf
= pcnt
+ 1;
2861 /* POSIX numbered argument index or zero when none. */
2862 unsigned dollar
= 0;
2864 /* With and precision. -1 when not specified, HOST_WIDE_INT_MIN
2865 when given by a va_list argument, and a non-negative value
2866 when specified in the format string itself. */
2867 HOST_WIDE_INT width
= -1;
2868 HOST_WIDE_INT precision
= -1;
2870 /* Width specified via the asterisk. Need not be INTEGER_CST.
2871 For vararg functions set to void_node. */
2872 tree star_width
= NULL_TREE
;
2874 /* Width specified via the asterisk. Need not be INTEGER_CST.
2875 For vararg functions set to void_node. */
2876 tree star_precision
= NULL_TREE
;
2880 /* This could be either a POSIX positional argument, the '0'
2881 flag, or a width, depending on what follows. Store it as
2882 width and sort it out later after the next character has
2885 width
= strtol (pf
, &end
, 10);
2888 else if ('*' == *pf
)
2890 /* Similarly to the block above, this could be either a POSIX
2891 positional argument or a width, depending on what follows. */
2892 if (*argno
< gimple_call_num_args (info
.callstmt
))
2893 star_width
= gimple_call_arg (info
.callstmt
, (*argno
)++);
2895 star_width
= void_node
;
2901 /* Handle the POSIX dollar sign which references the 1-based
2902 positional argument number. */
2904 dollar
= width
+ info
.argidx
;
2906 && TREE_CODE (star_width
) == INTEGER_CST
)
2907 dollar
= width
+ tree_to_shwi (star_width
);
2909 /* Bail when the numbered argument is out of range (it will
2910 have already been diagnosed by -Wformat). */
2912 || dollar
== info
.argidx
2913 || dollar
> gimple_call_num_args (info
.callstmt
))
2918 star_width
= NULL_TREE
;
2923 if (dollar
|| !star_width
)
2929 /* The '0' that has been interpreted as a width above is
2930 actually a flag. Reset HAVE_WIDTH, set the '0' flag,
2931 and continue processing other flags. */
2937 /* (Non-zero) width has been seen. The next character
2938 is either a period or a digit. */
2939 goto start_precision
;
2942 /* When either '$' has been seen, or width has not been seen,
2943 the next field is the optional flags followed by an optional
2953 dir
.set_flag (*pf
++);
2965 width
= strtol (pf
, &end
, 10);
2968 else if ('*' == *pf
)
2970 if (*argno
< gimple_call_num_args (info
.callstmt
))
2971 star_width
= gimple_call_arg (info
.callstmt
, (*argno
)++);
2974 /* This is (likely) a va_list. It could also be an invalid
2975 call with insufficient arguments. */
2976 star_width
= void_node
;
2980 else if ('\'' == *pf
)
2982 /* The POSIX apostrophe indicating a numeric grouping
2983 in the current locale. Even though it's possible to
2984 estimate the upper bound on the size of the output
2985 based on the number of digits it probably isn't worth
2999 precision
= strtol (pf
, &end
, 10);
3002 else if ('*' == *pf
)
3004 if (*argno
< gimple_call_num_args (info
.callstmt
))
3005 star_precision
= gimple_call_arg (info
.callstmt
, (*argno
)++);
3008 /* This is (likely) a va_list. It could also be an invalid
3009 call with insufficient arguments. */
3010 star_precision
= void_node
;
3016 /* The decimal precision or the asterisk are optional.
3017 When neither is dirified it's taken to be zero. */
3028 dir
.modifier
= FMT_LEN_hh
;
3031 dir
.modifier
= FMT_LEN_h
;
3036 dir
.modifier
= FMT_LEN_j
;
3041 dir
.modifier
= FMT_LEN_L
;
3049 dir
.modifier
= FMT_LEN_ll
;
3052 dir
.modifier
= FMT_LEN_l
;
3057 dir
.modifier
= FMT_LEN_t
;
3062 dir
.modifier
= FMT_LEN_z
;
3069 /* Handle a sole '%' character the same as "%%" but since it's
3070 undefined prevent the result from being folded. */
3073 res
->range
.min
= res
->range
.max
= HOST_WIDE_INT_M1U
;
3076 dir
.fmtfunc
= format_percent
;
3087 res
->floating
= true;
3088 dir
.fmtfunc
= format_floating
;
3097 dir
.fmtfunc
= format_integer
;
3101 /* The %p output is implementation-defined. It's possible
3102 to determine this format but due to extensions (edirially
3103 those of the Linux kernel -- see bug 78512) the first %p
3104 in the format string disables any further processing. */
3108 /* %n has side-effects even when nothing is actually printed to
3110 info
.nowrite
= false;
3111 dir
.fmtfunc
= format_none
;
3115 dir
.fmtfunc
= format_character
;
3120 dir
.fmtfunc
= format_string
;
3124 /* Unknown conversion specification. */
3128 dir
.specifier
= *pf
++;
3132 if (TREE_CODE (TREE_TYPE (star_width
)) == INTEGER_TYPE
)
3133 dir
.set_width (star_width
);
3136 /* Width specified by a va_list takes on the range [0, -INT_MIN]
3137 (width is the absolute value of that specified). */
3139 dir
.width
[1] = target_int_max () + 1;
3143 dir
.set_width (width
);
3147 if (TREE_CODE (TREE_TYPE (star_precision
)) == INTEGER_TYPE
)
3148 dir
.set_precision (star_precision
);
3151 /* Precision specified by a va_list takes on the range [-1, INT_MAX]
3152 (unlike width, negative precision is ignored). */
3154 dir
.prec
[1] = target_int_max ();
3158 dir
.set_precision (precision
);
3160 /* Extract the argument if the directive takes one and if it's
3161 available (e.g., the function doesn't take a va_list). Treat
3162 missing arguments the same as va_list, even though they will
3163 have likely already been diagnosed by -Wformat. */
3164 if (dir
.specifier
!= '%'
3165 && *argno
< gimple_call_num_args (info
.callstmt
))
3166 dir
.arg
= gimple_call_arg (info
.callstmt
, dollar
? dollar
: (*argno
)++);
3168 /* Return the length of the format directive. */
3169 dir
.len
= pf
- pcnt
;
3173 fprintf (dump_file
, " Directive %u at offset %llu: \"%.*s\"",
3174 dir
.dirno
, (unsigned long long)(size_t)(dir
.beg
- info
.fmtstr
),
3175 (int)dir
.len
, dir
.beg
);
3178 if (dir
.width
[0] == dir
.width
[1])
3179 fprintf (dump_file
, ", width = %lli", (long long)dir
.width
[0]);
3181 fprintf (dump_file
, ", width in range [%lli, %lli]",
3182 (long long)dir
.width
[0], (long long)dir
.width
[1]);
3187 if (dir
.prec
[0] == dir
.prec
[1])
3188 fprintf (dump_file
, ", precision = %lli", (long long)dir
.prec
[0]);
3190 fprintf (dump_file
, ", precision in range [%lli, %lli]",
3191 (long long)dir
.prec
[0], (long long)dir
.prec
[1]);
3193 fputc ('\n', dump_file
);
3199 /* Compute the length of the output resulting from the call to a formatted
3200 output function described by INFO and store the result of the call in
3201 *RES. Issue warnings for detected past the end writes. Return true
3202 if the complete format string has been processed and *RES can be relied
3203 on, false otherwise (e.g., when a unknown or unhandled directive was seen
3204 that caused the processing to be terminated early). */
3207 pass_sprintf_length::compute_format_length (call_info
&info
,
3212 location_t callloc
= gimple_location (info
.callstmt
);
3213 fprintf (dump_file
, "%s:%i: ",
3214 LOCATION_FILE (callloc
), LOCATION_LINE (callloc
));
3215 print_generic_expr (dump_file
, info
.func
, dump_flags
);
3217 fprintf (dump_file
, ": objsize = %llu, fmtstr = \"%s\"\n",
3218 (unsigned long long)info
.objsize
, info
.fmtstr
);
3221 /* Reset the minimum and maximum byte counters. */
3222 res
->range
.min
= res
->range
.max
= 0;
3224 /* No directive has been seen yet so the length of output is bounded
3225 by the known range [0, 0] (with no conversion producing more than
3226 4K bytes) until determined otherwise. */
3227 res
->knownrange
= true;
3228 res
->under4k
= true;
3229 res
->floating
= false;
3230 res
->warned
= false;
3232 /* 1-based directive counter. */
3235 /* The variadic argument counter. */
3236 unsigned argno
= info
.argidx
;
3238 for (const char *pf
= info
.fmtstr
; ; ++dirno
)
3240 directive dir
= directive ();
3243 size_t n
= parse_directive (info
, dir
, res
, pf
, &argno
);
3245 /* Return failure if the format function fails. */
3246 if (!format_directive (info
, res
, dir
))
3249 /* Return success the directive is zero bytes long and it's
3250 the last think in the format string (i.e., it's the terminating
3251 nul, which isn't really a directive but handling it as one makes
3259 /* The complete format string was processed (with or without warnings). */
3263 /* Return the size of the object referenced by the expression DEST if
3264 available, or -1 otherwise. */
3266 static unsigned HOST_WIDE_INT
3267 get_destination_size (tree dest
)
3269 /* Initialize object size info before trying to compute it. */
3270 init_object_sizes ();
3272 /* Use __builtin_object_size to determine the size of the destination
3273 object. When optimizing, determine the smallest object (such as
3274 a member array as opposed to the whole enclosing object), otherwise
3275 use type-zero object size to determine the size of the enclosing
3276 object (the function fails without optimization in this type). */
3277 int ost
= optimize
> 0;
3278 unsigned HOST_WIDE_INT size
;
3279 if (compute_builtin_object_size (dest
, ost
, &size
))
3282 return HOST_WIDE_INT_M1U
;
3285 /* Given a suitable result RES of a call to a formatted output function
3286 described by INFO, substitute the result for the return value of
3287 the call. The result is suitable if the number of bytes it represents
3288 is known and exact. A result that isn't suitable for substitution may
3289 have its range set to the range of return values, if that is known.
3290 Return true if the call is removed and gsi_next should not be performed
3294 try_substitute_return_value (gimple_stmt_iterator
*gsi
,
3295 const pass_sprintf_length::call_info
&info
,
3296 const format_result
&res
)
3298 tree lhs
= gimple_get_lhs (info
.callstmt
);
3300 /* Set to true when the entire call has been removed. */
3301 bool removed
= false;
3303 /* The minimum return value. */
3304 unsigned HOST_WIDE_INT minretval
= res
.range
.min
;
3306 /* The maximum return value is in most cases bounded by RES.RANGE.MAX
3307 but in cases involving multibyte characters could be as large as
3308 RES.RANGE.UNLIKELY. */
3309 unsigned HOST_WIDE_INT maxretval
3310 = res
.range
.unlikely
< res
.range
.max
? res
.range
.max
: res
.range
.unlikely
;
3312 /* Adjust the number of bytes which includes the terminating nul
3313 to reflect the return value of the function which does not.
3314 Because the valid range of the function is [INT_MIN, INT_MAX],
3315 a valid range before the adjustment below is [0, INT_MAX + 1]
3316 (the functions only return negative values on error or undefined
3318 if (minretval
<= target_int_max () + 1)
3320 if (maxretval
<= target_int_max () + 1)
3323 /* Avoid the return value optimization when the behavior of the call
3324 is undefined either because any directive may have produced 4K or
3325 more of output, or the return value exceeds INT_MAX, or because
3326 the output overflows the destination object (but leave it enabled
3327 when the function is bounded because then the behavior is well-
3330 && minretval
== maxretval
3331 && (info
.bounded
|| minretval
< info
.objsize
)
3332 && minretval
<= target_int_max ()
3333 /* Not prepared to handle possibly throwing calls here; they shouldn't
3334 appear in non-artificial testcases, except when the __*_chk routines
3335 are badly declared. */
3336 && !stmt_ends_bb_p (info
.callstmt
))
3338 tree cst
= build_int_cst (integer_type_node
, minretval
);
3340 if (lhs
== NULL_TREE
3343 /* Remove the call to the bounded function with a zero size
3344 (e.g., snprintf(0, 0, "%i", 123)) if there is no lhs. */
3345 unlink_stmt_vdef (info
.callstmt
);
3346 gsi_remove (gsi
, true);
3349 else if (info
.nowrite
)
3351 /* Replace the call to the bounded function with a zero size
3352 (e.g., snprintf(0, 0, "%i", 123) with the constant result
3354 if (!update_call_from_tree (gsi
, cst
))
3355 gimplify_and_update_call_from_tree (gsi
, cst
);
3356 gimple
*callstmt
= gsi_stmt (*gsi
);
3357 update_stmt (callstmt
);
3361 /* Replace the left-hand side of the call with the constant
3362 result of the formatted function. */
3363 gimple_call_set_lhs (info
.callstmt
, NULL_TREE
);
3364 gimple
*g
= gimple_build_assign (lhs
, cst
);
3365 gsi_insert_after (gsi
, g
, GSI_NEW_STMT
);
3366 update_stmt (info
.callstmt
);
3372 fprintf (dump_file
, " Removing call statement.");
3375 fprintf (dump_file
, " Substituting ");
3376 print_generic_expr (dump_file
, cst
, dump_flags
);
3377 fprintf (dump_file
, " for %s.\n",
3378 info
.nowrite
? "statement" : "return value");
3384 bool setrange
= false;
3386 if ((info
.bounded
|| maxretval
< info
.objsize
)
3388 && (minretval
< target_int_max ()
3389 && maxretval
< target_int_max ()))
3391 /* If the result is in a valid range bounded by the size of
3392 the destination set it so that it can be used for subsequent
3394 int prec
= TYPE_PRECISION (integer_type_node
);
3396 wide_int min
= wi::shwi (minretval
, prec
);
3397 wide_int max
= wi::shwi (maxretval
, prec
);
3398 set_range_info (lhs
, VR_RANGE
, min
, max
);
3405 const char *inbounds
3406 = (minretval
< info
.objsize
3407 ? (maxretval
< info
.objsize
3408 ? "in" : "potentially out-of")
3411 const char *what
= setrange
? "Setting" : "Discarding";
3412 if (minretval
!= maxretval
)
3414 " %s %s-bounds return value range [%llu, %llu].\n",
3416 (unsigned long long)minretval
,
3417 (unsigned long long)maxretval
);
3419 fprintf (dump_file
, " %s %s-bounds return value %llu.\n",
3420 what
, inbounds
, (unsigned long long)minretval
);
3425 fputc ('\n', dump_file
);
3430 /* Determine if a GIMPLE CALL is to one of the sprintf-like built-in
3431 functions and if so, handle it. Return true if the call is removed
3432 and gsi_next should not be performed in the caller. */
3435 pass_sprintf_length::handle_gimple_call (gimple_stmt_iterator
*gsi
)
3437 call_info info
= call_info ();
3439 info
.callstmt
= gsi_stmt (*gsi
);
3440 if (!gimple_call_builtin_p (info
.callstmt
, BUILT_IN_NORMAL
))
3443 info
.func
= gimple_call_fndecl (info
.callstmt
);
3444 info
.fncode
= DECL_FUNCTION_CODE (info
.func
);
3446 /* The size of the destination as in snprintf(dest, size, ...). */
3447 unsigned HOST_WIDE_INT dstsize
= HOST_WIDE_INT_M1U
;
3449 /* The size of the destination determined by __builtin_object_size. */
3450 unsigned HOST_WIDE_INT objsize
= HOST_WIDE_INT_M1U
;
3452 /* Buffer size argument number (snprintf and vsnprintf). */
3453 unsigned HOST_WIDE_INT idx_dstsize
= HOST_WIDE_INT_M1U
;
3455 /* Object size argument number (snprintf_chk and vsnprintf_chk). */
3456 unsigned HOST_WIDE_INT idx_objsize
= HOST_WIDE_INT_M1U
;
3458 /* Format string argument number (valid for all functions). */
3459 unsigned idx_format
;
3461 switch (info
.fncode
)
3463 case BUILT_IN_SPRINTF
:
3465 // __builtin_sprintf (dst, format, ...)
3470 case BUILT_IN_SPRINTF_CHK
:
3472 // __builtin___sprintf_chk (dst, ost, objsize, format, ...)
3478 case BUILT_IN_SNPRINTF
:
3480 // __builtin_snprintf (dst, size, format, ...)
3484 info
.bounded
= true;
3487 case BUILT_IN_SNPRINTF_CHK
:
3489 // __builtin___snprintf_chk (dst, size, ost, objsize, format, ...)
3494 info
.bounded
= true;
3497 case BUILT_IN_VSNPRINTF
:
3499 // __builtin_vsprintf (dst, size, format, va)
3503 info
.bounded
= true;
3506 case BUILT_IN_VSNPRINTF_CHK
:
3508 // __builtin___vsnprintf_chk (dst, size, ost, objsize, format, va)
3513 info
.bounded
= true;
3516 case BUILT_IN_VSPRINTF
:
3518 // __builtin_vsprintf (dst, format, va)
3523 case BUILT_IN_VSPRINTF_CHK
:
3525 // __builtin___vsprintf_chk (dst, ost, objsize, format, va)
3535 /* Set the global warning level for this function. */
3536 warn_level
= info
.bounded
? warn_format_trunc
: warn_format_overflow
;
3538 /* The first argument is a pointer to the destination. */
3539 tree dstptr
= gimple_call_arg (info
.callstmt
, 0);
3541 info
.format
= gimple_call_arg (info
.callstmt
, idx_format
);
3543 /* True when the destination size is constant as opposed to the lower
3544 or upper bound of a range. */
3545 bool dstsize_cst_p
= true;
3547 if (idx_dstsize
== HOST_WIDE_INT_M1U
)
3549 /* For non-bounded functions like sprintf, determine the size
3550 of the destination from the object or pointer passed to it
3551 as the first argument. */
3552 dstsize
= get_destination_size (dstptr
);
3554 else if (tree size
= gimple_call_arg (info
.callstmt
, idx_dstsize
))
3556 /* For bounded functions try to get the size argument. */
3558 if (TREE_CODE (size
) == INTEGER_CST
)
3560 dstsize
= tree_to_uhwi (size
);
3561 /* No object can be larger than SIZE_MAX bytes (half the address
3562 space) on the target.
3563 The functions are defined only for output of at most INT_MAX
3564 bytes. Specifying a bound in excess of that limit effectively
3565 defeats the bounds checking (and on some implementations such
3566 as Solaris cause the function to fail with EINVAL). */
3567 if (dstsize
> target_size_max () / 2)
3569 /* Avoid warning if -Wstringop-overflow is specified since
3570 it also warns for the same thing though only for the
3571 checking built-ins. */
3572 if ((idx_objsize
== HOST_WIDE_INT_M1U
3573 || !warn_stringop_overflow
))
3574 warning_at (gimple_location (info
.callstmt
), info
.warnopt (),
3575 "specified bound %wu exceeds maximum object size "
3577 dstsize
, target_size_max () / 2);
3579 else if (dstsize
> target_int_max ())
3580 warning_at (gimple_location (info
.callstmt
), info
.warnopt (),
3581 "specified bound %wu exceeds %<INT_MAX %>",
3584 else if (TREE_CODE (size
) == SSA_NAME
)
3586 /* Try to determine the range of values of the argument
3587 and use the greater of the two at level 1 and the smaller
3588 of them at level 2. */
3590 enum value_range_type range_type
3591 = get_range_info (size
, &min
, &max
);
3592 if (range_type
== VR_RANGE
)
3596 ? wi::fits_uhwi_p (max
) ? max
.to_uhwi () : max
.to_shwi ()
3597 : wi::fits_uhwi_p (min
) ? min
.to_uhwi () : min
.to_shwi ());
3600 /* The destination size is not constant. If the function is
3601 bounded (e.g., snprintf) a lower bound of zero doesn't
3602 necessarily imply it can be eliminated. */
3603 dstsize_cst_p
= false;
3607 if (idx_objsize
!= HOST_WIDE_INT_M1U
)
3608 if (tree size
= gimple_call_arg (info
.callstmt
, idx_objsize
))
3609 if (tree_fits_uhwi_p (size
))
3610 objsize
= tree_to_uhwi (size
);
3612 if (info
.bounded
&& !dstsize
)
3614 /* As a special case, when the explicitly specified destination
3615 size argument (to a bounded function like snprintf) is zero
3616 it is a request to determine the number of bytes on output
3617 without actually producing any. Pretend the size is
3618 unlimited in this case. */
3619 info
.objsize
= HOST_WIDE_INT_MAX
;
3620 info
.nowrite
= dstsize_cst_p
;
3624 /* For calls to non-bounded functions or to those of bounded
3625 functions with a non-zero size, warn if the destination
3627 if (integer_zerop (dstptr
))
3629 /* This is diagnosed with -Wformat only when the null is a constant
3630 pointer. The warning here diagnoses instances where the pointer
3632 location_t loc
= gimple_location (info
.callstmt
);
3633 warning_at (EXPR_LOC_OR_LOC (dstptr
, loc
),
3634 info
.warnopt (), "null destination pointer");
3638 /* Set the object size to the smaller of the two arguments
3639 of both have been specified and they're not equal. */
3640 info
.objsize
= dstsize
< objsize
? dstsize
: objsize
;
3643 && dstsize
< target_size_max () / 2 && objsize
< dstsize
3644 /* Avoid warning if -Wstringop-overflow is specified since
3645 it also warns for the same thing though only for the
3646 checking built-ins. */
3647 && (idx_objsize
== HOST_WIDE_INT_M1U
3648 || !warn_stringop_overflow
))
3650 warning_at (gimple_location (info
.callstmt
), info
.warnopt (),
3651 "specified bound %wu exceeds the size %wu "
3652 "of the destination object", dstsize
, objsize
);
3656 if (integer_zerop (info
.format
))
3658 /* This is diagnosed with -Wformat only when the null is a constant
3659 pointer. The warning here diagnoses instances where the pointer
3661 location_t loc
= gimple_location (info
.callstmt
);
3662 warning_at (EXPR_LOC_OR_LOC (info
.format
, loc
),
3663 info
.warnopt (), "null format string");
3667 info
.fmtstr
= get_format_string (info
.format
, &info
.fmtloc
);
3671 /* The result is the number of bytes output by the formatted function,
3672 including the terminating NUL. */
3673 format_result res
= format_result ();
3675 bool success
= compute_format_length (info
, &res
);
3677 /* When optimizing and the printf return value optimization is enabled,
3678 attempt to substitute the computed result for the return value of
3679 the call. Avoid this optimization when -frounding-math is in effect
3680 and the format string contains a floating point directive. */
3683 && flag_printf_return_value
3684 && (!flag_rounding_math
|| !res
.floating
))
3685 return try_substitute_return_value (gsi
, info
, res
);
3690 /* Execute the pass for function FUN. */
3693 pass_sprintf_length::execute (function
*fun
)
3696 FOR_EACH_BB_FN (bb
, fun
)
3698 for (gimple_stmt_iterator si
= gsi_start_bb (bb
); !gsi_end_p (si
); )
3700 /* Iterate over statements, looking for function calls. */
3701 gimple
*stmt
= gsi_stmt (si
);
3703 if (is_gimple_call (stmt
) && handle_gimple_call (&si
))
3704 /* If handle_gimple_call returns true, the iterator is
3705 already pointing to the next statement. */
3712 /* Clean up object size info. */
3713 fini_object_sizes ();
3718 } /* Unnamed namespace. */
3720 /* Return a pointer to a pass object newly constructed from the context
3724 make_pass_sprintf_length (gcc::context
*ctxt
)
3726 return new pass_sprintf_length (ctxt
);