1//===-- sanitizer_printf.cpp ----------------------------------------------===//
2//
3// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4// See https://llvm.org/LICENSE.txt for license information.
5// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6//
7//===----------------------------------------------------------------------===//
8//
9// This file is shared between AddressSanitizer and ThreadSanitizer.
10//
11// Internal printf function, used inside run-time libraries.
12// We can't use libc printf because we intercept some of the functions used
13// inside it.
14//===----------------------------------------------------------------------===//
15
16#include "sanitizer_common.h"
17#include "sanitizer_flags.h"
18#include "sanitizer_libc.h"
19
20#include <stdio.h>
21#include <stdarg.h>
22
23#if SANITIZER_WINDOWS && defined(_MSC_VER) && _MSC_VER < 1800 &&               \
24      !defined(va_copy)
25# define va_copy(dst, src) ((dst) = (src))
26#endif
27
28namespace __sanitizer {
29
30static int AppendChar(char **buff, const char *buff_end, char c) {
31  if (*buff < buff_end) {
32    **buff = c;
33    (*buff)++;
34  }
35  return 1;
36}
37
38// Appends number in a given base to buffer. If its length is less than
39// |minimal_num_length|, it is padded with leading zeroes or spaces, depending
40// on the value of |pad_with_zero|.
41static int AppendNumber(char **buff, const char *buff_end, u64 absolute_value,
42                        u8 base, u8 minimal_num_length, bool pad_with_zero,
43                        bool negative, bool uppercase) {
44  uptr const kMaxLen = 30;
45  RAW_CHECK(base == 10 || base == 16);
46  RAW_CHECK(base == 10 || !negative);
47  RAW_CHECK(absolute_value || !negative);
48  RAW_CHECK(minimal_num_length < kMaxLen);
49  int result = 0;
50  if (negative && minimal_num_length)
51    --minimal_num_length;
52  if (negative && pad_with_zero)
53    result += AppendChar(buff, buff_end, '-');
54  uptr num_buffer[kMaxLen];
55  int pos = 0;
56  do {
57    RAW_CHECK_MSG((uptr)pos < kMaxLen, "AppendNumber buffer overflow");
58    num_buffer[pos++] = absolute_value % base;
59    absolute_value /= base;
60  } while (absolute_value > 0);
61  if (pos < minimal_num_length) {
62    // Make sure compiler doesn't insert call to memset here.
63    internal_memset(&num_buffer[pos], 0,
64                    sizeof(num_buffer[0]) * (minimal_num_length - pos));
65    pos = minimal_num_length;
66  }
67  RAW_CHECK(pos > 0);
68  pos--;
69  for (; pos >= 0 && num_buffer[pos] == 0; pos--) {
70    char c = (pad_with_zero || pos == 0) ? '0' : ' ';
71    result += AppendChar(buff, buff_end, c);
72  }
73  if (negative && !pad_with_zero) result += AppendChar(buff, buff_end, '-');
74  for (; pos >= 0; pos--) {
75    char digit = static_cast<char>(num_buffer[pos]);
76    digit = (digit < 10) ? '0' + digit : (uppercase ? 'A' : 'a') + digit - 10;
77    result += AppendChar(buff, buff_end, digit);
78  }
79  return result;
80}
81
82static int AppendUnsigned(char **buff, const char *buff_end, u64 num, u8 base,
83                          u8 minimal_num_length, bool pad_with_zero,
84                          bool uppercase) {
85  return AppendNumber(buff, buff_end, num, base, minimal_num_length,
86                      pad_with_zero, false /* negative */, uppercase);
87}
88
89static int AppendSignedDecimal(char **buff, const char *buff_end, s64 num,
90                               u8 minimal_num_length, bool pad_with_zero) {
91  bool negative = (num < 0);
92  return AppendNumber(buff, buff_end, (u64)(negative ? -num : num), 10,
93                      minimal_num_length, pad_with_zero, negative,
94                      false /* uppercase */);
95}
96
97
98// Use the fact that explicitly requesting 0 width (%0s) results in UB and
99// interpret width == 0 as "no width requested":
100// width == 0 - no width requested
101// width  < 0 - left-justify s within and pad it to -width chars, if necessary
102// width  > 0 - right-justify s, not implemented yet
103static int AppendString(char **buff, const char *buff_end, int width,
104                        int max_chars, const char *s) {
105  if (!s)
106    s = "<null>";
107  int result = 0;
108  for (; *s; s++) {
109    if (max_chars >= 0 && result >= max_chars)
110      break;
111    result += AppendChar(buff, buff_end, *s);
112  }
113  // Only the left justified strings are supported.
114  while (width < -result)
115    result += AppendChar(buff, buff_end, ' ');
116  return result;
117}
118
119static int AppendPointer(char **buff, const char *buff_end, u64 ptr_value) {
120  int result = 0;
121  result += AppendString(buff, buff_end, 0, -1, "0x");
122  result += AppendUnsigned(buff, buff_end, ptr_value, 16,
123                           SANITIZER_POINTER_FORMAT_LENGTH,
124                           true /* pad_with_zero */, false /* uppercase */);
125  return result;
126}
127
128int VSNPrintf(char *buff, int buff_length,
129              const char *format, va_list args) {
130  static const char *kPrintfFormatsHelp =
131      "Supported Printf formats: %([0-9]*)?(z|l|ll)?{d,u,x,X}; %p; "
132      "%[-]([0-9]*)?(\\.\\*)?s; %c\nProvided format: ";
133  RAW_CHECK(format);
134  RAW_CHECK(buff_length > 0);
135  const char *buff_end = &buff[buff_length - 1];
136  const char *cur = format;
137  int result = 0;
138  for (; *cur; cur++) {
139    if (*cur != '%') {
140      result += AppendChar(&buff, buff_end, *cur);
141      continue;
142    }
143    cur++;
144    bool left_justified = *cur == '-';
145    if (left_justified)
146      cur++;
147    bool have_width = (*cur >= '0' && *cur <= '9');
148    bool pad_with_zero = (*cur == '0');
149    int width = 0;
150    if (have_width) {
151      while (*cur >= '0' && *cur <= '9') {
152        width = width * 10 + *cur++ - '0';
153      }
154    }
155    bool have_precision = (cur[0] == '.' && cur[1] == '*');
156    int precision = -1;
157    if (have_precision) {
158      cur += 2;
159      precision = va_arg(args, int);
160    }
161    bool have_z = (*cur == 'z');
162    cur += have_z;
163    bool have_l = cur[0] == 'l' && cur[1] != 'l';
164    cur += have_l;
165    bool have_ll = cur[0] == 'l' && cur[1] == 'l';
166    cur += have_ll * 2;
167    const bool have_length = have_z || have_l || have_ll;
168    const bool have_flags = have_width || have_length;
169    // At the moment only %s supports precision and left-justification.
170    CHECK(!((precision >= 0 || left_justified) && *cur != 's'));
171    switch (*cur) {
172      case 'd': {
173        s64 dval = have_ll  ? va_arg(args, s64)
174                   : have_z ? va_arg(args, sptr)
175                   : have_l ? va_arg(args, long)
176                            : va_arg(args, int);
177        result += AppendSignedDecimal(&buff, buff_end, dval, width,
178                                      pad_with_zero);
179        break;
180      }
181      case 'u':
182      case 'x':
183      case 'X': {
184        u64 uval = have_ll  ? va_arg(args, u64)
185                   : have_z ? va_arg(args, uptr)
186                   : have_l ? va_arg(args, unsigned long)
187                            : va_arg(args, unsigned);
188        bool uppercase = (*cur == 'X');
189        result += AppendUnsigned(&buff, buff_end, uval, (*cur == 'u') ? 10 : 16,
190                                 width, pad_with_zero, uppercase);
191        break;
192      }
193      case 'p': {
194        RAW_CHECK_VA(!have_flags, kPrintfFormatsHelp, format);
195        result += AppendPointer(&buff, buff_end, va_arg(args, uptr));
196        break;
197      }
198      case 's': {
199        RAW_CHECK_VA(!have_length, kPrintfFormatsHelp, format);
200        // Only left-justified width is supported.
201        CHECK(!have_width || left_justified);
202        result += AppendString(&buff, buff_end, left_justified ? -width : width,
203                               precision, va_arg(args, char*));
204        break;
205      }
206      case 'c': {
207        RAW_CHECK_VA(!have_flags, kPrintfFormatsHelp, format);
208        result += AppendChar(&buff, buff_end, va_arg(args, int));
209        break;
210      }
211      case '%' : {
212        RAW_CHECK_VA(!have_flags, kPrintfFormatsHelp, format);
213        result += AppendChar(&buff, buff_end, '%');
214        break;
215      }
216      default: {
217        RAW_CHECK_VA(false, kPrintfFormatsHelp, format);
218      }
219    }
220  }
221  RAW_CHECK(buff <= buff_end);
222  AppendChar(&buff, buff_end + 1, '\0');
223  return result;
224}
225
226static void (*PrintfAndReportCallback)(const char *);
227void SetPrintfAndReportCallback(void (*callback)(const char *)) {
228  PrintfAndReportCallback = callback;
229}
230
231// Can be overriden in frontend.
232#if SANITIZER_GO && defined(TSAN_EXTERNAL_HOOKS)
233// Implementation must be defined in frontend.
234extern "C" void __sanitizer_on_print(const char *str);
235#else
236SANITIZER_INTERFACE_WEAK_DEF(void, __sanitizer_on_print, const char *str) {
237  (void)str;
238}
239#endif
240
241static void CallPrintfAndReportCallback(const char *str) {
242  __sanitizer_on_print(str);
243  if (PrintfAndReportCallback)
244    PrintfAndReportCallback(str);
245}
246
247static void NOINLINE SharedPrintfCodeNoBuffer(bool append_pid,
248                                              char *local_buffer,
249                                              int buffer_size,
250                                              const char *format,
251                                              va_list args) {
252  va_list args2;
253  va_copy(args2, args);
254  InternalMmapVector<char> v;
255  int needed_length = 0;
256  char *buffer = local_buffer;
257  // First try to print a message using a local buffer, and then fall back to
258  // mmaped buffer.
259  for (int use_mmap = 0;; use_mmap++) {
260    if (use_mmap) {
261      va_end(args);
262      va_copy(args, args2);
263      v.resize(needed_length + 1);
264      buffer_size = v.capacity();
265      v.resize(buffer_size);
266      buffer = &v[0];
267    }
268    needed_length = 0;
269    // Fuchsia's logging infrastructure always keeps track of the logging
270    // process, thread, and timestamp, so never prepend such information.
271    if (!SANITIZER_FUCHSIA && append_pid) {
272      int pid = internal_getpid();
273      const char *exe_name = GetProcessName();
274      if (common_flags()->log_exe_name && exe_name) {
275        needed_length += internal_snprintf(buffer, buffer_size,
276                                           "==%s", exe_name);
277        if (needed_length >= buffer_size)
278          continue;
279      }
280      needed_length += internal_snprintf(
281          buffer + needed_length, buffer_size - needed_length, "==%d==", pid);
282      if (needed_length >= buffer_size)
283        continue;
284    }
285    needed_length += VSNPrintf(buffer + needed_length,
286                               buffer_size - needed_length, format, args);
287    if (needed_length >= buffer_size)
288      continue;
289    // If the message fit into the buffer, print it and exit.
290    break;
291  }
292  RawWrite(buffer);
293
294  // Remove color sequences from the message.
295  RemoveANSIEscapeSequencesFromString(buffer);
296  CallPrintfAndReportCallback(buffer);
297  LogMessageOnPrintf(buffer);
298
299  va_end(args2);
300}
301
302static void NOINLINE SharedPrintfCode(bool append_pid, const char *format,
303                                      va_list args) {
304  // |local_buffer| is small enough not to overflow the stack and/or violate
305  // the stack limit enforced by TSan (-Wframe-larger-than=512). On the other
306  // hand, the bigger the buffer is, the more the chance the error report will
307  // fit into it.
308  char local_buffer[400];
309  SharedPrintfCodeNoBuffer(append_pid, local_buffer, ARRAY_SIZE(local_buffer),
310                           format, args);
311}
312
313void Printf(const char *format, ...) {
314  va_list args;
315  va_start(args, format);
316  SharedPrintfCode(false, format, args);
317  va_end(args);
318}
319
320// Like Printf, but prints the current PID before the output string.
321void Report(const char *format, ...) {
322  va_list args;
323  va_start(args, format);
324  SharedPrintfCode(true, format, args);
325  va_end(args);
326}
327
328// Writes at most "length" symbols to "buffer" (including trailing '\0').
329// Returns the number of symbols that should have been written to buffer
330// (not including trailing '\0'). Thus, the string is truncated
331// iff return value is not less than "length".
332int internal_snprintf(char *buffer, uptr length, const char *format, ...) {
333  va_list args;
334  va_start(args, format);
335  int needed_length = VSNPrintf(buffer, length, format, args);
336  va_end(args);
337  return needed_length;
338}
339
340void InternalScopedString::Append(const char *str) {
341  uptr prev_len = length();
342  uptr str_len = internal_strlen(str);
343  buffer_.resize(prev_len + str_len + 1);
344  internal_memcpy(buffer_.data() + prev_len, str, str_len + 1);
345}
346
347void InternalScopedString::AppendF(const char *format, ...) {
348  uptr prev_len = length();
349
350  while (true) {
351    buffer_.resize(buffer_.capacity());
352
353    va_list args;
354    va_start(args, format);
355    uptr sz = VSNPrintf(buffer_.data() + prev_len, buffer_.size() - prev_len,
356                        format, args);
357    va_end(args);
358    if (sz < buffer_.size() - prev_len) {
359      buffer_.resize(prev_len + sz + 1);
360      break;
361    }
362
363    buffer_.reserve(buffer_.capacity() * 2);
364  }
365  CHECK_EQ(buffer_[length()], '\0');
366}
367
368} // namespace __sanitizer
369