blob: bccd32145c6f94fc4984a731f3398567e91261a3 [file] [log] [blame]
Victor Stinner4e314432010-10-07 21:45:39 +00001#include "Python.h"
Stefan Krah6df5cae2012-11-12 20:14:36 +01002#include "osdefs.h"
Stefan Krah6c01e382014-01-20 15:31:08 +01003#include <locale.h>
4
Victor Stinnerb306d752010-10-07 22:09:40 +00005#ifdef MS_WINDOWS
Steve Dowerd81431f2015-03-06 14:47:02 -08006# include <malloc.h>
Victor Stinnerb306d752010-10-07 22:09:40 +00007# include <windows.h>
Steve Dower8fc89802015-04-12 00:26:27 -04008extern int winerror_to_errno(int);
Victor Stinnerb306d752010-10-07 22:09:40 +00009#endif
Victor Stinner4e314432010-10-07 21:45:39 +000010
Brett Cannonefb00c02012-02-29 18:31:31 -050011#ifdef HAVE_LANGINFO_H
12#include <langinfo.h>
13#endif
14
Victor Stinnerdaf45552013-08-28 00:53:59 +020015#ifdef HAVE_SYS_IOCTL_H
16#include <sys/ioctl.h>
17#endif
18
19#ifdef HAVE_FCNTL_H
20#include <fcntl.h>
21#endif /* HAVE_FCNTL_H */
22
Victor Stinnere2623772012-11-12 23:04:02 +010023#ifdef __APPLE__
24extern wchar_t* _Py_DecodeUTF8_surrogateescape(const char *s, Py_ssize_t size);
25#endif
26
Victor Stinnerdaf45552013-08-28 00:53:59 +020027#ifdef O_CLOEXEC
Victor Stinnerb034eee2013-09-07 10:36:04 +020028/* Does open() support the O_CLOEXEC flag? Possible values:
Victor Stinnerdaf45552013-08-28 00:53:59 +020029
30 -1: unknown
31 0: open() ignores O_CLOEXEC flag, ex: Linux kernel older than 2.6.23
32 1: open() supports O_CLOEXEC flag, close-on-exec is set
33
Victor Stinnera555cfc2015-03-18 00:22:14 +010034 The flag is used by _Py_open(), _Py_open_noraise(), io.FileIO
35 and os.open(). */
Victor Stinnerdaf45552013-08-28 00:53:59 +020036int _Py_open_cloexec_works = -1;
37#endif
38
Brett Cannonefb00c02012-02-29 18:31:31 -050039PyObject *
40_Py_device_encoding(int fd)
41{
Victor Stinner14b9b112013-06-25 00:37:25 +020042#if defined(MS_WINDOWS)
Brett Cannonefb00c02012-02-29 18:31:31 -050043 UINT cp;
44#endif
Steve Dower8fc89802015-04-12 00:26:27 -040045 int valid;
46 _Py_BEGIN_SUPPRESS_IPH
47 valid = _PyVerify_fd(fd) && isatty(fd);
48 _Py_END_SUPPRESS_IPH
49 if (!valid)
Brett Cannonefb00c02012-02-29 18:31:31 -050050 Py_RETURN_NONE;
Steve Dower8fc89802015-04-12 00:26:27 -040051
Victor Stinner14b9b112013-06-25 00:37:25 +020052#if defined(MS_WINDOWS)
Brett Cannonefb00c02012-02-29 18:31:31 -050053 if (fd == 0)
54 cp = GetConsoleCP();
55 else if (fd == 1 || fd == 2)
56 cp = GetConsoleOutputCP();
57 else
58 cp = 0;
59 /* GetConsoleCP() and GetConsoleOutputCP() return 0 if the application
60 has no console */
61 if (cp != 0)
62 return PyUnicode_FromFormat("cp%u", (unsigned int)cp);
63#elif defined(CODESET)
64 {
65 char *codeset = nl_langinfo(CODESET);
66 if (codeset != NULL && codeset[0] != 0)
67 return PyUnicode_FromString(codeset);
68 }
69#endif
70 Py_RETURN_NONE;
71}
72
Victor Stinnerd45c7f82012-12-04 01:34:47 +010073#if !defined(__APPLE__) && !defined(MS_WINDOWS)
74extern int _Py_normalize_encoding(const char *, char *, size_t);
75
76/* Workaround FreeBSD and OpenIndiana locale encoding issue with the C locale.
77 On these operating systems, nl_langinfo(CODESET) announces an alias of the
78 ASCII encoding, whereas mbstowcs() and wcstombs() functions use the
79 ISO-8859-1 encoding. The problem is that os.fsencode() and os.fsdecode() use
80 locale.getpreferredencoding() codec. For example, if command line arguments
81 are decoded by mbstowcs() and encoded back by os.fsencode(), we get a
82 UnicodeEncodeError instead of retrieving the original byte string.
83
84 The workaround is enabled if setlocale(LC_CTYPE, NULL) returns "C",
85 nl_langinfo(CODESET) announces "ascii" (or an alias to ASCII), and at least
86 one byte in range 0x80-0xff can be decoded from the locale encoding. The
87 workaround is also enabled on error, for example if getting the locale
88 failed.
89
Philip Jenvey215c49a2013-01-15 13:24:12 -080090 Values of force_ascii:
Victor Stinnerd45c7f82012-12-04 01:34:47 +010091
Victor Stinnerf6a271a2014-08-01 12:28:48 +020092 1: the workaround is used: Py_EncodeLocale() uses
93 encode_ascii_surrogateescape() and Py_DecodeLocale() uses
Victor Stinnerd45c7f82012-12-04 01:34:47 +010094 decode_ascii_surrogateescape()
Victor Stinnerf6a271a2014-08-01 12:28:48 +020095 0: the workaround is not used: Py_EncodeLocale() uses wcstombs() and
96 Py_DecodeLocale() uses mbstowcs()
Victor Stinnerd45c7f82012-12-04 01:34:47 +010097 -1: unknown, need to call check_force_ascii() to get the value
98*/
99static int force_ascii = -1;
100
101static int
102check_force_ascii(void)
103{
104 char *loc;
105#if defined(HAVE_LANGINFO_H) && defined(CODESET)
106 char *codeset, **alias;
107 char encoding[100];
108 int is_ascii;
109 unsigned int i;
110 char* ascii_aliases[] = {
111 "ascii",
112 "646",
113 "ansi-x3.4-1968",
114 "ansi-x3-4-1968",
115 "ansi-x3.4-1986",
116 "cp367",
117 "csascii",
118 "ibm367",
119 "iso646-us",
120 "iso-646.irv-1991",
121 "iso-ir-6",
122 "us",
123 "us-ascii",
124 NULL
125 };
126#endif
127
128 loc = setlocale(LC_CTYPE, NULL);
129 if (loc == NULL)
130 goto error;
131 if (strcmp(loc, "C") != 0) {
132 /* the LC_CTYPE locale is different than C */
133 return 0;
134 }
135
136#if defined(HAVE_LANGINFO_H) && defined(CODESET)
137 codeset = nl_langinfo(CODESET);
138 if (!codeset || codeset[0] == '\0') {
139 /* CODESET is not set or empty */
140 goto error;
141 }
142 if (!_Py_normalize_encoding(codeset, encoding, sizeof(encoding)))
143 goto error;
144
145 is_ascii = 0;
146 for (alias=ascii_aliases; *alias != NULL; alias++) {
147 if (strcmp(encoding, *alias) == 0) {
148 is_ascii = 1;
149 break;
150 }
151 }
152 if (!is_ascii) {
153 /* nl_langinfo(CODESET) is not "ascii" or an alias of ASCII */
154 return 0;
155 }
156
157 for (i=0x80; i<0xff; i++) {
158 unsigned char ch;
159 wchar_t wch;
160 size_t res;
161
162 ch = (unsigned char)i;
163 res = mbstowcs(&wch, (char*)&ch, 1);
164 if (res != (size_t)-1) {
165 /* decoding a non-ASCII character from the locale encoding succeed:
166 the locale encoding is not ASCII, force ASCII */
167 return 1;
168 }
169 }
170 /* None of the bytes in the range 0x80-0xff can be decoded from the locale
171 encoding: the locale encoding is really ASCII */
172 return 0;
173#else
174 /* nl_langinfo(CODESET) is not available: always force ASCII */
175 return 1;
176#endif
177
178error:
179 /* if an error occured, force the ASCII encoding */
180 return 1;
181}
182
183static char*
184encode_ascii_surrogateescape(const wchar_t *text, size_t *error_pos)
185{
186 char *result = NULL, *out;
187 size_t len, i;
188 wchar_t ch;
189
190 if (error_pos != NULL)
191 *error_pos = (size_t)-1;
192
193 len = wcslen(text);
194
195 result = PyMem_Malloc(len + 1); /* +1 for NUL byte */
196 if (result == NULL)
197 return NULL;
198
199 out = result;
200 for (i=0; i<len; i++) {
201 ch = text[i];
202
203 if (ch <= 0x7f) {
204 /* ASCII character */
205 *out++ = (char)ch;
206 }
207 else if (0xdc80 <= ch && ch <= 0xdcff) {
208 /* UTF-8b surrogate */
209 *out++ = (char)(ch - 0xdc00);
210 }
211 else {
212 if (error_pos != NULL)
213 *error_pos = i;
214 PyMem_Free(result);
215 return NULL;
216 }
217 }
218 *out = '\0';
219 return result;
220}
221#endif /* !defined(__APPLE__) && !defined(MS_WINDOWS) */
222
223#if !defined(__APPLE__) && (!defined(MS_WINDOWS) || !defined(HAVE_MBRTOWC))
224static wchar_t*
225decode_ascii_surrogateescape(const char *arg, size_t *size)
226{
227 wchar_t *res;
228 unsigned char *in;
229 wchar_t *out;
Benjamin Petersonf18bf6f2015-01-04 16:03:17 -0600230 size_t argsize = strlen(arg) + 1;
Victor Stinnerd45c7f82012-12-04 01:34:47 +0100231
Benjamin Petersonf18bf6f2015-01-04 16:03:17 -0600232 if (argsize > PY_SSIZE_T_MAX/sizeof(wchar_t))
233 return NULL;
Benjamin Peterson10ecaa22015-01-04 16:05:39 -0600234 res = PyMem_RawMalloc(argsize*sizeof(wchar_t));
Victor Stinnerd45c7f82012-12-04 01:34:47 +0100235 if (!res)
236 return NULL;
237
238 in = (unsigned char*)arg;
239 out = res;
240 while(*in)
241 if(*in < 128)
242 *out++ = *in++;
243 else
244 *out++ = 0xdc00 + *in++;
245 *out = 0;
246 if (size != NULL)
247 *size = out - res;
248 return res;
249}
250#endif
251
Victor Stinner4e314432010-10-07 21:45:39 +0000252
253/* Decode a byte string from the locale encoding with the
Victor Stinnerf6a271a2014-08-01 12:28:48 +0200254 surrogateescape error handler: undecodable bytes are decoded as characters
255 in range U+DC80..U+DCFF. If a byte sequence can be decoded as a surrogate
Victor Stinner4e314432010-10-07 21:45:39 +0000256 character, escape the bytes using the surrogateescape error handler instead
257 of decoding them.
258
Victor Stinnerf6a271a2014-08-01 12:28:48 +0200259 Return a pointer to a newly allocated wide character string, use
260 PyMem_RawFree() to free the memory. If size is not NULL, write the number of
261 wide characters excluding the null character into *size
Victor Stinner4e314432010-10-07 21:45:39 +0000262
Victor Stinnerf6a271a2014-08-01 12:28:48 +0200263 Return NULL on decoding error or memory allocation error. If *size* is not
264 NULL, *size is set to (size_t)-1 on memory error or set to (size_t)-2 on
265 decoding error.
Victor Stinner19de4c32010-11-08 23:30:46 +0000266
Victor Stinnerf6a271a2014-08-01 12:28:48 +0200267 Decoding errors should never happen, unless there is a bug in the C
268 library.
269
270 Use the Py_EncodeLocale() function to encode the character string back to a
271 byte string. */
Victor Stinner4e314432010-10-07 21:45:39 +0000272wchar_t*
Victor Stinnerf6a271a2014-08-01 12:28:48 +0200273Py_DecodeLocale(const char* arg, size_t *size)
Victor Stinner4e314432010-10-07 21:45:39 +0000274{
Victor Stinnere2623772012-11-12 23:04:02 +0100275#ifdef __APPLE__
276 wchar_t *wstr;
277 wstr = _Py_DecodeUTF8_surrogateescape(arg, strlen(arg));
Victor Stinner0d92c4f2012-11-12 23:32:21 +0100278 if (size != NULL) {
279 if (wstr != NULL)
280 *size = wcslen(wstr);
281 else
282 *size = (size_t)-1;
283 }
Victor Stinnere2623772012-11-12 23:04:02 +0100284 return wstr;
285#else
Victor Stinner4e314432010-10-07 21:45:39 +0000286 wchar_t *res;
Victor Stinnerd45c7f82012-12-04 01:34:47 +0100287 size_t argsize;
Victor Stinner4e314432010-10-07 21:45:39 +0000288 size_t count;
Victor Stinner313f10c2013-05-07 23:48:56 +0200289#ifdef HAVE_MBRTOWC
Victor Stinner4e314432010-10-07 21:45:39 +0000290 unsigned char *in;
291 wchar_t *out;
Victor Stinner4e314432010-10-07 21:45:39 +0000292 mbstate_t mbs;
293#endif
Victor Stinnerd45c7f82012-12-04 01:34:47 +0100294
295#ifndef MS_WINDOWS
296 if (force_ascii == -1)
297 force_ascii = check_force_ascii();
298
299 if (force_ascii) {
300 /* force ASCII encoding to workaround mbstowcs() issue */
301 res = decode_ascii_surrogateescape(arg, size);
302 if (res == NULL)
303 goto oom;
304 return res;
305 }
306#endif
307
308#ifdef HAVE_BROKEN_MBSTOWCS
309 /* Some platforms have a broken implementation of
310 * mbstowcs which does not count the characters that
311 * would result from conversion. Use an upper bound.
312 */
313 argsize = strlen(arg);
314#else
315 argsize = mbstowcs(NULL, arg, 0);
316#endif
Victor Stinner4e314432010-10-07 21:45:39 +0000317 if (argsize != (size_t)-1) {
Benjamin Petersonf18bf6f2015-01-04 16:03:17 -0600318 if (argsize == PY_SSIZE_T_MAX)
319 goto oom;
320 argsize += 1;
321 if (argsize > PY_SSIZE_T_MAX/sizeof(wchar_t))
322 goto oom;
Benjamin Peterson10ecaa22015-01-04 16:05:39 -0600323 res = (wchar_t *)PyMem_RawMalloc(argsize*sizeof(wchar_t));
Victor Stinner4e314432010-10-07 21:45:39 +0000324 if (!res)
325 goto oom;
Benjamin Petersonf18bf6f2015-01-04 16:03:17 -0600326 count = mbstowcs(res, arg, argsize);
Victor Stinner4e314432010-10-07 21:45:39 +0000327 if (count != (size_t)-1) {
328 wchar_t *tmp;
329 /* Only use the result if it contains no
330 surrogate characters. */
331 for (tmp = res; *tmp != 0 &&
Victor Stinner76df43d2012-10-30 01:42:39 +0100332 !Py_UNICODE_IS_SURROGATE(*tmp); tmp++)
Victor Stinner4e314432010-10-07 21:45:39 +0000333 ;
Victor Stinner168e1172010-10-16 23:16:16 +0000334 if (*tmp == 0) {
335 if (size != NULL)
336 *size = count;
Victor Stinner4e314432010-10-07 21:45:39 +0000337 return res;
Victor Stinner168e1172010-10-16 23:16:16 +0000338 }
Victor Stinner4e314432010-10-07 21:45:39 +0000339 }
Victor Stinner1a7425f2013-07-07 16:25:15 +0200340 PyMem_RawFree(res);
Victor Stinner4e314432010-10-07 21:45:39 +0000341 }
342 /* Conversion failed. Fall back to escaping with surrogateescape. */
343#ifdef HAVE_MBRTOWC
344 /* Try conversion with mbrtwoc (C99), and escape non-decodable bytes. */
345
346 /* Overallocate; as multi-byte characters are in the argument, the
347 actual output could use less memory. */
348 argsize = strlen(arg) + 1;
Benjamin Petersonf18bf6f2015-01-04 16:03:17 -0600349 if (argsize > PY_SSIZE_T_MAX/sizeof(wchar_t))
350 goto oom;
Victor Stinner1a7425f2013-07-07 16:25:15 +0200351 res = (wchar_t*)PyMem_RawMalloc(argsize*sizeof(wchar_t));
Victor Stinner19de4c32010-11-08 23:30:46 +0000352 if (!res)
353 goto oom;
Victor Stinner4e314432010-10-07 21:45:39 +0000354 in = (unsigned char*)arg;
355 out = res;
356 memset(&mbs, 0, sizeof mbs);
357 while (argsize) {
358 size_t converted = mbrtowc(out, (char*)in, argsize, &mbs);
359 if (converted == 0)
360 /* Reached end of string; null char stored. */
361 break;
362 if (converted == (size_t)-2) {
363 /* Incomplete character. This should never happen,
364 since we provide everything that we have -
365 unless there is a bug in the C library, or I
366 misunderstood how mbrtowc works. */
Victor Stinner1a7425f2013-07-07 16:25:15 +0200367 PyMem_RawFree(res);
Victor Stinneraf02e1c2011-12-16 23:56:01 +0100368 if (size != NULL)
369 *size = (size_t)-2;
Victor Stinner4e314432010-10-07 21:45:39 +0000370 return NULL;
371 }
372 if (converted == (size_t)-1) {
373 /* Conversion error. Escape as UTF-8b, and start over
374 in the initial shift state. */
375 *out++ = 0xdc00 + *in++;
376 argsize--;
377 memset(&mbs, 0, sizeof mbs);
378 continue;
379 }
Victor Stinner76df43d2012-10-30 01:42:39 +0100380 if (Py_UNICODE_IS_SURROGATE(*out)) {
Victor Stinner4e314432010-10-07 21:45:39 +0000381 /* Surrogate character. Escape the original
382 byte sequence with surrogateescape. */
383 argsize -= converted;
384 while (converted--)
385 *out++ = 0xdc00 + *in++;
386 continue;
387 }
388 /* successfully converted some bytes */
389 in += converted;
390 argsize -= converted;
391 out++;
392 }
Victor Stinnerd45c7f82012-12-04 01:34:47 +0100393 if (size != NULL)
394 *size = out - res;
Victor Stinnere2623772012-11-12 23:04:02 +0100395#else /* HAVE_MBRTOWC */
Victor Stinner4e314432010-10-07 21:45:39 +0000396 /* Cannot use C locale for escaping; manually escape as if charset
397 is ASCII (i.e. escape all bytes > 128. This will still roundtrip
398 correctly in the locale's charset, which must be an ASCII superset. */
Victor Stinnerd45c7f82012-12-04 01:34:47 +0100399 res = decode_ascii_surrogateescape(arg, size);
400 if (res == NULL)
Victor Stinneraf02e1c2011-12-16 23:56:01 +0100401 goto oom;
Victor Stinnere2623772012-11-12 23:04:02 +0100402#endif /* HAVE_MBRTOWC */
Victor Stinner4e314432010-10-07 21:45:39 +0000403 return res;
404oom:
Victor Stinneraf02e1c2011-12-16 23:56:01 +0100405 if (size != NULL)
406 *size = (size_t)-1;
Victor Stinner4e314432010-10-07 21:45:39 +0000407 return NULL;
Victor Stinnere2623772012-11-12 23:04:02 +0100408#endif /* __APPLE__ */
Victor Stinner4e314432010-10-07 21:45:39 +0000409}
410
Victor Stinnerf6a271a2014-08-01 12:28:48 +0200411/* Encode a wide character string to the locale encoding with the
412 surrogateescape error handler: surrogate characters in the range
413 U+DC80..U+DCFF are converted to bytes 0x80..0xFF.
Victor Stinner4e314432010-10-07 21:45:39 +0000414
Victor Stinnerf6a271a2014-08-01 12:28:48 +0200415 Return a pointer to a newly allocated byte string, use PyMem_Free() to free
416 the memory. Return NULL on encoding or memory allocation error.
Victor Stinner4e314432010-10-07 21:45:39 +0000417
Victor Stinnerf6a271a2014-08-01 12:28:48 +0200418 If error_pos is not NULL, *error_pos is set to the index of the invalid
419 character on encoding error, or set to (size_t)-1 otherwise.
Victor Stinner2f02a512010-11-08 22:43:46 +0000420
Victor Stinnerf6a271a2014-08-01 12:28:48 +0200421 Use the Py_DecodeLocale() function to decode the bytes string back to a wide
422 character string. */
Victor Stinner4e314432010-10-07 21:45:39 +0000423char*
Victor Stinnerf6a271a2014-08-01 12:28:48 +0200424Py_EncodeLocale(const wchar_t *text, size_t *error_pos)
Victor Stinner4e314432010-10-07 21:45:39 +0000425{
Victor Stinnere2623772012-11-12 23:04:02 +0100426#ifdef __APPLE__
427 Py_ssize_t len;
428 PyObject *unicode, *bytes = NULL;
429 char *cpath;
430
431 unicode = PyUnicode_FromWideChar(text, wcslen(text));
Victor Stinner0d92c4f2012-11-12 23:32:21 +0100432 if (unicode == NULL)
Victor Stinnere2623772012-11-12 23:04:02 +0100433 return NULL;
Victor Stinnere2623772012-11-12 23:04:02 +0100434
435 bytes = _PyUnicode_AsUTF8String(unicode, "surrogateescape");
436 Py_DECREF(unicode);
437 if (bytes == NULL) {
438 PyErr_Clear();
Victor Stinner0d92c4f2012-11-12 23:32:21 +0100439 if (error_pos != NULL)
440 *error_pos = (size_t)-1;
Victor Stinnere2623772012-11-12 23:04:02 +0100441 return NULL;
442 }
443
444 len = PyBytes_GET_SIZE(bytes);
445 cpath = PyMem_Malloc(len+1);
446 if (cpath == NULL) {
Victor Stinner0d92c4f2012-11-12 23:32:21 +0100447 PyErr_Clear();
Victor Stinnere2623772012-11-12 23:04:02 +0100448 Py_DECREF(bytes);
Victor Stinner0d92c4f2012-11-12 23:32:21 +0100449 if (error_pos != NULL)
450 *error_pos = (size_t)-1;
Victor Stinnere2623772012-11-12 23:04:02 +0100451 return NULL;
452 }
453 memcpy(cpath, PyBytes_AsString(bytes), len + 1);
454 Py_DECREF(bytes);
455 return cpath;
456#else /* __APPLE__ */
Victor Stinner4e314432010-10-07 21:45:39 +0000457 const size_t len = wcslen(text);
458 char *result = NULL, *bytes = NULL;
459 size_t i, size, converted;
460 wchar_t c, buf[2];
461
Victor Stinnerd45c7f82012-12-04 01:34:47 +0100462#ifndef MS_WINDOWS
463 if (force_ascii == -1)
464 force_ascii = check_force_ascii();
465
466 if (force_ascii)
467 return encode_ascii_surrogateescape(text, error_pos);
468#endif
469
Victor Stinner4e314432010-10-07 21:45:39 +0000470 /* The function works in two steps:
471 1. compute the length of the output buffer in bytes (size)
472 2. outputs the bytes */
473 size = 0;
474 buf[1] = 0;
475 while (1) {
476 for (i=0; i < len; i++) {
477 c = text[i];
478 if (c >= 0xdc80 && c <= 0xdcff) {
479 /* UTF-8b surrogate */
480 if (bytes != NULL) {
481 *bytes++ = c - 0xdc00;
482 size--;
483 }
484 else
485 size++;
486 continue;
487 }
488 else {
489 buf[0] = c;
490 if (bytes != NULL)
491 converted = wcstombs(bytes, buf, size);
492 else
493 converted = wcstombs(NULL, buf, 0);
494 if (converted == (size_t)-1) {
495 if (result != NULL)
496 PyMem_Free(result);
Victor Stinner2f02a512010-11-08 22:43:46 +0000497 if (error_pos != NULL)
498 *error_pos = i;
Victor Stinner4e314432010-10-07 21:45:39 +0000499 return NULL;
500 }
501 if (bytes != NULL) {
502 bytes += converted;
503 size -= converted;
504 }
505 else
506 size += converted;
507 }
508 }
509 if (result != NULL) {
Victor Stinnerd45c7f82012-12-04 01:34:47 +0100510 *bytes = '\0';
Victor Stinner4e314432010-10-07 21:45:39 +0000511 break;
512 }
513
514 size += 1; /* nul byte at the end */
515 result = PyMem_Malloc(size);
Victor Stinner0d92c4f2012-11-12 23:32:21 +0100516 if (result == NULL) {
517 if (error_pos != NULL)
518 *error_pos = (size_t)-1;
Victor Stinner4e314432010-10-07 21:45:39 +0000519 return NULL;
Victor Stinner0d92c4f2012-11-12 23:32:21 +0100520 }
Victor Stinner4e314432010-10-07 21:45:39 +0000521 bytes = result;
522 }
523 return result;
Victor Stinnere2623772012-11-12 23:04:02 +0100524#endif /* __APPLE__ */
Victor Stinner4e314432010-10-07 21:45:39 +0000525}
526
Victor Stinner6672d0c2010-10-07 22:53:43 +0000527
Steve Dowerf2f373f2015-02-21 08:44:05 -0800528#ifdef MS_WINDOWS
529static __int64 secs_between_epochs = 11644473600; /* Seconds between 1.1.1601 and 1.1.1970 */
530
531static void
532FILE_TIME_to_time_t_nsec(FILETIME *in_ptr, time_t *time_out, int* nsec_out)
533{
534 /* XXX endianness. Shouldn't matter, as all Windows implementations are little-endian */
535 /* Cannot simply cast and dereference in_ptr,
536 since it might not be aligned properly */
537 __int64 in;
538 memcpy(&in, in_ptr, sizeof(in));
539 *nsec_out = (int)(in % 10000000) * 100; /* FILETIME is in units of 100 nsec. */
540 *time_out = Py_SAFE_DOWNCAST((in / 10000000) - secs_between_epochs, __int64, time_t);
541}
542
543void
Steve Dowerbf1f3762015-02-21 15:26:02 -0800544_Py_time_t_to_FILE_TIME(time_t time_in, int nsec_in, FILETIME *out_ptr)
Steve Dowerf2f373f2015-02-21 08:44:05 -0800545{
546 /* XXX endianness */
547 __int64 out;
548 out = time_in + secs_between_epochs;
549 out = out * 10000000 + nsec_in / 100;
550 memcpy(out_ptr, &out, sizeof(out));
551}
552
553/* Below, we *know* that ugo+r is 0444 */
554#if _S_IREAD != 0400
555#error Unsupported C library
556#endif
557static int
558attributes_to_mode(DWORD attr)
559{
560 int m = 0;
561 if (attr & FILE_ATTRIBUTE_DIRECTORY)
562 m |= _S_IFDIR | 0111; /* IFEXEC for user,group,other */
563 else
564 m |= _S_IFREG;
565 if (attr & FILE_ATTRIBUTE_READONLY)
566 m |= 0444;
567 else
568 m |= 0666;
569 return m;
570}
571
Steve Dowerbf1f3762015-02-21 15:26:02 -0800572void
Victor Stinnere134a7f2015-03-30 10:09:31 +0200573_Py_attribute_data_to_stat(BY_HANDLE_FILE_INFORMATION *info, ULONG reparse_tag,
574 struct _Py_stat_struct *result)
Steve Dowerf2f373f2015-02-21 08:44:05 -0800575{
576 memset(result, 0, sizeof(*result));
577 result->st_mode = attributes_to_mode(info->dwFileAttributes);
578 result->st_size = (((__int64)info->nFileSizeHigh)<<32) + info->nFileSizeLow;
579 result->st_dev = info->dwVolumeSerialNumber;
580 result->st_rdev = result->st_dev;
581 FILE_TIME_to_time_t_nsec(&info->ftCreationTime, &result->st_ctime, &result->st_ctime_nsec);
582 FILE_TIME_to_time_t_nsec(&info->ftLastWriteTime, &result->st_mtime, &result->st_mtime_nsec);
583 FILE_TIME_to_time_t_nsec(&info->ftLastAccessTime, &result->st_atime, &result->st_atime_nsec);
584 result->st_nlink = info->nNumberOfLinks;
585 result->st_ino = (((__int64)info->nFileIndexHigh)<<32) + info->nFileIndexLow;
586 if (reparse_tag == IO_REPARSE_TAG_SYMLINK) {
587 /* first clear the S_IFMT bits */
588 result->st_mode ^= (result->st_mode & S_IFMT);
589 /* now set the bits that make this a symlink */
590 result->st_mode |= S_IFLNK;
591 }
592 result->st_file_attributes = info->dwFileAttributes;
Steve Dowerf2f373f2015-02-21 08:44:05 -0800593}
594#endif
595
596/* Return information about a file.
597
598 On POSIX, use fstat().
599
600 On Windows, use GetFileType() and GetFileInformationByHandle() which support
601 files larger than 2 GB. fstat() may fail with EOVERFLOW on files larger
602 than 2 GB because the file size type is an signed 32-bit integer: see issue
603 #23152.
Victor Stinnere134a7f2015-03-30 10:09:31 +0200604
605 On Windows, set the last Windows error and return nonzero on error. On
606 POSIX, set errno and return nonzero on error. Fill status and return 0 on
607 success. */
Steve Dowerf2f373f2015-02-21 08:44:05 -0800608int
Victor Stinnere134a7f2015-03-30 10:09:31 +0200609_Py_fstat_noraise(int fd, struct _Py_stat_struct *status)
Steve Dowerf2f373f2015-02-21 08:44:05 -0800610{
611#ifdef MS_WINDOWS
612 BY_HANDLE_FILE_INFORMATION info;
613 HANDLE h;
614 int type;
615
616 if (!_PyVerify_fd(fd))
617 h = INVALID_HANDLE_VALUE;
Steve Dower8fc89802015-04-12 00:26:27 -0400618 else {
619 _Py_BEGIN_SUPPRESS_IPH
Steve Dowerf2f373f2015-02-21 08:44:05 -0800620 h = (HANDLE)_get_osfhandle(fd);
Steve Dower8fc89802015-04-12 00:26:27 -0400621 _Py_END_SUPPRESS_IPH
622 }
Steve Dowerf2f373f2015-02-21 08:44:05 -0800623
624 if (h == INVALID_HANDLE_VALUE) {
Steve Dower8fc89802015-04-12 00:26:27 -0400625 /* errno is already set by _get_osfhandle, but we also set
626 the Win32 error for callers who expect that */
Steve Dower8acde7d2015-03-07 18:14:07 -0800627 SetLastError(ERROR_INVALID_HANDLE);
Steve Dowerf2f373f2015-02-21 08:44:05 -0800628 return -1;
629 }
Victor Stinnere134a7f2015-03-30 10:09:31 +0200630 memset(status, 0, sizeof(*status));
Steve Dowerf2f373f2015-02-21 08:44:05 -0800631
632 type = GetFileType(h);
633 if (type == FILE_TYPE_UNKNOWN) {
634 DWORD error = GetLastError();
Steve Dower8fc89802015-04-12 00:26:27 -0400635 if (error != 0) {
636 errno = winerror_to_errno(error);
Steve Dowerf2f373f2015-02-21 08:44:05 -0800637 return -1;
Steve Dower8fc89802015-04-12 00:26:27 -0400638 }
Steve Dowerf2f373f2015-02-21 08:44:05 -0800639 /* else: valid but unknown file */
640 }
641
642 if (type != FILE_TYPE_DISK) {
643 if (type == FILE_TYPE_CHAR)
Victor Stinnere134a7f2015-03-30 10:09:31 +0200644 status->st_mode = _S_IFCHR;
Steve Dowerf2f373f2015-02-21 08:44:05 -0800645 else if (type == FILE_TYPE_PIPE)
Victor Stinnere134a7f2015-03-30 10:09:31 +0200646 status->st_mode = _S_IFIFO;
Steve Dowerf2f373f2015-02-21 08:44:05 -0800647 return 0;
648 }
649
650 if (!GetFileInformationByHandle(h, &info)) {
Steve Dower8fc89802015-04-12 00:26:27 -0400651 /* The Win32 error is already set, but we also set errno for
652 callers who expect it */
653 errno = winerror_to_errno(GetLastError());
Steve Dowerf2f373f2015-02-21 08:44:05 -0800654 return -1;
655 }
656
Victor Stinnere134a7f2015-03-30 10:09:31 +0200657 _Py_attribute_data_to_stat(&info, 0, status);
Steve Dowerf2f373f2015-02-21 08:44:05 -0800658 /* specific to fstat() */
Victor Stinnere134a7f2015-03-30 10:09:31 +0200659 status->st_ino = (((__int64)info.nFileIndexHigh)<<32) + info.nFileIndexLow;
Steve Dowerf2f373f2015-02-21 08:44:05 -0800660 return 0;
661#else
Victor Stinnere134a7f2015-03-30 10:09:31 +0200662 return fstat(fd, status);
Steve Dowerf2f373f2015-02-21 08:44:05 -0800663#endif
664}
Steve Dowerf2f373f2015-02-21 08:44:05 -0800665
Victor Stinnere134a7f2015-03-30 10:09:31 +0200666/* Return information about a file.
667
668 On POSIX, use fstat().
669
670 On Windows, use GetFileType() and GetFileInformationByHandle() which support
671 files larger than 2 GB. fstat() may fail with EOVERFLOW on files larger
672 than 2 GB because the file size type is an signed 32-bit integer: see issue
673 #23152.
674
675 Raise an exception and return -1 on error. On Windows, set the last Windows
676 error on error. On POSIX, set errno on error. Fill status and return 0 on
677 success.
678
Victor Stinner6f4fae82015-04-01 18:34:32 +0200679 Release the GIL to call GetFileType() and GetFileInformationByHandle(), or
680 to call fstat(). The caller must hold the GIL. */
Victor Stinnere134a7f2015-03-30 10:09:31 +0200681int
682_Py_fstat(int fd, struct _Py_stat_struct *status)
683{
684 int res;
685
686 Py_BEGIN_ALLOW_THREADS
687 res = _Py_fstat_noraise(fd, status);
688 Py_END_ALLOW_THREADS
689
690 if (res != 0) {
691#ifdef MS_WINDOWS
692 PyErr_SetFromWindowsErr(0);
693#else
694 PyErr_SetFromErrno(PyExc_OSError);
695#endif
696 return -1;
697 }
698 return 0;
699}
Steve Dowerf2f373f2015-02-21 08:44:05 -0800700
Victor Stinner6672d0c2010-10-07 22:53:43 +0000701/* Call _wstat() on Windows, or encode the path to the filesystem encoding and
702 call stat() otherwise. Only fill st_mode attribute on Windows.
703
Victor Stinnerbd0850b2011-12-18 20:47:30 +0100704 Return 0 on success, -1 on _wstat() / stat() error, -2 if an exception was
705 raised. */
Victor Stinner4e314432010-10-07 21:45:39 +0000706
707int
Victor Stinnera4a75952010-10-07 22:23:10 +0000708_Py_stat(PyObject *path, struct stat *statbuf)
Victor Stinner4e314432010-10-07 21:45:39 +0000709{
710#ifdef MS_WINDOWS
Victor Stinner4e314432010-10-07 21:45:39 +0000711 int err;
712 struct _stat wstatbuf;
Victor Stinneree587ea2011-11-17 00:51:38 +0100713 wchar_t *wpath;
Victor Stinner4e314432010-10-07 21:45:39 +0000714
Victor Stinneree587ea2011-11-17 00:51:38 +0100715 wpath = PyUnicode_AsUnicode(path);
716 if (wpath == NULL)
Victor Stinnerbd0850b2011-12-18 20:47:30 +0100717 return -2;
Victor Stinneree587ea2011-11-17 00:51:38 +0100718 err = _wstat(wpath, &wstatbuf);
Victor Stinner4e314432010-10-07 21:45:39 +0000719 if (!err)
720 statbuf->st_mode = wstatbuf.st_mode;
721 return err;
722#else
723 int ret;
Victor Stinnera4a75952010-10-07 22:23:10 +0000724 PyObject *bytes = PyUnicode_EncodeFSDefault(path);
Victor Stinner4e314432010-10-07 21:45:39 +0000725 if (bytes == NULL)
Victor Stinnerbd0850b2011-12-18 20:47:30 +0100726 return -2;
Victor Stinner4e314432010-10-07 21:45:39 +0000727 ret = stat(PyBytes_AS_STRING(bytes), statbuf);
728 Py_DECREF(bytes);
729 return ret;
730#endif
731}
732
Victor Stinnerd45c7f82012-12-04 01:34:47 +0100733
Antoine Pitrou409b5382013-10-12 22:41:17 +0200734static int
Victor Stinnerdaf45552013-08-28 00:53:59 +0200735get_inheritable(int fd, int raise)
736{
737#ifdef MS_WINDOWS
738 HANDLE handle;
739 DWORD flags;
Victor Stinner6672d0c2010-10-07 22:53:43 +0000740
Victor Stinnerdaf45552013-08-28 00:53:59 +0200741 if (!_PyVerify_fd(fd)) {
742 if (raise)
743 PyErr_SetFromErrno(PyExc_OSError);
744 return -1;
745 }
746
Steve Dower8fc89802015-04-12 00:26:27 -0400747 _Py_BEGIN_SUPPRESS_IPH
Victor Stinnerdaf45552013-08-28 00:53:59 +0200748 handle = (HANDLE)_get_osfhandle(fd);
Steve Dower8fc89802015-04-12 00:26:27 -0400749 _Py_END_SUPPRESS_IPH
Victor Stinnerdaf45552013-08-28 00:53:59 +0200750 if (handle == INVALID_HANDLE_VALUE) {
751 if (raise)
Steve Dower41e72442015-03-14 11:38:27 -0700752 PyErr_SetFromErrno(PyExc_OSError);
Victor Stinnerdaf45552013-08-28 00:53:59 +0200753 return -1;
754 }
755
756 if (!GetHandleInformation(handle, &flags)) {
757 if (raise)
758 PyErr_SetFromWindowsErr(0);
759 return -1;
760 }
761
762 return (flags & HANDLE_FLAG_INHERIT);
763#else
764 int flags;
765
766 flags = fcntl(fd, F_GETFD, 0);
767 if (flags == -1) {
768 if (raise)
769 PyErr_SetFromErrno(PyExc_OSError);
770 return -1;
771 }
772 return !(flags & FD_CLOEXEC);
773#endif
774}
775
776/* Get the inheritable flag of the specified file descriptor.
Victor Stinnerb034eee2013-09-07 10:36:04 +0200777 Return 1 if the file descriptor can be inherited, 0 if it cannot,
Victor Stinnerdaf45552013-08-28 00:53:59 +0200778 raise an exception and return -1 on error. */
779int
780_Py_get_inheritable(int fd)
781{
782 return get_inheritable(fd, 1);
783}
784
785static int
786set_inheritable(int fd, int inheritable, int raise, int *atomic_flag_works)
787{
788#ifdef MS_WINDOWS
789 HANDLE handle;
790 DWORD flags;
Victor Stinner282124b2014-09-02 11:41:04 +0200791#else
792#if defined(HAVE_SYS_IOCTL_H) && defined(FIOCLEX) && defined(FIONCLEX)
793 static int ioctl_works = -1;
Victor Stinnerdaf45552013-08-28 00:53:59 +0200794 int request;
795 int err;
Victor Stinner282124b2014-09-02 11:41:04 +0200796#endif
Victor Stinnerdaf45552013-08-28 00:53:59 +0200797 int flags;
798 int res;
799#endif
800
801 /* atomic_flag_works can only be used to make the file descriptor
802 non-inheritable */
803 assert(!(atomic_flag_works != NULL && inheritable));
804
805 if (atomic_flag_works != NULL && !inheritable) {
806 if (*atomic_flag_works == -1) {
Steve Dower41e72442015-03-14 11:38:27 -0700807 int isInheritable = get_inheritable(fd, raise);
808 if (isInheritable == -1)
Victor Stinnerdaf45552013-08-28 00:53:59 +0200809 return -1;
Steve Dower41e72442015-03-14 11:38:27 -0700810 *atomic_flag_works = !isInheritable;
Victor Stinnerdaf45552013-08-28 00:53:59 +0200811 }
812
813 if (*atomic_flag_works)
814 return 0;
815 }
816
817#ifdef MS_WINDOWS
818 if (!_PyVerify_fd(fd)) {
819 if (raise)
820 PyErr_SetFromErrno(PyExc_OSError);
821 return -1;
822 }
823
Steve Dower8fc89802015-04-12 00:26:27 -0400824 _Py_BEGIN_SUPPRESS_IPH
Victor Stinnerdaf45552013-08-28 00:53:59 +0200825 handle = (HANDLE)_get_osfhandle(fd);
Steve Dower8fc89802015-04-12 00:26:27 -0400826 _Py_END_SUPPRESS_IPH
Victor Stinnerdaf45552013-08-28 00:53:59 +0200827 if (handle == INVALID_HANDLE_VALUE) {
828 if (raise)
Steve Dower41e72442015-03-14 11:38:27 -0700829 PyErr_SetFromErrno(PyExc_OSError);
Victor Stinnerdaf45552013-08-28 00:53:59 +0200830 return -1;
831 }
832
833 if (inheritable)
834 flags = HANDLE_FLAG_INHERIT;
835 else
836 flags = 0;
837 if (!SetHandleInformation(handle, HANDLE_FLAG_INHERIT, flags)) {
838 if (raise)
839 PyErr_SetFromWindowsErr(0);
840 return -1;
841 }
842 return 0;
843
Victor Stinnerdaf45552013-08-28 00:53:59 +0200844#else
Victor Stinner282124b2014-09-02 11:41:04 +0200845
846#if defined(HAVE_SYS_IOCTL_H) && defined(FIOCLEX) && defined(FIONCLEX)
847 if (ioctl_works != 0) {
848 /* fast-path: ioctl() only requires one syscall */
849 if (inheritable)
850 request = FIONCLEX;
851 else
852 request = FIOCLEX;
853 err = ioctl(fd, request, NULL);
854 if (!err) {
855 ioctl_works = 1;
856 return 0;
857 }
858
859 if (errno != ENOTTY) {
860 if (raise)
861 PyErr_SetFromErrno(PyExc_OSError);
862 return -1;
863 }
864 else {
865 /* Issue #22258: Here, ENOTTY means "Inappropriate ioctl for
866 device". The ioctl is declared but not supported by the kernel.
867 Remember that ioctl() doesn't work. It is the case on
868 Illumos-based OS for example. */
869 ioctl_works = 0;
870 }
871 /* fallback to fcntl() if ioctl() does not work */
872 }
873#endif
874
875 /* slow-path: fcntl() requires two syscalls */
Victor Stinnerdaf45552013-08-28 00:53:59 +0200876 flags = fcntl(fd, F_GETFD);
877 if (flags < 0) {
878 if (raise)
879 PyErr_SetFromErrno(PyExc_OSError);
880 return -1;
881 }
882
883 if (inheritable)
884 flags &= ~FD_CLOEXEC;
885 else
886 flags |= FD_CLOEXEC;
887 res = fcntl(fd, F_SETFD, flags);
888 if (res < 0) {
889 if (raise)
890 PyErr_SetFromErrno(PyExc_OSError);
891 return -1;
892 }
893 return 0;
894#endif
895}
896
897/* Make the file descriptor non-inheritable.
Victor Stinnerb034eee2013-09-07 10:36:04 +0200898 Return 0 on success, set errno and return -1 on error. */
Victor Stinnerdaf45552013-08-28 00:53:59 +0200899static int
900make_non_inheritable(int fd)
901{
902 return set_inheritable(fd, 0, 0, NULL);
903}
904
905/* Set the inheritable flag of the specified file descriptor.
906 On success: return 0, on error: raise an exception if raise is nonzero
907 and return -1.
908
909 If atomic_flag_works is not NULL:
910
911 * if *atomic_flag_works==-1, check if the inheritable is set on the file
912 descriptor: if yes, set *atomic_flag_works to 1, otherwise set to 0 and
913 set the inheritable flag
914 * if *atomic_flag_works==1: do nothing
915 * if *atomic_flag_works==0: set inheritable flag to False
916
917 Set atomic_flag_works to NULL if no atomic flag was used to create the
918 file descriptor.
919
920 atomic_flag_works can only be used to make a file descriptor
921 non-inheritable: atomic_flag_works must be NULL if inheritable=1. */
922int
923_Py_set_inheritable(int fd, int inheritable, int *atomic_flag_works)
924{
925 return set_inheritable(fd, inheritable, 1, atomic_flag_works);
926}
927
Victor Stinnera555cfc2015-03-18 00:22:14 +0100928static int
929_Py_open_impl(const char *pathname, int flags, int gil_held)
Victor Stinnerdaf45552013-08-28 00:53:59 +0200930{
931 int fd;
Victor Stinnera47fc5c2015-03-18 09:52:54 +0100932 int async_err = 0;
Victor Stinnera555cfc2015-03-18 00:22:14 +0100933#ifndef MS_WINDOWS
Victor Stinnerdaf45552013-08-28 00:53:59 +0200934 int *atomic_flag_works;
Victor Stinnera555cfc2015-03-18 00:22:14 +0100935#endif
936
937#ifdef MS_WINDOWS
938 flags |= O_NOINHERIT;
939#elif defined(O_CLOEXEC)
Victor Stinnerdaf45552013-08-28 00:53:59 +0200940 atomic_flag_works = &_Py_open_cloexec_works;
941 flags |= O_CLOEXEC;
942#else
943 atomic_flag_works = NULL;
944#endif
Victor Stinnerdaf45552013-08-28 00:53:59 +0200945
Victor Stinnera555cfc2015-03-18 00:22:14 +0100946 if (gil_held) {
Victor Stinnera47fc5c2015-03-18 09:52:54 +0100947 do {
948 Py_BEGIN_ALLOW_THREADS
949 fd = open(pathname, flags);
950 Py_END_ALLOW_THREADS
951 } while (fd < 0
952 && errno == EINTR && !(async_err = PyErr_CheckSignals()));
953 if (async_err)
954 return -1;
Victor Stinnera555cfc2015-03-18 00:22:14 +0100955 if (fd < 0) {
956 PyErr_SetFromErrnoWithFilename(PyExc_OSError, pathname);
957 return -1;
958 }
959 }
960 else {
961 fd = open(pathname, flags);
962 if (fd < 0)
963 return -1;
964 }
965
966#ifndef MS_WINDOWS
967 if (set_inheritable(fd, 0, gil_held, atomic_flag_works) < 0) {
Victor Stinnerdaf45552013-08-28 00:53:59 +0200968 close(fd);
969 return -1;
970 }
Victor Stinnera555cfc2015-03-18 00:22:14 +0100971#endif
972
Victor Stinnerdaf45552013-08-28 00:53:59 +0200973 return fd;
974}
975
Victor Stinnera555cfc2015-03-18 00:22:14 +0100976/* Open a file with the specified flags (wrapper to open() function).
977 Return a file descriptor on success. Raise an exception and return -1 on
978 error.
979
980 The file descriptor is created non-inheritable.
981
Victor Stinnera47fc5c2015-03-18 09:52:54 +0100982 When interrupted by a signal (open() fails with EINTR), retry the syscall,
983 except if the Python signal handler raises an exception.
984
Victor Stinner6f4fae82015-04-01 18:34:32 +0200985 Release the GIL to call open(). The caller must hold the GIL. */
Victor Stinnera555cfc2015-03-18 00:22:14 +0100986int
987_Py_open(const char *pathname, int flags)
988{
989 /* _Py_open() must be called with the GIL held. */
990 assert(PyGILState_Check());
991 return _Py_open_impl(pathname, flags, 1);
992}
993
994/* Open a file with the specified flags (wrapper to open() function).
995 Return a file descriptor on success. Set errno and return -1 on error.
996
Victor Stinnera47fc5c2015-03-18 09:52:54 +0100997 The file descriptor is created non-inheritable.
998
999 If interrupted by a signal, fail with EINTR. */
Victor Stinnera555cfc2015-03-18 00:22:14 +01001000int
1001_Py_open_noraise(const char *pathname, int flags)
1002{
1003 return _Py_open_impl(pathname, flags, 0);
1004}
1005
Victor Stinnerdaf45552013-08-28 00:53:59 +02001006/* Open a file. Use _wfopen() on Windows, encode the path to the locale
Victor Stinnere42ccd22015-03-18 01:39:23 +01001007 encoding and use fopen() otherwise.
1008
Victor Stinnera47fc5c2015-03-18 09:52:54 +01001009 The file descriptor is created non-inheritable.
1010
1011 If interrupted by a signal, fail with EINTR. */
Victor Stinner4e314432010-10-07 21:45:39 +00001012FILE *
1013_Py_wfopen(const wchar_t *path, const wchar_t *mode)
1014{
Victor Stinner4e314432010-10-07 21:45:39 +00001015 FILE *f;
Victor Stinnerdaf45552013-08-28 00:53:59 +02001016#ifndef MS_WINDOWS
Victor Stinner4e314432010-10-07 21:45:39 +00001017 char *cpath;
1018 char cmode[10];
1019 size_t r;
1020 r = wcstombs(cmode, mode, 10);
1021 if (r == (size_t)-1 || r >= 10) {
1022 errno = EINVAL;
1023 return NULL;
1024 }
Victor Stinnerf6a271a2014-08-01 12:28:48 +02001025 cpath = Py_EncodeLocale(path, NULL);
Victor Stinner4e314432010-10-07 21:45:39 +00001026 if (cpath == NULL)
1027 return NULL;
1028 f = fopen(cpath, cmode);
1029 PyMem_Free(cpath);
Victor Stinner4e314432010-10-07 21:45:39 +00001030#else
Victor Stinnerdaf45552013-08-28 00:53:59 +02001031 f = _wfopen(path, mode);
Victor Stinner4e314432010-10-07 21:45:39 +00001032#endif
Victor Stinnerdaf45552013-08-28 00:53:59 +02001033 if (f == NULL)
1034 return NULL;
1035 if (make_non_inheritable(fileno(f)) < 0) {
1036 fclose(f);
1037 return NULL;
1038 }
1039 return f;
Victor Stinner4e314432010-10-07 21:45:39 +00001040}
1041
Victor Stinnere42ccd22015-03-18 01:39:23 +01001042/* Wrapper to fopen().
1043
Victor Stinnera47fc5c2015-03-18 09:52:54 +01001044 The file descriptor is created non-inheritable.
1045
1046 If interrupted by a signal, fail with EINTR. */
Victor Stinnerdaf45552013-08-28 00:53:59 +02001047FILE*
1048_Py_fopen(const char *pathname, const char *mode)
1049{
1050 FILE *f = fopen(pathname, mode);
1051 if (f == NULL)
1052 return NULL;
1053 if (make_non_inheritable(fileno(f)) < 0) {
1054 fclose(f);
1055 return NULL;
1056 }
1057 return f;
1058}
1059
1060/* Open a file. Call _wfopen() on Windows, or encode the path to the filesystem
Victor Stinnere42ccd22015-03-18 01:39:23 +01001061 encoding and call fopen() otherwise.
Victor Stinner6672d0c2010-10-07 22:53:43 +00001062
Victor Stinnere42ccd22015-03-18 01:39:23 +01001063 Return the new file object on success. Raise an exception and return NULL
1064 on error.
1065
1066 The file descriptor is created non-inheritable.
1067
Victor Stinnera47fc5c2015-03-18 09:52:54 +01001068 When interrupted by a signal (open() fails with EINTR), retry the syscall,
1069 except if the Python signal handler raises an exception.
1070
Victor Stinner6f4fae82015-04-01 18:34:32 +02001071 Release the GIL to call _wfopen() or fopen(). The caller must hold
1072 the GIL. */
Victor Stinner4e314432010-10-07 21:45:39 +00001073FILE*
Victor Stinnerdaf45552013-08-28 00:53:59 +02001074_Py_fopen_obj(PyObject *path, const char *mode)
Victor Stinner4e314432010-10-07 21:45:39 +00001075{
Victor Stinnerdaf45552013-08-28 00:53:59 +02001076 FILE *f;
Victor Stinnera47fc5c2015-03-18 09:52:54 +01001077 int async_err = 0;
Victor Stinner4e314432010-10-07 21:45:39 +00001078#ifdef MS_WINDOWS
Victor Stinneree587ea2011-11-17 00:51:38 +01001079 wchar_t *wpath;
Victor Stinner4e314432010-10-07 21:45:39 +00001080 wchar_t wmode[10];
1081 int usize;
Victor Stinner4e314432010-10-07 21:45:39 +00001082
Victor Stinnere42ccd22015-03-18 01:39:23 +01001083 assert(PyGILState_Check());
1084
Antoine Pitrou0e576f12011-12-22 10:03:38 +01001085 if (!PyUnicode_Check(path)) {
1086 PyErr_Format(PyExc_TypeError,
1087 "str file path expected under Windows, got %R",
1088 Py_TYPE(path));
1089 return NULL;
1090 }
Victor Stinneree587ea2011-11-17 00:51:38 +01001091 wpath = PyUnicode_AsUnicode(path);
1092 if (wpath == NULL)
1093 return NULL;
1094
Victor Stinner4e314432010-10-07 21:45:39 +00001095 usize = MultiByteToWideChar(CP_ACP, 0, mode, -1, wmode, sizeof(wmode));
Victor Stinnere42ccd22015-03-18 01:39:23 +01001096 if (usize == 0) {
1097 PyErr_SetFromWindowsErr(0);
Victor Stinner4e314432010-10-07 21:45:39 +00001098 return NULL;
Victor Stinnere42ccd22015-03-18 01:39:23 +01001099 }
Victor Stinner4e314432010-10-07 21:45:39 +00001100
Victor Stinnera47fc5c2015-03-18 09:52:54 +01001101 do {
1102 Py_BEGIN_ALLOW_THREADS
1103 f = _wfopen(wpath, wmode);
1104 Py_END_ALLOW_THREADS
1105 } while (f == NULL
1106 && errno == EINTR && !(async_err = PyErr_CheckSignals()));
Victor Stinner4e314432010-10-07 21:45:39 +00001107#else
Antoine Pitrou2b1cc892011-12-19 18:19:06 +01001108 PyObject *bytes;
Victor Stinnere42ccd22015-03-18 01:39:23 +01001109 char *path_bytes;
1110
1111 assert(PyGILState_Check());
1112
Antoine Pitrou2b1cc892011-12-19 18:19:06 +01001113 if (!PyUnicode_FSConverter(path, &bytes))
Victor Stinner4e314432010-10-07 21:45:39 +00001114 return NULL;
Victor Stinnere42ccd22015-03-18 01:39:23 +01001115 path_bytes = PyBytes_AS_STRING(bytes);
1116
Victor Stinnera47fc5c2015-03-18 09:52:54 +01001117 do {
1118 Py_BEGIN_ALLOW_THREADS
1119 f = fopen(path_bytes, mode);
1120 Py_END_ALLOW_THREADS
1121 } while (f == NULL
1122 && errno == EINTR && !(async_err = PyErr_CheckSignals()));
Victor Stinnere42ccd22015-03-18 01:39:23 +01001123
Victor Stinner4e314432010-10-07 21:45:39 +00001124 Py_DECREF(bytes);
Victor Stinner4e314432010-10-07 21:45:39 +00001125#endif
Victor Stinnera47fc5c2015-03-18 09:52:54 +01001126 if (async_err)
1127 return NULL;
1128
Victor Stinnere42ccd22015-03-18 01:39:23 +01001129 if (f == NULL) {
1130 PyErr_SetFromErrnoWithFilenameObject(PyExc_OSError, path);
Victor Stinnerdaf45552013-08-28 00:53:59 +02001131 return NULL;
Victor Stinnere42ccd22015-03-18 01:39:23 +01001132 }
1133
1134 if (set_inheritable(fileno(f), 0, 1, NULL) < 0) {
Victor Stinnerdaf45552013-08-28 00:53:59 +02001135 fclose(f);
1136 return NULL;
1137 }
1138 return f;
Victor Stinner4e314432010-10-07 21:45:39 +00001139}
1140
Victor Stinner66aab0c2015-03-19 22:53:20 +01001141/* Read count bytes from fd into buf.
Victor Stinner82c3e452015-04-01 18:34:45 +02001142
1143 On success, return the number of read bytes, it can be lower than count.
1144 If the current file offset is at or past the end of file, no bytes are read,
1145 and read() returns zero.
1146
1147 On error, raise an exception, set errno and return -1.
1148
1149 When interrupted by a signal (read() fails with EINTR), retry the syscall.
1150 If the Python signal handler raises an exception, the function returns -1
1151 (the syscall is not retried).
1152
1153 Release the GIL to call read(). The caller must hold the GIL. */
Victor Stinner66aab0c2015-03-19 22:53:20 +01001154Py_ssize_t
1155_Py_read(int fd, void *buf, size_t count)
1156{
1157 Py_ssize_t n;
Victor Stinnera3c02022015-03-20 11:58:18 +01001158 int err;
Victor Stinner66aab0c2015-03-19 22:53:20 +01001159 int async_err = 0;
1160
1161 /* _Py_read() must not be called with an exception set, otherwise the
1162 * caller may think that read() was interrupted by a signal and the signal
1163 * handler raised an exception. */
1164 assert(!PyErr_Occurred());
1165
Victor Stinnerc1cf4f72015-03-19 23:53:04 +01001166 if (!_PyVerify_fd(fd)) {
Victor Stinnera3c02022015-03-20 11:58:18 +01001167 /* save/restore errno because PyErr_SetFromErrno() can modify it */
1168 err = errno;
Victor Stinner66aab0c2015-03-19 22:53:20 +01001169 PyErr_SetFromErrno(PyExc_OSError);
Victor Stinnera3c02022015-03-20 11:58:18 +01001170 errno = err;
Victor Stinner66aab0c2015-03-19 22:53:20 +01001171 return -1;
1172 }
1173
1174#ifdef MS_WINDOWS
1175 if (count > INT_MAX) {
1176 /* On Windows, the count parameter of read() is an int */
1177 count = INT_MAX;
1178 }
1179#else
1180 if (count > PY_SSIZE_T_MAX) {
1181 /* if count is greater than PY_SSIZE_T_MAX,
1182 * read() result is undefined */
1183 count = PY_SSIZE_T_MAX;
1184 }
1185#endif
1186
Steve Dower8fc89802015-04-12 00:26:27 -04001187 _Py_BEGIN_SUPPRESS_IPH
Victor Stinner66aab0c2015-03-19 22:53:20 +01001188 do {
1189 Py_BEGIN_ALLOW_THREADS
1190 errno = 0;
1191#ifdef MS_WINDOWS
1192 n = read(fd, buf, (int)count);
1193#else
1194 n = read(fd, buf, count);
1195#endif
Victor Stinnera3c02022015-03-20 11:58:18 +01001196 /* save/restore errno because PyErr_CheckSignals()
1197 * and PyErr_SetFromErrno() can modify it */
1198 err = errno;
Victor Stinner66aab0c2015-03-19 22:53:20 +01001199 Py_END_ALLOW_THREADS
Victor Stinnera3c02022015-03-20 11:58:18 +01001200 } while (n < 0 && err == EINTR &&
Victor Stinner66aab0c2015-03-19 22:53:20 +01001201 !(async_err = PyErr_CheckSignals()));
Steve Dower8fc89802015-04-12 00:26:27 -04001202 _Py_END_SUPPRESS_IPH
Victor Stinner66aab0c2015-03-19 22:53:20 +01001203
1204 if (async_err) {
1205 /* read() was interrupted by a signal (failed with EINTR)
1206 * and the Python signal handler raised an exception */
Victor Stinnera3c02022015-03-20 11:58:18 +01001207 errno = err;
1208 assert(errno == EINTR && PyErr_Occurred());
Victor Stinner66aab0c2015-03-19 22:53:20 +01001209 return -1;
1210 }
1211 if (n < 0) {
Victor Stinner66aab0c2015-03-19 22:53:20 +01001212 PyErr_SetFromErrno(PyExc_OSError);
Victor Stinnera3c02022015-03-20 11:58:18 +01001213 errno = err;
Victor Stinner66aab0c2015-03-19 22:53:20 +01001214 return -1;
1215 }
1216
1217 return n;
1218}
1219
Victor Stinner82c3e452015-04-01 18:34:45 +02001220static Py_ssize_t
1221_Py_write_impl(int fd, const void *buf, size_t count, int gil_held)
Victor Stinner66aab0c2015-03-19 22:53:20 +01001222{
1223 Py_ssize_t n;
Victor Stinnera3c02022015-03-20 11:58:18 +01001224 int err;
Victor Stinner66aab0c2015-03-19 22:53:20 +01001225 int async_err = 0;
1226
Victor Stinner66aab0c2015-03-19 22:53:20 +01001227 if (!_PyVerify_fd(fd)) {
Victor Stinner82c3e452015-04-01 18:34:45 +02001228 if (gil_held) {
1229 /* save/restore errno because PyErr_SetFromErrno() can modify it */
1230 err = errno;
1231 PyErr_SetFromErrno(PyExc_OSError);
1232 errno = err;
1233 }
Victor Stinner66aab0c2015-03-19 22:53:20 +01001234 return -1;
1235 }
1236
Steve Dower8fc89802015-04-12 00:26:27 -04001237 _Py_BEGIN_SUPPRESS_IPH
Victor Stinner66aab0c2015-03-19 22:53:20 +01001238#ifdef MS_WINDOWS
1239 if (count > 32767 && isatty(fd)) {
1240 /* Issue #11395: the Windows console returns an error (12: not
1241 enough space error) on writing into stdout if stdout mode is
1242 binary and the length is greater than 66,000 bytes (or less,
1243 depending on heap usage). */
1244 count = 32767;
1245 }
1246 else if (count > INT_MAX)
1247 count = INT_MAX;
1248#else
1249 if (count > PY_SSIZE_T_MAX) {
1250 /* write() should truncate count to PY_SSIZE_T_MAX, but it's safer
1251 * to do it ourself to have a portable behaviour. */
1252 count = PY_SSIZE_T_MAX;
1253 }
1254#endif
1255
Victor Stinner82c3e452015-04-01 18:34:45 +02001256 if (gil_held) {
1257 do {
1258 Py_BEGIN_ALLOW_THREADS
1259 errno = 0;
Victor Stinner66aab0c2015-03-19 22:53:20 +01001260#ifdef MS_WINDOWS
Victor Stinner82c3e452015-04-01 18:34:45 +02001261 n = write(fd, buf, (int)count);
Victor Stinner66aab0c2015-03-19 22:53:20 +01001262#else
Victor Stinner82c3e452015-04-01 18:34:45 +02001263 n = write(fd, buf, count);
Victor Stinner66aab0c2015-03-19 22:53:20 +01001264#endif
Victor Stinner82c3e452015-04-01 18:34:45 +02001265 /* save/restore errno because PyErr_CheckSignals()
1266 * and PyErr_SetFromErrno() can modify it */
1267 err = errno;
1268 Py_END_ALLOW_THREADS
1269 } while (n < 0 && err == EINTR &&
1270 !(async_err = PyErr_CheckSignals()));
1271 }
1272 else {
1273 do {
1274 errno = 0;
1275#ifdef MS_WINDOWS
1276 n = write(fd, buf, (int)count);
1277#else
1278 n = write(fd, buf, count);
1279#endif
1280 err = errno;
1281 } while (n < 0 && err == EINTR);
1282 }
Steve Dower8fc89802015-04-12 00:26:27 -04001283 _Py_END_SUPPRESS_IPH
Victor Stinner66aab0c2015-03-19 22:53:20 +01001284
1285 if (async_err) {
1286 /* write() was interrupted by a signal (failed with EINTR)
Victor Stinner82c3e452015-04-01 18:34:45 +02001287 and the Python signal handler raised an exception (if gil_held is
1288 nonzero). */
Victor Stinnera3c02022015-03-20 11:58:18 +01001289 errno = err;
Victor Stinner82c3e452015-04-01 18:34:45 +02001290 assert(errno == EINTR && (!gil_held || PyErr_Occurred()));
Victor Stinner66aab0c2015-03-19 22:53:20 +01001291 return -1;
1292 }
1293 if (n < 0) {
Victor Stinner82c3e452015-04-01 18:34:45 +02001294 if (gil_held)
1295 PyErr_SetFromErrno(PyExc_OSError);
Victor Stinnera3c02022015-03-20 11:58:18 +01001296 errno = err;
Victor Stinner66aab0c2015-03-19 22:53:20 +01001297 return -1;
1298 }
1299
1300 return n;
1301}
1302
Victor Stinner82c3e452015-04-01 18:34:45 +02001303/* Write count bytes of buf into fd.
1304
1305 On success, return the number of written bytes, it can be lower than count
1306 including 0. On error, raise an exception, set errno and return -1.
1307
1308 When interrupted by a signal (write() fails with EINTR), retry the syscall.
1309 If the Python signal handler raises an exception, the function returns -1
1310 (the syscall is not retried).
1311
1312 Release the GIL to call write(). The caller must hold the GIL. */
1313Py_ssize_t
1314_Py_write(int fd, const void *buf, size_t count)
1315{
1316 /* _Py_write() must not be called with an exception set, otherwise the
1317 * caller may think that write() was interrupted by a signal and the signal
1318 * handler raised an exception. */
1319 assert(!PyErr_Occurred());
1320
1321 return _Py_write_impl(fd, buf, count, 1);
1322}
1323
1324/* Write count bytes of buf into fd.
1325 *
1326 * On success, return the number of written bytes, it can be lower than count
1327 * including 0. On error, set errno and return -1.
1328 *
1329 * When interrupted by a signal (write() fails with EINTR), retry the syscall
1330 * without calling the Python signal handler. */
1331Py_ssize_t
1332_Py_write_noraise(int fd, const void *buf, size_t count)
1333{
1334 return _Py_write_impl(fd, buf, count, 0);
1335}
1336
Victor Stinner4e314432010-10-07 21:45:39 +00001337#ifdef HAVE_READLINK
Victor Stinner6672d0c2010-10-07 22:53:43 +00001338
1339/* Read value of symbolic link. Encode the path to the locale encoding, decode
Victor Stinneraf02e1c2011-12-16 23:56:01 +01001340 the result from the locale encoding. Return -1 on error. */
Victor Stinner6672d0c2010-10-07 22:53:43 +00001341
Victor Stinner4e314432010-10-07 21:45:39 +00001342int
1343_Py_wreadlink(const wchar_t *path, wchar_t *buf, size_t bufsiz)
1344{
1345 char *cpath;
Victor Stinnerb11d6cb2013-11-15 18:14:11 +01001346 char cbuf[MAXPATHLEN];
Victor Stinner3f711f42010-10-16 22:47:37 +00001347 wchar_t *wbuf;
Victor Stinner4e314432010-10-07 21:45:39 +00001348 int res;
1349 size_t r1;
1350
Victor Stinnerf6a271a2014-08-01 12:28:48 +02001351 cpath = Py_EncodeLocale(path, NULL);
Victor Stinner4e314432010-10-07 21:45:39 +00001352 if (cpath == NULL) {
1353 errno = EINVAL;
1354 return -1;
1355 }
Victor Stinnerb11d6cb2013-11-15 18:14:11 +01001356 res = (int)readlink(cpath, cbuf, Py_ARRAY_LENGTH(cbuf));
Victor Stinner4e314432010-10-07 21:45:39 +00001357 PyMem_Free(cpath);
1358 if (res == -1)
1359 return -1;
Victor Stinnerb11d6cb2013-11-15 18:14:11 +01001360 if (res == Py_ARRAY_LENGTH(cbuf)) {
Victor Stinner4e314432010-10-07 21:45:39 +00001361 errno = EINVAL;
1362 return -1;
1363 }
1364 cbuf[res] = '\0'; /* buf will be null terminated */
Victor Stinnerf6a271a2014-08-01 12:28:48 +02001365 wbuf = Py_DecodeLocale(cbuf, &r1);
Victor Stinner350147b2010-10-16 22:52:09 +00001366 if (wbuf == NULL) {
1367 errno = EINVAL;
1368 return -1;
1369 }
Victor Stinner3f711f42010-10-16 22:47:37 +00001370 if (bufsiz <= r1) {
Victor Stinner1a7425f2013-07-07 16:25:15 +02001371 PyMem_RawFree(wbuf);
Victor Stinner4e314432010-10-07 21:45:39 +00001372 errno = EINVAL;
1373 return -1;
1374 }
Victor Stinner3f711f42010-10-16 22:47:37 +00001375 wcsncpy(buf, wbuf, bufsiz);
Victor Stinner1a7425f2013-07-07 16:25:15 +02001376 PyMem_RawFree(wbuf);
Victor Stinner4e314432010-10-07 21:45:39 +00001377 return (int)r1;
1378}
1379#endif
1380
1381#ifdef HAVE_REALPATH
Victor Stinner6672d0c2010-10-07 22:53:43 +00001382
1383/* Return the canonicalized absolute pathname. Encode path to the locale
Victor Stinneraf02e1c2011-12-16 23:56:01 +01001384 encoding, decode the result from the locale encoding.
1385 Return NULL on error. */
Victor Stinner6672d0c2010-10-07 22:53:43 +00001386
Victor Stinner4e314432010-10-07 21:45:39 +00001387wchar_t*
Victor Stinner015f4d82010-10-07 22:29:53 +00001388_Py_wrealpath(const wchar_t *path,
1389 wchar_t *resolved_path, size_t resolved_path_size)
Victor Stinner4e314432010-10-07 21:45:39 +00001390{
1391 char *cpath;
Victor Stinnerb11d6cb2013-11-15 18:14:11 +01001392 char cresolved_path[MAXPATHLEN];
Victor Stinner0a1b8cb2010-10-16 22:55:47 +00001393 wchar_t *wresolved_path;
Victor Stinner4e314432010-10-07 21:45:39 +00001394 char *res;
1395 size_t r;
Victor Stinnerf6a271a2014-08-01 12:28:48 +02001396 cpath = Py_EncodeLocale(path, NULL);
Victor Stinner4e314432010-10-07 21:45:39 +00001397 if (cpath == NULL) {
1398 errno = EINVAL;
1399 return NULL;
1400 }
1401 res = realpath(cpath, cresolved_path);
1402 PyMem_Free(cpath);
1403 if (res == NULL)
1404 return NULL;
Victor Stinner0a1b8cb2010-10-16 22:55:47 +00001405
Victor Stinnerf6a271a2014-08-01 12:28:48 +02001406 wresolved_path = Py_DecodeLocale(cresolved_path, &r);
Victor Stinner0a1b8cb2010-10-16 22:55:47 +00001407 if (wresolved_path == NULL) {
Victor Stinner4e314432010-10-07 21:45:39 +00001408 errno = EINVAL;
1409 return NULL;
1410 }
Victor Stinner0a1b8cb2010-10-16 22:55:47 +00001411 if (resolved_path_size <= r) {
Victor Stinner1a7425f2013-07-07 16:25:15 +02001412 PyMem_RawFree(wresolved_path);
Victor Stinner0a1b8cb2010-10-16 22:55:47 +00001413 errno = EINVAL;
1414 return NULL;
1415 }
1416 wcsncpy(resolved_path, wresolved_path, resolved_path_size);
Victor Stinner1a7425f2013-07-07 16:25:15 +02001417 PyMem_RawFree(wresolved_path);
Victor Stinner4e314432010-10-07 21:45:39 +00001418 return resolved_path;
1419}
1420#endif
1421
Victor Stinnerf4061da2010-10-14 12:37:19 +00001422/* Get the current directory. size is the buffer size in wide characters
Victor Stinneraf02e1c2011-12-16 23:56:01 +01001423 including the null character. Decode the path from the locale encoding.
1424 Return NULL on error. */
Victor Stinner6672d0c2010-10-07 22:53:43 +00001425
Victor Stinner4e314432010-10-07 21:45:39 +00001426wchar_t*
1427_Py_wgetcwd(wchar_t *buf, size_t size)
1428{
1429#ifdef MS_WINDOWS
Victor Stinner56785ea2013-06-05 00:46:29 +02001430 int isize = (int)Py_MIN(size, INT_MAX);
1431 return _wgetcwd(buf, isize);
Victor Stinner4e314432010-10-07 21:45:39 +00001432#else
Victor Stinnerb11d6cb2013-11-15 18:14:11 +01001433 char fname[MAXPATHLEN];
Victor Stinnerf4061da2010-10-14 12:37:19 +00001434 wchar_t *wname;
Victor Stinner168e1172010-10-16 23:16:16 +00001435 size_t len;
Victor Stinnerf4061da2010-10-14 12:37:19 +00001436
Victor Stinnerb11d6cb2013-11-15 18:14:11 +01001437 if (getcwd(fname, Py_ARRAY_LENGTH(fname)) == NULL)
Victor Stinner4e314432010-10-07 21:45:39 +00001438 return NULL;
Victor Stinnerf6a271a2014-08-01 12:28:48 +02001439 wname = Py_DecodeLocale(fname, &len);
Victor Stinnerf4061da2010-10-14 12:37:19 +00001440 if (wname == NULL)
1441 return NULL;
Victor Stinner168e1172010-10-16 23:16:16 +00001442 if (size <= len) {
Victor Stinner1a7425f2013-07-07 16:25:15 +02001443 PyMem_RawFree(wname);
Victor Stinner4e314432010-10-07 21:45:39 +00001444 return NULL;
1445 }
Victor Stinnerf4061da2010-10-14 12:37:19 +00001446 wcsncpy(buf, wname, size);
Victor Stinner1a7425f2013-07-07 16:25:15 +02001447 PyMem_RawFree(wname);
Victor Stinner4e314432010-10-07 21:45:39 +00001448 return buf;
1449#endif
1450}
1451
Victor Stinnerdaf45552013-08-28 00:53:59 +02001452/* Duplicate a file descriptor. The new file descriptor is created as
1453 non-inheritable. Return a new file descriptor on success, raise an OSError
1454 exception and return -1 on error.
1455
1456 The GIL is released to call dup(). The caller must hold the GIL. */
1457int
1458_Py_dup(int fd)
1459{
1460#ifdef MS_WINDOWS
1461 HANDLE handle;
1462 DWORD ftype;
1463#endif
1464
1465 if (!_PyVerify_fd(fd)) {
1466 PyErr_SetFromErrno(PyExc_OSError);
1467 return -1;
1468 }
1469
1470#ifdef MS_WINDOWS
Steve Dower8fc89802015-04-12 00:26:27 -04001471 _Py_BEGIN_SUPPRESS_IPH
Victor Stinnerdaf45552013-08-28 00:53:59 +02001472 handle = (HANDLE)_get_osfhandle(fd);
Steve Dower8fc89802015-04-12 00:26:27 -04001473 _Py_END_SUPPRESS_IPH
Victor Stinnerdaf45552013-08-28 00:53:59 +02001474 if (handle == INVALID_HANDLE_VALUE) {
Steve Dower41e72442015-03-14 11:38:27 -07001475 PyErr_SetFromErrno(PyExc_OSError);
Victor Stinnerdaf45552013-08-28 00:53:59 +02001476 return -1;
1477 }
1478
1479 /* get the file type, ignore the error if it failed */
1480 ftype = GetFileType(handle);
1481
1482 Py_BEGIN_ALLOW_THREADS
Steve Dower8fc89802015-04-12 00:26:27 -04001483 _Py_BEGIN_SUPPRESS_IPH
Victor Stinnerdaf45552013-08-28 00:53:59 +02001484 fd = dup(fd);
Steve Dower8fc89802015-04-12 00:26:27 -04001485 _Py_END_SUPPRESS_IPH
Victor Stinnerdaf45552013-08-28 00:53:59 +02001486 Py_END_ALLOW_THREADS
1487 if (fd < 0) {
1488 PyErr_SetFromErrno(PyExc_OSError);
1489 return -1;
1490 }
1491
1492 /* Character files like console cannot be make non-inheritable */
1493 if (ftype != FILE_TYPE_CHAR) {
1494 if (_Py_set_inheritable(fd, 0, NULL) < 0) {
Steve Dower8fc89802015-04-12 00:26:27 -04001495 _Py_BEGIN_SUPPRESS_IPH
Victor Stinnerdaf45552013-08-28 00:53:59 +02001496 close(fd);
Steve Dower8fc89802015-04-12 00:26:27 -04001497 _Py_END_SUPPRESS_IPH
Victor Stinnerdaf45552013-08-28 00:53:59 +02001498 return -1;
1499 }
1500 }
1501#elif defined(HAVE_FCNTL_H) && defined(F_DUPFD_CLOEXEC)
1502 Py_BEGIN_ALLOW_THREADS
Steve Dower8fc89802015-04-12 00:26:27 -04001503 _Py_BEGIN_SUPPRESS_IPH
Victor Stinnerdaf45552013-08-28 00:53:59 +02001504 fd = fcntl(fd, F_DUPFD_CLOEXEC, 0);
Steve Dower8fc89802015-04-12 00:26:27 -04001505 _Py_END_SUPPRESS_IPH
Victor Stinnerdaf45552013-08-28 00:53:59 +02001506 Py_END_ALLOW_THREADS
1507 if (fd < 0) {
1508 PyErr_SetFromErrno(PyExc_OSError);
1509 return -1;
1510 }
1511
1512#else
1513 Py_BEGIN_ALLOW_THREADS
Steve Dower8fc89802015-04-12 00:26:27 -04001514 _Py_BEGIN_SUPPRESS_IPH
Victor Stinnerdaf45552013-08-28 00:53:59 +02001515 fd = dup(fd);
Steve Dower8fc89802015-04-12 00:26:27 -04001516 _Py_END_SUPPRESS_IPH
Victor Stinnerdaf45552013-08-28 00:53:59 +02001517 Py_END_ALLOW_THREADS
1518 if (fd < 0) {
1519 PyErr_SetFromErrno(PyExc_OSError);
1520 return -1;
1521 }
1522
1523 if (_Py_set_inheritable(fd, 0, NULL) < 0) {
Steve Dower8fc89802015-04-12 00:26:27 -04001524 _Py_BEGIN_SUPPRESS_IPH
Victor Stinnerdaf45552013-08-28 00:53:59 +02001525 close(fd);
Steve Dower8fc89802015-04-12 00:26:27 -04001526 _Py_END_SUPPRESS_IPH
Victor Stinnerdaf45552013-08-28 00:53:59 +02001527 return -1;
1528 }
1529#endif
1530 return fd;
1531}
1532
Victor Stinner1db9e7b2014-07-29 22:32:47 +02001533#ifndef MS_WINDOWS
1534/* Get the blocking mode of the file descriptor.
1535 Return 0 if the O_NONBLOCK flag is set, 1 if the flag is cleared,
1536 raise an exception and return -1 on error. */
1537int
1538_Py_get_blocking(int fd)
1539{
Steve Dower8fc89802015-04-12 00:26:27 -04001540 int flags;
1541 _Py_BEGIN_SUPPRESS_IPH
1542 flags = fcntl(fd, F_GETFL, 0);
1543 _Py_END_SUPPRESS_IPH
Victor Stinner1db9e7b2014-07-29 22:32:47 +02001544 if (flags < 0) {
1545 PyErr_SetFromErrno(PyExc_OSError);
1546 return -1;
1547 }
1548
1549 return !(flags & O_NONBLOCK);
1550}
1551
1552/* Set the blocking mode of the specified file descriptor.
1553
1554 Set the O_NONBLOCK flag if blocking is False, clear the O_NONBLOCK flag
1555 otherwise.
1556
1557 Return 0 on success, raise an exception and return -1 on error. */
1558int
1559_Py_set_blocking(int fd, int blocking)
1560{
1561#if defined(HAVE_SYS_IOCTL_H) && defined(FIONBIO)
1562 int arg = !blocking;
1563 if (ioctl(fd, FIONBIO, &arg) < 0)
1564 goto error;
1565#else
1566 int flags, res;
1567
Steve Dower8fc89802015-04-12 00:26:27 -04001568 _Py_BEGIN_SUPPRESS_IPH
Victor Stinner1db9e7b2014-07-29 22:32:47 +02001569 flags = fcntl(fd, F_GETFL, 0);
Steve Dower8fc89802015-04-12 00:26:27 -04001570 if (flags >= 0) {
1571 if (blocking)
1572 flags = flags & (~O_NONBLOCK);
1573 else
1574 flags = flags | O_NONBLOCK;
Victor Stinner1db9e7b2014-07-29 22:32:47 +02001575
Steve Dower8fc89802015-04-12 00:26:27 -04001576 res = fcntl(fd, F_SETFL, flags);
1577 } else {
1578 res = -1;
1579 }
1580 _Py_END_SUPPRESS_IPH
Victor Stinner1db9e7b2014-07-29 22:32:47 +02001581
Victor Stinner1db9e7b2014-07-29 22:32:47 +02001582 if (res < 0)
1583 goto error;
1584#endif
1585 return 0;
1586
1587error:
1588 PyErr_SetFromErrno(PyExc_OSError);
1589 return -1;
1590}
1591#endif
1592
Steve Dower8fc89802015-04-12 00:26:27 -04001593#if defined _MSC_VER && _MSC_VER >= 1400 && _MSC_VER < 1900
Steve Dowerd81431f2015-03-06 14:47:02 -08001594/* Legacy implementation of _PyVerify_fd while transitioning to
1595 * MSVC 14.0. This should eventually be removed. (issue23524)
1596 */
1597
1598/* Microsoft CRT in VS2005 and higher will verify that a filehandle is
1599 * valid and raise an assertion if it isn't.
1600 * Normally, an invalid fd is likely to be a C program error and therefore
1601 * an assertion can be useful, but it does contradict the POSIX standard
1602 * which for write(2) states:
1603 * "Otherwise, -1 shall be returned and errno set to indicate the error."
1604 * "[EBADF] The fildes argument is not a valid file descriptor open for
1605 * writing."
1606 * Furthermore, python allows the user to enter any old integer
1607 * as a fd and should merely raise a python exception on error.
1608 * The Microsoft CRT doesn't provide an official way to check for the
1609 * validity of a file descriptor, but we can emulate its internal behaviour
1610 * by using the exported __pinfo data member and knowledge of the
1611 * internal structures involved.
1612 * The structures below must be updated for each version of visual studio
1613 * according to the file internal.h in the CRT source, until MS comes
1614 * up with a less hacky way to do this.
1615 * (all of this is to avoid globally modifying the CRT behaviour using
1616 * _set_invalid_parameter_handler() and _CrtSetReportMode())
1617 */
1618/* The actual size of the structure is determined at runtime.
1619 * Only the first items must be present.
1620 */
1621typedef struct {
1622 intptr_t osfhnd;
1623 char osfile;
1624} my_ioinfo;
1625
1626extern __declspec(dllimport) char * __pioinfo[];
1627#define IOINFO_L2E 5
1628#define IOINFO_ARRAYS 64
1629#define IOINFO_ARRAY_ELTS (1 << IOINFO_L2E)
1630#define _NHANDLE_ (IOINFO_ARRAYS * IOINFO_ARRAY_ELTS)
1631#define FOPEN 0x01
1632#define _NO_CONSOLE_FILENO (intptr_t)-2
1633
1634/* This function emulates what the windows CRT does to validate file handles */
1635int
1636_PyVerify_fd(int fd)
1637{
1638 const int i1 = fd >> IOINFO_L2E;
1639 const int i2 = fd & ((1 << IOINFO_L2E) - 1);
1640
1641 static size_t sizeof_ioinfo = 0;
1642
1643 /* Determine the actual size of the ioinfo structure,
1644 * as used by the CRT loaded in memory
1645 */
1646 if (sizeof_ioinfo == 0 && __pioinfo[0] != NULL) {
1647 sizeof_ioinfo = _msize(__pioinfo[0]) / IOINFO_ARRAY_ELTS;
1648 }
1649 if (sizeof_ioinfo == 0) {
1650 /* This should not happen... */
1651 goto fail;
1652 }
1653
1654 /* See that it isn't a special CLEAR fileno */
1655 if (fd != _NO_CONSOLE_FILENO) {
1656 /* Microsoft CRT would check that 0<=fd<_nhandle but we can't do that. Instead
1657 * we check pointer validity and other info
1658 */
1659 if (0 <= i1 && i1 < IOINFO_ARRAYS && __pioinfo[i1] != NULL) {
1660 /* finally, check that the file is open */
1661 my_ioinfo* info = (my_ioinfo*)(__pioinfo[i1] + i2 * sizeof_ioinfo);
1662 if (info->osfile & FOPEN) {
1663 return 1;
1664 }
1665 }
1666 }
1667 fail:
1668 errno = EBADF;
1669 return 0;
1670}
1671
Steve Dower8fc89802015-04-12 00:26:27 -04001672#endif /* defined _MSC_VER && _MSC_VER >= 1400 && _MSC_VER < 1900 */