/src/cpython/Python/fileutils.c
Line | Count | Source (jump to first uncovered line) |
1 | | #include "Python.h" |
2 | | #include "pycore_fileutils.h" // fileutils definitions |
3 | | #include "pycore_runtime.h" // _PyRuntime |
4 | | #include "pycore_pystate.h" // _Py_AssertHoldsTstate() |
5 | | #include "osdefs.h" // SEP |
6 | | |
7 | | #include <stdlib.h> // mbstowcs() |
8 | | #ifdef HAVE_UNISTD_H |
9 | | # include <unistd.h> // getcwd() |
10 | | #endif |
11 | | |
12 | | #ifdef MS_WINDOWS |
13 | | # include <malloc.h> |
14 | | # include <windows.h> |
15 | | # include <winioctl.h> // FILE_DEVICE_* constants |
16 | | # include "pycore_fileutils_windows.h" // FILE_STAT_BASIC_INFORMATION |
17 | | # if defined(MS_WINDOWS_GAMES) && !defined(MS_WINDOWS_DESKTOP) |
18 | | # define PATHCCH_ALLOW_LONG_PATHS 0x01 |
19 | | # else |
20 | | # include <pathcch.h> // PathCchCombineEx |
21 | | # endif |
22 | | extern int winerror_to_errno(int); |
23 | | #endif |
24 | | |
25 | | #ifdef HAVE_LANGINFO_H |
26 | | # include <langinfo.h> // nl_langinfo(CODESET) |
27 | | #endif |
28 | | |
29 | | #ifdef HAVE_SYS_IOCTL_H |
30 | | #include <sys/ioctl.h> |
31 | | #endif |
32 | | |
33 | | #ifdef HAVE_NON_UNICODE_WCHAR_T_REPRESENTATION |
34 | | # include <iconv.h> // iconv_open() |
35 | | #endif |
36 | | |
37 | | #ifdef HAVE_FCNTL_H |
38 | | # include <fcntl.h> // fcntl(F_GETFD) |
39 | | #endif |
40 | | |
41 | | #ifdef O_CLOEXEC |
42 | | /* Does open() support the O_CLOEXEC flag? Possible values: |
43 | | |
44 | | -1: unknown |
45 | | 0: open() ignores O_CLOEXEC flag, ex: Linux kernel older than 2.6.23 |
46 | | 1: open() supports O_CLOEXEC flag, close-on-exec is set |
47 | | |
48 | | The flag is used by _Py_open(), _Py_open_noraise(), io.FileIO |
49 | | and os.open(). */ |
50 | | int _Py_open_cloexec_works = -1; |
51 | | #endif |
52 | | |
53 | | // The value must be the same in unicodeobject.c. |
54 | 363k | #define MAX_UNICODE 0x10ffff |
55 | | |
56 | | // mbstowcs() and mbrtowc() errors |
57 | | static const size_t DECODE_ERROR = ((size_t)-1); |
58 | | #ifdef HAVE_MBRTOWC |
59 | | static const size_t INCOMPLETE_CHARACTER = (size_t)-2; |
60 | | #endif |
61 | | |
62 | | |
63 | | static int |
64 | | get_surrogateescape(_Py_error_handler errors, int *surrogateescape) |
65 | 18.0k | { |
66 | 18.0k | switch (errors) |
67 | 18.0k | { |
68 | 0 | case _Py_ERROR_STRICT: |
69 | 0 | *surrogateescape = 0; |
70 | 0 | return 0; |
71 | 18.0k | case _Py_ERROR_SURROGATEESCAPE: |
72 | 18.0k | *surrogateescape = 1; |
73 | 18.0k | return 0; |
74 | 0 | default: |
75 | 0 | return -1; |
76 | 18.0k | } |
77 | 18.0k | } |
78 | | |
79 | | |
80 | | PyObject * |
81 | | _Py_device_encoding(int fd) |
82 | 0 | { |
83 | 0 | int valid; |
84 | 0 | Py_BEGIN_ALLOW_THREADS |
85 | 0 | _Py_BEGIN_SUPPRESS_IPH |
86 | 0 | valid = isatty(fd); |
87 | 0 | _Py_END_SUPPRESS_IPH |
88 | 0 | Py_END_ALLOW_THREADS |
89 | 0 | if (!valid) |
90 | 0 | Py_RETURN_NONE; |
91 | | |
92 | | #ifdef MS_WINDOWS |
93 | | #ifdef HAVE_WINDOWS_CONSOLE_IO |
94 | | UINT cp; |
95 | | if (fd == 0) |
96 | | cp = GetConsoleCP(); |
97 | | else if (fd == 1 || fd == 2) |
98 | | cp = GetConsoleOutputCP(); |
99 | | else |
100 | | cp = 0; |
101 | | /* GetConsoleCP() and GetConsoleOutputCP() return 0 if the application |
102 | | has no console */ |
103 | | if (cp == 0) { |
104 | | Py_RETURN_NONE; |
105 | | } |
106 | | |
107 | | return PyUnicode_FromFormat("cp%u", (unsigned int)cp); |
108 | | #else |
109 | | Py_RETURN_NONE; |
110 | | #endif /* HAVE_WINDOWS_CONSOLE_IO */ |
111 | | #else |
112 | 0 | if (_PyRuntime.preconfig.utf8_mode) { |
113 | 0 | _Py_DECLARE_STR(utf_8, "utf-8"); |
114 | 0 | return &_Py_STR(utf_8); |
115 | 0 | } |
116 | 0 | return _Py_GetLocaleEncodingObject(); |
117 | 0 | #endif |
118 | 0 | } |
119 | | |
120 | | |
121 | | static int |
122 | | is_valid_wide_char(wchar_t ch) |
123 | 363k | { |
124 | | #ifdef HAVE_NON_UNICODE_WCHAR_T_REPRESENTATION |
125 | | /* Oracle Solaris doesn't use Unicode code points as wchar_t encoding |
126 | | for non-Unicode locales, which makes values higher than MAX_UNICODE |
127 | | possibly valid. */ |
128 | | return 1; |
129 | | #endif |
130 | 363k | if (Py_UNICODE_IS_SURROGATE(ch)) { |
131 | | // Reject lone surrogate characters |
132 | 0 | return 0; |
133 | 0 | } |
134 | 363k | #if SIZEOF_WCHAR_T > 2 |
135 | 363k | if (ch > MAX_UNICODE) { |
136 | | // bpo-35883: Reject characters outside [U+0000; U+10ffff] range. |
137 | | // The glibc mbstowcs() UTF-8 decoder does not respect the RFC 3629, |
138 | | // it creates characters outside the [U+0000; U+10ffff] range: |
139 | | // https://sourceware.org/bugzilla/show_bug.cgi?id=2373 |
140 | 0 | return 0; |
141 | 0 | } |
142 | 363k | #endif |
143 | 363k | return 1; |
144 | 363k | } |
145 | | |
146 | | |
147 | | static size_t |
148 | | _Py_mbstowcs(wchar_t *dest, const char *src, size_t n) |
149 | 36.9k | { |
150 | 36.9k | size_t count = mbstowcs(dest, src, n); |
151 | 36.9k | if (dest != NULL && count != DECODE_ERROR) { |
152 | 381k | for (size_t i=0; i < count; i++) { |
153 | 363k | wchar_t ch = dest[i]; |
154 | 363k | if (!is_valid_wide_char(ch)) { |
155 | 0 | return DECODE_ERROR; |
156 | 0 | } |
157 | 363k | } |
158 | 17.4k | } |
159 | 36.9k | return count; |
160 | 36.9k | } |
161 | | |
162 | | |
163 | | #ifdef HAVE_MBRTOWC |
164 | | static size_t |
165 | | _Py_mbrtowc(wchar_t *pwc, const char *str, size_t len, mbstate_t *pmbs) |
166 | 0 | { |
167 | 0 | assert(pwc != NULL); |
168 | 0 | size_t count = mbrtowc(pwc, str, len, pmbs); |
169 | 0 | if (count != 0 && count != DECODE_ERROR && count != INCOMPLETE_CHARACTER) { |
170 | 0 | if (!is_valid_wide_char(*pwc)) { |
171 | 0 | return DECODE_ERROR; |
172 | 0 | } |
173 | 0 | } |
174 | 0 | return count; |
175 | 0 | } |
176 | | #endif |
177 | | |
178 | | |
179 | | #if !defined(_Py_FORCE_UTF8_FS_ENCODING) && !defined(MS_WINDOWS) |
180 | | |
181 | | #define USE_FORCE_ASCII |
182 | | |
183 | | extern int _Py_normalize_encoding(const char *, char *, size_t); |
184 | | |
185 | | /* Workaround FreeBSD and OpenIndiana locale encoding issue with the C locale |
186 | | and POSIX locale. nl_langinfo(CODESET) announces an alias of the |
187 | | ASCII encoding, whereas mbstowcs() and wcstombs() functions use the |
188 | | ISO-8859-1 encoding. The problem is that os.fsencode() and os.fsdecode() use |
189 | | locale.getpreferredencoding() codec. For example, if command line arguments |
190 | | are decoded by mbstowcs() and encoded back by os.fsencode(), we get a |
191 | | UnicodeEncodeError instead of retrieving the original byte string. |
192 | | |
193 | | The workaround is enabled if setlocale(LC_CTYPE, NULL) returns "C", |
194 | | nl_langinfo(CODESET) announces "ascii" (or an alias to ASCII), and at least |
195 | | one byte in range 0x80-0xff can be decoded from the locale encoding. The |
196 | | workaround is also enabled on error, for example if getting the locale |
197 | | failed. |
198 | | |
199 | | On HP-UX with the C locale or the POSIX locale, nl_langinfo(CODESET) |
200 | | announces "roman8" but mbstowcs() uses Latin1 in practice. Force also the |
201 | | ASCII encoding in this case. |
202 | | |
203 | | Values of force_ascii: |
204 | | |
205 | | 1: the workaround is used: Py_EncodeLocale() uses |
206 | | encode_ascii_surrogateescape() and Py_DecodeLocale() uses |
207 | | decode_ascii() |
208 | | 0: the workaround is not used: Py_EncodeLocale() uses wcstombs() and |
209 | | Py_DecodeLocale() uses mbstowcs() |
210 | | -1: unknown, need to call check_force_ascii() to get the value |
211 | | */ |
212 | 11.6k | #define force_ascii (_PyRuntime.fileutils.force_ascii) |
213 | | |
214 | | static int |
215 | | check_force_ascii(void) |
216 | 16 | { |
217 | 16 | char *loc = setlocale(LC_CTYPE, NULL); |
218 | 16 | if (loc == NULL) { |
219 | 0 | goto error; |
220 | 0 | } |
221 | 16 | if (strcmp(loc, "C") != 0 && strcmp(loc, "POSIX") != 0) { |
222 | | /* the LC_CTYPE locale is different than C and POSIX */ |
223 | 0 | return 0; |
224 | 0 | } |
225 | | |
226 | 16 | #if defined(HAVE_LANGINFO_H) && defined(CODESET) |
227 | 16 | const char *codeset = nl_langinfo(CODESET); |
228 | 16 | if (!codeset || codeset[0] == '\0') { |
229 | | /* CODESET is not set or empty */ |
230 | 0 | goto error; |
231 | 0 | } |
232 | | |
233 | 16 | char encoding[20]; /* longest name: "iso_646.irv_1991\0" */ |
234 | 16 | if (!_Py_normalize_encoding(codeset, encoding, sizeof(encoding))) { |
235 | 0 | goto error; |
236 | 0 | } |
237 | | |
238 | | #ifdef __hpux |
239 | | if (strcmp(encoding, "roman8") == 0) { |
240 | | unsigned char ch; |
241 | | wchar_t wch; |
242 | | size_t res; |
243 | | |
244 | | ch = (unsigned char)0xA7; |
245 | | res = _Py_mbstowcs(&wch, (char*)&ch, 1); |
246 | | if (res != DECODE_ERROR && wch == L'\xA7') { |
247 | | /* On HP-UX with C locale or the POSIX locale, |
248 | | nl_langinfo(CODESET) announces "roman8", whereas mbstowcs() uses |
249 | | Latin1 encoding in practice. Force ASCII in this case. |
250 | | |
251 | | Roman8 decodes 0xA7 to U+00CF. Latin1 decodes 0xA7 to U+00A7. */ |
252 | | return 1; |
253 | | } |
254 | | } |
255 | | #else |
256 | 16 | const char* ascii_aliases[] = { |
257 | 16 | "ascii", |
258 | | /* Aliases from Lib/encodings/aliases.py */ |
259 | 16 | "646", |
260 | 16 | "ansi_x3.4_1968", |
261 | 16 | "ansi_x3.4_1986", |
262 | 16 | "ansi_x3_4_1968", |
263 | 16 | "cp367", |
264 | 16 | "csascii", |
265 | 16 | "ibm367", |
266 | 16 | "iso646_us", |
267 | 16 | "iso_646.irv_1991", |
268 | 16 | "iso_ir_6", |
269 | 16 | "us", |
270 | 16 | "us_ascii", |
271 | 16 | NULL |
272 | 16 | }; |
273 | | |
274 | 16 | int is_ascii = 0; |
275 | 48 | for (const char **alias=ascii_aliases; *alias != NULL; alias++) { |
276 | 48 | if (strcmp(encoding, *alias) == 0) { |
277 | 16 | is_ascii = 1; |
278 | 16 | break; |
279 | 16 | } |
280 | 48 | } |
281 | 16 | if (!is_ascii) { |
282 | | /* nl_langinfo(CODESET) is not "ascii" or an alias of ASCII */ |
283 | 0 | return 0; |
284 | 0 | } |
285 | | |
286 | 2.06k | for (unsigned int i=0x80; i<=0xff; i++) { |
287 | 2.04k | char ch[1]; |
288 | 2.04k | wchar_t wch[1]; |
289 | 2.04k | size_t res; |
290 | | |
291 | 2.04k | unsigned uch = (unsigned char)i; |
292 | 2.04k | ch[0] = (char)uch; |
293 | 2.04k | res = _Py_mbstowcs(wch, ch, 1); |
294 | 2.04k | if (res != DECODE_ERROR) { |
295 | | /* decoding a non-ASCII character from the locale encoding succeed: |
296 | | the locale encoding is not ASCII, force ASCII */ |
297 | 0 | return 1; |
298 | 0 | } |
299 | 2.04k | } |
300 | | /* None of the bytes in the range 0x80-0xff can be decoded from the locale |
301 | | encoding: the locale encoding is really ASCII */ |
302 | 16 | #endif /* !defined(__hpux) */ |
303 | 16 | return 0; |
304 | | #else |
305 | | /* nl_langinfo(CODESET) is not available: always force ASCII */ |
306 | | return 1; |
307 | | #endif /* defined(HAVE_LANGINFO_H) && defined(CODESET) */ |
308 | | |
309 | 0 | error: |
310 | | /* if an error occurred, force the ASCII encoding */ |
311 | 0 | return 1; |
312 | 16 | } |
313 | | |
314 | | |
315 | | int |
316 | | _Py_GetForceASCII(void) |
317 | 16 | { |
318 | 16 | if (force_ascii == -1) { |
319 | 16 | force_ascii = check_force_ascii(); |
320 | 16 | } |
321 | 16 | return force_ascii; |
322 | 16 | } |
323 | | |
324 | | |
325 | | void |
326 | | _Py_ResetForceASCII(void) |
327 | 32 | { |
328 | 32 | force_ascii = -1; |
329 | 32 | } |
330 | | |
331 | | |
332 | | static int |
333 | | encode_ascii(const wchar_t *text, char **str, |
334 | | size_t *error_pos, const char **reason, |
335 | | int raw_malloc, _Py_error_handler errors) |
336 | 0 | { |
337 | 0 | char *result = NULL, *out; |
338 | 0 | size_t len, i; |
339 | 0 | wchar_t ch; |
340 | |
|
341 | 0 | int surrogateescape; |
342 | 0 | if (get_surrogateescape(errors, &surrogateescape) < 0) { |
343 | 0 | return -3; |
344 | 0 | } |
345 | | |
346 | 0 | len = wcslen(text); |
347 | | |
348 | | /* +1 for NULL byte */ |
349 | 0 | if (raw_malloc) { |
350 | 0 | result = PyMem_RawMalloc(len + 1); |
351 | 0 | } |
352 | 0 | else { |
353 | 0 | result = PyMem_Malloc(len + 1); |
354 | 0 | } |
355 | 0 | if (result == NULL) { |
356 | 0 | return -1; |
357 | 0 | } |
358 | | |
359 | 0 | out = result; |
360 | 0 | for (i=0; i<len; i++) { |
361 | 0 | ch = text[i]; |
362 | |
|
363 | 0 | if (ch <= 0x7f) { |
364 | | /* ASCII character */ |
365 | 0 | *out++ = (char)ch; |
366 | 0 | } |
367 | 0 | else if (surrogateescape && 0xdc80 <= ch && ch <= 0xdcff) { |
368 | | /* UTF-8b surrogate */ |
369 | 0 | *out++ = (char)(ch - 0xdc00); |
370 | 0 | } |
371 | 0 | else { |
372 | 0 | if (raw_malloc) { |
373 | 0 | PyMem_RawFree(result); |
374 | 0 | } |
375 | 0 | else { |
376 | 0 | PyMem_Free(result); |
377 | 0 | } |
378 | 0 | if (error_pos != NULL) { |
379 | 0 | *error_pos = i; |
380 | 0 | } |
381 | 0 | if (reason) { |
382 | 0 | *reason = "encoding error"; |
383 | 0 | } |
384 | 0 | return -2; |
385 | 0 | } |
386 | 0 | } |
387 | 0 | *out = '\0'; |
388 | 0 | *str = result; |
389 | 0 | return 0; |
390 | 0 | } |
391 | | #else |
392 | | int |
393 | | _Py_GetForceASCII(void) |
394 | | { |
395 | | return 0; |
396 | | } |
397 | | |
398 | | void |
399 | | _Py_ResetForceASCII(void) |
400 | | { |
401 | | /* nothing to do */ |
402 | | } |
403 | | #endif /* !defined(_Py_FORCE_UTF8_FS_ENCODING) && !defined(MS_WINDOWS) */ |
404 | | |
405 | | |
406 | | #if !defined(HAVE_MBRTOWC) || defined(USE_FORCE_ASCII) |
407 | | static int |
408 | | decode_ascii(const char *arg, wchar_t **wstr, size_t *wlen, |
409 | | const char **reason, _Py_error_handler errors) |
410 | 0 | { |
411 | 0 | wchar_t *res; |
412 | 0 | unsigned char *in; |
413 | 0 | wchar_t *out; |
414 | 0 | size_t argsize = strlen(arg) + 1; |
415 | |
|
416 | 0 | int surrogateescape; |
417 | 0 | if (get_surrogateescape(errors, &surrogateescape) < 0) { |
418 | 0 | return -3; |
419 | 0 | } |
420 | | |
421 | 0 | if (argsize > PY_SSIZE_T_MAX / sizeof(wchar_t)) { |
422 | 0 | return -1; |
423 | 0 | } |
424 | 0 | res = PyMem_RawMalloc(argsize * sizeof(wchar_t)); |
425 | 0 | if (!res) { |
426 | 0 | return -1; |
427 | 0 | } |
428 | | |
429 | 0 | out = res; |
430 | 0 | for (in = (unsigned char*)arg; *in; in++) { |
431 | 0 | unsigned char ch = *in; |
432 | 0 | if (ch < 128) { |
433 | 0 | *out++ = ch; |
434 | 0 | } |
435 | 0 | else { |
436 | 0 | if (!surrogateescape) { |
437 | 0 | PyMem_RawFree(res); |
438 | 0 | if (wlen) { |
439 | 0 | *wlen = in - (unsigned char*)arg; |
440 | 0 | } |
441 | 0 | if (reason) { |
442 | 0 | *reason = "decoding error"; |
443 | 0 | } |
444 | 0 | return -2; |
445 | 0 | } |
446 | 0 | *out++ = 0xdc00 + ch; |
447 | 0 | } |
448 | 0 | } |
449 | 0 | *out = 0; |
450 | |
|
451 | 0 | if (wlen != NULL) { |
452 | 0 | *wlen = out - res; |
453 | 0 | } |
454 | 0 | *wstr = res; |
455 | 0 | return 0; |
456 | 0 | } |
457 | | #endif /* !HAVE_MBRTOWC */ |
458 | | |
459 | | static int |
460 | | decode_current_locale(const char* arg, wchar_t **wstr, size_t *wlen, |
461 | | const char **reason, _Py_error_handler errors) |
462 | 17.4k | { |
463 | 17.4k | wchar_t *res; |
464 | 17.4k | size_t argsize; |
465 | 17.4k | size_t count; |
466 | 17.4k | #ifdef HAVE_MBRTOWC |
467 | 17.4k | unsigned char *in; |
468 | 17.4k | wchar_t *out; |
469 | 17.4k | mbstate_t mbs; |
470 | 17.4k | #endif |
471 | | |
472 | 17.4k | int surrogateescape; |
473 | 17.4k | if (get_surrogateescape(errors, &surrogateescape) < 0) { |
474 | 0 | return -3; |
475 | 0 | } |
476 | | |
477 | | #ifdef HAVE_BROKEN_MBSTOWCS |
478 | | /* Some platforms have a broken implementation of |
479 | | * mbstowcs which does not count the characters that |
480 | | * would result from conversion. Use an upper bound. |
481 | | */ |
482 | | argsize = strlen(arg); |
483 | | #else |
484 | 17.4k | argsize = _Py_mbstowcs(NULL, arg, 0); |
485 | 17.4k | #endif |
486 | 17.4k | if (argsize != DECODE_ERROR) { |
487 | 17.4k | if (argsize > PY_SSIZE_T_MAX / sizeof(wchar_t) - 1) { |
488 | 0 | return -1; |
489 | 0 | } |
490 | 17.4k | res = (wchar_t *)PyMem_RawMalloc((argsize + 1) * sizeof(wchar_t)); |
491 | 17.4k | if (!res) { |
492 | 0 | return -1; |
493 | 0 | } |
494 | | |
495 | 17.4k | count = _Py_mbstowcs(res, arg, argsize + 1); |
496 | 17.4k | if (count != DECODE_ERROR) { |
497 | 17.4k | *wstr = res; |
498 | 17.4k | if (wlen != NULL) { |
499 | 17.4k | *wlen = count; |
500 | 17.4k | } |
501 | 17.4k | return 0; |
502 | 17.4k | } |
503 | 0 | PyMem_RawFree(res); |
504 | 0 | } |
505 | | |
506 | | /* Conversion failed. Fall back to escaping with surrogateescape. */ |
507 | 0 | #ifdef HAVE_MBRTOWC |
508 | | /* Try conversion with mbrtwoc (C99), and escape non-decodable bytes. */ |
509 | | |
510 | | /* Overallocate; as multi-byte characters are in the argument, the |
511 | | actual output could use less memory. */ |
512 | 0 | argsize = strlen(arg) + 1; |
513 | 0 | if (argsize > PY_SSIZE_T_MAX / sizeof(wchar_t)) { |
514 | 0 | return -1; |
515 | 0 | } |
516 | 0 | res = (wchar_t*)PyMem_RawMalloc(argsize * sizeof(wchar_t)); |
517 | 0 | if (!res) { |
518 | 0 | return -1; |
519 | 0 | } |
520 | | |
521 | 0 | in = (unsigned char*)arg; |
522 | 0 | out = res; |
523 | 0 | memset(&mbs, 0, sizeof mbs); |
524 | 0 | while (argsize) { |
525 | 0 | size_t converted = _Py_mbrtowc(out, (char*)in, argsize, &mbs); |
526 | 0 | if (converted == 0) { |
527 | | /* Reached end of string; null char stored. */ |
528 | 0 | break; |
529 | 0 | } |
530 | | |
531 | 0 | if (converted == DECODE_ERROR || converted == INCOMPLETE_CHARACTER) { |
532 | 0 | if (!surrogateescape) { |
533 | 0 | goto decode_error; |
534 | 0 | } |
535 | | |
536 | | /* Decoding error. Escape as UTF-8b, and start over in the initial |
537 | | shift state. */ |
538 | 0 | *out++ = 0xdc00 + *in++; |
539 | 0 | argsize--; |
540 | 0 | memset(&mbs, 0, sizeof mbs); |
541 | 0 | continue; |
542 | 0 | } |
543 | | |
544 | | // _Py_mbrtowc() reject lone surrogate characters |
545 | 0 | assert(!Py_UNICODE_IS_SURROGATE(*out)); |
546 | | |
547 | | /* successfully converted some bytes */ |
548 | 0 | in += converted; |
549 | 0 | argsize -= converted; |
550 | 0 | out++; |
551 | 0 | } |
552 | 0 | if (wlen != NULL) { |
553 | 0 | *wlen = out - res; |
554 | 0 | } |
555 | 0 | *wstr = res; |
556 | 0 | return 0; |
557 | | |
558 | 0 | decode_error: |
559 | 0 | PyMem_RawFree(res); |
560 | 0 | if (wlen) { |
561 | 0 | *wlen = in - (unsigned char*)arg; |
562 | 0 | } |
563 | 0 | if (reason) { |
564 | 0 | *reason = "decoding error"; |
565 | 0 | } |
566 | 0 | return -2; |
567 | | #else /* HAVE_MBRTOWC */ |
568 | | /* Cannot use C locale for escaping; manually escape as if charset |
569 | | is ASCII (i.e. escape all bytes > 128. This will still roundtrip |
570 | | correctly in the locale's charset, which must be an ASCII superset. */ |
571 | | return decode_ascii(arg, wstr, wlen, reason, errors); |
572 | | #endif /* HAVE_MBRTOWC */ |
573 | 0 | } |
574 | | |
575 | | |
576 | | /* Decode a byte string from the locale encoding. |
577 | | |
578 | | Use the strict error handler if 'surrogateescape' is zero. Use the |
579 | | surrogateescape error handler if 'surrogateescape' is non-zero: undecodable |
580 | | bytes are decoded as characters in range U+DC80..U+DCFF. If a byte sequence |
581 | | can be decoded as a surrogate character, escape the bytes using the |
582 | | surrogateescape error handler instead of decoding them. |
583 | | |
584 | | On success, return 0 and write the newly allocated wide character string into |
585 | | *wstr (use PyMem_RawFree() to free the memory). If wlen is not NULL, write |
586 | | the number of wide characters excluding the null character into *wlen. |
587 | | |
588 | | On memory allocation failure, return -1. |
589 | | |
590 | | On decoding error, return -2. If wlen is not NULL, write the start of |
591 | | invalid byte sequence in the input string into *wlen. If reason is not NULL, |
592 | | write the decoding error message into *reason. |
593 | | |
594 | | Return -3 if the error handler 'errors' is not supported. |
595 | | |
596 | | Use the Py_EncodeLocaleEx() function to encode the character string back to |
597 | | a byte string. */ |
598 | | int |
599 | | _Py_DecodeLocaleEx(const char* arg, wchar_t **wstr, size_t *wlen, |
600 | | const char **reason, |
601 | | int current_locale, _Py_error_handler errors) |
602 | 17.4k | { |
603 | 17.4k | if (current_locale) { |
604 | | #ifdef _Py_FORCE_UTF8_LOCALE |
605 | | return _Py_DecodeUTF8Ex(arg, strlen(arg), wstr, wlen, reason, |
606 | | errors); |
607 | | #else |
608 | 12.2k | return decode_current_locale(arg, wstr, wlen, reason, errors); |
609 | 12.2k | #endif |
610 | 12.2k | } |
611 | | |
612 | | #ifdef _Py_FORCE_UTF8_FS_ENCODING |
613 | | return _Py_DecodeUTF8Ex(arg, strlen(arg), wstr, wlen, reason, |
614 | | errors); |
615 | | #else |
616 | 5.23k | int use_utf8 = (_PyRuntime.preconfig.utf8_mode >= 1); |
617 | | #ifdef MS_WINDOWS |
618 | | use_utf8 |= (_PyRuntime.preconfig.legacy_windows_fs_encoding == 0); |
619 | | #endif |
620 | 5.23k | if (use_utf8) { |
621 | 0 | return _Py_DecodeUTF8Ex(arg, strlen(arg), wstr, wlen, reason, |
622 | 0 | errors); |
623 | 0 | } |
624 | | |
625 | 5.23k | #ifdef USE_FORCE_ASCII |
626 | 5.23k | if (force_ascii == -1) { |
627 | 0 | force_ascii = check_force_ascii(); |
628 | 0 | } |
629 | | |
630 | 5.23k | if (force_ascii) { |
631 | | /* force ASCII encoding to workaround mbstowcs() issue */ |
632 | 0 | return decode_ascii(arg, wstr, wlen, reason, errors); |
633 | 0 | } |
634 | 5.23k | #endif |
635 | | |
636 | 5.23k | return decode_current_locale(arg, wstr, wlen, reason, errors); |
637 | 5.23k | #endif /* !_Py_FORCE_UTF8_FS_ENCODING */ |
638 | 5.23k | } |
639 | | |
640 | | |
641 | | /* Decode a byte string from the locale encoding with the |
642 | | surrogateescape error handler: undecodable bytes are decoded as characters |
643 | | in range U+DC80..U+DCFF. If a byte sequence can be decoded as a surrogate |
644 | | character, escape the bytes using the surrogateescape error handler instead |
645 | | of decoding them. |
646 | | |
647 | | Return a pointer to a newly allocated wide character string, use |
648 | | PyMem_RawFree() to free the memory. If size is not NULL, write the number of |
649 | | wide characters excluding the null character into *size |
650 | | |
651 | | Return NULL on decoding error or memory allocation error. If *size* is not |
652 | | NULL, *size is set to (size_t)-1 on memory error or set to (size_t)-2 on |
653 | | decoding error. |
654 | | |
655 | | Decoding errors should never happen, unless there is a bug in the C |
656 | | library. |
657 | | |
658 | | Use the Py_EncodeLocale() function to encode the character string back to a |
659 | | byte string. */ |
660 | | wchar_t* |
661 | | Py_DecodeLocale(const char* arg, size_t *wlen) |
662 | 112 | { |
663 | 112 | wchar_t *wstr; |
664 | 112 | int res = _Py_DecodeLocaleEx(arg, &wstr, wlen, |
665 | 112 | NULL, 0, |
666 | 112 | _Py_ERROR_SURROGATEESCAPE); |
667 | 112 | if (res != 0) { |
668 | 0 | assert(res != -3); |
669 | 0 | if (wlen != NULL) { |
670 | 0 | *wlen = (size_t)res; |
671 | 0 | } |
672 | 0 | return NULL; |
673 | 0 | } |
674 | 112 | return wstr; |
675 | 112 | } |
676 | | |
677 | | |
678 | | static int |
679 | | encode_current_locale(const wchar_t *text, char **str, |
680 | | size_t *error_pos, const char **reason, |
681 | | int raw_malloc, _Py_error_handler errors) |
682 | 568 | { |
683 | 568 | const size_t len = wcslen(text); |
684 | 568 | char *result = NULL, *bytes = NULL; |
685 | 568 | size_t i, size, converted; |
686 | 568 | wchar_t c, buf[2]; |
687 | | |
688 | 568 | int surrogateescape; |
689 | 568 | if (get_surrogateescape(errors, &surrogateescape) < 0) { |
690 | 0 | return -3; |
691 | 0 | } |
692 | | |
693 | | /* The function works in two steps: |
694 | | 1. compute the length of the output buffer in bytes (size) |
695 | | 2. outputs the bytes */ |
696 | 568 | size = 0; |
697 | 568 | buf[1] = 0; |
698 | 1.13k | while (1) { |
699 | 82.1k | for (i=0; i < len; i++) { |
700 | 81.0k | c = text[i]; |
701 | 81.0k | if (c >= 0xdc80 && c <= 0xdcff) { |
702 | 0 | if (!surrogateescape) { |
703 | 0 | goto encode_error; |
704 | 0 | } |
705 | | /* UTF-8b surrogate */ |
706 | 0 | if (bytes != NULL) { |
707 | 0 | *bytes++ = c - 0xdc00; |
708 | 0 | size--; |
709 | 0 | } |
710 | 0 | else { |
711 | 0 | size++; |
712 | 0 | } |
713 | 0 | continue; |
714 | 0 | } |
715 | 81.0k | else { |
716 | 81.0k | buf[0] = c; |
717 | 81.0k | if (bytes != NULL) { |
718 | 40.5k | converted = wcstombs(bytes, buf, size); |
719 | 40.5k | } |
720 | 40.5k | else { |
721 | 40.5k | converted = wcstombs(NULL, buf, 0); |
722 | 40.5k | } |
723 | 81.0k | if (converted == DECODE_ERROR) { |
724 | 0 | goto encode_error; |
725 | 0 | } |
726 | 81.0k | if (bytes != NULL) { |
727 | 40.5k | bytes += converted; |
728 | 40.5k | size -= converted; |
729 | 40.5k | } |
730 | 40.5k | else { |
731 | 40.5k | size += converted; |
732 | 40.5k | } |
733 | 81.0k | } |
734 | 81.0k | } |
735 | 1.13k | if (result != NULL) { |
736 | 568 | *bytes = '\0'; |
737 | 568 | break; |
738 | 568 | } |
739 | | |
740 | 568 | size += 1; /* nul byte at the end */ |
741 | 568 | if (raw_malloc) { |
742 | 568 | result = PyMem_RawMalloc(size); |
743 | 568 | } |
744 | 0 | else { |
745 | 0 | result = PyMem_Malloc(size); |
746 | 0 | } |
747 | 568 | if (result == NULL) { |
748 | 0 | return -1; |
749 | 0 | } |
750 | 568 | bytes = result; |
751 | 568 | } |
752 | 568 | *str = result; |
753 | 568 | return 0; |
754 | | |
755 | 0 | encode_error: |
756 | 0 | if (raw_malloc) { |
757 | 0 | PyMem_RawFree(result); |
758 | 0 | } |
759 | 0 | else { |
760 | 0 | PyMem_Free(result); |
761 | 0 | } |
762 | 0 | if (error_pos != NULL) { |
763 | 0 | *error_pos = i; |
764 | 0 | } |
765 | 0 | if (reason) { |
766 | 0 | *reason = "encoding error"; |
767 | 0 | } |
768 | 0 | return -2; |
769 | 568 | } |
770 | | |
771 | | |
772 | | /* Encode a string to the locale encoding. |
773 | | |
774 | | Parameters: |
775 | | |
776 | | * raw_malloc: if non-zero, allocate memory using PyMem_RawMalloc() instead |
777 | | of PyMem_Malloc(). |
778 | | * current_locale: if non-zero, use the current LC_CTYPE, otherwise use |
779 | | Python filesystem encoding. |
780 | | * errors: error handler like "strict" or "surrogateescape". |
781 | | |
782 | | Return value: |
783 | | |
784 | | 0: success, *str is set to a newly allocated decoded string. |
785 | | -1: memory allocation failure |
786 | | -2: encoding error, set *error_pos and *reason (if set). |
787 | | -3: the error handler 'errors' is not supported. |
788 | | */ |
789 | | static int |
790 | | encode_locale_ex(const wchar_t *text, char **str, size_t *error_pos, |
791 | | const char **reason, |
792 | | int raw_malloc, int current_locale, _Py_error_handler errors) |
793 | 568 | { |
794 | 568 | if (current_locale) { |
795 | | #ifdef _Py_FORCE_UTF8_LOCALE |
796 | | return _Py_EncodeUTF8Ex(text, str, error_pos, reason, |
797 | | raw_malloc, errors); |
798 | | #else |
799 | 0 | return encode_current_locale(text, str, error_pos, reason, |
800 | 0 | raw_malloc, errors); |
801 | 0 | #endif |
802 | 0 | } |
803 | | |
804 | | #ifdef _Py_FORCE_UTF8_FS_ENCODING |
805 | | return _Py_EncodeUTF8Ex(text, str, error_pos, reason, |
806 | | raw_malloc, errors); |
807 | | #else |
808 | 568 | int use_utf8 = (_PyRuntime.preconfig.utf8_mode >= 1); |
809 | | #ifdef MS_WINDOWS |
810 | | use_utf8 |= (_PyRuntime.preconfig.legacy_windows_fs_encoding == 0); |
811 | | #endif |
812 | 568 | if (use_utf8) { |
813 | 0 | return _Py_EncodeUTF8Ex(text, str, error_pos, reason, |
814 | 0 | raw_malloc, errors); |
815 | 0 | } |
816 | | |
817 | 568 | #ifdef USE_FORCE_ASCII |
818 | 568 | if (force_ascii == -1) { |
819 | 0 | force_ascii = check_force_ascii(); |
820 | 0 | } |
821 | | |
822 | 568 | if (force_ascii) { |
823 | 0 | return encode_ascii(text, str, error_pos, reason, |
824 | 0 | raw_malloc, errors); |
825 | 0 | } |
826 | 568 | #endif |
827 | | |
828 | 568 | return encode_current_locale(text, str, error_pos, reason, |
829 | 568 | raw_malloc, errors); |
830 | 568 | #endif /* _Py_FORCE_UTF8_FS_ENCODING */ |
831 | 568 | } |
832 | | |
833 | | static char* |
834 | | encode_locale(const wchar_t *text, size_t *error_pos, |
835 | | int raw_malloc, int current_locale) |
836 | 160 | { |
837 | 160 | char *str; |
838 | 160 | int res = encode_locale_ex(text, &str, error_pos, NULL, |
839 | 160 | raw_malloc, current_locale, |
840 | 160 | _Py_ERROR_SURROGATEESCAPE); |
841 | 160 | if (res != -2 && error_pos) { |
842 | 0 | *error_pos = (size_t)-1; |
843 | 0 | } |
844 | 160 | if (res != 0) { |
845 | 0 | return NULL; |
846 | 0 | } |
847 | 160 | return str; |
848 | 160 | } |
849 | | |
850 | | /* Encode a wide character string to the locale encoding with the |
851 | | surrogateescape error handler: surrogate characters in the range |
852 | | U+DC80..U+DCFF are converted to bytes 0x80..0xFF. |
853 | | |
854 | | Return a pointer to a newly allocated byte string, use PyMem_Free() to free |
855 | | the memory. Return NULL on encoding or memory allocation error. |
856 | | |
857 | | If error_pos is not NULL, *error_pos is set to (size_t)-1 on success, or set |
858 | | to the index of the invalid character on encoding error. |
859 | | |
860 | | Use the Py_DecodeLocale() function to decode the bytes string back to a wide |
861 | | character string. */ |
862 | | char* |
863 | | Py_EncodeLocale(const wchar_t *text, size_t *error_pos) |
864 | 0 | { |
865 | 0 | return encode_locale(text, error_pos, 0, 0); |
866 | 0 | } |
867 | | |
868 | | |
869 | | /* Similar to Py_EncodeLocale(), but result must be freed by PyMem_RawFree() |
870 | | instead of PyMem_Free(). */ |
871 | | char* |
872 | | _Py_EncodeLocaleRaw(const wchar_t *text, size_t *error_pos) |
873 | 160 | { |
874 | 160 | return encode_locale(text, error_pos, 1, 0); |
875 | 160 | } |
876 | | |
877 | | |
878 | | int |
879 | | _Py_EncodeLocaleEx(const wchar_t *text, char **str, |
880 | | size_t *error_pos, const char **reason, |
881 | | int current_locale, _Py_error_handler errors) |
882 | 408 | { |
883 | 408 | return encode_locale_ex(text, str, error_pos, reason, 1, |
884 | 408 | current_locale, errors); |
885 | 408 | } |
886 | | |
887 | | |
888 | | // Get the current locale encoding name: |
889 | | // |
890 | | // - Return "utf-8" if _Py_FORCE_UTF8_LOCALE macro is defined (ex: on Android) |
891 | | // - Return "utf-8" if the UTF-8 Mode is enabled |
892 | | // - On Windows, return the ANSI code page (ex: "cp1250") |
893 | | // - Return "utf-8" if nl_langinfo(CODESET) returns an empty string. |
894 | | // - Otherwise, return nl_langinfo(CODESET). |
895 | | // |
896 | | // Return NULL on memory allocation failure. |
897 | | // |
898 | | // See also config_get_locale_encoding() |
899 | | wchar_t* |
900 | | _Py_GetLocaleEncoding(void) |
901 | 32 | { |
902 | | #ifdef _Py_FORCE_UTF8_LOCALE |
903 | | // On Android langinfo.h and CODESET are missing, |
904 | | // and UTF-8 is always used in mbstowcs() and wcstombs(). |
905 | | return _PyMem_RawWcsdup(L"utf-8"); |
906 | | #else |
907 | | |
908 | | #ifdef MS_WINDOWS |
909 | | wchar_t encoding[23]; |
910 | | unsigned int ansi_codepage = GetACP(); |
911 | | swprintf(encoding, Py_ARRAY_LENGTH(encoding), L"cp%u", ansi_codepage); |
912 | | encoding[Py_ARRAY_LENGTH(encoding) - 1] = 0; |
913 | | return _PyMem_RawWcsdup(encoding); |
914 | | #else |
915 | 32 | const char *encoding = nl_langinfo(CODESET); |
916 | 32 | if (!encoding || encoding[0] == '\0') { |
917 | | // Use UTF-8 if nl_langinfo() returns an empty string. It can happen on |
918 | | // macOS if the LC_CTYPE locale is not supported. |
919 | 0 | return _PyMem_RawWcsdup(L"utf-8"); |
920 | 0 | } |
921 | | |
922 | 32 | wchar_t *wstr; |
923 | 32 | int res = decode_current_locale(encoding, &wstr, NULL, |
924 | 32 | NULL, _Py_ERROR_SURROGATEESCAPE); |
925 | 32 | if (res < 0) { |
926 | 0 | return NULL; |
927 | 0 | } |
928 | 32 | return wstr; |
929 | 32 | #endif // !MS_WINDOWS |
930 | | |
931 | 32 | #endif // !_Py_FORCE_UTF8_LOCALE |
932 | 32 | } |
933 | | |
934 | | |
935 | | PyObject * |
936 | | _Py_GetLocaleEncodingObject(void) |
937 | 0 | { |
938 | 0 | wchar_t *encoding = _Py_GetLocaleEncoding(); |
939 | 0 | if (encoding == NULL) { |
940 | 0 | PyErr_NoMemory(); |
941 | 0 | return NULL; |
942 | 0 | } |
943 | | |
944 | 0 | PyObject *str = PyUnicode_FromWideChar(encoding, -1); |
945 | 0 | PyMem_RawFree(encoding); |
946 | 0 | return str; |
947 | 0 | } |
948 | | |
949 | | #ifdef HAVE_NON_UNICODE_WCHAR_T_REPRESENTATION |
950 | | |
951 | | /* Check whether current locale uses Unicode as internal wchar_t form. */ |
952 | | int |
953 | | _Py_LocaleUsesNonUnicodeWchar(void) |
954 | | { |
955 | | /* Oracle Solaris uses non-Unicode internal wchar_t form for |
956 | | non-Unicode locales and hence needs conversion to UTF first. */ |
957 | | char* codeset = nl_langinfo(CODESET); |
958 | | if (!codeset) { |
959 | | return 0; |
960 | | } |
961 | | /* 646 refers to ISO/IEC 646 standard that corresponds to ASCII encoding */ |
962 | | return (strcmp(codeset, "UTF-8") != 0 && strcmp(codeset, "646") != 0); |
963 | | } |
964 | | |
965 | | static wchar_t * |
966 | | _Py_ConvertWCharForm(const wchar_t *source, Py_ssize_t size, |
967 | | const char *tocode, const char *fromcode) |
968 | | { |
969 | | static_assert(sizeof(wchar_t) == 4, "wchar_t must be 32-bit"); |
970 | | |
971 | | /* Ensure we won't overflow the size. */ |
972 | | if (size > (PY_SSIZE_T_MAX / (Py_ssize_t)sizeof(wchar_t))) { |
973 | | PyErr_NoMemory(); |
974 | | return NULL; |
975 | | } |
976 | | |
977 | | /* the string doesn't have to be NULL terminated */ |
978 | | wchar_t* target = PyMem_Malloc(size * sizeof(wchar_t)); |
979 | | if (target == NULL) { |
980 | | PyErr_NoMemory(); |
981 | | return NULL; |
982 | | } |
983 | | |
984 | | iconv_t cd = iconv_open(tocode, fromcode); |
985 | | if (cd == (iconv_t)-1) { |
986 | | PyErr_Format(PyExc_ValueError, "iconv_open() failed"); |
987 | | PyMem_Free(target); |
988 | | return NULL; |
989 | | } |
990 | | |
991 | | char *inbuf = (char *) source; |
992 | | char *outbuf = (char *) target; |
993 | | size_t inbytesleft = sizeof(wchar_t) * size; |
994 | | size_t outbytesleft = inbytesleft; |
995 | | |
996 | | size_t ret = iconv(cd, &inbuf, &inbytesleft, &outbuf, &outbytesleft); |
997 | | if (ret == DECODE_ERROR) { |
998 | | PyErr_Format(PyExc_ValueError, "iconv() failed"); |
999 | | PyMem_Free(target); |
1000 | | iconv_close(cd); |
1001 | | return NULL; |
1002 | | } |
1003 | | |
1004 | | iconv_close(cd); |
1005 | | return target; |
1006 | | } |
1007 | | |
1008 | | /* Convert a wide character string to the UCS-4 encoded string. This |
1009 | | is necessary on systems where internal form of wchar_t are not Unicode |
1010 | | code points (e.g. Oracle Solaris). |
1011 | | |
1012 | | Return a pointer to a newly allocated string, use PyMem_Free() to free |
1013 | | the memory. Return NULL and raise exception on conversion or memory |
1014 | | allocation error. */ |
1015 | | wchar_t * |
1016 | | _Py_DecodeNonUnicodeWchar(const wchar_t *native, Py_ssize_t size) |
1017 | | { |
1018 | | return _Py_ConvertWCharForm(native, size, "UCS-4-INTERNAL", "wchar_t"); |
1019 | | } |
1020 | | |
1021 | | /* Convert a UCS-4 encoded string to native wide character string. This |
1022 | | is necessary on systems where internal form of wchar_t are not Unicode |
1023 | | code points (e.g. Oracle Solaris). |
1024 | | |
1025 | | The conversion is done in place. This can be done because both wchar_t |
1026 | | and UCS-4 use 4-byte encoding, and one wchar_t symbol always correspond |
1027 | | to a single UCS-4 symbol and vice versa. (This is true for Oracle Solaris, |
1028 | | which is currently the only system using these functions; it doesn't have |
1029 | | to be for other systems). |
1030 | | |
1031 | | Return 0 on success. Return -1 and raise exception on conversion |
1032 | | or memory allocation error. */ |
1033 | | int |
1034 | | _Py_EncodeNonUnicodeWchar_InPlace(wchar_t *unicode, Py_ssize_t size) |
1035 | | { |
1036 | | wchar_t* result = _Py_ConvertWCharForm(unicode, size, "wchar_t", "UCS-4-INTERNAL"); |
1037 | | if (!result) { |
1038 | | return -1; |
1039 | | } |
1040 | | memcpy(unicode, result, size * sizeof(wchar_t)); |
1041 | | PyMem_Free(result); |
1042 | | return 0; |
1043 | | } |
1044 | | #endif /* HAVE_NON_UNICODE_WCHAR_T_REPRESENTATION */ |
1045 | | |
1046 | | #ifdef MS_WINDOWS |
1047 | | static __int64 secs_between_epochs = 11644473600; /* Seconds between 1.1.1601 and 1.1.1970 */ |
1048 | | |
1049 | | static void |
1050 | | FILE_TIME_to_time_t_nsec(FILETIME *in_ptr, time_t *time_out, int* nsec_out) |
1051 | | { |
1052 | | /* XXX endianness. Shouldn't matter, as all Windows implementations are little-endian */ |
1053 | | /* Cannot simply cast and dereference in_ptr, |
1054 | | since it might not be aligned properly */ |
1055 | | __int64 in; |
1056 | | memcpy(&in, in_ptr, sizeof(in)); |
1057 | | *nsec_out = (int)(in % 10000000) * 100; /* FILETIME is in units of 100 nsec. */ |
1058 | | *time_out = Py_SAFE_DOWNCAST((in / 10000000) - secs_between_epochs, __int64, time_t); |
1059 | | } |
1060 | | |
1061 | | static void |
1062 | | LARGE_INTEGER_to_time_t_nsec(LARGE_INTEGER *in_ptr, time_t *time_out, int* nsec_out) |
1063 | | { |
1064 | | *nsec_out = (int)(in_ptr->QuadPart % 10000000) * 100; /* FILETIME is in units of 100 nsec. */ |
1065 | | *time_out = Py_SAFE_DOWNCAST((in_ptr->QuadPart / 10000000) - secs_between_epochs, __int64, time_t); |
1066 | | } |
1067 | | |
1068 | | void |
1069 | | _Py_time_t_to_FILE_TIME(time_t time_in, int nsec_in, FILETIME *out_ptr) |
1070 | | { |
1071 | | /* XXX endianness */ |
1072 | | __int64 out; |
1073 | | out = time_in + secs_between_epochs; |
1074 | | out = out * 10000000 + nsec_in / 100; |
1075 | | memcpy(out_ptr, &out, sizeof(out)); |
1076 | | } |
1077 | | |
1078 | | /* Below, we *know* that ugo+r is 0444 */ |
1079 | | #if _S_IREAD != 0400 |
1080 | | #error Unsupported C library |
1081 | | #endif |
1082 | | static int |
1083 | | attributes_to_mode(DWORD attr) |
1084 | | { |
1085 | | int m = 0; |
1086 | | if (attr & FILE_ATTRIBUTE_DIRECTORY) |
1087 | | m |= _S_IFDIR | 0111; /* IFEXEC for user,group,other */ |
1088 | | else |
1089 | | m |= _S_IFREG; |
1090 | | if (attr & FILE_ATTRIBUTE_READONLY) |
1091 | | m |= 0444; |
1092 | | else |
1093 | | m |= 0666; |
1094 | | return m; |
1095 | | } |
1096 | | |
1097 | | |
1098 | | typedef union { |
1099 | | FILE_ID_128 id; |
1100 | | struct { |
1101 | | uint64_t st_ino; |
1102 | | uint64_t st_ino_high; |
1103 | | }; |
1104 | | } id_128_to_ino; |
1105 | | |
1106 | | |
1107 | | void |
1108 | | _Py_attribute_data_to_stat(BY_HANDLE_FILE_INFORMATION *info, ULONG reparse_tag, |
1109 | | FILE_BASIC_INFO *basic_info, FILE_ID_INFO *id_info, |
1110 | | struct _Py_stat_struct *result) |
1111 | | { |
1112 | | memset(result, 0, sizeof(*result)); |
1113 | | result->st_mode = attributes_to_mode(info->dwFileAttributes); |
1114 | | result->st_size = (((__int64)info->nFileSizeHigh)<<32) + info->nFileSizeLow; |
1115 | | result->st_dev = id_info ? id_info->VolumeSerialNumber : info->dwVolumeSerialNumber; |
1116 | | result->st_rdev = 0; |
1117 | | /* st_ctime is deprecated, but we preserve the legacy value in our caller, not here */ |
1118 | | if (basic_info) { |
1119 | | LARGE_INTEGER_to_time_t_nsec(&basic_info->CreationTime, &result->st_birthtime, &result->st_birthtime_nsec); |
1120 | | LARGE_INTEGER_to_time_t_nsec(&basic_info->ChangeTime, &result->st_ctime, &result->st_ctime_nsec); |
1121 | | LARGE_INTEGER_to_time_t_nsec(&basic_info->LastWriteTime, &result->st_mtime, &result->st_mtime_nsec); |
1122 | | LARGE_INTEGER_to_time_t_nsec(&basic_info->LastAccessTime, &result->st_atime, &result->st_atime_nsec); |
1123 | | } else { |
1124 | | FILE_TIME_to_time_t_nsec(&info->ftCreationTime, &result->st_birthtime, &result->st_birthtime_nsec); |
1125 | | FILE_TIME_to_time_t_nsec(&info->ftLastWriteTime, &result->st_mtime, &result->st_mtime_nsec); |
1126 | | FILE_TIME_to_time_t_nsec(&info->ftLastAccessTime, &result->st_atime, &result->st_atime_nsec); |
1127 | | } |
1128 | | result->st_nlink = info->nNumberOfLinks; |
1129 | | |
1130 | | if (id_info) { |
1131 | | id_128_to_ino file_id; |
1132 | | file_id.id = id_info->FileId; |
1133 | | result->st_ino = file_id.st_ino; |
1134 | | result->st_ino_high = file_id.st_ino_high; |
1135 | | } |
1136 | | if (!result->st_ino && !result->st_ino_high) { |
1137 | | /* should only occur for DirEntry_from_find_data, in which case the |
1138 | | index is likely to be zero anyway. */ |
1139 | | result->st_ino = (((uint64_t)info->nFileIndexHigh) << 32) + info->nFileIndexLow; |
1140 | | } |
1141 | | |
1142 | | /* bpo-37834: Only actual symlinks set the S_IFLNK flag. But lstat() will |
1143 | | open other name surrogate reparse points without traversing them. To |
1144 | | detect/handle these, check st_file_attributes and st_reparse_tag. */ |
1145 | | result->st_reparse_tag = reparse_tag; |
1146 | | if (info->dwFileAttributes & FILE_ATTRIBUTE_REPARSE_POINT && |
1147 | | reparse_tag == IO_REPARSE_TAG_SYMLINK) { |
1148 | | /* set the bits that make this a symlink */ |
1149 | | result->st_mode = (result->st_mode & ~S_IFMT) | S_IFLNK; |
1150 | | } |
1151 | | result->st_file_attributes = info->dwFileAttributes; |
1152 | | } |
1153 | | |
1154 | | void |
1155 | | _Py_stat_basic_info_to_stat(FILE_STAT_BASIC_INFORMATION *info, |
1156 | | struct _Py_stat_struct *result) |
1157 | | { |
1158 | | memset(result, 0, sizeof(*result)); |
1159 | | result->st_mode = attributes_to_mode(info->FileAttributes); |
1160 | | result->st_size = info->EndOfFile.QuadPart; |
1161 | | LARGE_INTEGER_to_time_t_nsec(&info->CreationTime, &result->st_birthtime, &result->st_birthtime_nsec); |
1162 | | LARGE_INTEGER_to_time_t_nsec(&info->ChangeTime, &result->st_ctime, &result->st_ctime_nsec); |
1163 | | LARGE_INTEGER_to_time_t_nsec(&info->LastWriteTime, &result->st_mtime, &result->st_mtime_nsec); |
1164 | | LARGE_INTEGER_to_time_t_nsec(&info->LastAccessTime, &result->st_atime, &result->st_atime_nsec); |
1165 | | result->st_nlink = info->NumberOfLinks; |
1166 | | result->st_dev = info->VolumeSerialNumber.QuadPart; |
1167 | | /* File systems with less than 128-bits zero pad into this field */ |
1168 | | id_128_to_ino file_id; |
1169 | | file_id.id = info->FileId128; |
1170 | | result->st_ino = file_id.st_ino; |
1171 | | result->st_ino_high = file_id.st_ino_high; |
1172 | | /* bpo-37834: Only actual symlinks set the S_IFLNK flag. But lstat() will |
1173 | | open other name surrogate reparse points without traversing them. To |
1174 | | detect/handle these, check st_file_attributes and st_reparse_tag. */ |
1175 | | result->st_reparse_tag = info->ReparseTag; |
1176 | | if (info->FileAttributes & FILE_ATTRIBUTE_REPARSE_POINT && |
1177 | | info->ReparseTag == IO_REPARSE_TAG_SYMLINK) { |
1178 | | /* set the bits that make this a symlink */ |
1179 | | result->st_mode = (result->st_mode & ~S_IFMT) | S_IFLNK; |
1180 | | } |
1181 | | result->st_file_attributes = info->FileAttributes; |
1182 | | switch (info->DeviceType) { |
1183 | | case FILE_DEVICE_DISK: |
1184 | | case FILE_DEVICE_VIRTUAL_DISK: |
1185 | | case FILE_DEVICE_DFS: |
1186 | | case FILE_DEVICE_CD_ROM: |
1187 | | case FILE_DEVICE_CONTROLLER: |
1188 | | case FILE_DEVICE_DATALINK: |
1189 | | break; |
1190 | | case FILE_DEVICE_DISK_FILE_SYSTEM: |
1191 | | case FILE_DEVICE_CD_ROM_FILE_SYSTEM: |
1192 | | case FILE_DEVICE_NETWORK_FILE_SYSTEM: |
1193 | | result->st_mode = (result->st_mode & ~S_IFMT) | 0x6000; /* _S_IFBLK */ |
1194 | | break; |
1195 | | case FILE_DEVICE_CONSOLE: |
1196 | | case FILE_DEVICE_NULL: |
1197 | | case FILE_DEVICE_KEYBOARD: |
1198 | | case FILE_DEVICE_MODEM: |
1199 | | case FILE_DEVICE_MOUSE: |
1200 | | case FILE_DEVICE_PARALLEL_PORT: |
1201 | | case FILE_DEVICE_PRINTER: |
1202 | | case FILE_DEVICE_SCREEN: |
1203 | | case FILE_DEVICE_SERIAL_PORT: |
1204 | | case FILE_DEVICE_SOUND: |
1205 | | result->st_mode = (result->st_mode & ~S_IFMT) | _S_IFCHR; |
1206 | | break; |
1207 | | case FILE_DEVICE_NAMED_PIPE: |
1208 | | result->st_mode = (result->st_mode & ~S_IFMT) | _S_IFIFO; |
1209 | | break; |
1210 | | default: |
1211 | | if (info->FileAttributes & FILE_ATTRIBUTE_DIRECTORY) { |
1212 | | result->st_mode = (result->st_mode & ~S_IFMT) | _S_IFDIR; |
1213 | | } |
1214 | | break; |
1215 | | } |
1216 | | } |
1217 | | |
1218 | | #endif |
1219 | | |
1220 | | /* Return information about a file. |
1221 | | |
1222 | | On POSIX, use fstat(). |
1223 | | |
1224 | | On Windows, use GetFileType() and GetFileInformationByHandle() which support |
1225 | | files larger than 2 GiB. fstat() may fail with EOVERFLOW on files larger |
1226 | | than 2 GiB because the file size type is a signed 32-bit integer: see issue |
1227 | | #23152. |
1228 | | |
1229 | | On Windows, set the last Windows error and return nonzero on error. On |
1230 | | POSIX, set errno and return nonzero on error. Fill status and return 0 on |
1231 | | success. */ |
1232 | | int |
1233 | | _Py_fstat_noraise(int fd, struct _Py_stat_struct *status) |
1234 | 1.22k | { |
1235 | | #ifdef MS_WINDOWS |
1236 | | BY_HANDLE_FILE_INFORMATION info; |
1237 | | FILE_BASIC_INFO basicInfo; |
1238 | | FILE_ID_INFO idInfo; |
1239 | | FILE_ID_INFO *pIdInfo = &idInfo; |
1240 | | HANDLE h; |
1241 | | int type; |
1242 | | |
1243 | | h = _Py_get_osfhandle_noraise(fd); |
1244 | | |
1245 | | if (h == INVALID_HANDLE_VALUE) { |
1246 | | /* errno is already set by _get_osfhandle, but we also set |
1247 | | the Win32 error for callers who expect that */ |
1248 | | SetLastError(ERROR_INVALID_HANDLE); |
1249 | | return -1; |
1250 | | } |
1251 | | memset(status, 0, sizeof(*status)); |
1252 | | |
1253 | | type = GetFileType(h); |
1254 | | if (type == FILE_TYPE_UNKNOWN) { |
1255 | | DWORD error = GetLastError(); |
1256 | | if (error != 0) { |
1257 | | errno = winerror_to_errno(error); |
1258 | | return -1; |
1259 | | } |
1260 | | /* else: valid but unknown file */ |
1261 | | } |
1262 | | |
1263 | | if (type != FILE_TYPE_DISK) { |
1264 | | if (type == FILE_TYPE_CHAR) |
1265 | | status->st_mode = _S_IFCHR; |
1266 | | else if (type == FILE_TYPE_PIPE) |
1267 | | status->st_mode = _S_IFIFO; |
1268 | | return 0; |
1269 | | } |
1270 | | |
1271 | | if (!GetFileInformationByHandle(h, &info) || |
1272 | | !GetFileInformationByHandleEx(h, FileBasicInfo, &basicInfo, sizeof(basicInfo))) { |
1273 | | /* The Win32 error is already set, but we also set errno for |
1274 | | callers who expect it */ |
1275 | | errno = winerror_to_errno(GetLastError()); |
1276 | | return -1; |
1277 | | } |
1278 | | |
1279 | | if (!GetFileInformationByHandleEx(h, FileIdInfo, &idInfo, sizeof(idInfo))) { |
1280 | | /* Failed to get FileIdInfo, so do not pass it along */ |
1281 | | pIdInfo = NULL; |
1282 | | } |
1283 | | |
1284 | | _Py_attribute_data_to_stat(&info, 0, &basicInfo, pIdInfo, status); |
1285 | | return 0; |
1286 | | #else |
1287 | 1.22k | return fstat(fd, status); |
1288 | 1.22k | #endif |
1289 | 1.22k | } |
1290 | | |
1291 | | /* Return information about a file. |
1292 | | |
1293 | | On POSIX, use fstat(). |
1294 | | |
1295 | | On Windows, use GetFileType() and GetFileInformationByHandle() which support |
1296 | | files larger than 2 GiB. fstat() may fail with EOVERFLOW on files larger |
1297 | | than 2 GiB because the file size type is a signed 32-bit integer: see issue |
1298 | | #23152. |
1299 | | |
1300 | | Raise an exception and return -1 on error. On Windows, set the last Windows |
1301 | | error on error. On POSIX, set errno on error. Fill status and return 0 on |
1302 | | success. |
1303 | | |
1304 | | Release the GIL to call GetFileType() and GetFileInformationByHandle(), or |
1305 | | to call fstat(). The caller must hold the GIL. */ |
1306 | | int |
1307 | | _Py_fstat(int fd, struct _Py_stat_struct *status) |
1308 | 0 | { |
1309 | 0 | int res; |
1310 | |
|
1311 | 0 | _Py_AssertHoldsTstate(); |
1312 | |
|
1313 | 0 | Py_BEGIN_ALLOW_THREADS |
1314 | 0 | res = _Py_fstat_noraise(fd, status); |
1315 | 0 | Py_END_ALLOW_THREADS |
1316 | |
|
1317 | 0 | if (res != 0) { |
1318 | | #ifdef MS_WINDOWS |
1319 | | PyErr_SetFromWindowsErr(0); |
1320 | | #else |
1321 | 0 | PyErr_SetFromErrno(PyExc_OSError); |
1322 | 0 | #endif |
1323 | 0 | return -1; |
1324 | 0 | } |
1325 | 0 | return 0; |
1326 | 0 | } |
1327 | | |
1328 | | /* Like _Py_stat() but with a raw filename. */ |
1329 | | int |
1330 | | _Py_wstat(const wchar_t* path, struct stat *buf) |
1331 | 48 | { |
1332 | 48 | int err; |
1333 | | #ifdef MS_WINDOWS |
1334 | | struct _stat wstatbuf; |
1335 | | err = _wstat(path, &wstatbuf); |
1336 | | if (!err) { |
1337 | | buf->st_mode = wstatbuf.st_mode; |
1338 | | } |
1339 | | #else |
1340 | 48 | char *fname; |
1341 | 48 | fname = _Py_EncodeLocaleRaw(path, NULL); |
1342 | 48 | if (fname == NULL) { |
1343 | 0 | errno = EINVAL; |
1344 | 0 | return -1; |
1345 | 0 | } |
1346 | 48 | err = stat(fname, buf); |
1347 | 48 | PyMem_RawFree(fname); |
1348 | 48 | #endif |
1349 | 48 | return err; |
1350 | 48 | } |
1351 | | |
1352 | | |
1353 | | /* Call _wstat() on Windows, or encode the path to the filesystem encoding and |
1354 | | call stat() otherwise. Only fill st_mode attribute on Windows. |
1355 | | |
1356 | | Return 0 on success, -1 on _wstat() / stat() error, -2 if an exception was |
1357 | | raised. */ |
1358 | | |
1359 | | int |
1360 | | _Py_stat(PyObject *path, struct stat *statbuf) |
1361 | 0 | { |
1362 | | #ifdef MS_WINDOWS |
1363 | | int err; |
1364 | | |
1365 | | wchar_t *wpath = PyUnicode_AsWideCharString(path, NULL); |
1366 | | if (wpath == NULL) |
1367 | | return -2; |
1368 | | |
1369 | | err = _Py_wstat(wpath, statbuf); |
1370 | | PyMem_Free(wpath); |
1371 | | return err; |
1372 | | #else |
1373 | 0 | int ret; |
1374 | 0 | PyObject *bytes; |
1375 | 0 | char *cpath; |
1376 | |
|
1377 | 0 | bytes = PyUnicode_EncodeFSDefault(path); |
1378 | 0 | if (bytes == NULL) |
1379 | 0 | return -2; |
1380 | | |
1381 | | /* check for embedded null bytes */ |
1382 | 0 | if (PyBytes_AsStringAndSize(bytes, &cpath, NULL) == -1) { |
1383 | 0 | Py_DECREF(bytes); |
1384 | 0 | return -2; |
1385 | 0 | } |
1386 | | |
1387 | 0 | ret = stat(cpath, statbuf); |
1388 | 0 | Py_DECREF(bytes); |
1389 | 0 | return ret; |
1390 | 0 | #endif |
1391 | 0 | } |
1392 | | |
1393 | | #ifdef MS_WINDOWS |
1394 | | // For some Windows API partitions, SetHandleInformation() is declared |
1395 | | // but none of the handle flags are defined. |
1396 | | #ifndef HANDLE_FLAG_INHERIT |
1397 | | #define HANDLE_FLAG_INHERIT 0x00000001 |
1398 | | #endif |
1399 | | #endif |
1400 | | |
1401 | | /* This function MUST be kept async-signal-safe on POSIX when raise=0. */ |
1402 | | static int |
1403 | | get_inheritable(int fd, int raise) |
1404 | 16 | { |
1405 | | #ifdef MS_WINDOWS |
1406 | | HANDLE handle; |
1407 | | DWORD flags; |
1408 | | |
1409 | | handle = _Py_get_osfhandle_noraise(fd); |
1410 | | if (handle == INVALID_HANDLE_VALUE) { |
1411 | | if (raise) |
1412 | | PyErr_SetFromErrno(PyExc_OSError); |
1413 | | return -1; |
1414 | | } |
1415 | | |
1416 | | if (!GetHandleInformation(handle, &flags)) { |
1417 | | if (raise) |
1418 | | PyErr_SetFromWindowsErr(0); |
1419 | | return -1; |
1420 | | } |
1421 | | |
1422 | | return (flags & HANDLE_FLAG_INHERIT); |
1423 | | #else |
1424 | 16 | int flags; |
1425 | | |
1426 | 16 | flags = fcntl(fd, F_GETFD, 0); |
1427 | 16 | if (flags == -1) { |
1428 | 0 | if (raise) |
1429 | 0 | PyErr_SetFromErrno(PyExc_OSError); |
1430 | 0 | return -1; |
1431 | 0 | } |
1432 | 16 | return !(flags & FD_CLOEXEC); |
1433 | 16 | #endif |
1434 | 16 | } |
1435 | | |
1436 | | /* Get the inheritable flag of the specified file descriptor. |
1437 | | Return 1 if the file descriptor can be inherited, 0 if it cannot, |
1438 | | raise an exception and return -1 on error. */ |
1439 | | int |
1440 | | _Py_get_inheritable(int fd) |
1441 | 0 | { |
1442 | 0 | return get_inheritable(fd, 1); |
1443 | 0 | } |
1444 | | |
1445 | | |
1446 | | /* This function MUST be kept async-signal-safe on POSIX when raise=0. */ |
1447 | | static int |
1448 | | set_inheritable(int fd, int inheritable, int raise, int *atomic_flag_works) |
1449 | 1.16k | { |
1450 | | #ifdef MS_WINDOWS |
1451 | | HANDLE handle; |
1452 | | DWORD flags; |
1453 | | #else |
1454 | 1.16k | #if defined(HAVE_SYS_IOCTL_H) && defined(FIOCLEX) && defined(FIONCLEX) |
1455 | 1.16k | static int ioctl_works = -1; |
1456 | 1.16k | int request; |
1457 | 1.16k | int err; |
1458 | 1.16k | #endif |
1459 | 1.16k | int flags, new_flags; |
1460 | 1.16k | int res; |
1461 | 1.16k | #endif |
1462 | | |
1463 | | /* atomic_flag_works can only be used to make the file descriptor |
1464 | | non-inheritable */ |
1465 | 1.16k | assert(!(atomic_flag_works != NULL && inheritable)); |
1466 | | |
1467 | 1.16k | if (atomic_flag_works != NULL && !inheritable) { |
1468 | 1.16k | if (_Py_atomic_load_int_relaxed(atomic_flag_works) == -1) { |
1469 | 16 | int isInheritable = get_inheritable(fd, raise); |
1470 | 16 | if (isInheritable == -1) |
1471 | 0 | return -1; |
1472 | 16 | _Py_atomic_store_int_relaxed(atomic_flag_works, !isInheritable); |
1473 | 16 | } |
1474 | | |
1475 | 1.16k | if (_Py_atomic_load_int_relaxed(atomic_flag_works)) |
1476 | 1.16k | return 0; |
1477 | 1.16k | } |
1478 | | |
1479 | | #ifdef MS_WINDOWS |
1480 | | handle = _Py_get_osfhandle_noraise(fd); |
1481 | | if (handle == INVALID_HANDLE_VALUE) { |
1482 | | if (raise) |
1483 | | PyErr_SetFromErrno(PyExc_OSError); |
1484 | | return -1; |
1485 | | } |
1486 | | |
1487 | | if (inheritable) |
1488 | | flags = HANDLE_FLAG_INHERIT; |
1489 | | else |
1490 | | flags = 0; |
1491 | | |
1492 | | if (!SetHandleInformation(handle, HANDLE_FLAG_INHERIT, flags)) { |
1493 | | if (raise) |
1494 | | PyErr_SetFromWindowsErr(0); |
1495 | | return -1; |
1496 | | } |
1497 | | return 0; |
1498 | | |
1499 | | #else |
1500 | | |
1501 | 0 | #if defined(HAVE_SYS_IOCTL_H) && defined(FIOCLEX) && defined(FIONCLEX) |
1502 | 0 | if (raise != 0 && _Py_atomic_load_int_relaxed(&ioctl_works) != 0) { |
1503 | | /* fast-path: ioctl() only requires one syscall */ |
1504 | | /* caveat: raise=0 is an indicator that we must be async-signal-safe |
1505 | | * thus avoid using ioctl() so we skip the fast-path. */ |
1506 | 0 | if (inheritable) |
1507 | 0 | request = FIONCLEX; |
1508 | 0 | else |
1509 | 0 | request = FIOCLEX; |
1510 | 0 | err = ioctl(fd, request, NULL); |
1511 | 0 | if (!err) { |
1512 | 0 | if (_Py_atomic_load_int_relaxed(&ioctl_works) == -1) { |
1513 | 0 | _Py_atomic_store_int_relaxed(&ioctl_works, 1); |
1514 | 0 | } |
1515 | 0 | return 0; |
1516 | 0 | } |
1517 | | |
1518 | 0 | #ifdef O_PATH |
1519 | 0 | if (errno == EBADF) { |
1520 | | // bpo-44849: On Linux and FreeBSD, ioctl(FIOCLEX) fails with EBADF |
1521 | | // on O_PATH file descriptors. Fall through to the fcntl() |
1522 | | // implementation. |
1523 | 0 | } |
1524 | 0 | else |
1525 | 0 | #endif |
1526 | 0 | if (errno != ENOTTY && errno != EACCES) { |
1527 | 0 | if (raise) |
1528 | 0 | PyErr_SetFromErrno(PyExc_OSError); |
1529 | 0 | return -1; |
1530 | 0 | } |
1531 | 0 | else { |
1532 | | /* Issue #22258: Here, ENOTTY means "Inappropriate ioctl for |
1533 | | device". The ioctl is declared but not supported by the kernel. |
1534 | | Remember that ioctl() doesn't work. It is the case on |
1535 | | Illumos-based OS for example. |
1536 | | |
1537 | | Issue #27057: When SELinux policy disallows ioctl it will fail |
1538 | | with EACCES. While FIOCLEX is safe operation it may be |
1539 | | unavailable because ioctl was denied altogether. |
1540 | | This can be the case on Android. */ |
1541 | 0 | _Py_atomic_store_int_relaxed(&ioctl_works, 0); |
1542 | 0 | } |
1543 | | /* fallback to fcntl() if ioctl() does not work */ |
1544 | 0 | } |
1545 | 0 | #endif |
1546 | | |
1547 | | /* slow-path: fcntl() requires two syscalls */ |
1548 | 0 | flags = fcntl(fd, F_GETFD); |
1549 | 0 | if (flags < 0) { |
1550 | 0 | if (raise) |
1551 | 0 | PyErr_SetFromErrno(PyExc_OSError); |
1552 | 0 | return -1; |
1553 | 0 | } |
1554 | | |
1555 | 0 | if (inheritable) { |
1556 | 0 | new_flags = flags & ~FD_CLOEXEC; |
1557 | 0 | } |
1558 | 0 | else { |
1559 | 0 | new_flags = flags | FD_CLOEXEC; |
1560 | 0 | } |
1561 | |
|
1562 | 0 | if (new_flags == flags) { |
1563 | | /* FD_CLOEXEC flag already set/cleared: nothing to do */ |
1564 | 0 | return 0; |
1565 | 0 | } |
1566 | | |
1567 | 0 | res = fcntl(fd, F_SETFD, new_flags); |
1568 | 0 | if (res < 0) { |
1569 | 0 | if (raise) |
1570 | 0 | PyErr_SetFromErrno(PyExc_OSError); |
1571 | 0 | return -1; |
1572 | 0 | } |
1573 | 0 | return 0; |
1574 | 0 | #endif |
1575 | 0 | } |
1576 | | |
1577 | | /* Make the file descriptor non-inheritable. |
1578 | | Return 0 on success, set errno and return -1 on error. */ |
1579 | | static int |
1580 | | make_non_inheritable(int fd) |
1581 | 0 | { |
1582 | 0 | return set_inheritable(fd, 0, 0, NULL); |
1583 | 0 | } |
1584 | | |
1585 | | /* Set the inheritable flag of the specified file descriptor. |
1586 | | On success: return 0, on error: raise an exception and return -1. |
1587 | | |
1588 | | If atomic_flag_works is not NULL: |
1589 | | |
1590 | | * if *atomic_flag_works==-1, check if the inheritable is set on the file |
1591 | | descriptor: if yes, set *atomic_flag_works to 1, otherwise set to 0 and |
1592 | | set the inheritable flag |
1593 | | * if *atomic_flag_works==1: do nothing |
1594 | | * if *atomic_flag_works==0: set inheritable flag to False |
1595 | | |
1596 | | Set atomic_flag_works to NULL if no atomic flag was used to create the |
1597 | | file descriptor. |
1598 | | |
1599 | | atomic_flag_works can only be used to make a file descriptor |
1600 | | non-inheritable: atomic_flag_works must be NULL if inheritable=1. */ |
1601 | | int |
1602 | | _Py_set_inheritable(int fd, int inheritable, int *atomic_flag_works) |
1603 | 1.16k | { |
1604 | 1.16k | return set_inheritable(fd, inheritable, 1, atomic_flag_works); |
1605 | 1.16k | } |
1606 | | |
1607 | | /* Same as _Py_set_inheritable() but on error, set errno and |
1608 | | don't raise an exception. |
1609 | | This function is async-signal-safe. */ |
1610 | | int |
1611 | | _Py_set_inheritable_async_safe(int fd, int inheritable, int *atomic_flag_works) |
1612 | 0 | { |
1613 | 0 | return set_inheritable(fd, inheritable, 0, atomic_flag_works); |
1614 | 0 | } |
1615 | | |
1616 | | static int |
1617 | | _Py_open_impl(const char *pathname, int flags, int gil_held) |
1618 | 0 | { |
1619 | 0 | int fd; |
1620 | 0 | int async_err = 0; |
1621 | 0 | #ifndef MS_WINDOWS |
1622 | 0 | int *atomic_flag_works; |
1623 | 0 | #endif |
1624 | |
|
1625 | | #ifdef MS_WINDOWS |
1626 | | flags |= O_NOINHERIT; |
1627 | | #elif defined(O_CLOEXEC) |
1628 | | atomic_flag_works = &_Py_open_cloexec_works; |
1629 | 0 | flags |= O_CLOEXEC; |
1630 | | #else |
1631 | | atomic_flag_works = NULL; |
1632 | | #endif |
1633 | |
|
1634 | 0 | if (gil_held) { |
1635 | 0 | PyObject *pathname_obj = PyUnicode_DecodeFSDefault(pathname); |
1636 | 0 | if (pathname_obj == NULL) { |
1637 | 0 | return -1; |
1638 | 0 | } |
1639 | 0 | if (PySys_Audit("open", "OOi", pathname_obj, Py_None, flags) < 0) { |
1640 | 0 | Py_DECREF(pathname_obj); |
1641 | 0 | return -1; |
1642 | 0 | } |
1643 | | |
1644 | 0 | do { |
1645 | 0 | Py_BEGIN_ALLOW_THREADS |
1646 | 0 | fd = open(pathname, flags); |
1647 | 0 | Py_END_ALLOW_THREADS |
1648 | 0 | } while (fd < 0 |
1649 | 0 | && errno == EINTR && !(async_err = PyErr_CheckSignals())); |
1650 | 0 | if (async_err) { |
1651 | 0 | Py_DECREF(pathname_obj); |
1652 | 0 | return -1; |
1653 | 0 | } |
1654 | 0 | if (fd < 0) { |
1655 | 0 | PyErr_SetFromErrnoWithFilenameObjects(PyExc_OSError, pathname_obj, NULL); |
1656 | 0 | Py_DECREF(pathname_obj); |
1657 | 0 | return -1; |
1658 | 0 | } |
1659 | 0 | Py_DECREF(pathname_obj); |
1660 | 0 | } |
1661 | 0 | else { |
1662 | 0 | fd = open(pathname, flags); |
1663 | 0 | if (fd < 0) |
1664 | 0 | return -1; |
1665 | 0 | } |
1666 | | |
1667 | 0 | #ifndef MS_WINDOWS |
1668 | 0 | if (set_inheritable(fd, 0, gil_held, atomic_flag_works) < 0) { |
1669 | 0 | close(fd); |
1670 | 0 | return -1; |
1671 | 0 | } |
1672 | 0 | #endif |
1673 | | |
1674 | 0 | return fd; |
1675 | 0 | } |
1676 | | |
1677 | | /* Open a file with the specified flags (wrapper to open() function). |
1678 | | Return a file descriptor on success. Raise an exception and return -1 on |
1679 | | error. |
1680 | | |
1681 | | The file descriptor is created non-inheritable. |
1682 | | |
1683 | | When interrupted by a signal (open() fails with EINTR), retry the syscall, |
1684 | | except if the Python signal handler raises an exception. |
1685 | | |
1686 | | Release the GIL to call open(). The caller must hold the GIL. */ |
1687 | | int |
1688 | | _Py_open(const char *pathname, int flags) |
1689 | 0 | { |
1690 | | /* _Py_open() must be called with the GIL held. */ |
1691 | 0 | _Py_AssertHoldsTstate(); |
1692 | 0 | return _Py_open_impl(pathname, flags, 1); |
1693 | 0 | } |
1694 | | |
1695 | | /* Open a file with the specified flags (wrapper to open() function). |
1696 | | Return a file descriptor on success. Set errno and return -1 on error. |
1697 | | |
1698 | | The file descriptor is created non-inheritable. |
1699 | | |
1700 | | If interrupted by a signal, fail with EINTR. */ |
1701 | | int |
1702 | | _Py_open_noraise(const char *pathname, int flags) |
1703 | 0 | { |
1704 | 0 | return _Py_open_impl(pathname, flags, 0); |
1705 | 0 | } |
1706 | | |
1707 | | /* Open a file. Use _wfopen() on Windows, encode the path to the locale |
1708 | | encoding and use fopen() otherwise. |
1709 | | |
1710 | | The file descriptor is created non-inheritable. |
1711 | | |
1712 | | If interrupted by a signal, fail with EINTR. */ |
1713 | | FILE * |
1714 | | _Py_wfopen(const wchar_t *path, const wchar_t *mode) |
1715 | 80 | { |
1716 | 80 | FILE *f; |
1717 | 80 | if (PySys_Audit("open", "uui", path, mode, 0) < 0) { |
1718 | 0 | return NULL; |
1719 | 0 | } |
1720 | 80 | #ifndef MS_WINDOWS |
1721 | 80 | char *cpath; |
1722 | 80 | char cmode[10]; |
1723 | 80 | size_t r; |
1724 | 80 | r = wcstombs(cmode, mode, 10); |
1725 | 80 | if (r == DECODE_ERROR || r >= 10) { |
1726 | 0 | errno = EINVAL; |
1727 | 0 | return NULL; |
1728 | 0 | } |
1729 | 80 | cpath = _Py_EncodeLocaleRaw(path, NULL); |
1730 | 80 | if (cpath == NULL) { |
1731 | 0 | return NULL; |
1732 | 0 | } |
1733 | 80 | f = fopen(cpath, cmode); |
1734 | 80 | PyMem_RawFree(cpath); |
1735 | | #else |
1736 | | f = _wfopen(path, mode); |
1737 | | #endif |
1738 | 80 | if (f == NULL) |
1739 | 80 | return NULL; |
1740 | 0 | if (make_non_inheritable(fileno(f)) < 0) { |
1741 | 0 | fclose(f); |
1742 | 0 | return NULL; |
1743 | 0 | } |
1744 | 0 | return f; |
1745 | 0 | } |
1746 | | |
1747 | | |
1748 | | /* Open a file. |
1749 | | |
1750 | | On Windows, if 'path' is a Unicode string, call _wfopen(). Otherwise, encode |
1751 | | the path to the filesystem encoding and call fopen(). |
1752 | | |
1753 | | Return the new file object on success. Raise an exception and return NULL |
1754 | | on error. |
1755 | | |
1756 | | The file descriptor is created non-inheritable. |
1757 | | |
1758 | | When interrupted by a signal (open() fails with EINTR), retry the syscall, |
1759 | | except if the Python signal handler raises an exception. |
1760 | | |
1761 | | Release the GIL to call _wfopen() or fopen(). The caller must hold |
1762 | | the GIL. */ |
1763 | | FILE* |
1764 | | Py_fopen(PyObject *path, const char *mode) |
1765 | 11.8k | { |
1766 | 11.8k | _Py_AssertHoldsTstate(); |
1767 | | |
1768 | 11.8k | if (PySys_Audit("open", "Osi", path, mode, 0) < 0) { |
1769 | 0 | return NULL; |
1770 | 0 | } |
1771 | | |
1772 | 11.8k | FILE *f; |
1773 | 11.8k | int async_err = 0; |
1774 | 11.8k | int saved_errno; |
1775 | | #ifdef MS_WINDOWS |
1776 | | PyObject *unicode; |
1777 | | if (!PyUnicode_FSDecoder(path, &unicode)) { |
1778 | | return NULL; |
1779 | | } |
1780 | | |
1781 | | wchar_t *wpath = PyUnicode_AsWideCharString(unicode, NULL); |
1782 | | Py_DECREF(unicode); |
1783 | | if (wpath == NULL) { |
1784 | | return NULL; |
1785 | | } |
1786 | | |
1787 | | wchar_t wmode[10]; |
1788 | | int usize = MultiByteToWideChar(CP_ACP, 0, mode, -1, |
1789 | | wmode, Py_ARRAY_LENGTH(wmode)); |
1790 | | if (usize == 0) { |
1791 | | PyErr_SetFromWindowsErr(0); |
1792 | | PyMem_Free(wpath); |
1793 | | return NULL; |
1794 | | } |
1795 | | |
1796 | | do { |
1797 | | Py_BEGIN_ALLOW_THREADS |
1798 | | _Py_BEGIN_SUPPRESS_IPH |
1799 | | f = _wfopen(wpath, wmode); |
1800 | | _Py_END_SUPPRESS_IPH |
1801 | | Py_END_ALLOW_THREADS |
1802 | | } while (f == NULL |
1803 | | && errno == EINTR && !(async_err = PyErr_CheckSignals())); |
1804 | | saved_errno = errno; |
1805 | | PyMem_Free(wpath); |
1806 | | #else |
1807 | 11.8k | PyObject *bytes; |
1808 | 11.8k | if (!PyUnicode_FSConverter(path, &bytes)) { |
1809 | 0 | return NULL; |
1810 | 0 | } |
1811 | 11.8k | const char *path_bytes = PyBytes_AS_STRING(bytes); |
1812 | | |
1813 | 11.8k | do { |
1814 | 11.8k | Py_BEGIN_ALLOW_THREADS |
1815 | 11.8k | f = fopen(path_bytes, mode); |
1816 | 11.8k | Py_END_ALLOW_THREADS |
1817 | 11.8k | } while (f == NULL |
1818 | 11.8k | && errno == EINTR && !(async_err = PyErr_CheckSignals())); |
1819 | 11.8k | saved_errno = errno; |
1820 | 11.8k | Py_DECREF(bytes); |
1821 | 11.8k | #endif |
1822 | | |
1823 | 11.8k | if (async_err) { |
1824 | 0 | return NULL; |
1825 | 0 | } |
1826 | | |
1827 | 11.8k | if (f == NULL) { |
1828 | 11.8k | errno = saved_errno; |
1829 | 11.8k | PyErr_SetFromErrnoWithFilenameObject(PyExc_OSError, path); |
1830 | 11.8k | return NULL; |
1831 | 11.8k | } |
1832 | | |
1833 | 0 | if (set_inheritable(fileno(f), 0, 1, NULL) < 0) { |
1834 | 0 | fclose(f); |
1835 | 0 | return NULL; |
1836 | 0 | } |
1837 | 0 | return f; |
1838 | 0 | } |
1839 | | |
1840 | | |
1841 | | // Call fclose(). |
1842 | | // |
1843 | | // On Windows, files opened by Py_fopen() in the Python DLL must be closed by |
1844 | | // the Python DLL to use the same C runtime version. Otherwise, calling |
1845 | | // fclose() directly can cause undefined behavior. |
1846 | | int |
1847 | | Py_fclose(FILE *file) |
1848 | 0 | { |
1849 | 0 | return fclose(file); |
1850 | 0 | } |
1851 | | |
1852 | | |
1853 | | /* Read count bytes from fd into buf. |
1854 | | |
1855 | | On success, return the number of read bytes, it can be lower than count. |
1856 | | If the current file offset is at or past the end of file, no bytes are read, |
1857 | | and read() returns zero. |
1858 | | |
1859 | | On error, raise an exception, set errno and return -1. |
1860 | | |
1861 | | When interrupted by a signal (read() fails with EINTR), retry the syscall. |
1862 | | If the Python signal handler raises an exception, the function returns -1 |
1863 | | (the syscall is not retried). |
1864 | | |
1865 | | Release the GIL to call read(). The caller must hold the GIL. */ |
1866 | | Py_ssize_t |
1867 | | _Py_read(int fd, void *buf, size_t count) |
1868 | 1.88k | { |
1869 | 1.88k | Py_ssize_t n; |
1870 | 1.88k | int err; |
1871 | 1.88k | int async_err = 0; |
1872 | | |
1873 | 1.88k | _Py_AssertHoldsTstate(); |
1874 | | |
1875 | | /* _Py_read() must not be called with an exception set, otherwise the |
1876 | | * caller may think that read() was interrupted by a signal and the signal |
1877 | | * handler raised an exception. */ |
1878 | 1.88k | assert(!PyErr_Occurred()); |
1879 | | |
1880 | 1.88k | if (count > _PY_READ_MAX) { |
1881 | 0 | count = _PY_READ_MAX; |
1882 | 0 | } |
1883 | | |
1884 | 1.88k | _Py_BEGIN_SUPPRESS_IPH |
1885 | 1.88k | do { |
1886 | 1.88k | Py_BEGIN_ALLOW_THREADS |
1887 | 1.88k | errno = 0; |
1888 | | #ifdef MS_WINDOWS |
1889 | | _doserrno = 0; |
1890 | | n = read(fd, buf, (int)count); |
1891 | | // read() on a non-blocking empty pipe fails with EINVAL, which is |
1892 | | // mapped from the Windows error code ERROR_NO_DATA. |
1893 | | if (n < 0 && errno == EINVAL) { |
1894 | | if (_doserrno == ERROR_NO_DATA) { |
1895 | | errno = EAGAIN; |
1896 | | } |
1897 | | } |
1898 | | #else |
1899 | 1.88k | n = read(fd, buf, count); |
1900 | 1.88k | #endif |
1901 | | /* save/restore errno because PyErr_CheckSignals() |
1902 | | * and PyErr_SetFromErrno() can modify it */ |
1903 | 1.88k | err = errno; |
1904 | 1.88k | Py_END_ALLOW_THREADS |
1905 | 1.88k | } while (n < 0 && err == EINTR && |
1906 | 1.88k | !(async_err = PyErr_CheckSignals())); |
1907 | 1.88k | _Py_END_SUPPRESS_IPH |
1908 | | |
1909 | 1.88k | if (async_err) { |
1910 | | /* read() was interrupted by a signal (failed with EINTR) |
1911 | | * and the Python signal handler raised an exception */ |
1912 | 0 | errno = err; |
1913 | 0 | assert(errno == EINTR && PyErr_Occurred()); |
1914 | 0 | return -1; |
1915 | 0 | } |
1916 | 1.88k | if (n < 0) { |
1917 | 0 | PyErr_SetFromErrno(PyExc_OSError); |
1918 | 0 | errno = err; |
1919 | 0 | return -1; |
1920 | 0 | } |
1921 | | |
1922 | 1.88k | return n; |
1923 | 1.88k | } |
1924 | | |
1925 | | static Py_ssize_t |
1926 | | _Py_write_impl(int fd, const void *buf, size_t count, int gil_held) |
1927 | 216 | { |
1928 | 216 | Py_ssize_t n; |
1929 | 216 | int err; |
1930 | 216 | int async_err = 0; |
1931 | | |
1932 | 216 | _Py_BEGIN_SUPPRESS_IPH |
1933 | | #ifdef MS_WINDOWS |
1934 | | if (count > 32767) { |
1935 | | /* Issue #11395: the Windows console returns an error (12: not |
1936 | | enough space error) on writing into stdout if stdout mode is |
1937 | | binary and the length is greater than 66,000 bytes (or less, |
1938 | | depending on heap usage). */ |
1939 | | if (gil_held) { |
1940 | | Py_BEGIN_ALLOW_THREADS |
1941 | | if (isatty(fd)) { |
1942 | | count = 32767; |
1943 | | } |
1944 | | Py_END_ALLOW_THREADS |
1945 | | } else { |
1946 | | if (isatty(fd)) { |
1947 | | count = 32767; |
1948 | | } |
1949 | | } |
1950 | | } |
1951 | | |
1952 | | #endif |
1953 | 216 | if (count > _PY_WRITE_MAX) { |
1954 | 0 | count = _PY_WRITE_MAX; |
1955 | 0 | } |
1956 | | |
1957 | 216 | if (gil_held) { |
1958 | 216 | do { |
1959 | 216 | Py_BEGIN_ALLOW_THREADS |
1960 | 216 | errno = 0; |
1961 | | #ifdef MS_WINDOWS |
1962 | | // write() on a non-blocking pipe fails with ENOSPC on Windows if |
1963 | | // the pipe lacks available space for the entire buffer. |
1964 | | int c = (int)count; |
1965 | | do { |
1966 | | _doserrno = 0; |
1967 | | n = write(fd, buf, c); |
1968 | | if (n >= 0 || errno != ENOSPC || _doserrno != 0) { |
1969 | | break; |
1970 | | } |
1971 | | errno = EAGAIN; |
1972 | | c /= 2; |
1973 | | } while (c > 0); |
1974 | | #else |
1975 | 216 | n = write(fd, buf, count); |
1976 | 216 | #endif |
1977 | | /* save/restore errno because PyErr_CheckSignals() |
1978 | | * and PyErr_SetFromErrno() can modify it */ |
1979 | 216 | err = errno; |
1980 | 216 | Py_END_ALLOW_THREADS |
1981 | 216 | } while (n < 0 && err == EINTR && |
1982 | 216 | !(async_err = PyErr_CheckSignals())); |
1983 | 216 | } |
1984 | 0 | else { |
1985 | 0 | do { |
1986 | 0 | errno = 0; |
1987 | | #ifdef MS_WINDOWS |
1988 | | // write() on a non-blocking pipe fails with ENOSPC on Windows if |
1989 | | // the pipe lacks available space for the entire buffer. |
1990 | | int c = (int)count; |
1991 | | do { |
1992 | | _doserrno = 0; |
1993 | | n = write(fd, buf, c); |
1994 | | if (n >= 0 || errno != ENOSPC || _doserrno != 0) { |
1995 | | break; |
1996 | | } |
1997 | | errno = EAGAIN; |
1998 | | c /= 2; |
1999 | | } while (c > 0); |
2000 | | #else |
2001 | 0 | n = write(fd, buf, count); |
2002 | 0 | #endif |
2003 | 0 | err = errno; |
2004 | 0 | } while (n < 0 && err == EINTR); |
2005 | 0 | } |
2006 | 216 | _Py_END_SUPPRESS_IPH |
2007 | | |
2008 | 216 | if (async_err) { |
2009 | | /* write() was interrupted by a signal (failed with EINTR) |
2010 | | and the Python signal handler raised an exception (if gil_held is |
2011 | | nonzero). */ |
2012 | 0 | errno = err; |
2013 | 0 | assert(errno == EINTR && (!gil_held || PyErr_Occurred())); |
2014 | 0 | return -1; |
2015 | 0 | } |
2016 | 216 | if (n < 0) { |
2017 | 0 | if (gil_held) |
2018 | 0 | PyErr_SetFromErrno(PyExc_OSError); |
2019 | 0 | errno = err; |
2020 | 0 | return -1; |
2021 | 0 | } |
2022 | | |
2023 | 216 | return n; |
2024 | 216 | } |
2025 | | |
2026 | | /* Write count bytes of buf into fd. |
2027 | | |
2028 | | On success, return the number of written bytes, it can be lower than count |
2029 | | including 0. On error, raise an exception, set errno and return -1. |
2030 | | |
2031 | | When interrupted by a signal (write() fails with EINTR), retry the syscall. |
2032 | | If the Python signal handler raises an exception, the function returns -1 |
2033 | | (the syscall is not retried). |
2034 | | |
2035 | | Release the GIL to call write(). The caller must hold the GIL. */ |
2036 | | Py_ssize_t |
2037 | | _Py_write(int fd, const void *buf, size_t count) |
2038 | 216 | { |
2039 | 216 | _Py_AssertHoldsTstate(); |
2040 | | |
2041 | | /* _Py_write() must not be called with an exception set, otherwise the |
2042 | | * caller may think that write() was interrupted by a signal and the signal |
2043 | | * handler raised an exception. */ |
2044 | 216 | assert(!PyErr_Occurred()); |
2045 | | |
2046 | 216 | return _Py_write_impl(fd, buf, count, 1); |
2047 | 216 | } |
2048 | | |
2049 | | /* Write count bytes of buf into fd. |
2050 | | * |
2051 | | * On success, return the number of written bytes, it can be lower than count |
2052 | | * including 0. On error, set errno and return -1. |
2053 | | * |
2054 | | * When interrupted by a signal (write() fails with EINTR), retry the syscall |
2055 | | * without calling the Python signal handler. */ |
2056 | | Py_ssize_t |
2057 | | _Py_write_noraise(int fd, const void *buf, size_t count) |
2058 | 0 | { |
2059 | 0 | return _Py_write_impl(fd, buf, count, 0); |
2060 | 0 | } |
2061 | | |
2062 | | #ifdef HAVE_READLINK |
2063 | | |
2064 | | /* Read value of symbolic link. Encode the path to the locale encoding, decode |
2065 | | the result from the locale encoding. |
2066 | | |
2067 | | Return -1 on encoding error, on readlink() error, if the internal buffer is |
2068 | | too short, on decoding error, or if 'buf' is too short. */ |
2069 | | int |
2070 | | _Py_wreadlink(const wchar_t *path, wchar_t *buf, size_t buflen) |
2071 | 32 | { |
2072 | 32 | char *cpath; |
2073 | 32 | char cbuf[MAXPATHLEN]; |
2074 | 32 | size_t cbuf_len = Py_ARRAY_LENGTH(cbuf); |
2075 | 32 | wchar_t *wbuf; |
2076 | 32 | Py_ssize_t res; |
2077 | 32 | size_t r1; |
2078 | | |
2079 | 32 | cpath = _Py_EncodeLocaleRaw(path, NULL); |
2080 | 32 | if (cpath == NULL) { |
2081 | 0 | errno = EINVAL; |
2082 | 0 | return -1; |
2083 | 0 | } |
2084 | 32 | res = readlink(cpath, cbuf, cbuf_len); |
2085 | 32 | PyMem_RawFree(cpath); |
2086 | 32 | if (res == -1) { |
2087 | 16 | return -1; |
2088 | 16 | } |
2089 | 16 | if ((size_t)res == cbuf_len) { |
2090 | 0 | errno = EINVAL; |
2091 | 0 | return -1; |
2092 | 0 | } |
2093 | 16 | cbuf[res] = '\0'; /* buf will be null terminated */ |
2094 | 16 | wbuf = Py_DecodeLocale(cbuf, &r1); |
2095 | 16 | if (wbuf == NULL) { |
2096 | 0 | errno = EINVAL; |
2097 | 0 | return -1; |
2098 | 0 | } |
2099 | | /* wbuf must have space to store the trailing NUL character */ |
2100 | 16 | if (buflen <= r1) { |
2101 | 0 | PyMem_RawFree(wbuf); |
2102 | 0 | errno = EINVAL; |
2103 | 0 | return -1; |
2104 | 0 | } |
2105 | 16 | wcsncpy(buf, wbuf, buflen); |
2106 | 16 | PyMem_RawFree(wbuf); |
2107 | 16 | return (int)r1; |
2108 | 16 | } |
2109 | | #endif |
2110 | | |
2111 | | #ifdef HAVE_REALPATH |
2112 | | |
2113 | | /* Return the canonicalized absolute pathname. Encode path to the locale |
2114 | | encoding, decode the result from the locale encoding. |
2115 | | |
2116 | | Return NULL on encoding error, realpath() error, decoding error |
2117 | | or if 'resolved_path' is too short. */ |
2118 | | wchar_t* |
2119 | | _Py_wrealpath(const wchar_t *path, |
2120 | | wchar_t *resolved_path, size_t resolved_path_len) |
2121 | 0 | { |
2122 | 0 | char *cpath; |
2123 | 0 | char cresolved_path[MAXPATHLEN]; |
2124 | 0 | wchar_t *wresolved_path; |
2125 | 0 | char *res; |
2126 | 0 | size_t r; |
2127 | 0 | cpath = _Py_EncodeLocaleRaw(path, NULL); |
2128 | 0 | if (cpath == NULL) { |
2129 | 0 | errno = EINVAL; |
2130 | 0 | return NULL; |
2131 | 0 | } |
2132 | 0 | res = realpath(cpath, cresolved_path); |
2133 | 0 | PyMem_RawFree(cpath); |
2134 | 0 | if (res == NULL) |
2135 | 0 | return NULL; |
2136 | | |
2137 | 0 | wresolved_path = Py_DecodeLocale(cresolved_path, &r); |
2138 | 0 | if (wresolved_path == NULL) { |
2139 | 0 | errno = EINVAL; |
2140 | 0 | return NULL; |
2141 | 0 | } |
2142 | | /* wresolved_path must have space to store the trailing NUL character */ |
2143 | 0 | if (resolved_path_len <= r) { |
2144 | 0 | PyMem_RawFree(wresolved_path); |
2145 | 0 | errno = EINVAL; |
2146 | 0 | return NULL; |
2147 | 0 | } |
2148 | 0 | wcsncpy(resolved_path, wresolved_path, resolved_path_len); |
2149 | 0 | PyMem_RawFree(wresolved_path); |
2150 | 0 | return resolved_path; |
2151 | 0 | } |
2152 | | #endif |
2153 | | |
2154 | | |
2155 | | int |
2156 | | _Py_isabs(const wchar_t *path) |
2157 | 304 | { |
2158 | | #ifdef MS_WINDOWS |
2159 | | const wchar_t *tail; |
2160 | | HRESULT hr = PathCchSkipRoot(path, &tail); |
2161 | | if (FAILED(hr) || path == tail) { |
2162 | | return 0; |
2163 | | } |
2164 | | if (tail == &path[1] && (path[0] == SEP || path[0] == ALTSEP)) { |
2165 | | // Exclude paths with leading SEP |
2166 | | return 0; |
2167 | | } |
2168 | | if (tail == &path[2] && path[1] == L':') { |
2169 | | // Exclude drive-relative paths (e.g. C:filename.ext) |
2170 | | return 0; |
2171 | | } |
2172 | | return 1; |
2173 | | #else |
2174 | 304 | return (path[0] == SEP); |
2175 | 304 | #endif |
2176 | 304 | } |
2177 | | |
2178 | | |
2179 | | /* Get an absolute path. |
2180 | | On error (ex: fail to get the current directory), return -1. |
2181 | | On memory allocation failure, set *abspath_p to NULL and return 0. |
2182 | | On success, return a newly allocated to *abspath_p to and return 0. |
2183 | | The string must be freed by PyMem_RawFree(). */ |
2184 | | int |
2185 | | _Py_abspath(const wchar_t *path, wchar_t **abspath_p) |
2186 | 0 | { |
2187 | 0 | if (path[0] == '\0' || !wcscmp(path, L".")) { |
2188 | 0 | wchar_t cwd[MAXPATHLEN + 1]; |
2189 | 0 | cwd[Py_ARRAY_LENGTH(cwd) - 1] = 0; |
2190 | 0 | if (!_Py_wgetcwd(cwd, Py_ARRAY_LENGTH(cwd) - 1)) { |
2191 | | /* unable to get the current directory */ |
2192 | 0 | return -1; |
2193 | 0 | } |
2194 | 0 | *abspath_p = _PyMem_RawWcsdup(cwd); |
2195 | 0 | return 0; |
2196 | 0 | } |
2197 | | |
2198 | 0 | if (_Py_isabs(path)) { |
2199 | 0 | *abspath_p = _PyMem_RawWcsdup(path); |
2200 | 0 | return 0; |
2201 | 0 | } |
2202 | | |
2203 | | #ifdef MS_WINDOWS |
2204 | | return _PyOS_getfullpathname(path, abspath_p); |
2205 | | #else |
2206 | 0 | wchar_t cwd[MAXPATHLEN + 1]; |
2207 | 0 | cwd[Py_ARRAY_LENGTH(cwd) - 1] = 0; |
2208 | 0 | if (!_Py_wgetcwd(cwd, Py_ARRAY_LENGTH(cwd) - 1)) { |
2209 | | /* unable to get the current directory */ |
2210 | 0 | return -1; |
2211 | 0 | } |
2212 | | |
2213 | 0 | size_t cwd_len = wcslen(cwd); |
2214 | 0 | size_t path_len = wcslen(path); |
2215 | 0 | size_t len = cwd_len + 1 + path_len + 1; |
2216 | 0 | if (len <= (size_t)PY_SSIZE_T_MAX / sizeof(wchar_t)) { |
2217 | 0 | *abspath_p = PyMem_RawMalloc(len * sizeof(wchar_t)); |
2218 | 0 | } |
2219 | 0 | else { |
2220 | 0 | *abspath_p = NULL; |
2221 | 0 | } |
2222 | 0 | if (*abspath_p == NULL) { |
2223 | 0 | return 0; |
2224 | 0 | } |
2225 | | |
2226 | 0 | wchar_t *abspath = *abspath_p; |
2227 | 0 | memcpy(abspath, cwd, cwd_len * sizeof(wchar_t)); |
2228 | 0 | abspath += cwd_len; |
2229 | |
|
2230 | 0 | *abspath = (wchar_t)SEP; |
2231 | 0 | abspath++; |
2232 | |
|
2233 | 0 | memcpy(abspath, path, path_len * sizeof(wchar_t)); |
2234 | 0 | abspath += path_len; |
2235 | |
|
2236 | 0 | *abspath = 0; |
2237 | 0 | return 0; |
2238 | 0 | #endif |
2239 | 0 | } |
2240 | | |
2241 | | // The Windows Games API family implements the PathCch* APIs in the Xbox OS, |
2242 | | // but does not expose them yet. Load them dynamically until |
2243 | | // 1) they are officially exposed |
2244 | | // 2) we stop supporting older versions of the GDK which do not expose them |
2245 | | #if defined(MS_WINDOWS_GAMES) && !defined(MS_WINDOWS_DESKTOP) |
2246 | | HRESULT |
2247 | | PathCchSkipRoot(const wchar_t *path, const wchar_t **rootEnd) |
2248 | | { |
2249 | | static int initialized = 0; |
2250 | | typedef HRESULT(__stdcall *PPathCchSkipRoot) (PCWSTR pszPath, |
2251 | | PCWSTR *ppszRootEnd); |
2252 | | static PPathCchSkipRoot _PathCchSkipRoot; |
2253 | | |
2254 | | if (initialized == 0) { |
2255 | | HMODULE pathapi = LoadLibraryExW(L"api-ms-win-core-path-l1-1-0.dll", NULL, |
2256 | | LOAD_LIBRARY_SEARCH_SYSTEM32); |
2257 | | if (pathapi) { |
2258 | | _PathCchSkipRoot = (PPathCchSkipRoot)GetProcAddress( |
2259 | | pathapi, "PathCchSkipRoot"); |
2260 | | } |
2261 | | else { |
2262 | | _PathCchSkipRoot = NULL; |
2263 | | } |
2264 | | initialized = 1; |
2265 | | } |
2266 | | |
2267 | | if (!_PathCchSkipRoot) { |
2268 | | return E_NOINTERFACE; |
2269 | | } |
2270 | | |
2271 | | return _PathCchSkipRoot(path, rootEnd); |
2272 | | } |
2273 | | |
2274 | | static HRESULT |
2275 | | PathCchCombineEx(wchar_t *buffer, size_t bufsize, const wchar_t *dirname, |
2276 | | const wchar_t *relfile, unsigned long flags) |
2277 | | { |
2278 | | static int initialized = 0; |
2279 | | typedef HRESULT(__stdcall *PPathCchCombineEx) (PWSTR pszPathOut, |
2280 | | size_t cchPathOut, |
2281 | | PCWSTR pszPathIn, |
2282 | | PCWSTR pszMore, |
2283 | | unsigned long dwFlags); |
2284 | | static PPathCchCombineEx _PathCchCombineEx; |
2285 | | |
2286 | | if (initialized == 0) { |
2287 | | HMODULE pathapi = LoadLibraryExW(L"api-ms-win-core-path-l1-1-0.dll", NULL, |
2288 | | LOAD_LIBRARY_SEARCH_SYSTEM32); |
2289 | | if (pathapi) { |
2290 | | _PathCchCombineEx = (PPathCchCombineEx)GetProcAddress( |
2291 | | pathapi, "PathCchCombineEx"); |
2292 | | } |
2293 | | else { |
2294 | | _PathCchCombineEx = NULL; |
2295 | | } |
2296 | | initialized = 1; |
2297 | | } |
2298 | | |
2299 | | if (!_PathCchCombineEx) { |
2300 | | return E_NOINTERFACE; |
2301 | | } |
2302 | | |
2303 | | return _PathCchCombineEx(buffer, bufsize, dirname, relfile, flags); |
2304 | | } |
2305 | | |
2306 | | #endif /* defined(MS_WINDOWS_GAMES) && !defined(MS_WINDOWS_DESKTOP) */ |
2307 | | |
2308 | | void |
2309 | | _Py_skiproot(const wchar_t *path, Py_ssize_t size, Py_ssize_t *drvsize, |
2310 | | Py_ssize_t *rootsize) |
2311 | 224 | { |
2312 | 224 | assert(drvsize); |
2313 | 224 | assert(rootsize); |
2314 | 224 | #ifndef MS_WINDOWS |
2315 | 672 | #define IS_SEP(x) (*(x) == SEP) |
2316 | 224 | *drvsize = 0; |
2317 | 224 | if (!IS_SEP(&path[0])) { |
2318 | | // Relative path, e.g.: 'foo' |
2319 | 0 | *rootsize = 0; |
2320 | 0 | } |
2321 | 224 | else if (!IS_SEP(&path[1]) || IS_SEP(&path[2])) { |
2322 | | // Absolute path, e.g.: '/foo', '///foo', '////foo', etc. |
2323 | 224 | *rootsize = 1; |
2324 | 224 | } |
2325 | 0 | else { |
2326 | | // Precisely two leading slashes, e.g.: '//foo'. Implementation defined per POSIX, see |
2327 | | // https://pubs.opengroup.org/onlinepubs/9699919799/basedefs/V1_chap04.html#tag_04_13 |
2328 | 0 | *rootsize = 2; |
2329 | 0 | } |
2330 | 224 | #undef IS_SEP |
2331 | | #else |
2332 | | const wchar_t *pEnd = size >= 0 ? &path[size] : NULL; |
2333 | | #define IS_END(x) (pEnd ? (x) == pEnd : !*(x)) |
2334 | | #define IS_SEP(x) (*(x) == SEP || *(x) == ALTSEP) |
2335 | | #define SEP_OR_END(x) (IS_SEP(x) || IS_END(x)) |
2336 | | if (IS_SEP(&path[0])) { |
2337 | | if (IS_SEP(&path[1])) { |
2338 | | // Device drives, e.g. \\.\device or \\?\device |
2339 | | // UNC drives, e.g. \\server\share or \\?\UNC\server\share |
2340 | | Py_ssize_t idx; |
2341 | | if (path[2] == L'?' && IS_SEP(&path[3]) && |
2342 | | (path[4] == L'U' || path[4] == L'u') && |
2343 | | (path[5] == L'N' || path[5] == L'n') && |
2344 | | (path[6] == L'C' || path[6] == L'c') && |
2345 | | IS_SEP(&path[7])) |
2346 | | { |
2347 | | idx = 8; |
2348 | | } |
2349 | | else { |
2350 | | idx = 2; |
2351 | | } |
2352 | | while (!SEP_OR_END(&path[idx])) { |
2353 | | idx++; |
2354 | | } |
2355 | | if (IS_END(&path[idx])) { |
2356 | | *drvsize = idx; |
2357 | | *rootsize = 0; |
2358 | | } |
2359 | | else { |
2360 | | idx++; |
2361 | | while (!SEP_OR_END(&path[idx])) { |
2362 | | idx++; |
2363 | | } |
2364 | | *drvsize = idx; |
2365 | | if (IS_END(&path[idx])) { |
2366 | | *rootsize = 0; |
2367 | | } |
2368 | | else { |
2369 | | *rootsize = 1; |
2370 | | } |
2371 | | } |
2372 | | } |
2373 | | else { |
2374 | | // Relative path with root, e.g. \Windows |
2375 | | *drvsize = 0; |
2376 | | *rootsize = 1; |
2377 | | } |
2378 | | } |
2379 | | else if (!IS_END(&path[0]) && path[1] == L':') { |
2380 | | *drvsize = 2; |
2381 | | if (IS_SEP(&path[2])) { |
2382 | | // Absolute drive-letter path, e.g. X:\Windows |
2383 | | *rootsize = 1; |
2384 | | } |
2385 | | else { |
2386 | | // Relative path with drive, e.g. X:Windows |
2387 | | *rootsize = 0; |
2388 | | } |
2389 | | } |
2390 | | else { |
2391 | | // Relative path, e.g. Windows |
2392 | | *drvsize = 0; |
2393 | | *rootsize = 0; |
2394 | | } |
2395 | | #undef SEP_OR_END |
2396 | | #undef IS_SEP |
2397 | | #undef IS_END |
2398 | | #endif |
2399 | 224 | } |
2400 | | |
2401 | | // The caller must ensure "buffer" is big enough. |
2402 | | static int |
2403 | | join_relfile(wchar_t *buffer, size_t bufsize, |
2404 | | const wchar_t *dirname, const wchar_t *relfile) |
2405 | 144 | { |
2406 | | #ifdef MS_WINDOWS |
2407 | | if (FAILED(PathCchCombineEx(buffer, bufsize, dirname, relfile, |
2408 | | PATHCCH_ALLOW_LONG_PATHS))) { |
2409 | | return -1; |
2410 | | } |
2411 | | #else |
2412 | 144 | assert(!_Py_isabs(relfile)); |
2413 | 144 | size_t dirlen = wcslen(dirname); |
2414 | 144 | size_t rellen = wcslen(relfile); |
2415 | 144 | size_t maxlen = bufsize - 1; |
2416 | 144 | if (maxlen > MAXPATHLEN || dirlen >= maxlen || rellen >= maxlen - dirlen) { |
2417 | 0 | return -1; |
2418 | 0 | } |
2419 | 144 | if (dirlen == 0) { |
2420 | | // We do not add a leading separator. |
2421 | 0 | wcscpy(buffer, relfile); |
2422 | 0 | } |
2423 | 144 | else { |
2424 | 144 | if (dirname != buffer) { |
2425 | 0 | wcscpy(buffer, dirname); |
2426 | 0 | } |
2427 | 144 | size_t relstart = dirlen; |
2428 | 144 | if (dirlen > 1 && dirname[dirlen - 1] != SEP) { |
2429 | 144 | buffer[dirlen] = SEP; |
2430 | 144 | relstart += 1; |
2431 | 144 | } |
2432 | 144 | wcscpy(&buffer[relstart], relfile); |
2433 | 144 | } |
2434 | 144 | #endif |
2435 | 144 | return 0; |
2436 | 144 | } |
2437 | | |
2438 | | /* Join the two paths together, like os.path.join(). Return NULL |
2439 | | if memory could not be allocated. The caller is responsible |
2440 | | for calling PyMem_RawFree() on the result. */ |
2441 | | wchar_t * |
2442 | | _Py_join_relfile(const wchar_t *dirname, const wchar_t *relfile) |
2443 | 0 | { |
2444 | 0 | assert(dirname != NULL && relfile != NULL); |
2445 | 0 | #ifndef MS_WINDOWS |
2446 | 0 | assert(!_Py_isabs(relfile)); |
2447 | 0 | #endif |
2448 | 0 | size_t maxlen = wcslen(dirname) + 1 + wcslen(relfile); |
2449 | 0 | size_t bufsize = maxlen + 1; |
2450 | 0 | wchar_t *filename = PyMem_RawMalloc(bufsize * sizeof(wchar_t)); |
2451 | 0 | if (filename == NULL) { |
2452 | 0 | return NULL; |
2453 | 0 | } |
2454 | 0 | assert(wcslen(dirname) < MAXPATHLEN); |
2455 | 0 | assert(wcslen(relfile) < MAXPATHLEN - wcslen(dirname)); |
2456 | 0 | if (join_relfile(filename, bufsize, dirname, relfile) < 0) { |
2457 | 0 | PyMem_RawFree(filename); |
2458 | 0 | return NULL; |
2459 | 0 | } |
2460 | 0 | return filename; |
2461 | 0 | } |
2462 | | |
2463 | | /* Join the two paths together, like os.path.join(). |
2464 | | dirname: the target buffer with the dirname already in place, |
2465 | | including trailing NUL |
2466 | | relfile: this must be a relative path |
2467 | | bufsize: total allocated size of the buffer |
2468 | | Return -1 if anything is wrong with the path lengths. */ |
2469 | | int |
2470 | | _Py_add_relfile(wchar_t *dirname, const wchar_t *relfile, size_t bufsize) |
2471 | 144 | { |
2472 | 144 | assert(dirname != NULL && relfile != NULL); |
2473 | 144 | assert(bufsize > 0); |
2474 | 144 | return join_relfile(dirname, bufsize, dirname, relfile); |
2475 | 144 | } |
2476 | | |
2477 | | |
2478 | | size_t |
2479 | | _Py_find_basename(const wchar_t *filename) |
2480 | 0 | { |
2481 | 0 | for (size_t i = wcslen(filename); i > 0; --i) { |
2482 | 0 | if (filename[i] == SEP) { |
2483 | 0 | return i + 1; |
2484 | 0 | } |
2485 | 0 | } |
2486 | 0 | return 0; |
2487 | 0 | } |
2488 | | |
2489 | | /* In-place path normalisation. Returns the start of the normalized |
2490 | | path, which will be within the original buffer. Guaranteed to not |
2491 | | make the path longer, and will not fail. 'size' is the length of |
2492 | | the path, if known. If -1, the first null character will be assumed |
2493 | | to be the end of the path. 'normsize' will be set to contain the |
2494 | | length of the resulting normalized path. */ |
2495 | | wchar_t * |
2496 | | _Py_normpath_and_size(wchar_t *path, Py_ssize_t size, Py_ssize_t *normsize) |
2497 | 224 | { |
2498 | 224 | assert(path != NULL); |
2499 | 224 | if ((size < 0 && !path[0]) || size == 0) { |
2500 | 0 | *normsize = 0; |
2501 | 0 | return path; |
2502 | 0 | } |
2503 | 224 | wchar_t *pEnd = size >= 0 ? &path[size] : NULL; |
2504 | 224 | wchar_t *p1 = path; // sequentially scanned address in the path |
2505 | 224 | wchar_t *p2 = path; // destination of a scanned character to be ljusted |
2506 | 224 | wchar_t *minP2 = path; // the beginning of the destination range |
2507 | 224 | wchar_t lastC = L'\0'; // the last ljusted character, p2[-1] in most cases |
2508 | | |
2509 | 11.4k | #define IS_END(x) (pEnd ? (x) == pEnd : !*(x)) |
2510 | | #ifdef ALTSEP |
2511 | | #define IS_SEP(x) (*(x) == SEP || *(x) == ALTSEP) |
2512 | | #else |
2513 | 224 | #define IS_SEP(x) (*(x) == SEP) |
2514 | 224 | #endif |
2515 | 224 | #define SEP_OR_END(x) (IS_SEP(x) || IS_END(x)) |
2516 | | |
2517 | 224 | Py_ssize_t drvsize, rootsize; |
2518 | 224 | _Py_skiproot(path, size, &drvsize, &rootsize); |
2519 | 224 | if (drvsize || rootsize) { |
2520 | | // Skip past root and update minP2 |
2521 | 224 | p1 = &path[drvsize + rootsize]; |
2522 | 224 | #ifndef ALTSEP |
2523 | 224 | p2 = p1; |
2524 | | #else |
2525 | | for (; p2 < p1; ++p2) { |
2526 | | if (*p2 == ALTSEP) { |
2527 | | *p2 = SEP; |
2528 | | } |
2529 | | } |
2530 | | #endif |
2531 | 224 | minP2 = p2 - 1; |
2532 | 224 | lastC = *minP2; |
2533 | | #ifdef MS_WINDOWS |
2534 | | if (lastC != SEP) { |
2535 | | minP2++; |
2536 | | } |
2537 | | #endif |
2538 | 224 | } |
2539 | 224 | if (p1[0] == L'.' && SEP_OR_END(&p1[1])) { |
2540 | | // Skip leading '.\' |
2541 | 0 | lastC = *++p1; |
2542 | | #ifdef ALTSEP |
2543 | | if (lastC == ALTSEP) { |
2544 | | lastC = SEP; |
2545 | | } |
2546 | | #endif |
2547 | 0 | while (IS_SEP(p1)) { |
2548 | 0 | p1++; |
2549 | 0 | } |
2550 | 0 | } |
2551 | | |
2552 | | /* if pEnd is specified, check that. Else, check for null terminator */ |
2553 | 11.4k | for (; !IS_END(p1); ++p1) { |
2554 | 11.2k | wchar_t c = *p1; |
2555 | | #ifdef ALTSEP |
2556 | | if (c == ALTSEP) { |
2557 | | c = SEP; |
2558 | | } |
2559 | | #endif |
2560 | 11.2k | if (lastC == SEP) { |
2561 | 1.16k | if (c == L'.') { |
2562 | 0 | int sep_at_1 = SEP_OR_END(&p1[1]); |
2563 | 0 | int sep_at_2 = !sep_at_1 && SEP_OR_END(&p1[2]); |
2564 | 0 | if (sep_at_2 && p1[1] == L'.') { |
2565 | 0 | wchar_t *p3 = p2; |
2566 | 0 | while (p3 != minP2 && *--p3 == SEP) { } |
2567 | 0 | while (p3 != minP2 && *(p3 - 1) != SEP) { --p3; } |
2568 | 0 | if (p2 == minP2 |
2569 | 0 | || (p3[0] == L'.' && p3[1] == L'.' && IS_SEP(&p3[2]))) |
2570 | 0 | { |
2571 | | // Previous segment is also ../, so append instead. |
2572 | | // Relative path does not absorb ../ at minP2 as well. |
2573 | 0 | *p2++ = L'.'; |
2574 | 0 | *p2++ = L'.'; |
2575 | 0 | lastC = L'.'; |
2576 | 0 | } else if (p3[0] == SEP) { |
2577 | | // Absolute path, so absorb segment |
2578 | 0 | p2 = p3 + 1; |
2579 | 0 | } else { |
2580 | 0 | p2 = p3; |
2581 | 0 | } |
2582 | 0 | p1 += 1; |
2583 | 0 | } else if (sep_at_1) { |
2584 | 0 | } else { |
2585 | 0 | *p2++ = lastC = c; |
2586 | 0 | } |
2587 | 1.16k | } else if (c == SEP) { |
2588 | 1.16k | } else { |
2589 | 1.16k | *p2++ = lastC = c; |
2590 | 1.16k | } |
2591 | 10.0k | } else { |
2592 | 10.0k | *p2++ = lastC = c; |
2593 | 10.0k | } |
2594 | 11.2k | } |
2595 | 224 | *p2 = L'\0'; |
2596 | 224 | if (p2 != minP2) { |
2597 | 224 | while (--p2 != minP2 && *p2 == SEP) { |
2598 | 0 | *p2 = L'\0'; |
2599 | 0 | } |
2600 | 224 | } else { |
2601 | 0 | --p2; |
2602 | 0 | } |
2603 | 224 | *normsize = p2 - path + 1; |
2604 | 224 | #undef SEP_OR_END |
2605 | 224 | #undef IS_SEP |
2606 | 224 | #undef IS_END |
2607 | 224 | return path; |
2608 | 224 | } |
2609 | | |
2610 | | /* In-place path normalisation. Returns the start of the normalized |
2611 | | path, which will be within the original buffer. Guaranteed to not |
2612 | | make the path longer, and will not fail. 'size' is the length of |
2613 | | the path, if known. If -1, the first null character will be assumed |
2614 | | to be the end of the path. */ |
2615 | | wchar_t * |
2616 | | _Py_normpath(wchar_t *path, Py_ssize_t size) |
2617 | 144 | { |
2618 | 144 | Py_ssize_t norm_length; |
2619 | 144 | return _Py_normpath_and_size(path, size, &norm_length); |
2620 | 144 | } |
2621 | | |
2622 | | |
2623 | | /* Get the current directory. buflen is the buffer size in wide characters |
2624 | | including the null character. Decode the path from the locale encoding. |
2625 | | |
2626 | | Return NULL on getcwd() error, on decoding error, or if 'buf' is |
2627 | | too short. */ |
2628 | | wchar_t* |
2629 | | _Py_wgetcwd(wchar_t *buf, size_t buflen) |
2630 | 0 | { |
2631 | | #ifdef MS_WINDOWS |
2632 | | int ibuflen = (int)Py_MIN(buflen, INT_MAX); |
2633 | | return _wgetcwd(buf, ibuflen); |
2634 | | #else |
2635 | 0 | char fname[MAXPATHLEN]; |
2636 | 0 | wchar_t *wname; |
2637 | 0 | size_t len; |
2638 | |
|
2639 | 0 | if (getcwd(fname, Py_ARRAY_LENGTH(fname)) == NULL) |
2640 | 0 | return NULL; |
2641 | 0 | wname = Py_DecodeLocale(fname, &len); |
2642 | 0 | if (wname == NULL) |
2643 | 0 | return NULL; |
2644 | | /* wname must have space to store the trailing NUL character */ |
2645 | 0 | if (buflen <= len) { |
2646 | 0 | PyMem_RawFree(wname); |
2647 | 0 | return NULL; |
2648 | 0 | } |
2649 | 0 | wcsncpy(buf, wname, buflen); |
2650 | 0 | PyMem_RawFree(wname); |
2651 | 0 | return buf; |
2652 | 0 | #endif |
2653 | 0 | } |
2654 | | |
2655 | | /* Duplicate a file descriptor. The new file descriptor is created as |
2656 | | non-inheritable. Return a new file descriptor on success, raise an OSError |
2657 | | exception and return -1 on error. |
2658 | | |
2659 | | The GIL is released to call dup(). The caller must hold the GIL. */ |
2660 | | int |
2661 | | _Py_dup(int fd) |
2662 | 0 | { |
2663 | | #ifdef MS_WINDOWS |
2664 | | HANDLE handle; |
2665 | | #endif |
2666 | |
|
2667 | 0 | _Py_AssertHoldsTstate(); |
2668 | |
|
2669 | | #ifdef MS_WINDOWS |
2670 | | handle = _Py_get_osfhandle(fd); |
2671 | | if (handle == INVALID_HANDLE_VALUE) |
2672 | | return -1; |
2673 | | |
2674 | | Py_BEGIN_ALLOW_THREADS |
2675 | | _Py_BEGIN_SUPPRESS_IPH |
2676 | | fd = dup(fd); |
2677 | | _Py_END_SUPPRESS_IPH |
2678 | | Py_END_ALLOW_THREADS |
2679 | | if (fd < 0) { |
2680 | | PyErr_SetFromErrno(PyExc_OSError); |
2681 | | return -1; |
2682 | | } |
2683 | | |
2684 | | if (_Py_set_inheritable(fd, 0, NULL) < 0) { |
2685 | | _Py_BEGIN_SUPPRESS_IPH |
2686 | | close(fd); |
2687 | | _Py_END_SUPPRESS_IPH |
2688 | | return -1; |
2689 | | } |
2690 | | #elif defined(HAVE_FCNTL_H) && defined(F_DUPFD_CLOEXEC) |
2691 | 0 | Py_BEGIN_ALLOW_THREADS |
2692 | 0 | _Py_BEGIN_SUPPRESS_IPH |
2693 | 0 | fd = fcntl(fd, F_DUPFD_CLOEXEC, 0); |
2694 | 0 | _Py_END_SUPPRESS_IPH |
2695 | 0 | Py_END_ALLOW_THREADS |
2696 | 0 | if (fd < 0) { |
2697 | 0 | PyErr_SetFromErrno(PyExc_OSError); |
2698 | 0 | return -1; |
2699 | 0 | } |
2700 | | |
2701 | | #elif HAVE_DUP |
2702 | | Py_BEGIN_ALLOW_THREADS |
2703 | | _Py_BEGIN_SUPPRESS_IPH |
2704 | | fd = dup(fd); |
2705 | | _Py_END_SUPPRESS_IPH |
2706 | | Py_END_ALLOW_THREADS |
2707 | | if (fd < 0) { |
2708 | | PyErr_SetFromErrno(PyExc_OSError); |
2709 | | return -1; |
2710 | | } |
2711 | | |
2712 | | if (_Py_set_inheritable(fd, 0, NULL) < 0) { |
2713 | | _Py_BEGIN_SUPPRESS_IPH |
2714 | | close(fd); |
2715 | | _Py_END_SUPPRESS_IPH |
2716 | | return -1; |
2717 | | } |
2718 | | #else |
2719 | | errno = ENOTSUP; |
2720 | | PyErr_SetFromErrno(PyExc_OSError); |
2721 | | return -1; |
2722 | | #endif |
2723 | 0 | return fd; |
2724 | 0 | } |
2725 | | |
2726 | | #ifndef MS_WINDOWS |
2727 | | /* Get the blocking mode of the file descriptor. |
2728 | | Return 0 if the O_NONBLOCK flag is set, 1 if the flag is cleared, |
2729 | | raise an exception and return -1 on error. */ |
2730 | | int |
2731 | | _Py_get_blocking(int fd) |
2732 | 0 | { |
2733 | 0 | int flags; |
2734 | 0 | _Py_BEGIN_SUPPRESS_IPH |
2735 | 0 | flags = fcntl(fd, F_GETFL, 0); |
2736 | 0 | _Py_END_SUPPRESS_IPH |
2737 | 0 | if (flags < 0) { |
2738 | 0 | PyErr_SetFromErrno(PyExc_OSError); |
2739 | 0 | return -1; |
2740 | 0 | } |
2741 | | |
2742 | 0 | return !(flags & O_NONBLOCK); |
2743 | 0 | } |
2744 | | |
2745 | | /* Set the blocking mode of the specified file descriptor. |
2746 | | |
2747 | | Set the O_NONBLOCK flag if blocking is False, clear the O_NONBLOCK flag |
2748 | | otherwise. |
2749 | | |
2750 | | Return 0 on success, raise an exception and return -1 on error. */ |
2751 | | int |
2752 | | _Py_set_blocking(int fd, int blocking) |
2753 | 0 | { |
2754 | | /* bpo-41462: On VxWorks, ioctl(FIONBIO) only works on sockets. |
2755 | | Use fcntl() instead. */ |
2756 | 0 | #if defined(HAVE_SYS_IOCTL_H) && defined(FIONBIO) && !defined(__VXWORKS__) |
2757 | 0 | int arg = !blocking; |
2758 | 0 | if (ioctl(fd, FIONBIO, &arg) < 0) |
2759 | 0 | goto error; |
2760 | | #else |
2761 | | int flags, res; |
2762 | | |
2763 | | _Py_BEGIN_SUPPRESS_IPH |
2764 | | flags = fcntl(fd, F_GETFL, 0); |
2765 | | if (flags >= 0) { |
2766 | | if (blocking) |
2767 | | flags = flags & (~O_NONBLOCK); |
2768 | | else |
2769 | | flags = flags | O_NONBLOCK; |
2770 | | |
2771 | | res = fcntl(fd, F_SETFL, flags); |
2772 | | } else { |
2773 | | res = -1; |
2774 | | } |
2775 | | _Py_END_SUPPRESS_IPH |
2776 | | |
2777 | | if (res < 0) |
2778 | | goto error; |
2779 | | #endif |
2780 | 0 | return 0; |
2781 | | |
2782 | 0 | error: |
2783 | 0 | PyErr_SetFromErrno(PyExc_OSError); |
2784 | 0 | return -1; |
2785 | 0 | } |
2786 | | #else /* MS_WINDOWS */ |
2787 | | |
2788 | | // The Windows Games API family doesn't expose GetNamedPipeHandleStateW so attempt |
2789 | | // to load it directly from the Kernel32.dll |
2790 | | #if !defined(MS_WINDOWS_APP) && !defined(MS_WINDOWS_SYSTEM) |
2791 | | BOOL |
2792 | | GetNamedPipeHandleStateW(HANDLE hNamedPipe, LPDWORD lpState, LPDWORD lpCurInstances, LPDWORD lpMaxCollectionCount, |
2793 | | LPDWORD lpCollectDataTimeout, LPWSTR lpUserName, DWORD nMaxUserNameSize) |
2794 | | { |
2795 | | static int initialized = 0; |
2796 | | typedef BOOL(__stdcall* PGetNamedPipeHandleStateW) ( |
2797 | | HANDLE hNamedPipe, LPDWORD lpState, LPDWORD lpCurInstances, LPDWORD lpMaxCollectionCount, |
2798 | | LPDWORD lpCollectDataTimeout, LPWSTR lpUserName, DWORD nMaxUserNameSize); |
2799 | | static PGetNamedPipeHandleStateW _GetNamedPipeHandleStateW; |
2800 | | |
2801 | | if (initialized == 0) { |
2802 | | HMODULE api = LoadLibraryExW(L"Kernel32.dll", NULL, LOAD_LIBRARY_SEARCH_SYSTEM32); |
2803 | | if (api) { |
2804 | | _GetNamedPipeHandleStateW = (PGetNamedPipeHandleStateW)GetProcAddress( |
2805 | | api, "GetNamedPipeHandleStateW"); |
2806 | | } |
2807 | | else { |
2808 | | _GetNamedPipeHandleStateW = NULL; |
2809 | | } |
2810 | | initialized = 1; |
2811 | | } |
2812 | | |
2813 | | if (!_GetNamedPipeHandleStateW) { |
2814 | | SetLastError(E_NOINTERFACE); |
2815 | | return FALSE; |
2816 | | } |
2817 | | |
2818 | | return _GetNamedPipeHandleStateW( |
2819 | | hNamedPipe, lpState, lpCurInstances, lpMaxCollectionCount, lpCollectDataTimeout, lpUserName, nMaxUserNameSize |
2820 | | ); |
2821 | | } |
2822 | | #endif /* !MS_WINDOWS_APP && !MS_WINDOWS_SYSTEM */ |
2823 | | |
2824 | | int |
2825 | | _Py_get_blocking(int fd) |
2826 | | { |
2827 | | HANDLE handle; |
2828 | | DWORD mode; |
2829 | | BOOL success; |
2830 | | |
2831 | | handle = _Py_get_osfhandle(fd); |
2832 | | if (handle == INVALID_HANDLE_VALUE) { |
2833 | | return -1; |
2834 | | } |
2835 | | |
2836 | | Py_BEGIN_ALLOW_THREADS |
2837 | | success = GetNamedPipeHandleStateW(handle, &mode, |
2838 | | NULL, NULL, NULL, NULL, 0); |
2839 | | Py_END_ALLOW_THREADS |
2840 | | |
2841 | | if (!success) { |
2842 | | PyErr_SetFromWindowsErr(0); |
2843 | | return -1; |
2844 | | } |
2845 | | |
2846 | | return !(mode & PIPE_NOWAIT); |
2847 | | } |
2848 | | |
2849 | | int |
2850 | | _Py_set_blocking(int fd, int blocking) |
2851 | | { |
2852 | | HANDLE handle; |
2853 | | DWORD mode; |
2854 | | BOOL success; |
2855 | | |
2856 | | handle = _Py_get_osfhandle(fd); |
2857 | | if (handle == INVALID_HANDLE_VALUE) { |
2858 | | return -1; |
2859 | | } |
2860 | | |
2861 | | Py_BEGIN_ALLOW_THREADS |
2862 | | success = GetNamedPipeHandleStateW(handle, &mode, |
2863 | | NULL, NULL, NULL, NULL, 0); |
2864 | | if (success) { |
2865 | | if (blocking) { |
2866 | | mode &= ~PIPE_NOWAIT; |
2867 | | } |
2868 | | else { |
2869 | | mode |= PIPE_NOWAIT; |
2870 | | } |
2871 | | success = SetNamedPipeHandleState(handle, &mode, NULL, NULL); |
2872 | | } |
2873 | | Py_END_ALLOW_THREADS |
2874 | | |
2875 | | if (!success) { |
2876 | | PyErr_SetFromWindowsErr(0); |
2877 | | return -1; |
2878 | | } |
2879 | | return 0; |
2880 | | } |
2881 | | |
2882 | | void* |
2883 | | _Py_get_osfhandle_noraise(int fd) |
2884 | | { |
2885 | | void *handle; |
2886 | | _Py_BEGIN_SUPPRESS_IPH |
2887 | | handle = (void*)_get_osfhandle(fd); |
2888 | | _Py_END_SUPPRESS_IPH |
2889 | | return handle; |
2890 | | } |
2891 | | |
2892 | | void* |
2893 | | _Py_get_osfhandle(int fd) |
2894 | | { |
2895 | | void *handle = _Py_get_osfhandle_noraise(fd); |
2896 | | if (handle == INVALID_HANDLE_VALUE) |
2897 | | PyErr_SetFromErrno(PyExc_OSError); |
2898 | | |
2899 | | return handle; |
2900 | | } |
2901 | | |
2902 | | int |
2903 | | _Py_open_osfhandle_noraise(void *handle, int flags) |
2904 | | { |
2905 | | int fd; |
2906 | | _Py_BEGIN_SUPPRESS_IPH |
2907 | | fd = _open_osfhandle((intptr_t)handle, flags); |
2908 | | _Py_END_SUPPRESS_IPH |
2909 | | return fd; |
2910 | | } |
2911 | | |
2912 | | int |
2913 | | _Py_open_osfhandle(void *handle, int flags) |
2914 | | { |
2915 | | int fd = _Py_open_osfhandle_noraise(handle, flags); |
2916 | | if (fd == -1) |
2917 | | PyErr_SetFromErrno(PyExc_OSError); |
2918 | | |
2919 | | return fd; |
2920 | | } |
2921 | | #endif /* MS_WINDOWS */ |
2922 | | |
2923 | | int |
2924 | | _Py_GetLocaleconvNumeric(struct lconv *lc, |
2925 | | PyObject **decimal_point, PyObject **thousands_sep) |
2926 | 0 | { |
2927 | 0 | assert(decimal_point != NULL); |
2928 | 0 | assert(thousands_sep != NULL); |
2929 | |
|
2930 | 0 | #ifndef MS_WINDOWS |
2931 | 0 | int change_locale = 0; |
2932 | 0 | if ((strlen(lc->decimal_point) > 1 || ((unsigned char)lc->decimal_point[0]) > 127)) { |
2933 | 0 | change_locale = 1; |
2934 | 0 | } |
2935 | 0 | if ((strlen(lc->thousands_sep) > 1 || ((unsigned char)lc->thousands_sep[0]) > 127)) { |
2936 | 0 | change_locale = 1; |
2937 | 0 | } |
2938 | | |
2939 | | /* Keep a copy of the LC_CTYPE locale */ |
2940 | 0 | char *oldloc = NULL, *loc = NULL; |
2941 | 0 | if (change_locale) { |
2942 | 0 | oldloc = setlocale(LC_CTYPE, NULL); |
2943 | 0 | if (!oldloc) { |
2944 | 0 | PyErr_SetString(PyExc_RuntimeWarning, |
2945 | 0 | "failed to get LC_CTYPE locale"); |
2946 | 0 | return -1; |
2947 | 0 | } |
2948 | | |
2949 | 0 | oldloc = _PyMem_Strdup(oldloc); |
2950 | 0 | if (!oldloc) { |
2951 | 0 | PyErr_NoMemory(); |
2952 | 0 | return -1; |
2953 | 0 | } |
2954 | | |
2955 | 0 | loc = setlocale(LC_NUMERIC, NULL); |
2956 | 0 | if (loc != NULL && strcmp(loc, oldloc) == 0) { |
2957 | 0 | loc = NULL; |
2958 | 0 | } |
2959 | |
|
2960 | 0 | if (loc != NULL) { |
2961 | | /* Only set the locale temporarily the LC_CTYPE locale |
2962 | | if LC_NUMERIC locale is different than LC_CTYPE locale and |
2963 | | decimal_point and/or thousands_sep are non-ASCII or longer than |
2964 | | 1 byte */ |
2965 | 0 | setlocale(LC_CTYPE, loc); |
2966 | 0 | } |
2967 | 0 | } |
2968 | | |
2969 | 0 | #define GET_LOCALE_STRING(ATTR) PyUnicode_DecodeLocale(lc->ATTR, NULL) |
2970 | | #else /* MS_WINDOWS */ |
2971 | | /* Use _W_* fields of Windows strcut lconv */ |
2972 | | #define GET_LOCALE_STRING(ATTR) PyUnicode_FromWideChar(lc->_W_ ## ATTR, -1) |
2973 | | #endif /* MS_WINDOWS */ |
2974 | | |
2975 | 0 | int res = -1; |
2976 | |
|
2977 | 0 | *decimal_point = GET_LOCALE_STRING(decimal_point); |
2978 | 0 | if (*decimal_point == NULL) { |
2979 | 0 | goto done; |
2980 | 0 | } |
2981 | | |
2982 | 0 | *thousands_sep = GET_LOCALE_STRING(thousands_sep); |
2983 | 0 | if (*thousands_sep == NULL) { |
2984 | 0 | goto done; |
2985 | 0 | } |
2986 | | |
2987 | 0 | res = 0; |
2988 | |
|
2989 | 0 | done: |
2990 | 0 | #ifndef MS_WINDOWS |
2991 | 0 | if (loc != NULL) { |
2992 | 0 | setlocale(LC_CTYPE, oldloc); |
2993 | 0 | } |
2994 | 0 | PyMem_Free(oldloc); |
2995 | 0 | #endif |
2996 | 0 | return res; |
2997 | |
|
2998 | 0 | #undef GET_LOCALE_STRING |
2999 | 0 | } |
3000 | | |
3001 | | /* Our selection logic for which function to use is as follows: |
3002 | | * 1. If close_range(2) is available, always prefer that; it's better for |
3003 | | * contiguous ranges like this than fdwalk(3) which entails iterating over |
3004 | | * the entire fd space and simply doing nothing for those outside the range. |
3005 | | * 2. If closefrom(2) is available, we'll attempt to use that next if we're |
3006 | | * closing up to sysconf(_SC_OPEN_MAX). |
3007 | | * 2a. Fallback to fdwalk(3) if we're not closing up to sysconf(_SC_OPEN_MAX), |
3008 | | * as that will be more performant if the range happens to have any chunk of |
3009 | | * non-opened fd in the middle. |
3010 | | * 2b. If fdwalk(3) isn't available, just do a plain close(2) loop. |
3011 | | */ |
3012 | | #ifdef HAVE_CLOSEFROM |
3013 | | # define USE_CLOSEFROM |
3014 | | #endif /* HAVE_CLOSEFROM */ |
3015 | | |
3016 | | #ifdef HAVE_FDWALK |
3017 | | # define USE_FDWALK |
3018 | | #endif /* HAVE_FDWALK */ |
3019 | | |
3020 | | #ifdef USE_FDWALK |
3021 | | static int |
3022 | | _fdwalk_close_func(void *lohi, int fd) |
3023 | | { |
3024 | | int lo = ((int *)lohi)[0]; |
3025 | | int hi = ((int *)lohi)[1]; |
3026 | | |
3027 | | if (fd >= hi) { |
3028 | | return 1; |
3029 | | } |
3030 | | else if (fd >= lo) { |
3031 | | /* Ignore errors */ |
3032 | | (void)close(fd); |
3033 | | } |
3034 | | return 0; |
3035 | | } |
3036 | | #endif /* USE_FDWALK */ |
3037 | | |
3038 | | /* Closes all file descriptors in [first, last], ignoring errors. */ |
3039 | | void |
3040 | | _Py_closerange(int first, int last) |
3041 | 0 | { |
3042 | 0 | first = Py_MAX(first, 0); |
3043 | 0 | _Py_BEGIN_SUPPRESS_IPH |
3044 | | #ifdef HAVE_CLOSE_RANGE |
3045 | | if (close_range(first, last, 0) == 0) { |
3046 | | /* close_range() ignores errors when it closes file descriptors. |
3047 | | * Possible reasons of an error return are lack of kernel support |
3048 | | * or denial of the underlying syscall by a seccomp sandbox on Linux. |
3049 | | * Fallback to other methods in case of any error. */ |
3050 | | } |
3051 | | else |
3052 | | #endif /* HAVE_CLOSE_RANGE */ |
3053 | | #ifdef USE_CLOSEFROM |
3054 | | if (last >= sysconf(_SC_OPEN_MAX)) { |
3055 | | /* Any errors encountered while closing file descriptors are ignored */ |
3056 | | (void)closefrom(first); |
3057 | | } |
3058 | | else |
3059 | | #endif /* USE_CLOSEFROM */ |
3060 | | #ifdef USE_FDWALK |
3061 | | { |
3062 | | int lohi[2]; |
3063 | | lohi[0] = first; |
3064 | | lohi[1] = last + 1; |
3065 | | fdwalk(_fdwalk_close_func, lohi); |
3066 | | } |
3067 | | #else |
3068 | 0 | { |
3069 | 0 | for (int i = first; i <= last; i++) { |
3070 | | /* Ignore errors */ |
3071 | 0 | (void)close(i); |
3072 | 0 | } |
3073 | 0 | } |
3074 | 0 | #endif /* USE_FDWALK */ |
3075 | 0 | _Py_END_SUPPRESS_IPH |
3076 | 0 | } |
3077 | | |
3078 | | |
3079 | | #ifndef MS_WINDOWS |
3080 | | // Ticks per second used by clock() and times() functions. |
3081 | | // See os.times() and time.process_time() implementations. |
3082 | | int |
3083 | | _Py_GetTicksPerSecond(long *ticks_per_second) |
3084 | 32 | { |
3085 | 32 | #if defined(HAVE_SYSCONF) && defined(_SC_CLK_TCK) |
3086 | 32 | long value = sysconf(_SC_CLK_TCK); |
3087 | 32 | if (value < 1) { |
3088 | 0 | return -1; |
3089 | 0 | } |
3090 | 32 | *ticks_per_second = value; |
3091 | | #elif defined(HZ) |
3092 | | assert(HZ >= 1); |
3093 | | *ticks_per_second = HZ; |
3094 | | #else |
3095 | | // Magic fallback value; may be bogus |
3096 | | *ticks_per_second = 60; |
3097 | | #endif |
3098 | 32 | return 0; |
3099 | 32 | } |
3100 | | #endif |
3101 | | |
3102 | | |
3103 | | /* Check if a file descriptor is valid or not. |
3104 | | Return 0 if the file descriptor is invalid, return non-zero otherwise. */ |
3105 | | int |
3106 | | _Py_IsValidFD(int fd) |
3107 | 48 | { |
3108 | | /* dup() is faster than fstat(): fstat() can require input/output operations, |
3109 | | whereas dup() doesn't. There is a low risk of EMFILE/ENFILE at Python |
3110 | | startup. Problem: dup() doesn't check if the file descriptor is valid on |
3111 | | some platforms. |
3112 | | |
3113 | | fcntl(fd, F_GETFD) is even faster, because it only checks the process table. |
3114 | | It is preferred over dup() when available, since it cannot fail with the |
3115 | | "too many open files" error (EMFILE). |
3116 | | |
3117 | | bpo-30225: On macOS Tiger, when stdout is redirected to a pipe and the other |
3118 | | side of the pipe is closed, dup(1) succeed, whereas fstat(1, &st) fails with |
3119 | | EBADF. FreeBSD has similar issue (bpo-32849). |
3120 | | |
3121 | | Only use dup() on Linux where dup() is enough to detect invalid FD |
3122 | | (bpo-32849). |
3123 | | */ |
3124 | 48 | if (fd < 0) { |
3125 | 0 | return 0; |
3126 | 0 | } |
3127 | 48 | #if defined(F_GETFD) && ( \ |
3128 | 48 | defined(__linux__) || \ |
3129 | 48 | defined(__APPLE__) || \ |
3130 | 48 | (defined(__wasm__) && !defined(__wasi__))) |
3131 | 48 | return fcntl(fd, F_GETFD) >= 0; |
3132 | | #elif defined(__linux__) |
3133 | | int fd2 = dup(fd); |
3134 | | if (fd2 >= 0) { |
3135 | | close(fd2); |
3136 | | } |
3137 | | return (fd2 >= 0); |
3138 | | #elif defined(MS_WINDOWS) |
3139 | | HANDLE hfile; |
3140 | | _Py_BEGIN_SUPPRESS_IPH |
3141 | | hfile = (HANDLE)_get_osfhandle(fd); |
3142 | | _Py_END_SUPPRESS_IPH |
3143 | | return (hfile != INVALID_HANDLE_VALUE |
3144 | | && GetFileType(hfile) != FILE_TYPE_UNKNOWN); |
3145 | | #else |
3146 | | struct stat st; |
3147 | | return (fstat(fd, &st) == 0); |
3148 | | #endif |
3149 | 48 | } |