1 | /* Copyright (C) 1993-2024 Free Software Foundation, Inc. |
2 | This file is part of the GNU C Library. |
3 | |
4 | The GNU C Library is free software; you can redistribute it and/or |
5 | modify it under the terms of the GNU Lesser General Public |
6 | License as published by the Free Software Foundation; either |
7 | version 2.1 of the License, or (at your option) any later version. |
8 | |
9 | The GNU C Library is distributed in the hope that it will be useful, |
10 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
11 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
12 | Lesser General Public License for more details. |
13 | |
14 | You should have received a copy of the GNU Lesser General Public |
15 | License along with the GNU C Library; if not, see |
16 | <https://www.gnu.org/licenses/>. |
17 | |
18 | As a special exception, if you link the code in this file with |
19 | files compiled with a GNU compiler to produce an executable, |
20 | that does not cause the resulting executable to be covered by |
21 | the GNU Lesser General Public License. This exception does not |
22 | however invalidate any other reasons why the executable file |
23 | might be covered by the GNU Lesser General Public License. |
24 | This exception applies to code released by its copyright holders |
25 | in files containing the exception. */ |
26 | |
27 | #include <assert.h> |
28 | #include <libioP.h> |
29 | #include <wchar.h> |
30 | #include <gconv.h> |
31 | #include <stdlib.h> |
32 | #include <string.h> |
33 | |
34 | /* Convert TO_DO wide character from DATA to FP. |
35 | Then mark FP as having empty buffers. */ |
36 | int |
37 | _IO_wdo_write (FILE *fp, const wchar_t *data, size_t to_do) |
38 | { |
39 | struct _IO_codecvt *cc = fp->_codecvt; |
40 | |
41 | if (to_do > 0) |
42 | { |
43 | if (fp->_IO_write_end == fp->_IO_write_ptr |
44 | && fp->_IO_write_end != fp->_IO_write_base) |
45 | { |
46 | if (_IO_new_do_write (fp, fp->_IO_write_base, |
47 | fp->_IO_write_ptr - fp->_IO_write_base) == EOF) |
48 | return WEOF; |
49 | } |
50 | |
51 | do |
52 | { |
53 | enum __codecvt_result result; |
54 | const wchar_t *new_data; |
55 | char mb_buf[MB_LEN_MAX]; |
56 | char *write_base, *write_ptr, *buf_end; |
57 | |
58 | if (fp->_IO_buf_end - fp->_IO_write_ptr < sizeof (mb_buf)) |
59 | { |
60 | /* Make sure we have room for at least one multibyte |
61 | character. */ |
62 | write_ptr = write_base = mb_buf; |
63 | buf_end = mb_buf + sizeof (mb_buf); |
64 | } |
65 | else |
66 | { |
67 | write_ptr = fp->_IO_write_ptr; |
68 | write_base = fp->_IO_write_base; |
69 | buf_end = fp->_IO_buf_end; |
70 | } |
71 | |
72 | /* Now convert from the internal format into the external buffer. */ |
73 | result = __libio_codecvt_out (cc, &fp->_wide_data->_IO_state, |
74 | data, data + to_do, &new_data, |
75 | write_ptr, |
76 | buf_end, |
77 | &write_ptr); |
78 | |
79 | /* Write out what we produced so far. */ |
80 | if (_IO_new_do_write (fp, write_base, write_ptr - write_base) == EOF) |
81 | /* Something went wrong. */ |
82 | return WEOF; |
83 | |
84 | to_do -= new_data - data; |
85 | |
86 | /* Next see whether we had problems during the conversion. If yes, |
87 | we cannot go on. */ |
88 | if (result != __codecvt_ok |
89 | && (result != __codecvt_partial || new_data - data == 0)) |
90 | break; |
91 | |
92 | data = new_data; |
93 | } |
94 | while (to_do > 0); |
95 | } |
96 | |
97 | _IO_wsetg (fp, fp->_wide_data->_IO_buf_base, fp->_wide_data->_IO_buf_base, |
98 | fp->_wide_data->_IO_buf_base); |
99 | fp->_wide_data->_IO_write_base = fp->_wide_data->_IO_write_ptr |
100 | = fp->_wide_data->_IO_buf_base; |
101 | fp->_wide_data->_IO_write_end = ((fp->_flags & (_IO_LINE_BUF | _IO_UNBUFFERED)) |
102 | ? fp->_wide_data->_IO_buf_base |
103 | : fp->_wide_data->_IO_buf_end); |
104 | |
105 | return to_do == 0 ? 0 : WEOF; |
106 | } |
107 | libc_hidden_def (_IO_wdo_write) |
108 | |
109 | |
110 | wint_t |
111 | _IO_wfile_underflow (FILE *fp) |
112 | { |
113 | struct _IO_codecvt *cd; |
114 | enum __codecvt_result status; |
115 | ssize_t count; |
116 | |
117 | /* C99 requires EOF to be "sticky". */ |
118 | if (fp->_flags & _IO_EOF_SEEN) |
119 | return WEOF; |
120 | |
121 | if (__glibc_unlikely (fp->_flags & _IO_NO_READS)) |
122 | { |
123 | fp->_flags |= _IO_ERR_SEEN; |
124 | __set_errno (EBADF); |
125 | return WEOF; |
126 | } |
127 | if (fp->_wide_data->_IO_read_ptr < fp->_wide_data->_IO_read_end) |
128 | return *fp->_wide_data->_IO_read_ptr; |
129 | |
130 | cd = fp->_codecvt; |
131 | |
132 | /* Maybe there is something left in the external buffer. */ |
133 | if (fp->_IO_read_ptr < fp->_IO_read_end) |
134 | { |
135 | /* There is more in the external. Convert it. */ |
136 | const char *read_stop = (const char *) fp->_IO_read_ptr; |
137 | |
138 | fp->_wide_data->_IO_last_state = fp->_wide_data->_IO_state; |
139 | fp->_wide_data->_IO_read_base = fp->_wide_data->_IO_read_ptr = |
140 | fp->_wide_data->_IO_buf_base; |
141 | status = __libio_codecvt_in (cd, &fp->_wide_data->_IO_state, |
142 | fp->_IO_read_ptr, fp->_IO_read_end, |
143 | &read_stop, |
144 | fp->_wide_data->_IO_read_ptr, |
145 | fp->_wide_data->_IO_buf_end, |
146 | &fp->_wide_data->_IO_read_end); |
147 | |
148 | fp->_IO_read_base = fp->_IO_read_ptr; |
149 | fp->_IO_read_ptr = (char *) read_stop; |
150 | |
151 | /* If we managed to generate some text return the next character. */ |
152 | if (fp->_wide_data->_IO_read_ptr < fp->_wide_data->_IO_read_end) |
153 | return *fp->_wide_data->_IO_read_ptr; |
154 | |
155 | if (status == __codecvt_error) |
156 | { |
157 | __set_errno (EILSEQ); |
158 | fp->_flags |= _IO_ERR_SEEN; |
159 | return WEOF; |
160 | } |
161 | |
162 | /* Move the remaining content of the read buffer to the beginning. */ |
163 | memmove (fp->_IO_buf_base, fp->_IO_read_ptr, |
164 | fp->_IO_read_end - fp->_IO_read_ptr); |
165 | fp->_IO_read_end = (fp->_IO_buf_base |
166 | + (fp->_IO_read_end - fp->_IO_read_ptr)); |
167 | fp->_IO_read_base = fp->_IO_read_ptr = fp->_IO_buf_base; |
168 | } |
169 | else |
170 | fp->_IO_read_base = fp->_IO_read_ptr = fp->_IO_read_end = |
171 | fp->_IO_buf_base; |
172 | |
173 | if (fp->_IO_buf_base == NULL) |
174 | { |
175 | /* Maybe we already have a push back pointer. */ |
176 | if (fp->_IO_save_base != NULL) |
177 | { |
178 | free (ptr: fp->_IO_save_base); |
179 | fp->_flags &= ~_IO_IN_BACKUP; |
180 | } |
181 | _IO_doallocbuf (fp); |
182 | |
183 | fp->_IO_read_base = fp->_IO_read_ptr = fp->_IO_read_end = |
184 | fp->_IO_buf_base; |
185 | } |
186 | |
187 | fp->_IO_write_base = fp->_IO_write_ptr = fp->_IO_write_end = |
188 | fp->_IO_buf_base; |
189 | |
190 | if (fp->_wide_data->_IO_buf_base == NULL) |
191 | { |
192 | /* Maybe we already have a push back pointer. */ |
193 | if (fp->_wide_data->_IO_save_base != NULL) |
194 | { |
195 | free (ptr: fp->_wide_data->_IO_save_base); |
196 | fp->_flags &= ~_IO_IN_BACKUP; |
197 | } |
198 | _IO_wdoallocbuf (fp); |
199 | } |
200 | |
201 | /* FIXME This can/should be moved to genops ?? */ |
202 | if (fp->_flags & (_IO_LINE_BUF | _IO_UNBUFFERED)) |
203 | { |
204 | /* We used to flush all line-buffered stream. This really isn't |
205 | required by any standard. My recollection is that |
206 | traditional Unix systems did this for stdout. stderr better |
207 | not be line buffered. So we do just that here |
208 | explicitly. --drepper */ |
209 | _IO_acquire_lock (stdout); |
210 | |
211 | if ((stdout->_flags & (_IO_LINKED | _IO_NO_WRITES | _IO_LINE_BUF)) |
212 | == (_IO_LINKED | _IO_LINE_BUF)) |
213 | _IO_OVERFLOW (stdout, EOF); |
214 | |
215 | _IO_release_lock (stdout); |
216 | } |
217 | |
218 | _IO_switch_to_get_mode (fp); |
219 | |
220 | fp->_wide_data->_IO_read_base = fp->_wide_data->_IO_read_ptr = |
221 | fp->_wide_data->_IO_buf_base; |
222 | fp->_wide_data->_IO_read_end = fp->_wide_data->_IO_buf_base; |
223 | fp->_wide_data->_IO_write_base = fp->_wide_data->_IO_write_ptr = |
224 | fp->_wide_data->_IO_write_end = fp->_wide_data->_IO_buf_base; |
225 | |
226 | const char *read_ptr_copy; |
227 | char accbuf[MB_LEN_MAX]; |
228 | size_t naccbuf = 0; |
229 | again: |
230 | count = _IO_SYSREAD (fp, fp->_IO_read_end, |
231 | fp->_IO_buf_end - fp->_IO_read_end); |
232 | if (count <= 0) |
233 | { |
234 | if (count == 0 && naccbuf == 0) |
235 | { |
236 | fp->_flags |= _IO_EOF_SEEN; |
237 | fp->_offset = _IO_pos_BAD; |
238 | } |
239 | else |
240 | fp->_flags |= _IO_ERR_SEEN, count = 0; |
241 | } |
242 | fp->_IO_read_end += count; |
243 | if (count == 0) |
244 | { |
245 | if (naccbuf != 0) |
246 | /* There are some bytes in the external buffer but they don't |
247 | convert to anything. */ |
248 | __set_errno (EILSEQ); |
249 | return WEOF; |
250 | } |
251 | if (fp->_offset != _IO_pos_BAD) |
252 | _IO_pos_adjust (fp->_offset, count); |
253 | |
254 | /* Now convert the read input. */ |
255 | fp->_wide_data->_IO_last_state = fp->_wide_data->_IO_state; |
256 | fp->_IO_read_base = fp->_IO_read_ptr; |
257 | const char *from = fp->_IO_read_ptr; |
258 | const char *to = fp->_IO_read_end; |
259 | size_t to_copy = count; |
260 | if (__glibc_unlikely (naccbuf != 0)) |
261 | { |
262 | to_copy = MIN (sizeof (accbuf) - naccbuf, count); |
263 | to = __mempcpy (&accbuf[naccbuf], from, to_copy); |
264 | naccbuf += to_copy; |
265 | from = accbuf; |
266 | } |
267 | status = __libio_codecvt_in (cd, &fp->_wide_data->_IO_state, |
268 | from, to, &read_ptr_copy, |
269 | fp->_wide_data->_IO_read_end, |
270 | fp->_wide_data->_IO_buf_end, |
271 | &fp->_wide_data->_IO_read_end); |
272 | |
273 | if (__glibc_unlikely (naccbuf != 0)) |
274 | fp->_IO_read_ptr += MAX (0, read_ptr_copy - &accbuf[naccbuf - to_copy]); |
275 | else |
276 | fp->_IO_read_ptr = (char *) read_ptr_copy; |
277 | if (fp->_wide_data->_IO_read_end == fp->_wide_data->_IO_buf_base) |
278 | { |
279 | if (status == __codecvt_error) |
280 | { |
281 | out_eilseq: |
282 | __set_errno (EILSEQ); |
283 | fp->_flags |= _IO_ERR_SEEN; |
284 | return WEOF; |
285 | } |
286 | |
287 | /* The read bytes make no complete character. Try reading again. */ |
288 | assert (status == __codecvt_partial); |
289 | |
290 | if (naccbuf == 0) |
291 | { |
292 | if (fp->_IO_read_base < fp->_IO_read_ptr) |
293 | { |
294 | /* Partially used the buffer for some input data that |
295 | produces no output. */ |
296 | size_t avail = fp->_IO_read_end - fp->_IO_read_ptr; |
297 | memmove (fp->_IO_read_base, fp->_IO_read_ptr, avail); |
298 | fp->_IO_read_ptr = fp->_IO_read_base; |
299 | fp->_IO_read_end -= avail; |
300 | goto again; |
301 | } |
302 | naccbuf = fp->_IO_read_end - fp->_IO_read_ptr; |
303 | if (naccbuf >= sizeof (accbuf)) |
304 | goto out_eilseq; |
305 | |
306 | memcpy (accbuf, fp->_IO_read_ptr, naccbuf); |
307 | } |
308 | else |
309 | { |
310 | size_t used = read_ptr_copy - accbuf; |
311 | if (used > 0) |
312 | { |
313 | memmove (accbuf, read_ptr_copy, naccbuf - used); |
314 | naccbuf -= used; |
315 | } |
316 | |
317 | if (naccbuf == sizeof (accbuf)) |
318 | goto out_eilseq; |
319 | } |
320 | |
321 | fp->_IO_read_ptr = fp->_IO_read_end = fp->_IO_read_base; |
322 | |
323 | goto again; |
324 | } |
325 | |
326 | return *fp->_wide_data->_IO_read_ptr; |
327 | } |
328 | libc_hidden_def (_IO_wfile_underflow) |
329 | |
330 | |
331 | wint_t |
332 | _IO_wfile_underflow_mmap (FILE *fp) |
333 | { |
334 | struct _IO_codecvt *cd; |
335 | const char *read_stop; |
336 | |
337 | if (__glibc_unlikely (fp->_flags & _IO_NO_READS)) |
338 | { |
339 | fp->_flags |= _IO_ERR_SEEN; |
340 | __set_errno (EBADF); |
341 | return WEOF; |
342 | } |
343 | if (fp->_wide_data->_IO_read_ptr < fp->_wide_data->_IO_read_end) |
344 | return *fp->_wide_data->_IO_read_ptr; |
345 | |
346 | cd = fp->_codecvt; |
347 | |
348 | /* Maybe there is something left in the external buffer. */ |
349 | if (fp->_IO_read_ptr >= fp->_IO_read_end |
350 | /* No. But maybe the read buffer is not fully set up. */ |
351 | && _IO_file_underflow_mmap (fp) == EOF) |
352 | /* Nothing available. _IO_file_underflow_mmap has set the EOF or error |
353 | flags as appropriate. */ |
354 | return WEOF; |
355 | |
356 | /* There is more in the external. Convert it. */ |
357 | read_stop = (const char *) fp->_IO_read_ptr; |
358 | |
359 | if (fp->_wide_data->_IO_buf_base == NULL) |
360 | { |
361 | /* Maybe we already have a push back pointer. */ |
362 | if (fp->_wide_data->_IO_save_base != NULL) |
363 | { |
364 | free (ptr: fp->_wide_data->_IO_save_base); |
365 | fp->_flags &= ~_IO_IN_BACKUP; |
366 | } |
367 | _IO_wdoallocbuf (fp); |
368 | } |
369 | |
370 | fp->_wide_data->_IO_last_state = fp->_wide_data->_IO_state; |
371 | fp->_wide_data->_IO_read_base = fp->_wide_data->_IO_read_ptr = |
372 | fp->_wide_data->_IO_buf_base; |
373 | __libio_codecvt_in (cd, &fp->_wide_data->_IO_state, |
374 | fp->_IO_read_ptr, fp->_IO_read_end, |
375 | &read_stop, |
376 | fp->_wide_data->_IO_read_ptr, |
377 | fp->_wide_data->_IO_buf_end, |
378 | &fp->_wide_data->_IO_read_end); |
379 | |
380 | fp->_IO_read_ptr = (char *) read_stop; |
381 | |
382 | /* If we managed to generate some text return the next character. */ |
383 | if (fp->_wide_data->_IO_read_ptr < fp->_wide_data->_IO_read_end) |
384 | return *fp->_wide_data->_IO_read_ptr; |
385 | |
386 | /* There is some garbage at the end of the file. */ |
387 | __set_errno (EILSEQ); |
388 | fp->_flags |= _IO_ERR_SEEN; |
389 | return WEOF; |
390 | } |
391 | |
392 | wint_t |
393 | _IO_wfile_underflow_maybe_mmap (FILE *fp) |
394 | { |
395 | /* This is the first read attempt. Doing the underflow will choose mmap |
396 | or vanilla operations and then punt to the chosen underflow routine. |
397 | Then we can punt to ours. */ |
398 | if (_IO_file_underflow_maybe_mmap (fp) == EOF) |
399 | return WEOF; |
400 | |
401 | return _IO_WUNDERFLOW (fp); |
402 | } |
403 | |
404 | |
405 | wint_t |
406 | _IO_wfile_overflow (FILE *f, wint_t wch) |
407 | { |
408 | if (f->_flags & _IO_NO_WRITES) /* SET ERROR */ |
409 | { |
410 | f->_flags |= _IO_ERR_SEEN; |
411 | __set_errno (EBADF); |
412 | return WEOF; |
413 | } |
414 | /* If currently reading or no buffer allocated. */ |
415 | if ((f->_flags & _IO_CURRENTLY_PUTTING) == 0 |
416 | || f->_wide_data->_IO_write_base == NULL) |
417 | { |
418 | /* Allocate a buffer if needed. */ |
419 | if (f->_wide_data->_IO_write_base == 0) |
420 | { |
421 | _IO_wdoallocbuf (f); |
422 | _IO_free_wbackup_area (f); |
423 | _IO_wsetg (f, f->_wide_data->_IO_buf_base, |
424 | f->_wide_data->_IO_buf_base, f->_wide_data->_IO_buf_base); |
425 | |
426 | if (f->_IO_write_base == NULL) |
427 | { |
428 | _IO_doallocbuf (f); |
429 | _IO_setg (f, f->_IO_buf_base, f->_IO_buf_base, f->_IO_buf_base); |
430 | } |
431 | } |
432 | else |
433 | { |
434 | /* Otherwise must be currently reading. If _IO_read_ptr |
435 | (and hence also _IO_read_end) is at the buffer end, |
436 | logically slide the buffer forwards one block (by setting |
437 | the read pointers to all point at the beginning of the |
438 | block). This makes room for subsequent output. |
439 | Otherwise, set the read pointers to _IO_read_end (leaving |
440 | that alone, so it can continue to correspond to the |
441 | external position). */ |
442 | if (f->_wide_data->_IO_read_ptr == f->_wide_data->_IO_buf_end) |
443 | { |
444 | f->_IO_read_end = f->_IO_read_ptr = f->_IO_buf_base; |
445 | f->_wide_data->_IO_read_end = f->_wide_data->_IO_read_ptr = |
446 | f->_wide_data->_IO_buf_base; |
447 | } |
448 | } |
449 | f->_wide_data->_IO_write_ptr = f->_wide_data->_IO_read_ptr; |
450 | f->_wide_data->_IO_write_base = f->_wide_data->_IO_write_ptr; |
451 | f->_wide_data->_IO_write_end = f->_wide_data->_IO_buf_end; |
452 | f->_wide_data->_IO_read_base = f->_wide_data->_IO_read_ptr = |
453 | f->_wide_data->_IO_read_end; |
454 | |
455 | f->_IO_write_ptr = f->_IO_read_ptr; |
456 | f->_IO_write_base = f->_IO_write_ptr; |
457 | f->_IO_write_end = f->_IO_buf_end; |
458 | f->_IO_read_base = f->_IO_read_ptr = f->_IO_read_end; |
459 | |
460 | f->_flags |= _IO_CURRENTLY_PUTTING; |
461 | if (f->_flags & (_IO_LINE_BUF | _IO_UNBUFFERED)) |
462 | f->_wide_data->_IO_write_end = f->_wide_data->_IO_write_ptr; |
463 | } |
464 | if (wch == WEOF) |
465 | return _IO_do_flush (f); |
466 | if (f->_wide_data->_IO_write_ptr == f->_wide_data->_IO_buf_end) |
467 | /* Buffer is really full */ |
468 | if (_IO_do_flush (f) == EOF) |
469 | return WEOF; |
470 | *f->_wide_data->_IO_write_ptr++ = wch; |
471 | if ((f->_flags & _IO_UNBUFFERED) |
472 | || ((f->_flags & _IO_LINE_BUF) && wch == L'\n')) |
473 | if (_IO_do_flush (f) == EOF) |
474 | return WEOF; |
475 | return wch; |
476 | } |
477 | libc_hidden_def (_IO_wfile_overflow) |
478 | |
479 | wint_t |
480 | _IO_wfile_sync (FILE *fp) |
481 | { |
482 | ssize_t delta; |
483 | wint_t retval = 0; |
484 | |
485 | /* char* ptr = cur_ptr(); */ |
486 | if (fp->_wide_data->_IO_write_ptr > fp->_wide_data->_IO_write_base) |
487 | if (_IO_do_flush (fp)) |
488 | return WEOF; |
489 | delta = fp->_wide_data->_IO_read_ptr - fp->_wide_data->_IO_read_end; |
490 | if (delta != 0) |
491 | { |
492 | /* We have to find out how many bytes we have to go back in the |
493 | external buffer. */ |
494 | struct _IO_codecvt *cv = fp->_codecvt; |
495 | off64_t new_pos; |
496 | |
497 | int clen = __libio_codecvt_encoding (cv); |
498 | |
499 | if (clen > 0) |
500 | /* It is easy, a fixed number of input bytes are used for each |
501 | wide character. */ |
502 | delta *= clen; |
503 | else |
504 | { |
505 | /* We have to find out the hard way how much to back off. |
506 | To do this we determine how much input we needed to |
507 | generate the wide characters up to the current reading |
508 | position. */ |
509 | int nread; |
510 | size_t wnread = (fp->_wide_data->_IO_read_ptr |
511 | - fp->_wide_data->_IO_read_base); |
512 | fp->_wide_data->_IO_state = fp->_wide_data->_IO_last_state; |
513 | nread = __libio_codecvt_length (cv, &fp->_wide_data->_IO_state, |
514 | fp->_IO_read_base, |
515 | fp->_IO_read_end, wnread); |
516 | fp->_IO_read_ptr = fp->_IO_read_base + nread; |
517 | delta = -(fp->_IO_read_end - fp->_IO_read_base - nread); |
518 | } |
519 | |
520 | new_pos = _IO_SYSSEEK (fp, delta, 1); |
521 | if (new_pos != (off64_t) EOF) |
522 | { |
523 | fp->_wide_data->_IO_read_end = fp->_wide_data->_IO_read_ptr; |
524 | fp->_IO_read_end = fp->_IO_read_ptr; |
525 | } |
526 | else if (errno == ESPIPE) |
527 | ; /* Ignore error from unseekable devices. */ |
528 | else |
529 | retval = WEOF; |
530 | } |
531 | if (retval != WEOF) |
532 | fp->_offset = _IO_pos_BAD; |
533 | /* FIXME: Cleanup - can this be shared? */ |
534 | /* setg(base(), ptr, ptr); */ |
535 | return retval; |
536 | } |
537 | libc_hidden_def (_IO_wfile_sync) |
538 | |
539 | /* Adjust the internal buffer pointers to reflect the state in the external |
540 | buffer. The content between fp->_IO_read_base and fp->_IO_read_ptr is |
541 | assumed to be converted and available in the range |
542 | fp->_wide_data->_IO_read_base and fp->_wide_data->_IO_read_end. |
543 | |
544 | Returns 0 on success and -1 on error with the _IO_ERR_SEEN flag set. */ |
545 | static int |
546 | adjust_wide_data (FILE *fp, bool do_convert) |
547 | { |
548 | struct _IO_codecvt *cv = fp->_codecvt; |
549 | |
550 | int clen = __libio_codecvt_encoding (cv); |
551 | |
552 | /* Take the easy way out for constant length encodings if we don't need to |
553 | convert. */ |
554 | if (!do_convert && clen > 0) |
555 | { |
556 | fp->_wide_data->_IO_read_end += ((fp->_IO_read_ptr - fp->_IO_read_base) |
557 | / clen); |
558 | goto done; |
559 | } |
560 | |
561 | enum __codecvt_result status; |
562 | const char *read_stop = (const char *) fp->_IO_read_base; |
563 | do |
564 | { |
565 | |
566 | fp->_wide_data->_IO_last_state = fp->_wide_data->_IO_state; |
567 | status = __libio_codecvt_in (cv, &fp->_wide_data->_IO_state, |
568 | fp->_IO_read_base, fp->_IO_read_ptr, |
569 | &read_stop, |
570 | fp->_wide_data->_IO_read_base, |
571 | fp->_wide_data->_IO_buf_end, |
572 | &fp->_wide_data->_IO_read_end); |
573 | |
574 | /* Should we return EILSEQ? */ |
575 | if (__glibc_unlikely (status == __codecvt_error)) |
576 | { |
577 | fp->_flags |= _IO_ERR_SEEN; |
578 | return -1; |
579 | } |
580 | } |
581 | while (__builtin_expect (status == __codecvt_partial, 0)); |
582 | |
583 | done: |
584 | /* Now seek to _IO_read_end to behave as if we have read it all in. */ |
585 | fp->_wide_data->_IO_read_ptr = fp->_wide_data->_IO_read_end; |
586 | |
587 | return 0; |
588 | } |
589 | |
590 | /* ftell{,o} implementation for wide mode. Don't modify any state of the file |
591 | pointer while we try to get the current state of the stream except in one |
592 | case, which is when we have unflushed writes in append mode. */ |
593 | static off64_t |
594 | do_ftell_wide (FILE *fp) |
595 | { |
596 | off64_t result, offset = 0; |
597 | |
598 | /* No point looking for offsets in the buffer if it hasn't even been |
599 | allocated. */ |
600 | if (fp->_wide_data->_IO_buf_base != NULL) |
601 | { |
602 | const wchar_t *wide_read_base; |
603 | const wchar_t *wide_read_ptr; |
604 | const wchar_t *wide_read_end; |
605 | bool unflushed_writes = (fp->_wide_data->_IO_write_ptr |
606 | > fp->_wide_data->_IO_write_base); |
607 | |
608 | bool append_mode = (fp->_flags & _IO_IS_APPENDING) == _IO_IS_APPENDING; |
609 | |
610 | /* When we have unflushed writes in append mode, seek to the end of the |
611 | file and record that offset. This is the only time we change the file |
612 | stream state and it is safe since the file handle is active. */ |
613 | if (unflushed_writes && append_mode) |
614 | { |
615 | result = _IO_SYSSEEK (fp, 0, _IO_seek_end); |
616 | if (result == _IO_pos_BAD) |
617 | return EOF; |
618 | else |
619 | fp->_offset = result; |
620 | } |
621 | |
622 | /* XXX For wide stream with backup store it is not very |
623 | reasonable to determine the offset. The pushed-back |
624 | character might require a state change and we need not be |
625 | able to compute the initial state by reverse transformation |
626 | since there is no guarantee of symmetry. So we don't even |
627 | try and return an error. */ |
628 | if (_IO_in_backup (fp)) |
629 | { |
630 | if (fp->_wide_data->_IO_read_ptr < fp->_wide_data->_IO_read_end) |
631 | { |
632 | __set_errno (EINVAL); |
633 | return -1; |
634 | } |
635 | |
636 | /* Nothing in the backup store, so note the backed up pointers |
637 | without changing the state. */ |
638 | wide_read_base = fp->_wide_data->_IO_save_base; |
639 | wide_read_ptr = wide_read_base; |
640 | wide_read_end = fp->_wide_data->_IO_save_end; |
641 | } |
642 | else |
643 | { |
644 | wide_read_base = fp->_wide_data->_IO_read_base; |
645 | wide_read_ptr = fp->_wide_data->_IO_read_ptr; |
646 | wide_read_end = fp->_wide_data->_IO_read_end; |
647 | } |
648 | |
649 | struct _IO_codecvt *cv = fp->_codecvt; |
650 | int clen = __libio_codecvt_encoding (cv); |
651 | |
652 | if (!unflushed_writes) |
653 | { |
654 | if (clen > 0) |
655 | { |
656 | offset -= (wide_read_end - wide_read_ptr) * clen; |
657 | offset -= fp->_IO_read_end - fp->_IO_read_ptr; |
658 | } |
659 | else |
660 | { |
661 | int nread; |
662 | |
663 | size_t delta = wide_read_ptr - wide_read_base; |
664 | __mbstate_t state = fp->_wide_data->_IO_last_state; |
665 | nread = __libio_codecvt_length (cv, &state, |
666 | fp->_IO_read_base, |
667 | fp->_IO_read_end, delta); |
668 | offset -= fp->_IO_read_end - fp->_IO_read_base - nread; |
669 | } |
670 | } |
671 | else |
672 | { |
673 | if (clen > 0) |
674 | offset += (fp->_wide_data->_IO_write_ptr |
675 | - fp->_wide_data->_IO_write_base) * clen; |
676 | else |
677 | { |
678 | size_t delta = (fp->_wide_data->_IO_write_ptr |
679 | - fp->_wide_data->_IO_write_base); |
680 | |
681 | /* Allocate enough space for the conversion. */ |
682 | size_t outsize = delta * sizeof (wchar_t); |
683 | char *out = malloc (size: outsize); |
684 | char *outstop = out; |
685 | const wchar_t *in = fp->_wide_data->_IO_write_base; |
686 | |
687 | enum __codecvt_result status; |
688 | |
689 | __mbstate_t state = fp->_wide_data->_IO_last_state; |
690 | status = __libio_codecvt_out (cv, &state, in, in + delta, &in, |
691 | out, out + outsize, &outstop); |
692 | |
693 | /* We don't check for __codecvt_partial because it can be |
694 | returned on one of two conditions: either the output |
695 | buffer is full or the input sequence is incomplete. We |
696 | take care to allocate enough buffer and our input |
697 | sequences must be complete since they are accepted as |
698 | wchar_t; if not, then that is an error. */ |
699 | if (__glibc_unlikely (status != __codecvt_ok)) |
700 | { |
701 | free (ptr: out); |
702 | return WEOF; |
703 | } |
704 | |
705 | offset += outstop - out; |
706 | free (ptr: out); |
707 | } |
708 | |
709 | /* We don't trust _IO_read_end to represent the current file offset |
710 | when writing in append mode because the value would have to be |
711 | shifted to the end of the file during a flush. Use the write base |
712 | instead, along with the new offset we got above when we did a seek |
713 | to the end of the file. */ |
714 | if (append_mode) |
715 | offset += fp->_IO_write_ptr - fp->_IO_write_base; |
716 | /* For all other modes, _IO_read_end represents the file offset. */ |
717 | else |
718 | offset += fp->_IO_write_ptr - fp->_IO_read_end; |
719 | } |
720 | } |
721 | |
722 | if (fp->_offset != _IO_pos_BAD) |
723 | result = fp->_offset; |
724 | else |
725 | result = _IO_SYSSEEK (fp, 0, _IO_seek_cur); |
726 | |
727 | if (result == EOF) |
728 | return result; |
729 | |
730 | result += offset; |
731 | |
732 | if (result < 0) |
733 | { |
734 | __set_errno (EINVAL); |
735 | return EOF; |
736 | } |
737 | |
738 | return result; |
739 | } |
740 | |
741 | off64_t |
742 | _IO_wfile_seekoff (FILE *fp, off64_t offset, int dir, int mode) |
743 | { |
744 | off64_t result; |
745 | off64_t delta, new_offset; |
746 | long int count; |
747 | |
748 | /* Short-circuit into a separate function. We don't want to mix any |
749 | functionality and we don't want to touch anything inside the FILE |
750 | object. */ |
751 | if (mode == 0) |
752 | return do_ftell_wide (fp); |
753 | |
754 | /* POSIX.1 8.2.3.7 says that after a call the fflush() the file |
755 | offset of the underlying file must be exact. */ |
756 | int must_be_exact = ((fp->_wide_data->_IO_read_base |
757 | == fp->_wide_data->_IO_read_end) |
758 | && (fp->_wide_data->_IO_write_base |
759 | == fp->_wide_data->_IO_write_ptr)); |
760 | |
761 | bool was_writing = ((fp->_wide_data->_IO_write_ptr |
762 | > fp->_wide_data->_IO_write_base) |
763 | || _IO_in_put_mode (fp)); |
764 | |
765 | /* Flush unwritten characters. |
766 | (This may do an unneeded write if we seek within the buffer. |
767 | But to be able to switch to reading, we would need to set |
768 | egptr to pptr. That can't be done in the current design, |
769 | which assumes file_ptr() is eGptr. Anyway, since we probably |
770 | end up flushing when we close(), it doesn't make much difference.) |
771 | FIXME: simulate mem-mapped files. */ |
772 | if (was_writing && _IO_switch_to_wget_mode (fp)) |
773 | return WEOF; |
774 | |
775 | if (fp->_wide_data->_IO_buf_base == NULL) |
776 | { |
777 | /* It could be that we already have a pushback buffer. */ |
778 | if (fp->_wide_data->_IO_read_base != NULL) |
779 | { |
780 | free (ptr: fp->_wide_data->_IO_read_base); |
781 | fp->_flags &= ~_IO_IN_BACKUP; |
782 | } |
783 | _IO_doallocbuf (fp); |
784 | _IO_setp (fp, fp->_IO_buf_base, fp->_IO_buf_base); |
785 | _IO_setg (fp, fp->_IO_buf_base, fp->_IO_buf_base, fp->_IO_buf_base); |
786 | _IO_wsetp (fp, fp->_wide_data->_IO_buf_base, |
787 | fp->_wide_data->_IO_buf_base); |
788 | _IO_wsetg (fp, fp->_wide_data->_IO_buf_base, |
789 | fp->_wide_data->_IO_buf_base, fp->_wide_data->_IO_buf_base); |
790 | } |
791 | |
792 | switch (dir) |
793 | { |
794 | struct _IO_codecvt *cv; |
795 | int clen; |
796 | |
797 | case _IO_seek_cur: |
798 | /* Adjust for read-ahead (bytes is buffer). To do this we must |
799 | find out which position in the external buffer corresponds to |
800 | the current position in the internal buffer. */ |
801 | cv = fp->_codecvt; |
802 | clen = __libio_codecvt_encoding (cv); |
803 | |
804 | if (mode != 0 || !was_writing) |
805 | { |
806 | if (clen > 0) |
807 | { |
808 | offset -= (fp->_wide_data->_IO_read_end |
809 | - fp->_wide_data->_IO_read_ptr) * clen; |
810 | /* Adjust by readahead in external buffer. */ |
811 | offset -= fp->_IO_read_end - fp->_IO_read_ptr; |
812 | } |
813 | else |
814 | { |
815 | int nread; |
816 | |
817 | delta = (fp->_wide_data->_IO_read_ptr |
818 | - fp->_wide_data->_IO_read_base); |
819 | fp->_wide_data->_IO_state = fp->_wide_data->_IO_last_state; |
820 | nread = __libio_codecvt_length (cv, |
821 | &fp->_wide_data->_IO_state, |
822 | fp->_IO_read_base, |
823 | fp->_IO_read_end, delta); |
824 | fp->_IO_read_ptr = fp->_IO_read_base + nread; |
825 | fp->_wide_data->_IO_read_end = fp->_wide_data->_IO_read_ptr; |
826 | offset -= fp->_IO_read_end - fp->_IO_read_base - nread; |
827 | } |
828 | } |
829 | |
830 | if (fp->_offset == _IO_pos_BAD) |
831 | goto dumb; |
832 | |
833 | /* Make offset absolute, assuming current pointer is file_ptr(). */ |
834 | offset += fp->_offset; |
835 | |
836 | dir = _IO_seek_set; |
837 | break; |
838 | case _IO_seek_set: |
839 | break; |
840 | case _IO_seek_end: |
841 | { |
842 | struct __stat64_t64 st; |
843 | if (_IO_SYSSTAT (fp, &st) == 0 && S_ISREG (st.st_mode)) |
844 | { |
845 | offset += st.st_size; |
846 | dir = _IO_seek_set; |
847 | } |
848 | else |
849 | goto dumb; |
850 | } |
851 | } |
852 | |
853 | _IO_free_wbackup_area (fp); |
854 | |
855 | /* At this point, dir==_IO_seek_set. */ |
856 | |
857 | /* If destination is within current buffer, optimize: */ |
858 | if (fp->_offset != _IO_pos_BAD && fp->_IO_read_base != NULL |
859 | && !_IO_in_backup (fp)) |
860 | { |
861 | off64_t start_offset = (fp->_offset |
862 | - (fp->_IO_read_end - fp->_IO_buf_base)); |
863 | if (offset >= start_offset && offset < fp->_offset) |
864 | { |
865 | _IO_setg (fp, fp->_IO_buf_base, |
866 | fp->_IO_buf_base + (offset - start_offset), |
867 | fp->_IO_read_end); |
868 | _IO_setp (fp, fp->_IO_buf_base, fp->_IO_buf_base); |
869 | _IO_wsetg (fp, fp->_wide_data->_IO_buf_base, |
870 | fp->_wide_data->_IO_buf_base, |
871 | fp->_wide_data->_IO_buf_base); |
872 | _IO_wsetp (fp, fp->_wide_data->_IO_buf_base, |
873 | fp->_wide_data->_IO_buf_base); |
874 | |
875 | if (adjust_wide_data (fp, false)) |
876 | goto dumb; |
877 | |
878 | _IO_mask_flags (fp, 0, _IO_EOF_SEEN); |
879 | goto resync; |
880 | } |
881 | } |
882 | |
883 | if (fp->_flags & _IO_NO_READS) |
884 | goto dumb; |
885 | |
886 | /* Try to seek to a block boundary, to improve kernel page management. */ |
887 | new_offset = offset & ~(fp->_IO_buf_end - fp->_IO_buf_base - 1); |
888 | delta = offset - new_offset; |
889 | if (delta > fp->_IO_buf_end - fp->_IO_buf_base) |
890 | { |
891 | new_offset = offset; |
892 | delta = 0; |
893 | } |
894 | result = _IO_SYSSEEK (fp, new_offset, 0); |
895 | if (result < 0) |
896 | return EOF; |
897 | if (delta == 0) |
898 | count = 0; |
899 | else |
900 | { |
901 | count = _IO_SYSREAD (fp, fp->_IO_buf_base, |
902 | (must_be_exact |
903 | ? delta : fp->_IO_buf_end - fp->_IO_buf_base)); |
904 | if (count < delta) |
905 | { |
906 | /* We weren't allowed to read, but try to seek the remainder. */ |
907 | offset = count == EOF ? delta : delta-count; |
908 | dir = _IO_seek_cur; |
909 | goto dumb; |
910 | } |
911 | } |
912 | _IO_setg (fp, fp->_IO_buf_base, fp->_IO_buf_base + delta, |
913 | fp->_IO_buf_base + count); |
914 | _IO_setp (fp, fp->_IO_buf_base, fp->_IO_buf_base); |
915 | _IO_wsetg (fp, fp->_wide_data->_IO_buf_base, |
916 | fp->_wide_data->_IO_buf_base, fp->_wide_data->_IO_buf_base); |
917 | _IO_wsetp (fp, fp->_wide_data->_IO_buf_base, fp->_wide_data->_IO_buf_base); |
918 | |
919 | if (adjust_wide_data (fp, true)) |
920 | goto dumb; |
921 | |
922 | fp->_offset = result + count; |
923 | _IO_mask_flags (fp, 0, _IO_EOF_SEEN); |
924 | return offset; |
925 | dumb: |
926 | |
927 | _IO_unsave_markers (fp); |
928 | result = _IO_SYSSEEK (fp, offset, dir); |
929 | if (result != EOF) |
930 | { |
931 | _IO_mask_flags (fp, 0, _IO_EOF_SEEN); |
932 | fp->_offset = result; |
933 | _IO_setg (fp, fp->_IO_buf_base, fp->_IO_buf_base, fp->_IO_buf_base); |
934 | _IO_setp (fp, fp->_IO_buf_base, fp->_IO_buf_base); |
935 | _IO_wsetg (fp, fp->_wide_data->_IO_buf_base, |
936 | fp->_wide_data->_IO_buf_base, fp->_wide_data->_IO_buf_base); |
937 | _IO_wsetp (fp, fp->_wide_data->_IO_buf_base, |
938 | fp->_wide_data->_IO_buf_base); |
939 | } |
940 | return result; |
941 | |
942 | resync: |
943 | /* We need to do it since it is possible that the file offset in |
944 | the kernel may be changed behind our back. It may happen when |
945 | we fopen a file and then do a fork. One process may access the |
946 | file and the kernel file offset will be changed. */ |
947 | if (fp->_offset >= 0) |
948 | _IO_SYSSEEK (fp, fp->_offset, 0); |
949 | |
950 | return offset; |
951 | } |
952 | libc_hidden_def (_IO_wfile_seekoff) |
953 | |
954 | |
955 | size_t |
956 | _IO_wfile_xsputn (FILE *f, const void *data, size_t n) |
957 | { |
958 | const wchar_t *s = (const wchar_t *) data; |
959 | size_t to_do = n; |
960 | int must_flush = 0; |
961 | size_t count; |
962 | |
963 | if (n <= 0) |
964 | return 0; |
965 | /* This is an optimized implementation. |
966 | If the amount to be written straddles a block boundary |
967 | (or the filebuf is unbuffered), use sys_write directly. */ |
968 | |
969 | /* First figure out how much space is available in the buffer. */ |
970 | count = f->_wide_data->_IO_write_end - f->_wide_data->_IO_write_ptr; |
971 | if ((f->_flags & _IO_LINE_BUF) && (f->_flags & _IO_CURRENTLY_PUTTING)) |
972 | { |
973 | count = f->_wide_data->_IO_buf_end - f->_wide_data->_IO_write_ptr; |
974 | if (count >= n) |
975 | { |
976 | const wchar_t *p; |
977 | for (p = s + n; p > s; ) |
978 | { |
979 | if (*--p == L'\n') |
980 | { |
981 | count = p - s + 1; |
982 | must_flush = 1; |
983 | break; |
984 | } |
985 | } |
986 | } |
987 | } |
988 | /* Then fill the buffer. */ |
989 | if (count > 0) |
990 | { |
991 | if (count > to_do) |
992 | count = to_do; |
993 | if (count > 20) |
994 | { |
995 | f->_wide_data->_IO_write_ptr = |
996 | __wmempcpy (s1: f->_wide_data->_IO_write_ptr, s2: s, n: count); |
997 | s += count; |
998 | } |
999 | else |
1000 | { |
1001 | wchar_t *p = f->_wide_data->_IO_write_ptr; |
1002 | int i = (int) count; |
1003 | while (--i >= 0) |
1004 | *p++ = *s++; |
1005 | f->_wide_data->_IO_write_ptr = p; |
1006 | } |
1007 | to_do -= count; |
1008 | } |
1009 | if (to_do > 0) |
1010 | to_do -= _IO_wdefault_xsputn (f, s, to_do); |
1011 | if (must_flush |
1012 | && f->_wide_data->_IO_write_ptr != f->_wide_data->_IO_write_base) |
1013 | _IO_wdo_write (fp: f, data: f->_wide_data->_IO_write_base, |
1014 | to_do: f->_wide_data->_IO_write_ptr |
1015 | - f->_wide_data->_IO_write_base); |
1016 | |
1017 | return n - to_do; |
1018 | } |
1019 | libc_hidden_def (_IO_wfile_xsputn) |
1020 | |