iconv/loop.c

   1 /* Conversion loop frame work.
   2    Copyright (C) 1998-2022 Free Software Foundation, Inc.
   3    This file is part of the GNU C Library.
   4
   5    The GNU C Library is free software; you can redistribute it and/or
   6    modify it under the terms of the GNU Lesser General Public
   7    License as published by the Free Software Foundation; either
   8    version 2.1 of the License, or (at your option) any later version.
   9
  10    The GNU C Library is distributed in the hope that it will be useful,
  11    but WITHOUT ANY WARRANTY; without even the implied warranty of
  12    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13    Lesser General Public License for more details.
  14
  15    You should have received a copy of the GNU Lesser General Public
  16    License along with the GNU C Library; if not, see
  17    <https://www.gnu.org/licenses/>.  */
  18
  19 /* This file provides a frame for the reader loop in all conversion modules.
  20    The actual code must (of course) be provided in the actual module source
  21    code but certain actions can be written down generically, with some
  22    customization options which are these:
  23
  24      MIN_NEEDED_INPUT   minimal number of input bytes needed for the next
  25                         conversion.
  26      MIN_NEEDED_OUTPUT  minimal number of bytes produced by the next round
  27                         of conversion.
  28
  29      MAX_NEEDED_INPUT   you guess it, this is the maximal number of input
  30                         bytes needed.  It defaults to MIN_NEEDED_INPUT
  31      MAX_NEEDED_OUTPUT  likewise for output bytes.
  32
  33      LOOPFCT            name of the function created.  If not specified
  34                         the name is `loop' but this prevents the use
  35                         of multiple functions in the same file.
  36
  37      BODY               this is supposed to expand to the body of the loop.
  38                         The user must provide this.
  39
  40      EXTRA_LOOP_DECLS   extra arguments passed from conversion loop call.
  41
  42      INIT_PARAMS        code to define and initialize variables from params.
  43      UPDATE_PARAMS      code to store result in params.
  44
  45      ONEBYTE_BODY       body of the specialized conversion function for a
  46                         single byte from the current character set to INTERNAL.
  47 */
  48
  49 #include <assert.h>
  50 #include <endian.h>
  51 #include <iconv/gconv_int.h>
  52 #include <stdint.h>
  53 #include <string.h>
  54 #include <wchar.h>
  55 #include <sys/param.h>          /* For MIN.  */
  56 #define __need_size_t
  57 #include <stddef.h>
  58 #include <libc-diag.h>
  59
  60 /* We have to provide support for machines which are not able to handled
  61    unaligned memory accesses.  Some of the character encodings have
  62    representations with a fixed width of 2 or 4 bytes.  But if we cannot
  63    access unaligned memory we still have to read byte-wise.  */
  64 #undef FCTNAME2
  65 #if _STRING_ARCH_unaligned || !defined DEFINE_UNALIGNED
  66 /* We can handle unaligned memory access.  */
  67 # define get16(addr) *((const uint16_t *) (addr))
  68 # define get32(addr) *((const uint32_t *) (addr))
  69
  70 /* We need no special support for writing values either.  */
  71 # define put16(addr, val) *((uint16_t *) (addr)) = (val)
  72 # define put32(addr, val) *((uint32_t *) (addr)) = (val)
  73
  74 # define FCTNAME2(name) name
  75 #else
  76 /* Distinguish between big endian and little endian.  */
  77 # if __BYTE_ORDER == __LITTLE_ENDIAN
  78 #  define get16(addr) \
  79      (((const unsigned char *) (addr))[1] << 8                                \
  80       | ((const unsigned char *) (addr))[0])
  81 #  define get32(addr) \
  82      (((((const unsigned char *) (addr))[3] << 8                              \
  83         | ((const unsigned char *) (addr))[2]) << 8                           \
  84        | ((const unsigned char *) (addr))[1]) << 8                            \
  85       | ((const unsigned char *) (addr))[0])
  86
  87 #  define put16(addr, val) \
  88      ({ uint16_t __val = (val);                                               \
  89         ((unsigned char *) (addr))[0] = __val;                                \
  90         ((unsigned char *) (addr))[1] = __val >> 8;                           \
  91         (void) 0; })
  92 #  define put32(addr, val) \
  93      ({ uint32_t __val = (val);                                               \
  94         ((unsigned char *) (addr))[0] = __val;                                \
  95         __val >>= 8;                                                          \
  96         ((unsigned char *) (addr))[1] = __val;                                \
  97         __val >>= 8;                                                          \
  98         ((unsigned char *) (addr))[2] = __val;                                \
  99         __val >>= 8;                                                          \
 100         ((unsigned char *) (addr))[3] = __val;                                \
 101         (void) 0; })
 102 # else
 103 #  define get16(addr) \
 104      (((const unsigned char *) (addr))[0] << 8                                \
 105       | ((const unsigned char *) (addr))[1])
 106 #  define get32(addr) \
 107      (((((const unsigned char *) (addr))[0] << 8                              \
 108         | ((const unsigned char *) (addr))[1]) << 8                           \
 109        | ((const unsigned char *) (addr))[2]) << 8                            \
 110       | ((const unsigned char *) (addr))[3])
 111
 112 #  define put16(addr, val) \
 113      ({ uint16_t __val = (val);                                               \
 114         ((unsigned char *) (addr))[1] = __val;                                \
 115         ((unsigned char *) (addr))[0] = __val >> 8;                           \
 116         (void) 0; })
 117 #  define put32(addr, val) \
 118      ({ uint32_t __val = (val);                                               \
 119         ((unsigned char *) (addr))[3] = __val;                                \
 120         __val >>= 8;                                                          \
 121         ((unsigned char *) (addr))[2] = __val;                                \
 122         __val >>= 8;                                                          \
 123         ((unsigned char *) (addr))[1] = __val;                                \
 124         __val >>= 8;                                                          \
 125         ((unsigned char *) (addr))[0] = __val;                                \
 126         (void) 0; })
 127 # endif
 128
 129 # define FCTNAME2(name) name##_unaligned
 130 #endif
 131 #define FCTNAME(name) FCTNAME2(name)
 132
 133
 134 /* We need at least one byte for the next round.  */
 135 #ifndef MIN_NEEDED_INPUT
 136 # error "MIN_NEEDED_INPUT definition missing"
 137 #elif MIN_NEEDED_INPUT < 1
 138 # error "MIN_NEEDED_INPUT must be >= 1"
 139 #endif
 140
 141 /* Let's see how many bytes we produce.  */
 142 #ifndef MAX_NEEDED_INPUT
 143 # define MAX_NEEDED_INPUT       MIN_NEEDED_INPUT
 144 #endif
 145
 146 /* We produce at least one byte in the next round.  */
 147 #ifndef MIN_NEEDED_OUTPUT
 148 # error "MIN_NEEDED_OUTPUT definition missing"
 149 #elif MIN_NEEDED_OUTPUT < 1
 150 # error "MIN_NEEDED_OUTPUT must be >= 1"
 151 #endif
 152
 153 /* Let's see how many bytes we produce.  */
 154 #ifndef MAX_NEEDED_OUTPUT
 155 # define MAX_NEEDED_OUTPUT      MIN_NEEDED_OUTPUT
 156 #endif
 157
 158 /* Default name for the function.  */
 159 #ifndef LOOPFCT
 160 # define LOOPFCT                loop
 161 #endif
 162
 163 /* Make sure we have a loop body.  */
 164 #ifndef BODY
 165 # error "Definition of BODY missing for function" LOOPFCT
 166 #endif
 167
 168
 169 /* If no arguments have to passed to the loop function define the macro
 170    as empty.  */
 171 #ifndef EXTRA_LOOP_DECLS
 172 # define EXTRA_LOOP_DECLS
 173 #endif
 174
 175 /* Allow using UPDATE_PARAMS in macros where #ifdef UPDATE_PARAMS test
 176    isn't possible.  */
 177 #ifndef UPDATE_PARAMS
 178 # define UPDATE_PARAMS do { } while (0)
 179 #endif
 180 #ifndef REINIT_PARAMS
 181 # define REINIT_PARAMS do { } while (0)
 182 #endif
 183
 184
 185 /* To make it easier for the writers of the modules, we define a macro
 186    to test whether we have to ignore errors.  */
 187 #define ignore_errors_p() \
 188   (irreversible != NULL && (flags & __GCONV_IGNORE_ERRORS))
 189
 190
 191 /* Error handling for the FROM_LOOP direction, with ignoring of errors.
 192    Note that we cannot use the do while (0) trick since `break' and
 193    `continue' must reach certain points.  */
 194 #define STANDARD_FROM_LOOP_ERR_HANDLER(Incr) \
 195   {                                                                           \
 196     result = __GCONV_ILLEGAL_INPUT;                                           \
 197                                                                               \
 198     if (! ignore_errors_p ())                                                 \
 199       break;                                                                  \
 200                                                                               \
 201     /* We ignore the invalid input byte sequence.  */                         \
 202     inptr += (Incr);                                                          \
 203     ++*irreversible;                                                          \
 204     /* But we keep result == __GCONV_ILLEGAL_INPUT, because of the constraint \
 205        that "iconv -c" must give the same exitcode as "iconv".  */            \
 206     continue;                                                                 \
 207   }
 208
 209 /* Error handling for the TO_LOOP direction, with use of transliteration/
 210    transcription functions and ignoring of errors.  Note that we cannot use
 211    the do while (0) trick since `break' and `continue' must reach certain
 212    points.  */
 213 #define STANDARD_TO_LOOP_ERR_HANDLER(Incr) \
 214   {                                                                           \
 215     result = __GCONV_ILLEGAL_INPUT;                                           \
 216                                                                               \
 217     if (irreversible == NULL)                                                 \
 218       /* This means we are in call from __gconv_transliterate.  In this       \
 219          case we are not doing any error recovery outself.  */                \
 220       break;                                                                  \
 221                                                                               \
 222     /* If needed, flush any conversion state, so that __gconv_transliterate   \
 223        starts with current shift state.  */                                   \
 224     UPDATE_PARAMS;                                                            \
 225                                                                               \
 226     /* First try the transliteration methods.  */                             \
 227     if ((step_data->__flags & __GCONV_TRANSLIT) != 0)                         \
 228       result = __gconv_transliterate                                          \
 229         (step, step_data, *inptrp,                                            \
 230          &inptr, inend, &outptr, irreversible);                       \
 231                                                                               \
 232     REINIT_PARAMS;                                                            \
 233                                                                               \
 234     /* If any of them recognized the input continue with the loop.  */        \
 235     if (result != __GCONV_ILLEGAL_INPUT)                                      \
 236       {                                                                       \
 237         if (__glibc_unlikely (result == __GCONV_FULL_OUTPUT))                 \
 238           break;                                                              \
 239                                                                               \
 240         continue;                                                             \
 241       }                                                                       \
 242                                                                               \
 243     /* Next see whether we have to ignore the error.  If not, stop.  */       \
 244     if (! ignore_errors_p ())                                                 \
 245       break;                                                                  \
 246                                                                               \
 247     /* When we come here it means we ignore the character.  */                \
 248     ++*irreversible;                                                          \
 249     inptr += Incr;                                                            \
 250     /* But we keep result == __GCONV_ILLEGAL_INPUT, because of the constraint \
 251        that "iconv -c" must give the same exitcode as "iconv".  */            \
 252     continue;                                                                 \
 253   }
 254
 255
 256 /* With GCC 7 when compiling with -Os for 32-bit s390 the compiler
 257    warns that the variable 'ch', in the definition of BODY in
 258    sysdeps/s390/multiarch/8bit-generic.c, may be used uninitialized in
 259    the call to UNICODE_TAG_HANDLER in that macro.  This variable is
 260    actually always initialized before use, in the prior loop if INDEX
 261    is nonzero and in the following 'if' if INDEX is zero.  That code
 262    has a comment referencing this diagnostic disabling; updates in one
 263    place may require updates in the other.  */
 264 DIAG_PUSH_NEEDS_COMMENT;
 265 DIAG_IGNORE_Os_NEEDS_COMMENT (7, "-Wmaybe-uninitialized");
 266 /* Handling of Unicode 3.1 TAG characters.  Unicode recommends
 267    "If language codes are not relevant to the particular processing
 268     operation, then they should be ignored."  This macro is usually
 269    called right before  STANDARD_TO_LOOP_ERR_HANDLER (Incr).  */
 270 #define UNICODE_TAG_HANDLER(Character, Incr) \
 271   {                                                                           \
 272     /* TAG characters are those in the range U+E0000..U+E007F.  */            \
 273     if (((Character) >> 7) == (0xe0000 >> 7))                                 \
 274       {                                                                       \
 275         inptr += Incr;                                                        \
 276         continue;                                                             \
 277       }                                                                       \
 278   }
 279 DIAG_POP_NEEDS_COMMENT;
 280
 281
 282 /* The function returns the status, as defined in gconv.h.  */
 283 static inline int
 284 __attribute ((always_inline))
 285 FCTNAME (LOOPFCT) (struct __gconv_step *step,
 286                    struct __gconv_step_data *step_data,
 287                    const unsigned char **inptrp, const unsigned char *inend,
 288                    unsigned char **outptrp, const unsigned char *outend,
 289                    size_t *irreversible EXTRA_LOOP_DECLS)
 290 {
 291 #ifdef LOOP_NEED_STATE
 292   mbstate_t *state = step_data->__statep;
 293 #endif
 294 #ifdef LOOP_NEED_FLAGS
 295   int flags = step_data->__flags;
 296 #endif
 297 #ifdef LOOP_NEED_DATA
 298   void *data = step->__data;
 299 #endif
 300   int result = __GCONV_EMPTY_INPUT;
 301   const unsigned char *inptr = *inptrp;
 302   unsigned char *outptr = *outptrp;
 303
 304 #ifdef INIT_PARAMS
 305   INIT_PARAMS;
 306 #endif
 307
 308   while (inptr != inend)
 309     {
 310       /* `if' cases for MIN_NEEDED_OUTPUT ==/!= 1 is made to help the
 311          compiler generating better code.  They will be optimized away
 312          since MIN_NEEDED_OUTPUT is always a constant.  */
 313       if (MIN_NEEDED_INPUT > 1
 314           && __builtin_expect (inptr + MIN_NEEDED_INPUT > inend, 0))
 315         {
 316           /* We don't have enough input for another complete input
 317              character.  */
 318           result = __GCONV_INCOMPLETE_INPUT;
 319           break;
 320         }
 321       if ((MIN_NEEDED_OUTPUT != 1
 322            && __builtin_expect (outptr + MIN_NEEDED_OUTPUT > outend, 0))
 323           || (MIN_NEEDED_OUTPUT == 1
 324               && __builtin_expect (outptr >= outend, 0)))
 325         {
 326           /* Overflow in the output buffer.  */
 327           result = __GCONV_FULL_OUTPUT;
 328           break;
 329         }
 330
 331       /* Here comes the body the user provides.  It can stop with
 332          RESULT set to GCONV_INCOMPLETE_INPUT (if the size of the
 333          input characters vary in size), GCONV_ILLEGAL_INPUT, or
 334          GCONV_FULL_OUTPUT (if the output characters vary in size).  */
 335       BODY
 336     }
 337
 338   /* Update the pointers pointed to by the parameters.  */
 339   *inptrp = inptr;
 340   *outptrp = outptr;
 341   UPDATE_PARAMS;
 342
 343   return result;
 344 }
 345
 346
 347 /* Include the file a second time to define the function to handle
 348    unaligned access.  */
 349 #if !defined DEFINE_UNALIGNED && !_STRING_ARCH_unaligned \
 350     && MIN_NEEDED_INPUT != 1 && MAX_NEEDED_INPUT % MIN_NEEDED_INPUT == 0 \
 351     && MIN_NEEDED_OUTPUT != 1 && MAX_NEEDED_OUTPUT % MIN_NEEDED_OUTPUT == 0
 352 # undef get16
 353 # undef get32
 354 # undef put16
 355 # undef put32
 356 # undef unaligned
 357
 358 # define DEFINE_UNALIGNED
 359 # include "loop.c"
 360 # undef DEFINE_UNALIGNED
 361 #else
 362 # if MAX_NEEDED_INPUT > 1
 363 #  define SINGLE(fct) SINGLE2 (fct)
 364 #  define SINGLE2(fct) fct##_single
 365 static inline int
 366 __attribute ((always_inline))
 367 SINGLE(LOOPFCT) (struct __gconv_step *step,
 368                  struct __gconv_step_data *step_data,
 369                  const unsigned char **inptrp, const unsigned char *inend,
 370                  unsigned char **outptrp, unsigned char *outend,
 371                  size_t *irreversible EXTRA_LOOP_DECLS)
 372 {
 373   mbstate_t *state = step_data->__statep;
 374 #  ifdef LOOP_NEED_FLAGS
 375   int flags = step_data->__flags;
 376 #  endif
 377 #  ifdef LOOP_NEED_DATA
 378   void *data = step->__data;
 379 #  endif
 380   int result = __GCONV_OK;
 381   unsigned char bytebuf[MAX_NEEDED_INPUT];
 382   const unsigned char *inptr = *inptrp;
 383   unsigned char *outptr = *outptrp;
 384   size_t inlen;
 385
 386 #  ifdef INIT_PARAMS
 387   INIT_PARAMS;
 388 #  endif
 389
 390 #  ifdef UNPACK_BYTES
 391   UNPACK_BYTES
 392 #  else
 393   /* Add the bytes from the state to the input buffer.  */
 394   assert ((state->__count & 7) <= sizeof (state->__value));
 395   for (inlen = 0; inlen < (size_t) (state->__count & 7); ++inlen)
 396     bytebuf[inlen] = state->__value.__wchb[inlen];
 397 #  endif
 398
 399   /* Are there enough bytes in the input buffer?  */
 400   if (MIN_NEEDED_INPUT > 1
 401       && __builtin_expect (inptr + (MIN_NEEDED_INPUT - inlen) > inend, 0))
 402     {
 403       *inptrp = inend;
 404 #  ifdef STORE_REST
 405
 406       /* Building with -O3 GCC emits a `array subscript is above array
 407          bounds' warning.  GCC BZ #64739 has been opened for this.  */
 408       DIAG_PUSH_NEEDS_COMMENT;
 409       DIAG_IGNORE_NEEDS_COMMENT (4.9, "-Warray-bounds");
 410       while (inptr < inend)
 411         bytebuf[inlen++] = *inptr++;
 412       DIAG_POP_NEEDS_COMMENT;
 413
 414       inptr = bytebuf;
 415       inptrp = &inptr;
 416       inend = &bytebuf[inlen];
 417
 418       STORE_REST
 419 #  else
 420       /* We don't have enough input for another complete input
 421          character.  */
 422       size_t inlen_after = inlen + (inend - inptr);
 423       assert (inlen_after <= sizeof (state->__value.__wchb));
 424       for (; inlen < inlen_after; inlen++)
 425         state->__value.__wchb[inlen] = *inptr++;
 426 #  endif
 427
 428       return __GCONV_INCOMPLETE_INPUT;
 429     }
 430
 431   /* Enough space in output buffer.  */
 432   if ((MIN_NEEDED_OUTPUT != 1 && outptr + MIN_NEEDED_OUTPUT > outend)
 433       || (MIN_NEEDED_OUTPUT == 1 && outptr >= outend))
 434     /* Overflow in the output buffer.  */
 435     return __GCONV_FULL_OUTPUT;
 436
 437   /*  Now add characters from the normal input buffer.  */
 438   if (inlen >= MAX_NEEDED_INPUT || inptr >= inend)
 439     /* Avoid a -Wstringop-overflow= warning when this loop is
 440        unrolled.  The compiler cannot otherwise see that this is
 441        unreachable because it depends on (state->__count & 7) not
 442        being too large after a previous conversion step.
 443        Starting with GCC 12, we also have mark the inptr >= inend
 444        case as unreachable to omit the warning.  Note that this SINGLE
 445        function is only used to implement the mb*towc*() or wc*tomb*()
 446        functions.  Those functions use inptr and inend pointing to a
 447        variable on stack, compute the inend pointer or explicitly check
 448        the arguments which always leads to inptr < inend.  */
 449     __builtin_unreachable ();
 450   do
 451     bytebuf[inlen++] = *inptr++;
 452   while (inlen < MAX_NEEDED_INPUT && inptr < inend);
 453
 454   inptr = bytebuf;
 455   inend = &bytebuf[inlen];
 456
 457   do
 458     {
 459       BODY
 460     }
 461   while (0);
 462
 463   /* Now we either have produced an output character and consumed all the
 464      bytes from the state and at least one more, or the character is still
 465      incomplete, or we have some other error (like illegal input character,
 466      no space in output buffer).  */
 467   if (__glibc_likely (inptr != bytebuf))
 468     {
 469       /* We found a new character.  */
 470       assert (inptr - bytebuf > (state->__count & 7));
 471
 472       *inptrp += inptr - bytebuf - (state->__count & 7);
 473       *outptrp = outptr;
 474
 475       result = __GCONV_OK;
 476
 477       /* Clear the state buffer.  */
 478 #  ifdef CLEAR_STATE
 479       CLEAR_STATE;
 480 #  else
 481       state->__count &= ~7;
 482 #  endif
 483     }
 484   else if (result == __GCONV_INCOMPLETE_INPUT)
 485     {
 486       /* This can only happen if we have less than MAX_NEEDED_INPUT bytes
 487          available.  */
 488       assert (inend != &bytebuf[MAX_NEEDED_INPUT]);
 489
 490       *inptrp += inend - bytebuf - (state->__count & 7);
 491 #  ifdef STORE_REST
 492       inptrp = &inptr;
 493
 494       STORE_REST
 495 #  else
 496       /* We don't have enough input for another complete input
 497          character.  */
 498       assert (inend - inptr > (state->__count & ~7));
 499       assert (inend - inptr <= sizeof (state->__value.__wchb));
 500       state->__count = (state->__count & ~7) | (inend - inptr);
 501       for (inlen = 0; inlen < inend - inptr; inlen++)
 502         state->__value.__wchb[inlen] = inptr[inlen];
 503       inptr = inend;
 504 #  endif
 505     }
 506
 507   return result;
 508 }
 509 #  undef SINGLE
 510 #  undef SINGLE2
 511 # endif
 512
 513
 514 # ifdef ONEBYTE_BODY
 515 /* Define the shortcut function for btowc.  */
 516 static wint_t
 517 gconv_btowc (struct __gconv_step *step, unsigned char c)
 518   ONEBYTE_BODY
 519 #  define FROM_ONEBYTE gconv_btowc
 520 # endif
 521
 522 #endif
 523
 524 /* We remove the macro definitions so that we can include this file again
 525    for the definition of another function.  */
 526 #undef MIN_NEEDED_INPUT
 527 #undef MAX_NEEDED_INPUT
 528 #undef MIN_NEEDED_OUTPUT
 529 #undef MAX_NEEDED_OUTPUT
 530 #undef LOOPFCT
 531 #undef BODY
 532 #undef LOOPFCT
 533 #undef EXTRA_LOOP_DECLS
 534 #undef INIT_PARAMS
 535 #undef UPDATE_PARAMS
 536 #undef REINIT_PARAMS
 537 #undef ONEBYTE_BODY
 538 #undef UNPACK_BYTES
 539 #undef CLEAR_STATE
 540 #undef LOOP_NEED_STATE
 541 #undef LOOP_NEED_FLAGS
 542 #undef LOOP_NEED_DATA
 543 #undef get16
 544 #undef get32
 545 #undef put16
 546 #undef put32
 547 #undef unaligned