1 /* 2 * JSON Parser 3 * 4 * Copyright IBM, Corp. 2009 5 * 6 * Authors: 7 * Anthony Liguori <aliguori@us.ibm.com> 8 * 9 * This work is licensed under the terms of the GNU LGPL, version 2.1 or later. 10 * See the COPYING.LIB file in the top-level directory. 11 * 12 */ 13 14 #include "qemu/osdep.h" 15 #include "qemu/cutils.h" 16 #include "qemu/unicode.h" 17 #include "qapi/error.h" 18 #include "qemu-common.h" 19 #include "qapi/qmp/qbool.h" 20 #include "qapi/qmp/qdict.h" 21 #include "qapi/qmp/qlist.h" 22 #include "qapi/qmp/qnull.h" 23 #include "qapi/qmp/qnum.h" 24 #include "qapi/qmp/qstring.h" 25 #include "json-parser-int.h" 26 27 struct JSONToken { 28 JSONTokenType type; 29 int x; 30 int y; 31 char str[]; 32 }; 33 34 typedef struct JSONParserContext 35 { 36 Error *err; 37 JSONToken *current; 38 GQueue *buf; 39 va_list *ap; 40 } JSONParserContext; 41 42 #define BUG_ON(cond) assert(!(cond)) 43 44 /** 45 * TODO 46 * 47 * 0) make errors meaningful again 48 * 1) add geometry information to tokens 49 * 3) should we return a parsed size? 50 * 4) deal with premature EOI 51 */ 52 53 static QObject *parse_value(JSONParserContext *ctxt); 54 55 /** 56 * Error handler 57 */ 58 static void GCC_FMT_ATTR(3, 4) parse_error(JSONParserContext *ctxt, 59 JSONToken *token, const char *msg, ...) 60 { 61 va_list ap; 62 char message[1024]; 63 64 if (ctxt->err) { 65 return; 66 } 67 va_start(ap, msg); 68 vsnprintf(message, sizeof(message), msg, ap); 69 va_end(ap); 70 error_setg(&ctxt->err, "JSON parse error, %s", message); 71 } 72 73 static int cvt4hex(const char *s) 74 { 75 int cp, i; 76 77 cp = 0; 78 for (i = 0; i < 4; i++) { 79 if (!qemu_isxdigit(s[i])) { 80 return -1; 81 } 82 cp <<= 4; 83 if (s[i] >= '0' && s[i] <= '9') { 84 cp |= s[i] - '0'; 85 } else if (s[i] >= 'a' && s[i] <= 'f') { 86 cp |= 10 + s[i] - 'a'; 87 } else if (s[i] >= 'A' && s[i] <= 'F') { 88 cp |= 10 + s[i] - 'A'; 89 } else { 90 return -1; 91 } 92 } 93 return cp; 94 } 95 96 /** 97 * parse_string(): Parse a JSON string 98 * 99 * From RFC 8259 "The JavaScript Object Notation (JSON) Data 100 * Interchange Format": 101 * 102 * char = unescaped / 103 * escape ( 104 * %x22 / ; " quotation mark U+0022 105 * %x5C / ; \ reverse solidus U+005C 106 * %x2F / ; / solidus U+002F 107 * %x62 / ; b backspace U+0008 108 * %x66 / ; f form feed U+000C 109 * %x6E / ; n line feed U+000A 110 * %x72 / ; r carriage return U+000D 111 * %x74 / ; t tab U+0009 112 * %x75 4HEXDIG ) ; uXXXX U+XXXX 113 * escape = %x5C ; \ 114 * quotation-mark = %x22 ; " 115 * unescaped = %x20-21 / %x23-5B / %x5D-10FFFF 116 * 117 * Extensions over RFC 8259: 118 * - Extra escape sequence in strings: 119 * 0x27 (apostrophe) is recognized after escape, too 120 * - Single-quoted strings: 121 * Like double-quoted strings, except they're delimited by %x27 122 * (apostrophe) instead of %x22 (quotation mark), and can't contain 123 * unescaped apostrophe, but can contain unescaped quotation mark. 124 * 125 * Note: 126 * - Encoding is modified UTF-8. 127 * - Invalid Unicode characters are rejected. 128 * - Control characters \x00..\x1F are rejected by the lexer. 129 */ 130 static QString *parse_string(JSONParserContext *ctxt, JSONToken *token) 131 { 132 const char *ptr = token->str; 133 QString *str; 134 char quote; 135 const char *beg; 136 int cp, trailing; 137 char *end; 138 ssize_t len; 139 char utf8_buf[5]; 140 141 assert(*ptr == '"' || *ptr == '\''); 142 quote = *ptr++; 143 str = qstring_new(); 144 145 while (*ptr != quote) { 146 assert(*ptr); 147 if (*ptr == '\\') { 148 beg = ptr++; 149 switch (*ptr++) { 150 case '"': 151 qstring_append_chr(str, '"'); 152 break; 153 case '\'': 154 qstring_append_chr(str, '\''); 155 break; 156 case '\\': 157 qstring_append_chr(str, '\\'); 158 break; 159 case '/': 160 qstring_append_chr(str, '/'); 161 break; 162 case 'b': 163 qstring_append_chr(str, '\b'); 164 break; 165 case 'f': 166 qstring_append_chr(str, '\f'); 167 break; 168 case 'n': 169 qstring_append_chr(str, '\n'); 170 break; 171 case 'r': 172 qstring_append_chr(str, '\r'); 173 break; 174 case 't': 175 qstring_append_chr(str, '\t'); 176 break; 177 case 'u': 178 cp = cvt4hex(ptr); 179 ptr += 4; 180 181 /* handle surrogate pairs */ 182 if (cp >= 0xD800 && cp <= 0xDBFF 183 && ptr[0] == '\\' && ptr[1] == 'u') { 184 /* leading surrogate followed by \u */ 185 cp = 0x10000 + ((cp & 0x3FF) << 10); 186 trailing = cvt4hex(ptr + 2); 187 if (trailing >= 0xDC00 && trailing <= 0xDFFF) { 188 /* followed by trailing surrogate */ 189 cp |= trailing & 0x3FF; 190 ptr += 6; 191 } else { 192 cp = -1; /* invalid */ 193 } 194 } 195 196 if (mod_utf8_encode(utf8_buf, sizeof(utf8_buf), cp) < 0) { 197 parse_error(ctxt, token, 198 "%.*s is not a valid Unicode character", 199 (int)(ptr - beg), beg); 200 goto out; 201 } 202 qstring_append(str, utf8_buf); 203 break; 204 default: 205 parse_error(ctxt, token, "invalid escape sequence in string"); 206 goto out; 207 } 208 } else { 209 cp = mod_utf8_codepoint(ptr, 6, &end); 210 if (cp < 0) { 211 parse_error(ctxt, token, "invalid UTF-8 sequence in string"); 212 goto out; 213 } 214 ptr = end; 215 len = mod_utf8_encode(utf8_buf, sizeof(utf8_buf), cp); 216 assert(len >= 0); 217 qstring_append(str, utf8_buf); 218 } 219 } 220 221 return str; 222 223 out: 224 qobject_unref(str); 225 return NULL; 226 } 227 228 /* Note: the token object returned by parser_context_peek_token or 229 * parser_context_pop_token is deleted as soon as parser_context_pop_token 230 * is called again. 231 */ 232 static JSONToken *parser_context_pop_token(JSONParserContext *ctxt) 233 { 234 g_free(ctxt->current); 235 ctxt->current = g_queue_pop_head(ctxt->buf); 236 return ctxt->current; 237 } 238 239 static JSONToken *parser_context_peek_token(JSONParserContext *ctxt) 240 { 241 return g_queue_peek_head(ctxt->buf); 242 } 243 244 /** 245 * Parsing rules 246 */ 247 static int parse_pair(JSONParserContext *ctxt, QDict *dict) 248 { 249 QObject *value; 250 QString *key = NULL; 251 JSONToken *peek, *token; 252 253 peek = parser_context_peek_token(ctxt); 254 if (peek == NULL) { 255 parse_error(ctxt, NULL, "premature EOI"); 256 goto out; 257 } 258 259 key = qobject_to(QString, parse_value(ctxt)); 260 if (!key) { 261 parse_error(ctxt, peek, "key is not a string in object"); 262 goto out; 263 } 264 265 token = parser_context_pop_token(ctxt); 266 if (token == NULL) { 267 parse_error(ctxt, NULL, "premature EOI"); 268 goto out; 269 } 270 271 if (token->type != JSON_COLON) { 272 parse_error(ctxt, token, "missing : in object pair"); 273 goto out; 274 } 275 276 value = parse_value(ctxt); 277 if (value == NULL) { 278 parse_error(ctxt, token, "Missing value in dict"); 279 goto out; 280 } 281 282 qdict_put_obj(dict, qstring_get_str(key), value); 283 284 qobject_unref(key); 285 286 return 0; 287 288 out: 289 qobject_unref(key); 290 291 return -1; 292 } 293 294 static QObject *parse_object(JSONParserContext *ctxt) 295 { 296 QDict *dict = NULL; 297 JSONToken *token, *peek; 298 299 token = parser_context_pop_token(ctxt); 300 assert(token && token->type == JSON_LCURLY); 301 302 dict = qdict_new(); 303 304 peek = parser_context_peek_token(ctxt); 305 if (peek == NULL) { 306 parse_error(ctxt, NULL, "premature EOI"); 307 goto out; 308 } 309 310 if (peek->type != JSON_RCURLY) { 311 if (parse_pair(ctxt, dict) == -1) { 312 goto out; 313 } 314 315 token = parser_context_pop_token(ctxt); 316 if (token == NULL) { 317 parse_error(ctxt, NULL, "premature EOI"); 318 goto out; 319 } 320 321 while (token->type != JSON_RCURLY) { 322 if (token->type != JSON_COMMA) { 323 parse_error(ctxt, token, "expected separator in dict"); 324 goto out; 325 } 326 327 if (parse_pair(ctxt, dict) == -1) { 328 goto out; 329 } 330 331 token = parser_context_pop_token(ctxt); 332 if (token == NULL) { 333 parse_error(ctxt, NULL, "premature EOI"); 334 goto out; 335 } 336 } 337 } else { 338 (void)parser_context_pop_token(ctxt); 339 } 340 341 return QOBJECT(dict); 342 343 out: 344 qobject_unref(dict); 345 return NULL; 346 } 347 348 static QObject *parse_array(JSONParserContext *ctxt) 349 { 350 QList *list = NULL; 351 JSONToken *token, *peek; 352 353 token = parser_context_pop_token(ctxt); 354 assert(token && token->type == JSON_LSQUARE); 355 356 list = qlist_new(); 357 358 peek = parser_context_peek_token(ctxt); 359 if (peek == NULL) { 360 parse_error(ctxt, NULL, "premature EOI"); 361 goto out; 362 } 363 364 if (peek->type != JSON_RSQUARE) { 365 QObject *obj; 366 367 obj = parse_value(ctxt); 368 if (obj == NULL) { 369 parse_error(ctxt, token, "expecting value"); 370 goto out; 371 } 372 373 qlist_append_obj(list, obj); 374 375 token = parser_context_pop_token(ctxt); 376 if (token == NULL) { 377 parse_error(ctxt, NULL, "premature EOI"); 378 goto out; 379 } 380 381 while (token->type != JSON_RSQUARE) { 382 if (token->type != JSON_COMMA) { 383 parse_error(ctxt, token, "expected separator in list"); 384 goto out; 385 } 386 387 obj = parse_value(ctxt); 388 if (obj == NULL) { 389 parse_error(ctxt, token, "expecting value"); 390 goto out; 391 } 392 393 qlist_append_obj(list, obj); 394 395 token = parser_context_pop_token(ctxt); 396 if (token == NULL) { 397 parse_error(ctxt, NULL, "premature EOI"); 398 goto out; 399 } 400 } 401 } else { 402 (void)parser_context_pop_token(ctxt); 403 } 404 405 return QOBJECT(list); 406 407 out: 408 qobject_unref(list); 409 return NULL; 410 } 411 412 static QObject *parse_keyword(JSONParserContext *ctxt) 413 { 414 JSONToken *token; 415 416 token = parser_context_pop_token(ctxt); 417 assert(token && token->type == JSON_KEYWORD); 418 419 if (!strcmp(token->str, "true")) { 420 return QOBJECT(qbool_from_bool(true)); 421 } else if (!strcmp(token->str, "false")) { 422 return QOBJECT(qbool_from_bool(false)); 423 } else if (!strcmp(token->str, "null")) { 424 return QOBJECT(qnull()); 425 } 426 parse_error(ctxt, token, "invalid keyword '%s'", token->str); 427 return NULL; 428 } 429 430 static QObject *parse_interpolation(JSONParserContext *ctxt) 431 { 432 JSONToken *token; 433 434 token = parser_context_pop_token(ctxt); 435 assert(token && token->type == JSON_INTERP); 436 437 if (!strcmp(token->str, "%p")) { 438 return va_arg(*ctxt->ap, QObject *); 439 } else if (!strcmp(token->str, "%i")) { 440 return QOBJECT(qbool_from_bool(va_arg(*ctxt->ap, int))); 441 } else if (!strcmp(token->str, "%d")) { 442 return QOBJECT(qnum_from_int(va_arg(*ctxt->ap, int))); 443 } else if (!strcmp(token->str, "%ld")) { 444 return QOBJECT(qnum_from_int(va_arg(*ctxt->ap, long))); 445 } else if (!strcmp(token->str, "%lld")) { 446 return QOBJECT(qnum_from_int(va_arg(*ctxt->ap, long long))); 447 } else if (!strcmp(token->str, "%" PRId64)) { 448 return QOBJECT(qnum_from_int(va_arg(*ctxt->ap, int64_t))); 449 } else if (!strcmp(token->str, "%u")) { 450 return QOBJECT(qnum_from_uint(va_arg(*ctxt->ap, unsigned int))); 451 } else if (!strcmp(token->str, "%lu")) { 452 return QOBJECT(qnum_from_uint(va_arg(*ctxt->ap, unsigned long))); 453 } else if (!strcmp(token->str, "%llu")) { 454 return QOBJECT(qnum_from_uint(va_arg(*ctxt->ap, unsigned long long))); 455 } else if (!strcmp(token->str, "%" PRIu64)) { 456 return QOBJECT(qnum_from_uint(va_arg(*ctxt->ap, uint64_t))); 457 } else if (!strcmp(token->str, "%s")) { 458 return QOBJECT(qstring_from_str(va_arg(*ctxt->ap, const char *))); 459 } else if (!strcmp(token->str, "%f")) { 460 return QOBJECT(qnum_from_double(va_arg(*ctxt->ap, double))); 461 } 462 parse_error(ctxt, token, "invalid interpolation '%s'", token->str); 463 return NULL; 464 } 465 466 static QObject *parse_literal(JSONParserContext *ctxt) 467 { 468 JSONToken *token; 469 470 token = parser_context_pop_token(ctxt); 471 assert(token); 472 473 switch (token->type) { 474 case JSON_STRING: 475 return QOBJECT(parse_string(ctxt, token)); 476 case JSON_INTEGER: { 477 /* 478 * Represent JSON_INTEGER as QNUM_I64 if possible, else as 479 * QNUM_U64, else as QNUM_DOUBLE. Note that qemu_strtoi64() 480 * and qemu_strtou64() fail with ERANGE when it's not 481 * possible. 482 * 483 * qnum_get_int() will then work for any signed 64-bit 484 * JSON_INTEGER, qnum_get_uint() for any unsigned 64-bit 485 * integer, and qnum_get_double() both for any JSON_INTEGER 486 * and any JSON_FLOAT (with precision loss for integers beyond 487 * 53 bits) 488 */ 489 int ret; 490 int64_t value; 491 uint64_t uvalue; 492 493 ret = qemu_strtoi64(token->str, NULL, 10, &value); 494 if (!ret) { 495 return QOBJECT(qnum_from_int(value)); 496 } 497 assert(ret == -ERANGE); 498 499 if (token->str[0] != '-') { 500 ret = qemu_strtou64(token->str, NULL, 10, &uvalue); 501 if (!ret) { 502 return QOBJECT(qnum_from_uint(uvalue)); 503 } 504 assert(ret == -ERANGE); 505 } 506 /* fall through to JSON_FLOAT */ 507 } 508 case JSON_FLOAT: 509 /* FIXME dependent on locale; a pervasive issue in QEMU */ 510 /* FIXME our lexer matches RFC 7159 in forbidding Inf or NaN, 511 * but those might be useful extensions beyond JSON */ 512 return QOBJECT(qnum_from_double(strtod(token->str, NULL))); 513 default: 514 abort(); 515 } 516 } 517 518 static QObject *parse_value(JSONParserContext *ctxt) 519 { 520 JSONToken *token; 521 522 token = parser_context_peek_token(ctxt); 523 if (token == NULL) { 524 parse_error(ctxt, NULL, "premature EOI"); 525 return NULL; 526 } 527 528 switch (token->type) { 529 case JSON_LCURLY: 530 return parse_object(ctxt); 531 case JSON_LSQUARE: 532 return parse_array(ctxt); 533 case JSON_INTERP: 534 return parse_interpolation(ctxt); 535 case JSON_INTEGER: 536 case JSON_FLOAT: 537 case JSON_STRING: 538 return parse_literal(ctxt); 539 case JSON_KEYWORD: 540 return parse_keyword(ctxt); 541 default: 542 parse_error(ctxt, token, "expecting value"); 543 return NULL; 544 } 545 } 546 547 JSONToken *json_token(JSONTokenType type, int x, int y, GString *tokstr) 548 { 549 JSONToken *token = g_malloc(sizeof(JSONToken) + tokstr->len + 1); 550 551 token->type = type; 552 memcpy(token->str, tokstr->str, tokstr->len); 553 token->str[tokstr->len] = 0; 554 token->x = x; 555 token->y = y; 556 return token; 557 } 558 559 QObject *json_parser_parse(GQueue *tokens, va_list *ap, Error **errp) 560 { 561 JSONParserContext ctxt = { .buf = tokens, .ap = ap }; 562 QObject *result; 563 564 result = parse_value(&ctxt); 565 assert(ctxt.err || g_queue_is_empty(ctxt.buf)); 566 567 error_propagate(errp, ctxt.err); 568 569 while (!g_queue_is_empty(ctxt.buf)) { 570 parser_context_pop_token(&ctxt); 571 } 572 g_free(ctxt.current); 573 574 return result; 575 } 576