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 "qapi/qmp/json-parser.h" 26 #include "qapi/qmp/json-lexer.h" 27 #include "qapi/qmp/json-streamer.h" 28 29 typedef struct JSONParserContext 30 { 31 Error *err; 32 JSONToken *current; 33 GQueue *buf; 34 } JSONParserContext; 35 36 #define BUG_ON(cond) assert(!(cond)) 37 38 /** 39 * TODO 40 * 41 * 0) make errors meaningful again 42 * 1) add geometry information to tokens 43 * 3) should we return a parsed size? 44 * 4) deal with premature EOI 45 */ 46 47 static QObject *parse_value(JSONParserContext *ctxt, va_list *ap); 48 49 /** 50 * Error handler 51 */ 52 static void GCC_FMT_ATTR(3, 4) parse_error(JSONParserContext *ctxt, 53 JSONToken *token, const char *msg, ...) 54 { 55 va_list ap; 56 char message[1024]; 57 58 if (ctxt->err) { 59 return; 60 } 61 va_start(ap, msg); 62 vsnprintf(message, sizeof(message), msg, ap); 63 va_end(ap); 64 error_setg(&ctxt->err, "JSON parse error, %s", message); 65 } 66 67 static int cvt4hex(const char *s) 68 { 69 int cp, i; 70 71 cp = 0; 72 for (i = 0; i < 4; i++) { 73 if (!qemu_isxdigit(s[i])) { 74 return -1; 75 } 76 cp <<= 4; 77 if (s[i] >= '0' && s[i] <= '9') { 78 cp |= s[i] - '0'; 79 } else if (s[i] >= 'a' && s[i] <= 'f') { 80 cp |= 10 + s[i] - 'a'; 81 } else if (s[i] >= 'A' && s[i] <= 'F') { 82 cp |= 10 + s[i] - 'A'; 83 } else { 84 return -1; 85 } 86 } 87 return cp; 88 } 89 90 /** 91 * parse_string(): Parse a JSON string 92 * 93 * From RFC 8259 "The JavaScript Object Notation (JSON) Data 94 * Interchange Format": 95 * 96 * char = unescaped / 97 * escape ( 98 * %x22 / ; " quotation mark U+0022 99 * %x5C / ; \ reverse solidus U+005C 100 * %x2F / ; / solidus U+002F 101 * %x62 / ; b backspace U+0008 102 * %x66 / ; f form feed U+000C 103 * %x6E / ; n line feed U+000A 104 * %x72 / ; r carriage return U+000D 105 * %x74 / ; t tab U+0009 106 * %x75 4HEXDIG ) ; uXXXX U+XXXX 107 * escape = %x5C ; \ 108 * quotation-mark = %x22 ; " 109 * unescaped = %x20-21 / %x23-5B / %x5D-10FFFF 110 * 111 * Extensions over RFC 8259: 112 * - Extra escape sequence in strings: 113 * 0x27 (apostrophe) is recognized after escape, too 114 * - Single-quoted strings: 115 * Like double-quoted strings, except they're delimited by %x27 116 * (apostrophe) instead of %x22 (quotation mark), and can't contain 117 * unescaped apostrophe, but can contain unescaped quotation mark. 118 * 119 * Note: 120 * - Encoding is modified UTF-8. 121 * - Invalid Unicode characters are rejected. 122 * - Control characters \x00..\x1F are rejected by the lexer. 123 */ 124 static QString *parse_string(JSONParserContext *ctxt, JSONToken *token) 125 { 126 const char *ptr = token->str; 127 QString *str; 128 char quote; 129 const char *beg; 130 int cp, trailing; 131 char *end; 132 ssize_t len; 133 char utf8_buf[5]; 134 135 assert(*ptr == '"' || *ptr == '\''); 136 quote = *ptr++; 137 str = qstring_new(); 138 139 while (*ptr != quote) { 140 assert(*ptr); 141 if (*ptr == '\\') { 142 beg = ptr++; 143 switch (*ptr++) { 144 case '"': 145 qstring_append_chr(str, '"'); 146 break; 147 case '\'': 148 qstring_append_chr(str, '\''); 149 break; 150 case '\\': 151 qstring_append_chr(str, '\\'); 152 break; 153 case '/': 154 qstring_append_chr(str, '/'); 155 break; 156 case 'b': 157 qstring_append_chr(str, '\b'); 158 break; 159 case 'f': 160 qstring_append_chr(str, '\f'); 161 break; 162 case 'n': 163 qstring_append_chr(str, '\n'); 164 break; 165 case 'r': 166 qstring_append_chr(str, '\r'); 167 break; 168 case 't': 169 qstring_append_chr(str, '\t'); 170 break; 171 case 'u': 172 cp = cvt4hex(ptr); 173 ptr += 4; 174 175 /* handle surrogate pairs */ 176 if (cp >= 0xD800 && cp <= 0xDBFF 177 && ptr[0] == '\\' && ptr[1] == 'u') { 178 /* leading surrogate followed by \u */ 179 cp = 0x10000 + ((cp & 0x3FF) << 10); 180 trailing = cvt4hex(ptr + 2); 181 if (trailing >= 0xDC00 && trailing <= 0xDFFF) { 182 /* followed by trailing surrogate */ 183 cp |= trailing & 0x3FF; 184 ptr += 6; 185 } else { 186 cp = -1; /* invalid */ 187 } 188 } 189 190 if (mod_utf8_encode(utf8_buf, sizeof(utf8_buf), cp) < 0) { 191 parse_error(ctxt, token, 192 "%.*s is not a valid Unicode character", 193 (int)(ptr - beg), beg); 194 goto out; 195 } 196 qstring_append(str, utf8_buf); 197 break; 198 default: 199 parse_error(ctxt, token, "invalid escape sequence in string"); 200 goto out; 201 } 202 } else { 203 cp = mod_utf8_codepoint(ptr, 6, &end); 204 if (cp < 0) { 205 parse_error(ctxt, token, "invalid UTF-8 sequence in string"); 206 goto out; 207 } 208 ptr = end; 209 len = mod_utf8_encode(utf8_buf, sizeof(utf8_buf), cp); 210 assert(len >= 0); 211 qstring_append(str, utf8_buf); 212 } 213 } 214 215 return str; 216 217 out: 218 qobject_unref(str); 219 return NULL; 220 } 221 222 /* Note: the token object returned by parser_context_peek_token or 223 * parser_context_pop_token is deleted as soon as parser_context_pop_token 224 * is called again. 225 */ 226 static JSONToken *parser_context_pop_token(JSONParserContext *ctxt) 227 { 228 g_free(ctxt->current); 229 assert(!g_queue_is_empty(ctxt->buf)); 230 ctxt->current = g_queue_pop_head(ctxt->buf); 231 return ctxt->current; 232 } 233 234 static JSONToken *parser_context_peek_token(JSONParserContext *ctxt) 235 { 236 assert(!g_queue_is_empty(ctxt->buf)); 237 return g_queue_peek_head(ctxt->buf); 238 } 239 240 /** 241 * Parsing rules 242 */ 243 static int parse_pair(JSONParserContext *ctxt, QDict *dict, va_list *ap) 244 { 245 QObject *value; 246 QString *key = NULL; 247 JSONToken *peek, *token; 248 249 peek = parser_context_peek_token(ctxt); 250 if (peek == NULL) { 251 parse_error(ctxt, NULL, "premature EOI"); 252 goto out; 253 } 254 255 key = qobject_to(QString, parse_value(ctxt, ap)); 256 if (!key) { 257 parse_error(ctxt, peek, "key is not a string in object"); 258 goto out; 259 } 260 261 token = parser_context_pop_token(ctxt); 262 if (token == NULL) { 263 parse_error(ctxt, NULL, "premature EOI"); 264 goto out; 265 } 266 267 if (token->type != JSON_COLON) { 268 parse_error(ctxt, token, "missing : in object pair"); 269 goto out; 270 } 271 272 value = parse_value(ctxt, ap); 273 if (value == NULL) { 274 parse_error(ctxt, token, "Missing value in dict"); 275 goto out; 276 } 277 278 qdict_put_obj(dict, qstring_get_str(key), value); 279 280 qobject_unref(key); 281 282 return 0; 283 284 out: 285 qobject_unref(key); 286 287 return -1; 288 } 289 290 static QObject *parse_object(JSONParserContext *ctxt, va_list *ap) 291 { 292 QDict *dict = NULL; 293 JSONToken *token, *peek; 294 295 token = parser_context_pop_token(ctxt); 296 assert(token && token->type == JSON_LCURLY); 297 298 dict = qdict_new(); 299 300 peek = parser_context_peek_token(ctxt); 301 if (peek == NULL) { 302 parse_error(ctxt, NULL, "premature EOI"); 303 goto out; 304 } 305 306 if (peek->type != JSON_RCURLY) { 307 if (parse_pair(ctxt, dict, ap) == -1) { 308 goto out; 309 } 310 311 token = parser_context_pop_token(ctxt); 312 if (token == NULL) { 313 parse_error(ctxt, NULL, "premature EOI"); 314 goto out; 315 } 316 317 while (token->type != JSON_RCURLY) { 318 if (token->type != JSON_COMMA) { 319 parse_error(ctxt, token, "expected separator in dict"); 320 goto out; 321 } 322 323 if (parse_pair(ctxt, dict, ap) == -1) { 324 goto out; 325 } 326 327 token = parser_context_pop_token(ctxt); 328 if (token == NULL) { 329 parse_error(ctxt, NULL, "premature EOI"); 330 goto out; 331 } 332 } 333 } else { 334 (void)parser_context_pop_token(ctxt); 335 } 336 337 return QOBJECT(dict); 338 339 out: 340 qobject_unref(dict); 341 return NULL; 342 } 343 344 static QObject *parse_array(JSONParserContext *ctxt, va_list *ap) 345 { 346 QList *list = NULL; 347 JSONToken *token, *peek; 348 349 token = parser_context_pop_token(ctxt); 350 assert(token && token->type == JSON_LSQUARE); 351 352 list = qlist_new(); 353 354 peek = parser_context_peek_token(ctxt); 355 if (peek == NULL) { 356 parse_error(ctxt, NULL, "premature EOI"); 357 goto out; 358 } 359 360 if (peek->type != JSON_RSQUARE) { 361 QObject *obj; 362 363 obj = parse_value(ctxt, ap); 364 if (obj == NULL) { 365 parse_error(ctxt, token, "expecting value"); 366 goto out; 367 } 368 369 qlist_append_obj(list, obj); 370 371 token = parser_context_pop_token(ctxt); 372 if (token == NULL) { 373 parse_error(ctxt, NULL, "premature EOI"); 374 goto out; 375 } 376 377 while (token->type != JSON_RSQUARE) { 378 if (token->type != JSON_COMMA) { 379 parse_error(ctxt, token, "expected separator in list"); 380 goto out; 381 } 382 383 obj = parse_value(ctxt, ap); 384 if (obj == NULL) { 385 parse_error(ctxt, token, "expecting value"); 386 goto out; 387 } 388 389 qlist_append_obj(list, obj); 390 391 token = parser_context_pop_token(ctxt); 392 if (token == NULL) { 393 parse_error(ctxt, NULL, "premature EOI"); 394 goto out; 395 } 396 } 397 } else { 398 (void)parser_context_pop_token(ctxt); 399 } 400 401 return QOBJECT(list); 402 403 out: 404 qobject_unref(list); 405 return NULL; 406 } 407 408 static QObject *parse_keyword(JSONParserContext *ctxt) 409 { 410 JSONToken *token; 411 412 token = parser_context_pop_token(ctxt); 413 assert(token && token->type == JSON_KEYWORD); 414 415 if (!strcmp(token->str, "true")) { 416 return QOBJECT(qbool_from_bool(true)); 417 } else if (!strcmp(token->str, "false")) { 418 return QOBJECT(qbool_from_bool(false)); 419 } else if (!strcmp(token->str, "null")) { 420 return QOBJECT(qnull()); 421 } 422 parse_error(ctxt, token, "invalid keyword '%s'", token->str); 423 return NULL; 424 } 425 426 static QObject *parse_interpolation(JSONParserContext *ctxt, va_list *ap) 427 { 428 JSONToken *token; 429 430 token = parser_context_pop_token(ctxt); 431 assert(token && token->type == JSON_INTERP); 432 433 if (!strcmp(token->str, "%p")) { 434 return va_arg(*ap, QObject *); 435 } else if (!strcmp(token->str, "%i")) { 436 return QOBJECT(qbool_from_bool(va_arg(*ap, int))); 437 } else if (!strcmp(token->str, "%d")) { 438 return QOBJECT(qnum_from_int(va_arg(*ap, int))); 439 } else if (!strcmp(token->str, "%ld")) { 440 return QOBJECT(qnum_from_int(va_arg(*ap, long))); 441 } else if (!strcmp(token->str, "%lld") || 442 !strcmp(token->str, "%I64d")) { 443 return QOBJECT(qnum_from_int(va_arg(*ap, long long))); 444 } else if (!strcmp(token->str, "%u")) { 445 return QOBJECT(qnum_from_uint(va_arg(*ap, unsigned int))); 446 } else if (!strcmp(token->str, "%lu")) { 447 return QOBJECT(qnum_from_uint(va_arg(*ap, unsigned long))); 448 } else if (!strcmp(token->str, "%llu") || 449 !strcmp(token->str, "%I64u")) { 450 return QOBJECT(qnum_from_uint(va_arg(*ap, unsigned long long))); 451 } else if (!strcmp(token->str, "%s")) { 452 return QOBJECT(qstring_from_str(va_arg(*ap, const char *))); 453 } else if (!strcmp(token->str, "%f")) { 454 return QOBJECT(qnum_from_double(va_arg(*ap, double))); 455 } 456 return NULL; 457 } 458 459 static QObject *parse_literal(JSONParserContext *ctxt) 460 { 461 JSONToken *token; 462 463 token = parser_context_pop_token(ctxt); 464 assert(token); 465 466 switch (token->type) { 467 case JSON_STRING: 468 return QOBJECT(parse_string(ctxt, token)); 469 case JSON_INTEGER: { 470 /* 471 * Represent JSON_INTEGER as QNUM_I64 if possible, else as 472 * QNUM_U64, else as QNUM_DOUBLE. Note that qemu_strtoi64() 473 * and qemu_strtou64() fail with ERANGE when it's not 474 * possible. 475 * 476 * qnum_get_int() will then work for any signed 64-bit 477 * JSON_INTEGER, qnum_get_uint() for any unsigned 64-bit 478 * integer, and qnum_get_double() both for any JSON_INTEGER 479 * and any JSON_FLOAT (with precision loss for integers beyond 480 * 53 bits) 481 */ 482 int ret; 483 int64_t value; 484 uint64_t uvalue; 485 486 ret = qemu_strtoi64(token->str, NULL, 10, &value); 487 if (!ret) { 488 return QOBJECT(qnum_from_int(value)); 489 } 490 assert(ret == -ERANGE); 491 492 if (token->str[0] != '-') { 493 ret = qemu_strtou64(token->str, NULL, 10, &uvalue); 494 if (!ret) { 495 return QOBJECT(qnum_from_uint(uvalue)); 496 } 497 assert(ret == -ERANGE); 498 } 499 /* fall through to JSON_FLOAT */ 500 } 501 case JSON_FLOAT: 502 /* FIXME dependent on locale; a pervasive issue in QEMU */ 503 /* FIXME our lexer matches RFC 7159 in forbidding Inf or NaN, 504 * but those might be useful extensions beyond JSON */ 505 return QOBJECT(qnum_from_double(strtod(token->str, NULL))); 506 default: 507 abort(); 508 } 509 } 510 511 static QObject *parse_value(JSONParserContext *ctxt, va_list *ap) 512 { 513 JSONToken *token; 514 515 token = parser_context_peek_token(ctxt); 516 if (token == NULL) { 517 parse_error(ctxt, NULL, "premature EOI"); 518 return NULL; 519 } 520 521 switch (token->type) { 522 case JSON_LCURLY: 523 return parse_object(ctxt, ap); 524 case JSON_LSQUARE: 525 return parse_array(ctxt, ap); 526 case JSON_INTERP: 527 return parse_interpolation(ctxt, ap); 528 case JSON_INTEGER: 529 case JSON_FLOAT: 530 case JSON_STRING: 531 return parse_literal(ctxt); 532 case JSON_KEYWORD: 533 return parse_keyword(ctxt); 534 default: 535 parse_error(ctxt, token, "expecting value"); 536 return NULL; 537 } 538 } 539 540 QObject *json_parser_parse(GQueue *tokens, va_list *ap, Error **errp) 541 { 542 JSONParserContext ctxt = { .buf = tokens }; 543 QObject *result; 544 545 result = parse_value(&ctxt, ap); 546 547 error_propagate(errp, ctxt.err); 548 549 while (!g_queue_is_empty(ctxt.buf)) { 550 parser_context_pop_token(&ctxt); 551 } 552 g_free(ctxt.current); 553 g_queue_free(ctxt.buf); 554 555 return result; 556 } 557