src/or/parsecommon.c

   1 /* Copyright (c) 2016-2017, The Tor Project, Inc. */
   2 /* See LICENSE for licensing information */
   3
   4 /**
   5  * \file parsecommon.c
   6  * \brief Common code to parse and validate various type of descriptors.
   7  **/
   8
   9 #include "parsecommon.h"
  10 #include "torlog.h"
  11 #include "util_format.h"
  12
  13 #define MIN_ANNOTATION A_PURPOSE
  14 #define MAX_ANNOTATION A_UNKNOWN_
  15
  16 #define ALLOC_ZERO(sz) memarea_alloc_zero(area,sz)
  17 #define ALLOC(sz) memarea_alloc(area,sz)
  18 #define STRDUP(str) memarea_strdup(area,str)
  19 #define STRNDUP(str,n) memarea_strndup(area,(str),(n))
  20
  21 #define RET_ERR(msg)                                               \
  22   STMT_BEGIN                                                       \
  23     if (tok) token_clear(tok);                                      \
  24     tok = ALLOC_ZERO(sizeof(directory_token_t));                   \
  25     tok->tp = ERR_;                                                \
  26     tok->error = STRDUP(msg);                                      \
  27     goto done_tokenizing;                                          \
  28   STMT_END
  29
  30 /** Free all resources allocated for <b>tok</b> */
  31 void
  32 token_clear(directory_token_t *tok)
  33 {
  34   if (tok->key)
  35     crypto_pk_free(tok->key);
  36 }
  37
  38 /** Read all tokens from a string between <b>start</b> and <b>end</b>, and add
  39  * them to <b>out</b>.  Parse according to the token rules in <b>table</b>.
  40  * Caller must free tokens in <b>out</b>.  If <b>end</b> is NULL, use the
  41  * entire string.
  42  */
  43 int
  44 tokenize_string(memarea_t *area,
  45                 const char *start, const char *end, smartlist_t *out,
  46                 token_rule_t *table, int flags)
  47 {
  48   const char **s;
  49   directory_token_t *tok = NULL;
  50   int counts[NIL_];
  51   int i;
  52   int first_nonannotation;
  53   int prev_len = smartlist_len(out);
  54   tor_assert(area);
  55
  56   s = &start;
  57   if (!end) {
  58     end = start+strlen(start);
  59   } else {
  60     /* it's only meaningful to check for nuls if we got an end-of-string ptr */
  61     if (memchr(start, '\0', end-start)) {
  62       log_warn(LD_DIR, "parse error: internal NUL character.");
  63       return -1;
  64     }
  65   }
  66   for (i = 0; i < NIL_; ++i)
  67     counts[i] = 0;
  68
  69   SMARTLIST_FOREACH(out, const directory_token_t *, t, ++counts[t->tp]);
  70
  71   while (*s < end && (!tok || tok->tp != EOF_)) {
  72     tok = get_next_token(area, s, end, table);
  73     if (tok->tp == ERR_) {
  74       log_warn(LD_DIR, "parse error: %s", tok->error);
  75       token_clear(tok);
  76       return -1;
  77     }
  78     ++counts[tok->tp];
  79     smartlist_add(out, tok);
  80     *s = eat_whitespace_eos(*s, end);
  81   }
  82
  83   if (flags & TS_NOCHECK)
  84     return 0;
  85
  86   if ((flags & TS_ANNOTATIONS_OK)) {
  87     first_nonannotation = -1;
  88     for (i = 0; i < smartlist_len(out); ++i) {
  89       tok = smartlist_get(out, i);
  90       if (tok->tp < MIN_ANNOTATION || tok->tp > MAX_ANNOTATION) {
  91         first_nonannotation = i;
  92         break;
  93       }
  94     }
  95     if (first_nonannotation < 0) {
  96       log_warn(LD_DIR, "parse error: item contains only annotations");
  97       return -1;
  98     }
  99     for (i=first_nonannotation;  i < smartlist_len(out); ++i) {
 100       tok = smartlist_get(out, i);
 101       if (tok->tp >= MIN_ANNOTATION && tok->tp <= MAX_ANNOTATION) {
 102         log_warn(LD_DIR, "parse error: Annotations mixed with keywords");
 103         return -1;
 104       }
 105     }
 106     if ((flags & TS_NO_NEW_ANNOTATIONS)) {
 107       if (first_nonannotation != prev_len) {
 108         log_warn(LD_DIR, "parse error: Unexpected annotations.");
 109         return -1;
 110       }
 111     }
 112   } else {
 113     for (i=0;  i < smartlist_len(out); ++i) {
 114       tok = smartlist_get(out, i);
 115       if (tok->tp >= MIN_ANNOTATION && tok->tp <= MAX_ANNOTATION) {
 116         log_warn(LD_DIR, "parse error: no annotations allowed.");
 117         return -1;
 118       }
 119     }
 120     first_nonannotation = 0;
 121   }
 122   for (i = 0; table[i].t; ++i) {
 123     if (counts[table[i].v] < table[i].min_cnt) {
 124       log_warn(LD_DIR, "Parse error: missing %s element.", table[i].t);
 125       return -1;
 126     }
 127     if (counts[table[i].v] > table[i].max_cnt) {
 128       log_warn(LD_DIR, "Parse error: too many %s elements.", table[i].t);
 129       return -1;
 130     }
 131     if (table[i].pos & AT_START) {
 132       if (smartlist_len(out) < 1 ||
 133           (tok = smartlist_get(out, first_nonannotation))->tp != table[i].v) {
 134         log_warn(LD_DIR, "Parse error: first item is not %s.", table[i].t);
 135         return -1;
 136       }
 137     }
 138     if (table[i].pos & AT_END) {
 139       if (smartlist_len(out) < 1 ||
 140           (tok = smartlist_get(out, smartlist_len(out)-1))->tp != table[i].v) {
 141         log_warn(LD_DIR, "Parse error: last item is not %s.", table[i].t);
 142         return -1;
 143       }
 144     }
 145   }
 146   return 0;
 147 }
 148
 149 /** Helper: parse space-separated arguments from the string <b>s</b> ending at
 150  * <b>eol</b>, and store them in the args field of <b>tok</b>.  Store the
 151  * number of parsed elements into the n_args field of <b>tok</b>.  Allocate
 152  * all storage in <b>area</b>.  Return the number of arguments parsed, or
 153  * return -1 if there was an insanely high number of arguments. */
 154 static inline int
 155 get_token_arguments(memarea_t *area, directory_token_t *tok,
 156                     const char *s, const char *eol)
 157 {
 158 /** Largest number of arguments we'll accept to any token, ever. */
 159 #define MAX_ARGS 512
 160   char *mem = memarea_strndup(area, s, eol-s);
 161   char *cp = mem;
 162   int j = 0;
 163   char *args[MAX_ARGS];
 164   while (*cp) {
 165     if (j == MAX_ARGS)
 166       return -1;
 167     args[j++] = cp;
 168     cp = (char*)find_whitespace(cp);
 169     if (!cp || !*cp)
 170       break; /* End of the line. */
 171     *cp++ = '\0';
 172     cp = (char*)eat_whitespace(cp);
 173   }
 174   tok->n_args = j;
 175   tok->args = memarea_memdup(area, args, j*sizeof(char*));
 176   return j;
 177 #undef MAX_ARGS
 178 }
 179
 180 /** Helper: make sure that the token <b>tok</b> with keyword <b>kwd</b> obeys
 181  * the object syntax of <b>o_syn</b>.  Allocate all storage in <b>area</b>.
 182  * Return <b>tok</b> on success, or a new ERR_ token if the token didn't
 183  * conform to the syntax we wanted.
 184  **/
 185 static inline directory_token_t *
 186 token_check_object(memarea_t *area, const char *kwd,
 187                    directory_token_t *tok, obj_syntax o_syn)
 188 {
 189   char ebuf[128];
 190   switch (o_syn) {
 191     case NO_OBJ:
 192       /* No object is allowed for this token. */
 193       if (tok->object_body) {
 194         tor_snprintf(ebuf, sizeof(ebuf), "Unexpected object for %s", kwd);
 195         RET_ERR(ebuf);
 196       }
 197       if (tok->key) {
 198         tor_snprintf(ebuf, sizeof(ebuf), "Unexpected public key for %s", kwd);
 199         RET_ERR(ebuf);
 200       }
 201       break;
 202     case NEED_OBJ:
 203       /* There must be a (non-key) object. */
 204       if (!tok->object_body) {
 205         tor_snprintf(ebuf, sizeof(ebuf), "Missing object for %s", kwd);
 206         RET_ERR(ebuf);
 207       }
 208       break;
 209     case NEED_KEY_1024: /* There must be a 1024-bit public key. */
 210     case NEED_SKEY_1024: /* There must be a 1024-bit private key. */
 211       if (tok->key && crypto_pk_num_bits(tok->key) != PK_BYTES*8) {
 212         tor_snprintf(ebuf, sizeof(ebuf), "Wrong size on key for %s: %d bits",
 213                      kwd, crypto_pk_num_bits(tok->key));
 214         RET_ERR(ebuf);
 215       }
 216       /* fall through */
 217     case NEED_KEY: /* There must be some kind of key. */
 218       if (!tok->key) {
 219         tor_snprintf(ebuf, sizeof(ebuf), "Missing public key for %s", kwd);
 220         RET_ERR(ebuf);
 221       }
 222       if (o_syn != NEED_SKEY_1024) {
 223         if (crypto_pk_key_is_private(tok->key)) {
 224           tor_snprintf(ebuf, sizeof(ebuf),
 225                "Private key given for %s, which wants a public key", kwd);
 226           RET_ERR(ebuf);
 227         }
 228       } else { /* o_syn == NEED_SKEY_1024 */
 229         if (!crypto_pk_key_is_private(tok->key)) {
 230           tor_snprintf(ebuf, sizeof(ebuf),
 231                "Public key given for %s, which wants a private key", kwd);
 232           RET_ERR(ebuf);
 233         }
 234       }
 235       break;
 236     case OBJ_OK:
 237       /* Anything goes with this token. */
 238       break;
 239   }
 240
 241  done_tokenizing:
 242   return tok;
 243 }
 244
 245 /** Helper function: read the next token from *s, advance *s to the end of the
 246  * token, and return the parsed token.  Parse *<b>s</b> according to the list
 247  * of tokens in <b>table</b>.
 248  */
 249 directory_token_t *
 250 get_next_token(memarea_t *area,
 251                const char **s, const char *eos, token_rule_t *table)
 252 {
 253   /** Reject any object at least this big; it is probably an overflow, an
 254    * attack, a bug, or some other nonsense. */
 255 #define MAX_UNPARSED_OBJECT_SIZE (128*1024)
 256   /** Reject any line at least this big; it is probably an overflow, an
 257    * attack, a bug, or some other nonsense. */
 258 #define MAX_LINE_LENGTH (128*1024)
 259
 260   const char *next, *eol, *obstart;
 261   size_t obname_len;
 262   int i;
 263   directory_token_t *tok;
 264   obj_syntax o_syn = NO_OBJ;
 265   char ebuf[128];
 266   const char *kwd = "";
 267
 268   tor_assert(area);
 269   tok = ALLOC_ZERO(sizeof(directory_token_t));
 270   tok->tp = ERR_;
 271
 272   /* Set *s to first token, eol to end-of-line, next to after first token */
 273   *s = eat_whitespace_eos(*s, eos); /* eat multi-line whitespace */
 274   tor_assert(eos >= *s);
 275   eol = memchr(*s, '\n', eos-*s);
 276   if (!eol)
 277     eol = eos;
 278   if (eol - *s > MAX_LINE_LENGTH) {
 279     RET_ERR("Line far too long");
 280   }
 281
 282   next = find_whitespace_eos(*s, eol);
 283
 284   if (!strcmp_len(*s, "opt", next-*s)) {
 285     /* Skip past an "opt" at the start of the line. */
 286     *s = eat_whitespace_eos_no_nl(next, eol);
 287     next = find_whitespace_eos(*s, eol);
 288   } else if (*s == eos) {  /* If no "opt", and end-of-line, line is invalid */
 289     RET_ERR("Unexpected EOF");
 290   }
 291
 292   /* Search the table for the appropriate entry.  (I tried a binary search
 293    * instead, but it wasn't any faster.) */
 294   for (i = 0; table[i].t ; ++i) {
 295     if (!strcmp_len(*s, table[i].t, next-*s)) {
 296       /* We've found the keyword. */
 297       kwd = table[i].t;
 298       tok->tp = table[i].v;
 299       o_syn = table[i].os;
 300       *s = eat_whitespace_eos_no_nl(next, eol);
 301       /* We go ahead whether there are arguments or not, so that tok->args is
 302        * always set if we want arguments. */
 303       if (table[i].concat_args) {
 304         /* The keyword takes the line as a single argument */
 305         tok->args = ALLOC(sizeof(char*));
 306         tok->args[0] = STRNDUP(*s,eol-*s); /* Grab everything on line */
 307         tok->n_args = 1;
 308       } else {
 309         /* This keyword takes multiple arguments. */
 310         if (get_token_arguments(area, tok, *s, eol)<0) {
 311           tor_snprintf(ebuf, sizeof(ebuf),"Far too many arguments to %s", kwd);
 312           RET_ERR(ebuf);
 313         }
 314         *s = eol;
 315       }
 316       if (tok->n_args < table[i].min_args) {
 317         tor_snprintf(ebuf, sizeof(ebuf), "Too few arguments to %s", kwd);
 318         RET_ERR(ebuf);
 319       } else if (tok->n_args > table[i].max_args) {
 320         tor_snprintf(ebuf, sizeof(ebuf), "Too many arguments to %s", kwd);
 321         RET_ERR(ebuf);
 322       }
 323       break;
 324     }
 325   }
 326
 327   if (tok->tp == ERR_) {
 328     /* No keyword matched; call it an "K_opt" or "A_unrecognized" */
 329     if (*s < eol && **s == '@')
 330       tok->tp = A_UNKNOWN_;
 331     else
 332       tok->tp = K_OPT;
 333     tok->args = ALLOC(sizeof(char*));
 334     tok->args[0] = STRNDUP(*s, eol-*s);
 335     tok->n_args = 1;
 336     o_syn = OBJ_OK;
 337   }
 338
 339   /* Check whether there's an object present */
 340   *s = eat_whitespace_eos(eol, eos);  /* Scan from end of first line */
 341   tor_assert(eos >= *s);
 342   eol = memchr(*s, '\n', eos-*s);
 343   if (!eol || eol-*s<11 || strcmpstart(*s, "-----BEGIN ")) /* No object. */
 344     goto check_object;
 345
 346   obstart = *s; /* Set obstart to start of object spec */
 347   if (*s+16 >= eol || memchr(*s+11,'\0',eol-*s-16) || /* no short lines, */
 348       strcmp_len(eol-5, "-----", 5) ||           /* nuls or invalid endings */
 349       (eol-*s) > MAX_UNPARSED_OBJECT_SIZE) {     /* name too long */
 350     RET_ERR("Malformed object: bad begin line");
 351   }
 352   tok->object_type = STRNDUP(*s+11, eol-*s-16);
 353   obname_len = eol-*s-16; /* store objname length here to avoid a strlen() */
 354   *s = eol+1;    /* Set *s to possible start of object data (could be eos) */
 355
 356   /* Go to the end of the object */
 357   next = tor_memstr(*s, eos-*s, "-----END ");
 358   if (!next) {
 359     RET_ERR("Malformed object: missing object end line");
 360   }
 361   tor_assert(eos >= next);
 362   eol = memchr(next, '\n', eos-next);
 363   if (!eol)  /* end-of-line marker, or eos if there's no '\n' */
 364     eol = eos;
 365   /* Validate the ending tag, which should be 9 + NAME + 5 + eol */
 366   if ((size_t)(eol-next) != 9+obname_len+5 ||
 367       strcmp_len(next+9, tok->object_type, obname_len) ||
 368       strcmp_len(eol-5, "-----", 5)) {
 369     tor_snprintf(ebuf, sizeof(ebuf), "Malformed object: mismatched end tag %s",
 370              tok->object_type);
 371     ebuf[sizeof(ebuf)-1] = '\0';
 372     RET_ERR(ebuf);
 373   }
 374   if (next - *s > MAX_UNPARSED_OBJECT_SIZE)
 375     RET_ERR("Couldn't parse object: missing footer or object much too big.");
 376
 377   if (!strcmp(tok->object_type, "RSA PUBLIC KEY")) { /* If it's a public key */
 378     tok->key = crypto_pk_new();
 379     if (crypto_pk_read_public_key_from_string(tok->key, obstart, eol-obstart))
 380       RET_ERR("Couldn't parse public key.");
 381   } else if (!strcmp(tok->object_type, "RSA PRIVATE KEY")) { /* private key */
 382     tok->key = crypto_pk_new();
 383     if (crypto_pk_read_private_key_from_string(tok->key, obstart, eol-obstart))
 384       RET_ERR("Couldn't parse private key.");
 385   } else { /* If it's something else, try to base64-decode it */
 386     int r;
 387     tok->object_body = ALLOC(next-*s); /* really, this is too much RAM. */
 388     r = base64_decode(tok->object_body, next-*s, *s, next-*s);
 389     if (r<0)
 390       RET_ERR("Malformed object: bad base64-encoded data");
 391     tok->object_size = r;
 392   }
 393   *s = eol;
 394
 395  check_object:
 396   tok = token_check_object(area, kwd, tok, o_syn);
 397
 398  done_tokenizing:
 399   return tok;
 400
 401 #undef RET_ERR
 402 #undef ALLOC
 403 #undef ALLOC_ZERO
 404 #undef STRDUP
 405 #undef STRNDUP
 406 }
 407
 408 /** Find the first token in <b>s</b> whose keyword is <b>keyword</b>; fail
 409  * with an assert if no such keyword is found.
 410  */
 411 directory_token_t *
 412 find_by_keyword_(smartlist_t *s, directory_keyword keyword,
 413                  const char *keyword_as_string)
 414 {
 415   directory_token_t *tok = find_opt_by_keyword(s, keyword);
 416   if (PREDICT_UNLIKELY(!tok)) {
 417     log_err(LD_BUG, "Missing %s [%d] in directory object that should have "
 418          "been validated. Internal error.", keyword_as_string, (int)keyword);
 419     tor_assert(tok);
 420   }
 421   return tok;
 422 }
 423
 424 /** Find the first token in <b>s</b> whose keyword is <b>keyword</b>; return
 425  * NULL if no such keyword is found.
 426  */
 427 directory_token_t *
 428 find_opt_by_keyword(smartlist_t *s, directory_keyword keyword)
 429 {
 430   SMARTLIST_FOREACH(s, directory_token_t *, t, if (t->tp == keyword) return t);
 431   return NULL;
 432 }
 433
 434 /** If there are any directory_token_t entries in <b>s</b> whose keyword is
 435  * <b>k</b>, return a newly allocated smartlist_t containing all such entries,
 436  * in the same order in which they occur in <b>s</b>.  Otherwise return
 437  * NULL. */
 438 smartlist_t *
 439 find_all_by_keyword(const smartlist_t *s, directory_keyword k)
 440 {
 441   smartlist_t *out = NULL;
 442   SMARTLIST_FOREACH(s, directory_token_t *, t,
 443                     if (t->tp == k) {
 444                     if (!out)
 445                     out = smartlist_new();
 446                     smartlist_add(out, t);
 447                     });
 448   return out;
 449 }
 450