1 /*
   2  * Copyright (C) 2011 Dan Carpenter.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public License
   6  * as published by the Free Software Foundation; either version 2
   7  * of the License, or (at your option) any later version.
   8  *
   9  * This program 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
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, see http://www.gnu.org/copyleft/gpl.txt
  16  */
  17 
  18 /*
  19  * There are a couple checks that try to see if a variable
  20  * comes from the user.  It would be better to unify them
  21  * into one place.  Also it we should follow the data down
  22  * the call paths.  Hence this file.
  23  */
  24 
  25 #include "smatch.h"
  26 #include "smatch_slist.h"
  27 #include "smatch_extra.h"
  28 
  29 static int my_id;
  30 static int my_call_id;
  31 
  32 STATE(called);
  33 static bool func_gets_user_data;
  34 
  35 static const char * kstr_funcs[] = {
  36         "kstrtoull", "kstrtoll", "kstrtoul", "kstrtol", "kstrtouint",
  37         "kstrtoint", "kstrtou64", "kstrtos64", "kstrtou32", "kstrtos32",
  38         "kstrtou16", "kstrtos16", "kstrtou8", "kstrtos8", "kstrtoull_from_user"
  39         "kstrtoll_from_user", "kstrtoul_from_user", "kstrtol_from_user",
  40         "kstrtouint_from_user", "kstrtoint_from_user", "kstrtou16_from_user",
  41         "kstrtos16_from_user", "kstrtou8_from_user", "kstrtos8_from_user",
  42         "kstrtou64_from_user", "kstrtos64_from_user", "kstrtou32_from_user",
  43         "kstrtos32_from_user",
  44 };
  45 
  46 static const char *returns_user_data[] = {
  47         "simple_strtol", "simple_strtoll", "simple_strtoul", "simple_strtoull",
  48         "kvm_register_read", "nlmsg_data", "nla_data", "memdup_user",
  49         "kmap_atomic", "skb_network_header",
  50 };
  51 
  52 static void set_points_to_user_data(struct expression *expr);
  53 
  54 static struct stree *start_states;
  55 static struct stree_stack *saved_stack;
  56 static void save_start_states(struct statement *stmt)
  57 {
  58         start_states = clone_stree(__get_cur_stree());
  59 }
  60 
  61 static void free_start_states(void)
  62 {
  63         free_stree(&start_states);
  64 }
  65 
  66 static void match_save_states(struct expression *expr)
  67 {
  68         push_stree(&saved_stack, start_states);
  69         start_states = NULL;
  70 }
  71 
  72 static void match_restore_states(struct expression *expr)
  73 {
  74         free_stree(&start_states);
  75         start_states = pop_stree(&saved_stack);
  76 }
  77 
  78 static struct smatch_state *empty_state(struct sm_state *sm)
  79 {
  80         return alloc_estate_empty();
  81 }
  82 
  83 static void pre_merge_hook(struct sm_state *sm)
  84 {
  85         struct smatch_state *user;
  86         struct smatch_state *extra;
  87         struct range_list *rl;
  88         sval_t dummy;
  89         sval_t sval_100;
  90 
  91         sval_100.value = 100;
  92         sval_100.type = &int_ctype;
  93 
  94         user = get_state(my_id, sm->name, sm->sym);
  95         if (!user)
  96                 return;
  97         if (!__in_function_def && !estate_rl(sm->state)) {
  98                 /*
  99                  * If the one side is capped and the other side is empty then
 100                  * let's just mark it as not-user data because the information
 101                  * isn't going to be useful.  How this looks is:
 102                  *
 103                  * if (user_var > trusted)
 104                  *      user_var = trusted;  <-- empty state
 105                  * else
 106                  *      <-- capped
 107                  *
 108                  * The problem is that sometimes things are capped to a literal
 109                  * and we'd like to keep the state in that case...  Ugh.  I've
 110                  * added a check which assumes that everything less than 100 is
 111                  * probably capped against a literal.
 112                  *
 113                  */
 114                 if (is_capped_var_sym(sm->name, sm->sym) &&
 115                     sval_cmp(estate_max(user), sval_100) > 0)
 116                         set_state(my_id, sm->name, sm->sym, alloc_estate_empty());
 117                 return;
 118         }
 119         extra = get_state(SMATCH_EXTRA, sm->name, sm->sym);
 120         if (!extra || !estate_rl(extra))
 121                 return;
 122         rl = rl_intersection(estate_rl(user), estate_rl(extra));
 123         if (rl_to_sval(rl, &dummy))
 124                 rl = NULL;
 125         set_state(my_id, sm->name, sm->sym, alloc_estate_rl(clone_rl(rl)));
 126 }
 127 
 128 static void extra_nomod_hook(const char *name, struct symbol *sym, struct expression *expr, struct smatch_state *state)
 129 {
 130         struct smatch_state *user;
 131         struct range_list *rl;
 132 
 133         user = get_state(my_id, name, sym);
 134         if (!user)
 135                 return;
 136         rl = rl_intersection(estate_rl(user), estate_rl(state));
 137         if (rl_equiv(rl, estate_rl(user)))
 138                 return;
 139         set_state(my_id, name, sym, alloc_estate_rl(rl));
 140 }
 141 
 142 static void tag_inner_struct_members(struct expression *expr, struct symbol *member)
 143 {
 144         struct expression *edge_member;
 145         struct symbol *base = get_real_base_type(member);
 146         struct symbol *tmp;
 147 
 148         if (member->ident)
 149                 expr = member_expression(expr, '.', member->ident);
 150 
 151         FOR_EACH_PTR(base->symbol_list, tmp) {
 152                 struct symbol *type;
 153 
 154                 type = get_real_base_type(tmp);
 155                 if (!type)
 156                         continue;
 157 
 158                 if (type->type == SYM_UNION || type->type == SYM_STRUCT) {
 159                         tag_inner_struct_members(expr, tmp);
 160                         continue;
 161                 }
 162 
 163                 if (!tmp->ident)
 164                         continue;
 165 
 166                 edge_member = member_expression(expr, '.', tmp->ident);
 167                 set_state_expr(my_id, edge_member, alloc_estate_whole(type));
 168         } END_FOR_EACH_PTR(tmp);
 169 }
 170 
 171 static void tag_struct_members(struct symbol *type, struct expression *expr)
 172 {
 173         struct symbol *tmp;
 174         struct expression *member;
 175         int op = '*';
 176 
 177         if (expr->type == EXPR_PREOP && expr->op == '&') {
 178                 expr = strip_expr(expr->unop);
 179                 op = '.';
 180         }
 181 
 182         FOR_EACH_PTR(type->symbol_list, tmp) {
 183                 type = get_real_base_type(tmp);
 184                 if (!type)
 185                         continue;
 186 
 187                 if (type->type == SYM_UNION || type->type == SYM_STRUCT) {
 188                         tag_inner_struct_members(expr, tmp);
 189                         continue;
 190                 }
 191 
 192                 if (!tmp->ident)
 193                         continue;
 194 
 195                 member = member_expression(expr, op, tmp->ident);
 196                 set_state_expr(my_id, member, alloc_estate_whole(get_type(member)));
 197 
 198                 if (type->type == SYM_ARRAY)
 199                         set_points_to_user_data(member);
 200         } END_FOR_EACH_PTR(tmp);
 201 }
 202 
 203 static void tag_base_type(struct expression *expr)
 204 {
 205         if (expr->type == EXPR_PREOP && expr->op == '&')
 206                 expr = strip_expr(expr->unop);
 207         else
 208                 expr = deref_expression(expr);
 209         set_state_expr(my_id, expr, alloc_estate_whole(get_type(expr)));
 210 }
 211 
 212 static void tag_as_user_data(struct expression *expr)
 213 {
 214         struct symbol *type;
 215 
 216         expr = strip_expr(expr);
 217 
 218         type = get_type(expr);
 219         if (!type || type->type != SYM_PTR)
 220                 return;
 221         type = get_real_base_type(type);
 222         if (!type)
 223                 return;
 224         if (type == &void_ctype) {
 225                 set_state_expr(my_id, deref_expression(expr), alloc_estate_whole(&ulong_ctype));
 226                 return;
 227         }
 228         if (type->type == SYM_BASETYPE)
 229                 tag_base_type(expr);
 230         if (type->type == SYM_STRUCT || type->type == SYM_UNION) {
 231                 if (expr->type != EXPR_PREOP || expr->op != '&')
 232                         expr = deref_expression(expr);
 233                 else
 234                         set_state_expr(my_id, deref_expression(expr), alloc_estate_whole(&ulong_ctype));
 235                 tag_struct_members(type, expr);
 236         }
 237 }
 238 
 239 static void match_user_copy(const char *fn, struct expression *expr, void *_param)
 240 {
 241         int param = PTR_INT(_param);
 242         struct expression *dest;
 243 
 244         func_gets_user_data = true;
 245 
 246         dest = get_argument_from_call_expr(expr->args, param);
 247         dest = strip_expr(dest);
 248         if (!dest)
 249                 return;
 250         tag_as_user_data(dest);
 251 }
 252 
 253 static int is_dev_attr_name(struct expression *expr)
 254 {
 255         char *name;
 256         int ret = 0;
 257 
 258         name = expr_to_str(expr);
 259         if (!name)
 260                 return 0;
 261         if (strstr(name, "->attr.name"))
 262                 ret = 1;
 263         free_string(name);
 264         return ret;
 265 }
 266 
 267 static int ends_in_n(struct expression *expr)
 268 {
 269         struct string *str;
 270 
 271         if (!expr)
 272                 return 0;
 273         if (expr->type != EXPR_STRING || !expr->string)
 274                 return 0;
 275 
 276         str = expr->string;
 277         if (str->length < 3)
 278                 return 0;
 279 
 280         if (str->data[str->length - 3] == '%' &&
 281             str->data[str->length - 2] == 'n')
 282                 return 1;
 283         return 0;
 284 }
 285 
 286 static void match_sscanf(const char *fn, struct expression *expr, void *unused)
 287 {
 288         struct expression *str, *format, *arg;
 289         int i, last;
 290 
 291         func_gets_user_data = true;
 292 
 293         str = get_argument_from_call_expr(expr->args, 0);
 294         if (is_dev_attr_name(str))
 295                 return;
 296 
 297         format = get_argument_from_call_expr(expr->args, 1);
 298         if (is_dev_attr_name(format))
 299                 return;
 300 
 301         last = ptr_list_size((struct ptr_list *)expr->args) - 1;
 302 
 303         i = -1;
 304         FOR_EACH_PTR(expr->args, arg) {
 305                 i++;
 306                 if (i < 2)
 307                         continue;
 308                 if (i == last && ends_in_n(format))
 309                         continue;
 310                 tag_as_user_data(arg);
 311         } END_FOR_EACH_PTR(arg);
 312 }
 313 
 314 static int is_skb_data(struct expression *expr)
 315 {
 316         struct symbol *sym;
 317 
 318         if (!expr)
 319                 return 0;
 320 
 321         if (expr->type == EXPR_BINOP && expr->op == '+')
 322                 return is_skb_data(expr->left);
 323 
 324         expr = strip_expr(expr);
 325         if (!expr)
 326                 return 0;
 327         if (expr->type != EXPR_DEREF || expr->op != '.')
 328                 return 0;
 329 
 330         if (!expr->member)
 331                 return 0;
 332         if (strcmp(expr->member->name, "data") != 0)
 333                 return 0;
 334 
 335         sym = expr_to_sym(expr->deref);
 336         if (!sym)
 337                 return 0;
 338         sym = get_real_base_type(sym);
 339         if (!sym || sym->type != SYM_PTR)
 340                 return 0;
 341         sym = get_real_base_type(sym);
 342         if (!sym || sym->type != SYM_STRUCT || !sym->ident)
 343                 return 0;
 344         if (strcmp(sym->ident->name, "sk_buff") != 0)
 345                 return 0;
 346 
 347         return 1;
 348 }
 349 
 350 static int get_rl_from_function(struct expression *expr, struct range_list **rl)
 351 {
 352         int i;
 353 
 354         if (expr->type != EXPR_CALL || expr->fn->type != EXPR_SYMBOL ||
 355             !expr->fn->symbol_name || !expr->fn->symbol_name->name)
 356                 return 0;
 357 
 358         for (i = 0; i < ARRAY_SIZE(returns_user_data); i++) {
 359                 if (strcmp(expr->fn->symbol_name->name, returns_user_data[i]) == 0) {
 360                         *rl = alloc_whole_rl(get_type(expr));
 361                         return 1;
 362                 }
 363         }
 364         return 0;
 365 }
 366 
 367 int points_to_user_data(struct expression *expr)
 368 {
 369         struct smatch_state *state;
 370         struct range_list *rl;
 371         char buf[256];
 372         struct symbol *sym;
 373         char *name;
 374         int ret = 0;
 375 
 376         expr = strip_expr(expr);
 377         if (!expr)
 378                 return 0;
 379         if (is_skb_data(expr))
 380                 return 1;
 381         if (get_rl_from_function(expr, &rl))
 382                 return 1;
 383 
 384         if (expr->type == EXPR_BINOP && expr->op == '+') {
 385                 if (points_to_user_data(expr->left))
 386                         return 1;
 387                 if (points_to_user_data(expr->right))
 388                         return 1;
 389                 return 0;
 390         }
 391 
 392         name = expr_to_var_sym(expr, &sym);
 393         if (!name || !sym)
 394                 goto free;
 395         snprintf(buf, sizeof(buf), "*%s", name);
 396         state = get_state(my_id, buf, sym);
 397         if (state && estate_rl(state))
 398                 ret = 1;
 399 free:
 400         free_string(name);
 401         return ret;
 402 }
 403 
 404 static void set_points_to_user_data(struct expression *expr)
 405 {
 406         char *name;
 407         struct symbol *sym;
 408         char buf[256];
 409 
 410         name = expr_to_var_sym(expr, &sym);
 411         if (!name || !sym)
 412                 goto free;
 413         snprintf(buf, sizeof(buf), "*%s", name);
 414         set_state(my_id, buf, sym, alloc_estate_whole(&llong_ctype));
 415 free:
 416         free_string(name);
 417 }
 418 
 419 static int comes_from_skb_data(struct expression *expr)
 420 {
 421         expr = strip_expr(expr);
 422         if (!expr || expr->type != EXPR_PREOP || expr->op != '*')
 423                 return 0;
 424 
 425         expr = strip_expr(expr->unop);
 426         if (!expr)
 427                 return 0;
 428         if (expr->type == EXPR_BINOP && expr->op == '+')
 429                 expr = strip_expr(expr->left);
 430 
 431         return is_skb_data(expr);
 432 }
 433 
 434 static int handle_struct_assignment(struct expression *expr)
 435 {
 436         struct expression *right;
 437         struct symbol *left_type, *right_type;
 438 
 439         left_type = get_type(expr->left);
 440         if (!left_type || left_type->type != SYM_PTR)
 441                 return 0;
 442         left_type = get_real_base_type(left_type);
 443         if (!left_type)
 444                 return 0;
 445         if (left_type->type != SYM_STRUCT &&
 446             left_type->type != SYM_UNION)
 447                 return 0;
 448 
 449         /*
 450          * Ignore struct to struct assignments because for those we look at the
 451          * individual members.
 452          */
 453         right = strip_expr(expr->right);
 454         right_type = get_type(right);
 455         if (!right_type || right_type->type != SYM_PTR)
 456                 return 0;
 457 
 458         /* If we are assigning struct members then normally that is handled
 459          * by fake assignments, however if we cast one struct to a different
 460          * of struct then we handle that here.
 461          */
 462         right_type = get_real_base_type(right_type);
 463         if (right_type == left_type)
 464                 return 0;
 465 
 466         if (!points_to_user_data(right))
 467                 return 0;
 468 
 469         tag_as_user_data(expr->left);
 470         return 1;
 471 }
 472 
 473 static int handle_get_user(struct expression *expr)
 474 {
 475         char *name;
 476         int ret = 0;
 477 
 478         name = get_macro_name(expr->pos);
 479         if (!name || strcmp(name, "get_user") != 0)
 480                 return 0;
 481 
 482         name = expr_to_var(expr->right);
 483         if (!name || strcmp(name, "__val_gu") != 0)
 484                 goto free;
 485         set_state_expr(my_id, expr->left, alloc_estate_whole(get_type(expr->left)));
 486         ret = 1;
 487 free:
 488         free_string(name);
 489         return ret;
 490 }
 491 
 492 static void match_assign(struct expression *expr)
 493 {
 494         struct range_list *rl;
 495 
 496         if (is_fake_call(expr->right))
 497                 goto clear_old_state;
 498         if (handle_get_user(expr))
 499                 return;
 500         if (points_to_user_data(expr->right))
 501                 set_points_to_user_data(expr->left);
 502         if (handle_struct_assignment(expr))
 503                 return;
 504 
 505         if (!get_user_rl(expr->right, &rl))
 506                 goto clear_old_state;
 507 
 508         rl = cast_rl(get_type(expr->left), rl);
 509         set_state_expr(my_id, expr->left, alloc_estate_rl(rl));
 510 
 511         return;
 512 
 513 clear_old_state:
 514         if (get_state_expr(my_id, expr->left))
 515                 set_state_expr(my_id, expr->left, alloc_estate_empty());
 516 }
 517 
 518 static void handle_eq_noteq(struct expression *expr)
 519 {
 520         struct smatch_state *left_orig, *right_orig;
 521 
 522         left_orig = get_state_expr(my_id, expr->left);
 523         right_orig = get_state_expr(my_id, expr->right);
 524 
 525         if (!left_orig && !right_orig)
 526                 return;
 527         if (left_orig && right_orig)
 528                 return;
 529 
 530         if (left_orig) {
 531                 set_true_false_states_expr(my_id, expr->left,
 532                                 expr->op == SPECIAL_EQUAL ? alloc_estate_empty() : NULL,
 533                                 expr->op == SPECIAL_EQUAL ? NULL : alloc_estate_empty());
 534         } else {
 535                 set_true_false_states_expr(my_id, expr->right,
 536                                 expr->op == SPECIAL_EQUAL ? alloc_estate_empty() : NULL,
 537                                 expr->op == SPECIAL_EQUAL ? NULL : alloc_estate_empty());
 538         }
 539 }
 540 
 541 static void handle_unsigned_lt_gt(struct expression *expr)
 542 {
 543         struct symbol *type;
 544         struct range_list *left;
 545         struct range_list *right;
 546         struct range_list *non_negative;
 547         sval_t min, minus_one;
 548 
 549         /*
 550          * conditions are mostly handled by smatch_extra.c.  The special case
 551          * here is that say you have if (user_int < unknown_u32) {
 552          * In Smatch extra we say that, We have no idea what value
 553          * unknown_u32 is so the only thin we can say for sure is that
 554          * user_int is not -1 (UINT_MAX).  But in check_user_data2.c we should
 555          * assume that unless unknown_u32 is user data, it's probably less than
 556          * INT_MAX.
 557          *
 558          */
 559 
 560         type = get_type(expr);
 561         if (!type_unsigned(type))
 562                 return;
 563 
 564         /*
 565          * Assume if (user < trusted) { ... because I am lazy and because this
 566          * is the correct way to write code.
 567          */
 568         if (!get_user_rl(expr->left, &left))
 569                 return;
 570         if (get_user_rl(expr->right, &right))
 571                 return;
 572 
 573         if (!sval_is_negative(rl_min(left)))
 574                 return;
 575         min = rl_min(left);
 576         minus_one.type = rl_type(left);
 577         minus_one.value = -1;
 578         non_negative = remove_range(left, min, minus_one);
 579 
 580         switch (expr->op) {
 581         case '<':
 582         case SPECIAL_UNSIGNED_LT:
 583         case SPECIAL_LTE:
 584         case SPECIAL_UNSIGNED_LTE:
 585                 set_true_false_states_expr(my_id, expr->left,
 586                                            alloc_estate_rl(non_negative), NULL);
 587                 break;
 588         case '>':
 589         case SPECIAL_UNSIGNED_GT:
 590         case SPECIAL_GTE:
 591         case SPECIAL_UNSIGNED_GTE:
 592                 set_true_false_states_expr(my_id, expr->left,
 593                                            NULL, alloc_estate_rl(non_negative));
 594                 break;
 595         }
 596 }
 597 
 598 static void match_condition(struct expression *expr)
 599 {
 600         if (expr->type != EXPR_COMPARE)
 601                 return;
 602 
 603         if (expr->op == SPECIAL_EQUAL ||
 604             expr->op == SPECIAL_NOTEQUAL) {
 605                 handle_eq_noteq(expr);
 606                 return;
 607         }
 608 
 609         handle_unsigned_lt_gt(expr);
 610 }
 611 
 612 static void match_user_assign_function(const char *fn, struct expression *expr, void *unused)
 613 {
 614         tag_as_user_data(expr->left);
 615         set_points_to_user_data(expr->left);
 616 }
 617 
 618 static void match_returns_user_rl(const char *fn, struct expression *expr, void *unused)
 619 {
 620         func_gets_user_data = true;
 621 }
 622 
 623 static int get_user_macro_rl(struct expression *expr, struct range_list **rl)
 624 {
 625         struct expression *parent;
 626         char *macro;
 627 
 628         if (!expr)
 629                 return 0;
 630 
 631         macro = get_macro_name(expr->pos);
 632         if (!macro)
 633                 return 0;
 634 
 635         /* handle ntohl(foo[i]) where "i" is trusted */
 636         parent = expr_get_parent_expr(expr);
 637         while (parent && parent->type != EXPR_BINOP)
 638                 parent = expr_get_parent_expr(parent);
 639         if (parent && parent->type == EXPR_BINOP) {
 640                 char *parent_macro = get_macro_name(parent->pos);
 641 
 642                 if (parent_macro && strcmp(macro, parent_macro) == 0)
 643                         return 0;
 644         }
 645 
 646         if (strcmp(macro, "ntohl") == 0) {
 647                 *rl = alloc_whole_rl(&uint_ctype);
 648                 return 1;
 649         }
 650         if (strcmp(macro, "ntohs") == 0) {
 651                 *rl = alloc_whole_rl(&ushort_ctype);
 652                 return 1;
 653         }
 654         return 0;
 655 }
 656 
 657 struct db_info {
 658         struct range_list *rl;
 659         struct expression *call;
 660 };
 661 static int returned_rl_callback(void *_info, int argc, char **argv, char **azColName)
 662 {
 663         struct db_info *db_info = _info;
 664         struct range_list *rl;
 665         char *return_ranges = argv[0];
 666         char *user_ranges = argv[1];
 667         struct expression *arg;
 668         int comparison;
 669 
 670         if (argc != 2)
 671                 return 0;
 672 
 673         call_results_to_rl(db_info->call, get_type(db_info->call), user_ranges, &rl);
 674         if (str_to_comparison_arg(return_ranges, db_info->call, &comparison, &arg) &&
 675             comparison == SPECIAL_EQUAL) {
 676                 struct range_list *orig_rl;
 677 
 678                 if (!get_user_rl(arg, &orig_rl))
 679                         return 0;
 680                 rl = rl_intersection(rl, orig_rl);
 681                 if (!rl)
 682                         return 0;
 683         }
 684         db_info->rl = rl_union(db_info->rl, rl);
 685 
 686         return 0;
 687 }
 688 
 689 static int has_user_data(struct symbol *sym)
 690 {
 691         struct sm_state *tmp;
 692 
 693         FOR_EACH_MY_SM(my_id, __get_cur_stree(), tmp) {
 694                 if (tmp->sym == sym)
 695                         return 1;
 696         } END_FOR_EACH_SM(tmp);
 697         return 0;
 698 }
 699 
 700 static int we_pass_user_data(struct expression *call)
 701 {
 702         struct expression *arg;
 703         struct symbol *sym;
 704 
 705         FOR_EACH_PTR(call->args, arg) {
 706                 sym = expr_to_sym(arg);
 707                 if (!sym)
 708                         continue;
 709                 if (has_user_data(sym))
 710                         return 1;
 711         } END_FOR_EACH_PTR(arg);
 712 
 713         return 0;
 714 }
 715 
 716 static int db_returned_user_rl(struct expression *call, struct range_list **rl)
 717 {
 718         struct db_info db_info = {};
 719 
 720         /* for function pointers assume everything is used */
 721         if (call->fn->type != EXPR_SYMBOL)
 722                 return 0;
 723         if (is_fake_call(call))
 724                 return 0;
 725 
 726         db_info.call = call;
 727         run_sql(&returned_rl_callback, &db_info,
 728                 "select return, value from return_states where %s and type = %d and parameter = -1 and key = '$';",
 729                 get_static_filter(call->fn->symbol), USER_DATA3_SET);
 730         if (db_info.rl) {
 731                 func_gets_user_data = true;
 732                 *rl = db_info.rl;
 733                 return 1;
 734         }
 735 
 736         run_sql(&returned_rl_callback, &db_info,
 737                 "select return, value from return_states where %s and type = %d and parameter = -1 and key = '$';",
 738                 get_static_filter(call->fn->symbol), USER_DATA3);
 739         if (db_info.rl) {
 740                 if (!we_pass_user_data(call))
 741                         return 0;
 742                 *rl = db_info.rl;
 743                 return 1;
 744         }
 745 
 746         return 0;
 747 }
 748 
 749 struct stree *get_user_stree(void)
 750 {
 751         return get_all_states_stree(my_id);
 752 }
 753 
 754 static int user_data_flag;
 755 static int no_user_data_flag;
 756 static struct range_list *var_user_rl(struct expression *expr)
 757 {
 758         struct smatch_state *state;
 759         struct range_list *rl;
 760         struct range_list *absolute_rl;
 761 
 762         if (expr->type == EXPR_BINOP && expr->op == '%') {
 763                 struct range_list *left, *right;
 764 
 765                 if (!get_user_rl(expr->right, &right))
 766                         return NULL;
 767                 get_absolute_rl(expr->left, &left);
 768                 rl = rl_binop(left, '%', right);
 769                 goto found;
 770         }
 771 
 772         if (!option_spammy && expr->type == EXPR_BINOP && expr->op == '/') {
 773                 struct range_list *left = NULL;
 774                 struct range_list *right = NULL;
 775                 struct range_list *abs_right;
 776 
 777                 /*
 778                  * The specific bug I'm dealing with is:
 779                  *
 780                  * foo = capped_user / unknown;
 781                  *
 782                  * Instead of just saying foo is now entirely user_rl we should
 783                  * probably say instead that it is not at all user data.
 784                  *
 785                  */
 786 
 787                 get_user_rl(expr->left, &left);
 788                 get_user_rl(expr->right, &right);
 789                 get_absolute_rl(expr->right, &abs_right);
 790 
 791                 if (left && !right) {
 792                         rl = rl_binop(left, '/', abs_right);
 793                         if (sval_cmp(rl_max(left), rl_max(rl)) < 0)
 794                                 no_user_data_flag = 1;
 795                 }
 796 
 797                 return NULL;
 798         }
 799 
 800         if (get_rl_from_function(expr, &rl))
 801                 goto found;
 802 
 803         if (get_user_macro_rl(expr, &rl))
 804                 goto found;
 805 
 806         if (comes_from_skb_data(expr)) {
 807                 rl = alloc_whole_rl(get_type(expr));
 808                 goto found;
 809         }
 810 
 811         state = get_state_expr(my_id, expr);
 812         if (state && estate_rl(state)) {
 813                 rl = estate_rl(state);
 814                 goto found;
 815         }
 816 
 817         if (expr->type == EXPR_CALL && db_returned_user_rl(expr, &rl))
 818                 goto found;
 819 
 820         if (is_array(expr)) {
 821                 struct expression *array = get_array_base(expr);
 822 
 823                 if (!get_state_expr(my_id, array)) {
 824                         no_user_data_flag = 1;
 825                         return NULL;
 826                 }
 827         }
 828 
 829         if (expr->type == EXPR_PREOP && expr->op == '*' &&
 830             is_user_rl(expr->unop)) {
 831                 rl = var_to_absolute_rl(expr);
 832                 goto found;
 833         }
 834 
 835         return NULL;
 836 found:
 837         user_data_flag = 1;
 838         absolute_rl = var_to_absolute_rl(expr);
 839         return clone_rl(rl_intersection(rl, absolute_rl));
 840 }
 841 
 842 int get_user_rl(struct expression *expr, struct range_list **rl)
 843 {
 844         user_data_flag = 0;
 845         no_user_data_flag = 0;
 846         custom_get_absolute_rl(expr, &var_user_rl, rl);
 847         if (!user_data_flag || no_user_data_flag)
 848                 *rl = NULL;
 849 
 850         return !!*rl;
 851 }
 852 
 853 int get_user_rl_spammy(struct expression *expr, struct range_list **rl)
 854 {
 855         int ret;
 856 
 857         option_spammy++;
 858         ret = get_user_rl(expr, rl);
 859         option_spammy--;
 860 
 861         return ret;
 862 }
 863 
 864 int is_user_rl(struct expression *expr)
 865 {
 866         struct range_list *tmp;
 867 
 868         return get_user_rl_spammy(expr, &tmp);
 869 }
 870 
 871 int get_user_rl_var_sym(const char *name, struct symbol *sym, struct range_list **rl)
 872 {
 873         struct smatch_state *state;
 874 
 875         state = get_state(my_id, name, sym);
 876         if (state && estate_rl(state)) {
 877                 *rl = estate_rl(state);
 878                 return 1;
 879         }
 880         return 0;
 881 }
 882 
 883 static void match_call_info(struct expression *expr)
 884 {
 885         struct range_list *rl;
 886         struct expression *arg;
 887         struct symbol *type;
 888         int i = 0;
 889 
 890         i = -1;
 891         FOR_EACH_PTR(expr->args, arg) {
 892                 i++;
 893                 type = get_arg_type(expr->fn, i);
 894 
 895                 if (!get_user_rl(arg, &rl))
 896                         continue;
 897 
 898                 rl = cast_rl(type, rl);
 899                 sql_insert_caller_info(expr, USER_DATA3, i, "$", show_rl(rl));
 900         } END_FOR_EACH_PTR(arg);
 901 }
 902 
 903 static int is_struct_ptr(struct symbol *sym)
 904 {
 905         struct symbol *type;
 906 
 907         if (!sym)
 908                 return 0;
 909         type = get_real_base_type(sym);
 910         if (!type || type->type != SYM_PTR)
 911                 return 0;
 912         type = get_real_base_type(type);
 913         if (!type || type->type != SYM_STRUCT)
 914                 return 0;
 915         return 1;
 916 }
 917 
 918 static void struct_member_callback(struct expression *call, int param, char *printed_name, struct sm_state *sm)
 919 {
 920         struct smatch_state *state;
 921         struct range_list *rl;
 922         struct symbol *type;
 923 
 924         /*
 925          * Smatch uses a hack where if we get an unsigned long we say it's
 926          * both user data and it points to user data.  But if we pass it to a
 927          * function which takes an int, then it's just user data.  There's not
 928          * enough bytes for it to be a pointer.
 929          *
 930          */
 931         type = get_arg_type(call->fn, param);
 932         if (type && type_bits(type) < type_bits(&ptr_ctype))
 933                 return;
 934 
 935         if (strcmp(sm->state->name, "") == 0)
 936                 return;
 937 
 938         if (strcmp(printed_name, "*$") == 0 &&
 939             is_struct_ptr(sm->sym))
 940                 return;
 941 
 942         state = get_state(SMATCH_EXTRA, sm->name, sm->sym);
 943         if (!state || !estate_rl(state))
 944                 rl = estate_rl(sm->state);
 945         else
 946                 rl = rl_intersection(estate_rl(sm->state), estate_rl(state));
 947 
 948         sql_insert_caller_info(call, USER_DATA3, param, printed_name, show_rl(rl));
 949 }
 950 
 951 static void set_param_user_data(const char *name, struct symbol *sym, char *key, char *value)
 952 {
 953         struct range_list *rl = NULL;
 954         struct smatch_state *state;
 955         struct symbol *type;
 956         char fullname[256];
 957 
 958         if (strcmp(key, "*$") == 0)
 959                 snprintf(fullname, sizeof(fullname), "*%s", name);
 960         else if (strncmp(key, "$", 1) == 0)
 961                 snprintf(fullname, 256, "%s%s", name, key + 1);
 962         else
 963                 return;
 964 
 965         type = get_member_type_from_key(symbol_expression(sym), key);
 966 
 967         /* if the caller passes a void pointer with user data */
 968         if (strcmp(key, "*$") == 0 && type && type != &void_ctype) {
 969                 struct expression *expr = symbol_expression(sym);
 970 
 971                 tag_as_user_data(expr);
 972                 set_points_to_user_data(expr);
 973                 return;
 974         }
 975         str_to_rl(type, value, &rl);
 976         state = alloc_estate_rl(rl);
 977         set_state(my_id, fullname, sym, state);
 978 }
 979 
 980 static void set_called(const char *name, struct symbol *sym, char *key, char *value)
 981 {
 982         set_state(my_call_id, "this_function", NULL, &called);
 983 }
 984 
 985 static void match_syscall_definition(struct symbol *sym)
 986 {
 987         struct symbol *arg;
 988         char *macro;
 989         char *name;
 990         int is_syscall = 0;
 991 
 992         macro = get_macro_name(sym->pos);
 993         if (macro &&
 994             (strncmp("SYSCALL_DEFINE", macro, strlen("SYSCALL_DEFINE")) == 0 ||
 995              strncmp("COMPAT_SYSCALL_DEFINE", macro, strlen("COMPAT_SYSCALL_DEFINE")) == 0))
 996                 is_syscall = 1;
 997 
 998         name = get_function();
 999         if (!option_no_db && get_state(my_call_id, "this_function", NULL) != &called) {
1000                 if (name && strncmp(name, "sys_", 4) == 0)
1001                         is_syscall = 1;
1002         }
1003 
1004         if (name && strncmp(name, "compat_sys_", 11) == 0)
1005                 is_syscall = 1;
1006 
1007         if (!is_syscall)
1008                 return;
1009 
1010         FOR_EACH_PTR(sym->ctype.base_type->arguments, arg) {
1011                 set_state(my_id, arg->ident->name, arg, alloc_estate_whole(get_real_base_type(arg)));
1012         } END_FOR_EACH_PTR(arg);
1013 }
1014 
1015 static void set_to_user_data(struct expression *expr, char *key, char *value)
1016 {
1017         char *name;
1018         struct symbol *sym;
1019         struct symbol *type;
1020         struct range_list *rl = NULL;
1021 
1022         type = get_member_type_from_key(expr, key);
1023         name = get_variable_from_key(expr, key, &sym);
1024         if (!name || !sym)
1025                 goto free;
1026 
1027         call_results_to_rl(expr, type, value, &rl);
1028 
1029         set_state(my_id, name, sym, alloc_estate_rl(rl));
1030 free:
1031         free_string(name);
1032 
1033 }
1034 
1035 static void returns_param_user_data(struct expression *expr, int param, char *key, char *value)
1036 {
1037         struct expression *arg;
1038         struct expression *call;
1039 
1040         call = expr;
1041         while (call->type == EXPR_ASSIGNMENT)
1042                 call = strip_expr(call->right);
1043         if (call->type != EXPR_CALL)
1044                 return;
1045 
1046         if (!we_pass_user_data(call))
1047                 return;
1048 
1049         if (param == -1) {
1050                 if (expr->type != EXPR_ASSIGNMENT)
1051                         return;
1052                 set_to_user_data(expr->left, key, value);
1053                 return;
1054         }
1055 
1056         arg = get_argument_from_call_expr(call->args, param);
1057         if (!arg)
1058                 return;
1059         set_to_user_data(arg, key, value);
1060 }
1061 
1062 static void returns_param_user_data_set(struct expression *expr, int param, char *key, char *value)
1063 {
1064         struct expression *arg;
1065 
1066         func_gets_user_data = true;
1067 
1068         if (param == -1) {
1069                 if (expr->type != EXPR_ASSIGNMENT)
1070                         return;
1071                 if (strcmp(key, "*$") == 0) {
1072                         set_points_to_user_data(expr->left);
1073                         tag_as_user_data(expr->left);
1074                 } else {
1075                         set_to_user_data(expr->left, key, value);
1076                 }
1077                 return;
1078         }
1079 
1080         while (expr->type == EXPR_ASSIGNMENT)
1081                 expr = strip_expr(expr->right);
1082         if (expr->type != EXPR_CALL)
1083                 return;
1084 
1085         arg = get_argument_from_call_expr(expr->args, param);
1086         if (!arg)
1087                 return;
1088         set_to_user_data(arg, key, value);
1089 }
1090 
1091 static int has_empty_state(struct sm_state *sm)
1092 {
1093         struct sm_state *tmp;
1094 
1095         FOR_EACH_PTR(sm->possible, tmp) {
1096                 if (!estate_rl(tmp->state))
1097                         return 1;
1098         } END_FOR_EACH_PTR(tmp);
1099 
1100         return 0;
1101 }
1102 
1103 static void param_set_to_user_data(int return_id, char *return_ranges, struct expression *expr)
1104 {
1105         struct sm_state *sm;
1106         struct smatch_state *start_state;
1107         struct range_list *rl;
1108         int param;
1109         char *return_str;
1110         const char *param_name;
1111         struct symbol *ret_sym;
1112         bool return_found = false;
1113 
1114         expr = strip_expr(expr);
1115         return_str = expr_to_str(expr);
1116         ret_sym = expr_to_sym(expr);
1117 
1118         FOR_EACH_MY_SM(my_id, __get_cur_stree(), sm) {
1119                 if (has_empty_state(sm))
1120                         continue;
1121 
1122                 param = get_param_num_from_sym(sm->sym);
1123                 if (param < 0)
1124                         continue;
1125 
1126                 /* The logic here was that if we were passed in a user data then
1127                  * we don't record that.  It's like the difference between
1128                  * param_filter and param_set.  When I think about it, I'm not
1129                  * sure it actually works.  It's probably harmless because we
1130                  * checked earlier that we're not returning a parameter...
1131                  * Let's mark this as a TODO.
1132                  */
1133                 start_state = get_state_stree(start_states, my_id, sm->name, sm->sym);
1134                 if (start_state && rl_equiv(estate_rl(sm->state), estate_rl(start_state)))
1135                         continue;
1136 
1137                 param_name = get_param_name(sm);
1138                 if (!param_name)
1139                         continue;
1140                 if (strcmp(param_name, "$") == 0)  /* The -1 param is handled after the loop */
1141                         continue;
1142 
1143                 sql_insert_return_states(return_id, return_ranges,
1144                                          func_gets_user_data ? USER_DATA3_SET : USER_DATA3,
1145                                          param, param_name, show_rl(estate_rl(sm->state)));
1146         } END_FOR_EACH_SM(sm);
1147 
1148         if (points_to_user_data(expr)) {
1149                 sql_insert_return_states(return_id, return_ranges,
1150                                          (is_skb_data(expr) || !func_gets_user_data) ?
1151                                          USER_DATA3_SET : USER_DATA3,
1152                                          -1, "*$", "");
1153                 goto free_string;
1154         }
1155 
1156 
1157         FOR_EACH_MY_SM(my_id, __get_cur_stree(), sm) {
1158                 if (!ret_sym)
1159                         break;
1160                 if (ret_sym != sm->sym)
1161                         continue;
1162 
1163                 param_name = state_name_to_param_name(sm->name, return_str);
1164                 if (!param_name)
1165                         continue;
1166                 if (strcmp(param_name, "$") == 0)
1167                         return_found = true;
1168                 sql_insert_return_states(return_id, return_ranges,
1169                                          func_gets_user_data ? USER_DATA3_SET : USER_DATA3,
1170                                          -1, param_name, show_rl(estate_rl(sm->state)));
1171         } END_FOR_EACH_SM(sm);
1172 
1173 
1174         if (!return_found && get_user_rl(expr, &rl)) {
1175                 sql_insert_return_states(return_id, return_ranges,
1176                                          func_gets_user_data ? USER_DATA3_SET : USER_DATA3,
1177                                          -1, "$", show_rl(rl));
1178                 goto free_string;
1179         }
1180 
1181 free_string:
1182         free_string(return_str);
1183 }
1184 
1185 static struct int_stack *gets_data_stack;
1186 static void match_function_def(struct symbol *sym)
1187 {
1188         func_gets_user_data = false;
1189 }
1190 
1191 static void match_inline_start(struct expression *expr)
1192 {
1193         push_int(&gets_data_stack, func_gets_user_data);
1194 }
1195 
1196 static void match_inline_end(struct expression *expr)
1197 {
1198         func_gets_user_data = pop_int(&gets_data_stack);
1199 }
1200 
1201 void register_kernel_user_data2(int id)
1202 {
1203         int i;
1204 
1205         my_id = id;
1206 
1207         if (option_project != PROJ_KERNEL)
1208                 return;
1209 
1210         add_hook(&match_function_def, FUNC_DEF_HOOK);
1211         add_hook(&match_inline_start, INLINE_FN_START);
1212         add_hook(&match_inline_end, INLINE_FN_END);
1213 
1214         add_hook(&save_start_states, AFTER_DEF_HOOK);
1215         add_hook(&free_start_states, AFTER_FUNC_HOOK);
1216         add_hook(&match_save_states, INLINE_FN_START);
1217         add_hook(&match_restore_states, INLINE_FN_END);
1218 
1219         add_unmatched_state_hook(my_id, &empty_state);
1220         add_extra_nomod_hook(&extra_nomod_hook);
1221         add_pre_merge_hook(my_id, &pre_merge_hook);
1222         add_merge_hook(my_id, &merge_estates);
1223 
1224         add_function_hook("copy_from_user", &match_user_copy, INT_PTR(0));
1225         add_function_hook("__copy_from_user", &match_user_copy, INT_PTR(0));
1226         add_function_hook("memcpy_fromiovec", &match_user_copy, INT_PTR(0));
1227         for (i = 0; i < ARRAY_SIZE(kstr_funcs); i++)
1228                 add_function_hook(kstr_funcs[i], &match_user_copy, INT_PTR(2));
1229         add_function_hook("usb_control_msg", &match_user_copy, INT_PTR(6));
1230 
1231         for (i = 0; i < ARRAY_SIZE(returns_user_data); i++) {
1232                 add_function_assign_hook(returns_user_data[i], &match_user_assign_function, NULL);
1233                 add_function_hook(returns_user_data[i], &match_returns_user_rl, NULL);
1234         }
1235 
1236         add_function_hook("sscanf", &match_sscanf, NULL);
1237 
1238         add_hook(&match_syscall_definition, AFTER_DEF_HOOK);
1239 
1240         add_hook(&match_assign, ASSIGNMENT_HOOK);
1241         add_hook(&match_condition, CONDITION_HOOK);
1242 
1243         add_hook(&match_call_info, FUNCTION_CALL_HOOK);
1244         add_member_info_callback(my_id, struct_member_callback);
1245         select_caller_info_hook(set_param_user_data, USER_DATA3);
1246         select_return_states_hook(USER_DATA3, &returns_param_user_data);
1247         select_return_states_hook(USER_DATA3_SET, &returns_param_user_data_set);
1248         add_split_return_callback(&param_set_to_user_data);
1249 }
1250 
1251 void register_kernel_user_data3(int id)
1252 {
1253         my_call_id = id;
1254 
1255         if (option_project != PROJ_KERNEL)
1256                 return;
1257         select_caller_info_hook(set_called, INTERNAL);
1258 }
1259