1 /*
2 * Copyright (C) 2019 ARM.
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 #include "smatch.h"
19 #include "smatch_extra.h"
20 #include "smatch_function_hashtable.h"
21
22 static bool expr_has_memory_addr(struct expression *expr);
23
24 static DEFINE_HASHTABLE_SEARCH(search_symbol, char, char);
25 static DEFINE_HASHTABLE_INSERT(insert_symbol, char, char);
26 static struct hashtable *symbols;
27
match_assign(struct expression * expr)28 static void match_assign(struct expression *expr)
29 {
30 char *left_name;
31 struct symbol *left_sym;
32
33 left_name = expr_to_var_sym(expr->left, &left_sym);
34 if (!left_name || !left_sym)
35 return;
36
37 /*
38 * Once we have spotted a symbol of interest (one that may hold
39 * an untagged memory address), we keep track of any assignments
40 * made, such that we can also treat the assigned symbol as something
41 * of interest. This tracking is limited in scope to the function.
42 */
43 if (expr_has_memory_addr(expr->right))
44 insert_symbol(symbols, left_name, left_name);
45 }
46
match_endfunc(struct symbol * sym)47 static void match_endfunc(struct symbol *sym)
48 {
49 destroy_function_hashtable(symbols);
50 symbols = create_function_hashtable(4000);
51 }
52
expr_has_untagged_symbol(struct expression * expr)53 static bool expr_has_untagged_symbol(struct expression *expr)
54 {
55 char *name;
56 struct symbol *sym;
57
58 if (expr->type != EXPR_SYMBOL)
59 return false;
60
61 name = expr_to_var_sym(expr, &sym);
62 if (!name || !sym)
63 return false;
64
65 /* See if this is something we already know is of interest */
66 if (search_symbol(symbols, name))
67 return true;
68
69 return false;
70 }
71
expr_has_untagged_member(struct expression * expr)72 static bool expr_has_untagged_member(struct expression *expr)
73 {
74 if (expr->type != EXPR_DEREF)
75 return false;
76
77 if (!strcmp(expr->member->name, "vm_start") ||
78 !strcmp(expr->member->name, "vm_end") ||
79 !strcmp(expr->member->name, "addr_limit"))
80 return true;
81
82 return false;
83 }
84
expr_has_macro_with_name(struct expression * expr,const char * macro_name)85 static bool expr_has_macro_with_name(struct expression *expr, const char *macro_name)
86 {
87 char *name;
88
89 name = get_macro_name(expr->pos);
90 return (name && !strcmp(name, macro_name));
91 }
92
expr_has_untagged_macro(struct expression * expr)93 static bool expr_has_untagged_macro(struct expression *expr)
94 {
95 if (expr_has_macro_with_name(expr, "PAGE_SIZE") ||
96 expr_has_macro_with_name(expr, "PAGE_MASK") ||
97 expr_has_macro_with_name(expr, "TASK_SIZE"))
98 return true;
99
100 /**
101 * We can't detect a marco (such as PAGE_MASK) inside another macro
102 * such as offset_in_page, therefore we have to detect the outer macro
103 * instead.
104 */
105 if (expr_has_macro_with_name(expr, "offset_in_page"))
106 return true;
107
108 return false;
109 }
110
111 /*
112 * Identify expressions that contain memory addresses, in the future
113 * we may use annotations on symbols or function parameters.
114 */
expr_has_memory_addr(struct expression * expr)115 static bool expr_has_memory_addr(struct expression *expr)
116 {
117 if (expr->type == EXPR_PREOP || expr->type == EXPR_POSTOP)
118 expr = strip_expr(expr->unop);
119
120 if (expr_has_untagged_member(expr))
121 return true;
122
123 if (expr_has_untagged_macro(expr))
124 return true;
125
126 if (expr_has_untagged_symbol(expr))
127 return true;
128
129 return false;
130 }
131
rl_is_larger_or_equal(struct range_list * rl,sval_t sval)132 int rl_is_larger_or_equal(struct range_list *rl, sval_t sval)
133 {
134 struct data_range *tmp;
135
136 FOR_EACH_PTR(rl, tmp) {
137 if (sval_cmp(tmp->max, sval) >= 0)
138 return 1;
139 } END_FOR_EACH_PTR(tmp);
140 return 0;
141 }
142
rl_range_has_min_value(struct range_list * rl,sval_t sval)143 int rl_range_has_min_value(struct range_list *rl, sval_t sval)
144 {
145 struct data_range *tmp;
146
147 FOR_EACH_PTR(rl, tmp) {
148 if (!sval_cmp(tmp->min, sval)) {
149 return 1;
150 }
151 } END_FOR_EACH_PTR(tmp);
152 return 0;
153 }
154
rl_is_tagged(struct range_list * rl)155 static bool rl_is_tagged(struct range_list *rl)
156 {
157 sval_t invalid;
158 sval_t invalid_kernel;
159
160 invalid.type = &ullong_ctype;
161 invalid.value = 1ULL << 56;
162 invalid_kernel.type = &ullong_ctype;
163 invalid_kernel.value = 0xff8ULL << 52;
164
165 /*
166 * We only care for tagged addresses, thus ignore anything where the
167 * ranges of potential values cannot possibly have any of the top byte
168 * bits set.
169 */
170 if (!rl_is_larger_or_equal(rl, invalid))
171 return false;
172
173 /*
174 * Tagged addresses are untagged in the kernel by using sign_extend64 in
175 * the untagged_addr macro. For userspace addresses bit 55 will always
176 * be 0 and thus this has the effect of clearing the top byte. However
177 * for kernel addresses this is not true and the top bits end up set to
178 * all 1s. The untagged_addr macro results in leaving a gap in the range
179 * of possible values which can exist, thus let's look for a tell-tale
180 * range which starts from (0xff8ULL << 52).
181 */
182 if (rl_range_has_min_value(rl, invalid_kernel))
183 return false;
184
185 return true;
186 }
187
match_condition(struct expression * expr)188 static void match_condition(struct expression *expr)
189 {
190 struct range_list *rl = NULL;
191 struct expression *val = NULL;
192 struct symbol *type;
193 char *var_name;
194
195 /*
196 * Match instances where something is compared against something
197 * else - we include binary operators as these are commonly used
198 * to make a comparison, e.g. if (start & ~PAGE_MASK).
199 */
200 if (expr->type != EXPR_COMPARE &&
201 expr->type != EXPR_BINOP)
202 return;
203
204 /*
205 * Look on both sides of the comparison for something that shouldn't
206 * be compared with a tagged address, e.g. macros such as PAGE_MASK
207 * or struct members named .vm_start.
208 */
209 if (expr_has_memory_addr(expr->left))
210 val = expr->right;
211
212 /*
213 * The macro 'offset_in_page' has the PAGE_MASK macro inside it, this
214 * results in 'expr_has_memory_addr' returning true for both sides. To
215 * work around this we assume PAGE_MASK (or similar) is on the right
216 * side, thus we do the following test last.
217 */
218 if (expr_has_memory_addr(expr->right))
219 val = expr->left;
220
221 if (!val)
222 return;
223
224 /* We only care about memory addresses which are 64 bits */
225 type = get_type(val);
226 if (!type || type_bits(type) != 64)
227 return;
228
229 /* We only care for comparison against user originated data */
230 if (!get_user_rl(val, &rl))
231 return;
232
233 /* We only care for tagged addresses */
234 if (!rl_is_tagged(rl))
235 return;
236
237 /* Finally, we believe we may have spotted a risky comparison */
238 var_name = expr_to_var(val);
239 if (var_name)
240 sm_warning("comparison of a potentially tagged address (%s, %d, %s)", get_function(), get_param_num(val), var_name);
241 }
242
check_arm64_tagged(int id)243 void check_arm64_tagged(int id)
244 {
245 char *arch;
246
247 if (option_project != PROJ_KERNEL)
248 return;
249
250 /* Limit to aarch64 */
251 arch = getenv("ARCH");
252 if (!arch || strcmp(arch, "arm64"))
253 return;
254
255 symbols = create_function_hashtable(4000);
256
257 add_hook(&match_assign, ASSIGNMENT_HOOK);
258 add_hook(&match_condition, CONDITION_HOOK);
259 add_hook(&match_endfunc, END_FUNC_HOOK);
260 }
261