1 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
2 *
3 * This program is free software; you can redistribute it and/or
4 * modify it under the terms of version 2 of the GNU General Public
5 * License as published by the Free Software Foundation.
6 *
7 * This program is distributed in the hope that it will be useful, but
8 * WITHOUT ANY WARRANTY; without even the implied warranty of
9 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
10 * General Public License for more details.
11 */
12 #include <linux/kernel.h>
13 #include <linux/types.h>
14 #include <linux/slab.h>
15 #include <linux/bpf.h>
16 #include <linux/filter.h>
17 #include <net/netlink.h>
18 #include <linux/file.h>
19 #include <linux/vmalloc.h>
20
21 /* bpf_check() is a static code analyzer that walks eBPF program
22 * instruction by instruction and updates register/stack state.
23 * All paths of conditional branches are analyzed until 'bpf_exit' insn.
24 *
25 * The first pass is depth-first-search to check that the program is a DAG.
26 * It rejects the following programs:
27 * - larger than BPF_MAXINSNS insns
28 * - if loop is present (detected via back-edge)
29 * - unreachable insns exist (shouldn't be a forest. program = one function)
30 * - out of bounds or malformed jumps
31 * The second pass is all possible path descent from the 1st insn.
32 * Since it's analyzing all pathes through the program, the length of the
33 * analysis is limited to 32k insn, which may be hit even if total number of
34 * insn is less then 4K, but there are too many branches that change stack/regs.
35 * Number of 'branches to be analyzed' is limited to 1k
36 *
37 * On entry to each instruction, each register has a type, and the instruction
38 * changes the types of the registers depending on instruction semantics.
39 * If instruction is BPF_MOV64_REG(BPF_REG_1, BPF_REG_5), then type of R5 is
40 * copied to R1.
41 *
42 * All registers are 64-bit.
43 * R0 - return register
44 * R1-R5 argument passing registers
45 * R6-R9 callee saved registers
46 * R10 - frame pointer read-only
47 *
48 * At the start of BPF program the register R1 contains a pointer to bpf_context
49 * and has type PTR_TO_CTX.
50 *
51 * Verifier tracks arithmetic operations on pointers in case:
52 * BPF_MOV64_REG(BPF_REG_1, BPF_REG_10),
53 * BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -20),
54 * 1st insn copies R10 (which has FRAME_PTR) type into R1
55 * and 2nd arithmetic instruction is pattern matched to recognize
56 * that it wants to construct a pointer to some element within stack.
57 * So after 2nd insn, the register R1 has type PTR_TO_STACK
58 * (and -20 constant is saved for further stack bounds checking).
59 * Meaning that this reg is a pointer to stack plus known immediate constant.
60 *
61 * Most of the time the registers have UNKNOWN_VALUE type, which
62 * means the register has some value, but it's not a valid pointer.
63 * (like pointer plus pointer becomes UNKNOWN_VALUE type)
64 *
65 * When verifier sees load or store instructions the type of base register
66 * can be: PTR_TO_MAP_VALUE, PTR_TO_CTX, FRAME_PTR. These are three pointer
67 * types recognized by check_mem_access() function.
68 *
69 * PTR_TO_MAP_VALUE means that this register is pointing to 'map element value'
70 * and the range of [ptr, ptr + map's value_size) is accessible.
71 *
72 * registers used to pass values to function calls are checked against
73 * function argument constraints.
74 *
75 * ARG_PTR_TO_MAP_KEY is one of such argument constraints.
76 * It means that the register type passed to this function must be
77 * PTR_TO_STACK and it will be used inside the function as
78 * 'pointer to map element key'
79 *
80 * For example the argument constraints for bpf_map_lookup_elem():
81 * .ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL,
82 * .arg1_type = ARG_CONST_MAP_PTR,
83 * .arg2_type = ARG_PTR_TO_MAP_KEY,
84 *
85 * ret_type says that this function returns 'pointer to map elem value or null'
86 * function expects 1st argument to be a const pointer to 'struct bpf_map' and
87 * 2nd argument should be a pointer to stack, which will be used inside
88 * the helper function as a pointer to map element key.
89 *
90 * On the kernel side the helper function looks like:
91 * u64 bpf_map_lookup_elem(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
92 * {
93 * struct bpf_map *map = (struct bpf_map *) (unsigned long) r1;
94 * void *key = (void *) (unsigned long) r2;
95 * void *value;
96 *
97 * here kernel can access 'key' and 'map' pointers safely, knowing that
98 * [key, key + map->key_size) bytes are valid and were initialized on
99 * the stack of eBPF program.
100 * }
101 *
102 * Corresponding eBPF program may look like:
103 * BPF_MOV64_REG(BPF_REG_2, BPF_REG_10), // after this insn R2 type is FRAME_PTR
104 * BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -4), // after this insn R2 type is PTR_TO_STACK
105 * BPF_LD_MAP_FD(BPF_REG_1, map_fd), // after this insn R1 type is CONST_PTR_TO_MAP
106 * BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem),
107 * here verifier looks at prototype of map_lookup_elem() and sees:
108 * .arg1_type == ARG_CONST_MAP_PTR and R1->type == CONST_PTR_TO_MAP, which is ok,
109 * Now verifier knows that this map has key of R1->map_ptr->key_size bytes
110 *
111 * Then .arg2_type == ARG_PTR_TO_MAP_KEY and R2->type == PTR_TO_STACK, ok so far,
112 * Now verifier checks that [R2, R2 + map's key_size) are within stack limits
113 * and were initialized prior to this call.
114 * If it's ok, then verifier allows this BPF_CALL insn and looks at
115 * .ret_type which is RET_PTR_TO_MAP_VALUE_OR_NULL, so it sets
116 * R0->type = PTR_TO_MAP_VALUE_OR_NULL which means bpf_map_lookup_elem() function
117 * returns ether pointer to map value or NULL.
118 *
119 * When type PTR_TO_MAP_VALUE_OR_NULL passes through 'if (reg != 0) goto +off'
120 * insn, the register holding that pointer in the true branch changes state to
121 * PTR_TO_MAP_VALUE and the same register changes state to CONST_IMM in the false
122 * branch. See check_cond_jmp_op().
123 *
124 * After the call R0 is set to return type of the function and registers R1-R5
125 * are set to NOT_INIT to indicate that they are no longer readable.
126 */
127
128 /* types of values stored in eBPF registers */
129 enum bpf_reg_type {
130 NOT_INIT = 0, /* nothing was written into register */
131 UNKNOWN_VALUE, /* reg doesn't contain a valid pointer */
132 PTR_TO_CTX, /* reg points to bpf_context */
133 CONST_PTR_TO_MAP, /* reg points to struct bpf_map */
134 PTR_TO_MAP_VALUE, /* reg points to map element value */
135 PTR_TO_MAP_VALUE_OR_NULL,/* points to map elem value or NULL */
136 FRAME_PTR, /* reg == frame_pointer */
137 PTR_TO_STACK, /* reg == frame_pointer + imm */
138 CONST_IMM, /* constant integer value */
139 };
140
141 struct reg_state {
142 enum bpf_reg_type type;
143 union {
144 /* valid when type == CONST_IMM | PTR_TO_STACK */
145 int imm;
146
147 /* valid when type == CONST_PTR_TO_MAP | PTR_TO_MAP_VALUE |
148 * PTR_TO_MAP_VALUE_OR_NULL
149 */
150 struct bpf_map *map_ptr;
151 };
152 };
153
154 enum bpf_stack_slot_type {
155 STACK_INVALID, /* nothing was stored in this stack slot */
156 STACK_SPILL, /* register spilled into stack */
157 STACK_MISC /* BPF program wrote some data into this slot */
158 };
159
160 #define BPF_REG_SIZE 8 /* size of eBPF register in bytes */
161
162 /* state of the program:
163 * type of all registers and stack info
164 */
165 struct verifier_state {
166 struct reg_state regs[MAX_BPF_REG];
167 u8 stack_slot_type[MAX_BPF_STACK];
168 struct reg_state spilled_regs[MAX_BPF_STACK / BPF_REG_SIZE];
169 };
170
171 /* linked list of verifier states used to prune search */
172 struct verifier_state_list {
173 struct verifier_state state;
174 struct verifier_state_list *next;
175 };
176
177 /* verifier_state + insn_idx are pushed to stack when branch is encountered */
178 struct verifier_stack_elem {
179 /* verifer state is 'st'
180 * before processing instruction 'insn_idx'
181 * and after processing instruction 'prev_insn_idx'
182 */
183 struct verifier_state st;
184 int insn_idx;
185 int prev_insn_idx;
186 struct verifier_stack_elem *next;
187 };
188
189 #define MAX_USED_MAPS 64 /* max number of maps accessed by one eBPF program */
190
191 /* single container for all structs
192 * one verifier_env per bpf_check() call
193 */
194 struct verifier_env {
195 struct bpf_prog *prog; /* eBPF program being verified */
196 struct verifier_stack_elem *head; /* stack of verifier states to be processed */
197 int stack_size; /* number of states to be processed */
198 struct verifier_state cur_state; /* current verifier state */
199 struct verifier_state_list **explored_states; /* search pruning optimization */
200 struct bpf_map *used_maps[MAX_USED_MAPS]; /* array of map's used by eBPF program */
201 u32 used_map_cnt; /* number of used maps */
202 bool allow_ptr_leaks;
203 };
204
205 /* verbose verifier prints what it's seeing
206 * bpf_check() is called under lock, so no race to access these global vars
207 */
208 static u32 log_level, log_size, log_len;
209 static char *log_buf;
210
211 static DEFINE_MUTEX(bpf_verifier_lock);
212
213 /* log_level controls verbosity level of eBPF verifier.
214 * verbose() is used to dump the verification trace to the log, so the user
215 * can figure out what's wrong with the program
216 */
verbose(const char * fmt,...)217 static __printf(1, 2) void verbose(const char *fmt, ...)
218 {
219 va_list args;
220
221 if (log_level == 0 || log_len >= log_size - 1)
222 return;
223
224 va_start(args, fmt);
225 log_len += vscnprintf(log_buf + log_len, log_size - log_len, fmt, args);
226 va_end(args);
227 }
228
229 /* string representation of 'enum bpf_reg_type' */
230 static const char * const reg_type_str[] = {
231 [NOT_INIT] = "?",
232 [UNKNOWN_VALUE] = "inv",
233 [PTR_TO_CTX] = "ctx",
234 [CONST_PTR_TO_MAP] = "map_ptr",
235 [PTR_TO_MAP_VALUE] = "map_value",
236 [PTR_TO_MAP_VALUE_OR_NULL] = "map_value_or_null",
237 [FRAME_PTR] = "fp",
238 [PTR_TO_STACK] = "fp",
239 [CONST_IMM] = "imm",
240 };
241
print_verifier_state(struct verifier_env * env)242 static void print_verifier_state(struct verifier_env *env)
243 {
244 enum bpf_reg_type t;
245 int i;
246
247 for (i = 0; i < MAX_BPF_REG; i++) {
248 t = env->cur_state.regs[i].type;
249 if (t == NOT_INIT)
250 continue;
251 verbose(" R%d=%s", i, reg_type_str[t]);
252 if (t == CONST_IMM || t == PTR_TO_STACK)
253 verbose("%d", env->cur_state.regs[i].imm);
254 else if (t == CONST_PTR_TO_MAP || t == PTR_TO_MAP_VALUE ||
255 t == PTR_TO_MAP_VALUE_OR_NULL)
256 verbose("(ks=%d,vs=%d)",
257 env->cur_state.regs[i].map_ptr->key_size,
258 env->cur_state.regs[i].map_ptr->value_size);
259 }
260 for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
261 if (env->cur_state.stack_slot_type[i] == STACK_SPILL)
262 verbose(" fp%d=%s", -MAX_BPF_STACK + i,
263 reg_type_str[env->cur_state.spilled_regs[i / BPF_REG_SIZE].type]);
264 }
265 verbose("\n");
266 }
267
268 static const char *const bpf_class_string[] = {
269 [BPF_LD] = "ld",
270 [BPF_LDX] = "ldx",
271 [BPF_ST] = "st",
272 [BPF_STX] = "stx",
273 [BPF_ALU] = "alu",
274 [BPF_JMP] = "jmp",
275 [BPF_RET] = "BUG",
276 [BPF_ALU64] = "alu64",
277 };
278
279 static const char *const bpf_alu_string[16] = {
280 [BPF_ADD >> 4] = "+=",
281 [BPF_SUB >> 4] = "-=",
282 [BPF_MUL >> 4] = "*=",
283 [BPF_DIV >> 4] = "/=",
284 [BPF_OR >> 4] = "|=",
285 [BPF_AND >> 4] = "&=",
286 [BPF_LSH >> 4] = "<<=",
287 [BPF_RSH >> 4] = ">>=",
288 [BPF_NEG >> 4] = "neg",
289 [BPF_MOD >> 4] = "%=",
290 [BPF_XOR >> 4] = "^=",
291 [BPF_MOV >> 4] = "=",
292 [BPF_ARSH >> 4] = "s>>=",
293 [BPF_END >> 4] = "endian",
294 };
295
296 static const char *const bpf_ldst_string[] = {
297 [BPF_W >> 3] = "u32",
298 [BPF_H >> 3] = "u16",
299 [BPF_B >> 3] = "u8",
300 [BPF_DW >> 3] = "u64",
301 };
302
303 static const char *const bpf_jmp_string[16] = {
304 [BPF_JA >> 4] = "jmp",
305 [BPF_JEQ >> 4] = "==",
306 [BPF_JGT >> 4] = ">",
307 [BPF_JGE >> 4] = ">=",
308 [BPF_JSET >> 4] = "&",
309 [BPF_JNE >> 4] = "!=",
310 [BPF_JSGT >> 4] = "s>",
311 [BPF_JSGE >> 4] = "s>=",
312 [BPF_CALL >> 4] = "call",
313 [BPF_EXIT >> 4] = "exit",
314 };
315
print_bpf_insn(struct bpf_insn * insn)316 static void print_bpf_insn(struct bpf_insn *insn)
317 {
318 u8 class = BPF_CLASS(insn->code);
319
320 if (class == BPF_ALU || class == BPF_ALU64) {
321 if (BPF_SRC(insn->code) == BPF_X)
322 verbose("(%02x) %sr%d %s %sr%d\n",
323 insn->code, class == BPF_ALU ? "(u32) " : "",
324 insn->dst_reg,
325 bpf_alu_string[BPF_OP(insn->code) >> 4],
326 class == BPF_ALU ? "(u32) " : "",
327 insn->src_reg);
328 else
329 verbose("(%02x) %sr%d %s %s%d\n",
330 insn->code, class == BPF_ALU ? "(u32) " : "",
331 insn->dst_reg,
332 bpf_alu_string[BPF_OP(insn->code) >> 4],
333 class == BPF_ALU ? "(u32) " : "",
334 insn->imm);
335 } else if (class == BPF_STX) {
336 if (BPF_MODE(insn->code) == BPF_MEM)
337 verbose("(%02x) *(%s *)(r%d %+d) = r%d\n",
338 insn->code,
339 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
340 insn->dst_reg,
341 insn->off, insn->src_reg);
342 else if (BPF_MODE(insn->code) == BPF_XADD)
343 verbose("(%02x) lock *(%s *)(r%d %+d) += r%d\n",
344 insn->code,
345 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
346 insn->dst_reg, insn->off,
347 insn->src_reg);
348 else
349 verbose("BUG_%02x\n", insn->code);
350 } else if (class == BPF_ST) {
351 if (BPF_MODE(insn->code) != BPF_MEM) {
352 verbose("BUG_st_%02x\n", insn->code);
353 return;
354 }
355 verbose("(%02x) *(%s *)(r%d %+d) = %d\n",
356 insn->code,
357 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
358 insn->dst_reg,
359 insn->off, insn->imm);
360 } else if (class == BPF_LDX) {
361 if (BPF_MODE(insn->code) != BPF_MEM) {
362 verbose("BUG_ldx_%02x\n", insn->code);
363 return;
364 }
365 verbose("(%02x) r%d = *(%s *)(r%d %+d)\n",
366 insn->code, insn->dst_reg,
367 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
368 insn->src_reg, insn->off);
369 } else if (class == BPF_LD) {
370 if (BPF_MODE(insn->code) == BPF_ABS) {
371 verbose("(%02x) r0 = *(%s *)skb[%d]\n",
372 insn->code,
373 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
374 insn->imm);
375 } else if (BPF_MODE(insn->code) == BPF_IND) {
376 verbose("(%02x) r0 = *(%s *)skb[r%d + %d]\n",
377 insn->code,
378 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
379 insn->src_reg, insn->imm);
380 } else if (BPF_MODE(insn->code) == BPF_IMM) {
381 verbose("(%02x) r%d = 0x%x\n",
382 insn->code, insn->dst_reg, insn->imm);
383 } else {
384 verbose("BUG_ld_%02x\n", insn->code);
385 return;
386 }
387 } else if (class == BPF_JMP) {
388 u8 opcode = BPF_OP(insn->code);
389
390 if (opcode == BPF_CALL) {
391 verbose("(%02x) call %d\n", insn->code, insn->imm);
392 } else if (insn->code == (BPF_JMP | BPF_JA)) {
393 verbose("(%02x) goto pc%+d\n",
394 insn->code, insn->off);
395 } else if (insn->code == (BPF_JMP | BPF_EXIT)) {
396 verbose("(%02x) exit\n", insn->code);
397 } else if (BPF_SRC(insn->code) == BPF_X) {
398 verbose("(%02x) if r%d %s r%d goto pc%+d\n",
399 insn->code, insn->dst_reg,
400 bpf_jmp_string[BPF_OP(insn->code) >> 4],
401 insn->src_reg, insn->off);
402 } else {
403 verbose("(%02x) if r%d %s 0x%x goto pc%+d\n",
404 insn->code, insn->dst_reg,
405 bpf_jmp_string[BPF_OP(insn->code) >> 4],
406 insn->imm, insn->off);
407 }
408 } else {
409 verbose("(%02x) %s\n", insn->code, bpf_class_string[class]);
410 }
411 }
412
pop_stack(struct verifier_env * env,int * prev_insn_idx)413 static int pop_stack(struct verifier_env *env, int *prev_insn_idx)
414 {
415 struct verifier_stack_elem *elem;
416 int insn_idx;
417
418 if (env->head == NULL)
419 return -1;
420
421 memcpy(&env->cur_state, &env->head->st, sizeof(env->cur_state));
422 insn_idx = env->head->insn_idx;
423 if (prev_insn_idx)
424 *prev_insn_idx = env->head->prev_insn_idx;
425 elem = env->head->next;
426 kfree(env->head);
427 env->head = elem;
428 env->stack_size--;
429 return insn_idx;
430 }
431
push_stack(struct verifier_env * env,int insn_idx,int prev_insn_idx)432 static struct verifier_state *push_stack(struct verifier_env *env, int insn_idx,
433 int prev_insn_idx)
434 {
435 struct verifier_stack_elem *elem;
436
437 elem = kmalloc(sizeof(struct verifier_stack_elem), GFP_KERNEL);
438 if (!elem)
439 goto err;
440
441 memcpy(&elem->st, &env->cur_state, sizeof(env->cur_state));
442 elem->insn_idx = insn_idx;
443 elem->prev_insn_idx = prev_insn_idx;
444 elem->next = env->head;
445 env->head = elem;
446 env->stack_size++;
447 if (env->stack_size > 1024) {
448 verbose("BPF program is too complex\n");
449 goto err;
450 }
451 return &elem->st;
452 err:
453 /* pop all elements and return */
454 while (pop_stack(env, NULL) >= 0);
455 return NULL;
456 }
457
458 #define CALLER_SAVED_REGS 6
459 static const int caller_saved[CALLER_SAVED_REGS] = {
460 BPF_REG_0, BPF_REG_1, BPF_REG_2, BPF_REG_3, BPF_REG_4, BPF_REG_5
461 };
462
init_reg_state(struct reg_state * regs)463 static void init_reg_state(struct reg_state *regs)
464 {
465 int i;
466
467 for (i = 0; i < MAX_BPF_REG; i++) {
468 regs[i].type = NOT_INIT;
469 regs[i].imm = 0;
470 regs[i].map_ptr = NULL;
471 }
472
473 /* frame pointer */
474 regs[BPF_REG_FP].type = FRAME_PTR;
475
476 /* 1st arg to a function */
477 regs[BPF_REG_1].type = PTR_TO_CTX;
478 }
479
mark_reg_unknown_value(struct reg_state * regs,u32 regno)480 static void mark_reg_unknown_value(struct reg_state *regs, u32 regno)
481 {
482 BUG_ON(regno >= MAX_BPF_REG);
483 regs[regno].type = UNKNOWN_VALUE;
484 regs[regno].imm = 0;
485 regs[regno].map_ptr = NULL;
486 }
487
488 enum reg_arg_type {
489 SRC_OP, /* register is used as source operand */
490 DST_OP, /* register is used as destination operand */
491 DST_OP_NO_MARK /* same as above, check only, don't mark */
492 };
493
check_reg_arg(struct reg_state * regs,u32 regno,enum reg_arg_type t)494 static int check_reg_arg(struct reg_state *regs, u32 regno,
495 enum reg_arg_type t)
496 {
497 if (regno >= MAX_BPF_REG) {
498 verbose("R%d is invalid\n", regno);
499 return -EINVAL;
500 }
501
502 if (t == SRC_OP) {
503 /* check whether register used as source operand can be read */
504 if (regs[regno].type == NOT_INIT) {
505 verbose("R%d !read_ok\n", regno);
506 return -EACCES;
507 }
508 } else {
509 /* check whether register used as dest operand can be written to */
510 if (regno == BPF_REG_FP) {
511 verbose("frame pointer is read only\n");
512 return -EACCES;
513 }
514 if (t == DST_OP)
515 mark_reg_unknown_value(regs, regno);
516 }
517 return 0;
518 }
519
bpf_size_to_bytes(int bpf_size)520 static int bpf_size_to_bytes(int bpf_size)
521 {
522 if (bpf_size == BPF_W)
523 return 4;
524 else if (bpf_size == BPF_H)
525 return 2;
526 else if (bpf_size == BPF_B)
527 return 1;
528 else if (bpf_size == BPF_DW)
529 return 8;
530 else
531 return -EINVAL;
532 }
533
is_spillable_regtype(enum bpf_reg_type type)534 static bool is_spillable_regtype(enum bpf_reg_type type)
535 {
536 switch (type) {
537 case PTR_TO_MAP_VALUE:
538 case PTR_TO_MAP_VALUE_OR_NULL:
539 case PTR_TO_STACK:
540 case PTR_TO_CTX:
541 case FRAME_PTR:
542 case CONST_PTR_TO_MAP:
543 return true;
544 default:
545 return false;
546 }
547 }
548
549 /* check_stack_read/write functions track spill/fill of registers,
550 * stack boundary and alignment are checked in check_mem_access()
551 */
check_stack_write(struct verifier_state * state,int off,int size,int value_regno)552 static int check_stack_write(struct verifier_state *state, int off, int size,
553 int value_regno)
554 {
555 int i;
556 /* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0,
557 * so it's aligned access and [off, off + size) are within stack limits
558 */
559
560 if (value_regno >= 0 &&
561 is_spillable_regtype(state->regs[value_regno].type)) {
562
563 /* register containing pointer is being spilled into stack */
564 if (size != BPF_REG_SIZE) {
565 verbose("invalid size of register spill\n");
566 return -EACCES;
567 }
568
569 /* save register state */
570 state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] =
571 state->regs[value_regno];
572
573 for (i = 0; i < BPF_REG_SIZE; i++)
574 state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_SPILL;
575 } else {
576 /* regular write of data into stack */
577 state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] =
578 (struct reg_state) {};
579
580 for (i = 0; i < size; i++)
581 state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_MISC;
582 }
583 return 0;
584 }
585
check_stack_read(struct verifier_state * state,int off,int size,int value_regno)586 static int check_stack_read(struct verifier_state *state, int off, int size,
587 int value_regno)
588 {
589 u8 *slot_type;
590 int i;
591
592 slot_type = &state->stack_slot_type[MAX_BPF_STACK + off];
593
594 if (slot_type[0] == STACK_SPILL) {
595 if (size != BPF_REG_SIZE) {
596 verbose("invalid size of register spill\n");
597 return -EACCES;
598 }
599 for (i = 1; i < BPF_REG_SIZE; i++) {
600 if (slot_type[i] != STACK_SPILL) {
601 verbose("corrupted spill memory\n");
602 return -EACCES;
603 }
604 }
605
606 if (value_regno >= 0)
607 /* restore register state from stack */
608 state->regs[value_regno] =
609 state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE];
610 return 0;
611 } else {
612 for (i = 0; i < size; i++) {
613 if (slot_type[i] != STACK_MISC) {
614 verbose("invalid read from stack off %d+%d size %d\n",
615 off, i, size);
616 return -EACCES;
617 }
618 }
619 if (value_regno >= 0)
620 /* have read misc data from the stack */
621 mark_reg_unknown_value(state->regs, value_regno);
622 return 0;
623 }
624 }
625
626 /* check read/write into map element returned by bpf_map_lookup_elem() */
check_map_access(struct verifier_env * env,u32 regno,int off,int size)627 static int check_map_access(struct verifier_env *env, u32 regno, int off,
628 int size)
629 {
630 struct bpf_map *map = env->cur_state.regs[regno].map_ptr;
631
632 if (off < 0 || off + size > map->value_size) {
633 verbose("invalid access to map value, value_size=%d off=%d size=%d\n",
634 map->value_size, off, size);
635 return -EACCES;
636 }
637 return 0;
638 }
639
640 /* check access to 'struct bpf_context' fields */
check_ctx_access(struct verifier_env * env,int off,int size,enum bpf_access_type t)641 static int check_ctx_access(struct verifier_env *env, int off, int size,
642 enum bpf_access_type t)
643 {
644 if (env->prog->aux->ops->is_valid_access &&
645 env->prog->aux->ops->is_valid_access(off, size, t))
646 return 0;
647
648 verbose("invalid bpf_context access off=%d size=%d\n", off, size);
649 return -EACCES;
650 }
651
is_pointer_value(struct verifier_env * env,int regno)652 static bool is_pointer_value(struct verifier_env *env, int regno)
653 {
654 if (env->allow_ptr_leaks)
655 return false;
656
657 switch (env->cur_state.regs[regno].type) {
658 case UNKNOWN_VALUE:
659 case CONST_IMM:
660 return false;
661 default:
662 return true;
663 }
664 }
665
666 /* check whether memory at (regno + off) is accessible for t = (read | write)
667 * if t==write, value_regno is a register which value is stored into memory
668 * if t==read, value_regno is a register which will receive the value from memory
669 * if t==write && value_regno==-1, some unknown value is stored into memory
670 * if t==read && value_regno==-1, don't care what we read from memory
671 */
check_mem_access(struct verifier_env * env,u32 regno,int off,int bpf_size,enum bpf_access_type t,int value_regno)672 static int check_mem_access(struct verifier_env *env, u32 regno, int off,
673 int bpf_size, enum bpf_access_type t,
674 int value_regno)
675 {
676 struct verifier_state *state = &env->cur_state;
677 int size, err = 0;
678
679 if (state->regs[regno].type == PTR_TO_STACK)
680 off += state->regs[regno].imm;
681
682 size = bpf_size_to_bytes(bpf_size);
683 if (size < 0)
684 return size;
685
686 if (off % size != 0) {
687 verbose("misaligned access off %d size %d\n", off, size);
688 return -EACCES;
689 }
690
691 if (state->regs[regno].type == PTR_TO_MAP_VALUE) {
692 if (t == BPF_WRITE && value_regno >= 0 &&
693 is_pointer_value(env, value_regno)) {
694 verbose("R%d leaks addr into map\n", value_regno);
695 return -EACCES;
696 }
697 err = check_map_access(env, regno, off, size);
698 if (!err && t == BPF_READ && value_regno >= 0)
699 mark_reg_unknown_value(state->regs, value_regno);
700
701 } else if (state->regs[regno].type == PTR_TO_CTX) {
702 if (t == BPF_WRITE && value_regno >= 0 &&
703 is_pointer_value(env, value_regno)) {
704 verbose("R%d leaks addr into ctx\n", value_regno);
705 return -EACCES;
706 }
707 err = check_ctx_access(env, off, size, t);
708 if (!err && t == BPF_READ && value_regno >= 0)
709 mark_reg_unknown_value(state->regs, value_regno);
710
711 } else if (state->regs[regno].type == FRAME_PTR ||
712 state->regs[regno].type == PTR_TO_STACK) {
713 if (off >= 0 || off < -MAX_BPF_STACK) {
714 verbose("invalid stack off=%d size=%d\n", off, size);
715 return -EACCES;
716 }
717 if (t == BPF_WRITE) {
718 if (!env->allow_ptr_leaks &&
719 state->stack_slot_type[MAX_BPF_STACK + off] == STACK_SPILL &&
720 size != BPF_REG_SIZE) {
721 verbose("attempt to corrupt spilled pointer on stack\n");
722 return -EACCES;
723 }
724 err = check_stack_write(state, off, size, value_regno);
725 } else {
726 err = check_stack_read(state, off, size, value_regno);
727 }
728 } else {
729 verbose("R%d invalid mem access '%s'\n",
730 regno, reg_type_str[state->regs[regno].type]);
731 return -EACCES;
732 }
733 return err;
734 }
735
check_xadd(struct verifier_env * env,struct bpf_insn * insn)736 static int check_xadd(struct verifier_env *env, struct bpf_insn *insn)
737 {
738 struct reg_state *regs = env->cur_state.regs;
739 int err;
740
741 if ((BPF_SIZE(insn->code) != BPF_W && BPF_SIZE(insn->code) != BPF_DW) ||
742 insn->imm != 0) {
743 verbose("BPF_XADD uses reserved fields\n");
744 return -EINVAL;
745 }
746
747 /* check src1 operand */
748 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
749 if (err)
750 return err;
751
752 /* check src2 operand */
753 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
754 if (err)
755 return err;
756
757 /* check whether atomic_add can read the memory */
758 err = check_mem_access(env, insn->dst_reg, insn->off,
759 BPF_SIZE(insn->code), BPF_READ, -1);
760 if (err)
761 return err;
762
763 /* check whether atomic_add can write into the same memory */
764 return check_mem_access(env, insn->dst_reg, insn->off,
765 BPF_SIZE(insn->code), BPF_WRITE, -1);
766 }
767
768 /* when register 'regno' is passed into function that will read 'access_size'
769 * bytes from that pointer, make sure that it's within stack boundary
770 * and all elements of stack are initialized
771 */
check_stack_boundary(struct verifier_env * env,int regno,int access_size)772 static int check_stack_boundary(struct verifier_env *env,
773 int regno, int access_size)
774 {
775 struct verifier_state *state = &env->cur_state;
776 struct reg_state *regs = state->regs;
777 int off, i;
778
779 if (regs[regno].type != PTR_TO_STACK)
780 return -EACCES;
781
782 off = regs[regno].imm;
783 if (off >= 0 || off < -MAX_BPF_STACK || off + access_size > 0 ||
784 access_size <= 0) {
785 verbose("invalid stack type R%d off=%d access_size=%d\n",
786 regno, off, access_size);
787 return -EACCES;
788 }
789
790 for (i = 0; i < access_size; i++) {
791 if (state->stack_slot_type[MAX_BPF_STACK + off + i] != STACK_MISC) {
792 verbose("invalid indirect read from stack off %d+%d size %d\n",
793 off, i, access_size);
794 return -EACCES;
795 }
796 }
797 return 0;
798 }
799
check_func_arg(struct verifier_env * env,u32 regno,enum bpf_arg_type arg_type,struct bpf_map ** mapp)800 static int check_func_arg(struct verifier_env *env, u32 regno,
801 enum bpf_arg_type arg_type, struct bpf_map **mapp)
802 {
803 struct reg_state *reg = env->cur_state.regs + regno;
804 enum bpf_reg_type expected_type;
805 int err = 0;
806
807 if (arg_type == ARG_DONTCARE)
808 return 0;
809
810 if (reg->type == NOT_INIT) {
811 verbose("R%d !read_ok\n", regno);
812 return -EACCES;
813 }
814
815 if (arg_type == ARG_ANYTHING) {
816 if (is_pointer_value(env, regno)) {
817 verbose("R%d leaks addr into helper function\n", regno);
818 return -EACCES;
819 }
820 return 0;
821 }
822
823 if (arg_type == ARG_PTR_TO_STACK || arg_type == ARG_PTR_TO_MAP_KEY ||
824 arg_type == ARG_PTR_TO_MAP_VALUE) {
825 expected_type = PTR_TO_STACK;
826 } else if (arg_type == ARG_CONST_STACK_SIZE) {
827 expected_type = CONST_IMM;
828 } else if (arg_type == ARG_CONST_MAP_PTR) {
829 expected_type = CONST_PTR_TO_MAP;
830 } else if (arg_type == ARG_PTR_TO_CTX) {
831 expected_type = PTR_TO_CTX;
832 } else {
833 verbose("unsupported arg_type %d\n", arg_type);
834 return -EFAULT;
835 }
836
837 if (reg->type != expected_type) {
838 verbose("R%d type=%s expected=%s\n", regno,
839 reg_type_str[reg->type], reg_type_str[expected_type]);
840 return -EACCES;
841 }
842
843 if (arg_type == ARG_CONST_MAP_PTR) {
844 /* bpf_map_xxx(map_ptr) call: remember that map_ptr */
845 *mapp = reg->map_ptr;
846
847 } else if (arg_type == ARG_PTR_TO_MAP_KEY) {
848 /* bpf_map_xxx(..., map_ptr, ..., key) call:
849 * check that [key, key + map->key_size) are within
850 * stack limits and initialized
851 */
852 if (!*mapp) {
853 /* in function declaration map_ptr must come before
854 * map_key, so that it's verified and known before
855 * we have to check map_key here. Otherwise it means
856 * that kernel subsystem misconfigured verifier
857 */
858 verbose("invalid map_ptr to access map->key\n");
859 return -EACCES;
860 }
861 err = check_stack_boundary(env, regno, (*mapp)->key_size);
862
863 } else if (arg_type == ARG_PTR_TO_MAP_VALUE) {
864 /* bpf_map_xxx(..., map_ptr, ..., value) call:
865 * check [value, value + map->value_size) validity
866 */
867 if (!*mapp) {
868 /* kernel subsystem misconfigured verifier */
869 verbose("invalid map_ptr to access map->value\n");
870 return -EACCES;
871 }
872 err = check_stack_boundary(env, regno, (*mapp)->value_size);
873
874 } else if (arg_type == ARG_CONST_STACK_SIZE) {
875 /* bpf_xxx(..., buf, len) call will access 'len' bytes
876 * from stack pointer 'buf'. Check it
877 * note: regno == len, regno - 1 == buf
878 */
879 if (regno == 0) {
880 /* kernel subsystem misconfigured verifier */
881 verbose("ARG_CONST_STACK_SIZE cannot be first argument\n");
882 return -EACCES;
883 }
884 err = check_stack_boundary(env, regno - 1, reg->imm);
885 }
886
887 return err;
888 }
889
check_map_func_compatibility(struct bpf_map * map,int func_id)890 static int check_map_func_compatibility(struct bpf_map *map, int func_id)
891 {
892 if (!map)
893 return 0;
894
895 /* We need a two way check, first is from map perspective ... */
896 switch (map->map_type) {
897 case BPF_MAP_TYPE_PROG_ARRAY:
898 if (func_id != BPF_FUNC_tail_call)
899 goto error;
900 break;
901 case BPF_MAP_TYPE_PERF_EVENT_ARRAY:
902 if (func_id != BPF_FUNC_perf_event_read &&
903 func_id != BPF_FUNC_perf_event_output)
904 goto error;
905 break;
906 default:
907 break;
908 }
909
910 /* ... and second from the function itself. */
911 switch (func_id) {
912 case BPF_FUNC_tail_call:
913 if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY)
914 goto error;
915 break;
916 case BPF_FUNC_perf_event_read:
917 case BPF_FUNC_perf_event_output:
918 if (map->map_type != BPF_MAP_TYPE_PERF_EVENT_ARRAY)
919 goto error;
920 break;
921 default:
922 break;
923 }
924
925 return 0;
926 error:
927 verbose("cannot pass map_type %d into func %d\n",
928 map->map_type, func_id);
929 return -EINVAL;
930 }
931
check_call(struct verifier_env * env,int func_id)932 static int check_call(struct verifier_env *env, int func_id)
933 {
934 struct verifier_state *state = &env->cur_state;
935 const struct bpf_func_proto *fn = NULL;
936 struct reg_state *regs = state->regs;
937 struct bpf_map *map = NULL;
938 struct reg_state *reg;
939 int i, err;
940
941 /* find function prototype */
942 if (func_id < 0 || func_id >= __BPF_FUNC_MAX_ID) {
943 verbose("invalid func %d\n", func_id);
944 return -EINVAL;
945 }
946
947 if (env->prog->aux->ops->get_func_proto)
948 fn = env->prog->aux->ops->get_func_proto(func_id);
949
950 if (!fn) {
951 verbose("unknown func %d\n", func_id);
952 return -EINVAL;
953 }
954
955 /* eBPF programs must be GPL compatible to use GPL-ed functions */
956 if (!env->prog->gpl_compatible && fn->gpl_only) {
957 verbose("cannot call GPL only function from proprietary program\n");
958 return -EINVAL;
959 }
960
961 /* check args */
962 err = check_func_arg(env, BPF_REG_1, fn->arg1_type, &map);
963 if (err)
964 return err;
965 err = check_func_arg(env, BPF_REG_2, fn->arg2_type, &map);
966 if (err)
967 return err;
968 err = check_func_arg(env, BPF_REG_3, fn->arg3_type, &map);
969 if (err)
970 return err;
971 err = check_func_arg(env, BPF_REG_4, fn->arg4_type, &map);
972 if (err)
973 return err;
974 err = check_func_arg(env, BPF_REG_5, fn->arg5_type, &map);
975 if (err)
976 return err;
977
978 /* reset caller saved regs */
979 for (i = 0; i < CALLER_SAVED_REGS; i++) {
980 reg = regs + caller_saved[i];
981 reg->type = NOT_INIT;
982 reg->imm = 0;
983 }
984
985 /* update return register */
986 if (fn->ret_type == RET_INTEGER) {
987 regs[BPF_REG_0].type = UNKNOWN_VALUE;
988 } else if (fn->ret_type == RET_VOID) {
989 regs[BPF_REG_0].type = NOT_INIT;
990 } else if (fn->ret_type == RET_PTR_TO_MAP_VALUE_OR_NULL) {
991 regs[BPF_REG_0].type = PTR_TO_MAP_VALUE_OR_NULL;
992 /* remember map_ptr, so that check_map_access()
993 * can check 'value_size' boundary of memory access
994 * to map element returned from bpf_map_lookup_elem()
995 */
996 if (map == NULL) {
997 verbose("kernel subsystem misconfigured verifier\n");
998 return -EINVAL;
999 }
1000 regs[BPF_REG_0].map_ptr = map;
1001 } else {
1002 verbose("unknown return type %d of func %d\n",
1003 fn->ret_type, func_id);
1004 return -EINVAL;
1005 }
1006
1007 err = check_map_func_compatibility(map, func_id);
1008 if (err)
1009 return err;
1010
1011 return 0;
1012 }
1013
1014 /* check validity of 32-bit and 64-bit arithmetic operations */
check_alu_op(struct verifier_env * env,struct bpf_insn * insn)1015 static int check_alu_op(struct verifier_env *env, struct bpf_insn *insn)
1016 {
1017 struct reg_state *regs = env->cur_state.regs;
1018 u8 opcode = BPF_OP(insn->code);
1019 int err;
1020
1021 if (opcode == BPF_END || opcode == BPF_NEG) {
1022 if (opcode == BPF_NEG) {
1023 if (BPF_SRC(insn->code) != 0 ||
1024 insn->src_reg != BPF_REG_0 ||
1025 insn->off != 0 || insn->imm != 0) {
1026 verbose("BPF_NEG uses reserved fields\n");
1027 return -EINVAL;
1028 }
1029 } else {
1030 if (insn->src_reg != BPF_REG_0 || insn->off != 0 ||
1031 (insn->imm != 16 && insn->imm != 32 && insn->imm != 64)) {
1032 verbose("BPF_END uses reserved fields\n");
1033 return -EINVAL;
1034 }
1035 }
1036
1037 /* check src operand */
1038 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1039 if (err)
1040 return err;
1041
1042 if (is_pointer_value(env, insn->dst_reg)) {
1043 verbose("R%d pointer arithmetic prohibited\n",
1044 insn->dst_reg);
1045 return -EACCES;
1046 }
1047
1048 /* check dest operand */
1049 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1050 if (err)
1051 return err;
1052
1053 } else if (opcode == BPF_MOV) {
1054
1055 if (BPF_SRC(insn->code) == BPF_X) {
1056 if (insn->imm != 0 || insn->off != 0) {
1057 verbose("BPF_MOV uses reserved fields\n");
1058 return -EINVAL;
1059 }
1060
1061 /* check src operand */
1062 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1063 if (err)
1064 return err;
1065 } else {
1066 if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
1067 verbose("BPF_MOV uses reserved fields\n");
1068 return -EINVAL;
1069 }
1070 }
1071
1072 /* check dest operand */
1073 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1074 if (err)
1075 return err;
1076
1077 if (BPF_SRC(insn->code) == BPF_X) {
1078 if (BPF_CLASS(insn->code) == BPF_ALU64) {
1079 /* case: R1 = R2
1080 * copy register state to dest reg
1081 */
1082 regs[insn->dst_reg] = regs[insn->src_reg];
1083 } else {
1084 if (is_pointer_value(env, insn->src_reg)) {
1085 verbose("R%d partial copy of pointer\n",
1086 insn->src_reg);
1087 return -EACCES;
1088 }
1089 regs[insn->dst_reg].type = UNKNOWN_VALUE;
1090 regs[insn->dst_reg].map_ptr = NULL;
1091 }
1092 } else {
1093 /* case: R = imm
1094 * remember the value we stored into this reg
1095 */
1096 regs[insn->dst_reg].type = CONST_IMM;
1097 regs[insn->dst_reg].imm = insn->imm;
1098 }
1099
1100 } else if (opcode > BPF_END) {
1101 verbose("invalid BPF_ALU opcode %x\n", opcode);
1102 return -EINVAL;
1103
1104 } else { /* all other ALU ops: and, sub, xor, add, ... */
1105
1106 bool stack_relative = false;
1107
1108 if (BPF_SRC(insn->code) == BPF_X) {
1109 if (insn->imm != 0 || insn->off != 0) {
1110 verbose("BPF_ALU uses reserved fields\n");
1111 return -EINVAL;
1112 }
1113 /* check src1 operand */
1114 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1115 if (err)
1116 return err;
1117 } else {
1118 if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
1119 verbose("BPF_ALU uses reserved fields\n");
1120 return -EINVAL;
1121 }
1122 }
1123
1124 /* check src2 operand */
1125 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1126 if (err)
1127 return err;
1128
1129 if ((opcode == BPF_MOD || opcode == BPF_DIV) &&
1130 BPF_SRC(insn->code) == BPF_K && insn->imm == 0) {
1131 verbose("div by zero\n");
1132 return -EINVAL;
1133 }
1134
1135 if ((opcode == BPF_LSH || opcode == BPF_RSH ||
1136 opcode == BPF_ARSH) && BPF_SRC(insn->code) == BPF_K) {
1137 int size = BPF_CLASS(insn->code) == BPF_ALU64 ? 64 : 32;
1138
1139 if (insn->imm < 0 || insn->imm >= size) {
1140 verbose("invalid shift %d\n", insn->imm);
1141 return -EINVAL;
1142 }
1143 }
1144
1145 /* pattern match 'bpf_add Rx, imm' instruction */
1146 if (opcode == BPF_ADD && BPF_CLASS(insn->code) == BPF_ALU64 &&
1147 regs[insn->dst_reg].type == FRAME_PTR &&
1148 BPF_SRC(insn->code) == BPF_K) {
1149 stack_relative = true;
1150 } else if (is_pointer_value(env, insn->dst_reg)) {
1151 verbose("R%d pointer arithmetic prohibited\n",
1152 insn->dst_reg);
1153 return -EACCES;
1154 } else if (BPF_SRC(insn->code) == BPF_X &&
1155 is_pointer_value(env, insn->src_reg)) {
1156 verbose("R%d pointer arithmetic prohibited\n",
1157 insn->src_reg);
1158 return -EACCES;
1159 }
1160
1161 /* check dest operand */
1162 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1163 if (err)
1164 return err;
1165
1166 if (stack_relative) {
1167 regs[insn->dst_reg].type = PTR_TO_STACK;
1168 regs[insn->dst_reg].imm = insn->imm;
1169 }
1170 }
1171
1172 return 0;
1173 }
1174
check_cond_jmp_op(struct verifier_env * env,struct bpf_insn * insn,int * insn_idx)1175 static int check_cond_jmp_op(struct verifier_env *env,
1176 struct bpf_insn *insn, int *insn_idx)
1177 {
1178 struct reg_state *regs = env->cur_state.regs;
1179 struct verifier_state *other_branch;
1180 u8 opcode = BPF_OP(insn->code);
1181 int err;
1182
1183 if (opcode > BPF_EXIT) {
1184 verbose("invalid BPF_JMP opcode %x\n", opcode);
1185 return -EINVAL;
1186 }
1187
1188 if (BPF_SRC(insn->code) == BPF_X) {
1189 if (insn->imm != 0) {
1190 verbose("BPF_JMP uses reserved fields\n");
1191 return -EINVAL;
1192 }
1193
1194 /* check src1 operand */
1195 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1196 if (err)
1197 return err;
1198
1199 if (is_pointer_value(env, insn->src_reg)) {
1200 verbose("R%d pointer comparison prohibited\n",
1201 insn->src_reg);
1202 return -EACCES;
1203 }
1204 } else {
1205 if (insn->src_reg != BPF_REG_0) {
1206 verbose("BPF_JMP uses reserved fields\n");
1207 return -EINVAL;
1208 }
1209 }
1210
1211 /* check src2 operand */
1212 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1213 if (err)
1214 return err;
1215
1216 /* detect if R == 0 where R was initialized to zero earlier */
1217 if (BPF_SRC(insn->code) == BPF_K &&
1218 (opcode == BPF_JEQ || opcode == BPF_JNE) &&
1219 regs[insn->dst_reg].type == CONST_IMM &&
1220 regs[insn->dst_reg].imm == insn->imm) {
1221 if (opcode == BPF_JEQ) {
1222 /* if (imm == imm) goto pc+off;
1223 * only follow the goto, ignore fall-through
1224 */
1225 *insn_idx += insn->off;
1226 return 0;
1227 } else {
1228 /* if (imm != imm) goto pc+off;
1229 * only follow fall-through branch, since
1230 * that's where the program will go
1231 */
1232 return 0;
1233 }
1234 }
1235
1236 other_branch = push_stack(env, *insn_idx + insn->off + 1, *insn_idx);
1237 if (!other_branch)
1238 return -EFAULT;
1239
1240 /* detect if R == 0 where R is returned value from bpf_map_lookup_elem() */
1241 if (BPF_SRC(insn->code) == BPF_K &&
1242 insn->imm == 0 && (opcode == BPF_JEQ ||
1243 opcode == BPF_JNE) &&
1244 regs[insn->dst_reg].type == PTR_TO_MAP_VALUE_OR_NULL) {
1245 if (opcode == BPF_JEQ) {
1246 /* next fallthrough insn can access memory via
1247 * this register
1248 */
1249 regs[insn->dst_reg].type = PTR_TO_MAP_VALUE;
1250 /* branch targer cannot access it, since reg == 0 */
1251 other_branch->regs[insn->dst_reg].type = CONST_IMM;
1252 other_branch->regs[insn->dst_reg].imm = 0;
1253 } else {
1254 other_branch->regs[insn->dst_reg].type = PTR_TO_MAP_VALUE;
1255 regs[insn->dst_reg].type = CONST_IMM;
1256 regs[insn->dst_reg].imm = 0;
1257 }
1258 } else if (is_pointer_value(env, insn->dst_reg)) {
1259 verbose("R%d pointer comparison prohibited\n", insn->dst_reg);
1260 return -EACCES;
1261 } else if (BPF_SRC(insn->code) == BPF_K &&
1262 (opcode == BPF_JEQ || opcode == BPF_JNE)) {
1263
1264 if (opcode == BPF_JEQ) {
1265 /* detect if (R == imm) goto
1266 * and in the target state recognize that R = imm
1267 */
1268 other_branch->regs[insn->dst_reg].type = CONST_IMM;
1269 other_branch->regs[insn->dst_reg].imm = insn->imm;
1270 } else {
1271 /* detect if (R != imm) goto
1272 * and in the fall-through state recognize that R = imm
1273 */
1274 regs[insn->dst_reg].type = CONST_IMM;
1275 regs[insn->dst_reg].imm = insn->imm;
1276 }
1277 }
1278 if (log_level)
1279 print_verifier_state(env);
1280 return 0;
1281 }
1282
1283 /* return the map pointer stored inside BPF_LD_IMM64 instruction */
ld_imm64_to_map_ptr(struct bpf_insn * insn)1284 static struct bpf_map *ld_imm64_to_map_ptr(struct bpf_insn *insn)
1285 {
1286 u64 imm64 = ((u64) (u32) insn[0].imm) | ((u64) (u32) insn[1].imm) << 32;
1287
1288 return (struct bpf_map *) (unsigned long) imm64;
1289 }
1290
1291 /* verify BPF_LD_IMM64 instruction */
check_ld_imm(struct verifier_env * env,struct bpf_insn * insn)1292 static int check_ld_imm(struct verifier_env *env, struct bpf_insn *insn)
1293 {
1294 struct reg_state *regs = env->cur_state.regs;
1295 int err;
1296
1297 if (BPF_SIZE(insn->code) != BPF_DW) {
1298 verbose("invalid BPF_LD_IMM insn\n");
1299 return -EINVAL;
1300 }
1301 if (insn->off != 0) {
1302 verbose("BPF_LD_IMM64 uses reserved fields\n");
1303 return -EINVAL;
1304 }
1305
1306 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1307 if (err)
1308 return err;
1309
1310 if (insn->src_reg == 0)
1311 /* generic move 64-bit immediate into a register */
1312 return 0;
1313
1314 /* replace_map_fd_with_map_ptr() should have caught bad ld_imm64 */
1315 BUG_ON(insn->src_reg != BPF_PSEUDO_MAP_FD);
1316
1317 regs[insn->dst_reg].type = CONST_PTR_TO_MAP;
1318 regs[insn->dst_reg].map_ptr = ld_imm64_to_map_ptr(insn);
1319 return 0;
1320 }
1321
may_access_skb(enum bpf_prog_type type)1322 static bool may_access_skb(enum bpf_prog_type type)
1323 {
1324 switch (type) {
1325 case BPF_PROG_TYPE_SOCKET_FILTER:
1326 case BPF_PROG_TYPE_SCHED_CLS:
1327 case BPF_PROG_TYPE_SCHED_ACT:
1328 return true;
1329 default:
1330 return false;
1331 }
1332 }
1333
1334 /* verify safety of LD_ABS|LD_IND instructions:
1335 * - they can only appear in the programs where ctx == skb
1336 * - since they are wrappers of function calls, they scratch R1-R5 registers,
1337 * preserve R6-R9, and store return value into R0
1338 *
1339 * Implicit input:
1340 * ctx == skb == R6 == CTX
1341 *
1342 * Explicit input:
1343 * SRC == any register
1344 * IMM == 32-bit immediate
1345 *
1346 * Output:
1347 * R0 - 8/16/32-bit skb data converted to cpu endianness
1348 */
check_ld_abs(struct verifier_env * env,struct bpf_insn * insn)1349 static int check_ld_abs(struct verifier_env *env, struct bpf_insn *insn)
1350 {
1351 struct reg_state *regs = env->cur_state.regs;
1352 u8 mode = BPF_MODE(insn->code);
1353 struct reg_state *reg;
1354 int i, err;
1355
1356 if (!may_access_skb(env->prog->type)) {
1357 verbose("BPF_LD_ABS|IND instructions not allowed for this program type\n");
1358 return -EINVAL;
1359 }
1360
1361 if (insn->dst_reg != BPF_REG_0 || insn->off != 0 ||
1362 BPF_SIZE(insn->code) == BPF_DW ||
1363 (mode == BPF_ABS && insn->src_reg != BPF_REG_0)) {
1364 verbose("BPF_LD_ABS uses reserved fields\n");
1365 return -EINVAL;
1366 }
1367
1368 /* check whether implicit source operand (register R6) is readable */
1369 err = check_reg_arg(regs, BPF_REG_6, SRC_OP);
1370 if (err)
1371 return err;
1372
1373 if (regs[BPF_REG_6].type != PTR_TO_CTX) {
1374 verbose("at the time of BPF_LD_ABS|IND R6 != pointer to skb\n");
1375 return -EINVAL;
1376 }
1377
1378 if (mode == BPF_IND) {
1379 /* check explicit source operand */
1380 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1381 if (err)
1382 return err;
1383 }
1384
1385 /* reset caller saved regs to unreadable */
1386 for (i = 0; i < CALLER_SAVED_REGS; i++) {
1387 reg = regs + caller_saved[i];
1388 reg->type = NOT_INIT;
1389 reg->imm = 0;
1390 }
1391
1392 /* mark destination R0 register as readable, since it contains
1393 * the value fetched from the packet
1394 */
1395 regs[BPF_REG_0].type = UNKNOWN_VALUE;
1396 return 0;
1397 }
1398
1399 /* non-recursive DFS pseudo code
1400 * 1 procedure DFS-iterative(G,v):
1401 * 2 label v as discovered
1402 * 3 let S be a stack
1403 * 4 S.push(v)
1404 * 5 while S is not empty
1405 * 6 t <- S.pop()
1406 * 7 if t is what we're looking for:
1407 * 8 return t
1408 * 9 for all edges e in G.adjacentEdges(t) do
1409 * 10 if edge e is already labelled
1410 * 11 continue with the next edge
1411 * 12 w <- G.adjacentVertex(t,e)
1412 * 13 if vertex w is not discovered and not explored
1413 * 14 label e as tree-edge
1414 * 15 label w as discovered
1415 * 16 S.push(w)
1416 * 17 continue at 5
1417 * 18 else if vertex w is discovered
1418 * 19 label e as back-edge
1419 * 20 else
1420 * 21 // vertex w is explored
1421 * 22 label e as forward- or cross-edge
1422 * 23 label t as explored
1423 * 24 S.pop()
1424 *
1425 * convention:
1426 * 0x10 - discovered
1427 * 0x11 - discovered and fall-through edge labelled
1428 * 0x12 - discovered and fall-through and branch edges labelled
1429 * 0x20 - explored
1430 */
1431
1432 enum {
1433 DISCOVERED = 0x10,
1434 EXPLORED = 0x20,
1435 FALLTHROUGH = 1,
1436 BRANCH = 2,
1437 };
1438
1439 #define STATE_LIST_MARK ((struct verifier_state_list *) -1L)
1440
1441 static int *insn_stack; /* stack of insns to process */
1442 static int cur_stack; /* current stack index */
1443 static int *insn_state;
1444
1445 /* t, w, e - match pseudo-code above:
1446 * t - index of current instruction
1447 * w - next instruction
1448 * e - edge
1449 */
push_insn(int t,int w,int e,struct verifier_env * env)1450 static int push_insn(int t, int w, int e, struct verifier_env *env)
1451 {
1452 if (e == FALLTHROUGH && insn_state[t] >= (DISCOVERED | FALLTHROUGH))
1453 return 0;
1454
1455 if (e == BRANCH && insn_state[t] >= (DISCOVERED | BRANCH))
1456 return 0;
1457
1458 if (w < 0 || w >= env->prog->len) {
1459 verbose("jump out of range from insn %d to %d\n", t, w);
1460 return -EINVAL;
1461 }
1462
1463 if (e == BRANCH)
1464 /* mark branch target for state pruning */
1465 env->explored_states[w] = STATE_LIST_MARK;
1466
1467 if (insn_state[w] == 0) {
1468 /* tree-edge */
1469 insn_state[t] = DISCOVERED | e;
1470 insn_state[w] = DISCOVERED;
1471 if (cur_stack >= env->prog->len)
1472 return -E2BIG;
1473 insn_stack[cur_stack++] = w;
1474 return 1;
1475 } else if ((insn_state[w] & 0xF0) == DISCOVERED) {
1476 verbose("back-edge from insn %d to %d\n", t, w);
1477 return -EINVAL;
1478 } else if (insn_state[w] == EXPLORED) {
1479 /* forward- or cross-edge */
1480 insn_state[t] = DISCOVERED | e;
1481 } else {
1482 verbose("insn state internal bug\n");
1483 return -EFAULT;
1484 }
1485 return 0;
1486 }
1487
1488 /* non-recursive depth-first-search to detect loops in BPF program
1489 * loop == back-edge in directed graph
1490 */
check_cfg(struct verifier_env * env)1491 static int check_cfg(struct verifier_env *env)
1492 {
1493 struct bpf_insn *insns = env->prog->insnsi;
1494 int insn_cnt = env->prog->len;
1495 int ret = 0;
1496 int i, t;
1497
1498 insn_state = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
1499 if (!insn_state)
1500 return -ENOMEM;
1501
1502 insn_stack = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
1503 if (!insn_stack) {
1504 kfree(insn_state);
1505 return -ENOMEM;
1506 }
1507
1508 insn_state[0] = DISCOVERED; /* mark 1st insn as discovered */
1509 insn_stack[0] = 0; /* 0 is the first instruction */
1510 cur_stack = 1;
1511
1512 peek_stack:
1513 if (cur_stack == 0)
1514 goto check_state;
1515 t = insn_stack[cur_stack - 1];
1516
1517 if (BPF_CLASS(insns[t].code) == BPF_JMP) {
1518 u8 opcode = BPF_OP(insns[t].code);
1519
1520 if (opcode == BPF_EXIT) {
1521 goto mark_explored;
1522 } else if (opcode == BPF_CALL) {
1523 ret = push_insn(t, t + 1, FALLTHROUGH, env);
1524 if (ret == 1)
1525 goto peek_stack;
1526 else if (ret < 0)
1527 goto err_free;
1528 } else if (opcode == BPF_JA) {
1529 if (BPF_SRC(insns[t].code) != BPF_K) {
1530 ret = -EINVAL;
1531 goto err_free;
1532 }
1533 /* unconditional jump with single edge */
1534 ret = push_insn(t, t + insns[t].off + 1,
1535 FALLTHROUGH, env);
1536 if (ret == 1)
1537 goto peek_stack;
1538 else if (ret < 0)
1539 goto err_free;
1540 /* tell verifier to check for equivalent states
1541 * after every call and jump
1542 */
1543 if (t + 1 < insn_cnt)
1544 env->explored_states[t + 1] = STATE_LIST_MARK;
1545 } else {
1546 /* conditional jump with two edges */
1547 ret = push_insn(t, t + 1, FALLTHROUGH, env);
1548 if (ret == 1)
1549 goto peek_stack;
1550 else if (ret < 0)
1551 goto err_free;
1552
1553 ret = push_insn(t, t + insns[t].off + 1, BRANCH, env);
1554 if (ret == 1)
1555 goto peek_stack;
1556 else if (ret < 0)
1557 goto err_free;
1558 }
1559 } else {
1560 /* all other non-branch instructions with single
1561 * fall-through edge
1562 */
1563 ret = push_insn(t, t + 1, FALLTHROUGH, env);
1564 if (ret == 1)
1565 goto peek_stack;
1566 else if (ret < 0)
1567 goto err_free;
1568 }
1569
1570 mark_explored:
1571 insn_state[t] = EXPLORED;
1572 if (cur_stack-- <= 0) {
1573 verbose("pop stack internal bug\n");
1574 ret = -EFAULT;
1575 goto err_free;
1576 }
1577 goto peek_stack;
1578
1579 check_state:
1580 for (i = 0; i < insn_cnt; i++) {
1581 if (insn_state[i] != EXPLORED) {
1582 verbose("unreachable insn %d\n", i);
1583 ret = -EINVAL;
1584 goto err_free;
1585 }
1586 }
1587 ret = 0; /* cfg looks good */
1588
1589 err_free:
1590 kfree(insn_state);
1591 kfree(insn_stack);
1592 return ret;
1593 }
1594
1595 /* compare two verifier states
1596 *
1597 * all states stored in state_list are known to be valid, since
1598 * verifier reached 'bpf_exit' instruction through them
1599 *
1600 * this function is called when verifier exploring different branches of
1601 * execution popped from the state stack. If it sees an old state that has
1602 * more strict register state and more strict stack state then this execution
1603 * branch doesn't need to be explored further, since verifier already
1604 * concluded that more strict state leads to valid finish.
1605 *
1606 * Therefore two states are equivalent if register state is more conservative
1607 * and explored stack state is more conservative than the current one.
1608 * Example:
1609 * explored current
1610 * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC)
1611 * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC)
1612 *
1613 * In other words if current stack state (one being explored) has more
1614 * valid slots than old one that already passed validation, it means
1615 * the verifier can stop exploring and conclude that current state is valid too
1616 *
1617 * Similarly with registers. If explored state has register type as invalid
1618 * whereas register type in current state is meaningful, it means that
1619 * the current state will reach 'bpf_exit' instruction safely
1620 */
states_equal(struct verifier_state * old,struct verifier_state * cur)1621 static bool states_equal(struct verifier_state *old, struct verifier_state *cur)
1622 {
1623 int i;
1624
1625 for (i = 0; i < MAX_BPF_REG; i++) {
1626 if (memcmp(&old->regs[i], &cur->regs[i],
1627 sizeof(old->regs[0])) != 0) {
1628 if (old->regs[i].type == NOT_INIT ||
1629 (old->regs[i].type == UNKNOWN_VALUE &&
1630 cur->regs[i].type != NOT_INIT))
1631 continue;
1632 return false;
1633 }
1634 }
1635
1636 for (i = 0; i < MAX_BPF_STACK; i++) {
1637 if (old->stack_slot_type[i] == STACK_INVALID)
1638 continue;
1639 if (old->stack_slot_type[i] != cur->stack_slot_type[i])
1640 /* Ex: old explored (safe) state has STACK_SPILL in
1641 * this stack slot, but current has has STACK_MISC ->
1642 * this verifier states are not equivalent,
1643 * return false to continue verification of this path
1644 */
1645 return false;
1646 if (i % BPF_REG_SIZE)
1647 continue;
1648 if (memcmp(&old->spilled_regs[i / BPF_REG_SIZE],
1649 &cur->spilled_regs[i / BPF_REG_SIZE],
1650 sizeof(old->spilled_regs[0])))
1651 /* when explored and current stack slot types are
1652 * the same, check that stored pointers types
1653 * are the same as well.
1654 * Ex: explored safe path could have stored
1655 * (struct reg_state) {.type = PTR_TO_STACK, .imm = -8}
1656 * but current path has stored:
1657 * (struct reg_state) {.type = PTR_TO_STACK, .imm = -16}
1658 * such verifier states are not equivalent.
1659 * return false to continue verification of this path
1660 */
1661 return false;
1662 else
1663 continue;
1664 }
1665 return true;
1666 }
1667
is_state_visited(struct verifier_env * env,int insn_idx)1668 static int is_state_visited(struct verifier_env *env, int insn_idx)
1669 {
1670 struct verifier_state_list *new_sl;
1671 struct verifier_state_list *sl;
1672
1673 sl = env->explored_states[insn_idx];
1674 if (!sl)
1675 /* this 'insn_idx' instruction wasn't marked, so we will not
1676 * be doing state search here
1677 */
1678 return 0;
1679
1680 while (sl != STATE_LIST_MARK) {
1681 if (states_equal(&sl->state, &env->cur_state))
1682 /* reached equivalent register/stack state,
1683 * prune the search
1684 */
1685 return 1;
1686 sl = sl->next;
1687 }
1688
1689 /* there were no equivalent states, remember current one.
1690 * technically the current state is not proven to be safe yet,
1691 * but it will either reach bpf_exit (which means it's safe) or
1692 * it will be rejected. Since there are no loops, we won't be
1693 * seeing this 'insn_idx' instruction again on the way to bpf_exit
1694 */
1695 new_sl = kmalloc(sizeof(struct verifier_state_list), GFP_USER);
1696 if (!new_sl)
1697 return -ENOMEM;
1698
1699 /* add new state to the head of linked list */
1700 memcpy(&new_sl->state, &env->cur_state, sizeof(env->cur_state));
1701 new_sl->next = env->explored_states[insn_idx];
1702 env->explored_states[insn_idx] = new_sl;
1703 return 0;
1704 }
1705
do_check(struct verifier_env * env)1706 static int do_check(struct verifier_env *env)
1707 {
1708 struct verifier_state *state = &env->cur_state;
1709 struct bpf_insn *insns = env->prog->insnsi;
1710 struct reg_state *regs = state->regs;
1711 int insn_cnt = env->prog->len;
1712 int insn_idx, prev_insn_idx = 0;
1713 int insn_processed = 0;
1714 bool do_print_state = false;
1715
1716 init_reg_state(regs);
1717 insn_idx = 0;
1718 for (;;) {
1719 struct bpf_insn *insn;
1720 u8 class;
1721 int err;
1722
1723 if (insn_idx >= insn_cnt) {
1724 verbose("invalid insn idx %d insn_cnt %d\n",
1725 insn_idx, insn_cnt);
1726 return -EFAULT;
1727 }
1728
1729 insn = &insns[insn_idx];
1730 class = BPF_CLASS(insn->code);
1731
1732 if (++insn_processed > 32768) {
1733 verbose("BPF program is too large. Proccessed %d insn\n",
1734 insn_processed);
1735 return -E2BIG;
1736 }
1737
1738 err = is_state_visited(env, insn_idx);
1739 if (err < 0)
1740 return err;
1741 if (err == 1) {
1742 /* found equivalent state, can prune the search */
1743 if (log_level) {
1744 if (do_print_state)
1745 verbose("\nfrom %d to %d: safe\n",
1746 prev_insn_idx, insn_idx);
1747 else
1748 verbose("%d: safe\n", insn_idx);
1749 }
1750 goto process_bpf_exit;
1751 }
1752
1753 if (log_level && do_print_state) {
1754 verbose("\nfrom %d to %d:", prev_insn_idx, insn_idx);
1755 print_verifier_state(env);
1756 do_print_state = false;
1757 }
1758
1759 if (log_level) {
1760 verbose("%d: ", insn_idx);
1761 print_bpf_insn(insn);
1762 }
1763
1764 if (class == BPF_ALU || class == BPF_ALU64) {
1765 err = check_alu_op(env, insn);
1766 if (err)
1767 return err;
1768
1769 } else if (class == BPF_LDX) {
1770 enum bpf_reg_type src_reg_type;
1771
1772 /* check for reserved fields is already done */
1773
1774 /* check src operand */
1775 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1776 if (err)
1777 return err;
1778
1779 err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK);
1780 if (err)
1781 return err;
1782
1783 src_reg_type = regs[insn->src_reg].type;
1784
1785 /* check that memory (src_reg + off) is readable,
1786 * the state of dst_reg will be updated by this func
1787 */
1788 err = check_mem_access(env, insn->src_reg, insn->off,
1789 BPF_SIZE(insn->code), BPF_READ,
1790 insn->dst_reg);
1791 if (err)
1792 return err;
1793
1794 if (BPF_SIZE(insn->code) != BPF_W) {
1795 insn_idx++;
1796 continue;
1797 }
1798
1799 if (insn->imm == 0) {
1800 /* saw a valid insn
1801 * dst_reg = *(u32 *)(src_reg + off)
1802 * use reserved 'imm' field to mark this insn
1803 */
1804 insn->imm = src_reg_type;
1805
1806 } else if (src_reg_type != insn->imm &&
1807 (src_reg_type == PTR_TO_CTX ||
1808 insn->imm == PTR_TO_CTX)) {
1809 /* ABuser program is trying to use the same insn
1810 * dst_reg = *(u32*) (src_reg + off)
1811 * with different pointer types:
1812 * src_reg == ctx in one branch and
1813 * src_reg == stack|map in some other branch.
1814 * Reject it.
1815 */
1816 verbose("same insn cannot be used with different pointers\n");
1817 return -EINVAL;
1818 }
1819
1820 } else if (class == BPF_STX) {
1821 enum bpf_reg_type dst_reg_type;
1822
1823 if (BPF_MODE(insn->code) == BPF_XADD) {
1824 err = check_xadd(env, insn);
1825 if (err)
1826 return err;
1827 insn_idx++;
1828 continue;
1829 }
1830
1831 /* check src1 operand */
1832 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1833 if (err)
1834 return err;
1835 /* check src2 operand */
1836 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1837 if (err)
1838 return err;
1839
1840 dst_reg_type = regs[insn->dst_reg].type;
1841
1842 /* check that memory (dst_reg + off) is writeable */
1843 err = check_mem_access(env, insn->dst_reg, insn->off,
1844 BPF_SIZE(insn->code), BPF_WRITE,
1845 insn->src_reg);
1846 if (err)
1847 return err;
1848
1849 if (insn->imm == 0) {
1850 insn->imm = dst_reg_type;
1851 } else if (dst_reg_type != insn->imm &&
1852 (dst_reg_type == PTR_TO_CTX ||
1853 insn->imm == PTR_TO_CTX)) {
1854 verbose("same insn cannot be used with different pointers\n");
1855 return -EINVAL;
1856 }
1857
1858 } else if (class == BPF_ST) {
1859 if (BPF_MODE(insn->code) != BPF_MEM ||
1860 insn->src_reg != BPF_REG_0) {
1861 verbose("BPF_ST uses reserved fields\n");
1862 return -EINVAL;
1863 }
1864 /* check src operand */
1865 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1866 if (err)
1867 return err;
1868
1869 /* check that memory (dst_reg + off) is writeable */
1870 err = check_mem_access(env, insn->dst_reg, insn->off,
1871 BPF_SIZE(insn->code), BPF_WRITE,
1872 -1);
1873 if (err)
1874 return err;
1875
1876 } else if (class == BPF_JMP) {
1877 u8 opcode = BPF_OP(insn->code);
1878
1879 if (opcode == BPF_CALL) {
1880 if (BPF_SRC(insn->code) != BPF_K ||
1881 insn->off != 0 ||
1882 insn->src_reg != BPF_REG_0 ||
1883 insn->dst_reg != BPF_REG_0) {
1884 verbose("BPF_CALL uses reserved fields\n");
1885 return -EINVAL;
1886 }
1887
1888 err = check_call(env, insn->imm);
1889 if (err)
1890 return err;
1891
1892 } else if (opcode == BPF_JA) {
1893 if (BPF_SRC(insn->code) != BPF_K ||
1894 insn->imm != 0 ||
1895 insn->src_reg != BPF_REG_0 ||
1896 insn->dst_reg != BPF_REG_0) {
1897 verbose("BPF_JA uses reserved fields\n");
1898 return -EINVAL;
1899 }
1900
1901 insn_idx += insn->off + 1;
1902 continue;
1903
1904 } else if (opcode == BPF_EXIT) {
1905 if (BPF_SRC(insn->code) != BPF_K ||
1906 insn->imm != 0 ||
1907 insn->src_reg != BPF_REG_0 ||
1908 insn->dst_reg != BPF_REG_0) {
1909 verbose("BPF_EXIT uses reserved fields\n");
1910 return -EINVAL;
1911 }
1912
1913 /* eBPF calling convetion is such that R0 is used
1914 * to return the value from eBPF program.
1915 * Make sure that it's readable at this time
1916 * of bpf_exit, which means that program wrote
1917 * something into it earlier
1918 */
1919 err = check_reg_arg(regs, BPF_REG_0, SRC_OP);
1920 if (err)
1921 return err;
1922
1923 if (is_pointer_value(env, BPF_REG_0)) {
1924 verbose("R0 leaks addr as return value\n");
1925 return -EACCES;
1926 }
1927
1928 process_bpf_exit:
1929 insn_idx = pop_stack(env, &prev_insn_idx);
1930 if (insn_idx < 0) {
1931 break;
1932 } else {
1933 do_print_state = true;
1934 continue;
1935 }
1936 } else {
1937 err = check_cond_jmp_op(env, insn, &insn_idx);
1938 if (err)
1939 return err;
1940 }
1941 } else if (class == BPF_LD) {
1942 u8 mode = BPF_MODE(insn->code);
1943
1944 if (mode == BPF_ABS || mode == BPF_IND) {
1945 err = check_ld_abs(env, insn);
1946 if (err)
1947 return err;
1948
1949 } else if (mode == BPF_IMM) {
1950 err = check_ld_imm(env, insn);
1951 if (err)
1952 return err;
1953
1954 insn_idx++;
1955 } else {
1956 verbose("invalid BPF_LD mode\n");
1957 return -EINVAL;
1958 }
1959 } else {
1960 verbose("unknown insn class %d\n", class);
1961 return -EINVAL;
1962 }
1963
1964 insn_idx++;
1965 }
1966
1967 return 0;
1968 }
1969
1970 /* look for pseudo eBPF instructions that access map FDs and
1971 * replace them with actual map pointers
1972 */
replace_map_fd_with_map_ptr(struct verifier_env * env)1973 static int replace_map_fd_with_map_ptr(struct verifier_env *env)
1974 {
1975 struct bpf_insn *insn = env->prog->insnsi;
1976 int insn_cnt = env->prog->len;
1977 int i, j;
1978
1979 for (i = 0; i < insn_cnt; i++, insn++) {
1980 if (BPF_CLASS(insn->code) == BPF_LDX &&
1981 (BPF_MODE(insn->code) != BPF_MEM || insn->imm != 0)) {
1982 verbose("BPF_LDX uses reserved fields\n");
1983 return -EINVAL;
1984 }
1985
1986 if (BPF_CLASS(insn->code) == BPF_STX &&
1987 ((BPF_MODE(insn->code) != BPF_MEM &&
1988 BPF_MODE(insn->code) != BPF_XADD) || insn->imm != 0)) {
1989 verbose("BPF_STX uses reserved fields\n");
1990 return -EINVAL;
1991 }
1992
1993 if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW)) {
1994 struct bpf_map *map;
1995 struct fd f;
1996
1997 if (i == insn_cnt - 1 || insn[1].code != 0 ||
1998 insn[1].dst_reg != 0 || insn[1].src_reg != 0 ||
1999 insn[1].off != 0) {
2000 verbose("invalid bpf_ld_imm64 insn\n");
2001 return -EINVAL;
2002 }
2003
2004 if (insn->src_reg == 0)
2005 /* valid generic load 64-bit imm */
2006 goto next_insn;
2007
2008 if (insn->src_reg != BPF_PSEUDO_MAP_FD) {
2009 verbose("unrecognized bpf_ld_imm64 insn\n");
2010 return -EINVAL;
2011 }
2012
2013 f = fdget(insn->imm);
2014 map = __bpf_map_get(f);
2015 if (IS_ERR(map)) {
2016 verbose("fd %d is not pointing to valid bpf_map\n",
2017 insn->imm);
2018 return PTR_ERR(map);
2019 }
2020
2021 /* store map pointer inside BPF_LD_IMM64 instruction */
2022 insn[0].imm = (u32) (unsigned long) map;
2023 insn[1].imm = ((u64) (unsigned long) map) >> 32;
2024
2025 /* check whether we recorded this map already */
2026 for (j = 0; j < env->used_map_cnt; j++)
2027 if (env->used_maps[j] == map) {
2028 fdput(f);
2029 goto next_insn;
2030 }
2031
2032 if (env->used_map_cnt >= MAX_USED_MAPS) {
2033 fdput(f);
2034 return -E2BIG;
2035 }
2036
2037 /* hold the map. If the program is rejected by verifier,
2038 * the map will be released by release_maps() or it
2039 * will be used by the valid program until it's unloaded
2040 * and all maps are released in free_bpf_prog_info()
2041 */
2042 map = bpf_map_inc(map, false);
2043 if (IS_ERR(map)) {
2044 fdput(f);
2045 return PTR_ERR(map);
2046 }
2047 env->used_maps[env->used_map_cnt++] = map;
2048
2049 fdput(f);
2050 next_insn:
2051 insn++;
2052 i++;
2053 }
2054 }
2055
2056 /* now all pseudo BPF_LD_IMM64 instructions load valid
2057 * 'struct bpf_map *' into a register instead of user map_fd.
2058 * These pointers will be used later by verifier to validate map access.
2059 */
2060 return 0;
2061 }
2062
2063 /* drop refcnt of maps used by the rejected program */
release_maps(struct verifier_env * env)2064 static void release_maps(struct verifier_env *env)
2065 {
2066 int i;
2067
2068 for (i = 0; i < env->used_map_cnt; i++)
2069 bpf_map_put(env->used_maps[i]);
2070 }
2071
2072 /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */
convert_pseudo_ld_imm64(struct verifier_env * env)2073 static void convert_pseudo_ld_imm64(struct verifier_env *env)
2074 {
2075 struct bpf_insn *insn = env->prog->insnsi;
2076 int insn_cnt = env->prog->len;
2077 int i;
2078
2079 for (i = 0; i < insn_cnt; i++, insn++)
2080 if (insn->code == (BPF_LD | BPF_IMM | BPF_DW))
2081 insn->src_reg = 0;
2082 }
2083
adjust_branches(struct bpf_prog * prog,int pos,int delta)2084 static void adjust_branches(struct bpf_prog *prog, int pos, int delta)
2085 {
2086 struct bpf_insn *insn = prog->insnsi;
2087 int insn_cnt = prog->len;
2088 int i;
2089
2090 for (i = 0; i < insn_cnt; i++, insn++) {
2091 if (BPF_CLASS(insn->code) != BPF_JMP ||
2092 BPF_OP(insn->code) == BPF_CALL ||
2093 BPF_OP(insn->code) == BPF_EXIT)
2094 continue;
2095
2096 /* adjust offset of jmps if necessary */
2097 if (i < pos && i + insn->off + 1 > pos)
2098 insn->off += delta;
2099 else if (i > pos + delta && i + insn->off + 1 <= pos + delta)
2100 insn->off -= delta;
2101 }
2102 }
2103
2104 /* convert load instructions that access fields of 'struct __sk_buff'
2105 * into sequence of instructions that access fields of 'struct sk_buff'
2106 */
convert_ctx_accesses(struct verifier_env * env)2107 static int convert_ctx_accesses(struct verifier_env *env)
2108 {
2109 struct bpf_insn *insn = env->prog->insnsi;
2110 int insn_cnt = env->prog->len;
2111 struct bpf_insn insn_buf[16];
2112 struct bpf_prog *new_prog;
2113 u32 cnt;
2114 int i;
2115 enum bpf_access_type type;
2116
2117 if (!env->prog->aux->ops->convert_ctx_access)
2118 return 0;
2119
2120 for (i = 0; i < insn_cnt; i++, insn++) {
2121 if (insn->code == (BPF_LDX | BPF_MEM | BPF_W))
2122 type = BPF_READ;
2123 else if (insn->code == (BPF_STX | BPF_MEM | BPF_W))
2124 type = BPF_WRITE;
2125 else
2126 continue;
2127
2128 if (insn->imm != PTR_TO_CTX) {
2129 /* clear internal mark */
2130 insn->imm = 0;
2131 continue;
2132 }
2133
2134 cnt = env->prog->aux->ops->
2135 convert_ctx_access(type, insn->dst_reg, insn->src_reg,
2136 insn->off, insn_buf, env->prog);
2137 if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) {
2138 verbose("bpf verifier is misconfigured\n");
2139 return -EINVAL;
2140 }
2141
2142 if (cnt == 1) {
2143 memcpy(insn, insn_buf, sizeof(*insn));
2144 continue;
2145 }
2146
2147 /* several new insns need to be inserted. Make room for them */
2148 insn_cnt += cnt - 1;
2149 new_prog = bpf_prog_realloc(env->prog,
2150 bpf_prog_size(insn_cnt),
2151 GFP_USER);
2152 if (!new_prog)
2153 return -ENOMEM;
2154
2155 new_prog->len = insn_cnt;
2156
2157 memmove(new_prog->insnsi + i + cnt, new_prog->insns + i + 1,
2158 sizeof(*insn) * (insn_cnt - i - cnt));
2159
2160 /* copy substitute insns in place of load instruction */
2161 memcpy(new_prog->insnsi + i, insn_buf, sizeof(*insn) * cnt);
2162
2163 /* adjust branches in the whole program */
2164 adjust_branches(new_prog, i, cnt - 1);
2165
2166 /* keep walking new program and skip insns we just inserted */
2167 env->prog = new_prog;
2168 insn = new_prog->insnsi + i + cnt - 1;
2169 i += cnt - 1;
2170 }
2171
2172 return 0;
2173 }
2174
free_states(struct verifier_env * env)2175 static void free_states(struct verifier_env *env)
2176 {
2177 struct verifier_state_list *sl, *sln;
2178 int i;
2179
2180 if (!env->explored_states)
2181 return;
2182
2183 for (i = 0; i < env->prog->len; i++) {
2184 sl = env->explored_states[i];
2185
2186 if (sl)
2187 while (sl != STATE_LIST_MARK) {
2188 sln = sl->next;
2189 kfree(sl);
2190 sl = sln;
2191 }
2192 }
2193
2194 kfree(env->explored_states);
2195 }
2196
bpf_check(struct bpf_prog ** prog,union bpf_attr * attr)2197 int bpf_check(struct bpf_prog **prog, union bpf_attr *attr)
2198 {
2199 char __user *log_ubuf = NULL;
2200 struct verifier_env *env;
2201 int ret = -EINVAL;
2202
2203 if ((*prog)->len <= 0 || (*prog)->len > BPF_MAXINSNS)
2204 return -E2BIG;
2205
2206 /* 'struct verifier_env' can be global, but since it's not small,
2207 * allocate/free it every time bpf_check() is called
2208 */
2209 env = kzalloc(sizeof(struct verifier_env), GFP_KERNEL);
2210 if (!env)
2211 return -ENOMEM;
2212
2213 env->prog = *prog;
2214
2215 /* grab the mutex to protect few globals used by verifier */
2216 mutex_lock(&bpf_verifier_lock);
2217
2218 if (attr->log_level || attr->log_buf || attr->log_size) {
2219 /* user requested verbose verifier output
2220 * and supplied buffer to store the verification trace
2221 */
2222 log_level = attr->log_level;
2223 log_ubuf = (char __user *) (unsigned long) attr->log_buf;
2224 log_size = attr->log_size;
2225 log_len = 0;
2226
2227 ret = -EINVAL;
2228 /* log_* values have to be sane */
2229 if (log_size < 128 || log_size > UINT_MAX >> 8 ||
2230 log_level == 0 || log_ubuf == NULL)
2231 goto free_env;
2232
2233 ret = -ENOMEM;
2234 log_buf = vmalloc(log_size);
2235 if (!log_buf)
2236 goto free_env;
2237 } else {
2238 log_level = 0;
2239 }
2240
2241 ret = replace_map_fd_with_map_ptr(env);
2242 if (ret < 0)
2243 goto skip_full_check;
2244
2245 env->explored_states = kcalloc(env->prog->len,
2246 sizeof(struct verifier_state_list *),
2247 GFP_USER);
2248 ret = -ENOMEM;
2249 if (!env->explored_states)
2250 goto skip_full_check;
2251
2252 ret = check_cfg(env);
2253 if (ret < 0)
2254 goto skip_full_check;
2255
2256 env->allow_ptr_leaks = capable(CAP_SYS_ADMIN);
2257
2258 ret = do_check(env);
2259
2260 skip_full_check:
2261 while (pop_stack(env, NULL) >= 0);
2262 free_states(env);
2263
2264 if (ret == 0)
2265 /* program is valid, convert *(u32*)(ctx + off) accesses */
2266 ret = convert_ctx_accesses(env);
2267
2268 if (log_level && log_len >= log_size - 1) {
2269 BUG_ON(log_len >= log_size);
2270 /* verifier log exceeded user supplied buffer */
2271 ret = -ENOSPC;
2272 /* fall through to return what was recorded */
2273 }
2274
2275 /* copy verifier log back to user space including trailing zero */
2276 if (log_level && copy_to_user(log_ubuf, log_buf, log_len + 1) != 0) {
2277 ret = -EFAULT;
2278 goto free_log_buf;
2279 }
2280
2281 if (ret == 0 && env->used_map_cnt) {
2282 /* if program passed verifier, update used_maps in bpf_prog_info */
2283 env->prog->aux->used_maps = kmalloc_array(env->used_map_cnt,
2284 sizeof(env->used_maps[0]),
2285 GFP_KERNEL);
2286
2287 if (!env->prog->aux->used_maps) {
2288 ret = -ENOMEM;
2289 goto free_log_buf;
2290 }
2291
2292 memcpy(env->prog->aux->used_maps, env->used_maps,
2293 sizeof(env->used_maps[0]) * env->used_map_cnt);
2294 env->prog->aux->used_map_cnt = env->used_map_cnt;
2295
2296 /* program is valid. Convert pseudo bpf_ld_imm64 into generic
2297 * bpf_ld_imm64 instructions
2298 */
2299 convert_pseudo_ld_imm64(env);
2300 }
2301
2302 free_log_buf:
2303 if (log_level)
2304 vfree(log_buf);
2305 free_env:
2306 if (!env->prog->aux->used_maps)
2307 /* if we didn't copy map pointers into bpf_prog_info, release
2308 * them now. Otherwise free_bpf_prog_info() will release them.
2309 */
2310 release_maps(env);
2311 *prog = env->prog;
2312 kfree(env);
2313 mutex_unlock(&bpf_verifier_lock);
2314 return ret;
2315 }
2316