]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - kernel/bpf/verifier.c
projects / mirror_ubuntu-bionic-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge tag 'wireless-drivers-for-davem-2016-11-29' of git://git.kernel.org/pub/scm...
[mirror_ubuntu-bionic-kernel.git] / kernel / bpf / verifier.c
1 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
2 * Copyright (c) 2016 Facebook
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of version 2 of the GNU General Public
6 * License as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 */
13 #include <linux/kernel.h>
14 #include <linux/types.h>
15 #include <linux/slab.h>
16 #include <linux/bpf.h>
17 #include <linux/bpf_verifier.h>
18 #include <linux/filter.h>
19 #include <net/netlink.h>
20 #include <linux/file.h>
21 #include <linux/vmalloc.h>
22
23 /* bpf_check() is a static code analyzer that walks eBPF program
24 * instruction by instruction and updates register/stack state.
25 * All paths of conditional branches are analyzed until 'bpf_exit' insn.
26 *
27 * The first pass is depth-first-search to check that the program is a DAG.
28 * It rejects the following programs:
29 * - larger than BPF_MAXINSNS insns
30 * - if loop is present (detected via back-edge)
31 * - unreachable insns exist (shouldn't be a forest. program = one function)
32 * - out of bounds or malformed jumps
33 * The second pass is all possible path descent from the 1st insn.
34 * Since it's analyzing all pathes through the program, the length of the
35 * analysis is limited to 32k insn, which may be hit even if total number of
36 * insn is less then 4K, but there are too many branches that change stack/regs.
37 * Number of 'branches to be analyzed' is limited to 1k
38 *
39 * On entry to each instruction, each register has a type, and the instruction
40 * changes the types of the registers depending on instruction semantics.
41 * If instruction is BPF_MOV64_REG(BPF_REG_1, BPF_REG_5), then type of R5 is
42 * copied to R1.
43 *
44 * All registers are 64-bit.
45 * R0 - return register
46 * R1-R5 argument passing registers
47 * R6-R9 callee saved registers
48 * R10 - frame pointer read-only
49 *
50 * At the start of BPF program the register R1 contains a pointer to bpf_context
51 * and has type PTR_TO_CTX.
52 *
53 * Verifier tracks arithmetic operations on pointers in case:
54 * BPF_MOV64_REG(BPF_REG_1, BPF_REG_10),
55 * BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -20),
56 * 1st insn copies R10 (which has FRAME_PTR) type into R1
57 * and 2nd arithmetic instruction is pattern matched to recognize
58 * that it wants to construct a pointer to some element within stack.
59 * So after 2nd insn, the register R1 has type PTR_TO_STACK
60 * (and -20 constant is saved for further stack bounds checking).
61 * Meaning that this reg is a pointer to stack plus known immediate constant.
62 *
63 * Most of the time the registers have UNKNOWN_VALUE type, which
64 * means the register has some value, but it's not a valid pointer.
65 * (like pointer plus pointer becomes UNKNOWN_VALUE type)
66 *
67 * When verifier sees load or store instructions the type of base register
68 * can be: PTR_TO_MAP_VALUE, PTR_TO_CTX, FRAME_PTR. These are three pointer
69 * types recognized by check_mem_access() function.
70 *
71 * PTR_TO_MAP_VALUE means that this register is pointing to 'map element value'
72 * and the range of [ptr, ptr + map's value_size) is accessible.
73 *
74 * registers used to pass values to function calls are checked against
75 * function argument constraints.
76 *
77 * ARG_PTR_TO_MAP_KEY is one of such argument constraints.
78 * It means that the register type passed to this function must be
79 * PTR_TO_STACK and it will be used inside the function as
80 * 'pointer to map element key'
81 *
82 * For example the argument constraints for bpf_map_lookup_elem():
83 * .ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL,
84 * .arg1_type = ARG_CONST_MAP_PTR,
85 * .arg2_type = ARG_PTR_TO_MAP_KEY,
86 *
87 * ret_type says that this function returns 'pointer to map elem value or null'
88 * function expects 1st argument to be a const pointer to 'struct bpf_map' and
89 * 2nd argument should be a pointer to stack, which will be used inside
90 * the helper function as a pointer to map element key.
91 *
92 * On the kernel side the helper function looks like:
93 * u64 bpf_map_lookup_elem(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
94 * {
95 * struct bpf_map *map = (struct bpf_map *) (unsigned long) r1;
96 * void *key = (void *) (unsigned long) r2;
97 * void *value;
98 *
99 * here kernel can access 'key' and 'map' pointers safely, knowing that
100 * [key, key + map->key_size) bytes are valid and were initialized on
101 * the stack of eBPF program.
102 * }
103 *
104 * Corresponding eBPF program may look like:
105 * BPF_MOV64_REG(BPF_REG_2, BPF_REG_10), // after this insn R2 type is FRAME_PTR
106 * BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -4), // after this insn R2 type is PTR_TO_STACK
107 * BPF_LD_MAP_FD(BPF_REG_1, map_fd), // after this insn R1 type is CONST_PTR_TO_MAP
108 * BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem),
109 * here verifier looks at prototype of map_lookup_elem() and sees:
110 * .arg1_type == ARG_CONST_MAP_PTR and R1->type == CONST_PTR_TO_MAP, which is ok,
111 * Now verifier knows that this map has key of R1->map_ptr->key_size bytes
112 *
113 * Then .arg2_type == ARG_PTR_TO_MAP_KEY and R2->type == PTR_TO_STACK, ok so far,
114 * Now verifier checks that [R2, R2 + map's key_size) are within stack limits
115 * and were initialized prior to this call.
116 * If it's ok, then verifier allows this BPF_CALL insn and looks at
117 * .ret_type which is RET_PTR_TO_MAP_VALUE_OR_NULL, so it sets
118 * R0->type = PTR_TO_MAP_VALUE_OR_NULL which means bpf_map_lookup_elem() function
119 * returns ether pointer to map value or NULL.
120 *
121 * When type PTR_TO_MAP_VALUE_OR_NULL passes through 'if (reg != 0) goto +off'
122 * insn, the register holding that pointer in the true branch changes state to
123 * PTR_TO_MAP_VALUE and the same register changes state to CONST_IMM in the false
124 * branch. See check_cond_jmp_op().
125 *
126 * After the call R0 is set to return type of the function and registers R1-R5
127 * are set to NOT_INIT to indicate that they are no longer readable.
128 */
129
130 /* verifier_state + insn_idx are pushed to stack when branch is encountered */
131 struct bpf_verifier_stack_elem {
132 /* verifer state is 'st'
133 * before processing instruction 'insn_idx'
134 * and after processing instruction 'prev_insn_idx'
135 */
136 struct bpf_verifier_state st;
137 int insn_idx;
138 int prev_insn_idx;
139 struct bpf_verifier_stack_elem *next;
140 };
141
142 #define BPF_COMPLEXITY_LIMIT_INSNS 65536
143 #define BPF_COMPLEXITY_LIMIT_STACK 1024
144
145 struct bpf_call_arg_meta {
146 struct bpf_map *map_ptr;
147 bool raw_mode;
148 bool pkt_access;
149 int regno;
150 int access_size;
151 };
152
153 /* verbose verifier prints what it's seeing
154 * bpf_check() is called under lock, so no race to access these global vars
155 */
156 static u32 log_level, log_size, log_len;
157 static char *log_buf;
158
159 static DEFINE_MUTEX(bpf_verifier_lock);
160
161 /* log_level controls verbosity level of eBPF verifier.
162 * verbose() is used to dump the verification trace to the log, so the user
163 * can figure out what's wrong with the program
164 */
165 static __printf(1, 2) void verbose(const char *fmt, ...)
166 {
167 va_list args;
168
169 if (log_level == 0 || log_len >= log_size - 1)
170 return;
171
172 va_start(args, fmt);
173 log_len += vscnprintf(log_buf + log_len, log_size - log_len, fmt, args);
174 va_end(args);
175 }
176
177 /* string representation of 'enum bpf_reg_type' */
178 static const char * const reg_type_str[] = {
179 [NOT_INIT] = "?",
180 [UNKNOWN_VALUE] = "inv",
181 [PTR_TO_CTX] = "ctx",
182 [CONST_PTR_TO_MAP] = "map_ptr",
183 [PTR_TO_MAP_VALUE] = "map_value",
184 [PTR_TO_MAP_VALUE_OR_NULL] = "map_value_or_null",
185 [PTR_TO_MAP_VALUE_ADJ] = "map_value_adj",
186 [FRAME_PTR] = "fp",
187 [PTR_TO_STACK] = "fp",
188 [CONST_IMM] = "imm",
189 [PTR_TO_PACKET] = "pkt",
190 [PTR_TO_PACKET_END] = "pkt_end",
191 };
192
193 static void print_verifier_state(struct bpf_verifier_state *state)
194 {
195 struct bpf_reg_state *reg;
196 enum bpf_reg_type t;
197 int i;
198
199 for (i = 0; i < MAX_BPF_REG; i++) {
200 reg = &state->regs[i];
201 t = reg->type;
202 if (t == NOT_INIT)
203 continue;
204 verbose(" R%d=%s", i, reg_type_str[t]);
205 if (t == CONST_IMM || t == PTR_TO_STACK)
206 verbose("%lld", reg->imm);
207 else if (t == PTR_TO_PACKET)
208 verbose("(id=%d,off=%d,r=%d)",
209 reg->id, reg->off, reg->range);
210 else if (t == UNKNOWN_VALUE && reg->imm)
211 verbose("%lld", reg->imm);
212 else if (t == CONST_PTR_TO_MAP || t == PTR_TO_MAP_VALUE ||
213 t == PTR_TO_MAP_VALUE_OR_NULL ||
214 t == PTR_TO_MAP_VALUE_ADJ)
215 verbose("(ks=%d,vs=%d)",
216 reg->map_ptr->key_size,
217 reg->map_ptr->value_size);
218 if (reg->min_value != BPF_REGISTER_MIN_RANGE)
219 verbose(",min_value=%lld",
220 (long long)reg->min_value);
221 if (reg->max_value != BPF_REGISTER_MAX_RANGE)
222 verbose(",max_value=%llu",
223 (unsigned long long)reg->max_value);
224 }
225 for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
226 if (state->stack_slot_type[i] == STACK_SPILL)
227 verbose(" fp%d=%s", -MAX_BPF_STACK + i,
228 reg_type_str[state->spilled_regs[i / BPF_REG_SIZE].type]);
229 }
230 verbose("\n");
231 }
232
233 static const char *const bpf_class_string[] = {
234 [BPF_LD] = "ld",
235 [BPF_LDX] = "ldx",
236 [BPF_ST] = "st",
237 [BPF_STX] = "stx",
238 [BPF_ALU] = "alu",
239 [BPF_JMP] = "jmp",
240 [BPF_RET] = "BUG",
241 [BPF_ALU64] = "alu64",
242 };
243
244 static const char *const bpf_alu_string[16] = {
245 [BPF_ADD >> 4] = "+=",
246 [BPF_SUB >> 4] = "-=",
247 [BPF_MUL >> 4] = "*=",
248 [BPF_DIV >> 4] = "/=",
249 [BPF_OR >> 4] = "|=",
250 [BPF_AND >> 4] = "&=",
251 [BPF_LSH >> 4] = "<<=",
252 [BPF_RSH >> 4] = ">>=",
253 [BPF_NEG >> 4] = "neg",
254 [BPF_MOD >> 4] = "%=",
255 [BPF_XOR >> 4] = "^=",
256 [BPF_MOV >> 4] = "=",
257 [BPF_ARSH >> 4] = "s>>=",
258 [BPF_END >> 4] = "endian",
259 };
260
261 static const char *const bpf_ldst_string[] = {
262 [BPF_W >> 3] = "u32",
263 [BPF_H >> 3] = "u16",
264 [BPF_B >> 3] = "u8",
265 [BPF_DW >> 3] = "u64",
266 };
267
268 static const char *const bpf_jmp_string[16] = {
269 [BPF_JA >> 4] = "jmp",
270 [BPF_JEQ >> 4] = "==",
271 [BPF_JGT >> 4] = ">",
272 [BPF_JGE >> 4] = ">=",
273 [BPF_JSET >> 4] = "&",
274 [BPF_JNE >> 4] = "!=",
275 [BPF_JSGT >> 4] = "s>",
276 [BPF_JSGE >> 4] = "s>=",
277 [BPF_CALL >> 4] = "call",
278 [BPF_EXIT >> 4] = "exit",
279 };
280
281 static void print_bpf_insn(struct bpf_insn *insn)
282 {
283 u8 class = BPF_CLASS(insn->code);
284
285 if (class == BPF_ALU || class == BPF_ALU64) {
286 if (BPF_SRC(insn->code) == BPF_X)
287 verbose("(%02x) %sr%d %s %sr%d\n",
288 insn->code, class == BPF_ALU ? "(u32) " : "",
289 insn->dst_reg,
290 bpf_alu_string[BPF_OP(insn->code) >> 4],
291 class == BPF_ALU ? "(u32) " : "",
292 insn->src_reg);
293 else
294 verbose("(%02x) %sr%d %s %s%d\n",
295 insn->code, class == BPF_ALU ? "(u32) " : "",
296 insn->dst_reg,
297 bpf_alu_string[BPF_OP(insn->code) >> 4],
298 class == BPF_ALU ? "(u32) " : "",
299 insn->imm);
300 } else if (class == BPF_STX) {
301 if (BPF_MODE(insn->code) == BPF_MEM)
302 verbose("(%02x) *(%s *)(r%d %+d) = r%d\n",
303 insn->code,
304 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
305 insn->dst_reg,
306 insn->off, insn->src_reg);
307 else if (BPF_MODE(insn->code) == BPF_XADD)
308 verbose("(%02x) lock *(%s *)(r%d %+d) += r%d\n",
309 insn->code,
310 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
311 insn->dst_reg, insn->off,
312 insn->src_reg);
313 else
314 verbose("BUG_%02x\n", insn->code);
315 } else if (class == BPF_ST) {
316 if (BPF_MODE(insn->code) != BPF_MEM) {
317 verbose("BUG_st_%02x\n", insn->code);
318 return;
319 }
320 verbose("(%02x) *(%s *)(r%d %+d) = %d\n",
321 insn->code,
322 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
323 insn->dst_reg,
324 insn->off, insn->imm);
325 } else if (class == BPF_LDX) {
326 if (BPF_MODE(insn->code) != BPF_MEM) {
327 verbose("BUG_ldx_%02x\n", insn->code);
328 return;
329 }
330 verbose("(%02x) r%d = *(%s *)(r%d %+d)\n",
331 insn->code, insn->dst_reg,
332 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
333 insn->src_reg, insn->off);
334 } else if (class == BPF_LD) {
335 if (BPF_MODE(insn->code) == BPF_ABS) {
336 verbose("(%02x) r0 = *(%s *)skb[%d]\n",
337 insn->code,
338 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
339 insn->imm);
340 } else if (BPF_MODE(insn->code) == BPF_IND) {
341 verbose("(%02x) r0 = *(%s *)skb[r%d + %d]\n",
342 insn->code,
343 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
344 insn->src_reg, insn->imm);
345 } else if (BPF_MODE(insn->code) == BPF_IMM) {
346 verbose("(%02x) r%d = 0x%x\n",
347 insn->code, insn->dst_reg, insn->imm);
348 } else {
349 verbose("BUG_ld_%02x\n", insn->code);
350 return;
351 }
352 } else if (class == BPF_JMP) {
353 u8 opcode = BPF_OP(insn->code);
354
355 if (opcode == BPF_CALL) {
356 verbose("(%02x) call %d\n", insn->code, insn->imm);
357 } else if (insn->code == (BPF_JMP | BPF_JA)) {
358 verbose("(%02x) goto pc%+d\n",
359 insn->code, insn->off);
360 } else if (insn->code == (BPF_JMP | BPF_EXIT)) {
361 verbose("(%02x) exit\n", insn->code);
362 } else if (BPF_SRC(insn->code) == BPF_X) {
363 verbose("(%02x) if r%d %s r%d goto pc%+d\n",
364 insn->code, insn->dst_reg,
365 bpf_jmp_string[BPF_OP(insn->code) >> 4],
366 insn->src_reg, insn->off);
367 } else {
368 verbose("(%02x) if r%d %s 0x%x goto pc%+d\n",
369 insn->code, insn->dst_reg,
370 bpf_jmp_string[BPF_OP(insn->code) >> 4],
371 insn->imm, insn->off);
372 }
373 } else {
374 verbose("(%02x) %s\n", insn->code, bpf_class_string[class]);
375 }
376 }
377
378 static int pop_stack(struct bpf_verifier_env *env, int *prev_insn_idx)
379 {
380 struct bpf_verifier_stack_elem *elem;
381 int insn_idx;
382
383 if (env->head == NULL)
384 return -1;
385
386 memcpy(&env->cur_state, &env->head->st, sizeof(env->cur_state));
387 insn_idx = env->head->insn_idx;
388 if (prev_insn_idx)
389 *prev_insn_idx = env->head->prev_insn_idx;
390 elem = env->head->next;
391 kfree(env->head);
392 env->head = elem;
393 env->stack_size--;
394 return insn_idx;
395 }
396
397 static struct bpf_verifier_state *push_stack(struct bpf_verifier_env *env,
398 int insn_idx, int prev_insn_idx)
399 {
400 struct bpf_verifier_stack_elem *elem;
401
402 elem = kmalloc(sizeof(struct bpf_verifier_stack_elem), GFP_KERNEL);
403 if (!elem)
404 goto err;
405
406 memcpy(&elem->st, &env->cur_state, sizeof(env->cur_state));
407 elem->insn_idx = insn_idx;
408 elem->prev_insn_idx = prev_insn_idx;
409 elem->next = env->head;
410 env->head = elem;
411 env->stack_size++;
412 if (env->stack_size > BPF_COMPLEXITY_LIMIT_STACK) {
413 verbose("BPF program is too complex\n");
414 goto err;
415 }
416 return &elem->st;
417 err:
418 /* pop all elements and return */
419 while (pop_stack(env, NULL) >= 0);
420 return NULL;
421 }
422
423 #define CALLER_SAVED_REGS 6
424 static const int caller_saved[CALLER_SAVED_REGS] = {
425 BPF_REG_0, BPF_REG_1, BPF_REG_2, BPF_REG_3, BPF_REG_4, BPF_REG_5
426 };
427
428 static void init_reg_state(struct bpf_reg_state *regs)
429 {
430 int i;
431
432 for (i = 0; i < MAX_BPF_REG; i++) {
433 regs[i].type = NOT_INIT;
434 regs[i].imm = 0;
435 regs[i].min_value = BPF_REGISTER_MIN_RANGE;
436 regs[i].max_value = BPF_REGISTER_MAX_RANGE;
437 }
438
439 /* frame pointer */
440 regs[BPF_REG_FP].type = FRAME_PTR;
441
442 /* 1st arg to a function */
443 regs[BPF_REG_1].type = PTR_TO_CTX;
444 }
445
446 static void mark_reg_unknown_value(struct bpf_reg_state *regs, u32 regno)
447 {
448 BUG_ON(regno >= MAX_BPF_REG);
449 regs[regno].type = UNKNOWN_VALUE;
450 regs[regno].imm = 0;
451 }
452
453 static void reset_reg_range_values(struct bpf_reg_state *regs, u32 regno)
454 {
455 regs[regno].min_value = BPF_REGISTER_MIN_RANGE;
456 regs[regno].max_value = BPF_REGISTER_MAX_RANGE;
457 }
458
459 enum reg_arg_type {
460 SRC_OP, /* register is used as source operand */
461 DST_OP, /* register is used as destination operand */
462 DST_OP_NO_MARK /* same as above, check only, don't mark */
463 };
464
465 static int check_reg_arg(struct bpf_reg_state *regs, u32 regno,
466 enum reg_arg_type t)
467 {
468 if (regno >= MAX_BPF_REG) {
469 verbose("R%d is invalid\n", regno);
470 return -EINVAL;
471 }
472
473 if (t == SRC_OP) {
474 /* check whether register used as source operand can be read */
475 if (regs[regno].type == NOT_INIT) {
476 verbose("R%d !read_ok\n", regno);
477 return -EACCES;
478 }
479 } else {
480 /* check whether register used as dest operand can be written to */
481 if (regno == BPF_REG_FP) {
482 verbose("frame pointer is read only\n");
483 return -EACCES;
484 }
485 if (t == DST_OP)
486 mark_reg_unknown_value(regs, regno);
487 }
488 return 0;
489 }
490
491 static int bpf_size_to_bytes(int bpf_size)
492 {
493 if (bpf_size == BPF_W)
494 return 4;
495 else if (bpf_size == BPF_H)
496 return 2;
497 else if (bpf_size == BPF_B)
498 return 1;
499 else if (bpf_size == BPF_DW)
500 return 8;
501 else
502 return -EINVAL;
503 }
504
505 static bool is_spillable_regtype(enum bpf_reg_type type)
506 {
507 switch (type) {
508 case PTR_TO_MAP_VALUE:
509 case PTR_TO_MAP_VALUE_OR_NULL:
510 case PTR_TO_STACK:
511 case PTR_TO_CTX:
512 case PTR_TO_PACKET:
513 case PTR_TO_PACKET_END:
514 case FRAME_PTR:
515 case CONST_PTR_TO_MAP:
516 return true;
517 default:
518 return false;
519 }
520 }
521
522 /* check_stack_read/write functions track spill/fill of registers,
523 * stack boundary and alignment are checked in check_mem_access()
524 */
525 static int check_stack_write(struct bpf_verifier_state *state, int off,
526 int size, int value_regno)
527 {
528 int i;
529 /* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0,
530 * so it's aligned access and [off, off + size) are within stack limits
531 */
532
533 if (value_regno >= 0 &&
534 is_spillable_regtype(state->regs[value_regno].type)) {
535
536 /* register containing pointer is being spilled into stack */
537 if (size != BPF_REG_SIZE) {
538 verbose("invalid size of register spill\n");
539 return -EACCES;
540 }
541
542 /* save register state */
543 state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] =
544 state->regs[value_regno];
545
546 for (i = 0; i < BPF_REG_SIZE; i++)
547 state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_SPILL;
548 } else {
549 /* regular write of data into stack */
550 state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] =
551 (struct bpf_reg_state) {};
552
553 for (i = 0; i < size; i++)
554 state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_MISC;
555 }
556 return 0;
557 }
558
559 static int check_stack_read(struct bpf_verifier_state *state, int off, int size,
560 int value_regno)
561 {
562 u8 *slot_type;
563 int i;
564
565 slot_type = &state->stack_slot_type[MAX_BPF_STACK + off];
566
567 if (slot_type[0] == STACK_SPILL) {
568 if (size != BPF_REG_SIZE) {
569 verbose("invalid size of register spill\n");
570 return -EACCES;
571 }
572 for (i = 1; i < BPF_REG_SIZE; i++) {
573 if (slot_type[i] != STACK_SPILL) {
574 verbose("corrupted spill memory\n");
575 return -EACCES;
576 }
577 }
578
579 if (value_regno >= 0)
580 /* restore register state from stack */
581 state->regs[value_regno] =
582 state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE];
583 return 0;
584 } else {
585 for (i = 0; i < size; i++) {
586 if (slot_type[i] != STACK_MISC) {
587 verbose("invalid read from stack off %d+%d size %d\n",
588 off, i, size);
589 return -EACCES;
590 }
591 }
592 if (value_regno >= 0)
593 /* have read misc data from the stack */
594 mark_reg_unknown_value(state->regs, value_regno);
595 return 0;
596 }
597 }
598
599 /* check read/write into map element returned by bpf_map_lookup_elem() */
600 static int check_map_access(struct bpf_verifier_env *env, u32 regno, int off,
601 int size)
602 {
603 struct bpf_map *map = env->cur_state.regs[regno].map_ptr;
604
605 if (off < 0 || off + size > map->value_size) {
606 verbose("invalid access to map value, value_size=%d off=%d size=%d\n",
607 map->value_size, off, size);
608 return -EACCES;
609 }
610 return 0;
611 }
612
613 #define MAX_PACKET_OFF 0xffff
614
615 static bool may_access_direct_pkt_data(struct bpf_verifier_env *env,
616 const struct bpf_call_arg_meta *meta)
617 {
618 switch (env->prog->type) {
619 case BPF_PROG_TYPE_SCHED_CLS:
620 case BPF_PROG_TYPE_SCHED_ACT:
621 case BPF_PROG_TYPE_XDP:
622 if (meta)
623 return meta->pkt_access;
624
625 env->seen_direct_write = true;
626 return true;
627 default:
628 return false;
629 }
630 }
631
632 static int check_packet_access(struct bpf_verifier_env *env, u32 regno, int off,
633 int size)
634 {
635 struct bpf_reg_state *regs = env->cur_state.regs;
636 struct bpf_reg_state *reg = &regs[regno];
637
638 off += reg->off;
639 if (off < 0 || size <= 0 || off + size > reg->range) {
640 verbose("invalid access to packet, off=%d size=%d, R%d(id=%d,off=%d,r=%d)\n",
641 off, size, regno, reg->id, reg->off, reg->range);
642 return -EACCES;
643 }
644 return 0;
645 }
646
647 /* check access to 'struct bpf_context' fields */
648 static int check_ctx_access(struct bpf_verifier_env *env, int off, int size,
649 enum bpf_access_type t, enum bpf_reg_type *reg_type)
650 {
651 /* for analyzer ctx accesses are already validated and converted */
652 if (env->analyzer_ops)
653 return 0;
654
655 if (env->prog->aux->ops->is_valid_access &&
656 env->prog->aux->ops->is_valid_access(off, size, t, reg_type)) {
657 /* remember the offset of last byte accessed in ctx */
658 if (env->prog->aux->max_ctx_offset < off + size)
659 env->prog->aux->max_ctx_offset = off + size;
660 return 0;
661 }
662
663 verbose("invalid bpf_context access off=%d size=%d\n", off, size);
664 return -EACCES;
665 }
666
667 static bool is_pointer_value(struct bpf_verifier_env *env, int regno)
668 {
669 if (env->allow_ptr_leaks)
670 return false;
671
672 switch (env->cur_state.regs[regno].type) {
673 case UNKNOWN_VALUE:
674 case CONST_IMM:
675 return false;
676 default:
677 return true;
678 }
679 }
680
681 static int check_ptr_alignment(struct bpf_verifier_env *env,
682 struct bpf_reg_state *reg, int off, int size)
683 {
684 if (reg->type != PTR_TO_PACKET && reg->type != PTR_TO_MAP_VALUE_ADJ) {
685 if (off % size != 0) {
686 verbose("misaligned access off %d size %d\n",
687 off, size);
688 return -EACCES;
689 } else {
690 return 0;
691 }
692 }
693
694 if (IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS))
695 /* misaligned access to packet is ok on x86,arm,arm64 */
696 return 0;
697
698 if (reg->id && size != 1) {
699 verbose("Unknown packet alignment. Only byte-sized access allowed\n");
700 return -EACCES;
701 }
702
703 /* skb->data is NET_IP_ALIGN-ed */
704 if (reg->type == PTR_TO_PACKET &&
705 (NET_IP_ALIGN + reg->off + off) % size != 0) {
706 verbose("misaligned packet access off %d+%d+%d size %d\n",
707 NET_IP_ALIGN, reg->off, off, size);
708 return -EACCES;
709 }
710 return 0;
711 }
712
713 /* check whether memory at (regno + off) is accessible for t = (read | write)
714 * if t==write, value_regno is a register which value is stored into memory
715 * if t==read, value_regno is a register which will receive the value from memory
716 * if t==write && value_regno==-1, some unknown value is stored into memory
717 * if t==read && value_regno==-1, don't care what we read from memory
718 */
719 static int check_mem_access(struct bpf_verifier_env *env, u32 regno, int off,
720 int bpf_size, enum bpf_access_type t,
721 int value_regno)
722 {
723 struct bpf_verifier_state *state = &env->cur_state;
724 struct bpf_reg_state *reg = &state->regs[regno];
725 int size, err = 0;
726
727 if (reg->type == PTR_TO_STACK)
728 off += reg->imm;
729
730 size = bpf_size_to_bytes(bpf_size);
731 if (size < 0)
732 return size;
733
734 err = check_ptr_alignment(env, reg, off, size);
735 if (err)
736 return err;
737
738 if (reg->type == PTR_TO_MAP_VALUE ||
739 reg->type == PTR_TO_MAP_VALUE_ADJ) {
740 if (t == BPF_WRITE && value_regno >= 0 &&
741 is_pointer_value(env, value_regno)) {
742 verbose("R%d leaks addr into map\n", value_regno);
743 return -EACCES;
744 }
745
746 /* If we adjusted the register to this map value at all then we
747 * need to change off and size to min_value and max_value
748 * respectively to make sure our theoretical access will be
749 * safe.
750 */
751 if (reg->type == PTR_TO_MAP_VALUE_ADJ) {
752 if (log_level)
753 print_verifier_state(state);
754 env->varlen_map_value_access = true;
755 /* The minimum value is only important with signed
756 * comparisons where we can't assume the floor of a
757 * value is 0. If we are using signed variables for our
758 * index'es we need to make sure that whatever we use
759 * will have a set floor within our range.
760 */
761 if (reg->min_value < 0) {
762 verbose("R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n",
763 regno);
764 return -EACCES;
765 }
766 err = check_map_access(env, regno, reg->min_value + off,
767 size);
768 if (err) {
769 verbose("R%d min value is outside of the array range\n",
770 regno);
771 return err;
772 }
773
774 /* If we haven't set a max value then we need to bail
775 * since we can't be sure we won't do bad things.
776 */
777 if (reg->max_value == BPF_REGISTER_MAX_RANGE) {
778 verbose("R%d unbounded memory access, make sure to bounds check any array access into a map\n",
779 regno);
780 return -EACCES;
781 }
782 off += reg->max_value;
783 }
784 err = check_map_access(env, regno, off, size);
785 if (!err && t == BPF_READ && value_regno >= 0)
786 mark_reg_unknown_value(state->regs, value_regno);
787
788 } else if (reg->type == PTR_TO_CTX) {
789 enum bpf_reg_type reg_type = UNKNOWN_VALUE;
790
791 if (t == BPF_WRITE && value_regno >= 0 &&
792 is_pointer_value(env, value_regno)) {
793 verbose("R%d leaks addr into ctx\n", value_regno);
794 return -EACCES;
795 }
796 err = check_ctx_access(env, off, size, t, &reg_type);
797 if (!err && t == BPF_READ && value_regno >= 0) {
798 mark_reg_unknown_value(state->regs, value_regno);
799 /* note that reg.[id|off|range] == 0 */
800 state->regs[value_regno].type = reg_type;
801 }
802
803 } else if (reg->type == FRAME_PTR || reg->type == PTR_TO_STACK) {
804 if (off >= 0 || off < -MAX_BPF_STACK) {
805 verbose("invalid stack off=%d size=%d\n", off, size);
806 return -EACCES;
807 }
808 if (t == BPF_WRITE) {
809 if (!env->allow_ptr_leaks &&
810 state->stack_slot_type[MAX_BPF_STACK + off] == STACK_SPILL &&
811 size != BPF_REG_SIZE) {
812 verbose("attempt to corrupt spilled pointer on stack\n");
813 return -EACCES;
814 }
815 err = check_stack_write(state, off, size, value_regno);
816 } else {
817 err = check_stack_read(state, off, size, value_regno);
818 }
819 } else if (state->regs[regno].type == PTR_TO_PACKET) {
820 if (t == BPF_WRITE && !may_access_direct_pkt_data(env, NULL)) {
821 verbose("cannot write into packet\n");
822 return -EACCES;
823 }
824 if (t == BPF_WRITE && value_regno >= 0 &&
825 is_pointer_value(env, value_regno)) {
826 verbose("R%d leaks addr into packet\n", value_regno);
827 return -EACCES;
828 }
829 err = check_packet_access(env, regno, off, size);
830 if (!err && t == BPF_READ && value_regno >= 0)
831 mark_reg_unknown_value(state->regs, value_regno);
832 } else {
833 verbose("R%d invalid mem access '%s'\n",
834 regno, reg_type_str[reg->type]);
835 return -EACCES;
836 }
837
838 if (!err && size <= 2 && value_regno >= 0 && env->allow_ptr_leaks &&
839 state->regs[value_regno].type == UNKNOWN_VALUE) {
840 /* 1 or 2 byte load zero-extends, determine the number of
841 * zero upper bits. Not doing it fo 4 byte load, since
842 * such values cannot be added to ptr_to_packet anyway.
843 */
844 state->regs[value_regno].imm = 64 - size * 8;
845 }
846 return err;
847 }
848
849 static int check_xadd(struct bpf_verifier_env *env, struct bpf_insn *insn)
850 {
851 struct bpf_reg_state *regs = env->cur_state.regs;
852 int err;
853
854 if ((BPF_SIZE(insn->code) != BPF_W && BPF_SIZE(insn->code) != BPF_DW) ||
855 insn->imm != 0) {
856 verbose("BPF_XADD uses reserved fields\n");
857 return -EINVAL;
858 }
859
860 /* check src1 operand */
861 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
862 if (err)
863 return err;
864
865 /* check src2 operand */
866 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
867 if (err)
868 return err;
869
870 /* check whether atomic_add can read the memory */
871 err = check_mem_access(env, insn->dst_reg, insn->off,
872 BPF_SIZE(insn->code), BPF_READ, -1);
873 if (err)
874 return err;
875
876 /* check whether atomic_add can write into the same memory */
877 return check_mem_access(env, insn->dst_reg, insn->off,
878 BPF_SIZE(insn->code), BPF_WRITE, -1);
879 }
880
881 /* when register 'regno' is passed into function that will read 'access_size'
882 * bytes from that pointer, make sure that it's within stack boundary
883 * and all elements of stack are initialized
884 */
885 static int check_stack_boundary(struct bpf_verifier_env *env, int regno,
886 int access_size, bool zero_size_allowed,
887 struct bpf_call_arg_meta *meta)
888 {
889 struct bpf_verifier_state *state = &env->cur_state;
890 struct bpf_reg_state *regs = state->regs;
891 int off, i;
892
893 if (regs[regno].type != PTR_TO_STACK) {
894 if (zero_size_allowed && access_size == 0 &&
895 regs[regno].type == CONST_IMM &&
896 regs[regno].imm == 0)
897 return 0;
898
899 verbose("R%d type=%s expected=%s\n", regno,
900 reg_type_str[regs[regno].type],
901 reg_type_str[PTR_TO_STACK]);
902 return -EACCES;
903 }
904
905 off = regs[regno].imm;
906 if (off >= 0 || off < -MAX_BPF_STACK || off + access_size > 0 ||
907 access_size <= 0) {
908 verbose("invalid stack type R%d off=%d access_size=%d\n",
909 regno, off, access_size);
910 return -EACCES;
911 }
912
913 if (meta && meta->raw_mode) {
914 meta->access_size = access_size;
915 meta->regno = regno;
916 return 0;
917 }
918
919 for (i = 0; i < access_size; i++) {
920 if (state->stack_slot_type[MAX_BPF_STACK + off + i] != STACK_MISC) {
921 verbose("invalid indirect read from stack off %d+%d size %d\n",
922 off, i, access_size);
923 return -EACCES;
924 }
925 }
926 return 0;
927 }
928
929 static int check_func_arg(struct bpf_verifier_env *env, u32 regno,
930 enum bpf_arg_type arg_type,
931 struct bpf_call_arg_meta *meta)
932 {
933 struct bpf_reg_state *regs = env->cur_state.regs, *reg = &regs[regno];
934 enum bpf_reg_type expected_type, type = reg->type;
935 int err = 0;
936
937 if (arg_type == ARG_DONTCARE)
938 return 0;
939
940 if (type == NOT_INIT) {
941 verbose("R%d !read_ok\n", regno);
942 return -EACCES;
943 }
944
945 if (arg_type == ARG_ANYTHING) {
946 if (is_pointer_value(env, regno)) {
947 verbose("R%d leaks addr into helper function\n", regno);
948 return -EACCES;
949 }
950 return 0;
951 }
952
953 if (type == PTR_TO_PACKET && !may_access_direct_pkt_data(env, meta)) {
954 verbose("helper access to the packet is not allowed\n");
955 return -EACCES;
956 }
957
958 if (arg_type == ARG_PTR_TO_MAP_KEY ||
959 arg_type == ARG_PTR_TO_MAP_VALUE) {
960 expected_type = PTR_TO_STACK;
961 if (type != PTR_TO_PACKET && type != expected_type)
962 goto err_type;
963 } else if (arg_type == ARG_CONST_STACK_SIZE ||
964 arg_type == ARG_CONST_STACK_SIZE_OR_ZERO) {
965 expected_type = CONST_IMM;
966 if (type != expected_type)
967 goto err_type;
968 } else if (arg_type == ARG_CONST_MAP_PTR) {
969 expected_type = CONST_PTR_TO_MAP;
970 if (type != expected_type)
971 goto err_type;
972 } else if (arg_type == ARG_PTR_TO_CTX) {
973 expected_type = PTR_TO_CTX;
974 if (type != expected_type)
975 goto err_type;
976 } else if (arg_type == ARG_PTR_TO_STACK ||
977 arg_type == ARG_PTR_TO_RAW_STACK) {
978 expected_type = PTR_TO_STACK;
979 /* One exception here. In case function allows for NULL to be
980 * passed in as argument, it's a CONST_IMM type. Final test
981 * happens during stack boundary checking.
982 */
983 if (type == CONST_IMM && reg->imm == 0)
984 /* final test in check_stack_boundary() */;
985 else if (type != PTR_TO_PACKET && type != expected_type)
986 goto err_type;
987 meta->raw_mode = arg_type == ARG_PTR_TO_RAW_STACK;
988 } else {
989 verbose("unsupported arg_type %d\n", arg_type);
990 return -EFAULT;
991 }
992
993 if (arg_type == ARG_CONST_MAP_PTR) {
994 /* bpf_map_xxx(map_ptr) call: remember that map_ptr */
995 meta->map_ptr = reg->map_ptr;
996 } else if (arg_type == ARG_PTR_TO_MAP_KEY) {
997 /* bpf_map_xxx(..., map_ptr, ..., key) call:
998 * check that [key, key + map->key_size) are within
999 * stack limits and initialized
1000 */
1001 if (!meta->map_ptr) {
1002 /* in function declaration map_ptr must come before
1003 * map_key, so that it's verified and known before
1004 * we have to check map_key here. Otherwise it means
1005 * that kernel subsystem misconfigured verifier
1006 */
1007 verbose("invalid map_ptr to access map->key\n");
1008 return -EACCES;
1009 }
1010 if (type == PTR_TO_PACKET)
1011 err = check_packet_access(env, regno, 0,
1012 meta->map_ptr->key_size);
1013 else
1014 err = check_stack_boundary(env, regno,
1015 meta->map_ptr->key_size,
1016 false, NULL);
1017 } else if (arg_type == ARG_PTR_TO_MAP_VALUE) {
1018 /* bpf_map_xxx(..., map_ptr, ..., value) call:
1019 * check [value, value + map->value_size) validity
1020 */
1021 if (!meta->map_ptr) {
1022 /* kernel subsystem misconfigured verifier */
1023 verbose("invalid map_ptr to access map->value\n");
1024 return -EACCES;
1025 }
1026 if (type == PTR_TO_PACKET)
1027 err = check_packet_access(env, regno, 0,
1028 meta->map_ptr->value_size);
1029 else
1030 err = check_stack_boundary(env, regno,
1031 meta->map_ptr->value_size,
1032 false, NULL);
1033 } else if (arg_type == ARG_CONST_STACK_SIZE ||
1034 arg_type == ARG_CONST_STACK_SIZE_OR_ZERO) {
1035 bool zero_size_allowed = (arg_type == ARG_CONST_STACK_SIZE_OR_ZERO);
1036
1037 /* bpf_xxx(..., buf, len) call will access 'len' bytes
1038 * from stack pointer 'buf'. Check it
1039 * note: regno == len, regno - 1 == buf
1040 */
1041 if (regno == 0) {
1042 /* kernel subsystem misconfigured verifier */
1043 verbose("ARG_CONST_STACK_SIZE cannot be first argument\n");
1044 return -EACCES;
1045 }
1046 if (regs[regno - 1].type == PTR_TO_PACKET)
1047 err = check_packet_access(env, regno - 1, 0, reg->imm);
1048 else
1049 err = check_stack_boundary(env, regno - 1, reg->imm,
1050 zero_size_allowed, meta);
1051 }
1052
1053 return err;
1054 err_type:
1055 verbose("R%d type=%s expected=%s\n", regno,
1056 reg_type_str[type], reg_type_str[expected_type]);
1057 return -EACCES;
1058 }
1059
1060 static int check_map_func_compatibility(struct bpf_map *map, int func_id)
1061 {
1062 if (!map)
1063 return 0;
1064
1065 /* We need a two way check, first is from map perspective ... */
1066 switch (map->map_type) {
1067 case BPF_MAP_TYPE_PROG_ARRAY:
1068 if (func_id != BPF_FUNC_tail_call)
1069 goto error;
1070 break;
1071 case BPF_MAP_TYPE_PERF_EVENT_ARRAY:
1072 if (func_id != BPF_FUNC_perf_event_read &&
1073 func_id != BPF_FUNC_perf_event_output)
1074 goto error;
1075 break;
1076 case BPF_MAP_TYPE_STACK_TRACE:
1077 if (func_id != BPF_FUNC_get_stackid)
1078 goto error;
1079 break;
1080 case BPF_MAP_TYPE_CGROUP_ARRAY:
1081 if (func_id != BPF_FUNC_skb_under_cgroup &&
1082 func_id != BPF_FUNC_current_task_under_cgroup)
1083 goto error;
1084 break;
1085 default:
1086 break;
1087 }
1088
1089 /* ... and second from the function itself. */
1090 switch (func_id) {
1091 case BPF_FUNC_tail_call:
1092 if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY)
1093 goto error;
1094 break;
1095 case BPF_FUNC_perf_event_read:
1096 case BPF_FUNC_perf_event_output:
1097 if (map->map_type != BPF_MAP_TYPE_PERF_EVENT_ARRAY)
1098 goto error;
1099 break;
1100 case BPF_FUNC_get_stackid:
1101 if (map->map_type != BPF_MAP_TYPE_STACK_TRACE)
1102 goto error;
1103 break;
1104 case BPF_FUNC_current_task_under_cgroup:
1105 case BPF_FUNC_skb_under_cgroup:
1106 if (map->map_type != BPF_MAP_TYPE_CGROUP_ARRAY)
1107 goto error;
1108 break;
1109 default:
1110 break;
1111 }
1112
1113 return 0;
1114 error:
1115 verbose("cannot pass map_type %d into func %d\n",
1116 map->map_type, func_id);
1117 return -EINVAL;
1118 }
1119
1120 static int check_raw_mode(const struct bpf_func_proto *fn)
1121 {
1122 int count = 0;
1123
1124 if (fn->arg1_type == ARG_PTR_TO_RAW_STACK)
1125 count++;
1126 if (fn->arg2_type == ARG_PTR_TO_RAW_STACK)
1127 count++;
1128 if (fn->arg3_type == ARG_PTR_TO_RAW_STACK)
1129 count++;
1130 if (fn->arg4_type == ARG_PTR_TO_RAW_STACK)
1131 count++;
1132 if (fn->arg5_type == ARG_PTR_TO_RAW_STACK)
1133 count++;
1134
1135 return count > 1 ? -EINVAL : 0;
1136 }
1137
1138 static void clear_all_pkt_pointers(struct bpf_verifier_env *env)
1139 {
1140 struct bpf_verifier_state *state = &env->cur_state;
1141 struct bpf_reg_state *regs = state->regs, *reg;
1142 int i;
1143
1144 for (i = 0; i < MAX_BPF_REG; i++)
1145 if (regs[i].type == PTR_TO_PACKET ||
1146 regs[i].type == PTR_TO_PACKET_END)
1147 mark_reg_unknown_value(regs, i);
1148
1149 for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
1150 if (state->stack_slot_type[i] != STACK_SPILL)
1151 continue;
1152 reg = &state->spilled_regs[i / BPF_REG_SIZE];
1153 if (reg->type != PTR_TO_PACKET &&
1154 reg->type != PTR_TO_PACKET_END)
1155 continue;
1156 reg->type = UNKNOWN_VALUE;
1157 reg->imm = 0;
1158 }
1159 }
1160
1161 static int check_call(struct bpf_verifier_env *env, int func_id)
1162 {
1163 struct bpf_verifier_state *state = &env->cur_state;
1164 const struct bpf_func_proto *fn = NULL;
1165 struct bpf_reg_state *regs = state->regs;
1166 struct bpf_reg_state *reg;
1167 struct bpf_call_arg_meta meta;
1168 bool changes_data;
1169 int i, err;
1170
1171 /* find function prototype */
1172 if (func_id < 0 || func_id >= __BPF_FUNC_MAX_ID) {
1173 verbose("invalid func %d\n", func_id);
1174 return -EINVAL;
1175 }
1176
1177 if (env->prog->aux->ops->get_func_proto)
1178 fn = env->prog->aux->ops->get_func_proto(func_id);
1179
1180 if (!fn) {
1181 verbose("unknown func %d\n", func_id);
1182 return -EINVAL;
1183 }
1184
1185 /* eBPF programs must be GPL compatible to use GPL-ed functions */
1186 if (!env->prog->gpl_compatible && fn->gpl_only) {
1187 verbose("cannot call GPL only function from proprietary program\n");
1188 return -EINVAL;
1189 }
1190
1191 changes_data = bpf_helper_changes_skb_data(fn->func);
1192
1193 memset(&meta, 0, sizeof(meta));
1194 meta.pkt_access = fn->pkt_access;
1195
1196 /* We only support one arg being in raw mode at the moment, which
1197 * is sufficient for the helper functions we have right now.
1198 */
1199 err = check_raw_mode(fn);
1200 if (err) {
1201 verbose("kernel subsystem misconfigured func %d\n", func_id);
1202 return err;
1203 }
1204
1205 /* check args */
1206 err = check_func_arg(env, BPF_REG_1, fn->arg1_type, &meta);
1207 if (err)
1208 return err;
1209 err = check_func_arg(env, BPF_REG_2, fn->arg2_type, &meta);
1210 if (err)
1211 return err;
1212 err = check_func_arg(env, BPF_REG_3, fn->arg3_type, &meta);
1213 if (err)
1214 return err;
1215 err = check_func_arg(env, BPF_REG_4, fn->arg4_type, &meta);
1216 if (err)
1217 return err;
1218 err = check_func_arg(env, BPF_REG_5, fn->arg5_type, &meta);
1219 if (err)
1220 return err;
1221
1222 /* Mark slots with STACK_MISC in case of raw mode, stack offset
1223 * is inferred from register state.
1224 */
1225 for (i = 0; i < meta.access_size; i++) {
1226 err = check_mem_access(env, meta.regno, i, BPF_B, BPF_WRITE, -1);
1227 if (err)
1228 return err;
1229 }
1230
1231 /* reset caller saved regs */
1232 for (i = 0; i < CALLER_SAVED_REGS; i++) {
1233 reg = regs + caller_saved[i];
1234 reg->type = NOT_INIT;
1235 reg->imm = 0;
1236 }
1237
1238 /* update return register */
1239 if (fn->ret_type == RET_INTEGER) {
1240 regs[BPF_REG_0].type = UNKNOWN_VALUE;
1241 } else if (fn->ret_type == RET_VOID) {
1242 regs[BPF_REG_0].type = NOT_INIT;
1243 } else if (fn->ret_type == RET_PTR_TO_MAP_VALUE_OR_NULL) {
1244 regs[BPF_REG_0].type = PTR_TO_MAP_VALUE_OR_NULL;
1245 regs[BPF_REG_0].max_value = regs[BPF_REG_0].min_value = 0;
1246 /* remember map_ptr, so that check_map_access()
1247 * can check 'value_size' boundary of memory access
1248 * to map element returned from bpf_map_lookup_elem()
1249 */
1250 if (meta.map_ptr == NULL) {
1251 verbose("kernel subsystem misconfigured verifier\n");
1252 return -EINVAL;
1253 }
1254 regs[BPF_REG_0].map_ptr = meta.map_ptr;
1255 } else {
1256 verbose("unknown return type %d of func %d\n",
1257 fn->ret_type, func_id);
1258 return -EINVAL;
1259 }
1260
1261 err = check_map_func_compatibility(meta.map_ptr, func_id);
1262 if (err)
1263 return err;
1264
1265 if (changes_data)
1266 clear_all_pkt_pointers(env);
1267 return 0;
1268 }
1269
1270 static int check_packet_ptr_add(struct bpf_verifier_env *env,
1271 struct bpf_insn *insn)
1272 {
1273 struct bpf_reg_state *regs = env->cur_state.regs;
1274 struct bpf_reg_state *dst_reg = &regs[insn->dst_reg];
1275 struct bpf_reg_state *src_reg = &regs[insn->src_reg];
1276 struct bpf_reg_state tmp_reg;
1277 s32 imm;
1278
1279 if (BPF_SRC(insn->code) == BPF_K) {
1280 /* pkt_ptr += imm */
1281 imm = insn->imm;
1282
1283 add_imm:
1284 if (imm <= 0) {
1285 verbose("addition of negative constant to packet pointer is not allowed\n");
1286 return -EACCES;
1287 }
1288 if (imm >= MAX_PACKET_OFF ||
1289 imm + dst_reg->off >= MAX_PACKET_OFF) {
1290 verbose("constant %d is too large to add to packet pointer\n",
1291 imm);
1292 return -EACCES;
1293 }
1294 /* a constant was added to pkt_ptr.
1295 * Remember it while keeping the same 'id'
1296 */
1297 dst_reg->off += imm;
1298 } else {
1299 if (src_reg->type == PTR_TO_PACKET) {
1300 /* R6=pkt(id=0,off=0,r=62) R7=imm22; r7 += r6 */
1301 tmp_reg = *dst_reg; /* save r7 state */
1302 *dst_reg = *src_reg; /* copy pkt_ptr state r6 into r7 */
1303 src_reg = &tmp_reg; /* pretend it's src_reg state */
1304 /* if the checks below reject it, the copy won't matter,
1305 * since we're rejecting the whole program. If all ok,
1306 * then imm22 state will be added to r7
1307 * and r7 will be pkt(id=0,off=22,r=62) while
1308 * r6 will stay as pkt(id=0,off=0,r=62)
1309 */
1310 }
1311
1312 if (src_reg->type == CONST_IMM) {
1313 /* pkt_ptr += reg where reg is known constant */
1314 imm = src_reg->imm;
1315 goto add_imm;
1316 }
1317 /* disallow pkt_ptr += reg
1318 * if reg is not uknown_value with guaranteed zero upper bits
1319 * otherwise pkt_ptr may overflow and addition will become
1320 * subtraction which is not allowed
1321 */
1322 if (src_reg->type != UNKNOWN_VALUE) {
1323 verbose("cannot add '%s' to ptr_to_packet\n",
1324 reg_type_str[src_reg->type]);
1325 return -EACCES;
1326 }
1327 if (src_reg->imm < 48) {
1328 verbose("cannot add integer value with %lld upper zero bits to ptr_to_packet\n",
1329 src_reg->imm);
1330 return -EACCES;
1331 }
1332 /* dst_reg stays as pkt_ptr type and since some positive
1333 * integer value was added to the pointer, increment its 'id'
1334 */
1335 dst_reg->id = ++env->id_gen;
1336
1337 /* something was added to pkt_ptr, set range and off to zero */
1338 dst_reg->off = 0;
1339 dst_reg->range = 0;
1340 }
1341 return 0;
1342 }
1343
1344 static int evaluate_reg_alu(struct bpf_verifier_env *env, struct bpf_insn *insn)
1345 {
1346 struct bpf_reg_state *regs = env->cur_state.regs;
1347 struct bpf_reg_state *dst_reg = &regs[insn->dst_reg];
1348 u8 opcode = BPF_OP(insn->code);
1349 s64 imm_log2;
1350
1351 /* for type == UNKNOWN_VALUE:
1352 * imm > 0 -> number of zero upper bits
1353 * imm == 0 -> don't track which is the same as all bits can be non-zero
1354 */
1355
1356 if (BPF_SRC(insn->code) == BPF_X) {
1357 struct bpf_reg_state *src_reg = &regs[insn->src_reg];
1358
1359 if (src_reg->type == UNKNOWN_VALUE && src_reg->imm > 0 &&
1360 dst_reg->imm && opcode == BPF_ADD) {
1361 /* dreg += sreg
1362 * where both have zero upper bits. Adding them
1363 * can only result making one more bit non-zero
1364 * in the larger value.
1365 * Ex. 0xffff (imm=48) + 1 (imm=63) = 0x10000 (imm=47)
1366 * 0xffff (imm=48) + 0xffff = 0x1fffe (imm=47)
1367 */
1368 dst_reg->imm = min(dst_reg->imm, src_reg->imm);
1369 dst_reg->imm--;
1370 return 0;
1371 }
1372 if (src_reg->type == CONST_IMM && src_reg->imm > 0 &&
1373 dst_reg->imm && opcode == BPF_ADD) {
1374 /* dreg += sreg
1375 * where dreg has zero upper bits and sreg is const.
1376 * Adding them can only result making one more bit
1377 * non-zero in the larger value.
1378 */
1379 imm_log2 = __ilog2_u64((long long)src_reg->imm);
1380 dst_reg->imm = min(dst_reg->imm, 63 - imm_log2);
1381 dst_reg->imm--;
1382 return 0;
1383 }
1384 /* all other cases non supported yet, just mark dst_reg */
1385 dst_reg->imm = 0;
1386 return 0;
1387 }
1388
1389 /* sign extend 32-bit imm into 64-bit to make sure that
1390 * negative values occupy bit 63. Note ilog2() would have
1391 * been incorrect, since sizeof(insn->imm) == 4
1392 */
1393 imm_log2 = __ilog2_u64((long long)insn->imm);
1394
1395 if (dst_reg->imm && opcode == BPF_LSH) {
1396 /* reg <<= imm
1397 * if reg was a result of 2 byte load, then its imm == 48
1398 * which means that upper 48 bits are zero and shifting this reg
1399 * left by 4 would mean that upper 44 bits are still zero
1400 */
1401 dst_reg->imm -= insn->imm;
1402 } else if (dst_reg->imm && opcode == BPF_MUL) {
1403 /* reg *= imm
1404 * if multiplying by 14 subtract 4
1405 * This is conservative calculation of upper zero bits.
1406 * It's not trying to special case insn->imm == 1 or 0 cases
1407 */
1408 dst_reg->imm -= imm_log2 + 1;
1409 } else if (opcode == BPF_AND) {
1410 /* reg &= imm */
1411 dst_reg->imm = 63 - imm_log2;
1412 } else if (dst_reg->imm && opcode == BPF_ADD) {
1413 /* reg += imm */
1414 dst_reg->imm = min(dst_reg->imm, 63 - imm_log2);
1415 dst_reg->imm--;
1416 } else if (opcode == BPF_RSH) {
1417 /* reg >>= imm
1418 * which means that after right shift, upper bits will be zero
1419 * note that verifier already checked that
1420 * 0 <= imm < 64 for shift insn
1421 */
1422 dst_reg->imm += insn->imm;
1423 if (unlikely(dst_reg->imm > 64))
1424 /* some dumb code did:
1425 * r2 = *(u32 *)mem;
1426 * r2 >>= 32;
1427 * and all bits are zero now */
1428 dst_reg->imm = 64;
1429 } else {
1430 /* all other alu ops, means that we don't know what will
1431 * happen to the value, mark it with unknown number of zero bits
1432 */
1433 dst_reg->imm = 0;
1434 }
1435
1436 if (dst_reg->imm < 0) {
1437 /* all 64 bits of the register can contain non-zero bits
1438 * and such value cannot be added to ptr_to_packet, since it
1439 * may overflow, mark it as unknown to avoid further eval
1440 */
1441 dst_reg->imm = 0;
1442 }
1443 return 0;
1444 }
1445
1446 static int evaluate_reg_imm_alu(struct bpf_verifier_env *env,
1447 struct bpf_insn *insn)
1448 {
1449 struct bpf_reg_state *regs = env->cur_state.regs;
1450 struct bpf_reg_state *dst_reg = &regs[insn->dst_reg];
1451 struct bpf_reg_state *src_reg = &regs[insn->src_reg];
1452 u8 opcode = BPF_OP(insn->code);
1453
1454 /* dst_reg->type == CONST_IMM here, simulate execution of 'add' insn.
1455 * Don't care about overflow or negative values, just add them
1456 */
1457 if (opcode == BPF_ADD && BPF_SRC(insn->code) == BPF_K)
1458 dst_reg->imm += insn->imm;
1459 else if (opcode == BPF_ADD && BPF_SRC(insn->code) == BPF_X &&
1460 src_reg->type == CONST_IMM)
1461 dst_reg->imm += src_reg->imm;
1462 else
1463 mark_reg_unknown_value(regs, insn->dst_reg);
1464 return 0;
1465 }
1466
1467 static void check_reg_overflow(struct bpf_reg_state *reg)
1468 {
1469 if (reg->max_value > BPF_REGISTER_MAX_RANGE)
1470 reg->max_value = BPF_REGISTER_MAX_RANGE;
1471 if (reg->min_value < BPF_REGISTER_MIN_RANGE ||
1472 reg->min_value > BPF_REGISTER_MAX_RANGE)
1473 reg->min_value = BPF_REGISTER_MIN_RANGE;
1474 }
1475
1476 static void adjust_reg_min_max_vals(struct bpf_verifier_env *env,
1477 struct bpf_insn *insn)
1478 {
1479 struct bpf_reg_state *regs = env->cur_state.regs, *dst_reg;
1480 s64 min_val = BPF_REGISTER_MIN_RANGE;
1481 u64 max_val = BPF_REGISTER_MAX_RANGE;
1482 bool min_set = false, max_set = false;
1483 u8 opcode = BPF_OP(insn->code);
1484
1485 dst_reg = &regs[insn->dst_reg];
1486 if (BPF_SRC(insn->code) == BPF_X) {
1487 check_reg_overflow(&regs[insn->src_reg]);
1488 min_val = regs[insn->src_reg].min_value;
1489 max_val = regs[insn->src_reg].max_value;
1490
1491 /* If the source register is a random pointer then the
1492 * min_value/max_value values represent the range of the known
1493 * accesses into that value, not the actual min/max value of the
1494 * register itself. In this case we have to reset the reg range
1495 * values so we know it is not safe to look at.
1496 */
1497 if (regs[insn->src_reg].type != CONST_IMM &&
1498 regs[insn->src_reg].type != UNKNOWN_VALUE) {
1499 min_val = BPF_REGISTER_MIN_RANGE;
1500 max_val = BPF_REGISTER_MAX_RANGE;
1501 }
1502 } else if (insn->imm < BPF_REGISTER_MAX_RANGE &&
1503 (s64)insn->imm > BPF_REGISTER_MIN_RANGE) {
1504 min_val = max_val = insn->imm;
1505 min_set = max_set = true;
1506 }
1507
1508 /* We don't know anything about what was done to this register, mark it
1509 * as unknown.
1510 */
1511 if (min_val == BPF_REGISTER_MIN_RANGE &&
1512 max_val == BPF_REGISTER_MAX_RANGE) {
1513 reset_reg_range_values(regs, insn->dst_reg);
1514 return;
1515 }
1516
1517 /* If one of our values was at the end of our ranges then we can't just
1518 * do our normal operations to the register, we need to set the values
1519 * to the min/max since they are undefined.
1520 */
1521 if (min_val == BPF_REGISTER_MIN_RANGE)
1522 dst_reg->min_value = BPF_REGISTER_MIN_RANGE;
1523 if (max_val == BPF_REGISTER_MAX_RANGE)
1524 dst_reg->max_value = BPF_REGISTER_MAX_RANGE;
1525
1526 switch (opcode) {
1527 case BPF_ADD:
1528 if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE)
1529 dst_reg->min_value += min_val;
1530 if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1531 dst_reg->max_value += max_val;
1532 break;
1533 case BPF_SUB:
1534 if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE)
1535 dst_reg->min_value -= min_val;
1536 if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1537 dst_reg->max_value -= max_val;
1538 break;
1539 case BPF_MUL:
1540 if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE)
1541 dst_reg->min_value *= min_val;
1542 if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1543 dst_reg->max_value *= max_val;
1544 break;
1545 case BPF_AND:
1546 /* Disallow AND'ing of negative numbers, ain't nobody got time
1547 * for that. Otherwise the minimum is 0 and the max is the max
1548 * value we could AND against.
1549 */
1550 if (min_val < 0)
1551 dst_reg->min_value = BPF_REGISTER_MIN_RANGE;
1552 else
1553 dst_reg->min_value = 0;
1554 dst_reg->max_value = max_val;
1555 break;
1556 case BPF_LSH:
1557 /* Gotta have special overflow logic here, if we're shifting
1558 * more than MAX_RANGE then just assume we have an invalid
1559 * range.
1560 */
1561 if (min_val > ilog2(BPF_REGISTER_MAX_RANGE))
1562 dst_reg->min_value = BPF_REGISTER_MIN_RANGE;
1563 else if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE)
1564 dst_reg->min_value <<= min_val;
1565
1566 if (max_val > ilog2(BPF_REGISTER_MAX_RANGE))
1567 dst_reg->max_value = BPF_REGISTER_MAX_RANGE;
1568 else if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1569 dst_reg->max_value <<= max_val;
1570 break;
1571 case BPF_RSH:
1572 /* RSH by a negative number is undefined, and the BPF_RSH is an
1573 * unsigned shift, so make the appropriate casts.
1574 */
1575 if (min_val < 0 || dst_reg->min_value < 0)
1576 dst_reg->min_value = BPF_REGISTER_MIN_RANGE;
1577 else
1578 dst_reg->min_value =
1579 (u64)(dst_reg->min_value) >> min_val;
1580 if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1581 dst_reg->max_value >>= max_val;
1582 break;
1583 default:
1584 reset_reg_range_values(regs, insn->dst_reg);
1585 break;
1586 }
1587
1588 check_reg_overflow(dst_reg);
1589 }
1590
1591 /* check validity of 32-bit and 64-bit arithmetic operations */
1592 static int check_alu_op(struct bpf_verifier_env *env, struct bpf_insn *insn)
1593 {
1594 struct bpf_reg_state *regs = env->cur_state.regs, *dst_reg;
1595 u8 opcode = BPF_OP(insn->code);
1596 int err;
1597
1598 if (opcode == BPF_END || opcode == BPF_NEG) {
1599 if (opcode == BPF_NEG) {
1600 if (BPF_SRC(insn->code) != 0 ||
1601 insn->src_reg != BPF_REG_0 ||
1602 insn->off != 0 || insn->imm != 0) {
1603 verbose("BPF_NEG uses reserved fields\n");
1604 return -EINVAL;
1605 }
1606 } else {
1607 if (insn->src_reg != BPF_REG_0 || insn->off != 0 ||
1608 (insn->imm != 16 && insn->imm != 32 && insn->imm != 64)) {
1609 verbose("BPF_END uses reserved fields\n");
1610 return -EINVAL;
1611 }
1612 }
1613
1614 /* check src operand */
1615 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1616 if (err)
1617 return err;
1618
1619 if (is_pointer_value(env, insn->dst_reg)) {
1620 verbose("R%d pointer arithmetic prohibited\n",
1621 insn->dst_reg);
1622 return -EACCES;
1623 }
1624
1625 /* check dest operand */
1626 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1627 if (err)
1628 return err;
1629
1630 } else if (opcode == BPF_MOV) {
1631
1632 if (BPF_SRC(insn->code) == BPF_X) {
1633 if (insn->imm != 0 || insn->off != 0) {
1634 verbose("BPF_MOV uses reserved fields\n");
1635 return -EINVAL;
1636 }
1637
1638 /* check src operand */
1639 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1640 if (err)
1641 return err;
1642 } else {
1643 if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
1644 verbose("BPF_MOV uses reserved fields\n");
1645 return -EINVAL;
1646 }
1647 }
1648
1649 /* check dest operand */
1650 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1651 if (err)
1652 return err;
1653
1654 /* we are setting our register to something new, we need to
1655 * reset its range values.
1656 */
1657 reset_reg_range_values(regs, insn->dst_reg);
1658
1659 if (BPF_SRC(insn->code) == BPF_X) {
1660 if (BPF_CLASS(insn->code) == BPF_ALU64) {
1661 /* case: R1 = R2
1662 * copy register state to dest reg
1663 */
1664 regs[insn->dst_reg] = regs[insn->src_reg];
1665 } else {
1666 if (is_pointer_value(env, insn->src_reg)) {
1667 verbose("R%d partial copy of pointer\n",
1668 insn->src_reg);
1669 return -EACCES;
1670 }
1671 regs[insn->dst_reg].type = UNKNOWN_VALUE;
1672 regs[insn->dst_reg].map_ptr = NULL;
1673 }
1674 } else {
1675 /* case: R = imm
1676 * remember the value we stored into this reg
1677 */
1678 regs[insn->dst_reg].type = CONST_IMM;
1679 regs[insn->dst_reg].imm = insn->imm;
1680 regs[insn->dst_reg].max_value = insn->imm;
1681 regs[insn->dst_reg].min_value = insn->imm;
1682 }
1683
1684 } else if (opcode > BPF_END) {
1685 verbose("invalid BPF_ALU opcode %x\n", opcode);
1686 return -EINVAL;
1687
1688 } else { /* all other ALU ops: and, sub, xor, add, ... */
1689
1690 if (BPF_SRC(insn->code) == BPF_X) {
1691 if (insn->imm != 0 || insn->off != 0) {
1692 verbose("BPF_ALU uses reserved fields\n");
1693 return -EINVAL;
1694 }
1695 /* check src1 operand */
1696 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1697 if (err)
1698 return err;
1699 } else {
1700 if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
1701 verbose("BPF_ALU uses reserved fields\n");
1702 return -EINVAL;
1703 }
1704 }
1705
1706 /* check src2 operand */
1707 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1708 if (err)
1709 return err;
1710
1711 if ((opcode == BPF_MOD || opcode == BPF_DIV) &&
1712 BPF_SRC(insn->code) == BPF_K && insn->imm == 0) {
1713 verbose("div by zero\n");
1714 return -EINVAL;
1715 }
1716
1717 if ((opcode == BPF_LSH || opcode == BPF_RSH ||
1718 opcode == BPF_ARSH) && BPF_SRC(insn->code) == BPF_K) {
1719 int size = BPF_CLASS(insn->code) == BPF_ALU64 ? 64 : 32;
1720
1721 if (insn->imm < 0 || insn->imm >= size) {
1722 verbose("invalid shift %d\n", insn->imm);
1723 return -EINVAL;
1724 }
1725 }
1726
1727 /* check dest operand */
1728 err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK);
1729 if (err)
1730 return err;
1731
1732 dst_reg = &regs[insn->dst_reg];
1733
1734 /* first we want to adjust our ranges. */
1735 adjust_reg_min_max_vals(env, insn);
1736
1737 /* pattern match 'bpf_add Rx, imm' instruction */
1738 if (opcode == BPF_ADD && BPF_CLASS(insn->code) == BPF_ALU64 &&
1739 dst_reg->type == FRAME_PTR && BPF_SRC(insn->code) == BPF_K) {
1740 dst_reg->type = PTR_TO_STACK;
1741 dst_reg->imm = insn->imm;
1742 return 0;
1743 } else if (opcode == BPF_ADD &&
1744 BPF_CLASS(insn->code) == BPF_ALU64 &&
1745 (dst_reg->type == PTR_TO_PACKET ||
1746 (BPF_SRC(insn->code) == BPF_X &&
1747 regs[insn->src_reg].type == PTR_TO_PACKET))) {
1748 /* ptr_to_packet += K|X */
1749 return check_packet_ptr_add(env, insn);
1750 } else if (BPF_CLASS(insn->code) == BPF_ALU64 &&
1751 dst_reg->type == UNKNOWN_VALUE &&
1752 env->allow_ptr_leaks) {
1753 /* unknown += K|X */
1754 return evaluate_reg_alu(env, insn);
1755 } else if (BPF_CLASS(insn->code) == BPF_ALU64 &&
1756 dst_reg->type == CONST_IMM &&
1757 env->allow_ptr_leaks) {
1758 /* reg_imm += K|X */
1759 return evaluate_reg_imm_alu(env, insn);
1760 } else if (is_pointer_value(env, insn->dst_reg)) {
1761 verbose("R%d pointer arithmetic prohibited\n",
1762 insn->dst_reg);
1763 return -EACCES;
1764 } else if (BPF_SRC(insn->code) == BPF_X &&
1765 is_pointer_value(env, insn->src_reg)) {
1766 verbose("R%d pointer arithmetic prohibited\n",
1767 insn->src_reg);
1768 return -EACCES;
1769 }
1770
1771 /* If we did pointer math on a map value then just set it to our
1772 * PTR_TO_MAP_VALUE_ADJ type so we can deal with any stores or
1773 * loads to this register appropriately, otherwise just mark the
1774 * register as unknown.
1775 */
1776 if (env->allow_ptr_leaks &&
1777 (dst_reg->type == PTR_TO_MAP_VALUE ||
1778 dst_reg->type == PTR_TO_MAP_VALUE_ADJ))
1779 dst_reg->type = PTR_TO_MAP_VALUE_ADJ;
1780 else
1781 mark_reg_unknown_value(regs, insn->dst_reg);
1782 }
1783
1784 return 0;
1785 }
1786
1787 static void find_good_pkt_pointers(struct bpf_verifier_state *state,
1788 struct bpf_reg_state *dst_reg)
1789 {
1790 struct bpf_reg_state *regs = state->regs, *reg;
1791 int i;
1792
1793 /* LLVM can generate two kind of checks:
1794 *
1795 * Type 1:
1796 *
1797 * r2 = r3;
1798 * r2 += 8;
1799 * if (r2 > pkt_end) goto <handle exception>
1800 * <access okay>
1801 *
1802 * Where:
1803 * r2 == dst_reg, pkt_end == src_reg
1804 * r2=pkt(id=n,off=8,r=0)
1805 * r3=pkt(id=n,off=0,r=0)
1806 *
1807 * Type 2:
1808 *
1809 * r2 = r3;
1810 * r2 += 8;
1811 * if (pkt_end >= r2) goto <access okay>
1812 * <handle exception>
1813 *
1814 * Where:
1815 * pkt_end == dst_reg, r2 == src_reg
1816 * r2=pkt(id=n,off=8,r=0)
1817 * r3=pkt(id=n,off=0,r=0)
1818 *
1819 * Find register r3 and mark its range as r3=pkt(id=n,off=0,r=8)
1820 * so that range of bytes [r3, r3 + 8) is safe to access.
1821 */
1822
1823 for (i = 0; i < MAX_BPF_REG; i++)
1824 if (regs[i].type == PTR_TO_PACKET && regs[i].id == dst_reg->id)
1825 regs[i].range = dst_reg->off;
1826
1827 for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
1828 if (state->stack_slot_type[i] != STACK_SPILL)
1829 continue;
1830 reg = &state->spilled_regs[i / BPF_REG_SIZE];
1831 if (reg->type == PTR_TO_PACKET && reg->id == dst_reg->id)
1832 reg->range = dst_reg->off;
1833 }
1834 }
1835
1836 /* Adjusts the register min/max values in the case that the dst_reg is the
1837 * variable register that we are working on, and src_reg is a constant or we're
1838 * simply doing a BPF_K check.
1839 */
1840 static void reg_set_min_max(struct bpf_reg_state *true_reg,
1841 struct bpf_reg_state *false_reg, u64 val,
1842 u8 opcode)
1843 {
1844 switch (opcode) {
1845 case BPF_JEQ:
1846 /* If this is false then we know nothing Jon Snow, but if it is
1847 * true then we know for sure.
1848 */
1849 true_reg->max_value = true_reg->min_value = val;
1850 break;
1851 case BPF_JNE:
1852 /* If this is true we know nothing Jon Snow, but if it is false
1853 * we know the value for sure;
1854 */
1855 false_reg->max_value = false_reg->min_value = val;
1856 break;
1857 case BPF_JGT:
1858 /* Unsigned comparison, the minimum value is 0. */
1859 false_reg->min_value = 0;
1860 case BPF_JSGT:
1861 /* If this is false then we know the maximum val is val,
1862 * otherwise we know the min val is val+1.
1863 */
1864 false_reg->max_value = val;
1865 true_reg->min_value = val + 1;
1866 break;
1867 case BPF_JGE:
1868 /* Unsigned comparison, the minimum value is 0. */
1869 false_reg->min_value = 0;
1870 case BPF_JSGE:
1871 /* If this is false then we know the maximum value is val - 1,
1872 * otherwise we know the mimimum value is val.
1873 */
1874 false_reg->max_value = val - 1;
1875 true_reg->min_value = val;
1876 break;
1877 default:
1878 break;
1879 }
1880
1881 check_reg_overflow(false_reg);
1882 check_reg_overflow(true_reg);
1883 }
1884
1885 /* Same as above, but for the case that dst_reg is a CONST_IMM reg and src_reg
1886 * is the variable reg.
1887 */
1888 static void reg_set_min_max_inv(struct bpf_reg_state *true_reg,
1889 struct bpf_reg_state *false_reg, u64 val,
1890 u8 opcode)
1891 {
1892 switch (opcode) {
1893 case BPF_JEQ:
1894 /* If this is false then we know nothing Jon Snow, but if it is
1895 * true then we know for sure.
1896 */
1897 true_reg->max_value = true_reg->min_value = val;
1898 break;
1899 case BPF_JNE:
1900 /* If this is true we know nothing Jon Snow, but if it is false
1901 * we know the value for sure;
1902 */
1903 false_reg->max_value = false_reg->min_value = val;
1904 break;
1905 case BPF_JGT:
1906 /* Unsigned comparison, the minimum value is 0. */
1907 true_reg->min_value = 0;
1908 case BPF_JSGT:
1909 /*
1910 * If this is false, then the val is <= the register, if it is
1911 * true the register <= to the val.
1912 */
1913 false_reg->min_value = val;
1914 true_reg->max_value = val - 1;
1915 break;
1916 case BPF_JGE:
1917 /* Unsigned comparison, the minimum value is 0. */
1918 true_reg->min_value = 0;
1919 case BPF_JSGE:
1920 /* If this is false then constant < register, if it is true then
1921 * the register < constant.
1922 */
1923 false_reg->min_value = val + 1;
1924 true_reg->max_value = val;
1925 break;
1926 default:
1927 break;
1928 }
1929
1930 check_reg_overflow(false_reg);
1931 check_reg_overflow(true_reg);
1932 }
1933
1934 static int check_cond_jmp_op(struct bpf_verifier_env *env,
1935 struct bpf_insn *insn, int *insn_idx)
1936 {
1937 struct bpf_verifier_state *other_branch, *this_branch = &env->cur_state;
1938 struct bpf_reg_state *regs = this_branch->regs, *dst_reg;
1939 u8 opcode = BPF_OP(insn->code);
1940 int err;
1941
1942 if (opcode > BPF_EXIT) {
1943 verbose("invalid BPF_JMP opcode %x\n", opcode);
1944 return -EINVAL;
1945 }
1946
1947 if (BPF_SRC(insn->code) == BPF_X) {
1948 if (insn->imm != 0) {
1949 verbose("BPF_JMP uses reserved fields\n");
1950 return -EINVAL;
1951 }
1952
1953 /* check src1 operand */
1954 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1955 if (err)
1956 return err;
1957
1958 if (is_pointer_value(env, insn->src_reg)) {
1959 verbose("R%d pointer comparison prohibited\n",
1960 insn->src_reg);
1961 return -EACCES;
1962 }
1963 } else {
1964 if (insn->src_reg != BPF_REG_0) {
1965 verbose("BPF_JMP uses reserved fields\n");
1966 return -EINVAL;
1967 }
1968 }
1969
1970 /* check src2 operand */
1971 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1972 if (err)
1973 return err;
1974
1975 dst_reg = &regs[insn->dst_reg];
1976
1977 /* detect if R == 0 where R was initialized to zero earlier */
1978 if (BPF_SRC(insn->code) == BPF_K &&
1979 (opcode == BPF_JEQ || opcode == BPF_JNE) &&
1980 dst_reg->type == CONST_IMM && dst_reg->imm == insn->imm) {
1981 if (opcode == BPF_JEQ) {
1982 /* if (imm == imm) goto pc+off;
1983 * only follow the goto, ignore fall-through
1984 */
1985 *insn_idx += insn->off;
1986 return 0;
1987 } else {
1988 /* if (imm != imm) goto pc+off;
1989 * only follow fall-through branch, since
1990 * that's where the program will go
1991 */
1992 return 0;
1993 }
1994 }
1995
1996 other_branch = push_stack(env, *insn_idx + insn->off + 1, *insn_idx);
1997 if (!other_branch)
1998 return -EFAULT;
1999
2000 /* detect if we are comparing against a constant value so we can adjust
2001 * our min/max values for our dst register.
2002 */
2003 if (BPF_SRC(insn->code) == BPF_X) {
2004 if (regs[insn->src_reg].type == CONST_IMM)
2005 reg_set_min_max(&other_branch->regs[insn->dst_reg],
2006 dst_reg, regs[insn->src_reg].imm,
2007 opcode);
2008 else if (dst_reg->type == CONST_IMM)
2009 reg_set_min_max_inv(&other_branch->regs[insn->src_reg],
2010 &regs[insn->src_reg], dst_reg->imm,
2011 opcode);
2012 } else {
2013 reg_set_min_max(&other_branch->regs[insn->dst_reg],
2014 dst_reg, insn->imm, opcode);
2015 }
2016
2017 /* detect if R == 0 where R is returned from bpf_map_lookup_elem() */
2018 if (BPF_SRC(insn->code) == BPF_K &&
2019 insn->imm == 0 && (opcode == BPF_JEQ || opcode == BPF_JNE) &&
2020 dst_reg->type == PTR_TO_MAP_VALUE_OR_NULL) {
2021 if (opcode == BPF_JEQ) {
2022 /* next fallthrough insn can access memory via
2023 * this register
2024 */
2025 regs[insn->dst_reg].type = PTR_TO_MAP_VALUE;
2026 /* branch targer cannot access it, since reg == 0 */
2027 mark_reg_unknown_value(other_branch->regs,
2028 insn->dst_reg);
2029 } else {
2030 other_branch->regs[insn->dst_reg].type = PTR_TO_MAP_VALUE;
2031 mark_reg_unknown_value(regs, insn->dst_reg);
2032 }
2033 } else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JGT &&
2034 dst_reg->type == PTR_TO_PACKET &&
2035 regs[insn->src_reg].type == PTR_TO_PACKET_END) {
2036 find_good_pkt_pointers(this_branch, dst_reg);
2037 } else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JGE &&
2038 dst_reg->type == PTR_TO_PACKET_END &&
2039 regs[insn->src_reg].type == PTR_TO_PACKET) {
2040 find_good_pkt_pointers(other_branch, &regs[insn->src_reg]);
2041 } else if (is_pointer_value(env, insn->dst_reg)) {
2042 verbose("R%d pointer comparison prohibited\n", insn->dst_reg);
2043 return -EACCES;
2044 }
2045 if (log_level)
2046 print_verifier_state(this_branch);
2047 return 0;
2048 }
2049
2050 /* return the map pointer stored inside BPF_LD_IMM64 instruction */
2051 static struct bpf_map *ld_imm64_to_map_ptr(struct bpf_insn *insn)
2052 {
2053 u64 imm64 = ((u64) (u32) insn[0].imm) | ((u64) (u32) insn[1].imm) << 32;
2054
2055 return (struct bpf_map *) (unsigned long) imm64;
2056 }
2057
2058 /* verify BPF_LD_IMM64 instruction */
2059 static int check_ld_imm(struct bpf_verifier_env *env, struct bpf_insn *insn)
2060 {
2061 struct bpf_reg_state *regs = env->cur_state.regs;
2062 int err;
2063
2064 if (BPF_SIZE(insn->code) != BPF_DW) {
2065 verbose("invalid BPF_LD_IMM insn\n");
2066 return -EINVAL;
2067 }
2068 if (insn->off != 0) {
2069 verbose("BPF_LD_IMM64 uses reserved fields\n");
2070 return -EINVAL;
2071 }
2072
2073 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
2074 if (err)
2075 return err;
2076
2077 if (insn->src_reg == 0) {
2078 /* generic move 64-bit immediate into a register,
2079 * only analyzer needs to collect the ld_imm value.
2080 */
2081 u64 imm = ((u64)(insn + 1)->imm << 32) | (u32)insn->imm;
2082
2083 if (!env->analyzer_ops)
2084 return 0;
2085
2086 regs[insn->dst_reg].type = CONST_IMM;
2087 regs[insn->dst_reg].imm = imm;
2088 return 0;
2089 }
2090
2091 /* replace_map_fd_with_map_ptr() should have caught bad ld_imm64 */
2092 BUG_ON(insn->src_reg != BPF_PSEUDO_MAP_FD);
2093
2094 regs[insn->dst_reg].type = CONST_PTR_TO_MAP;
2095 regs[insn->dst_reg].map_ptr = ld_imm64_to_map_ptr(insn);
2096 return 0;
2097 }
2098
2099 static bool may_access_skb(enum bpf_prog_type type)
2100 {
2101 switch (type) {
2102 case BPF_PROG_TYPE_SOCKET_FILTER:
2103 case BPF_PROG_TYPE_SCHED_CLS:
2104 case BPF_PROG_TYPE_SCHED_ACT:
2105 return true;
2106 default:
2107 return false;
2108 }
2109 }
2110
2111 /* verify safety of LD_ABS|LD_IND instructions:
2112 * - they can only appear in the programs where ctx == skb
2113 * - since they are wrappers of function calls, they scratch R1-R5 registers,
2114 * preserve R6-R9, and store return value into R0
2115 *
2116 * Implicit input:
2117 * ctx == skb == R6 == CTX
2118 *
2119 * Explicit input:
2120 * SRC == any register
2121 * IMM == 32-bit immediate
2122 *
2123 * Output:
2124 * R0 - 8/16/32-bit skb data converted to cpu endianness
2125 */
2126 static int check_ld_abs(struct bpf_verifier_env *env, struct bpf_insn *insn)
2127 {
2128 struct bpf_reg_state *regs = env->cur_state.regs;
2129 u8 mode = BPF_MODE(insn->code);
2130 struct bpf_reg_state *reg;
2131 int i, err;
2132
2133 if (!may_access_skb(env->prog->type)) {
2134 verbose("BPF_LD_[ABS|IND] instructions not allowed for this program type\n");
2135 return -EINVAL;
2136 }
2137
2138 if (insn->dst_reg != BPF_REG_0 || insn->off != 0 ||
2139 BPF_SIZE(insn->code) == BPF_DW ||
2140 (mode == BPF_ABS && insn->src_reg != BPF_REG_0)) {
2141 verbose("BPF_LD_[ABS|IND] uses reserved fields\n");
2142 return -EINVAL;
2143 }
2144
2145 /* check whether implicit source operand (register R6) is readable */
2146 err = check_reg_arg(regs, BPF_REG_6, SRC_OP);
2147 if (err)
2148 return err;
2149
2150 if (regs[BPF_REG_6].type != PTR_TO_CTX) {
2151 verbose("at the time of BPF_LD_ABS|IND R6 != pointer to skb\n");
2152 return -EINVAL;
2153 }
2154
2155 if (mode == BPF_IND) {
2156 /* check explicit source operand */
2157 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
2158 if (err)
2159 return err;
2160 }
2161
2162 /* reset caller saved regs to unreadable */
2163 for (i = 0; i < CALLER_SAVED_REGS; i++) {
2164 reg = regs + caller_saved[i];
2165 reg->type = NOT_INIT;
2166 reg->imm = 0;
2167 }
2168
2169 /* mark destination R0 register as readable, since it contains
2170 * the value fetched from the packet
2171 */
2172 regs[BPF_REG_0].type = UNKNOWN_VALUE;
2173 return 0;
2174 }
2175
2176 /* non-recursive DFS pseudo code
2177 * 1 procedure DFS-iterative(G,v):
2178 * 2 label v as discovered
2179 * 3 let S be a stack
2180 * 4 S.push(v)
2181 * 5 while S is not empty
2182 * 6 t <- S.pop()
2183 * 7 if t is what we're looking for:
2184 * 8 return t
2185 * 9 for all edges e in G.adjacentEdges(t) do
2186 * 10 if edge e is already labelled
2187 * 11 continue with the next edge
2188 * 12 w <- G.adjacentVertex(t,e)
2189 * 13 if vertex w is not discovered and not explored
2190 * 14 label e as tree-edge
2191 * 15 label w as discovered
2192 * 16 S.push(w)
2193 * 17 continue at 5
2194 * 18 else if vertex w is discovered
2195 * 19 label e as back-edge
2196 * 20 else
2197 * 21 // vertex w is explored
2198 * 22 label e as forward- or cross-edge
2199 * 23 label t as explored
2200 * 24 S.pop()
2201 *
2202 * convention:
2203 * 0x10 - discovered
2204 * 0x11 - discovered and fall-through edge labelled
2205 * 0x12 - discovered and fall-through and branch edges labelled
2206 * 0x20 - explored
2207 */
2208
2209 enum {
2210 DISCOVERED = 0x10,
2211 EXPLORED = 0x20,
2212 FALLTHROUGH = 1,
2213 BRANCH = 2,
2214 };
2215
2216 #define STATE_LIST_MARK ((struct bpf_verifier_state_list *) -1L)
2217
2218 static int *insn_stack; /* stack of insns to process */
2219 static int cur_stack; /* current stack index */
2220 static int *insn_state;
2221
2222 /* t, w, e - match pseudo-code above:
2223 * t - index of current instruction
2224 * w - next instruction
2225 * e - edge
2226 */
2227 static int push_insn(int t, int w, int e, struct bpf_verifier_env *env)
2228 {
2229 if (e == FALLTHROUGH && insn_state[t] >= (DISCOVERED | FALLTHROUGH))
2230 return 0;
2231
2232 if (e == BRANCH && insn_state[t] >= (DISCOVERED | BRANCH))
2233 return 0;
2234
2235 if (w < 0 || w >= env->prog->len) {
2236 verbose("jump out of range from insn %d to %d\n", t, w);
2237 return -EINVAL;
2238 }
2239
2240 if (e == BRANCH)
2241 /* mark branch target for state pruning */
2242 env->explored_states[w] = STATE_LIST_MARK;
2243
2244 if (insn_state[w] == 0) {
2245 /* tree-edge */
2246 insn_state[t] = DISCOVERED | e;
2247 insn_state[w] = DISCOVERED;
2248 if (cur_stack >= env->prog->len)
2249 return -E2BIG;
2250 insn_stack[cur_stack++] = w;
2251 return 1;
2252 } else if ((insn_state[w] & 0xF0) == DISCOVERED) {
2253 verbose("back-edge from insn %d to %d\n", t, w);
2254 return -EINVAL;
2255 } else if (insn_state[w] == EXPLORED) {
2256 /* forward- or cross-edge */
2257 insn_state[t] = DISCOVERED | e;
2258 } else {
2259 verbose("insn state internal bug\n");
2260 return -EFAULT;
2261 }
2262 return 0;
2263 }
2264
2265 /* non-recursive depth-first-search to detect loops in BPF program
2266 * loop == back-edge in directed graph
2267 */
2268 static int check_cfg(struct bpf_verifier_env *env)
2269 {
2270 struct bpf_insn *insns = env->prog->insnsi;
2271 int insn_cnt = env->prog->len;
2272 int ret = 0;
2273 int i, t;
2274
2275 insn_state = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
2276 if (!insn_state)
2277 return -ENOMEM;
2278
2279 insn_stack = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
2280 if (!insn_stack) {
2281 kfree(insn_state);
2282 return -ENOMEM;
2283 }
2284
2285 insn_state[0] = DISCOVERED; /* mark 1st insn as discovered */
2286 insn_stack[0] = 0; /* 0 is the first instruction */
2287 cur_stack = 1;
2288
2289 peek_stack:
2290 if (cur_stack == 0)
2291 goto check_state;
2292 t = insn_stack[cur_stack - 1];
2293
2294 if (BPF_CLASS(insns[t].code) == BPF_JMP) {
2295 u8 opcode = BPF_OP(insns[t].code);
2296
2297 if (opcode == BPF_EXIT) {
2298 goto mark_explored;
2299 } else if (opcode == BPF_CALL) {
2300 ret = push_insn(t, t + 1, FALLTHROUGH, env);
2301 if (ret == 1)
2302 goto peek_stack;
2303 else if (ret < 0)
2304 goto err_free;
2305 if (t + 1 < insn_cnt)
2306 env->explored_states[t + 1] = STATE_LIST_MARK;
2307 } else if (opcode == BPF_JA) {
2308 if (BPF_SRC(insns[t].code) != BPF_K) {
2309 ret = -EINVAL;
2310 goto err_free;
2311 }
2312 /* unconditional jump with single edge */
2313 ret = push_insn(t, t + insns[t].off + 1,
2314 FALLTHROUGH, env);
2315 if (ret == 1)
2316 goto peek_stack;
2317 else if (ret < 0)
2318 goto err_free;
2319 /* tell verifier to check for equivalent states
2320 * after every call and jump
2321 */
2322 if (t + 1 < insn_cnt)
2323 env->explored_states[t + 1] = STATE_LIST_MARK;
2324 } else {
2325 /* conditional jump with two edges */
2326 ret = push_insn(t, t + 1, FALLTHROUGH, env);
2327 if (ret == 1)
2328 goto peek_stack;
2329 else if (ret < 0)
2330 goto err_free;
2331
2332 ret = push_insn(t, t + insns[t].off + 1, BRANCH, env);
2333 if (ret == 1)
2334 goto peek_stack;
2335 else if (ret < 0)
2336 goto err_free;
2337 }
2338 } else {
2339 /* all other non-branch instructions with single
2340 * fall-through edge
2341 */
2342 ret = push_insn(t, t + 1, FALLTHROUGH, env);
2343 if (ret == 1)
2344 goto peek_stack;
2345 else if (ret < 0)
2346 goto err_free;
2347 }
2348
2349 mark_explored:
2350 insn_state[t] = EXPLORED;
2351 if (cur_stack-- <= 0) {
2352 verbose("pop stack internal bug\n");
2353 ret = -EFAULT;
2354 goto err_free;
2355 }
2356 goto peek_stack;
2357
2358 check_state:
2359 for (i = 0; i < insn_cnt; i++) {
2360 if (insn_state[i] != EXPLORED) {
2361 verbose("unreachable insn %d\n", i);
2362 ret = -EINVAL;
2363 goto err_free;
2364 }
2365 }
2366 ret = 0; /* cfg looks good */
2367
2368 err_free:
2369 kfree(insn_state);
2370 kfree(insn_stack);
2371 return ret;
2372 }
2373
2374 /* the following conditions reduce the number of explored insns
2375 * from ~140k to ~80k for ultra large programs that use a lot of ptr_to_packet
2376 */
2377 static bool compare_ptrs_to_packet(struct bpf_reg_state *old,
2378 struct bpf_reg_state *cur)
2379 {
2380 if (old->id != cur->id)
2381 return false;
2382
2383 /* old ptr_to_packet is more conservative, since it allows smaller
2384 * range. Ex:
2385 * old(off=0,r=10) is equal to cur(off=0,r=20), because
2386 * old(off=0,r=10) means that with range=10 the verifier proceeded
2387 * further and found no issues with the program. Now we're in the same
2388 * spot with cur(off=0,r=20), so we're safe too, since anything further
2389 * will only be looking at most 10 bytes after this pointer.
2390 */
2391 if (old->off == cur->off && old->range < cur->range)
2392 return true;
2393
2394 /* old(off=20,r=10) is equal to cur(off=22,re=22 or 5 or 0)
2395 * since both cannot be used for packet access and safe(old)
2396 * pointer has smaller off that could be used for further
2397 * 'if (ptr > data_end)' check
2398 * Ex:
2399 * old(off=20,r=10) and cur(off=22,r=22) and cur(off=22,r=0) mean
2400 * that we cannot access the packet.
2401 * The safe range is:
2402 * [ptr, ptr + range - off)
2403 * so whenever off >=range, it means no safe bytes from this pointer.
2404 * When comparing old->off <= cur->off, it means that older code
2405 * went with smaller offset and that offset was later
2406 * used to figure out the safe range after 'if (ptr > data_end)' check
2407 * Say, 'old' state was explored like:
2408 * ... R3(off=0, r=0)
2409 * R4 = R3 + 20
2410 * ... now R4(off=20,r=0) <-- here
2411 * if (R4 > data_end)
2412 * ... R4(off=20,r=20), R3(off=0,r=20) and R3 can be used to access.
2413 * ... the code further went all the way to bpf_exit.
2414 * Now the 'cur' state at the mark 'here' has R4(off=30,r=0).
2415 * old_R4(off=20,r=0) equal to cur_R4(off=30,r=0), since if the verifier
2416 * goes further, such cur_R4 will give larger safe packet range after
2417 * 'if (R4 > data_end)' and all further insn were already good with r=20,
2418 * so they will be good with r=30 and we can prune the search.
2419 */
2420 if (old->off <= cur->off &&
2421 old->off >= old->range && cur->off >= cur->range)
2422 return true;
2423
2424 return false;
2425 }
2426
2427 /* compare two verifier states
2428 *
2429 * all states stored in state_list are known to be valid, since
2430 * verifier reached 'bpf_exit' instruction through them
2431 *
2432 * this function is called when verifier exploring different branches of
2433 * execution popped from the state stack. If it sees an old state that has
2434 * more strict register state and more strict stack state then this execution
2435 * branch doesn't need to be explored further, since verifier already
2436 * concluded that more strict state leads to valid finish.
2437 *
2438 * Therefore two states are equivalent if register state is more conservative
2439 * and explored stack state is more conservative than the current one.
2440 * Example:
2441 * explored current
2442 * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC)
2443 * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC)
2444 *
2445 * In other words if current stack state (one being explored) has more
2446 * valid slots than old one that already passed validation, it means
2447 * the verifier can stop exploring and conclude that current state is valid too
2448 *
2449 * Similarly with registers. If explored state has register type as invalid
2450 * whereas register type in current state is meaningful, it means that
2451 * the current state will reach 'bpf_exit' instruction safely
2452 */
2453 static bool states_equal(struct bpf_verifier_env *env,
2454 struct bpf_verifier_state *old,
2455 struct bpf_verifier_state *cur)
2456 {
2457 struct bpf_reg_state *rold, *rcur;
2458 int i;
2459
2460 for (i = 0; i < MAX_BPF_REG; i++) {
2461 rold = &old->regs[i];
2462 rcur = &cur->regs[i];
2463
2464 if (memcmp(rold, rcur, sizeof(*rold)) == 0)
2465 continue;
2466
2467 /* If the ranges were not the same, but everything else was and
2468 * we didn't do a variable access into a map then we are a-ok.
2469 */
2470 if (!env->varlen_map_value_access &&
2471 rold->type == rcur->type && rold->imm == rcur->imm)
2472 continue;
2473
2474 if (rold->type == NOT_INIT ||
2475 (rold->type == UNKNOWN_VALUE && rcur->type != NOT_INIT))
2476 continue;
2477
2478 if (rold->type == PTR_TO_PACKET && rcur->type == PTR_TO_PACKET &&
2479 compare_ptrs_to_packet(rold, rcur))
2480 continue;
2481
2482 return false;
2483 }
2484
2485 for (i = 0; i < MAX_BPF_STACK; i++) {
2486 if (old->stack_slot_type[i] == STACK_INVALID)
2487 continue;
2488 if (old->stack_slot_type[i] != cur->stack_slot_type[i])
2489 /* Ex: old explored (safe) state has STACK_SPILL in
2490 * this stack slot, but current has has STACK_MISC ->
2491 * this verifier states are not equivalent,
2492 * return false to continue verification of this path
2493 */
2494 return false;
2495 if (i % BPF_REG_SIZE)
2496 continue;
2497 if (memcmp(&old->spilled_regs[i / BPF_REG_SIZE],
2498 &cur->spilled_regs[i / BPF_REG_SIZE],
2499 sizeof(old->spilled_regs[0])))
2500 /* when explored and current stack slot types are
2501 * the same, check that stored pointers types
2502 * are the same as well.
2503 * Ex: explored safe path could have stored
2504 * (bpf_reg_state) {.type = PTR_TO_STACK, .imm = -8}
2505 * but current path has stored:
2506 * (bpf_reg_state) {.type = PTR_TO_STACK, .imm = -16}
2507 * such verifier states are not equivalent.
2508 * return false to continue verification of this path
2509 */
2510 return false;
2511 else
2512 continue;
2513 }
2514 return true;
2515 }
2516
2517 static int is_state_visited(struct bpf_verifier_env *env, int insn_idx)
2518 {
2519 struct bpf_verifier_state_list *new_sl;
2520 struct bpf_verifier_state_list *sl;
2521
2522 sl = env->explored_states[insn_idx];
2523 if (!sl)
2524 /* this 'insn_idx' instruction wasn't marked, so we will not
2525 * be doing state search here
2526 */
2527 return 0;
2528
2529 while (sl != STATE_LIST_MARK) {
2530 if (states_equal(env, &sl->state, &env->cur_state))
2531 /* reached equivalent register/stack state,
2532 * prune the search
2533 */
2534 return 1;
2535 sl = sl->next;
2536 }
2537
2538 /* there were no equivalent states, remember current one.
2539 * technically the current state is not proven to be safe yet,
2540 * but it will either reach bpf_exit (which means it's safe) or
2541 * it will be rejected. Since there are no loops, we won't be
2542 * seeing this 'insn_idx' instruction again on the way to bpf_exit
2543 */
2544 new_sl = kmalloc(sizeof(struct bpf_verifier_state_list), GFP_USER);
2545 if (!new_sl)
2546 return -ENOMEM;
2547
2548 /* add new state to the head of linked list */
2549 memcpy(&new_sl->state, &env->cur_state, sizeof(env->cur_state));
2550 new_sl->next = env->explored_states[insn_idx];
2551 env->explored_states[insn_idx] = new_sl;
2552 return 0;
2553 }
2554
2555 static int ext_analyzer_insn_hook(struct bpf_verifier_env *env,
2556 int insn_idx, int prev_insn_idx)
2557 {
2558 if (!env->analyzer_ops || !env->analyzer_ops->insn_hook)
2559 return 0;
2560
2561 return env->analyzer_ops->insn_hook(env, insn_idx, prev_insn_idx);
2562 }
2563
2564 static int do_check(struct bpf_verifier_env *env)
2565 {
2566 struct bpf_verifier_state *state = &env->cur_state;
2567 struct bpf_insn *insns = env->prog->insnsi;
2568 struct bpf_reg_state *regs = state->regs;
2569 int insn_cnt = env->prog->len;
2570 int insn_idx, prev_insn_idx = 0;
2571 int insn_processed = 0;
2572 bool do_print_state = false;
2573
2574 init_reg_state(regs);
2575 insn_idx = 0;
2576 env->varlen_map_value_access = false;
2577 for (;;) {
2578 struct bpf_insn *insn;
2579 u8 class;
2580 int err;
2581
2582 if (insn_idx >= insn_cnt) {
2583 verbose("invalid insn idx %d insn_cnt %d\n",
2584 insn_idx, insn_cnt);
2585 return -EFAULT;
2586 }
2587
2588 insn = &insns[insn_idx];
2589 class = BPF_CLASS(insn->code);
2590
2591 if (++insn_processed > BPF_COMPLEXITY_LIMIT_INSNS) {
2592 verbose("BPF program is too large. Proccessed %d insn\n",
2593 insn_processed);
2594 return -E2BIG;
2595 }
2596
2597 err = is_state_visited(env, insn_idx);
2598 if (err < 0)
2599 return err;
2600 if (err == 1) {
2601 /* found equivalent state, can prune the search */
2602 if (log_level) {
2603 if (do_print_state)
2604 verbose("\nfrom %d to %d: safe\n",
2605 prev_insn_idx, insn_idx);
2606 else
2607 verbose("%d: safe\n", insn_idx);
2608 }
2609 goto process_bpf_exit;
2610 }
2611
2612 if (log_level && do_print_state) {
2613 verbose("\nfrom %d to %d:", prev_insn_idx, insn_idx);
2614 print_verifier_state(&env->cur_state);
2615 do_print_state = false;
2616 }
2617
2618 if (log_level) {
2619 verbose("%d: ", insn_idx);
2620 print_bpf_insn(insn);
2621 }
2622
2623 err = ext_analyzer_insn_hook(env, insn_idx, prev_insn_idx);
2624 if (err)
2625 return err;
2626
2627 if (class == BPF_ALU || class == BPF_ALU64) {
2628 err = check_alu_op(env, insn);
2629 if (err)
2630 return err;
2631
2632 } else if (class == BPF_LDX) {
2633 enum bpf_reg_type *prev_src_type, src_reg_type;
2634
2635 /* check for reserved fields is already done */
2636
2637 /* check src operand */
2638 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
2639 if (err)
2640 return err;
2641
2642 err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK);
2643 if (err)
2644 return err;
2645
2646 src_reg_type = regs[insn->src_reg].type;
2647
2648 /* check that memory (src_reg + off) is readable,
2649 * the state of dst_reg will be updated by this func
2650 */
2651 err = check_mem_access(env, insn->src_reg, insn->off,
2652 BPF_SIZE(insn->code), BPF_READ,
2653 insn->dst_reg);
2654 if (err)
2655 return err;
2656
2657 reset_reg_range_values(regs, insn->dst_reg);
2658 if (BPF_SIZE(insn->code) != BPF_W &&
2659 BPF_SIZE(insn->code) != BPF_DW) {
2660 insn_idx++;
2661 continue;
2662 }
2663
2664 prev_src_type = &env->insn_aux_data[insn_idx].ptr_type;
2665
2666 if (*prev_src_type == NOT_INIT) {
2667 /* saw a valid insn
2668 * dst_reg = *(u32 *)(src_reg + off)
2669 * save type to validate intersecting paths
2670 */
2671 *prev_src_type = src_reg_type;
2672
2673 } else if (src_reg_type != *prev_src_type &&
2674 (src_reg_type == PTR_TO_CTX ||
2675 *prev_src_type == PTR_TO_CTX)) {
2676 /* ABuser program is trying to use the same insn
2677 * dst_reg = *(u32*) (src_reg + off)
2678 * with different pointer types:
2679 * src_reg == ctx in one branch and
2680 * src_reg == stack|map in some other branch.
2681 * Reject it.
2682 */
2683 verbose("same insn cannot be used with different pointers\n");
2684 return -EINVAL;
2685 }
2686
2687 } else if (class == BPF_STX) {
2688 enum bpf_reg_type *prev_dst_type, dst_reg_type;
2689
2690 if (BPF_MODE(insn->code) == BPF_XADD) {
2691 err = check_xadd(env, insn);
2692 if (err)
2693 return err;
2694 insn_idx++;
2695 continue;
2696 }
2697
2698 /* check src1 operand */
2699 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
2700 if (err)
2701 return err;
2702 /* check src2 operand */
2703 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
2704 if (err)
2705 return err;
2706
2707 dst_reg_type = regs[insn->dst_reg].type;
2708
2709 /* check that memory (dst_reg + off) is writeable */
2710 err = check_mem_access(env, insn->dst_reg, insn->off,
2711 BPF_SIZE(insn->code), BPF_WRITE,
2712 insn->src_reg);
2713 if (err)
2714 return err;
2715
2716 prev_dst_type = &env->insn_aux_data[insn_idx].ptr_type;
2717
2718 if (*prev_dst_type == NOT_INIT) {
2719 *prev_dst_type = dst_reg_type;
2720 } else if (dst_reg_type != *prev_dst_type &&
2721 (dst_reg_type == PTR_TO_CTX ||
2722 *prev_dst_type == PTR_TO_CTX)) {
2723 verbose("same insn cannot be used with different pointers\n");
2724 return -EINVAL;
2725 }
2726
2727 } else if (class == BPF_ST) {
2728 if (BPF_MODE(insn->code) != BPF_MEM ||
2729 insn->src_reg != BPF_REG_0) {
2730 verbose("BPF_ST uses reserved fields\n");
2731 return -EINVAL;
2732 }
2733 /* check src operand */
2734 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
2735 if (err)
2736 return err;
2737
2738 /* check that memory (dst_reg + off) is writeable */
2739 err = check_mem_access(env, insn->dst_reg, insn->off,
2740 BPF_SIZE(insn->code), BPF_WRITE,
2741 -1);
2742 if (err)
2743 return err;
2744
2745 } else if (class == BPF_JMP) {
2746 u8 opcode = BPF_OP(insn->code);
2747
2748 if (opcode == BPF_CALL) {
2749 if (BPF_SRC(insn->code) != BPF_K ||
2750 insn->off != 0 ||
2751 insn->src_reg != BPF_REG_0 ||
2752 insn->dst_reg != BPF_REG_0) {
2753 verbose("BPF_CALL uses reserved fields\n");
2754 return -EINVAL;
2755 }
2756
2757 err = check_call(env, insn->imm);
2758 if (err)
2759 return err;
2760
2761 } else if (opcode == BPF_JA) {
2762 if (BPF_SRC(insn->code) != BPF_K ||
2763 insn->imm != 0 ||
2764 insn->src_reg != BPF_REG_0 ||
2765 insn->dst_reg != BPF_REG_0) {
2766 verbose("BPF_JA uses reserved fields\n");
2767 return -EINVAL;
2768 }
2769
2770 insn_idx += insn->off + 1;
2771 continue;
2772
2773 } else if (opcode == BPF_EXIT) {
2774 if (BPF_SRC(insn->code) != BPF_K ||
2775 insn->imm != 0 ||
2776 insn->src_reg != BPF_REG_0 ||
2777 insn->dst_reg != BPF_REG_0) {
2778 verbose("BPF_EXIT uses reserved fields\n");
2779 return -EINVAL;
2780 }
2781
2782 /* eBPF calling convetion is such that R0 is used
2783 * to return the value from eBPF program.
2784 * Make sure that it's readable at this time
2785 * of bpf_exit, which means that program wrote
2786 * something into it earlier
2787 */
2788 err = check_reg_arg(regs, BPF_REG_0, SRC_OP);
2789 if (err)
2790 return err;
2791
2792 if (is_pointer_value(env, BPF_REG_0)) {
2793 verbose("R0 leaks addr as return value\n");
2794 return -EACCES;
2795 }
2796
2797 process_bpf_exit:
2798 insn_idx = pop_stack(env, &prev_insn_idx);
2799 if (insn_idx < 0) {
2800 break;
2801 } else {
2802 do_print_state = true;
2803 continue;
2804 }
2805 } else {
2806 err = check_cond_jmp_op(env, insn, &insn_idx);
2807 if (err)
2808 return err;
2809 }
2810 } else if (class == BPF_LD) {
2811 u8 mode = BPF_MODE(insn->code);
2812
2813 if (mode == BPF_ABS || mode == BPF_IND) {
2814 err = check_ld_abs(env, insn);
2815 if (err)
2816 return err;
2817
2818 } else if (mode == BPF_IMM) {
2819 err = check_ld_imm(env, insn);
2820 if (err)
2821 return err;
2822
2823 insn_idx++;
2824 } else {
2825 verbose("invalid BPF_LD mode\n");
2826 return -EINVAL;
2827 }
2828 reset_reg_range_values(regs, insn->dst_reg);
2829 } else {
2830 verbose("unknown insn class %d\n", class);
2831 return -EINVAL;
2832 }
2833
2834 insn_idx++;
2835 }
2836
2837 verbose("processed %d insns\n", insn_processed);
2838 return 0;
2839 }
2840
2841 static int check_map_prog_compatibility(struct bpf_map *map,
2842 struct bpf_prog *prog)
2843
2844 {
2845 if (prog->type == BPF_PROG_TYPE_PERF_EVENT &&
2846 (map->map_type == BPF_MAP_TYPE_HASH ||
2847 map->map_type == BPF_MAP_TYPE_PERCPU_HASH) &&
2848 (map->map_flags & BPF_F_NO_PREALLOC)) {
2849 verbose("perf_event programs can only use preallocated hash map\n");
2850 return -EINVAL;
2851 }
2852 return 0;
2853 }
2854
2855 /* look for pseudo eBPF instructions that access map FDs and
2856 * replace them with actual map pointers
2857 */
2858 static int replace_map_fd_with_map_ptr(struct bpf_verifier_env *env)
2859 {
2860 struct bpf_insn *insn = env->prog->insnsi;
2861 int insn_cnt = env->prog->len;
2862 int i, j, err;
2863
2864 for (i = 0; i < insn_cnt; i++, insn++) {
2865 if (BPF_CLASS(insn->code) == BPF_LDX &&
2866 (BPF_MODE(insn->code) != BPF_MEM || insn->imm != 0)) {
2867 verbose("BPF_LDX uses reserved fields\n");
2868 return -EINVAL;
2869 }
2870
2871 if (BPF_CLASS(insn->code) == BPF_STX &&
2872 ((BPF_MODE(insn->code) != BPF_MEM &&
2873 BPF_MODE(insn->code) != BPF_XADD) || insn->imm != 0)) {
2874 verbose("BPF_STX uses reserved fields\n");
2875 return -EINVAL;
2876 }
2877
2878 if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW)) {
2879 struct bpf_map *map;
2880 struct fd f;
2881
2882 if (i == insn_cnt - 1 || insn[1].code != 0 ||
2883 insn[1].dst_reg != 0 || insn[1].src_reg != 0 ||
2884 insn[1].off != 0) {
2885 verbose("invalid bpf_ld_imm64 insn\n");
2886 return -EINVAL;
2887 }
2888
2889 if (insn->src_reg == 0)
2890 /* valid generic load 64-bit imm */
2891 goto next_insn;
2892
2893 if (insn->src_reg != BPF_PSEUDO_MAP_FD) {
2894 verbose("unrecognized bpf_ld_imm64 insn\n");
2895 return -EINVAL;
2896 }
2897
2898 f = fdget(insn->imm);
2899 map = __bpf_map_get(f);
2900 if (IS_ERR(map)) {
2901 verbose("fd %d is not pointing to valid bpf_map\n",
2902 insn->imm);
2903 return PTR_ERR(map);
2904 }
2905
2906 err = check_map_prog_compatibility(map, env->prog);
2907 if (err) {
2908 fdput(f);
2909 return err;
2910 }
2911
2912 /* store map pointer inside BPF_LD_IMM64 instruction */
2913 insn[0].imm = (u32) (unsigned long) map;
2914 insn[1].imm = ((u64) (unsigned long) map) >> 32;
2915
2916 /* check whether we recorded this map already */
2917 for (j = 0; j < env->used_map_cnt; j++)
2918 if (env->used_maps[j] == map) {
2919 fdput(f);
2920 goto next_insn;
2921 }
2922
2923 if (env->used_map_cnt >= MAX_USED_MAPS) {
2924 fdput(f);
2925 return -E2BIG;
2926 }
2927
2928 /* hold the map. If the program is rejected by verifier,
2929 * the map will be released by release_maps() or it
2930 * will be used by the valid program until it's unloaded
2931 * and all maps are released in free_bpf_prog_info()
2932 */
2933 map = bpf_map_inc(map, false);
2934 if (IS_ERR(map)) {
2935 fdput(f);
2936 return PTR_ERR(map);
2937 }
2938 env->used_maps[env->used_map_cnt++] = map;
2939
2940 fdput(f);
2941 next_insn:
2942 insn++;
2943 i++;
2944 }
2945 }
2946
2947 /* now all pseudo BPF_LD_IMM64 instructions load valid
2948 * 'struct bpf_map *' into a register instead of user map_fd.
2949 * These pointers will be used later by verifier to validate map access.
2950 */
2951 return 0;
2952 }
2953
2954 /* drop refcnt of maps used by the rejected program */
2955 static void release_maps(struct bpf_verifier_env *env)
2956 {
2957 int i;
2958
2959 for (i = 0; i < env->used_map_cnt; i++)
2960 bpf_map_put(env->used_maps[i]);
2961 }
2962
2963 /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */
2964 static void convert_pseudo_ld_imm64(struct bpf_verifier_env *env)
2965 {
2966 struct bpf_insn *insn = env->prog->insnsi;
2967 int insn_cnt = env->prog->len;
2968 int i;
2969
2970 for (i = 0; i < insn_cnt; i++, insn++)
2971 if (insn->code == (BPF_LD | BPF_IMM | BPF_DW))
2972 insn->src_reg = 0;
2973 }
2974
2975 /* convert load instructions that access fields of 'struct __sk_buff'
2976 * into sequence of instructions that access fields of 'struct sk_buff'
2977 */
2978 static int convert_ctx_accesses(struct bpf_verifier_env *env)
2979 {
2980 const struct bpf_verifier_ops *ops = env->prog->aux->ops;
2981 const int insn_cnt = env->prog->len;
2982 struct bpf_insn insn_buf[16], *insn;
2983 struct bpf_prog *new_prog;
2984 enum bpf_access_type type;
2985 int i, cnt, delta = 0;
2986
2987 if (ops->gen_prologue) {
2988 cnt = ops->gen_prologue(insn_buf, env->seen_direct_write,
2989 env->prog);
2990 if (cnt >= ARRAY_SIZE(insn_buf)) {
2991 verbose("bpf verifier is misconfigured\n");
2992 return -EINVAL;
2993 } else if (cnt) {
2994 new_prog = bpf_patch_insn_single(env->prog, 0,
2995 insn_buf, cnt);
2996 if (!new_prog)
2997 return -ENOMEM;
2998 env->prog = new_prog;
2999 delta += cnt - 1;
3000 }
3001 }
3002
3003 if (!ops->convert_ctx_access)
3004 return 0;
3005
3006 insn = env->prog->insnsi + delta;
3007
3008 for (i = 0; i < insn_cnt; i++, insn++) {
3009 if (insn->code == (BPF_LDX | BPF_MEM | BPF_W) ||
3010 insn->code == (BPF_LDX | BPF_MEM | BPF_DW))
3011 type = BPF_READ;
3012 else if (insn->code == (BPF_STX | BPF_MEM | BPF_W) ||
3013 insn->code == (BPF_STX | BPF_MEM | BPF_DW))
3014 type = BPF_WRITE;
3015 else
3016 continue;
3017
3018 if (env->insn_aux_data[i].ptr_type != PTR_TO_CTX)
3019 continue;
3020
3021 cnt = ops->convert_ctx_access(type, insn->dst_reg, insn->src_reg,
3022 insn->off, insn_buf, env->prog);
3023 if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) {
3024 verbose("bpf verifier is misconfigured\n");
3025 return -EINVAL;
3026 }
3027
3028 new_prog = bpf_patch_insn_single(env->prog, i + delta, insn_buf,
3029 cnt);
3030 if (!new_prog)
3031 return -ENOMEM;
3032
3033 delta += cnt - 1;
3034
3035 /* keep walking new program and skip insns we just inserted */
3036 env->prog = new_prog;
3037 insn = new_prog->insnsi + i + delta;
3038 }
3039
3040 return 0;
3041 }
3042
3043 static void free_states(struct bpf_verifier_env *env)
3044 {
3045 struct bpf_verifier_state_list *sl, *sln;
3046 int i;
3047
3048 if (!env->explored_states)
3049 return;
3050
3051 for (i = 0; i < env->prog->len; i++) {
3052 sl = env->explored_states[i];
3053
3054 if (sl)
3055 while (sl != STATE_LIST_MARK) {
3056 sln = sl->next;
3057 kfree(sl);
3058 sl = sln;
3059 }
3060 }
3061
3062 kfree(env->explored_states);
3063 }
3064
3065 int bpf_check(struct bpf_prog **prog, union bpf_attr *attr)
3066 {
3067 char __user *log_ubuf = NULL;
3068 struct bpf_verifier_env *env;
3069 int ret = -EINVAL;
3070
3071 if ((*prog)->len <= 0 || (*prog)->len > BPF_MAXINSNS)
3072 return -E2BIG;
3073
3074 /* 'struct bpf_verifier_env' can be global, but since it's not small,
3075 * allocate/free it every time bpf_check() is called
3076 */
3077 env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL);
3078 if (!env)
3079 return -ENOMEM;
3080
3081 env->insn_aux_data = vzalloc(sizeof(struct bpf_insn_aux_data) *
3082 (*prog)->len);
3083 ret = -ENOMEM;
3084 if (!env->insn_aux_data)
3085 goto err_free_env;
3086 env->prog = *prog;
3087
3088 /* grab the mutex to protect few globals used by verifier */
3089 mutex_lock(&bpf_verifier_lock);
3090
3091 if (attr->log_level || attr->log_buf || attr->log_size) {
3092 /* user requested verbose verifier output
3093 * and supplied buffer to store the verification trace
3094 */
3095 log_level = attr->log_level;
3096 log_ubuf = (char __user *) (unsigned long) attr->log_buf;
3097 log_size = attr->log_size;
3098 log_len = 0;
3099
3100 ret = -EINVAL;
3101 /* log_* values have to be sane */
3102 if (log_size < 128 || log_size > UINT_MAX >> 8 ||
3103 log_level == 0 || log_ubuf == NULL)
3104 goto err_unlock;
3105
3106 ret = -ENOMEM;
3107 log_buf = vmalloc(log_size);
3108 if (!log_buf)
3109 goto err_unlock;
3110 } else {
3111 log_level = 0;
3112 }
3113
3114 ret = replace_map_fd_with_map_ptr(env);
3115 if (ret < 0)
3116 goto skip_full_check;
3117
3118 env->explored_states = kcalloc(env->prog->len,
3119 sizeof(struct bpf_verifier_state_list *),
3120 GFP_USER);
3121 ret = -ENOMEM;
3122 if (!env->explored_states)
3123 goto skip_full_check;
3124
3125 ret = check_cfg(env);
3126 if (ret < 0)
3127 goto skip_full_check;
3128
3129 env->allow_ptr_leaks = capable(CAP_SYS_ADMIN);
3130
3131 ret = do_check(env);
3132
3133 skip_full_check:
3134 while (pop_stack(env, NULL) >= 0);
3135 free_states(env);
3136
3137 if (ret == 0)
3138 /* program is valid, convert *(u32*)(ctx + off) accesses */
3139 ret = convert_ctx_accesses(env);
3140
3141 if (log_level && log_len >= log_size - 1) {
3142 BUG_ON(log_len >= log_size);
3143 /* verifier log exceeded user supplied buffer */
3144 ret = -ENOSPC;
3145 /* fall through to return what was recorded */
3146 }
3147
3148 /* copy verifier log back to user space including trailing zero */
3149 if (log_level && copy_to_user(log_ubuf, log_buf, log_len + 1) != 0) {
3150 ret = -EFAULT;
3151 goto free_log_buf;
3152 }
3153
3154 if (ret == 0 && env->used_map_cnt) {
3155 /* if program passed verifier, update used_maps in bpf_prog_info */
3156 env->prog->aux->used_maps = kmalloc_array(env->used_map_cnt,
3157 sizeof(env->used_maps[0]),
3158 GFP_KERNEL);
3159
3160 if (!env->prog->aux->used_maps) {
3161 ret = -ENOMEM;
3162 goto free_log_buf;
3163 }
3164
3165 memcpy(env->prog->aux->used_maps, env->used_maps,
3166 sizeof(env->used_maps[0]) * env->used_map_cnt);
3167 env->prog->aux->used_map_cnt = env->used_map_cnt;
3168
3169 /* program is valid. Convert pseudo bpf_ld_imm64 into generic
3170 * bpf_ld_imm64 instructions
3171 */
3172 convert_pseudo_ld_imm64(env);
3173 }
3174
3175 free_log_buf:
3176 if (log_level)
3177 vfree(log_buf);
3178 if (!env->prog->aux->used_maps)
3179 /* if we didn't copy map pointers into bpf_prog_info, release
3180 * them now. Otherwise free_bpf_prog_info() will release them.
3181 */
3182 release_maps(env);
3183 *prog = env->prog;
3184 err_unlock:
3185 mutex_unlock(&bpf_verifier_lock);
3186 vfree(env->insn_aux_data);
3187 err_free_env:
3188 kfree(env);
3189 return ret;
3190 }
3191
3192 int bpf_analyzer(struct bpf_prog *prog, const struct bpf_ext_analyzer_ops *ops,
3193 void *priv)
3194 {
3195 struct bpf_verifier_env *env;
3196 int ret;
3197
3198 env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL);
3199 if (!env)
3200 return -ENOMEM;
3201
3202 env->insn_aux_data = vzalloc(sizeof(struct bpf_insn_aux_data) *
3203 prog->len);
3204 ret = -ENOMEM;
3205 if (!env->insn_aux_data)
3206 goto err_free_env;
3207 env->prog = prog;
3208 env->analyzer_ops = ops;
3209 env->analyzer_priv = priv;
3210
3211 /* grab the mutex to protect few globals used by verifier */
3212 mutex_lock(&bpf_verifier_lock);
3213
3214 log_level = 0;
3215
3216 env->explored_states = kcalloc(env->prog->len,
3217 sizeof(struct bpf_verifier_state_list *),
3218 GFP_KERNEL);
3219 ret = -ENOMEM;
3220 if (!env->explored_states)
3221 goto skip_full_check;
3222
3223 ret = check_cfg(env);
3224 if (ret < 0)
3225 goto skip_full_check;
3226
3227 env->allow_ptr_leaks = capable(CAP_SYS_ADMIN);
3228
3229 ret = do_check(env);
3230
3231 skip_full_check:
3232 while (pop_stack(env, NULL) >= 0);
3233 free_states(env);
3234
3235 mutex_unlock(&bpf_verifier_lock);
3236 vfree(env->insn_aux_data);
3237 err_free_env:
3238 kfree(env);
3239 return ret;
3240 }
3241 EXPORT_SYMBOL_GPL(bpf_analyzer);
ubuntu-bionic-kernel mirror