]> 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
bpf: Add hash of maps support
[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 #include <linux/stringify.h>
23
24 /* bpf_check() is a static code analyzer that walks eBPF program
25 * instruction by instruction and updates register/stack state.
26 * All paths of conditional branches are analyzed until 'bpf_exit' insn.
27 *
28 * The first pass is depth-first-search to check that the program is a DAG.
29 * It rejects the following programs:
30 * - larger than BPF_MAXINSNS insns
31 * - if loop is present (detected via back-edge)
32 * - unreachable insns exist (shouldn't be a forest. program = one function)
33 * - out of bounds or malformed jumps
34 * The second pass is all possible path descent from the 1st insn.
35 * Since it's analyzing all pathes through the program, the length of the
36 * analysis is limited to 64k insn, which may be hit even if total number of
37 * insn is less then 4K, but there are too many branches that change stack/regs.
38 * Number of 'branches to be analyzed' is limited to 1k
39 *
40 * On entry to each instruction, each register has a type, and the instruction
41 * changes the types of the registers depending on instruction semantics.
42 * If instruction is BPF_MOV64_REG(BPF_REG_1, BPF_REG_5), then type of R5 is
43 * copied to R1.
44 *
45 * All registers are 64-bit.
46 * R0 - return register
47 * R1-R5 argument passing registers
48 * R6-R9 callee saved registers
49 * R10 - frame pointer read-only
50 *
51 * At the start of BPF program the register R1 contains a pointer to bpf_context
52 * and has type PTR_TO_CTX.
53 *
54 * Verifier tracks arithmetic operations on pointers in case:
55 * BPF_MOV64_REG(BPF_REG_1, BPF_REG_10),
56 * BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -20),
57 * 1st insn copies R10 (which has FRAME_PTR) type into R1
58 * and 2nd arithmetic instruction is pattern matched to recognize
59 * that it wants to construct a pointer to some element within stack.
60 * So after 2nd insn, the register R1 has type PTR_TO_STACK
61 * (and -20 constant is saved for further stack bounds checking).
62 * Meaning that this reg is a pointer to stack plus known immediate constant.
63 *
64 * Most of the time the registers have UNKNOWN_VALUE type, which
65 * means the register has some value, but it's not a valid pointer.
66 * (like pointer plus pointer becomes UNKNOWN_VALUE type)
67 *
68 * When verifier sees load or store instructions the type of base register
69 * can be: PTR_TO_MAP_VALUE, PTR_TO_CTX, FRAME_PTR. These are three pointer
70 * types recognized by check_mem_access() function.
71 *
72 * PTR_TO_MAP_VALUE means that this register is pointing to 'map element value'
73 * and the range of [ptr, ptr + map's value_size) is accessible.
74 *
75 * registers used to pass values to function calls are checked against
76 * function argument constraints.
77 *
78 * ARG_PTR_TO_MAP_KEY is one of such argument constraints.
79 * It means that the register type passed to this function must be
80 * PTR_TO_STACK and it will be used inside the function as
81 * 'pointer to map element key'
82 *
83 * For example the argument constraints for bpf_map_lookup_elem():
84 * .ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL,
85 * .arg1_type = ARG_CONST_MAP_PTR,
86 * .arg2_type = ARG_PTR_TO_MAP_KEY,
87 *
88 * ret_type says that this function returns 'pointer to map elem value or null'
89 * function expects 1st argument to be a const pointer to 'struct bpf_map' and
90 * 2nd argument should be a pointer to stack, which will be used inside
91 * the helper function as a pointer to map element key.
92 *
93 * On the kernel side the helper function looks like:
94 * u64 bpf_map_lookup_elem(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
95 * {
96 * struct bpf_map *map = (struct bpf_map *) (unsigned long) r1;
97 * void *key = (void *) (unsigned long) r2;
98 * void *value;
99 *
100 * here kernel can access 'key' and 'map' pointers safely, knowing that
101 * [key, key + map->key_size) bytes are valid and were initialized on
102 * the stack of eBPF program.
103 * }
104 *
105 * Corresponding eBPF program may look like:
106 * BPF_MOV64_REG(BPF_REG_2, BPF_REG_10), // after this insn R2 type is FRAME_PTR
107 * BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -4), // after this insn R2 type is PTR_TO_STACK
108 * BPF_LD_MAP_FD(BPF_REG_1, map_fd), // after this insn R1 type is CONST_PTR_TO_MAP
109 * BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem),
110 * here verifier looks at prototype of map_lookup_elem() and sees:
111 * .arg1_type == ARG_CONST_MAP_PTR and R1->type == CONST_PTR_TO_MAP, which is ok,
112 * Now verifier knows that this map has key of R1->map_ptr->key_size bytes
113 *
114 * Then .arg2_type == ARG_PTR_TO_MAP_KEY and R2->type == PTR_TO_STACK, ok so far,
115 * Now verifier checks that [R2, R2 + map's key_size) are within stack limits
116 * and were initialized prior to this call.
117 * If it's ok, then verifier allows this BPF_CALL insn and looks at
118 * .ret_type which is RET_PTR_TO_MAP_VALUE_OR_NULL, so it sets
119 * R0->type = PTR_TO_MAP_VALUE_OR_NULL which means bpf_map_lookup_elem() function
120 * returns ether pointer to map value or NULL.
121 *
122 * When type PTR_TO_MAP_VALUE_OR_NULL passes through 'if (reg != 0) goto +off'
123 * insn, the register holding that pointer in the true branch changes state to
124 * PTR_TO_MAP_VALUE and the same register changes state to CONST_IMM in the false
125 * branch. See check_cond_jmp_op().
126 *
127 * After the call R0 is set to return type of the function and registers R1-R5
128 * are set to NOT_INIT to indicate that they are no longer readable.
129 */
130
131 /* verifier_state + insn_idx are pushed to stack when branch is encountered */
132 struct bpf_verifier_stack_elem {
133 /* verifer state is 'st'
134 * before processing instruction 'insn_idx'
135 * and after processing instruction 'prev_insn_idx'
136 */
137 struct bpf_verifier_state st;
138 int insn_idx;
139 int prev_insn_idx;
140 struct bpf_verifier_stack_elem *next;
141 };
142
143 #define BPF_COMPLEXITY_LIMIT_INSNS 65536
144 #define BPF_COMPLEXITY_LIMIT_STACK 1024
145
146 #define BPF_MAP_PTR_POISON ((void *)0xeB9F + POISON_POINTER_DELTA)
147
148 struct bpf_call_arg_meta {
149 struct bpf_map *map_ptr;
150 bool raw_mode;
151 bool pkt_access;
152 int regno;
153 int access_size;
154 };
155
156 /* verbose verifier prints what it's seeing
157 * bpf_check() is called under lock, so no race to access these global vars
158 */
159 static u32 log_level, log_size, log_len;
160 static char *log_buf;
161
162 static DEFINE_MUTEX(bpf_verifier_lock);
163
164 /* log_level controls verbosity level of eBPF verifier.
165 * verbose() is used to dump the verification trace to the log, so the user
166 * can figure out what's wrong with the program
167 */
168 static __printf(1, 2) void verbose(const char *fmt, ...)
169 {
170 va_list args;
171
172 if (log_level == 0 || log_len >= log_size - 1)
173 return;
174
175 va_start(args, fmt);
176 log_len += vscnprintf(log_buf + log_len, log_size - log_len, fmt, args);
177 va_end(args);
178 }
179
180 /* string representation of 'enum bpf_reg_type' */
181 static const char * const reg_type_str[] = {
182 [NOT_INIT] = "?",
183 [UNKNOWN_VALUE] = "inv",
184 [PTR_TO_CTX] = "ctx",
185 [CONST_PTR_TO_MAP] = "map_ptr",
186 [PTR_TO_MAP_VALUE] = "map_value",
187 [PTR_TO_MAP_VALUE_OR_NULL] = "map_value_or_null",
188 [PTR_TO_MAP_VALUE_ADJ] = "map_value_adj",
189 [FRAME_PTR] = "fp",
190 [PTR_TO_STACK] = "fp",
191 [CONST_IMM] = "imm",
192 [PTR_TO_PACKET] = "pkt",
193 [PTR_TO_PACKET_END] = "pkt_end",
194 };
195
196 #define __BPF_FUNC_STR_FN(x) [BPF_FUNC_ ## x] = __stringify(bpf_ ## x)
197 static const char * const func_id_str[] = {
198 __BPF_FUNC_MAPPER(__BPF_FUNC_STR_FN)
199 };
200 #undef __BPF_FUNC_STR_FN
201
202 static const char *func_id_name(int id)
203 {
204 BUILD_BUG_ON(ARRAY_SIZE(func_id_str) != __BPF_FUNC_MAX_ID);
205
206 if (id >= 0 && id < __BPF_FUNC_MAX_ID && func_id_str[id])
207 return func_id_str[id];
208 else
209 return "unknown";
210 }
211
212 static void print_verifier_state(struct bpf_verifier_state *state)
213 {
214 struct bpf_reg_state *reg;
215 enum bpf_reg_type t;
216 int i;
217
218 for (i = 0; i < MAX_BPF_REG; i++) {
219 reg = &state->regs[i];
220 t = reg->type;
221 if (t == NOT_INIT)
222 continue;
223 verbose(" R%d=%s", i, reg_type_str[t]);
224 if (t == CONST_IMM || t == PTR_TO_STACK)
225 verbose("%lld", reg->imm);
226 else if (t == PTR_TO_PACKET)
227 verbose("(id=%d,off=%d,r=%d)",
228 reg->id, reg->off, reg->range);
229 else if (t == UNKNOWN_VALUE && reg->imm)
230 verbose("%lld", reg->imm);
231 else if (t == CONST_PTR_TO_MAP || t == PTR_TO_MAP_VALUE ||
232 t == PTR_TO_MAP_VALUE_OR_NULL ||
233 t == PTR_TO_MAP_VALUE_ADJ)
234 verbose("(ks=%d,vs=%d,id=%u)",
235 reg->map_ptr->key_size,
236 reg->map_ptr->value_size,
237 reg->id);
238 if (reg->min_value != BPF_REGISTER_MIN_RANGE)
239 verbose(",min_value=%lld",
240 (long long)reg->min_value);
241 if (reg->max_value != BPF_REGISTER_MAX_RANGE)
242 verbose(",max_value=%llu",
243 (unsigned long long)reg->max_value);
244 }
245 for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
246 if (state->stack_slot_type[i] == STACK_SPILL)
247 verbose(" fp%d=%s", -MAX_BPF_STACK + i,
248 reg_type_str[state->spilled_regs[i / BPF_REG_SIZE].type]);
249 }
250 verbose("\n");
251 }
252
253 static const char *const bpf_class_string[] = {
254 [BPF_LD] = "ld",
255 [BPF_LDX] = "ldx",
256 [BPF_ST] = "st",
257 [BPF_STX] = "stx",
258 [BPF_ALU] = "alu",
259 [BPF_JMP] = "jmp",
260 [BPF_RET] = "BUG",
261 [BPF_ALU64] = "alu64",
262 };
263
264 static const char *const bpf_alu_string[16] = {
265 [BPF_ADD >> 4] = "+=",
266 [BPF_SUB >> 4] = "-=",
267 [BPF_MUL >> 4] = "*=",
268 [BPF_DIV >> 4] = "/=",
269 [BPF_OR >> 4] = "|=",
270 [BPF_AND >> 4] = "&=",
271 [BPF_LSH >> 4] = "<<=",
272 [BPF_RSH >> 4] = ">>=",
273 [BPF_NEG >> 4] = "neg",
274 [BPF_MOD >> 4] = "%=",
275 [BPF_XOR >> 4] = "^=",
276 [BPF_MOV >> 4] = "=",
277 [BPF_ARSH >> 4] = "s>>=",
278 [BPF_END >> 4] = "endian",
279 };
280
281 static const char *const bpf_ldst_string[] = {
282 [BPF_W >> 3] = "u32",
283 [BPF_H >> 3] = "u16",
284 [BPF_B >> 3] = "u8",
285 [BPF_DW >> 3] = "u64",
286 };
287
288 static const char *const bpf_jmp_string[16] = {
289 [BPF_JA >> 4] = "jmp",
290 [BPF_JEQ >> 4] = "==",
291 [BPF_JGT >> 4] = ">",
292 [BPF_JGE >> 4] = ">=",
293 [BPF_JSET >> 4] = "&",
294 [BPF_JNE >> 4] = "!=",
295 [BPF_JSGT >> 4] = "s>",
296 [BPF_JSGE >> 4] = "s>=",
297 [BPF_CALL >> 4] = "call",
298 [BPF_EXIT >> 4] = "exit",
299 };
300
301 static void print_bpf_insn(struct bpf_insn *insn)
302 {
303 u8 class = BPF_CLASS(insn->code);
304
305 if (class == BPF_ALU || class == BPF_ALU64) {
306 if (BPF_SRC(insn->code) == BPF_X)
307 verbose("(%02x) %sr%d %s %sr%d\n",
308 insn->code, class == BPF_ALU ? "(u32) " : "",
309 insn->dst_reg,
310 bpf_alu_string[BPF_OP(insn->code) >> 4],
311 class == BPF_ALU ? "(u32) " : "",
312 insn->src_reg);
313 else
314 verbose("(%02x) %sr%d %s %s%d\n",
315 insn->code, class == BPF_ALU ? "(u32) " : "",
316 insn->dst_reg,
317 bpf_alu_string[BPF_OP(insn->code) >> 4],
318 class == BPF_ALU ? "(u32) " : "",
319 insn->imm);
320 } else if (class == BPF_STX) {
321 if (BPF_MODE(insn->code) == BPF_MEM)
322 verbose("(%02x) *(%s *)(r%d %+d) = r%d\n",
323 insn->code,
324 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
325 insn->dst_reg,
326 insn->off, insn->src_reg);
327 else if (BPF_MODE(insn->code) == BPF_XADD)
328 verbose("(%02x) lock *(%s *)(r%d %+d) += r%d\n",
329 insn->code,
330 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
331 insn->dst_reg, insn->off,
332 insn->src_reg);
333 else
334 verbose("BUG_%02x\n", insn->code);
335 } else if (class == BPF_ST) {
336 if (BPF_MODE(insn->code) != BPF_MEM) {
337 verbose("BUG_st_%02x\n", insn->code);
338 return;
339 }
340 verbose("(%02x) *(%s *)(r%d %+d) = %d\n",
341 insn->code,
342 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
343 insn->dst_reg,
344 insn->off, insn->imm);
345 } else if (class == BPF_LDX) {
346 if (BPF_MODE(insn->code) != BPF_MEM) {
347 verbose("BUG_ldx_%02x\n", insn->code);
348 return;
349 }
350 verbose("(%02x) r%d = *(%s *)(r%d %+d)\n",
351 insn->code, insn->dst_reg,
352 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
353 insn->src_reg, insn->off);
354 } else if (class == BPF_LD) {
355 if (BPF_MODE(insn->code) == BPF_ABS) {
356 verbose("(%02x) r0 = *(%s *)skb[%d]\n",
357 insn->code,
358 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
359 insn->imm);
360 } else if (BPF_MODE(insn->code) == BPF_IND) {
361 verbose("(%02x) r0 = *(%s *)skb[r%d + %d]\n",
362 insn->code,
363 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
364 insn->src_reg, insn->imm);
365 } else if (BPF_MODE(insn->code) == BPF_IMM) {
366 verbose("(%02x) r%d = 0x%x\n",
367 insn->code, insn->dst_reg, insn->imm);
368 } else {
369 verbose("BUG_ld_%02x\n", insn->code);
370 return;
371 }
372 } else if (class == BPF_JMP) {
373 u8 opcode = BPF_OP(insn->code);
374
375 if (opcode == BPF_CALL) {
376 verbose("(%02x) call %s#%d\n", insn->code,
377 func_id_name(insn->imm), insn->imm);
378 } else if (insn->code == (BPF_JMP | BPF_JA)) {
379 verbose("(%02x) goto pc%+d\n",
380 insn->code, insn->off);
381 } else if (insn->code == (BPF_JMP | BPF_EXIT)) {
382 verbose("(%02x) exit\n", insn->code);
383 } else if (BPF_SRC(insn->code) == BPF_X) {
384 verbose("(%02x) if r%d %s r%d goto pc%+d\n",
385 insn->code, insn->dst_reg,
386 bpf_jmp_string[BPF_OP(insn->code) >> 4],
387 insn->src_reg, insn->off);
388 } else {
389 verbose("(%02x) if r%d %s 0x%x goto pc%+d\n",
390 insn->code, insn->dst_reg,
391 bpf_jmp_string[BPF_OP(insn->code) >> 4],
392 insn->imm, insn->off);
393 }
394 } else {
395 verbose("(%02x) %s\n", insn->code, bpf_class_string[class]);
396 }
397 }
398
399 static int pop_stack(struct bpf_verifier_env *env, int *prev_insn_idx)
400 {
401 struct bpf_verifier_stack_elem *elem;
402 int insn_idx;
403
404 if (env->head == NULL)
405 return -1;
406
407 memcpy(&env->cur_state, &env->head->st, sizeof(env->cur_state));
408 insn_idx = env->head->insn_idx;
409 if (prev_insn_idx)
410 *prev_insn_idx = env->head->prev_insn_idx;
411 elem = env->head->next;
412 kfree(env->head);
413 env->head = elem;
414 env->stack_size--;
415 return insn_idx;
416 }
417
418 static struct bpf_verifier_state *push_stack(struct bpf_verifier_env *env,
419 int insn_idx, int prev_insn_idx)
420 {
421 struct bpf_verifier_stack_elem *elem;
422
423 elem = kmalloc(sizeof(struct bpf_verifier_stack_elem), GFP_KERNEL);
424 if (!elem)
425 goto err;
426
427 memcpy(&elem->st, &env->cur_state, sizeof(env->cur_state));
428 elem->insn_idx = insn_idx;
429 elem->prev_insn_idx = prev_insn_idx;
430 elem->next = env->head;
431 env->head = elem;
432 env->stack_size++;
433 if (env->stack_size > BPF_COMPLEXITY_LIMIT_STACK) {
434 verbose("BPF program is too complex\n");
435 goto err;
436 }
437 return &elem->st;
438 err:
439 /* pop all elements and return */
440 while (pop_stack(env, NULL) >= 0);
441 return NULL;
442 }
443
444 #define CALLER_SAVED_REGS 6
445 static const int caller_saved[CALLER_SAVED_REGS] = {
446 BPF_REG_0, BPF_REG_1, BPF_REG_2, BPF_REG_3, BPF_REG_4, BPF_REG_5
447 };
448
449 static void init_reg_state(struct bpf_reg_state *regs)
450 {
451 int i;
452
453 for (i = 0; i < MAX_BPF_REG; i++) {
454 regs[i].type = NOT_INIT;
455 regs[i].imm = 0;
456 regs[i].min_value = BPF_REGISTER_MIN_RANGE;
457 regs[i].max_value = BPF_REGISTER_MAX_RANGE;
458 }
459
460 /* frame pointer */
461 regs[BPF_REG_FP].type = FRAME_PTR;
462
463 /* 1st arg to a function */
464 regs[BPF_REG_1].type = PTR_TO_CTX;
465 }
466
467 static void __mark_reg_unknown_value(struct bpf_reg_state *regs, u32 regno)
468 {
469 regs[regno].type = UNKNOWN_VALUE;
470 regs[regno].id = 0;
471 regs[regno].imm = 0;
472 }
473
474 static void mark_reg_unknown_value(struct bpf_reg_state *regs, u32 regno)
475 {
476 BUG_ON(regno >= MAX_BPF_REG);
477 __mark_reg_unknown_value(regs, regno);
478 }
479
480 static void reset_reg_range_values(struct bpf_reg_state *regs, u32 regno)
481 {
482 regs[regno].min_value = BPF_REGISTER_MIN_RANGE;
483 regs[regno].max_value = BPF_REGISTER_MAX_RANGE;
484 }
485
486 static void mark_reg_unknown_value_and_range(struct bpf_reg_state *regs,
487 u32 regno)
488 {
489 mark_reg_unknown_value(regs, regno);
490 reset_reg_range_values(regs, regno);
491 }
492
493 enum reg_arg_type {
494 SRC_OP, /* register is used as source operand */
495 DST_OP, /* register is used as destination operand */
496 DST_OP_NO_MARK /* same as above, check only, don't mark */
497 };
498
499 static int check_reg_arg(struct bpf_reg_state *regs, u32 regno,
500 enum reg_arg_type t)
501 {
502 if (regno >= MAX_BPF_REG) {
503 verbose("R%d is invalid\n", regno);
504 return -EINVAL;
505 }
506
507 if (t == SRC_OP) {
508 /* check whether register used as source operand can be read */
509 if (regs[regno].type == NOT_INIT) {
510 verbose("R%d !read_ok\n", regno);
511 return -EACCES;
512 }
513 } else {
514 /* check whether register used as dest operand can be written to */
515 if (regno == BPF_REG_FP) {
516 verbose("frame pointer is read only\n");
517 return -EACCES;
518 }
519 if (t == DST_OP)
520 mark_reg_unknown_value(regs, regno);
521 }
522 return 0;
523 }
524
525 static int bpf_size_to_bytes(int bpf_size)
526 {
527 if (bpf_size == BPF_W)
528 return 4;
529 else if (bpf_size == BPF_H)
530 return 2;
531 else if (bpf_size == BPF_B)
532 return 1;
533 else if (bpf_size == BPF_DW)
534 return 8;
535 else
536 return -EINVAL;
537 }
538
539 static bool is_spillable_regtype(enum bpf_reg_type type)
540 {
541 switch (type) {
542 case PTR_TO_MAP_VALUE:
543 case PTR_TO_MAP_VALUE_OR_NULL:
544 case PTR_TO_MAP_VALUE_ADJ:
545 case PTR_TO_STACK:
546 case PTR_TO_CTX:
547 case PTR_TO_PACKET:
548 case PTR_TO_PACKET_END:
549 case FRAME_PTR:
550 case CONST_PTR_TO_MAP:
551 return true;
552 default:
553 return false;
554 }
555 }
556
557 /* check_stack_read/write functions track spill/fill of registers,
558 * stack boundary and alignment are checked in check_mem_access()
559 */
560 static int check_stack_write(struct bpf_verifier_state *state, int off,
561 int size, int value_regno)
562 {
563 int i;
564 /* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0,
565 * so it's aligned access and [off, off + size) are within stack limits
566 */
567
568 if (value_regno >= 0 &&
569 is_spillable_regtype(state->regs[value_regno].type)) {
570
571 /* register containing pointer is being spilled into stack */
572 if (size != BPF_REG_SIZE) {
573 verbose("invalid size of register spill\n");
574 return -EACCES;
575 }
576
577 /* save register state */
578 state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] =
579 state->regs[value_regno];
580
581 for (i = 0; i < BPF_REG_SIZE; i++)
582 state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_SPILL;
583 } else {
584 /* regular write of data into stack */
585 state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] =
586 (struct bpf_reg_state) {};
587
588 for (i = 0; i < size; i++)
589 state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_MISC;
590 }
591 return 0;
592 }
593
594 static int check_stack_read(struct bpf_verifier_state *state, int off, int size,
595 int value_regno)
596 {
597 u8 *slot_type;
598 int i;
599
600 slot_type = &state->stack_slot_type[MAX_BPF_STACK + off];
601
602 if (slot_type[0] == STACK_SPILL) {
603 if (size != BPF_REG_SIZE) {
604 verbose("invalid size of register spill\n");
605 return -EACCES;
606 }
607 for (i = 1; i < BPF_REG_SIZE; i++) {
608 if (slot_type[i] != STACK_SPILL) {
609 verbose("corrupted spill memory\n");
610 return -EACCES;
611 }
612 }
613
614 if (value_regno >= 0)
615 /* restore register state from stack */
616 state->regs[value_regno] =
617 state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE];
618 return 0;
619 } else {
620 for (i = 0; i < size; i++) {
621 if (slot_type[i] != STACK_MISC) {
622 verbose("invalid read from stack off %d+%d size %d\n",
623 off, i, size);
624 return -EACCES;
625 }
626 }
627 if (value_regno >= 0)
628 /* have read misc data from the stack */
629 mark_reg_unknown_value_and_range(state->regs,
630 value_regno);
631 return 0;
632 }
633 }
634
635 /* check read/write into map element returned by bpf_map_lookup_elem() */
636 static int check_map_access(struct bpf_verifier_env *env, u32 regno, int off,
637 int size)
638 {
639 struct bpf_map *map = env->cur_state.regs[regno].map_ptr;
640
641 if (off < 0 || size <= 0 || off + size > map->value_size) {
642 verbose("invalid access to map value, value_size=%d off=%d size=%d\n",
643 map->value_size, off, size);
644 return -EACCES;
645 }
646 return 0;
647 }
648
649 /* check read/write into an adjusted map element */
650 static int check_map_access_adj(struct bpf_verifier_env *env, u32 regno,
651 int off, int size)
652 {
653 struct bpf_verifier_state *state = &env->cur_state;
654 struct bpf_reg_state *reg = &state->regs[regno];
655 int err;
656
657 /* We adjusted the register to this map value, so we
658 * need to change off and size to min_value and max_value
659 * respectively to make sure our theoretical access will be
660 * safe.
661 */
662 if (log_level)
663 print_verifier_state(state);
664 env->varlen_map_value_access = true;
665 /* The minimum value is only important with signed
666 * comparisons where we can't assume the floor of a
667 * value is 0. If we are using signed variables for our
668 * index'es we need to make sure that whatever we use
669 * will have a set floor within our range.
670 */
671 if (reg->min_value < 0) {
672 verbose("R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n",
673 regno);
674 return -EACCES;
675 }
676 err = check_map_access(env, regno, reg->min_value + off, size);
677 if (err) {
678 verbose("R%d min value is outside of the array range\n",
679 regno);
680 return err;
681 }
682
683 /* If we haven't set a max value then we need to bail
684 * since we can't be sure we won't do bad things.
685 */
686 if (reg->max_value == BPF_REGISTER_MAX_RANGE) {
687 verbose("R%d unbounded memory access, make sure to bounds check any array access into a map\n",
688 regno);
689 return -EACCES;
690 }
691 return check_map_access(env, regno, reg->max_value + off, size);
692 }
693
694 #define MAX_PACKET_OFF 0xffff
695
696 static bool may_access_direct_pkt_data(struct bpf_verifier_env *env,
697 const struct bpf_call_arg_meta *meta,
698 enum bpf_access_type t)
699 {
700 switch (env->prog->type) {
701 case BPF_PROG_TYPE_LWT_IN:
702 case BPF_PROG_TYPE_LWT_OUT:
703 /* dst_input() and dst_output() can't write for now */
704 if (t == BPF_WRITE)
705 return false;
706 /* fallthrough */
707 case BPF_PROG_TYPE_SCHED_CLS:
708 case BPF_PROG_TYPE_SCHED_ACT:
709 case BPF_PROG_TYPE_XDP:
710 case BPF_PROG_TYPE_LWT_XMIT:
711 if (meta)
712 return meta->pkt_access;
713
714 env->seen_direct_write = true;
715 return true;
716 default:
717 return false;
718 }
719 }
720
721 static int check_packet_access(struct bpf_verifier_env *env, u32 regno, int off,
722 int size)
723 {
724 struct bpf_reg_state *regs = env->cur_state.regs;
725 struct bpf_reg_state *reg = &regs[regno];
726
727 off += reg->off;
728 if (off < 0 || size <= 0 || off + size > reg->range) {
729 verbose("invalid access to packet, off=%d size=%d, R%d(id=%d,off=%d,r=%d)\n",
730 off, size, regno, reg->id, reg->off, reg->range);
731 return -EACCES;
732 }
733 return 0;
734 }
735
736 /* check access to 'struct bpf_context' fields */
737 static int check_ctx_access(struct bpf_verifier_env *env, int off, int size,
738 enum bpf_access_type t, enum bpf_reg_type *reg_type)
739 {
740 /* for analyzer ctx accesses are already validated and converted */
741 if (env->analyzer_ops)
742 return 0;
743
744 if (env->prog->aux->ops->is_valid_access &&
745 env->prog->aux->ops->is_valid_access(off, size, t, reg_type)) {
746 /* remember the offset of last byte accessed in ctx */
747 if (env->prog->aux->max_ctx_offset < off + size)
748 env->prog->aux->max_ctx_offset = off + size;
749 return 0;
750 }
751
752 verbose("invalid bpf_context access off=%d size=%d\n", off, size);
753 return -EACCES;
754 }
755
756 static bool is_pointer_value(struct bpf_verifier_env *env, int regno)
757 {
758 if (env->allow_ptr_leaks)
759 return false;
760
761 switch (env->cur_state.regs[regno].type) {
762 case UNKNOWN_VALUE:
763 case CONST_IMM:
764 return false;
765 default:
766 return true;
767 }
768 }
769
770 static int check_ptr_alignment(struct bpf_verifier_env *env,
771 struct bpf_reg_state *reg, int off, int size)
772 {
773 if (reg->type != PTR_TO_PACKET && reg->type != PTR_TO_MAP_VALUE_ADJ) {
774 if (off % size != 0) {
775 verbose("misaligned access off %d size %d\n",
776 off, size);
777 return -EACCES;
778 } else {
779 return 0;
780 }
781 }
782
783 if (IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS))
784 /* misaligned access to packet is ok on x86,arm,arm64 */
785 return 0;
786
787 if (reg->id && size != 1) {
788 verbose("Unknown packet alignment. Only byte-sized access allowed\n");
789 return -EACCES;
790 }
791
792 /* skb->data is NET_IP_ALIGN-ed */
793 if (reg->type == PTR_TO_PACKET &&
794 (NET_IP_ALIGN + reg->off + off) % size != 0) {
795 verbose("misaligned packet access off %d+%d+%d size %d\n",
796 NET_IP_ALIGN, reg->off, off, size);
797 return -EACCES;
798 }
799 return 0;
800 }
801
802 /* check whether memory at (regno + off) is accessible for t = (read | write)
803 * if t==write, value_regno is a register which value is stored into memory
804 * if t==read, value_regno is a register which will receive the value from memory
805 * if t==write && value_regno==-1, some unknown value is stored into memory
806 * if t==read && value_regno==-1, don't care what we read from memory
807 */
808 static int check_mem_access(struct bpf_verifier_env *env, u32 regno, int off,
809 int bpf_size, enum bpf_access_type t,
810 int value_regno)
811 {
812 struct bpf_verifier_state *state = &env->cur_state;
813 struct bpf_reg_state *reg = &state->regs[regno];
814 int size, err = 0;
815
816 if (reg->type == PTR_TO_STACK)
817 off += reg->imm;
818
819 size = bpf_size_to_bytes(bpf_size);
820 if (size < 0)
821 return size;
822
823 err = check_ptr_alignment(env, reg, off, size);
824 if (err)
825 return err;
826
827 if (reg->type == PTR_TO_MAP_VALUE ||
828 reg->type == PTR_TO_MAP_VALUE_ADJ) {
829 if (t == BPF_WRITE && value_regno >= 0 &&
830 is_pointer_value(env, value_regno)) {
831 verbose("R%d leaks addr into map\n", value_regno);
832 return -EACCES;
833 }
834
835 if (reg->type == PTR_TO_MAP_VALUE_ADJ)
836 err = check_map_access_adj(env, regno, off, size);
837 else
838 err = check_map_access(env, regno, off, size);
839 if (!err && t == BPF_READ && value_regno >= 0)
840 mark_reg_unknown_value_and_range(state->regs,
841 value_regno);
842
843 } else if (reg->type == PTR_TO_CTX) {
844 enum bpf_reg_type reg_type = UNKNOWN_VALUE;
845
846 if (t == BPF_WRITE && value_regno >= 0 &&
847 is_pointer_value(env, value_regno)) {
848 verbose("R%d leaks addr into ctx\n", value_regno);
849 return -EACCES;
850 }
851 err = check_ctx_access(env, off, size, t, &reg_type);
852 if (!err && t == BPF_READ && value_regno >= 0) {
853 mark_reg_unknown_value_and_range(state->regs,
854 value_regno);
855 /* note that reg.[id|off|range] == 0 */
856 state->regs[value_regno].type = reg_type;
857 }
858
859 } else if (reg->type == FRAME_PTR || reg->type == PTR_TO_STACK) {
860 if (off >= 0 || off < -MAX_BPF_STACK) {
861 verbose("invalid stack off=%d size=%d\n", off, size);
862 return -EACCES;
863 }
864 if (t == BPF_WRITE) {
865 if (!env->allow_ptr_leaks &&
866 state->stack_slot_type[MAX_BPF_STACK + off] == STACK_SPILL &&
867 size != BPF_REG_SIZE) {
868 verbose("attempt to corrupt spilled pointer on stack\n");
869 return -EACCES;
870 }
871 err = check_stack_write(state, off, size, value_regno);
872 } else {
873 err = check_stack_read(state, off, size, value_regno);
874 }
875 } else if (state->regs[regno].type == PTR_TO_PACKET) {
876 if (t == BPF_WRITE && !may_access_direct_pkt_data(env, NULL, t)) {
877 verbose("cannot write into packet\n");
878 return -EACCES;
879 }
880 if (t == BPF_WRITE && value_regno >= 0 &&
881 is_pointer_value(env, value_regno)) {
882 verbose("R%d leaks addr into packet\n", value_regno);
883 return -EACCES;
884 }
885 err = check_packet_access(env, regno, off, size);
886 if (!err && t == BPF_READ && value_regno >= 0)
887 mark_reg_unknown_value_and_range(state->regs,
888 value_regno);
889 } else {
890 verbose("R%d invalid mem access '%s'\n",
891 regno, reg_type_str[reg->type]);
892 return -EACCES;
893 }
894
895 if (!err && size <= 2 && value_regno >= 0 && env->allow_ptr_leaks &&
896 state->regs[value_regno].type == UNKNOWN_VALUE) {
897 /* 1 or 2 byte load zero-extends, determine the number of
898 * zero upper bits. Not doing it fo 4 byte load, since
899 * such values cannot be added to ptr_to_packet anyway.
900 */
901 state->regs[value_regno].imm = 64 - size * 8;
902 }
903 return err;
904 }
905
906 static int check_xadd(struct bpf_verifier_env *env, struct bpf_insn *insn)
907 {
908 struct bpf_reg_state *regs = env->cur_state.regs;
909 int err;
910
911 if ((BPF_SIZE(insn->code) != BPF_W && BPF_SIZE(insn->code) != BPF_DW) ||
912 insn->imm != 0) {
913 verbose("BPF_XADD uses reserved fields\n");
914 return -EINVAL;
915 }
916
917 /* check src1 operand */
918 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
919 if (err)
920 return err;
921
922 /* check src2 operand */
923 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
924 if (err)
925 return err;
926
927 /* check whether atomic_add can read the memory */
928 err = check_mem_access(env, insn->dst_reg, insn->off,
929 BPF_SIZE(insn->code), BPF_READ, -1);
930 if (err)
931 return err;
932
933 /* check whether atomic_add can write into the same memory */
934 return check_mem_access(env, insn->dst_reg, insn->off,
935 BPF_SIZE(insn->code), BPF_WRITE, -1);
936 }
937
938 /* when register 'regno' is passed into function that will read 'access_size'
939 * bytes from that pointer, make sure that it's within stack boundary
940 * and all elements of stack are initialized
941 */
942 static int check_stack_boundary(struct bpf_verifier_env *env, int regno,
943 int access_size, bool zero_size_allowed,
944 struct bpf_call_arg_meta *meta)
945 {
946 struct bpf_verifier_state *state = &env->cur_state;
947 struct bpf_reg_state *regs = state->regs;
948 int off, i;
949
950 if (regs[regno].type != PTR_TO_STACK) {
951 if (zero_size_allowed && access_size == 0 &&
952 regs[regno].type == CONST_IMM &&
953 regs[regno].imm == 0)
954 return 0;
955
956 verbose("R%d type=%s expected=%s\n", regno,
957 reg_type_str[regs[regno].type],
958 reg_type_str[PTR_TO_STACK]);
959 return -EACCES;
960 }
961
962 off = regs[regno].imm;
963 if (off >= 0 || off < -MAX_BPF_STACK || off + access_size > 0 ||
964 access_size <= 0) {
965 verbose("invalid stack type R%d off=%d access_size=%d\n",
966 regno, off, access_size);
967 return -EACCES;
968 }
969
970 if (meta && meta->raw_mode) {
971 meta->access_size = access_size;
972 meta->regno = regno;
973 return 0;
974 }
975
976 for (i = 0; i < access_size; i++) {
977 if (state->stack_slot_type[MAX_BPF_STACK + off + i] != STACK_MISC) {
978 verbose("invalid indirect read from stack off %d+%d size %d\n",
979 off, i, access_size);
980 return -EACCES;
981 }
982 }
983 return 0;
984 }
985
986 static int check_helper_mem_access(struct bpf_verifier_env *env, int regno,
987 int access_size, bool zero_size_allowed,
988 struct bpf_call_arg_meta *meta)
989 {
990 struct bpf_reg_state *regs = env->cur_state.regs;
991
992 switch (regs[regno].type) {
993 case PTR_TO_PACKET:
994 return check_packet_access(env, regno, 0, access_size);
995 case PTR_TO_MAP_VALUE:
996 return check_map_access(env, regno, 0, access_size);
997 case PTR_TO_MAP_VALUE_ADJ:
998 return check_map_access_adj(env, regno, 0, access_size);
999 default: /* const_imm|ptr_to_stack or invalid ptr */
1000 return check_stack_boundary(env, regno, access_size,
1001 zero_size_allowed, meta);
1002 }
1003 }
1004
1005 static int check_func_arg(struct bpf_verifier_env *env, u32 regno,
1006 enum bpf_arg_type arg_type,
1007 struct bpf_call_arg_meta *meta)
1008 {
1009 struct bpf_reg_state *regs = env->cur_state.regs, *reg = &regs[regno];
1010 enum bpf_reg_type expected_type, type = reg->type;
1011 int err = 0;
1012
1013 if (arg_type == ARG_DONTCARE)
1014 return 0;
1015
1016 if (type == NOT_INIT) {
1017 verbose("R%d !read_ok\n", regno);
1018 return -EACCES;
1019 }
1020
1021 if (arg_type == ARG_ANYTHING) {
1022 if (is_pointer_value(env, regno)) {
1023 verbose("R%d leaks addr into helper function\n", regno);
1024 return -EACCES;
1025 }
1026 return 0;
1027 }
1028
1029 if (type == PTR_TO_PACKET &&
1030 !may_access_direct_pkt_data(env, meta, BPF_READ)) {
1031 verbose("helper access to the packet is not allowed\n");
1032 return -EACCES;
1033 }
1034
1035 if (arg_type == ARG_PTR_TO_MAP_KEY ||
1036 arg_type == ARG_PTR_TO_MAP_VALUE) {
1037 expected_type = PTR_TO_STACK;
1038 if (type != PTR_TO_PACKET && type != expected_type)
1039 goto err_type;
1040 } else if (arg_type == ARG_CONST_SIZE ||
1041 arg_type == ARG_CONST_SIZE_OR_ZERO) {
1042 expected_type = CONST_IMM;
1043 /* One exception. Allow UNKNOWN_VALUE registers when the
1044 * boundaries are known and don't cause unsafe memory accesses
1045 */
1046 if (type != UNKNOWN_VALUE && type != expected_type)
1047 goto err_type;
1048 } else if (arg_type == ARG_CONST_MAP_PTR) {
1049 expected_type = CONST_PTR_TO_MAP;
1050 if (type != expected_type)
1051 goto err_type;
1052 } else if (arg_type == ARG_PTR_TO_CTX) {
1053 expected_type = PTR_TO_CTX;
1054 if (type != expected_type)
1055 goto err_type;
1056 } else if (arg_type == ARG_PTR_TO_MEM ||
1057 arg_type == ARG_PTR_TO_UNINIT_MEM) {
1058 expected_type = PTR_TO_STACK;
1059 /* One exception here. In case function allows for NULL to be
1060 * passed in as argument, it's a CONST_IMM type. Final test
1061 * happens during stack boundary checking.
1062 */
1063 if (type == CONST_IMM && reg->imm == 0)
1064 /* final test in check_stack_boundary() */;
1065 else if (type != PTR_TO_PACKET && type != PTR_TO_MAP_VALUE &&
1066 type != PTR_TO_MAP_VALUE_ADJ && type != expected_type)
1067 goto err_type;
1068 meta->raw_mode = arg_type == ARG_PTR_TO_UNINIT_MEM;
1069 } else {
1070 verbose("unsupported arg_type %d\n", arg_type);
1071 return -EFAULT;
1072 }
1073
1074 if (arg_type == ARG_CONST_MAP_PTR) {
1075 /* bpf_map_xxx(map_ptr) call: remember that map_ptr */
1076 meta->map_ptr = reg->map_ptr;
1077 } else if (arg_type == ARG_PTR_TO_MAP_KEY) {
1078 /* bpf_map_xxx(..., map_ptr, ..., key) call:
1079 * check that [key, key + map->key_size) are within
1080 * stack limits and initialized
1081 */
1082 if (!meta->map_ptr) {
1083 /* in function declaration map_ptr must come before
1084 * map_key, so that it's verified and known before
1085 * we have to check map_key here. Otherwise it means
1086 * that kernel subsystem misconfigured verifier
1087 */
1088 verbose("invalid map_ptr to access map->key\n");
1089 return -EACCES;
1090 }
1091 if (type == PTR_TO_PACKET)
1092 err = check_packet_access(env, regno, 0,
1093 meta->map_ptr->key_size);
1094 else
1095 err = check_stack_boundary(env, regno,
1096 meta->map_ptr->key_size,
1097 false, NULL);
1098 } else if (arg_type == ARG_PTR_TO_MAP_VALUE) {
1099 /* bpf_map_xxx(..., map_ptr, ..., value) call:
1100 * check [value, value + map->value_size) validity
1101 */
1102 if (!meta->map_ptr) {
1103 /* kernel subsystem misconfigured verifier */
1104 verbose("invalid map_ptr to access map->value\n");
1105 return -EACCES;
1106 }
1107 if (type == PTR_TO_PACKET)
1108 err = check_packet_access(env, regno, 0,
1109 meta->map_ptr->value_size);
1110 else
1111 err = check_stack_boundary(env, regno,
1112 meta->map_ptr->value_size,
1113 false, NULL);
1114 } else if (arg_type == ARG_CONST_SIZE ||
1115 arg_type == ARG_CONST_SIZE_OR_ZERO) {
1116 bool zero_size_allowed = (arg_type == ARG_CONST_SIZE_OR_ZERO);
1117
1118 /* bpf_xxx(..., buf, len) call will access 'len' bytes
1119 * from stack pointer 'buf'. Check it
1120 * note: regno == len, regno - 1 == buf
1121 */
1122 if (regno == 0) {
1123 /* kernel subsystem misconfigured verifier */
1124 verbose("ARG_CONST_SIZE cannot be first argument\n");
1125 return -EACCES;
1126 }
1127
1128 /* If the register is UNKNOWN_VALUE, the access check happens
1129 * using its boundaries. Otherwise, just use its imm
1130 */
1131 if (type == UNKNOWN_VALUE) {
1132 /* For unprivileged variable accesses, disable raw
1133 * mode so that the program is required to
1134 * initialize all the memory that the helper could
1135 * just partially fill up.
1136 */
1137 meta = NULL;
1138
1139 if (reg->min_value < 0) {
1140 verbose("R%d min value is negative, either use unsigned or 'var &= const'\n",
1141 regno);
1142 return -EACCES;
1143 }
1144
1145 if (reg->min_value == 0) {
1146 err = check_helper_mem_access(env, regno - 1, 0,
1147 zero_size_allowed,
1148 meta);
1149 if (err)
1150 return err;
1151 }
1152
1153 if (reg->max_value == BPF_REGISTER_MAX_RANGE) {
1154 verbose("R%d unbounded memory access, use 'var &= const' or 'if (var < const)'\n",
1155 regno);
1156 return -EACCES;
1157 }
1158 err = check_helper_mem_access(env, regno - 1,
1159 reg->max_value,
1160 zero_size_allowed, meta);
1161 if (err)
1162 return err;
1163 } else {
1164 /* register is CONST_IMM */
1165 err = check_helper_mem_access(env, regno - 1, reg->imm,
1166 zero_size_allowed, meta);
1167 }
1168 }
1169
1170 return err;
1171 err_type:
1172 verbose("R%d type=%s expected=%s\n", regno,
1173 reg_type_str[type], reg_type_str[expected_type]);
1174 return -EACCES;
1175 }
1176
1177 static int check_map_func_compatibility(struct bpf_map *map, int func_id)
1178 {
1179 if (!map)
1180 return 0;
1181
1182 /* We need a two way check, first is from map perspective ... */
1183 switch (map->map_type) {
1184 case BPF_MAP_TYPE_PROG_ARRAY:
1185 if (func_id != BPF_FUNC_tail_call)
1186 goto error;
1187 break;
1188 case BPF_MAP_TYPE_PERF_EVENT_ARRAY:
1189 if (func_id != BPF_FUNC_perf_event_read &&
1190 func_id != BPF_FUNC_perf_event_output)
1191 goto error;
1192 break;
1193 case BPF_MAP_TYPE_STACK_TRACE:
1194 if (func_id != BPF_FUNC_get_stackid)
1195 goto error;
1196 break;
1197 case BPF_MAP_TYPE_CGROUP_ARRAY:
1198 if (func_id != BPF_FUNC_skb_under_cgroup &&
1199 func_id != BPF_FUNC_current_task_under_cgroup)
1200 goto error;
1201 break;
1202 case BPF_MAP_TYPE_ARRAY_OF_MAPS:
1203 case BPF_MAP_TYPE_HASH_OF_MAPS:
1204 if (func_id != BPF_FUNC_map_lookup_elem)
1205 goto error;
1206 default:
1207 break;
1208 }
1209
1210 /* ... and second from the function itself. */
1211 switch (func_id) {
1212 case BPF_FUNC_tail_call:
1213 if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY)
1214 goto error;
1215 break;
1216 case BPF_FUNC_perf_event_read:
1217 case BPF_FUNC_perf_event_output:
1218 if (map->map_type != BPF_MAP_TYPE_PERF_EVENT_ARRAY)
1219 goto error;
1220 break;
1221 case BPF_FUNC_get_stackid:
1222 if (map->map_type != BPF_MAP_TYPE_STACK_TRACE)
1223 goto error;
1224 break;
1225 case BPF_FUNC_current_task_under_cgroup:
1226 case BPF_FUNC_skb_under_cgroup:
1227 if (map->map_type != BPF_MAP_TYPE_CGROUP_ARRAY)
1228 goto error;
1229 break;
1230 default:
1231 break;
1232 }
1233
1234 return 0;
1235 error:
1236 verbose("cannot pass map_type %d into func %s#%d\n",
1237 map->map_type, func_id_name(func_id), func_id);
1238 return -EINVAL;
1239 }
1240
1241 static int check_raw_mode(const struct bpf_func_proto *fn)
1242 {
1243 int count = 0;
1244
1245 if (fn->arg1_type == ARG_PTR_TO_UNINIT_MEM)
1246 count++;
1247 if (fn->arg2_type == ARG_PTR_TO_UNINIT_MEM)
1248 count++;
1249 if (fn->arg3_type == ARG_PTR_TO_UNINIT_MEM)
1250 count++;
1251 if (fn->arg4_type == ARG_PTR_TO_UNINIT_MEM)
1252 count++;
1253 if (fn->arg5_type == ARG_PTR_TO_UNINIT_MEM)
1254 count++;
1255
1256 return count > 1 ? -EINVAL : 0;
1257 }
1258
1259 static void clear_all_pkt_pointers(struct bpf_verifier_env *env)
1260 {
1261 struct bpf_verifier_state *state = &env->cur_state;
1262 struct bpf_reg_state *regs = state->regs, *reg;
1263 int i;
1264
1265 for (i = 0; i < MAX_BPF_REG; i++)
1266 if (regs[i].type == PTR_TO_PACKET ||
1267 regs[i].type == PTR_TO_PACKET_END)
1268 mark_reg_unknown_value(regs, i);
1269
1270 for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
1271 if (state->stack_slot_type[i] != STACK_SPILL)
1272 continue;
1273 reg = &state->spilled_regs[i / BPF_REG_SIZE];
1274 if (reg->type != PTR_TO_PACKET &&
1275 reg->type != PTR_TO_PACKET_END)
1276 continue;
1277 reg->type = UNKNOWN_VALUE;
1278 reg->imm = 0;
1279 }
1280 }
1281
1282 static int check_call(struct bpf_verifier_env *env, int func_id, int insn_idx)
1283 {
1284 struct bpf_verifier_state *state = &env->cur_state;
1285 const struct bpf_func_proto *fn = NULL;
1286 struct bpf_reg_state *regs = state->regs;
1287 struct bpf_reg_state *reg;
1288 struct bpf_call_arg_meta meta;
1289 bool changes_data;
1290 int i, err;
1291
1292 /* find function prototype */
1293 if (func_id < 0 || func_id >= __BPF_FUNC_MAX_ID) {
1294 verbose("invalid func %s#%d\n", func_id_name(func_id), func_id);
1295 return -EINVAL;
1296 }
1297
1298 if (env->prog->aux->ops->get_func_proto)
1299 fn = env->prog->aux->ops->get_func_proto(func_id);
1300
1301 if (!fn) {
1302 verbose("unknown func %s#%d\n", func_id_name(func_id), func_id);
1303 return -EINVAL;
1304 }
1305
1306 /* eBPF programs must be GPL compatible to use GPL-ed functions */
1307 if (!env->prog->gpl_compatible && fn->gpl_only) {
1308 verbose("cannot call GPL only function from proprietary program\n");
1309 return -EINVAL;
1310 }
1311
1312 changes_data = bpf_helper_changes_pkt_data(fn->func);
1313
1314 memset(&meta, 0, sizeof(meta));
1315 meta.pkt_access = fn->pkt_access;
1316
1317 /* We only support one arg being in raw mode at the moment, which
1318 * is sufficient for the helper functions we have right now.
1319 */
1320 err = check_raw_mode(fn);
1321 if (err) {
1322 verbose("kernel subsystem misconfigured func %s#%d\n",
1323 func_id_name(func_id), func_id);
1324 return err;
1325 }
1326
1327 /* check args */
1328 err = check_func_arg(env, BPF_REG_1, fn->arg1_type, &meta);
1329 if (err)
1330 return err;
1331 err = check_func_arg(env, BPF_REG_2, fn->arg2_type, &meta);
1332 if (err)
1333 return err;
1334 err = check_func_arg(env, BPF_REG_3, fn->arg3_type, &meta);
1335 if (err)
1336 return err;
1337 err = check_func_arg(env, BPF_REG_4, fn->arg4_type, &meta);
1338 if (err)
1339 return err;
1340 err = check_func_arg(env, BPF_REG_5, fn->arg5_type, &meta);
1341 if (err)
1342 return err;
1343
1344 /* Mark slots with STACK_MISC in case of raw mode, stack offset
1345 * is inferred from register state.
1346 */
1347 for (i = 0; i < meta.access_size; i++) {
1348 err = check_mem_access(env, meta.regno, i, BPF_B, BPF_WRITE, -1);
1349 if (err)
1350 return err;
1351 }
1352
1353 /* reset caller saved regs */
1354 for (i = 0; i < CALLER_SAVED_REGS; i++) {
1355 reg = regs + caller_saved[i];
1356 reg->type = NOT_INIT;
1357 reg->imm = 0;
1358 }
1359
1360 /* update return register */
1361 if (fn->ret_type == RET_INTEGER) {
1362 regs[BPF_REG_0].type = UNKNOWN_VALUE;
1363 } else if (fn->ret_type == RET_VOID) {
1364 regs[BPF_REG_0].type = NOT_INIT;
1365 } else if (fn->ret_type == RET_PTR_TO_MAP_VALUE_OR_NULL) {
1366 struct bpf_insn_aux_data *insn_aux;
1367
1368 regs[BPF_REG_0].type = PTR_TO_MAP_VALUE_OR_NULL;
1369 regs[BPF_REG_0].max_value = regs[BPF_REG_0].min_value = 0;
1370 /* remember map_ptr, so that check_map_access()
1371 * can check 'value_size' boundary of memory access
1372 * to map element returned from bpf_map_lookup_elem()
1373 */
1374 if (meta.map_ptr == NULL) {
1375 verbose("kernel subsystem misconfigured verifier\n");
1376 return -EINVAL;
1377 }
1378 regs[BPF_REG_0].map_ptr = meta.map_ptr;
1379 regs[BPF_REG_0].id = ++env->id_gen;
1380 insn_aux = &env->insn_aux_data[insn_idx];
1381 if (!insn_aux->map_ptr)
1382 insn_aux->map_ptr = meta.map_ptr;
1383 else if (insn_aux->map_ptr != meta.map_ptr)
1384 insn_aux->map_ptr = BPF_MAP_PTR_POISON;
1385 } else {
1386 verbose("unknown return type %d of func %s#%d\n",
1387 fn->ret_type, func_id_name(func_id), func_id);
1388 return -EINVAL;
1389 }
1390
1391 err = check_map_func_compatibility(meta.map_ptr, func_id);
1392 if (err)
1393 return err;
1394
1395 if (changes_data)
1396 clear_all_pkt_pointers(env);
1397 return 0;
1398 }
1399
1400 static int check_packet_ptr_add(struct bpf_verifier_env *env,
1401 struct bpf_insn *insn)
1402 {
1403 struct bpf_reg_state *regs = env->cur_state.regs;
1404 struct bpf_reg_state *dst_reg = &regs[insn->dst_reg];
1405 struct bpf_reg_state *src_reg = &regs[insn->src_reg];
1406 struct bpf_reg_state tmp_reg;
1407 s32 imm;
1408
1409 if (BPF_SRC(insn->code) == BPF_K) {
1410 /* pkt_ptr += imm */
1411 imm = insn->imm;
1412
1413 add_imm:
1414 if (imm < 0) {
1415 verbose("addition of negative constant to packet pointer is not allowed\n");
1416 return -EACCES;
1417 }
1418 if (imm >= MAX_PACKET_OFF ||
1419 imm + dst_reg->off >= MAX_PACKET_OFF) {
1420 verbose("constant %d is too large to add to packet pointer\n",
1421 imm);
1422 return -EACCES;
1423 }
1424 /* a constant was added to pkt_ptr.
1425 * Remember it while keeping the same 'id'
1426 */
1427 dst_reg->off += imm;
1428 } else {
1429 if (src_reg->type == PTR_TO_PACKET) {
1430 /* R6=pkt(id=0,off=0,r=62) R7=imm22; r7 += r6 */
1431 tmp_reg = *dst_reg; /* save r7 state */
1432 *dst_reg = *src_reg; /* copy pkt_ptr state r6 into r7 */
1433 src_reg = &tmp_reg; /* pretend it's src_reg state */
1434 /* if the checks below reject it, the copy won't matter,
1435 * since we're rejecting the whole program. If all ok,
1436 * then imm22 state will be added to r7
1437 * and r7 will be pkt(id=0,off=22,r=62) while
1438 * r6 will stay as pkt(id=0,off=0,r=62)
1439 */
1440 }
1441
1442 if (src_reg->type == CONST_IMM) {
1443 /* pkt_ptr += reg where reg is known constant */
1444 imm = src_reg->imm;
1445 goto add_imm;
1446 }
1447 /* disallow pkt_ptr += reg
1448 * if reg is not uknown_value with guaranteed zero upper bits
1449 * otherwise pkt_ptr may overflow and addition will become
1450 * subtraction which is not allowed
1451 */
1452 if (src_reg->type != UNKNOWN_VALUE) {
1453 verbose("cannot add '%s' to ptr_to_packet\n",
1454 reg_type_str[src_reg->type]);
1455 return -EACCES;
1456 }
1457 if (src_reg->imm < 48) {
1458 verbose("cannot add integer value with %lld upper zero bits to ptr_to_packet\n",
1459 src_reg->imm);
1460 return -EACCES;
1461 }
1462 /* dst_reg stays as pkt_ptr type and since some positive
1463 * integer value was added to the pointer, increment its 'id'
1464 */
1465 dst_reg->id = ++env->id_gen;
1466
1467 /* something was added to pkt_ptr, set range and off to zero */
1468 dst_reg->off = 0;
1469 dst_reg->range = 0;
1470 }
1471 return 0;
1472 }
1473
1474 static int evaluate_reg_alu(struct bpf_verifier_env *env, struct bpf_insn *insn)
1475 {
1476 struct bpf_reg_state *regs = env->cur_state.regs;
1477 struct bpf_reg_state *dst_reg = &regs[insn->dst_reg];
1478 u8 opcode = BPF_OP(insn->code);
1479 s64 imm_log2;
1480
1481 /* for type == UNKNOWN_VALUE:
1482 * imm > 0 -> number of zero upper bits
1483 * imm == 0 -> don't track which is the same as all bits can be non-zero
1484 */
1485
1486 if (BPF_SRC(insn->code) == BPF_X) {
1487 struct bpf_reg_state *src_reg = &regs[insn->src_reg];
1488
1489 if (src_reg->type == UNKNOWN_VALUE && src_reg->imm > 0 &&
1490 dst_reg->imm && opcode == BPF_ADD) {
1491 /* dreg += sreg
1492 * where both have zero upper bits. Adding them
1493 * can only result making one more bit non-zero
1494 * in the larger value.
1495 * Ex. 0xffff (imm=48) + 1 (imm=63) = 0x10000 (imm=47)
1496 * 0xffff (imm=48) + 0xffff = 0x1fffe (imm=47)
1497 */
1498 dst_reg->imm = min(dst_reg->imm, src_reg->imm);
1499 dst_reg->imm--;
1500 return 0;
1501 }
1502 if (src_reg->type == CONST_IMM && src_reg->imm > 0 &&
1503 dst_reg->imm && opcode == BPF_ADD) {
1504 /* dreg += sreg
1505 * where dreg has zero upper bits and sreg is const.
1506 * Adding them can only result making one more bit
1507 * non-zero in the larger value.
1508 */
1509 imm_log2 = __ilog2_u64((long long)src_reg->imm);
1510 dst_reg->imm = min(dst_reg->imm, 63 - imm_log2);
1511 dst_reg->imm--;
1512 return 0;
1513 }
1514 /* all other cases non supported yet, just mark dst_reg */
1515 dst_reg->imm = 0;
1516 return 0;
1517 }
1518
1519 /* sign extend 32-bit imm into 64-bit to make sure that
1520 * negative values occupy bit 63. Note ilog2() would have
1521 * been incorrect, since sizeof(insn->imm) == 4
1522 */
1523 imm_log2 = __ilog2_u64((long long)insn->imm);
1524
1525 if (dst_reg->imm && opcode == BPF_LSH) {
1526 /* reg <<= imm
1527 * if reg was a result of 2 byte load, then its imm == 48
1528 * which means that upper 48 bits are zero and shifting this reg
1529 * left by 4 would mean that upper 44 bits are still zero
1530 */
1531 dst_reg->imm -= insn->imm;
1532 } else if (dst_reg->imm && opcode == BPF_MUL) {
1533 /* reg *= imm
1534 * if multiplying by 14 subtract 4
1535 * This is conservative calculation of upper zero bits.
1536 * It's not trying to special case insn->imm == 1 or 0 cases
1537 */
1538 dst_reg->imm -= imm_log2 + 1;
1539 } else if (opcode == BPF_AND) {
1540 /* reg &= imm */
1541 dst_reg->imm = 63 - imm_log2;
1542 } else if (dst_reg->imm && opcode == BPF_ADD) {
1543 /* reg += imm */
1544 dst_reg->imm = min(dst_reg->imm, 63 - imm_log2);
1545 dst_reg->imm--;
1546 } else if (opcode == BPF_RSH) {
1547 /* reg >>= imm
1548 * which means that after right shift, upper bits will be zero
1549 * note that verifier already checked that
1550 * 0 <= imm < 64 for shift insn
1551 */
1552 dst_reg->imm += insn->imm;
1553 if (unlikely(dst_reg->imm > 64))
1554 /* some dumb code did:
1555 * r2 = *(u32 *)mem;
1556 * r2 >>= 32;
1557 * and all bits are zero now */
1558 dst_reg->imm = 64;
1559 } else {
1560 /* all other alu ops, means that we don't know what will
1561 * happen to the value, mark it with unknown number of zero bits
1562 */
1563 dst_reg->imm = 0;
1564 }
1565
1566 if (dst_reg->imm < 0) {
1567 /* all 64 bits of the register can contain non-zero bits
1568 * and such value cannot be added to ptr_to_packet, since it
1569 * may overflow, mark it as unknown to avoid further eval
1570 */
1571 dst_reg->imm = 0;
1572 }
1573 return 0;
1574 }
1575
1576 static int evaluate_reg_imm_alu(struct bpf_verifier_env *env,
1577 struct bpf_insn *insn)
1578 {
1579 struct bpf_reg_state *regs = env->cur_state.regs;
1580 struct bpf_reg_state *dst_reg = &regs[insn->dst_reg];
1581 struct bpf_reg_state *src_reg = &regs[insn->src_reg];
1582 u8 opcode = BPF_OP(insn->code);
1583 u64 dst_imm = dst_reg->imm;
1584
1585 /* dst_reg->type == CONST_IMM here. Simulate execution of insns
1586 * containing ALU ops. Don't care about overflow or negative
1587 * values, just add/sub/... them; registers are in u64.
1588 */
1589 if (opcode == BPF_ADD && BPF_SRC(insn->code) == BPF_K) {
1590 dst_imm += insn->imm;
1591 } else if (opcode == BPF_ADD && BPF_SRC(insn->code) == BPF_X &&
1592 src_reg->type == CONST_IMM) {
1593 dst_imm += src_reg->imm;
1594 } else if (opcode == BPF_SUB && BPF_SRC(insn->code) == BPF_K) {
1595 dst_imm -= insn->imm;
1596 } else if (opcode == BPF_SUB && BPF_SRC(insn->code) == BPF_X &&
1597 src_reg->type == CONST_IMM) {
1598 dst_imm -= src_reg->imm;
1599 } else if (opcode == BPF_MUL && BPF_SRC(insn->code) == BPF_K) {
1600 dst_imm *= insn->imm;
1601 } else if (opcode == BPF_MUL && BPF_SRC(insn->code) == BPF_X &&
1602 src_reg->type == CONST_IMM) {
1603 dst_imm *= src_reg->imm;
1604 } else if (opcode == BPF_OR && BPF_SRC(insn->code) == BPF_K) {
1605 dst_imm |= insn->imm;
1606 } else if (opcode == BPF_OR && BPF_SRC(insn->code) == BPF_X &&
1607 src_reg->type == CONST_IMM) {
1608 dst_imm |= src_reg->imm;
1609 } else if (opcode == BPF_AND && BPF_SRC(insn->code) == BPF_K) {
1610 dst_imm &= insn->imm;
1611 } else if (opcode == BPF_AND && BPF_SRC(insn->code) == BPF_X &&
1612 src_reg->type == CONST_IMM) {
1613 dst_imm &= src_reg->imm;
1614 } else if (opcode == BPF_RSH && BPF_SRC(insn->code) == BPF_K) {
1615 dst_imm >>= insn->imm;
1616 } else if (opcode == BPF_RSH && BPF_SRC(insn->code) == BPF_X &&
1617 src_reg->type == CONST_IMM) {
1618 dst_imm >>= src_reg->imm;
1619 } else if (opcode == BPF_LSH && BPF_SRC(insn->code) == BPF_K) {
1620 dst_imm <<= insn->imm;
1621 } else if (opcode == BPF_LSH && BPF_SRC(insn->code) == BPF_X &&
1622 src_reg->type == CONST_IMM) {
1623 dst_imm <<= src_reg->imm;
1624 } else {
1625 mark_reg_unknown_value(regs, insn->dst_reg);
1626 goto out;
1627 }
1628
1629 dst_reg->imm = dst_imm;
1630 out:
1631 return 0;
1632 }
1633
1634 static void check_reg_overflow(struct bpf_reg_state *reg)
1635 {
1636 if (reg->max_value > BPF_REGISTER_MAX_RANGE)
1637 reg->max_value = BPF_REGISTER_MAX_RANGE;
1638 if (reg->min_value < BPF_REGISTER_MIN_RANGE ||
1639 reg->min_value > BPF_REGISTER_MAX_RANGE)
1640 reg->min_value = BPF_REGISTER_MIN_RANGE;
1641 }
1642
1643 static void adjust_reg_min_max_vals(struct bpf_verifier_env *env,
1644 struct bpf_insn *insn)
1645 {
1646 struct bpf_reg_state *regs = env->cur_state.regs, *dst_reg;
1647 s64 min_val = BPF_REGISTER_MIN_RANGE;
1648 u64 max_val = BPF_REGISTER_MAX_RANGE;
1649 u8 opcode = BPF_OP(insn->code);
1650
1651 dst_reg = &regs[insn->dst_reg];
1652 if (BPF_SRC(insn->code) == BPF_X) {
1653 check_reg_overflow(&regs[insn->src_reg]);
1654 min_val = regs[insn->src_reg].min_value;
1655 max_val = regs[insn->src_reg].max_value;
1656
1657 /* If the source register is a random pointer then the
1658 * min_value/max_value values represent the range of the known
1659 * accesses into that value, not the actual min/max value of the
1660 * register itself. In this case we have to reset the reg range
1661 * values so we know it is not safe to look at.
1662 */
1663 if (regs[insn->src_reg].type != CONST_IMM &&
1664 regs[insn->src_reg].type != UNKNOWN_VALUE) {
1665 min_val = BPF_REGISTER_MIN_RANGE;
1666 max_val = BPF_REGISTER_MAX_RANGE;
1667 }
1668 } else if (insn->imm < BPF_REGISTER_MAX_RANGE &&
1669 (s64)insn->imm > BPF_REGISTER_MIN_RANGE) {
1670 min_val = max_val = insn->imm;
1671 }
1672
1673 /* We don't know anything about what was done to this register, mark it
1674 * as unknown.
1675 */
1676 if (min_val == BPF_REGISTER_MIN_RANGE &&
1677 max_val == BPF_REGISTER_MAX_RANGE) {
1678 reset_reg_range_values(regs, insn->dst_reg);
1679 return;
1680 }
1681
1682 /* If one of our values was at the end of our ranges then we can't just
1683 * do our normal operations to the register, we need to set the values
1684 * to the min/max since they are undefined.
1685 */
1686 if (min_val == BPF_REGISTER_MIN_RANGE)
1687 dst_reg->min_value = BPF_REGISTER_MIN_RANGE;
1688 if (max_val == BPF_REGISTER_MAX_RANGE)
1689 dst_reg->max_value = BPF_REGISTER_MAX_RANGE;
1690
1691 switch (opcode) {
1692 case BPF_ADD:
1693 if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE)
1694 dst_reg->min_value += min_val;
1695 if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1696 dst_reg->max_value += max_val;
1697 break;
1698 case BPF_SUB:
1699 if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE)
1700 dst_reg->min_value -= min_val;
1701 if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1702 dst_reg->max_value -= max_val;
1703 break;
1704 case BPF_MUL:
1705 if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE)
1706 dst_reg->min_value *= min_val;
1707 if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1708 dst_reg->max_value *= max_val;
1709 break;
1710 case BPF_AND:
1711 /* Disallow AND'ing of negative numbers, ain't nobody got time
1712 * for that. Otherwise the minimum is 0 and the max is the max
1713 * value we could AND against.
1714 */
1715 if (min_val < 0)
1716 dst_reg->min_value = BPF_REGISTER_MIN_RANGE;
1717 else
1718 dst_reg->min_value = 0;
1719 dst_reg->max_value = max_val;
1720 break;
1721 case BPF_LSH:
1722 /* Gotta have special overflow logic here, if we're shifting
1723 * more than MAX_RANGE then just assume we have an invalid
1724 * range.
1725 */
1726 if (min_val > ilog2(BPF_REGISTER_MAX_RANGE))
1727 dst_reg->min_value = BPF_REGISTER_MIN_RANGE;
1728 else if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE)
1729 dst_reg->min_value <<= min_val;
1730
1731 if (max_val > ilog2(BPF_REGISTER_MAX_RANGE))
1732 dst_reg->max_value = BPF_REGISTER_MAX_RANGE;
1733 else if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1734 dst_reg->max_value <<= max_val;
1735 break;
1736 case BPF_RSH:
1737 /* RSH by a negative number is undefined, and the BPF_RSH is an
1738 * unsigned shift, so make the appropriate casts.
1739 */
1740 if (min_val < 0 || dst_reg->min_value < 0)
1741 dst_reg->min_value = BPF_REGISTER_MIN_RANGE;
1742 else
1743 dst_reg->min_value =
1744 (u64)(dst_reg->min_value) >> min_val;
1745 if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1746 dst_reg->max_value >>= max_val;
1747 break;
1748 default:
1749 reset_reg_range_values(regs, insn->dst_reg);
1750 break;
1751 }
1752
1753 check_reg_overflow(dst_reg);
1754 }
1755
1756 /* check validity of 32-bit and 64-bit arithmetic operations */
1757 static int check_alu_op(struct bpf_verifier_env *env, struct bpf_insn *insn)
1758 {
1759 struct bpf_reg_state *regs = env->cur_state.regs, *dst_reg;
1760 u8 opcode = BPF_OP(insn->code);
1761 int err;
1762
1763 if (opcode == BPF_END || opcode == BPF_NEG) {
1764 if (opcode == BPF_NEG) {
1765 if (BPF_SRC(insn->code) != 0 ||
1766 insn->src_reg != BPF_REG_0 ||
1767 insn->off != 0 || insn->imm != 0) {
1768 verbose("BPF_NEG uses reserved fields\n");
1769 return -EINVAL;
1770 }
1771 } else {
1772 if (insn->src_reg != BPF_REG_0 || insn->off != 0 ||
1773 (insn->imm != 16 && insn->imm != 32 && insn->imm != 64)) {
1774 verbose("BPF_END uses reserved fields\n");
1775 return -EINVAL;
1776 }
1777 }
1778
1779 /* check src operand */
1780 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1781 if (err)
1782 return err;
1783
1784 if (is_pointer_value(env, insn->dst_reg)) {
1785 verbose("R%d pointer arithmetic prohibited\n",
1786 insn->dst_reg);
1787 return -EACCES;
1788 }
1789
1790 /* check dest operand */
1791 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1792 if (err)
1793 return err;
1794
1795 } else if (opcode == BPF_MOV) {
1796
1797 if (BPF_SRC(insn->code) == BPF_X) {
1798 if (insn->imm != 0 || insn->off != 0) {
1799 verbose("BPF_MOV uses reserved fields\n");
1800 return -EINVAL;
1801 }
1802
1803 /* check src operand */
1804 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1805 if (err)
1806 return err;
1807 } else {
1808 if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
1809 verbose("BPF_MOV uses reserved fields\n");
1810 return -EINVAL;
1811 }
1812 }
1813
1814 /* check dest operand */
1815 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1816 if (err)
1817 return err;
1818
1819 /* we are setting our register to something new, we need to
1820 * reset its range values.
1821 */
1822 reset_reg_range_values(regs, insn->dst_reg);
1823
1824 if (BPF_SRC(insn->code) == BPF_X) {
1825 if (BPF_CLASS(insn->code) == BPF_ALU64) {
1826 /* case: R1 = R2
1827 * copy register state to dest reg
1828 */
1829 regs[insn->dst_reg] = regs[insn->src_reg];
1830 } else {
1831 if (is_pointer_value(env, insn->src_reg)) {
1832 verbose("R%d partial copy of pointer\n",
1833 insn->src_reg);
1834 return -EACCES;
1835 }
1836 mark_reg_unknown_value(regs, insn->dst_reg);
1837 }
1838 } else {
1839 /* case: R = imm
1840 * remember the value we stored into this reg
1841 */
1842 regs[insn->dst_reg].type = CONST_IMM;
1843 regs[insn->dst_reg].imm = insn->imm;
1844 regs[insn->dst_reg].max_value = insn->imm;
1845 regs[insn->dst_reg].min_value = insn->imm;
1846 }
1847
1848 } else if (opcode > BPF_END) {
1849 verbose("invalid BPF_ALU opcode %x\n", opcode);
1850 return -EINVAL;
1851
1852 } else { /* all other ALU ops: and, sub, xor, add, ... */
1853
1854 if (BPF_SRC(insn->code) == BPF_X) {
1855 if (insn->imm != 0 || insn->off != 0) {
1856 verbose("BPF_ALU uses reserved fields\n");
1857 return -EINVAL;
1858 }
1859 /* check src1 operand */
1860 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1861 if (err)
1862 return err;
1863 } else {
1864 if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
1865 verbose("BPF_ALU uses reserved fields\n");
1866 return -EINVAL;
1867 }
1868 }
1869
1870 /* check src2 operand */
1871 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1872 if (err)
1873 return err;
1874
1875 if ((opcode == BPF_MOD || opcode == BPF_DIV) &&
1876 BPF_SRC(insn->code) == BPF_K && insn->imm == 0) {
1877 verbose("div by zero\n");
1878 return -EINVAL;
1879 }
1880
1881 if ((opcode == BPF_LSH || opcode == BPF_RSH ||
1882 opcode == BPF_ARSH) && BPF_SRC(insn->code) == BPF_K) {
1883 int size = BPF_CLASS(insn->code) == BPF_ALU64 ? 64 : 32;
1884
1885 if (insn->imm < 0 || insn->imm >= size) {
1886 verbose("invalid shift %d\n", insn->imm);
1887 return -EINVAL;
1888 }
1889 }
1890
1891 /* check dest operand */
1892 err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK);
1893 if (err)
1894 return err;
1895
1896 dst_reg = &regs[insn->dst_reg];
1897
1898 /* first we want to adjust our ranges. */
1899 adjust_reg_min_max_vals(env, insn);
1900
1901 /* pattern match 'bpf_add Rx, imm' instruction */
1902 if (opcode == BPF_ADD && BPF_CLASS(insn->code) == BPF_ALU64 &&
1903 dst_reg->type == FRAME_PTR && BPF_SRC(insn->code) == BPF_K) {
1904 dst_reg->type = PTR_TO_STACK;
1905 dst_reg->imm = insn->imm;
1906 return 0;
1907 } else if (opcode == BPF_ADD &&
1908 BPF_CLASS(insn->code) == BPF_ALU64 &&
1909 (dst_reg->type == PTR_TO_PACKET ||
1910 (BPF_SRC(insn->code) == BPF_X &&
1911 regs[insn->src_reg].type == PTR_TO_PACKET))) {
1912 /* ptr_to_packet += K|X */
1913 return check_packet_ptr_add(env, insn);
1914 } else if (BPF_CLASS(insn->code) == BPF_ALU64 &&
1915 dst_reg->type == UNKNOWN_VALUE &&
1916 env->allow_ptr_leaks) {
1917 /* unknown += K|X */
1918 return evaluate_reg_alu(env, insn);
1919 } else if (BPF_CLASS(insn->code) == BPF_ALU64 &&
1920 dst_reg->type == CONST_IMM &&
1921 env->allow_ptr_leaks) {
1922 /* reg_imm += K|X */
1923 return evaluate_reg_imm_alu(env, insn);
1924 } else if (is_pointer_value(env, insn->dst_reg)) {
1925 verbose("R%d pointer arithmetic prohibited\n",
1926 insn->dst_reg);
1927 return -EACCES;
1928 } else if (BPF_SRC(insn->code) == BPF_X &&
1929 is_pointer_value(env, insn->src_reg)) {
1930 verbose("R%d pointer arithmetic prohibited\n",
1931 insn->src_reg);
1932 return -EACCES;
1933 }
1934
1935 /* If we did pointer math on a map value then just set it to our
1936 * PTR_TO_MAP_VALUE_ADJ type so we can deal with any stores or
1937 * loads to this register appropriately, otherwise just mark the
1938 * register as unknown.
1939 */
1940 if (env->allow_ptr_leaks &&
1941 (dst_reg->type == PTR_TO_MAP_VALUE ||
1942 dst_reg->type == PTR_TO_MAP_VALUE_ADJ))
1943 dst_reg->type = PTR_TO_MAP_VALUE_ADJ;
1944 else
1945 mark_reg_unknown_value(regs, insn->dst_reg);
1946 }
1947
1948 return 0;
1949 }
1950
1951 static void find_good_pkt_pointers(struct bpf_verifier_state *state,
1952 struct bpf_reg_state *dst_reg)
1953 {
1954 struct bpf_reg_state *regs = state->regs, *reg;
1955 int i;
1956
1957 /* LLVM can generate two kind of checks:
1958 *
1959 * Type 1:
1960 *
1961 * r2 = r3;
1962 * r2 += 8;
1963 * if (r2 > pkt_end) goto <handle exception>
1964 * <access okay>
1965 *
1966 * Where:
1967 * r2 == dst_reg, pkt_end == src_reg
1968 * r2=pkt(id=n,off=8,r=0)
1969 * r3=pkt(id=n,off=0,r=0)
1970 *
1971 * Type 2:
1972 *
1973 * r2 = r3;
1974 * r2 += 8;
1975 * if (pkt_end >= r2) goto <access okay>
1976 * <handle exception>
1977 *
1978 * Where:
1979 * pkt_end == dst_reg, r2 == src_reg
1980 * r2=pkt(id=n,off=8,r=0)
1981 * r3=pkt(id=n,off=0,r=0)
1982 *
1983 * Find register r3 and mark its range as r3=pkt(id=n,off=0,r=8)
1984 * so that range of bytes [r3, r3 + 8) is safe to access.
1985 */
1986
1987 for (i = 0; i < MAX_BPF_REG; i++)
1988 if (regs[i].type == PTR_TO_PACKET && regs[i].id == dst_reg->id)
1989 regs[i].range = dst_reg->off;
1990
1991 for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
1992 if (state->stack_slot_type[i] != STACK_SPILL)
1993 continue;
1994 reg = &state->spilled_regs[i / BPF_REG_SIZE];
1995 if (reg->type == PTR_TO_PACKET && reg->id == dst_reg->id)
1996 reg->range = dst_reg->off;
1997 }
1998 }
1999
2000 /* Adjusts the register min/max values in the case that the dst_reg is the
2001 * variable register that we are working on, and src_reg is a constant or we're
2002 * simply doing a BPF_K check.
2003 */
2004 static void reg_set_min_max(struct bpf_reg_state *true_reg,
2005 struct bpf_reg_state *false_reg, u64 val,
2006 u8 opcode)
2007 {
2008 switch (opcode) {
2009 case BPF_JEQ:
2010 /* If this is false then we know nothing Jon Snow, but if it is
2011 * true then we know for sure.
2012 */
2013 true_reg->max_value = true_reg->min_value = val;
2014 break;
2015 case BPF_JNE:
2016 /* If this is true we know nothing Jon Snow, but if it is false
2017 * we know the value for sure;
2018 */
2019 false_reg->max_value = false_reg->min_value = val;
2020 break;
2021 case BPF_JGT:
2022 /* Unsigned comparison, the minimum value is 0. */
2023 false_reg->min_value = 0;
2024 /* fallthrough */
2025 case BPF_JSGT:
2026 /* If this is false then we know the maximum val is val,
2027 * otherwise we know the min val is val+1.
2028 */
2029 false_reg->max_value = val;
2030 true_reg->min_value = val + 1;
2031 break;
2032 case BPF_JGE:
2033 /* Unsigned comparison, the minimum value is 0. */
2034 false_reg->min_value = 0;
2035 /* fallthrough */
2036 case BPF_JSGE:
2037 /* If this is false then we know the maximum value is val - 1,
2038 * otherwise we know the mimimum value is val.
2039 */
2040 false_reg->max_value = val - 1;
2041 true_reg->min_value = val;
2042 break;
2043 default:
2044 break;
2045 }
2046
2047 check_reg_overflow(false_reg);
2048 check_reg_overflow(true_reg);
2049 }
2050
2051 /* Same as above, but for the case that dst_reg is a CONST_IMM reg and src_reg
2052 * is the variable reg.
2053 */
2054 static void reg_set_min_max_inv(struct bpf_reg_state *true_reg,
2055 struct bpf_reg_state *false_reg, u64 val,
2056 u8 opcode)
2057 {
2058 switch (opcode) {
2059 case BPF_JEQ:
2060 /* If this is false then we know nothing Jon Snow, but if it is
2061 * true then we know for sure.
2062 */
2063 true_reg->max_value = true_reg->min_value = val;
2064 break;
2065 case BPF_JNE:
2066 /* If this is true we know nothing Jon Snow, but if it is false
2067 * we know the value for sure;
2068 */
2069 false_reg->max_value = false_reg->min_value = val;
2070 break;
2071 case BPF_JGT:
2072 /* Unsigned comparison, the minimum value is 0. */
2073 true_reg->min_value = 0;
2074 /* fallthrough */
2075 case BPF_JSGT:
2076 /*
2077 * If this is false, then the val is <= the register, if it is
2078 * true the register <= to the val.
2079 */
2080 false_reg->min_value = val;
2081 true_reg->max_value = val - 1;
2082 break;
2083 case BPF_JGE:
2084 /* Unsigned comparison, the minimum value is 0. */
2085 true_reg->min_value = 0;
2086 /* fallthrough */
2087 case BPF_JSGE:
2088 /* If this is false then constant < register, if it is true then
2089 * the register < constant.
2090 */
2091 false_reg->min_value = val + 1;
2092 true_reg->max_value = val;
2093 break;
2094 default:
2095 break;
2096 }
2097
2098 check_reg_overflow(false_reg);
2099 check_reg_overflow(true_reg);
2100 }
2101
2102 static void mark_map_reg(struct bpf_reg_state *regs, u32 regno, u32 id,
2103 enum bpf_reg_type type)
2104 {
2105 struct bpf_reg_state *reg = &regs[regno];
2106
2107 if (reg->type == PTR_TO_MAP_VALUE_OR_NULL && reg->id == id) {
2108 if (type == UNKNOWN_VALUE) {
2109 __mark_reg_unknown_value(regs, regno);
2110 } else if (reg->map_ptr->inner_map_meta) {
2111 reg->type = CONST_PTR_TO_MAP;
2112 reg->map_ptr = reg->map_ptr->inner_map_meta;
2113 } else {
2114 reg->type = type;
2115 }
2116 /* We don't need id from this point onwards anymore, thus we
2117 * should better reset it, so that state pruning has chances
2118 * to take effect.
2119 */
2120 reg->id = 0;
2121 }
2122 }
2123
2124 /* The logic is similar to find_good_pkt_pointers(), both could eventually
2125 * be folded together at some point.
2126 */
2127 static void mark_map_regs(struct bpf_verifier_state *state, u32 regno,
2128 enum bpf_reg_type type)
2129 {
2130 struct bpf_reg_state *regs = state->regs;
2131 u32 id = regs[regno].id;
2132 int i;
2133
2134 for (i = 0; i < MAX_BPF_REG; i++)
2135 mark_map_reg(regs, i, id, type);
2136
2137 for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
2138 if (state->stack_slot_type[i] != STACK_SPILL)
2139 continue;
2140 mark_map_reg(state->spilled_regs, i / BPF_REG_SIZE, id, type);
2141 }
2142 }
2143
2144 static int check_cond_jmp_op(struct bpf_verifier_env *env,
2145 struct bpf_insn *insn, int *insn_idx)
2146 {
2147 struct bpf_verifier_state *other_branch, *this_branch = &env->cur_state;
2148 struct bpf_reg_state *regs = this_branch->regs, *dst_reg;
2149 u8 opcode = BPF_OP(insn->code);
2150 int err;
2151
2152 if (opcode > BPF_EXIT) {
2153 verbose("invalid BPF_JMP opcode %x\n", opcode);
2154 return -EINVAL;
2155 }
2156
2157 if (BPF_SRC(insn->code) == BPF_X) {
2158 if (insn->imm != 0) {
2159 verbose("BPF_JMP uses reserved fields\n");
2160 return -EINVAL;
2161 }
2162
2163 /* check src1 operand */
2164 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
2165 if (err)
2166 return err;
2167
2168 if (is_pointer_value(env, insn->src_reg)) {
2169 verbose("R%d pointer comparison prohibited\n",
2170 insn->src_reg);
2171 return -EACCES;
2172 }
2173 } else {
2174 if (insn->src_reg != BPF_REG_0) {
2175 verbose("BPF_JMP uses reserved fields\n");
2176 return -EINVAL;
2177 }
2178 }
2179
2180 /* check src2 operand */
2181 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
2182 if (err)
2183 return err;
2184
2185 dst_reg = &regs[insn->dst_reg];
2186
2187 /* detect if R == 0 where R was initialized to zero earlier */
2188 if (BPF_SRC(insn->code) == BPF_K &&
2189 (opcode == BPF_JEQ || opcode == BPF_JNE) &&
2190 dst_reg->type == CONST_IMM && dst_reg->imm == insn->imm) {
2191 if (opcode == BPF_JEQ) {
2192 /* if (imm == imm) goto pc+off;
2193 * only follow the goto, ignore fall-through
2194 */
2195 *insn_idx += insn->off;
2196 return 0;
2197 } else {
2198 /* if (imm != imm) goto pc+off;
2199 * only follow fall-through branch, since
2200 * that's where the program will go
2201 */
2202 return 0;
2203 }
2204 }
2205
2206 other_branch = push_stack(env, *insn_idx + insn->off + 1, *insn_idx);
2207 if (!other_branch)
2208 return -EFAULT;
2209
2210 /* detect if we are comparing against a constant value so we can adjust
2211 * our min/max values for our dst register.
2212 */
2213 if (BPF_SRC(insn->code) == BPF_X) {
2214 if (regs[insn->src_reg].type == CONST_IMM)
2215 reg_set_min_max(&other_branch->regs[insn->dst_reg],
2216 dst_reg, regs[insn->src_reg].imm,
2217 opcode);
2218 else if (dst_reg->type == CONST_IMM)
2219 reg_set_min_max_inv(&other_branch->regs[insn->src_reg],
2220 &regs[insn->src_reg], dst_reg->imm,
2221 opcode);
2222 } else {
2223 reg_set_min_max(&other_branch->regs[insn->dst_reg],
2224 dst_reg, insn->imm, opcode);
2225 }
2226
2227 /* detect if R == 0 where R is returned from bpf_map_lookup_elem() */
2228 if (BPF_SRC(insn->code) == BPF_K &&
2229 insn->imm == 0 && (opcode == BPF_JEQ || opcode == BPF_JNE) &&
2230 dst_reg->type == PTR_TO_MAP_VALUE_OR_NULL) {
2231 /* Mark all identical map registers in each branch as either
2232 * safe or unknown depending R == 0 or R != 0 conditional.
2233 */
2234 mark_map_regs(this_branch, insn->dst_reg,
2235 opcode == BPF_JEQ ? PTR_TO_MAP_VALUE : UNKNOWN_VALUE);
2236 mark_map_regs(other_branch, insn->dst_reg,
2237 opcode == BPF_JEQ ? UNKNOWN_VALUE : PTR_TO_MAP_VALUE);
2238 } else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JGT &&
2239 dst_reg->type == PTR_TO_PACKET &&
2240 regs[insn->src_reg].type == PTR_TO_PACKET_END) {
2241 find_good_pkt_pointers(this_branch, dst_reg);
2242 } else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JGE &&
2243 dst_reg->type == PTR_TO_PACKET_END &&
2244 regs[insn->src_reg].type == PTR_TO_PACKET) {
2245 find_good_pkt_pointers(other_branch, &regs[insn->src_reg]);
2246 } else if (is_pointer_value(env, insn->dst_reg)) {
2247 verbose("R%d pointer comparison prohibited\n", insn->dst_reg);
2248 return -EACCES;
2249 }
2250 if (log_level)
2251 print_verifier_state(this_branch);
2252 return 0;
2253 }
2254
2255 /* return the map pointer stored inside BPF_LD_IMM64 instruction */
2256 static struct bpf_map *ld_imm64_to_map_ptr(struct bpf_insn *insn)
2257 {
2258 u64 imm64 = ((u64) (u32) insn[0].imm) | ((u64) (u32) insn[1].imm) << 32;
2259
2260 return (struct bpf_map *) (unsigned long) imm64;
2261 }
2262
2263 /* verify BPF_LD_IMM64 instruction */
2264 static int check_ld_imm(struct bpf_verifier_env *env, struct bpf_insn *insn)
2265 {
2266 struct bpf_reg_state *regs = env->cur_state.regs;
2267 int err;
2268
2269 if (BPF_SIZE(insn->code) != BPF_DW) {
2270 verbose("invalid BPF_LD_IMM insn\n");
2271 return -EINVAL;
2272 }
2273 if (insn->off != 0) {
2274 verbose("BPF_LD_IMM64 uses reserved fields\n");
2275 return -EINVAL;
2276 }
2277
2278 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
2279 if (err)
2280 return err;
2281
2282 if (insn->src_reg == 0) {
2283 u64 imm = ((u64)(insn + 1)->imm << 32) | (u32)insn->imm;
2284
2285 regs[insn->dst_reg].type = CONST_IMM;
2286 regs[insn->dst_reg].imm = imm;
2287 return 0;
2288 }
2289
2290 /* replace_map_fd_with_map_ptr() should have caught bad ld_imm64 */
2291 BUG_ON(insn->src_reg != BPF_PSEUDO_MAP_FD);
2292
2293 regs[insn->dst_reg].type = CONST_PTR_TO_MAP;
2294 regs[insn->dst_reg].map_ptr = ld_imm64_to_map_ptr(insn);
2295 return 0;
2296 }
2297
2298 static bool may_access_skb(enum bpf_prog_type type)
2299 {
2300 switch (type) {
2301 case BPF_PROG_TYPE_SOCKET_FILTER:
2302 case BPF_PROG_TYPE_SCHED_CLS:
2303 case BPF_PROG_TYPE_SCHED_ACT:
2304 return true;
2305 default:
2306 return false;
2307 }
2308 }
2309
2310 /* verify safety of LD_ABS|LD_IND instructions:
2311 * - they can only appear in the programs where ctx == skb
2312 * - since they are wrappers of function calls, they scratch R1-R5 registers,
2313 * preserve R6-R9, and store return value into R0
2314 *
2315 * Implicit input:
2316 * ctx == skb == R6 == CTX
2317 *
2318 * Explicit input:
2319 * SRC == any register
2320 * IMM == 32-bit immediate
2321 *
2322 * Output:
2323 * R0 - 8/16/32-bit skb data converted to cpu endianness
2324 */
2325 static int check_ld_abs(struct bpf_verifier_env *env, struct bpf_insn *insn)
2326 {
2327 struct bpf_reg_state *regs = env->cur_state.regs;
2328 u8 mode = BPF_MODE(insn->code);
2329 struct bpf_reg_state *reg;
2330 int i, err;
2331
2332 if (!may_access_skb(env->prog->type)) {
2333 verbose("BPF_LD_[ABS|IND] instructions not allowed for this program type\n");
2334 return -EINVAL;
2335 }
2336
2337 if (insn->dst_reg != BPF_REG_0 || insn->off != 0 ||
2338 BPF_SIZE(insn->code) == BPF_DW ||
2339 (mode == BPF_ABS && insn->src_reg != BPF_REG_0)) {
2340 verbose("BPF_LD_[ABS|IND] uses reserved fields\n");
2341 return -EINVAL;
2342 }
2343
2344 /* check whether implicit source operand (register R6) is readable */
2345 err = check_reg_arg(regs, BPF_REG_6, SRC_OP);
2346 if (err)
2347 return err;
2348
2349 if (regs[BPF_REG_6].type != PTR_TO_CTX) {
2350 verbose("at the time of BPF_LD_ABS|IND R6 != pointer to skb\n");
2351 return -EINVAL;
2352 }
2353
2354 if (mode == BPF_IND) {
2355 /* check explicit source operand */
2356 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
2357 if (err)
2358 return err;
2359 }
2360
2361 /* reset caller saved regs to unreadable */
2362 for (i = 0; i < CALLER_SAVED_REGS; i++) {
2363 reg = regs + caller_saved[i];
2364 reg->type = NOT_INIT;
2365 reg->imm = 0;
2366 }
2367
2368 /* mark destination R0 register as readable, since it contains
2369 * the value fetched from the packet
2370 */
2371 regs[BPF_REG_0].type = UNKNOWN_VALUE;
2372 return 0;
2373 }
2374
2375 /* non-recursive DFS pseudo code
2376 * 1 procedure DFS-iterative(G,v):
2377 * 2 label v as discovered
2378 * 3 let S be a stack
2379 * 4 S.push(v)
2380 * 5 while S is not empty
2381 * 6 t <- S.pop()
2382 * 7 if t is what we're looking for:
2383 * 8 return t
2384 * 9 for all edges e in G.adjacentEdges(t) do
2385 * 10 if edge e is already labelled
2386 * 11 continue with the next edge
2387 * 12 w <- G.adjacentVertex(t,e)
2388 * 13 if vertex w is not discovered and not explored
2389 * 14 label e as tree-edge
2390 * 15 label w as discovered
2391 * 16 S.push(w)
2392 * 17 continue at 5
2393 * 18 else if vertex w is discovered
2394 * 19 label e as back-edge
2395 * 20 else
2396 * 21 // vertex w is explored
2397 * 22 label e as forward- or cross-edge
2398 * 23 label t as explored
2399 * 24 S.pop()
2400 *
2401 * convention:
2402 * 0x10 - discovered
2403 * 0x11 - discovered and fall-through edge labelled
2404 * 0x12 - discovered and fall-through and branch edges labelled
2405 * 0x20 - explored
2406 */
2407
2408 enum {
2409 DISCOVERED = 0x10,
2410 EXPLORED = 0x20,
2411 FALLTHROUGH = 1,
2412 BRANCH = 2,
2413 };
2414
2415 #define STATE_LIST_MARK ((struct bpf_verifier_state_list *) -1L)
2416
2417 static int *insn_stack; /* stack of insns to process */
2418 static int cur_stack; /* current stack index */
2419 static int *insn_state;
2420
2421 /* t, w, e - match pseudo-code above:
2422 * t - index of current instruction
2423 * w - next instruction
2424 * e - edge
2425 */
2426 static int push_insn(int t, int w, int e, struct bpf_verifier_env *env)
2427 {
2428 if (e == FALLTHROUGH && insn_state[t] >= (DISCOVERED | FALLTHROUGH))
2429 return 0;
2430
2431 if (e == BRANCH && insn_state[t] >= (DISCOVERED | BRANCH))
2432 return 0;
2433
2434 if (w < 0 || w >= env->prog->len) {
2435 verbose("jump out of range from insn %d to %d\n", t, w);
2436 return -EINVAL;
2437 }
2438
2439 if (e == BRANCH)
2440 /* mark branch target for state pruning */
2441 env->explored_states[w] = STATE_LIST_MARK;
2442
2443 if (insn_state[w] == 0) {
2444 /* tree-edge */
2445 insn_state[t] = DISCOVERED | e;
2446 insn_state[w] = DISCOVERED;
2447 if (cur_stack >= env->prog->len)
2448 return -E2BIG;
2449 insn_stack[cur_stack++] = w;
2450 return 1;
2451 } else if ((insn_state[w] & 0xF0) == DISCOVERED) {
2452 verbose("back-edge from insn %d to %d\n", t, w);
2453 return -EINVAL;
2454 } else if (insn_state[w] == EXPLORED) {
2455 /* forward- or cross-edge */
2456 insn_state[t] = DISCOVERED | e;
2457 } else {
2458 verbose("insn state internal bug\n");
2459 return -EFAULT;
2460 }
2461 return 0;
2462 }
2463
2464 /* non-recursive depth-first-search to detect loops in BPF program
2465 * loop == back-edge in directed graph
2466 */
2467 static int check_cfg(struct bpf_verifier_env *env)
2468 {
2469 struct bpf_insn *insns = env->prog->insnsi;
2470 int insn_cnt = env->prog->len;
2471 int ret = 0;
2472 int i, t;
2473
2474 insn_state = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
2475 if (!insn_state)
2476 return -ENOMEM;
2477
2478 insn_stack = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
2479 if (!insn_stack) {
2480 kfree(insn_state);
2481 return -ENOMEM;
2482 }
2483
2484 insn_state[0] = DISCOVERED; /* mark 1st insn as discovered */
2485 insn_stack[0] = 0; /* 0 is the first instruction */
2486 cur_stack = 1;
2487
2488 peek_stack:
2489 if (cur_stack == 0)
2490 goto check_state;
2491 t = insn_stack[cur_stack - 1];
2492
2493 if (BPF_CLASS(insns[t].code) == BPF_JMP) {
2494 u8 opcode = BPF_OP(insns[t].code);
2495
2496 if (opcode == BPF_EXIT) {
2497 goto mark_explored;
2498 } else if (opcode == BPF_CALL) {
2499 ret = push_insn(t, t + 1, FALLTHROUGH, env);
2500 if (ret == 1)
2501 goto peek_stack;
2502 else if (ret < 0)
2503 goto err_free;
2504 if (t + 1 < insn_cnt)
2505 env->explored_states[t + 1] = STATE_LIST_MARK;
2506 } else if (opcode == BPF_JA) {
2507 if (BPF_SRC(insns[t].code) != BPF_K) {
2508 ret = -EINVAL;
2509 goto err_free;
2510 }
2511 /* unconditional jump with single edge */
2512 ret = push_insn(t, t + insns[t].off + 1,
2513 FALLTHROUGH, env);
2514 if (ret == 1)
2515 goto peek_stack;
2516 else if (ret < 0)
2517 goto err_free;
2518 /* tell verifier to check for equivalent states
2519 * after every call and jump
2520 */
2521 if (t + 1 < insn_cnt)
2522 env->explored_states[t + 1] = STATE_LIST_MARK;
2523 } else {
2524 /* conditional jump with two edges */
2525 ret = push_insn(t, t + 1, FALLTHROUGH, env);
2526 if (ret == 1)
2527 goto peek_stack;
2528 else if (ret < 0)
2529 goto err_free;
2530
2531 ret = push_insn(t, t + insns[t].off + 1, BRANCH, env);
2532 if (ret == 1)
2533 goto peek_stack;
2534 else if (ret < 0)
2535 goto err_free;
2536 }
2537 } else {
2538 /* all other non-branch instructions with single
2539 * fall-through edge
2540 */
2541 ret = push_insn(t, t + 1, FALLTHROUGH, env);
2542 if (ret == 1)
2543 goto peek_stack;
2544 else if (ret < 0)
2545 goto err_free;
2546 }
2547
2548 mark_explored:
2549 insn_state[t] = EXPLORED;
2550 if (cur_stack-- <= 0) {
2551 verbose("pop stack internal bug\n");
2552 ret = -EFAULT;
2553 goto err_free;
2554 }
2555 goto peek_stack;
2556
2557 check_state:
2558 for (i = 0; i < insn_cnt; i++) {
2559 if (insn_state[i] != EXPLORED) {
2560 verbose("unreachable insn %d\n", i);
2561 ret = -EINVAL;
2562 goto err_free;
2563 }
2564 }
2565 ret = 0; /* cfg looks good */
2566
2567 err_free:
2568 kfree(insn_state);
2569 kfree(insn_stack);
2570 return ret;
2571 }
2572
2573 /* the following conditions reduce the number of explored insns
2574 * from ~140k to ~80k for ultra large programs that use a lot of ptr_to_packet
2575 */
2576 static bool compare_ptrs_to_packet(struct bpf_reg_state *old,
2577 struct bpf_reg_state *cur)
2578 {
2579 if (old->id != cur->id)
2580 return false;
2581
2582 /* old ptr_to_packet is more conservative, since it allows smaller
2583 * range. Ex:
2584 * old(off=0,r=10) is equal to cur(off=0,r=20), because
2585 * old(off=0,r=10) means that with range=10 the verifier proceeded
2586 * further and found no issues with the program. Now we're in the same
2587 * spot with cur(off=0,r=20), so we're safe too, since anything further
2588 * will only be looking at most 10 bytes after this pointer.
2589 */
2590 if (old->off == cur->off && old->range < cur->range)
2591 return true;
2592
2593 /* old(off=20,r=10) is equal to cur(off=22,re=22 or 5 or 0)
2594 * since both cannot be used for packet access and safe(old)
2595 * pointer has smaller off that could be used for further
2596 * 'if (ptr > data_end)' check
2597 * Ex:
2598 * old(off=20,r=10) and cur(off=22,r=22) and cur(off=22,r=0) mean
2599 * that we cannot access the packet.
2600 * The safe range is:
2601 * [ptr, ptr + range - off)
2602 * so whenever off >=range, it means no safe bytes from this pointer.
2603 * When comparing old->off <= cur->off, it means that older code
2604 * went with smaller offset and that offset was later
2605 * used to figure out the safe range after 'if (ptr > data_end)' check
2606 * Say, 'old' state was explored like:
2607 * ... R3(off=0, r=0)
2608 * R4 = R3 + 20
2609 * ... now R4(off=20,r=0) <-- here
2610 * if (R4 > data_end)
2611 * ... R4(off=20,r=20), R3(off=0,r=20) and R3 can be used to access.
2612 * ... the code further went all the way to bpf_exit.
2613 * Now the 'cur' state at the mark 'here' has R4(off=30,r=0).
2614 * old_R4(off=20,r=0) equal to cur_R4(off=30,r=0), since if the verifier
2615 * goes further, such cur_R4 will give larger safe packet range after
2616 * 'if (R4 > data_end)' and all further insn were already good with r=20,
2617 * so they will be good with r=30 and we can prune the search.
2618 */
2619 if (old->off <= cur->off &&
2620 old->off >= old->range && cur->off >= cur->range)
2621 return true;
2622
2623 return false;
2624 }
2625
2626 /* compare two verifier states
2627 *
2628 * all states stored in state_list are known to be valid, since
2629 * verifier reached 'bpf_exit' instruction through them
2630 *
2631 * this function is called when verifier exploring different branches of
2632 * execution popped from the state stack. If it sees an old state that has
2633 * more strict register state and more strict stack state then this execution
2634 * branch doesn't need to be explored further, since verifier already
2635 * concluded that more strict state leads to valid finish.
2636 *
2637 * Therefore two states are equivalent if register state is more conservative
2638 * and explored stack state is more conservative than the current one.
2639 * Example:
2640 * explored current
2641 * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC)
2642 * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC)
2643 *
2644 * In other words if current stack state (one being explored) has more
2645 * valid slots than old one that already passed validation, it means
2646 * the verifier can stop exploring and conclude that current state is valid too
2647 *
2648 * Similarly with registers. If explored state has register type as invalid
2649 * whereas register type in current state is meaningful, it means that
2650 * the current state will reach 'bpf_exit' instruction safely
2651 */
2652 static bool states_equal(struct bpf_verifier_env *env,
2653 struct bpf_verifier_state *old,
2654 struct bpf_verifier_state *cur)
2655 {
2656 bool varlen_map_access = env->varlen_map_value_access;
2657 struct bpf_reg_state *rold, *rcur;
2658 int i;
2659
2660 for (i = 0; i < MAX_BPF_REG; i++) {
2661 rold = &old->regs[i];
2662 rcur = &cur->regs[i];
2663
2664 if (memcmp(rold, rcur, sizeof(*rold)) == 0)
2665 continue;
2666
2667 /* If the ranges were not the same, but everything else was and
2668 * we didn't do a variable access into a map then we are a-ok.
2669 */
2670 if (!varlen_map_access &&
2671 memcmp(rold, rcur, offsetofend(struct bpf_reg_state, id)) == 0)
2672 continue;
2673
2674 /* If we didn't map access then again we don't care about the
2675 * mismatched range values and it's ok if our old type was
2676 * UNKNOWN and we didn't go to a NOT_INIT'ed reg.
2677 */
2678 if (rold->type == NOT_INIT ||
2679 (!varlen_map_access && rold->type == UNKNOWN_VALUE &&
2680 rcur->type != NOT_INIT))
2681 continue;
2682
2683 if (rold->type == PTR_TO_PACKET && rcur->type == PTR_TO_PACKET &&
2684 compare_ptrs_to_packet(rold, rcur))
2685 continue;
2686
2687 return false;
2688 }
2689
2690 for (i = 0; i < MAX_BPF_STACK; i++) {
2691 if (old->stack_slot_type[i] == STACK_INVALID)
2692 continue;
2693 if (old->stack_slot_type[i] != cur->stack_slot_type[i])
2694 /* Ex: old explored (safe) state has STACK_SPILL in
2695 * this stack slot, but current has has STACK_MISC ->
2696 * this verifier states are not equivalent,
2697 * return false to continue verification of this path
2698 */
2699 return false;
2700 if (i % BPF_REG_SIZE)
2701 continue;
2702 if (memcmp(&old->spilled_regs[i / BPF_REG_SIZE],
2703 &cur->spilled_regs[i / BPF_REG_SIZE],
2704 sizeof(old->spilled_regs[0])))
2705 /* when explored and current stack slot types are
2706 * the same, check that stored pointers types
2707 * are the same as well.
2708 * Ex: explored safe path could have stored
2709 * (bpf_reg_state) {.type = PTR_TO_STACK, .imm = -8}
2710 * but current path has stored:
2711 * (bpf_reg_state) {.type = PTR_TO_STACK, .imm = -16}
2712 * such verifier states are not equivalent.
2713 * return false to continue verification of this path
2714 */
2715 return false;
2716 else
2717 continue;
2718 }
2719 return true;
2720 }
2721
2722 static int is_state_visited(struct bpf_verifier_env *env, int insn_idx)
2723 {
2724 struct bpf_verifier_state_list *new_sl;
2725 struct bpf_verifier_state_list *sl;
2726
2727 sl = env->explored_states[insn_idx];
2728 if (!sl)
2729 /* this 'insn_idx' instruction wasn't marked, so we will not
2730 * be doing state search here
2731 */
2732 return 0;
2733
2734 while (sl != STATE_LIST_MARK) {
2735 if (states_equal(env, &sl->state, &env->cur_state))
2736 /* reached equivalent register/stack state,
2737 * prune the search
2738 */
2739 return 1;
2740 sl = sl->next;
2741 }
2742
2743 /* there were no equivalent states, remember current one.
2744 * technically the current state is not proven to be safe yet,
2745 * but it will either reach bpf_exit (which means it's safe) or
2746 * it will be rejected. Since there are no loops, we won't be
2747 * seeing this 'insn_idx' instruction again on the way to bpf_exit
2748 */
2749 new_sl = kmalloc(sizeof(struct bpf_verifier_state_list), GFP_USER);
2750 if (!new_sl)
2751 return -ENOMEM;
2752
2753 /* add new state to the head of linked list */
2754 memcpy(&new_sl->state, &env->cur_state, sizeof(env->cur_state));
2755 new_sl->next = env->explored_states[insn_idx];
2756 env->explored_states[insn_idx] = new_sl;
2757 return 0;
2758 }
2759
2760 static int ext_analyzer_insn_hook(struct bpf_verifier_env *env,
2761 int insn_idx, int prev_insn_idx)
2762 {
2763 if (!env->analyzer_ops || !env->analyzer_ops->insn_hook)
2764 return 0;
2765
2766 return env->analyzer_ops->insn_hook(env, insn_idx, prev_insn_idx);
2767 }
2768
2769 static int do_check(struct bpf_verifier_env *env)
2770 {
2771 struct bpf_verifier_state *state = &env->cur_state;
2772 struct bpf_insn *insns = env->prog->insnsi;
2773 struct bpf_reg_state *regs = state->regs;
2774 int insn_cnt = env->prog->len;
2775 int insn_idx, prev_insn_idx = 0;
2776 int insn_processed = 0;
2777 bool do_print_state = false;
2778
2779 init_reg_state(regs);
2780 insn_idx = 0;
2781 env->varlen_map_value_access = false;
2782 for (;;) {
2783 struct bpf_insn *insn;
2784 u8 class;
2785 int err;
2786
2787 if (insn_idx >= insn_cnt) {
2788 verbose("invalid insn idx %d insn_cnt %d\n",
2789 insn_idx, insn_cnt);
2790 return -EFAULT;
2791 }
2792
2793 insn = &insns[insn_idx];
2794 class = BPF_CLASS(insn->code);
2795
2796 if (++insn_processed > BPF_COMPLEXITY_LIMIT_INSNS) {
2797 verbose("BPF program is too large. Processed %d insn\n",
2798 insn_processed);
2799 return -E2BIG;
2800 }
2801
2802 err = is_state_visited(env, insn_idx);
2803 if (err < 0)
2804 return err;
2805 if (err == 1) {
2806 /* found equivalent state, can prune the search */
2807 if (log_level) {
2808 if (do_print_state)
2809 verbose("\nfrom %d to %d: safe\n",
2810 prev_insn_idx, insn_idx);
2811 else
2812 verbose("%d: safe\n", insn_idx);
2813 }
2814 goto process_bpf_exit;
2815 }
2816
2817 if (log_level && do_print_state) {
2818 verbose("\nfrom %d to %d:", prev_insn_idx, insn_idx);
2819 print_verifier_state(&env->cur_state);
2820 do_print_state = false;
2821 }
2822
2823 if (log_level) {
2824 verbose("%d: ", insn_idx);
2825 print_bpf_insn(insn);
2826 }
2827
2828 err = ext_analyzer_insn_hook(env, insn_idx, prev_insn_idx);
2829 if (err)
2830 return err;
2831
2832 if (class == BPF_ALU || class == BPF_ALU64) {
2833 err = check_alu_op(env, insn);
2834 if (err)
2835 return err;
2836
2837 } else if (class == BPF_LDX) {
2838 enum bpf_reg_type *prev_src_type, src_reg_type;
2839
2840 /* check for reserved fields is already done */
2841
2842 /* check src operand */
2843 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
2844 if (err)
2845 return err;
2846
2847 err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK);
2848 if (err)
2849 return err;
2850
2851 src_reg_type = regs[insn->src_reg].type;
2852
2853 /* check that memory (src_reg + off) is readable,
2854 * the state of dst_reg will be updated by this func
2855 */
2856 err = check_mem_access(env, insn->src_reg, insn->off,
2857 BPF_SIZE(insn->code), BPF_READ,
2858 insn->dst_reg);
2859 if (err)
2860 return err;
2861
2862 if (BPF_SIZE(insn->code) != BPF_W &&
2863 BPF_SIZE(insn->code) != BPF_DW) {
2864 insn_idx++;
2865 continue;
2866 }
2867
2868 prev_src_type = &env->insn_aux_data[insn_idx].ptr_type;
2869
2870 if (*prev_src_type == NOT_INIT) {
2871 /* saw a valid insn
2872 * dst_reg = *(u32 *)(src_reg + off)
2873 * save type to validate intersecting paths
2874 */
2875 *prev_src_type = src_reg_type;
2876
2877 } else if (src_reg_type != *prev_src_type &&
2878 (src_reg_type == PTR_TO_CTX ||
2879 *prev_src_type == PTR_TO_CTX)) {
2880 /* ABuser program is trying to use the same insn
2881 * dst_reg = *(u32*) (src_reg + off)
2882 * with different pointer types:
2883 * src_reg == ctx in one branch and
2884 * src_reg == stack|map in some other branch.
2885 * Reject it.
2886 */
2887 verbose("same insn cannot be used with different pointers\n");
2888 return -EINVAL;
2889 }
2890
2891 } else if (class == BPF_STX) {
2892 enum bpf_reg_type *prev_dst_type, dst_reg_type;
2893
2894 if (BPF_MODE(insn->code) == BPF_XADD) {
2895 err = check_xadd(env, insn);
2896 if (err)
2897 return err;
2898 insn_idx++;
2899 continue;
2900 }
2901
2902 /* check src1 operand */
2903 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
2904 if (err)
2905 return err;
2906 /* check src2 operand */
2907 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
2908 if (err)
2909 return err;
2910
2911 dst_reg_type = regs[insn->dst_reg].type;
2912
2913 /* check that memory (dst_reg + off) is writeable */
2914 err = check_mem_access(env, insn->dst_reg, insn->off,
2915 BPF_SIZE(insn->code), BPF_WRITE,
2916 insn->src_reg);
2917 if (err)
2918 return err;
2919
2920 prev_dst_type = &env->insn_aux_data[insn_idx].ptr_type;
2921
2922 if (*prev_dst_type == NOT_INIT) {
2923 *prev_dst_type = dst_reg_type;
2924 } else if (dst_reg_type != *prev_dst_type &&
2925 (dst_reg_type == PTR_TO_CTX ||
2926 *prev_dst_type == PTR_TO_CTX)) {
2927 verbose("same insn cannot be used with different pointers\n");
2928 return -EINVAL;
2929 }
2930
2931 } else if (class == BPF_ST) {
2932 if (BPF_MODE(insn->code) != BPF_MEM ||
2933 insn->src_reg != BPF_REG_0) {
2934 verbose("BPF_ST uses reserved fields\n");
2935 return -EINVAL;
2936 }
2937 /* check src operand */
2938 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
2939 if (err)
2940 return err;
2941
2942 /* check that memory (dst_reg + off) is writeable */
2943 err = check_mem_access(env, insn->dst_reg, insn->off,
2944 BPF_SIZE(insn->code), BPF_WRITE,
2945 -1);
2946 if (err)
2947 return err;
2948
2949 } else if (class == BPF_JMP) {
2950 u8 opcode = BPF_OP(insn->code);
2951
2952 if (opcode == BPF_CALL) {
2953 if (BPF_SRC(insn->code) != BPF_K ||
2954 insn->off != 0 ||
2955 insn->src_reg != BPF_REG_0 ||
2956 insn->dst_reg != BPF_REG_0) {
2957 verbose("BPF_CALL uses reserved fields\n");
2958 return -EINVAL;
2959 }
2960
2961 err = check_call(env, insn->imm, insn_idx);
2962 if (err)
2963 return err;
2964
2965 } else if (opcode == BPF_JA) {
2966 if (BPF_SRC(insn->code) != BPF_K ||
2967 insn->imm != 0 ||
2968 insn->src_reg != BPF_REG_0 ||
2969 insn->dst_reg != BPF_REG_0) {
2970 verbose("BPF_JA uses reserved fields\n");
2971 return -EINVAL;
2972 }
2973
2974 insn_idx += insn->off + 1;
2975 continue;
2976
2977 } else if (opcode == BPF_EXIT) {
2978 if (BPF_SRC(insn->code) != BPF_K ||
2979 insn->imm != 0 ||
2980 insn->src_reg != BPF_REG_0 ||
2981 insn->dst_reg != BPF_REG_0) {
2982 verbose("BPF_EXIT uses reserved fields\n");
2983 return -EINVAL;
2984 }
2985
2986 /* eBPF calling convetion is such that R0 is used
2987 * to return the value from eBPF program.
2988 * Make sure that it's readable at this time
2989 * of bpf_exit, which means that program wrote
2990 * something into it earlier
2991 */
2992 err = check_reg_arg(regs, BPF_REG_0, SRC_OP);
2993 if (err)
2994 return err;
2995
2996 if (is_pointer_value(env, BPF_REG_0)) {
2997 verbose("R0 leaks addr as return value\n");
2998 return -EACCES;
2999 }
3000
3001 process_bpf_exit:
3002 insn_idx = pop_stack(env, &prev_insn_idx);
3003 if (insn_idx < 0) {
3004 break;
3005 } else {
3006 do_print_state = true;
3007 continue;
3008 }
3009 } else {
3010 err = check_cond_jmp_op(env, insn, &insn_idx);
3011 if (err)
3012 return err;
3013 }
3014 } else if (class == BPF_LD) {
3015 u8 mode = BPF_MODE(insn->code);
3016
3017 if (mode == BPF_ABS || mode == BPF_IND) {
3018 err = check_ld_abs(env, insn);
3019 if (err)
3020 return err;
3021
3022 } else if (mode == BPF_IMM) {
3023 err = check_ld_imm(env, insn);
3024 if (err)
3025 return err;
3026
3027 insn_idx++;
3028 } else {
3029 verbose("invalid BPF_LD mode\n");
3030 return -EINVAL;
3031 }
3032 reset_reg_range_values(regs, insn->dst_reg);
3033 } else {
3034 verbose("unknown insn class %d\n", class);
3035 return -EINVAL;
3036 }
3037
3038 insn_idx++;
3039 }
3040
3041 verbose("processed %d insns\n", insn_processed);
3042 return 0;
3043 }
3044
3045 static int check_map_prealloc(struct bpf_map *map)
3046 {
3047 return (map->map_type != BPF_MAP_TYPE_HASH &&
3048 map->map_type != BPF_MAP_TYPE_PERCPU_HASH &&
3049 map->map_type != BPF_MAP_TYPE_HASH_OF_MAPS) ||
3050 !(map->map_flags & BPF_F_NO_PREALLOC);
3051 }
3052
3053 static int check_map_prog_compatibility(struct bpf_map *map,
3054 struct bpf_prog *prog)
3055
3056 {
3057 /* Make sure that BPF_PROG_TYPE_PERF_EVENT programs only use
3058 * preallocated hash maps, since doing memory allocation
3059 * in overflow_handler can crash depending on where nmi got
3060 * triggered.
3061 */
3062 if (prog->type == BPF_PROG_TYPE_PERF_EVENT) {
3063 if (!check_map_prealloc(map)) {
3064 verbose("perf_event programs can only use preallocated hash map\n");
3065 return -EINVAL;
3066 }
3067 if (map->inner_map_meta &&
3068 !check_map_prealloc(map->inner_map_meta)) {
3069 verbose("perf_event programs can only use preallocated inner hash map\n");
3070 return -EINVAL;
3071 }
3072 }
3073 return 0;
3074 }
3075
3076 /* look for pseudo eBPF instructions that access map FDs and
3077 * replace them with actual map pointers
3078 */
3079 static int replace_map_fd_with_map_ptr(struct bpf_verifier_env *env)
3080 {
3081 struct bpf_insn *insn = env->prog->insnsi;
3082 int insn_cnt = env->prog->len;
3083 int i, j, err;
3084
3085 err = bpf_prog_calc_tag(env->prog);
3086 if (err)
3087 return err;
3088
3089 for (i = 0; i < insn_cnt; i++, insn++) {
3090 if (BPF_CLASS(insn->code) == BPF_LDX &&
3091 (BPF_MODE(insn->code) != BPF_MEM || insn->imm != 0)) {
3092 verbose("BPF_LDX uses reserved fields\n");
3093 return -EINVAL;
3094 }
3095
3096 if (BPF_CLASS(insn->code) == BPF_STX &&
3097 ((BPF_MODE(insn->code) != BPF_MEM &&
3098 BPF_MODE(insn->code) != BPF_XADD) || insn->imm != 0)) {
3099 verbose("BPF_STX uses reserved fields\n");
3100 return -EINVAL;
3101 }
3102
3103 if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW)) {
3104 struct bpf_map *map;
3105 struct fd f;
3106
3107 if (i == insn_cnt - 1 || insn[1].code != 0 ||
3108 insn[1].dst_reg != 0 || insn[1].src_reg != 0 ||
3109 insn[1].off != 0) {
3110 verbose("invalid bpf_ld_imm64 insn\n");
3111 return -EINVAL;
3112 }
3113
3114 if (insn->src_reg == 0)
3115 /* valid generic load 64-bit imm */
3116 goto next_insn;
3117
3118 if (insn->src_reg != BPF_PSEUDO_MAP_FD) {
3119 verbose("unrecognized bpf_ld_imm64 insn\n");
3120 return -EINVAL;
3121 }
3122
3123 f = fdget(insn->imm);
3124 map = __bpf_map_get(f);
3125 if (IS_ERR(map)) {
3126 verbose("fd %d is not pointing to valid bpf_map\n",
3127 insn->imm);
3128 return PTR_ERR(map);
3129 }
3130
3131 err = check_map_prog_compatibility(map, env->prog);
3132 if (err) {
3133 fdput(f);
3134 return err;
3135 }
3136
3137 /* store map pointer inside BPF_LD_IMM64 instruction */
3138 insn[0].imm = (u32) (unsigned long) map;
3139 insn[1].imm = ((u64) (unsigned long) map) >> 32;
3140
3141 /* check whether we recorded this map already */
3142 for (j = 0; j < env->used_map_cnt; j++)
3143 if (env->used_maps[j] == map) {
3144 fdput(f);
3145 goto next_insn;
3146 }
3147
3148 if (env->used_map_cnt >= MAX_USED_MAPS) {
3149 fdput(f);
3150 return -E2BIG;
3151 }
3152
3153 /* hold the map. If the program is rejected by verifier,
3154 * the map will be released by release_maps() or it
3155 * will be used by the valid program until it's unloaded
3156 * and all maps are released in free_bpf_prog_info()
3157 */
3158 map = bpf_map_inc(map, false);
3159 if (IS_ERR(map)) {
3160 fdput(f);
3161 return PTR_ERR(map);
3162 }
3163 env->used_maps[env->used_map_cnt++] = map;
3164
3165 fdput(f);
3166 next_insn:
3167 insn++;
3168 i++;
3169 }
3170 }
3171
3172 /* now all pseudo BPF_LD_IMM64 instructions load valid
3173 * 'struct bpf_map *' into a register instead of user map_fd.
3174 * These pointers will be used later by verifier to validate map access.
3175 */
3176 return 0;
3177 }
3178
3179 /* drop refcnt of maps used by the rejected program */
3180 static void release_maps(struct bpf_verifier_env *env)
3181 {
3182 int i;
3183
3184 for (i = 0; i < env->used_map_cnt; i++)
3185 bpf_map_put(env->used_maps[i]);
3186 }
3187
3188 /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */
3189 static void convert_pseudo_ld_imm64(struct bpf_verifier_env *env)
3190 {
3191 struct bpf_insn *insn = env->prog->insnsi;
3192 int insn_cnt = env->prog->len;
3193 int i;
3194
3195 for (i = 0; i < insn_cnt; i++, insn++)
3196 if (insn->code == (BPF_LD | BPF_IMM | BPF_DW))
3197 insn->src_reg = 0;
3198 }
3199
3200 /* single env->prog->insni[off] instruction was replaced with the range
3201 * insni[off, off + cnt). Adjust corresponding insn_aux_data by copying
3202 * [0, off) and [off, end) to new locations, so the patched range stays zero
3203 */
3204 static int adjust_insn_aux_data(struct bpf_verifier_env *env, u32 prog_len,
3205 u32 off, u32 cnt)
3206 {
3207 struct bpf_insn_aux_data *new_data, *old_data = env->insn_aux_data;
3208
3209 if (cnt == 1)
3210 return 0;
3211 new_data = vzalloc(sizeof(struct bpf_insn_aux_data) * prog_len);
3212 if (!new_data)
3213 return -ENOMEM;
3214 memcpy(new_data, old_data, sizeof(struct bpf_insn_aux_data) * off);
3215 memcpy(new_data + off + cnt - 1, old_data + off,
3216 sizeof(struct bpf_insn_aux_data) * (prog_len - off - cnt + 1));
3217 env->insn_aux_data = new_data;
3218 vfree(old_data);
3219 return 0;
3220 }
3221
3222 static struct bpf_prog *bpf_patch_insn_data(struct bpf_verifier_env *env, u32 off,
3223 const struct bpf_insn *patch, u32 len)
3224 {
3225 struct bpf_prog *new_prog;
3226
3227 new_prog = bpf_patch_insn_single(env->prog, off, patch, len);
3228 if (!new_prog)
3229 return NULL;
3230 if (adjust_insn_aux_data(env, new_prog->len, off, len))
3231 return NULL;
3232 return new_prog;
3233 }
3234
3235 /* convert load instructions that access fields of 'struct __sk_buff'
3236 * into sequence of instructions that access fields of 'struct sk_buff'
3237 */
3238 static int convert_ctx_accesses(struct bpf_verifier_env *env)
3239 {
3240 const struct bpf_verifier_ops *ops = env->prog->aux->ops;
3241 const int insn_cnt = env->prog->len;
3242 struct bpf_insn insn_buf[16], *insn;
3243 struct bpf_prog *new_prog;
3244 enum bpf_access_type type;
3245 int i, cnt, delta = 0;
3246
3247 if (ops->gen_prologue) {
3248 cnt = ops->gen_prologue(insn_buf, env->seen_direct_write,
3249 env->prog);
3250 if (cnt >= ARRAY_SIZE(insn_buf)) {
3251 verbose("bpf verifier is misconfigured\n");
3252 return -EINVAL;
3253 } else if (cnt) {
3254 new_prog = bpf_patch_insn_data(env, 0, insn_buf, cnt);
3255 if (!new_prog)
3256 return -ENOMEM;
3257
3258 env->prog = new_prog;
3259 delta += cnt - 1;
3260 }
3261 }
3262
3263 if (!ops->convert_ctx_access)
3264 return 0;
3265
3266 insn = env->prog->insnsi + delta;
3267
3268 for (i = 0; i < insn_cnt; i++, insn++) {
3269 if (insn->code == (BPF_LDX | BPF_MEM | BPF_B) ||
3270 insn->code == (BPF_LDX | BPF_MEM | BPF_H) ||
3271 insn->code == (BPF_LDX | BPF_MEM | BPF_W) ||
3272 insn->code == (BPF_LDX | BPF_MEM | BPF_DW))
3273 type = BPF_READ;
3274 else if (insn->code == (BPF_STX | BPF_MEM | BPF_B) ||
3275 insn->code == (BPF_STX | BPF_MEM | BPF_H) ||
3276 insn->code == (BPF_STX | BPF_MEM | BPF_W) ||
3277 insn->code == (BPF_STX | BPF_MEM | BPF_DW))
3278 type = BPF_WRITE;
3279 else
3280 continue;
3281
3282 if (env->insn_aux_data[i + delta].ptr_type != PTR_TO_CTX)
3283 continue;
3284
3285 cnt = ops->convert_ctx_access(type, insn, insn_buf, env->prog);
3286 if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) {
3287 verbose("bpf verifier is misconfigured\n");
3288 return -EINVAL;
3289 }
3290
3291 new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt);
3292 if (!new_prog)
3293 return -ENOMEM;
3294
3295 delta += cnt - 1;
3296
3297 /* keep walking new program and skip insns we just inserted */
3298 env->prog = new_prog;
3299 insn = new_prog->insnsi + i + delta;
3300 }
3301
3302 return 0;
3303 }
3304
3305 /* fixup insn->imm field of bpf_call instructions
3306 * and inline eligible helpers as explicit sequence of BPF instructions
3307 *
3308 * this function is called after eBPF program passed verification
3309 */
3310 static int fixup_bpf_calls(struct bpf_verifier_env *env)
3311 {
3312 struct bpf_prog *prog = env->prog;
3313 struct bpf_insn *insn = prog->insnsi;
3314 const struct bpf_func_proto *fn;
3315 const int insn_cnt = prog->len;
3316 struct bpf_insn insn_buf[16];
3317 struct bpf_prog *new_prog;
3318 struct bpf_map *map_ptr;
3319 int i, cnt, delta = 0;
3320
3321 for (i = 0; i < insn_cnt; i++, insn++) {
3322 if (insn->code != (BPF_JMP | BPF_CALL))
3323 continue;
3324
3325 if (insn->imm == BPF_FUNC_get_route_realm)
3326 prog->dst_needed = 1;
3327 if (insn->imm == BPF_FUNC_get_prandom_u32)
3328 bpf_user_rnd_init_once();
3329 if (insn->imm == BPF_FUNC_xdp_adjust_head)
3330 prog->xdp_adjust_head = 1;
3331 if (insn->imm == BPF_FUNC_tail_call) {
3332 /* mark bpf_tail_call as different opcode to avoid
3333 * conditional branch in the interpeter for every normal
3334 * call and to prevent accidental JITing by JIT compiler
3335 * that doesn't support bpf_tail_call yet
3336 */
3337 insn->imm = 0;
3338 insn->code |= BPF_X;
3339 continue;
3340 }
3341
3342 if (ebpf_jit_enabled() && insn->imm == BPF_FUNC_map_lookup_elem) {
3343 map_ptr = env->insn_aux_data[i + delta].map_ptr;
3344 if (map_ptr == BPF_MAP_PTR_POISON ||
3345 !map_ptr->ops->map_gen_lookup)
3346 goto patch_call_imm;
3347
3348 cnt = map_ptr->ops->map_gen_lookup(map_ptr, insn_buf);
3349 if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) {
3350 verbose("bpf verifier is misconfigured\n");
3351 return -EINVAL;
3352 }
3353
3354 new_prog = bpf_patch_insn_data(env, i + delta, insn_buf,
3355 cnt);
3356 if (!new_prog)
3357 return -ENOMEM;
3358
3359 delta += cnt - 1;
3360
3361 /* keep walking new program and skip insns we just inserted */
3362 env->prog = prog = new_prog;
3363 insn = new_prog->insnsi + i + delta;
3364 continue;
3365 }
3366
3367 patch_call_imm:
3368 fn = prog->aux->ops->get_func_proto(insn->imm);
3369 /* all functions that have prototype and verifier allowed
3370 * programs to call them, must be real in-kernel functions
3371 */
3372 if (!fn->func) {
3373 verbose("kernel subsystem misconfigured func %s#%d\n",
3374 func_id_name(insn->imm), insn->imm);
3375 return -EFAULT;
3376 }
3377 insn->imm = fn->func - __bpf_call_base;
3378 }
3379
3380 return 0;
3381 }
3382
3383 static void free_states(struct bpf_verifier_env *env)
3384 {
3385 struct bpf_verifier_state_list *sl, *sln;
3386 int i;
3387
3388 if (!env->explored_states)
3389 return;
3390
3391 for (i = 0; i < env->prog->len; i++) {
3392 sl = env->explored_states[i];
3393
3394 if (sl)
3395 while (sl != STATE_LIST_MARK) {
3396 sln = sl->next;
3397 kfree(sl);
3398 sl = sln;
3399 }
3400 }
3401
3402 kfree(env->explored_states);
3403 }
3404
3405 int bpf_check(struct bpf_prog **prog, union bpf_attr *attr)
3406 {
3407 char __user *log_ubuf = NULL;
3408 struct bpf_verifier_env *env;
3409 int ret = -EINVAL;
3410
3411 /* 'struct bpf_verifier_env' can be global, but since it's not small,
3412 * allocate/free it every time bpf_check() is called
3413 */
3414 env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL);
3415 if (!env)
3416 return -ENOMEM;
3417
3418 env->insn_aux_data = vzalloc(sizeof(struct bpf_insn_aux_data) *
3419 (*prog)->len);
3420 ret = -ENOMEM;
3421 if (!env->insn_aux_data)
3422 goto err_free_env;
3423 env->prog = *prog;
3424
3425 /* grab the mutex to protect few globals used by verifier */
3426 mutex_lock(&bpf_verifier_lock);
3427
3428 if (attr->log_level || attr->log_buf || attr->log_size) {
3429 /* user requested verbose verifier output
3430 * and supplied buffer to store the verification trace
3431 */
3432 log_level = attr->log_level;
3433 log_ubuf = (char __user *) (unsigned long) attr->log_buf;
3434 log_size = attr->log_size;
3435 log_len = 0;
3436
3437 ret = -EINVAL;
3438 /* log_* values have to be sane */
3439 if (log_size < 128 || log_size > UINT_MAX >> 8 ||
3440 log_level == 0 || log_ubuf == NULL)
3441 goto err_unlock;
3442
3443 ret = -ENOMEM;
3444 log_buf = vmalloc(log_size);
3445 if (!log_buf)
3446 goto err_unlock;
3447 } else {
3448 log_level = 0;
3449 }
3450
3451 ret = replace_map_fd_with_map_ptr(env);
3452 if (ret < 0)
3453 goto skip_full_check;
3454
3455 env->explored_states = kcalloc(env->prog->len,
3456 sizeof(struct bpf_verifier_state_list *),
3457 GFP_USER);
3458 ret = -ENOMEM;
3459 if (!env->explored_states)
3460 goto skip_full_check;
3461
3462 ret = check_cfg(env);
3463 if (ret < 0)
3464 goto skip_full_check;
3465
3466 env->allow_ptr_leaks = capable(CAP_SYS_ADMIN);
3467
3468 ret = do_check(env);
3469
3470 skip_full_check:
3471 while (pop_stack(env, NULL) >= 0);
3472 free_states(env);
3473
3474 if (ret == 0)
3475 /* program is valid, convert *(u32*)(ctx + off) accesses */
3476 ret = convert_ctx_accesses(env);
3477
3478 if (ret == 0)
3479 ret = fixup_bpf_calls(env);
3480
3481 if (log_level && log_len >= log_size - 1) {
3482 BUG_ON(log_len >= log_size);
3483 /* verifier log exceeded user supplied buffer */
3484 ret = -ENOSPC;
3485 /* fall through to return what was recorded */
3486 }
3487
3488 /* copy verifier log back to user space including trailing zero */
3489 if (log_level && copy_to_user(log_ubuf, log_buf, log_len + 1) != 0) {
3490 ret = -EFAULT;
3491 goto free_log_buf;
3492 }
3493
3494 if (ret == 0 && env->used_map_cnt) {
3495 /* if program passed verifier, update used_maps in bpf_prog_info */
3496 env->prog->aux->used_maps = kmalloc_array(env->used_map_cnt,
3497 sizeof(env->used_maps[0]),
3498 GFP_KERNEL);
3499
3500 if (!env->prog->aux->used_maps) {
3501 ret = -ENOMEM;
3502 goto free_log_buf;
3503 }
3504
3505 memcpy(env->prog->aux->used_maps, env->used_maps,
3506 sizeof(env->used_maps[0]) * env->used_map_cnt);
3507 env->prog->aux->used_map_cnt = env->used_map_cnt;
3508
3509 /* program is valid. Convert pseudo bpf_ld_imm64 into generic
3510 * bpf_ld_imm64 instructions
3511 */
3512 convert_pseudo_ld_imm64(env);
3513 }
3514
3515 free_log_buf:
3516 if (log_level)
3517 vfree(log_buf);
3518 if (!env->prog->aux->used_maps)
3519 /* if we didn't copy map pointers into bpf_prog_info, release
3520 * them now. Otherwise free_bpf_prog_info() will release them.
3521 */
3522 release_maps(env);
3523 *prog = env->prog;
3524 err_unlock:
3525 mutex_unlock(&bpf_verifier_lock);
3526 vfree(env->insn_aux_data);
3527 err_free_env:
3528 kfree(env);
3529 return ret;
3530 }
3531
3532 int bpf_analyzer(struct bpf_prog *prog, const struct bpf_ext_analyzer_ops *ops,
3533 void *priv)
3534 {
3535 struct bpf_verifier_env *env;
3536 int ret;
3537
3538 env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL);
3539 if (!env)
3540 return -ENOMEM;
3541
3542 env->insn_aux_data = vzalloc(sizeof(struct bpf_insn_aux_data) *
3543 prog->len);
3544 ret = -ENOMEM;
3545 if (!env->insn_aux_data)
3546 goto err_free_env;
3547 env->prog = prog;
3548 env->analyzer_ops = ops;
3549 env->analyzer_priv = priv;
3550
3551 /* grab the mutex to protect few globals used by verifier */
3552 mutex_lock(&bpf_verifier_lock);
3553
3554 log_level = 0;
3555
3556 env->explored_states = kcalloc(env->prog->len,
3557 sizeof(struct bpf_verifier_state_list *),
3558 GFP_KERNEL);
3559 ret = -ENOMEM;
3560 if (!env->explored_states)
3561 goto skip_full_check;
3562
3563 ret = check_cfg(env);
3564 if (ret < 0)
3565 goto skip_full_check;
3566
3567 env->allow_ptr_leaks = capable(CAP_SYS_ADMIN);
3568
3569 ret = do_check(env);
3570
3571 skip_full_check:
3572 while (pop_stack(env, NULL) >= 0);
3573 free_states(env);
3574
3575 mutex_unlock(&bpf_verifier_lock);
3576 vfree(env->insn_aux_data);
3577 err_free_env:
3578 kfree(env);
3579 return ret;
3580 }
3581 EXPORT_SYMBOL_GPL(bpf_analyzer);
ubuntu-bionic-kernel mirror