xref: /linux/kernel/bpf/verifier.c (revision 80d443e8876602be2c130f79c4de81e12e2a700d)
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 32k 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 struct bpf_call_arg_meta {
147 	struct bpf_map *map_ptr;
148 	bool raw_mode;
149 	bool pkt_access;
150 	int regno;
151 	int access_size;
152 };
153 
154 /* verbose verifier prints what it's seeing
155  * bpf_check() is called under lock, so no race to access these global vars
156  */
157 static u32 log_level, log_size, log_len;
158 static char *log_buf;
159 
160 static DEFINE_MUTEX(bpf_verifier_lock);
161 
162 /* log_level controls verbosity level of eBPF verifier.
163  * verbose() is used to dump the verification trace to the log, so the user
164  * can figure out what's wrong with the program
165  */
166 static __printf(1, 2) void verbose(const char *fmt, ...)
167 {
168 	va_list args;
169 
170 	if (log_level == 0 || log_len >= log_size - 1)
171 		return;
172 
173 	va_start(args, fmt);
174 	log_len += vscnprintf(log_buf + log_len, log_size - log_len, fmt, args);
175 	va_end(args);
176 }
177 
178 /* string representation of 'enum bpf_reg_type' */
179 static const char * const reg_type_str[] = {
180 	[NOT_INIT]		= "?",
181 	[UNKNOWN_VALUE]		= "inv",
182 	[PTR_TO_CTX]		= "ctx",
183 	[CONST_PTR_TO_MAP]	= "map_ptr",
184 	[PTR_TO_MAP_VALUE]	= "map_value",
185 	[PTR_TO_MAP_VALUE_OR_NULL] = "map_value_or_null",
186 	[PTR_TO_MAP_VALUE_ADJ]	= "map_value_adj",
187 	[FRAME_PTR]		= "fp",
188 	[PTR_TO_STACK]		= "fp",
189 	[CONST_IMM]		= "imm",
190 	[PTR_TO_PACKET]		= "pkt",
191 	[PTR_TO_PACKET_END]	= "pkt_end",
192 };
193 
194 #define __BPF_FUNC_STR_FN(x) [BPF_FUNC_ ## x] = __stringify(bpf_ ## x)
195 static const char * const func_id_str[] = {
196 	__BPF_FUNC_MAPPER(__BPF_FUNC_STR_FN)
197 };
198 #undef __BPF_FUNC_STR_FN
199 
200 static const char *func_id_name(int id)
201 {
202 	BUILD_BUG_ON(ARRAY_SIZE(func_id_str) != __BPF_FUNC_MAX_ID);
203 
204 	if (id >= 0 && id < __BPF_FUNC_MAX_ID && func_id_str[id])
205 		return func_id_str[id];
206 	else
207 		return "unknown";
208 }
209 
210 static void print_verifier_state(struct bpf_verifier_state *state)
211 {
212 	struct bpf_reg_state *reg;
213 	enum bpf_reg_type t;
214 	int i;
215 
216 	for (i = 0; i < MAX_BPF_REG; i++) {
217 		reg = &state->regs[i];
218 		t = reg->type;
219 		if (t == NOT_INIT)
220 			continue;
221 		verbose(" R%d=%s", i, reg_type_str[t]);
222 		if (t == CONST_IMM || t == PTR_TO_STACK)
223 			verbose("%lld", reg->imm);
224 		else if (t == PTR_TO_PACKET)
225 			verbose("(id=%d,off=%d,r=%d)",
226 				reg->id, reg->off, reg->range);
227 		else if (t == UNKNOWN_VALUE && reg->imm)
228 			verbose("%lld", reg->imm);
229 		else if (t == CONST_PTR_TO_MAP || t == PTR_TO_MAP_VALUE ||
230 			 t == PTR_TO_MAP_VALUE_OR_NULL ||
231 			 t == PTR_TO_MAP_VALUE_ADJ)
232 			verbose("(ks=%d,vs=%d,id=%u)",
233 				reg->map_ptr->key_size,
234 				reg->map_ptr->value_size,
235 				reg->id);
236 		if (reg->min_value != BPF_REGISTER_MIN_RANGE)
237 			verbose(",min_value=%lld",
238 				(long long)reg->min_value);
239 		if (reg->max_value != BPF_REGISTER_MAX_RANGE)
240 			verbose(",max_value=%llu",
241 				(unsigned long long)reg->max_value);
242 	}
243 	for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
244 		if (state->stack_slot_type[i] == STACK_SPILL)
245 			verbose(" fp%d=%s", -MAX_BPF_STACK + i,
246 				reg_type_str[state->spilled_regs[i / BPF_REG_SIZE].type]);
247 	}
248 	verbose("\n");
249 }
250 
251 static const char *const bpf_class_string[] = {
252 	[BPF_LD]    = "ld",
253 	[BPF_LDX]   = "ldx",
254 	[BPF_ST]    = "st",
255 	[BPF_STX]   = "stx",
256 	[BPF_ALU]   = "alu",
257 	[BPF_JMP]   = "jmp",
258 	[BPF_RET]   = "BUG",
259 	[BPF_ALU64] = "alu64",
260 };
261 
262 static const char *const bpf_alu_string[16] = {
263 	[BPF_ADD >> 4]  = "+=",
264 	[BPF_SUB >> 4]  = "-=",
265 	[BPF_MUL >> 4]  = "*=",
266 	[BPF_DIV >> 4]  = "/=",
267 	[BPF_OR  >> 4]  = "|=",
268 	[BPF_AND >> 4]  = "&=",
269 	[BPF_LSH >> 4]  = "<<=",
270 	[BPF_RSH >> 4]  = ">>=",
271 	[BPF_NEG >> 4]  = "neg",
272 	[BPF_MOD >> 4]  = "%=",
273 	[BPF_XOR >> 4]  = "^=",
274 	[BPF_MOV >> 4]  = "=",
275 	[BPF_ARSH >> 4] = "s>>=",
276 	[BPF_END >> 4]  = "endian",
277 };
278 
279 static const char *const bpf_ldst_string[] = {
280 	[BPF_W >> 3]  = "u32",
281 	[BPF_H >> 3]  = "u16",
282 	[BPF_B >> 3]  = "u8",
283 	[BPF_DW >> 3] = "u64",
284 };
285 
286 static const char *const bpf_jmp_string[16] = {
287 	[BPF_JA >> 4]   = "jmp",
288 	[BPF_JEQ >> 4]  = "==",
289 	[BPF_JGT >> 4]  = ">",
290 	[BPF_JGE >> 4]  = ">=",
291 	[BPF_JSET >> 4] = "&",
292 	[BPF_JNE >> 4]  = "!=",
293 	[BPF_JSGT >> 4] = "s>",
294 	[BPF_JSGE >> 4] = "s>=",
295 	[BPF_CALL >> 4] = "call",
296 	[BPF_EXIT >> 4] = "exit",
297 };
298 
299 static void print_bpf_insn(struct bpf_insn *insn)
300 {
301 	u8 class = BPF_CLASS(insn->code);
302 
303 	if (class == BPF_ALU || class == BPF_ALU64) {
304 		if (BPF_SRC(insn->code) == BPF_X)
305 			verbose("(%02x) %sr%d %s %sr%d\n",
306 				insn->code, class == BPF_ALU ? "(u32) " : "",
307 				insn->dst_reg,
308 				bpf_alu_string[BPF_OP(insn->code) >> 4],
309 				class == BPF_ALU ? "(u32) " : "",
310 				insn->src_reg);
311 		else
312 			verbose("(%02x) %sr%d %s %s%d\n",
313 				insn->code, class == BPF_ALU ? "(u32) " : "",
314 				insn->dst_reg,
315 				bpf_alu_string[BPF_OP(insn->code) >> 4],
316 				class == BPF_ALU ? "(u32) " : "",
317 				insn->imm);
318 	} else if (class == BPF_STX) {
319 		if (BPF_MODE(insn->code) == BPF_MEM)
320 			verbose("(%02x) *(%s *)(r%d %+d) = r%d\n",
321 				insn->code,
322 				bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
323 				insn->dst_reg,
324 				insn->off, insn->src_reg);
325 		else if (BPF_MODE(insn->code) == BPF_XADD)
326 			verbose("(%02x) lock *(%s *)(r%d %+d) += r%d\n",
327 				insn->code,
328 				bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
329 				insn->dst_reg, insn->off,
330 				insn->src_reg);
331 		else
332 			verbose("BUG_%02x\n", insn->code);
333 	} else if (class == BPF_ST) {
334 		if (BPF_MODE(insn->code) != BPF_MEM) {
335 			verbose("BUG_st_%02x\n", insn->code);
336 			return;
337 		}
338 		verbose("(%02x) *(%s *)(r%d %+d) = %d\n",
339 			insn->code,
340 			bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
341 			insn->dst_reg,
342 			insn->off, insn->imm);
343 	} else if (class == BPF_LDX) {
344 		if (BPF_MODE(insn->code) != BPF_MEM) {
345 			verbose("BUG_ldx_%02x\n", insn->code);
346 			return;
347 		}
348 		verbose("(%02x) r%d = *(%s *)(r%d %+d)\n",
349 			insn->code, insn->dst_reg,
350 			bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
351 			insn->src_reg, insn->off);
352 	} else if (class == BPF_LD) {
353 		if (BPF_MODE(insn->code) == BPF_ABS) {
354 			verbose("(%02x) r0 = *(%s *)skb[%d]\n",
355 				insn->code,
356 				bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
357 				insn->imm);
358 		} else if (BPF_MODE(insn->code) == BPF_IND) {
359 			verbose("(%02x) r0 = *(%s *)skb[r%d + %d]\n",
360 				insn->code,
361 				bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
362 				insn->src_reg, insn->imm);
363 		} else if (BPF_MODE(insn->code) == BPF_IMM) {
364 			verbose("(%02x) r%d = 0x%x\n",
365 				insn->code, insn->dst_reg, insn->imm);
366 		} else {
367 			verbose("BUG_ld_%02x\n", insn->code);
368 			return;
369 		}
370 	} else if (class == BPF_JMP) {
371 		u8 opcode = BPF_OP(insn->code);
372 
373 		if (opcode == BPF_CALL) {
374 			verbose("(%02x) call %s#%d\n", insn->code,
375 				func_id_name(insn->imm), insn->imm);
376 		} else if (insn->code == (BPF_JMP | BPF_JA)) {
377 			verbose("(%02x) goto pc%+d\n",
378 				insn->code, insn->off);
379 		} else if (insn->code == (BPF_JMP | BPF_EXIT)) {
380 			verbose("(%02x) exit\n", insn->code);
381 		} else if (BPF_SRC(insn->code) == BPF_X) {
382 			verbose("(%02x) if r%d %s r%d goto pc%+d\n",
383 				insn->code, insn->dst_reg,
384 				bpf_jmp_string[BPF_OP(insn->code) >> 4],
385 				insn->src_reg, insn->off);
386 		} else {
387 			verbose("(%02x) if r%d %s 0x%x goto pc%+d\n",
388 				insn->code, insn->dst_reg,
389 				bpf_jmp_string[BPF_OP(insn->code) >> 4],
390 				insn->imm, insn->off);
391 		}
392 	} else {
393 		verbose("(%02x) %s\n", insn->code, bpf_class_string[class]);
394 	}
395 }
396 
397 static int pop_stack(struct bpf_verifier_env *env, int *prev_insn_idx)
398 {
399 	struct bpf_verifier_stack_elem *elem;
400 	int insn_idx;
401 
402 	if (env->head == NULL)
403 		return -1;
404 
405 	memcpy(&env->cur_state, &env->head->st, sizeof(env->cur_state));
406 	insn_idx = env->head->insn_idx;
407 	if (prev_insn_idx)
408 		*prev_insn_idx = env->head->prev_insn_idx;
409 	elem = env->head->next;
410 	kfree(env->head);
411 	env->head = elem;
412 	env->stack_size--;
413 	return insn_idx;
414 }
415 
416 static struct bpf_verifier_state *push_stack(struct bpf_verifier_env *env,
417 					     int insn_idx, int prev_insn_idx)
418 {
419 	struct bpf_verifier_stack_elem *elem;
420 
421 	elem = kmalloc(sizeof(struct bpf_verifier_stack_elem), GFP_KERNEL);
422 	if (!elem)
423 		goto err;
424 
425 	memcpy(&elem->st, &env->cur_state, sizeof(env->cur_state));
426 	elem->insn_idx = insn_idx;
427 	elem->prev_insn_idx = prev_insn_idx;
428 	elem->next = env->head;
429 	env->head = elem;
430 	env->stack_size++;
431 	if (env->stack_size > BPF_COMPLEXITY_LIMIT_STACK) {
432 		verbose("BPF program is too complex\n");
433 		goto err;
434 	}
435 	return &elem->st;
436 err:
437 	/* pop all elements and return */
438 	while (pop_stack(env, NULL) >= 0);
439 	return NULL;
440 }
441 
442 #define CALLER_SAVED_REGS 6
443 static const int caller_saved[CALLER_SAVED_REGS] = {
444 	BPF_REG_0, BPF_REG_1, BPF_REG_2, BPF_REG_3, BPF_REG_4, BPF_REG_5
445 };
446 
447 static void init_reg_state(struct bpf_reg_state *regs)
448 {
449 	int i;
450 
451 	for (i = 0; i < MAX_BPF_REG; i++) {
452 		regs[i].type = NOT_INIT;
453 		regs[i].imm = 0;
454 		regs[i].min_value = BPF_REGISTER_MIN_RANGE;
455 		regs[i].max_value = BPF_REGISTER_MAX_RANGE;
456 	}
457 
458 	/* frame pointer */
459 	regs[BPF_REG_FP].type = FRAME_PTR;
460 
461 	/* 1st arg to a function */
462 	regs[BPF_REG_1].type = PTR_TO_CTX;
463 }
464 
465 static void __mark_reg_unknown_value(struct bpf_reg_state *regs, u32 regno)
466 {
467 	regs[regno].type = UNKNOWN_VALUE;
468 	regs[regno].id = 0;
469 	regs[regno].imm = 0;
470 }
471 
472 static void mark_reg_unknown_value(struct bpf_reg_state *regs, u32 regno)
473 {
474 	BUG_ON(regno >= MAX_BPF_REG);
475 	__mark_reg_unknown_value(regs, regno);
476 }
477 
478 static void reset_reg_range_values(struct bpf_reg_state *regs, u32 regno)
479 {
480 	regs[regno].min_value = BPF_REGISTER_MIN_RANGE;
481 	regs[regno].max_value = BPF_REGISTER_MAX_RANGE;
482 }
483 
484 enum reg_arg_type {
485 	SRC_OP,		/* register is used as source operand */
486 	DST_OP,		/* register is used as destination operand */
487 	DST_OP_NO_MARK	/* same as above, check only, don't mark */
488 };
489 
490 static int check_reg_arg(struct bpf_reg_state *regs, u32 regno,
491 			 enum reg_arg_type t)
492 {
493 	if (regno >= MAX_BPF_REG) {
494 		verbose("R%d is invalid\n", regno);
495 		return -EINVAL;
496 	}
497 
498 	if (t == SRC_OP) {
499 		/* check whether register used as source operand can be read */
500 		if (regs[regno].type == NOT_INIT) {
501 			verbose("R%d !read_ok\n", regno);
502 			return -EACCES;
503 		}
504 	} else {
505 		/* check whether register used as dest operand can be written to */
506 		if (regno == BPF_REG_FP) {
507 			verbose("frame pointer is read only\n");
508 			return -EACCES;
509 		}
510 		if (t == DST_OP)
511 			mark_reg_unknown_value(regs, regno);
512 	}
513 	return 0;
514 }
515 
516 static int bpf_size_to_bytes(int bpf_size)
517 {
518 	if (bpf_size == BPF_W)
519 		return 4;
520 	else if (bpf_size == BPF_H)
521 		return 2;
522 	else if (bpf_size == BPF_B)
523 		return 1;
524 	else if (bpf_size == BPF_DW)
525 		return 8;
526 	else
527 		return -EINVAL;
528 }
529 
530 static bool is_spillable_regtype(enum bpf_reg_type type)
531 {
532 	switch (type) {
533 	case PTR_TO_MAP_VALUE:
534 	case PTR_TO_MAP_VALUE_OR_NULL:
535 	case PTR_TO_STACK:
536 	case PTR_TO_CTX:
537 	case PTR_TO_PACKET:
538 	case PTR_TO_PACKET_END:
539 	case FRAME_PTR:
540 	case CONST_PTR_TO_MAP:
541 		return true;
542 	default:
543 		return false;
544 	}
545 }
546 
547 /* check_stack_read/write functions track spill/fill of registers,
548  * stack boundary and alignment are checked in check_mem_access()
549  */
550 static int check_stack_write(struct bpf_verifier_state *state, int off,
551 			     int size, int value_regno)
552 {
553 	int i;
554 	/* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0,
555 	 * so it's aligned access and [off, off + size) are within stack limits
556 	 */
557 
558 	if (value_regno >= 0 &&
559 	    is_spillable_regtype(state->regs[value_regno].type)) {
560 
561 		/* register containing pointer is being spilled into stack */
562 		if (size != BPF_REG_SIZE) {
563 			verbose("invalid size of register spill\n");
564 			return -EACCES;
565 		}
566 
567 		/* save register state */
568 		state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] =
569 			state->regs[value_regno];
570 
571 		for (i = 0; i < BPF_REG_SIZE; i++)
572 			state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_SPILL;
573 	} else {
574 		/* regular write of data into stack */
575 		state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] =
576 			(struct bpf_reg_state) {};
577 
578 		for (i = 0; i < size; i++)
579 			state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_MISC;
580 	}
581 	return 0;
582 }
583 
584 static int check_stack_read(struct bpf_verifier_state *state, int off, int size,
585 			    int value_regno)
586 {
587 	u8 *slot_type;
588 	int i;
589 
590 	slot_type = &state->stack_slot_type[MAX_BPF_STACK + off];
591 
592 	if (slot_type[0] == STACK_SPILL) {
593 		if (size != BPF_REG_SIZE) {
594 			verbose("invalid size of register spill\n");
595 			return -EACCES;
596 		}
597 		for (i = 1; i < BPF_REG_SIZE; i++) {
598 			if (slot_type[i] != STACK_SPILL) {
599 				verbose("corrupted spill memory\n");
600 				return -EACCES;
601 			}
602 		}
603 
604 		if (value_regno >= 0)
605 			/* restore register state from stack */
606 			state->regs[value_regno] =
607 				state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE];
608 		return 0;
609 	} else {
610 		for (i = 0; i < size; i++) {
611 			if (slot_type[i] != STACK_MISC) {
612 				verbose("invalid read from stack off %d+%d size %d\n",
613 					off, i, size);
614 				return -EACCES;
615 			}
616 		}
617 		if (value_regno >= 0)
618 			/* have read misc data from the stack */
619 			mark_reg_unknown_value(state->regs, value_regno);
620 		return 0;
621 	}
622 }
623 
624 /* check read/write into map element returned by bpf_map_lookup_elem() */
625 static int check_map_access(struct bpf_verifier_env *env, u32 regno, int off,
626 			    int size)
627 {
628 	struct bpf_map *map = env->cur_state.regs[regno].map_ptr;
629 
630 	if (off < 0 || off + size > map->value_size) {
631 		verbose("invalid access to map value, value_size=%d off=%d size=%d\n",
632 			map->value_size, off, size);
633 		return -EACCES;
634 	}
635 	return 0;
636 }
637 
638 #define MAX_PACKET_OFF 0xffff
639 
640 static bool may_access_direct_pkt_data(struct bpf_verifier_env *env,
641 				       const struct bpf_call_arg_meta *meta,
642 				       enum bpf_access_type t)
643 {
644 	switch (env->prog->type) {
645 	case BPF_PROG_TYPE_LWT_IN:
646 	case BPF_PROG_TYPE_LWT_OUT:
647 		/* dst_input() and dst_output() can't write for now */
648 		if (t == BPF_WRITE)
649 			return false;
650 	case BPF_PROG_TYPE_SCHED_CLS:
651 	case BPF_PROG_TYPE_SCHED_ACT:
652 	case BPF_PROG_TYPE_XDP:
653 	case BPF_PROG_TYPE_LWT_XMIT:
654 		if (meta)
655 			return meta->pkt_access;
656 
657 		env->seen_direct_write = true;
658 		return true;
659 	default:
660 		return false;
661 	}
662 }
663 
664 static int check_packet_access(struct bpf_verifier_env *env, u32 regno, int off,
665 			       int size)
666 {
667 	struct bpf_reg_state *regs = env->cur_state.regs;
668 	struct bpf_reg_state *reg = &regs[regno];
669 
670 	off += reg->off;
671 	if (off < 0 || size <= 0 || off + size > reg->range) {
672 		verbose("invalid access to packet, off=%d size=%d, R%d(id=%d,off=%d,r=%d)\n",
673 			off, size, regno, reg->id, reg->off, reg->range);
674 		return -EACCES;
675 	}
676 	return 0;
677 }
678 
679 /* check access to 'struct bpf_context' fields */
680 static int check_ctx_access(struct bpf_verifier_env *env, int off, int size,
681 			    enum bpf_access_type t, enum bpf_reg_type *reg_type)
682 {
683 	/* for analyzer ctx accesses are already validated and converted */
684 	if (env->analyzer_ops)
685 		return 0;
686 
687 	if (env->prog->aux->ops->is_valid_access &&
688 	    env->prog->aux->ops->is_valid_access(off, size, t, reg_type)) {
689 		/* remember the offset of last byte accessed in ctx */
690 		if (env->prog->aux->max_ctx_offset < off + size)
691 			env->prog->aux->max_ctx_offset = off + size;
692 		return 0;
693 	}
694 
695 	verbose("invalid bpf_context access off=%d size=%d\n", off, size);
696 	return -EACCES;
697 }
698 
699 static bool is_pointer_value(struct bpf_verifier_env *env, int regno)
700 {
701 	if (env->allow_ptr_leaks)
702 		return false;
703 
704 	switch (env->cur_state.regs[regno].type) {
705 	case UNKNOWN_VALUE:
706 	case CONST_IMM:
707 		return false;
708 	default:
709 		return true;
710 	}
711 }
712 
713 static int check_ptr_alignment(struct bpf_verifier_env *env,
714 			       struct bpf_reg_state *reg, int off, int size)
715 {
716 	if (reg->type != PTR_TO_PACKET && reg->type != PTR_TO_MAP_VALUE_ADJ) {
717 		if (off % size != 0) {
718 			verbose("misaligned access off %d size %d\n",
719 				off, size);
720 			return -EACCES;
721 		} else {
722 			return 0;
723 		}
724 	}
725 
726 	if (IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS))
727 		/* misaligned access to packet is ok on x86,arm,arm64 */
728 		return 0;
729 
730 	if (reg->id && size != 1) {
731 		verbose("Unknown packet alignment. Only byte-sized access allowed\n");
732 		return -EACCES;
733 	}
734 
735 	/* skb->data is NET_IP_ALIGN-ed */
736 	if (reg->type == PTR_TO_PACKET &&
737 	    (NET_IP_ALIGN + reg->off + off) % size != 0) {
738 		verbose("misaligned packet access off %d+%d+%d size %d\n",
739 			NET_IP_ALIGN, reg->off, off, size);
740 		return -EACCES;
741 	}
742 	return 0;
743 }
744 
745 /* check whether memory at (regno + off) is accessible for t = (read | write)
746  * if t==write, value_regno is a register which value is stored into memory
747  * if t==read, value_regno is a register which will receive the value from memory
748  * if t==write && value_regno==-1, some unknown value is stored into memory
749  * if t==read && value_regno==-1, don't care what we read from memory
750  */
751 static int check_mem_access(struct bpf_verifier_env *env, u32 regno, int off,
752 			    int bpf_size, enum bpf_access_type t,
753 			    int value_regno)
754 {
755 	struct bpf_verifier_state *state = &env->cur_state;
756 	struct bpf_reg_state *reg = &state->regs[regno];
757 	int size, err = 0;
758 
759 	if (reg->type == PTR_TO_STACK)
760 		off += reg->imm;
761 
762 	size = bpf_size_to_bytes(bpf_size);
763 	if (size < 0)
764 		return size;
765 
766 	err = check_ptr_alignment(env, reg, off, size);
767 	if (err)
768 		return err;
769 
770 	if (reg->type == PTR_TO_MAP_VALUE ||
771 	    reg->type == PTR_TO_MAP_VALUE_ADJ) {
772 		if (t == BPF_WRITE && value_regno >= 0 &&
773 		    is_pointer_value(env, value_regno)) {
774 			verbose("R%d leaks addr into map\n", value_regno);
775 			return -EACCES;
776 		}
777 
778 		/* If we adjusted the register to this map value at all then we
779 		 * need to change off and size to min_value and max_value
780 		 * respectively to make sure our theoretical access will be
781 		 * safe.
782 		 */
783 		if (reg->type == PTR_TO_MAP_VALUE_ADJ) {
784 			if (log_level)
785 				print_verifier_state(state);
786 			env->varlen_map_value_access = true;
787 			/* The minimum value is only important with signed
788 			 * comparisons where we can't assume the floor of a
789 			 * value is 0.  If we are using signed variables for our
790 			 * index'es we need to make sure that whatever we use
791 			 * will have a set floor within our range.
792 			 */
793 			if (reg->min_value < 0) {
794 				verbose("R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n",
795 					regno);
796 				return -EACCES;
797 			}
798 			err = check_map_access(env, regno, reg->min_value + off,
799 					       size);
800 			if (err) {
801 				verbose("R%d min value is outside of the array range\n",
802 					regno);
803 				return err;
804 			}
805 
806 			/* If we haven't set a max value then we need to bail
807 			 * since we can't be sure we won't do bad things.
808 			 */
809 			if (reg->max_value == BPF_REGISTER_MAX_RANGE) {
810 				verbose("R%d unbounded memory access, make sure to bounds check any array access into a map\n",
811 					regno);
812 				return -EACCES;
813 			}
814 			off += reg->max_value;
815 		}
816 		err = check_map_access(env, regno, off, size);
817 		if (!err && t == BPF_READ && value_regno >= 0)
818 			mark_reg_unknown_value(state->regs, value_regno);
819 
820 	} else if (reg->type == PTR_TO_CTX) {
821 		enum bpf_reg_type reg_type = UNKNOWN_VALUE;
822 
823 		if (t == BPF_WRITE && value_regno >= 0 &&
824 		    is_pointer_value(env, value_regno)) {
825 			verbose("R%d leaks addr into ctx\n", value_regno);
826 			return -EACCES;
827 		}
828 		err = check_ctx_access(env, off, size, t, &reg_type);
829 		if (!err && t == BPF_READ && value_regno >= 0) {
830 			mark_reg_unknown_value(state->regs, value_regno);
831 			/* note that reg.[id|off|range] == 0 */
832 			state->regs[value_regno].type = reg_type;
833 		}
834 
835 	} else if (reg->type == FRAME_PTR || reg->type == PTR_TO_STACK) {
836 		if (off >= 0 || off < -MAX_BPF_STACK) {
837 			verbose("invalid stack off=%d size=%d\n", off, size);
838 			return -EACCES;
839 		}
840 		if (t == BPF_WRITE) {
841 			if (!env->allow_ptr_leaks &&
842 			    state->stack_slot_type[MAX_BPF_STACK + off] == STACK_SPILL &&
843 			    size != BPF_REG_SIZE) {
844 				verbose("attempt to corrupt spilled pointer on stack\n");
845 				return -EACCES;
846 			}
847 			err = check_stack_write(state, off, size, value_regno);
848 		} else {
849 			err = check_stack_read(state, off, size, value_regno);
850 		}
851 	} else if (state->regs[regno].type == PTR_TO_PACKET) {
852 		if (t == BPF_WRITE && !may_access_direct_pkt_data(env, NULL, t)) {
853 			verbose("cannot write into packet\n");
854 			return -EACCES;
855 		}
856 		if (t == BPF_WRITE && value_regno >= 0 &&
857 		    is_pointer_value(env, value_regno)) {
858 			verbose("R%d leaks addr into packet\n", value_regno);
859 			return -EACCES;
860 		}
861 		err = check_packet_access(env, regno, off, size);
862 		if (!err && t == BPF_READ && value_regno >= 0)
863 			mark_reg_unknown_value(state->regs, value_regno);
864 	} else {
865 		verbose("R%d invalid mem access '%s'\n",
866 			regno, reg_type_str[reg->type]);
867 		return -EACCES;
868 	}
869 
870 	if (!err && size <= 2 && value_regno >= 0 && env->allow_ptr_leaks &&
871 	    state->regs[value_regno].type == UNKNOWN_VALUE) {
872 		/* 1 or 2 byte load zero-extends, determine the number of
873 		 * zero upper bits. Not doing it fo 4 byte load, since
874 		 * such values cannot be added to ptr_to_packet anyway.
875 		 */
876 		state->regs[value_regno].imm = 64 - size * 8;
877 	}
878 	return err;
879 }
880 
881 static int check_xadd(struct bpf_verifier_env *env, struct bpf_insn *insn)
882 {
883 	struct bpf_reg_state *regs = env->cur_state.regs;
884 	int err;
885 
886 	if ((BPF_SIZE(insn->code) != BPF_W && BPF_SIZE(insn->code) != BPF_DW) ||
887 	    insn->imm != 0) {
888 		verbose("BPF_XADD uses reserved fields\n");
889 		return -EINVAL;
890 	}
891 
892 	/* check src1 operand */
893 	err = check_reg_arg(regs, insn->src_reg, SRC_OP);
894 	if (err)
895 		return err;
896 
897 	/* check src2 operand */
898 	err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
899 	if (err)
900 		return err;
901 
902 	/* check whether atomic_add can read the memory */
903 	err = check_mem_access(env, insn->dst_reg, insn->off,
904 			       BPF_SIZE(insn->code), BPF_READ, -1);
905 	if (err)
906 		return err;
907 
908 	/* check whether atomic_add can write into the same memory */
909 	return check_mem_access(env, insn->dst_reg, insn->off,
910 				BPF_SIZE(insn->code), BPF_WRITE, -1);
911 }
912 
913 /* when register 'regno' is passed into function that will read 'access_size'
914  * bytes from that pointer, make sure that it's within stack boundary
915  * and all elements of stack are initialized
916  */
917 static int check_stack_boundary(struct bpf_verifier_env *env, int regno,
918 				int access_size, bool zero_size_allowed,
919 				struct bpf_call_arg_meta *meta)
920 {
921 	struct bpf_verifier_state *state = &env->cur_state;
922 	struct bpf_reg_state *regs = state->regs;
923 	int off, i;
924 
925 	if (regs[regno].type != PTR_TO_STACK) {
926 		if (zero_size_allowed && access_size == 0 &&
927 		    regs[regno].type == CONST_IMM &&
928 		    regs[regno].imm  == 0)
929 			return 0;
930 
931 		verbose("R%d type=%s expected=%s\n", regno,
932 			reg_type_str[regs[regno].type],
933 			reg_type_str[PTR_TO_STACK]);
934 		return -EACCES;
935 	}
936 
937 	off = regs[regno].imm;
938 	if (off >= 0 || off < -MAX_BPF_STACK || off + access_size > 0 ||
939 	    access_size <= 0) {
940 		verbose("invalid stack type R%d off=%d access_size=%d\n",
941 			regno, off, access_size);
942 		return -EACCES;
943 	}
944 
945 	if (meta && meta->raw_mode) {
946 		meta->access_size = access_size;
947 		meta->regno = regno;
948 		return 0;
949 	}
950 
951 	for (i = 0; i < access_size; i++) {
952 		if (state->stack_slot_type[MAX_BPF_STACK + off + i] != STACK_MISC) {
953 			verbose("invalid indirect read from stack off %d+%d size %d\n",
954 				off, i, access_size);
955 			return -EACCES;
956 		}
957 	}
958 	return 0;
959 }
960 
961 static int check_func_arg(struct bpf_verifier_env *env, u32 regno,
962 			  enum bpf_arg_type arg_type,
963 			  struct bpf_call_arg_meta *meta)
964 {
965 	struct bpf_reg_state *regs = env->cur_state.regs, *reg = &regs[regno];
966 	enum bpf_reg_type expected_type, type = reg->type;
967 	int err = 0;
968 
969 	if (arg_type == ARG_DONTCARE)
970 		return 0;
971 
972 	if (type == NOT_INIT) {
973 		verbose("R%d !read_ok\n", regno);
974 		return -EACCES;
975 	}
976 
977 	if (arg_type == ARG_ANYTHING) {
978 		if (is_pointer_value(env, regno)) {
979 			verbose("R%d leaks addr into helper function\n", regno);
980 			return -EACCES;
981 		}
982 		return 0;
983 	}
984 
985 	if (type == PTR_TO_PACKET &&
986 	    !may_access_direct_pkt_data(env, meta, BPF_READ)) {
987 		verbose("helper access to the packet is not allowed\n");
988 		return -EACCES;
989 	}
990 
991 	if (arg_type == ARG_PTR_TO_MAP_KEY ||
992 	    arg_type == ARG_PTR_TO_MAP_VALUE) {
993 		expected_type = PTR_TO_STACK;
994 		if (type != PTR_TO_PACKET && type != expected_type)
995 			goto err_type;
996 	} else if (arg_type == ARG_CONST_STACK_SIZE ||
997 		   arg_type == ARG_CONST_STACK_SIZE_OR_ZERO) {
998 		expected_type = CONST_IMM;
999 		if (type != expected_type)
1000 			goto err_type;
1001 	} else if (arg_type == ARG_CONST_MAP_PTR) {
1002 		expected_type = CONST_PTR_TO_MAP;
1003 		if (type != expected_type)
1004 			goto err_type;
1005 	} else if (arg_type == ARG_PTR_TO_CTX) {
1006 		expected_type = PTR_TO_CTX;
1007 		if (type != expected_type)
1008 			goto err_type;
1009 	} else if (arg_type == ARG_PTR_TO_STACK ||
1010 		   arg_type == ARG_PTR_TO_RAW_STACK) {
1011 		expected_type = PTR_TO_STACK;
1012 		/* One exception here. In case function allows for NULL to be
1013 		 * passed in as argument, it's a CONST_IMM type. Final test
1014 		 * happens during stack boundary checking.
1015 		 */
1016 		if (type == CONST_IMM && reg->imm == 0)
1017 			/* final test in check_stack_boundary() */;
1018 		else if (type != PTR_TO_PACKET && type != expected_type)
1019 			goto err_type;
1020 		meta->raw_mode = arg_type == ARG_PTR_TO_RAW_STACK;
1021 	} else {
1022 		verbose("unsupported arg_type %d\n", arg_type);
1023 		return -EFAULT;
1024 	}
1025 
1026 	if (arg_type == ARG_CONST_MAP_PTR) {
1027 		/* bpf_map_xxx(map_ptr) call: remember that map_ptr */
1028 		meta->map_ptr = reg->map_ptr;
1029 	} else if (arg_type == ARG_PTR_TO_MAP_KEY) {
1030 		/* bpf_map_xxx(..., map_ptr, ..., key) call:
1031 		 * check that [key, key + map->key_size) are within
1032 		 * stack limits and initialized
1033 		 */
1034 		if (!meta->map_ptr) {
1035 			/* in function declaration map_ptr must come before
1036 			 * map_key, so that it's verified and known before
1037 			 * we have to check map_key here. Otherwise it means
1038 			 * that kernel subsystem misconfigured verifier
1039 			 */
1040 			verbose("invalid map_ptr to access map->key\n");
1041 			return -EACCES;
1042 		}
1043 		if (type == PTR_TO_PACKET)
1044 			err = check_packet_access(env, regno, 0,
1045 						  meta->map_ptr->key_size);
1046 		else
1047 			err = check_stack_boundary(env, regno,
1048 						   meta->map_ptr->key_size,
1049 						   false, NULL);
1050 	} else if (arg_type == ARG_PTR_TO_MAP_VALUE) {
1051 		/* bpf_map_xxx(..., map_ptr, ..., value) call:
1052 		 * check [value, value + map->value_size) validity
1053 		 */
1054 		if (!meta->map_ptr) {
1055 			/* kernel subsystem misconfigured verifier */
1056 			verbose("invalid map_ptr to access map->value\n");
1057 			return -EACCES;
1058 		}
1059 		if (type == PTR_TO_PACKET)
1060 			err = check_packet_access(env, regno, 0,
1061 						  meta->map_ptr->value_size);
1062 		else
1063 			err = check_stack_boundary(env, regno,
1064 						   meta->map_ptr->value_size,
1065 						   false, NULL);
1066 	} else if (arg_type == ARG_CONST_STACK_SIZE ||
1067 		   arg_type == ARG_CONST_STACK_SIZE_OR_ZERO) {
1068 		bool zero_size_allowed = (arg_type == ARG_CONST_STACK_SIZE_OR_ZERO);
1069 
1070 		/* bpf_xxx(..., buf, len) call will access 'len' bytes
1071 		 * from stack pointer 'buf'. Check it
1072 		 * note: regno == len, regno - 1 == buf
1073 		 */
1074 		if (regno == 0) {
1075 			/* kernel subsystem misconfigured verifier */
1076 			verbose("ARG_CONST_STACK_SIZE cannot be first argument\n");
1077 			return -EACCES;
1078 		}
1079 		if (regs[regno - 1].type == PTR_TO_PACKET)
1080 			err = check_packet_access(env, regno - 1, 0, reg->imm);
1081 		else
1082 			err = check_stack_boundary(env, regno - 1, reg->imm,
1083 						   zero_size_allowed, meta);
1084 	}
1085 
1086 	return err;
1087 err_type:
1088 	verbose("R%d type=%s expected=%s\n", regno,
1089 		reg_type_str[type], reg_type_str[expected_type]);
1090 	return -EACCES;
1091 }
1092 
1093 static int check_map_func_compatibility(struct bpf_map *map, int func_id)
1094 {
1095 	if (!map)
1096 		return 0;
1097 
1098 	/* We need a two way check, first is from map perspective ... */
1099 	switch (map->map_type) {
1100 	case BPF_MAP_TYPE_PROG_ARRAY:
1101 		if (func_id != BPF_FUNC_tail_call)
1102 			goto error;
1103 		break;
1104 	case BPF_MAP_TYPE_PERF_EVENT_ARRAY:
1105 		if (func_id != BPF_FUNC_perf_event_read &&
1106 		    func_id != BPF_FUNC_perf_event_output)
1107 			goto error;
1108 		break;
1109 	case BPF_MAP_TYPE_STACK_TRACE:
1110 		if (func_id != BPF_FUNC_get_stackid)
1111 			goto error;
1112 		break;
1113 	case BPF_MAP_TYPE_CGROUP_ARRAY:
1114 		if (func_id != BPF_FUNC_skb_under_cgroup &&
1115 		    func_id != BPF_FUNC_current_task_under_cgroup)
1116 			goto error;
1117 		break;
1118 	default:
1119 		break;
1120 	}
1121 
1122 	/* ... and second from the function itself. */
1123 	switch (func_id) {
1124 	case BPF_FUNC_tail_call:
1125 		if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY)
1126 			goto error;
1127 		break;
1128 	case BPF_FUNC_perf_event_read:
1129 	case BPF_FUNC_perf_event_output:
1130 		if (map->map_type != BPF_MAP_TYPE_PERF_EVENT_ARRAY)
1131 			goto error;
1132 		break;
1133 	case BPF_FUNC_get_stackid:
1134 		if (map->map_type != BPF_MAP_TYPE_STACK_TRACE)
1135 			goto error;
1136 		break;
1137 	case BPF_FUNC_current_task_under_cgroup:
1138 	case BPF_FUNC_skb_under_cgroup:
1139 		if (map->map_type != BPF_MAP_TYPE_CGROUP_ARRAY)
1140 			goto error;
1141 		break;
1142 	default:
1143 		break;
1144 	}
1145 
1146 	return 0;
1147 error:
1148 	verbose("cannot pass map_type %d into func %s#%d\n",
1149 		map->map_type, func_id_name(func_id), func_id);
1150 	return -EINVAL;
1151 }
1152 
1153 static int check_raw_mode(const struct bpf_func_proto *fn)
1154 {
1155 	int count = 0;
1156 
1157 	if (fn->arg1_type == ARG_PTR_TO_RAW_STACK)
1158 		count++;
1159 	if (fn->arg2_type == ARG_PTR_TO_RAW_STACK)
1160 		count++;
1161 	if (fn->arg3_type == ARG_PTR_TO_RAW_STACK)
1162 		count++;
1163 	if (fn->arg4_type == ARG_PTR_TO_RAW_STACK)
1164 		count++;
1165 	if (fn->arg5_type == ARG_PTR_TO_RAW_STACK)
1166 		count++;
1167 
1168 	return count > 1 ? -EINVAL : 0;
1169 }
1170 
1171 static void clear_all_pkt_pointers(struct bpf_verifier_env *env)
1172 {
1173 	struct bpf_verifier_state *state = &env->cur_state;
1174 	struct bpf_reg_state *regs = state->regs, *reg;
1175 	int i;
1176 
1177 	for (i = 0; i < MAX_BPF_REG; i++)
1178 		if (regs[i].type == PTR_TO_PACKET ||
1179 		    regs[i].type == PTR_TO_PACKET_END)
1180 			mark_reg_unknown_value(regs, i);
1181 
1182 	for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
1183 		if (state->stack_slot_type[i] != STACK_SPILL)
1184 			continue;
1185 		reg = &state->spilled_regs[i / BPF_REG_SIZE];
1186 		if (reg->type != PTR_TO_PACKET &&
1187 		    reg->type != PTR_TO_PACKET_END)
1188 			continue;
1189 		reg->type = UNKNOWN_VALUE;
1190 		reg->imm = 0;
1191 	}
1192 }
1193 
1194 static int check_call(struct bpf_verifier_env *env, int func_id)
1195 {
1196 	struct bpf_verifier_state *state = &env->cur_state;
1197 	const struct bpf_func_proto *fn = NULL;
1198 	struct bpf_reg_state *regs = state->regs;
1199 	struct bpf_reg_state *reg;
1200 	struct bpf_call_arg_meta meta;
1201 	bool changes_data;
1202 	int i, err;
1203 
1204 	/* find function prototype */
1205 	if (func_id < 0 || func_id >= __BPF_FUNC_MAX_ID) {
1206 		verbose("invalid func %s#%d\n", func_id_name(func_id), func_id);
1207 		return -EINVAL;
1208 	}
1209 
1210 	if (env->prog->aux->ops->get_func_proto)
1211 		fn = env->prog->aux->ops->get_func_proto(func_id);
1212 
1213 	if (!fn) {
1214 		verbose("unknown func %s#%d\n", func_id_name(func_id), func_id);
1215 		return -EINVAL;
1216 	}
1217 
1218 	/* eBPF programs must be GPL compatible to use GPL-ed functions */
1219 	if (!env->prog->gpl_compatible && fn->gpl_only) {
1220 		verbose("cannot call GPL only function from proprietary program\n");
1221 		return -EINVAL;
1222 	}
1223 
1224 	changes_data = bpf_helper_changes_pkt_data(fn->func);
1225 
1226 	memset(&meta, 0, sizeof(meta));
1227 	meta.pkt_access = fn->pkt_access;
1228 
1229 	/* We only support one arg being in raw mode at the moment, which
1230 	 * is sufficient for the helper functions we have right now.
1231 	 */
1232 	err = check_raw_mode(fn);
1233 	if (err) {
1234 		verbose("kernel subsystem misconfigured func %s#%d\n",
1235 			func_id_name(func_id), func_id);
1236 		return err;
1237 	}
1238 
1239 	/* check args */
1240 	err = check_func_arg(env, BPF_REG_1, fn->arg1_type, &meta);
1241 	if (err)
1242 		return err;
1243 	err = check_func_arg(env, BPF_REG_2, fn->arg2_type, &meta);
1244 	if (err)
1245 		return err;
1246 	err = check_func_arg(env, BPF_REG_3, fn->arg3_type, &meta);
1247 	if (err)
1248 		return err;
1249 	err = check_func_arg(env, BPF_REG_4, fn->arg4_type, &meta);
1250 	if (err)
1251 		return err;
1252 	err = check_func_arg(env, BPF_REG_5, fn->arg5_type, &meta);
1253 	if (err)
1254 		return err;
1255 
1256 	/* Mark slots with STACK_MISC in case of raw mode, stack offset
1257 	 * is inferred from register state.
1258 	 */
1259 	for (i = 0; i < meta.access_size; i++) {
1260 		err = check_mem_access(env, meta.regno, i, BPF_B, BPF_WRITE, -1);
1261 		if (err)
1262 			return err;
1263 	}
1264 
1265 	/* reset caller saved regs */
1266 	for (i = 0; i < CALLER_SAVED_REGS; i++) {
1267 		reg = regs + caller_saved[i];
1268 		reg->type = NOT_INIT;
1269 		reg->imm = 0;
1270 	}
1271 
1272 	/* update return register */
1273 	if (fn->ret_type == RET_INTEGER) {
1274 		regs[BPF_REG_0].type = UNKNOWN_VALUE;
1275 	} else if (fn->ret_type == RET_VOID) {
1276 		regs[BPF_REG_0].type = NOT_INIT;
1277 	} else if (fn->ret_type == RET_PTR_TO_MAP_VALUE_OR_NULL) {
1278 		regs[BPF_REG_0].type = PTR_TO_MAP_VALUE_OR_NULL;
1279 		regs[BPF_REG_0].max_value = regs[BPF_REG_0].min_value = 0;
1280 		/* remember map_ptr, so that check_map_access()
1281 		 * can check 'value_size' boundary of memory access
1282 		 * to map element returned from bpf_map_lookup_elem()
1283 		 */
1284 		if (meta.map_ptr == NULL) {
1285 			verbose("kernel subsystem misconfigured verifier\n");
1286 			return -EINVAL;
1287 		}
1288 		regs[BPF_REG_0].map_ptr = meta.map_ptr;
1289 		regs[BPF_REG_0].id = ++env->id_gen;
1290 	} else {
1291 		verbose("unknown return type %d of func %s#%d\n",
1292 			fn->ret_type, func_id_name(func_id), func_id);
1293 		return -EINVAL;
1294 	}
1295 
1296 	err = check_map_func_compatibility(meta.map_ptr, func_id);
1297 	if (err)
1298 		return err;
1299 
1300 	if (changes_data)
1301 		clear_all_pkt_pointers(env);
1302 	return 0;
1303 }
1304 
1305 static int check_packet_ptr_add(struct bpf_verifier_env *env,
1306 				struct bpf_insn *insn)
1307 {
1308 	struct bpf_reg_state *regs = env->cur_state.regs;
1309 	struct bpf_reg_state *dst_reg = &regs[insn->dst_reg];
1310 	struct bpf_reg_state *src_reg = &regs[insn->src_reg];
1311 	struct bpf_reg_state tmp_reg;
1312 	s32 imm;
1313 
1314 	if (BPF_SRC(insn->code) == BPF_K) {
1315 		/* pkt_ptr += imm */
1316 		imm = insn->imm;
1317 
1318 add_imm:
1319 		if (imm <= 0) {
1320 			verbose("addition of negative constant to packet pointer is not allowed\n");
1321 			return -EACCES;
1322 		}
1323 		if (imm >= MAX_PACKET_OFF ||
1324 		    imm + dst_reg->off >= MAX_PACKET_OFF) {
1325 			verbose("constant %d is too large to add to packet pointer\n",
1326 				imm);
1327 			return -EACCES;
1328 		}
1329 		/* a constant was added to pkt_ptr.
1330 		 * Remember it while keeping the same 'id'
1331 		 */
1332 		dst_reg->off += imm;
1333 	} else {
1334 		if (src_reg->type == PTR_TO_PACKET) {
1335 			/* R6=pkt(id=0,off=0,r=62) R7=imm22; r7 += r6 */
1336 			tmp_reg = *dst_reg;  /* save r7 state */
1337 			*dst_reg = *src_reg; /* copy pkt_ptr state r6 into r7 */
1338 			src_reg = &tmp_reg;  /* pretend it's src_reg state */
1339 			/* if the checks below reject it, the copy won't matter,
1340 			 * since we're rejecting the whole program. If all ok,
1341 			 * then imm22 state will be added to r7
1342 			 * and r7 will be pkt(id=0,off=22,r=62) while
1343 			 * r6 will stay as pkt(id=0,off=0,r=62)
1344 			 */
1345 		}
1346 
1347 		if (src_reg->type == CONST_IMM) {
1348 			/* pkt_ptr += reg where reg is known constant */
1349 			imm = src_reg->imm;
1350 			goto add_imm;
1351 		}
1352 		/* disallow pkt_ptr += reg
1353 		 * if reg is not uknown_value with guaranteed zero upper bits
1354 		 * otherwise pkt_ptr may overflow and addition will become
1355 		 * subtraction which is not allowed
1356 		 */
1357 		if (src_reg->type != UNKNOWN_VALUE) {
1358 			verbose("cannot add '%s' to ptr_to_packet\n",
1359 				reg_type_str[src_reg->type]);
1360 			return -EACCES;
1361 		}
1362 		if (src_reg->imm < 48) {
1363 			verbose("cannot add integer value with %lld upper zero bits to ptr_to_packet\n",
1364 				src_reg->imm);
1365 			return -EACCES;
1366 		}
1367 		/* dst_reg stays as pkt_ptr type and since some positive
1368 		 * integer value was added to the pointer, increment its 'id'
1369 		 */
1370 		dst_reg->id = ++env->id_gen;
1371 
1372 		/* something was added to pkt_ptr, set range and off to zero */
1373 		dst_reg->off = 0;
1374 		dst_reg->range = 0;
1375 	}
1376 	return 0;
1377 }
1378 
1379 static int evaluate_reg_alu(struct bpf_verifier_env *env, struct bpf_insn *insn)
1380 {
1381 	struct bpf_reg_state *regs = env->cur_state.regs;
1382 	struct bpf_reg_state *dst_reg = &regs[insn->dst_reg];
1383 	u8 opcode = BPF_OP(insn->code);
1384 	s64 imm_log2;
1385 
1386 	/* for type == UNKNOWN_VALUE:
1387 	 * imm > 0 -> number of zero upper bits
1388 	 * imm == 0 -> don't track which is the same as all bits can be non-zero
1389 	 */
1390 
1391 	if (BPF_SRC(insn->code) == BPF_X) {
1392 		struct bpf_reg_state *src_reg = &regs[insn->src_reg];
1393 
1394 		if (src_reg->type == UNKNOWN_VALUE && src_reg->imm > 0 &&
1395 		    dst_reg->imm && opcode == BPF_ADD) {
1396 			/* dreg += sreg
1397 			 * where both have zero upper bits. Adding them
1398 			 * can only result making one more bit non-zero
1399 			 * in the larger value.
1400 			 * Ex. 0xffff (imm=48) + 1 (imm=63) = 0x10000 (imm=47)
1401 			 *     0xffff (imm=48) + 0xffff = 0x1fffe (imm=47)
1402 			 */
1403 			dst_reg->imm = min(dst_reg->imm, src_reg->imm);
1404 			dst_reg->imm--;
1405 			return 0;
1406 		}
1407 		if (src_reg->type == CONST_IMM && src_reg->imm > 0 &&
1408 		    dst_reg->imm && opcode == BPF_ADD) {
1409 			/* dreg += sreg
1410 			 * where dreg has zero upper bits and sreg is const.
1411 			 * Adding them can only result making one more bit
1412 			 * non-zero in the larger value.
1413 			 */
1414 			imm_log2 = __ilog2_u64((long long)src_reg->imm);
1415 			dst_reg->imm = min(dst_reg->imm, 63 - imm_log2);
1416 			dst_reg->imm--;
1417 			return 0;
1418 		}
1419 		/* all other cases non supported yet, just mark dst_reg */
1420 		dst_reg->imm = 0;
1421 		return 0;
1422 	}
1423 
1424 	/* sign extend 32-bit imm into 64-bit to make sure that
1425 	 * negative values occupy bit 63. Note ilog2() would have
1426 	 * been incorrect, since sizeof(insn->imm) == 4
1427 	 */
1428 	imm_log2 = __ilog2_u64((long long)insn->imm);
1429 
1430 	if (dst_reg->imm && opcode == BPF_LSH) {
1431 		/* reg <<= imm
1432 		 * if reg was a result of 2 byte load, then its imm == 48
1433 		 * which means that upper 48 bits are zero and shifting this reg
1434 		 * left by 4 would mean that upper 44 bits are still zero
1435 		 */
1436 		dst_reg->imm -= insn->imm;
1437 	} else if (dst_reg->imm && opcode == BPF_MUL) {
1438 		/* reg *= imm
1439 		 * if multiplying by 14 subtract 4
1440 		 * This is conservative calculation of upper zero bits.
1441 		 * It's not trying to special case insn->imm == 1 or 0 cases
1442 		 */
1443 		dst_reg->imm -= imm_log2 + 1;
1444 	} else if (opcode == BPF_AND) {
1445 		/* reg &= imm */
1446 		dst_reg->imm = 63 - imm_log2;
1447 	} else if (dst_reg->imm && opcode == BPF_ADD) {
1448 		/* reg += imm */
1449 		dst_reg->imm = min(dst_reg->imm, 63 - imm_log2);
1450 		dst_reg->imm--;
1451 	} else if (opcode == BPF_RSH) {
1452 		/* reg >>= imm
1453 		 * which means that after right shift, upper bits will be zero
1454 		 * note that verifier already checked that
1455 		 * 0 <= imm < 64 for shift insn
1456 		 */
1457 		dst_reg->imm += insn->imm;
1458 		if (unlikely(dst_reg->imm > 64))
1459 			/* some dumb code did:
1460 			 * r2 = *(u32 *)mem;
1461 			 * r2 >>= 32;
1462 			 * and all bits are zero now */
1463 			dst_reg->imm = 64;
1464 	} else {
1465 		/* all other alu ops, means that we don't know what will
1466 		 * happen to the value, mark it with unknown number of zero bits
1467 		 */
1468 		dst_reg->imm = 0;
1469 	}
1470 
1471 	if (dst_reg->imm < 0) {
1472 		/* all 64 bits of the register can contain non-zero bits
1473 		 * and such value cannot be added to ptr_to_packet, since it
1474 		 * may overflow, mark it as unknown to avoid further eval
1475 		 */
1476 		dst_reg->imm = 0;
1477 	}
1478 	return 0;
1479 }
1480 
1481 static int evaluate_reg_imm_alu(struct bpf_verifier_env *env,
1482 				struct bpf_insn *insn)
1483 {
1484 	struct bpf_reg_state *regs = env->cur_state.regs;
1485 	struct bpf_reg_state *dst_reg = &regs[insn->dst_reg];
1486 	struct bpf_reg_state *src_reg = &regs[insn->src_reg];
1487 	u8 opcode = BPF_OP(insn->code);
1488 
1489 	/* dst_reg->type == CONST_IMM here, simulate execution of 'add'/'or'
1490 	 * insn. Don't care about overflow or negative values, just add them
1491 	 */
1492 	if (opcode == BPF_ADD && BPF_SRC(insn->code) == BPF_K)
1493 		dst_reg->imm += insn->imm;
1494 	else if (opcode == BPF_ADD && BPF_SRC(insn->code) == BPF_X &&
1495 		 src_reg->type == CONST_IMM)
1496 		dst_reg->imm += src_reg->imm;
1497 	else if (opcode == BPF_OR && BPF_SRC(insn->code) == BPF_K)
1498 		dst_reg->imm |= insn->imm;
1499 	else if (opcode == BPF_OR && BPF_SRC(insn->code) == BPF_X &&
1500 		 src_reg->type == CONST_IMM)
1501 		dst_reg->imm |= src_reg->imm;
1502 	else
1503 		mark_reg_unknown_value(regs, insn->dst_reg);
1504 	return 0;
1505 }
1506 
1507 static void check_reg_overflow(struct bpf_reg_state *reg)
1508 {
1509 	if (reg->max_value > BPF_REGISTER_MAX_RANGE)
1510 		reg->max_value = BPF_REGISTER_MAX_RANGE;
1511 	if (reg->min_value < BPF_REGISTER_MIN_RANGE ||
1512 	    reg->min_value > BPF_REGISTER_MAX_RANGE)
1513 		reg->min_value = BPF_REGISTER_MIN_RANGE;
1514 }
1515 
1516 static void adjust_reg_min_max_vals(struct bpf_verifier_env *env,
1517 				    struct bpf_insn *insn)
1518 {
1519 	struct bpf_reg_state *regs = env->cur_state.regs, *dst_reg;
1520 	s64 min_val = BPF_REGISTER_MIN_RANGE;
1521 	u64 max_val = BPF_REGISTER_MAX_RANGE;
1522 	u8 opcode = BPF_OP(insn->code);
1523 
1524 	dst_reg = &regs[insn->dst_reg];
1525 	if (BPF_SRC(insn->code) == BPF_X) {
1526 		check_reg_overflow(&regs[insn->src_reg]);
1527 		min_val = regs[insn->src_reg].min_value;
1528 		max_val = regs[insn->src_reg].max_value;
1529 
1530 		/* If the source register is a random pointer then the
1531 		 * min_value/max_value values represent the range of the known
1532 		 * accesses into that value, not the actual min/max value of the
1533 		 * register itself.  In this case we have to reset the reg range
1534 		 * values so we know it is not safe to look at.
1535 		 */
1536 		if (regs[insn->src_reg].type != CONST_IMM &&
1537 		    regs[insn->src_reg].type != UNKNOWN_VALUE) {
1538 			min_val = BPF_REGISTER_MIN_RANGE;
1539 			max_val = BPF_REGISTER_MAX_RANGE;
1540 		}
1541 	} else if (insn->imm < BPF_REGISTER_MAX_RANGE &&
1542 		   (s64)insn->imm > BPF_REGISTER_MIN_RANGE) {
1543 		min_val = max_val = insn->imm;
1544 	}
1545 
1546 	/* We don't know anything about what was done to this register, mark it
1547 	 * as unknown.
1548 	 */
1549 	if (min_val == BPF_REGISTER_MIN_RANGE &&
1550 	    max_val == BPF_REGISTER_MAX_RANGE) {
1551 		reset_reg_range_values(regs, insn->dst_reg);
1552 		return;
1553 	}
1554 
1555 	/* If one of our values was at the end of our ranges then we can't just
1556 	 * do our normal operations to the register, we need to set the values
1557 	 * to the min/max since they are undefined.
1558 	 */
1559 	if (min_val == BPF_REGISTER_MIN_RANGE)
1560 		dst_reg->min_value = BPF_REGISTER_MIN_RANGE;
1561 	if (max_val == BPF_REGISTER_MAX_RANGE)
1562 		dst_reg->max_value = BPF_REGISTER_MAX_RANGE;
1563 
1564 	switch (opcode) {
1565 	case BPF_ADD:
1566 		if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE)
1567 			dst_reg->min_value += min_val;
1568 		if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1569 			dst_reg->max_value += max_val;
1570 		break;
1571 	case BPF_SUB:
1572 		if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE)
1573 			dst_reg->min_value -= min_val;
1574 		if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1575 			dst_reg->max_value -= max_val;
1576 		break;
1577 	case BPF_MUL:
1578 		if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE)
1579 			dst_reg->min_value *= min_val;
1580 		if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1581 			dst_reg->max_value *= max_val;
1582 		break;
1583 	case BPF_AND:
1584 		/* Disallow AND'ing of negative numbers, ain't nobody got time
1585 		 * for that.  Otherwise the minimum is 0 and the max is the max
1586 		 * value we could AND against.
1587 		 */
1588 		if (min_val < 0)
1589 			dst_reg->min_value = BPF_REGISTER_MIN_RANGE;
1590 		else
1591 			dst_reg->min_value = 0;
1592 		dst_reg->max_value = max_val;
1593 		break;
1594 	case BPF_LSH:
1595 		/* Gotta have special overflow logic here, if we're shifting
1596 		 * more than MAX_RANGE then just assume we have an invalid
1597 		 * range.
1598 		 */
1599 		if (min_val > ilog2(BPF_REGISTER_MAX_RANGE))
1600 			dst_reg->min_value = BPF_REGISTER_MIN_RANGE;
1601 		else if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE)
1602 			dst_reg->min_value <<= min_val;
1603 
1604 		if (max_val > ilog2(BPF_REGISTER_MAX_RANGE))
1605 			dst_reg->max_value = BPF_REGISTER_MAX_RANGE;
1606 		else if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1607 			dst_reg->max_value <<= max_val;
1608 		break;
1609 	case BPF_RSH:
1610 		/* RSH by a negative number is undefined, and the BPF_RSH is an
1611 		 * unsigned shift, so make the appropriate casts.
1612 		 */
1613 		if (min_val < 0 || dst_reg->min_value < 0)
1614 			dst_reg->min_value = BPF_REGISTER_MIN_RANGE;
1615 		else
1616 			dst_reg->min_value =
1617 				(u64)(dst_reg->min_value) >> min_val;
1618 		if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1619 			dst_reg->max_value >>= max_val;
1620 		break;
1621 	default:
1622 		reset_reg_range_values(regs, insn->dst_reg);
1623 		break;
1624 	}
1625 
1626 	check_reg_overflow(dst_reg);
1627 }
1628 
1629 /* check validity of 32-bit and 64-bit arithmetic operations */
1630 static int check_alu_op(struct bpf_verifier_env *env, struct bpf_insn *insn)
1631 {
1632 	struct bpf_reg_state *regs = env->cur_state.regs, *dst_reg;
1633 	u8 opcode = BPF_OP(insn->code);
1634 	int err;
1635 
1636 	if (opcode == BPF_END || opcode == BPF_NEG) {
1637 		if (opcode == BPF_NEG) {
1638 			if (BPF_SRC(insn->code) != 0 ||
1639 			    insn->src_reg != BPF_REG_0 ||
1640 			    insn->off != 0 || insn->imm != 0) {
1641 				verbose("BPF_NEG uses reserved fields\n");
1642 				return -EINVAL;
1643 			}
1644 		} else {
1645 			if (insn->src_reg != BPF_REG_0 || insn->off != 0 ||
1646 			    (insn->imm != 16 && insn->imm != 32 && insn->imm != 64)) {
1647 				verbose("BPF_END uses reserved fields\n");
1648 				return -EINVAL;
1649 			}
1650 		}
1651 
1652 		/* check src operand */
1653 		err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1654 		if (err)
1655 			return err;
1656 
1657 		if (is_pointer_value(env, insn->dst_reg)) {
1658 			verbose("R%d pointer arithmetic prohibited\n",
1659 				insn->dst_reg);
1660 			return -EACCES;
1661 		}
1662 
1663 		/* check dest operand */
1664 		err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1665 		if (err)
1666 			return err;
1667 
1668 	} else if (opcode == BPF_MOV) {
1669 
1670 		if (BPF_SRC(insn->code) == BPF_X) {
1671 			if (insn->imm != 0 || insn->off != 0) {
1672 				verbose("BPF_MOV uses reserved fields\n");
1673 				return -EINVAL;
1674 			}
1675 
1676 			/* check src operand */
1677 			err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1678 			if (err)
1679 				return err;
1680 		} else {
1681 			if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
1682 				verbose("BPF_MOV uses reserved fields\n");
1683 				return -EINVAL;
1684 			}
1685 		}
1686 
1687 		/* check dest operand */
1688 		err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1689 		if (err)
1690 			return err;
1691 
1692 		/* we are setting our register to something new, we need to
1693 		 * reset its range values.
1694 		 */
1695 		reset_reg_range_values(regs, insn->dst_reg);
1696 
1697 		if (BPF_SRC(insn->code) == BPF_X) {
1698 			if (BPF_CLASS(insn->code) == BPF_ALU64) {
1699 				/* case: R1 = R2
1700 				 * copy register state to dest reg
1701 				 */
1702 				regs[insn->dst_reg] = regs[insn->src_reg];
1703 			} else {
1704 				if (is_pointer_value(env, insn->src_reg)) {
1705 					verbose("R%d partial copy of pointer\n",
1706 						insn->src_reg);
1707 					return -EACCES;
1708 				}
1709 				mark_reg_unknown_value(regs, insn->dst_reg);
1710 			}
1711 		} else {
1712 			/* case: R = imm
1713 			 * remember the value we stored into this reg
1714 			 */
1715 			regs[insn->dst_reg].type = CONST_IMM;
1716 			regs[insn->dst_reg].imm = insn->imm;
1717 			regs[insn->dst_reg].max_value = insn->imm;
1718 			regs[insn->dst_reg].min_value = insn->imm;
1719 		}
1720 
1721 	} else if (opcode > BPF_END) {
1722 		verbose("invalid BPF_ALU opcode %x\n", opcode);
1723 		return -EINVAL;
1724 
1725 	} else {	/* all other ALU ops: and, sub, xor, add, ... */
1726 
1727 		if (BPF_SRC(insn->code) == BPF_X) {
1728 			if (insn->imm != 0 || insn->off != 0) {
1729 				verbose("BPF_ALU uses reserved fields\n");
1730 				return -EINVAL;
1731 			}
1732 			/* check src1 operand */
1733 			err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1734 			if (err)
1735 				return err;
1736 		} else {
1737 			if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
1738 				verbose("BPF_ALU uses reserved fields\n");
1739 				return -EINVAL;
1740 			}
1741 		}
1742 
1743 		/* check src2 operand */
1744 		err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1745 		if (err)
1746 			return err;
1747 
1748 		if ((opcode == BPF_MOD || opcode == BPF_DIV) &&
1749 		    BPF_SRC(insn->code) == BPF_K && insn->imm == 0) {
1750 			verbose("div by zero\n");
1751 			return -EINVAL;
1752 		}
1753 
1754 		if ((opcode == BPF_LSH || opcode == BPF_RSH ||
1755 		     opcode == BPF_ARSH) && BPF_SRC(insn->code) == BPF_K) {
1756 			int size = BPF_CLASS(insn->code) == BPF_ALU64 ? 64 : 32;
1757 
1758 			if (insn->imm < 0 || insn->imm >= size) {
1759 				verbose("invalid shift %d\n", insn->imm);
1760 				return -EINVAL;
1761 			}
1762 		}
1763 
1764 		/* check dest operand */
1765 		err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK);
1766 		if (err)
1767 			return err;
1768 
1769 		dst_reg = &regs[insn->dst_reg];
1770 
1771 		/* first we want to adjust our ranges. */
1772 		adjust_reg_min_max_vals(env, insn);
1773 
1774 		/* pattern match 'bpf_add Rx, imm' instruction */
1775 		if (opcode == BPF_ADD && BPF_CLASS(insn->code) == BPF_ALU64 &&
1776 		    dst_reg->type == FRAME_PTR && BPF_SRC(insn->code) == BPF_K) {
1777 			dst_reg->type = PTR_TO_STACK;
1778 			dst_reg->imm = insn->imm;
1779 			return 0;
1780 		} else if (opcode == BPF_ADD &&
1781 			   BPF_CLASS(insn->code) == BPF_ALU64 &&
1782 			   (dst_reg->type == PTR_TO_PACKET ||
1783 			    (BPF_SRC(insn->code) == BPF_X &&
1784 			     regs[insn->src_reg].type == PTR_TO_PACKET))) {
1785 			/* ptr_to_packet += K|X */
1786 			return check_packet_ptr_add(env, insn);
1787 		} else if (BPF_CLASS(insn->code) == BPF_ALU64 &&
1788 			   dst_reg->type == UNKNOWN_VALUE &&
1789 			   env->allow_ptr_leaks) {
1790 			/* unknown += K|X */
1791 			return evaluate_reg_alu(env, insn);
1792 		} else if (BPF_CLASS(insn->code) == BPF_ALU64 &&
1793 			   dst_reg->type == CONST_IMM &&
1794 			   env->allow_ptr_leaks) {
1795 			/* reg_imm += K|X */
1796 			return evaluate_reg_imm_alu(env, insn);
1797 		} else if (is_pointer_value(env, insn->dst_reg)) {
1798 			verbose("R%d pointer arithmetic prohibited\n",
1799 				insn->dst_reg);
1800 			return -EACCES;
1801 		} else if (BPF_SRC(insn->code) == BPF_X &&
1802 			   is_pointer_value(env, insn->src_reg)) {
1803 			verbose("R%d pointer arithmetic prohibited\n",
1804 				insn->src_reg);
1805 			return -EACCES;
1806 		}
1807 
1808 		/* If we did pointer math on a map value then just set it to our
1809 		 * PTR_TO_MAP_VALUE_ADJ type so we can deal with any stores or
1810 		 * loads to this register appropriately, otherwise just mark the
1811 		 * register as unknown.
1812 		 */
1813 		if (env->allow_ptr_leaks &&
1814 		    (dst_reg->type == PTR_TO_MAP_VALUE ||
1815 		     dst_reg->type == PTR_TO_MAP_VALUE_ADJ))
1816 			dst_reg->type = PTR_TO_MAP_VALUE_ADJ;
1817 		else
1818 			mark_reg_unknown_value(regs, insn->dst_reg);
1819 	}
1820 
1821 	return 0;
1822 }
1823 
1824 static void find_good_pkt_pointers(struct bpf_verifier_state *state,
1825 				   struct bpf_reg_state *dst_reg)
1826 {
1827 	struct bpf_reg_state *regs = state->regs, *reg;
1828 	int i;
1829 
1830 	/* LLVM can generate two kind of checks:
1831 	 *
1832 	 * Type 1:
1833 	 *
1834 	 *   r2 = r3;
1835 	 *   r2 += 8;
1836 	 *   if (r2 > pkt_end) goto <handle exception>
1837 	 *   <access okay>
1838 	 *
1839 	 *   Where:
1840 	 *     r2 == dst_reg, pkt_end == src_reg
1841 	 *     r2=pkt(id=n,off=8,r=0)
1842 	 *     r3=pkt(id=n,off=0,r=0)
1843 	 *
1844 	 * Type 2:
1845 	 *
1846 	 *   r2 = r3;
1847 	 *   r2 += 8;
1848 	 *   if (pkt_end >= r2) goto <access okay>
1849 	 *   <handle exception>
1850 	 *
1851 	 *   Where:
1852 	 *     pkt_end == dst_reg, r2 == src_reg
1853 	 *     r2=pkt(id=n,off=8,r=0)
1854 	 *     r3=pkt(id=n,off=0,r=0)
1855 	 *
1856 	 * Find register r3 and mark its range as r3=pkt(id=n,off=0,r=8)
1857 	 * so that range of bytes [r3, r3 + 8) is safe to access.
1858 	 */
1859 
1860 	for (i = 0; i < MAX_BPF_REG; i++)
1861 		if (regs[i].type == PTR_TO_PACKET && regs[i].id == dst_reg->id)
1862 			regs[i].range = dst_reg->off;
1863 
1864 	for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
1865 		if (state->stack_slot_type[i] != STACK_SPILL)
1866 			continue;
1867 		reg = &state->spilled_regs[i / BPF_REG_SIZE];
1868 		if (reg->type == PTR_TO_PACKET && reg->id == dst_reg->id)
1869 			reg->range = dst_reg->off;
1870 	}
1871 }
1872 
1873 /* Adjusts the register min/max values in the case that the dst_reg is the
1874  * variable register that we are working on, and src_reg is a constant or we're
1875  * simply doing a BPF_K check.
1876  */
1877 static void reg_set_min_max(struct bpf_reg_state *true_reg,
1878 			    struct bpf_reg_state *false_reg, u64 val,
1879 			    u8 opcode)
1880 {
1881 	switch (opcode) {
1882 	case BPF_JEQ:
1883 		/* If this is false then we know nothing Jon Snow, but if it is
1884 		 * true then we know for sure.
1885 		 */
1886 		true_reg->max_value = true_reg->min_value = val;
1887 		break;
1888 	case BPF_JNE:
1889 		/* If this is true we know nothing Jon Snow, but if it is false
1890 		 * we know the value for sure;
1891 		 */
1892 		false_reg->max_value = false_reg->min_value = val;
1893 		break;
1894 	case BPF_JGT:
1895 		/* Unsigned comparison, the minimum value is 0. */
1896 		false_reg->min_value = 0;
1897 	case BPF_JSGT:
1898 		/* If this is false then we know the maximum val is val,
1899 		 * otherwise we know the min val is val+1.
1900 		 */
1901 		false_reg->max_value = val;
1902 		true_reg->min_value = val + 1;
1903 		break;
1904 	case BPF_JGE:
1905 		/* Unsigned comparison, the minimum value is 0. */
1906 		false_reg->min_value = 0;
1907 	case BPF_JSGE:
1908 		/* If this is false then we know the maximum value is val - 1,
1909 		 * otherwise we know the mimimum value is val.
1910 		 */
1911 		false_reg->max_value = val - 1;
1912 		true_reg->min_value = val;
1913 		break;
1914 	default:
1915 		break;
1916 	}
1917 
1918 	check_reg_overflow(false_reg);
1919 	check_reg_overflow(true_reg);
1920 }
1921 
1922 /* Same as above, but for the case that dst_reg is a CONST_IMM reg and src_reg
1923  * is the variable reg.
1924  */
1925 static void reg_set_min_max_inv(struct bpf_reg_state *true_reg,
1926 				struct bpf_reg_state *false_reg, u64 val,
1927 				u8 opcode)
1928 {
1929 	switch (opcode) {
1930 	case BPF_JEQ:
1931 		/* If this is false then we know nothing Jon Snow, but if it is
1932 		 * true then we know for sure.
1933 		 */
1934 		true_reg->max_value = true_reg->min_value = val;
1935 		break;
1936 	case BPF_JNE:
1937 		/* If this is true we know nothing Jon Snow, but if it is false
1938 		 * we know the value for sure;
1939 		 */
1940 		false_reg->max_value = false_reg->min_value = val;
1941 		break;
1942 	case BPF_JGT:
1943 		/* Unsigned comparison, the minimum value is 0. */
1944 		true_reg->min_value = 0;
1945 	case BPF_JSGT:
1946 		/*
1947 		 * If this is false, then the val is <= the register, if it is
1948 		 * true the register <= to the val.
1949 		 */
1950 		false_reg->min_value = val;
1951 		true_reg->max_value = val - 1;
1952 		break;
1953 	case BPF_JGE:
1954 		/* Unsigned comparison, the minimum value is 0. */
1955 		true_reg->min_value = 0;
1956 	case BPF_JSGE:
1957 		/* If this is false then constant < register, if it is true then
1958 		 * the register < constant.
1959 		 */
1960 		false_reg->min_value = val + 1;
1961 		true_reg->max_value = val;
1962 		break;
1963 	default:
1964 		break;
1965 	}
1966 
1967 	check_reg_overflow(false_reg);
1968 	check_reg_overflow(true_reg);
1969 }
1970 
1971 static void mark_map_reg(struct bpf_reg_state *regs, u32 regno, u32 id,
1972 			 enum bpf_reg_type type)
1973 {
1974 	struct bpf_reg_state *reg = &regs[regno];
1975 
1976 	if (reg->type == PTR_TO_MAP_VALUE_OR_NULL && reg->id == id) {
1977 		reg->type = type;
1978 		/* We don't need id from this point onwards anymore, thus we
1979 		 * should better reset it, so that state pruning has chances
1980 		 * to take effect.
1981 		 */
1982 		reg->id = 0;
1983 		if (type == UNKNOWN_VALUE)
1984 			__mark_reg_unknown_value(regs, regno);
1985 	}
1986 }
1987 
1988 /* The logic is similar to find_good_pkt_pointers(), both could eventually
1989  * be folded together at some point.
1990  */
1991 static void mark_map_regs(struct bpf_verifier_state *state, u32 regno,
1992 			  enum bpf_reg_type type)
1993 {
1994 	struct bpf_reg_state *regs = state->regs;
1995 	u32 id = regs[regno].id;
1996 	int i;
1997 
1998 	for (i = 0; i < MAX_BPF_REG; i++)
1999 		mark_map_reg(regs, i, id, type);
2000 
2001 	for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
2002 		if (state->stack_slot_type[i] != STACK_SPILL)
2003 			continue;
2004 		mark_map_reg(state->spilled_regs, i / BPF_REG_SIZE, id, type);
2005 	}
2006 }
2007 
2008 static int check_cond_jmp_op(struct bpf_verifier_env *env,
2009 			     struct bpf_insn *insn, int *insn_idx)
2010 {
2011 	struct bpf_verifier_state *other_branch, *this_branch = &env->cur_state;
2012 	struct bpf_reg_state *regs = this_branch->regs, *dst_reg;
2013 	u8 opcode = BPF_OP(insn->code);
2014 	int err;
2015 
2016 	if (opcode > BPF_EXIT) {
2017 		verbose("invalid BPF_JMP opcode %x\n", opcode);
2018 		return -EINVAL;
2019 	}
2020 
2021 	if (BPF_SRC(insn->code) == BPF_X) {
2022 		if (insn->imm != 0) {
2023 			verbose("BPF_JMP uses reserved fields\n");
2024 			return -EINVAL;
2025 		}
2026 
2027 		/* check src1 operand */
2028 		err = check_reg_arg(regs, insn->src_reg, SRC_OP);
2029 		if (err)
2030 			return err;
2031 
2032 		if (is_pointer_value(env, insn->src_reg)) {
2033 			verbose("R%d pointer comparison prohibited\n",
2034 				insn->src_reg);
2035 			return -EACCES;
2036 		}
2037 	} else {
2038 		if (insn->src_reg != BPF_REG_0) {
2039 			verbose("BPF_JMP uses reserved fields\n");
2040 			return -EINVAL;
2041 		}
2042 	}
2043 
2044 	/* check src2 operand */
2045 	err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
2046 	if (err)
2047 		return err;
2048 
2049 	dst_reg = &regs[insn->dst_reg];
2050 
2051 	/* detect if R == 0 where R was initialized to zero earlier */
2052 	if (BPF_SRC(insn->code) == BPF_K &&
2053 	    (opcode == BPF_JEQ || opcode == BPF_JNE) &&
2054 	    dst_reg->type == CONST_IMM && dst_reg->imm == insn->imm) {
2055 		if (opcode == BPF_JEQ) {
2056 			/* if (imm == imm) goto pc+off;
2057 			 * only follow the goto, ignore fall-through
2058 			 */
2059 			*insn_idx += insn->off;
2060 			return 0;
2061 		} else {
2062 			/* if (imm != imm) goto pc+off;
2063 			 * only follow fall-through branch, since
2064 			 * that's where the program will go
2065 			 */
2066 			return 0;
2067 		}
2068 	}
2069 
2070 	other_branch = push_stack(env, *insn_idx + insn->off + 1, *insn_idx);
2071 	if (!other_branch)
2072 		return -EFAULT;
2073 
2074 	/* detect if we are comparing against a constant value so we can adjust
2075 	 * our min/max values for our dst register.
2076 	 */
2077 	if (BPF_SRC(insn->code) == BPF_X) {
2078 		if (regs[insn->src_reg].type == CONST_IMM)
2079 			reg_set_min_max(&other_branch->regs[insn->dst_reg],
2080 					dst_reg, regs[insn->src_reg].imm,
2081 					opcode);
2082 		else if (dst_reg->type == CONST_IMM)
2083 			reg_set_min_max_inv(&other_branch->regs[insn->src_reg],
2084 					    &regs[insn->src_reg], dst_reg->imm,
2085 					    opcode);
2086 	} else {
2087 		reg_set_min_max(&other_branch->regs[insn->dst_reg],
2088 					dst_reg, insn->imm, opcode);
2089 	}
2090 
2091 	/* detect if R == 0 where R is returned from bpf_map_lookup_elem() */
2092 	if (BPF_SRC(insn->code) == BPF_K &&
2093 	    insn->imm == 0 && (opcode == BPF_JEQ || opcode == BPF_JNE) &&
2094 	    dst_reg->type == PTR_TO_MAP_VALUE_OR_NULL) {
2095 		/* Mark all identical map registers in each branch as either
2096 		 * safe or unknown depending R == 0 or R != 0 conditional.
2097 		 */
2098 		mark_map_regs(this_branch, insn->dst_reg,
2099 			      opcode == BPF_JEQ ? PTR_TO_MAP_VALUE : UNKNOWN_VALUE);
2100 		mark_map_regs(other_branch, insn->dst_reg,
2101 			      opcode == BPF_JEQ ? UNKNOWN_VALUE : PTR_TO_MAP_VALUE);
2102 	} else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JGT &&
2103 		   dst_reg->type == PTR_TO_PACKET &&
2104 		   regs[insn->src_reg].type == PTR_TO_PACKET_END) {
2105 		find_good_pkt_pointers(this_branch, dst_reg);
2106 	} else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JGE &&
2107 		   dst_reg->type == PTR_TO_PACKET_END &&
2108 		   regs[insn->src_reg].type == PTR_TO_PACKET) {
2109 		find_good_pkt_pointers(other_branch, &regs[insn->src_reg]);
2110 	} else if (is_pointer_value(env, insn->dst_reg)) {
2111 		verbose("R%d pointer comparison prohibited\n", insn->dst_reg);
2112 		return -EACCES;
2113 	}
2114 	if (log_level)
2115 		print_verifier_state(this_branch);
2116 	return 0;
2117 }
2118 
2119 /* return the map pointer stored inside BPF_LD_IMM64 instruction */
2120 static struct bpf_map *ld_imm64_to_map_ptr(struct bpf_insn *insn)
2121 {
2122 	u64 imm64 = ((u64) (u32) insn[0].imm) | ((u64) (u32) insn[1].imm) << 32;
2123 
2124 	return (struct bpf_map *) (unsigned long) imm64;
2125 }
2126 
2127 /* verify BPF_LD_IMM64 instruction */
2128 static int check_ld_imm(struct bpf_verifier_env *env, struct bpf_insn *insn)
2129 {
2130 	struct bpf_reg_state *regs = env->cur_state.regs;
2131 	int err;
2132 
2133 	if (BPF_SIZE(insn->code) != BPF_DW) {
2134 		verbose("invalid BPF_LD_IMM insn\n");
2135 		return -EINVAL;
2136 	}
2137 	if (insn->off != 0) {
2138 		verbose("BPF_LD_IMM64 uses reserved fields\n");
2139 		return -EINVAL;
2140 	}
2141 
2142 	err = check_reg_arg(regs, insn->dst_reg, DST_OP);
2143 	if (err)
2144 		return err;
2145 
2146 	if (insn->src_reg == 0) {
2147 		/* generic move 64-bit immediate into a register,
2148 		 * only analyzer needs to collect the ld_imm value.
2149 		 */
2150 		u64 imm = ((u64)(insn + 1)->imm << 32) | (u32)insn->imm;
2151 
2152 		if (!env->analyzer_ops)
2153 			return 0;
2154 
2155 		regs[insn->dst_reg].type = CONST_IMM;
2156 		regs[insn->dst_reg].imm = imm;
2157 		return 0;
2158 	}
2159 
2160 	/* replace_map_fd_with_map_ptr() should have caught bad ld_imm64 */
2161 	BUG_ON(insn->src_reg != BPF_PSEUDO_MAP_FD);
2162 
2163 	regs[insn->dst_reg].type = CONST_PTR_TO_MAP;
2164 	regs[insn->dst_reg].map_ptr = ld_imm64_to_map_ptr(insn);
2165 	return 0;
2166 }
2167 
2168 static bool may_access_skb(enum bpf_prog_type type)
2169 {
2170 	switch (type) {
2171 	case BPF_PROG_TYPE_SOCKET_FILTER:
2172 	case BPF_PROG_TYPE_SCHED_CLS:
2173 	case BPF_PROG_TYPE_SCHED_ACT:
2174 		return true;
2175 	default:
2176 		return false;
2177 	}
2178 }
2179 
2180 /* verify safety of LD_ABS|LD_IND instructions:
2181  * - they can only appear in the programs where ctx == skb
2182  * - since they are wrappers of function calls, they scratch R1-R5 registers,
2183  *   preserve R6-R9, and store return value into R0
2184  *
2185  * Implicit input:
2186  *   ctx == skb == R6 == CTX
2187  *
2188  * Explicit input:
2189  *   SRC == any register
2190  *   IMM == 32-bit immediate
2191  *
2192  * Output:
2193  *   R0 - 8/16/32-bit skb data converted to cpu endianness
2194  */
2195 static int check_ld_abs(struct bpf_verifier_env *env, struct bpf_insn *insn)
2196 {
2197 	struct bpf_reg_state *regs = env->cur_state.regs;
2198 	u8 mode = BPF_MODE(insn->code);
2199 	struct bpf_reg_state *reg;
2200 	int i, err;
2201 
2202 	if (!may_access_skb(env->prog->type)) {
2203 		verbose("BPF_LD_[ABS|IND] instructions not allowed for this program type\n");
2204 		return -EINVAL;
2205 	}
2206 
2207 	if (insn->dst_reg != BPF_REG_0 || insn->off != 0 ||
2208 	    BPF_SIZE(insn->code) == BPF_DW ||
2209 	    (mode == BPF_ABS && insn->src_reg != BPF_REG_0)) {
2210 		verbose("BPF_LD_[ABS|IND] uses reserved fields\n");
2211 		return -EINVAL;
2212 	}
2213 
2214 	/* check whether implicit source operand (register R6) is readable */
2215 	err = check_reg_arg(regs, BPF_REG_6, SRC_OP);
2216 	if (err)
2217 		return err;
2218 
2219 	if (regs[BPF_REG_6].type != PTR_TO_CTX) {
2220 		verbose("at the time of BPF_LD_ABS|IND R6 != pointer to skb\n");
2221 		return -EINVAL;
2222 	}
2223 
2224 	if (mode == BPF_IND) {
2225 		/* check explicit source operand */
2226 		err = check_reg_arg(regs, insn->src_reg, SRC_OP);
2227 		if (err)
2228 			return err;
2229 	}
2230 
2231 	/* reset caller saved regs to unreadable */
2232 	for (i = 0; i < CALLER_SAVED_REGS; i++) {
2233 		reg = regs + caller_saved[i];
2234 		reg->type = NOT_INIT;
2235 		reg->imm = 0;
2236 	}
2237 
2238 	/* mark destination R0 register as readable, since it contains
2239 	 * the value fetched from the packet
2240 	 */
2241 	regs[BPF_REG_0].type = UNKNOWN_VALUE;
2242 	return 0;
2243 }
2244 
2245 /* non-recursive DFS pseudo code
2246  * 1  procedure DFS-iterative(G,v):
2247  * 2      label v as discovered
2248  * 3      let S be a stack
2249  * 4      S.push(v)
2250  * 5      while S is not empty
2251  * 6            t <- S.pop()
2252  * 7            if t is what we're looking for:
2253  * 8                return t
2254  * 9            for all edges e in G.adjacentEdges(t) do
2255  * 10               if edge e is already labelled
2256  * 11                   continue with the next edge
2257  * 12               w <- G.adjacentVertex(t,e)
2258  * 13               if vertex w is not discovered and not explored
2259  * 14                   label e as tree-edge
2260  * 15                   label w as discovered
2261  * 16                   S.push(w)
2262  * 17                   continue at 5
2263  * 18               else if vertex w is discovered
2264  * 19                   label e as back-edge
2265  * 20               else
2266  * 21                   // vertex w is explored
2267  * 22                   label e as forward- or cross-edge
2268  * 23           label t as explored
2269  * 24           S.pop()
2270  *
2271  * convention:
2272  * 0x10 - discovered
2273  * 0x11 - discovered and fall-through edge labelled
2274  * 0x12 - discovered and fall-through and branch edges labelled
2275  * 0x20 - explored
2276  */
2277 
2278 enum {
2279 	DISCOVERED = 0x10,
2280 	EXPLORED = 0x20,
2281 	FALLTHROUGH = 1,
2282 	BRANCH = 2,
2283 };
2284 
2285 #define STATE_LIST_MARK ((struct bpf_verifier_state_list *) -1L)
2286 
2287 static int *insn_stack;	/* stack of insns to process */
2288 static int cur_stack;	/* current stack index */
2289 static int *insn_state;
2290 
2291 /* t, w, e - match pseudo-code above:
2292  * t - index of current instruction
2293  * w - next instruction
2294  * e - edge
2295  */
2296 static int push_insn(int t, int w, int e, struct bpf_verifier_env *env)
2297 {
2298 	if (e == FALLTHROUGH && insn_state[t] >= (DISCOVERED | FALLTHROUGH))
2299 		return 0;
2300 
2301 	if (e == BRANCH && insn_state[t] >= (DISCOVERED | BRANCH))
2302 		return 0;
2303 
2304 	if (w < 0 || w >= env->prog->len) {
2305 		verbose("jump out of range from insn %d to %d\n", t, w);
2306 		return -EINVAL;
2307 	}
2308 
2309 	if (e == BRANCH)
2310 		/* mark branch target for state pruning */
2311 		env->explored_states[w] = STATE_LIST_MARK;
2312 
2313 	if (insn_state[w] == 0) {
2314 		/* tree-edge */
2315 		insn_state[t] = DISCOVERED | e;
2316 		insn_state[w] = DISCOVERED;
2317 		if (cur_stack >= env->prog->len)
2318 			return -E2BIG;
2319 		insn_stack[cur_stack++] = w;
2320 		return 1;
2321 	} else if ((insn_state[w] & 0xF0) == DISCOVERED) {
2322 		verbose("back-edge from insn %d to %d\n", t, w);
2323 		return -EINVAL;
2324 	} else if (insn_state[w] == EXPLORED) {
2325 		/* forward- or cross-edge */
2326 		insn_state[t] = DISCOVERED | e;
2327 	} else {
2328 		verbose("insn state internal bug\n");
2329 		return -EFAULT;
2330 	}
2331 	return 0;
2332 }
2333 
2334 /* non-recursive depth-first-search to detect loops in BPF program
2335  * loop == back-edge in directed graph
2336  */
2337 static int check_cfg(struct bpf_verifier_env *env)
2338 {
2339 	struct bpf_insn *insns = env->prog->insnsi;
2340 	int insn_cnt = env->prog->len;
2341 	int ret = 0;
2342 	int i, t;
2343 
2344 	insn_state = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
2345 	if (!insn_state)
2346 		return -ENOMEM;
2347 
2348 	insn_stack = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
2349 	if (!insn_stack) {
2350 		kfree(insn_state);
2351 		return -ENOMEM;
2352 	}
2353 
2354 	insn_state[0] = DISCOVERED; /* mark 1st insn as discovered */
2355 	insn_stack[0] = 0; /* 0 is the first instruction */
2356 	cur_stack = 1;
2357 
2358 peek_stack:
2359 	if (cur_stack == 0)
2360 		goto check_state;
2361 	t = insn_stack[cur_stack - 1];
2362 
2363 	if (BPF_CLASS(insns[t].code) == BPF_JMP) {
2364 		u8 opcode = BPF_OP(insns[t].code);
2365 
2366 		if (opcode == BPF_EXIT) {
2367 			goto mark_explored;
2368 		} else if (opcode == BPF_CALL) {
2369 			ret = push_insn(t, t + 1, FALLTHROUGH, env);
2370 			if (ret == 1)
2371 				goto peek_stack;
2372 			else if (ret < 0)
2373 				goto err_free;
2374 			if (t + 1 < insn_cnt)
2375 				env->explored_states[t + 1] = STATE_LIST_MARK;
2376 		} else if (opcode == BPF_JA) {
2377 			if (BPF_SRC(insns[t].code) != BPF_K) {
2378 				ret = -EINVAL;
2379 				goto err_free;
2380 			}
2381 			/* unconditional jump with single edge */
2382 			ret = push_insn(t, t + insns[t].off + 1,
2383 					FALLTHROUGH, env);
2384 			if (ret == 1)
2385 				goto peek_stack;
2386 			else if (ret < 0)
2387 				goto err_free;
2388 			/* tell verifier to check for equivalent states
2389 			 * after every call and jump
2390 			 */
2391 			if (t + 1 < insn_cnt)
2392 				env->explored_states[t + 1] = STATE_LIST_MARK;
2393 		} else {
2394 			/* conditional jump with two edges */
2395 			ret = push_insn(t, t + 1, FALLTHROUGH, env);
2396 			if (ret == 1)
2397 				goto peek_stack;
2398 			else if (ret < 0)
2399 				goto err_free;
2400 
2401 			ret = push_insn(t, t + insns[t].off + 1, BRANCH, env);
2402 			if (ret == 1)
2403 				goto peek_stack;
2404 			else if (ret < 0)
2405 				goto err_free;
2406 		}
2407 	} else {
2408 		/* all other non-branch instructions with single
2409 		 * fall-through edge
2410 		 */
2411 		ret = push_insn(t, t + 1, FALLTHROUGH, env);
2412 		if (ret == 1)
2413 			goto peek_stack;
2414 		else if (ret < 0)
2415 			goto err_free;
2416 	}
2417 
2418 mark_explored:
2419 	insn_state[t] = EXPLORED;
2420 	if (cur_stack-- <= 0) {
2421 		verbose("pop stack internal bug\n");
2422 		ret = -EFAULT;
2423 		goto err_free;
2424 	}
2425 	goto peek_stack;
2426 
2427 check_state:
2428 	for (i = 0; i < insn_cnt; i++) {
2429 		if (insn_state[i] != EXPLORED) {
2430 			verbose("unreachable insn %d\n", i);
2431 			ret = -EINVAL;
2432 			goto err_free;
2433 		}
2434 	}
2435 	ret = 0; /* cfg looks good */
2436 
2437 err_free:
2438 	kfree(insn_state);
2439 	kfree(insn_stack);
2440 	return ret;
2441 }
2442 
2443 /* the following conditions reduce the number of explored insns
2444  * from ~140k to ~80k for ultra large programs that use a lot of ptr_to_packet
2445  */
2446 static bool compare_ptrs_to_packet(struct bpf_reg_state *old,
2447 				   struct bpf_reg_state *cur)
2448 {
2449 	if (old->id != cur->id)
2450 		return false;
2451 
2452 	/* old ptr_to_packet is more conservative, since it allows smaller
2453 	 * range. Ex:
2454 	 * old(off=0,r=10) is equal to cur(off=0,r=20), because
2455 	 * old(off=0,r=10) means that with range=10 the verifier proceeded
2456 	 * further and found no issues with the program. Now we're in the same
2457 	 * spot with cur(off=0,r=20), so we're safe too, since anything further
2458 	 * will only be looking at most 10 bytes after this pointer.
2459 	 */
2460 	if (old->off == cur->off && old->range < cur->range)
2461 		return true;
2462 
2463 	/* old(off=20,r=10) is equal to cur(off=22,re=22 or 5 or 0)
2464 	 * since both cannot be used for packet access and safe(old)
2465 	 * pointer has smaller off that could be used for further
2466 	 * 'if (ptr > data_end)' check
2467 	 * Ex:
2468 	 * old(off=20,r=10) and cur(off=22,r=22) and cur(off=22,r=0) mean
2469 	 * that we cannot access the packet.
2470 	 * The safe range is:
2471 	 * [ptr, ptr + range - off)
2472 	 * so whenever off >=range, it means no safe bytes from this pointer.
2473 	 * When comparing old->off <= cur->off, it means that older code
2474 	 * went with smaller offset and that offset was later
2475 	 * used to figure out the safe range after 'if (ptr > data_end)' check
2476 	 * Say, 'old' state was explored like:
2477 	 * ... R3(off=0, r=0)
2478 	 * R4 = R3 + 20
2479 	 * ... now R4(off=20,r=0)  <-- here
2480 	 * if (R4 > data_end)
2481 	 * ... R4(off=20,r=20), R3(off=0,r=20) and R3 can be used to access.
2482 	 * ... the code further went all the way to bpf_exit.
2483 	 * Now the 'cur' state at the mark 'here' has R4(off=30,r=0).
2484 	 * old_R4(off=20,r=0) equal to cur_R4(off=30,r=0), since if the verifier
2485 	 * goes further, such cur_R4 will give larger safe packet range after
2486 	 * 'if (R4 > data_end)' and all further insn were already good with r=20,
2487 	 * so they will be good with r=30 and we can prune the search.
2488 	 */
2489 	if (old->off <= cur->off &&
2490 	    old->off >= old->range && cur->off >= cur->range)
2491 		return true;
2492 
2493 	return false;
2494 }
2495 
2496 /* compare two verifier states
2497  *
2498  * all states stored in state_list are known to be valid, since
2499  * verifier reached 'bpf_exit' instruction through them
2500  *
2501  * this function is called when verifier exploring different branches of
2502  * execution popped from the state stack. If it sees an old state that has
2503  * more strict register state and more strict stack state then this execution
2504  * branch doesn't need to be explored further, since verifier already
2505  * concluded that more strict state leads to valid finish.
2506  *
2507  * Therefore two states are equivalent if register state is more conservative
2508  * and explored stack state is more conservative than the current one.
2509  * Example:
2510  *       explored                   current
2511  * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC)
2512  * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC)
2513  *
2514  * In other words if current stack state (one being explored) has more
2515  * valid slots than old one that already passed validation, it means
2516  * the verifier can stop exploring and conclude that current state is valid too
2517  *
2518  * Similarly with registers. If explored state has register type as invalid
2519  * whereas register type in current state is meaningful, it means that
2520  * the current state will reach 'bpf_exit' instruction safely
2521  */
2522 static bool states_equal(struct bpf_verifier_env *env,
2523 			 struct bpf_verifier_state *old,
2524 			 struct bpf_verifier_state *cur)
2525 {
2526 	bool varlen_map_access = env->varlen_map_value_access;
2527 	struct bpf_reg_state *rold, *rcur;
2528 	int i;
2529 
2530 	for (i = 0; i < MAX_BPF_REG; i++) {
2531 		rold = &old->regs[i];
2532 		rcur = &cur->regs[i];
2533 
2534 		if (memcmp(rold, rcur, sizeof(*rold)) == 0)
2535 			continue;
2536 
2537 		/* If the ranges were not the same, but everything else was and
2538 		 * we didn't do a variable access into a map then we are a-ok.
2539 		 */
2540 		if (!varlen_map_access &&
2541 		    memcmp(rold, rcur, offsetofend(struct bpf_reg_state, id)) == 0)
2542 			continue;
2543 
2544 		/* If we didn't map access then again we don't care about the
2545 		 * mismatched range values and it's ok if our old type was
2546 		 * UNKNOWN and we didn't go to a NOT_INIT'ed reg.
2547 		 */
2548 		if (rold->type == NOT_INIT ||
2549 		    (!varlen_map_access && rold->type == UNKNOWN_VALUE &&
2550 		     rcur->type != NOT_INIT))
2551 			continue;
2552 
2553 		if (rold->type == PTR_TO_PACKET && rcur->type == PTR_TO_PACKET &&
2554 		    compare_ptrs_to_packet(rold, rcur))
2555 			continue;
2556 
2557 		return false;
2558 	}
2559 
2560 	for (i = 0; i < MAX_BPF_STACK; i++) {
2561 		if (old->stack_slot_type[i] == STACK_INVALID)
2562 			continue;
2563 		if (old->stack_slot_type[i] != cur->stack_slot_type[i])
2564 			/* Ex: old explored (safe) state has STACK_SPILL in
2565 			 * this stack slot, but current has has STACK_MISC ->
2566 			 * this verifier states are not equivalent,
2567 			 * return false to continue verification of this path
2568 			 */
2569 			return false;
2570 		if (i % BPF_REG_SIZE)
2571 			continue;
2572 		if (memcmp(&old->spilled_regs[i / BPF_REG_SIZE],
2573 			   &cur->spilled_regs[i / BPF_REG_SIZE],
2574 			   sizeof(old->spilled_regs[0])))
2575 			/* when explored and current stack slot types are
2576 			 * the same, check that stored pointers types
2577 			 * are the same as well.
2578 			 * Ex: explored safe path could have stored
2579 			 * (bpf_reg_state) {.type = PTR_TO_STACK, .imm = -8}
2580 			 * but current path has stored:
2581 			 * (bpf_reg_state) {.type = PTR_TO_STACK, .imm = -16}
2582 			 * such verifier states are not equivalent.
2583 			 * return false to continue verification of this path
2584 			 */
2585 			return false;
2586 		else
2587 			continue;
2588 	}
2589 	return true;
2590 }
2591 
2592 static int is_state_visited(struct bpf_verifier_env *env, int insn_idx)
2593 {
2594 	struct bpf_verifier_state_list *new_sl;
2595 	struct bpf_verifier_state_list *sl;
2596 
2597 	sl = env->explored_states[insn_idx];
2598 	if (!sl)
2599 		/* this 'insn_idx' instruction wasn't marked, so we will not
2600 		 * be doing state search here
2601 		 */
2602 		return 0;
2603 
2604 	while (sl != STATE_LIST_MARK) {
2605 		if (states_equal(env, &sl->state, &env->cur_state))
2606 			/* reached equivalent register/stack state,
2607 			 * prune the search
2608 			 */
2609 			return 1;
2610 		sl = sl->next;
2611 	}
2612 
2613 	/* there were no equivalent states, remember current one.
2614 	 * technically the current state is not proven to be safe yet,
2615 	 * but it will either reach bpf_exit (which means it's safe) or
2616 	 * it will be rejected. Since there are no loops, we won't be
2617 	 * seeing this 'insn_idx' instruction again on the way to bpf_exit
2618 	 */
2619 	new_sl = kmalloc(sizeof(struct bpf_verifier_state_list), GFP_USER);
2620 	if (!new_sl)
2621 		return -ENOMEM;
2622 
2623 	/* add new state to the head of linked list */
2624 	memcpy(&new_sl->state, &env->cur_state, sizeof(env->cur_state));
2625 	new_sl->next = env->explored_states[insn_idx];
2626 	env->explored_states[insn_idx] = new_sl;
2627 	return 0;
2628 }
2629 
2630 static int ext_analyzer_insn_hook(struct bpf_verifier_env *env,
2631 				  int insn_idx, int prev_insn_idx)
2632 {
2633 	if (!env->analyzer_ops || !env->analyzer_ops->insn_hook)
2634 		return 0;
2635 
2636 	return env->analyzer_ops->insn_hook(env, insn_idx, prev_insn_idx);
2637 }
2638 
2639 static int do_check(struct bpf_verifier_env *env)
2640 {
2641 	struct bpf_verifier_state *state = &env->cur_state;
2642 	struct bpf_insn *insns = env->prog->insnsi;
2643 	struct bpf_reg_state *regs = state->regs;
2644 	int insn_cnt = env->prog->len;
2645 	int insn_idx, prev_insn_idx = 0;
2646 	int insn_processed = 0;
2647 	bool do_print_state = false;
2648 
2649 	init_reg_state(regs);
2650 	insn_idx = 0;
2651 	env->varlen_map_value_access = false;
2652 	for (;;) {
2653 		struct bpf_insn *insn;
2654 		u8 class;
2655 		int err;
2656 
2657 		if (insn_idx >= insn_cnt) {
2658 			verbose("invalid insn idx %d insn_cnt %d\n",
2659 				insn_idx, insn_cnt);
2660 			return -EFAULT;
2661 		}
2662 
2663 		insn = &insns[insn_idx];
2664 		class = BPF_CLASS(insn->code);
2665 
2666 		if (++insn_processed > BPF_COMPLEXITY_LIMIT_INSNS) {
2667 			verbose("BPF program is too large. Proccessed %d insn\n",
2668 				insn_processed);
2669 			return -E2BIG;
2670 		}
2671 
2672 		err = is_state_visited(env, insn_idx);
2673 		if (err < 0)
2674 			return err;
2675 		if (err == 1) {
2676 			/* found equivalent state, can prune the search */
2677 			if (log_level) {
2678 				if (do_print_state)
2679 					verbose("\nfrom %d to %d: safe\n",
2680 						prev_insn_idx, insn_idx);
2681 				else
2682 					verbose("%d: safe\n", insn_idx);
2683 			}
2684 			goto process_bpf_exit;
2685 		}
2686 
2687 		if (log_level && do_print_state) {
2688 			verbose("\nfrom %d to %d:", prev_insn_idx, insn_idx);
2689 			print_verifier_state(&env->cur_state);
2690 			do_print_state = false;
2691 		}
2692 
2693 		if (log_level) {
2694 			verbose("%d: ", insn_idx);
2695 			print_bpf_insn(insn);
2696 		}
2697 
2698 		err = ext_analyzer_insn_hook(env, insn_idx, prev_insn_idx);
2699 		if (err)
2700 			return err;
2701 
2702 		if (class == BPF_ALU || class == BPF_ALU64) {
2703 			err = check_alu_op(env, insn);
2704 			if (err)
2705 				return err;
2706 
2707 		} else if (class == BPF_LDX) {
2708 			enum bpf_reg_type *prev_src_type, src_reg_type;
2709 
2710 			/* check for reserved fields is already done */
2711 
2712 			/* check src operand */
2713 			err = check_reg_arg(regs, insn->src_reg, SRC_OP);
2714 			if (err)
2715 				return err;
2716 
2717 			err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK);
2718 			if (err)
2719 				return err;
2720 
2721 			src_reg_type = regs[insn->src_reg].type;
2722 
2723 			/* check that memory (src_reg + off) is readable,
2724 			 * the state of dst_reg will be updated by this func
2725 			 */
2726 			err = check_mem_access(env, insn->src_reg, insn->off,
2727 					       BPF_SIZE(insn->code), BPF_READ,
2728 					       insn->dst_reg);
2729 			if (err)
2730 				return err;
2731 
2732 			reset_reg_range_values(regs, insn->dst_reg);
2733 			if (BPF_SIZE(insn->code) != BPF_W &&
2734 			    BPF_SIZE(insn->code) != BPF_DW) {
2735 				insn_idx++;
2736 				continue;
2737 			}
2738 
2739 			prev_src_type = &env->insn_aux_data[insn_idx].ptr_type;
2740 
2741 			if (*prev_src_type == NOT_INIT) {
2742 				/* saw a valid insn
2743 				 * dst_reg = *(u32 *)(src_reg + off)
2744 				 * save type to validate intersecting paths
2745 				 */
2746 				*prev_src_type = src_reg_type;
2747 
2748 			} else if (src_reg_type != *prev_src_type &&
2749 				   (src_reg_type == PTR_TO_CTX ||
2750 				    *prev_src_type == PTR_TO_CTX)) {
2751 				/* ABuser program is trying to use the same insn
2752 				 * dst_reg = *(u32*) (src_reg + off)
2753 				 * with different pointer types:
2754 				 * src_reg == ctx in one branch and
2755 				 * src_reg == stack|map in some other branch.
2756 				 * Reject it.
2757 				 */
2758 				verbose("same insn cannot be used with different pointers\n");
2759 				return -EINVAL;
2760 			}
2761 
2762 		} else if (class == BPF_STX) {
2763 			enum bpf_reg_type *prev_dst_type, dst_reg_type;
2764 
2765 			if (BPF_MODE(insn->code) == BPF_XADD) {
2766 				err = check_xadd(env, insn);
2767 				if (err)
2768 					return err;
2769 				insn_idx++;
2770 				continue;
2771 			}
2772 
2773 			/* check src1 operand */
2774 			err = check_reg_arg(regs, insn->src_reg, SRC_OP);
2775 			if (err)
2776 				return err;
2777 			/* check src2 operand */
2778 			err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
2779 			if (err)
2780 				return err;
2781 
2782 			dst_reg_type = regs[insn->dst_reg].type;
2783 
2784 			/* check that memory (dst_reg + off) is writeable */
2785 			err = check_mem_access(env, insn->dst_reg, insn->off,
2786 					       BPF_SIZE(insn->code), BPF_WRITE,
2787 					       insn->src_reg);
2788 			if (err)
2789 				return err;
2790 
2791 			prev_dst_type = &env->insn_aux_data[insn_idx].ptr_type;
2792 
2793 			if (*prev_dst_type == NOT_INIT) {
2794 				*prev_dst_type = dst_reg_type;
2795 			} else if (dst_reg_type != *prev_dst_type &&
2796 				   (dst_reg_type == PTR_TO_CTX ||
2797 				    *prev_dst_type == PTR_TO_CTX)) {
2798 				verbose("same insn cannot be used with different pointers\n");
2799 				return -EINVAL;
2800 			}
2801 
2802 		} else if (class == BPF_ST) {
2803 			if (BPF_MODE(insn->code) != BPF_MEM ||
2804 			    insn->src_reg != BPF_REG_0) {
2805 				verbose("BPF_ST uses reserved fields\n");
2806 				return -EINVAL;
2807 			}
2808 			/* check src operand */
2809 			err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
2810 			if (err)
2811 				return err;
2812 
2813 			/* check that memory (dst_reg + off) is writeable */
2814 			err = check_mem_access(env, insn->dst_reg, insn->off,
2815 					       BPF_SIZE(insn->code), BPF_WRITE,
2816 					       -1);
2817 			if (err)
2818 				return err;
2819 
2820 		} else if (class == BPF_JMP) {
2821 			u8 opcode = BPF_OP(insn->code);
2822 
2823 			if (opcode == BPF_CALL) {
2824 				if (BPF_SRC(insn->code) != BPF_K ||
2825 				    insn->off != 0 ||
2826 				    insn->src_reg != BPF_REG_0 ||
2827 				    insn->dst_reg != BPF_REG_0) {
2828 					verbose("BPF_CALL uses reserved fields\n");
2829 					return -EINVAL;
2830 				}
2831 
2832 				err = check_call(env, insn->imm);
2833 				if (err)
2834 					return err;
2835 
2836 			} else if (opcode == BPF_JA) {
2837 				if (BPF_SRC(insn->code) != BPF_K ||
2838 				    insn->imm != 0 ||
2839 				    insn->src_reg != BPF_REG_0 ||
2840 				    insn->dst_reg != BPF_REG_0) {
2841 					verbose("BPF_JA uses reserved fields\n");
2842 					return -EINVAL;
2843 				}
2844 
2845 				insn_idx += insn->off + 1;
2846 				continue;
2847 
2848 			} else if (opcode == BPF_EXIT) {
2849 				if (BPF_SRC(insn->code) != BPF_K ||
2850 				    insn->imm != 0 ||
2851 				    insn->src_reg != BPF_REG_0 ||
2852 				    insn->dst_reg != BPF_REG_0) {
2853 					verbose("BPF_EXIT uses reserved fields\n");
2854 					return -EINVAL;
2855 				}
2856 
2857 				/* eBPF calling convetion is such that R0 is used
2858 				 * to return the value from eBPF program.
2859 				 * Make sure that it's readable at this time
2860 				 * of bpf_exit, which means that program wrote
2861 				 * something into it earlier
2862 				 */
2863 				err = check_reg_arg(regs, BPF_REG_0, SRC_OP);
2864 				if (err)
2865 					return err;
2866 
2867 				if (is_pointer_value(env, BPF_REG_0)) {
2868 					verbose("R0 leaks addr as return value\n");
2869 					return -EACCES;
2870 				}
2871 
2872 process_bpf_exit:
2873 				insn_idx = pop_stack(env, &prev_insn_idx);
2874 				if (insn_idx < 0) {
2875 					break;
2876 				} else {
2877 					do_print_state = true;
2878 					continue;
2879 				}
2880 			} else {
2881 				err = check_cond_jmp_op(env, insn, &insn_idx);
2882 				if (err)
2883 					return err;
2884 			}
2885 		} else if (class == BPF_LD) {
2886 			u8 mode = BPF_MODE(insn->code);
2887 
2888 			if (mode == BPF_ABS || mode == BPF_IND) {
2889 				err = check_ld_abs(env, insn);
2890 				if (err)
2891 					return err;
2892 
2893 			} else if (mode == BPF_IMM) {
2894 				err = check_ld_imm(env, insn);
2895 				if (err)
2896 					return err;
2897 
2898 				insn_idx++;
2899 			} else {
2900 				verbose("invalid BPF_LD mode\n");
2901 				return -EINVAL;
2902 			}
2903 			reset_reg_range_values(regs, insn->dst_reg);
2904 		} else {
2905 			verbose("unknown insn class %d\n", class);
2906 			return -EINVAL;
2907 		}
2908 
2909 		insn_idx++;
2910 	}
2911 
2912 	verbose("processed %d insns\n", insn_processed);
2913 	return 0;
2914 }
2915 
2916 static int check_map_prog_compatibility(struct bpf_map *map,
2917 					struct bpf_prog *prog)
2918 
2919 {
2920 	if (prog->type == BPF_PROG_TYPE_PERF_EVENT &&
2921 	    (map->map_type == BPF_MAP_TYPE_HASH ||
2922 	     map->map_type == BPF_MAP_TYPE_PERCPU_HASH) &&
2923 	    (map->map_flags & BPF_F_NO_PREALLOC)) {
2924 		verbose("perf_event programs can only use preallocated hash map\n");
2925 		return -EINVAL;
2926 	}
2927 	return 0;
2928 }
2929 
2930 /* look for pseudo eBPF instructions that access map FDs and
2931  * replace them with actual map pointers
2932  */
2933 static int replace_map_fd_with_map_ptr(struct bpf_verifier_env *env)
2934 {
2935 	struct bpf_insn *insn = env->prog->insnsi;
2936 	int insn_cnt = env->prog->len;
2937 	int i, j, err;
2938 
2939 	err = bpf_prog_calc_digest(env->prog);
2940 	if (err)
2941 		return err;
2942 
2943 	for (i = 0; i < insn_cnt; i++, insn++) {
2944 		if (BPF_CLASS(insn->code) == BPF_LDX &&
2945 		    (BPF_MODE(insn->code) != BPF_MEM || insn->imm != 0)) {
2946 			verbose("BPF_LDX uses reserved fields\n");
2947 			return -EINVAL;
2948 		}
2949 
2950 		if (BPF_CLASS(insn->code) == BPF_STX &&
2951 		    ((BPF_MODE(insn->code) != BPF_MEM &&
2952 		      BPF_MODE(insn->code) != BPF_XADD) || insn->imm != 0)) {
2953 			verbose("BPF_STX uses reserved fields\n");
2954 			return -EINVAL;
2955 		}
2956 
2957 		if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW)) {
2958 			struct bpf_map *map;
2959 			struct fd f;
2960 
2961 			if (i == insn_cnt - 1 || insn[1].code != 0 ||
2962 			    insn[1].dst_reg != 0 || insn[1].src_reg != 0 ||
2963 			    insn[1].off != 0) {
2964 				verbose("invalid bpf_ld_imm64 insn\n");
2965 				return -EINVAL;
2966 			}
2967 
2968 			if (insn->src_reg == 0)
2969 				/* valid generic load 64-bit imm */
2970 				goto next_insn;
2971 
2972 			if (insn->src_reg != BPF_PSEUDO_MAP_FD) {
2973 				verbose("unrecognized bpf_ld_imm64 insn\n");
2974 				return -EINVAL;
2975 			}
2976 
2977 			f = fdget(insn->imm);
2978 			map = __bpf_map_get(f);
2979 			if (IS_ERR(map)) {
2980 				verbose("fd %d is not pointing to valid bpf_map\n",
2981 					insn->imm);
2982 				return PTR_ERR(map);
2983 			}
2984 
2985 			err = check_map_prog_compatibility(map, env->prog);
2986 			if (err) {
2987 				fdput(f);
2988 				return err;
2989 			}
2990 
2991 			/* store map pointer inside BPF_LD_IMM64 instruction */
2992 			insn[0].imm = (u32) (unsigned long) map;
2993 			insn[1].imm = ((u64) (unsigned long) map) >> 32;
2994 
2995 			/* check whether we recorded this map already */
2996 			for (j = 0; j < env->used_map_cnt; j++)
2997 				if (env->used_maps[j] == map) {
2998 					fdput(f);
2999 					goto next_insn;
3000 				}
3001 
3002 			if (env->used_map_cnt >= MAX_USED_MAPS) {
3003 				fdput(f);
3004 				return -E2BIG;
3005 			}
3006 
3007 			/* hold the map. If the program is rejected by verifier,
3008 			 * the map will be released by release_maps() or it
3009 			 * will be used by the valid program until it's unloaded
3010 			 * and all maps are released in free_bpf_prog_info()
3011 			 */
3012 			map = bpf_map_inc(map, false);
3013 			if (IS_ERR(map)) {
3014 				fdput(f);
3015 				return PTR_ERR(map);
3016 			}
3017 			env->used_maps[env->used_map_cnt++] = map;
3018 
3019 			fdput(f);
3020 next_insn:
3021 			insn++;
3022 			i++;
3023 		}
3024 	}
3025 
3026 	/* now all pseudo BPF_LD_IMM64 instructions load valid
3027 	 * 'struct bpf_map *' into a register instead of user map_fd.
3028 	 * These pointers will be used later by verifier to validate map access.
3029 	 */
3030 	return 0;
3031 }
3032 
3033 /* drop refcnt of maps used by the rejected program */
3034 static void release_maps(struct bpf_verifier_env *env)
3035 {
3036 	int i;
3037 
3038 	for (i = 0; i < env->used_map_cnt; i++)
3039 		bpf_map_put(env->used_maps[i]);
3040 }
3041 
3042 /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */
3043 static void convert_pseudo_ld_imm64(struct bpf_verifier_env *env)
3044 {
3045 	struct bpf_insn *insn = env->prog->insnsi;
3046 	int insn_cnt = env->prog->len;
3047 	int i;
3048 
3049 	for (i = 0; i < insn_cnt; i++, insn++)
3050 		if (insn->code == (BPF_LD | BPF_IMM | BPF_DW))
3051 			insn->src_reg = 0;
3052 }
3053 
3054 /* convert load instructions that access fields of 'struct __sk_buff'
3055  * into sequence of instructions that access fields of 'struct sk_buff'
3056  */
3057 static int convert_ctx_accesses(struct bpf_verifier_env *env)
3058 {
3059 	const struct bpf_verifier_ops *ops = env->prog->aux->ops;
3060 	const int insn_cnt = env->prog->len;
3061 	struct bpf_insn insn_buf[16], *insn;
3062 	struct bpf_prog *new_prog;
3063 	enum bpf_access_type type;
3064 	int i, cnt, delta = 0;
3065 
3066 	if (ops->gen_prologue) {
3067 		cnt = ops->gen_prologue(insn_buf, env->seen_direct_write,
3068 					env->prog);
3069 		if (cnt >= ARRAY_SIZE(insn_buf)) {
3070 			verbose("bpf verifier is misconfigured\n");
3071 			return -EINVAL;
3072 		} else if (cnt) {
3073 			new_prog = bpf_patch_insn_single(env->prog, 0,
3074 							 insn_buf, cnt);
3075 			if (!new_prog)
3076 				return -ENOMEM;
3077 			env->prog = new_prog;
3078 			delta += cnt - 1;
3079 		}
3080 	}
3081 
3082 	if (!ops->convert_ctx_access)
3083 		return 0;
3084 
3085 	insn = env->prog->insnsi + delta;
3086 
3087 	for (i = 0; i < insn_cnt; i++, insn++) {
3088 		if (insn->code == (BPF_LDX | BPF_MEM | BPF_W) ||
3089 		    insn->code == (BPF_LDX | BPF_MEM | BPF_DW))
3090 			type = BPF_READ;
3091 		else if (insn->code == (BPF_STX | BPF_MEM | BPF_W) ||
3092 			 insn->code == (BPF_STX | BPF_MEM | BPF_DW))
3093 			type = BPF_WRITE;
3094 		else
3095 			continue;
3096 
3097 		if (env->insn_aux_data[i].ptr_type != PTR_TO_CTX)
3098 			continue;
3099 
3100 		cnt = ops->convert_ctx_access(type, insn->dst_reg, insn->src_reg,
3101 					      insn->off, insn_buf, env->prog);
3102 		if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) {
3103 			verbose("bpf verifier is misconfigured\n");
3104 			return -EINVAL;
3105 		}
3106 
3107 		new_prog = bpf_patch_insn_single(env->prog, i + delta, insn_buf,
3108 						 cnt);
3109 		if (!new_prog)
3110 			return -ENOMEM;
3111 
3112 		delta += cnt - 1;
3113 
3114 		/* keep walking new program and skip insns we just inserted */
3115 		env->prog = new_prog;
3116 		insn      = new_prog->insnsi + i + delta;
3117 	}
3118 
3119 	return 0;
3120 }
3121 
3122 static void free_states(struct bpf_verifier_env *env)
3123 {
3124 	struct bpf_verifier_state_list *sl, *sln;
3125 	int i;
3126 
3127 	if (!env->explored_states)
3128 		return;
3129 
3130 	for (i = 0; i < env->prog->len; i++) {
3131 		sl = env->explored_states[i];
3132 
3133 		if (sl)
3134 			while (sl != STATE_LIST_MARK) {
3135 				sln = sl->next;
3136 				kfree(sl);
3137 				sl = sln;
3138 			}
3139 	}
3140 
3141 	kfree(env->explored_states);
3142 }
3143 
3144 int bpf_check(struct bpf_prog **prog, union bpf_attr *attr)
3145 {
3146 	char __user *log_ubuf = NULL;
3147 	struct bpf_verifier_env *env;
3148 	int ret = -EINVAL;
3149 
3150 	/* 'struct bpf_verifier_env' can be global, but since it's not small,
3151 	 * allocate/free it every time bpf_check() is called
3152 	 */
3153 	env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL);
3154 	if (!env)
3155 		return -ENOMEM;
3156 
3157 	env->insn_aux_data = vzalloc(sizeof(struct bpf_insn_aux_data) *
3158 				     (*prog)->len);
3159 	ret = -ENOMEM;
3160 	if (!env->insn_aux_data)
3161 		goto err_free_env;
3162 	env->prog = *prog;
3163 
3164 	/* grab the mutex to protect few globals used by verifier */
3165 	mutex_lock(&bpf_verifier_lock);
3166 
3167 	if (attr->log_level || attr->log_buf || attr->log_size) {
3168 		/* user requested verbose verifier output
3169 		 * and supplied buffer to store the verification trace
3170 		 */
3171 		log_level = attr->log_level;
3172 		log_ubuf = (char __user *) (unsigned long) attr->log_buf;
3173 		log_size = attr->log_size;
3174 		log_len = 0;
3175 
3176 		ret = -EINVAL;
3177 		/* log_* values have to be sane */
3178 		if (log_size < 128 || log_size > UINT_MAX >> 8 ||
3179 		    log_level == 0 || log_ubuf == NULL)
3180 			goto err_unlock;
3181 
3182 		ret = -ENOMEM;
3183 		log_buf = vmalloc(log_size);
3184 		if (!log_buf)
3185 			goto err_unlock;
3186 	} else {
3187 		log_level = 0;
3188 	}
3189 
3190 	ret = replace_map_fd_with_map_ptr(env);
3191 	if (ret < 0)
3192 		goto skip_full_check;
3193 
3194 	env->explored_states = kcalloc(env->prog->len,
3195 				       sizeof(struct bpf_verifier_state_list *),
3196 				       GFP_USER);
3197 	ret = -ENOMEM;
3198 	if (!env->explored_states)
3199 		goto skip_full_check;
3200 
3201 	ret = check_cfg(env);
3202 	if (ret < 0)
3203 		goto skip_full_check;
3204 
3205 	env->allow_ptr_leaks = capable(CAP_SYS_ADMIN);
3206 
3207 	ret = do_check(env);
3208 
3209 skip_full_check:
3210 	while (pop_stack(env, NULL) >= 0);
3211 	free_states(env);
3212 
3213 	if (ret == 0)
3214 		/* program is valid, convert *(u32*)(ctx + off) accesses */
3215 		ret = convert_ctx_accesses(env);
3216 
3217 	if (log_level && log_len >= log_size - 1) {
3218 		BUG_ON(log_len >= log_size);
3219 		/* verifier log exceeded user supplied buffer */
3220 		ret = -ENOSPC;
3221 		/* fall through to return what was recorded */
3222 	}
3223 
3224 	/* copy verifier log back to user space including trailing zero */
3225 	if (log_level && copy_to_user(log_ubuf, log_buf, log_len + 1) != 0) {
3226 		ret = -EFAULT;
3227 		goto free_log_buf;
3228 	}
3229 
3230 	if (ret == 0 && env->used_map_cnt) {
3231 		/* if program passed verifier, update used_maps in bpf_prog_info */
3232 		env->prog->aux->used_maps = kmalloc_array(env->used_map_cnt,
3233 							  sizeof(env->used_maps[0]),
3234 							  GFP_KERNEL);
3235 
3236 		if (!env->prog->aux->used_maps) {
3237 			ret = -ENOMEM;
3238 			goto free_log_buf;
3239 		}
3240 
3241 		memcpy(env->prog->aux->used_maps, env->used_maps,
3242 		       sizeof(env->used_maps[0]) * env->used_map_cnt);
3243 		env->prog->aux->used_map_cnt = env->used_map_cnt;
3244 
3245 		/* program is valid. Convert pseudo bpf_ld_imm64 into generic
3246 		 * bpf_ld_imm64 instructions
3247 		 */
3248 		convert_pseudo_ld_imm64(env);
3249 	}
3250 
3251 free_log_buf:
3252 	if (log_level)
3253 		vfree(log_buf);
3254 	if (!env->prog->aux->used_maps)
3255 		/* if we didn't copy map pointers into bpf_prog_info, release
3256 		 * them now. Otherwise free_bpf_prog_info() will release them.
3257 		 */
3258 		release_maps(env);
3259 	*prog = env->prog;
3260 err_unlock:
3261 	mutex_unlock(&bpf_verifier_lock);
3262 	vfree(env->insn_aux_data);
3263 err_free_env:
3264 	kfree(env);
3265 	return ret;
3266 }
3267 
3268 int bpf_analyzer(struct bpf_prog *prog, const struct bpf_ext_analyzer_ops *ops,
3269 		 void *priv)
3270 {
3271 	struct bpf_verifier_env *env;
3272 	int ret;
3273 
3274 	env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL);
3275 	if (!env)
3276 		return -ENOMEM;
3277 
3278 	env->insn_aux_data = vzalloc(sizeof(struct bpf_insn_aux_data) *
3279 				     prog->len);
3280 	ret = -ENOMEM;
3281 	if (!env->insn_aux_data)
3282 		goto err_free_env;
3283 	env->prog = prog;
3284 	env->analyzer_ops = ops;
3285 	env->analyzer_priv = priv;
3286 
3287 	/* grab the mutex to protect few globals used by verifier */
3288 	mutex_lock(&bpf_verifier_lock);
3289 
3290 	log_level = 0;
3291 
3292 	env->explored_states = kcalloc(env->prog->len,
3293 				       sizeof(struct bpf_verifier_state_list *),
3294 				       GFP_KERNEL);
3295 	ret = -ENOMEM;
3296 	if (!env->explored_states)
3297 		goto skip_full_check;
3298 
3299 	ret = check_cfg(env);
3300 	if (ret < 0)
3301 		goto skip_full_check;
3302 
3303 	env->allow_ptr_leaks = capable(CAP_SYS_ADMIN);
3304 
3305 	ret = do_check(env);
3306 
3307 skip_full_check:
3308 	while (pop_stack(env, NULL) >= 0);
3309 	free_states(env);
3310 
3311 	mutex_unlock(&bpf_verifier_lock);
3312 	vfree(env->insn_aux_data);
3313 err_free_env:
3314 	kfree(env);
3315 	return ret;
3316 }
3317 EXPORT_SYMBOL_GPL(bpf_analyzer);
3318