xref: /linux/kernel/bpf/verifier.c (revision e5c86679d5e864947a52fb31e45a425dea3e7fa9)
1 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
2  * Copyright (c) 2016 Facebook
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of version 2 of the GNU General Public
6  * License as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful, but
9  * WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11  * General Public License for more details.
12  */
13 #include <linux/kernel.h>
14 #include <linux/types.h>
15 #include <linux/slab.h>
16 #include <linux/bpf.h>
17 #include <linux/bpf_verifier.h>
18 #include <linux/filter.h>
19 #include <net/netlink.h>
20 #include <linux/file.h>
21 #include <linux/vmalloc.h>
22 #include <linux/stringify.h>
23 
24 /* bpf_check() is a static code analyzer that walks eBPF program
25  * instruction by instruction and updates register/stack state.
26  * All paths of conditional branches are analyzed until 'bpf_exit' insn.
27  *
28  * The first pass is depth-first-search to check that the program is a DAG.
29  * It rejects the following programs:
30  * - larger than BPF_MAXINSNS insns
31  * - if loop is present (detected via back-edge)
32  * - unreachable insns exist (shouldn't be a forest. program = one function)
33  * - out of bounds or malformed jumps
34  * The second pass is all possible path descent from the 1st insn.
35  * Since it's analyzing all pathes through the program, the length of the
36  * analysis is limited to 64k insn, which may be hit even if total number of
37  * insn is less then 4K, but there are too many branches that change stack/regs.
38  * Number of 'branches to be analyzed' is limited to 1k
39  *
40  * On entry to each instruction, each register has a type, and the instruction
41  * changes the types of the registers depending on instruction semantics.
42  * If instruction is BPF_MOV64_REG(BPF_REG_1, BPF_REG_5), then type of R5 is
43  * copied to R1.
44  *
45  * All registers are 64-bit.
46  * R0 - return register
47  * R1-R5 argument passing registers
48  * R6-R9 callee saved registers
49  * R10 - frame pointer read-only
50  *
51  * At the start of BPF program the register R1 contains a pointer to bpf_context
52  * and has type PTR_TO_CTX.
53  *
54  * Verifier tracks arithmetic operations on pointers in case:
55  *    BPF_MOV64_REG(BPF_REG_1, BPF_REG_10),
56  *    BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -20),
57  * 1st insn copies R10 (which has FRAME_PTR) type into R1
58  * and 2nd arithmetic instruction is pattern matched to recognize
59  * that it wants to construct a pointer to some element within stack.
60  * So after 2nd insn, the register R1 has type PTR_TO_STACK
61  * (and -20 constant is saved for further stack bounds checking).
62  * Meaning that this reg is a pointer to stack plus known immediate constant.
63  *
64  * Most of the time the registers have UNKNOWN_VALUE type, which
65  * means the register has some value, but it's not a valid pointer.
66  * (like pointer plus pointer becomes UNKNOWN_VALUE type)
67  *
68  * When verifier sees load or store instructions the type of base register
69  * can be: PTR_TO_MAP_VALUE, PTR_TO_CTX, FRAME_PTR. These are three pointer
70  * types recognized by check_mem_access() function.
71  *
72  * PTR_TO_MAP_VALUE means that this register is pointing to 'map element value'
73  * and the range of [ptr, ptr + map's value_size) is accessible.
74  *
75  * registers used to pass values to function calls are checked against
76  * function argument constraints.
77  *
78  * ARG_PTR_TO_MAP_KEY is one of such argument constraints.
79  * It means that the register type passed to this function must be
80  * PTR_TO_STACK and it will be used inside the function as
81  * 'pointer to map element key'
82  *
83  * For example the argument constraints for bpf_map_lookup_elem():
84  *   .ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL,
85  *   .arg1_type = ARG_CONST_MAP_PTR,
86  *   .arg2_type = ARG_PTR_TO_MAP_KEY,
87  *
88  * ret_type says that this function returns 'pointer to map elem value or null'
89  * function expects 1st argument to be a const pointer to 'struct bpf_map' and
90  * 2nd argument should be a pointer to stack, which will be used inside
91  * the helper function as a pointer to map element key.
92  *
93  * On the kernel side the helper function looks like:
94  * u64 bpf_map_lookup_elem(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
95  * {
96  *    struct bpf_map *map = (struct bpf_map *) (unsigned long) r1;
97  *    void *key = (void *) (unsigned long) r2;
98  *    void *value;
99  *
100  *    here kernel can access 'key' and 'map' pointers safely, knowing that
101  *    [key, key + map->key_size) bytes are valid and were initialized on
102  *    the stack of eBPF program.
103  * }
104  *
105  * Corresponding eBPF program may look like:
106  *    BPF_MOV64_REG(BPF_REG_2, BPF_REG_10),  // after this insn R2 type is FRAME_PTR
107  *    BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -4), // after this insn R2 type is PTR_TO_STACK
108  *    BPF_LD_MAP_FD(BPF_REG_1, map_fd),      // after this insn R1 type is CONST_PTR_TO_MAP
109  *    BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem),
110  * here verifier looks at prototype of map_lookup_elem() and sees:
111  * .arg1_type == ARG_CONST_MAP_PTR and R1->type == CONST_PTR_TO_MAP, which is ok,
112  * Now verifier knows that this map has key of R1->map_ptr->key_size bytes
113  *
114  * Then .arg2_type == ARG_PTR_TO_MAP_KEY and R2->type == PTR_TO_STACK, ok so far,
115  * Now verifier checks that [R2, R2 + map's key_size) are within stack limits
116  * and were initialized prior to this call.
117  * If it's ok, then verifier allows this BPF_CALL insn and looks at
118  * .ret_type which is RET_PTR_TO_MAP_VALUE_OR_NULL, so it sets
119  * R0->type = PTR_TO_MAP_VALUE_OR_NULL which means bpf_map_lookup_elem() function
120  * returns ether pointer to map value or NULL.
121  *
122  * When type PTR_TO_MAP_VALUE_OR_NULL passes through 'if (reg != 0) goto +off'
123  * insn, the register holding that pointer in the true branch changes state to
124  * PTR_TO_MAP_VALUE and the same register changes state to CONST_IMM in the false
125  * branch. See check_cond_jmp_op().
126  *
127  * After the call R0 is set to return type of the function and registers R1-R5
128  * are set to NOT_INIT to indicate that they are no longer readable.
129  */
130 
131 /* verifier_state + insn_idx are pushed to stack when branch is encountered */
132 struct bpf_verifier_stack_elem {
133 	/* verifer state is 'st'
134 	 * before processing instruction 'insn_idx'
135 	 * and after processing instruction 'prev_insn_idx'
136 	 */
137 	struct bpf_verifier_state st;
138 	int insn_idx;
139 	int prev_insn_idx;
140 	struct bpf_verifier_stack_elem *next;
141 };
142 
143 #define BPF_COMPLEXITY_LIMIT_INSNS	65536
144 #define BPF_COMPLEXITY_LIMIT_STACK	1024
145 
146 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 static void mark_reg_unknown_value_and_range(struct bpf_reg_state *regs,
485 					     u32 regno)
486 {
487 	mark_reg_unknown_value(regs, regno);
488 	reset_reg_range_values(regs, regno);
489 }
490 
491 enum reg_arg_type {
492 	SRC_OP,		/* register is used as source operand */
493 	DST_OP,		/* register is used as destination operand */
494 	DST_OP_NO_MARK	/* same as above, check only, don't mark */
495 };
496 
497 static int check_reg_arg(struct bpf_reg_state *regs, u32 regno,
498 			 enum reg_arg_type t)
499 {
500 	if (regno >= MAX_BPF_REG) {
501 		verbose("R%d is invalid\n", regno);
502 		return -EINVAL;
503 	}
504 
505 	if (t == SRC_OP) {
506 		/* check whether register used as source operand can be read */
507 		if (regs[regno].type == NOT_INIT) {
508 			verbose("R%d !read_ok\n", regno);
509 			return -EACCES;
510 		}
511 	} else {
512 		/* check whether register used as dest operand can be written to */
513 		if (regno == BPF_REG_FP) {
514 			verbose("frame pointer is read only\n");
515 			return -EACCES;
516 		}
517 		if (t == DST_OP)
518 			mark_reg_unknown_value(regs, regno);
519 	}
520 	return 0;
521 }
522 
523 static int bpf_size_to_bytes(int bpf_size)
524 {
525 	if (bpf_size == BPF_W)
526 		return 4;
527 	else if (bpf_size == BPF_H)
528 		return 2;
529 	else if (bpf_size == BPF_B)
530 		return 1;
531 	else if (bpf_size == BPF_DW)
532 		return 8;
533 	else
534 		return -EINVAL;
535 }
536 
537 static bool is_spillable_regtype(enum bpf_reg_type type)
538 {
539 	switch (type) {
540 	case PTR_TO_MAP_VALUE:
541 	case PTR_TO_MAP_VALUE_OR_NULL:
542 	case PTR_TO_MAP_VALUE_ADJ:
543 	case PTR_TO_STACK:
544 	case PTR_TO_CTX:
545 	case PTR_TO_PACKET:
546 	case PTR_TO_PACKET_END:
547 	case FRAME_PTR:
548 	case CONST_PTR_TO_MAP:
549 		return true;
550 	default:
551 		return false;
552 	}
553 }
554 
555 /* check_stack_read/write functions track spill/fill of registers,
556  * stack boundary and alignment are checked in check_mem_access()
557  */
558 static int check_stack_write(struct bpf_verifier_state *state, int off,
559 			     int size, int value_regno)
560 {
561 	int i;
562 	/* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0,
563 	 * so it's aligned access and [off, off + size) are within stack limits
564 	 */
565 
566 	if (value_regno >= 0 &&
567 	    is_spillable_regtype(state->regs[value_regno].type)) {
568 
569 		/* register containing pointer is being spilled into stack */
570 		if (size != BPF_REG_SIZE) {
571 			verbose("invalid size of register spill\n");
572 			return -EACCES;
573 		}
574 
575 		/* save register state */
576 		state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] =
577 			state->regs[value_regno];
578 
579 		for (i = 0; i < BPF_REG_SIZE; i++)
580 			state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_SPILL;
581 	} else {
582 		/* regular write of data into stack */
583 		state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] =
584 			(struct bpf_reg_state) {};
585 
586 		for (i = 0; i < size; i++)
587 			state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_MISC;
588 	}
589 	return 0;
590 }
591 
592 static int check_stack_read(struct bpf_verifier_state *state, int off, int size,
593 			    int value_regno)
594 {
595 	u8 *slot_type;
596 	int i;
597 
598 	slot_type = &state->stack_slot_type[MAX_BPF_STACK + off];
599 
600 	if (slot_type[0] == STACK_SPILL) {
601 		if (size != BPF_REG_SIZE) {
602 			verbose("invalid size of register spill\n");
603 			return -EACCES;
604 		}
605 		for (i = 1; i < BPF_REG_SIZE; i++) {
606 			if (slot_type[i] != STACK_SPILL) {
607 				verbose("corrupted spill memory\n");
608 				return -EACCES;
609 			}
610 		}
611 
612 		if (value_regno >= 0)
613 			/* restore register state from stack */
614 			state->regs[value_regno] =
615 				state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE];
616 		return 0;
617 	} else {
618 		for (i = 0; i < size; i++) {
619 			if (slot_type[i] != STACK_MISC) {
620 				verbose("invalid read from stack off %d+%d size %d\n",
621 					off, i, size);
622 				return -EACCES;
623 			}
624 		}
625 		if (value_regno >= 0)
626 			/* have read misc data from the stack */
627 			mark_reg_unknown_value_and_range(state->regs,
628 							 value_regno);
629 		return 0;
630 	}
631 }
632 
633 /* check read/write into map element returned by bpf_map_lookup_elem() */
634 static int check_map_access(struct bpf_verifier_env *env, u32 regno, int off,
635 			    int size)
636 {
637 	struct bpf_map *map = env->cur_state.regs[regno].map_ptr;
638 
639 	if (off < 0 || size <= 0 || off + size > map->value_size) {
640 		verbose("invalid access to map value, value_size=%d off=%d size=%d\n",
641 			map->value_size, off, size);
642 		return -EACCES;
643 	}
644 	return 0;
645 }
646 
647 /* check read/write into an adjusted map element */
648 static int check_map_access_adj(struct bpf_verifier_env *env, u32 regno,
649 				int off, int size)
650 {
651 	struct bpf_verifier_state *state = &env->cur_state;
652 	struct bpf_reg_state *reg = &state->regs[regno];
653 	int err;
654 
655 	/* We adjusted the register to this map value, so we
656 	 * need to change off and size to min_value and max_value
657 	 * respectively to make sure our theoretical access will be
658 	 * safe.
659 	 */
660 	if (log_level)
661 		print_verifier_state(state);
662 	env->varlen_map_value_access = true;
663 	/* The minimum value is only important with signed
664 	 * comparisons where we can't assume the floor of a
665 	 * value is 0.  If we are using signed variables for our
666 	 * index'es we need to make sure that whatever we use
667 	 * will have a set floor within our range.
668 	 */
669 	if (reg->min_value < 0) {
670 		verbose("R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n",
671 			regno);
672 		return -EACCES;
673 	}
674 	err = check_map_access(env, regno, reg->min_value + off, size);
675 	if (err) {
676 		verbose("R%d min value is outside of the array range\n",
677 			regno);
678 		return err;
679 	}
680 
681 	/* If we haven't set a max value then we need to bail
682 	 * since we can't be sure we won't do bad things.
683 	 */
684 	if (reg->max_value == BPF_REGISTER_MAX_RANGE) {
685 		verbose("R%d unbounded memory access, make sure to bounds check any array access into a map\n",
686 			regno);
687 		return -EACCES;
688 	}
689 	return check_map_access(env, regno, reg->max_value + off, size);
690 }
691 
692 #define MAX_PACKET_OFF 0xffff
693 
694 static bool may_access_direct_pkt_data(struct bpf_verifier_env *env,
695 				       const struct bpf_call_arg_meta *meta,
696 				       enum bpf_access_type t)
697 {
698 	switch (env->prog->type) {
699 	case BPF_PROG_TYPE_LWT_IN:
700 	case BPF_PROG_TYPE_LWT_OUT:
701 		/* dst_input() and dst_output() can't write for now */
702 		if (t == BPF_WRITE)
703 			return false;
704 		/* fallthrough */
705 	case BPF_PROG_TYPE_SCHED_CLS:
706 	case BPF_PROG_TYPE_SCHED_ACT:
707 	case BPF_PROG_TYPE_XDP:
708 	case BPF_PROG_TYPE_LWT_XMIT:
709 		if (meta)
710 			return meta->pkt_access;
711 
712 		env->seen_direct_write = true;
713 		return true;
714 	default:
715 		return false;
716 	}
717 }
718 
719 static int check_packet_access(struct bpf_verifier_env *env, u32 regno, int off,
720 			       int size)
721 {
722 	struct bpf_reg_state *regs = env->cur_state.regs;
723 	struct bpf_reg_state *reg = &regs[regno];
724 
725 	off += reg->off;
726 	if (off < 0 || size <= 0 || off + size > reg->range) {
727 		verbose("invalid access to packet, off=%d size=%d, R%d(id=%d,off=%d,r=%d)\n",
728 			off, size, regno, reg->id, reg->off, reg->range);
729 		return -EACCES;
730 	}
731 	return 0;
732 }
733 
734 /* check access to 'struct bpf_context' fields */
735 static int check_ctx_access(struct bpf_verifier_env *env, int off, int size,
736 			    enum bpf_access_type t, enum bpf_reg_type *reg_type)
737 {
738 	/* for analyzer ctx accesses are already validated and converted */
739 	if (env->analyzer_ops)
740 		return 0;
741 
742 	if (env->prog->aux->ops->is_valid_access &&
743 	    env->prog->aux->ops->is_valid_access(off, size, t, reg_type)) {
744 		/* remember the offset of last byte accessed in ctx */
745 		if (env->prog->aux->max_ctx_offset < off + size)
746 			env->prog->aux->max_ctx_offset = off + size;
747 		return 0;
748 	}
749 
750 	verbose("invalid bpf_context access off=%d size=%d\n", off, size);
751 	return -EACCES;
752 }
753 
754 static bool is_pointer_value(struct bpf_verifier_env *env, int regno)
755 {
756 	if (env->allow_ptr_leaks)
757 		return false;
758 
759 	switch (env->cur_state.regs[regno].type) {
760 	case UNKNOWN_VALUE:
761 	case CONST_IMM:
762 		return false;
763 	default:
764 		return true;
765 	}
766 }
767 
768 static int check_pkt_ptr_alignment(const struct bpf_reg_state *reg,
769 				   int off, int size)
770 {
771 	if (reg->id && size != 1) {
772 		verbose("Unknown alignment. Only byte-sized access allowed in packet access.\n");
773 		return -EACCES;
774 	}
775 
776 	/* skb->data is NET_IP_ALIGN-ed */
777 	if ((NET_IP_ALIGN + reg->off + off) % size != 0) {
778 		verbose("misaligned packet access off %d+%d+%d size %d\n",
779 			NET_IP_ALIGN, reg->off, off, size);
780 		return -EACCES;
781 	}
782 
783 	return 0;
784 }
785 
786 static int check_val_ptr_alignment(const struct bpf_reg_state *reg,
787 				   int size)
788 {
789 	if (size != 1) {
790 		verbose("Unknown alignment. Only byte-sized access allowed in value access.\n");
791 		return -EACCES;
792 	}
793 
794 	return 0;
795 }
796 
797 static int check_ptr_alignment(const struct bpf_reg_state *reg,
798 			       int off, int size)
799 {
800 	switch (reg->type) {
801 	case PTR_TO_PACKET:
802 		return IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) ? 0 :
803 		       check_pkt_ptr_alignment(reg, off, size);
804 	case PTR_TO_MAP_VALUE_ADJ:
805 		return IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) ? 0 :
806 		       check_val_ptr_alignment(reg, size);
807 	default:
808 		if (off % size != 0) {
809 			verbose("misaligned access off %d size %d\n",
810 				off, size);
811 			return -EACCES;
812 		}
813 
814 		return 0;
815 	}
816 }
817 
818 /* check whether memory at (regno + off) is accessible for t = (read | write)
819  * if t==write, value_regno is a register which value is stored into memory
820  * if t==read, value_regno is a register which will receive the value from memory
821  * if t==write && value_regno==-1, some unknown value is stored into memory
822  * if t==read && value_regno==-1, don't care what we read from memory
823  */
824 static int check_mem_access(struct bpf_verifier_env *env, u32 regno, int off,
825 			    int bpf_size, enum bpf_access_type t,
826 			    int value_regno)
827 {
828 	struct bpf_verifier_state *state = &env->cur_state;
829 	struct bpf_reg_state *reg = &state->regs[regno];
830 	int size, err = 0;
831 
832 	if (reg->type == PTR_TO_STACK)
833 		off += reg->imm;
834 
835 	size = bpf_size_to_bytes(bpf_size);
836 	if (size < 0)
837 		return size;
838 
839 	err = check_ptr_alignment(reg, off, size);
840 	if (err)
841 		return err;
842 
843 	if (reg->type == PTR_TO_MAP_VALUE ||
844 	    reg->type == PTR_TO_MAP_VALUE_ADJ) {
845 		if (t == BPF_WRITE && value_regno >= 0 &&
846 		    is_pointer_value(env, value_regno)) {
847 			verbose("R%d leaks addr into map\n", value_regno);
848 			return -EACCES;
849 		}
850 
851 		if (reg->type == PTR_TO_MAP_VALUE_ADJ)
852 			err = check_map_access_adj(env, regno, off, size);
853 		else
854 			err = check_map_access(env, regno, off, size);
855 		if (!err && t == BPF_READ && value_regno >= 0)
856 			mark_reg_unknown_value_and_range(state->regs,
857 							 value_regno);
858 
859 	} else if (reg->type == PTR_TO_CTX) {
860 		enum bpf_reg_type reg_type = UNKNOWN_VALUE;
861 
862 		if (t == BPF_WRITE && value_regno >= 0 &&
863 		    is_pointer_value(env, value_regno)) {
864 			verbose("R%d leaks addr into ctx\n", value_regno);
865 			return -EACCES;
866 		}
867 		err = check_ctx_access(env, off, size, t, &reg_type);
868 		if (!err && t == BPF_READ && value_regno >= 0) {
869 			mark_reg_unknown_value_and_range(state->regs,
870 							 value_regno);
871 			/* note that reg.[id|off|range] == 0 */
872 			state->regs[value_regno].type = reg_type;
873 		}
874 
875 	} else if (reg->type == FRAME_PTR || reg->type == PTR_TO_STACK) {
876 		if (off >= 0 || off < -MAX_BPF_STACK) {
877 			verbose("invalid stack off=%d size=%d\n", off, size);
878 			return -EACCES;
879 		}
880 		if (t == BPF_WRITE) {
881 			if (!env->allow_ptr_leaks &&
882 			    state->stack_slot_type[MAX_BPF_STACK + off] == STACK_SPILL &&
883 			    size != BPF_REG_SIZE) {
884 				verbose("attempt to corrupt spilled pointer on stack\n");
885 				return -EACCES;
886 			}
887 			err = check_stack_write(state, off, size, value_regno);
888 		} else {
889 			err = check_stack_read(state, off, size, value_regno);
890 		}
891 	} else if (state->regs[regno].type == PTR_TO_PACKET) {
892 		if (t == BPF_WRITE && !may_access_direct_pkt_data(env, NULL, t)) {
893 			verbose("cannot write into packet\n");
894 			return -EACCES;
895 		}
896 		if (t == BPF_WRITE && value_regno >= 0 &&
897 		    is_pointer_value(env, value_regno)) {
898 			verbose("R%d leaks addr into packet\n", value_regno);
899 			return -EACCES;
900 		}
901 		err = check_packet_access(env, regno, off, size);
902 		if (!err && t == BPF_READ && value_regno >= 0)
903 			mark_reg_unknown_value_and_range(state->regs,
904 							 value_regno);
905 	} else {
906 		verbose("R%d invalid mem access '%s'\n",
907 			regno, reg_type_str[reg->type]);
908 		return -EACCES;
909 	}
910 
911 	if (!err && size <= 2 && value_regno >= 0 && env->allow_ptr_leaks &&
912 	    state->regs[value_regno].type == UNKNOWN_VALUE) {
913 		/* 1 or 2 byte load zero-extends, determine the number of
914 		 * zero upper bits. Not doing it fo 4 byte load, since
915 		 * such values cannot be added to ptr_to_packet anyway.
916 		 */
917 		state->regs[value_regno].imm = 64 - size * 8;
918 	}
919 	return err;
920 }
921 
922 static int check_xadd(struct bpf_verifier_env *env, struct bpf_insn *insn)
923 {
924 	struct bpf_reg_state *regs = env->cur_state.regs;
925 	int err;
926 
927 	if ((BPF_SIZE(insn->code) != BPF_W && BPF_SIZE(insn->code) != BPF_DW) ||
928 	    insn->imm != 0) {
929 		verbose("BPF_XADD uses reserved fields\n");
930 		return -EINVAL;
931 	}
932 
933 	/* check src1 operand */
934 	err = check_reg_arg(regs, insn->src_reg, SRC_OP);
935 	if (err)
936 		return err;
937 
938 	/* check src2 operand */
939 	err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
940 	if (err)
941 		return err;
942 
943 	/* check whether atomic_add can read the memory */
944 	err = check_mem_access(env, insn->dst_reg, insn->off,
945 			       BPF_SIZE(insn->code), BPF_READ, -1);
946 	if (err)
947 		return err;
948 
949 	/* check whether atomic_add can write into the same memory */
950 	return check_mem_access(env, insn->dst_reg, insn->off,
951 				BPF_SIZE(insn->code), BPF_WRITE, -1);
952 }
953 
954 /* when register 'regno' is passed into function that will read 'access_size'
955  * bytes from that pointer, make sure that it's within stack boundary
956  * and all elements of stack are initialized
957  */
958 static int check_stack_boundary(struct bpf_verifier_env *env, int regno,
959 				int access_size, bool zero_size_allowed,
960 				struct bpf_call_arg_meta *meta)
961 {
962 	struct bpf_verifier_state *state = &env->cur_state;
963 	struct bpf_reg_state *regs = state->regs;
964 	int off, i;
965 
966 	if (regs[regno].type != PTR_TO_STACK) {
967 		if (zero_size_allowed && access_size == 0 &&
968 		    regs[regno].type == CONST_IMM &&
969 		    regs[regno].imm  == 0)
970 			return 0;
971 
972 		verbose("R%d type=%s expected=%s\n", regno,
973 			reg_type_str[regs[regno].type],
974 			reg_type_str[PTR_TO_STACK]);
975 		return -EACCES;
976 	}
977 
978 	off = regs[regno].imm;
979 	if (off >= 0 || off < -MAX_BPF_STACK || off + access_size > 0 ||
980 	    access_size <= 0) {
981 		verbose("invalid stack type R%d off=%d access_size=%d\n",
982 			regno, off, access_size);
983 		return -EACCES;
984 	}
985 
986 	if (meta && meta->raw_mode) {
987 		meta->access_size = access_size;
988 		meta->regno = regno;
989 		return 0;
990 	}
991 
992 	for (i = 0; i < access_size; i++) {
993 		if (state->stack_slot_type[MAX_BPF_STACK + off + i] != STACK_MISC) {
994 			verbose("invalid indirect read from stack off %d+%d size %d\n",
995 				off, i, access_size);
996 			return -EACCES;
997 		}
998 	}
999 	return 0;
1000 }
1001 
1002 static int check_helper_mem_access(struct bpf_verifier_env *env, int regno,
1003 				   int access_size, bool zero_size_allowed,
1004 				   struct bpf_call_arg_meta *meta)
1005 {
1006 	struct bpf_reg_state *regs = env->cur_state.regs;
1007 
1008 	switch (regs[regno].type) {
1009 	case PTR_TO_PACKET:
1010 		return check_packet_access(env, regno, 0, access_size);
1011 	case PTR_TO_MAP_VALUE:
1012 		return check_map_access(env, regno, 0, access_size);
1013 	case PTR_TO_MAP_VALUE_ADJ:
1014 		return check_map_access_adj(env, regno, 0, access_size);
1015 	default: /* const_imm|ptr_to_stack or invalid ptr */
1016 		return check_stack_boundary(env, regno, access_size,
1017 					    zero_size_allowed, meta);
1018 	}
1019 }
1020 
1021 static int check_func_arg(struct bpf_verifier_env *env, u32 regno,
1022 			  enum bpf_arg_type arg_type,
1023 			  struct bpf_call_arg_meta *meta)
1024 {
1025 	struct bpf_reg_state *regs = env->cur_state.regs, *reg = &regs[regno];
1026 	enum bpf_reg_type expected_type, type = reg->type;
1027 	int err = 0;
1028 
1029 	if (arg_type == ARG_DONTCARE)
1030 		return 0;
1031 
1032 	if (type == NOT_INIT) {
1033 		verbose("R%d !read_ok\n", regno);
1034 		return -EACCES;
1035 	}
1036 
1037 	if (arg_type == ARG_ANYTHING) {
1038 		if (is_pointer_value(env, regno)) {
1039 			verbose("R%d leaks addr into helper function\n", regno);
1040 			return -EACCES;
1041 		}
1042 		return 0;
1043 	}
1044 
1045 	if (type == PTR_TO_PACKET &&
1046 	    !may_access_direct_pkt_data(env, meta, BPF_READ)) {
1047 		verbose("helper access to the packet is not allowed\n");
1048 		return -EACCES;
1049 	}
1050 
1051 	if (arg_type == ARG_PTR_TO_MAP_KEY ||
1052 	    arg_type == ARG_PTR_TO_MAP_VALUE) {
1053 		expected_type = PTR_TO_STACK;
1054 		if (type != PTR_TO_PACKET && type != expected_type)
1055 			goto err_type;
1056 	} else if (arg_type == ARG_CONST_SIZE ||
1057 		   arg_type == ARG_CONST_SIZE_OR_ZERO) {
1058 		expected_type = CONST_IMM;
1059 		/* One exception. Allow UNKNOWN_VALUE registers when the
1060 		 * boundaries are known and don't cause unsafe memory accesses
1061 		 */
1062 		if (type != UNKNOWN_VALUE && type != expected_type)
1063 			goto err_type;
1064 	} else if (arg_type == ARG_CONST_MAP_PTR) {
1065 		expected_type = CONST_PTR_TO_MAP;
1066 		if (type != expected_type)
1067 			goto err_type;
1068 	} else if (arg_type == ARG_PTR_TO_CTX) {
1069 		expected_type = PTR_TO_CTX;
1070 		if (type != expected_type)
1071 			goto err_type;
1072 	} else if (arg_type == ARG_PTR_TO_MEM ||
1073 		   arg_type == ARG_PTR_TO_UNINIT_MEM) {
1074 		expected_type = PTR_TO_STACK;
1075 		/* One exception here. In case function allows for NULL to be
1076 		 * passed in as argument, it's a CONST_IMM type. Final test
1077 		 * happens during stack boundary checking.
1078 		 */
1079 		if (type == CONST_IMM && reg->imm == 0)
1080 			/* final test in check_stack_boundary() */;
1081 		else if (type != PTR_TO_PACKET && type != PTR_TO_MAP_VALUE &&
1082 			 type != PTR_TO_MAP_VALUE_ADJ && type != expected_type)
1083 			goto err_type;
1084 		meta->raw_mode = arg_type == ARG_PTR_TO_UNINIT_MEM;
1085 	} else {
1086 		verbose("unsupported arg_type %d\n", arg_type);
1087 		return -EFAULT;
1088 	}
1089 
1090 	if (arg_type == ARG_CONST_MAP_PTR) {
1091 		/* bpf_map_xxx(map_ptr) call: remember that map_ptr */
1092 		meta->map_ptr = reg->map_ptr;
1093 	} else if (arg_type == ARG_PTR_TO_MAP_KEY) {
1094 		/* bpf_map_xxx(..., map_ptr, ..., key) call:
1095 		 * check that [key, key + map->key_size) are within
1096 		 * stack limits and initialized
1097 		 */
1098 		if (!meta->map_ptr) {
1099 			/* in function declaration map_ptr must come before
1100 			 * map_key, so that it's verified and known before
1101 			 * we have to check map_key here. Otherwise it means
1102 			 * that kernel subsystem misconfigured verifier
1103 			 */
1104 			verbose("invalid map_ptr to access map->key\n");
1105 			return -EACCES;
1106 		}
1107 		if (type == PTR_TO_PACKET)
1108 			err = check_packet_access(env, regno, 0,
1109 						  meta->map_ptr->key_size);
1110 		else
1111 			err = check_stack_boundary(env, regno,
1112 						   meta->map_ptr->key_size,
1113 						   false, NULL);
1114 	} else if (arg_type == ARG_PTR_TO_MAP_VALUE) {
1115 		/* bpf_map_xxx(..., map_ptr, ..., value) call:
1116 		 * check [value, value + map->value_size) validity
1117 		 */
1118 		if (!meta->map_ptr) {
1119 			/* kernel subsystem misconfigured verifier */
1120 			verbose("invalid map_ptr to access map->value\n");
1121 			return -EACCES;
1122 		}
1123 		if (type == PTR_TO_PACKET)
1124 			err = check_packet_access(env, regno, 0,
1125 						  meta->map_ptr->value_size);
1126 		else
1127 			err = check_stack_boundary(env, regno,
1128 						   meta->map_ptr->value_size,
1129 						   false, NULL);
1130 	} else if (arg_type == ARG_CONST_SIZE ||
1131 		   arg_type == ARG_CONST_SIZE_OR_ZERO) {
1132 		bool zero_size_allowed = (arg_type == ARG_CONST_SIZE_OR_ZERO);
1133 
1134 		/* bpf_xxx(..., buf, len) call will access 'len' bytes
1135 		 * from stack pointer 'buf'. Check it
1136 		 * note: regno == len, regno - 1 == buf
1137 		 */
1138 		if (regno == 0) {
1139 			/* kernel subsystem misconfigured verifier */
1140 			verbose("ARG_CONST_SIZE cannot be first argument\n");
1141 			return -EACCES;
1142 		}
1143 
1144 		/* If the register is UNKNOWN_VALUE, the access check happens
1145 		 * using its boundaries. Otherwise, just use its imm
1146 		 */
1147 		if (type == UNKNOWN_VALUE) {
1148 			/* For unprivileged variable accesses, disable raw
1149 			 * mode so that the program is required to
1150 			 * initialize all the memory that the helper could
1151 			 * just partially fill up.
1152 			 */
1153 			meta = NULL;
1154 
1155 			if (reg->min_value < 0) {
1156 				verbose("R%d min value is negative, either use unsigned or 'var &= const'\n",
1157 					regno);
1158 				return -EACCES;
1159 			}
1160 
1161 			if (reg->min_value == 0) {
1162 				err = check_helper_mem_access(env, regno - 1, 0,
1163 							      zero_size_allowed,
1164 							      meta);
1165 				if (err)
1166 					return err;
1167 			}
1168 
1169 			if (reg->max_value == BPF_REGISTER_MAX_RANGE) {
1170 				verbose("R%d unbounded memory access, use 'var &= const' or 'if (var < const)'\n",
1171 					regno);
1172 				return -EACCES;
1173 			}
1174 			err = check_helper_mem_access(env, regno - 1,
1175 						      reg->max_value,
1176 						      zero_size_allowed, meta);
1177 			if (err)
1178 				return err;
1179 		} else {
1180 			/* register is CONST_IMM */
1181 			err = check_helper_mem_access(env, regno - 1, reg->imm,
1182 						      zero_size_allowed, meta);
1183 		}
1184 	}
1185 
1186 	return err;
1187 err_type:
1188 	verbose("R%d type=%s expected=%s\n", regno,
1189 		reg_type_str[type], reg_type_str[expected_type]);
1190 	return -EACCES;
1191 }
1192 
1193 static int check_map_func_compatibility(struct bpf_map *map, int func_id)
1194 {
1195 	if (!map)
1196 		return 0;
1197 
1198 	/* We need a two way check, first is from map perspective ... */
1199 	switch (map->map_type) {
1200 	case BPF_MAP_TYPE_PROG_ARRAY:
1201 		if (func_id != BPF_FUNC_tail_call)
1202 			goto error;
1203 		break;
1204 	case BPF_MAP_TYPE_PERF_EVENT_ARRAY:
1205 		if (func_id != BPF_FUNC_perf_event_read &&
1206 		    func_id != BPF_FUNC_perf_event_output)
1207 			goto error;
1208 		break;
1209 	case BPF_MAP_TYPE_STACK_TRACE:
1210 		if (func_id != BPF_FUNC_get_stackid)
1211 			goto error;
1212 		break;
1213 	case BPF_MAP_TYPE_CGROUP_ARRAY:
1214 		if (func_id != BPF_FUNC_skb_under_cgroup &&
1215 		    func_id != BPF_FUNC_current_task_under_cgroup)
1216 			goto error;
1217 		break;
1218 	default:
1219 		break;
1220 	}
1221 
1222 	/* ... and second from the function itself. */
1223 	switch (func_id) {
1224 	case BPF_FUNC_tail_call:
1225 		if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY)
1226 			goto error;
1227 		break;
1228 	case BPF_FUNC_perf_event_read:
1229 	case BPF_FUNC_perf_event_output:
1230 		if (map->map_type != BPF_MAP_TYPE_PERF_EVENT_ARRAY)
1231 			goto error;
1232 		break;
1233 	case BPF_FUNC_get_stackid:
1234 		if (map->map_type != BPF_MAP_TYPE_STACK_TRACE)
1235 			goto error;
1236 		break;
1237 	case BPF_FUNC_current_task_under_cgroup:
1238 	case BPF_FUNC_skb_under_cgroup:
1239 		if (map->map_type != BPF_MAP_TYPE_CGROUP_ARRAY)
1240 			goto error;
1241 		break;
1242 	default:
1243 		break;
1244 	}
1245 
1246 	return 0;
1247 error:
1248 	verbose("cannot pass map_type %d into func %s#%d\n",
1249 		map->map_type, func_id_name(func_id), func_id);
1250 	return -EINVAL;
1251 }
1252 
1253 static int check_raw_mode(const struct bpf_func_proto *fn)
1254 {
1255 	int count = 0;
1256 
1257 	if (fn->arg1_type == ARG_PTR_TO_UNINIT_MEM)
1258 		count++;
1259 	if (fn->arg2_type == ARG_PTR_TO_UNINIT_MEM)
1260 		count++;
1261 	if (fn->arg3_type == ARG_PTR_TO_UNINIT_MEM)
1262 		count++;
1263 	if (fn->arg4_type == ARG_PTR_TO_UNINIT_MEM)
1264 		count++;
1265 	if (fn->arg5_type == ARG_PTR_TO_UNINIT_MEM)
1266 		count++;
1267 
1268 	return count > 1 ? -EINVAL : 0;
1269 }
1270 
1271 static void clear_all_pkt_pointers(struct bpf_verifier_env *env)
1272 {
1273 	struct bpf_verifier_state *state = &env->cur_state;
1274 	struct bpf_reg_state *regs = state->regs, *reg;
1275 	int i;
1276 
1277 	for (i = 0; i < MAX_BPF_REG; i++)
1278 		if (regs[i].type == PTR_TO_PACKET ||
1279 		    regs[i].type == PTR_TO_PACKET_END)
1280 			mark_reg_unknown_value(regs, i);
1281 
1282 	for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
1283 		if (state->stack_slot_type[i] != STACK_SPILL)
1284 			continue;
1285 		reg = &state->spilled_regs[i / BPF_REG_SIZE];
1286 		if (reg->type != PTR_TO_PACKET &&
1287 		    reg->type != PTR_TO_PACKET_END)
1288 			continue;
1289 		reg->type = UNKNOWN_VALUE;
1290 		reg->imm = 0;
1291 	}
1292 }
1293 
1294 static int check_call(struct bpf_verifier_env *env, int func_id)
1295 {
1296 	struct bpf_verifier_state *state = &env->cur_state;
1297 	const struct bpf_func_proto *fn = NULL;
1298 	struct bpf_reg_state *regs = state->regs;
1299 	struct bpf_reg_state *reg;
1300 	struct bpf_call_arg_meta meta;
1301 	bool changes_data;
1302 	int i, err;
1303 
1304 	/* find function prototype */
1305 	if (func_id < 0 || func_id >= __BPF_FUNC_MAX_ID) {
1306 		verbose("invalid func %s#%d\n", func_id_name(func_id), func_id);
1307 		return -EINVAL;
1308 	}
1309 
1310 	if (env->prog->aux->ops->get_func_proto)
1311 		fn = env->prog->aux->ops->get_func_proto(func_id);
1312 
1313 	if (!fn) {
1314 		verbose("unknown func %s#%d\n", func_id_name(func_id), func_id);
1315 		return -EINVAL;
1316 	}
1317 
1318 	/* eBPF programs must be GPL compatible to use GPL-ed functions */
1319 	if (!env->prog->gpl_compatible && fn->gpl_only) {
1320 		verbose("cannot call GPL only function from proprietary program\n");
1321 		return -EINVAL;
1322 	}
1323 
1324 	changes_data = bpf_helper_changes_pkt_data(fn->func);
1325 
1326 	memset(&meta, 0, sizeof(meta));
1327 	meta.pkt_access = fn->pkt_access;
1328 
1329 	/* We only support one arg being in raw mode at the moment, which
1330 	 * is sufficient for the helper functions we have right now.
1331 	 */
1332 	err = check_raw_mode(fn);
1333 	if (err) {
1334 		verbose("kernel subsystem misconfigured func %s#%d\n",
1335 			func_id_name(func_id), func_id);
1336 		return err;
1337 	}
1338 
1339 	/* check args */
1340 	err = check_func_arg(env, BPF_REG_1, fn->arg1_type, &meta);
1341 	if (err)
1342 		return err;
1343 	err = check_func_arg(env, BPF_REG_2, fn->arg2_type, &meta);
1344 	if (err)
1345 		return err;
1346 	err = check_func_arg(env, BPF_REG_3, fn->arg3_type, &meta);
1347 	if (err)
1348 		return err;
1349 	err = check_func_arg(env, BPF_REG_4, fn->arg4_type, &meta);
1350 	if (err)
1351 		return err;
1352 	err = check_func_arg(env, BPF_REG_5, fn->arg5_type, &meta);
1353 	if (err)
1354 		return err;
1355 
1356 	/* Mark slots with STACK_MISC in case of raw mode, stack offset
1357 	 * is inferred from register state.
1358 	 */
1359 	for (i = 0; i < meta.access_size; i++) {
1360 		err = check_mem_access(env, meta.regno, i, BPF_B, BPF_WRITE, -1);
1361 		if (err)
1362 			return err;
1363 	}
1364 
1365 	/* reset caller saved regs */
1366 	for (i = 0; i < CALLER_SAVED_REGS; i++) {
1367 		reg = regs + caller_saved[i];
1368 		reg->type = NOT_INIT;
1369 		reg->imm = 0;
1370 	}
1371 
1372 	/* update return register */
1373 	if (fn->ret_type == RET_INTEGER) {
1374 		regs[BPF_REG_0].type = UNKNOWN_VALUE;
1375 	} else if (fn->ret_type == RET_VOID) {
1376 		regs[BPF_REG_0].type = NOT_INIT;
1377 	} else if (fn->ret_type == RET_PTR_TO_MAP_VALUE_OR_NULL) {
1378 		regs[BPF_REG_0].type = PTR_TO_MAP_VALUE_OR_NULL;
1379 		regs[BPF_REG_0].max_value = regs[BPF_REG_0].min_value = 0;
1380 		/* remember map_ptr, so that check_map_access()
1381 		 * can check 'value_size' boundary of memory access
1382 		 * to map element returned from bpf_map_lookup_elem()
1383 		 */
1384 		if (meta.map_ptr == NULL) {
1385 			verbose("kernel subsystem misconfigured verifier\n");
1386 			return -EINVAL;
1387 		}
1388 		regs[BPF_REG_0].map_ptr = meta.map_ptr;
1389 		regs[BPF_REG_0].id = ++env->id_gen;
1390 	} else {
1391 		verbose("unknown return type %d of func %s#%d\n",
1392 			fn->ret_type, func_id_name(func_id), func_id);
1393 		return -EINVAL;
1394 	}
1395 
1396 	err = check_map_func_compatibility(meta.map_ptr, func_id);
1397 	if (err)
1398 		return err;
1399 
1400 	if (changes_data)
1401 		clear_all_pkt_pointers(env);
1402 	return 0;
1403 }
1404 
1405 static int check_packet_ptr_add(struct bpf_verifier_env *env,
1406 				struct bpf_insn *insn)
1407 {
1408 	struct bpf_reg_state *regs = env->cur_state.regs;
1409 	struct bpf_reg_state *dst_reg = &regs[insn->dst_reg];
1410 	struct bpf_reg_state *src_reg = &regs[insn->src_reg];
1411 	struct bpf_reg_state tmp_reg;
1412 	s32 imm;
1413 
1414 	if (BPF_SRC(insn->code) == BPF_K) {
1415 		/* pkt_ptr += imm */
1416 		imm = insn->imm;
1417 
1418 add_imm:
1419 		if (imm < 0) {
1420 			verbose("addition of negative constant to packet pointer is not allowed\n");
1421 			return -EACCES;
1422 		}
1423 		if (imm >= MAX_PACKET_OFF ||
1424 		    imm + dst_reg->off >= MAX_PACKET_OFF) {
1425 			verbose("constant %d is too large to add to packet pointer\n",
1426 				imm);
1427 			return -EACCES;
1428 		}
1429 		/* a constant was added to pkt_ptr.
1430 		 * Remember it while keeping the same 'id'
1431 		 */
1432 		dst_reg->off += imm;
1433 	} else {
1434 		if (src_reg->type == PTR_TO_PACKET) {
1435 			/* R6=pkt(id=0,off=0,r=62) R7=imm22; r7 += r6 */
1436 			tmp_reg = *dst_reg;  /* save r7 state */
1437 			*dst_reg = *src_reg; /* copy pkt_ptr state r6 into r7 */
1438 			src_reg = &tmp_reg;  /* pretend it's src_reg state */
1439 			/* if the checks below reject it, the copy won't matter,
1440 			 * since we're rejecting the whole program. If all ok,
1441 			 * then imm22 state will be added to r7
1442 			 * and r7 will be pkt(id=0,off=22,r=62) while
1443 			 * r6 will stay as pkt(id=0,off=0,r=62)
1444 			 */
1445 		}
1446 
1447 		if (src_reg->type == CONST_IMM) {
1448 			/* pkt_ptr += reg where reg is known constant */
1449 			imm = src_reg->imm;
1450 			goto add_imm;
1451 		}
1452 		/* disallow pkt_ptr += reg
1453 		 * if reg is not uknown_value with guaranteed zero upper bits
1454 		 * otherwise pkt_ptr may overflow and addition will become
1455 		 * subtraction which is not allowed
1456 		 */
1457 		if (src_reg->type != UNKNOWN_VALUE) {
1458 			verbose("cannot add '%s' to ptr_to_packet\n",
1459 				reg_type_str[src_reg->type]);
1460 			return -EACCES;
1461 		}
1462 		if (src_reg->imm < 48) {
1463 			verbose("cannot add integer value with %lld upper zero bits to ptr_to_packet\n",
1464 				src_reg->imm);
1465 			return -EACCES;
1466 		}
1467 		/* dst_reg stays as pkt_ptr type and since some positive
1468 		 * integer value was added to the pointer, increment its 'id'
1469 		 */
1470 		dst_reg->id = ++env->id_gen;
1471 
1472 		/* something was added to pkt_ptr, set range and off to zero */
1473 		dst_reg->off = 0;
1474 		dst_reg->range = 0;
1475 	}
1476 	return 0;
1477 }
1478 
1479 static int evaluate_reg_alu(struct bpf_verifier_env *env, struct bpf_insn *insn)
1480 {
1481 	struct bpf_reg_state *regs = env->cur_state.regs;
1482 	struct bpf_reg_state *dst_reg = &regs[insn->dst_reg];
1483 	u8 opcode = BPF_OP(insn->code);
1484 	s64 imm_log2;
1485 
1486 	/* for type == UNKNOWN_VALUE:
1487 	 * imm > 0 -> number of zero upper bits
1488 	 * imm == 0 -> don't track which is the same as all bits can be non-zero
1489 	 */
1490 
1491 	if (BPF_SRC(insn->code) == BPF_X) {
1492 		struct bpf_reg_state *src_reg = &regs[insn->src_reg];
1493 
1494 		if (src_reg->type == UNKNOWN_VALUE && src_reg->imm > 0 &&
1495 		    dst_reg->imm && opcode == BPF_ADD) {
1496 			/* dreg += sreg
1497 			 * where both have zero upper bits. Adding them
1498 			 * can only result making one more bit non-zero
1499 			 * in the larger value.
1500 			 * Ex. 0xffff (imm=48) + 1 (imm=63) = 0x10000 (imm=47)
1501 			 *     0xffff (imm=48) + 0xffff = 0x1fffe (imm=47)
1502 			 */
1503 			dst_reg->imm = min(dst_reg->imm, src_reg->imm);
1504 			dst_reg->imm--;
1505 			return 0;
1506 		}
1507 		if (src_reg->type == CONST_IMM && src_reg->imm > 0 &&
1508 		    dst_reg->imm && opcode == BPF_ADD) {
1509 			/* dreg += sreg
1510 			 * where dreg has zero upper bits and sreg is const.
1511 			 * Adding them can only result making one more bit
1512 			 * non-zero in the larger value.
1513 			 */
1514 			imm_log2 = __ilog2_u64((long long)src_reg->imm);
1515 			dst_reg->imm = min(dst_reg->imm, 63 - imm_log2);
1516 			dst_reg->imm--;
1517 			return 0;
1518 		}
1519 		/* all other cases non supported yet, just mark dst_reg */
1520 		dst_reg->imm = 0;
1521 		return 0;
1522 	}
1523 
1524 	/* sign extend 32-bit imm into 64-bit to make sure that
1525 	 * negative values occupy bit 63. Note ilog2() would have
1526 	 * been incorrect, since sizeof(insn->imm) == 4
1527 	 */
1528 	imm_log2 = __ilog2_u64((long long)insn->imm);
1529 
1530 	if (dst_reg->imm && opcode == BPF_LSH) {
1531 		/* reg <<= imm
1532 		 * if reg was a result of 2 byte load, then its imm == 48
1533 		 * which means that upper 48 bits are zero and shifting this reg
1534 		 * left by 4 would mean that upper 44 bits are still zero
1535 		 */
1536 		dst_reg->imm -= insn->imm;
1537 	} else if (dst_reg->imm && opcode == BPF_MUL) {
1538 		/* reg *= imm
1539 		 * if multiplying by 14 subtract 4
1540 		 * This is conservative calculation of upper zero bits.
1541 		 * It's not trying to special case insn->imm == 1 or 0 cases
1542 		 */
1543 		dst_reg->imm -= imm_log2 + 1;
1544 	} else if (opcode == BPF_AND) {
1545 		/* reg &= imm */
1546 		dst_reg->imm = 63 - imm_log2;
1547 	} else if (dst_reg->imm && opcode == BPF_ADD) {
1548 		/* reg += imm */
1549 		dst_reg->imm = min(dst_reg->imm, 63 - imm_log2);
1550 		dst_reg->imm--;
1551 	} else if (opcode == BPF_RSH) {
1552 		/* reg >>= imm
1553 		 * which means that after right shift, upper bits will be zero
1554 		 * note that verifier already checked that
1555 		 * 0 <= imm < 64 for shift insn
1556 		 */
1557 		dst_reg->imm += insn->imm;
1558 		if (unlikely(dst_reg->imm > 64))
1559 			/* some dumb code did:
1560 			 * r2 = *(u32 *)mem;
1561 			 * r2 >>= 32;
1562 			 * and all bits are zero now */
1563 			dst_reg->imm = 64;
1564 	} else {
1565 		/* all other alu ops, means that we don't know what will
1566 		 * happen to the value, mark it with unknown number of zero bits
1567 		 */
1568 		dst_reg->imm = 0;
1569 	}
1570 
1571 	if (dst_reg->imm < 0) {
1572 		/* all 64 bits of the register can contain non-zero bits
1573 		 * and such value cannot be added to ptr_to_packet, since it
1574 		 * may overflow, mark it as unknown to avoid further eval
1575 		 */
1576 		dst_reg->imm = 0;
1577 	}
1578 	return 0;
1579 }
1580 
1581 static int evaluate_reg_imm_alu(struct bpf_verifier_env *env,
1582 				struct bpf_insn *insn)
1583 {
1584 	struct bpf_reg_state *regs = env->cur_state.regs;
1585 	struct bpf_reg_state *dst_reg = &regs[insn->dst_reg];
1586 	struct bpf_reg_state *src_reg = &regs[insn->src_reg];
1587 	u8 opcode = BPF_OP(insn->code);
1588 	u64 dst_imm = dst_reg->imm;
1589 
1590 	/* dst_reg->type == CONST_IMM here. Simulate execution of insns
1591 	 * containing ALU ops. Don't care about overflow or negative
1592 	 * values, just add/sub/... them; registers are in u64.
1593 	 */
1594 	if (opcode == BPF_ADD && BPF_SRC(insn->code) == BPF_K) {
1595 		dst_imm += insn->imm;
1596 	} else if (opcode == BPF_ADD && BPF_SRC(insn->code) == BPF_X &&
1597 		   src_reg->type == CONST_IMM) {
1598 		dst_imm += src_reg->imm;
1599 	} else if (opcode == BPF_SUB && BPF_SRC(insn->code) == BPF_K) {
1600 		dst_imm -= insn->imm;
1601 	} else if (opcode == BPF_SUB && BPF_SRC(insn->code) == BPF_X &&
1602 		   src_reg->type == CONST_IMM) {
1603 		dst_imm -= src_reg->imm;
1604 	} else if (opcode == BPF_MUL && BPF_SRC(insn->code) == BPF_K) {
1605 		dst_imm *= insn->imm;
1606 	} else if (opcode == BPF_MUL && BPF_SRC(insn->code) == BPF_X &&
1607 		   src_reg->type == CONST_IMM) {
1608 		dst_imm *= src_reg->imm;
1609 	} else if (opcode == BPF_OR && BPF_SRC(insn->code) == BPF_K) {
1610 		dst_imm |= insn->imm;
1611 	} else if (opcode == BPF_OR && BPF_SRC(insn->code) == BPF_X &&
1612 		   src_reg->type == CONST_IMM) {
1613 		dst_imm |= src_reg->imm;
1614 	} else if (opcode == BPF_AND && BPF_SRC(insn->code) == BPF_K) {
1615 		dst_imm &= insn->imm;
1616 	} else if (opcode == BPF_AND && BPF_SRC(insn->code) == BPF_X &&
1617 		   src_reg->type == CONST_IMM) {
1618 		dst_imm &= src_reg->imm;
1619 	} else if (opcode == BPF_RSH && BPF_SRC(insn->code) == BPF_K) {
1620 		dst_imm >>= insn->imm;
1621 	} else if (opcode == BPF_RSH && BPF_SRC(insn->code) == BPF_X &&
1622 		   src_reg->type == CONST_IMM) {
1623 		dst_imm >>= src_reg->imm;
1624 	} else if (opcode == BPF_LSH && BPF_SRC(insn->code) == BPF_K) {
1625 		dst_imm <<= insn->imm;
1626 	} else if (opcode == BPF_LSH && BPF_SRC(insn->code) == BPF_X &&
1627 		   src_reg->type == CONST_IMM) {
1628 		dst_imm <<= src_reg->imm;
1629 	} else {
1630 		mark_reg_unknown_value(regs, insn->dst_reg);
1631 		goto out;
1632 	}
1633 
1634 	dst_reg->imm = dst_imm;
1635 out:
1636 	return 0;
1637 }
1638 
1639 static void check_reg_overflow(struct bpf_reg_state *reg)
1640 {
1641 	if (reg->max_value > BPF_REGISTER_MAX_RANGE)
1642 		reg->max_value = BPF_REGISTER_MAX_RANGE;
1643 	if (reg->min_value < BPF_REGISTER_MIN_RANGE ||
1644 	    reg->min_value > BPF_REGISTER_MAX_RANGE)
1645 		reg->min_value = BPF_REGISTER_MIN_RANGE;
1646 }
1647 
1648 static void adjust_reg_min_max_vals(struct bpf_verifier_env *env,
1649 				    struct bpf_insn *insn)
1650 {
1651 	struct bpf_reg_state *regs = env->cur_state.regs, *dst_reg;
1652 	s64 min_val = BPF_REGISTER_MIN_RANGE;
1653 	u64 max_val = BPF_REGISTER_MAX_RANGE;
1654 	u8 opcode = BPF_OP(insn->code);
1655 
1656 	dst_reg = &regs[insn->dst_reg];
1657 	if (BPF_SRC(insn->code) == BPF_X) {
1658 		check_reg_overflow(&regs[insn->src_reg]);
1659 		min_val = regs[insn->src_reg].min_value;
1660 		max_val = regs[insn->src_reg].max_value;
1661 
1662 		/* If the source register is a random pointer then the
1663 		 * min_value/max_value values represent the range of the known
1664 		 * accesses into that value, not the actual min/max value of the
1665 		 * register itself.  In this case we have to reset the reg range
1666 		 * values so we know it is not safe to look at.
1667 		 */
1668 		if (regs[insn->src_reg].type != CONST_IMM &&
1669 		    regs[insn->src_reg].type != UNKNOWN_VALUE) {
1670 			min_val = BPF_REGISTER_MIN_RANGE;
1671 			max_val = BPF_REGISTER_MAX_RANGE;
1672 		}
1673 	} else if (insn->imm < BPF_REGISTER_MAX_RANGE &&
1674 		   (s64)insn->imm > BPF_REGISTER_MIN_RANGE) {
1675 		min_val = max_val = insn->imm;
1676 	}
1677 
1678 	/* We don't know anything about what was done to this register, mark it
1679 	 * as unknown.
1680 	 */
1681 	if (min_val == BPF_REGISTER_MIN_RANGE &&
1682 	    max_val == BPF_REGISTER_MAX_RANGE) {
1683 		reset_reg_range_values(regs, insn->dst_reg);
1684 		return;
1685 	}
1686 
1687 	/* If one of our values was at the end of our ranges then we can't just
1688 	 * do our normal operations to the register, we need to set the values
1689 	 * to the min/max since they are undefined.
1690 	 */
1691 	if (min_val == BPF_REGISTER_MIN_RANGE)
1692 		dst_reg->min_value = BPF_REGISTER_MIN_RANGE;
1693 	if (max_val == BPF_REGISTER_MAX_RANGE)
1694 		dst_reg->max_value = BPF_REGISTER_MAX_RANGE;
1695 
1696 	switch (opcode) {
1697 	case BPF_ADD:
1698 		if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE)
1699 			dst_reg->min_value += min_val;
1700 		if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1701 			dst_reg->max_value += max_val;
1702 		break;
1703 	case BPF_SUB:
1704 		if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE)
1705 			dst_reg->min_value -= min_val;
1706 		if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1707 			dst_reg->max_value -= max_val;
1708 		break;
1709 	case BPF_MUL:
1710 		if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE)
1711 			dst_reg->min_value *= min_val;
1712 		if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1713 			dst_reg->max_value *= max_val;
1714 		break;
1715 	case BPF_AND:
1716 		/* Disallow AND'ing of negative numbers, ain't nobody got time
1717 		 * for that.  Otherwise the minimum is 0 and the max is the max
1718 		 * value we could AND against.
1719 		 */
1720 		if (min_val < 0)
1721 			dst_reg->min_value = BPF_REGISTER_MIN_RANGE;
1722 		else
1723 			dst_reg->min_value = 0;
1724 		dst_reg->max_value = max_val;
1725 		break;
1726 	case BPF_LSH:
1727 		/* Gotta have special overflow logic here, if we're shifting
1728 		 * more than MAX_RANGE then just assume we have an invalid
1729 		 * range.
1730 		 */
1731 		if (min_val > ilog2(BPF_REGISTER_MAX_RANGE))
1732 			dst_reg->min_value = BPF_REGISTER_MIN_RANGE;
1733 		else if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE)
1734 			dst_reg->min_value <<= min_val;
1735 
1736 		if (max_val > ilog2(BPF_REGISTER_MAX_RANGE))
1737 			dst_reg->max_value = BPF_REGISTER_MAX_RANGE;
1738 		else if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1739 			dst_reg->max_value <<= max_val;
1740 		break;
1741 	case BPF_RSH:
1742 		/* RSH by a negative number is undefined, and the BPF_RSH is an
1743 		 * unsigned shift, so make the appropriate casts.
1744 		 */
1745 		if (min_val < 0 || dst_reg->min_value < 0)
1746 			dst_reg->min_value = BPF_REGISTER_MIN_RANGE;
1747 		else
1748 			dst_reg->min_value =
1749 				(u64)(dst_reg->min_value) >> min_val;
1750 		if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1751 			dst_reg->max_value >>= max_val;
1752 		break;
1753 	default:
1754 		reset_reg_range_values(regs, insn->dst_reg);
1755 		break;
1756 	}
1757 
1758 	check_reg_overflow(dst_reg);
1759 }
1760 
1761 /* check validity of 32-bit and 64-bit arithmetic operations */
1762 static int check_alu_op(struct bpf_verifier_env *env, struct bpf_insn *insn)
1763 {
1764 	struct bpf_reg_state *regs = env->cur_state.regs, *dst_reg;
1765 	u8 opcode = BPF_OP(insn->code);
1766 	int err;
1767 
1768 	if (opcode == BPF_END || opcode == BPF_NEG) {
1769 		if (opcode == BPF_NEG) {
1770 			if (BPF_SRC(insn->code) != 0 ||
1771 			    insn->src_reg != BPF_REG_0 ||
1772 			    insn->off != 0 || insn->imm != 0) {
1773 				verbose("BPF_NEG uses reserved fields\n");
1774 				return -EINVAL;
1775 			}
1776 		} else {
1777 			if (insn->src_reg != BPF_REG_0 || insn->off != 0 ||
1778 			    (insn->imm != 16 && insn->imm != 32 && insn->imm != 64)) {
1779 				verbose("BPF_END uses reserved fields\n");
1780 				return -EINVAL;
1781 			}
1782 		}
1783 
1784 		/* check src operand */
1785 		err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1786 		if (err)
1787 			return err;
1788 
1789 		if (is_pointer_value(env, insn->dst_reg)) {
1790 			verbose("R%d pointer arithmetic prohibited\n",
1791 				insn->dst_reg);
1792 			return -EACCES;
1793 		}
1794 
1795 		/* check dest operand */
1796 		err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1797 		if (err)
1798 			return err;
1799 
1800 	} else if (opcode == BPF_MOV) {
1801 
1802 		if (BPF_SRC(insn->code) == BPF_X) {
1803 			if (insn->imm != 0 || insn->off != 0) {
1804 				verbose("BPF_MOV uses reserved fields\n");
1805 				return -EINVAL;
1806 			}
1807 
1808 			/* check src operand */
1809 			err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1810 			if (err)
1811 				return err;
1812 		} else {
1813 			if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
1814 				verbose("BPF_MOV uses reserved fields\n");
1815 				return -EINVAL;
1816 			}
1817 		}
1818 
1819 		/* check dest operand */
1820 		err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1821 		if (err)
1822 			return err;
1823 
1824 		/* we are setting our register to something new, we need to
1825 		 * reset its range values.
1826 		 */
1827 		reset_reg_range_values(regs, insn->dst_reg);
1828 
1829 		if (BPF_SRC(insn->code) == BPF_X) {
1830 			if (BPF_CLASS(insn->code) == BPF_ALU64) {
1831 				/* case: R1 = R2
1832 				 * copy register state to dest reg
1833 				 */
1834 				regs[insn->dst_reg] = regs[insn->src_reg];
1835 			} else {
1836 				if (is_pointer_value(env, insn->src_reg)) {
1837 					verbose("R%d partial copy of pointer\n",
1838 						insn->src_reg);
1839 					return -EACCES;
1840 				}
1841 				mark_reg_unknown_value(regs, insn->dst_reg);
1842 			}
1843 		} else {
1844 			/* case: R = imm
1845 			 * remember the value we stored into this reg
1846 			 */
1847 			regs[insn->dst_reg].type = CONST_IMM;
1848 			regs[insn->dst_reg].imm = insn->imm;
1849 			regs[insn->dst_reg].max_value = insn->imm;
1850 			regs[insn->dst_reg].min_value = insn->imm;
1851 		}
1852 
1853 	} else if (opcode > BPF_END) {
1854 		verbose("invalid BPF_ALU opcode %x\n", opcode);
1855 		return -EINVAL;
1856 
1857 	} else {	/* all other ALU ops: and, sub, xor, add, ... */
1858 
1859 		if (BPF_SRC(insn->code) == BPF_X) {
1860 			if (insn->imm != 0 || insn->off != 0) {
1861 				verbose("BPF_ALU uses reserved fields\n");
1862 				return -EINVAL;
1863 			}
1864 			/* check src1 operand */
1865 			err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1866 			if (err)
1867 				return err;
1868 		} else {
1869 			if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
1870 				verbose("BPF_ALU uses reserved fields\n");
1871 				return -EINVAL;
1872 			}
1873 		}
1874 
1875 		/* check src2 operand */
1876 		err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1877 		if (err)
1878 			return err;
1879 
1880 		if ((opcode == BPF_MOD || opcode == BPF_DIV) &&
1881 		    BPF_SRC(insn->code) == BPF_K && insn->imm == 0) {
1882 			verbose("div by zero\n");
1883 			return -EINVAL;
1884 		}
1885 
1886 		if ((opcode == BPF_LSH || opcode == BPF_RSH ||
1887 		     opcode == BPF_ARSH) && BPF_SRC(insn->code) == BPF_K) {
1888 			int size = BPF_CLASS(insn->code) == BPF_ALU64 ? 64 : 32;
1889 
1890 			if (insn->imm < 0 || insn->imm >= size) {
1891 				verbose("invalid shift %d\n", insn->imm);
1892 				return -EINVAL;
1893 			}
1894 		}
1895 
1896 		/* check dest operand */
1897 		err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK);
1898 		if (err)
1899 			return err;
1900 
1901 		dst_reg = &regs[insn->dst_reg];
1902 
1903 		/* first we want to adjust our ranges. */
1904 		adjust_reg_min_max_vals(env, insn);
1905 
1906 		/* pattern match 'bpf_add Rx, imm' instruction */
1907 		if (opcode == BPF_ADD && BPF_CLASS(insn->code) == BPF_ALU64 &&
1908 		    dst_reg->type == FRAME_PTR && BPF_SRC(insn->code) == BPF_K) {
1909 			dst_reg->type = PTR_TO_STACK;
1910 			dst_reg->imm = insn->imm;
1911 			return 0;
1912 		} else if (opcode == BPF_ADD &&
1913 			   BPF_CLASS(insn->code) == BPF_ALU64 &&
1914 			   (dst_reg->type == PTR_TO_PACKET ||
1915 			    (BPF_SRC(insn->code) == BPF_X &&
1916 			     regs[insn->src_reg].type == PTR_TO_PACKET))) {
1917 			/* ptr_to_packet += K|X */
1918 			return check_packet_ptr_add(env, insn);
1919 		} else if (BPF_CLASS(insn->code) == BPF_ALU64 &&
1920 			   dst_reg->type == UNKNOWN_VALUE &&
1921 			   env->allow_ptr_leaks) {
1922 			/* unknown += K|X */
1923 			return evaluate_reg_alu(env, insn);
1924 		} else if (BPF_CLASS(insn->code) == BPF_ALU64 &&
1925 			   dst_reg->type == CONST_IMM &&
1926 			   env->allow_ptr_leaks) {
1927 			/* reg_imm += K|X */
1928 			return evaluate_reg_imm_alu(env, insn);
1929 		} else if (is_pointer_value(env, insn->dst_reg)) {
1930 			verbose("R%d pointer arithmetic prohibited\n",
1931 				insn->dst_reg);
1932 			return -EACCES;
1933 		} else if (BPF_SRC(insn->code) == BPF_X &&
1934 			   is_pointer_value(env, insn->src_reg)) {
1935 			verbose("R%d pointer arithmetic prohibited\n",
1936 				insn->src_reg);
1937 			return -EACCES;
1938 		}
1939 
1940 		/* If we did pointer math on a map value then just set it to our
1941 		 * PTR_TO_MAP_VALUE_ADJ type so we can deal with any stores or
1942 		 * loads to this register appropriately, otherwise just mark the
1943 		 * register as unknown.
1944 		 */
1945 		if (env->allow_ptr_leaks &&
1946 		    BPF_CLASS(insn->code) == BPF_ALU64 && opcode == BPF_ADD &&
1947 		    (dst_reg->type == PTR_TO_MAP_VALUE ||
1948 		     dst_reg->type == PTR_TO_MAP_VALUE_ADJ))
1949 			dst_reg->type = PTR_TO_MAP_VALUE_ADJ;
1950 		else
1951 			mark_reg_unknown_value(regs, insn->dst_reg);
1952 	}
1953 
1954 	return 0;
1955 }
1956 
1957 static void find_good_pkt_pointers(struct bpf_verifier_state *state,
1958 				   struct bpf_reg_state *dst_reg)
1959 {
1960 	struct bpf_reg_state *regs = state->regs, *reg;
1961 	int i;
1962 
1963 	/* LLVM can generate two kind of checks:
1964 	 *
1965 	 * Type 1:
1966 	 *
1967 	 *   r2 = r3;
1968 	 *   r2 += 8;
1969 	 *   if (r2 > pkt_end) goto <handle exception>
1970 	 *   <access okay>
1971 	 *
1972 	 *   Where:
1973 	 *     r2 == dst_reg, pkt_end == src_reg
1974 	 *     r2=pkt(id=n,off=8,r=0)
1975 	 *     r3=pkt(id=n,off=0,r=0)
1976 	 *
1977 	 * Type 2:
1978 	 *
1979 	 *   r2 = r3;
1980 	 *   r2 += 8;
1981 	 *   if (pkt_end >= r2) goto <access okay>
1982 	 *   <handle exception>
1983 	 *
1984 	 *   Where:
1985 	 *     pkt_end == dst_reg, r2 == src_reg
1986 	 *     r2=pkt(id=n,off=8,r=0)
1987 	 *     r3=pkt(id=n,off=0,r=0)
1988 	 *
1989 	 * Find register r3 and mark its range as r3=pkt(id=n,off=0,r=8)
1990 	 * so that range of bytes [r3, r3 + 8) is safe to access.
1991 	 */
1992 
1993 	for (i = 0; i < MAX_BPF_REG; i++)
1994 		if (regs[i].type == PTR_TO_PACKET && regs[i].id == dst_reg->id)
1995 			/* keep the maximum range already checked */
1996 			regs[i].range = max(regs[i].range, dst_reg->off);
1997 
1998 	for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
1999 		if (state->stack_slot_type[i] != STACK_SPILL)
2000 			continue;
2001 		reg = &state->spilled_regs[i / BPF_REG_SIZE];
2002 		if (reg->type == PTR_TO_PACKET && reg->id == dst_reg->id)
2003 			reg->range = max(reg->range, dst_reg->off);
2004 	}
2005 }
2006 
2007 /* Adjusts the register min/max values in the case that the dst_reg is the
2008  * variable register that we are working on, and src_reg is a constant or we're
2009  * simply doing a BPF_K check.
2010  */
2011 static void reg_set_min_max(struct bpf_reg_state *true_reg,
2012 			    struct bpf_reg_state *false_reg, u64 val,
2013 			    u8 opcode)
2014 {
2015 	switch (opcode) {
2016 	case BPF_JEQ:
2017 		/* If this is false then we know nothing Jon Snow, but if it is
2018 		 * true then we know for sure.
2019 		 */
2020 		true_reg->max_value = true_reg->min_value = val;
2021 		break;
2022 	case BPF_JNE:
2023 		/* If this is true we know nothing Jon Snow, but if it is false
2024 		 * we know the value for sure;
2025 		 */
2026 		false_reg->max_value = false_reg->min_value = val;
2027 		break;
2028 	case BPF_JGT:
2029 		/* Unsigned comparison, the minimum value is 0. */
2030 		false_reg->min_value = 0;
2031 		/* fallthrough */
2032 	case BPF_JSGT:
2033 		/* If this is false then we know the maximum val is val,
2034 		 * otherwise we know the min val is val+1.
2035 		 */
2036 		false_reg->max_value = val;
2037 		true_reg->min_value = val + 1;
2038 		break;
2039 	case BPF_JGE:
2040 		/* Unsigned comparison, the minimum value is 0. */
2041 		false_reg->min_value = 0;
2042 		/* fallthrough */
2043 	case BPF_JSGE:
2044 		/* If this is false then we know the maximum value is val - 1,
2045 		 * otherwise we know the mimimum value is val.
2046 		 */
2047 		false_reg->max_value = val - 1;
2048 		true_reg->min_value = val;
2049 		break;
2050 	default:
2051 		break;
2052 	}
2053 
2054 	check_reg_overflow(false_reg);
2055 	check_reg_overflow(true_reg);
2056 }
2057 
2058 /* Same as above, but for the case that dst_reg is a CONST_IMM reg and src_reg
2059  * is the variable reg.
2060  */
2061 static void reg_set_min_max_inv(struct bpf_reg_state *true_reg,
2062 				struct bpf_reg_state *false_reg, u64 val,
2063 				u8 opcode)
2064 {
2065 	switch (opcode) {
2066 	case BPF_JEQ:
2067 		/* If this is false then we know nothing Jon Snow, but if it is
2068 		 * true then we know for sure.
2069 		 */
2070 		true_reg->max_value = true_reg->min_value = val;
2071 		break;
2072 	case BPF_JNE:
2073 		/* If this is true we know nothing Jon Snow, but if it is false
2074 		 * we know the value for sure;
2075 		 */
2076 		false_reg->max_value = false_reg->min_value = val;
2077 		break;
2078 	case BPF_JGT:
2079 		/* Unsigned comparison, the minimum value is 0. */
2080 		true_reg->min_value = 0;
2081 		/* fallthrough */
2082 	case BPF_JSGT:
2083 		/*
2084 		 * If this is false, then the val is <= the register, if it is
2085 		 * true the register <= to the val.
2086 		 */
2087 		false_reg->min_value = val;
2088 		true_reg->max_value = val - 1;
2089 		break;
2090 	case BPF_JGE:
2091 		/* Unsigned comparison, the minimum value is 0. */
2092 		true_reg->min_value = 0;
2093 		/* fallthrough */
2094 	case BPF_JSGE:
2095 		/* If this is false then constant < register, if it is true then
2096 		 * the register < constant.
2097 		 */
2098 		false_reg->min_value = val + 1;
2099 		true_reg->max_value = val;
2100 		break;
2101 	default:
2102 		break;
2103 	}
2104 
2105 	check_reg_overflow(false_reg);
2106 	check_reg_overflow(true_reg);
2107 }
2108 
2109 static void mark_map_reg(struct bpf_reg_state *regs, u32 regno, u32 id,
2110 			 enum bpf_reg_type type)
2111 {
2112 	struct bpf_reg_state *reg = &regs[regno];
2113 
2114 	if (reg->type == PTR_TO_MAP_VALUE_OR_NULL && reg->id == id) {
2115 		reg->type = type;
2116 		/* We don't need id from this point onwards anymore, thus we
2117 		 * should better reset it, so that state pruning has chances
2118 		 * to take effect.
2119 		 */
2120 		reg->id = 0;
2121 		if (type == UNKNOWN_VALUE)
2122 			__mark_reg_unknown_value(regs, regno);
2123 	}
2124 }
2125 
2126 /* The logic is similar to find_good_pkt_pointers(), both could eventually
2127  * be folded together at some point.
2128  */
2129 static void mark_map_regs(struct bpf_verifier_state *state, u32 regno,
2130 			  enum bpf_reg_type type)
2131 {
2132 	struct bpf_reg_state *regs = state->regs;
2133 	u32 id = regs[regno].id;
2134 	int i;
2135 
2136 	for (i = 0; i < MAX_BPF_REG; i++)
2137 		mark_map_reg(regs, i, id, type);
2138 
2139 	for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
2140 		if (state->stack_slot_type[i] != STACK_SPILL)
2141 			continue;
2142 		mark_map_reg(state->spilled_regs, i / BPF_REG_SIZE, id, type);
2143 	}
2144 }
2145 
2146 static int check_cond_jmp_op(struct bpf_verifier_env *env,
2147 			     struct bpf_insn *insn, int *insn_idx)
2148 {
2149 	struct bpf_verifier_state *other_branch, *this_branch = &env->cur_state;
2150 	struct bpf_reg_state *regs = this_branch->regs, *dst_reg;
2151 	u8 opcode = BPF_OP(insn->code);
2152 	int err;
2153 
2154 	if (opcode > BPF_EXIT) {
2155 		verbose("invalid BPF_JMP opcode %x\n", opcode);
2156 		return -EINVAL;
2157 	}
2158 
2159 	if (BPF_SRC(insn->code) == BPF_X) {
2160 		if (insn->imm != 0) {
2161 			verbose("BPF_JMP uses reserved fields\n");
2162 			return -EINVAL;
2163 		}
2164 
2165 		/* check src1 operand */
2166 		err = check_reg_arg(regs, insn->src_reg, SRC_OP);
2167 		if (err)
2168 			return err;
2169 
2170 		if (is_pointer_value(env, insn->src_reg)) {
2171 			verbose("R%d pointer comparison prohibited\n",
2172 				insn->src_reg);
2173 			return -EACCES;
2174 		}
2175 	} else {
2176 		if (insn->src_reg != BPF_REG_0) {
2177 			verbose("BPF_JMP uses reserved fields\n");
2178 			return -EINVAL;
2179 		}
2180 	}
2181 
2182 	/* check src2 operand */
2183 	err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
2184 	if (err)
2185 		return err;
2186 
2187 	dst_reg = &regs[insn->dst_reg];
2188 
2189 	/* detect if R == 0 where R was initialized to zero earlier */
2190 	if (BPF_SRC(insn->code) == BPF_K &&
2191 	    (opcode == BPF_JEQ || opcode == BPF_JNE) &&
2192 	    dst_reg->type == CONST_IMM && dst_reg->imm == insn->imm) {
2193 		if (opcode == BPF_JEQ) {
2194 			/* if (imm == imm) goto pc+off;
2195 			 * only follow the goto, ignore fall-through
2196 			 */
2197 			*insn_idx += insn->off;
2198 			return 0;
2199 		} else {
2200 			/* if (imm != imm) goto pc+off;
2201 			 * only follow fall-through branch, since
2202 			 * that's where the program will go
2203 			 */
2204 			return 0;
2205 		}
2206 	}
2207 
2208 	other_branch = push_stack(env, *insn_idx + insn->off + 1, *insn_idx);
2209 	if (!other_branch)
2210 		return -EFAULT;
2211 
2212 	/* detect if we are comparing against a constant value so we can adjust
2213 	 * our min/max values for our dst register.
2214 	 */
2215 	if (BPF_SRC(insn->code) == BPF_X) {
2216 		if (regs[insn->src_reg].type == CONST_IMM)
2217 			reg_set_min_max(&other_branch->regs[insn->dst_reg],
2218 					dst_reg, regs[insn->src_reg].imm,
2219 					opcode);
2220 		else if (dst_reg->type == CONST_IMM)
2221 			reg_set_min_max_inv(&other_branch->regs[insn->src_reg],
2222 					    &regs[insn->src_reg], dst_reg->imm,
2223 					    opcode);
2224 	} else {
2225 		reg_set_min_max(&other_branch->regs[insn->dst_reg],
2226 					dst_reg, insn->imm, opcode);
2227 	}
2228 
2229 	/* detect if R == 0 where R is returned from bpf_map_lookup_elem() */
2230 	if (BPF_SRC(insn->code) == BPF_K &&
2231 	    insn->imm == 0 && (opcode == BPF_JEQ || opcode == BPF_JNE) &&
2232 	    dst_reg->type == PTR_TO_MAP_VALUE_OR_NULL) {
2233 		/* Mark all identical map registers in each branch as either
2234 		 * safe or unknown depending R == 0 or R != 0 conditional.
2235 		 */
2236 		mark_map_regs(this_branch, insn->dst_reg,
2237 			      opcode == BPF_JEQ ? PTR_TO_MAP_VALUE : UNKNOWN_VALUE);
2238 		mark_map_regs(other_branch, insn->dst_reg,
2239 			      opcode == BPF_JEQ ? UNKNOWN_VALUE : PTR_TO_MAP_VALUE);
2240 	} else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JGT &&
2241 		   dst_reg->type == PTR_TO_PACKET &&
2242 		   regs[insn->src_reg].type == PTR_TO_PACKET_END) {
2243 		find_good_pkt_pointers(this_branch, dst_reg);
2244 	} else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JGE &&
2245 		   dst_reg->type == PTR_TO_PACKET_END &&
2246 		   regs[insn->src_reg].type == PTR_TO_PACKET) {
2247 		find_good_pkt_pointers(other_branch, &regs[insn->src_reg]);
2248 	} else if (is_pointer_value(env, insn->dst_reg)) {
2249 		verbose("R%d pointer comparison prohibited\n", insn->dst_reg);
2250 		return -EACCES;
2251 	}
2252 	if (log_level)
2253 		print_verifier_state(this_branch);
2254 	return 0;
2255 }
2256 
2257 /* return the map pointer stored inside BPF_LD_IMM64 instruction */
2258 static struct bpf_map *ld_imm64_to_map_ptr(struct bpf_insn *insn)
2259 {
2260 	u64 imm64 = ((u64) (u32) insn[0].imm) | ((u64) (u32) insn[1].imm) << 32;
2261 
2262 	return (struct bpf_map *) (unsigned long) imm64;
2263 }
2264 
2265 /* verify BPF_LD_IMM64 instruction */
2266 static int check_ld_imm(struct bpf_verifier_env *env, struct bpf_insn *insn)
2267 {
2268 	struct bpf_reg_state *regs = env->cur_state.regs;
2269 	int err;
2270 
2271 	if (BPF_SIZE(insn->code) != BPF_DW) {
2272 		verbose("invalid BPF_LD_IMM insn\n");
2273 		return -EINVAL;
2274 	}
2275 	if (insn->off != 0) {
2276 		verbose("BPF_LD_IMM64 uses reserved fields\n");
2277 		return -EINVAL;
2278 	}
2279 
2280 	err = check_reg_arg(regs, insn->dst_reg, DST_OP);
2281 	if (err)
2282 		return err;
2283 
2284 	if (insn->src_reg == 0) {
2285 		u64 imm = ((u64)(insn + 1)->imm << 32) | (u32)insn->imm;
2286 
2287 		regs[insn->dst_reg].type = CONST_IMM;
2288 		regs[insn->dst_reg].imm = imm;
2289 		return 0;
2290 	}
2291 
2292 	/* replace_map_fd_with_map_ptr() should have caught bad ld_imm64 */
2293 	BUG_ON(insn->src_reg != BPF_PSEUDO_MAP_FD);
2294 
2295 	regs[insn->dst_reg].type = CONST_PTR_TO_MAP;
2296 	regs[insn->dst_reg].map_ptr = ld_imm64_to_map_ptr(insn);
2297 	return 0;
2298 }
2299 
2300 static bool may_access_skb(enum bpf_prog_type type)
2301 {
2302 	switch (type) {
2303 	case BPF_PROG_TYPE_SOCKET_FILTER:
2304 	case BPF_PROG_TYPE_SCHED_CLS:
2305 	case BPF_PROG_TYPE_SCHED_ACT:
2306 		return true;
2307 	default:
2308 		return false;
2309 	}
2310 }
2311 
2312 /* verify safety of LD_ABS|LD_IND instructions:
2313  * - they can only appear in the programs where ctx == skb
2314  * - since they are wrappers of function calls, they scratch R1-R5 registers,
2315  *   preserve R6-R9, and store return value into R0
2316  *
2317  * Implicit input:
2318  *   ctx == skb == R6 == CTX
2319  *
2320  * Explicit input:
2321  *   SRC == any register
2322  *   IMM == 32-bit immediate
2323  *
2324  * Output:
2325  *   R0 - 8/16/32-bit skb data converted to cpu endianness
2326  */
2327 static int check_ld_abs(struct bpf_verifier_env *env, struct bpf_insn *insn)
2328 {
2329 	struct bpf_reg_state *regs = env->cur_state.regs;
2330 	u8 mode = BPF_MODE(insn->code);
2331 	struct bpf_reg_state *reg;
2332 	int i, err;
2333 
2334 	if (!may_access_skb(env->prog->type)) {
2335 		verbose("BPF_LD_[ABS|IND] instructions not allowed for this program type\n");
2336 		return -EINVAL;
2337 	}
2338 
2339 	if (insn->dst_reg != BPF_REG_0 || insn->off != 0 ||
2340 	    BPF_SIZE(insn->code) == BPF_DW ||
2341 	    (mode == BPF_ABS && insn->src_reg != BPF_REG_0)) {
2342 		verbose("BPF_LD_[ABS|IND] uses reserved fields\n");
2343 		return -EINVAL;
2344 	}
2345 
2346 	/* check whether implicit source operand (register R6) is readable */
2347 	err = check_reg_arg(regs, BPF_REG_6, SRC_OP);
2348 	if (err)
2349 		return err;
2350 
2351 	if (regs[BPF_REG_6].type != PTR_TO_CTX) {
2352 		verbose("at the time of BPF_LD_ABS|IND R6 != pointer to skb\n");
2353 		return -EINVAL;
2354 	}
2355 
2356 	if (mode == BPF_IND) {
2357 		/* check explicit source operand */
2358 		err = check_reg_arg(regs, insn->src_reg, SRC_OP);
2359 		if (err)
2360 			return err;
2361 	}
2362 
2363 	/* reset caller saved regs to unreadable */
2364 	for (i = 0; i < CALLER_SAVED_REGS; i++) {
2365 		reg = regs + caller_saved[i];
2366 		reg->type = NOT_INIT;
2367 		reg->imm = 0;
2368 	}
2369 
2370 	/* mark destination R0 register as readable, since it contains
2371 	 * the value fetched from the packet
2372 	 */
2373 	regs[BPF_REG_0].type = UNKNOWN_VALUE;
2374 	return 0;
2375 }
2376 
2377 /* non-recursive DFS pseudo code
2378  * 1  procedure DFS-iterative(G,v):
2379  * 2      label v as discovered
2380  * 3      let S be a stack
2381  * 4      S.push(v)
2382  * 5      while S is not empty
2383  * 6            t <- S.pop()
2384  * 7            if t is what we're looking for:
2385  * 8                return t
2386  * 9            for all edges e in G.adjacentEdges(t) do
2387  * 10               if edge e is already labelled
2388  * 11                   continue with the next edge
2389  * 12               w <- G.adjacentVertex(t,e)
2390  * 13               if vertex w is not discovered and not explored
2391  * 14                   label e as tree-edge
2392  * 15                   label w as discovered
2393  * 16                   S.push(w)
2394  * 17                   continue at 5
2395  * 18               else if vertex w is discovered
2396  * 19                   label e as back-edge
2397  * 20               else
2398  * 21                   // vertex w is explored
2399  * 22                   label e as forward- or cross-edge
2400  * 23           label t as explored
2401  * 24           S.pop()
2402  *
2403  * convention:
2404  * 0x10 - discovered
2405  * 0x11 - discovered and fall-through edge labelled
2406  * 0x12 - discovered and fall-through and branch edges labelled
2407  * 0x20 - explored
2408  */
2409 
2410 enum {
2411 	DISCOVERED = 0x10,
2412 	EXPLORED = 0x20,
2413 	FALLTHROUGH = 1,
2414 	BRANCH = 2,
2415 };
2416 
2417 #define STATE_LIST_MARK ((struct bpf_verifier_state_list *) -1L)
2418 
2419 static int *insn_stack;	/* stack of insns to process */
2420 static int cur_stack;	/* current stack index */
2421 static int *insn_state;
2422 
2423 /* t, w, e - match pseudo-code above:
2424  * t - index of current instruction
2425  * w - next instruction
2426  * e - edge
2427  */
2428 static int push_insn(int t, int w, int e, struct bpf_verifier_env *env)
2429 {
2430 	if (e == FALLTHROUGH && insn_state[t] >= (DISCOVERED | FALLTHROUGH))
2431 		return 0;
2432 
2433 	if (e == BRANCH && insn_state[t] >= (DISCOVERED | BRANCH))
2434 		return 0;
2435 
2436 	if (w < 0 || w >= env->prog->len) {
2437 		verbose("jump out of range from insn %d to %d\n", t, w);
2438 		return -EINVAL;
2439 	}
2440 
2441 	if (e == BRANCH)
2442 		/* mark branch target for state pruning */
2443 		env->explored_states[w] = STATE_LIST_MARK;
2444 
2445 	if (insn_state[w] == 0) {
2446 		/* tree-edge */
2447 		insn_state[t] = DISCOVERED | e;
2448 		insn_state[w] = DISCOVERED;
2449 		if (cur_stack >= env->prog->len)
2450 			return -E2BIG;
2451 		insn_stack[cur_stack++] = w;
2452 		return 1;
2453 	} else if ((insn_state[w] & 0xF0) == DISCOVERED) {
2454 		verbose("back-edge from insn %d to %d\n", t, w);
2455 		return -EINVAL;
2456 	} else if (insn_state[w] == EXPLORED) {
2457 		/* forward- or cross-edge */
2458 		insn_state[t] = DISCOVERED | e;
2459 	} else {
2460 		verbose("insn state internal bug\n");
2461 		return -EFAULT;
2462 	}
2463 	return 0;
2464 }
2465 
2466 /* non-recursive depth-first-search to detect loops in BPF program
2467  * loop == back-edge in directed graph
2468  */
2469 static int check_cfg(struct bpf_verifier_env *env)
2470 {
2471 	struct bpf_insn *insns = env->prog->insnsi;
2472 	int insn_cnt = env->prog->len;
2473 	int ret = 0;
2474 	int i, t;
2475 
2476 	insn_state = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
2477 	if (!insn_state)
2478 		return -ENOMEM;
2479 
2480 	insn_stack = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
2481 	if (!insn_stack) {
2482 		kfree(insn_state);
2483 		return -ENOMEM;
2484 	}
2485 
2486 	insn_state[0] = DISCOVERED; /* mark 1st insn as discovered */
2487 	insn_stack[0] = 0; /* 0 is the first instruction */
2488 	cur_stack = 1;
2489 
2490 peek_stack:
2491 	if (cur_stack == 0)
2492 		goto check_state;
2493 	t = insn_stack[cur_stack - 1];
2494 
2495 	if (BPF_CLASS(insns[t].code) == BPF_JMP) {
2496 		u8 opcode = BPF_OP(insns[t].code);
2497 
2498 		if (opcode == BPF_EXIT) {
2499 			goto mark_explored;
2500 		} else if (opcode == BPF_CALL) {
2501 			ret = push_insn(t, t + 1, FALLTHROUGH, env);
2502 			if (ret == 1)
2503 				goto peek_stack;
2504 			else if (ret < 0)
2505 				goto err_free;
2506 			if (t + 1 < insn_cnt)
2507 				env->explored_states[t + 1] = STATE_LIST_MARK;
2508 		} else if (opcode == BPF_JA) {
2509 			if (BPF_SRC(insns[t].code) != BPF_K) {
2510 				ret = -EINVAL;
2511 				goto err_free;
2512 			}
2513 			/* unconditional jump with single edge */
2514 			ret = push_insn(t, t + insns[t].off + 1,
2515 					FALLTHROUGH, env);
2516 			if (ret == 1)
2517 				goto peek_stack;
2518 			else if (ret < 0)
2519 				goto err_free;
2520 			/* tell verifier to check for equivalent states
2521 			 * after every call and jump
2522 			 */
2523 			if (t + 1 < insn_cnt)
2524 				env->explored_states[t + 1] = STATE_LIST_MARK;
2525 		} else {
2526 			/* conditional jump with two edges */
2527 			ret = push_insn(t, t + 1, FALLTHROUGH, env);
2528 			if (ret == 1)
2529 				goto peek_stack;
2530 			else if (ret < 0)
2531 				goto err_free;
2532 
2533 			ret = push_insn(t, t + insns[t].off + 1, BRANCH, env);
2534 			if (ret == 1)
2535 				goto peek_stack;
2536 			else if (ret < 0)
2537 				goto err_free;
2538 		}
2539 	} else {
2540 		/* all other non-branch instructions with single
2541 		 * fall-through edge
2542 		 */
2543 		ret = push_insn(t, t + 1, FALLTHROUGH, env);
2544 		if (ret == 1)
2545 			goto peek_stack;
2546 		else if (ret < 0)
2547 			goto err_free;
2548 	}
2549 
2550 mark_explored:
2551 	insn_state[t] = EXPLORED;
2552 	if (cur_stack-- <= 0) {
2553 		verbose("pop stack internal bug\n");
2554 		ret = -EFAULT;
2555 		goto err_free;
2556 	}
2557 	goto peek_stack;
2558 
2559 check_state:
2560 	for (i = 0; i < insn_cnt; i++) {
2561 		if (insn_state[i] != EXPLORED) {
2562 			verbose("unreachable insn %d\n", i);
2563 			ret = -EINVAL;
2564 			goto err_free;
2565 		}
2566 	}
2567 	ret = 0; /* cfg looks good */
2568 
2569 err_free:
2570 	kfree(insn_state);
2571 	kfree(insn_stack);
2572 	return ret;
2573 }
2574 
2575 /* the following conditions reduce the number of explored insns
2576  * from ~140k to ~80k for ultra large programs that use a lot of ptr_to_packet
2577  */
2578 static bool compare_ptrs_to_packet(struct bpf_reg_state *old,
2579 				   struct bpf_reg_state *cur)
2580 {
2581 	if (old->id != cur->id)
2582 		return false;
2583 
2584 	/* old ptr_to_packet is more conservative, since it allows smaller
2585 	 * range. Ex:
2586 	 * old(off=0,r=10) is equal to cur(off=0,r=20), because
2587 	 * old(off=0,r=10) means that with range=10 the verifier proceeded
2588 	 * further and found no issues with the program. Now we're in the same
2589 	 * spot with cur(off=0,r=20), so we're safe too, since anything further
2590 	 * will only be looking at most 10 bytes after this pointer.
2591 	 */
2592 	if (old->off == cur->off && old->range < cur->range)
2593 		return true;
2594 
2595 	/* old(off=20,r=10) is equal to cur(off=22,re=22 or 5 or 0)
2596 	 * since both cannot be used for packet access and safe(old)
2597 	 * pointer has smaller off that could be used for further
2598 	 * 'if (ptr > data_end)' check
2599 	 * Ex:
2600 	 * old(off=20,r=10) and cur(off=22,r=22) and cur(off=22,r=0) mean
2601 	 * that we cannot access the packet.
2602 	 * The safe range is:
2603 	 * [ptr, ptr + range - off)
2604 	 * so whenever off >=range, it means no safe bytes from this pointer.
2605 	 * When comparing old->off <= cur->off, it means that older code
2606 	 * went with smaller offset and that offset was later
2607 	 * used to figure out the safe range after 'if (ptr > data_end)' check
2608 	 * Say, 'old' state was explored like:
2609 	 * ... R3(off=0, r=0)
2610 	 * R4 = R3 + 20
2611 	 * ... now R4(off=20,r=0)  <-- here
2612 	 * if (R4 > data_end)
2613 	 * ... R4(off=20,r=20), R3(off=0,r=20) and R3 can be used to access.
2614 	 * ... the code further went all the way to bpf_exit.
2615 	 * Now the 'cur' state at the mark 'here' has R4(off=30,r=0).
2616 	 * old_R4(off=20,r=0) equal to cur_R4(off=30,r=0), since if the verifier
2617 	 * goes further, such cur_R4 will give larger safe packet range after
2618 	 * 'if (R4 > data_end)' and all further insn were already good with r=20,
2619 	 * so they will be good with r=30 and we can prune the search.
2620 	 */
2621 	if (old->off <= cur->off &&
2622 	    old->off >= old->range && cur->off >= cur->range)
2623 		return true;
2624 
2625 	return false;
2626 }
2627 
2628 /* compare two verifier states
2629  *
2630  * all states stored in state_list are known to be valid, since
2631  * verifier reached 'bpf_exit' instruction through them
2632  *
2633  * this function is called when verifier exploring different branches of
2634  * execution popped from the state stack. If it sees an old state that has
2635  * more strict register state and more strict stack state then this execution
2636  * branch doesn't need to be explored further, since verifier already
2637  * concluded that more strict state leads to valid finish.
2638  *
2639  * Therefore two states are equivalent if register state is more conservative
2640  * and explored stack state is more conservative than the current one.
2641  * Example:
2642  *       explored                   current
2643  * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC)
2644  * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC)
2645  *
2646  * In other words if current stack state (one being explored) has more
2647  * valid slots than old one that already passed validation, it means
2648  * the verifier can stop exploring and conclude that current state is valid too
2649  *
2650  * Similarly with registers. If explored state has register type as invalid
2651  * whereas register type in current state is meaningful, it means that
2652  * the current state will reach 'bpf_exit' instruction safely
2653  */
2654 static bool states_equal(struct bpf_verifier_env *env,
2655 			 struct bpf_verifier_state *old,
2656 			 struct bpf_verifier_state *cur)
2657 {
2658 	bool varlen_map_access = env->varlen_map_value_access;
2659 	struct bpf_reg_state *rold, *rcur;
2660 	int i;
2661 
2662 	for (i = 0; i < MAX_BPF_REG; i++) {
2663 		rold = &old->regs[i];
2664 		rcur = &cur->regs[i];
2665 
2666 		if (memcmp(rold, rcur, sizeof(*rold)) == 0)
2667 			continue;
2668 
2669 		/* If the ranges were not the same, but everything else was and
2670 		 * we didn't do a variable access into a map then we are a-ok.
2671 		 */
2672 		if (!varlen_map_access &&
2673 		    memcmp(rold, rcur, offsetofend(struct bpf_reg_state, id)) == 0)
2674 			continue;
2675 
2676 		/* If we didn't map access then again we don't care about the
2677 		 * mismatched range values and it's ok if our old type was
2678 		 * UNKNOWN and we didn't go to a NOT_INIT'ed reg.
2679 		 */
2680 		if (rold->type == NOT_INIT ||
2681 		    (!varlen_map_access && rold->type == UNKNOWN_VALUE &&
2682 		     rcur->type != NOT_INIT))
2683 			continue;
2684 
2685 		if (rold->type == PTR_TO_PACKET && rcur->type == PTR_TO_PACKET &&
2686 		    compare_ptrs_to_packet(rold, rcur))
2687 			continue;
2688 
2689 		return false;
2690 	}
2691 
2692 	for (i = 0; i < MAX_BPF_STACK; i++) {
2693 		if (old->stack_slot_type[i] == STACK_INVALID)
2694 			continue;
2695 		if (old->stack_slot_type[i] != cur->stack_slot_type[i])
2696 			/* Ex: old explored (safe) state has STACK_SPILL in
2697 			 * this stack slot, but current has has STACK_MISC ->
2698 			 * this verifier states are not equivalent,
2699 			 * return false to continue verification of this path
2700 			 */
2701 			return false;
2702 		if (i % BPF_REG_SIZE)
2703 			continue;
2704 		if (memcmp(&old->spilled_regs[i / BPF_REG_SIZE],
2705 			   &cur->spilled_regs[i / BPF_REG_SIZE],
2706 			   sizeof(old->spilled_regs[0])))
2707 			/* when explored and current stack slot types are
2708 			 * the same, check that stored pointers types
2709 			 * are the same as well.
2710 			 * Ex: explored safe path could have stored
2711 			 * (bpf_reg_state) {.type = PTR_TO_STACK, .imm = -8}
2712 			 * but current path has stored:
2713 			 * (bpf_reg_state) {.type = PTR_TO_STACK, .imm = -16}
2714 			 * such verifier states are not equivalent.
2715 			 * return false to continue verification of this path
2716 			 */
2717 			return false;
2718 		else
2719 			continue;
2720 	}
2721 	return true;
2722 }
2723 
2724 static int is_state_visited(struct bpf_verifier_env *env, int insn_idx)
2725 {
2726 	struct bpf_verifier_state_list *new_sl;
2727 	struct bpf_verifier_state_list *sl;
2728 
2729 	sl = env->explored_states[insn_idx];
2730 	if (!sl)
2731 		/* this 'insn_idx' instruction wasn't marked, so we will not
2732 		 * be doing state search here
2733 		 */
2734 		return 0;
2735 
2736 	while (sl != STATE_LIST_MARK) {
2737 		if (states_equal(env, &sl->state, &env->cur_state))
2738 			/* reached equivalent register/stack state,
2739 			 * prune the search
2740 			 */
2741 			return 1;
2742 		sl = sl->next;
2743 	}
2744 
2745 	/* there were no equivalent states, remember current one.
2746 	 * technically the current state is not proven to be safe yet,
2747 	 * but it will either reach bpf_exit (which means it's safe) or
2748 	 * it will be rejected. Since there are no loops, we won't be
2749 	 * seeing this 'insn_idx' instruction again on the way to bpf_exit
2750 	 */
2751 	new_sl = kmalloc(sizeof(struct bpf_verifier_state_list), GFP_USER);
2752 	if (!new_sl)
2753 		return -ENOMEM;
2754 
2755 	/* add new state to the head of linked list */
2756 	memcpy(&new_sl->state, &env->cur_state, sizeof(env->cur_state));
2757 	new_sl->next = env->explored_states[insn_idx];
2758 	env->explored_states[insn_idx] = new_sl;
2759 	return 0;
2760 }
2761 
2762 static int ext_analyzer_insn_hook(struct bpf_verifier_env *env,
2763 				  int insn_idx, int prev_insn_idx)
2764 {
2765 	if (!env->analyzer_ops || !env->analyzer_ops->insn_hook)
2766 		return 0;
2767 
2768 	return env->analyzer_ops->insn_hook(env, insn_idx, prev_insn_idx);
2769 }
2770 
2771 static int do_check(struct bpf_verifier_env *env)
2772 {
2773 	struct bpf_verifier_state *state = &env->cur_state;
2774 	struct bpf_insn *insns = env->prog->insnsi;
2775 	struct bpf_reg_state *regs = state->regs;
2776 	int insn_cnt = env->prog->len;
2777 	int insn_idx, prev_insn_idx = 0;
2778 	int insn_processed = 0;
2779 	bool do_print_state = false;
2780 
2781 	init_reg_state(regs);
2782 	insn_idx = 0;
2783 	env->varlen_map_value_access = false;
2784 	for (;;) {
2785 		struct bpf_insn *insn;
2786 		u8 class;
2787 		int err;
2788 
2789 		if (insn_idx >= insn_cnt) {
2790 			verbose("invalid insn idx %d insn_cnt %d\n",
2791 				insn_idx, insn_cnt);
2792 			return -EFAULT;
2793 		}
2794 
2795 		insn = &insns[insn_idx];
2796 		class = BPF_CLASS(insn->code);
2797 
2798 		if (++insn_processed > BPF_COMPLEXITY_LIMIT_INSNS) {
2799 			verbose("BPF program is too large. Processed %d insn\n",
2800 				insn_processed);
2801 			return -E2BIG;
2802 		}
2803 
2804 		err = is_state_visited(env, insn_idx);
2805 		if (err < 0)
2806 			return err;
2807 		if (err == 1) {
2808 			/* found equivalent state, can prune the search */
2809 			if (log_level) {
2810 				if (do_print_state)
2811 					verbose("\nfrom %d to %d: safe\n",
2812 						prev_insn_idx, insn_idx);
2813 				else
2814 					verbose("%d: safe\n", insn_idx);
2815 			}
2816 			goto process_bpf_exit;
2817 		}
2818 
2819 		if (log_level && do_print_state) {
2820 			verbose("\nfrom %d to %d:", prev_insn_idx, insn_idx);
2821 			print_verifier_state(&env->cur_state);
2822 			do_print_state = false;
2823 		}
2824 
2825 		if (log_level) {
2826 			verbose("%d: ", insn_idx);
2827 			print_bpf_insn(insn);
2828 		}
2829 
2830 		err = ext_analyzer_insn_hook(env, insn_idx, prev_insn_idx);
2831 		if (err)
2832 			return err;
2833 
2834 		if (class == BPF_ALU || class == BPF_ALU64) {
2835 			err = check_alu_op(env, insn);
2836 			if (err)
2837 				return err;
2838 
2839 		} else if (class == BPF_LDX) {
2840 			enum bpf_reg_type *prev_src_type, src_reg_type;
2841 
2842 			/* check for reserved fields is already done */
2843 
2844 			/* check src operand */
2845 			err = check_reg_arg(regs, insn->src_reg, SRC_OP);
2846 			if (err)
2847 				return err;
2848 
2849 			err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK);
2850 			if (err)
2851 				return err;
2852 
2853 			src_reg_type = regs[insn->src_reg].type;
2854 
2855 			/* check that memory (src_reg + off) is readable,
2856 			 * the state of dst_reg will be updated by this func
2857 			 */
2858 			err = check_mem_access(env, insn->src_reg, insn->off,
2859 					       BPF_SIZE(insn->code), BPF_READ,
2860 					       insn->dst_reg);
2861 			if (err)
2862 				return err;
2863 
2864 			if (BPF_SIZE(insn->code) != BPF_W &&
2865 			    BPF_SIZE(insn->code) != BPF_DW) {
2866 				insn_idx++;
2867 				continue;
2868 			}
2869 
2870 			prev_src_type = &env->insn_aux_data[insn_idx].ptr_type;
2871 
2872 			if (*prev_src_type == NOT_INIT) {
2873 				/* saw a valid insn
2874 				 * dst_reg = *(u32 *)(src_reg + off)
2875 				 * save type to validate intersecting paths
2876 				 */
2877 				*prev_src_type = src_reg_type;
2878 
2879 			} else if (src_reg_type != *prev_src_type &&
2880 				   (src_reg_type == PTR_TO_CTX ||
2881 				    *prev_src_type == PTR_TO_CTX)) {
2882 				/* ABuser program is trying to use the same insn
2883 				 * dst_reg = *(u32*) (src_reg + off)
2884 				 * with different pointer types:
2885 				 * src_reg == ctx in one branch and
2886 				 * src_reg == stack|map in some other branch.
2887 				 * Reject it.
2888 				 */
2889 				verbose("same insn cannot be used with different pointers\n");
2890 				return -EINVAL;
2891 			}
2892 
2893 		} else if (class == BPF_STX) {
2894 			enum bpf_reg_type *prev_dst_type, dst_reg_type;
2895 
2896 			if (BPF_MODE(insn->code) == BPF_XADD) {
2897 				err = check_xadd(env, insn);
2898 				if (err)
2899 					return err;
2900 				insn_idx++;
2901 				continue;
2902 			}
2903 
2904 			/* check src1 operand */
2905 			err = check_reg_arg(regs, insn->src_reg, SRC_OP);
2906 			if (err)
2907 				return err;
2908 			/* check src2 operand */
2909 			err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
2910 			if (err)
2911 				return err;
2912 
2913 			dst_reg_type = regs[insn->dst_reg].type;
2914 
2915 			/* check that memory (dst_reg + off) is writeable */
2916 			err = check_mem_access(env, insn->dst_reg, insn->off,
2917 					       BPF_SIZE(insn->code), BPF_WRITE,
2918 					       insn->src_reg);
2919 			if (err)
2920 				return err;
2921 
2922 			prev_dst_type = &env->insn_aux_data[insn_idx].ptr_type;
2923 
2924 			if (*prev_dst_type == NOT_INIT) {
2925 				*prev_dst_type = dst_reg_type;
2926 			} else if (dst_reg_type != *prev_dst_type &&
2927 				   (dst_reg_type == PTR_TO_CTX ||
2928 				    *prev_dst_type == PTR_TO_CTX)) {
2929 				verbose("same insn cannot be used with different pointers\n");
2930 				return -EINVAL;
2931 			}
2932 
2933 		} else if (class == BPF_ST) {
2934 			if (BPF_MODE(insn->code) != BPF_MEM ||
2935 			    insn->src_reg != BPF_REG_0) {
2936 				verbose("BPF_ST uses reserved fields\n");
2937 				return -EINVAL;
2938 			}
2939 			/* check src operand */
2940 			err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
2941 			if (err)
2942 				return err;
2943 
2944 			/* check that memory (dst_reg + off) is writeable */
2945 			err = check_mem_access(env, insn->dst_reg, insn->off,
2946 					       BPF_SIZE(insn->code), BPF_WRITE,
2947 					       -1);
2948 			if (err)
2949 				return err;
2950 
2951 		} else if (class == BPF_JMP) {
2952 			u8 opcode = BPF_OP(insn->code);
2953 
2954 			if (opcode == BPF_CALL) {
2955 				if (BPF_SRC(insn->code) != BPF_K ||
2956 				    insn->off != 0 ||
2957 				    insn->src_reg != BPF_REG_0 ||
2958 				    insn->dst_reg != BPF_REG_0) {
2959 					verbose("BPF_CALL uses reserved fields\n");
2960 					return -EINVAL;
2961 				}
2962 
2963 				err = check_call(env, insn->imm);
2964 				if (err)
2965 					return err;
2966 
2967 			} else if (opcode == BPF_JA) {
2968 				if (BPF_SRC(insn->code) != BPF_K ||
2969 				    insn->imm != 0 ||
2970 				    insn->src_reg != BPF_REG_0 ||
2971 				    insn->dst_reg != BPF_REG_0) {
2972 					verbose("BPF_JA uses reserved fields\n");
2973 					return -EINVAL;
2974 				}
2975 
2976 				insn_idx += insn->off + 1;
2977 				continue;
2978 
2979 			} else if (opcode == BPF_EXIT) {
2980 				if (BPF_SRC(insn->code) != BPF_K ||
2981 				    insn->imm != 0 ||
2982 				    insn->src_reg != BPF_REG_0 ||
2983 				    insn->dst_reg != BPF_REG_0) {
2984 					verbose("BPF_EXIT uses reserved fields\n");
2985 					return -EINVAL;
2986 				}
2987 
2988 				/* eBPF calling convetion is such that R0 is used
2989 				 * to return the value from eBPF program.
2990 				 * Make sure that it's readable at this time
2991 				 * of bpf_exit, which means that program wrote
2992 				 * something into it earlier
2993 				 */
2994 				err = check_reg_arg(regs, BPF_REG_0, SRC_OP);
2995 				if (err)
2996 					return err;
2997 
2998 				if (is_pointer_value(env, BPF_REG_0)) {
2999 					verbose("R0 leaks addr as return value\n");
3000 					return -EACCES;
3001 				}
3002 
3003 process_bpf_exit:
3004 				insn_idx = pop_stack(env, &prev_insn_idx);
3005 				if (insn_idx < 0) {
3006 					break;
3007 				} else {
3008 					do_print_state = true;
3009 					continue;
3010 				}
3011 			} else {
3012 				err = check_cond_jmp_op(env, insn, &insn_idx);
3013 				if (err)
3014 					return err;
3015 			}
3016 		} else if (class == BPF_LD) {
3017 			u8 mode = BPF_MODE(insn->code);
3018 
3019 			if (mode == BPF_ABS || mode == BPF_IND) {
3020 				err = check_ld_abs(env, insn);
3021 				if (err)
3022 					return err;
3023 
3024 			} else if (mode == BPF_IMM) {
3025 				err = check_ld_imm(env, insn);
3026 				if (err)
3027 					return err;
3028 
3029 				insn_idx++;
3030 			} else {
3031 				verbose("invalid BPF_LD mode\n");
3032 				return -EINVAL;
3033 			}
3034 			reset_reg_range_values(regs, insn->dst_reg);
3035 		} else {
3036 			verbose("unknown insn class %d\n", class);
3037 			return -EINVAL;
3038 		}
3039 
3040 		insn_idx++;
3041 	}
3042 
3043 	verbose("processed %d insns\n", insn_processed);
3044 	return 0;
3045 }
3046 
3047 static int check_map_prog_compatibility(struct bpf_map *map,
3048 					struct bpf_prog *prog)
3049 
3050 {
3051 	if (prog->type == BPF_PROG_TYPE_PERF_EVENT &&
3052 	    (map->map_type == BPF_MAP_TYPE_HASH ||
3053 	     map->map_type == BPF_MAP_TYPE_PERCPU_HASH) &&
3054 	    (map->map_flags & BPF_F_NO_PREALLOC)) {
3055 		verbose("perf_event programs can only use preallocated hash map\n");
3056 		return -EINVAL;
3057 	}
3058 	return 0;
3059 }
3060 
3061 /* look for pseudo eBPF instructions that access map FDs and
3062  * replace them with actual map pointers
3063  */
3064 static int replace_map_fd_with_map_ptr(struct bpf_verifier_env *env)
3065 {
3066 	struct bpf_insn *insn = env->prog->insnsi;
3067 	int insn_cnt = env->prog->len;
3068 	int i, j, err;
3069 
3070 	err = bpf_prog_calc_tag(env->prog);
3071 	if (err)
3072 		return err;
3073 
3074 	for (i = 0; i < insn_cnt; i++, insn++) {
3075 		if (BPF_CLASS(insn->code) == BPF_LDX &&
3076 		    (BPF_MODE(insn->code) != BPF_MEM || insn->imm != 0)) {
3077 			verbose("BPF_LDX uses reserved fields\n");
3078 			return -EINVAL;
3079 		}
3080 
3081 		if (BPF_CLASS(insn->code) == BPF_STX &&
3082 		    ((BPF_MODE(insn->code) != BPF_MEM &&
3083 		      BPF_MODE(insn->code) != BPF_XADD) || insn->imm != 0)) {
3084 			verbose("BPF_STX uses reserved fields\n");
3085 			return -EINVAL;
3086 		}
3087 
3088 		if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW)) {
3089 			struct bpf_map *map;
3090 			struct fd f;
3091 
3092 			if (i == insn_cnt - 1 || insn[1].code != 0 ||
3093 			    insn[1].dst_reg != 0 || insn[1].src_reg != 0 ||
3094 			    insn[1].off != 0) {
3095 				verbose("invalid bpf_ld_imm64 insn\n");
3096 				return -EINVAL;
3097 			}
3098 
3099 			if (insn->src_reg == 0)
3100 				/* valid generic load 64-bit imm */
3101 				goto next_insn;
3102 
3103 			if (insn->src_reg != BPF_PSEUDO_MAP_FD) {
3104 				verbose("unrecognized bpf_ld_imm64 insn\n");
3105 				return -EINVAL;
3106 			}
3107 
3108 			f = fdget(insn->imm);
3109 			map = __bpf_map_get(f);
3110 			if (IS_ERR(map)) {
3111 				verbose("fd %d is not pointing to valid bpf_map\n",
3112 					insn->imm);
3113 				return PTR_ERR(map);
3114 			}
3115 
3116 			err = check_map_prog_compatibility(map, env->prog);
3117 			if (err) {
3118 				fdput(f);
3119 				return err;
3120 			}
3121 
3122 			/* store map pointer inside BPF_LD_IMM64 instruction */
3123 			insn[0].imm = (u32) (unsigned long) map;
3124 			insn[1].imm = ((u64) (unsigned long) map) >> 32;
3125 
3126 			/* check whether we recorded this map already */
3127 			for (j = 0; j < env->used_map_cnt; j++)
3128 				if (env->used_maps[j] == map) {
3129 					fdput(f);
3130 					goto next_insn;
3131 				}
3132 
3133 			if (env->used_map_cnt >= MAX_USED_MAPS) {
3134 				fdput(f);
3135 				return -E2BIG;
3136 			}
3137 
3138 			/* hold the map. If the program is rejected by verifier,
3139 			 * the map will be released by release_maps() or it
3140 			 * will be used by the valid program until it's unloaded
3141 			 * and all maps are released in free_bpf_prog_info()
3142 			 */
3143 			map = bpf_map_inc(map, false);
3144 			if (IS_ERR(map)) {
3145 				fdput(f);
3146 				return PTR_ERR(map);
3147 			}
3148 			env->used_maps[env->used_map_cnt++] = map;
3149 
3150 			fdput(f);
3151 next_insn:
3152 			insn++;
3153 			i++;
3154 		}
3155 	}
3156 
3157 	/* now all pseudo BPF_LD_IMM64 instructions load valid
3158 	 * 'struct bpf_map *' into a register instead of user map_fd.
3159 	 * These pointers will be used later by verifier to validate map access.
3160 	 */
3161 	return 0;
3162 }
3163 
3164 /* drop refcnt of maps used by the rejected program */
3165 static void release_maps(struct bpf_verifier_env *env)
3166 {
3167 	int i;
3168 
3169 	for (i = 0; i < env->used_map_cnt; i++)
3170 		bpf_map_put(env->used_maps[i]);
3171 }
3172 
3173 /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */
3174 static void convert_pseudo_ld_imm64(struct bpf_verifier_env *env)
3175 {
3176 	struct bpf_insn *insn = env->prog->insnsi;
3177 	int insn_cnt = env->prog->len;
3178 	int i;
3179 
3180 	for (i = 0; i < insn_cnt; i++, insn++)
3181 		if (insn->code == (BPF_LD | BPF_IMM | BPF_DW))
3182 			insn->src_reg = 0;
3183 }
3184 
3185 /* convert load instructions that access fields of 'struct __sk_buff'
3186  * into sequence of instructions that access fields of 'struct sk_buff'
3187  */
3188 static int convert_ctx_accesses(struct bpf_verifier_env *env)
3189 {
3190 	const struct bpf_verifier_ops *ops = env->prog->aux->ops;
3191 	const int insn_cnt = env->prog->len;
3192 	struct bpf_insn insn_buf[16], *insn;
3193 	struct bpf_prog *new_prog;
3194 	enum bpf_access_type type;
3195 	int i, cnt, delta = 0;
3196 
3197 	if (ops->gen_prologue) {
3198 		cnt = ops->gen_prologue(insn_buf, env->seen_direct_write,
3199 					env->prog);
3200 		if (cnt >= ARRAY_SIZE(insn_buf)) {
3201 			verbose("bpf verifier is misconfigured\n");
3202 			return -EINVAL;
3203 		} else if (cnt) {
3204 			new_prog = bpf_patch_insn_single(env->prog, 0,
3205 							 insn_buf, cnt);
3206 			if (!new_prog)
3207 				return -ENOMEM;
3208 			env->prog = new_prog;
3209 			delta += cnt - 1;
3210 		}
3211 	}
3212 
3213 	if (!ops->convert_ctx_access)
3214 		return 0;
3215 
3216 	insn = env->prog->insnsi + delta;
3217 
3218 	for (i = 0; i < insn_cnt; i++, insn++) {
3219 		if (insn->code == (BPF_LDX | BPF_MEM | BPF_B) ||
3220 		    insn->code == (BPF_LDX | BPF_MEM | BPF_H) ||
3221 		    insn->code == (BPF_LDX | BPF_MEM | BPF_W) ||
3222 		    insn->code == (BPF_LDX | BPF_MEM | BPF_DW))
3223 			type = BPF_READ;
3224 		else if (insn->code == (BPF_STX | BPF_MEM | BPF_B) ||
3225 			 insn->code == (BPF_STX | BPF_MEM | BPF_H) ||
3226 			 insn->code == (BPF_STX | BPF_MEM | BPF_W) ||
3227 			 insn->code == (BPF_STX | BPF_MEM | BPF_DW))
3228 			type = BPF_WRITE;
3229 		else
3230 			continue;
3231 
3232 		if (env->insn_aux_data[i].ptr_type != PTR_TO_CTX)
3233 			continue;
3234 
3235 		cnt = ops->convert_ctx_access(type, insn, insn_buf, env->prog);
3236 		if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) {
3237 			verbose("bpf verifier is misconfigured\n");
3238 			return -EINVAL;
3239 		}
3240 
3241 		new_prog = bpf_patch_insn_single(env->prog, i + delta, insn_buf,
3242 						 cnt);
3243 		if (!new_prog)
3244 			return -ENOMEM;
3245 
3246 		delta += cnt - 1;
3247 
3248 		/* keep walking new program and skip insns we just inserted */
3249 		env->prog = new_prog;
3250 		insn      = new_prog->insnsi + i + delta;
3251 	}
3252 
3253 	return 0;
3254 }
3255 
3256 static void free_states(struct bpf_verifier_env *env)
3257 {
3258 	struct bpf_verifier_state_list *sl, *sln;
3259 	int i;
3260 
3261 	if (!env->explored_states)
3262 		return;
3263 
3264 	for (i = 0; i < env->prog->len; i++) {
3265 		sl = env->explored_states[i];
3266 
3267 		if (sl)
3268 			while (sl != STATE_LIST_MARK) {
3269 				sln = sl->next;
3270 				kfree(sl);
3271 				sl = sln;
3272 			}
3273 	}
3274 
3275 	kfree(env->explored_states);
3276 }
3277 
3278 int bpf_check(struct bpf_prog **prog, union bpf_attr *attr)
3279 {
3280 	char __user *log_ubuf = NULL;
3281 	struct bpf_verifier_env *env;
3282 	int ret = -EINVAL;
3283 
3284 	/* 'struct bpf_verifier_env' can be global, but since it's not small,
3285 	 * allocate/free it every time bpf_check() is called
3286 	 */
3287 	env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL);
3288 	if (!env)
3289 		return -ENOMEM;
3290 
3291 	env->insn_aux_data = vzalloc(sizeof(struct bpf_insn_aux_data) *
3292 				     (*prog)->len);
3293 	ret = -ENOMEM;
3294 	if (!env->insn_aux_data)
3295 		goto err_free_env;
3296 	env->prog = *prog;
3297 
3298 	/* grab the mutex to protect few globals used by verifier */
3299 	mutex_lock(&bpf_verifier_lock);
3300 
3301 	if (attr->log_level || attr->log_buf || attr->log_size) {
3302 		/* user requested verbose verifier output
3303 		 * and supplied buffer to store the verification trace
3304 		 */
3305 		log_level = attr->log_level;
3306 		log_ubuf = (char __user *) (unsigned long) attr->log_buf;
3307 		log_size = attr->log_size;
3308 		log_len = 0;
3309 
3310 		ret = -EINVAL;
3311 		/* log_* values have to be sane */
3312 		if (log_size < 128 || log_size > UINT_MAX >> 8 ||
3313 		    log_level == 0 || log_ubuf == NULL)
3314 			goto err_unlock;
3315 
3316 		ret = -ENOMEM;
3317 		log_buf = vmalloc(log_size);
3318 		if (!log_buf)
3319 			goto err_unlock;
3320 	} else {
3321 		log_level = 0;
3322 	}
3323 
3324 	ret = replace_map_fd_with_map_ptr(env);
3325 	if (ret < 0)
3326 		goto skip_full_check;
3327 
3328 	env->explored_states = kcalloc(env->prog->len,
3329 				       sizeof(struct bpf_verifier_state_list *),
3330 				       GFP_USER);
3331 	ret = -ENOMEM;
3332 	if (!env->explored_states)
3333 		goto skip_full_check;
3334 
3335 	ret = check_cfg(env);
3336 	if (ret < 0)
3337 		goto skip_full_check;
3338 
3339 	env->allow_ptr_leaks = capable(CAP_SYS_ADMIN);
3340 
3341 	ret = do_check(env);
3342 
3343 skip_full_check:
3344 	while (pop_stack(env, NULL) >= 0);
3345 	free_states(env);
3346 
3347 	if (ret == 0)
3348 		/* program is valid, convert *(u32*)(ctx + off) accesses */
3349 		ret = convert_ctx_accesses(env);
3350 
3351 	if (log_level && log_len >= log_size - 1) {
3352 		BUG_ON(log_len >= log_size);
3353 		/* verifier log exceeded user supplied buffer */
3354 		ret = -ENOSPC;
3355 		/* fall through to return what was recorded */
3356 	}
3357 
3358 	/* copy verifier log back to user space including trailing zero */
3359 	if (log_level && copy_to_user(log_ubuf, log_buf, log_len + 1) != 0) {
3360 		ret = -EFAULT;
3361 		goto free_log_buf;
3362 	}
3363 
3364 	if (ret == 0 && env->used_map_cnt) {
3365 		/* if program passed verifier, update used_maps in bpf_prog_info */
3366 		env->prog->aux->used_maps = kmalloc_array(env->used_map_cnt,
3367 							  sizeof(env->used_maps[0]),
3368 							  GFP_KERNEL);
3369 
3370 		if (!env->prog->aux->used_maps) {
3371 			ret = -ENOMEM;
3372 			goto free_log_buf;
3373 		}
3374 
3375 		memcpy(env->prog->aux->used_maps, env->used_maps,
3376 		       sizeof(env->used_maps[0]) * env->used_map_cnt);
3377 		env->prog->aux->used_map_cnt = env->used_map_cnt;
3378 
3379 		/* program is valid. Convert pseudo bpf_ld_imm64 into generic
3380 		 * bpf_ld_imm64 instructions
3381 		 */
3382 		convert_pseudo_ld_imm64(env);
3383 	}
3384 
3385 free_log_buf:
3386 	if (log_level)
3387 		vfree(log_buf);
3388 	if (!env->prog->aux->used_maps)
3389 		/* if we didn't copy map pointers into bpf_prog_info, release
3390 		 * them now. Otherwise free_bpf_prog_info() will release them.
3391 		 */
3392 		release_maps(env);
3393 	*prog = env->prog;
3394 err_unlock:
3395 	mutex_unlock(&bpf_verifier_lock);
3396 	vfree(env->insn_aux_data);
3397 err_free_env:
3398 	kfree(env);
3399 	return ret;
3400 }
3401 
3402 int bpf_analyzer(struct bpf_prog *prog, const struct bpf_ext_analyzer_ops *ops,
3403 		 void *priv)
3404 {
3405 	struct bpf_verifier_env *env;
3406 	int ret;
3407 
3408 	env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL);
3409 	if (!env)
3410 		return -ENOMEM;
3411 
3412 	env->insn_aux_data = vzalloc(sizeof(struct bpf_insn_aux_data) *
3413 				     prog->len);
3414 	ret = -ENOMEM;
3415 	if (!env->insn_aux_data)
3416 		goto err_free_env;
3417 	env->prog = prog;
3418 	env->analyzer_ops = ops;
3419 	env->analyzer_priv = priv;
3420 
3421 	/* grab the mutex to protect few globals used by verifier */
3422 	mutex_lock(&bpf_verifier_lock);
3423 
3424 	log_level = 0;
3425 
3426 	env->explored_states = kcalloc(env->prog->len,
3427 				       sizeof(struct bpf_verifier_state_list *),
3428 				       GFP_KERNEL);
3429 	ret = -ENOMEM;
3430 	if (!env->explored_states)
3431 		goto skip_full_check;
3432 
3433 	ret = check_cfg(env);
3434 	if (ret < 0)
3435 		goto skip_full_check;
3436 
3437 	env->allow_ptr_leaks = capable(CAP_SYS_ADMIN);
3438 
3439 	ret = do_check(env);
3440 
3441 skip_full_check:
3442 	while (pop_stack(env, NULL) >= 0);
3443 	free_states(env);
3444 
3445 	mutex_unlock(&bpf_verifier_lock);
3446 	vfree(env->insn_aux_data);
3447 err_free_env:
3448 	kfree(env);
3449 	return ret;
3450 }
3451 EXPORT_SYMBOL_GPL(bpf_analyzer);
3452