xref: /linux/kernel/bpf/verifier.c (revision 372e2db7210df7c45ead46429aeb1443ba148060)
1 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
2  * Copyright (c) 2016 Facebook
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of version 2 of the GNU General Public
6  * License as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful, but
9  * WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11  * General Public License for more details.
12  */
13 #include <linux/kernel.h>
14 #include <linux/types.h>
15 #include <linux/slab.h>
16 #include <linux/bpf.h>
17 #include <linux/bpf_verifier.h>
18 #include <linux/filter.h>
19 #include <net/netlink.h>
20 #include <linux/file.h>
21 #include <linux/vmalloc.h>
22 #include <linux/stringify.h>
23 
24 /* bpf_check() is a static code analyzer that walks eBPF program
25  * instruction by instruction and updates register/stack state.
26  * All paths of conditional branches are analyzed until 'bpf_exit' insn.
27  *
28  * The first pass is depth-first-search to check that the program is a DAG.
29  * It rejects the following programs:
30  * - larger than BPF_MAXINSNS insns
31  * - if loop is present (detected via back-edge)
32  * - unreachable insns exist (shouldn't be a forest. program = one function)
33  * - out of bounds or malformed jumps
34  * The second pass is all possible path descent from the 1st insn.
35  * Since it's analyzing all pathes through the program, the length of the
36  * analysis is limited to 32k insn, which may be hit even if total number of
37  * insn is less then 4K, but there are too many branches that change stack/regs.
38  * Number of 'branches to be analyzed' is limited to 1k
39  *
40  * On entry to each instruction, each register has a type, and the instruction
41  * changes the types of the registers depending on instruction semantics.
42  * If instruction is BPF_MOV64_REG(BPF_REG_1, BPF_REG_5), then type of R5 is
43  * copied to R1.
44  *
45  * All registers are 64-bit.
46  * R0 - return register
47  * R1-R5 argument passing registers
48  * R6-R9 callee saved registers
49  * R10 - frame pointer read-only
50  *
51  * At the start of BPF program the register R1 contains a pointer to bpf_context
52  * and has type PTR_TO_CTX.
53  *
54  * Verifier tracks arithmetic operations on pointers in case:
55  *    BPF_MOV64_REG(BPF_REG_1, BPF_REG_10),
56  *    BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -20),
57  * 1st insn copies R10 (which has FRAME_PTR) type into R1
58  * and 2nd arithmetic instruction is pattern matched to recognize
59  * that it wants to construct a pointer to some element within stack.
60  * So after 2nd insn, the register R1 has type PTR_TO_STACK
61  * (and -20 constant is saved for further stack bounds checking).
62  * Meaning that this reg is a pointer to stack plus known immediate constant.
63  *
64  * Most of the time the registers have UNKNOWN_VALUE type, which
65  * means the register has some value, but it's not a valid pointer.
66  * (like pointer plus pointer becomes UNKNOWN_VALUE type)
67  *
68  * When verifier sees load or store instructions the type of base register
69  * can be: PTR_TO_MAP_VALUE, PTR_TO_CTX, FRAME_PTR. These are three pointer
70  * types recognized by check_mem_access() function.
71  *
72  * PTR_TO_MAP_VALUE means that this register is pointing to 'map element value'
73  * and the range of [ptr, ptr + map's value_size) is accessible.
74  *
75  * registers used to pass values to function calls are checked against
76  * function argument constraints.
77  *
78  * ARG_PTR_TO_MAP_KEY is one of such argument constraints.
79  * It means that the register type passed to this function must be
80  * PTR_TO_STACK and it will be used inside the function as
81  * 'pointer to map element key'
82  *
83  * For example the argument constraints for bpf_map_lookup_elem():
84  *   .ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL,
85  *   .arg1_type = ARG_CONST_MAP_PTR,
86  *   .arg2_type = ARG_PTR_TO_MAP_KEY,
87  *
88  * ret_type says that this function returns 'pointer to map elem value or null'
89  * function expects 1st argument to be a const pointer to 'struct bpf_map' and
90  * 2nd argument should be a pointer to stack, which will be used inside
91  * the helper function as a pointer to map element key.
92  *
93  * On the kernel side the helper function looks like:
94  * u64 bpf_map_lookup_elem(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
95  * {
96  *    struct bpf_map *map = (struct bpf_map *) (unsigned long) r1;
97  *    void *key = (void *) (unsigned long) r2;
98  *    void *value;
99  *
100  *    here kernel can access 'key' and 'map' pointers safely, knowing that
101  *    [key, key + map->key_size) bytes are valid and were initialized on
102  *    the stack of eBPF program.
103  * }
104  *
105  * Corresponding eBPF program may look like:
106  *    BPF_MOV64_REG(BPF_REG_2, BPF_REG_10),  // after this insn R2 type is FRAME_PTR
107  *    BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -4), // after this insn R2 type is PTR_TO_STACK
108  *    BPF_LD_MAP_FD(BPF_REG_1, map_fd),      // after this insn R1 type is CONST_PTR_TO_MAP
109  *    BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem),
110  * here verifier looks at prototype of map_lookup_elem() and sees:
111  * .arg1_type == ARG_CONST_MAP_PTR and R1->type == CONST_PTR_TO_MAP, which is ok,
112  * Now verifier knows that this map has key of R1->map_ptr->key_size bytes
113  *
114  * Then .arg2_type == ARG_PTR_TO_MAP_KEY and R2->type == PTR_TO_STACK, ok so far,
115  * Now verifier checks that [R2, R2 + map's key_size) are within stack limits
116  * and were initialized prior to this call.
117  * If it's ok, then verifier allows this BPF_CALL insn and looks at
118  * .ret_type which is RET_PTR_TO_MAP_VALUE_OR_NULL, so it sets
119  * R0->type = PTR_TO_MAP_VALUE_OR_NULL which means bpf_map_lookup_elem() function
120  * returns ether pointer to map value or NULL.
121  *
122  * When type PTR_TO_MAP_VALUE_OR_NULL passes through 'if (reg != 0) goto +off'
123  * insn, the register holding that pointer in the true branch changes state to
124  * PTR_TO_MAP_VALUE and the same register changes state to CONST_IMM in the false
125  * branch. See check_cond_jmp_op().
126  *
127  * After the call R0 is set to return type of the function and registers R1-R5
128  * are set to NOT_INIT to indicate that they are no longer readable.
129  */
130 
131 /* verifier_state + insn_idx are pushed to stack when branch is encountered */
132 struct bpf_verifier_stack_elem {
133 	/* verifer state is 'st'
134 	 * before processing instruction 'insn_idx'
135 	 * and after processing instruction 'prev_insn_idx'
136 	 */
137 	struct bpf_verifier_state st;
138 	int insn_idx;
139 	int prev_insn_idx;
140 	struct bpf_verifier_stack_elem *next;
141 };
142 
143 #define BPF_COMPLEXITY_LIMIT_INSNS	65536
144 #define BPF_COMPLEXITY_LIMIT_STACK	1024
145 
146 struct bpf_call_arg_meta {
147 	struct bpf_map *map_ptr;
148 	bool raw_mode;
149 	bool pkt_access;
150 	int regno;
151 	int access_size;
152 };
153 
154 /* verbose verifier prints what it's seeing
155  * bpf_check() is called under lock, so no race to access these global vars
156  */
157 static u32 log_level, log_size, log_len;
158 static char *log_buf;
159 
160 static DEFINE_MUTEX(bpf_verifier_lock);
161 
162 /* log_level controls verbosity level of eBPF verifier.
163  * verbose() is used to dump the verification trace to the log, so the user
164  * can figure out what's wrong with the program
165  */
166 static __printf(1, 2) void verbose(const char *fmt, ...)
167 {
168 	va_list args;
169 
170 	if (log_level == 0 || log_len >= log_size - 1)
171 		return;
172 
173 	va_start(args, fmt);
174 	log_len += vscnprintf(log_buf + log_len, log_size - log_len, fmt, args);
175 	va_end(args);
176 }
177 
178 /* string representation of 'enum bpf_reg_type' */
179 static const char * const reg_type_str[] = {
180 	[NOT_INIT]		= "?",
181 	[UNKNOWN_VALUE]		= "inv",
182 	[PTR_TO_CTX]		= "ctx",
183 	[CONST_PTR_TO_MAP]	= "map_ptr",
184 	[PTR_TO_MAP_VALUE]	= "map_value",
185 	[PTR_TO_MAP_VALUE_OR_NULL] = "map_value_or_null",
186 	[PTR_TO_MAP_VALUE_ADJ]	= "map_value_adj",
187 	[FRAME_PTR]		= "fp",
188 	[PTR_TO_STACK]		= "fp",
189 	[CONST_IMM]		= "imm",
190 	[PTR_TO_PACKET]		= "pkt",
191 	[PTR_TO_PACKET_END]	= "pkt_end",
192 };
193 
194 #define __BPF_FUNC_STR_FN(x) [BPF_FUNC_ ## x] = __stringify(bpf_ ## x)
195 static const char * const func_id_str[] = {
196 	__BPF_FUNC_MAPPER(__BPF_FUNC_STR_FN)
197 };
198 #undef __BPF_FUNC_STR_FN
199 
200 static const char *func_id_name(int id)
201 {
202 	BUILD_BUG_ON(ARRAY_SIZE(func_id_str) != __BPF_FUNC_MAX_ID);
203 
204 	if (id >= 0 && id < __BPF_FUNC_MAX_ID && func_id_str[id])
205 		return func_id_str[id];
206 	else
207 		return "unknown";
208 }
209 
210 static void print_verifier_state(struct bpf_verifier_state *state)
211 {
212 	struct bpf_reg_state *reg;
213 	enum bpf_reg_type t;
214 	int i;
215 
216 	for (i = 0; i < MAX_BPF_REG; i++) {
217 		reg = &state->regs[i];
218 		t = reg->type;
219 		if (t == NOT_INIT)
220 			continue;
221 		verbose(" R%d=%s", i, reg_type_str[t]);
222 		if (t == CONST_IMM || t == PTR_TO_STACK)
223 			verbose("%lld", reg->imm);
224 		else if (t == PTR_TO_PACKET)
225 			verbose("(id=%d,off=%d,r=%d)",
226 				reg->id, reg->off, reg->range);
227 		else if (t == UNKNOWN_VALUE && reg->imm)
228 			verbose("%lld", reg->imm);
229 		else if (t == CONST_PTR_TO_MAP || t == PTR_TO_MAP_VALUE ||
230 			 t == PTR_TO_MAP_VALUE_OR_NULL ||
231 			 t == PTR_TO_MAP_VALUE_ADJ)
232 			verbose("(ks=%d,vs=%d,id=%u)",
233 				reg->map_ptr->key_size,
234 				reg->map_ptr->value_size,
235 				reg->id);
236 		if (reg->min_value != BPF_REGISTER_MIN_RANGE)
237 			verbose(",min_value=%lld",
238 				(long long)reg->min_value);
239 		if (reg->max_value != BPF_REGISTER_MAX_RANGE)
240 			verbose(",max_value=%llu",
241 				(unsigned long long)reg->max_value);
242 	}
243 	for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
244 		if (state->stack_slot_type[i] == STACK_SPILL)
245 			verbose(" fp%d=%s", -MAX_BPF_STACK + i,
246 				reg_type_str[state->spilled_regs[i / BPF_REG_SIZE].type]);
247 	}
248 	verbose("\n");
249 }
250 
251 static const char *const bpf_class_string[] = {
252 	[BPF_LD]    = "ld",
253 	[BPF_LDX]   = "ldx",
254 	[BPF_ST]    = "st",
255 	[BPF_STX]   = "stx",
256 	[BPF_ALU]   = "alu",
257 	[BPF_JMP]   = "jmp",
258 	[BPF_RET]   = "BUG",
259 	[BPF_ALU64] = "alu64",
260 };
261 
262 static const char *const bpf_alu_string[16] = {
263 	[BPF_ADD >> 4]  = "+=",
264 	[BPF_SUB >> 4]  = "-=",
265 	[BPF_MUL >> 4]  = "*=",
266 	[BPF_DIV >> 4]  = "/=",
267 	[BPF_OR  >> 4]  = "|=",
268 	[BPF_AND >> 4]  = "&=",
269 	[BPF_LSH >> 4]  = "<<=",
270 	[BPF_RSH >> 4]  = ">>=",
271 	[BPF_NEG >> 4]  = "neg",
272 	[BPF_MOD >> 4]  = "%=",
273 	[BPF_XOR >> 4]  = "^=",
274 	[BPF_MOV >> 4]  = "=",
275 	[BPF_ARSH >> 4] = "s>>=",
276 	[BPF_END >> 4]  = "endian",
277 };
278 
279 static const char *const bpf_ldst_string[] = {
280 	[BPF_W >> 3]  = "u32",
281 	[BPF_H >> 3]  = "u16",
282 	[BPF_B >> 3]  = "u8",
283 	[BPF_DW >> 3] = "u64",
284 };
285 
286 static const char *const bpf_jmp_string[16] = {
287 	[BPF_JA >> 4]   = "jmp",
288 	[BPF_JEQ >> 4]  = "==",
289 	[BPF_JGT >> 4]  = ">",
290 	[BPF_JGE >> 4]  = ">=",
291 	[BPF_JSET >> 4] = "&",
292 	[BPF_JNE >> 4]  = "!=",
293 	[BPF_JSGT >> 4] = "s>",
294 	[BPF_JSGE >> 4] = "s>=",
295 	[BPF_CALL >> 4] = "call",
296 	[BPF_EXIT >> 4] = "exit",
297 };
298 
299 static void print_bpf_insn(struct bpf_insn *insn)
300 {
301 	u8 class = BPF_CLASS(insn->code);
302 
303 	if (class == BPF_ALU || class == BPF_ALU64) {
304 		if (BPF_SRC(insn->code) == BPF_X)
305 			verbose("(%02x) %sr%d %s %sr%d\n",
306 				insn->code, class == BPF_ALU ? "(u32) " : "",
307 				insn->dst_reg,
308 				bpf_alu_string[BPF_OP(insn->code) >> 4],
309 				class == BPF_ALU ? "(u32) " : "",
310 				insn->src_reg);
311 		else
312 			verbose("(%02x) %sr%d %s %s%d\n",
313 				insn->code, class == BPF_ALU ? "(u32) " : "",
314 				insn->dst_reg,
315 				bpf_alu_string[BPF_OP(insn->code) >> 4],
316 				class == BPF_ALU ? "(u32) " : "",
317 				insn->imm);
318 	} else if (class == BPF_STX) {
319 		if (BPF_MODE(insn->code) == BPF_MEM)
320 			verbose("(%02x) *(%s *)(r%d %+d) = r%d\n",
321 				insn->code,
322 				bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
323 				insn->dst_reg,
324 				insn->off, insn->src_reg);
325 		else if (BPF_MODE(insn->code) == BPF_XADD)
326 			verbose("(%02x) lock *(%s *)(r%d %+d) += r%d\n",
327 				insn->code,
328 				bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
329 				insn->dst_reg, insn->off,
330 				insn->src_reg);
331 		else
332 			verbose("BUG_%02x\n", insn->code);
333 	} else if (class == BPF_ST) {
334 		if (BPF_MODE(insn->code) != BPF_MEM) {
335 			verbose("BUG_st_%02x\n", insn->code);
336 			return;
337 		}
338 		verbose("(%02x) *(%s *)(r%d %+d) = %d\n",
339 			insn->code,
340 			bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
341 			insn->dst_reg,
342 			insn->off, insn->imm);
343 	} else if (class == BPF_LDX) {
344 		if (BPF_MODE(insn->code) != BPF_MEM) {
345 			verbose("BUG_ldx_%02x\n", insn->code);
346 			return;
347 		}
348 		verbose("(%02x) r%d = *(%s *)(r%d %+d)\n",
349 			insn->code, insn->dst_reg,
350 			bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
351 			insn->src_reg, insn->off);
352 	} else if (class == BPF_LD) {
353 		if (BPF_MODE(insn->code) == BPF_ABS) {
354 			verbose("(%02x) r0 = *(%s *)skb[%d]\n",
355 				insn->code,
356 				bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
357 				insn->imm);
358 		} else if (BPF_MODE(insn->code) == BPF_IND) {
359 			verbose("(%02x) r0 = *(%s *)skb[r%d + %d]\n",
360 				insn->code,
361 				bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
362 				insn->src_reg, insn->imm);
363 		} else if (BPF_MODE(insn->code) == BPF_IMM) {
364 			verbose("(%02x) r%d = 0x%x\n",
365 				insn->code, insn->dst_reg, insn->imm);
366 		} else {
367 			verbose("BUG_ld_%02x\n", insn->code);
368 			return;
369 		}
370 	} else if (class == BPF_JMP) {
371 		u8 opcode = BPF_OP(insn->code);
372 
373 		if (opcode == BPF_CALL) {
374 			verbose("(%02x) call %s#%d\n", insn->code,
375 				func_id_name(insn->imm), insn->imm);
376 		} else if (insn->code == (BPF_JMP | BPF_JA)) {
377 			verbose("(%02x) goto pc%+d\n",
378 				insn->code, insn->off);
379 		} else if (insn->code == (BPF_JMP | BPF_EXIT)) {
380 			verbose("(%02x) exit\n", insn->code);
381 		} else if (BPF_SRC(insn->code) == BPF_X) {
382 			verbose("(%02x) if r%d %s r%d goto pc%+d\n",
383 				insn->code, insn->dst_reg,
384 				bpf_jmp_string[BPF_OP(insn->code) >> 4],
385 				insn->src_reg, insn->off);
386 		} else {
387 			verbose("(%02x) if r%d %s 0x%x goto pc%+d\n",
388 				insn->code, insn->dst_reg,
389 				bpf_jmp_string[BPF_OP(insn->code) >> 4],
390 				insn->imm, insn->off);
391 		}
392 	} else {
393 		verbose("(%02x) %s\n", insn->code, bpf_class_string[class]);
394 	}
395 }
396 
397 static int pop_stack(struct bpf_verifier_env *env, int *prev_insn_idx)
398 {
399 	struct bpf_verifier_stack_elem *elem;
400 	int insn_idx;
401 
402 	if (env->head == NULL)
403 		return -1;
404 
405 	memcpy(&env->cur_state, &env->head->st, sizeof(env->cur_state));
406 	insn_idx = env->head->insn_idx;
407 	if (prev_insn_idx)
408 		*prev_insn_idx = env->head->prev_insn_idx;
409 	elem = env->head->next;
410 	kfree(env->head);
411 	env->head = elem;
412 	env->stack_size--;
413 	return insn_idx;
414 }
415 
416 static struct bpf_verifier_state *push_stack(struct bpf_verifier_env *env,
417 					     int insn_idx, int prev_insn_idx)
418 {
419 	struct bpf_verifier_stack_elem *elem;
420 
421 	elem = kmalloc(sizeof(struct bpf_verifier_stack_elem), GFP_KERNEL);
422 	if (!elem)
423 		goto err;
424 
425 	memcpy(&elem->st, &env->cur_state, sizeof(env->cur_state));
426 	elem->insn_idx = insn_idx;
427 	elem->prev_insn_idx = prev_insn_idx;
428 	elem->next = env->head;
429 	env->head = elem;
430 	env->stack_size++;
431 	if (env->stack_size > BPF_COMPLEXITY_LIMIT_STACK) {
432 		verbose("BPF program is too complex\n");
433 		goto err;
434 	}
435 	return &elem->st;
436 err:
437 	/* pop all elements and return */
438 	while (pop_stack(env, NULL) >= 0);
439 	return NULL;
440 }
441 
442 #define CALLER_SAVED_REGS 6
443 static const int caller_saved[CALLER_SAVED_REGS] = {
444 	BPF_REG_0, BPF_REG_1, BPF_REG_2, BPF_REG_3, BPF_REG_4, BPF_REG_5
445 };
446 
447 static void init_reg_state(struct bpf_reg_state *regs)
448 {
449 	int i;
450 
451 	for (i = 0; i < MAX_BPF_REG; i++) {
452 		regs[i].type = NOT_INIT;
453 		regs[i].imm = 0;
454 		regs[i].min_value = BPF_REGISTER_MIN_RANGE;
455 		regs[i].max_value = BPF_REGISTER_MAX_RANGE;
456 	}
457 
458 	/* frame pointer */
459 	regs[BPF_REG_FP].type = FRAME_PTR;
460 
461 	/* 1st arg to a function */
462 	regs[BPF_REG_1].type = PTR_TO_CTX;
463 }
464 
465 static void mark_reg_unknown_value(struct bpf_reg_state *regs, u32 regno)
466 {
467 	BUG_ON(regno >= MAX_BPF_REG);
468 	regs[regno].type = UNKNOWN_VALUE;
469 	regs[regno].id = 0;
470 	regs[regno].imm = 0;
471 }
472 
473 static void reset_reg_range_values(struct bpf_reg_state *regs, u32 regno)
474 {
475 	regs[regno].min_value = BPF_REGISTER_MIN_RANGE;
476 	regs[regno].max_value = BPF_REGISTER_MAX_RANGE;
477 }
478 
479 enum reg_arg_type {
480 	SRC_OP,		/* register is used as source operand */
481 	DST_OP,		/* register is used as destination operand */
482 	DST_OP_NO_MARK	/* same as above, check only, don't mark */
483 };
484 
485 static int check_reg_arg(struct bpf_reg_state *regs, u32 regno,
486 			 enum reg_arg_type t)
487 {
488 	if (regno >= MAX_BPF_REG) {
489 		verbose("R%d is invalid\n", regno);
490 		return -EINVAL;
491 	}
492 
493 	if (t == SRC_OP) {
494 		/* check whether register used as source operand can be read */
495 		if (regs[regno].type == NOT_INIT) {
496 			verbose("R%d !read_ok\n", regno);
497 			return -EACCES;
498 		}
499 	} else {
500 		/* check whether register used as dest operand can be written to */
501 		if (regno == BPF_REG_FP) {
502 			verbose("frame pointer is read only\n");
503 			return -EACCES;
504 		}
505 		if (t == DST_OP)
506 			mark_reg_unknown_value(regs, regno);
507 	}
508 	return 0;
509 }
510 
511 static int bpf_size_to_bytes(int bpf_size)
512 {
513 	if (bpf_size == BPF_W)
514 		return 4;
515 	else if (bpf_size == BPF_H)
516 		return 2;
517 	else if (bpf_size == BPF_B)
518 		return 1;
519 	else if (bpf_size == BPF_DW)
520 		return 8;
521 	else
522 		return -EINVAL;
523 }
524 
525 static bool is_spillable_regtype(enum bpf_reg_type type)
526 {
527 	switch (type) {
528 	case PTR_TO_MAP_VALUE:
529 	case PTR_TO_MAP_VALUE_OR_NULL:
530 	case PTR_TO_STACK:
531 	case PTR_TO_CTX:
532 	case PTR_TO_PACKET:
533 	case PTR_TO_PACKET_END:
534 	case FRAME_PTR:
535 	case CONST_PTR_TO_MAP:
536 		return true;
537 	default:
538 		return false;
539 	}
540 }
541 
542 /* check_stack_read/write functions track spill/fill of registers,
543  * stack boundary and alignment are checked in check_mem_access()
544  */
545 static int check_stack_write(struct bpf_verifier_state *state, int off,
546 			     int size, int value_regno)
547 {
548 	int i;
549 	/* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0,
550 	 * so it's aligned access and [off, off + size) are within stack limits
551 	 */
552 
553 	if (value_regno >= 0 &&
554 	    is_spillable_regtype(state->regs[value_regno].type)) {
555 
556 		/* register containing pointer is being spilled into stack */
557 		if (size != BPF_REG_SIZE) {
558 			verbose("invalid size of register spill\n");
559 			return -EACCES;
560 		}
561 
562 		/* save register state */
563 		state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] =
564 			state->regs[value_regno];
565 
566 		for (i = 0; i < BPF_REG_SIZE; i++)
567 			state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_SPILL;
568 	} else {
569 		/* regular write of data into stack */
570 		state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] =
571 			(struct bpf_reg_state) {};
572 
573 		for (i = 0; i < size; i++)
574 			state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_MISC;
575 	}
576 	return 0;
577 }
578 
579 static int check_stack_read(struct bpf_verifier_state *state, int off, int size,
580 			    int value_regno)
581 {
582 	u8 *slot_type;
583 	int i;
584 
585 	slot_type = &state->stack_slot_type[MAX_BPF_STACK + off];
586 
587 	if (slot_type[0] == STACK_SPILL) {
588 		if (size != BPF_REG_SIZE) {
589 			verbose("invalid size of register spill\n");
590 			return -EACCES;
591 		}
592 		for (i = 1; i < BPF_REG_SIZE; i++) {
593 			if (slot_type[i] != STACK_SPILL) {
594 				verbose("corrupted spill memory\n");
595 				return -EACCES;
596 			}
597 		}
598 
599 		if (value_regno >= 0)
600 			/* restore register state from stack */
601 			state->regs[value_regno] =
602 				state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE];
603 		return 0;
604 	} else {
605 		for (i = 0; i < size; i++) {
606 			if (slot_type[i] != STACK_MISC) {
607 				verbose("invalid read from stack off %d+%d size %d\n",
608 					off, i, size);
609 				return -EACCES;
610 			}
611 		}
612 		if (value_regno >= 0)
613 			/* have read misc data from the stack */
614 			mark_reg_unknown_value(state->regs, value_regno);
615 		return 0;
616 	}
617 }
618 
619 /* check read/write into map element returned by bpf_map_lookup_elem() */
620 static int check_map_access(struct bpf_verifier_env *env, u32 regno, int off,
621 			    int size)
622 {
623 	struct bpf_map *map = env->cur_state.regs[regno].map_ptr;
624 
625 	if (off < 0 || off + size > map->value_size) {
626 		verbose("invalid access to map value, value_size=%d off=%d size=%d\n",
627 			map->value_size, off, size);
628 		return -EACCES;
629 	}
630 	return 0;
631 }
632 
633 #define MAX_PACKET_OFF 0xffff
634 
635 static bool may_access_direct_pkt_data(struct bpf_verifier_env *env,
636 				       const struct bpf_call_arg_meta *meta,
637 				       enum bpf_access_type t)
638 {
639 	switch (env->prog->type) {
640 	case BPF_PROG_TYPE_LWT_IN:
641 	case BPF_PROG_TYPE_LWT_OUT:
642 		/* dst_input() and dst_output() can't write for now */
643 		if (t == BPF_WRITE)
644 			return false;
645 	case BPF_PROG_TYPE_SCHED_CLS:
646 	case BPF_PROG_TYPE_SCHED_ACT:
647 	case BPF_PROG_TYPE_XDP:
648 	case BPF_PROG_TYPE_LWT_XMIT:
649 		if (meta)
650 			return meta->pkt_access;
651 
652 		env->seen_direct_write = true;
653 		return true;
654 	default:
655 		return false;
656 	}
657 }
658 
659 static int check_packet_access(struct bpf_verifier_env *env, u32 regno, int off,
660 			       int size)
661 {
662 	struct bpf_reg_state *regs = env->cur_state.regs;
663 	struct bpf_reg_state *reg = &regs[regno];
664 
665 	off += reg->off;
666 	if (off < 0 || size <= 0 || off + size > reg->range) {
667 		verbose("invalid access to packet, off=%d size=%d, R%d(id=%d,off=%d,r=%d)\n",
668 			off, size, regno, reg->id, reg->off, reg->range);
669 		return -EACCES;
670 	}
671 	return 0;
672 }
673 
674 /* check access to 'struct bpf_context' fields */
675 static int check_ctx_access(struct bpf_verifier_env *env, int off, int size,
676 			    enum bpf_access_type t, enum bpf_reg_type *reg_type)
677 {
678 	/* for analyzer ctx accesses are already validated and converted */
679 	if (env->analyzer_ops)
680 		return 0;
681 
682 	if (env->prog->aux->ops->is_valid_access &&
683 	    env->prog->aux->ops->is_valid_access(off, size, t, reg_type)) {
684 		/* remember the offset of last byte accessed in ctx */
685 		if (env->prog->aux->max_ctx_offset < off + size)
686 			env->prog->aux->max_ctx_offset = off + size;
687 		return 0;
688 	}
689 
690 	verbose("invalid bpf_context access off=%d size=%d\n", off, size);
691 	return -EACCES;
692 }
693 
694 static bool is_pointer_value(struct bpf_verifier_env *env, int regno)
695 {
696 	if (env->allow_ptr_leaks)
697 		return false;
698 
699 	switch (env->cur_state.regs[regno].type) {
700 	case UNKNOWN_VALUE:
701 	case CONST_IMM:
702 		return false;
703 	default:
704 		return true;
705 	}
706 }
707 
708 static int check_ptr_alignment(struct bpf_verifier_env *env,
709 			       struct bpf_reg_state *reg, int off, int size)
710 {
711 	if (reg->type != PTR_TO_PACKET && reg->type != PTR_TO_MAP_VALUE_ADJ) {
712 		if (off % size != 0) {
713 			verbose("misaligned access off %d size %d\n",
714 				off, size);
715 			return -EACCES;
716 		} else {
717 			return 0;
718 		}
719 	}
720 
721 	if (IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS))
722 		/* misaligned access to packet is ok on x86,arm,arm64 */
723 		return 0;
724 
725 	if (reg->id && size != 1) {
726 		verbose("Unknown packet alignment. Only byte-sized access allowed\n");
727 		return -EACCES;
728 	}
729 
730 	/* skb->data is NET_IP_ALIGN-ed */
731 	if (reg->type == PTR_TO_PACKET &&
732 	    (NET_IP_ALIGN + reg->off + off) % size != 0) {
733 		verbose("misaligned packet access off %d+%d+%d size %d\n",
734 			NET_IP_ALIGN, reg->off, off, size);
735 		return -EACCES;
736 	}
737 	return 0;
738 }
739 
740 /* check whether memory at (regno + off) is accessible for t = (read | write)
741  * if t==write, value_regno is a register which value is stored into memory
742  * if t==read, value_regno is a register which will receive the value from memory
743  * if t==write && value_regno==-1, some unknown value is stored into memory
744  * if t==read && value_regno==-1, don't care what we read from memory
745  */
746 static int check_mem_access(struct bpf_verifier_env *env, u32 regno, int off,
747 			    int bpf_size, enum bpf_access_type t,
748 			    int value_regno)
749 {
750 	struct bpf_verifier_state *state = &env->cur_state;
751 	struct bpf_reg_state *reg = &state->regs[regno];
752 	int size, err = 0;
753 
754 	if (reg->type == PTR_TO_STACK)
755 		off += reg->imm;
756 
757 	size = bpf_size_to_bytes(bpf_size);
758 	if (size < 0)
759 		return size;
760 
761 	err = check_ptr_alignment(env, reg, off, size);
762 	if (err)
763 		return err;
764 
765 	if (reg->type == PTR_TO_MAP_VALUE ||
766 	    reg->type == PTR_TO_MAP_VALUE_ADJ) {
767 		if (t == BPF_WRITE && value_regno >= 0 &&
768 		    is_pointer_value(env, value_regno)) {
769 			verbose("R%d leaks addr into map\n", value_regno);
770 			return -EACCES;
771 		}
772 
773 		/* If we adjusted the register to this map value at all then we
774 		 * need to change off and size to min_value and max_value
775 		 * respectively to make sure our theoretical access will be
776 		 * safe.
777 		 */
778 		if (reg->type == PTR_TO_MAP_VALUE_ADJ) {
779 			if (log_level)
780 				print_verifier_state(state);
781 			env->varlen_map_value_access = true;
782 			/* The minimum value is only important with signed
783 			 * comparisons where we can't assume the floor of a
784 			 * value is 0.  If we are using signed variables for our
785 			 * index'es we need to make sure that whatever we use
786 			 * will have a set floor within our range.
787 			 */
788 			if (reg->min_value < 0) {
789 				verbose("R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n",
790 					regno);
791 				return -EACCES;
792 			}
793 			err = check_map_access(env, regno, reg->min_value + off,
794 					       size);
795 			if (err) {
796 				verbose("R%d min value is outside of the array range\n",
797 					regno);
798 				return err;
799 			}
800 
801 			/* If we haven't set a max value then we need to bail
802 			 * since we can't be sure we won't do bad things.
803 			 */
804 			if (reg->max_value == BPF_REGISTER_MAX_RANGE) {
805 				verbose("R%d unbounded memory access, make sure to bounds check any array access into a map\n",
806 					regno);
807 				return -EACCES;
808 			}
809 			off += reg->max_value;
810 		}
811 		err = check_map_access(env, regno, off, size);
812 		if (!err && t == BPF_READ && value_regno >= 0)
813 			mark_reg_unknown_value(state->regs, value_regno);
814 
815 	} else if (reg->type == PTR_TO_CTX) {
816 		enum bpf_reg_type reg_type = UNKNOWN_VALUE;
817 
818 		if (t == BPF_WRITE && value_regno >= 0 &&
819 		    is_pointer_value(env, value_regno)) {
820 			verbose("R%d leaks addr into ctx\n", value_regno);
821 			return -EACCES;
822 		}
823 		err = check_ctx_access(env, off, size, t, &reg_type);
824 		if (!err && t == BPF_READ && value_regno >= 0) {
825 			mark_reg_unknown_value(state->regs, value_regno);
826 			/* note that reg.[id|off|range] == 0 */
827 			state->regs[value_regno].type = reg_type;
828 		}
829 
830 	} else if (reg->type == FRAME_PTR || reg->type == PTR_TO_STACK) {
831 		if (off >= 0 || off < -MAX_BPF_STACK) {
832 			verbose("invalid stack off=%d size=%d\n", off, size);
833 			return -EACCES;
834 		}
835 		if (t == BPF_WRITE) {
836 			if (!env->allow_ptr_leaks &&
837 			    state->stack_slot_type[MAX_BPF_STACK + off] == STACK_SPILL &&
838 			    size != BPF_REG_SIZE) {
839 				verbose("attempt to corrupt spilled pointer on stack\n");
840 				return -EACCES;
841 			}
842 			err = check_stack_write(state, off, size, value_regno);
843 		} else {
844 			err = check_stack_read(state, off, size, value_regno);
845 		}
846 	} else if (state->regs[regno].type == PTR_TO_PACKET) {
847 		if (t == BPF_WRITE && !may_access_direct_pkt_data(env, NULL, t)) {
848 			verbose("cannot write into packet\n");
849 			return -EACCES;
850 		}
851 		if (t == BPF_WRITE && value_regno >= 0 &&
852 		    is_pointer_value(env, value_regno)) {
853 			verbose("R%d leaks addr into packet\n", value_regno);
854 			return -EACCES;
855 		}
856 		err = check_packet_access(env, regno, off, size);
857 		if (!err && t == BPF_READ && value_regno >= 0)
858 			mark_reg_unknown_value(state->regs, value_regno);
859 	} else {
860 		verbose("R%d invalid mem access '%s'\n",
861 			regno, reg_type_str[reg->type]);
862 		return -EACCES;
863 	}
864 
865 	if (!err && size <= 2 && value_regno >= 0 && env->allow_ptr_leaks &&
866 	    state->regs[value_regno].type == UNKNOWN_VALUE) {
867 		/* 1 or 2 byte load zero-extends, determine the number of
868 		 * zero upper bits. Not doing it fo 4 byte load, since
869 		 * such values cannot be added to ptr_to_packet anyway.
870 		 */
871 		state->regs[value_regno].imm = 64 - size * 8;
872 	}
873 	return err;
874 }
875 
876 static int check_xadd(struct bpf_verifier_env *env, struct bpf_insn *insn)
877 {
878 	struct bpf_reg_state *regs = env->cur_state.regs;
879 	int err;
880 
881 	if ((BPF_SIZE(insn->code) != BPF_W && BPF_SIZE(insn->code) != BPF_DW) ||
882 	    insn->imm != 0) {
883 		verbose("BPF_XADD uses reserved fields\n");
884 		return -EINVAL;
885 	}
886 
887 	/* check src1 operand */
888 	err = check_reg_arg(regs, insn->src_reg, SRC_OP);
889 	if (err)
890 		return err;
891 
892 	/* check src2 operand */
893 	err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
894 	if (err)
895 		return err;
896 
897 	/* check whether atomic_add can read the memory */
898 	err = check_mem_access(env, insn->dst_reg, insn->off,
899 			       BPF_SIZE(insn->code), BPF_READ, -1);
900 	if (err)
901 		return err;
902 
903 	/* check whether atomic_add can write into the same memory */
904 	return check_mem_access(env, insn->dst_reg, insn->off,
905 				BPF_SIZE(insn->code), BPF_WRITE, -1);
906 }
907 
908 /* when register 'regno' is passed into function that will read 'access_size'
909  * bytes from that pointer, make sure that it's within stack boundary
910  * and all elements of stack are initialized
911  */
912 static int check_stack_boundary(struct bpf_verifier_env *env, int regno,
913 				int access_size, bool zero_size_allowed,
914 				struct bpf_call_arg_meta *meta)
915 {
916 	struct bpf_verifier_state *state = &env->cur_state;
917 	struct bpf_reg_state *regs = state->regs;
918 	int off, i;
919 
920 	if (regs[regno].type != PTR_TO_STACK) {
921 		if (zero_size_allowed && access_size == 0 &&
922 		    regs[regno].type == CONST_IMM &&
923 		    regs[regno].imm  == 0)
924 			return 0;
925 
926 		verbose("R%d type=%s expected=%s\n", regno,
927 			reg_type_str[regs[regno].type],
928 			reg_type_str[PTR_TO_STACK]);
929 		return -EACCES;
930 	}
931 
932 	off = regs[regno].imm;
933 	if (off >= 0 || off < -MAX_BPF_STACK || off + access_size > 0 ||
934 	    access_size <= 0) {
935 		verbose("invalid stack type R%d off=%d access_size=%d\n",
936 			regno, off, access_size);
937 		return -EACCES;
938 	}
939 
940 	if (meta && meta->raw_mode) {
941 		meta->access_size = access_size;
942 		meta->regno = regno;
943 		return 0;
944 	}
945 
946 	for (i = 0; i < access_size; i++) {
947 		if (state->stack_slot_type[MAX_BPF_STACK + off + i] != STACK_MISC) {
948 			verbose("invalid indirect read from stack off %d+%d size %d\n",
949 				off, i, access_size);
950 			return -EACCES;
951 		}
952 	}
953 	return 0;
954 }
955 
956 static int check_func_arg(struct bpf_verifier_env *env, u32 regno,
957 			  enum bpf_arg_type arg_type,
958 			  struct bpf_call_arg_meta *meta)
959 {
960 	struct bpf_reg_state *regs = env->cur_state.regs, *reg = &regs[regno];
961 	enum bpf_reg_type expected_type, type = reg->type;
962 	int err = 0;
963 
964 	if (arg_type == ARG_DONTCARE)
965 		return 0;
966 
967 	if (type == NOT_INIT) {
968 		verbose("R%d !read_ok\n", regno);
969 		return -EACCES;
970 	}
971 
972 	if (arg_type == ARG_ANYTHING) {
973 		if (is_pointer_value(env, regno)) {
974 			verbose("R%d leaks addr into helper function\n", regno);
975 			return -EACCES;
976 		}
977 		return 0;
978 	}
979 
980 	if (type == PTR_TO_PACKET &&
981 	    !may_access_direct_pkt_data(env, meta, BPF_READ)) {
982 		verbose("helper access to the packet is not allowed\n");
983 		return -EACCES;
984 	}
985 
986 	if (arg_type == ARG_PTR_TO_MAP_KEY ||
987 	    arg_type == ARG_PTR_TO_MAP_VALUE) {
988 		expected_type = PTR_TO_STACK;
989 		if (type != PTR_TO_PACKET && type != expected_type)
990 			goto err_type;
991 	} else if (arg_type == ARG_CONST_STACK_SIZE ||
992 		   arg_type == ARG_CONST_STACK_SIZE_OR_ZERO) {
993 		expected_type = CONST_IMM;
994 		if (type != expected_type)
995 			goto err_type;
996 	} else if (arg_type == ARG_CONST_MAP_PTR) {
997 		expected_type = CONST_PTR_TO_MAP;
998 		if (type != expected_type)
999 			goto err_type;
1000 	} else if (arg_type == ARG_PTR_TO_CTX) {
1001 		expected_type = PTR_TO_CTX;
1002 		if (type != expected_type)
1003 			goto err_type;
1004 	} else if (arg_type == ARG_PTR_TO_STACK ||
1005 		   arg_type == ARG_PTR_TO_RAW_STACK) {
1006 		expected_type = PTR_TO_STACK;
1007 		/* One exception here. In case function allows for NULL to be
1008 		 * passed in as argument, it's a CONST_IMM type. Final test
1009 		 * happens during stack boundary checking.
1010 		 */
1011 		if (type == CONST_IMM && reg->imm == 0)
1012 			/* final test in check_stack_boundary() */;
1013 		else if (type != PTR_TO_PACKET && type != expected_type)
1014 			goto err_type;
1015 		meta->raw_mode = arg_type == ARG_PTR_TO_RAW_STACK;
1016 	} else {
1017 		verbose("unsupported arg_type %d\n", arg_type);
1018 		return -EFAULT;
1019 	}
1020 
1021 	if (arg_type == ARG_CONST_MAP_PTR) {
1022 		/* bpf_map_xxx(map_ptr) call: remember that map_ptr */
1023 		meta->map_ptr = reg->map_ptr;
1024 	} else if (arg_type == ARG_PTR_TO_MAP_KEY) {
1025 		/* bpf_map_xxx(..., map_ptr, ..., key) call:
1026 		 * check that [key, key + map->key_size) are within
1027 		 * stack limits and initialized
1028 		 */
1029 		if (!meta->map_ptr) {
1030 			/* in function declaration map_ptr must come before
1031 			 * map_key, so that it's verified and known before
1032 			 * we have to check map_key here. Otherwise it means
1033 			 * that kernel subsystem misconfigured verifier
1034 			 */
1035 			verbose("invalid map_ptr to access map->key\n");
1036 			return -EACCES;
1037 		}
1038 		if (type == PTR_TO_PACKET)
1039 			err = check_packet_access(env, regno, 0,
1040 						  meta->map_ptr->key_size);
1041 		else
1042 			err = check_stack_boundary(env, regno,
1043 						   meta->map_ptr->key_size,
1044 						   false, NULL);
1045 	} else if (arg_type == ARG_PTR_TO_MAP_VALUE) {
1046 		/* bpf_map_xxx(..., map_ptr, ..., value) call:
1047 		 * check [value, value + map->value_size) validity
1048 		 */
1049 		if (!meta->map_ptr) {
1050 			/* kernel subsystem misconfigured verifier */
1051 			verbose("invalid map_ptr to access map->value\n");
1052 			return -EACCES;
1053 		}
1054 		if (type == PTR_TO_PACKET)
1055 			err = check_packet_access(env, regno, 0,
1056 						  meta->map_ptr->value_size);
1057 		else
1058 			err = check_stack_boundary(env, regno,
1059 						   meta->map_ptr->value_size,
1060 						   false, NULL);
1061 	} else if (arg_type == ARG_CONST_STACK_SIZE ||
1062 		   arg_type == ARG_CONST_STACK_SIZE_OR_ZERO) {
1063 		bool zero_size_allowed = (arg_type == ARG_CONST_STACK_SIZE_OR_ZERO);
1064 
1065 		/* bpf_xxx(..., buf, len) call will access 'len' bytes
1066 		 * from stack pointer 'buf'. Check it
1067 		 * note: regno == len, regno - 1 == buf
1068 		 */
1069 		if (regno == 0) {
1070 			/* kernel subsystem misconfigured verifier */
1071 			verbose("ARG_CONST_STACK_SIZE cannot be first argument\n");
1072 			return -EACCES;
1073 		}
1074 		if (regs[regno - 1].type == PTR_TO_PACKET)
1075 			err = check_packet_access(env, regno - 1, 0, reg->imm);
1076 		else
1077 			err = check_stack_boundary(env, regno - 1, reg->imm,
1078 						   zero_size_allowed, meta);
1079 	}
1080 
1081 	return err;
1082 err_type:
1083 	verbose("R%d type=%s expected=%s\n", regno,
1084 		reg_type_str[type], reg_type_str[expected_type]);
1085 	return -EACCES;
1086 }
1087 
1088 static int check_map_func_compatibility(struct bpf_map *map, int func_id)
1089 {
1090 	if (!map)
1091 		return 0;
1092 
1093 	/* We need a two way check, first is from map perspective ... */
1094 	switch (map->map_type) {
1095 	case BPF_MAP_TYPE_PROG_ARRAY:
1096 		if (func_id != BPF_FUNC_tail_call)
1097 			goto error;
1098 		break;
1099 	case BPF_MAP_TYPE_PERF_EVENT_ARRAY:
1100 		if (func_id != BPF_FUNC_perf_event_read &&
1101 		    func_id != BPF_FUNC_perf_event_output)
1102 			goto error;
1103 		break;
1104 	case BPF_MAP_TYPE_STACK_TRACE:
1105 		if (func_id != BPF_FUNC_get_stackid)
1106 			goto error;
1107 		break;
1108 	case BPF_MAP_TYPE_CGROUP_ARRAY:
1109 		if (func_id != BPF_FUNC_skb_under_cgroup &&
1110 		    func_id != BPF_FUNC_current_task_under_cgroup)
1111 			goto error;
1112 		break;
1113 	default:
1114 		break;
1115 	}
1116 
1117 	/* ... and second from the function itself. */
1118 	switch (func_id) {
1119 	case BPF_FUNC_tail_call:
1120 		if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY)
1121 			goto error;
1122 		break;
1123 	case BPF_FUNC_perf_event_read:
1124 	case BPF_FUNC_perf_event_output:
1125 		if (map->map_type != BPF_MAP_TYPE_PERF_EVENT_ARRAY)
1126 			goto error;
1127 		break;
1128 	case BPF_FUNC_get_stackid:
1129 		if (map->map_type != BPF_MAP_TYPE_STACK_TRACE)
1130 			goto error;
1131 		break;
1132 	case BPF_FUNC_current_task_under_cgroup:
1133 	case BPF_FUNC_skb_under_cgroup:
1134 		if (map->map_type != BPF_MAP_TYPE_CGROUP_ARRAY)
1135 			goto error;
1136 		break;
1137 	default:
1138 		break;
1139 	}
1140 
1141 	return 0;
1142 error:
1143 	verbose("cannot pass map_type %d into func %s#%d\n",
1144 		map->map_type, func_id_name(func_id), func_id);
1145 	return -EINVAL;
1146 }
1147 
1148 static int check_raw_mode(const struct bpf_func_proto *fn)
1149 {
1150 	int count = 0;
1151 
1152 	if (fn->arg1_type == ARG_PTR_TO_RAW_STACK)
1153 		count++;
1154 	if (fn->arg2_type == ARG_PTR_TO_RAW_STACK)
1155 		count++;
1156 	if (fn->arg3_type == ARG_PTR_TO_RAW_STACK)
1157 		count++;
1158 	if (fn->arg4_type == ARG_PTR_TO_RAW_STACK)
1159 		count++;
1160 	if (fn->arg5_type == ARG_PTR_TO_RAW_STACK)
1161 		count++;
1162 
1163 	return count > 1 ? -EINVAL : 0;
1164 }
1165 
1166 static void clear_all_pkt_pointers(struct bpf_verifier_env *env)
1167 {
1168 	struct bpf_verifier_state *state = &env->cur_state;
1169 	struct bpf_reg_state *regs = state->regs, *reg;
1170 	int i;
1171 
1172 	for (i = 0; i < MAX_BPF_REG; i++)
1173 		if (regs[i].type == PTR_TO_PACKET ||
1174 		    regs[i].type == PTR_TO_PACKET_END)
1175 			mark_reg_unknown_value(regs, i);
1176 
1177 	for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
1178 		if (state->stack_slot_type[i] != STACK_SPILL)
1179 			continue;
1180 		reg = &state->spilled_regs[i / BPF_REG_SIZE];
1181 		if (reg->type != PTR_TO_PACKET &&
1182 		    reg->type != PTR_TO_PACKET_END)
1183 			continue;
1184 		reg->type = UNKNOWN_VALUE;
1185 		reg->imm = 0;
1186 	}
1187 }
1188 
1189 static int check_call(struct bpf_verifier_env *env, int func_id)
1190 {
1191 	struct bpf_verifier_state *state = &env->cur_state;
1192 	const struct bpf_func_proto *fn = NULL;
1193 	struct bpf_reg_state *regs = state->regs;
1194 	struct bpf_reg_state *reg;
1195 	struct bpf_call_arg_meta meta;
1196 	bool changes_data;
1197 	int i, err;
1198 
1199 	/* find function prototype */
1200 	if (func_id < 0 || func_id >= __BPF_FUNC_MAX_ID) {
1201 		verbose("invalid func %s#%d\n", func_id_name(func_id), func_id);
1202 		return -EINVAL;
1203 	}
1204 
1205 	if (env->prog->aux->ops->get_func_proto)
1206 		fn = env->prog->aux->ops->get_func_proto(func_id);
1207 
1208 	if (!fn) {
1209 		verbose("unknown func %s#%d\n", func_id_name(func_id), func_id);
1210 		return -EINVAL;
1211 	}
1212 
1213 	/* eBPF programs must be GPL compatible to use GPL-ed functions */
1214 	if (!env->prog->gpl_compatible && fn->gpl_only) {
1215 		verbose("cannot call GPL only function from proprietary program\n");
1216 		return -EINVAL;
1217 	}
1218 
1219 	changes_data = bpf_helper_changes_pkt_data(fn->func);
1220 
1221 	memset(&meta, 0, sizeof(meta));
1222 	meta.pkt_access = fn->pkt_access;
1223 
1224 	/* We only support one arg being in raw mode at the moment, which
1225 	 * is sufficient for the helper functions we have right now.
1226 	 */
1227 	err = check_raw_mode(fn);
1228 	if (err) {
1229 		verbose("kernel subsystem misconfigured func %s#%d\n",
1230 			func_id_name(func_id), func_id);
1231 		return err;
1232 	}
1233 
1234 	/* check args */
1235 	err = check_func_arg(env, BPF_REG_1, fn->arg1_type, &meta);
1236 	if (err)
1237 		return err;
1238 	err = check_func_arg(env, BPF_REG_2, fn->arg2_type, &meta);
1239 	if (err)
1240 		return err;
1241 	err = check_func_arg(env, BPF_REG_3, fn->arg3_type, &meta);
1242 	if (err)
1243 		return err;
1244 	err = check_func_arg(env, BPF_REG_4, fn->arg4_type, &meta);
1245 	if (err)
1246 		return err;
1247 	err = check_func_arg(env, BPF_REG_5, fn->arg5_type, &meta);
1248 	if (err)
1249 		return err;
1250 
1251 	/* Mark slots with STACK_MISC in case of raw mode, stack offset
1252 	 * is inferred from register state.
1253 	 */
1254 	for (i = 0; i < meta.access_size; i++) {
1255 		err = check_mem_access(env, meta.regno, i, BPF_B, BPF_WRITE, -1);
1256 		if (err)
1257 			return err;
1258 	}
1259 
1260 	/* reset caller saved regs */
1261 	for (i = 0; i < CALLER_SAVED_REGS; i++) {
1262 		reg = regs + caller_saved[i];
1263 		reg->type = NOT_INIT;
1264 		reg->imm = 0;
1265 	}
1266 
1267 	/* update return register */
1268 	if (fn->ret_type == RET_INTEGER) {
1269 		regs[BPF_REG_0].type = UNKNOWN_VALUE;
1270 	} else if (fn->ret_type == RET_VOID) {
1271 		regs[BPF_REG_0].type = NOT_INIT;
1272 	} else if (fn->ret_type == RET_PTR_TO_MAP_VALUE_OR_NULL) {
1273 		regs[BPF_REG_0].type = PTR_TO_MAP_VALUE_OR_NULL;
1274 		regs[BPF_REG_0].max_value = regs[BPF_REG_0].min_value = 0;
1275 		/* remember map_ptr, so that check_map_access()
1276 		 * can check 'value_size' boundary of memory access
1277 		 * to map element returned from bpf_map_lookup_elem()
1278 		 */
1279 		if (meta.map_ptr == NULL) {
1280 			verbose("kernel subsystem misconfigured verifier\n");
1281 			return -EINVAL;
1282 		}
1283 		regs[BPF_REG_0].map_ptr = meta.map_ptr;
1284 		regs[BPF_REG_0].id = ++env->id_gen;
1285 	} else {
1286 		verbose("unknown return type %d of func %s#%d\n",
1287 			fn->ret_type, func_id_name(func_id), func_id);
1288 		return -EINVAL;
1289 	}
1290 
1291 	err = check_map_func_compatibility(meta.map_ptr, func_id);
1292 	if (err)
1293 		return err;
1294 
1295 	if (changes_data)
1296 		clear_all_pkt_pointers(env);
1297 	return 0;
1298 }
1299 
1300 static int check_packet_ptr_add(struct bpf_verifier_env *env,
1301 				struct bpf_insn *insn)
1302 {
1303 	struct bpf_reg_state *regs = env->cur_state.regs;
1304 	struct bpf_reg_state *dst_reg = &regs[insn->dst_reg];
1305 	struct bpf_reg_state *src_reg = &regs[insn->src_reg];
1306 	struct bpf_reg_state tmp_reg;
1307 	s32 imm;
1308 
1309 	if (BPF_SRC(insn->code) == BPF_K) {
1310 		/* pkt_ptr += imm */
1311 		imm = insn->imm;
1312 
1313 add_imm:
1314 		if (imm <= 0) {
1315 			verbose("addition of negative constant to packet pointer is not allowed\n");
1316 			return -EACCES;
1317 		}
1318 		if (imm >= MAX_PACKET_OFF ||
1319 		    imm + dst_reg->off >= MAX_PACKET_OFF) {
1320 			verbose("constant %d is too large to add to packet pointer\n",
1321 				imm);
1322 			return -EACCES;
1323 		}
1324 		/* a constant was added to pkt_ptr.
1325 		 * Remember it while keeping the same 'id'
1326 		 */
1327 		dst_reg->off += imm;
1328 	} else {
1329 		if (src_reg->type == PTR_TO_PACKET) {
1330 			/* R6=pkt(id=0,off=0,r=62) R7=imm22; r7 += r6 */
1331 			tmp_reg = *dst_reg;  /* save r7 state */
1332 			*dst_reg = *src_reg; /* copy pkt_ptr state r6 into r7 */
1333 			src_reg = &tmp_reg;  /* pretend it's src_reg state */
1334 			/* if the checks below reject it, the copy won't matter,
1335 			 * since we're rejecting the whole program. If all ok,
1336 			 * then imm22 state will be added to r7
1337 			 * and r7 will be pkt(id=0,off=22,r=62) while
1338 			 * r6 will stay as pkt(id=0,off=0,r=62)
1339 			 */
1340 		}
1341 
1342 		if (src_reg->type == CONST_IMM) {
1343 			/* pkt_ptr += reg where reg is known constant */
1344 			imm = src_reg->imm;
1345 			goto add_imm;
1346 		}
1347 		/* disallow pkt_ptr += reg
1348 		 * if reg is not uknown_value with guaranteed zero upper bits
1349 		 * otherwise pkt_ptr may overflow and addition will become
1350 		 * subtraction which is not allowed
1351 		 */
1352 		if (src_reg->type != UNKNOWN_VALUE) {
1353 			verbose("cannot add '%s' to ptr_to_packet\n",
1354 				reg_type_str[src_reg->type]);
1355 			return -EACCES;
1356 		}
1357 		if (src_reg->imm < 48) {
1358 			verbose("cannot add integer value with %lld upper zero bits to ptr_to_packet\n",
1359 				src_reg->imm);
1360 			return -EACCES;
1361 		}
1362 		/* dst_reg stays as pkt_ptr type and since some positive
1363 		 * integer value was added to the pointer, increment its 'id'
1364 		 */
1365 		dst_reg->id = ++env->id_gen;
1366 
1367 		/* something was added to pkt_ptr, set range and off to zero */
1368 		dst_reg->off = 0;
1369 		dst_reg->range = 0;
1370 	}
1371 	return 0;
1372 }
1373 
1374 static int evaluate_reg_alu(struct bpf_verifier_env *env, struct bpf_insn *insn)
1375 {
1376 	struct bpf_reg_state *regs = env->cur_state.regs;
1377 	struct bpf_reg_state *dst_reg = &regs[insn->dst_reg];
1378 	u8 opcode = BPF_OP(insn->code);
1379 	s64 imm_log2;
1380 
1381 	/* for type == UNKNOWN_VALUE:
1382 	 * imm > 0 -> number of zero upper bits
1383 	 * imm == 0 -> don't track which is the same as all bits can be non-zero
1384 	 */
1385 
1386 	if (BPF_SRC(insn->code) == BPF_X) {
1387 		struct bpf_reg_state *src_reg = &regs[insn->src_reg];
1388 
1389 		if (src_reg->type == UNKNOWN_VALUE && src_reg->imm > 0 &&
1390 		    dst_reg->imm && opcode == BPF_ADD) {
1391 			/* dreg += sreg
1392 			 * where both have zero upper bits. Adding them
1393 			 * can only result making one more bit non-zero
1394 			 * in the larger value.
1395 			 * Ex. 0xffff (imm=48) + 1 (imm=63) = 0x10000 (imm=47)
1396 			 *     0xffff (imm=48) + 0xffff = 0x1fffe (imm=47)
1397 			 */
1398 			dst_reg->imm = min(dst_reg->imm, src_reg->imm);
1399 			dst_reg->imm--;
1400 			return 0;
1401 		}
1402 		if (src_reg->type == CONST_IMM && src_reg->imm > 0 &&
1403 		    dst_reg->imm && opcode == BPF_ADD) {
1404 			/* dreg += sreg
1405 			 * where dreg has zero upper bits and sreg is const.
1406 			 * Adding them can only result making one more bit
1407 			 * non-zero in the larger value.
1408 			 */
1409 			imm_log2 = __ilog2_u64((long long)src_reg->imm);
1410 			dst_reg->imm = min(dst_reg->imm, 63 - imm_log2);
1411 			dst_reg->imm--;
1412 			return 0;
1413 		}
1414 		/* all other cases non supported yet, just mark dst_reg */
1415 		dst_reg->imm = 0;
1416 		return 0;
1417 	}
1418 
1419 	/* sign extend 32-bit imm into 64-bit to make sure that
1420 	 * negative values occupy bit 63. Note ilog2() would have
1421 	 * been incorrect, since sizeof(insn->imm) == 4
1422 	 */
1423 	imm_log2 = __ilog2_u64((long long)insn->imm);
1424 
1425 	if (dst_reg->imm && opcode == BPF_LSH) {
1426 		/* reg <<= imm
1427 		 * if reg was a result of 2 byte load, then its imm == 48
1428 		 * which means that upper 48 bits are zero and shifting this reg
1429 		 * left by 4 would mean that upper 44 bits are still zero
1430 		 */
1431 		dst_reg->imm -= insn->imm;
1432 	} else if (dst_reg->imm && opcode == BPF_MUL) {
1433 		/* reg *= imm
1434 		 * if multiplying by 14 subtract 4
1435 		 * This is conservative calculation of upper zero bits.
1436 		 * It's not trying to special case insn->imm == 1 or 0 cases
1437 		 */
1438 		dst_reg->imm -= imm_log2 + 1;
1439 	} else if (opcode == BPF_AND) {
1440 		/* reg &= imm */
1441 		dst_reg->imm = 63 - imm_log2;
1442 	} else if (dst_reg->imm && opcode == BPF_ADD) {
1443 		/* reg += imm */
1444 		dst_reg->imm = min(dst_reg->imm, 63 - imm_log2);
1445 		dst_reg->imm--;
1446 	} else if (opcode == BPF_RSH) {
1447 		/* reg >>= imm
1448 		 * which means that after right shift, upper bits will be zero
1449 		 * note that verifier already checked that
1450 		 * 0 <= imm < 64 for shift insn
1451 		 */
1452 		dst_reg->imm += insn->imm;
1453 		if (unlikely(dst_reg->imm > 64))
1454 			/* some dumb code did:
1455 			 * r2 = *(u32 *)mem;
1456 			 * r2 >>= 32;
1457 			 * and all bits are zero now */
1458 			dst_reg->imm = 64;
1459 	} else {
1460 		/* all other alu ops, means that we don't know what will
1461 		 * happen to the value, mark it with unknown number of zero bits
1462 		 */
1463 		dst_reg->imm = 0;
1464 	}
1465 
1466 	if (dst_reg->imm < 0) {
1467 		/* all 64 bits of the register can contain non-zero bits
1468 		 * and such value cannot be added to ptr_to_packet, since it
1469 		 * may overflow, mark it as unknown to avoid further eval
1470 		 */
1471 		dst_reg->imm = 0;
1472 	}
1473 	return 0;
1474 }
1475 
1476 static int evaluate_reg_imm_alu(struct bpf_verifier_env *env,
1477 				struct bpf_insn *insn)
1478 {
1479 	struct bpf_reg_state *regs = env->cur_state.regs;
1480 	struct bpf_reg_state *dst_reg = &regs[insn->dst_reg];
1481 	struct bpf_reg_state *src_reg = &regs[insn->src_reg];
1482 	u8 opcode = BPF_OP(insn->code);
1483 
1484 	/* dst_reg->type == CONST_IMM here, simulate execution of 'add'/'or'
1485 	 * insn. Don't care about overflow or negative values, just add them
1486 	 */
1487 	if (opcode == BPF_ADD && BPF_SRC(insn->code) == BPF_K)
1488 		dst_reg->imm += insn->imm;
1489 	else if (opcode == BPF_ADD && BPF_SRC(insn->code) == BPF_X &&
1490 		 src_reg->type == CONST_IMM)
1491 		dst_reg->imm += src_reg->imm;
1492 	else if (opcode == BPF_OR && BPF_SRC(insn->code) == BPF_K)
1493 		dst_reg->imm |= insn->imm;
1494 	else if (opcode == BPF_OR && BPF_SRC(insn->code) == BPF_X &&
1495 		 src_reg->type == CONST_IMM)
1496 		dst_reg->imm |= src_reg->imm;
1497 	else
1498 		mark_reg_unknown_value(regs, insn->dst_reg);
1499 	return 0;
1500 }
1501 
1502 static void check_reg_overflow(struct bpf_reg_state *reg)
1503 {
1504 	if (reg->max_value > BPF_REGISTER_MAX_RANGE)
1505 		reg->max_value = BPF_REGISTER_MAX_RANGE;
1506 	if (reg->min_value < BPF_REGISTER_MIN_RANGE ||
1507 	    reg->min_value > BPF_REGISTER_MAX_RANGE)
1508 		reg->min_value = BPF_REGISTER_MIN_RANGE;
1509 }
1510 
1511 static void adjust_reg_min_max_vals(struct bpf_verifier_env *env,
1512 				    struct bpf_insn *insn)
1513 {
1514 	struct bpf_reg_state *regs = env->cur_state.regs, *dst_reg;
1515 	s64 min_val = BPF_REGISTER_MIN_RANGE;
1516 	u64 max_val = BPF_REGISTER_MAX_RANGE;
1517 	u8 opcode = BPF_OP(insn->code);
1518 
1519 	dst_reg = &regs[insn->dst_reg];
1520 	if (BPF_SRC(insn->code) == BPF_X) {
1521 		check_reg_overflow(&regs[insn->src_reg]);
1522 		min_val = regs[insn->src_reg].min_value;
1523 		max_val = regs[insn->src_reg].max_value;
1524 
1525 		/* If the source register is a random pointer then the
1526 		 * min_value/max_value values represent the range of the known
1527 		 * accesses into that value, not the actual min/max value of the
1528 		 * register itself.  In this case we have to reset the reg range
1529 		 * values so we know it is not safe to look at.
1530 		 */
1531 		if (regs[insn->src_reg].type != CONST_IMM &&
1532 		    regs[insn->src_reg].type != UNKNOWN_VALUE) {
1533 			min_val = BPF_REGISTER_MIN_RANGE;
1534 			max_val = BPF_REGISTER_MAX_RANGE;
1535 		}
1536 	} else if (insn->imm < BPF_REGISTER_MAX_RANGE &&
1537 		   (s64)insn->imm > BPF_REGISTER_MIN_RANGE) {
1538 		min_val = max_val = insn->imm;
1539 	}
1540 
1541 	/* We don't know anything about what was done to this register, mark it
1542 	 * as unknown.
1543 	 */
1544 	if (min_val == BPF_REGISTER_MIN_RANGE &&
1545 	    max_val == BPF_REGISTER_MAX_RANGE) {
1546 		reset_reg_range_values(regs, insn->dst_reg);
1547 		return;
1548 	}
1549 
1550 	/* If one of our values was at the end of our ranges then we can't just
1551 	 * do our normal operations to the register, we need to set the values
1552 	 * to the min/max since they are undefined.
1553 	 */
1554 	if (min_val == BPF_REGISTER_MIN_RANGE)
1555 		dst_reg->min_value = BPF_REGISTER_MIN_RANGE;
1556 	if (max_val == BPF_REGISTER_MAX_RANGE)
1557 		dst_reg->max_value = BPF_REGISTER_MAX_RANGE;
1558 
1559 	switch (opcode) {
1560 	case BPF_ADD:
1561 		if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE)
1562 			dst_reg->min_value += min_val;
1563 		if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1564 			dst_reg->max_value += max_val;
1565 		break;
1566 	case BPF_SUB:
1567 		if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE)
1568 			dst_reg->min_value -= min_val;
1569 		if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1570 			dst_reg->max_value -= max_val;
1571 		break;
1572 	case BPF_MUL:
1573 		if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE)
1574 			dst_reg->min_value *= min_val;
1575 		if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1576 			dst_reg->max_value *= max_val;
1577 		break;
1578 	case BPF_AND:
1579 		/* Disallow AND'ing of negative numbers, ain't nobody got time
1580 		 * for that.  Otherwise the minimum is 0 and the max is the max
1581 		 * value we could AND against.
1582 		 */
1583 		if (min_val < 0)
1584 			dst_reg->min_value = BPF_REGISTER_MIN_RANGE;
1585 		else
1586 			dst_reg->min_value = 0;
1587 		dst_reg->max_value = max_val;
1588 		break;
1589 	case BPF_LSH:
1590 		/* Gotta have special overflow logic here, if we're shifting
1591 		 * more than MAX_RANGE then just assume we have an invalid
1592 		 * range.
1593 		 */
1594 		if (min_val > ilog2(BPF_REGISTER_MAX_RANGE))
1595 			dst_reg->min_value = BPF_REGISTER_MIN_RANGE;
1596 		else if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE)
1597 			dst_reg->min_value <<= min_val;
1598 
1599 		if (max_val > ilog2(BPF_REGISTER_MAX_RANGE))
1600 			dst_reg->max_value = BPF_REGISTER_MAX_RANGE;
1601 		else if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1602 			dst_reg->max_value <<= max_val;
1603 		break;
1604 	case BPF_RSH:
1605 		/* RSH by a negative number is undefined, and the BPF_RSH is an
1606 		 * unsigned shift, so make the appropriate casts.
1607 		 */
1608 		if (min_val < 0 || dst_reg->min_value < 0)
1609 			dst_reg->min_value = BPF_REGISTER_MIN_RANGE;
1610 		else
1611 			dst_reg->min_value =
1612 				(u64)(dst_reg->min_value) >> min_val;
1613 		if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1614 			dst_reg->max_value >>= max_val;
1615 		break;
1616 	default:
1617 		reset_reg_range_values(regs, insn->dst_reg);
1618 		break;
1619 	}
1620 
1621 	check_reg_overflow(dst_reg);
1622 }
1623 
1624 /* check validity of 32-bit and 64-bit arithmetic operations */
1625 static int check_alu_op(struct bpf_verifier_env *env, struct bpf_insn *insn)
1626 {
1627 	struct bpf_reg_state *regs = env->cur_state.regs, *dst_reg;
1628 	u8 opcode = BPF_OP(insn->code);
1629 	int err;
1630 
1631 	if (opcode == BPF_END || opcode == BPF_NEG) {
1632 		if (opcode == BPF_NEG) {
1633 			if (BPF_SRC(insn->code) != 0 ||
1634 			    insn->src_reg != BPF_REG_0 ||
1635 			    insn->off != 0 || insn->imm != 0) {
1636 				verbose("BPF_NEG uses reserved fields\n");
1637 				return -EINVAL;
1638 			}
1639 		} else {
1640 			if (insn->src_reg != BPF_REG_0 || insn->off != 0 ||
1641 			    (insn->imm != 16 && insn->imm != 32 && insn->imm != 64)) {
1642 				verbose("BPF_END uses reserved fields\n");
1643 				return -EINVAL;
1644 			}
1645 		}
1646 
1647 		/* check src operand */
1648 		err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1649 		if (err)
1650 			return err;
1651 
1652 		if (is_pointer_value(env, insn->dst_reg)) {
1653 			verbose("R%d pointer arithmetic prohibited\n",
1654 				insn->dst_reg);
1655 			return -EACCES;
1656 		}
1657 
1658 		/* check dest operand */
1659 		err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1660 		if (err)
1661 			return err;
1662 
1663 	} else if (opcode == BPF_MOV) {
1664 
1665 		if (BPF_SRC(insn->code) == BPF_X) {
1666 			if (insn->imm != 0 || insn->off != 0) {
1667 				verbose("BPF_MOV uses reserved fields\n");
1668 				return -EINVAL;
1669 			}
1670 
1671 			/* check src operand */
1672 			err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1673 			if (err)
1674 				return err;
1675 		} else {
1676 			if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
1677 				verbose("BPF_MOV uses reserved fields\n");
1678 				return -EINVAL;
1679 			}
1680 		}
1681 
1682 		/* check dest operand */
1683 		err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1684 		if (err)
1685 			return err;
1686 
1687 		/* we are setting our register to something new, we need to
1688 		 * reset its range values.
1689 		 */
1690 		reset_reg_range_values(regs, insn->dst_reg);
1691 
1692 		if (BPF_SRC(insn->code) == BPF_X) {
1693 			if (BPF_CLASS(insn->code) == BPF_ALU64) {
1694 				/* case: R1 = R2
1695 				 * copy register state to dest reg
1696 				 */
1697 				regs[insn->dst_reg] = regs[insn->src_reg];
1698 			} else {
1699 				if (is_pointer_value(env, insn->src_reg)) {
1700 					verbose("R%d partial copy of pointer\n",
1701 						insn->src_reg);
1702 					return -EACCES;
1703 				}
1704 				mark_reg_unknown_value(regs, insn->dst_reg);
1705 			}
1706 		} else {
1707 			/* case: R = imm
1708 			 * remember the value we stored into this reg
1709 			 */
1710 			regs[insn->dst_reg].type = CONST_IMM;
1711 			regs[insn->dst_reg].imm = insn->imm;
1712 			regs[insn->dst_reg].max_value = insn->imm;
1713 			regs[insn->dst_reg].min_value = insn->imm;
1714 		}
1715 
1716 	} else if (opcode > BPF_END) {
1717 		verbose("invalid BPF_ALU opcode %x\n", opcode);
1718 		return -EINVAL;
1719 
1720 	} else {	/* all other ALU ops: and, sub, xor, add, ... */
1721 
1722 		if (BPF_SRC(insn->code) == BPF_X) {
1723 			if (insn->imm != 0 || insn->off != 0) {
1724 				verbose("BPF_ALU uses reserved fields\n");
1725 				return -EINVAL;
1726 			}
1727 			/* check src1 operand */
1728 			err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1729 			if (err)
1730 				return err;
1731 		} else {
1732 			if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
1733 				verbose("BPF_ALU uses reserved fields\n");
1734 				return -EINVAL;
1735 			}
1736 		}
1737 
1738 		/* check src2 operand */
1739 		err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1740 		if (err)
1741 			return err;
1742 
1743 		if ((opcode == BPF_MOD || opcode == BPF_DIV) &&
1744 		    BPF_SRC(insn->code) == BPF_K && insn->imm == 0) {
1745 			verbose("div by zero\n");
1746 			return -EINVAL;
1747 		}
1748 
1749 		if ((opcode == BPF_LSH || opcode == BPF_RSH ||
1750 		     opcode == BPF_ARSH) && BPF_SRC(insn->code) == BPF_K) {
1751 			int size = BPF_CLASS(insn->code) == BPF_ALU64 ? 64 : 32;
1752 
1753 			if (insn->imm < 0 || insn->imm >= size) {
1754 				verbose("invalid shift %d\n", insn->imm);
1755 				return -EINVAL;
1756 			}
1757 		}
1758 
1759 		/* check dest operand */
1760 		err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK);
1761 		if (err)
1762 			return err;
1763 
1764 		dst_reg = &regs[insn->dst_reg];
1765 
1766 		/* first we want to adjust our ranges. */
1767 		adjust_reg_min_max_vals(env, insn);
1768 
1769 		/* pattern match 'bpf_add Rx, imm' instruction */
1770 		if (opcode == BPF_ADD && BPF_CLASS(insn->code) == BPF_ALU64 &&
1771 		    dst_reg->type == FRAME_PTR && BPF_SRC(insn->code) == BPF_K) {
1772 			dst_reg->type = PTR_TO_STACK;
1773 			dst_reg->imm = insn->imm;
1774 			return 0;
1775 		} else if (opcode == BPF_ADD &&
1776 			   BPF_CLASS(insn->code) == BPF_ALU64 &&
1777 			   (dst_reg->type == PTR_TO_PACKET ||
1778 			    (BPF_SRC(insn->code) == BPF_X &&
1779 			     regs[insn->src_reg].type == PTR_TO_PACKET))) {
1780 			/* ptr_to_packet += K|X */
1781 			return check_packet_ptr_add(env, insn);
1782 		} else if (BPF_CLASS(insn->code) == BPF_ALU64 &&
1783 			   dst_reg->type == UNKNOWN_VALUE &&
1784 			   env->allow_ptr_leaks) {
1785 			/* unknown += K|X */
1786 			return evaluate_reg_alu(env, insn);
1787 		} else if (BPF_CLASS(insn->code) == BPF_ALU64 &&
1788 			   dst_reg->type == CONST_IMM &&
1789 			   env->allow_ptr_leaks) {
1790 			/* reg_imm += K|X */
1791 			return evaluate_reg_imm_alu(env, insn);
1792 		} else if (is_pointer_value(env, insn->dst_reg)) {
1793 			verbose("R%d pointer arithmetic prohibited\n",
1794 				insn->dst_reg);
1795 			return -EACCES;
1796 		} else if (BPF_SRC(insn->code) == BPF_X &&
1797 			   is_pointer_value(env, insn->src_reg)) {
1798 			verbose("R%d pointer arithmetic prohibited\n",
1799 				insn->src_reg);
1800 			return -EACCES;
1801 		}
1802 
1803 		/* If we did pointer math on a map value then just set it to our
1804 		 * PTR_TO_MAP_VALUE_ADJ type so we can deal with any stores or
1805 		 * loads to this register appropriately, otherwise just mark the
1806 		 * register as unknown.
1807 		 */
1808 		if (env->allow_ptr_leaks &&
1809 		    (dst_reg->type == PTR_TO_MAP_VALUE ||
1810 		     dst_reg->type == PTR_TO_MAP_VALUE_ADJ))
1811 			dst_reg->type = PTR_TO_MAP_VALUE_ADJ;
1812 		else
1813 			mark_reg_unknown_value(regs, insn->dst_reg);
1814 	}
1815 
1816 	return 0;
1817 }
1818 
1819 static void find_good_pkt_pointers(struct bpf_verifier_state *state,
1820 				   struct bpf_reg_state *dst_reg)
1821 {
1822 	struct bpf_reg_state *regs = state->regs, *reg;
1823 	int i;
1824 
1825 	/* LLVM can generate two kind of checks:
1826 	 *
1827 	 * Type 1:
1828 	 *
1829 	 *   r2 = r3;
1830 	 *   r2 += 8;
1831 	 *   if (r2 > pkt_end) goto <handle exception>
1832 	 *   <access okay>
1833 	 *
1834 	 *   Where:
1835 	 *     r2 == dst_reg, pkt_end == src_reg
1836 	 *     r2=pkt(id=n,off=8,r=0)
1837 	 *     r3=pkt(id=n,off=0,r=0)
1838 	 *
1839 	 * Type 2:
1840 	 *
1841 	 *   r2 = r3;
1842 	 *   r2 += 8;
1843 	 *   if (pkt_end >= r2) goto <access okay>
1844 	 *   <handle exception>
1845 	 *
1846 	 *   Where:
1847 	 *     pkt_end == dst_reg, r2 == src_reg
1848 	 *     r2=pkt(id=n,off=8,r=0)
1849 	 *     r3=pkt(id=n,off=0,r=0)
1850 	 *
1851 	 * Find register r3 and mark its range as r3=pkt(id=n,off=0,r=8)
1852 	 * so that range of bytes [r3, r3 + 8) is safe to access.
1853 	 */
1854 
1855 	for (i = 0; i < MAX_BPF_REG; i++)
1856 		if (regs[i].type == PTR_TO_PACKET && regs[i].id == dst_reg->id)
1857 			regs[i].range = dst_reg->off;
1858 
1859 	for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
1860 		if (state->stack_slot_type[i] != STACK_SPILL)
1861 			continue;
1862 		reg = &state->spilled_regs[i / BPF_REG_SIZE];
1863 		if (reg->type == PTR_TO_PACKET && reg->id == dst_reg->id)
1864 			reg->range = dst_reg->off;
1865 	}
1866 }
1867 
1868 /* Adjusts the register min/max values in the case that the dst_reg is the
1869  * variable register that we are working on, and src_reg is a constant or we're
1870  * simply doing a BPF_K check.
1871  */
1872 static void reg_set_min_max(struct bpf_reg_state *true_reg,
1873 			    struct bpf_reg_state *false_reg, u64 val,
1874 			    u8 opcode)
1875 {
1876 	switch (opcode) {
1877 	case BPF_JEQ:
1878 		/* If this is false then we know nothing Jon Snow, but if it is
1879 		 * true then we know for sure.
1880 		 */
1881 		true_reg->max_value = true_reg->min_value = val;
1882 		break;
1883 	case BPF_JNE:
1884 		/* If this is true we know nothing Jon Snow, but if it is false
1885 		 * we know the value for sure;
1886 		 */
1887 		false_reg->max_value = false_reg->min_value = val;
1888 		break;
1889 	case BPF_JGT:
1890 		/* Unsigned comparison, the minimum value is 0. */
1891 		false_reg->min_value = 0;
1892 	case BPF_JSGT:
1893 		/* If this is false then we know the maximum val is val,
1894 		 * otherwise we know the min val is val+1.
1895 		 */
1896 		false_reg->max_value = val;
1897 		true_reg->min_value = val + 1;
1898 		break;
1899 	case BPF_JGE:
1900 		/* Unsigned comparison, the minimum value is 0. */
1901 		false_reg->min_value = 0;
1902 	case BPF_JSGE:
1903 		/* If this is false then we know the maximum value is val - 1,
1904 		 * otherwise we know the mimimum value is val.
1905 		 */
1906 		false_reg->max_value = val - 1;
1907 		true_reg->min_value = val;
1908 		break;
1909 	default:
1910 		break;
1911 	}
1912 
1913 	check_reg_overflow(false_reg);
1914 	check_reg_overflow(true_reg);
1915 }
1916 
1917 /* Same as above, but for the case that dst_reg is a CONST_IMM reg and src_reg
1918  * is the variable reg.
1919  */
1920 static void reg_set_min_max_inv(struct bpf_reg_state *true_reg,
1921 				struct bpf_reg_state *false_reg, u64 val,
1922 				u8 opcode)
1923 {
1924 	switch (opcode) {
1925 	case BPF_JEQ:
1926 		/* If this is false then we know nothing Jon Snow, but if it is
1927 		 * true then we know for sure.
1928 		 */
1929 		true_reg->max_value = true_reg->min_value = val;
1930 		break;
1931 	case BPF_JNE:
1932 		/* If this is true we know nothing Jon Snow, but if it is false
1933 		 * we know the value for sure;
1934 		 */
1935 		false_reg->max_value = false_reg->min_value = val;
1936 		break;
1937 	case BPF_JGT:
1938 		/* Unsigned comparison, the minimum value is 0. */
1939 		true_reg->min_value = 0;
1940 	case BPF_JSGT:
1941 		/*
1942 		 * If this is false, then the val is <= the register, if it is
1943 		 * true the register <= to the val.
1944 		 */
1945 		false_reg->min_value = val;
1946 		true_reg->max_value = val - 1;
1947 		break;
1948 	case BPF_JGE:
1949 		/* Unsigned comparison, the minimum value is 0. */
1950 		true_reg->min_value = 0;
1951 	case BPF_JSGE:
1952 		/* If this is false then constant < register, if it is true then
1953 		 * the register < constant.
1954 		 */
1955 		false_reg->min_value = val + 1;
1956 		true_reg->max_value = val;
1957 		break;
1958 	default:
1959 		break;
1960 	}
1961 
1962 	check_reg_overflow(false_reg);
1963 	check_reg_overflow(true_reg);
1964 }
1965 
1966 static void mark_map_reg(struct bpf_reg_state *regs, u32 regno, u32 id,
1967 			 enum bpf_reg_type type)
1968 {
1969 	struct bpf_reg_state *reg = &regs[regno];
1970 
1971 	if (reg->type == PTR_TO_MAP_VALUE_OR_NULL && reg->id == id) {
1972 		reg->type = type;
1973 		if (type == UNKNOWN_VALUE)
1974 			mark_reg_unknown_value(regs, regno);
1975 	}
1976 }
1977 
1978 /* The logic is similar to find_good_pkt_pointers(), both could eventually
1979  * be folded together at some point.
1980  */
1981 static void mark_map_regs(struct bpf_verifier_state *state, u32 regno,
1982 			  enum bpf_reg_type type)
1983 {
1984 	struct bpf_reg_state *regs = state->regs;
1985 	int i;
1986 
1987 	for (i = 0; i < MAX_BPF_REG; i++)
1988 		mark_map_reg(regs, i, regs[regno].id, type);
1989 
1990 	for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
1991 		if (state->stack_slot_type[i] != STACK_SPILL)
1992 			continue;
1993 		mark_map_reg(state->spilled_regs, i / BPF_REG_SIZE,
1994 			     regs[regno].id, type);
1995 	}
1996 }
1997 
1998 static int check_cond_jmp_op(struct bpf_verifier_env *env,
1999 			     struct bpf_insn *insn, int *insn_idx)
2000 {
2001 	struct bpf_verifier_state *other_branch, *this_branch = &env->cur_state;
2002 	struct bpf_reg_state *regs = this_branch->regs, *dst_reg;
2003 	u8 opcode = BPF_OP(insn->code);
2004 	int err;
2005 
2006 	if (opcode > BPF_EXIT) {
2007 		verbose("invalid BPF_JMP opcode %x\n", opcode);
2008 		return -EINVAL;
2009 	}
2010 
2011 	if (BPF_SRC(insn->code) == BPF_X) {
2012 		if (insn->imm != 0) {
2013 			verbose("BPF_JMP uses reserved fields\n");
2014 			return -EINVAL;
2015 		}
2016 
2017 		/* check src1 operand */
2018 		err = check_reg_arg(regs, insn->src_reg, SRC_OP);
2019 		if (err)
2020 			return err;
2021 
2022 		if (is_pointer_value(env, insn->src_reg)) {
2023 			verbose("R%d pointer comparison prohibited\n",
2024 				insn->src_reg);
2025 			return -EACCES;
2026 		}
2027 	} else {
2028 		if (insn->src_reg != BPF_REG_0) {
2029 			verbose("BPF_JMP uses reserved fields\n");
2030 			return -EINVAL;
2031 		}
2032 	}
2033 
2034 	/* check src2 operand */
2035 	err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
2036 	if (err)
2037 		return err;
2038 
2039 	dst_reg = &regs[insn->dst_reg];
2040 
2041 	/* detect if R == 0 where R was initialized to zero earlier */
2042 	if (BPF_SRC(insn->code) == BPF_K &&
2043 	    (opcode == BPF_JEQ || opcode == BPF_JNE) &&
2044 	    dst_reg->type == CONST_IMM && dst_reg->imm == insn->imm) {
2045 		if (opcode == BPF_JEQ) {
2046 			/* if (imm == imm) goto pc+off;
2047 			 * only follow the goto, ignore fall-through
2048 			 */
2049 			*insn_idx += insn->off;
2050 			return 0;
2051 		} else {
2052 			/* if (imm != imm) goto pc+off;
2053 			 * only follow fall-through branch, since
2054 			 * that's where the program will go
2055 			 */
2056 			return 0;
2057 		}
2058 	}
2059 
2060 	other_branch = push_stack(env, *insn_idx + insn->off + 1, *insn_idx);
2061 	if (!other_branch)
2062 		return -EFAULT;
2063 
2064 	/* detect if we are comparing against a constant value so we can adjust
2065 	 * our min/max values for our dst register.
2066 	 */
2067 	if (BPF_SRC(insn->code) == BPF_X) {
2068 		if (regs[insn->src_reg].type == CONST_IMM)
2069 			reg_set_min_max(&other_branch->regs[insn->dst_reg],
2070 					dst_reg, regs[insn->src_reg].imm,
2071 					opcode);
2072 		else if (dst_reg->type == CONST_IMM)
2073 			reg_set_min_max_inv(&other_branch->regs[insn->src_reg],
2074 					    &regs[insn->src_reg], dst_reg->imm,
2075 					    opcode);
2076 	} else {
2077 		reg_set_min_max(&other_branch->regs[insn->dst_reg],
2078 					dst_reg, insn->imm, opcode);
2079 	}
2080 
2081 	/* detect if R == 0 where R is returned from bpf_map_lookup_elem() */
2082 	if (BPF_SRC(insn->code) == BPF_K &&
2083 	    insn->imm == 0 && (opcode == BPF_JEQ || opcode == BPF_JNE) &&
2084 	    dst_reg->type == PTR_TO_MAP_VALUE_OR_NULL) {
2085 		/* Mark all identical map registers in each branch as either
2086 		 * safe or unknown depending R == 0 or R != 0 conditional.
2087 		 */
2088 		mark_map_regs(this_branch, insn->dst_reg,
2089 			      opcode == BPF_JEQ ? PTR_TO_MAP_VALUE : UNKNOWN_VALUE);
2090 		mark_map_regs(other_branch, insn->dst_reg,
2091 			      opcode == BPF_JEQ ? UNKNOWN_VALUE : PTR_TO_MAP_VALUE);
2092 	} else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JGT &&
2093 		   dst_reg->type == PTR_TO_PACKET &&
2094 		   regs[insn->src_reg].type == PTR_TO_PACKET_END) {
2095 		find_good_pkt_pointers(this_branch, dst_reg);
2096 	} else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JGE &&
2097 		   dst_reg->type == PTR_TO_PACKET_END &&
2098 		   regs[insn->src_reg].type == PTR_TO_PACKET) {
2099 		find_good_pkt_pointers(other_branch, &regs[insn->src_reg]);
2100 	} else if (is_pointer_value(env, insn->dst_reg)) {
2101 		verbose("R%d pointer comparison prohibited\n", insn->dst_reg);
2102 		return -EACCES;
2103 	}
2104 	if (log_level)
2105 		print_verifier_state(this_branch);
2106 	return 0;
2107 }
2108 
2109 /* return the map pointer stored inside BPF_LD_IMM64 instruction */
2110 static struct bpf_map *ld_imm64_to_map_ptr(struct bpf_insn *insn)
2111 {
2112 	u64 imm64 = ((u64) (u32) insn[0].imm) | ((u64) (u32) insn[1].imm) << 32;
2113 
2114 	return (struct bpf_map *) (unsigned long) imm64;
2115 }
2116 
2117 /* verify BPF_LD_IMM64 instruction */
2118 static int check_ld_imm(struct bpf_verifier_env *env, struct bpf_insn *insn)
2119 {
2120 	struct bpf_reg_state *regs = env->cur_state.regs;
2121 	int err;
2122 
2123 	if (BPF_SIZE(insn->code) != BPF_DW) {
2124 		verbose("invalid BPF_LD_IMM insn\n");
2125 		return -EINVAL;
2126 	}
2127 	if (insn->off != 0) {
2128 		verbose("BPF_LD_IMM64 uses reserved fields\n");
2129 		return -EINVAL;
2130 	}
2131 
2132 	err = check_reg_arg(regs, insn->dst_reg, DST_OP);
2133 	if (err)
2134 		return err;
2135 
2136 	if (insn->src_reg == 0) {
2137 		/* generic move 64-bit immediate into a register,
2138 		 * only analyzer needs to collect the ld_imm value.
2139 		 */
2140 		u64 imm = ((u64)(insn + 1)->imm << 32) | (u32)insn->imm;
2141 
2142 		if (!env->analyzer_ops)
2143 			return 0;
2144 
2145 		regs[insn->dst_reg].type = CONST_IMM;
2146 		regs[insn->dst_reg].imm = imm;
2147 		return 0;
2148 	}
2149 
2150 	/* replace_map_fd_with_map_ptr() should have caught bad ld_imm64 */
2151 	BUG_ON(insn->src_reg != BPF_PSEUDO_MAP_FD);
2152 
2153 	regs[insn->dst_reg].type = CONST_PTR_TO_MAP;
2154 	regs[insn->dst_reg].map_ptr = ld_imm64_to_map_ptr(insn);
2155 	return 0;
2156 }
2157 
2158 static bool may_access_skb(enum bpf_prog_type type)
2159 {
2160 	switch (type) {
2161 	case BPF_PROG_TYPE_SOCKET_FILTER:
2162 	case BPF_PROG_TYPE_SCHED_CLS:
2163 	case BPF_PROG_TYPE_SCHED_ACT:
2164 		return true;
2165 	default:
2166 		return false;
2167 	}
2168 }
2169 
2170 /* verify safety of LD_ABS|LD_IND instructions:
2171  * - they can only appear in the programs where ctx == skb
2172  * - since they are wrappers of function calls, they scratch R1-R5 registers,
2173  *   preserve R6-R9, and store return value into R0
2174  *
2175  * Implicit input:
2176  *   ctx == skb == R6 == CTX
2177  *
2178  * Explicit input:
2179  *   SRC == any register
2180  *   IMM == 32-bit immediate
2181  *
2182  * Output:
2183  *   R0 - 8/16/32-bit skb data converted to cpu endianness
2184  */
2185 static int check_ld_abs(struct bpf_verifier_env *env, struct bpf_insn *insn)
2186 {
2187 	struct bpf_reg_state *regs = env->cur_state.regs;
2188 	u8 mode = BPF_MODE(insn->code);
2189 	struct bpf_reg_state *reg;
2190 	int i, err;
2191 
2192 	if (!may_access_skb(env->prog->type)) {
2193 		verbose("BPF_LD_[ABS|IND] instructions not allowed for this program type\n");
2194 		return -EINVAL;
2195 	}
2196 
2197 	if (insn->dst_reg != BPF_REG_0 || insn->off != 0 ||
2198 	    BPF_SIZE(insn->code) == BPF_DW ||
2199 	    (mode == BPF_ABS && insn->src_reg != BPF_REG_0)) {
2200 		verbose("BPF_LD_[ABS|IND] uses reserved fields\n");
2201 		return -EINVAL;
2202 	}
2203 
2204 	/* check whether implicit source operand (register R6) is readable */
2205 	err = check_reg_arg(regs, BPF_REG_6, SRC_OP);
2206 	if (err)
2207 		return err;
2208 
2209 	if (regs[BPF_REG_6].type != PTR_TO_CTX) {
2210 		verbose("at the time of BPF_LD_ABS|IND R6 != pointer to skb\n");
2211 		return -EINVAL;
2212 	}
2213 
2214 	if (mode == BPF_IND) {
2215 		/* check explicit source operand */
2216 		err = check_reg_arg(regs, insn->src_reg, SRC_OP);
2217 		if (err)
2218 			return err;
2219 	}
2220 
2221 	/* reset caller saved regs to unreadable */
2222 	for (i = 0; i < CALLER_SAVED_REGS; i++) {
2223 		reg = regs + caller_saved[i];
2224 		reg->type = NOT_INIT;
2225 		reg->imm = 0;
2226 	}
2227 
2228 	/* mark destination R0 register as readable, since it contains
2229 	 * the value fetched from the packet
2230 	 */
2231 	regs[BPF_REG_0].type = UNKNOWN_VALUE;
2232 	return 0;
2233 }
2234 
2235 /* non-recursive DFS pseudo code
2236  * 1  procedure DFS-iterative(G,v):
2237  * 2      label v as discovered
2238  * 3      let S be a stack
2239  * 4      S.push(v)
2240  * 5      while S is not empty
2241  * 6            t <- S.pop()
2242  * 7            if t is what we're looking for:
2243  * 8                return t
2244  * 9            for all edges e in G.adjacentEdges(t) do
2245  * 10               if edge e is already labelled
2246  * 11                   continue with the next edge
2247  * 12               w <- G.adjacentVertex(t,e)
2248  * 13               if vertex w is not discovered and not explored
2249  * 14                   label e as tree-edge
2250  * 15                   label w as discovered
2251  * 16                   S.push(w)
2252  * 17                   continue at 5
2253  * 18               else if vertex w is discovered
2254  * 19                   label e as back-edge
2255  * 20               else
2256  * 21                   // vertex w is explored
2257  * 22                   label e as forward- or cross-edge
2258  * 23           label t as explored
2259  * 24           S.pop()
2260  *
2261  * convention:
2262  * 0x10 - discovered
2263  * 0x11 - discovered and fall-through edge labelled
2264  * 0x12 - discovered and fall-through and branch edges labelled
2265  * 0x20 - explored
2266  */
2267 
2268 enum {
2269 	DISCOVERED = 0x10,
2270 	EXPLORED = 0x20,
2271 	FALLTHROUGH = 1,
2272 	BRANCH = 2,
2273 };
2274 
2275 #define STATE_LIST_MARK ((struct bpf_verifier_state_list *) -1L)
2276 
2277 static int *insn_stack;	/* stack of insns to process */
2278 static int cur_stack;	/* current stack index */
2279 static int *insn_state;
2280 
2281 /* t, w, e - match pseudo-code above:
2282  * t - index of current instruction
2283  * w - next instruction
2284  * e - edge
2285  */
2286 static int push_insn(int t, int w, int e, struct bpf_verifier_env *env)
2287 {
2288 	if (e == FALLTHROUGH && insn_state[t] >= (DISCOVERED | FALLTHROUGH))
2289 		return 0;
2290 
2291 	if (e == BRANCH && insn_state[t] >= (DISCOVERED | BRANCH))
2292 		return 0;
2293 
2294 	if (w < 0 || w >= env->prog->len) {
2295 		verbose("jump out of range from insn %d to %d\n", t, w);
2296 		return -EINVAL;
2297 	}
2298 
2299 	if (e == BRANCH)
2300 		/* mark branch target for state pruning */
2301 		env->explored_states[w] = STATE_LIST_MARK;
2302 
2303 	if (insn_state[w] == 0) {
2304 		/* tree-edge */
2305 		insn_state[t] = DISCOVERED | e;
2306 		insn_state[w] = DISCOVERED;
2307 		if (cur_stack >= env->prog->len)
2308 			return -E2BIG;
2309 		insn_stack[cur_stack++] = w;
2310 		return 1;
2311 	} else if ((insn_state[w] & 0xF0) == DISCOVERED) {
2312 		verbose("back-edge from insn %d to %d\n", t, w);
2313 		return -EINVAL;
2314 	} else if (insn_state[w] == EXPLORED) {
2315 		/* forward- or cross-edge */
2316 		insn_state[t] = DISCOVERED | e;
2317 	} else {
2318 		verbose("insn state internal bug\n");
2319 		return -EFAULT;
2320 	}
2321 	return 0;
2322 }
2323 
2324 /* non-recursive depth-first-search to detect loops in BPF program
2325  * loop == back-edge in directed graph
2326  */
2327 static int check_cfg(struct bpf_verifier_env *env)
2328 {
2329 	struct bpf_insn *insns = env->prog->insnsi;
2330 	int insn_cnt = env->prog->len;
2331 	int ret = 0;
2332 	int i, t;
2333 
2334 	insn_state = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
2335 	if (!insn_state)
2336 		return -ENOMEM;
2337 
2338 	insn_stack = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
2339 	if (!insn_stack) {
2340 		kfree(insn_state);
2341 		return -ENOMEM;
2342 	}
2343 
2344 	insn_state[0] = DISCOVERED; /* mark 1st insn as discovered */
2345 	insn_stack[0] = 0; /* 0 is the first instruction */
2346 	cur_stack = 1;
2347 
2348 peek_stack:
2349 	if (cur_stack == 0)
2350 		goto check_state;
2351 	t = insn_stack[cur_stack - 1];
2352 
2353 	if (BPF_CLASS(insns[t].code) == BPF_JMP) {
2354 		u8 opcode = BPF_OP(insns[t].code);
2355 
2356 		if (opcode == BPF_EXIT) {
2357 			goto mark_explored;
2358 		} else if (opcode == BPF_CALL) {
2359 			ret = push_insn(t, t + 1, FALLTHROUGH, env);
2360 			if (ret == 1)
2361 				goto peek_stack;
2362 			else if (ret < 0)
2363 				goto err_free;
2364 			if (t + 1 < insn_cnt)
2365 				env->explored_states[t + 1] = STATE_LIST_MARK;
2366 		} else if (opcode == BPF_JA) {
2367 			if (BPF_SRC(insns[t].code) != BPF_K) {
2368 				ret = -EINVAL;
2369 				goto err_free;
2370 			}
2371 			/* unconditional jump with single edge */
2372 			ret = push_insn(t, t + insns[t].off + 1,
2373 					FALLTHROUGH, env);
2374 			if (ret == 1)
2375 				goto peek_stack;
2376 			else if (ret < 0)
2377 				goto err_free;
2378 			/* tell verifier to check for equivalent states
2379 			 * after every call and jump
2380 			 */
2381 			if (t + 1 < insn_cnt)
2382 				env->explored_states[t + 1] = STATE_LIST_MARK;
2383 		} else {
2384 			/* conditional jump with two edges */
2385 			ret = push_insn(t, t + 1, FALLTHROUGH, env);
2386 			if (ret == 1)
2387 				goto peek_stack;
2388 			else if (ret < 0)
2389 				goto err_free;
2390 
2391 			ret = push_insn(t, t + insns[t].off + 1, BRANCH, env);
2392 			if (ret == 1)
2393 				goto peek_stack;
2394 			else if (ret < 0)
2395 				goto err_free;
2396 		}
2397 	} else {
2398 		/* all other non-branch instructions with single
2399 		 * fall-through edge
2400 		 */
2401 		ret = push_insn(t, t + 1, FALLTHROUGH, env);
2402 		if (ret == 1)
2403 			goto peek_stack;
2404 		else if (ret < 0)
2405 			goto err_free;
2406 	}
2407 
2408 mark_explored:
2409 	insn_state[t] = EXPLORED;
2410 	if (cur_stack-- <= 0) {
2411 		verbose("pop stack internal bug\n");
2412 		ret = -EFAULT;
2413 		goto err_free;
2414 	}
2415 	goto peek_stack;
2416 
2417 check_state:
2418 	for (i = 0; i < insn_cnt; i++) {
2419 		if (insn_state[i] != EXPLORED) {
2420 			verbose("unreachable insn %d\n", i);
2421 			ret = -EINVAL;
2422 			goto err_free;
2423 		}
2424 	}
2425 	ret = 0; /* cfg looks good */
2426 
2427 err_free:
2428 	kfree(insn_state);
2429 	kfree(insn_stack);
2430 	return ret;
2431 }
2432 
2433 /* the following conditions reduce the number of explored insns
2434  * from ~140k to ~80k for ultra large programs that use a lot of ptr_to_packet
2435  */
2436 static bool compare_ptrs_to_packet(struct bpf_reg_state *old,
2437 				   struct bpf_reg_state *cur)
2438 {
2439 	if (old->id != cur->id)
2440 		return false;
2441 
2442 	/* old ptr_to_packet is more conservative, since it allows smaller
2443 	 * range. Ex:
2444 	 * old(off=0,r=10) is equal to cur(off=0,r=20), because
2445 	 * old(off=0,r=10) means that with range=10 the verifier proceeded
2446 	 * further and found no issues with the program. Now we're in the same
2447 	 * spot with cur(off=0,r=20), so we're safe too, since anything further
2448 	 * will only be looking at most 10 bytes after this pointer.
2449 	 */
2450 	if (old->off == cur->off && old->range < cur->range)
2451 		return true;
2452 
2453 	/* old(off=20,r=10) is equal to cur(off=22,re=22 or 5 or 0)
2454 	 * since both cannot be used for packet access and safe(old)
2455 	 * pointer has smaller off that could be used for further
2456 	 * 'if (ptr > data_end)' check
2457 	 * Ex:
2458 	 * old(off=20,r=10) and cur(off=22,r=22) and cur(off=22,r=0) mean
2459 	 * that we cannot access the packet.
2460 	 * The safe range is:
2461 	 * [ptr, ptr + range - off)
2462 	 * so whenever off >=range, it means no safe bytes from this pointer.
2463 	 * When comparing old->off <= cur->off, it means that older code
2464 	 * went with smaller offset and that offset was later
2465 	 * used to figure out the safe range after 'if (ptr > data_end)' check
2466 	 * Say, 'old' state was explored like:
2467 	 * ... R3(off=0, r=0)
2468 	 * R4 = R3 + 20
2469 	 * ... now R4(off=20,r=0)  <-- here
2470 	 * if (R4 > data_end)
2471 	 * ... R4(off=20,r=20), R3(off=0,r=20) and R3 can be used to access.
2472 	 * ... the code further went all the way to bpf_exit.
2473 	 * Now the 'cur' state at the mark 'here' has R4(off=30,r=0).
2474 	 * old_R4(off=20,r=0) equal to cur_R4(off=30,r=0), since if the verifier
2475 	 * goes further, such cur_R4 will give larger safe packet range after
2476 	 * 'if (R4 > data_end)' and all further insn were already good with r=20,
2477 	 * so they will be good with r=30 and we can prune the search.
2478 	 */
2479 	if (old->off <= cur->off &&
2480 	    old->off >= old->range && cur->off >= cur->range)
2481 		return true;
2482 
2483 	return false;
2484 }
2485 
2486 /* compare two verifier states
2487  *
2488  * all states stored in state_list are known to be valid, since
2489  * verifier reached 'bpf_exit' instruction through them
2490  *
2491  * this function is called when verifier exploring different branches of
2492  * execution popped from the state stack. If it sees an old state that has
2493  * more strict register state and more strict stack state then this execution
2494  * branch doesn't need to be explored further, since verifier already
2495  * concluded that more strict state leads to valid finish.
2496  *
2497  * Therefore two states are equivalent if register state is more conservative
2498  * and explored stack state is more conservative than the current one.
2499  * Example:
2500  *       explored                   current
2501  * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC)
2502  * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC)
2503  *
2504  * In other words if current stack state (one being explored) has more
2505  * valid slots than old one that already passed validation, it means
2506  * the verifier can stop exploring and conclude that current state is valid too
2507  *
2508  * Similarly with registers. If explored state has register type as invalid
2509  * whereas register type in current state is meaningful, it means that
2510  * the current state will reach 'bpf_exit' instruction safely
2511  */
2512 static bool states_equal(struct bpf_verifier_env *env,
2513 			 struct bpf_verifier_state *old,
2514 			 struct bpf_verifier_state *cur)
2515 {
2516 	bool varlen_map_access = env->varlen_map_value_access;
2517 	struct bpf_reg_state *rold, *rcur;
2518 	int i;
2519 
2520 	for (i = 0; i < MAX_BPF_REG; i++) {
2521 		rold = &old->regs[i];
2522 		rcur = &cur->regs[i];
2523 
2524 		if (memcmp(rold, rcur, sizeof(*rold)) == 0)
2525 			continue;
2526 
2527 		/* If the ranges were not the same, but everything else was and
2528 		 * we didn't do a variable access into a map then we are a-ok.
2529 		 */
2530 		if (!varlen_map_access &&
2531 		    memcmp(rold, rcur, offsetofend(struct bpf_reg_state, id)) == 0)
2532 			continue;
2533 
2534 		/* If we didn't map access then again we don't care about the
2535 		 * mismatched range values and it's ok if our old type was
2536 		 * UNKNOWN and we didn't go to a NOT_INIT'ed reg.
2537 		 */
2538 		if (rold->type == NOT_INIT ||
2539 		    (!varlen_map_access && rold->type == UNKNOWN_VALUE &&
2540 		     rcur->type != NOT_INIT))
2541 			continue;
2542 
2543 		if (rold->type == PTR_TO_PACKET && rcur->type == PTR_TO_PACKET &&
2544 		    compare_ptrs_to_packet(rold, rcur))
2545 			continue;
2546 
2547 		return false;
2548 	}
2549 
2550 	for (i = 0; i < MAX_BPF_STACK; i++) {
2551 		if (old->stack_slot_type[i] == STACK_INVALID)
2552 			continue;
2553 		if (old->stack_slot_type[i] != cur->stack_slot_type[i])
2554 			/* Ex: old explored (safe) state has STACK_SPILL in
2555 			 * this stack slot, but current has has STACK_MISC ->
2556 			 * this verifier states are not equivalent,
2557 			 * return false to continue verification of this path
2558 			 */
2559 			return false;
2560 		if (i % BPF_REG_SIZE)
2561 			continue;
2562 		if (memcmp(&old->spilled_regs[i / BPF_REG_SIZE],
2563 			   &cur->spilled_regs[i / BPF_REG_SIZE],
2564 			   sizeof(old->spilled_regs[0])))
2565 			/* when explored and current stack slot types are
2566 			 * the same, check that stored pointers types
2567 			 * are the same as well.
2568 			 * Ex: explored safe path could have stored
2569 			 * (bpf_reg_state) {.type = PTR_TO_STACK, .imm = -8}
2570 			 * but current path has stored:
2571 			 * (bpf_reg_state) {.type = PTR_TO_STACK, .imm = -16}
2572 			 * such verifier states are not equivalent.
2573 			 * return false to continue verification of this path
2574 			 */
2575 			return false;
2576 		else
2577 			continue;
2578 	}
2579 	return true;
2580 }
2581 
2582 static int is_state_visited(struct bpf_verifier_env *env, int insn_idx)
2583 {
2584 	struct bpf_verifier_state_list *new_sl;
2585 	struct bpf_verifier_state_list *sl;
2586 
2587 	sl = env->explored_states[insn_idx];
2588 	if (!sl)
2589 		/* this 'insn_idx' instruction wasn't marked, so we will not
2590 		 * be doing state search here
2591 		 */
2592 		return 0;
2593 
2594 	while (sl != STATE_LIST_MARK) {
2595 		if (states_equal(env, &sl->state, &env->cur_state))
2596 			/* reached equivalent register/stack state,
2597 			 * prune the search
2598 			 */
2599 			return 1;
2600 		sl = sl->next;
2601 	}
2602 
2603 	/* there were no equivalent states, remember current one.
2604 	 * technically the current state is not proven to be safe yet,
2605 	 * but it will either reach bpf_exit (which means it's safe) or
2606 	 * it will be rejected. Since there are no loops, we won't be
2607 	 * seeing this 'insn_idx' instruction again on the way to bpf_exit
2608 	 */
2609 	new_sl = kmalloc(sizeof(struct bpf_verifier_state_list), GFP_USER);
2610 	if (!new_sl)
2611 		return -ENOMEM;
2612 
2613 	/* add new state to the head of linked list */
2614 	memcpy(&new_sl->state, &env->cur_state, sizeof(env->cur_state));
2615 	new_sl->next = env->explored_states[insn_idx];
2616 	env->explored_states[insn_idx] = new_sl;
2617 	return 0;
2618 }
2619 
2620 static int ext_analyzer_insn_hook(struct bpf_verifier_env *env,
2621 				  int insn_idx, int prev_insn_idx)
2622 {
2623 	if (!env->analyzer_ops || !env->analyzer_ops->insn_hook)
2624 		return 0;
2625 
2626 	return env->analyzer_ops->insn_hook(env, insn_idx, prev_insn_idx);
2627 }
2628 
2629 static int do_check(struct bpf_verifier_env *env)
2630 {
2631 	struct bpf_verifier_state *state = &env->cur_state;
2632 	struct bpf_insn *insns = env->prog->insnsi;
2633 	struct bpf_reg_state *regs = state->regs;
2634 	int insn_cnt = env->prog->len;
2635 	int insn_idx, prev_insn_idx = 0;
2636 	int insn_processed = 0;
2637 	bool do_print_state = false;
2638 
2639 	init_reg_state(regs);
2640 	insn_idx = 0;
2641 	env->varlen_map_value_access = false;
2642 	for (;;) {
2643 		struct bpf_insn *insn;
2644 		u8 class;
2645 		int err;
2646 
2647 		if (insn_idx >= insn_cnt) {
2648 			verbose("invalid insn idx %d insn_cnt %d\n",
2649 				insn_idx, insn_cnt);
2650 			return -EFAULT;
2651 		}
2652 
2653 		insn = &insns[insn_idx];
2654 		class = BPF_CLASS(insn->code);
2655 
2656 		if (++insn_processed > BPF_COMPLEXITY_LIMIT_INSNS) {
2657 			verbose("BPF program is too large. Proccessed %d insn\n",
2658 				insn_processed);
2659 			return -E2BIG;
2660 		}
2661 
2662 		err = is_state_visited(env, insn_idx);
2663 		if (err < 0)
2664 			return err;
2665 		if (err == 1) {
2666 			/* found equivalent state, can prune the search */
2667 			if (log_level) {
2668 				if (do_print_state)
2669 					verbose("\nfrom %d to %d: safe\n",
2670 						prev_insn_idx, insn_idx);
2671 				else
2672 					verbose("%d: safe\n", insn_idx);
2673 			}
2674 			goto process_bpf_exit;
2675 		}
2676 
2677 		if (log_level && do_print_state) {
2678 			verbose("\nfrom %d to %d:", prev_insn_idx, insn_idx);
2679 			print_verifier_state(&env->cur_state);
2680 			do_print_state = false;
2681 		}
2682 
2683 		if (log_level) {
2684 			verbose("%d: ", insn_idx);
2685 			print_bpf_insn(insn);
2686 		}
2687 
2688 		err = ext_analyzer_insn_hook(env, insn_idx, prev_insn_idx);
2689 		if (err)
2690 			return err;
2691 
2692 		if (class == BPF_ALU || class == BPF_ALU64) {
2693 			err = check_alu_op(env, insn);
2694 			if (err)
2695 				return err;
2696 
2697 		} else if (class == BPF_LDX) {
2698 			enum bpf_reg_type *prev_src_type, src_reg_type;
2699 
2700 			/* check for reserved fields is already done */
2701 
2702 			/* check src operand */
2703 			err = check_reg_arg(regs, insn->src_reg, SRC_OP);
2704 			if (err)
2705 				return err;
2706 
2707 			err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK);
2708 			if (err)
2709 				return err;
2710 
2711 			src_reg_type = regs[insn->src_reg].type;
2712 
2713 			/* check that memory (src_reg + off) is readable,
2714 			 * the state of dst_reg will be updated by this func
2715 			 */
2716 			err = check_mem_access(env, insn->src_reg, insn->off,
2717 					       BPF_SIZE(insn->code), BPF_READ,
2718 					       insn->dst_reg);
2719 			if (err)
2720 				return err;
2721 
2722 			reset_reg_range_values(regs, insn->dst_reg);
2723 			if (BPF_SIZE(insn->code) != BPF_W &&
2724 			    BPF_SIZE(insn->code) != BPF_DW) {
2725 				insn_idx++;
2726 				continue;
2727 			}
2728 
2729 			prev_src_type = &env->insn_aux_data[insn_idx].ptr_type;
2730 
2731 			if (*prev_src_type == NOT_INIT) {
2732 				/* saw a valid insn
2733 				 * dst_reg = *(u32 *)(src_reg + off)
2734 				 * save type to validate intersecting paths
2735 				 */
2736 				*prev_src_type = src_reg_type;
2737 
2738 			} else if (src_reg_type != *prev_src_type &&
2739 				   (src_reg_type == PTR_TO_CTX ||
2740 				    *prev_src_type == PTR_TO_CTX)) {
2741 				/* ABuser program is trying to use the same insn
2742 				 * dst_reg = *(u32*) (src_reg + off)
2743 				 * with different pointer types:
2744 				 * src_reg == ctx in one branch and
2745 				 * src_reg == stack|map in some other branch.
2746 				 * Reject it.
2747 				 */
2748 				verbose("same insn cannot be used with different pointers\n");
2749 				return -EINVAL;
2750 			}
2751 
2752 		} else if (class == BPF_STX) {
2753 			enum bpf_reg_type *prev_dst_type, dst_reg_type;
2754 
2755 			if (BPF_MODE(insn->code) == BPF_XADD) {
2756 				err = check_xadd(env, insn);
2757 				if (err)
2758 					return err;
2759 				insn_idx++;
2760 				continue;
2761 			}
2762 
2763 			/* check src1 operand */
2764 			err = check_reg_arg(regs, insn->src_reg, SRC_OP);
2765 			if (err)
2766 				return err;
2767 			/* check src2 operand */
2768 			err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
2769 			if (err)
2770 				return err;
2771 
2772 			dst_reg_type = regs[insn->dst_reg].type;
2773 
2774 			/* check that memory (dst_reg + off) is writeable */
2775 			err = check_mem_access(env, insn->dst_reg, insn->off,
2776 					       BPF_SIZE(insn->code), BPF_WRITE,
2777 					       insn->src_reg);
2778 			if (err)
2779 				return err;
2780 
2781 			prev_dst_type = &env->insn_aux_data[insn_idx].ptr_type;
2782 
2783 			if (*prev_dst_type == NOT_INIT) {
2784 				*prev_dst_type = dst_reg_type;
2785 			} else if (dst_reg_type != *prev_dst_type &&
2786 				   (dst_reg_type == PTR_TO_CTX ||
2787 				    *prev_dst_type == PTR_TO_CTX)) {
2788 				verbose("same insn cannot be used with different pointers\n");
2789 				return -EINVAL;
2790 			}
2791 
2792 		} else if (class == BPF_ST) {
2793 			if (BPF_MODE(insn->code) != BPF_MEM ||
2794 			    insn->src_reg != BPF_REG_0) {
2795 				verbose("BPF_ST uses reserved fields\n");
2796 				return -EINVAL;
2797 			}
2798 			/* check src operand */
2799 			err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
2800 			if (err)
2801 				return err;
2802 
2803 			/* check that memory (dst_reg + off) is writeable */
2804 			err = check_mem_access(env, insn->dst_reg, insn->off,
2805 					       BPF_SIZE(insn->code), BPF_WRITE,
2806 					       -1);
2807 			if (err)
2808 				return err;
2809 
2810 		} else if (class == BPF_JMP) {
2811 			u8 opcode = BPF_OP(insn->code);
2812 
2813 			if (opcode == BPF_CALL) {
2814 				if (BPF_SRC(insn->code) != BPF_K ||
2815 				    insn->off != 0 ||
2816 				    insn->src_reg != BPF_REG_0 ||
2817 				    insn->dst_reg != BPF_REG_0) {
2818 					verbose("BPF_CALL uses reserved fields\n");
2819 					return -EINVAL;
2820 				}
2821 
2822 				err = check_call(env, insn->imm);
2823 				if (err)
2824 					return err;
2825 
2826 			} else if (opcode == BPF_JA) {
2827 				if (BPF_SRC(insn->code) != BPF_K ||
2828 				    insn->imm != 0 ||
2829 				    insn->src_reg != BPF_REG_0 ||
2830 				    insn->dst_reg != BPF_REG_0) {
2831 					verbose("BPF_JA uses reserved fields\n");
2832 					return -EINVAL;
2833 				}
2834 
2835 				insn_idx += insn->off + 1;
2836 				continue;
2837 
2838 			} else if (opcode == BPF_EXIT) {
2839 				if (BPF_SRC(insn->code) != BPF_K ||
2840 				    insn->imm != 0 ||
2841 				    insn->src_reg != BPF_REG_0 ||
2842 				    insn->dst_reg != BPF_REG_0) {
2843 					verbose("BPF_EXIT uses reserved fields\n");
2844 					return -EINVAL;
2845 				}
2846 
2847 				/* eBPF calling convetion is such that R0 is used
2848 				 * to return the value from eBPF program.
2849 				 * Make sure that it's readable at this time
2850 				 * of bpf_exit, which means that program wrote
2851 				 * something into it earlier
2852 				 */
2853 				err = check_reg_arg(regs, BPF_REG_0, SRC_OP);
2854 				if (err)
2855 					return err;
2856 
2857 				if (is_pointer_value(env, BPF_REG_0)) {
2858 					verbose("R0 leaks addr as return value\n");
2859 					return -EACCES;
2860 				}
2861 
2862 process_bpf_exit:
2863 				insn_idx = pop_stack(env, &prev_insn_idx);
2864 				if (insn_idx < 0) {
2865 					break;
2866 				} else {
2867 					do_print_state = true;
2868 					continue;
2869 				}
2870 			} else {
2871 				err = check_cond_jmp_op(env, insn, &insn_idx);
2872 				if (err)
2873 					return err;
2874 			}
2875 		} else if (class == BPF_LD) {
2876 			u8 mode = BPF_MODE(insn->code);
2877 
2878 			if (mode == BPF_ABS || mode == BPF_IND) {
2879 				err = check_ld_abs(env, insn);
2880 				if (err)
2881 					return err;
2882 
2883 			} else if (mode == BPF_IMM) {
2884 				err = check_ld_imm(env, insn);
2885 				if (err)
2886 					return err;
2887 
2888 				insn_idx++;
2889 			} else {
2890 				verbose("invalid BPF_LD mode\n");
2891 				return -EINVAL;
2892 			}
2893 			reset_reg_range_values(regs, insn->dst_reg);
2894 		} else {
2895 			verbose("unknown insn class %d\n", class);
2896 			return -EINVAL;
2897 		}
2898 
2899 		insn_idx++;
2900 	}
2901 
2902 	verbose("processed %d insns\n", insn_processed);
2903 	return 0;
2904 }
2905 
2906 static int check_map_prog_compatibility(struct bpf_map *map,
2907 					struct bpf_prog *prog)
2908 
2909 {
2910 	if (prog->type == BPF_PROG_TYPE_PERF_EVENT &&
2911 	    (map->map_type == BPF_MAP_TYPE_HASH ||
2912 	     map->map_type == BPF_MAP_TYPE_PERCPU_HASH) &&
2913 	    (map->map_flags & BPF_F_NO_PREALLOC)) {
2914 		verbose("perf_event programs can only use preallocated hash map\n");
2915 		return -EINVAL;
2916 	}
2917 	return 0;
2918 }
2919 
2920 /* look for pseudo eBPF instructions that access map FDs and
2921  * replace them with actual map pointers
2922  */
2923 static int replace_map_fd_with_map_ptr(struct bpf_verifier_env *env)
2924 {
2925 	struct bpf_insn *insn = env->prog->insnsi;
2926 	int insn_cnt = env->prog->len;
2927 	int i, j, err;
2928 
2929 	for (i = 0; i < insn_cnt; i++, insn++) {
2930 		if (BPF_CLASS(insn->code) == BPF_LDX &&
2931 		    (BPF_MODE(insn->code) != BPF_MEM || insn->imm != 0)) {
2932 			verbose("BPF_LDX uses reserved fields\n");
2933 			return -EINVAL;
2934 		}
2935 
2936 		if (BPF_CLASS(insn->code) == BPF_STX &&
2937 		    ((BPF_MODE(insn->code) != BPF_MEM &&
2938 		      BPF_MODE(insn->code) != BPF_XADD) || insn->imm != 0)) {
2939 			verbose("BPF_STX uses reserved fields\n");
2940 			return -EINVAL;
2941 		}
2942 
2943 		if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW)) {
2944 			struct bpf_map *map;
2945 			struct fd f;
2946 
2947 			if (i == insn_cnt - 1 || insn[1].code != 0 ||
2948 			    insn[1].dst_reg != 0 || insn[1].src_reg != 0 ||
2949 			    insn[1].off != 0) {
2950 				verbose("invalid bpf_ld_imm64 insn\n");
2951 				return -EINVAL;
2952 			}
2953 
2954 			if (insn->src_reg == 0)
2955 				/* valid generic load 64-bit imm */
2956 				goto next_insn;
2957 
2958 			if (insn->src_reg != BPF_PSEUDO_MAP_FD) {
2959 				verbose("unrecognized bpf_ld_imm64 insn\n");
2960 				return -EINVAL;
2961 			}
2962 
2963 			f = fdget(insn->imm);
2964 			map = __bpf_map_get(f);
2965 			if (IS_ERR(map)) {
2966 				verbose("fd %d is not pointing to valid bpf_map\n",
2967 					insn->imm);
2968 				return PTR_ERR(map);
2969 			}
2970 
2971 			err = check_map_prog_compatibility(map, env->prog);
2972 			if (err) {
2973 				fdput(f);
2974 				return err;
2975 			}
2976 
2977 			/* store map pointer inside BPF_LD_IMM64 instruction */
2978 			insn[0].imm = (u32) (unsigned long) map;
2979 			insn[1].imm = ((u64) (unsigned long) map) >> 32;
2980 
2981 			/* check whether we recorded this map already */
2982 			for (j = 0; j < env->used_map_cnt; j++)
2983 				if (env->used_maps[j] == map) {
2984 					fdput(f);
2985 					goto next_insn;
2986 				}
2987 
2988 			if (env->used_map_cnt >= MAX_USED_MAPS) {
2989 				fdput(f);
2990 				return -E2BIG;
2991 			}
2992 
2993 			/* hold the map. If the program is rejected by verifier,
2994 			 * the map will be released by release_maps() or it
2995 			 * will be used by the valid program until it's unloaded
2996 			 * and all maps are released in free_bpf_prog_info()
2997 			 */
2998 			map = bpf_map_inc(map, false);
2999 			if (IS_ERR(map)) {
3000 				fdput(f);
3001 				return PTR_ERR(map);
3002 			}
3003 			env->used_maps[env->used_map_cnt++] = map;
3004 
3005 			fdput(f);
3006 next_insn:
3007 			insn++;
3008 			i++;
3009 		}
3010 	}
3011 
3012 	/* now all pseudo BPF_LD_IMM64 instructions load valid
3013 	 * 'struct bpf_map *' into a register instead of user map_fd.
3014 	 * These pointers will be used later by verifier to validate map access.
3015 	 */
3016 	return 0;
3017 }
3018 
3019 /* drop refcnt of maps used by the rejected program */
3020 static void release_maps(struct bpf_verifier_env *env)
3021 {
3022 	int i;
3023 
3024 	for (i = 0; i < env->used_map_cnt; i++)
3025 		bpf_map_put(env->used_maps[i]);
3026 }
3027 
3028 /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */
3029 static void convert_pseudo_ld_imm64(struct bpf_verifier_env *env)
3030 {
3031 	struct bpf_insn *insn = env->prog->insnsi;
3032 	int insn_cnt = env->prog->len;
3033 	int i;
3034 
3035 	for (i = 0; i < insn_cnt; i++, insn++)
3036 		if (insn->code == (BPF_LD | BPF_IMM | BPF_DW))
3037 			insn->src_reg = 0;
3038 }
3039 
3040 /* convert load instructions that access fields of 'struct __sk_buff'
3041  * into sequence of instructions that access fields of 'struct sk_buff'
3042  */
3043 static int convert_ctx_accesses(struct bpf_verifier_env *env)
3044 {
3045 	const struct bpf_verifier_ops *ops = env->prog->aux->ops;
3046 	const int insn_cnt = env->prog->len;
3047 	struct bpf_insn insn_buf[16], *insn;
3048 	struct bpf_prog *new_prog;
3049 	enum bpf_access_type type;
3050 	int i, cnt, delta = 0;
3051 
3052 	if (ops->gen_prologue) {
3053 		cnt = ops->gen_prologue(insn_buf, env->seen_direct_write,
3054 					env->prog);
3055 		if (cnt >= ARRAY_SIZE(insn_buf)) {
3056 			verbose("bpf verifier is misconfigured\n");
3057 			return -EINVAL;
3058 		} else if (cnt) {
3059 			new_prog = bpf_patch_insn_single(env->prog, 0,
3060 							 insn_buf, cnt);
3061 			if (!new_prog)
3062 				return -ENOMEM;
3063 			env->prog = new_prog;
3064 			delta += cnt - 1;
3065 		}
3066 	}
3067 
3068 	if (!ops->convert_ctx_access)
3069 		return 0;
3070 
3071 	insn = env->prog->insnsi + delta;
3072 
3073 	for (i = 0; i < insn_cnt; i++, insn++) {
3074 		if (insn->code == (BPF_LDX | BPF_MEM | BPF_W) ||
3075 		    insn->code == (BPF_LDX | BPF_MEM | BPF_DW))
3076 			type = BPF_READ;
3077 		else if (insn->code == (BPF_STX | BPF_MEM | BPF_W) ||
3078 			 insn->code == (BPF_STX | BPF_MEM | BPF_DW))
3079 			type = BPF_WRITE;
3080 		else
3081 			continue;
3082 
3083 		if (env->insn_aux_data[i].ptr_type != PTR_TO_CTX)
3084 			continue;
3085 
3086 		cnt = ops->convert_ctx_access(type, insn->dst_reg, insn->src_reg,
3087 					      insn->off, insn_buf, env->prog);
3088 		if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) {
3089 			verbose("bpf verifier is misconfigured\n");
3090 			return -EINVAL;
3091 		}
3092 
3093 		new_prog = bpf_patch_insn_single(env->prog, i + delta, insn_buf,
3094 						 cnt);
3095 		if (!new_prog)
3096 			return -ENOMEM;
3097 
3098 		delta += cnt - 1;
3099 
3100 		/* keep walking new program and skip insns we just inserted */
3101 		env->prog = new_prog;
3102 		insn      = new_prog->insnsi + i + delta;
3103 	}
3104 
3105 	return 0;
3106 }
3107 
3108 static void free_states(struct bpf_verifier_env *env)
3109 {
3110 	struct bpf_verifier_state_list *sl, *sln;
3111 	int i;
3112 
3113 	if (!env->explored_states)
3114 		return;
3115 
3116 	for (i = 0; i < env->prog->len; i++) {
3117 		sl = env->explored_states[i];
3118 
3119 		if (sl)
3120 			while (sl != STATE_LIST_MARK) {
3121 				sln = sl->next;
3122 				kfree(sl);
3123 				sl = sln;
3124 			}
3125 	}
3126 
3127 	kfree(env->explored_states);
3128 }
3129 
3130 int bpf_check(struct bpf_prog **prog, union bpf_attr *attr)
3131 {
3132 	char __user *log_ubuf = NULL;
3133 	struct bpf_verifier_env *env;
3134 	int ret = -EINVAL;
3135 
3136 	/* 'struct bpf_verifier_env' can be global, but since it's not small,
3137 	 * allocate/free it every time bpf_check() is called
3138 	 */
3139 	env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL);
3140 	if (!env)
3141 		return -ENOMEM;
3142 
3143 	env->insn_aux_data = vzalloc(sizeof(struct bpf_insn_aux_data) *
3144 				     (*prog)->len);
3145 	ret = -ENOMEM;
3146 	if (!env->insn_aux_data)
3147 		goto err_free_env;
3148 	env->prog = *prog;
3149 
3150 	/* grab the mutex to protect few globals used by verifier */
3151 	mutex_lock(&bpf_verifier_lock);
3152 
3153 	if (attr->log_level || attr->log_buf || attr->log_size) {
3154 		/* user requested verbose verifier output
3155 		 * and supplied buffer to store the verification trace
3156 		 */
3157 		log_level = attr->log_level;
3158 		log_ubuf = (char __user *) (unsigned long) attr->log_buf;
3159 		log_size = attr->log_size;
3160 		log_len = 0;
3161 
3162 		ret = -EINVAL;
3163 		/* log_* values have to be sane */
3164 		if (log_size < 128 || log_size > UINT_MAX >> 8 ||
3165 		    log_level == 0 || log_ubuf == NULL)
3166 			goto err_unlock;
3167 
3168 		ret = -ENOMEM;
3169 		log_buf = vmalloc(log_size);
3170 		if (!log_buf)
3171 			goto err_unlock;
3172 	} else {
3173 		log_level = 0;
3174 	}
3175 
3176 	bpf_prog_calc_digest(env->prog);
3177 
3178 	ret = replace_map_fd_with_map_ptr(env);
3179 	if (ret < 0)
3180 		goto skip_full_check;
3181 
3182 	env->explored_states = kcalloc(env->prog->len,
3183 				       sizeof(struct bpf_verifier_state_list *),
3184 				       GFP_USER);
3185 	ret = -ENOMEM;
3186 	if (!env->explored_states)
3187 		goto skip_full_check;
3188 
3189 	ret = check_cfg(env);
3190 	if (ret < 0)
3191 		goto skip_full_check;
3192 
3193 	env->allow_ptr_leaks = capable(CAP_SYS_ADMIN);
3194 
3195 	ret = do_check(env);
3196 
3197 skip_full_check:
3198 	while (pop_stack(env, NULL) >= 0);
3199 	free_states(env);
3200 
3201 	if (ret == 0)
3202 		/* program is valid, convert *(u32*)(ctx + off) accesses */
3203 		ret = convert_ctx_accesses(env);
3204 
3205 	if (log_level && log_len >= log_size - 1) {
3206 		BUG_ON(log_len >= log_size);
3207 		/* verifier log exceeded user supplied buffer */
3208 		ret = -ENOSPC;
3209 		/* fall through to return what was recorded */
3210 	}
3211 
3212 	/* copy verifier log back to user space including trailing zero */
3213 	if (log_level && copy_to_user(log_ubuf, log_buf, log_len + 1) != 0) {
3214 		ret = -EFAULT;
3215 		goto free_log_buf;
3216 	}
3217 
3218 	if (ret == 0 && env->used_map_cnt) {
3219 		/* if program passed verifier, update used_maps in bpf_prog_info */
3220 		env->prog->aux->used_maps = kmalloc_array(env->used_map_cnt,
3221 							  sizeof(env->used_maps[0]),
3222 							  GFP_KERNEL);
3223 
3224 		if (!env->prog->aux->used_maps) {
3225 			ret = -ENOMEM;
3226 			goto free_log_buf;
3227 		}
3228 
3229 		memcpy(env->prog->aux->used_maps, env->used_maps,
3230 		       sizeof(env->used_maps[0]) * env->used_map_cnt);
3231 		env->prog->aux->used_map_cnt = env->used_map_cnt;
3232 
3233 		/* program is valid. Convert pseudo bpf_ld_imm64 into generic
3234 		 * bpf_ld_imm64 instructions
3235 		 */
3236 		convert_pseudo_ld_imm64(env);
3237 	}
3238 
3239 free_log_buf:
3240 	if (log_level)
3241 		vfree(log_buf);
3242 	if (!env->prog->aux->used_maps)
3243 		/* if we didn't copy map pointers into bpf_prog_info, release
3244 		 * them now. Otherwise free_bpf_prog_info() will release them.
3245 		 */
3246 		release_maps(env);
3247 	*prog = env->prog;
3248 err_unlock:
3249 	mutex_unlock(&bpf_verifier_lock);
3250 	vfree(env->insn_aux_data);
3251 err_free_env:
3252 	kfree(env);
3253 	return ret;
3254 }
3255 
3256 int bpf_analyzer(struct bpf_prog *prog, const struct bpf_ext_analyzer_ops *ops,
3257 		 void *priv)
3258 {
3259 	struct bpf_verifier_env *env;
3260 	int ret;
3261 
3262 	env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL);
3263 	if (!env)
3264 		return -ENOMEM;
3265 
3266 	env->insn_aux_data = vzalloc(sizeof(struct bpf_insn_aux_data) *
3267 				     prog->len);
3268 	ret = -ENOMEM;
3269 	if (!env->insn_aux_data)
3270 		goto err_free_env;
3271 	env->prog = prog;
3272 	env->analyzer_ops = ops;
3273 	env->analyzer_priv = priv;
3274 
3275 	/* grab the mutex to protect few globals used by verifier */
3276 	mutex_lock(&bpf_verifier_lock);
3277 
3278 	log_level = 0;
3279 
3280 	env->explored_states = kcalloc(env->prog->len,
3281 				       sizeof(struct bpf_verifier_state_list *),
3282 				       GFP_KERNEL);
3283 	ret = -ENOMEM;
3284 	if (!env->explored_states)
3285 		goto skip_full_check;
3286 
3287 	ret = check_cfg(env);
3288 	if (ret < 0)
3289 		goto skip_full_check;
3290 
3291 	env->allow_ptr_leaks = capable(CAP_SYS_ADMIN);
3292 
3293 	ret = do_check(env);
3294 
3295 skip_full_check:
3296 	while (pop_stack(env, NULL) >= 0);
3297 	free_states(env);
3298 
3299 	mutex_unlock(&bpf_verifier_lock);
3300 	vfree(env->insn_aux_data);
3301 err_free_env:
3302 	kfree(env);
3303 	return ret;
3304 }
3305 EXPORT_SYMBOL_GPL(bpf_analyzer);
3306