xref: /openbmc/linux/kernel/bpf/verifier.c (revision a977d045)
1 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
2  * Copyright (c) 2016 Facebook
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of version 2 of the GNU General Public
6  * License as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful, but
9  * WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11  * General Public License for more details.
12  */
13 #include <linux/kernel.h>
14 #include <linux/types.h>
15 #include <linux/slab.h>
16 #include <linux/bpf.h>
17 #include <linux/bpf_verifier.h>
18 #include <linux/filter.h>
19 #include <net/netlink.h>
20 #include <linux/file.h>
21 #include <linux/vmalloc.h>
22 #include <linux/stringify.h>
23 
24 /* bpf_check() is a static code analyzer that walks eBPF program
25  * instruction by instruction and updates register/stack state.
26  * All paths of conditional branches are analyzed until 'bpf_exit' insn.
27  *
28  * The first pass is depth-first-search to check that the program is a DAG.
29  * It rejects the following programs:
30  * - larger than BPF_MAXINSNS insns
31  * - if loop is present (detected via back-edge)
32  * - unreachable insns exist (shouldn't be a forest. program = one function)
33  * - out of bounds or malformed jumps
34  * The second pass is all possible path descent from the 1st insn.
35  * Since it's analyzing all pathes through the program, the length of the
36  * analysis is limited to 64k insn, which may be hit even if total number of
37  * insn is less then 4K, but there are too many branches that change stack/regs.
38  * Number of 'branches to be analyzed' is limited to 1k
39  *
40  * On entry to each instruction, each register has a type, and the instruction
41  * changes the types of the registers depending on instruction semantics.
42  * If instruction is BPF_MOV64_REG(BPF_REG_1, BPF_REG_5), then type of R5 is
43  * copied to R1.
44  *
45  * All registers are 64-bit.
46  * R0 - return register
47  * R1-R5 argument passing registers
48  * R6-R9 callee saved registers
49  * R10 - frame pointer read-only
50  *
51  * At the start of BPF program the register R1 contains a pointer to bpf_context
52  * and has type PTR_TO_CTX.
53  *
54  * Verifier tracks arithmetic operations on pointers in case:
55  *    BPF_MOV64_REG(BPF_REG_1, BPF_REG_10),
56  *    BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -20),
57  * 1st insn copies R10 (which has FRAME_PTR) type into R1
58  * and 2nd arithmetic instruction is pattern matched to recognize
59  * that it wants to construct a pointer to some element within stack.
60  * So after 2nd insn, the register R1 has type PTR_TO_STACK
61  * (and -20 constant is saved for further stack bounds checking).
62  * Meaning that this reg is a pointer to stack plus known immediate constant.
63  *
64  * Most of the time the registers have UNKNOWN_VALUE type, which
65  * means the register has some value, but it's not a valid pointer.
66  * (like pointer plus pointer becomes UNKNOWN_VALUE type)
67  *
68  * When verifier sees load or store instructions the type of base register
69  * can be: PTR_TO_MAP_VALUE, PTR_TO_CTX, FRAME_PTR. These are three pointer
70  * types recognized by check_mem_access() function.
71  *
72  * PTR_TO_MAP_VALUE means that this register is pointing to 'map element value'
73  * and the range of [ptr, ptr + map's value_size) is accessible.
74  *
75  * registers used to pass values to function calls are checked against
76  * function argument constraints.
77  *
78  * ARG_PTR_TO_MAP_KEY is one of such argument constraints.
79  * It means that the register type passed to this function must be
80  * PTR_TO_STACK and it will be used inside the function as
81  * 'pointer to map element key'
82  *
83  * For example the argument constraints for bpf_map_lookup_elem():
84  *   .ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL,
85  *   .arg1_type = ARG_CONST_MAP_PTR,
86  *   .arg2_type = ARG_PTR_TO_MAP_KEY,
87  *
88  * ret_type says that this function returns 'pointer to map elem value or null'
89  * function expects 1st argument to be a const pointer to 'struct bpf_map' and
90  * 2nd argument should be a pointer to stack, which will be used inside
91  * the helper function as a pointer to map element key.
92  *
93  * On the kernel side the helper function looks like:
94  * u64 bpf_map_lookup_elem(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
95  * {
96  *    struct bpf_map *map = (struct bpf_map *) (unsigned long) r1;
97  *    void *key = (void *) (unsigned long) r2;
98  *    void *value;
99  *
100  *    here kernel can access 'key' and 'map' pointers safely, knowing that
101  *    [key, key + map->key_size) bytes are valid and were initialized on
102  *    the stack of eBPF program.
103  * }
104  *
105  * Corresponding eBPF program may look like:
106  *    BPF_MOV64_REG(BPF_REG_2, BPF_REG_10),  // after this insn R2 type is FRAME_PTR
107  *    BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -4), // after this insn R2 type is PTR_TO_STACK
108  *    BPF_LD_MAP_FD(BPF_REG_1, map_fd),      // after this insn R1 type is CONST_PTR_TO_MAP
109  *    BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem),
110  * here verifier looks at prototype of map_lookup_elem() and sees:
111  * .arg1_type == ARG_CONST_MAP_PTR and R1->type == CONST_PTR_TO_MAP, which is ok,
112  * Now verifier knows that this map has key of R1->map_ptr->key_size bytes
113  *
114  * Then .arg2_type == ARG_PTR_TO_MAP_KEY and R2->type == PTR_TO_STACK, ok so far,
115  * Now verifier checks that [R2, R2 + map's key_size) are within stack limits
116  * and were initialized prior to this call.
117  * If it's ok, then verifier allows this BPF_CALL insn and looks at
118  * .ret_type which is RET_PTR_TO_MAP_VALUE_OR_NULL, so it sets
119  * R0->type = PTR_TO_MAP_VALUE_OR_NULL which means bpf_map_lookup_elem() function
120  * returns ether pointer to map value or NULL.
121  *
122  * When type PTR_TO_MAP_VALUE_OR_NULL passes through 'if (reg != 0) goto +off'
123  * insn, the register holding that pointer in the true branch changes state to
124  * PTR_TO_MAP_VALUE and the same register changes state to CONST_IMM in the false
125  * branch. See check_cond_jmp_op().
126  *
127  * After the call R0 is set to return type of the function and registers R1-R5
128  * are set to NOT_INIT to indicate that they are no longer readable.
129  */
130 
131 /* verifier_state + insn_idx are pushed to stack when branch is encountered */
132 struct bpf_verifier_stack_elem {
133 	/* verifer state is 'st'
134 	 * before processing instruction 'insn_idx'
135 	 * and after processing instruction 'prev_insn_idx'
136 	 */
137 	struct bpf_verifier_state st;
138 	int insn_idx;
139 	int prev_insn_idx;
140 	struct bpf_verifier_stack_elem *next;
141 };
142 
143 #define BPF_COMPLEXITY_LIMIT_INSNS	98304
144 #define BPF_COMPLEXITY_LIMIT_STACK	1024
145 
146 #define BPF_MAP_PTR_POISON ((void *)0xeB9F + POISON_POINTER_DELTA)
147 
148 struct bpf_call_arg_meta {
149 	struct bpf_map *map_ptr;
150 	bool raw_mode;
151 	bool pkt_access;
152 	int regno;
153 	int access_size;
154 };
155 
156 /* verbose verifier prints what it's seeing
157  * bpf_check() is called under lock, so no race to access these global vars
158  */
159 static u32 log_level, log_size, log_len;
160 static char *log_buf;
161 
162 static DEFINE_MUTEX(bpf_verifier_lock);
163 
164 /* log_level controls verbosity level of eBPF verifier.
165  * verbose() is used to dump the verification trace to the log, so the user
166  * can figure out what's wrong with the program
167  */
168 static __printf(1, 2) void verbose(const char *fmt, ...)
169 {
170 	va_list args;
171 
172 	if (log_level == 0 || log_len >= log_size - 1)
173 		return;
174 
175 	va_start(args, fmt);
176 	log_len += vscnprintf(log_buf + log_len, log_size - log_len, fmt, args);
177 	va_end(args);
178 }
179 
180 /* string representation of 'enum bpf_reg_type' */
181 static const char * const reg_type_str[] = {
182 	[NOT_INIT]		= "?",
183 	[UNKNOWN_VALUE]		= "inv",
184 	[PTR_TO_CTX]		= "ctx",
185 	[CONST_PTR_TO_MAP]	= "map_ptr",
186 	[PTR_TO_MAP_VALUE]	= "map_value",
187 	[PTR_TO_MAP_VALUE_OR_NULL] = "map_value_or_null",
188 	[PTR_TO_MAP_VALUE_ADJ]	= "map_value_adj",
189 	[FRAME_PTR]		= "fp",
190 	[PTR_TO_STACK]		= "fp",
191 	[CONST_IMM]		= "imm",
192 	[PTR_TO_PACKET]		= "pkt",
193 	[PTR_TO_PACKET_END]	= "pkt_end",
194 };
195 
196 #define __BPF_FUNC_STR_FN(x) [BPF_FUNC_ ## x] = __stringify(bpf_ ## x)
197 static const char * const func_id_str[] = {
198 	__BPF_FUNC_MAPPER(__BPF_FUNC_STR_FN)
199 };
200 #undef __BPF_FUNC_STR_FN
201 
202 static const char *func_id_name(int id)
203 {
204 	BUILD_BUG_ON(ARRAY_SIZE(func_id_str) != __BPF_FUNC_MAX_ID);
205 
206 	if (id >= 0 && id < __BPF_FUNC_MAX_ID && func_id_str[id])
207 		return func_id_str[id];
208 	else
209 		return "unknown";
210 }
211 
212 static void print_verifier_state(struct bpf_verifier_state *state)
213 {
214 	struct bpf_reg_state *reg;
215 	enum bpf_reg_type t;
216 	int i;
217 
218 	for (i = 0; i < MAX_BPF_REG; i++) {
219 		reg = &state->regs[i];
220 		t = reg->type;
221 		if (t == NOT_INIT)
222 			continue;
223 		verbose(" R%d=%s", i, reg_type_str[t]);
224 		if (t == CONST_IMM || t == PTR_TO_STACK)
225 			verbose("%lld", reg->imm);
226 		else if (t == PTR_TO_PACKET)
227 			verbose("(id=%d,off=%d,r=%d)",
228 				reg->id, reg->off, reg->range);
229 		else if (t == UNKNOWN_VALUE && reg->imm)
230 			verbose("%lld", reg->imm);
231 		else if (t == CONST_PTR_TO_MAP || t == PTR_TO_MAP_VALUE ||
232 			 t == PTR_TO_MAP_VALUE_OR_NULL ||
233 			 t == PTR_TO_MAP_VALUE_ADJ)
234 			verbose("(ks=%d,vs=%d,id=%u)",
235 				reg->map_ptr->key_size,
236 				reg->map_ptr->value_size,
237 				reg->id);
238 		if (reg->min_value != BPF_REGISTER_MIN_RANGE)
239 			verbose(",min_value=%lld",
240 				(long long)reg->min_value);
241 		if (reg->max_value != BPF_REGISTER_MAX_RANGE)
242 			verbose(",max_value=%llu",
243 				(unsigned long long)reg->max_value);
244 		if (reg->min_align)
245 			verbose(",min_align=%u", reg->min_align);
246 		if (reg->aux_off)
247 			verbose(",aux_off=%u", reg->aux_off);
248 		if (reg->aux_off_align)
249 			verbose(",aux_off_align=%u", reg->aux_off_align);
250 	}
251 	for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
252 		if (state->stack_slot_type[i] == STACK_SPILL)
253 			verbose(" fp%d=%s", -MAX_BPF_STACK + i,
254 				reg_type_str[state->spilled_regs[i / BPF_REG_SIZE].type]);
255 	}
256 	verbose("\n");
257 }
258 
259 static const char *const bpf_class_string[] = {
260 	[BPF_LD]    = "ld",
261 	[BPF_LDX]   = "ldx",
262 	[BPF_ST]    = "st",
263 	[BPF_STX]   = "stx",
264 	[BPF_ALU]   = "alu",
265 	[BPF_JMP]   = "jmp",
266 	[BPF_RET]   = "BUG",
267 	[BPF_ALU64] = "alu64",
268 };
269 
270 static const char *const bpf_alu_string[16] = {
271 	[BPF_ADD >> 4]  = "+=",
272 	[BPF_SUB >> 4]  = "-=",
273 	[BPF_MUL >> 4]  = "*=",
274 	[BPF_DIV >> 4]  = "/=",
275 	[BPF_OR  >> 4]  = "|=",
276 	[BPF_AND >> 4]  = "&=",
277 	[BPF_LSH >> 4]  = "<<=",
278 	[BPF_RSH >> 4]  = ">>=",
279 	[BPF_NEG >> 4]  = "neg",
280 	[BPF_MOD >> 4]  = "%=",
281 	[BPF_XOR >> 4]  = "^=",
282 	[BPF_MOV >> 4]  = "=",
283 	[BPF_ARSH >> 4] = "s>>=",
284 	[BPF_END >> 4]  = "endian",
285 };
286 
287 static const char *const bpf_ldst_string[] = {
288 	[BPF_W >> 3]  = "u32",
289 	[BPF_H >> 3]  = "u16",
290 	[BPF_B >> 3]  = "u8",
291 	[BPF_DW >> 3] = "u64",
292 };
293 
294 static const char *const bpf_jmp_string[16] = {
295 	[BPF_JA >> 4]   = "jmp",
296 	[BPF_JEQ >> 4]  = "==",
297 	[BPF_JGT >> 4]  = ">",
298 	[BPF_JGE >> 4]  = ">=",
299 	[BPF_JSET >> 4] = "&",
300 	[BPF_JNE >> 4]  = "!=",
301 	[BPF_JSGT >> 4] = "s>",
302 	[BPF_JSGE >> 4] = "s>=",
303 	[BPF_CALL >> 4] = "call",
304 	[BPF_EXIT >> 4] = "exit",
305 };
306 
307 static void print_bpf_insn(const struct bpf_verifier_env *env,
308 			   const struct bpf_insn *insn)
309 {
310 	u8 class = BPF_CLASS(insn->code);
311 
312 	if (class == BPF_ALU || class == BPF_ALU64) {
313 		if (BPF_SRC(insn->code) == BPF_X)
314 			verbose("(%02x) %sr%d %s %sr%d\n",
315 				insn->code, class == BPF_ALU ? "(u32) " : "",
316 				insn->dst_reg,
317 				bpf_alu_string[BPF_OP(insn->code) >> 4],
318 				class == BPF_ALU ? "(u32) " : "",
319 				insn->src_reg);
320 		else
321 			verbose("(%02x) %sr%d %s %s%d\n",
322 				insn->code, class == BPF_ALU ? "(u32) " : "",
323 				insn->dst_reg,
324 				bpf_alu_string[BPF_OP(insn->code) >> 4],
325 				class == BPF_ALU ? "(u32) " : "",
326 				insn->imm);
327 	} else if (class == BPF_STX) {
328 		if (BPF_MODE(insn->code) == BPF_MEM)
329 			verbose("(%02x) *(%s *)(r%d %+d) = r%d\n",
330 				insn->code,
331 				bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
332 				insn->dst_reg,
333 				insn->off, insn->src_reg);
334 		else if (BPF_MODE(insn->code) == BPF_XADD)
335 			verbose("(%02x) lock *(%s *)(r%d %+d) += r%d\n",
336 				insn->code,
337 				bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
338 				insn->dst_reg, insn->off,
339 				insn->src_reg);
340 		else
341 			verbose("BUG_%02x\n", insn->code);
342 	} else if (class == BPF_ST) {
343 		if (BPF_MODE(insn->code) != BPF_MEM) {
344 			verbose("BUG_st_%02x\n", insn->code);
345 			return;
346 		}
347 		verbose("(%02x) *(%s *)(r%d %+d) = %d\n",
348 			insn->code,
349 			bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
350 			insn->dst_reg,
351 			insn->off, insn->imm);
352 	} else if (class == BPF_LDX) {
353 		if (BPF_MODE(insn->code) != BPF_MEM) {
354 			verbose("BUG_ldx_%02x\n", insn->code);
355 			return;
356 		}
357 		verbose("(%02x) r%d = *(%s *)(r%d %+d)\n",
358 			insn->code, insn->dst_reg,
359 			bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
360 			insn->src_reg, insn->off);
361 	} else if (class == BPF_LD) {
362 		if (BPF_MODE(insn->code) == BPF_ABS) {
363 			verbose("(%02x) r0 = *(%s *)skb[%d]\n",
364 				insn->code,
365 				bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
366 				insn->imm);
367 		} else if (BPF_MODE(insn->code) == BPF_IND) {
368 			verbose("(%02x) r0 = *(%s *)skb[r%d + %d]\n",
369 				insn->code,
370 				bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
371 				insn->src_reg, insn->imm);
372 		} else if (BPF_MODE(insn->code) == BPF_IMM &&
373 			   BPF_SIZE(insn->code) == BPF_DW) {
374 			/* At this point, we already made sure that the second
375 			 * part of the ldimm64 insn is accessible.
376 			 */
377 			u64 imm = ((u64)(insn + 1)->imm << 32) | (u32)insn->imm;
378 			bool map_ptr = insn->src_reg == BPF_PSEUDO_MAP_FD;
379 
380 			if (map_ptr && !env->allow_ptr_leaks)
381 				imm = 0;
382 
383 			verbose("(%02x) r%d = 0x%llx\n", insn->code,
384 				insn->dst_reg, (unsigned long long)imm);
385 		} else {
386 			verbose("BUG_ld_%02x\n", insn->code);
387 			return;
388 		}
389 	} else if (class == BPF_JMP) {
390 		u8 opcode = BPF_OP(insn->code);
391 
392 		if (opcode == BPF_CALL) {
393 			verbose("(%02x) call %s#%d\n", insn->code,
394 				func_id_name(insn->imm), insn->imm);
395 		} else if (insn->code == (BPF_JMP | BPF_JA)) {
396 			verbose("(%02x) goto pc%+d\n",
397 				insn->code, insn->off);
398 		} else if (insn->code == (BPF_JMP | BPF_EXIT)) {
399 			verbose("(%02x) exit\n", insn->code);
400 		} else if (BPF_SRC(insn->code) == BPF_X) {
401 			verbose("(%02x) if r%d %s r%d goto pc%+d\n",
402 				insn->code, insn->dst_reg,
403 				bpf_jmp_string[BPF_OP(insn->code) >> 4],
404 				insn->src_reg, insn->off);
405 		} else {
406 			verbose("(%02x) if r%d %s 0x%x goto pc%+d\n",
407 				insn->code, insn->dst_reg,
408 				bpf_jmp_string[BPF_OP(insn->code) >> 4],
409 				insn->imm, insn->off);
410 		}
411 	} else {
412 		verbose("(%02x) %s\n", insn->code, bpf_class_string[class]);
413 	}
414 }
415 
416 static int pop_stack(struct bpf_verifier_env *env, int *prev_insn_idx)
417 {
418 	struct bpf_verifier_stack_elem *elem;
419 	int insn_idx;
420 
421 	if (env->head == NULL)
422 		return -1;
423 
424 	memcpy(&env->cur_state, &env->head->st, sizeof(env->cur_state));
425 	insn_idx = env->head->insn_idx;
426 	if (prev_insn_idx)
427 		*prev_insn_idx = env->head->prev_insn_idx;
428 	elem = env->head->next;
429 	kfree(env->head);
430 	env->head = elem;
431 	env->stack_size--;
432 	return insn_idx;
433 }
434 
435 static struct bpf_verifier_state *push_stack(struct bpf_verifier_env *env,
436 					     int insn_idx, int prev_insn_idx)
437 {
438 	struct bpf_verifier_stack_elem *elem;
439 
440 	elem = kmalloc(sizeof(struct bpf_verifier_stack_elem), GFP_KERNEL);
441 	if (!elem)
442 		goto err;
443 
444 	memcpy(&elem->st, &env->cur_state, sizeof(env->cur_state));
445 	elem->insn_idx = insn_idx;
446 	elem->prev_insn_idx = prev_insn_idx;
447 	elem->next = env->head;
448 	env->head = elem;
449 	env->stack_size++;
450 	if (env->stack_size > BPF_COMPLEXITY_LIMIT_STACK) {
451 		verbose("BPF program is too complex\n");
452 		goto err;
453 	}
454 	return &elem->st;
455 err:
456 	/* pop all elements and return */
457 	while (pop_stack(env, NULL) >= 0);
458 	return NULL;
459 }
460 
461 #define CALLER_SAVED_REGS 6
462 static const int caller_saved[CALLER_SAVED_REGS] = {
463 	BPF_REG_0, BPF_REG_1, BPF_REG_2, BPF_REG_3, BPF_REG_4, BPF_REG_5
464 };
465 
466 static void mark_reg_not_init(struct bpf_reg_state *regs, u32 regno)
467 {
468 	BUG_ON(regno >= MAX_BPF_REG);
469 
470 	memset(&regs[regno], 0, sizeof(regs[regno]));
471 	regs[regno].type = NOT_INIT;
472 	regs[regno].min_value = BPF_REGISTER_MIN_RANGE;
473 	regs[regno].max_value = BPF_REGISTER_MAX_RANGE;
474 }
475 
476 static void init_reg_state(struct bpf_reg_state *regs)
477 {
478 	int i;
479 
480 	for (i = 0; i < MAX_BPF_REG; i++)
481 		mark_reg_not_init(regs, i);
482 
483 	/* frame pointer */
484 	regs[BPF_REG_FP].type = FRAME_PTR;
485 
486 	/* 1st arg to a function */
487 	regs[BPF_REG_1].type = PTR_TO_CTX;
488 }
489 
490 static void __mark_reg_unknown_value(struct bpf_reg_state *regs, u32 regno)
491 {
492 	regs[regno].type = UNKNOWN_VALUE;
493 	regs[regno].id = 0;
494 	regs[regno].imm = 0;
495 }
496 
497 static void mark_reg_unknown_value(struct bpf_reg_state *regs, u32 regno)
498 {
499 	BUG_ON(regno >= MAX_BPF_REG);
500 	__mark_reg_unknown_value(regs, regno);
501 }
502 
503 static void reset_reg_range_values(struct bpf_reg_state *regs, u32 regno)
504 {
505 	regs[regno].min_value = BPF_REGISTER_MIN_RANGE;
506 	regs[regno].max_value = BPF_REGISTER_MAX_RANGE;
507 	regs[regno].value_from_signed = false;
508 	regs[regno].min_align = 0;
509 }
510 
511 static void mark_reg_unknown_value_and_range(struct bpf_reg_state *regs,
512 					     u32 regno)
513 {
514 	mark_reg_unknown_value(regs, regno);
515 	reset_reg_range_values(regs, regno);
516 }
517 
518 enum reg_arg_type {
519 	SRC_OP,		/* register is used as source operand */
520 	DST_OP,		/* register is used as destination operand */
521 	DST_OP_NO_MARK	/* same as above, check only, don't mark */
522 };
523 
524 static int check_reg_arg(struct bpf_reg_state *regs, u32 regno,
525 			 enum reg_arg_type t)
526 {
527 	if (regno >= MAX_BPF_REG) {
528 		verbose("R%d is invalid\n", regno);
529 		return -EINVAL;
530 	}
531 
532 	if (t == SRC_OP) {
533 		/* check whether register used as source operand can be read */
534 		if (regs[regno].type == NOT_INIT) {
535 			verbose("R%d !read_ok\n", regno);
536 			return -EACCES;
537 		}
538 	} else {
539 		/* check whether register used as dest operand can be written to */
540 		if (regno == BPF_REG_FP) {
541 			verbose("frame pointer is read only\n");
542 			return -EACCES;
543 		}
544 		if (t == DST_OP)
545 			mark_reg_unknown_value(regs, regno);
546 	}
547 	return 0;
548 }
549 
550 static bool is_spillable_regtype(enum bpf_reg_type type)
551 {
552 	switch (type) {
553 	case PTR_TO_MAP_VALUE:
554 	case PTR_TO_MAP_VALUE_OR_NULL:
555 	case PTR_TO_MAP_VALUE_ADJ:
556 	case PTR_TO_STACK:
557 	case PTR_TO_CTX:
558 	case PTR_TO_PACKET:
559 	case PTR_TO_PACKET_END:
560 	case FRAME_PTR:
561 	case CONST_PTR_TO_MAP:
562 		return true;
563 	default:
564 		return false;
565 	}
566 }
567 
568 /* check_stack_read/write functions track spill/fill of registers,
569  * stack boundary and alignment are checked in check_mem_access()
570  */
571 static int check_stack_write(struct bpf_verifier_state *state, int off,
572 			     int size, int value_regno)
573 {
574 	int i;
575 	/* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0,
576 	 * so it's aligned access and [off, off + size) are within stack limits
577 	 */
578 
579 	if (value_regno >= 0 &&
580 	    is_spillable_regtype(state->regs[value_regno].type)) {
581 
582 		/* register containing pointer is being spilled into stack */
583 		if (size != BPF_REG_SIZE) {
584 			verbose("invalid size of register spill\n");
585 			return -EACCES;
586 		}
587 
588 		/* save register state */
589 		state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] =
590 			state->regs[value_regno];
591 
592 		for (i = 0; i < BPF_REG_SIZE; i++)
593 			state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_SPILL;
594 	} else {
595 		/* regular write of data into stack */
596 		state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] =
597 			(struct bpf_reg_state) {};
598 
599 		for (i = 0; i < size; i++)
600 			state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_MISC;
601 	}
602 	return 0;
603 }
604 
605 static int check_stack_read(struct bpf_verifier_state *state, int off, int size,
606 			    int value_regno)
607 {
608 	u8 *slot_type;
609 	int i;
610 
611 	slot_type = &state->stack_slot_type[MAX_BPF_STACK + off];
612 
613 	if (slot_type[0] == STACK_SPILL) {
614 		if (size != BPF_REG_SIZE) {
615 			verbose("invalid size of register spill\n");
616 			return -EACCES;
617 		}
618 		for (i = 1; i < BPF_REG_SIZE; i++) {
619 			if (slot_type[i] != STACK_SPILL) {
620 				verbose("corrupted spill memory\n");
621 				return -EACCES;
622 			}
623 		}
624 
625 		if (value_regno >= 0)
626 			/* restore register state from stack */
627 			state->regs[value_regno] =
628 				state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE];
629 		return 0;
630 	} else {
631 		for (i = 0; i < size; i++) {
632 			if (slot_type[i] != STACK_MISC) {
633 				verbose("invalid read from stack off %d+%d size %d\n",
634 					off, i, size);
635 				return -EACCES;
636 			}
637 		}
638 		if (value_regno >= 0)
639 			/* have read misc data from the stack */
640 			mark_reg_unknown_value_and_range(state->regs,
641 							 value_regno);
642 		return 0;
643 	}
644 }
645 
646 /* check read/write into map element returned by bpf_map_lookup_elem() */
647 static int check_map_access(struct bpf_verifier_env *env, u32 regno, int off,
648 			    int size)
649 {
650 	struct bpf_map *map = env->cur_state.regs[regno].map_ptr;
651 
652 	if (off < 0 || size <= 0 || off + size > map->value_size) {
653 		verbose("invalid access to map value, value_size=%d off=%d size=%d\n",
654 			map->value_size, off, size);
655 		return -EACCES;
656 	}
657 	return 0;
658 }
659 
660 /* check read/write into an adjusted map element */
661 static int check_map_access_adj(struct bpf_verifier_env *env, u32 regno,
662 				int off, int size)
663 {
664 	struct bpf_verifier_state *state = &env->cur_state;
665 	struct bpf_reg_state *reg = &state->regs[regno];
666 	int err;
667 
668 	/* We adjusted the register to this map value, so we
669 	 * need to change off and size to min_value and max_value
670 	 * respectively to make sure our theoretical access will be
671 	 * safe.
672 	 */
673 	if (log_level)
674 		print_verifier_state(state);
675 	env->varlen_map_value_access = true;
676 	/* The minimum value is only important with signed
677 	 * comparisons where we can't assume the floor of a
678 	 * value is 0.  If we are using signed variables for our
679 	 * index'es we need to make sure that whatever we use
680 	 * will have a set floor within our range.
681 	 */
682 	if (reg->min_value < 0) {
683 		verbose("R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n",
684 			regno);
685 		return -EACCES;
686 	}
687 	err = check_map_access(env, regno, reg->min_value + off, size);
688 	if (err) {
689 		verbose("R%d min value is outside of the array range\n",
690 			regno);
691 		return err;
692 	}
693 
694 	/* If we haven't set a max value then we need to bail
695 	 * since we can't be sure we won't do bad things.
696 	 */
697 	if (reg->max_value == BPF_REGISTER_MAX_RANGE) {
698 		verbose("R%d unbounded memory access, make sure to bounds check any array access into a map\n",
699 			regno);
700 		return -EACCES;
701 	}
702 	return check_map_access(env, regno, reg->max_value + off, size);
703 }
704 
705 #define MAX_PACKET_OFF 0xffff
706 
707 static bool may_access_direct_pkt_data(struct bpf_verifier_env *env,
708 				       const struct bpf_call_arg_meta *meta,
709 				       enum bpf_access_type t)
710 {
711 	switch (env->prog->type) {
712 	case BPF_PROG_TYPE_LWT_IN:
713 	case BPF_PROG_TYPE_LWT_OUT:
714 		/* dst_input() and dst_output() can't write for now */
715 		if (t == BPF_WRITE)
716 			return false;
717 		/* fallthrough */
718 	case BPF_PROG_TYPE_SCHED_CLS:
719 	case BPF_PROG_TYPE_SCHED_ACT:
720 	case BPF_PROG_TYPE_XDP:
721 	case BPF_PROG_TYPE_LWT_XMIT:
722 		if (meta)
723 			return meta->pkt_access;
724 
725 		env->seen_direct_write = true;
726 		return true;
727 	default:
728 		return false;
729 	}
730 }
731 
732 static int check_packet_access(struct bpf_verifier_env *env, u32 regno, int off,
733 			       int size)
734 {
735 	struct bpf_reg_state *regs = env->cur_state.regs;
736 	struct bpf_reg_state *reg = &regs[regno];
737 
738 	off += reg->off;
739 	if (off < 0 || size <= 0 || off + size > reg->range) {
740 		verbose("invalid access to packet, off=%d size=%d, R%d(id=%d,off=%d,r=%d)\n",
741 			off, size, regno, reg->id, reg->off, reg->range);
742 		return -EACCES;
743 	}
744 	return 0;
745 }
746 
747 /* check access to 'struct bpf_context' fields */
748 static int check_ctx_access(struct bpf_verifier_env *env, int insn_idx, int off, int size,
749 			    enum bpf_access_type t, enum bpf_reg_type *reg_type)
750 {
751 	struct bpf_insn_access_aux info = {
752 		.reg_type = *reg_type,
753 	};
754 
755 	/* for analyzer ctx accesses are already validated and converted */
756 	if (env->analyzer_ops)
757 		return 0;
758 
759 	if (env->prog->aux->ops->is_valid_access &&
760 	    env->prog->aux->ops->is_valid_access(off, size, t, &info)) {
761 		/* A non zero info.ctx_field_size indicates that this field is a
762 		 * candidate for later verifier transformation to load the whole
763 		 * field and then apply a mask when accessed with a narrower
764 		 * access than actual ctx access size. A zero info.ctx_field_size
765 		 * will only allow for whole field access and rejects any other
766 		 * type of narrower access.
767 		 */
768 		env->insn_aux_data[insn_idx].ctx_field_size = info.ctx_field_size;
769 		*reg_type = info.reg_type;
770 
771 		/* remember the offset of last byte accessed in ctx */
772 		if (env->prog->aux->max_ctx_offset < off + size)
773 			env->prog->aux->max_ctx_offset = off + size;
774 		return 0;
775 	}
776 
777 	verbose("invalid bpf_context access off=%d size=%d\n", off, size);
778 	return -EACCES;
779 }
780 
781 static bool __is_pointer_value(bool allow_ptr_leaks,
782 			       const struct bpf_reg_state *reg)
783 {
784 	if (allow_ptr_leaks)
785 		return false;
786 
787 	switch (reg->type) {
788 	case UNKNOWN_VALUE:
789 	case CONST_IMM:
790 		return false;
791 	default:
792 		return true;
793 	}
794 }
795 
796 static bool is_pointer_value(struct bpf_verifier_env *env, int regno)
797 {
798 	return __is_pointer_value(env->allow_ptr_leaks, &env->cur_state.regs[regno]);
799 }
800 
801 static int check_pkt_ptr_alignment(const struct bpf_reg_state *reg,
802 				   int off, int size, bool strict)
803 {
804 	int ip_align;
805 	int reg_off;
806 
807 	/* Byte size accesses are always allowed. */
808 	if (!strict || size == 1)
809 		return 0;
810 
811 	reg_off = reg->off;
812 	if (reg->id) {
813 		if (reg->aux_off_align % size) {
814 			verbose("Packet access is only %u byte aligned, %d byte access not allowed\n",
815 				reg->aux_off_align, size);
816 			return -EACCES;
817 		}
818 		reg_off += reg->aux_off;
819 	}
820 
821 	/* For platforms that do not have a Kconfig enabling
822 	 * CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS the value of
823 	 * NET_IP_ALIGN is universally set to '2'.  And on platforms
824 	 * that do set CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS, we get
825 	 * to this code only in strict mode where we want to emulate
826 	 * the NET_IP_ALIGN==2 checking.  Therefore use an
827 	 * unconditional IP align value of '2'.
828 	 */
829 	ip_align = 2;
830 	if ((ip_align + reg_off + off) % size != 0) {
831 		verbose("misaligned packet access off %d+%d+%d size %d\n",
832 			ip_align, reg_off, off, size);
833 		return -EACCES;
834 	}
835 
836 	return 0;
837 }
838 
839 static int check_val_ptr_alignment(const struct bpf_reg_state *reg,
840 				   int size, bool strict)
841 {
842 	if (strict && size != 1) {
843 		verbose("Unknown alignment. Only byte-sized access allowed in value access.\n");
844 		return -EACCES;
845 	}
846 
847 	return 0;
848 }
849 
850 static int check_ptr_alignment(struct bpf_verifier_env *env,
851 			       const struct bpf_reg_state *reg,
852 			       int off, int size)
853 {
854 	bool strict = env->strict_alignment;
855 
856 	switch (reg->type) {
857 	case PTR_TO_PACKET:
858 		return check_pkt_ptr_alignment(reg, off, size, strict);
859 	case PTR_TO_MAP_VALUE_ADJ:
860 		return check_val_ptr_alignment(reg, size, strict);
861 	default:
862 		if (off % size != 0) {
863 			verbose("misaligned access off %d size %d\n",
864 				off, size);
865 			return -EACCES;
866 		}
867 
868 		return 0;
869 	}
870 }
871 
872 /* check whether memory at (regno + off) is accessible for t = (read | write)
873  * if t==write, value_regno is a register which value is stored into memory
874  * if t==read, value_regno is a register which will receive the value from memory
875  * if t==write && value_regno==-1, some unknown value is stored into memory
876  * if t==read && value_regno==-1, don't care what we read from memory
877  */
878 static int check_mem_access(struct bpf_verifier_env *env, int insn_idx, u32 regno, int off,
879 			    int bpf_size, enum bpf_access_type t,
880 			    int value_regno)
881 {
882 	struct bpf_verifier_state *state = &env->cur_state;
883 	struct bpf_reg_state *reg = &state->regs[regno];
884 	int size, err = 0;
885 
886 	if (reg->type == PTR_TO_STACK)
887 		off += reg->imm;
888 
889 	size = bpf_size_to_bytes(bpf_size);
890 	if (size < 0)
891 		return size;
892 
893 	err = check_ptr_alignment(env, reg, off, size);
894 	if (err)
895 		return err;
896 
897 	if (reg->type == PTR_TO_MAP_VALUE ||
898 	    reg->type == PTR_TO_MAP_VALUE_ADJ) {
899 		if (t == BPF_WRITE && value_regno >= 0 &&
900 		    is_pointer_value(env, value_regno)) {
901 			verbose("R%d leaks addr into map\n", value_regno);
902 			return -EACCES;
903 		}
904 
905 		if (reg->type == PTR_TO_MAP_VALUE_ADJ)
906 			err = check_map_access_adj(env, regno, off, size);
907 		else
908 			err = check_map_access(env, regno, off, size);
909 		if (!err && t == BPF_READ && value_regno >= 0)
910 			mark_reg_unknown_value_and_range(state->regs,
911 							 value_regno);
912 
913 	} else if (reg->type == PTR_TO_CTX) {
914 		enum bpf_reg_type reg_type = UNKNOWN_VALUE;
915 
916 		if (t == BPF_WRITE && value_regno >= 0 &&
917 		    is_pointer_value(env, value_regno)) {
918 			verbose("R%d leaks addr into ctx\n", value_regno);
919 			return -EACCES;
920 		}
921 		err = check_ctx_access(env, insn_idx, off, size, t, &reg_type);
922 		if (!err && t == BPF_READ && value_regno >= 0) {
923 			mark_reg_unknown_value_and_range(state->regs,
924 							 value_regno);
925 			/* note that reg.[id|off|range] == 0 */
926 			state->regs[value_regno].type = reg_type;
927 			state->regs[value_regno].aux_off = 0;
928 			state->regs[value_regno].aux_off_align = 0;
929 		}
930 
931 	} else if (reg->type == FRAME_PTR || reg->type == PTR_TO_STACK) {
932 		if (off >= 0 || off < -MAX_BPF_STACK) {
933 			verbose("invalid stack off=%d size=%d\n", off, size);
934 			return -EACCES;
935 		}
936 
937 		if (env->prog->aux->stack_depth < -off)
938 			env->prog->aux->stack_depth = -off;
939 
940 		if (t == BPF_WRITE) {
941 			if (!env->allow_ptr_leaks &&
942 			    state->stack_slot_type[MAX_BPF_STACK + off] == STACK_SPILL &&
943 			    size != BPF_REG_SIZE) {
944 				verbose("attempt to corrupt spilled pointer on stack\n");
945 				return -EACCES;
946 			}
947 			err = check_stack_write(state, off, size, value_regno);
948 		} else {
949 			err = check_stack_read(state, off, size, value_regno);
950 		}
951 	} else if (state->regs[regno].type == PTR_TO_PACKET) {
952 		if (t == BPF_WRITE && !may_access_direct_pkt_data(env, NULL, t)) {
953 			verbose("cannot write into packet\n");
954 			return -EACCES;
955 		}
956 		if (t == BPF_WRITE && value_regno >= 0 &&
957 		    is_pointer_value(env, value_regno)) {
958 			verbose("R%d leaks addr into packet\n", value_regno);
959 			return -EACCES;
960 		}
961 		err = check_packet_access(env, regno, off, size);
962 		if (!err && t == BPF_READ && value_regno >= 0)
963 			mark_reg_unknown_value_and_range(state->regs,
964 							 value_regno);
965 	} else {
966 		verbose("R%d invalid mem access '%s'\n",
967 			regno, reg_type_str[reg->type]);
968 		return -EACCES;
969 	}
970 
971 	if (!err && size <= 2 && value_regno >= 0 && env->allow_ptr_leaks &&
972 	    state->regs[value_regno].type == UNKNOWN_VALUE) {
973 		/* 1 or 2 byte load zero-extends, determine the number of
974 		 * zero upper bits. Not doing it fo 4 byte load, since
975 		 * such values cannot be added to ptr_to_packet anyway.
976 		 */
977 		state->regs[value_regno].imm = 64 - size * 8;
978 	}
979 	return err;
980 }
981 
982 static int check_xadd(struct bpf_verifier_env *env, int insn_idx, struct bpf_insn *insn)
983 {
984 	struct bpf_reg_state *regs = env->cur_state.regs;
985 	int err;
986 
987 	if ((BPF_SIZE(insn->code) != BPF_W && BPF_SIZE(insn->code) != BPF_DW) ||
988 	    insn->imm != 0) {
989 		verbose("BPF_XADD uses reserved fields\n");
990 		return -EINVAL;
991 	}
992 
993 	/* check src1 operand */
994 	err = check_reg_arg(regs, insn->src_reg, SRC_OP);
995 	if (err)
996 		return err;
997 
998 	/* check src2 operand */
999 	err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1000 	if (err)
1001 		return err;
1002 
1003 	if (is_pointer_value(env, insn->src_reg)) {
1004 		verbose("R%d leaks addr into mem\n", insn->src_reg);
1005 		return -EACCES;
1006 	}
1007 
1008 	/* check whether atomic_add can read the memory */
1009 	err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off,
1010 			       BPF_SIZE(insn->code), BPF_READ, -1);
1011 	if (err)
1012 		return err;
1013 
1014 	/* check whether atomic_add can write into the same memory */
1015 	return check_mem_access(env, insn_idx, insn->dst_reg, insn->off,
1016 				BPF_SIZE(insn->code), BPF_WRITE, -1);
1017 }
1018 
1019 /* when register 'regno' is passed into function that will read 'access_size'
1020  * bytes from that pointer, make sure that it's within stack boundary
1021  * and all elements of stack are initialized
1022  */
1023 static int check_stack_boundary(struct bpf_verifier_env *env, int regno,
1024 				int access_size, bool zero_size_allowed,
1025 				struct bpf_call_arg_meta *meta)
1026 {
1027 	struct bpf_verifier_state *state = &env->cur_state;
1028 	struct bpf_reg_state *regs = state->regs;
1029 	int off, i;
1030 
1031 	if (regs[regno].type != PTR_TO_STACK) {
1032 		if (zero_size_allowed && access_size == 0 &&
1033 		    regs[regno].type == CONST_IMM &&
1034 		    regs[regno].imm  == 0)
1035 			return 0;
1036 
1037 		verbose("R%d type=%s expected=%s\n", regno,
1038 			reg_type_str[regs[regno].type],
1039 			reg_type_str[PTR_TO_STACK]);
1040 		return -EACCES;
1041 	}
1042 
1043 	off = regs[regno].imm;
1044 	if (off >= 0 || off < -MAX_BPF_STACK || off + access_size > 0 ||
1045 	    access_size <= 0) {
1046 		verbose("invalid stack type R%d off=%d access_size=%d\n",
1047 			regno, off, access_size);
1048 		return -EACCES;
1049 	}
1050 
1051 	if (env->prog->aux->stack_depth < -off)
1052 		env->prog->aux->stack_depth = -off;
1053 
1054 	if (meta && meta->raw_mode) {
1055 		meta->access_size = access_size;
1056 		meta->regno = regno;
1057 		return 0;
1058 	}
1059 
1060 	for (i = 0; i < access_size; i++) {
1061 		if (state->stack_slot_type[MAX_BPF_STACK + off + i] != STACK_MISC) {
1062 			verbose("invalid indirect read from stack off %d+%d size %d\n",
1063 				off, i, access_size);
1064 			return -EACCES;
1065 		}
1066 	}
1067 	return 0;
1068 }
1069 
1070 static int check_helper_mem_access(struct bpf_verifier_env *env, int regno,
1071 				   int access_size, bool zero_size_allowed,
1072 				   struct bpf_call_arg_meta *meta)
1073 {
1074 	struct bpf_reg_state *regs = env->cur_state.regs;
1075 
1076 	switch (regs[regno].type) {
1077 	case PTR_TO_PACKET:
1078 		return check_packet_access(env, regno, 0, access_size);
1079 	case PTR_TO_MAP_VALUE:
1080 		return check_map_access(env, regno, 0, access_size);
1081 	case PTR_TO_MAP_VALUE_ADJ:
1082 		return check_map_access_adj(env, regno, 0, access_size);
1083 	default: /* const_imm|ptr_to_stack or invalid ptr */
1084 		return check_stack_boundary(env, regno, access_size,
1085 					    zero_size_allowed, meta);
1086 	}
1087 }
1088 
1089 static int check_func_arg(struct bpf_verifier_env *env, u32 regno,
1090 			  enum bpf_arg_type arg_type,
1091 			  struct bpf_call_arg_meta *meta)
1092 {
1093 	struct bpf_reg_state *regs = env->cur_state.regs, *reg = &regs[regno];
1094 	enum bpf_reg_type expected_type, type = reg->type;
1095 	int err = 0;
1096 
1097 	if (arg_type == ARG_DONTCARE)
1098 		return 0;
1099 
1100 	if (type == NOT_INIT) {
1101 		verbose("R%d !read_ok\n", regno);
1102 		return -EACCES;
1103 	}
1104 
1105 	if (arg_type == ARG_ANYTHING) {
1106 		if (is_pointer_value(env, regno)) {
1107 			verbose("R%d leaks addr into helper function\n", regno);
1108 			return -EACCES;
1109 		}
1110 		return 0;
1111 	}
1112 
1113 	if (type == PTR_TO_PACKET &&
1114 	    !may_access_direct_pkt_data(env, meta, BPF_READ)) {
1115 		verbose("helper access to the packet is not allowed\n");
1116 		return -EACCES;
1117 	}
1118 
1119 	if (arg_type == ARG_PTR_TO_MAP_KEY ||
1120 	    arg_type == ARG_PTR_TO_MAP_VALUE) {
1121 		expected_type = PTR_TO_STACK;
1122 		if (type != PTR_TO_PACKET && type != expected_type)
1123 			goto err_type;
1124 	} else if (arg_type == ARG_CONST_SIZE ||
1125 		   arg_type == ARG_CONST_SIZE_OR_ZERO) {
1126 		expected_type = CONST_IMM;
1127 		/* One exception. Allow UNKNOWN_VALUE registers when the
1128 		 * boundaries are known and don't cause unsafe memory accesses
1129 		 */
1130 		if (type != UNKNOWN_VALUE && type != expected_type)
1131 			goto err_type;
1132 	} else if (arg_type == ARG_CONST_MAP_PTR) {
1133 		expected_type = CONST_PTR_TO_MAP;
1134 		if (type != expected_type)
1135 			goto err_type;
1136 	} else if (arg_type == ARG_PTR_TO_CTX) {
1137 		expected_type = PTR_TO_CTX;
1138 		if (type != expected_type)
1139 			goto err_type;
1140 	} else if (arg_type == ARG_PTR_TO_MEM ||
1141 		   arg_type == ARG_PTR_TO_UNINIT_MEM) {
1142 		expected_type = PTR_TO_STACK;
1143 		/* One exception here. In case function allows for NULL to be
1144 		 * passed in as argument, it's a CONST_IMM type. Final test
1145 		 * happens during stack boundary checking.
1146 		 */
1147 		if (type == CONST_IMM && reg->imm == 0)
1148 			/* final test in check_stack_boundary() */;
1149 		else if (type != PTR_TO_PACKET && type != PTR_TO_MAP_VALUE &&
1150 			 type != PTR_TO_MAP_VALUE_ADJ && type != expected_type)
1151 			goto err_type;
1152 		meta->raw_mode = arg_type == ARG_PTR_TO_UNINIT_MEM;
1153 	} else {
1154 		verbose("unsupported arg_type %d\n", arg_type);
1155 		return -EFAULT;
1156 	}
1157 
1158 	if (arg_type == ARG_CONST_MAP_PTR) {
1159 		/* bpf_map_xxx(map_ptr) call: remember that map_ptr */
1160 		meta->map_ptr = reg->map_ptr;
1161 	} else if (arg_type == ARG_PTR_TO_MAP_KEY) {
1162 		/* bpf_map_xxx(..., map_ptr, ..., key) call:
1163 		 * check that [key, key + map->key_size) are within
1164 		 * stack limits and initialized
1165 		 */
1166 		if (!meta->map_ptr) {
1167 			/* in function declaration map_ptr must come before
1168 			 * map_key, so that it's verified and known before
1169 			 * we have to check map_key here. Otherwise it means
1170 			 * that kernel subsystem misconfigured verifier
1171 			 */
1172 			verbose("invalid map_ptr to access map->key\n");
1173 			return -EACCES;
1174 		}
1175 		if (type == PTR_TO_PACKET)
1176 			err = check_packet_access(env, regno, 0,
1177 						  meta->map_ptr->key_size);
1178 		else
1179 			err = check_stack_boundary(env, regno,
1180 						   meta->map_ptr->key_size,
1181 						   false, NULL);
1182 	} else if (arg_type == ARG_PTR_TO_MAP_VALUE) {
1183 		/* bpf_map_xxx(..., map_ptr, ..., value) call:
1184 		 * check [value, value + map->value_size) validity
1185 		 */
1186 		if (!meta->map_ptr) {
1187 			/* kernel subsystem misconfigured verifier */
1188 			verbose("invalid map_ptr to access map->value\n");
1189 			return -EACCES;
1190 		}
1191 		if (type == PTR_TO_PACKET)
1192 			err = check_packet_access(env, regno, 0,
1193 						  meta->map_ptr->value_size);
1194 		else
1195 			err = check_stack_boundary(env, regno,
1196 						   meta->map_ptr->value_size,
1197 						   false, NULL);
1198 	} else if (arg_type == ARG_CONST_SIZE ||
1199 		   arg_type == ARG_CONST_SIZE_OR_ZERO) {
1200 		bool zero_size_allowed = (arg_type == ARG_CONST_SIZE_OR_ZERO);
1201 
1202 		/* bpf_xxx(..., buf, len) call will access 'len' bytes
1203 		 * from stack pointer 'buf'. Check it
1204 		 * note: regno == len, regno - 1 == buf
1205 		 */
1206 		if (regno == 0) {
1207 			/* kernel subsystem misconfigured verifier */
1208 			verbose("ARG_CONST_SIZE cannot be first argument\n");
1209 			return -EACCES;
1210 		}
1211 
1212 		/* If the register is UNKNOWN_VALUE, the access check happens
1213 		 * using its boundaries. Otherwise, just use its imm
1214 		 */
1215 		if (type == UNKNOWN_VALUE) {
1216 			/* For unprivileged variable accesses, disable raw
1217 			 * mode so that the program is required to
1218 			 * initialize all the memory that the helper could
1219 			 * just partially fill up.
1220 			 */
1221 			meta = NULL;
1222 
1223 			if (reg->min_value < 0) {
1224 				verbose("R%d min value is negative, either use unsigned or 'var &= const'\n",
1225 					regno);
1226 				return -EACCES;
1227 			}
1228 
1229 			if (reg->min_value == 0) {
1230 				err = check_helper_mem_access(env, regno - 1, 0,
1231 							      zero_size_allowed,
1232 							      meta);
1233 				if (err)
1234 					return err;
1235 			}
1236 
1237 			if (reg->max_value == BPF_REGISTER_MAX_RANGE) {
1238 				verbose("R%d unbounded memory access, use 'var &= const' or 'if (var < const)'\n",
1239 					regno);
1240 				return -EACCES;
1241 			}
1242 			err = check_helper_mem_access(env, regno - 1,
1243 						      reg->max_value,
1244 						      zero_size_allowed, meta);
1245 			if (err)
1246 				return err;
1247 		} else {
1248 			/* register is CONST_IMM */
1249 			err = check_helper_mem_access(env, regno - 1, reg->imm,
1250 						      zero_size_allowed, meta);
1251 		}
1252 	}
1253 
1254 	return err;
1255 err_type:
1256 	verbose("R%d type=%s expected=%s\n", regno,
1257 		reg_type_str[type], reg_type_str[expected_type]);
1258 	return -EACCES;
1259 }
1260 
1261 static int check_map_func_compatibility(struct bpf_map *map, int func_id)
1262 {
1263 	if (!map)
1264 		return 0;
1265 
1266 	/* We need a two way check, first is from map perspective ... */
1267 	switch (map->map_type) {
1268 	case BPF_MAP_TYPE_PROG_ARRAY:
1269 		if (func_id != BPF_FUNC_tail_call)
1270 			goto error;
1271 		break;
1272 	case BPF_MAP_TYPE_PERF_EVENT_ARRAY:
1273 		if (func_id != BPF_FUNC_perf_event_read &&
1274 		    func_id != BPF_FUNC_perf_event_output)
1275 			goto error;
1276 		break;
1277 	case BPF_MAP_TYPE_STACK_TRACE:
1278 		if (func_id != BPF_FUNC_get_stackid)
1279 			goto error;
1280 		break;
1281 	case BPF_MAP_TYPE_CGROUP_ARRAY:
1282 		if (func_id != BPF_FUNC_skb_under_cgroup &&
1283 		    func_id != BPF_FUNC_current_task_under_cgroup)
1284 			goto error;
1285 		break;
1286 	case BPF_MAP_TYPE_ARRAY_OF_MAPS:
1287 	case BPF_MAP_TYPE_HASH_OF_MAPS:
1288 		if (func_id != BPF_FUNC_map_lookup_elem)
1289 			goto error;
1290 	default:
1291 		break;
1292 	}
1293 
1294 	/* ... and second from the function itself. */
1295 	switch (func_id) {
1296 	case BPF_FUNC_tail_call:
1297 		if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY)
1298 			goto error;
1299 		break;
1300 	case BPF_FUNC_perf_event_read:
1301 	case BPF_FUNC_perf_event_output:
1302 		if (map->map_type != BPF_MAP_TYPE_PERF_EVENT_ARRAY)
1303 			goto error;
1304 		break;
1305 	case BPF_FUNC_get_stackid:
1306 		if (map->map_type != BPF_MAP_TYPE_STACK_TRACE)
1307 			goto error;
1308 		break;
1309 	case BPF_FUNC_current_task_under_cgroup:
1310 	case BPF_FUNC_skb_under_cgroup:
1311 		if (map->map_type != BPF_MAP_TYPE_CGROUP_ARRAY)
1312 			goto error;
1313 		break;
1314 	default:
1315 		break;
1316 	}
1317 
1318 	return 0;
1319 error:
1320 	verbose("cannot pass map_type %d into func %s#%d\n",
1321 		map->map_type, func_id_name(func_id), func_id);
1322 	return -EINVAL;
1323 }
1324 
1325 static int check_raw_mode(const struct bpf_func_proto *fn)
1326 {
1327 	int count = 0;
1328 
1329 	if (fn->arg1_type == ARG_PTR_TO_UNINIT_MEM)
1330 		count++;
1331 	if (fn->arg2_type == ARG_PTR_TO_UNINIT_MEM)
1332 		count++;
1333 	if (fn->arg3_type == ARG_PTR_TO_UNINIT_MEM)
1334 		count++;
1335 	if (fn->arg4_type == ARG_PTR_TO_UNINIT_MEM)
1336 		count++;
1337 	if (fn->arg5_type == ARG_PTR_TO_UNINIT_MEM)
1338 		count++;
1339 
1340 	return count > 1 ? -EINVAL : 0;
1341 }
1342 
1343 static void clear_all_pkt_pointers(struct bpf_verifier_env *env)
1344 {
1345 	struct bpf_verifier_state *state = &env->cur_state;
1346 	struct bpf_reg_state *regs = state->regs, *reg;
1347 	int i;
1348 
1349 	for (i = 0; i < MAX_BPF_REG; i++)
1350 		if (regs[i].type == PTR_TO_PACKET ||
1351 		    regs[i].type == PTR_TO_PACKET_END)
1352 			mark_reg_unknown_value(regs, i);
1353 
1354 	for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
1355 		if (state->stack_slot_type[i] != STACK_SPILL)
1356 			continue;
1357 		reg = &state->spilled_regs[i / BPF_REG_SIZE];
1358 		if (reg->type != PTR_TO_PACKET &&
1359 		    reg->type != PTR_TO_PACKET_END)
1360 			continue;
1361 		__mark_reg_unknown_value(state->spilled_regs,
1362 					 i / BPF_REG_SIZE);
1363 	}
1364 }
1365 
1366 static int check_call(struct bpf_verifier_env *env, int func_id, int insn_idx)
1367 {
1368 	struct bpf_verifier_state *state = &env->cur_state;
1369 	const struct bpf_func_proto *fn = NULL;
1370 	struct bpf_reg_state *regs = state->regs;
1371 	struct bpf_call_arg_meta meta;
1372 	bool changes_data;
1373 	int i, err;
1374 
1375 	/* find function prototype */
1376 	if (func_id < 0 || func_id >= __BPF_FUNC_MAX_ID) {
1377 		verbose("invalid func %s#%d\n", func_id_name(func_id), func_id);
1378 		return -EINVAL;
1379 	}
1380 
1381 	if (env->prog->aux->ops->get_func_proto)
1382 		fn = env->prog->aux->ops->get_func_proto(func_id);
1383 
1384 	if (!fn) {
1385 		verbose("unknown func %s#%d\n", func_id_name(func_id), func_id);
1386 		return -EINVAL;
1387 	}
1388 
1389 	/* eBPF programs must be GPL compatible to use GPL-ed functions */
1390 	if (!env->prog->gpl_compatible && fn->gpl_only) {
1391 		verbose("cannot call GPL only function from proprietary program\n");
1392 		return -EINVAL;
1393 	}
1394 
1395 	changes_data = bpf_helper_changes_pkt_data(fn->func);
1396 
1397 	memset(&meta, 0, sizeof(meta));
1398 	meta.pkt_access = fn->pkt_access;
1399 
1400 	/* We only support one arg being in raw mode at the moment, which
1401 	 * is sufficient for the helper functions we have right now.
1402 	 */
1403 	err = check_raw_mode(fn);
1404 	if (err) {
1405 		verbose("kernel subsystem misconfigured func %s#%d\n",
1406 			func_id_name(func_id), func_id);
1407 		return err;
1408 	}
1409 
1410 	/* check args */
1411 	err = check_func_arg(env, BPF_REG_1, fn->arg1_type, &meta);
1412 	if (err)
1413 		return err;
1414 	err = check_func_arg(env, BPF_REG_2, fn->arg2_type, &meta);
1415 	if (err)
1416 		return err;
1417 	err = check_func_arg(env, BPF_REG_3, fn->arg3_type, &meta);
1418 	if (err)
1419 		return err;
1420 	err = check_func_arg(env, BPF_REG_4, fn->arg4_type, &meta);
1421 	if (err)
1422 		return err;
1423 	err = check_func_arg(env, BPF_REG_5, fn->arg5_type, &meta);
1424 	if (err)
1425 		return err;
1426 
1427 	/* Mark slots with STACK_MISC in case of raw mode, stack offset
1428 	 * is inferred from register state.
1429 	 */
1430 	for (i = 0; i < meta.access_size; i++) {
1431 		err = check_mem_access(env, insn_idx, meta.regno, i, BPF_B, BPF_WRITE, -1);
1432 		if (err)
1433 			return err;
1434 	}
1435 
1436 	/* reset caller saved regs */
1437 	for (i = 0; i < CALLER_SAVED_REGS; i++)
1438 		mark_reg_not_init(regs, caller_saved[i]);
1439 
1440 	/* update return register */
1441 	if (fn->ret_type == RET_INTEGER) {
1442 		regs[BPF_REG_0].type = UNKNOWN_VALUE;
1443 	} else if (fn->ret_type == RET_VOID) {
1444 		regs[BPF_REG_0].type = NOT_INIT;
1445 	} else if (fn->ret_type == RET_PTR_TO_MAP_VALUE_OR_NULL) {
1446 		struct bpf_insn_aux_data *insn_aux;
1447 
1448 		regs[BPF_REG_0].type = PTR_TO_MAP_VALUE_OR_NULL;
1449 		regs[BPF_REG_0].max_value = regs[BPF_REG_0].min_value = 0;
1450 		/* remember map_ptr, so that check_map_access()
1451 		 * can check 'value_size' boundary of memory access
1452 		 * to map element returned from bpf_map_lookup_elem()
1453 		 */
1454 		if (meta.map_ptr == NULL) {
1455 			verbose("kernel subsystem misconfigured verifier\n");
1456 			return -EINVAL;
1457 		}
1458 		regs[BPF_REG_0].map_ptr = meta.map_ptr;
1459 		regs[BPF_REG_0].id = ++env->id_gen;
1460 		insn_aux = &env->insn_aux_data[insn_idx];
1461 		if (!insn_aux->map_ptr)
1462 			insn_aux->map_ptr = meta.map_ptr;
1463 		else if (insn_aux->map_ptr != meta.map_ptr)
1464 			insn_aux->map_ptr = BPF_MAP_PTR_POISON;
1465 	} else {
1466 		verbose("unknown return type %d of func %s#%d\n",
1467 			fn->ret_type, func_id_name(func_id), func_id);
1468 		return -EINVAL;
1469 	}
1470 
1471 	err = check_map_func_compatibility(meta.map_ptr, func_id);
1472 	if (err)
1473 		return err;
1474 
1475 	if (changes_data)
1476 		clear_all_pkt_pointers(env);
1477 	return 0;
1478 }
1479 
1480 static int check_packet_ptr_add(struct bpf_verifier_env *env,
1481 				struct bpf_insn *insn)
1482 {
1483 	struct bpf_reg_state *regs = env->cur_state.regs;
1484 	struct bpf_reg_state *dst_reg = &regs[insn->dst_reg];
1485 	struct bpf_reg_state *src_reg = &regs[insn->src_reg];
1486 	struct bpf_reg_state tmp_reg;
1487 	s32 imm;
1488 
1489 	if (BPF_SRC(insn->code) == BPF_K) {
1490 		/* pkt_ptr += imm */
1491 		imm = insn->imm;
1492 
1493 add_imm:
1494 		if (imm < 0) {
1495 			verbose("addition of negative constant to packet pointer is not allowed\n");
1496 			return -EACCES;
1497 		}
1498 		if (imm >= MAX_PACKET_OFF ||
1499 		    imm + dst_reg->off >= MAX_PACKET_OFF) {
1500 			verbose("constant %d is too large to add to packet pointer\n",
1501 				imm);
1502 			return -EACCES;
1503 		}
1504 		/* a constant was added to pkt_ptr.
1505 		 * Remember it while keeping the same 'id'
1506 		 */
1507 		dst_reg->off += imm;
1508 	} else {
1509 		bool had_id;
1510 
1511 		if (src_reg->type == PTR_TO_PACKET) {
1512 			/* R6=pkt(id=0,off=0,r=62) R7=imm22; r7 += r6 */
1513 			tmp_reg = *dst_reg;  /* save r7 state */
1514 			*dst_reg = *src_reg; /* copy pkt_ptr state r6 into r7 */
1515 			src_reg = &tmp_reg;  /* pretend it's src_reg state */
1516 			/* if the checks below reject it, the copy won't matter,
1517 			 * since we're rejecting the whole program. If all ok,
1518 			 * then imm22 state will be added to r7
1519 			 * and r7 will be pkt(id=0,off=22,r=62) while
1520 			 * r6 will stay as pkt(id=0,off=0,r=62)
1521 			 */
1522 		}
1523 
1524 		if (src_reg->type == CONST_IMM) {
1525 			/* pkt_ptr += reg where reg is known constant */
1526 			imm = src_reg->imm;
1527 			goto add_imm;
1528 		}
1529 		/* disallow pkt_ptr += reg
1530 		 * if reg is not uknown_value with guaranteed zero upper bits
1531 		 * otherwise pkt_ptr may overflow and addition will become
1532 		 * subtraction which is not allowed
1533 		 */
1534 		if (src_reg->type != UNKNOWN_VALUE) {
1535 			verbose("cannot add '%s' to ptr_to_packet\n",
1536 				reg_type_str[src_reg->type]);
1537 			return -EACCES;
1538 		}
1539 		if (src_reg->imm < 48) {
1540 			verbose("cannot add integer value with %lld upper zero bits to ptr_to_packet\n",
1541 				src_reg->imm);
1542 			return -EACCES;
1543 		}
1544 
1545 		had_id = (dst_reg->id != 0);
1546 
1547 		/* dst_reg stays as pkt_ptr type and since some positive
1548 		 * integer value was added to the pointer, increment its 'id'
1549 		 */
1550 		dst_reg->id = ++env->id_gen;
1551 
1552 		/* something was added to pkt_ptr, set range to zero */
1553 		dst_reg->aux_off += dst_reg->off;
1554 		dst_reg->off = 0;
1555 		dst_reg->range = 0;
1556 		if (had_id)
1557 			dst_reg->aux_off_align = min(dst_reg->aux_off_align,
1558 						     src_reg->min_align);
1559 		else
1560 			dst_reg->aux_off_align = src_reg->min_align;
1561 	}
1562 	return 0;
1563 }
1564 
1565 static int evaluate_reg_alu(struct bpf_verifier_env *env, struct bpf_insn *insn)
1566 {
1567 	struct bpf_reg_state *regs = env->cur_state.regs;
1568 	struct bpf_reg_state *dst_reg = &regs[insn->dst_reg];
1569 	u8 opcode = BPF_OP(insn->code);
1570 	s64 imm_log2;
1571 
1572 	/* for type == UNKNOWN_VALUE:
1573 	 * imm > 0 -> number of zero upper bits
1574 	 * imm == 0 -> don't track which is the same as all bits can be non-zero
1575 	 */
1576 
1577 	if (BPF_SRC(insn->code) == BPF_X) {
1578 		struct bpf_reg_state *src_reg = &regs[insn->src_reg];
1579 
1580 		if (src_reg->type == UNKNOWN_VALUE && src_reg->imm > 0 &&
1581 		    dst_reg->imm && opcode == BPF_ADD) {
1582 			/* dreg += sreg
1583 			 * where both have zero upper bits. Adding them
1584 			 * can only result making one more bit non-zero
1585 			 * in the larger value.
1586 			 * Ex. 0xffff (imm=48) + 1 (imm=63) = 0x10000 (imm=47)
1587 			 *     0xffff (imm=48) + 0xffff = 0x1fffe (imm=47)
1588 			 */
1589 			dst_reg->imm = min(dst_reg->imm, src_reg->imm);
1590 			dst_reg->imm--;
1591 			return 0;
1592 		}
1593 		if (src_reg->type == CONST_IMM && src_reg->imm > 0 &&
1594 		    dst_reg->imm && opcode == BPF_ADD) {
1595 			/* dreg += sreg
1596 			 * where dreg has zero upper bits and sreg is const.
1597 			 * Adding them can only result making one more bit
1598 			 * non-zero in the larger value.
1599 			 */
1600 			imm_log2 = __ilog2_u64((long long)src_reg->imm);
1601 			dst_reg->imm = min(dst_reg->imm, 63 - imm_log2);
1602 			dst_reg->imm--;
1603 			return 0;
1604 		}
1605 		/* all other cases non supported yet, just mark dst_reg */
1606 		dst_reg->imm = 0;
1607 		return 0;
1608 	}
1609 
1610 	/* sign extend 32-bit imm into 64-bit to make sure that
1611 	 * negative values occupy bit 63. Note ilog2() would have
1612 	 * been incorrect, since sizeof(insn->imm) == 4
1613 	 */
1614 	imm_log2 = __ilog2_u64((long long)insn->imm);
1615 
1616 	if (dst_reg->imm && opcode == BPF_LSH) {
1617 		/* reg <<= imm
1618 		 * if reg was a result of 2 byte load, then its imm == 48
1619 		 * which means that upper 48 bits are zero and shifting this reg
1620 		 * left by 4 would mean that upper 44 bits are still zero
1621 		 */
1622 		dst_reg->imm -= insn->imm;
1623 	} else if (dst_reg->imm && opcode == BPF_MUL) {
1624 		/* reg *= imm
1625 		 * if multiplying by 14 subtract 4
1626 		 * This is conservative calculation of upper zero bits.
1627 		 * It's not trying to special case insn->imm == 1 or 0 cases
1628 		 */
1629 		dst_reg->imm -= imm_log2 + 1;
1630 	} else if (opcode == BPF_AND) {
1631 		/* reg &= imm */
1632 		dst_reg->imm = 63 - imm_log2;
1633 	} else if (dst_reg->imm && opcode == BPF_ADD) {
1634 		/* reg += imm */
1635 		dst_reg->imm = min(dst_reg->imm, 63 - imm_log2);
1636 		dst_reg->imm--;
1637 	} else if (opcode == BPF_RSH) {
1638 		/* reg >>= imm
1639 		 * which means that after right shift, upper bits will be zero
1640 		 * note that verifier already checked that
1641 		 * 0 <= imm < 64 for shift insn
1642 		 */
1643 		dst_reg->imm += insn->imm;
1644 		if (unlikely(dst_reg->imm > 64))
1645 			/* some dumb code did:
1646 			 * r2 = *(u32 *)mem;
1647 			 * r2 >>= 32;
1648 			 * and all bits are zero now */
1649 			dst_reg->imm = 64;
1650 	} else {
1651 		/* all other alu ops, means that we don't know what will
1652 		 * happen to the value, mark it with unknown number of zero bits
1653 		 */
1654 		dst_reg->imm = 0;
1655 	}
1656 
1657 	if (dst_reg->imm < 0) {
1658 		/* all 64 bits of the register can contain non-zero bits
1659 		 * and such value cannot be added to ptr_to_packet, since it
1660 		 * may overflow, mark it as unknown to avoid further eval
1661 		 */
1662 		dst_reg->imm = 0;
1663 	}
1664 	return 0;
1665 }
1666 
1667 static int evaluate_reg_imm_alu_unknown(struct bpf_verifier_env *env,
1668 					struct bpf_insn *insn)
1669 {
1670 	struct bpf_reg_state *regs = env->cur_state.regs;
1671 	struct bpf_reg_state *dst_reg = &regs[insn->dst_reg];
1672 	struct bpf_reg_state *src_reg = &regs[insn->src_reg];
1673 	u8 opcode = BPF_OP(insn->code);
1674 	s64 imm_log2 = __ilog2_u64((long long)dst_reg->imm);
1675 
1676 	/* BPF_X code with src_reg->type UNKNOWN_VALUE here. */
1677 	if (src_reg->imm > 0 && dst_reg->imm) {
1678 		switch (opcode) {
1679 		case BPF_ADD:
1680 			/* dreg += sreg
1681 			 * where both have zero upper bits. Adding them
1682 			 * can only result making one more bit non-zero
1683 			 * in the larger value.
1684 			 * Ex. 0xffff (imm=48) + 1 (imm=63) = 0x10000 (imm=47)
1685 			 *     0xffff (imm=48) + 0xffff = 0x1fffe (imm=47)
1686 			 */
1687 			dst_reg->imm = min(src_reg->imm, 63 - imm_log2);
1688 			dst_reg->imm--;
1689 			break;
1690 		case BPF_AND:
1691 			/* dreg &= sreg
1692 			 * AND can not extend zero bits only shrink
1693 			 * Ex.  0x00..00ffffff
1694 			 *    & 0x0f..ffffffff
1695 			 *     ----------------
1696 			 *      0x00..00ffffff
1697 			 */
1698 			dst_reg->imm = max(src_reg->imm, 63 - imm_log2);
1699 			break;
1700 		case BPF_OR:
1701 			/* dreg |= sreg
1702 			 * OR can only extend zero bits
1703 			 * Ex.  0x00..00ffffff
1704 			 *    | 0x0f..ffffffff
1705 			 *     ----------------
1706 			 *      0x0f..00ffffff
1707 			 */
1708 			dst_reg->imm = min(src_reg->imm, 63 - imm_log2);
1709 			break;
1710 		case BPF_SUB:
1711 		case BPF_MUL:
1712 		case BPF_RSH:
1713 		case BPF_LSH:
1714 			/* These may be flushed out later */
1715 		default:
1716 			mark_reg_unknown_value(regs, insn->dst_reg);
1717 		}
1718 	} else {
1719 		mark_reg_unknown_value(regs, insn->dst_reg);
1720 	}
1721 
1722 	dst_reg->type = UNKNOWN_VALUE;
1723 	return 0;
1724 }
1725 
1726 static int evaluate_reg_imm_alu(struct bpf_verifier_env *env,
1727 				struct bpf_insn *insn)
1728 {
1729 	struct bpf_reg_state *regs = env->cur_state.regs;
1730 	struct bpf_reg_state *dst_reg = &regs[insn->dst_reg];
1731 	struct bpf_reg_state *src_reg = &regs[insn->src_reg];
1732 	u8 opcode = BPF_OP(insn->code);
1733 	u64 dst_imm = dst_reg->imm;
1734 
1735 	if (BPF_SRC(insn->code) == BPF_X && src_reg->type == UNKNOWN_VALUE)
1736 		return evaluate_reg_imm_alu_unknown(env, insn);
1737 
1738 	/* dst_reg->type == CONST_IMM here. Simulate execution of insns
1739 	 * containing ALU ops. Don't care about overflow or negative
1740 	 * values, just add/sub/... them; registers are in u64.
1741 	 */
1742 	if (opcode == BPF_ADD && BPF_SRC(insn->code) == BPF_K) {
1743 		dst_imm += insn->imm;
1744 	} else if (opcode == BPF_ADD && BPF_SRC(insn->code) == BPF_X &&
1745 		   src_reg->type == CONST_IMM) {
1746 		dst_imm += src_reg->imm;
1747 	} else if (opcode == BPF_SUB && BPF_SRC(insn->code) == BPF_K) {
1748 		dst_imm -= insn->imm;
1749 	} else if (opcode == BPF_SUB && BPF_SRC(insn->code) == BPF_X &&
1750 		   src_reg->type == CONST_IMM) {
1751 		dst_imm -= src_reg->imm;
1752 	} else if (opcode == BPF_MUL && BPF_SRC(insn->code) == BPF_K) {
1753 		dst_imm *= insn->imm;
1754 	} else if (opcode == BPF_MUL && BPF_SRC(insn->code) == BPF_X &&
1755 		   src_reg->type == CONST_IMM) {
1756 		dst_imm *= src_reg->imm;
1757 	} else if (opcode == BPF_OR && BPF_SRC(insn->code) == BPF_K) {
1758 		dst_imm |= insn->imm;
1759 	} else if (opcode == BPF_OR && BPF_SRC(insn->code) == BPF_X &&
1760 		   src_reg->type == CONST_IMM) {
1761 		dst_imm |= src_reg->imm;
1762 	} else if (opcode == BPF_AND && BPF_SRC(insn->code) == BPF_K) {
1763 		dst_imm &= insn->imm;
1764 	} else if (opcode == BPF_AND && BPF_SRC(insn->code) == BPF_X &&
1765 		   src_reg->type == CONST_IMM) {
1766 		dst_imm &= src_reg->imm;
1767 	} else if (opcode == BPF_RSH && BPF_SRC(insn->code) == BPF_K) {
1768 		dst_imm >>= insn->imm;
1769 	} else if (opcode == BPF_RSH && BPF_SRC(insn->code) == BPF_X &&
1770 		   src_reg->type == CONST_IMM) {
1771 		dst_imm >>= src_reg->imm;
1772 	} else if (opcode == BPF_LSH && BPF_SRC(insn->code) == BPF_K) {
1773 		dst_imm <<= insn->imm;
1774 	} else if (opcode == BPF_LSH && BPF_SRC(insn->code) == BPF_X &&
1775 		   src_reg->type == CONST_IMM) {
1776 		dst_imm <<= src_reg->imm;
1777 	} else {
1778 		mark_reg_unknown_value(regs, insn->dst_reg);
1779 		goto out;
1780 	}
1781 
1782 	dst_reg->imm = dst_imm;
1783 out:
1784 	return 0;
1785 }
1786 
1787 static void check_reg_overflow(struct bpf_reg_state *reg)
1788 {
1789 	if (reg->max_value > BPF_REGISTER_MAX_RANGE)
1790 		reg->max_value = BPF_REGISTER_MAX_RANGE;
1791 	if (reg->min_value < BPF_REGISTER_MIN_RANGE ||
1792 	    reg->min_value > BPF_REGISTER_MAX_RANGE)
1793 		reg->min_value = BPF_REGISTER_MIN_RANGE;
1794 }
1795 
1796 static u32 calc_align(u32 imm)
1797 {
1798 	if (!imm)
1799 		return 1U << 31;
1800 	return imm - ((imm - 1) & imm);
1801 }
1802 
1803 static void adjust_reg_min_max_vals(struct bpf_verifier_env *env,
1804 				    struct bpf_insn *insn)
1805 {
1806 	struct bpf_reg_state *regs = env->cur_state.regs, *dst_reg;
1807 	s64 min_val = BPF_REGISTER_MIN_RANGE;
1808 	u64 max_val = BPF_REGISTER_MAX_RANGE;
1809 	u8 opcode = BPF_OP(insn->code);
1810 	u32 dst_align, src_align;
1811 
1812 	dst_reg = &regs[insn->dst_reg];
1813 	src_align = 0;
1814 	if (BPF_SRC(insn->code) == BPF_X) {
1815 		check_reg_overflow(&regs[insn->src_reg]);
1816 		min_val = regs[insn->src_reg].min_value;
1817 		max_val = regs[insn->src_reg].max_value;
1818 
1819 		/* If the source register is a random pointer then the
1820 		 * min_value/max_value values represent the range of the known
1821 		 * accesses into that value, not the actual min/max value of the
1822 		 * register itself.  In this case we have to reset the reg range
1823 		 * values so we know it is not safe to look at.
1824 		 */
1825 		if (regs[insn->src_reg].type != CONST_IMM &&
1826 		    regs[insn->src_reg].type != UNKNOWN_VALUE) {
1827 			min_val = BPF_REGISTER_MIN_RANGE;
1828 			max_val = BPF_REGISTER_MAX_RANGE;
1829 			src_align = 0;
1830 		} else {
1831 			src_align = regs[insn->src_reg].min_align;
1832 		}
1833 	} else if (insn->imm < BPF_REGISTER_MAX_RANGE &&
1834 		   (s64)insn->imm > BPF_REGISTER_MIN_RANGE) {
1835 		min_val = max_val = insn->imm;
1836 		src_align = calc_align(insn->imm);
1837 	}
1838 
1839 	dst_align = dst_reg->min_align;
1840 
1841 	/* We don't know anything about what was done to this register, mark it
1842 	 * as unknown. Also, if both derived bounds came from signed/unsigned
1843 	 * mixed compares and one side is unbounded, we cannot really do anything
1844 	 * with them as boundaries cannot be trusted. Thus, arithmetic of two
1845 	 * regs of such kind will get invalidated bounds on the dst side.
1846 	 */
1847 	if ((min_val == BPF_REGISTER_MIN_RANGE &&
1848 	     max_val == BPF_REGISTER_MAX_RANGE) ||
1849 	    (BPF_SRC(insn->code) == BPF_X &&
1850 	     ((min_val != BPF_REGISTER_MIN_RANGE &&
1851 	       max_val == BPF_REGISTER_MAX_RANGE) ||
1852 	      (min_val == BPF_REGISTER_MIN_RANGE &&
1853 	       max_val != BPF_REGISTER_MAX_RANGE) ||
1854 	      (dst_reg->min_value != BPF_REGISTER_MIN_RANGE &&
1855 	       dst_reg->max_value == BPF_REGISTER_MAX_RANGE) ||
1856 	      (dst_reg->min_value == BPF_REGISTER_MIN_RANGE &&
1857 	       dst_reg->max_value != BPF_REGISTER_MAX_RANGE)) &&
1858 	     regs[insn->dst_reg].value_from_signed !=
1859 	     regs[insn->src_reg].value_from_signed)) {
1860 		reset_reg_range_values(regs, insn->dst_reg);
1861 		return;
1862 	}
1863 
1864 	/* If one of our values was at the end of our ranges then we can't just
1865 	 * do our normal operations to the register, we need to set the values
1866 	 * to the min/max since they are undefined.
1867 	 */
1868 	if (min_val == BPF_REGISTER_MIN_RANGE)
1869 		dst_reg->min_value = BPF_REGISTER_MIN_RANGE;
1870 	if (max_val == BPF_REGISTER_MAX_RANGE)
1871 		dst_reg->max_value = BPF_REGISTER_MAX_RANGE;
1872 
1873 	switch (opcode) {
1874 	case BPF_ADD:
1875 		if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE)
1876 			dst_reg->min_value += min_val;
1877 		if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1878 			dst_reg->max_value += max_val;
1879 		dst_reg->min_align = min(src_align, dst_align);
1880 		break;
1881 	case BPF_SUB:
1882 		if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE)
1883 			dst_reg->min_value -= min_val;
1884 		if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1885 			dst_reg->max_value -= max_val;
1886 		dst_reg->min_align = min(src_align, dst_align);
1887 		break;
1888 	case BPF_MUL:
1889 		if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE)
1890 			dst_reg->min_value *= min_val;
1891 		if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1892 			dst_reg->max_value *= max_val;
1893 		dst_reg->min_align = max(src_align, dst_align);
1894 		break;
1895 	case BPF_AND:
1896 		/* Disallow AND'ing of negative numbers, ain't nobody got time
1897 		 * for that.  Otherwise the minimum is 0 and the max is the max
1898 		 * value we could AND against.
1899 		 */
1900 		if (min_val < 0)
1901 			dst_reg->min_value = BPF_REGISTER_MIN_RANGE;
1902 		else
1903 			dst_reg->min_value = 0;
1904 		dst_reg->max_value = max_val;
1905 		dst_reg->min_align = max(src_align, dst_align);
1906 		break;
1907 	case BPF_LSH:
1908 		/* Gotta have special overflow logic here, if we're shifting
1909 		 * more than MAX_RANGE then just assume we have an invalid
1910 		 * range.
1911 		 */
1912 		if (min_val > ilog2(BPF_REGISTER_MAX_RANGE)) {
1913 			dst_reg->min_value = BPF_REGISTER_MIN_RANGE;
1914 			dst_reg->min_align = 1;
1915 		} else {
1916 			if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE)
1917 				dst_reg->min_value <<= min_val;
1918 			if (!dst_reg->min_align)
1919 				dst_reg->min_align = 1;
1920 			dst_reg->min_align <<= min_val;
1921 		}
1922 		if (max_val > ilog2(BPF_REGISTER_MAX_RANGE))
1923 			dst_reg->max_value = BPF_REGISTER_MAX_RANGE;
1924 		else if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1925 			dst_reg->max_value <<= max_val;
1926 		break;
1927 	case BPF_RSH:
1928 		/* RSH by a negative number is undefined, and the BPF_RSH is an
1929 		 * unsigned shift, so make the appropriate casts.
1930 		 */
1931 		if (min_val < 0 || dst_reg->min_value < 0) {
1932 			dst_reg->min_value = BPF_REGISTER_MIN_RANGE;
1933 		} else {
1934 			dst_reg->min_value =
1935 				(u64)(dst_reg->min_value) >> min_val;
1936 		}
1937 		if (min_val < 0) {
1938 			dst_reg->min_align = 1;
1939 		} else {
1940 			dst_reg->min_align >>= (u64) min_val;
1941 			if (!dst_reg->min_align)
1942 				dst_reg->min_align = 1;
1943 		}
1944 		if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1945 			dst_reg->max_value >>= max_val;
1946 		break;
1947 	default:
1948 		reset_reg_range_values(regs, insn->dst_reg);
1949 		break;
1950 	}
1951 
1952 	check_reg_overflow(dst_reg);
1953 }
1954 
1955 /* check validity of 32-bit and 64-bit arithmetic operations */
1956 static int check_alu_op(struct bpf_verifier_env *env, struct bpf_insn *insn)
1957 {
1958 	struct bpf_reg_state *regs = env->cur_state.regs, *dst_reg;
1959 	u8 opcode = BPF_OP(insn->code);
1960 	int err;
1961 
1962 	if (opcode == BPF_END || opcode == BPF_NEG) {
1963 		if (opcode == BPF_NEG) {
1964 			if (BPF_SRC(insn->code) != 0 ||
1965 			    insn->src_reg != BPF_REG_0 ||
1966 			    insn->off != 0 || insn->imm != 0) {
1967 				verbose("BPF_NEG uses reserved fields\n");
1968 				return -EINVAL;
1969 			}
1970 		} else {
1971 			if (insn->src_reg != BPF_REG_0 || insn->off != 0 ||
1972 			    (insn->imm != 16 && insn->imm != 32 && insn->imm != 64)) {
1973 				verbose("BPF_END uses reserved fields\n");
1974 				return -EINVAL;
1975 			}
1976 		}
1977 
1978 		/* check src operand */
1979 		err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1980 		if (err)
1981 			return err;
1982 
1983 		if (is_pointer_value(env, insn->dst_reg)) {
1984 			verbose("R%d pointer arithmetic prohibited\n",
1985 				insn->dst_reg);
1986 			return -EACCES;
1987 		}
1988 
1989 		/* check dest operand */
1990 		err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1991 		if (err)
1992 			return err;
1993 
1994 	} else if (opcode == BPF_MOV) {
1995 
1996 		if (BPF_SRC(insn->code) == BPF_X) {
1997 			if (insn->imm != 0 || insn->off != 0) {
1998 				verbose("BPF_MOV uses reserved fields\n");
1999 				return -EINVAL;
2000 			}
2001 
2002 			/* check src operand */
2003 			err = check_reg_arg(regs, insn->src_reg, SRC_OP);
2004 			if (err)
2005 				return err;
2006 		} else {
2007 			if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
2008 				verbose("BPF_MOV uses reserved fields\n");
2009 				return -EINVAL;
2010 			}
2011 		}
2012 
2013 		/* check dest operand */
2014 		err = check_reg_arg(regs, insn->dst_reg, DST_OP);
2015 		if (err)
2016 			return err;
2017 
2018 		/* we are setting our register to something new, we need to
2019 		 * reset its range values.
2020 		 */
2021 		reset_reg_range_values(regs, insn->dst_reg);
2022 
2023 		if (BPF_SRC(insn->code) == BPF_X) {
2024 			if (BPF_CLASS(insn->code) == BPF_ALU64) {
2025 				/* case: R1 = R2
2026 				 * copy register state to dest reg
2027 				 */
2028 				regs[insn->dst_reg] = regs[insn->src_reg];
2029 			} else {
2030 				if (is_pointer_value(env, insn->src_reg)) {
2031 					verbose("R%d partial copy of pointer\n",
2032 						insn->src_reg);
2033 					return -EACCES;
2034 				}
2035 				mark_reg_unknown_value(regs, insn->dst_reg);
2036 			}
2037 		} else {
2038 			/* case: R = imm
2039 			 * remember the value we stored into this reg
2040 			 */
2041 			regs[insn->dst_reg].type = CONST_IMM;
2042 			regs[insn->dst_reg].imm = insn->imm;
2043 			regs[insn->dst_reg].id = 0;
2044 			regs[insn->dst_reg].max_value = insn->imm;
2045 			regs[insn->dst_reg].min_value = insn->imm;
2046 			regs[insn->dst_reg].min_align = calc_align(insn->imm);
2047 			regs[insn->dst_reg].value_from_signed = false;
2048 		}
2049 
2050 	} else if (opcode > BPF_END) {
2051 		verbose("invalid BPF_ALU opcode %x\n", opcode);
2052 		return -EINVAL;
2053 
2054 	} else {	/* all other ALU ops: and, sub, xor, add, ... */
2055 
2056 		if (BPF_SRC(insn->code) == BPF_X) {
2057 			if (insn->imm != 0 || insn->off != 0) {
2058 				verbose("BPF_ALU uses reserved fields\n");
2059 				return -EINVAL;
2060 			}
2061 			/* check src1 operand */
2062 			err = check_reg_arg(regs, insn->src_reg, SRC_OP);
2063 			if (err)
2064 				return err;
2065 		} else {
2066 			if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
2067 				verbose("BPF_ALU uses reserved fields\n");
2068 				return -EINVAL;
2069 			}
2070 		}
2071 
2072 		/* check src2 operand */
2073 		err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
2074 		if (err)
2075 			return err;
2076 
2077 		if ((opcode == BPF_MOD || opcode == BPF_DIV) &&
2078 		    BPF_SRC(insn->code) == BPF_K && insn->imm == 0) {
2079 			verbose("div by zero\n");
2080 			return -EINVAL;
2081 		}
2082 
2083 		if ((opcode == BPF_LSH || opcode == BPF_RSH ||
2084 		     opcode == BPF_ARSH) && BPF_SRC(insn->code) == BPF_K) {
2085 			int size = BPF_CLASS(insn->code) == BPF_ALU64 ? 64 : 32;
2086 
2087 			if (insn->imm < 0 || insn->imm >= size) {
2088 				verbose("invalid shift %d\n", insn->imm);
2089 				return -EINVAL;
2090 			}
2091 		}
2092 
2093 		/* check dest operand */
2094 		err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK);
2095 		if (err)
2096 			return err;
2097 
2098 		dst_reg = &regs[insn->dst_reg];
2099 
2100 		/* first we want to adjust our ranges. */
2101 		adjust_reg_min_max_vals(env, insn);
2102 
2103 		/* pattern match 'bpf_add Rx, imm' instruction */
2104 		if (opcode == BPF_ADD && BPF_CLASS(insn->code) == BPF_ALU64 &&
2105 		    dst_reg->type == FRAME_PTR && BPF_SRC(insn->code) == BPF_K) {
2106 			dst_reg->type = PTR_TO_STACK;
2107 			dst_reg->imm = insn->imm;
2108 			return 0;
2109 		} else if (opcode == BPF_ADD &&
2110 			   BPF_CLASS(insn->code) == BPF_ALU64 &&
2111 			   dst_reg->type == PTR_TO_STACK &&
2112 			   ((BPF_SRC(insn->code) == BPF_X &&
2113 			     regs[insn->src_reg].type == CONST_IMM) ||
2114 			    BPF_SRC(insn->code) == BPF_K)) {
2115 			if (BPF_SRC(insn->code) == BPF_X)
2116 				dst_reg->imm += regs[insn->src_reg].imm;
2117 			else
2118 				dst_reg->imm += insn->imm;
2119 			return 0;
2120 		} else if (opcode == BPF_ADD &&
2121 			   BPF_CLASS(insn->code) == BPF_ALU64 &&
2122 			   (dst_reg->type == PTR_TO_PACKET ||
2123 			    (BPF_SRC(insn->code) == BPF_X &&
2124 			     regs[insn->src_reg].type == PTR_TO_PACKET))) {
2125 			/* ptr_to_packet += K|X */
2126 			return check_packet_ptr_add(env, insn);
2127 		} else if (BPF_CLASS(insn->code) == BPF_ALU64 &&
2128 			   dst_reg->type == UNKNOWN_VALUE &&
2129 			   env->allow_ptr_leaks) {
2130 			/* unknown += K|X */
2131 			return evaluate_reg_alu(env, insn);
2132 		} else if (BPF_CLASS(insn->code) == BPF_ALU64 &&
2133 			   dst_reg->type == CONST_IMM &&
2134 			   env->allow_ptr_leaks) {
2135 			/* reg_imm += K|X */
2136 			return evaluate_reg_imm_alu(env, insn);
2137 		} else if (is_pointer_value(env, insn->dst_reg)) {
2138 			verbose("R%d pointer arithmetic prohibited\n",
2139 				insn->dst_reg);
2140 			return -EACCES;
2141 		} else if (BPF_SRC(insn->code) == BPF_X &&
2142 			   is_pointer_value(env, insn->src_reg)) {
2143 			verbose("R%d pointer arithmetic prohibited\n",
2144 				insn->src_reg);
2145 			return -EACCES;
2146 		}
2147 
2148 		/* If we did pointer math on a map value then just set it to our
2149 		 * PTR_TO_MAP_VALUE_ADJ type so we can deal with any stores or
2150 		 * loads to this register appropriately, otherwise just mark the
2151 		 * register as unknown.
2152 		 */
2153 		if (env->allow_ptr_leaks &&
2154 		    BPF_CLASS(insn->code) == BPF_ALU64 && opcode == BPF_ADD &&
2155 		    (dst_reg->type == PTR_TO_MAP_VALUE ||
2156 		     dst_reg->type == PTR_TO_MAP_VALUE_ADJ))
2157 			dst_reg->type = PTR_TO_MAP_VALUE_ADJ;
2158 		else
2159 			mark_reg_unknown_value(regs, insn->dst_reg);
2160 	}
2161 
2162 	return 0;
2163 }
2164 
2165 static void find_good_pkt_pointers(struct bpf_verifier_state *state,
2166 				   struct bpf_reg_state *dst_reg)
2167 {
2168 	struct bpf_reg_state *regs = state->regs, *reg;
2169 	int i;
2170 
2171 	/* LLVM can generate two kind of checks:
2172 	 *
2173 	 * Type 1:
2174 	 *
2175 	 *   r2 = r3;
2176 	 *   r2 += 8;
2177 	 *   if (r2 > pkt_end) goto <handle exception>
2178 	 *   <access okay>
2179 	 *
2180 	 *   Where:
2181 	 *     r2 == dst_reg, pkt_end == src_reg
2182 	 *     r2=pkt(id=n,off=8,r=0)
2183 	 *     r3=pkt(id=n,off=0,r=0)
2184 	 *
2185 	 * Type 2:
2186 	 *
2187 	 *   r2 = r3;
2188 	 *   r2 += 8;
2189 	 *   if (pkt_end >= r2) goto <access okay>
2190 	 *   <handle exception>
2191 	 *
2192 	 *   Where:
2193 	 *     pkt_end == dst_reg, r2 == src_reg
2194 	 *     r2=pkt(id=n,off=8,r=0)
2195 	 *     r3=pkt(id=n,off=0,r=0)
2196 	 *
2197 	 * Find register r3 and mark its range as r3=pkt(id=n,off=0,r=8)
2198 	 * so that range of bytes [r3, r3 + 8) is safe to access.
2199 	 */
2200 
2201 	for (i = 0; i < MAX_BPF_REG; i++)
2202 		if (regs[i].type == PTR_TO_PACKET && regs[i].id == dst_reg->id)
2203 			/* keep the maximum range already checked */
2204 			regs[i].range = max(regs[i].range, dst_reg->off);
2205 
2206 	for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
2207 		if (state->stack_slot_type[i] != STACK_SPILL)
2208 			continue;
2209 		reg = &state->spilled_regs[i / BPF_REG_SIZE];
2210 		if (reg->type == PTR_TO_PACKET && reg->id == dst_reg->id)
2211 			reg->range = max(reg->range, dst_reg->off);
2212 	}
2213 }
2214 
2215 /* Adjusts the register min/max values in the case that the dst_reg is the
2216  * variable register that we are working on, and src_reg is a constant or we're
2217  * simply doing a BPF_K check.
2218  */
2219 static void reg_set_min_max(struct bpf_reg_state *true_reg,
2220 			    struct bpf_reg_state *false_reg, u64 val,
2221 			    u8 opcode)
2222 {
2223 	bool value_from_signed = true;
2224 	bool is_range = true;
2225 
2226 	switch (opcode) {
2227 	case BPF_JEQ:
2228 		/* If this is false then we know nothing Jon Snow, but if it is
2229 		 * true then we know for sure.
2230 		 */
2231 		true_reg->max_value = true_reg->min_value = val;
2232 		is_range = false;
2233 		break;
2234 	case BPF_JNE:
2235 		/* If this is true we know nothing Jon Snow, but if it is false
2236 		 * we know the value for sure;
2237 		 */
2238 		false_reg->max_value = false_reg->min_value = val;
2239 		is_range = false;
2240 		break;
2241 	case BPF_JGT:
2242 		value_from_signed = false;
2243 		/* fallthrough */
2244 	case BPF_JSGT:
2245 		if (true_reg->value_from_signed != value_from_signed)
2246 			reset_reg_range_values(true_reg, 0);
2247 		if (false_reg->value_from_signed != value_from_signed)
2248 			reset_reg_range_values(false_reg, 0);
2249 		if (opcode == BPF_JGT) {
2250 			/* Unsigned comparison, the minimum value is 0. */
2251 			false_reg->min_value = 0;
2252 		}
2253 		/* If this is false then we know the maximum val is val,
2254 		 * otherwise we know the min val is val+1.
2255 		 */
2256 		false_reg->max_value = val;
2257 		false_reg->value_from_signed = value_from_signed;
2258 		true_reg->min_value = val + 1;
2259 		true_reg->value_from_signed = value_from_signed;
2260 		break;
2261 	case BPF_JGE:
2262 		value_from_signed = false;
2263 		/* fallthrough */
2264 	case BPF_JSGE:
2265 		if (true_reg->value_from_signed != value_from_signed)
2266 			reset_reg_range_values(true_reg, 0);
2267 		if (false_reg->value_from_signed != value_from_signed)
2268 			reset_reg_range_values(false_reg, 0);
2269 		if (opcode == BPF_JGE) {
2270 			/* Unsigned comparison, the minimum value is 0. */
2271 			false_reg->min_value = 0;
2272 		}
2273 		/* If this is false then we know the maximum value is val - 1,
2274 		 * otherwise we know the mimimum value is val.
2275 		 */
2276 		false_reg->max_value = val - 1;
2277 		false_reg->value_from_signed = value_from_signed;
2278 		true_reg->min_value = val;
2279 		true_reg->value_from_signed = value_from_signed;
2280 		break;
2281 	default:
2282 		break;
2283 	}
2284 
2285 	check_reg_overflow(false_reg);
2286 	check_reg_overflow(true_reg);
2287 	if (is_range) {
2288 		if (__is_pointer_value(false, false_reg))
2289 			reset_reg_range_values(false_reg, 0);
2290 		if (__is_pointer_value(false, true_reg))
2291 			reset_reg_range_values(true_reg, 0);
2292 	}
2293 }
2294 
2295 /* Same as above, but for the case that dst_reg is a CONST_IMM reg and src_reg
2296  * is the variable reg.
2297  */
2298 static void reg_set_min_max_inv(struct bpf_reg_state *true_reg,
2299 				struct bpf_reg_state *false_reg, u64 val,
2300 				u8 opcode)
2301 {
2302 	bool value_from_signed = true;
2303 	bool is_range = true;
2304 
2305 	switch (opcode) {
2306 	case BPF_JEQ:
2307 		/* If this is false then we know nothing Jon Snow, but if it is
2308 		 * true then we know for sure.
2309 		 */
2310 		true_reg->max_value = true_reg->min_value = val;
2311 		is_range = false;
2312 		break;
2313 	case BPF_JNE:
2314 		/* If this is true we know nothing Jon Snow, but if it is false
2315 		 * we know the value for sure;
2316 		 */
2317 		false_reg->max_value = false_reg->min_value = val;
2318 		is_range = false;
2319 		break;
2320 	case BPF_JGT:
2321 		value_from_signed = false;
2322 		/* fallthrough */
2323 	case BPF_JSGT:
2324 		if (true_reg->value_from_signed != value_from_signed)
2325 			reset_reg_range_values(true_reg, 0);
2326 		if (false_reg->value_from_signed != value_from_signed)
2327 			reset_reg_range_values(false_reg, 0);
2328 		if (opcode == BPF_JGT) {
2329 			/* Unsigned comparison, the minimum value is 0. */
2330 			true_reg->min_value = 0;
2331 		}
2332 		/*
2333 		 * If this is false, then the val is <= the register, if it is
2334 		 * true the register <= to the val.
2335 		 */
2336 		false_reg->min_value = val;
2337 		false_reg->value_from_signed = value_from_signed;
2338 		true_reg->max_value = val - 1;
2339 		true_reg->value_from_signed = value_from_signed;
2340 		break;
2341 	case BPF_JGE:
2342 		value_from_signed = false;
2343 		/* fallthrough */
2344 	case BPF_JSGE:
2345 		if (true_reg->value_from_signed != value_from_signed)
2346 			reset_reg_range_values(true_reg, 0);
2347 		if (false_reg->value_from_signed != value_from_signed)
2348 			reset_reg_range_values(false_reg, 0);
2349 		if (opcode == BPF_JGE) {
2350 			/* Unsigned comparison, the minimum value is 0. */
2351 			true_reg->min_value = 0;
2352 		}
2353 		/* If this is false then constant < register, if it is true then
2354 		 * the register < constant.
2355 		 */
2356 		false_reg->min_value = val + 1;
2357 		false_reg->value_from_signed = value_from_signed;
2358 		true_reg->max_value = val;
2359 		true_reg->value_from_signed = value_from_signed;
2360 		break;
2361 	default:
2362 		break;
2363 	}
2364 
2365 	check_reg_overflow(false_reg);
2366 	check_reg_overflow(true_reg);
2367 	if (is_range) {
2368 		if (__is_pointer_value(false, false_reg))
2369 			reset_reg_range_values(false_reg, 0);
2370 		if (__is_pointer_value(false, true_reg))
2371 			reset_reg_range_values(true_reg, 0);
2372 	}
2373 }
2374 
2375 static void mark_map_reg(struct bpf_reg_state *regs, u32 regno, u32 id,
2376 			 enum bpf_reg_type type)
2377 {
2378 	struct bpf_reg_state *reg = &regs[regno];
2379 
2380 	if (reg->type == PTR_TO_MAP_VALUE_OR_NULL && reg->id == id) {
2381 		if (type == UNKNOWN_VALUE) {
2382 			__mark_reg_unknown_value(regs, regno);
2383 		} else if (reg->map_ptr->inner_map_meta) {
2384 			reg->type = CONST_PTR_TO_MAP;
2385 			reg->map_ptr = reg->map_ptr->inner_map_meta;
2386 		} else {
2387 			reg->type = type;
2388 		}
2389 		/* We don't need id from this point onwards anymore, thus we
2390 		 * should better reset it, so that state pruning has chances
2391 		 * to take effect.
2392 		 */
2393 		reg->id = 0;
2394 	}
2395 }
2396 
2397 /* The logic is similar to find_good_pkt_pointers(), both could eventually
2398  * be folded together at some point.
2399  */
2400 static void mark_map_regs(struct bpf_verifier_state *state, u32 regno,
2401 			  enum bpf_reg_type type)
2402 {
2403 	struct bpf_reg_state *regs = state->regs;
2404 	u32 id = regs[regno].id;
2405 	int i;
2406 
2407 	for (i = 0; i < MAX_BPF_REG; i++)
2408 		mark_map_reg(regs, i, id, type);
2409 
2410 	for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
2411 		if (state->stack_slot_type[i] != STACK_SPILL)
2412 			continue;
2413 		mark_map_reg(state->spilled_regs, i / BPF_REG_SIZE, id, type);
2414 	}
2415 }
2416 
2417 static int check_cond_jmp_op(struct bpf_verifier_env *env,
2418 			     struct bpf_insn *insn, int *insn_idx)
2419 {
2420 	struct bpf_verifier_state *other_branch, *this_branch = &env->cur_state;
2421 	struct bpf_reg_state *regs = this_branch->regs, *dst_reg;
2422 	u8 opcode = BPF_OP(insn->code);
2423 	int err;
2424 
2425 	if (opcode > BPF_EXIT) {
2426 		verbose("invalid BPF_JMP opcode %x\n", opcode);
2427 		return -EINVAL;
2428 	}
2429 
2430 	if (BPF_SRC(insn->code) == BPF_X) {
2431 		if (insn->imm != 0) {
2432 			verbose("BPF_JMP uses reserved fields\n");
2433 			return -EINVAL;
2434 		}
2435 
2436 		/* check src1 operand */
2437 		err = check_reg_arg(regs, insn->src_reg, SRC_OP);
2438 		if (err)
2439 			return err;
2440 
2441 		if (is_pointer_value(env, insn->src_reg)) {
2442 			verbose("R%d pointer comparison prohibited\n",
2443 				insn->src_reg);
2444 			return -EACCES;
2445 		}
2446 	} else {
2447 		if (insn->src_reg != BPF_REG_0) {
2448 			verbose("BPF_JMP uses reserved fields\n");
2449 			return -EINVAL;
2450 		}
2451 	}
2452 
2453 	/* check src2 operand */
2454 	err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
2455 	if (err)
2456 		return err;
2457 
2458 	dst_reg = &regs[insn->dst_reg];
2459 
2460 	/* detect if R == 0 where R was initialized to zero earlier */
2461 	if (BPF_SRC(insn->code) == BPF_K &&
2462 	    (opcode == BPF_JEQ || opcode == BPF_JNE) &&
2463 	    dst_reg->type == CONST_IMM && dst_reg->imm == insn->imm) {
2464 		if (opcode == BPF_JEQ) {
2465 			/* if (imm == imm) goto pc+off;
2466 			 * only follow the goto, ignore fall-through
2467 			 */
2468 			*insn_idx += insn->off;
2469 			return 0;
2470 		} else {
2471 			/* if (imm != imm) goto pc+off;
2472 			 * only follow fall-through branch, since
2473 			 * that's where the program will go
2474 			 */
2475 			return 0;
2476 		}
2477 	}
2478 
2479 	other_branch = push_stack(env, *insn_idx + insn->off + 1, *insn_idx);
2480 	if (!other_branch)
2481 		return -EFAULT;
2482 
2483 	/* detect if we are comparing against a constant value so we can adjust
2484 	 * our min/max values for our dst register.
2485 	 */
2486 	if (BPF_SRC(insn->code) == BPF_X) {
2487 		if (regs[insn->src_reg].type == CONST_IMM)
2488 			reg_set_min_max(&other_branch->regs[insn->dst_reg],
2489 					dst_reg, regs[insn->src_reg].imm,
2490 					opcode);
2491 		else if (dst_reg->type == CONST_IMM)
2492 			reg_set_min_max_inv(&other_branch->regs[insn->src_reg],
2493 					    &regs[insn->src_reg], dst_reg->imm,
2494 					    opcode);
2495 	} else {
2496 		reg_set_min_max(&other_branch->regs[insn->dst_reg],
2497 					dst_reg, insn->imm, opcode);
2498 	}
2499 
2500 	/* detect if R == 0 where R is returned from bpf_map_lookup_elem() */
2501 	if (BPF_SRC(insn->code) == BPF_K &&
2502 	    insn->imm == 0 && (opcode == BPF_JEQ || opcode == BPF_JNE) &&
2503 	    dst_reg->type == PTR_TO_MAP_VALUE_OR_NULL) {
2504 		/* Mark all identical map registers in each branch as either
2505 		 * safe or unknown depending R == 0 or R != 0 conditional.
2506 		 */
2507 		mark_map_regs(this_branch, insn->dst_reg,
2508 			      opcode == BPF_JEQ ? PTR_TO_MAP_VALUE : UNKNOWN_VALUE);
2509 		mark_map_regs(other_branch, insn->dst_reg,
2510 			      opcode == BPF_JEQ ? UNKNOWN_VALUE : PTR_TO_MAP_VALUE);
2511 	} else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JGT &&
2512 		   dst_reg->type == PTR_TO_PACKET &&
2513 		   regs[insn->src_reg].type == PTR_TO_PACKET_END) {
2514 		find_good_pkt_pointers(this_branch, dst_reg);
2515 	} else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JGE &&
2516 		   dst_reg->type == PTR_TO_PACKET_END &&
2517 		   regs[insn->src_reg].type == PTR_TO_PACKET) {
2518 		find_good_pkt_pointers(other_branch, &regs[insn->src_reg]);
2519 	} else if (is_pointer_value(env, insn->dst_reg)) {
2520 		verbose("R%d pointer comparison prohibited\n", insn->dst_reg);
2521 		return -EACCES;
2522 	}
2523 	if (log_level)
2524 		print_verifier_state(this_branch);
2525 	return 0;
2526 }
2527 
2528 /* return the map pointer stored inside BPF_LD_IMM64 instruction */
2529 static struct bpf_map *ld_imm64_to_map_ptr(struct bpf_insn *insn)
2530 {
2531 	u64 imm64 = ((u64) (u32) insn[0].imm) | ((u64) (u32) insn[1].imm) << 32;
2532 
2533 	return (struct bpf_map *) (unsigned long) imm64;
2534 }
2535 
2536 /* verify BPF_LD_IMM64 instruction */
2537 static int check_ld_imm(struct bpf_verifier_env *env, struct bpf_insn *insn)
2538 {
2539 	struct bpf_reg_state *regs = env->cur_state.regs;
2540 	int err;
2541 
2542 	if (BPF_SIZE(insn->code) != BPF_DW) {
2543 		verbose("invalid BPF_LD_IMM insn\n");
2544 		return -EINVAL;
2545 	}
2546 	if (insn->off != 0) {
2547 		verbose("BPF_LD_IMM64 uses reserved fields\n");
2548 		return -EINVAL;
2549 	}
2550 
2551 	err = check_reg_arg(regs, insn->dst_reg, DST_OP);
2552 	if (err)
2553 		return err;
2554 
2555 	if (insn->src_reg == 0) {
2556 		u64 imm = ((u64)(insn + 1)->imm << 32) | (u32)insn->imm;
2557 
2558 		regs[insn->dst_reg].type = CONST_IMM;
2559 		regs[insn->dst_reg].imm = imm;
2560 		regs[insn->dst_reg].id = 0;
2561 		return 0;
2562 	}
2563 
2564 	/* replace_map_fd_with_map_ptr() should have caught bad ld_imm64 */
2565 	BUG_ON(insn->src_reg != BPF_PSEUDO_MAP_FD);
2566 
2567 	regs[insn->dst_reg].type = CONST_PTR_TO_MAP;
2568 	regs[insn->dst_reg].map_ptr = ld_imm64_to_map_ptr(insn);
2569 	return 0;
2570 }
2571 
2572 static bool may_access_skb(enum bpf_prog_type type)
2573 {
2574 	switch (type) {
2575 	case BPF_PROG_TYPE_SOCKET_FILTER:
2576 	case BPF_PROG_TYPE_SCHED_CLS:
2577 	case BPF_PROG_TYPE_SCHED_ACT:
2578 		return true;
2579 	default:
2580 		return false;
2581 	}
2582 }
2583 
2584 /* verify safety of LD_ABS|LD_IND instructions:
2585  * - they can only appear in the programs where ctx == skb
2586  * - since they are wrappers of function calls, they scratch R1-R5 registers,
2587  *   preserve R6-R9, and store return value into R0
2588  *
2589  * Implicit input:
2590  *   ctx == skb == R6 == CTX
2591  *
2592  * Explicit input:
2593  *   SRC == any register
2594  *   IMM == 32-bit immediate
2595  *
2596  * Output:
2597  *   R0 - 8/16/32-bit skb data converted to cpu endianness
2598  */
2599 static int check_ld_abs(struct bpf_verifier_env *env, struct bpf_insn *insn)
2600 {
2601 	struct bpf_reg_state *regs = env->cur_state.regs;
2602 	u8 mode = BPF_MODE(insn->code);
2603 	int i, err;
2604 
2605 	if (!may_access_skb(env->prog->type)) {
2606 		verbose("BPF_LD_[ABS|IND] instructions not allowed for this program type\n");
2607 		return -EINVAL;
2608 	}
2609 
2610 	if (insn->dst_reg != BPF_REG_0 || insn->off != 0 ||
2611 	    BPF_SIZE(insn->code) == BPF_DW ||
2612 	    (mode == BPF_ABS && insn->src_reg != BPF_REG_0)) {
2613 		verbose("BPF_LD_[ABS|IND] uses reserved fields\n");
2614 		return -EINVAL;
2615 	}
2616 
2617 	/* check whether implicit source operand (register R6) is readable */
2618 	err = check_reg_arg(regs, BPF_REG_6, SRC_OP);
2619 	if (err)
2620 		return err;
2621 
2622 	if (regs[BPF_REG_6].type != PTR_TO_CTX) {
2623 		verbose("at the time of BPF_LD_ABS|IND R6 != pointer to skb\n");
2624 		return -EINVAL;
2625 	}
2626 
2627 	if (mode == BPF_IND) {
2628 		/* check explicit source operand */
2629 		err = check_reg_arg(regs, insn->src_reg, SRC_OP);
2630 		if (err)
2631 			return err;
2632 	}
2633 
2634 	/* reset caller saved regs to unreadable */
2635 	for (i = 0; i < CALLER_SAVED_REGS; i++)
2636 		mark_reg_not_init(regs, caller_saved[i]);
2637 
2638 	/* mark destination R0 register as readable, since it contains
2639 	 * the value fetched from the packet
2640 	 */
2641 	regs[BPF_REG_0].type = UNKNOWN_VALUE;
2642 	return 0;
2643 }
2644 
2645 /* non-recursive DFS pseudo code
2646  * 1  procedure DFS-iterative(G,v):
2647  * 2      label v as discovered
2648  * 3      let S be a stack
2649  * 4      S.push(v)
2650  * 5      while S is not empty
2651  * 6            t <- S.pop()
2652  * 7            if t is what we're looking for:
2653  * 8                return t
2654  * 9            for all edges e in G.adjacentEdges(t) do
2655  * 10               if edge e is already labelled
2656  * 11                   continue with the next edge
2657  * 12               w <- G.adjacentVertex(t,e)
2658  * 13               if vertex w is not discovered and not explored
2659  * 14                   label e as tree-edge
2660  * 15                   label w as discovered
2661  * 16                   S.push(w)
2662  * 17                   continue at 5
2663  * 18               else if vertex w is discovered
2664  * 19                   label e as back-edge
2665  * 20               else
2666  * 21                   // vertex w is explored
2667  * 22                   label e as forward- or cross-edge
2668  * 23           label t as explored
2669  * 24           S.pop()
2670  *
2671  * convention:
2672  * 0x10 - discovered
2673  * 0x11 - discovered and fall-through edge labelled
2674  * 0x12 - discovered and fall-through and branch edges labelled
2675  * 0x20 - explored
2676  */
2677 
2678 enum {
2679 	DISCOVERED = 0x10,
2680 	EXPLORED = 0x20,
2681 	FALLTHROUGH = 1,
2682 	BRANCH = 2,
2683 };
2684 
2685 #define STATE_LIST_MARK ((struct bpf_verifier_state_list *) -1L)
2686 
2687 static int *insn_stack;	/* stack of insns to process */
2688 static int cur_stack;	/* current stack index */
2689 static int *insn_state;
2690 
2691 /* t, w, e - match pseudo-code above:
2692  * t - index of current instruction
2693  * w - next instruction
2694  * e - edge
2695  */
2696 static int push_insn(int t, int w, int e, struct bpf_verifier_env *env)
2697 {
2698 	if (e == FALLTHROUGH && insn_state[t] >= (DISCOVERED | FALLTHROUGH))
2699 		return 0;
2700 
2701 	if (e == BRANCH && insn_state[t] >= (DISCOVERED | BRANCH))
2702 		return 0;
2703 
2704 	if (w < 0 || w >= env->prog->len) {
2705 		verbose("jump out of range from insn %d to %d\n", t, w);
2706 		return -EINVAL;
2707 	}
2708 
2709 	if (e == BRANCH)
2710 		/* mark branch target for state pruning */
2711 		env->explored_states[w] = STATE_LIST_MARK;
2712 
2713 	if (insn_state[w] == 0) {
2714 		/* tree-edge */
2715 		insn_state[t] = DISCOVERED | e;
2716 		insn_state[w] = DISCOVERED;
2717 		if (cur_stack >= env->prog->len)
2718 			return -E2BIG;
2719 		insn_stack[cur_stack++] = w;
2720 		return 1;
2721 	} else if ((insn_state[w] & 0xF0) == DISCOVERED) {
2722 		verbose("back-edge from insn %d to %d\n", t, w);
2723 		return -EINVAL;
2724 	} else if (insn_state[w] == EXPLORED) {
2725 		/* forward- or cross-edge */
2726 		insn_state[t] = DISCOVERED | e;
2727 	} else {
2728 		verbose("insn state internal bug\n");
2729 		return -EFAULT;
2730 	}
2731 	return 0;
2732 }
2733 
2734 /* non-recursive depth-first-search to detect loops in BPF program
2735  * loop == back-edge in directed graph
2736  */
2737 static int check_cfg(struct bpf_verifier_env *env)
2738 {
2739 	struct bpf_insn *insns = env->prog->insnsi;
2740 	int insn_cnt = env->prog->len;
2741 	int ret = 0;
2742 	int i, t;
2743 
2744 	insn_state = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
2745 	if (!insn_state)
2746 		return -ENOMEM;
2747 
2748 	insn_stack = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
2749 	if (!insn_stack) {
2750 		kfree(insn_state);
2751 		return -ENOMEM;
2752 	}
2753 
2754 	insn_state[0] = DISCOVERED; /* mark 1st insn as discovered */
2755 	insn_stack[0] = 0; /* 0 is the first instruction */
2756 	cur_stack = 1;
2757 
2758 peek_stack:
2759 	if (cur_stack == 0)
2760 		goto check_state;
2761 	t = insn_stack[cur_stack - 1];
2762 
2763 	if (BPF_CLASS(insns[t].code) == BPF_JMP) {
2764 		u8 opcode = BPF_OP(insns[t].code);
2765 
2766 		if (opcode == BPF_EXIT) {
2767 			goto mark_explored;
2768 		} else if (opcode == BPF_CALL) {
2769 			ret = push_insn(t, t + 1, FALLTHROUGH, env);
2770 			if (ret == 1)
2771 				goto peek_stack;
2772 			else if (ret < 0)
2773 				goto err_free;
2774 			if (t + 1 < insn_cnt)
2775 				env->explored_states[t + 1] = STATE_LIST_MARK;
2776 		} else if (opcode == BPF_JA) {
2777 			if (BPF_SRC(insns[t].code) != BPF_K) {
2778 				ret = -EINVAL;
2779 				goto err_free;
2780 			}
2781 			/* unconditional jump with single edge */
2782 			ret = push_insn(t, t + insns[t].off + 1,
2783 					FALLTHROUGH, env);
2784 			if (ret == 1)
2785 				goto peek_stack;
2786 			else if (ret < 0)
2787 				goto err_free;
2788 			/* tell verifier to check for equivalent states
2789 			 * after every call and jump
2790 			 */
2791 			if (t + 1 < insn_cnt)
2792 				env->explored_states[t + 1] = STATE_LIST_MARK;
2793 		} else {
2794 			/* conditional jump with two edges */
2795 			env->explored_states[t] = STATE_LIST_MARK;
2796 			ret = push_insn(t, t + 1, FALLTHROUGH, env);
2797 			if (ret == 1)
2798 				goto peek_stack;
2799 			else if (ret < 0)
2800 				goto err_free;
2801 
2802 			ret = push_insn(t, t + insns[t].off + 1, BRANCH, env);
2803 			if (ret == 1)
2804 				goto peek_stack;
2805 			else if (ret < 0)
2806 				goto err_free;
2807 		}
2808 	} else {
2809 		/* all other non-branch instructions with single
2810 		 * fall-through edge
2811 		 */
2812 		ret = push_insn(t, t + 1, FALLTHROUGH, env);
2813 		if (ret == 1)
2814 			goto peek_stack;
2815 		else if (ret < 0)
2816 			goto err_free;
2817 	}
2818 
2819 mark_explored:
2820 	insn_state[t] = EXPLORED;
2821 	if (cur_stack-- <= 0) {
2822 		verbose("pop stack internal bug\n");
2823 		ret = -EFAULT;
2824 		goto err_free;
2825 	}
2826 	goto peek_stack;
2827 
2828 check_state:
2829 	for (i = 0; i < insn_cnt; i++) {
2830 		if (insn_state[i] != EXPLORED) {
2831 			verbose("unreachable insn %d\n", i);
2832 			ret = -EINVAL;
2833 			goto err_free;
2834 		}
2835 	}
2836 	ret = 0; /* cfg looks good */
2837 
2838 err_free:
2839 	kfree(insn_state);
2840 	kfree(insn_stack);
2841 	return ret;
2842 }
2843 
2844 /* the following conditions reduce the number of explored insns
2845  * from ~140k to ~80k for ultra large programs that use a lot of ptr_to_packet
2846  */
2847 static bool compare_ptrs_to_packet(struct bpf_verifier_env *env,
2848 				   struct bpf_reg_state *old,
2849 				   struct bpf_reg_state *cur)
2850 {
2851 	if (old->id != cur->id)
2852 		return false;
2853 
2854 	/* old ptr_to_packet is more conservative, since it allows smaller
2855 	 * range. Ex:
2856 	 * old(off=0,r=10) is equal to cur(off=0,r=20), because
2857 	 * old(off=0,r=10) means that with range=10 the verifier proceeded
2858 	 * further and found no issues with the program. Now we're in the same
2859 	 * spot with cur(off=0,r=20), so we're safe too, since anything further
2860 	 * will only be looking at most 10 bytes after this pointer.
2861 	 */
2862 	if (old->off == cur->off && old->range < cur->range)
2863 		return true;
2864 
2865 	/* old(off=20,r=10) is equal to cur(off=22,re=22 or 5 or 0)
2866 	 * since both cannot be used for packet access and safe(old)
2867 	 * pointer has smaller off that could be used for further
2868 	 * 'if (ptr > data_end)' check
2869 	 * Ex:
2870 	 * old(off=20,r=10) and cur(off=22,r=22) and cur(off=22,r=0) mean
2871 	 * that we cannot access the packet.
2872 	 * The safe range is:
2873 	 * [ptr, ptr + range - off)
2874 	 * so whenever off >=range, it means no safe bytes from this pointer.
2875 	 * When comparing old->off <= cur->off, it means that older code
2876 	 * went with smaller offset and that offset was later
2877 	 * used to figure out the safe range after 'if (ptr > data_end)' check
2878 	 * Say, 'old' state was explored like:
2879 	 * ... R3(off=0, r=0)
2880 	 * R4 = R3 + 20
2881 	 * ... now R4(off=20,r=0)  <-- here
2882 	 * if (R4 > data_end)
2883 	 * ... R4(off=20,r=20), R3(off=0,r=20) and R3 can be used to access.
2884 	 * ... the code further went all the way to bpf_exit.
2885 	 * Now the 'cur' state at the mark 'here' has R4(off=30,r=0).
2886 	 * old_R4(off=20,r=0) equal to cur_R4(off=30,r=0), since if the verifier
2887 	 * goes further, such cur_R4 will give larger safe packet range after
2888 	 * 'if (R4 > data_end)' and all further insn were already good with r=20,
2889 	 * so they will be good with r=30 and we can prune the search.
2890 	 */
2891 	if (!env->strict_alignment && old->off <= cur->off &&
2892 	    old->off >= old->range && cur->off >= cur->range)
2893 		return true;
2894 
2895 	return false;
2896 }
2897 
2898 /* compare two verifier states
2899  *
2900  * all states stored in state_list are known to be valid, since
2901  * verifier reached 'bpf_exit' instruction through them
2902  *
2903  * this function is called when verifier exploring different branches of
2904  * execution popped from the state stack. If it sees an old state that has
2905  * more strict register state and more strict stack state then this execution
2906  * branch doesn't need to be explored further, since verifier already
2907  * concluded that more strict state leads to valid finish.
2908  *
2909  * Therefore two states are equivalent if register state is more conservative
2910  * and explored stack state is more conservative than the current one.
2911  * Example:
2912  *       explored                   current
2913  * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC)
2914  * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC)
2915  *
2916  * In other words if current stack state (one being explored) has more
2917  * valid slots than old one that already passed validation, it means
2918  * the verifier can stop exploring and conclude that current state is valid too
2919  *
2920  * Similarly with registers. If explored state has register type as invalid
2921  * whereas register type in current state is meaningful, it means that
2922  * the current state will reach 'bpf_exit' instruction safely
2923  */
2924 static bool states_equal(struct bpf_verifier_env *env,
2925 			 struct bpf_verifier_state *old,
2926 			 struct bpf_verifier_state *cur)
2927 {
2928 	bool varlen_map_access = env->varlen_map_value_access;
2929 	struct bpf_reg_state *rold, *rcur;
2930 	int i;
2931 
2932 	for (i = 0; i < MAX_BPF_REG; i++) {
2933 		rold = &old->regs[i];
2934 		rcur = &cur->regs[i];
2935 
2936 		if (memcmp(rold, rcur, sizeof(*rold)) == 0)
2937 			continue;
2938 
2939 		/* If the ranges were not the same, but everything else was and
2940 		 * we didn't do a variable access into a map then we are a-ok.
2941 		 */
2942 		if (!varlen_map_access &&
2943 		    memcmp(rold, rcur, offsetofend(struct bpf_reg_state, id)) == 0)
2944 			continue;
2945 
2946 		/* If we didn't map access then again we don't care about the
2947 		 * mismatched range values and it's ok if our old type was
2948 		 * UNKNOWN and we didn't go to a NOT_INIT'ed reg.
2949 		 */
2950 		if (rold->type == NOT_INIT ||
2951 		    (!varlen_map_access && rold->type == UNKNOWN_VALUE &&
2952 		     rcur->type != NOT_INIT))
2953 			continue;
2954 
2955 		/* Don't care about the reg->id in this case. */
2956 		if (rold->type == PTR_TO_MAP_VALUE_OR_NULL &&
2957 		    rcur->type == PTR_TO_MAP_VALUE_OR_NULL &&
2958 		    rold->map_ptr == rcur->map_ptr)
2959 			continue;
2960 
2961 		if (rold->type == PTR_TO_PACKET && rcur->type == PTR_TO_PACKET &&
2962 		    compare_ptrs_to_packet(env, rold, rcur))
2963 			continue;
2964 
2965 		return false;
2966 	}
2967 
2968 	for (i = 0; i < MAX_BPF_STACK; i++) {
2969 		if (old->stack_slot_type[i] == STACK_INVALID)
2970 			continue;
2971 		if (old->stack_slot_type[i] != cur->stack_slot_type[i])
2972 			/* Ex: old explored (safe) state has STACK_SPILL in
2973 			 * this stack slot, but current has has STACK_MISC ->
2974 			 * this verifier states are not equivalent,
2975 			 * return false to continue verification of this path
2976 			 */
2977 			return false;
2978 		if (i % BPF_REG_SIZE)
2979 			continue;
2980 		if (old->stack_slot_type[i] != STACK_SPILL)
2981 			continue;
2982 		if (memcmp(&old->spilled_regs[i / BPF_REG_SIZE],
2983 			   &cur->spilled_regs[i / BPF_REG_SIZE],
2984 			   sizeof(old->spilled_regs[0])))
2985 			/* when explored and current stack slot types are
2986 			 * the same, check that stored pointers types
2987 			 * are the same as well.
2988 			 * Ex: explored safe path could have stored
2989 			 * (bpf_reg_state) {.type = PTR_TO_STACK, .imm = -8}
2990 			 * but current path has stored:
2991 			 * (bpf_reg_state) {.type = PTR_TO_STACK, .imm = -16}
2992 			 * such verifier states are not equivalent.
2993 			 * return false to continue verification of this path
2994 			 */
2995 			return false;
2996 		else
2997 			continue;
2998 	}
2999 	return true;
3000 }
3001 
3002 static int is_state_visited(struct bpf_verifier_env *env, int insn_idx)
3003 {
3004 	struct bpf_verifier_state_list *new_sl;
3005 	struct bpf_verifier_state_list *sl;
3006 
3007 	sl = env->explored_states[insn_idx];
3008 	if (!sl)
3009 		/* this 'insn_idx' instruction wasn't marked, so we will not
3010 		 * be doing state search here
3011 		 */
3012 		return 0;
3013 
3014 	while (sl != STATE_LIST_MARK) {
3015 		if (states_equal(env, &sl->state, &env->cur_state))
3016 			/* reached equivalent register/stack state,
3017 			 * prune the search
3018 			 */
3019 			return 1;
3020 		sl = sl->next;
3021 	}
3022 
3023 	/* there were no equivalent states, remember current one.
3024 	 * technically the current state is not proven to be safe yet,
3025 	 * but it will either reach bpf_exit (which means it's safe) or
3026 	 * it will be rejected. Since there are no loops, we won't be
3027 	 * seeing this 'insn_idx' instruction again on the way to bpf_exit
3028 	 */
3029 	new_sl = kmalloc(sizeof(struct bpf_verifier_state_list), GFP_USER);
3030 	if (!new_sl)
3031 		return -ENOMEM;
3032 
3033 	/* add new state to the head of linked list */
3034 	memcpy(&new_sl->state, &env->cur_state, sizeof(env->cur_state));
3035 	new_sl->next = env->explored_states[insn_idx];
3036 	env->explored_states[insn_idx] = new_sl;
3037 	return 0;
3038 }
3039 
3040 static int ext_analyzer_insn_hook(struct bpf_verifier_env *env,
3041 				  int insn_idx, int prev_insn_idx)
3042 {
3043 	if (!env->analyzer_ops || !env->analyzer_ops->insn_hook)
3044 		return 0;
3045 
3046 	return env->analyzer_ops->insn_hook(env, insn_idx, prev_insn_idx);
3047 }
3048 
3049 static int do_check(struct bpf_verifier_env *env)
3050 {
3051 	struct bpf_verifier_state *state = &env->cur_state;
3052 	struct bpf_insn *insns = env->prog->insnsi;
3053 	struct bpf_reg_state *regs = state->regs;
3054 	int insn_cnt = env->prog->len;
3055 	int insn_idx, prev_insn_idx = 0;
3056 	int insn_processed = 0;
3057 	bool do_print_state = false;
3058 
3059 	init_reg_state(regs);
3060 	insn_idx = 0;
3061 	env->varlen_map_value_access = false;
3062 	for (;;) {
3063 		struct bpf_insn *insn;
3064 		u8 class;
3065 		int err;
3066 
3067 		if (insn_idx >= insn_cnt) {
3068 			verbose("invalid insn idx %d insn_cnt %d\n",
3069 				insn_idx, insn_cnt);
3070 			return -EFAULT;
3071 		}
3072 
3073 		insn = &insns[insn_idx];
3074 		class = BPF_CLASS(insn->code);
3075 
3076 		if (++insn_processed > BPF_COMPLEXITY_LIMIT_INSNS) {
3077 			verbose("BPF program is too large. Processed %d insn\n",
3078 				insn_processed);
3079 			return -E2BIG;
3080 		}
3081 
3082 		err = is_state_visited(env, insn_idx);
3083 		if (err < 0)
3084 			return err;
3085 		if (err == 1) {
3086 			/* found equivalent state, can prune the search */
3087 			if (log_level) {
3088 				if (do_print_state)
3089 					verbose("\nfrom %d to %d: safe\n",
3090 						prev_insn_idx, insn_idx);
3091 				else
3092 					verbose("%d: safe\n", insn_idx);
3093 			}
3094 			goto process_bpf_exit;
3095 		}
3096 
3097 		if (need_resched())
3098 			cond_resched();
3099 
3100 		if (log_level > 1 || (log_level && do_print_state)) {
3101 			if (log_level > 1)
3102 				verbose("%d:", insn_idx);
3103 			else
3104 				verbose("\nfrom %d to %d:",
3105 					prev_insn_idx, insn_idx);
3106 			print_verifier_state(&env->cur_state);
3107 			do_print_state = false;
3108 		}
3109 
3110 		if (log_level) {
3111 			verbose("%d: ", insn_idx);
3112 			print_bpf_insn(env, insn);
3113 		}
3114 
3115 		err = ext_analyzer_insn_hook(env, insn_idx, prev_insn_idx);
3116 		if (err)
3117 			return err;
3118 
3119 		if (class == BPF_ALU || class == BPF_ALU64) {
3120 			err = check_alu_op(env, insn);
3121 			if (err)
3122 				return err;
3123 
3124 		} else if (class == BPF_LDX) {
3125 			enum bpf_reg_type *prev_src_type, src_reg_type;
3126 
3127 			/* check for reserved fields is already done */
3128 
3129 			/* check src operand */
3130 			err = check_reg_arg(regs, insn->src_reg, SRC_OP);
3131 			if (err)
3132 				return err;
3133 
3134 			err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK);
3135 			if (err)
3136 				return err;
3137 
3138 			src_reg_type = regs[insn->src_reg].type;
3139 
3140 			/* check that memory (src_reg + off) is readable,
3141 			 * the state of dst_reg will be updated by this func
3142 			 */
3143 			err = check_mem_access(env, insn_idx, insn->src_reg, insn->off,
3144 					       BPF_SIZE(insn->code), BPF_READ,
3145 					       insn->dst_reg);
3146 			if (err)
3147 				return err;
3148 
3149 			prev_src_type = &env->insn_aux_data[insn_idx].ptr_type;
3150 
3151 			if (*prev_src_type == NOT_INIT) {
3152 				/* saw a valid insn
3153 				 * dst_reg = *(u32 *)(src_reg + off)
3154 				 * save type to validate intersecting paths
3155 				 */
3156 				*prev_src_type = src_reg_type;
3157 
3158 			} else if (src_reg_type != *prev_src_type &&
3159 				   (src_reg_type == PTR_TO_CTX ||
3160 				    *prev_src_type == PTR_TO_CTX)) {
3161 				/* ABuser program is trying to use the same insn
3162 				 * dst_reg = *(u32*) (src_reg + off)
3163 				 * with different pointer types:
3164 				 * src_reg == ctx in one branch and
3165 				 * src_reg == stack|map in some other branch.
3166 				 * Reject it.
3167 				 */
3168 				verbose("same insn cannot be used with different pointers\n");
3169 				return -EINVAL;
3170 			}
3171 
3172 		} else if (class == BPF_STX) {
3173 			enum bpf_reg_type *prev_dst_type, dst_reg_type;
3174 
3175 			if (BPF_MODE(insn->code) == BPF_XADD) {
3176 				err = check_xadd(env, insn_idx, insn);
3177 				if (err)
3178 					return err;
3179 				insn_idx++;
3180 				continue;
3181 			}
3182 
3183 			/* check src1 operand */
3184 			err = check_reg_arg(regs, insn->src_reg, SRC_OP);
3185 			if (err)
3186 				return err;
3187 			/* check src2 operand */
3188 			err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
3189 			if (err)
3190 				return err;
3191 
3192 			dst_reg_type = regs[insn->dst_reg].type;
3193 
3194 			/* check that memory (dst_reg + off) is writeable */
3195 			err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off,
3196 					       BPF_SIZE(insn->code), BPF_WRITE,
3197 					       insn->src_reg);
3198 			if (err)
3199 				return err;
3200 
3201 			prev_dst_type = &env->insn_aux_data[insn_idx].ptr_type;
3202 
3203 			if (*prev_dst_type == NOT_INIT) {
3204 				*prev_dst_type = dst_reg_type;
3205 			} else if (dst_reg_type != *prev_dst_type &&
3206 				   (dst_reg_type == PTR_TO_CTX ||
3207 				    *prev_dst_type == PTR_TO_CTX)) {
3208 				verbose("same insn cannot be used with different pointers\n");
3209 				return -EINVAL;
3210 			}
3211 
3212 		} else if (class == BPF_ST) {
3213 			if (BPF_MODE(insn->code) != BPF_MEM ||
3214 			    insn->src_reg != BPF_REG_0) {
3215 				verbose("BPF_ST uses reserved fields\n");
3216 				return -EINVAL;
3217 			}
3218 			/* check src operand */
3219 			err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
3220 			if (err)
3221 				return err;
3222 
3223 			/* check that memory (dst_reg + off) is writeable */
3224 			err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off,
3225 					       BPF_SIZE(insn->code), BPF_WRITE,
3226 					       -1);
3227 			if (err)
3228 				return err;
3229 
3230 		} else if (class == BPF_JMP) {
3231 			u8 opcode = BPF_OP(insn->code);
3232 
3233 			if (opcode == BPF_CALL) {
3234 				if (BPF_SRC(insn->code) != BPF_K ||
3235 				    insn->off != 0 ||
3236 				    insn->src_reg != BPF_REG_0 ||
3237 				    insn->dst_reg != BPF_REG_0) {
3238 					verbose("BPF_CALL uses reserved fields\n");
3239 					return -EINVAL;
3240 				}
3241 
3242 				err = check_call(env, insn->imm, insn_idx);
3243 				if (err)
3244 					return err;
3245 
3246 			} else if (opcode == BPF_JA) {
3247 				if (BPF_SRC(insn->code) != BPF_K ||
3248 				    insn->imm != 0 ||
3249 				    insn->src_reg != BPF_REG_0 ||
3250 				    insn->dst_reg != BPF_REG_0) {
3251 					verbose("BPF_JA uses reserved fields\n");
3252 					return -EINVAL;
3253 				}
3254 
3255 				insn_idx += insn->off + 1;
3256 				continue;
3257 
3258 			} else if (opcode == BPF_EXIT) {
3259 				if (BPF_SRC(insn->code) != BPF_K ||
3260 				    insn->imm != 0 ||
3261 				    insn->src_reg != BPF_REG_0 ||
3262 				    insn->dst_reg != BPF_REG_0) {
3263 					verbose("BPF_EXIT uses reserved fields\n");
3264 					return -EINVAL;
3265 				}
3266 
3267 				/* eBPF calling convetion is such that R0 is used
3268 				 * to return the value from eBPF program.
3269 				 * Make sure that it's readable at this time
3270 				 * of bpf_exit, which means that program wrote
3271 				 * something into it earlier
3272 				 */
3273 				err = check_reg_arg(regs, BPF_REG_0, SRC_OP);
3274 				if (err)
3275 					return err;
3276 
3277 				if (is_pointer_value(env, BPF_REG_0)) {
3278 					verbose("R0 leaks addr as return value\n");
3279 					return -EACCES;
3280 				}
3281 
3282 process_bpf_exit:
3283 				insn_idx = pop_stack(env, &prev_insn_idx);
3284 				if (insn_idx < 0) {
3285 					break;
3286 				} else {
3287 					do_print_state = true;
3288 					continue;
3289 				}
3290 			} else {
3291 				err = check_cond_jmp_op(env, insn, &insn_idx);
3292 				if (err)
3293 					return err;
3294 			}
3295 		} else if (class == BPF_LD) {
3296 			u8 mode = BPF_MODE(insn->code);
3297 
3298 			if (mode == BPF_ABS || mode == BPF_IND) {
3299 				err = check_ld_abs(env, insn);
3300 				if (err)
3301 					return err;
3302 
3303 			} else if (mode == BPF_IMM) {
3304 				err = check_ld_imm(env, insn);
3305 				if (err)
3306 					return err;
3307 
3308 				insn_idx++;
3309 			} else {
3310 				verbose("invalid BPF_LD mode\n");
3311 				return -EINVAL;
3312 			}
3313 			reset_reg_range_values(regs, insn->dst_reg);
3314 		} else {
3315 			verbose("unknown insn class %d\n", class);
3316 			return -EINVAL;
3317 		}
3318 
3319 		insn_idx++;
3320 	}
3321 
3322 	verbose("processed %d insns, stack depth %d\n",
3323 		insn_processed, env->prog->aux->stack_depth);
3324 	return 0;
3325 }
3326 
3327 static int check_map_prealloc(struct bpf_map *map)
3328 {
3329 	return (map->map_type != BPF_MAP_TYPE_HASH &&
3330 		map->map_type != BPF_MAP_TYPE_PERCPU_HASH &&
3331 		map->map_type != BPF_MAP_TYPE_HASH_OF_MAPS) ||
3332 		!(map->map_flags & BPF_F_NO_PREALLOC);
3333 }
3334 
3335 static int check_map_prog_compatibility(struct bpf_map *map,
3336 					struct bpf_prog *prog)
3337 
3338 {
3339 	/* Make sure that BPF_PROG_TYPE_PERF_EVENT programs only use
3340 	 * preallocated hash maps, since doing memory allocation
3341 	 * in overflow_handler can crash depending on where nmi got
3342 	 * triggered.
3343 	 */
3344 	if (prog->type == BPF_PROG_TYPE_PERF_EVENT) {
3345 		if (!check_map_prealloc(map)) {
3346 			verbose("perf_event programs can only use preallocated hash map\n");
3347 			return -EINVAL;
3348 		}
3349 		if (map->inner_map_meta &&
3350 		    !check_map_prealloc(map->inner_map_meta)) {
3351 			verbose("perf_event programs can only use preallocated inner hash map\n");
3352 			return -EINVAL;
3353 		}
3354 	}
3355 	return 0;
3356 }
3357 
3358 /* look for pseudo eBPF instructions that access map FDs and
3359  * replace them with actual map pointers
3360  */
3361 static int replace_map_fd_with_map_ptr(struct bpf_verifier_env *env)
3362 {
3363 	struct bpf_insn *insn = env->prog->insnsi;
3364 	int insn_cnt = env->prog->len;
3365 	int i, j, err;
3366 
3367 	err = bpf_prog_calc_tag(env->prog);
3368 	if (err)
3369 		return err;
3370 
3371 	for (i = 0; i < insn_cnt; i++, insn++) {
3372 		if (BPF_CLASS(insn->code) == BPF_LDX &&
3373 		    (BPF_MODE(insn->code) != BPF_MEM || insn->imm != 0)) {
3374 			verbose("BPF_LDX uses reserved fields\n");
3375 			return -EINVAL;
3376 		}
3377 
3378 		if (BPF_CLASS(insn->code) == BPF_STX &&
3379 		    ((BPF_MODE(insn->code) != BPF_MEM &&
3380 		      BPF_MODE(insn->code) != BPF_XADD) || insn->imm != 0)) {
3381 			verbose("BPF_STX uses reserved fields\n");
3382 			return -EINVAL;
3383 		}
3384 
3385 		if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW)) {
3386 			struct bpf_map *map;
3387 			struct fd f;
3388 
3389 			if (i == insn_cnt - 1 || insn[1].code != 0 ||
3390 			    insn[1].dst_reg != 0 || insn[1].src_reg != 0 ||
3391 			    insn[1].off != 0) {
3392 				verbose("invalid bpf_ld_imm64 insn\n");
3393 				return -EINVAL;
3394 			}
3395 
3396 			if (insn->src_reg == 0)
3397 				/* valid generic load 64-bit imm */
3398 				goto next_insn;
3399 
3400 			if (insn->src_reg != BPF_PSEUDO_MAP_FD) {
3401 				verbose("unrecognized bpf_ld_imm64 insn\n");
3402 				return -EINVAL;
3403 			}
3404 
3405 			f = fdget(insn->imm);
3406 			map = __bpf_map_get(f);
3407 			if (IS_ERR(map)) {
3408 				verbose("fd %d is not pointing to valid bpf_map\n",
3409 					insn->imm);
3410 				return PTR_ERR(map);
3411 			}
3412 
3413 			err = check_map_prog_compatibility(map, env->prog);
3414 			if (err) {
3415 				fdput(f);
3416 				return err;
3417 			}
3418 
3419 			/* store map pointer inside BPF_LD_IMM64 instruction */
3420 			insn[0].imm = (u32) (unsigned long) map;
3421 			insn[1].imm = ((u64) (unsigned long) map) >> 32;
3422 
3423 			/* check whether we recorded this map already */
3424 			for (j = 0; j < env->used_map_cnt; j++)
3425 				if (env->used_maps[j] == map) {
3426 					fdput(f);
3427 					goto next_insn;
3428 				}
3429 
3430 			if (env->used_map_cnt >= MAX_USED_MAPS) {
3431 				fdput(f);
3432 				return -E2BIG;
3433 			}
3434 
3435 			/* hold the map. If the program is rejected by verifier,
3436 			 * the map will be released by release_maps() or it
3437 			 * will be used by the valid program until it's unloaded
3438 			 * and all maps are released in free_bpf_prog_info()
3439 			 */
3440 			map = bpf_map_inc(map, false);
3441 			if (IS_ERR(map)) {
3442 				fdput(f);
3443 				return PTR_ERR(map);
3444 			}
3445 			env->used_maps[env->used_map_cnt++] = map;
3446 
3447 			fdput(f);
3448 next_insn:
3449 			insn++;
3450 			i++;
3451 		}
3452 	}
3453 
3454 	/* now all pseudo BPF_LD_IMM64 instructions load valid
3455 	 * 'struct bpf_map *' into a register instead of user map_fd.
3456 	 * These pointers will be used later by verifier to validate map access.
3457 	 */
3458 	return 0;
3459 }
3460 
3461 /* drop refcnt of maps used by the rejected program */
3462 static void release_maps(struct bpf_verifier_env *env)
3463 {
3464 	int i;
3465 
3466 	for (i = 0; i < env->used_map_cnt; i++)
3467 		bpf_map_put(env->used_maps[i]);
3468 }
3469 
3470 /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */
3471 static void convert_pseudo_ld_imm64(struct bpf_verifier_env *env)
3472 {
3473 	struct bpf_insn *insn = env->prog->insnsi;
3474 	int insn_cnt = env->prog->len;
3475 	int i;
3476 
3477 	for (i = 0; i < insn_cnt; i++, insn++)
3478 		if (insn->code == (BPF_LD | BPF_IMM | BPF_DW))
3479 			insn->src_reg = 0;
3480 }
3481 
3482 /* single env->prog->insni[off] instruction was replaced with the range
3483  * insni[off, off + cnt).  Adjust corresponding insn_aux_data by copying
3484  * [0, off) and [off, end) to new locations, so the patched range stays zero
3485  */
3486 static int adjust_insn_aux_data(struct bpf_verifier_env *env, u32 prog_len,
3487 				u32 off, u32 cnt)
3488 {
3489 	struct bpf_insn_aux_data *new_data, *old_data = env->insn_aux_data;
3490 
3491 	if (cnt == 1)
3492 		return 0;
3493 	new_data = vzalloc(sizeof(struct bpf_insn_aux_data) * prog_len);
3494 	if (!new_data)
3495 		return -ENOMEM;
3496 	memcpy(new_data, old_data, sizeof(struct bpf_insn_aux_data) * off);
3497 	memcpy(new_data + off + cnt - 1, old_data + off,
3498 	       sizeof(struct bpf_insn_aux_data) * (prog_len - off - cnt + 1));
3499 	env->insn_aux_data = new_data;
3500 	vfree(old_data);
3501 	return 0;
3502 }
3503 
3504 static struct bpf_prog *bpf_patch_insn_data(struct bpf_verifier_env *env, u32 off,
3505 					    const struct bpf_insn *patch, u32 len)
3506 {
3507 	struct bpf_prog *new_prog;
3508 
3509 	new_prog = bpf_patch_insn_single(env->prog, off, patch, len);
3510 	if (!new_prog)
3511 		return NULL;
3512 	if (adjust_insn_aux_data(env, new_prog->len, off, len))
3513 		return NULL;
3514 	return new_prog;
3515 }
3516 
3517 /* convert load instructions that access fields of 'struct __sk_buff'
3518  * into sequence of instructions that access fields of 'struct sk_buff'
3519  */
3520 static int convert_ctx_accesses(struct bpf_verifier_env *env)
3521 {
3522 	const struct bpf_verifier_ops *ops = env->prog->aux->ops;
3523 	int i, cnt, size, ctx_field_size, delta = 0;
3524 	const int insn_cnt = env->prog->len;
3525 	struct bpf_insn insn_buf[16], *insn;
3526 	struct bpf_prog *new_prog;
3527 	enum bpf_access_type type;
3528 	bool is_narrower_load;
3529 	u32 target_size;
3530 
3531 	if (ops->gen_prologue) {
3532 		cnt = ops->gen_prologue(insn_buf, env->seen_direct_write,
3533 					env->prog);
3534 		if (cnt >= ARRAY_SIZE(insn_buf)) {
3535 			verbose("bpf verifier is misconfigured\n");
3536 			return -EINVAL;
3537 		} else if (cnt) {
3538 			new_prog = bpf_patch_insn_data(env, 0, insn_buf, cnt);
3539 			if (!new_prog)
3540 				return -ENOMEM;
3541 
3542 			env->prog = new_prog;
3543 			delta += cnt - 1;
3544 		}
3545 	}
3546 
3547 	if (!ops->convert_ctx_access)
3548 		return 0;
3549 
3550 	insn = env->prog->insnsi + delta;
3551 
3552 	for (i = 0; i < insn_cnt; i++, insn++) {
3553 		if (insn->code == (BPF_LDX | BPF_MEM | BPF_B) ||
3554 		    insn->code == (BPF_LDX | BPF_MEM | BPF_H) ||
3555 		    insn->code == (BPF_LDX | BPF_MEM | BPF_W) ||
3556 		    insn->code == (BPF_LDX | BPF_MEM | BPF_DW))
3557 			type = BPF_READ;
3558 		else if (insn->code == (BPF_STX | BPF_MEM | BPF_B) ||
3559 			 insn->code == (BPF_STX | BPF_MEM | BPF_H) ||
3560 			 insn->code == (BPF_STX | BPF_MEM | BPF_W) ||
3561 			 insn->code == (BPF_STX | BPF_MEM | BPF_DW))
3562 			type = BPF_WRITE;
3563 		else
3564 			continue;
3565 
3566 		if (env->insn_aux_data[i + delta].ptr_type != PTR_TO_CTX)
3567 			continue;
3568 
3569 		ctx_field_size = env->insn_aux_data[i + delta].ctx_field_size;
3570 		size = BPF_LDST_BYTES(insn);
3571 
3572 		/* If the read access is a narrower load of the field,
3573 		 * convert to a 4/8-byte load, to minimum program type specific
3574 		 * convert_ctx_access changes. If conversion is successful,
3575 		 * we will apply proper mask to the result.
3576 		 */
3577 		is_narrower_load = size < ctx_field_size;
3578 		if (is_narrower_load) {
3579 			u32 off = insn->off;
3580 			u8 size_code;
3581 
3582 			if (type == BPF_WRITE) {
3583 				verbose("bpf verifier narrow ctx access misconfigured\n");
3584 				return -EINVAL;
3585 			}
3586 
3587 			size_code = BPF_H;
3588 			if (ctx_field_size == 4)
3589 				size_code = BPF_W;
3590 			else if (ctx_field_size == 8)
3591 				size_code = BPF_DW;
3592 
3593 			insn->off = off & ~(ctx_field_size - 1);
3594 			insn->code = BPF_LDX | BPF_MEM | size_code;
3595 		}
3596 
3597 		target_size = 0;
3598 		cnt = ops->convert_ctx_access(type, insn, insn_buf, env->prog,
3599 					      &target_size);
3600 		if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf) ||
3601 		    (ctx_field_size && !target_size)) {
3602 			verbose("bpf verifier is misconfigured\n");
3603 			return -EINVAL;
3604 		}
3605 
3606 		if (is_narrower_load && size < target_size) {
3607 			if (ctx_field_size <= 4)
3608 				insn_buf[cnt++] = BPF_ALU32_IMM(BPF_AND, insn->dst_reg,
3609 								(1 << size * 8) - 1);
3610 			else
3611 				insn_buf[cnt++] = BPF_ALU64_IMM(BPF_AND, insn->dst_reg,
3612 								(1 << size * 8) - 1);
3613 		}
3614 
3615 		new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt);
3616 		if (!new_prog)
3617 			return -ENOMEM;
3618 
3619 		delta += cnt - 1;
3620 
3621 		/* keep walking new program and skip insns we just inserted */
3622 		env->prog = new_prog;
3623 		insn      = new_prog->insnsi + i + delta;
3624 	}
3625 
3626 	return 0;
3627 }
3628 
3629 /* fixup insn->imm field of bpf_call instructions
3630  * and inline eligible helpers as explicit sequence of BPF instructions
3631  *
3632  * this function is called after eBPF program passed verification
3633  */
3634 static int fixup_bpf_calls(struct bpf_verifier_env *env)
3635 {
3636 	struct bpf_prog *prog = env->prog;
3637 	struct bpf_insn *insn = prog->insnsi;
3638 	const struct bpf_func_proto *fn;
3639 	const int insn_cnt = prog->len;
3640 	struct bpf_insn insn_buf[16];
3641 	struct bpf_prog *new_prog;
3642 	struct bpf_map *map_ptr;
3643 	int i, cnt, delta = 0;
3644 
3645 	for (i = 0; i < insn_cnt; i++, insn++) {
3646 		if (insn->code != (BPF_JMP | BPF_CALL))
3647 			continue;
3648 
3649 		if (insn->imm == BPF_FUNC_get_route_realm)
3650 			prog->dst_needed = 1;
3651 		if (insn->imm == BPF_FUNC_get_prandom_u32)
3652 			bpf_user_rnd_init_once();
3653 		if (insn->imm == BPF_FUNC_tail_call) {
3654 			/* If we tail call into other programs, we
3655 			 * cannot make any assumptions since they can
3656 			 * be replaced dynamically during runtime in
3657 			 * the program array.
3658 			 */
3659 			prog->cb_access = 1;
3660 			env->prog->aux->stack_depth = MAX_BPF_STACK;
3661 
3662 			/* mark bpf_tail_call as different opcode to avoid
3663 			 * conditional branch in the interpeter for every normal
3664 			 * call and to prevent accidental JITing by JIT compiler
3665 			 * that doesn't support bpf_tail_call yet
3666 			 */
3667 			insn->imm = 0;
3668 			insn->code = BPF_JMP | BPF_TAIL_CALL;
3669 			continue;
3670 		}
3671 
3672 		if (ebpf_jit_enabled() && insn->imm == BPF_FUNC_map_lookup_elem) {
3673 			map_ptr = env->insn_aux_data[i + delta].map_ptr;
3674 			if (map_ptr == BPF_MAP_PTR_POISON ||
3675 			    !map_ptr->ops->map_gen_lookup)
3676 				goto patch_call_imm;
3677 
3678 			cnt = map_ptr->ops->map_gen_lookup(map_ptr, insn_buf);
3679 			if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) {
3680 				verbose("bpf verifier is misconfigured\n");
3681 				return -EINVAL;
3682 			}
3683 
3684 			new_prog = bpf_patch_insn_data(env, i + delta, insn_buf,
3685 						       cnt);
3686 			if (!new_prog)
3687 				return -ENOMEM;
3688 
3689 			delta += cnt - 1;
3690 
3691 			/* keep walking new program and skip insns we just inserted */
3692 			env->prog = prog = new_prog;
3693 			insn      = new_prog->insnsi + i + delta;
3694 			continue;
3695 		}
3696 
3697 patch_call_imm:
3698 		fn = prog->aux->ops->get_func_proto(insn->imm);
3699 		/* all functions that have prototype and verifier allowed
3700 		 * programs to call them, must be real in-kernel functions
3701 		 */
3702 		if (!fn->func) {
3703 			verbose("kernel subsystem misconfigured func %s#%d\n",
3704 				func_id_name(insn->imm), insn->imm);
3705 			return -EFAULT;
3706 		}
3707 		insn->imm = fn->func - __bpf_call_base;
3708 	}
3709 
3710 	return 0;
3711 }
3712 
3713 static void free_states(struct bpf_verifier_env *env)
3714 {
3715 	struct bpf_verifier_state_list *sl, *sln;
3716 	int i;
3717 
3718 	if (!env->explored_states)
3719 		return;
3720 
3721 	for (i = 0; i < env->prog->len; i++) {
3722 		sl = env->explored_states[i];
3723 
3724 		if (sl)
3725 			while (sl != STATE_LIST_MARK) {
3726 				sln = sl->next;
3727 				kfree(sl);
3728 				sl = sln;
3729 			}
3730 	}
3731 
3732 	kfree(env->explored_states);
3733 }
3734 
3735 int bpf_check(struct bpf_prog **prog, union bpf_attr *attr)
3736 {
3737 	char __user *log_ubuf = NULL;
3738 	struct bpf_verifier_env *env;
3739 	int ret = -EINVAL;
3740 
3741 	/* 'struct bpf_verifier_env' can be global, but since it's not small,
3742 	 * allocate/free it every time bpf_check() is called
3743 	 */
3744 	env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL);
3745 	if (!env)
3746 		return -ENOMEM;
3747 
3748 	env->insn_aux_data = vzalloc(sizeof(struct bpf_insn_aux_data) *
3749 				     (*prog)->len);
3750 	ret = -ENOMEM;
3751 	if (!env->insn_aux_data)
3752 		goto err_free_env;
3753 	env->prog = *prog;
3754 
3755 	/* grab the mutex to protect few globals used by verifier */
3756 	mutex_lock(&bpf_verifier_lock);
3757 
3758 	if (attr->log_level || attr->log_buf || attr->log_size) {
3759 		/* user requested verbose verifier output
3760 		 * and supplied buffer to store the verification trace
3761 		 */
3762 		log_level = attr->log_level;
3763 		log_ubuf = (char __user *) (unsigned long) attr->log_buf;
3764 		log_size = attr->log_size;
3765 		log_len = 0;
3766 
3767 		ret = -EINVAL;
3768 		/* log_* values have to be sane */
3769 		if (log_size < 128 || log_size > UINT_MAX >> 8 ||
3770 		    log_level == 0 || log_ubuf == NULL)
3771 			goto err_unlock;
3772 
3773 		ret = -ENOMEM;
3774 		log_buf = vmalloc(log_size);
3775 		if (!log_buf)
3776 			goto err_unlock;
3777 	} else {
3778 		log_level = 0;
3779 	}
3780 
3781 	env->strict_alignment = !!(attr->prog_flags & BPF_F_STRICT_ALIGNMENT);
3782 	if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS))
3783 		env->strict_alignment = true;
3784 
3785 	ret = replace_map_fd_with_map_ptr(env);
3786 	if (ret < 0)
3787 		goto skip_full_check;
3788 
3789 	env->explored_states = kcalloc(env->prog->len,
3790 				       sizeof(struct bpf_verifier_state_list *),
3791 				       GFP_USER);
3792 	ret = -ENOMEM;
3793 	if (!env->explored_states)
3794 		goto skip_full_check;
3795 
3796 	ret = check_cfg(env);
3797 	if (ret < 0)
3798 		goto skip_full_check;
3799 
3800 	env->allow_ptr_leaks = capable(CAP_SYS_ADMIN);
3801 
3802 	ret = do_check(env);
3803 
3804 skip_full_check:
3805 	while (pop_stack(env, NULL) >= 0);
3806 	free_states(env);
3807 
3808 	if (ret == 0)
3809 		/* program is valid, convert *(u32*)(ctx + off) accesses */
3810 		ret = convert_ctx_accesses(env);
3811 
3812 	if (ret == 0)
3813 		ret = fixup_bpf_calls(env);
3814 
3815 	if (log_level && log_len >= log_size - 1) {
3816 		BUG_ON(log_len >= log_size);
3817 		/* verifier log exceeded user supplied buffer */
3818 		ret = -ENOSPC;
3819 		/* fall through to return what was recorded */
3820 	}
3821 
3822 	/* copy verifier log back to user space including trailing zero */
3823 	if (log_level && copy_to_user(log_ubuf, log_buf, log_len + 1) != 0) {
3824 		ret = -EFAULT;
3825 		goto free_log_buf;
3826 	}
3827 
3828 	if (ret == 0 && env->used_map_cnt) {
3829 		/* if program passed verifier, update used_maps in bpf_prog_info */
3830 		env->prog->aux->used_maps = kmalloc_array(env->used_map_cnt,
3831 							  sizeof(env->used_maps[0]),
3832 							  GFP_KERNEL);
3833 
3834 		if (!env->prog->aux->used_maps) {
3835 			ret = -ENOMEM;
3836 			goto free_log_buf;
3837 		}
3838 
3839 		memcpy(env->prog->aux->used_maps, env->used_maps,
3840 		       sizeof(env->used_maps[0]) * env->used_map_cnt);
3841 		env->prog->aux->used_map_cnt = env->used_map_cnt;
3842 
3843 		/* program is valid. Convert pseudo bpf_ld_imm64 into generic
3844 		 * bpf_ld_imm64 instructions
3845 		 */
3846 		convert_pseudo_ld_imm64(env);
3847 	}
3848 
3849 free_log_buf:
3850 	if (log_level)
3851 		vfree(log_buf);
3852 	if (!env->prog->aux->used_maps)
3853 		/* if we didn't copy map pointers into bpf_prog_info, release
3854 		 * them now. Otherwise free_bpf_prog_info() will release them.
3855 		 */
3856 		release_maps(env);
3857 	*prog = env->prog;
3858 err_unlock:
3859 	mutex_unlock(&bpf_verifier_lock);
3860 	vfree(env->insn_aux_data);
3861 err_free_env:
3862 	kfree(env);
3863 	return ret;
3864 }
3865 
3866 int bpf_analyzer(struct bpf_prog *prog, const struct bpf_ext_analyzer_ops *ops,
3867 		 void *priv)
3868 {
3869 	struct bpf_verifier_env *env;
3870 	int ret;
3871 
3872 	env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL);
3873 	if (!env)
3874 		return -ENOMEM;
3875 
3876 	env->insn_aux_data = vzalloc(sizeof(struct bpf_insn_aux_data) *
3877 				     prog->len);
3878 	ret = -ENOMEM;
3879 	if (!env->insn_aux_data)
3880 		goto err_free_env;
3881 	env->prog = prog;
3882 	env->analyzer_ops = ops;
3883 	env->analyzer_priv = priv;
3884 
3885 	/* grab the mutex to protect few globals used by verifier */
3886 	mutex_lock(&bpf_verifier_lock);
3887 
3888 	log_level = 0;
3889 
3890 	env->strict_alignment = false;
3891 	if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS))
3892 		env->strict_alignment = true;
3893 
3894 	env->explored_states = kcalloc(env->prog->len,
3895 				       sizeof(struct bpf_verifier_state_list *),
3896 				       GFP_KERNEL);
3897 	ret = -ENOMEM;
3898 	if (!env->explored_states)
3899 		goto skip_full_check;
3900 
3901 	ret = check_cfg(env);
3902 	if (ret < 0)
3903 		goto skip_full_check;
3904 
3905 	env->allow_ptr_leaks = capable(CAP_SYS_ADMIN);
3906 
3907 	ret = do_check(env);
3908 
3909 skip_full_check:
3910 	while (pop_stack(env, NULL) >= 0);
3911 	free_states(env);
3912 
3913 	mutex_unlock(&bpf_verifier_lock);
3914 	vfree(env->insn_aux_data);
3915 err_free_env:
3916 	kfree(env);
3917 	return ret;
3918 }
3919 EXPORT_SYMBOL_GPL(bpf_analyzer);
3920