xref: /openbmc/linux/kernel/bpf/verifier.c (revision bc5aa3a0)
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/filter.h>
18 #include <net/netlink.h>
19 #include <linux/file.h>
20 #include <linux/vmalloc.h>
21 
22 /* bpf_check() is a static code analyzer that walks eBPF program
23  * instruction by instruction and updates register/stack state.
24  * All paths of conditional branches are analyzed until 'bpf_exit' insn.
25  *
26  * The first pass is depth-first-search to check that the program is a DAG.
27  * It rejects the following programs:
28  * - larger than BPF_MAXINSNS insns
29  * - if loop is present (detected via back-edge)
30  * - unreachable insns exist (shouldn't be a forest. program = one function)
31  * - out of bounds or malformed jumps
32  * The second pass is all possible path descent from the 1st insn.
33  * Since it's analyzing all pathes through the program, the length of the
34  * analysis is limited to 32k insn, which may be hit even if total number of
35  * insn is less then 4K, but there are too many branches that change stack/regs.
36  * Number of 'branches to be analyzed' is limited to 1k
37  *
38  * On entry to each instruction, each register has a type, and the instruction
39  * changes the types of the registers depending on instruction semantics.
40  * If instruction is BPF_MOV64_REG(BPF_REG_1, BPF_REG_5), then type of R5 is
41  * copied to R1.
42  *
43  * All registers are 64-bit.
44  * R0 - return register
45  * R1-R5 argument passing registers
46  * R6-R9 callee saved registers
47  * R10 - frame pointer read-only
48  *
49  * At the start of BPF program the register R1 contains a pointer to bpf_context
50  * and has type PTR_TO_CTX.
51  *
52  * Verifier tracks arithmetic operations on pointers in case:
53  *    BPF_MOV64_REG(BPF_REG_1, BPF_REG_10),
54  *    BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -20),
55  * 1st insn copies R10 (which has FRAME_PTR) type into R1
56  * and 2nd arithmetic instruction is pattern matched to recognize
57  * that it wants to construct a pointer to some element within stack.
58  * So after 2nd insn, the register R1 has type PTR_TO_STACK
59  * (and -20 constant is saved for further stack bounds checking).
60  * Meaning that this reg is a pointer to stack plus known immediate constant.
61  *
62  * Most of the time the registers have UNKNOWN_VALUE type, which
63  * means the register has some value, but it's not a valid pointer.
64  * (like pointer plus pointer becomes UNKNOWN_VALUE type)
65  *
66  * When verifier sees load or store instructions the type of base register
67  * can be: PTR_TO_MAP_VALUE, PTR_TO_CTX, FRAME_PTR. These are three pointer
68  * types recognized by check_mem_access() function.
69  *
70  * PTR_TO_MAP_VALUE means that this register is pointing to 'map element value'
71  * and the range of [ptr, ptr + map's value_size) is accessible.
72  *
73  * registers used to pass values to function calls are checked against
74  * function argument constraints.
75  *
76  * ARG_PTR_TO_MAP_KEY is one of such argument constraints.
77  * It means that the register type passed to this function must be
78  * PTR_TO_STACK and it will be used inside the function as
79  * 'pointer to map element key'
80  *
81  * For example the argument constraints for bpf_map_lookup_elem():
82  *   .ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL,
83  *   .arg1_type = ARG_CONST_MAP_PTR,
84  *   .arg2_type = ARG_PTR_TO_MAP_KEY,
85  *
86  * ret_type says that this function returns 'pointer to map elem value or null'
87  * function expects 1st argument to be a const pointer to 'struct bpf_map' and
88  * 2nd argument should be a pointer to stack, which will be used inside
89  * the helper function as a pointer to map element key.
90  *
91  * On the kernel side the helper function looks like:
92  * u64 bpf_map_lookup_elem(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
93  * {
94  *    struct bpf_map *map = (struct bpf_map *) (unsigned long) r1;
95  *    void *key = (void *) (unsigned long) r2;
96  *    void *value;
97  *
98  *    here kernel can access 'key' and 'map' pointers safely, knowing that
99  *    [key, key + map->key_size) bytes are valid and were initialized on
100  *    the stack of eBPF program.
101  * }
102  *
103  * Corresponding eBPF program may look like:
104  *    BPF_MOV64_REG(BPF_REG_2, BPF_REG_10),  // after this insn R2 type is FRAME_PTR
105  *    BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -4), // after this insn R2 type is PTR_TO_STACK
106  *    BPF_LD_MAP_FD(BPF_REG_1, map_fd),      // after this insn R1 type is CONST_PTR_TO_MAP
107  *    BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem),
108  * here verifier looks at prototype of map_lookup_elem() and sees:
109  * .arg1_type == ARG_CONST_MAP_PTR and R1->type == CONST_PTR_TO_MAP, which is ok,
110  * Now verifier knows that this map has key of R1->map_ptr->key_size bytes
111  *
112  * Then .arg2_type == ARG_PTR_TO_MAP_KEY and R2->type == PTR_TO_STACK, ok so far,
113  * Now verifier checks that [R2, R2 + map's key_size) are within stack limits
114  * and were initialized prior to this call.
115  * If it's ok, then verifier allows this BPF_CALL insn and looks at
116  * .ret_type which is RET_PTR_TO_MAP_VALUE_OR_NULL, so it sets
117  * R0->type = PTR_TO_MAP_VALUE_OR_NULL which means bpf_map_lookup_elem() function
118  * returns ether pointer to map value or NULL.
119  *
120  * When type PTR_TO_MAP_VALUE_OR_NULL passes through 'if (reg != 0) goto +off'
121  * insn, the register holding that pointer in the true branch changes state to
122  * PTR_TO_MAP_VALUE and the same register changes state to CONST_IMM in the false
123  * branch. See check_cond_jmp_op().
124  *
125  * After the call R0 is set to return type of the function and registers R1-R5
126  * are set to NOT_INIT to indicate that they are no longer readable.
127  */
128 
129 struct reg_state {
130 	enum bpf_reg_type type;
131 	union {
132 		/* valid when type == CONST_IMM | PTR_TO_STACK | UNKNOWN_VALUE */
133 		s64 imm;
134 
135 		/* valid when type == PTR_TO_PACKET* */
136 		struct {
137 			u32 id;
138 			u16 off;
139 			u16 range;
140 		};
141 
142 		/* valid when type == CONST_PTR_TO_MAP | PTR_TO_MAP_VALUE |
143 		 *   PTR_TO_MAP_VALUE_OR_NULL
144 		 */
145 		struct bpf_map *map_ptr;
146 	};
147 };
148 
149 enum bpf_stack_slot_type {
150 	STACK_INVALID,    /* nothing was stored in this stack slot */
151 	STACK_SPILL,      /* register spilled into stack */
152 	STACK_MISC	  /* BPF program wrote some data into this slot */
153 };
154 
155 #define BPF_REG_SIZE 8	/* size of eBPF register in bytes */
156 
157 /* state of the program:
158  * type of all registers and stack info
159  */
160 struct verifier_state {
161 	struct reg_state regs[MAX_BPF_REG];
162 	u8 stack_slot_type[MAX_BPF_STACK];
163 	struct reg_state spilled_regs[MAX_BPF_STACK / BPF_REG_SIZE];
164 };
165 
166 /* linked list of verifier states used to prune search */
167 struct verifier_state_list {
168 	struct verifier_state state;
169 	struct verifier_state_list *next;
170 };
171 
172 /* verifier_state + insn_idx are pushed to stack when branch is encountered */
173 struct verifier_stack_elem {
174 	/* verifer state is 'st'
175 	 * before processing instruction 'insn_idx'
176 	 * and after processing instruction 'prev_insn_idx'
177 	 */
178 	struct verifier_state st;
179 	int insn_idx;
180 	int prev_insn_idx;
181 	struct verifier_stack_elem *next;
182 };
183 
184 #define MAX_USED_MAPS 64 /* max number of maps accessed by one eBPF program */
185 
186 /* single container for all structs
187  * one verifier_env per bpf_check() call
188  */
189 struct verifier_env {
190 	struct bpf_prog *prog;		/* eBPF program being verified */
191 	struct verifier_stack_elem *head; /* stack of verifier states to be processed */
192 	int stack_size;			/* number of states to be processed */
193 	struct verifier_state cur_state; /* current verifier state */
194 	struct verifier_state_list **explored_states; /* search pruning optimization */
195 	struct bpf_map *used_maps[MAX_USED_MAPS]; /* array of map's used by eBPF program */
196 	u32 used_map_cnt;		/* number of used maps */
197 	u32 id_gen;			/* used to generate unique reg IDs */
198 	bool allow_ptr_leaks;
199 };
200 
201 #define BPF_COMPLEXITY_LIMIT_INSNS	65536
202 #define BPF_COMPLEXITY_LIMIT_STACK	1024
203 
204 struct bpf_call_arg_meta {
205 	struct bpf_map *map_ptr;
206 	bool raw_mode;
207 	int regno;
208 	int access_size;
209 };
210 
211 /* verbose verifier prints what it's seeing
212  * bpf_check() is called under lock, so no race to access these global vars
213  */
214 static u32 log_level, log_size, log_len;
215 static char *log_buf;
216 
217 static DEFINE_MUTEX(bpf_verifier_lock);
218 
219 /* log_level controls verbosity level of eBPF verifier.
220  * verbose() is used to dump the verification trace to the log, so the user
221  * can figure out what's wrong with the program
222  */
223 static __printf(1, 2) void verbose(const char *fmt, ...)
224 {
225 	va_list args;
226 
227 	if (log_level == 0 || log_len >= log_size - 1)
228 		return;
229 
230 	va_start(args, fmt);
231 	log_len += vscnprintf(log_buf + log_len, log_size - log_len, fmt, args);
232 	va_end(args);
233 }
234 
235 /* string representation of 'enum bpf_reg_type' */
236 static const char * const reg_type_str[] = {
237 	[NOT_INIT]		= "?",
238 	[UNKNOWN_VALUE]		= "inv",
239 	[PTR_TO_CTX]		= "ctx",
240 	[CONST_PTR_TO_MAP]	= "map_ptr",
241 	[PTR_TO_MAP_VALUE]	= "map_value",
242 	[PTR_TO_MAP_VALUE_OR_NULL] = "map_value_or_null",
243 	[FRAME_PTR]		= "fp",
244 	[PTR_TO_STACK]		= "fp",
245 	[CONST_IMM]		= "imm",
246 	[PTR_TO_PACKET]		= "pkt",
247 	[PTR_TO_PACKET_END]	= "pkt_end",
248 };
249 
250 static void print_verifier_state(struct verifier_state *state)
251 {
252 	struct reg_state *reg;
253 	enum bpf_reg_type t;
254 	int i;
255 
256 	for (i = 0; i < MAX_BPF_REG; i++) {
257 		reg = &state->regs[i];
258 		t = reg->type;
259 		if (t == NOT_INIT)
260 			continue;
261 		verbose(" R%d=%s", i, reg_type_str[t]);
262 		if (t == CONST_IMM || t == PTR_TO_STACK)
263 			verbose("%lld", reg->imm);
264 		else if (t == PTR_TO_PACKET)
265 			verbose("(id=%d,off=%d,r=%d)",
266 				reg->id, reg->off, reg->range);
267 		else if (t == UNKNOWN_VALUE && reg->imm)
268 			verbose("%lld", reg->imm);
269 		else if (t == CONST_PTR_TO_MAP || t == PTR_TO_MAP_VALUE ||
270 			 t == PTR_TO_MAP_VALUE_OR_NULL)
271 			verbose("(ks=%d,vs=%d)",
272 				reg->map_ptr->key_size,
273 				reg->map_ptr->value_size);
274 	}
275 	for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
276 		if (state->stack_slot_type[i] == STACK_SPILL)
277 			verbose(" fp%d=%s", -MAX_BPF_STACK + i,
278 				reg_type_str[state->spilled_regs[i / BPF_REG_SIZE].type]);
279 	}
280 	verbose("\n");
281 }
282 
283 static const char *const bpf_class_string[] = {
284 	[BPF_LD]    = "ld",
285 	[BPF_LDX]   = "ldx",
286 	[BPF_ST]    = "st",
287 	[BPF_STX]   = "stx",
288 	[BPF_ALU]   = "alu",
289 	[BPF_JMP]   = "jmp",
290 	[BPF_RET]   = "BUG",
291 	[BPF_ALU64] = "alu64",
292 };
293 
294 static const char *const bpf_alu_string[16] = {
295 	[BPF_ADD >> 4]  = "+=",
296 	[BPF_SUB >> 4]  = "-=",
297 	[BPF_MUL >> 4]  = "*=",
298 	[BPF_DIV >> 4]  = "/=",
299 	[BPF_OR  >> 4]  = "|=",
300 	[BPF_AND >> 4]  = "&=",
301 	[BPF_LSH >> 4]  = "<<=",
302 	[BPF_RSH >> 4]  = ">>=",
303 	[BPF_NEG >> 4]  = "neg",
304 	[BPF_MOD >> 4]  = "%=",
305 	[BPF_XOR >> 4]  = "^=",
306 	[BPF_MOV >> 4]  = "=",
307 	[BPF_ARSH >> 4] = "s>>=",
308 	[BPF_END >> 4]  = "endian",
309 };
310 
311 static const char *const bpf_ldst_string[] = {
312 	[BPF_W >> 3]  = "u32",
313 	[BPF_H >> 3]  = "u16",
314 	[BPF_B >> 3]  = "u8",
315 	[BPF_DW >> 3] = "u64",
316 };
317 
318 static const char *const bpf_jmp_string[16] = {
319 	[BPF_JA >> 4]   = "jmp",
320 	[BPF_JEQ >> 4]  = "==",
321 	[BPF_JGT >> 4]  = ">",
322 	[BPF_JGE >> 4]  = ">=",
323 	[BPF_JSET >> 4] = "&",
324 	[BPF_JNE >> 4]  = "!=",
325 	[BPF_JSGT >> 4] = "s>",
326 	[BPF_JSGE >> 4] = "s>=",
327 	[BPF_CALL >> 4] = "call",
328 	[BPF_EXIT >> 4] = "exit",
329 };
330 
331 static void print_bpf_insn(struct bpf_insn *insn)
332 {
333 	u8 class = BPF_CLASS(insn->code);
334 
335 	if (class == BPF_ALU || class == BPF_ALU64) {
336 		if (BPF_SRC(insn->code) == BPF_X)
337 			verbose("(%02x) %sr%d %s %sr%d\n",
338 				insn->code, class == BPF_ALU ? "(u32) " : "",
339 				insn->dst_reg,
340 				bpf_alu_string[BPF_OP(insn->code) >> 4],
341 				class == BPF_ALU ? "(u32) " : "",
342 				insn->src_reg);
343 		else
344 			verbose("(%02x) %sr%d %s %s%d\n",
345 				insn->code, class == BPF_ALU ? "(u32) " : "",
346 				insn->dst_reg,
347 				bpf_alu_string[BPF_OP(insn->code) >> 4],
348 				class == BPF_ALU ? "(u32) " : "",
349 				insn->imm);
350 	} else if (class == BPF_STX) {
351 		if (BPF_MODE(insn->code) == BPF_MEM)
352 			verbose("(%02x) *(%s *)(r%d %+d) = r%d\n",
353 				insn->code,
354 				bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
355 				insn->dst_reg,
356 				insn->off, insn->src_reg);
357 		else if (BPF_MODE(insn->code) == BPF_XADD)
358 			verbose("(%02x) lock *(%s *)(r%d %+d) += r%d\n",
359 				insn->code,
360 				bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
361 				insn->dst_reg, insn->off,
362 				insn->src_reg);
363 		else
364 			verbose("BUG_%02x\n", insn->code);
365 	} else if (class == BPF_ST) {
366 		if (BPF_MODE(insn->code) != BPF_MEM) {
367 			verbose("BUG_st_%02x\n", insn->code);
368 			return;
369 		}
370 		verbose("(%02x) *(%s *)(r%d %+d) = %d\n",
371 			insn->code,
372 			bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
373 			insn->dst_reg,
374 			insn->off, insn->imm);
375 	} else if (class == BPF_LDX) {
376 		if (BPF_MODE(insn->code) != BPF_MEM) {
377 			verbose("BUG_ldx_%02x\n", insn->code);
378 			return;
379 		}
380 		verbose("(%02x) r%d = *(%s *)(r%d %+d)\n",
381 			insn->code, insn->dst_reg,
382 			bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
383 			insn->src_reg, insn->off);
384 	} else if (class == BPF_LD) {
385 		if (BPF_MODE(insn->code) == BPF_ABS) {
386 			verbose("(%02x) r0 = *(%s *)skb[%d]\n",
387 				insn->code,
388 				bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
389 				insn->imm);
390 		} else if (BPF_MODE(insn->code) == BPF_IND) {
391 			verbose("(%02x) r0 = *(%s *)skb[r%d + %d]\n",
392 				insn->code,
393 				bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
394 				insn->src_reg, insn->imm);
395 		} else if (BPF_MODE(insn->code) == BPF_IMM) {
396 			verbose("(%02x) r%d = 0x%x\n",
397 				insn->code, insn->dst_reg, insn->imm);
398 		} else {
399 			verbose("BUG_ld_%02x\n", insn->code);
400 			return;
401 		}
402 	} else if (class == BPF_JMP) {
403 		u8 opcode = BPF_OP(insn->code);
404 
405 		if (opcode == BPF_CALL) {
406 			verbose("(%02x) call %d\n", insn->code, insn->imm);
407 		} else if (insn->code == (BPF_JMP | BPF_JA)) {
408 			verbose("(%02x) goto pc%+d\n",
409 				insn->code, insn->off);
410 		} else if (insn->code == (BPF_JMP | BPF_EXIT)) {
411 			verbose("(%02x) exit\n", insn->code);
412 		} else if (BPF_SRC(insn->code) == BPF_X) {
413 			verbose("(%02x) if r%d %s r%d goto pc%+d\n",
414 				insn->code, insn->dst_reg,
415 				bpf_jmp_string[BPF_OP(insn->code) >> 4],
416 				insn->src_reg, insn->off);
417 		} else {
418 			verbose("(%02x) if r%d %s 0x%x goto pc%+d\n",
419 				insn->code, insn->dst_reg,
420 				bpf_jmp_string[BPF_OP(insn->code) >> 4],
421 				insn->imm, insn->off);
422 		}
423 	} else {
424 		verbose("(%02x) %s\n", insn->code, bpf_class_string[class]);
425 	}
426 }
427 
428 static int pop_stack(struct verifier_env *env, int *prev_insn_idx)
429 {
430 	struct verifier_stack_elem *elem;
431 	int insn_idx;
432 
433 	if (env->head == NULL)
434 		return -1;
435 
436 	memcpy(&env->cur_state, &env->head->st, sizeof(env->cur_state));
437 	insn_idx = env->head->insn_idx;
438 	if (prev_insn_idx)
439 		*prev_insn_idx = env->head->prev_insn_idx;
440 	elem = env->head->next;
441 	kfree(env->head);
442 	env->head = elem;
443 	env->stack_size--;
444 	return insn_idx;
445 }
446 
447 static struct verifier_state *push_stack(struct verifier_env *env, int insn_idx,
448 					 int prev_insn_idx)
449 {
450 	struct verifier_stack_elem *elem;
451 
452 	elem = kmalloc(sizeof(struct verifier_stack_elem), GFP_KERNEL);
453 	if (!elem)
454 		goto err;
455 
456 	memcpy(&elem->st, &env->cur_state, sizeof(env->cur_state));
457 	elem->insn_idx = insn_idx;
458 	elem->prev_insn_idx = prev_insn_idx;
459 	elem->next = env->head;
460 	env->head = elem;
461 	env->stack_size++;
462 	if (env->stack_size > BPF_COMPLEXITY_LIMIT_STACK) {
463 		verbose("BPF program is too complex\n");
464 		goto err;
465 	}
466 	return &elem->st;
467 err:
468 	/* pop all elements and return */
469 	while (pop_stack(env, NULL) >= 0);
470 	return NULL;
471 }
472 
473 #define CALLER_SAVED_REGS 6
474 static const int caller_saved[CALLER_SAVED_REGS] = {
475 	BPF_REG_0, BPF_REG_1, BPF_REG_2, BPF_REG_3, BPF_REG_4, BPF_REG_5
476 };
477 
478 static void init_reg_state(struct reg_state *regs)
479 {
480 	int i;
481 
482 	for (i = 0; i < MAX_BPF_REG; i++) {
483 		regs[i].type = NOT_INIT;
484 		regs[i].imm = 0;
485 	}
486 
487 	/* frame pointer */
488 	regs[BPF_REG_FP].type = FRAME_PTR;
489 
490 	/* 1st arg to a function */
491 	regs[BPF_REG_1].type = PTR_TO_CTX;
492 }
493 
494 static void mark_reg_unknown_value(struct reg_state *regs, u32 regno)
495 {
496 	BUG_ON(regno >= MAX_BPF_REG);
497 	regs[regno].type = UNKNOWN_VALUE;
498 	regs[regno].imm = 0;
499 }
500 
501 enum reg_arg_type {
502 	SRC_OP,		/* register is used as source operand */
503 	DST_OP,		/* register is used as destination operand */
504 	DST_OP_NO_MARK	/* same as above, check only, don't mark */
505 };
506 
507 static int check_reg_arg(struct reg_state *regs, u32 regno,
508 			 enum reg_arg_type t)
509 {
510 	if (regno >= MAX_BPF_REG) {
511 		verbose("R%d is invalid\n", regno);
512 		return -EINVAL;
513 	}
514 
515 	if (t == SRC_OP) {
516 		/* check whether register used as source operand can be read */
517 		if (regs[regno].type == NOT_INIT) {
518 			verbose("R%d !read_ok\n", regno);
519 			return -EACCES;
520 		}
521 	} else {
522 		/* check whether register used as dest operand can be written to */
523 		if (regno == BPF_REG_FP) {
524 			verbose("frame pointer is read only\n");
525 			return -EACCES;
526 		}
527 		if (t == DST_OP)
528 			mark_reg_unknown_value(regs, regno);
529 	}
530 	return 0;
531 }
532 
533 static int bpf_size_to_bytes(int bpf_size)
534 {
535 	if (bpf_size == BPF_W)
536 		return 4;
537 	else if (bpf_size == BPF_H)
538 		return 2;
539 	else if (bpf_size == BPF_B)
540 		return 1;
541 	else if (bpf_size == BPF_DW)
542 		return 8;
543 	else
544 		return -EINVAL;
545 }
546 
547 static bool is_spillable_regtype(enum bpf_reg_type type)
548 {
549 	switch (type) {
550 	case PTR_TO_MAP_VALUE:
551 	case PTR_TO_MAP_VALUE_OR_NULL:
552 	case PTR_TO_STACK:
553 	case PTR_TO_CTX:
554 	case PTR_TO_PACKET:
555 	case PTR_TO_PACKET_END:
556 	case FRAME_PTR:
557 	case CONST_PTR_TO_MAP:
558 		return true;
559 	default:
560 		return false;
561 	}
562 }
563 
564 /* check_stack_read/write functions track spill/fill of registers,
565  * stack boundary and alignment are checked in check_mem_access()
566  */
567 static int check_stack_write(struct verifier_state *state, int off, int size,
568 			     int value_regno)
569 {
570 	int i;
571 	/* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0,
572 	 * so it's aligned access and [off, off + size) are within stack limits
573 	 */
574 
575 	if (value_regno >= 0 &&
576 	    is_spillable_regtype(state->regs[value_regno].type)) {
577 
578 		/* register containing pointer is being spilled into stack */
579 		if (size != BPF_REG_SIZE) {
580 			verbose("invalid size of register spill\n");
581 			return -EACCES;
582 		}
583 
584 		/* save register state */
585 		state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] =
586 			state->regs[value_regno];
587 
588 		for (i = 0; i < BPF_REG_SIZE; i++)
589 			state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_SPILL;
590 	} else {
591 		/* regular write of data into stack */
592 		state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] =
593 			(struct reg_state) {};
594 
595 		for (i = 0; i < size; i++)
596 			state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_MISC;
597 	}
598 	return 0;
599 }
600 
601 static int check_stack_read(struct verifier_state *state, int off, int size,
602 			    int value_regno)
603 {
604 	u8 *slot_type;
605 	int i;
606 
607 	slot_type = &state->stack_slot_type[MAX_BPF_STACK + off];
608 
609 	if (slot_type[0] == STACK_SPILL) {
610 		if (size != BPF_REG_SIZE) {
611 			verbose("invalid size of register spill\n");
612 			return -EACCES;
613 		}
614 		for (i = 1; i < BPF_REG_SIZE; i++) {
615 			if (slot_type[i] != STACK_SPILL) {
616 				verbose("corrupted spill memory\n");
617 				return -EACCES;
618 			}
619 		}
620 
621 		if (value_regno >= 0)
622 			/* restore register state from stack */
623 			state->regs[value_regno] =
624 				state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE];
625 		return 0;
626 	} else {
627 		for (i = 0; i < size; i++) {
628 			if (slot_type[i] != STACK_MISC) {
629 				verbose("invalid read from stack off %d+%d size %d\n",
630 					off, i, size);
631 				return -EACCES;
632 			}
633 		}
634 		if (value_regno >= 0)
635 			/* have read misc data from the stack */
636 			mark_reg_unknown_value(state->regs, value_regno);
637 		return 0;
638 	}
639 }
640 
641 /* check read/write into map element returned by bpf_map_lookup_elem() */
642 static int check_map_access(struct verifier_env *env, u32 regno, int off,
643 			    int size)
644 {
645 	struct bpf_map *map = env->cur_state.regs[regno].map_ptr;
646 
647 	if (off < 0 || off + size > map->value_size) {
648 		verbose("invalid access to map value, value_size=%d off=%d size=%d\n",
649 			map->value_size, off, size);
650 		return -EACCES;
651 	}
652 	return 0;
653 }
654 
655 #define MAX_PACKET_OFF 0xffff
656 
657 static bool may_write_pkt_data(enum bpf_prog_type type)
658 {
659 	switch (type) {
660 	case BPF_PROG_TYPE_XDP:
661 		return true;
662 	default:
663 		return false;
664 	}
665 }
666 
667 static int check_packet_access(struct verifier_env *env, u32 regno, int off,
668 			       int size)
669 {
670 	struct reg_state *regs = env->cur_state.regs;
671 	struct reg_state *reg = &regs[regno];
672 
673 	off += reg->off;
674 	if (off < 0 || off + size > reg->range) {
675 		verbose("invalid access to packet, off=%d size=%d, R%d(id=%d,off=%d,r=%d)\n",
676 			off, size, regno, reg->id, reg->off, reg->range);
677 		return -EACCES;
678 	}
679 	return 0;
680 }
681 
682 /* check access to 'struct bpf_context' fields */
683 static int check_ctx_access(struct verifier_env *env, int off, int size,
684 			    enum bpf_access_type t, enum bpf_reg_type *reg_type)
685 {
686 	if (env->prog->aux->ops->is_valid_access &&
687 	    env->prog->aux->ops->is_valid_access(off, size, t, reg_type)) {
688 		/* remember the offset of last byte accessed in ctx */
689 		if (env->prog->aux->max_ctx_offset < off + size)
690 			env->prog->aux->max_ctx_offset = off + size;
691 		return 0;
692 	}
693 
694 	verbose("invalid bpf_context access off=%d size=%d\n", off, size);
695 	return -EACCES;
696 }
697 
698 static bool is_pointer_value(struct verifier_env *env, int regno)
699 {
700 	if (env->allow_ptr_leaks)
701 		return false;
702 
703 	switch (env->cur_state.regs[regno].type) {
704 	case UNKNOWN_VALUE:
705 	case CONST_IMM:
706 		return false;
707 	default:
708 		return true;
709 	}
710 }
711 
712 static int check_ptr_alignment(struct verifier_env *env, struct reg_state *reg,
713 			       int off, int size)
714 {
715 	if (reg->type != PTR_TO_PACKET) {
716 		if (off % size != 0) {
717 			verbose("misaligned access off %d size %d\n", off, size);
718 			return -EACCES;
719 		} else {
720 			return 0;
721 		}
722 	}
723 
724 	switch (env->prog->type) {
725 	case BPF_PROG_TYPE_SCHED_CLS:
726 	case BPF_PROG_TYPE_SCHED_ACT:
727 	case BPF_PROG_TYPE_XDP:
728 		break;
729 	default:
730 		verbose("verifier is misconfigured\n");
731 		return -EACCES;
732 	}
733 
734 	if (IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS))
735 		/* misaligned access to packet is ok on x86,arm,arm64 */
736 		return 0;
737 
738 	if (reg->id && size != 1) {
739 		verbose("Unknown packet alignment. Only byte-sized access allowed\n");
740 		return -EACCES;
741 	}
742 
743 	/* skb->data is NET_IP_ALIGN-ed */
744 	if ((NET_IP_ALIGN + reg->off + off) % size != 0) {
745 		verbose("misaligned packet access off %d+%d+%d size %d\n",
746 			NET_IP_ALIGN, reg->off, off, size);
747 		return -EACCES;
748 	}
749 	return 0;
750 }
751 
752 /* check whether memory at (regno + off) is accessible for t = (read | write)
753  * if t==write, value_regno is a register which value is stored into memory
754  * if t==read, value_regno is a register which will receive the value from memory
755  * if t==write && value_regno==-1, some unknown value is stored into memory
756  * if t==read && value_regno==-1, don't care what we read from memory
757  */
758 static int check_mem_access(struct verifier_env *env, u32 regno, int off,
759 			    int bpf_size, enum bpf_access_type t,
760 			    int value_regno)
761 {
762 	struct verifier_state *state = &env->cur_state;
763 	struct reg_state *reg = &state->regs[regno];
764 	int size, err = 0;
765 
766 	if (reg->type == PTR_TO_STACK)
767 		off += reg->imm;
768 
769 	size = bpf_size_to_bytes(bpf_size);
770 	if (size < 0)
771 		return size;
772 
773 	err = check_ptr_alignment(env, reg, off, size);
774 	if (err)
775 		return err;
776 
777 	if (reg->type == PTR_TO_MAP_VALUE) {
778 		if (t == BPF_WRITE && value_regno >= 0 &&
779 		    is_pointer_value(env, value_regno)) {
780 			verbose("R%d leaks addr into map\n", value_regno);
781 			return -EACCES;
782 		}
783 		err = check_map_access(env, regno, off, size);
784 		if (!err && t == BPF_READ && value_regno >= 0)
785 			mark_reg_unknown_value(state->regs, value_regno);
786 
787 	} else if (reg->type == PTR_TO_CTX) {
788 		enum bpf_reg_type reg_type = UNKNOWN_VALUE;
789 
790 		if (t == BPF_WRITE && value_regno >= 0 &&
791 		    is_pointer_value(env, value_regno)) {
792 			verbose("R%d leaks addr into ctx\n", value_regno);
793 			return -EACCES;
794 		}
795 		err = check_ctx_access(env, off, size, t, &reg_type);
796 		if (!err && t == BPF_READ && value_regno >= 0) {
797 			mark_reg_unknown_value(state->regs, value_regno);
798 			if (env->allow_ptr_leaks)
799 				/* note that reg.[id|off|range] == 0 */
800 				state->regs[value_regno].type = reg_type;
801 		}
802 
803 	} else if (reg->type == FRAME_PTR || reg->type == PTR_TO_STACK) {
804 		if (off >= 0 || off < -MAX_BPF_STACK) {
805 			verbose("invalid stack off=%d size=%d\n", off, size);
806 			return -EACCES;
807 		}
808 		if (t == BPF_WRITE) {
809 			if (!env->allow_ptr_leaks &&
810 			    state->stack_slot_type[MAX_BPF_STACK + off] == STACK_SPILL &&
811 			    size != BPF_REG_SIZE) {
812 				verbose("attempt to corrupt spilled pointer on stack\n");
813 				return -EACCES;
814 			}
815 			err = check_stack_write(state, off, size, value_regno);
816 		} else {
817 			err = check_stack_read(state, off, size, value_regno);
818 		}
819 	} else if (state->regs[regno].type == PTR_TO_PACKET) {
820 		if (t == BPF_WRITE && !may_write_pkt_data(env->prog->type)) {
821 			verbose("cannot write into packet\n");
822 			return -EACCES;
823 		}
824 		if (t == BPF_WRITE && value_regno >= 0 &&
825 		    is_pointer_value(env, value_regno)) {
826 			verbose("R%d leaks addr into packet\n", value_regno);
827 			return -EACCES;
828 		}
829 		err = check_packet_access(env, regno, off, size);
830 		if (!err && t == BPF_READ && value_regno >= 0)
831 			mark_reg_unknown_value(state->regs, value_regno);
832 	} else {
833 		verbose("R%d invalid mem access '%s'\n",
834 			regno, reg_type_str[reg->type]);
835 		return -EACCES;
836 	}
837 
838 	if (!err && size <= 2 && value_regno >= 0 && env->allow_ptr_leaks &&
839 	    state->regs[value_regno].type == UNKNOWN_VALUE) {
840 		/* 1 or 2 byte load zero-extends, determine the number of
841 		 * zero upper bits. Not doing it fo 4 byte load, since
842 		 * such values cannot be added to ptr_to_packet anyway.
843 		 */
844 		state->regs[value_regno].imm = 64 - size * 8;
845 	}
846 	return err;
847 }
848 
849 static int check_xadd(struct verifier_env *env, struct bpf_insn *insn)
850 {
851 	struct reg_state *regs = env->cur_state.regs;
852 	int err;
853 
854 	if ((BPF_SIZE(insn->code) != BPF_W && BPF_SIZE(insn->code) != BPF_DW) ||
855 	    insn->imm != 0) {
856 		verbose("BPF_XADD uses reserved fields\n");
857 		return -EINVAL;
858 	}
859 
860 	/* check src1 operand */
861 	err = check_reg_arg(regs, insn->src_reg, SRC_OP);
862 	if (err)
863 		return err;
864 
865 	/* check src2 operand */
866 	err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
867 	if (err)
868 		return err;
869 
870 	/* check whether atomic_add can read the memory */
871 	err = check_mem_access(env, insn->dst_reg, insn->off,
872 			       BPF_SIZE(insn->code), BPF_READ, -1);
873 	if (err)
874 		return err;
875 
876 	/* check whether atomic_add can write into the same memory */
877 	return check_mem_access(env, insn->dst_reg, insn->off,
878 				BPF_SIZE(insn->code), BPF_WRITE, -1);
879 }
880 
881 /* when register 'regno' is passed into function that will read 'access_size'
882  * bytes from that pointer, make sure that it's within stack boundary
883  * and all elements of stack are initialized
884  */
885 static int check_stack_boundary(struct verifier_env *env, int regno,
886 				int access_size, bool zero_size_allowed,
887 				struct bpf_call_arg_meta *meta)
888 {
889 	struct verifier_state *state = &env->cur_state;
890 	struct reg_state *regs = state->regs;
891 	int off, i;
892 
893 	if (regs[regno].type != PTR_TO_STACK) {
894 		if (zero_size_allowed && access_size == 0 &&
895 		    regs[regno].type == CONST_IMM &&
896 		    regs[regno].imm  == 0)
897 			return 0;
898 
899 		verbose("R%d type=%s expected=%s\n", regno,
900 			reg_type_str[regs[regno].type],
901 			reg_type_str[PTR_TO_STACK]);
902 		return -EACCES;
903 	}
904 
905 	off = regs[regno].imm;
906 	if (off >= 0 || off < -MAX_BPF_STACK || off + access_size > 0 ||
907 	    access_size <= 0) {
908 		verbose("invalid stack type R%d off=%d access_size=%d\n",
909 			regno, off, access_size);
910 		return -EACCES;
911 	}
912 
913 	if (meta && meta->raw_mode) {
914 		meta->access_size = access_size;
915 		meta->regno = regno;
916 		return 0;
917 	}
918 
919 	for (i = 0; i < access_size; i++) {
920 		if (state->stack_slot_type[MAX_BPF_STACK + off + i] != STACK_MISC) {
921 			verbose("invalid indirect read from stack off %d+%d size %d\n",
922 				off, i, access_size);
923 			return -EACCES;
924 		}
925 	}
926 	return 0;
927 }
928 
929 static int check_func_arg(struct verifier_env *env, u32 regno,
930 			  enum bpf_arg_type arg_type,
931 			  struct bpf_call_arg_meta *meta)
932 {
933 	struct reg_state *reg = env->cur_state.regs + regno;
934 	enum bpf_reg_type expected_type;
935 	int err = 0;
936 
937 	if (arg_type == ARG_DONTCARE)
938 		return 0;
939 
940 	if (reg->type == NOT_INIT) {
941 		verbose("R%d !read_ok\n", regno);
942 		return -EACCES;
943 	}
944 
945 	if (arg_type == ARG_ANYTHING) {
946 		if (is_pointer_value(env, regno)) {
947 			verbose("R%d leaks addr into helper function\n", regno);
948 			return -EACCES;
949 		}
950 		return 0;
951 	}
952 
953 	if (arg_type == ARG_PTR_TO_MAP_KEY ||
954 	    arg_type == ARG_PTR_TO_MAP_VALUE) {
955 		expected_type = PTR_TO_STACK;
956 	} else if (arg_type == ARG_CONST_STACK_SIZE ||
957 		   arg_type == ARG_CONST_STACK_SIZE_OR_ZERO) {
958 		expected_type = CONST_IMM;
959 	} else if (arg_type == ARG_CONST_MAP_PTR) {
960 		expected_type = CONST_PTR_TO_MAP;
961 	} else if (arg_type == ARG_PTR_TO_CTX) {
962 		expected_type = PTR_TO_CTX;
963 	} else if (arg_type == ARG_PTR_TO_STACK ||
964 		   arg_type == ARG_PTR_TO_RAW_STACK) {
965 		expected_type = PTR_TO_STACK;
966 		/* One exception here. In case function allows for NULL to be
967 		 * passed in as argument, it's a CONST_IMM type. Final test
968 		 * happens during stack boundary checking.
969 		 */
970 		if (reg->type == CONST_IMM && reg->imm == 0)
971 			expected_type = CONST_IMM;
972 		meta->raw_mode = arg_type == ARG_PTR_TO_RAW_STACK;
973 	} else {
974 		verbose("unsupported arg_type %d\n", arg_type);
975 		return -EFAULT;
976 	}
977 
978 	if (reg->type != expected_type) {
979 		verbose("R%d type=%s expected=%s\n", regno,
980 			reg_type_str[reg->type], reg_type_str[expected_type]);
981 		return -EACCES;
982 	}
983 
984 	if (arg_type == ARG_CONST_MAP_PTR) {
985 		/* bpf_map_xxx(map_ptr) call: remember that map_ptr */
986 		meta->map_ptr = reg->map_ptr;
987 	} else if (arg_type == ARG_PTR_TO_MAP_KEY) {
988 		/* bpf_map_xxx(..., map_ptr, ..., key) call:
989 		 * check that [key, key + map->key_size) are within
990 		 * stack limits and initialized
991 		 */
992 		if (!meta->map_ptr) {
993 			/* in function declaration map_ptr must come before
994 			 * map_key, so that it's verified and known before
995 			 * we have to check map_key here. Otherwise it means
996 			 * that kernel subsystem misconfigured verifier
997 			 */
998 			verbose("invalid map_ptr to access map->key\n");
999 			return -EACCES;
1000 		}
1001 		err = check_stack_boundary(env, regno, meta->map_ptr->key_size,
1002 					   false, NULL);
1003 	} else if (arg_type == ARG_PTR_TO_MAP_VALUE) {
1004 		/* bpf_map_xxx(..., map_ptr, ..., value) call:
1005 		 * check [value, value + map->value_size) validity
1006 		 */
1007 		if (!meta->map_ptr) {
1008 			/* kernel subsystem misconfigured verifier */
1009 			verbose("invalid map_ptr to access map->value\n");
1010 			return -EACCES;
1011 		}
1012 		err = check_stack_boundary(env, regno,
1013 					   meta->map_ptr->value_size,
1014 					   false, NULL);
1015 	} else if (arg_type == ARG_CONST_STACK_SIZE ||
1016 		   arg_type == ARG_CONST_STACK_SIZE_OR_ZERO) {
1017 		bool zero_size_allowed = (arg_type == ARG_CONST_STACK_SIZE_OR_ZERO);
1018 
1019 		/* bpf_xxx(..., buf, len) call will access 'len' bytes
1020 		 * from stack pointer 'buf'. Check it
1021 		 * note: regno == len, regno - 1 == buf
1022 		 */
1023 		if (regno == 0) {
1024 			/* kernel subsystem misconfigured verifier */
1025 			verbose("ARG_CONST_STACK_SIZE cannot be first argument\n");
1026 			return -EACCES;
1027 		}
1028 		err = check_stack_boundary(env, regno - 1, reg->imm,
1029 					   zero_size_allowed, meta);
1030 	}
1031 
1032 	return err;
1033 }
1034 
1035 static int check_map_func_compatibility(struct bpf_map *map, int func_id)
1036 {
1037 	if (!map)
1038 		return 0;
1039 
1040 	/* We need a two way check, first is from map perspective ... */
1041 	switch (map->map_type) {
1042 	case BPF_MAP_TYPE_PROG_ARRAY:
1043 		if (func_id != BPF_FUNC_tail_call)
1044 			goto error;
1045 		break;
1046 	case BPF_MAP_TYPE_PERF_EVENT_ARRAY:
1047 		if (func_id != BPF_FUNC_perf_event_read &&
1048 		    func_id != BPF_FUNC_perf_event_output)
1049 			goto error;
1050 		break;
1051 	case BPF_MAP_TYPE_STACK_TRACE:
1052 		if (func_id != BPF_FUNC_get_stackid)
1053 			goto error;
1054 		break;
1055 	case BPF_MAP_TYPE_CGROUP_ARRAY:
1056 		if (func_id != BPF_FUNC_skb_under_cgroup)
1057 			goto error;
1058 		break;
1059 	default:
1060 		break;
1061 	}
1062 
1063 	/* ... and second from the function itself. */
1064 	switch (func_id) {
1065 	case BPF_FUNC_tail_call:
1066 		if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY)
1067 			goto error;
1068 		break;
1069 	case BPF_FUNC_perf_event_read:
1070 	case BPF_FUNC_perf_event_output:
1071 		if (map->map_type != BPF_MAP_TYPE_PERF_EVENT_ARRAY)
1072 			goto error;
1073 		break;
1074 	case BPF_FUNC_get_stackid:
1075 		if (map->map_type != BPF_MAP_TYPE_STACK_TRACE)
1076 			goto error;
1077 		break;
1078 	case BPF_FUNC_skb_under_cgroup:
1079 		if (map->map_type != BPF_MAP_TYPE_CGROUP_ARRAY)
1080 			goto error;
1081 		break;
1082 	default:
1083 		break;
1084 	}
1085 
1086 	return 0;
1087 error:
1088 	verbose("cannot pass map_type %d into func %d\n",
1089 		map->map_type, func_id);
1090 	return -EINVAL;
1091 }
1092 
1093 static int check_raw_mode(const struct bpf_func_proto *fn)
1094 {
1095 	int count = 0;
1096 
1097 	if (fn->arg1_type == ARG_PTR_TO_RAW_STACK)
1098 		count++;
1099 	if (fn->arg2_type == ARG_PTR_TO_RAW_STACK)
1100 		count++;
1101 	if (fn->arg3_type == ARG_PTR_TO_RAW_STACK)
1102 		count++;
1103 	if (fn->arg4_type == ARG_PTR_TO_RAW_STACK)
1104 		count++;
1105 	if (fn->arg5_type == ARG_PTR_TO_RAW_STACK)
1106 		count++;
1107 
1108 	return count > 1 ? -EINVAL : 0;
1109 }
1110 
1111 static void clear_all_pkt_pointers(struct verifier_env *env)
1112 {
1113 	struct verifier_state *state = &env->cur_state;
1114 	struct reg_state *regs = state->regs, *reg;
1115 	int i;
1116 
1117 	for (i = 0; i < MAX_BPF_REG; i++)
1118 		if (regs[i].type == PTR_TO_PACKET ||
1119 		    regs[i].type == PTR_TO_PACKET_END)
1120 			mark_reg_unknown_value(regs, i);
1121 
1122 	for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
1123 		if (state->stack_slot_type[i] != STACK_SPILL)
1124 			continue;
1125 		reg = &state->spilled_regs[i / BPF_REG_SIZE];
1126 		if (reg->type != PTR_TO_PACKET &&
1127 		    reg->type != PTR_TO_PACKET_END)
1128 			continue;
1129 		reg->type = UNKNOWN_VALUE;
1130 		reg->imm = 0;
1131 	}
1132 }
1133 
1134 static int check_call(struct verifier_env *env, int func_id)
1135 {
1136 	struct verifier_state *state = &env->cur_state;
1137 	const struct bpf_func_proto *fn = NULL;
1138 	struct reg_state *regs = state->regs;
1139 	struct reg_state *reg;
1140 	struct bpf_call_arg_meta meta;
1141 	bool changes_data;
1142 	int i, err;
1143 
1144 	/* find function prototype */
1145 	if (func_id < 0 || func_id >= __BPF_FUNC_MAX_ID) {
1146 		verbose("invalid func %d\n", func_id);
1147 		return -EINVAL;
1148 	}
1149 
1150 	if (env->prog->aux->ops->get_func_proto)
1151 		fn = env->prog->aux->ops->get_func_proto(func_id);
1152 
1153 	if (!fn) {
1154 		verbose("unknown func %d\n", func_id);
1155 		return -EINVAL;
1156 	}
1157 
1158 	/* eBPF programs must be GPL compatible to use GPL-ed functions */
1159 	if (!env->prog->gpl_compatible && fn->gpl_only) {
1160 		verbose("cannot call GPL only function from proprietary program\n");
1161 		return -EINVAL;
1162 	}
1163 
1164 	changes_data = bpf_helper_changes_skb_data(fn->func);
1165 
1166 	memset(&meta, 0, sizeof(meta));
1167 
1168 	/* We only support one arg being in raw mode at the moment, which
1169 	 * is sufficient for the helper functions we have right now.
1170 	 */
1171 	err = check_raw_mode(fn);
1172 	if (err) {
1173 		verbose("kernel subsystem misconfigured func %d\n", func_id);
1174 		return err;
1175 	}
1176 
1177 	/* check args */
1178 	err = check_func_arg(env, BPF_REG_1, fn->arg1_type, &meta);
1179 	if (err)
1180 		return err;
1181 	err = check_func_arg(env, BPF_REG_2, fn->arg2_type, &meta);
1182 	if (err)
1183 		return err;
1184 	err = check_func_arg(env, BPF_REG_3, fn->arg3_type, &meta);
1185 	if (err)
1186 		return err;
1187 	err = check_func_arg(env, BPF_REG_4, fn->arg4_type, &meta);
1188 	if (err)
1189 		return err;
1190 	err = check_func_arg(env, BPF_REG_5, fn->arg5_type, &meta);
1191 	if (err)
1192 		return err;
1193 
1194 	/* Mark slots with STACK_MISC in case of raw mode, stack offset
1195 	 * is inferred from register state.
1196 	 */
1197 	for (i = 0; i < meta.access_size; i++) {
1198 		err = check_mem_access(env, meta.regno, i, BPF_B, BPF_WRITE, -1);
1199 		if (err)
1200 			return err;
1201 	}
1202 
1203 	/* reset caller saved regs */
1204 	for (i = 0; i < CALLER_SAVED_REGS; i++) {
1205 		reg = regs + caller_saved[i];
1206 		reg->type = NOT_INIT;
1207 		reg->imm = 0;
1208 	}
1209 
1210 	/* update return register */
1211 	if (fn->ret_type == RET_INTEGER) {
1212 		regs[BPF_REG_0].type = UNKNOWN_VALUE;
1213 	} else if (fn->ret_type == RET_VOID) {
1214 		regs[BPF_REG_0].type = NOT_INIT;
1215 	} else if (fn->ret_type == RET_PTR_TO_MAP_VALUE_OR_NULL) {
1216 		regs[BPF_REG_0].type = PTR_TO_MAP_VALUE_OR_NULL;
1217 		/* remember map_ptr, so that check_map_access()
1218 		 * can check 'value_size' boundary of memory access
1219 		 * to map element returned from bpf_map_lookup_elem()
1220 		 */
1221 		if (meta.map_ptr == NULL) {
1222 			verbose("kernel subsystem misconfigured verifier\n");
1223 			return -EINVAL;
1224 		}
1225 		regs[BPF_REG_0].map_ptr = meta.map_ptr;
1226 	} else {
1227 		verbose("unknown return type %d of func %d\n",
1228 			fn->ret_type, func_id);
1229 		return -EINVAL;
1230 	}
1231 
1232 	err = check_map_func_compatibility(meta.map_ptr, func_id);
1233 	if (err)
1234 		return err;
1235 
1236 	if (changes_data)
1237 		clear_all_pkt_pointers(env);
1238 	return 0;
1239 }
1240 
1241 static int check_packet_ptr_add(struct verifier_env *env, struct bpf_insn *insn)
1242 {
1243 	struct reg_state *regs = env->cur_state.regs;
1244 	struct reg_state *dst_reg = &regs[insn->dst_reg];
1245 	struct reg_state *src_reg = &regs[insn->src_reg];
1246 	struct reg_state tmp_reg;
1247 	s32 imm;
1248 
1249 	if (BPF_SRC(insn->code) == BPF_K) {
1250 		/* pkt_ptr += imm */
1251 		imm = insn->imm;
1252 
1253 add_imm:
1254 		if (imm <= 0) {
1255 			verbose("addition of negative constant to packet pointer is not allowed\n");
1256 			return -EACCES;
1257 		}
1258 		if (imm >= MAX_PACKET_OFF ||
1259 		    imm + dst_reg->off >= MAX_PACKET_OFF) {
1260 			verbose("constant %d is too large to add to packet pointer\n",
1261 				imm);
1262 			return -EACCES;
1263 		}
1264 		/* a constant was added to pkt_ptr.
1265 		 * Remember it while keeping the same 'id'
1266 		 */
1267 		dst_reg->off += imm;
1268 	} else {
1269 		if (src_reg->type == PTR_TO_PACKET) {
1270 			/* R6=pkt(id=0,off=0,r=62) R7=imm22; r7 += r6 */
1271 			tmp_reg = *dst_reg;  /* save r7 state */
1272 			*dst_reg = *src_reg; /* copy pkt_ptr state r6 into r7 */
1273 			src_reg = &tmp_reg;  /* pretend it's src_reg state */
1274 			/* if the checks below reject it, the copy won't matter,
1275 			 * since we're rejecting the whole program. If all ok,
1276 			 * then imm22 state will be added to r7
1277 			 * and r7 will be pkt(id=0,off=22,r=62) while
1278 			 * r6 will stay as pkt(id=0,off=0,r=62)
1279 			 */
1280 		}
1281 
1282 		if (src_reg->type == CONST_IMM) {
1283 			/* pkt_ptr += reg where reg is known constant */
1284 			imm = src_reg->imm;
1285 			goto add_imm;
1286 		}
1287 		/* disallow pkt_ptr += reg
1288 		 * if reg is not uknown_value with guaranteed zero upper bits
1289 		 * otherwise pkt_ptr may overflow and addition will become
1290 		 * subtraction which is not allowed
1291 		 */
1292 		if (src_reg->type != UNKNOWN_VALUE) {
1293 			verbose("cannot add '%s' to ptr_to_packet\n",
1294 				reg_type_str[src_reg->type]);
1295 			return -EACCES;
1296 		}
1297 		if (src_reg->imm < 48) {
1298 			verbose("cannot add integer value with %lld upper zero bits to ptr_to_packet\n",
1299 				src_reg->imm);
1300 			return -EACCES;
1301 		}
1302 		/* dst_reg stays as pkt_ptr type and since some positive
1303 		 * integer value was added to the pointer, increment its 'id'
1304 		 */
1305 		dst_reg->id = ++env->id_gen;
1306 
1307 		/* something was added to pkt_ptr, set range and off to zero */
1308 		dst_reg->off = 0;
1309 		dst_reg->range = 0;
1310 	}
1311 	return 0;
1312 }
1313 
1314 static int evaluate_reg_alu(struct verifier_env *env, struct bpf_insn *insn)
1315 {
1316 	struct reg_state *regs = env->cur_state.regs;
1317 	struct reg_state *dst_reg = &regs[insn->dst_reg];
1318 	u8 opcode = BPF_OP(insn->code);
1319 	s64 imm_log2;
1320 
1321 	/* for type == UNKNOWN_VALUE:
1322 	 * imm > 0 -> number of zero upper bits
1323 	 * imm == 0 -> don't track which is the same as all bits can be non-zero
1324 	 */
1325 
1326 	if (BPF_SRC(insn->code) == BPF_X) {
1327 		struct reg_state *src_reg = &regs[insn->src_reg];
1328 
1329 		if (src_reg->type == UNKNOWN_VALUE && src_reg->imm > 0 &&
1330 		    dst_reg->imm && opcode == BPF_ADD) {
1331 			/* dreg += sreg
1332 			 * where both have zero upper bits. Adding them
1333 			 * can only result making one more bit non-zero
1334 			 * in the larger value.
1335 			 * Ex. 0xffff (imm=48) + 1 (imm=63) = 0x10000 (imm=47)
1336 			 *     0xffff (imm=48) + 0xffff = 0x1fffe (imm=47)
1337 			 */
1338 			dst_reg->imm = min(dst_reg->imm, src_reg->imm);
1339 			dst_reg->imm--;
1340 			return 0;
1341 		}
1342 		if (src_reg->type == CONST_IMM && src_reg->imm > 0 &&
1343 		    dst_reg->imm && opcode == BPF_ADD) {
1344 			/* dreg += sreg
1345 			 * where dreg has zero upper bits and sreg is const.
1346 			 * Adding them can only result making one more bit
1347 			 * non-zero in the larger value.
1348 			 */
1349 			imm_log2 = __ilog2_u64((long long)src_reg->imm);
1350 			dst_reg->imm = min(dst_reg->imm, 63 - imm_log2);
1351 			dst_reg->imm--;
1352 			return 0;
1353 		}
1354 		/* all other cases non supported yet, just mark dst_reg */
1355 		dst_reg->imm = 0;
1356 		return 0;
1357 	}
1358 
1359 	/* sign extend 32-bit imm into 64-bit to make sure that
1360 	 * negative values occupy bit 63. Note ilog2() would have
1361 	 * been incorrect, since sizeof(insn->imm) == 4
1362 	 */
1363 	imm_log2 = __ilog2_u64((long long)insn->imm);
1364 
1365 	if (dst_reg->imm && opcode == BPF_LSH) {
1366 		/* reg <<= imm
1367 		 * if reg was a result of 2 byte load, then its imm == 48
1368 		 * which means that upper 48 bits are zero and shifting this reg
1369 		 * left by 4 would mean that upper 44 bits are still zero
1370 		 */
1371 		dst_reg->imm -= insn->imm;
1372 	} else if (dst_reg->imm && opcode == BPF_MUL) {
1373 		/* reg *= imm
1374 		 * if multiplying by 14 subtract 4
1375 		 * This is conservative calculation of upper zero bits.
1376 		 * It's not trying to special case insn->imm == 1 or 0 cases
1377 		 */
1378 		dst_reg->imm -= imm_log2 + 1;
1379 	} else if (opcode == BPF_AND) {
1380 		/* reg &= imm */
1381 		dst_reg->imm = 63 - imm_log2;
1382 	} else if (dst_reg->imm && opcode == BPF_ADD) {
1383 		/* reg += imm */
1384 		dst_reg->imm = min(dst_reg->imm, 63 - imm_log2);
1385 		dst_reg->imm--;
1386 	} else if (opcode == BPF_RSH) {
1387 		/* reg >>= imm
1388 		 * which means that after right shift, upper bits will be zero
1389 		 * note that verifier already checked that
1390 		 * 0 <= imm < 64 for shift insn
1391 		 */
1392 		dst_reg->imm += insn->imm;
1393 		if (unlikely(dst_reg->imm > 64))
1394 			/* some dumb code did:
1395 			 * r2 = *(u32 *)mem;
1396 			 * r2 >>= 32;
1397 			 * and all bits are zero now */
1398 			dst_reg->imm = 64;
1399 	} else {
1400 		/* all other alu ops, means that we don't know what will
1401 		 * happen to the value, mark it with unknown number of zero bits
1402 		 */
1403 		dst_reg->imm = 0;
1404 	}
1405 
1406 	if (dst_reg->imm < 0) {
1407 		/* all 64 bits of the register can contain non-zero bits
1408 		 * and such value cannot be added to ptr_to_packet, since it
1409 		 * may overflow, mark it as unknown to avoid further eval
1410 		 */
1411 		dst_reg->imm = 0;
1412 	}
1413 	return 0;
1414 }
1415 
1416 static int evaluate_reg_imm_alu(struct verifier_env *env, struct bpf_insn *insn)
1417 {
1418 	struct reg_state *regs = env->cur_state.regs;
1419 	struct reg_state *dst_reg = &regs[insn->dst_reg];
1420 	struct reg_state *src_reg = &regs[insn->src_reg];
1421 	u8 opcode = BPF_OP(insn->code);
1422 
1423 	/* dst_reg->type == CONST_IMM here, simulate execution of 'add' insn.
1424 	 * Don't care about overflow or negative values, just add them
1425 	 */
1426 	if (opcode == BPF_ADD && BPF_SRC(insn->code) == BPF_K)
1427 		dst_reg->imm += insn->imm;
1428 	else if (opcode == BPF_ADD && BPF_SRC(insn->code) == BPF_X &&
1429 		 src_reg->type == CONST_IMM)
1430 		dst_reg->imm += src_reg->imm;
1431 	else
1432 		mark_reg_unknown_value(regs, insn->dst_reg);
1433 	return 0;
1434 }
1435 
1436 /* check validity of 32-bit and 64-bit arithmetic operations */
1437 static int check_alu_op(struct verifier_env *env, struct bpf_insn *insn)
1438 {
1439 	struct reg_state *regs = env->cur_state.regs, *dst_reg;
1440 	u8 opcode = BPF_OP(insn->code);
1441 	int err;
1442 
1443 	if (opcode == BPF_END || opcode == BPF_NEG) {
1444 		if (opcode == BPF_NEG) {
1445 			if (BPF_SRC(insn->code) != 0 ||
1446 			    insn->src_reg != BPF_REG_0 ||
1447 			    insn->off != 0 || insn->imm != 0) {
1448 				verbose("BPF_NEG uses reserved fields\n");
1449 				return -EINVAL;
1450 			}
1451 		} else {
1452 			if (insn->src_reg != BPF_REG_0 || insn->off != 0 ||
1453 			    (insn->imm != 16 && insn->imm != 32 && insn->imm != 64)) {
1454 				verbose("BPF_END uses reserved fields\n");
1455 				return -EINVAL;
1456 			}
1457 		}
1458 
1459 		/* check src operand */
1460 		err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1461 		if (err)
1462 			return err;
1463 
1464 		if (is_pointer_value(env, insn->dst_reg)) {
1465 			verbose("R%d pointer arithmetic prohibited\n",
1466 				insn->dst_reg);
1467 			return -EACCES;
1468 		}
1469 
1470 		/* check dest operand */
1471 		err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1472 		if (err)
1473 			return err;
1474 
1475 	} else if (opcode == BPF_MOV) {
1476 
1477 		if (BPF_SRC(insn->code) == BPF_X) {
1478 			if (insn->imm != 0 || insn->off != 0) {
1479 				verbose("BPF_MOV uses reserved fields\n");
1480 				return -EINVAL;
1481 			}
1482 
1483 			/* check src operand */
1484 			err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1485 			if (err)
1486 				return err;
1487 		} else {
1488 			if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
1489 				verbose("BPF_MOV uses reserved fields\n");
1490 				return -EINVAL;
1491 			}
1492 		}
1493 
1494 		/* check dest operand */
1495 		err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1496 		if (err)
1497 			return err;
1498 
1499 		if (BPF_SRC(insn->code) == BPF_X) {
1500 			if (BPF_CLASS(insn->code) == BPF_ALU64) {
1501 				/* case: R1 = R2
1502 				 * copy register state to dest reg
1503 				 */
1504 				regs[insn->dst_reg] = regs[insn->src_reg];
1505 			} else {
1506 				if (is_pointer_value(env, insn->src_reg)) {
1507 					verbose("R%d partial copy of pointer\n",
1508 						insn->src_reg);
1509 					return -EACCES;
1510 				}
1511 				regs[insn->dst_reg].type = UNKNOWN_VALUE;
1512 				regs[insn->dst_reg].map_ptr = NULL;
1513 			}
1514 		} else {
1515 			/* case: R = imm
1516 			 * remember the value we stored into this reg
1517 			 */
1518 			regs[insn->dst_reg].type = CONST_IMM;
1519 			regs[insn->dst_reg].imm = insn->imm;
1520 		}
1521 
1522 	} else if (opcode > BPF_END) {
1523 		verbose("invalid BPF_ALU opcode %x\n", opcode);
1524 		return -EINVAL;
1525 
1526 	} else {	/* all other ALU ops: and, sub, xor, add, ... */
1527 
1528 		if (BPF_SRC(insn->code) == BPF_X) {
1529 			if (insn->imm != 0 || insn->off != 0) {
1530 				verbose("BPF_ALU uses reserved fields\n");
1531 				return -EINVAL;
1532 			}
1533 			/* check src1 operand */
1534 			err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1535 			if (err)
1536 				return err;
1537 		} else {
1538 			if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
1539 				verbose("BPF_ALU uses reserved fields\n");
1540 				return -EINVAL;
1541 			}
1542 		}
1543 
1544 		/* check src2 operand */
1545 		err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1546 		if (err)
1547 			return err;
1548 
1549 		if ((opcode == BPF_MOD || opcode == BPF_DIV) &&
1550 		    BPF_SRC(insn->code) == BPF_K && insn->imm == 0) {
1551 			verbose("div by zero\n");
1552 			return -EINVAL;
1553 		}
1554 
1555 		if ((opcode == BPF_LSH || opcode == BPF_RSH ||
1556 		     opcode == BPF_ARSH) && BPF_SRC(insn->code) == BPF_K) {
1557 			int size = BPF_CLASS(insn->code) == BPF_ALU64 ? 64 : 32;
1558 
1559 			if (insn->imm < 0 || insn->imm >= size) {
1560 				verbose("invalid shift %d\n", insn->imm);
1561 				return -EINVAL;
1562 			}
1563 		}
1564 
1565 		/* check dest operand */
1566 		err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK);
1567 		if (err)
1568 			return err;
1569 
1570 		dst_reg = &regs[insn->dst_reg];
1571 
1572 		/* pattern match 'bpf_add Rx, imm' instruction */
1573 		if (opcode == BPF_ADD && BPF_CLASS(insn->code) == BPF_ALU64 &&
1574 		    dst_reg->type == FRAME_PTR && BPF_SRC(insn->code) == BPF_K) {
1575 			dst_reg->type = PTR_TO_STACK;
1576 			dst_reg->imm = insn->imm;
1577 			return 0;
1578 		} else if (opcode == BPF_ADD &&
1579 			   BPF_CLASS(insn->code) == BPF_ALU64 &&
1580 			   (dst_reg->type == PTR_TO_PACKET ||
1581 			    (BPF_SRC(insn->code) == BPF_X &&
1582 			     regs[insn->src_reg].type == PTR_TO_PACKET))) {
1583 			/* ptr_to_packet += K|X */
1584 			return check_packet_ptr_add(env, insn);
1585 		} else if (BPF_CLASS(insn->code) == BPF_ALU64 &&
1586 			   dst_reg->type == UNKNOWN_VALUE &&
1587 			   env->allow_ptr_leaks) {
1588 			/* unknown += K|X */
1589 			return evaluate_reg_alu(env, insn);
1590 		} else if (BPF_CLASS(insn->code) == BPF_ALU64 &&
1591 			   dst_reg->type == CONST_IMM &&
1592 			   env->allow_ptr_leaks) {
1593 			/* reg_imm += K|X */
1594 			return evaluate_reg_imm_alu(env, insn);
1595 		} else if (is_pointer_value(env, insn->dst_reg)) {
1596 			verbose("R%d pointer arithmetic prohibited\n",
1597 				insn->dst_reg);
1598 			return -EACCES;
1599 		} else if (BPF_SRC(insn->code) == BPF_X &&
1600 			   is_pointer_value(env, insn->src_reg)) {
1601 			verbose("R%d pointer arithmetic prohibited\n",
1602 				insn->src_reg);
1603 			return -EACCES;
1604 		}
1605 
1606 		/* mark dest operand */
1607 		mark_reg_unknown_value(regs, insn->dst_reg);
1608 	}
1609 
1610 	return 0;
1611 }
1612 
1613 static void find_good_pkt_pointers(struct verifier_env *env,
1614 				   struct reg_state *dst_reg)
1615 {
1616 	struct verifier_state *state = &env->cur_state;
1617 	struct reg_state *regs = state->regs, *reg;
1618 	int i;
1619 	/* r2 = r3;
1620 	 * r2 += 8
1621 	 * if (r2 > pkt_end) goto somewhere
1622 	 * r2 == dst_reg, pkt_end == src_reg,
1623 	 * r2=pkt(id=n,off=8,r=0)
1624 	 * r3=pkt(id=n,off=0,r=0)
1625 	 * find register r3 and mark its range as r3=pkt(id=n,off=0,r=8)
1626 	 * so that range of bytes [r3, r3 + 8) is safe to access
1627 	 */
1628 	for (i = 0; i < MAX_BPF_REG; i++)
1629 		if (regs[i].type == PTR_TO_PACKET && regs[i].id == dst_reg->id)
1630 			regs[i].range = dst_reg->off;
1631 
1632 	for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
1633 		if (state->stack_slot_type[i] != STACK_SPILL)
1634 			continue;
1635 		reg = &state->spilled_regs[i / BPF_REG_SIZE];
1636 		if (reg->type == PTR_TO_PACKET && reg->id == dst_reg->id)
1637 			reg->range = dst_reg->off;
1638 	}
1639 }
1640 
1641 static int check_cond_jmp_op(struct verifier_env *env,
1642 			     struct bpf_insn *insn, int *insn_idx)
1643 {
1644 	struct reg_state *regs = env->cur_state.regs, *dst_reg;
1645 	struct verifier_state *other_branch;
1646 	u8 opcode = BPF_OP(insn->code);
1647 	int err;
1648 
1649 	if (opcode > BPF_EXIT) {
1650 		verbose("invalid BPF_JMP opcode %x\n", opcode);
1651 		return -EINVAL;
1652 	}
1653 
1654 	if (BPF_SRC(insn->code) == BPF_X) {
1655 		if (insn->imm != 0) {
1656 			verbose("BPF_JMP uses reserved fields\n");
1657 			return -EINVAL;
1658 		}
1659 
1660 		/* check src1 operand */
1661 		err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1662 		if (err)
1663 			return err;
1664 
1665 		if (is_pointer_value(env, insn->src_reg)) {
1666 			verbose("R%d pointer comparison prohibited\n",
1667 				insn->src_reg);
1668 			return -EACCES;
1669 		}
1670 	} else {
1671 		if (insn->src_reg != BPF_REG_0) {
1672 			verbose("BPF_JMP uses reserved fields\n");
1673 			return -EINVAL;
1674 		}
1675 	}
1676 
1677 	/* check src2 operand */
1678 	err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1679 	if (err)
1680 		return err;
1681 
1682 	dst_reg = &regs[insn->dst_reg];
1683 
1684 	/* detect if R == 0 where R was initialized to zero earlier */
1685 	if (BPF_SRC(insn->code) == BPF_K &&
1686 	    (opcode == BPF_JEQ || opcode == BPF_JNE) &&
1687 	    dst_reg->type == CONST_IMM && dst_reg->imm == insn->imm) {
1688 		if (opcode == BPF_JEQ) {
1689 			/* if (imm == imm) goto pc+off;
1690 			 * only follow the goto, ignore fall-through
1691 			 */
1692 			*insn_idx += insn->off;
1693 			return 0;
1694 		} else {
1695 			/* if (imm != imm) goto pc+off;
1696 			 * only follow fall-through branch, since
1697 			 * that's where the program will go
1698 			 */
1699 			return 0;
1700 		}
1701 	}
1702 
1703 	other_branch = push_stack(env, *insn_idx + insn->off + 1, *insn_idx);
1704 	if (!other_branch)
1705 		return -EFAULT;
1706 
1707 	/* detect if R == 0 where R is returned value from bpf_map_lookup_elem() */
1708 	if (BPF_SRC(insn->code) == BPF_K &&
1709 	    insn->imm == 0 && (opcode == BPF_JEQ || opcode == BPF_JNE) &&
1710 	    dst_reg->type == PTR_TO_MAP_VALUE_OR_NULL) {
1711 		if (opcode == BPF_JEQ) {
1712 			/* next fallthrough insn can access memory via
1713 			 * this register
1714 			 */
1715 			regs[insn->dst_reg].type = PTR_TO_MAP_VALUE;
1716 			/* branch targer cannot access it, since reg == 0 */
1717 			mark_reg_unknown_value(other_branch->regs,
1718 					       insn->dst_reg);
1719 		} else {
1720 			other_branch->regs[insn->dst_reg].type = PTR_TO_MAP_VALUE;
1721 			mark_reg_unknown_value(regs, insn->dst_reg);
1722 		}
1723 	} else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JGT &&
1724 		   dst_reg->type == PTR_TO_PACKET &&
1725 		   regs[insn->src_reg].type == PTR_TO_PACKET_END) {
1726 		find_good_pkt_pointers(env, dst_reg);
1727 	} else if (is_pointer_value(env, insn->dst_reg)) {
1728 		verbose("R%d pointer comparison prohibited\n", insn->dst_reg);
1729 		return -EACCES;
1730 	}
1731 	if (log_level)
1732 		print_verifier_state(&env->cur_state);
1733 	return 0;
1734 }
1735 
1736 /* return the map pointer stored inside BPF_LD_IMM64 instruction */
1737 static struct bpf_map *ld_imm64_to_map_ptr(struct bpf_insn *insn)
1738 {
1739 	u64 imm64 = ((u64) (u32) insn[0].imm) | ((u64) (u32) insn[1].imm) << 32;
1740 
1741 	return (struct bpf_map *) (unsigned long) imm64;
1742 }
1743 
1744 /* verify BPF_LD_IMM64 instruction */
1745 static int check_ld_imm(struct verifier_env *env, struct bpf_insn *insn)
1746 {
1747 	struct reg_state *regs = env->cur_state.regs;
1748 	int err;
1749 
1750 	if (BPF_SIZE(insn->code) != BPF_DW) {
1751 		verbose("invalid BPF_LD_IMM insn\n");
1752 		return -EINVAL;
1753 	}
1754 	if (insn->off != 0) {
1755 		verbose("BPF_LD_IMM64 uses reserved fields\n");
1756 		return -EINVAL;
1757 	}
1758 
1759 	err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1760 	if (err)
1761 		return err;
1762 
1763 	if (insn->src_reg == 0)
1764 		/* generic move 64-bit immediate into a register */
1765 		return 0;
1766 
1767 	/* replace_map_fd_with_map_ptr() should have caught bad ld_imm64 */
1768 	BUG_ON(insn->src_reg != BPF_PSEUDO_MAP_FD);
1769 
1770 	regs[insn->dst_reg].type = CONST_PTR_TO_MAP;
1771 	regs[insn->dst_reg].map_ptr = ld_imm64_to_map_ptr(insn);
1772 	return 0;
1773 }
1774 
1775 static bool may_access_skb(enum bpf_prog_type type)
1776 {
1777 	switch (type) {
1778 	case BPF_PROG_TYPE_SOCKET_FILTER:
1779 	case BPF_PROG_TYPE_SCHED_CLS:
1780 	case BPF_PROG_TYPE_SCHED_ACT:
1781 		return true;
1782 	default:
1783 		return false;
1784 	}
1785 }
1786 
1787 /* verify safety of LD_ABS|LD_IND instructions:
1788  * - they can only appear in the programs where ctx == skb
1789  * - since they are wrappers of function calls, they scratch R1-R5 registers,
1790  *   preserve R6-R9, and store return value into R0
1791  *
1792  * Implicit input:
1793  *   ctx == skb == R6 == CTX
1794  *
1795  * Explicit input:
1796  *   SRC == any register
1797  *   IMM == 32-bit immediate
1798  *
1799  * Output:
1800  *   R0 - 8/16/32-bit skb data converted to cpu endianness
1801  */
1802 static int check_ld_abs(struct verifier_env *env, struct bpf_insn *insn)
1803 {
1804 	struct reg_state *regs = env->cur_state.regs;
1805 	u8 mode = BPF_MODE(insn->code);
1806 	struct reg_state *reg;
1807 	int i, err;
1808 
1809 	if (!may_access_skb(env->prog->type)) {
1810 		verbose("BPF_LD_[ABS|IND] instructions not allowed for this program type\n");
1811 		return -EINVAL;
1812 	}
1813 
1814 	if (insn->dst_reg != BPF_REG_0 || insn->off != 0 ||
1815 	    BPF_SIZE(insn->code) == BPF_DW ||
1816 	    (mode == BPF_ABS && insn->src_reg != BPF_REG_0)) {
1817 		verbose("BPF_LD_[ABS|IND] uses reserved fields\n");
1818 		return -EINVAL;
1819 	}
1820 
1821 	/* check whether implicit source operand (register R6) is readable */
1822 	err = check_reg_arg(regs, BPF_REG_6, SRC_OP);
1823 	if (err)
1824 		return err;
1825 
1826 	if (regs[BPF_REG_6].type != PTR_TO_CTX) {
1827 		verbose("at the time of BPF_LD_ABS|IND R6 != pointer to skb\n");
1828 		return -EINVAL;
1829 	}
1830 
1831 	if (mode == BPF_IND) {
1832 		/* check explicit source operand */
1833 		err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1834 		if (err)
1835 			return err;
1836 	}
1837 
1838 	/* reset caller saved regs to unreadable */
1839 	for (i = 0; i < CALLER_SAVED_REGS; i++) {
1840 		reg = regs + caller_saved[i];
1841 		reg->type = NOT_INIT;
1842 		reg->imm = 0;
1843 	}
1844 
1845 	/* mark destination R0 register as readable, since it contains
1846 	 * the value fetched from the packet
1847 	 */
1848 	regs[BPF_REG_0].type = UNKNOWN_VALUE;
1849 	return 0;
1850 }
1851 
1852 /* non-recursive DFS pseudo code
1853  * 1  procedure DFS-iterative(G,v):
1854  * 2      label v as discovered
1855  * 3      let S be a stack
1856  * 4      S.push(v)
1857  * 5      while S is not empty
1858  * 6            t <- S.pop()
1859  * 7            if t is what we're looking for:
1860  * 8                return t
1861  * 9            for all edges e in G.adjacentEdges(t) do
1862  * 10               if edge e is already labelled
1863  * 11                   continue with the next edge
1864  * 12               w <- G.adjacentVertex(t,e)
1865  * 13               if vertex w is not discovered and not explored
1866  * 14                   label e as tree-edge
1867  * 15                   label w as discovered
1868  * 16                   S.push(w)
1869  * 17                   continue at 5
1870  * 18               else if vertex w is discovered
1871  * 19                   label e as back-edge
1872  * 20               else
1873  * 21                   // vertex w is explored
1874  * 22                   label e as forward- or cross-edge
1875  * 23           label t as explored
1876  * 24           S.pop()
1877  *
1878  * convention:
1879  * 0x10 - discovered
1880  * 0x11 - discovered and fall-through edge labelled
1881  * 0x12 - discovered and fall-through and branch edges labelled
1882  * 0x20 - explored
1883  */
1884 
1885 enum {
1886 	DISCOVERED = 0x10,
1887 	EXPLORED = 0x20,
1888 	FALLTHROUGH = 1,
1889 	BRANCH = 2,
1890 };
1891 
1892 #define STATE_LIST_MARK ((struct verifier_state_list *) -1L)
1893 
1894 static int *insn_stack;	/* stack of insns to process */
1895 static int cur_stack;	/* current stack index */
1896 static int *insn_state;
1897 
1898 /* t, w, e - match pseudo-code above:
1899  * t - index of current instruction
1900  * w - next instruction
1901  * e - edge
1902  */
1903 static int push_insn(int t, int w, int e, struct verifier_env *env)
1904 {
1905 	if (e == FALLTHROUGH && insn_state[t] >= (DISCOVERED | FALLTHROUGH))
1906 		return 0;
1907 
1908 	if (e == BRANCH && insn_state[t] >= (DISCOVERED | BRANCH))
1909 		return 0;
1910 
1911 	if (w < 0 || w >= env->prog->len) {
1912 		verbose("jump out of range from insn %d to %d\n", t, w);
1913 		return -EINVAL;
1914 	}
1915 
1916 	if (e == BRANCH)
1917 		/* mark branch target for state pruning */
1918 		env->explored_states[w] = STATE_LIST_MARK;
1919 
1920 	if (insn_state[w] == 0) {
1921 		/* tree-edge */
1922 		insn_state[t] = DISCOVERED | e;
1923 		insn_state[w] = DISCOVERED;
1924 		if (cur_stack >= env->prog->len)
1925 			return -E2BIG;
1926 		insn_stack[cur_stack++] = w;
1927 		return 1;
1928 	} else if ((insn_state[w] & 0xF0) == DISCOVERED) {
1929 		verbose("back-edge from insn %d to %d\n", t, w);
1930 		return -EINVAL;
1931 	} else if (insn_state[w] == EXPLORED) {
1932 		/* forward- or cross-edge */
1933 		insn_state[t] = DISCOVERED | e;
1934 	} else {
1935 		verbose("insn state internal bug\n");
1936 		return -EFAULT;
1937 	}
1938 	return 0;
1939 }
1940 
1941 /* non-recursive depth-first-search to detect loops in BPF program
1942  * loop == back-edge in directed graph
1943  */
1944 static int check_cfg(struct verifier_env *env)
1945 {
1946 	struct bpf_insn *insns = env->prog->insnsi;
1947 	int insn_cnt = env->prog->len;
1948 	int ret = 0;
1949 	int i, t;
1950 
1951 	insn_state = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
1952 	if (!insn_state)
1953 		return -ENOMEM;
1954 
1955 	insn_stack = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
1956 	if (!insn_stack) {
1957 		kfree(insn_state);
1958 		return -ENOMEM;
1959 	}
1960 
1961 	insn_state[0] = DISCOVERED; /* mark 1st insn as discovered */
1962 	insn_stack[0] = 0; /* 0 is the first instruction */
1963 	cur_stack = 1;
1964 
1965 peek_stack:
1966 	if (cur_stack == 0)
1967 		goto check_state;
1968 	t = insn_stack[cur_stack - 1];
1969 
1970 	if (BPF_CLASS(insns[t].code) == BPF_JMP) {
1971 		u8 opcode = BPF_OP(insns[t].code);
1972 
1973 		if (opcode == BPF_EXIT) {
1974 			goto mark_explored;
1975 		} else if (opcode == BPF_CALL) {
1976 			ret = push_insn(t, t + 1, FALLTHROUGH, env);
1977 			if (ret == 1)
1978 				goto peek_stack;
1979 			else if (ret < 0)
1980 				goto err_free;
1981 			if (t + 1 < insn_cnt)
1982 				env->explored_states[t + 1] = STATE_LIST_MARK;
1983 		} else if (opcode == BPF_JA) {
1984 			if (BPF_SRC(insns[t].code) != BPF_K) {
1985 				ret = -EINVAL;
1986 				goto err_free;
1987 			}
1988 			/* unconditional jump with single edge */
1989 			ret = push_insn(t, t + insns[t].off + 1,
1990 					FALLTHROUGH, env);
1991 			if (ret == 1)
1992 				goto peek_stack;
1993 			else if (ret < 0)
1994 				goto err_free;
1995 			/* tell verifier to check for equivalent states
1996 			 * after every call and jump
1997 			 */
1998 			if (t + 1 < insn_cnt)
1999 				env->explored_states[t + 1] = STATE_LIST_MARK;
2000 		} else {
2001 			/* conditional jump with two edges */
2002 			ret = push_insn(t, t + 1, FALLTHROUGH, env);
2003 			if (ret == 1)
2004 				goto peek_stack;
2005 			else if (ret < 0)
2006 				goto err_free;
2007 
2008 			ret = push_insn(t, t + insns[t].off + 1, BRANCH, env);
2009 			if (ret == 1)
2010 				goto peek_stack;
2011 			else if (ret < 0)
2012 				goto err_free;
2013 		}
2014 	} else {
2015 		/* all other non-branch instructions with single
2016 		 * fall-through edge
2017 		 */
2018 		ret = push_insn(t, t + 1, FALLTHROUGH, env);
2019 		if (ret == 1)
2020 			goto peek_stack;
2021 		else if (ret < 0)
2022 			goto err_free;
2023 	}
2024 
2025 mark_explored:
2026 	insn_state[t] = EXPLORED;
2027 	if (cur_stack-- <= 0) {
2028 		verbose("pop stack internal bug\n");
2029 		ret = -EFAULT;
2030 		goto err_free;
2031 	}
2032 	goto peek_stack;
2033 
2034 check_state:
2035 	for (i = 0; i < insn_cnt; i++) {
2036 		if (insn_state[i] != EXPLORED) {
2037 			verbose("unreachable insn %d\n", i);
2038 			ret = -EINVAL;
2039 			goto err_free;
2040 		}
2041 	}
2042 	ret = 0; /* cfg looks good */
2043 
2044 err_free:
2045 	kfree(insn_state);
2046 	kfree(insn_stack);
2047 	return ret;
2048 }
2049 
2050 /* the following conditions reduce the number of explored insns
2051  * from ~140k to ~80k for ultra large programs that use a lot of ptr_to_packet
2052  */
2053 static bool compare_ptrs_to_packet(struct reg_state *old, struct reg_state *cur)
2054 {
2055 	if (old->id != cur->id)
2056 		return false;
2057 
2058 	/* old ptr_to_packet is more conservative, since it allows smaller
2059 	 * range. Ex:
2060 	 * old(off=0,r=10) is equal to cur(off=0,r=20), because
2061 	 * old(off=0,r=10) means that with range=10 the verifier proceeded
2062 	 * further and found no issues with the program. Now we're in the same
2063 	 * spot with cur(off=0,r=20), so we're safe too, since anything further
2064 	 * will only be looking at most 10 bytes after this pointer.
2065 	 */
2066 	if (old->off == cur->off && old->range < cur->range)
2067 		return true;
2068 
2069 	/* old(off=20,r=10) is equal to cur(off=22,re=22 or 5 or 0)
2070 	 * since both cannot be used for packet access and safe(old)
2071 	 * pointer has smaller off that could be used for further
2072 	 * 'if (ptr > data_end)' check
2073 	 * Ex:
2074 	 * old(off=20,r=10) and cur(off=22,r=22) and cur(off=22,r=0) mean
2075 	 * that we cannot access the packet.
2076 	 * The safe range is:
2077 	 * [ptr, ptr + range - off)
2078 	 * so whenever off >=range, it means no safe bytes from this pointer.
2079 	 * When comparing old->off <= cur->off, it means that older code
2080 	 * went with smaller offset and that offset was later
2081 	 * used to figure out the safe range after 'if (ptr > data_end)' check
2082 	 * Say, 'old' state was explored like:
2083 	 * ... R3(off=0, r=0)
2084 	 * R4 = R3 + 20
2085 	 * ... now R4(off=20,r=0)  <-- here
2086 	 * if (R4 > data_end)
2087 	 * ... R4(off=20,r=20), R3(off=0,r=20) and R3 can be used to access.
2088 	 * ... the code further went all the way to bpf_exit.
2089 	 * Now the 'cur' state at the mark 'here' has R4(off=30,r=0).
2090 	 * old_R4(off=20,r=0) equal to cur_R4(off=30,r=0), since if the verifier
2091 	 * goes further, such cur_R4 will give larger safe packet range after
2092 	 * 'if (R4 > data_end)' and all further insn were already good with r=20,
2093 	 * so they will be good with r=30 and we can prune the search.
2094 	 */
2095 	if (old->off <= cur->off &&
2096 	    old->off >= old->range && cur->off >= cur->range)
2097 		return true;
2098 
2099 	return false;
2100 }
2101 
2102 /* compare two verifier states
2103  *
2104  * all states stored in state_list are known to be valid, since
2105  * verifier reached 'bpf_exit' instruction through them
2106  *
2107  * this function is called when verifier exploring different branches of
2108  * execution popped from the state stack. If it sees an old state that has
2109  * more strict register state and more strict stack state then this execution
2110  * branch doesn't need to be explored further, since verifier already
2111  * concluded that more strict state leads to valid finish.
2112  *
2113  * Therefore two states are equivalent if register state is more conservative
2114  * and explored stack state is more conservative than the current one.
2115  * Example:
2116  *       explored                   current
2117  * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC)
2118  * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC)
2119  *
2120  * In other words if current stack state (one being explored) has more
2121  * valid slots than old one that already passed validation, it means
2122  * the verifier can stop exploring and conclude that current state is valid too
2123  *
2124  * Similarly with registers. If explored state has register type as invalid
2125  * whereas register type in current state is meaningful, it means that
2126  * the current state will reach 'bpf_exit' instruction safely
2127  */
2128 static bool states_equal(struct verifier_state *old, struct verifier_state *cur)
2129 {
2130 	struct reg_state *rold, *rcur;
2131 	int i;
2132 
2133 	for (i = 0; i < MAX_BPF_REG; i++) {
2134 		rold = &old->regs[i];
2135 		rcur = &cur->regs[i];
2136 
2137 		if (memcmp(rold, rcur, sizeof(*rold)) == 0)
2138 			continue;
2139 
2140 		if (rold->type == NOT_INIT ||
2141 		    (rold->type == UNKNOWN_VALUE && rcur->type != NOT_INIT))
2142 			continue;
2143 
2144 		if (rold->type == PTR_TO_PACKET && rcur->type == PTR_TO_PACKET &&
2145 		    compare_ptrs_to_packet(rold, rcur))
2146 			continue;
2147 
2148 		return false;
2149 	}
2150 
2151 	for (i = 0; i < MAX_BPF_STACK; i++) {
2152 		if (old->stack_slot_type[i] == STACK_INVALID)
2153 			continue;
2154 		if (old->stack_slot_type[i] != cur->stack_slot_type[i])
2155 			/* Ex: old explored (safe) state has STACK_SPILL in
2156 			 * this stack slot, but current has has STACK_MISC ->
2157 			 * this verifier states are not equivalent,
2158 			 * return false to continue verification of this path
2159 			 */
2160 			return false;
2161 		if (i % BPF_REG_SIZE)
2162 			continue;
2163 		if (memcmp(&old->spilled_regs[i / BPF_REG_SIZE],
2164 			   &cur->spilled_regs[i / BPF_REG_SIZE],
2165 			   sizeof(old->spilled_regs[0])))
2166 			/* when explored and current stack slot types are
2167 			 * the same, check that stored pointers types
2168 			 * are the same as well.
2169 			 * Ex: explored safe path could have stored
2170 			 * (struct reg_state) {.type = PTR_TO_STACK, .imm = -8}
2171 			 * but current path has stored:
2172 			 * (struct reg_state) {.type = PTR_TO_STACK, .imm = -16}
2173 			 * such verifier states are not equivalent.
2174 			 * return false to continue verification of this path
2175 			 */
2176 			return false;
2177 		else
2178 			continue;
2179 	}
2180 	return true;
2181 }
2182 
2183 static int is_state_visited(struct verifier_env *env, int insn_idx)
2184 {
2185 	struct verifier_state_list *new_sl;
2186 	struct verifier_state_list *sl;
2187 
2188 	sl = env->explored_states[insn_idx];
2189 	if (!sl)
2190 		/* this 'insn_idx' instruction wasn't marked, so we will not
2191 		 * be doing state search here
2192 		 */
2193 		return 0;
2194 
2195 	while (sl != STATE_LIST_MARK) {
2196 		if (states_equal(&sl->state, &env->cur_state))
2197 			/* reached equivalent register/stack state,
2198 			 * prune the search
2199 			 */
2200 			return 1;
2201 		sl = sl->next;
2202 	}
2203 
2204 	/* there were no equivalent states, remember current one.
2205 	 * technically the current state is not proven to be safe yet,
2206 	 * but it will either reach bpf_exit (which means it's safe) or
2207 	 * it will be rejected. Since there are no loops, we won't be
2208 	 * seeing this 'insn_idx' instruction again on the way to bpf_exit
2209 	 */
2210 	new_sl = kmalloc(sizeof(struct verifier_state_list), GFP_USER);
2211 	if (!new_sl)
2212 		return -ENOMEM;
2213 
2214 	/* add new state to the head of linked list */
2215 	memcpy(&new_sl->state, &env->cur_state, sizeof(env->cur_state));
2216 	new_sl->next = env->explored_states[insn_idx];
2217 	env->explored_states[insn_idx] = new_sl;
2218 	return 0;
2219 }
2220 
2221 static int do_check(struct verifier_env *env)
2222 {
2223 	struct verifier_state *state = &env->cur_state;
2224 	struct bpf_insn *insns = env->prog->insnsi;
2225 	struct reg_state *regs = state->regs;
2226 	int insn_cnt = env->prog->len;
2227 	int insn_idx, prev_insn_idx = 0;
2228 	int insn_processed = 0;
2229 	bool do_print_state = false;
2230 
2231 	init_reg_state(regs);
2232 	insn_idx = 0;
2233 	for (;;) {
2234 		struct bpf_insn *insn;
2235 		u8 class;
2236 		int err;
2237 
2238 		if (insn_idx >= insn_cnt) {
2239 			verbose("invalid insn idx %d insn_cnt %d\n",
2240 				insn_idx, insn_cnt);
2241 			return -EFAULT;
2242 		}
2243 
2244 		insn = &insns[insn_idx];
2245 		class = BPF_CLASS(insn->code);
2246 
2247 		if (++insn_processed > BPF_COMPLEXITY_LIMIT_INSNS) {
2248 			verbose("BPF program is too large. Proccessed %d insn\n",
2249 				insn_processed);
2250 			return -E2BIG;
2251 		}
2252 
2253 		err = is_state_visited(env, insn_idx);
2254 		if (err < 0)
2255 			return err;
2256 		if (err == 1) {
2257 			/* found equivalent state, can prune the search */
2258 			if (log_level) {
2259 				if (do_print_state)
2260 					verbose("\nfrom %d to %d: safe\n",
2261 						prev_insn_idx, insn_idx);
2262 				else
2263 					verbose("%d: safe\n", insn_idx);
2264 			}
2265 			goto process_bpf_exit;
2266 		}
2267 
2268 		if (log_level && do_print_state) {
2269 			verbose("\nfrom %d to %d:", prev_insn_idx, insn_idx);
2270 			print_verifier_state(&env->cur_state);
2271 			do_print_state = false;
2272 		}
2273 
2274 		if (log_level) {
2275 			verbose("%d: ", insn_idx);
2276 			print_bpf_insn(insn);
2277 		}
2278 
2279 		if (class == BPF_ALU || class == BPF_ALU64) {
2280 			err = check_alu_op(env, insn);
2281 			if (err)
2282 				return err;
2283 
2284 		} else if (class == BPF_LDX) {
2285 			enum bpf_reg_type src_reg_type;
2286 
2287 			/* check for reserved fields is already done */
2288 
2289 			/* check src operand */
2290 			err = check_reg_arg(regs, insn->src_reg, SRC_OP);
2291 			if (err)
2292 				return err;
2293 
2294 			err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK);
2295 			if (err)
2296 				return err;
2297 
2298 			src_reg_type = regs[insn->src_reg].type;
2299 
2300 			/* check that memory (src_reg + off) is readable,
2301 			 * the state of dst_reg will be updated by this func
2302 			 */
2303 			err = check_mem_access(env, insn->src_reg, insn->off,
2304 					       BPF_SIZE(insn->code), BPF_READ,
2305 					       insn->dst_reg);
2306 			if (err)
2307 				return err;
2308 
2309 			if (BPF_SIZE(insn->code) != BPF_W) {
2310 				insn_idx++;
2311 				continue;
2312 			}
2313 
2314 			if (insn->imm == 0) {
2315 				/* saw a valid insn
2316 				 * dst_reg = *(u32 *)(src_reg + off)
2317 				 * use reserved 'imm' field to mark this insn
2318 				 */
2319 				insn->imm = src_reg_type;
2320 
2321 			} else if (src_reg_type != insn->imm &&
2322 				   (src_reg_type == PTR_TO_CTX ||
2323 				    insn->imm == PTR_TO_CTX)) {
2324 				/* ABuser program is trying to use the same insn
2325 				 * dst_reg = *(u32*) (src_reg + off)
2326 				 * with different pointer types:
2327 				 * src_reg == ctx in one branch and
2328 				 * src_reg == stack|map in some other branch.
2329 				 * Reject it.
2330 				 */
2331 				verbose("same insn cannot be used with different pointers\n");
2332 				return -EINVAL;
2333 			}
2334 
2335 		} else if (class == BPF_STX) {
2336 			enum bpf_reg_type dst_reg_type;
2337 
2338 			if (BPF_MODE(insn->code) == BPF_XADD) {
2339 				err = check_xadd(env, insn);
2340 				if (err)
2341 					return err;
2342 				insn_idx++;
2343 				continue;
2344 			}
2345 
2346 			/* check src1 operand */
2347 			err = check_reg_arg(regs, insn->src_reg, SRC_OP);
2348 			if (err)
2349 				return err;
2350 			/* check src2 operand */
2351 			err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
2352 			if (err)
2353 				return err;
2354 
2355 			dst_reg_type = regs[insn->dst_reg].type;
2356 
2357 			/* check that memory (dst_reg + off) is writeable */
2358 			err = check_mem_access(env, insn->dst_reg, insn->off,
2359 					       BPF_SIZE(insn->code), BPF_WRITE,
2360 					       insn->src_reg);
2361 			if (err)
2362 				return err;
2363 
2364 			if (insn->imm == 0) {
2365 				insn->imm = dst_reg_type;
2366 			} else if (dst_reg_type != insn->imm &&
2367 				   (dst_reg_type == PTR_TO_CTX ||
2368 				    insn->imm == PTR_TO_CTX)) {
2369 				verbose("same insn cannot be used with different pointers\n");
2370 				return -EINVAL;
2371 			}
2372 
2373 		} else if (class == BPF_ST) {
2374 			if (BPF_MODE(insn->code) != BPF_MEM ||
2375 			    insn->src_reg != BPF_REG_0) {
2376 				verbose("BPF_ST uses reserved fields\n");
2377 				return -EINVAL;
2378 			}
2379 			/* check src operand */
2380 			err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
2381 			if (err)
2382 				return err;
2383 
2384 			/* check that memory (dst_reg + off) is writeable */
2385 			err = check_mem_access(env, insn->dst_reg, insn->off,
2386 					       BPF_SIZE(insn->code), BPF_WRITE,
2387 					       -1);
2388 			if (err)
2389 				return err;
2390 
2391 		} else if (class == BPF_JMP) {
2392 			u8 opcode = BPF_OP(insn->code);
2393 
2394 			if (opcode == BPF_CALL) {
2395 				if (BPF_SRC(insn->code) != BPF_K ||
2396 				    insn->off != 0 ||
2397 				    insn->src_reg != BPF_REG_0 ||
2398 				    insn->dst_reg != BPF_REG_0) {
2399 					verbose("BPF_CALL uses reserved fields\n");
2400 					return -EINVAL;
2401 				}
2402 
2403 				err = check_call(env, insn->imm);
2404 				if (err)
2405 					return err;
2406 
2407 			} else if (opcode == BPF_JA) {
2408 				if (BPF_SRC(insn->code) != BPF_K ||
2409 				    insn->imm != 0 ||
2410 				    insn->src_reg != BPF_REG_0 ||
2411 				    insn->dst_reg != BPF_REG_0) {
2412 					verbose("BPF_JA uses reserved fields\n");
2413 					return -EINVAL;
2414 				}
2415 
2416 				insn_idx += insn->off + 1;
2417 				continue;
2418 
2419 			} else if (opcode == BPF_EXIT) {
2420 				if (BPF_SRC(insn->code) != BPF_K ||
2421 				    insn->imm != 0 ||
2422 				    insn->src_reg != BPF_REG_0 ||
2423 				    insn->dst_reg != BPF_REG_0) {
2424 					verbose("BPF_EXIT uses reserved fields\n");
2425 					return -EINVAL;
2426 				}
2427 
2428 				/* eBPF calling convetion is such that R0 is used
2429 				 * to return the value from eBPF program.
2430 				 * Make sure that it's readable at this time
2431 				 * of bpf_exit, which means that program wrote
2432 				 * something into it earlier
2433 				 */
2434 				err = check_reg_arg(regs, BPF_REG_0, SRC_OP);
2435 				if (err)
2436 					return err;
2437 
2438 				if (is_pointer_value(env, BPF_REG_0)) {
2439 					verbose("R0 leaks addr as return value\n");
2440 					return -EACCES;
2441 				}
2442 
2443 process_bpf_exit:
2444 				insn_idx = pop_stack(env, &prev_insn_idx);
2445 				if (insn_idx < 0) {
2446 					break;
2447 				} else {
2448 					do_print_state = true;
2449 					continue;
2450 				}
2451 			} else {
2452 				err = check_cond_jmp_op(env, insn, &insn_idx);
2453 				if (err)
2454 					return err;
2455 			}
2456 		} else if (class == BPF_LD) {
2457 			u8 mode = BPF_MODE(insn->code);
2458 
2459 			if (mode == BPF_ABS || mode == BPF_IND) {
2460 				err = check_ld_abs(env, insn);
2461 				if (err)
2462 					return err;
2463 
2464 			} else if (mode == BPF_IMM) {
2465 				err = check_ld_imm(env, insn);
2466 				if (err)
2467 					return err;
2468 
2469 				insn_idx++;
2470 			} else {
2471 				verbose("invalid BPF_LD mode\n");
2472 				return -EINVAL;
2473 			}
2474 		} else {
2475 			verbose("unknown insn class %d\n", class);
2476 			return -EINVAL;
2477 		}
2478 
2479 		insn_idx++;
2480 	}
2481 
2482 	verbose("processed %d insns\n", insn_processed);
2483 	return 0;
2484 }
2485 
2486 /* look for pseudo eBPF instructions that access map FDs and
2487  * replace them with actual map pointers
2488  */
2489 static int replace_map_fd_with_map_ptr(struct verifier_env *env)
2490 {
2491 	struct bpf_insn *insn = env->prog->insnsi;
2492 	int insn_cnt = env->prog->len;
2493 	int i, j;
2494 
2495 	for (i = 0; i < insn_cnt; i++, insn++) {
2496 		if (BPF_CLASS(insn->code) == BPF_LDX &&
2497 		    (BPF_MODE(insn->code) != BPF_MEM || insn->imm != 0)) {
2498 			verbose("BPF_LDX uses reserved fields\n");
2499 			return -EINVAL;
2500 		}
2501 
2502 		if (BPF_CLASS(insn->code) == BPF_STX &&
2503 		    ((BPF_MODE(insn->code) != BPF_MEM &&
2504 		      BPF_MODE(insn->code) != BPF_XADD) || insn->imm != 0)) {
2505 			verbose("BPF_STX uses reserved fields\n");
2506 			return -EINVAL;
2507 		}
2508 
2509 		if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW)) {
2510 			struct bpf_map *map;
2511 			struct fd f;
2512 
2513 			if (i == insn_cnt - 1 || insn[1].code != 0 ||
2514 			    insn[1].dst_reg != 0 || insn[1].src_reg != 0 ||
2515 			    insn[1].off != 0) {
2516 				verbose("invalid bpf_ld_imm64 insn\n");
2517 				return -EINVAL;
2518 			}
2519 
2520 			if (insn->src_reg == 0)
2521 				/* valid generic load 64-bit imm */
2522 				goto next_insn;
2523 
2524 			if (insn->src_reg != BPF_PSEUDO_MAP_FD) {
2525 				verbose("unrecognized bpf_ld_imm64 insn\n");
2526 				return -EINVAL;
2527 			}
2528 
2529 			f = fdget(insn->imm);
2530 			map = __bpf_map_get(f);
2531 			if (IS_ERR(map)) {
2532 				verbose("fd %d is not pointing to valid bpf_map\n",
2533 					insn->imm);
2534 				return PTR_ERR(map);
2535 			}
2536 
2537 			/* store map pointer inside BPF_LD_IMM64 instruction */
2538 			insn[0].imm = (u32) (unsigned long) map;
2539 			insn[1].imm = ((u64) (unsigned long) map) >> 32;
2540 
2541 			/* check whether we recorded this map already */
2542 			for (j = 0; j < env->used_map_cnt; j++)
2543 				if (env->used_maps[j] == map) {
2544 					fdput(f);
2545 					goto next_insn;
2546 				}
2547 
2548 			if (env->used_map_cnt >= MAX_USED_MAPS) {
2549 				fdput(f);
2550 				return -E2BIG;
2551 			}
2552 
2553 			/* hold the map. If the program is rejected by verifier,
2554 			 * the map will be released by release_maps() or it
2555 			 * will be used by the valid program until it's unloaded
2556 			 * and all maps are released in free_bpf_prog_info()
2557 			 */
2558 			map = bpf_map_inc(map, false);
2559 			if (IS_ERR(map)) {
2560 				fdput(f);
2561 				return PTR_ERR(map);
2562 			}
2563 			env->used_maps[env->used_map_cnt++] = map;
2564 
2565 			fdput(f);
2566 next_insn:
2567 			insn++;
2568 			i++;
2569 		}
2570 	}
2571 
2572 	/* now all pseudo BPF_LD_IMM64 instructions load valid
2573 	 * 'struct bpf_map *' into a register instead of user map_fd.
2574 	 * These pointers will be used later by verifier to validate map access.
2575 	 */
2576 	return 0;
2577 }
2578 
2579 /* drop refcnt of maps used by the rejected program */
2580 static void release_maps(struct verifier_env *env)
2581 {
2582 	int i;
2583 
2584 	for (i = 0; i < env->used_map_cnt; i++)
2585 		bpf_map_put(env->used_maps[i]);
2586 }
2587 
2588 /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */
2589 static void convert_pseudo_ld_imm64(struct verifier_env *env)
2590 {
2591 	struct bpf_insn *insn = env->prog->insnsi;
2592 	int insn_cnt = env->prog->len;
2593 	int i;
2594 
2595 	for (i = 0; i < insn_cnt; i++, insn++)
2596 		if (insn->code == (BPF_LD | BPF_IMM | BPF_DW))
2597 			insn->src_reg = 0;
2598 }
2599 
2600 /* convert load instructions that access fields of 'struct __sk_buff'
2601  * into sequence of instructions that access fields of 'struct sk_buff'
2602  */
2603 static int convert_ctx_accesses(struct verifier_env *env)
2604 {
2605 	struct bpf_insn *insn = env->prog->insnsi;
2606 	int insn_cnt = env->prog->len;
2607 	struct bpf_insn insn_buf[16];
2608 	struct bpf_prog *new_prog;
2609 	enum bpf_access_type type;
2610 	int i;
2611 
2612 	if (!env->prog->aux->ops->convert_ctx_access)
2613 		return 0;
2614 
2615 	for (i = 0; i < insn_cnt; i++, insn++) {
2616 		u32 insn_delta, cnt;
2617 
2618 		if (insn->code == (BPF_LDX | BPF_MEM | BPF_W))
2619 			type = BPF_READ;
2620 		else if (insn->code == (BPF_STX | BPF_MEM | BPF_W))
2621 			type = BPF_WRITE;
2622 		else
2623 			continue;
2624 
2625 		if (insn->imm != PTR_TO_CTX) {
2626 			/* clear internal mark */
2627 			insn->imm = 0;
2628 			continue;
2629 		}
2630 
2631 		cnt = env->prog->aux->ops->
2632 			convert_ctx_access(type, insn->dst_reg, insn->src_reg,
2633 					   insn->off, insn_buf, env->prog);
2634 		if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) {
2635 			verbose("bpf verifier is misconfigured\n");
2636 			return -EINVAL;
2637 		}
2638 
2639 		new_prog = bpf_patch_insn_single(env->prog, i, insn_buf, cnt);
2640 		if (!new_prog)
2641 			return -ENOMEM;
2642 
2643 		insn_delta = cnt - 1;
2644 
2645 		/* keep walking new program and skip insns we just inserted */
2646 		env->prog = new_prog;
2647 		insn      = new_prog->insnsi + i + insn_delta;
2648 
2649 		insn_cnt += insn_delta;
2650 		i        += insn_delta;
2651 	}
2652 
2653 	return 0;
2654 }
2655 
2656 static void free_states(struct verifier_env *env)
2657 {
2658 	struct verifier_state_list *sl, *sln;
2659 	int i;
2660 
2661 	if (!env->explored_states)
2662 		return;
2663 
2664 	for (i = 0; i < env->prog->len; i++) {
2665 		sl = env->explored_states[i];
2666 
2667 		if (sl)
2668 			while (sl != STATE_LIST_MARK) {
2669 				sln = sl->next;
2670 				kfree(sl);
2671 				sl = sln;
2672 			}
2673 	}
2674 
2675 	kfree(env->explored_states);
2676 }
2677 
2678 int bpf_check(struct bpf_prog **prog, union bpf_attr *attr)
2679 {
2680 	char __user *log_ubuf = NULL;
2681 	struct verifier_env *env;
2682 	int ret = -EINVAL;
2683 
2684 	if ((*prog)->len <= 0 || (*prog)->len > BPF_MAXINSNS)
2685 		return -E2BIG;
2686 
2687 	/* 'struct verifier_env' can be global, but since it's not small,
2688 	 * allocate/free it every time bpf_check() is called
2689 	 */
2690 	env = kzalloc(sizeof(struct verifier_env), GFP_KERNEL);
2691 	if (!env)
2692 		return -ENOMEM;
2693 
2694 	env->prog = *prog;
2695 
2696 	/* grab the mutex to protect few globals used by verifier */
2697 	mutex_lock(&bpf_verifier_lock);
2698 
2699 	if (attr->log_level || attr->log_buf || attr->log_size) {
2700 		/* user requested verbose verifier output
2701 		 * and supplied buffer to store the verification trace
2702 		 */
2703 		log_level = attr->log_level;
2704 		log_ubuf = (char __user *) (unsigned long) attr->log_buf;
2705 		log_size = attr->log_size;
2706 		log_len = 0;
2707 
2708 		ret = -EINVAL;
2709 		/* log_* values have to be sane */
2710 		if (log_size < 128 || log_size > UINT_MAX >> 8 ||
2711 		    log_level == 0 || log_ubuf == NULL)
2712 			goto free_env;
2713 
2714 		ret = -ENOMEM;
2715 		log_buf = vmalloc(log_size);
2716 		if (!log_buf)
2717 			goto free_env;
2718 	} else {
2719 		log_level = 0;
2720 	}
2721 
2722 	ret = replace_map_fd_with_map_ptr(env);
2723 	if (ret < 0)
2724 		goto skip_full_check;
2725 
2726 	env->explored_states = kcalloc(env->prog->len,
2727 				       sizeof(struct verifier_state_list *),
2728 				       GFP_USER);
2729 	ret = -ENOMEM;
2730 	if (!env->explored_states)
2731 		goto skip_full_check;
2732 
2733 	ret = check_cfg(env);
2734 	if (ret < 0)
2735 		goto skip_full_check;
2736 
2737 	env->allow_ptr_leaks = capable(CAP_SYS_ADMIN);
2738 
2739 	ret = do_check(env);
2740 
2741 skip_full_check:
2742 	while (pop_stack(env, NULL) >= 0);
2743 	free_states(env);
2744 
2745 	if (ret == 0)
2746 		/* program is valid, convert *(u32*)(ctx + off) accesses */
2747 		ret = convert_ctx_accesses(env);
2748 
2749 	if (log_level && log_len >= log_size - 1) {
2750 		BUG_ON(log_len >= log_size);
2751 		/* verifier log exceeded user supplied buffer */
2752 		ret = -ENOSPC;
2753 		/* fall through to return what was recorded */
2754 	}
2755 
2756 	/* copy verifier log back to user space including trailing zero */
2757 	if (log_level && copy_to_user(log_ubuf, log_buf, log_len + 1) != 0) {
2758 		ret = -EFAULT;
2759 		goto free_log_buf;
2760 	}
2761 
2762 	if (ret == 0 && env->used_map_cnt) {
2763 		/* if program passed verifier, update used_maps in bpf_prog_info */
2764 		env->prog->aux->used_maps = kmalloc_array(env->used_map_cnt,
2765 							  sizeof(env->used_maps[0]),
2766 							  GFP_KERNEL);
2767 
2768 		if (!env->prog->aux->used_maps) {
2769 			ret = -ENOMEM;
2770 			goto free_log_buf;
2771 		}
2772 
2773 		memcpy(env->prog->aux->used_maps, env->used_maps,
2774 		       sizeof(env->used_maps[0]) * env->used_map_cnt);
2775 		env->prog->aux->used_map_cnt = env->used_map_cnt;
2776 
2777 		/* program is valid. Convert pseudo bpf_ld_imm64 into generic
2778 		 * bpf_ld_imm64 instructions
2779 		 */
2780 		convert_pseudo_ld_imm64(env);
2781 	}
2782 
2783 free_log_buf:
2784 	if (log_level)
2785 		vfree(log_buf);
2786 free_env:
2787 	if (!env->prog->aux->used_maps)
2788 		/* if we didn't copy map pointers into bpf_prog_info, release
2789 		 * them now. Otherwise free_bpf_prog_info() will release them.
2790 		 */
2791 		release_maps(env);
2792 	*prog = env->prog;
2793 	kfree(env);
2794 	mutex_unlock(&bpf_verifier_lock);
2795 	return ret;
2796 }
2797