xref: /openbmc/linux/kernel/bpf/verifier.c (revision 82e6fdd6)
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 #include <linux/bsearch.h>
24 #include <linux/sort.h>
25 
26 #include "disasm.h"
27 
28 static const struct bpf_verifier_ops * const bpf_verifier_ops[] = {
29 #define BPF_PROG_TYPE(_id, _name) \
30 	[_id] = & _name ## _verifier_ops,
31 #define BPF_MAP_TYPE(_id, _ops)
32 #include <linux/bpf_types.h>
33 #undef BPF_PROG_TYPE
34 #undef BPF_MAP_TYPE
35 };
36 
37 /* bpf_check() is a static code analyzer that walks eBPF program
38  * instruction by instruction and updates register/stack state.
39  * All paths of conditional branches are analyzed until 'bpf_exit' insn.
40  *
41  * The first pass is depth-first-search to check that the program is a DAG.
42  * It rejects the following programs:
43  * - larger than BPF_MAXINSNS insns
44  * - if loop is present (detected via back-edge)
45  * - unreachable insns exist (shouldn't be a forest. program = one function)
46  * - out of bounds or malformed jumps
47  * The second pass is all possible path descent from the 1st insn.
48  * Since it's analyzing all pathes through the program, the length of the
49  * analysis is limited to 64k insn, which may be hit even if total number of
50  * insn is less then 4K, but there are too many branches that change stack/regs.
51  * Number of 'branches to be analyzed' is limited to 1k
52  *
53  * On entry to each instruction, each register has a type, and the instruction
54  * changes the types of the registers depending on instruction semantics.
55  * If instruction is BPF_MOV64_REG(BPF_REG_1, BPF_REG_5), then type of R5 is
56  * copied to R1.
57  *
58  * All registers are 64-bit.
59  * R0 - return register
60  * R1-R5 argument passing registers
61  * R6-R9 callee saved registers
62  * R10 - frame pointer read-only
63  *
64  * At the start of BPF program the register R1 contains a pointer to bpf_context
65  * and has type PTR_TO_CTX.
66  *
67  * Verifier tracks arithmetic operations on pointers in case:
68  *    BPF_MOV64_REG(BPF_REG_1, BPF_REG_10),
69  *    BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -20),
70  * 1st insn copies R10 (which has FRAME_PTR) type into R1
71  * and 2nd arithmetic instruction is pattern matched to recognize
72  * that it wants to construct a pointer to some element within stack.
73  * So after 2nd insn, the register R1 has type PTR_TO_STACK
74  * (and -20 constant is saved for further stack bounds checking).
75  * Meaning that this reg is a pointer to stack plus known immediate constant.
76  *
77  * Most of the time the registers have SCALAR_VALUE type, which
78  * means the register has some value, but it's not a valid pointer.
79  * (like pointer plus pointer becomes SCALAR_VALUE type)
80  *
81  * When verifier sees load or store instructions the type of base register
82  * can be: PTR_TO_MAP_VALUE, PTR_TO_CTX, PTR_TO_STACK. These are three pointer
83  * types recognized by check_mem_access() function.
84  *
85  * PTR_TO_MAP_VALUE means that this register is pointing to 'map element value'
86  * and the range of [ptr, ptr + map's value_size) is accessible.
87  *
88  * registers used to pass values to function calls are checked against
89  * function argument constraints.
90  *
91  * ARG_PTR_TO_MAP_KEY is one of such argument constraints.
92  * It means that the register type passed to this function must be
93  * PTR_TO_STACK and it will be used inside the function as
94  * 'pointer to map element key'
95  *
96  * For example the argument constraints for bpf_map_lookup_elem():
97  *   .ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL,
98  *   .arg1_type = ARG_CONST_MAP_PTR,
99  *   .arg2_type = ARG_PTR_TO_MAP_KEY,
100  *
101  * ret_type says that this function returns 'pointer to map elem value or null'
102  * function expects 1st argument to be a const pointer to 'struct bpf_map' and
103  * 2nd argument should be a pointer to stack, which will be used inside
104  * the helper function as a pointer to map element key.
105  *
106  * On the kernel side the helper function looks like:
107  * u64 bpf_map_lookup_elem(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
108  * {
109  *    struct bpf_map *map = (struct bpf_map *) (unsigned long) r1;
110  *    void *key = (void *) (unsigned long) r2;
111  *    void *value;
112  *
113  *    here kernel can access 'key' and 'map' pointers safely, knowing that
114  *    [key, key + map->key_size) bytes are valid and were initialized on
115  *    the stack of eBPF program.
116  * }
117  *
118  * Corresponding eBPF program may look like:
119  *    BPF_MOV64_REG(BPF_REG_2, BPF_REG_10),  // after this insn R2 type is FRAME_PTR
120  *    BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -4), // after this insn R2 type is PTR_TO_STACK
121  *    BPF_LD_MAP_FD(BPF_REG_1, map_fd),      // after this insn R1 type is CONST_PTR_TO_MAP
122  *    BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem),
123  * here verifier looks at prototype of map_lookup_elem() and sees:
124  * .arg1_type == ARG_CONST_MAP_PTR and R1->type == CONST_PTR_TO_MAP, which is ok,
125  * Now verifier knows that this map has key of R1->map_ptr->key_size bytes
126  *
127  * Then .arg2_type == ARG_PTR_TO_MAP_KEY and R2->type == PTR_TO_STACK, ok so far,
128  * Now verifier checks that [R2, R2 + map's key_size) are within stack limits
129  * and were initialized prior to this call.
130  * If it's ok, then verifier allows this BPF_CALL insn and looks at
131  * .ret_type which is RET_PTR_TO_MAP_VALUE_OR_NULL, so it sets
132  * R0->type = PTR_TO_MAP_VALUE_OR_NULL which means bpf_map_lookup_elem() function
133  * returns ether pointer to map value or NULL.
134  *
135  * When type PTR_TO_MAP_VALUE_OR_NULL passes through 'if (reg != 0) goto +off'
136  * insn, the register holding that pointer in the true branch changes state to
137  * PTR_TO_MAP_VALUE and the same register changes state to CONST_IMM in the false
138  * branch. See check_cond_jmp_op().
139  *
140  * After the call R0 is set to return type of the function and registers R1-R5
141  * are set to NOT_INIT to indicate that they are no longer readable.
142  */
143 
144 /* verifier_state + insn_idx are pushed to stack when branch is encountered */
145 struct bpf_verifier_stack_elem {
146 	/* verifer state is 'st'
147 	 * before processing instruction 'insn_idx'
148 	 * and after processing instruction 'prev_insn_idx'
149 	 */
150 	struct bpf_verifier_state st;
151 	int insn_idx;
152 	int prev_insn_idx;
153 	struct bpf_verifier_stack_elem *next;
154 };
155 
156 #define BPF_COMPLEXITY_LIMIT_INSNS	131072
157 #define BPF_COMPLEXITY_LIMIT_STACK	1024
158 
159 #define BPF_MAP_PTR_POISON ((void *)0xeB9F + POISON_POINTER_DELTA)
160 
161 struct bpf_call_arg_meta {
162 	struct bpf_map *map_ptr;
163 	bool raw_mode;
164 	bool pkt_access;
165 	int regno;
166 	int access_size;
167 };
168 
169 static DEFINE_MUTEX(bpf_verifier_lock);
170 
171 /* log_level controls verbosity level of eBPF verifier.
172  * bpf_verifier_log_write() is used to dump the verification trace to the log,
173  * so the user can figure out what's wrong with the program
174  */
175 __printf(2, 3) void bpf_verifier_log_write(struct bpf_verifier_env *env,
176 					   const char *fmt, ...)
177 {
178 	struct bpf_verifer_log *log = &env->log;
179 	unsigned int n;
180 	va_list args;
181 
182 	if (!log->level || !log->ubuf || bpf_verifier_log_full(log))
183 		return;
184 
185 	va_start(args, fmt);
186 	n = vscnprintf(log->kbuf, BPF_VERIFIER_TMP_LOG_SIZE, fmt, args);
187 	va_end(args);
188 
189 	WARN_ONCE(n >= BPF_VERIFIER_TMP_LOG_SIZE - 1,
190 		  "verifier log line truncated - local buffer too short\n");
191 
192 	n = min(log->len_total - log->len_used - 1, n);
193 	log->kbuf[n] = '\0';
194 
195 	if (!copy_to_user(log->ubuf + log->len_used, log->kbuf, n + 1))
196 		log->len_used += n;
197 	else
198 		log->ubuf = NULL;
199 }
200 EXPORT_SYMBOL_GPL(bpf_verifier_log_write);
201 /* Historically bpf_verifier_log_write was called verbose, but the name was too
202  * generic for symbol export. The function was renamed, but not the calls in
203  * the verifier to avoid complicating backports. Hence the alias below.
204  */
205 static __printf(2, 3) void verbose(struct bpf_verifier_env *env,
206 				   const char *fmt, ...)
207 	__attribute__((alias("bpf_verifier_log_write")));
208 
209 static bool type_is_pkt_pointer(enum bpf_reg_type type)
210 {
211 	return type == PTR_TO_PACKET ||
212 	       type == PTR_TO_PACKET_META;
213 }
214 
215 /* string representation of 'enum bpf_reg_type' */
216 static const char * const reg_type_str[] = {
217 	[NOT_INIT]		= "?",
218 	[SCALAR_VALUE]		= "inv",
219 	[PTR_TO_CTX]		= "ctx",
220 	[CONST_PTR_TO_MAP]	= "map_ptr",
221 	[PTR_TO_MAP_VALUE]	= "map_value",
222 	[PTR_TO_MAP_VALUE_OR_NULL] = "map_value_or_null",
223 	[PTR_TO_STACK]		= "fp",
224 	[PTR_TO_PACKET]		= "pkt",
225 	[PTR_TO_PACKET_META]	= "pkt_meta",
226 	[PTR_TO_PACKET_END]	= "pkt_end",
227 };
228 
229 static void print_liveness(struct bpf_verifier_env *env,
230 			   enum bpf_reg_liveness live)
231 {
232 	if (live & (REG_LIVE_READ | REG_LIVE_WRITTEN))
233 	    verbose(env, "_");
234 	if (live & REG_LIVE_READ)
235 		verbose(env, "r");
236 	if (live & REG_LIVE_WRITTEN)
237 		verbose(env, "w");
238 }
239 
240 static struct bpf_func_state *func(struct bpf_verifier_env *env,
241 				   const struct bpf_reg_state *reg)
242 {
243 	struct bpf_verifier_state *cur = env->cur_state;
244 
245 	return cur->frame[reg->frameno];
246 }
247 
248 static void print_verifier_state(struct bpf_verifier_env *env,
249 				 const struct bpf_func_state *state)
250 {
251 	const struct bpf_reg_state *reg;
252 	enum bpf_reg_type t;
253 	int i;
254 
255 	if (state->frameno)
256 		verbose(env, " frame%d:", state->frameno);
257 	for (i = 0; i < MAX_BPF_REG; i++) {
258 		reg = &state->regs[i];
259 		t = reg->type;
260 		if (t == NOT_INIT)
261 			continue;
262 		verbose(env, " R%d", i);
263 		print_liveness(env, reg->live);
264 		verbose(env, "=%s", reg_type_str[t]);
265 		if ((t == SCALAR_VALUE || t == PTR_TO_STACK) &&
266 		    tnum_is_const(reg->var_off)) {
267 			/* reg->off should be 0 for SCALAR_VALUE */
268 			verbose(env, "%lld", reg->var_off.value + reg->off);
269 			if (t == PTR_TO_STACK)
270 				verbose(env, ",call_%d", func(env, reg)->callsite);
271 		} else {
272 			verbose(env, "(id=%d", reg->id);
273 			if (t != SCALAR_VALUE)
274 				verbose(env, ",off=%d", reg->off);
275 			if (type_is_pkt_pointer(t))
276 				verbose(env, ",r=%d", reg->range);
277 			else if (t == CONST_PTR_TO_MAP ||
278 				 t == PTR_TO_MAP_VALUE ||
279 				 t == PTR_TO_MAP_VALUE_OR_NULL)
280 				verbose(env, ",ks=%d,vs=%d",
281 					reg->map_ptr->key_size,
282 					reg->map_ptr->value_size);
283 			if (tnum_is_const(reg->var_off)) {
284 				/* Typically an immediate SCALAR_VALUE, but
285 				 * could be a pointer whose offset is too big
286 				 * for reg->off
287 				 */
288 				verbose(env, ",imm=%llx", reg->var_off.value);
289 			} else {
290 				if (reg->smin_value != reg->umin_value &&
291 				    reg->smin_value != S64_MIN)
292 					verbose(env, ",smin_value=%lld",
293 						(long long)reg->smin_value);
294 				if (reg->smax_value != reg->umax_value &&
295 				    reg->smax_value != S64_MAX)
296 					verbose(env, ",smax_value=%lld",
297 						(long long)reg->smax_value);
298 				if (reg->umin_value != 0)
299 					verbose(env, ",umin_value=%llu",
300 						(unsigned long long)reg->umin_value);
301 				if (reg->umax_value != U64_MAX)
302 					verbose(env, ",umax_value=%llu",
303 						(unsigned long long)reg->umax_value);
304 				if (!tnum_is_unknown(reg->var_off)) {
305 					char tn_buf[48];
306 
307 					tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off);
308 					verbose(env, ",var_off=%s", tn_buf);
309 				}
310 			}
311 			verbose(env, ")");
312 		}
313 	}
314 	for (i = 0; i < state->allocated_stack / BPF_REG_SIZE; i++) {
315 		if (state->stack[i].slot_type[0] == STACK_SPILL) {
316 			verbose(env, " fp%d",
317 				(-i - 1) * BPF_REG_SIZE);
318 			print_liveness(env, state->stack[i].spilled_ptr.live);
319 			verbose(env, "=%s",
320 				reg_type_str[state->stack[i].spilled_ptr.type]);
321 		}
322 		if (state->stack[i].slot_type[0] == STACK_ZERO)
323 			verbose(env, " fp%d=0", (-i - 1) * BPF_REG_SIZE);
324 	}
325 	verbose(env, "\n");
326 }
327 
328 static int copy_stack_state(struct bpf_func_state *dst,
329 			    const struct bpf_func_state *src)
330 {
331 	if (!src->stack)
332 		return 0;
333 	if (WARN_ON_ONCE(dst->allocated_stack < src->allocated_stack)) {
334 		/* internal bug, make state invalid to reject the program */
335 		memset(dst, 0, sizeof(*dst));
336 		return -EFAULT;
337 	}
338 	memcpy(dst->stack, src->stack,
339 	       sizeof(*src->stack) * (src->allocated_stack / BPF_REG_SIZE));
340 	return 0;
341 }
342 
343 /* do_check() starts with zero-sized stack in struct bpf_verifier_state to
344  * make it consume minimal amount of memory. check_stack_write() access from
345  * the program calls into realloc_func_state() to grow the stack size.
346  * Note there is a non-zero 'parent' pointer inside bpf_verifier_state
347  * which this function copies over. It points to previous bpf_verifier_state
348  * which is never reallocated
349  */
350 static int realloc_func_state(struct bpf_func_state *state, int size,
351 			      bool copy_old)
352 {
353 	u32 old_size = state->allocated_stack;
354 	struct bpf_stack_state *new_stack;
355 	int slot = size / BPF_REG_SIZE;
356 
357 	if (size <= old_size || !size) {
358 		if (copy_old)
359 			return 0;
360 		state->allocated_stack = slot * BPF_REG_SIZE;
361 		if (!size && old_size) {
362 			kfree(state->stack);
363 			state->stack = NULL;
364 		}
365 		return 0;
366 	}
367 	new_stack = kmalloc_array(slot, sizeof(struct bpf_stack_state),
368 				  GFP_KERNEL);
369 	if (!new_stack)
370 		return -ENOMEM;
371 	if (copy_old) {
372 		if (state->stack)
373 			memcpy(new_stack, state->stack,
374 			       sizeof(*new_stack) * (old_size / BPF_REG_SIZE));
375 		memset(new_stack + old_size / BPF_REG_SIZE, 0,
376 		       sizeof(*new_stack) * (size - old_size) / BPF_REG_SIZE);
377 	}
378 	state->allocated_stack = slot * BPF_REG_SIZE;
379 	kfree(state->stack);
380 	state->stack = new_stack;
381 	return 0;
382 }
383 
384 static void free_func_state(struct bpf_func_state *state)
385 {
386 	if (!state)
387 		return;
388 	kfree(state->stack);
389 	kfree(state);
390 }
391 
392 static void free_verifier_state(struct bpf_verifier_state *state,
393 				bool free_self)
394 {
395 	int i;
396 
397 	for (i = 0; i <= state->curframe; i++) {
398 		free_func_state(state->frame[i]);
399 		state->frame[i] = NULL;
400 	}
401 	if (free_self)
402 		kfree(state);
403 }
404 
405 /* copy verifier state from src to dst growing dst stack space
406  * when necessary to accommodate larger src stack
407  */
408 static int copy_func_state(struct bpf_func_state *dst,
409 			   const struct bpf_func_state *src)
410 {
411 	int err;
412 
413 	err = realloc_func_state(dst, src->allocated_stack, false);
414 	if (err)
415 		return err;
416 	memcpy(dst, src, offsetof(struct bpf_func_state, allocated_stack));
417 	return copy_stack_state(dst, src);
418 }
419 
420 static int copy_verifier_state(struct bpf_verifier_state *dst_state,
421 			       const struct bpf_verifier_state *src)
422 {
423 	struct bpf_func_state *dst;
424 	int i, err;
425 
426 	/* if dst has more stack frames then src frame, free them */
427 	for (i = src->curframe + 1; i <= dst_state->curframe; i++) {
428 		free_func_state(dst_state->frame[i]);
429 		dst_state->frame[i] = NULL;
430 	}
431 	dst_state->curframe = src->curframe;
432 	dst_state->parent = src->parent;
433 	for (i = 0; i <= src->curframe; i++) {
434 		dst = dst_state->frame[i];
435 		if (!dst) {
436 			dst = kzalloc(sizeof(*dst), GFP_KERNEL);
437 			if (!dst)
438 				return -ENOMEM;
439 			dst_state->frame[i] = dst;
440 		}
441 		err = copy_func_state(dst, src->frame[i]);
442 		if (err)
443 			return err;
444 	}
445 	return 0;
446 }
447 
448 static int pop_stack(struct bpf_verifier_env *env, int *prev_insn_idx,
449 		     int *insn_idx)
450 {
451 	struct bpf_verifier_state *cur = env->cur_state;
452 	struct bpf_verifier_stack_elem *elem, *head = env->head;
453 	int err;
454 
455 	if (env->head == NULL)
456 		return -ENOENT;
457 
458 	if (cur) {
459 		err = copy_verifier_state(cur, &head->st);
460 		if (err)
461 			return err;
462 	}
463 	if (insn_idx)
464 		*insn_idx = head->insn_idx;
465 	if (prev_insn_idx)
466 		*prev_insn_idx = head->prev_insn_idx;
467 	elem = head->next;
468 	free_verifier_state(&head->st, false);
469 	kfree(head);
470 	env->head = elem;
471 	env->stack_size--;
472 	return 0;
473 }
474 
475 static struct bpf_verifier_state *push_stack(struct bpf_verifier_env *env,
476 					     int insn_idx, int prev_insn_idx)
477 {
478 	struct bpf_verifier_state *cur = env->cur_state;
479 	struct bpf_verifier_stack_elem *elem;
480 	int err;
481 
482 	elem = kzalloc(sizeof(struct bpf_verifier_stack_elem), GFP_KERNEL);
483 	if (!elem)
484 		goto err;
485 
486 	elem->insn_idx = insn_idx;
487 	elem->prev_insn_idx = prev_insn_idx;
488 	elem->next = env->head;
489 	env->head = elem;
490 	env->stack_size++;
491 	err = copy_verifier_state(&elem->st, cur);
492 	if (err)
493 		goto err;
494 	if (env->stack_size > BPF_COMPLEXITY_LIMIT_STACK) {
495 		verbose(env, "BPF program is too complex\n");
496 		goto err;
497 	}
498 	return &elem->st;
499 err:
500 	free_verifier_state(env->cur_state, true);
501 	env->cur_state = NULL;
502 	/* pop all elements and return */
503 	while (!pop_stack(env, NULL, NULL));
504 	return NULL;
505 }
506 
507 #define CALLER_SAVED_REGS 6
508 static const int caller_saved[CALLER_SAVED_REGS] = {
509 	BPF_REG_0, BPF_REG_1, BPF_REG_2, BPF_REG_3, BPF_REG_4, BPF_REG_5
510 };
511 #define CALLEE_SAVED_REGS 5
512 static const int callee_saved[CALLEE_SAVED_REGS] = {
513 	BPF_REG_6, BPF_REG_7, BPF_REG_8, BPF_REG_9
514 };
515 
516 static void __mark_reg_not_init(struct bpf_reg_state *reg);
517 
518 /* Mark the unknown part of a register (variable offset or scalar value) as
519  * known to have the value @imm.
520  */
521 static void __mark_reg_known(struct bpf_reg_state *reg, u64 imm)
522 {
523 	reg->id = 0;
524 	reg->var_off = tnum_const(imm);
525 	reg->smin_value = (s64)imm;
526 	reg->smax_value = (s64)imm;
527 	reg->umin_value = imm;
528 	reg->umax_value = imm;
529 }
530 
531 /* Mark the 'variable offset' part of a register as zero.  This should be
532  * used only on registers holding a pointer type.
533  */
534 static void __mark_reg_known_zero(struct bpf_reg_state *reg)
535 {
536 	__mark_reg_known(reg, 0);
537 }
538 
539 static void __mark_reg_const_zero(struct bpf_reg_state *reg)
540 {
541 	__mark_reg_known(reg, 0);
542 	reg->off = 0;
543 	reg->type = SCALAR_VALUE;
544 }
545 
546 static void mark_reg_known_zero(struct bpf_verifier_env *env,
547 				struct bpf_reg_state *regs, u32 regno)
548 {
549 	if (WARN_ON(regno >= MAX_BPF_REG)) {
550 		verbose(env, "mark_reg_known_zero(regs, %u)\n", regno);
551 		/* Something bad happened, let's kill all regs */
552 		for (regno = 0; regno < MAX_BPF_REG; regno++)
553 			__mark_reg_not_init(regs + regno);
554 		return;
555 	}
556 	__mark_reg_known_zero(regs + regno);
557 }
558 
559 static bool reg_is_pkt_pointer(const struct bpf_reg_state *reg)
560 {
561 	return type_is_pkt_pointer(reg->type);
562 }
563 
564 static bool reg_is_pkt_pointer_any(const struct bpf_reg_state *reg)
565 {
566 	return reg_is_pkt_pointer(reg) ||
567 	       reg->type == PTR_TO_PACKET_END;
568 }
569 
570 /* Unmodified PTR_TO_PACKET[_META,_END] register from ctx access. */
571 static bool reg_is_init_pkt_pointer(const struct bpf_reg_state *reg,
572 				    enum bpf_reg_type which)
573 {
574 	/* The register can already have a range from prior markings.
575 	 * This is fine as long as it hasn't been advanced from its
576 	 * origin.
577 	 */
578 	return reg->type == which &&
579 	       reg->id == 0 &&
580 	       reg->off == 0 &&
581 	       tnum_equals_const(reg->var_off, 0);
582 }
583 
584 /* Attempts to improve min/max values based on var_off information */
585 static void __update_reg_bounds(struct bpf_reg_state *reg)
586 {
587 	/* min signed is max(sign bit) | min(other bits) */
588 	reg->smin_value = max_t(s64, reg->smin_value,
589 				reg->var_off.value | (reg->var_off.mask & S64_MIN));
590 	/* max signed is min(sign bit) | max(other bits) */
591 	reg->smax_value = min_t(s64, reg->smax_value,
592 				reg->var_off.value | (reg->var_off.mask & S64_MAX));
593 	reg->umin_value = max(reg->umin_value, reg->var_off.value);
594 	reg->umax_value = min(reg->umax_value,
595 			      reg->var_off.value | reg->var_off.mask);
596 }
597 
598 /* Uses signed min/max values to inform unsigned, and vice-versa */
599 static void __reg_deduce_bounds(struct bpf_reg_state *reg)
600 {
601 	/* Learn sign from signed bounds.
602 	 * If we cannot cross the sign boundary, then signed and unsigned bounds
603 	 * are the same, so combine.  This works even in the negative case, e.g.
604 	 * -3 s<= x s<= -1 implies 0xf...fd u<= x u<= 0xf...ff.
605 	 */
606 	if (reg->smin_value >= 0 || reg->smax_value < 0) {
607 		reg->smin_value = reg->umin_value = max_t(u64, reg->smin_value,
608 							  reg->umin_value);
609 		reg->smax_value = reg->umax_value = min_t(u64, reg->smax_value,
610 							  reg->umax_value);
611 		return;
612 	}
613 	/* Learn sign from unsigned bounds.  Signed bounds cross the sign
614 	 * boundary, so we must be careful.
615 	 */
616 	if ((s64)reg->umax_value >= 0) {
617 		/* Positive.  We can't learn anything from the smin, but smax
618 		 * is positive, hence safe.
619 		 */
620 		reg->smin_value = reg->umin_value;
621 		reg->smax_value = reg->umax_value = min_t(u64, reg->smax_value,
622 							  reg->umax_value);
623 	} else if ((s64)reg->umin_value < 0) {
624 		/* Negative.  We can't learn anything from the smax, but smin
625 		 * is negative, hence safe.
626 		 */
627 		reg->smin_value = reg->umin_value = max_t(u64, reg->smin_value,
628 							  reg->umin_value);
629 		reg->smax_value = reg->umax_value;
630 	}
631 }
632 
633 /* Attempts to improve var_off based on unsigned min/max information */
634 static void __reg_bound_offset(struct bpf_reg_state *reg)
635 {
636 	reg->var_off = tnum_intersect(reg->var_off,
637 				      tnum_range(reg->umin_value,
638 						 reg->umax_value));
639 }
640 
641 /* Reset the min/max bounds of a register */
642 static void __mark_reg_unbounded(struct bpf_reg_state *reg)
643 {
644 	reg->smin_value = S64_MIN;
645 	reg->smax_value = S64_MAX;
646 	reg->umin_value = 0;
647 	reg->umax_value = U64_MAX;
648 }
649 
650 /* Mark a register as having a completely unknown (scalar) value. */
651 static void __mark_reg_unknown(struct bpf_reg_state *reg)
652 {
653 	reg->type = SCALAR_VALUE;
654 	reg->id = 0;
655 	reg->off = 0;
656 	reg->var_off = tnum_unknown;
657 	reg->frameno = 0;
658 	__mark_reg_unbounded(reg);
659 }
660 
661 static void mark_reg_unknown(struct bpf_verifier_env *env,
662 			     struct bpf_reg_state *regs, u32 regno)
663 {
664 	if (WARN_ON(regno >= MAX_BPF_REG)) {
665 		verbose(env, "mark_reg_unknown(regs, %u)\n", regno);
666 		/* Something bad happened, let's kill all regs except FP */
667 		for (regno = 0; regno < BPF_REG_FP; regno++)
668 			__mark_reg_not_init(regs + regno);
669 		return;
670 	}
671 	__mark_reg_unknown(regs + regno);
672 }
673 
674 static void __mark_reg_not_init(struct bpf_reg_state *reg)
675 {
676 	__mark_reg_unknown(reg);
677 	reg->type = NOT_INIT;
678 }
679 
680 static void mark_reg_not_init(struct bpf_verifier_env *env,
681 			      struct bpf_reg_state *regs, u32 regno)
682 {
683 	if (WARN_ON(regno >= MAX_BPF_REG)) {
684 		verbose(env, "mark_reg_not_init(regs, %u)\n", regno);
685 		/* Something bad happened, let's kill all regs except FP */
686 		for (regno = 0; regno < BPF_REG_FP; regno++)
687 			__mark_reg_not_init(regs + regno);
688 		return;
689 	}
690 	__mark_reg_not_init(regs + regno);
691 }
692 
693 static void init_reg_state(struct bpf_verifier_env *env,
694 			   struct bpf_func_state *state)
695 {
696 	struct bpf_reg_state *regs = state->regs;
697 	int i;
698 
699 	for (i = 0; i < MAX_BPF_REG; i++) {
700 		mark_reg_not_init(env, regs, i);
701 		regs[i].live = REG_LIVE_NONE;
702 	}
703 
704 	/* frame pointer */
705 	regs[BPF_REG_FP].type = PTR_TO_STACK;
706 	mark_reg_known_zero(env, regs, BPF_REG_FP);
707 	regs[BPF_REG_FP].frameno = state->frameno;
708 
709 	/* 1st arg to a function */
710 	regs[BPF_REG_1].type = PTR_TO_CTX;
711 	mark_reg_known_zero(env, regs, BPF_REG_1);
712 }
713 
714 #define BPF_MAIN_FUNC (-1)
715 static void init_func_state(struct bpf_verifier_env *env,
716 			    struct bpf_func_state *state,
717 			    int callsite, int frameno, int subprogno)
718 {
719 	state->callsite = callsite;
720 	state->frameno = frameno;
721 	state->subprogno = subprogno;
722 	init_reg_state(env, state);
723 }
724 
725 enum reg_arg_type {
726 	SRC_OP,		/* register is used as source operand */
727 	DST_OP,		/* register is used as destination operand */
728 	DST_OP_NO_MARK	/* same as above, check only, don't mark */
729 };
730 
731 static int cmp_subprogs(const void *a, const void *b)
732 {
733 	return *(int *)a - *(int *)b;
734 }
735 
736 static int find_subprog(struct bpf_verifier_env *env, int off)
737 {
738 	u32 *p;
739 
740 	p = bsearch(&off, env->subprog_starts, env->subprog_cnt,
741 		    sizeof(env->subprog_starts[0]), cmp_subprogs);
742 	if (!p)
743 		return -ENOENT;
744 	return p - env->subprog_starts;
745 
746 }
747 
748 static int add_subprog(struct bpf_verifier_env *env, int off)
749 {
750 	int insn_cnt = env->prog->len;
751 	int ret;
752 
753 	if (off >= insn_cnt || off < 0) {
754 		verbose(env, "call to invalid destination\n");
755 		return -EINVAL;
756 	}
757 	ret = find_subprog(env, off);
758 	if (ret >= 0)
759 		return 0;
760 	if (env->subprog_cnt >= BPF_MAX_SUBPROGS) {
761 		verbose(env, "too many subprograms\n");
762 		return -E2BIG;
763 	}
764 	env->subprog_starts[env->subprog_cnt++] = off;
765 	sort(env->subprog_starts, env->subprog_cnt,
766 	     sizeof(env->subprog_starts[0]), cmp_subprogs, NULL);
767 	return 0;
768 }
769 
770 static int check_subprogs(struct bpf_verifier_env *env)
771 {
772 	int i, ret, subprog_start, subprog_end, off, cur_subprog = 0;
773 	struct bpf_insn *insn = env->prog->insnsi;
774 	int insn_cnt = env->prog->len;
775 
776 	/* determine subprog starts. The end is one before the next starts */
777 	for (i = 0; i < insn_cnt; i++) {
778 		if (insn[i].code != (BPF_JMP | BPF_CALL))
779 			continue;
780 		if (insn[i].src_reg != BPF_PSEUDO_CALL)
781 			continue;
782 		if (!env->allow_ptr_leaks) {
783 			verbose(env, "function calls to other bpf functions are allowed for root only\n");
784 			return -EPERM;
785 		}
786 		if (bpf_prog_is_dev_bound(env->prog->aux)) {
787 			verbose(env, "function calls in offloaded programs are not supported yet\n");
788 			return -EINVAL;
789 		}
790 		ret = add_subprog(env, i + insn[i].imm + 1);
791 		if (ret < 0)
792 			return ret;
793 	}
794 
795 	if (env->log.level > 1)
796 		for (i = 0; i < env->subprog_cnt; i++)
797 			verbose(env, "func#%d @%d\n", i, env->subprog_starts[i]);
798 
799 	/* now check that all jumps are within the same subprog */
800 	subprog_start = 0;
801 	if (env->subprog_cnt == cur_subprog)
802 		subprog_end = insn_cnt;
803 	else
804 		subprog_end = env->subprog_starts[cur_subprog++];
805 	for (i = 0; i < insn_cnt; i++) {
806 		u8 code = insn[i].code;
807 
808 		if (BPF_CLASS(code) != BPF_JMP)
809 			goto next;
810 		if (BPF_OP(code) == BPF_EXIT || BPF_OP(code) == BPF_CALL)
811 			goto next;
812 		off = i + insn[i].off + 1;
813 		if (off < subprog_start || off >= subprog_end) {
814 			verbose(env, "jump out of range from insn %d to %d\n", i, off);
815 			return -EINVAL;
816 		}
817 next:
818 		if (i == subprog_end - 1) {
819 			/* to avoid fall-through from one subprog into another
820 			 * the last insn of the subprog should be either exit
821 			 * or unconditional jump back
822 			 */
823 			if (code != (BPF_JMP | BPF_EXIT) &&
824 			    code != (BPF_JMP | BPF_JA)) {
825 				verbose(env, "last insn is not an exit or jmp\n");
826 				return -EINVAL;
827 			}
828 			subprog_start = subprog_end;
829 			if (env->subprog_cnt == cur_subprog)
830 				subprog_end = insn_cnt;
831 			else
832 				subprog_end = env->subprog_starts[cur_subprog++];
833 		}
834 	}
835 	return 0;
836 }
837 
838 static
839 struct bpf_verifier_state *skip_callee(struct bpf_verifier_env *env,
840 				       const struct bpf_verifier_state *state,
841 				       struct bpf_verifier_state *parent,
842 				       u32 regno)
843 {
844 	struct bpf_verifier_state *tmp = NULL;
845 
846 	/* 'parent' could be a state of caller and
847 	 * 'state' could be a state of callee. In such case
848 	 * parent->curframe < state->curframe
849 	 * and it's ok for r1 - r5 registers
850 	 *
851 	 * 'parent' could be a callee's state after it bpf_exit-ed.
852 	 * In such case parent->curframe > state->curframe
853 	 * and it's ok for r0 only
854 	 */
855 	if (parent->curframe == state->curframe ||
856 	    (parent->curframe < state->curframe &&
857 	     regno >= BPF_REG_1 && regno <= BPF_REG_5) ||
858 	    (parent->curframe > state->curframe &&
859 	       regno == BPF_REG_0))
860 		return parent;
861 
862 	if (parent->curframe > state->curframe &&
863 	    regno >= BPF_REG_6) {
864 		/* for callee saved regs we have to skip the whole chain
865 		 * of states that belong to callee and mark as LIVE_READ
866 		 * the registers before the call
867 		 */
868 		tmp = parent;
869 		while (tmp && tmp->curframe != state->curframe) {
870 			tmp = tmp->parent;
871 		}
872 		if (!tmp)
873 			goto bug;
874 		parent = tmp;
875 	} else {
876 		goto bug;
877 	}
878 	return parent;
879 bug:
880 	verbose(env, "verifier bug regno %d tmp %p\n", regno, tmp);
881 	verbose(env, "regno %d parent frame %d current frame %d\n",
882 		regno, parent->curframe, state->curframe);
883 	return NULL;
884 }
885 
886 static int mark_reg_read(struct bpf_verifier_env *env,
887 			 const struct bpf_verifier_state *state,
888 			 struct bpf_verifier_state *parent,
889 			 u32 regno)
890 {
891 	bool writes = parent == state->parent; /* Observe write marks */
892 
893 	if (regno == BPF_REG_FP)
894 		/* We don't need to worry about FP liveness because it's read-only */
895 		return 0;
896 
897 	while (parent) {
898 		/* if read wasn't screened by an earlier write ... */
899 		if (writes && state->frame[state->curframe]->regs[regno].live & REG_LIVE_WRITTEN)
900 			break;
901 		parent = skip_callee(env, state, parent, regno);
902 		if (!parent)
903 			return -EFAULT;
904 		/* ... then we depend on parent's value */
905 		parent->frame[parent->curframe]->regs[regno].live |= REG_LIVE_READ;
906 		state = parent;
907 		parent = state->parent;
908 		writes = true;
909 	}
910 	return 0;
911 }
912 
913 static int check_reg_arg(struct bpf_verifier_env *env, u32 regno,
914 			 enum reg_arg_type t)
915 {
916 	struct bpf_verifier_state *vstate = env->cur_state;
917 	struct bpf_func_state *state = vstate->frame[vstate->curframe];
918 	struct bpf_reg_state *regs = state->regs;
919 
920 	if (regno >= MAX_BPF_REG) {
921 		verbose(env, "R%d is invalid\n", regno);
922 		return -EINVAL;
923 	}
924 
925 	if (t == SRC_OP) {
926 		/* check whether register used as source operand can be read */
927 		if (regs[regno].type == NOT_INIT) {
928 			verbose(env, "R%d !read_ok\n", regno);
929 			return -EACCES;
930 		}
931 		return mark_reg_read(env, vstate, vstate->parent, regno);
932 	} else {
933 		/* check whether register used as dest operand can be written to */
934 		if (regno == BPF_REG_FP) {
935 			verbose(env, "frame pointer is read only\n");
936 			return -EACCES;
937 		}
938 		regs[regno].live |= REG_LIVE_WRITTEN;
939 		if (t == DST_OP)
940 			mark_reg_unknown(env, regs, regno);
941 	}
942 	return 0;
943 }
944 
945 static bool is_spillable_regtype(enum bpf_reg_type type)
946 {
947 	switch (type) {
948 	case PTR_TO_MAP_VALUE:
949 	case PTR_TO_MAP_VALUE_OR_NULL:
950 	case PTR_TO_STACK:
951 	case PTR_TO_CTX:
952 	case PTR_TO_PACKET:
953 	case PTR_TO_PACKET_META:
954 	case PTR_TO_PACKET_END:
955 	case CONST_PTR_TO_MAP:
956 		return true;
957 	default:
958 		return false;
959 	}
960 }
961 
962 /* Does this register contain a constant zero? */
963 static bool register_is_null(struct bpf_reg_state *reg)
964 {
965 	return reg->type == SCALAR_VALUE && tnum_equals_const(reg->var_off, 0);
966 }
967 
968 /* check_stack_read/write functions track spill/fill of registers,
969  * stack boundary and alignment are checked in check_mem_access()
970  */
971 static int check_stack_write(struct bpf_verifier_env *env,
972 			     struct bpf_func_state *state, /* func where register points to */
973 			     int off, int size, int value_regno)
974 {
975 	struct bpf_func_state *cur; /* state of the current function */
976 	int i, slot = -off - 1, spi = slot / BPF_REG_SIZE, err;
977 	enum bpf_reg_type type;
978 
979 	err = realloc_func_state(state, round_up(slot + 1, BPF_REG_SIZE),
980 				 true);
981 	if (err)
982 		return err;
983 	/* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0,
984 	 * so it's aligned access and [off, off + size) are within stack limits
985 	 */
986 	if (!env->allow_ptr_leaks &&
987 	    state->stack[spi].slot_type[0] == STACK_SPILL &&
988 	    size != BPF_REG_SIZE) {
989 		verbose(env, "attempt to corrupt spilled pointer on stack\n");
990 		return -EACCES;
991 	}
992 
993 	cur = env->cur_state->frame[env->cur_state->curframe];
994 	if (value_regno >= 0 &&
995 	    is_spillable_regtype((type = cur->regs[value_regno].type))) {
996 
997 		/* register containing pointer is being spilled into stack */
998 		if (size != BPF_REG_SIZE) {
999 			verbose(env, "invalid size of register spill\n");
1000 			return -EACCES;
1001 		}
1002 
1003 		if (state != cur && type == PTR_TO_STACK) {
1004 			verbose(env, "cannot spill pointers to stack into stack frame of the caller\n");
1005 			return -EINVAL;
1006 		}
1007 
1008 		/* save register state */
1009 		state->stack[spi].spilled_ptr = cur->regs[value_regno];
1010 		state->stack[spi].spilled_ptr.live |= REG_LIVE_WRITTEN;
1011 
1012 		for (i = 0; i < BPF_REG_SIZE; i++)
1013 			state->stack[spi].slot_type[i] = STACK_SPILL;
1014 	} else {
1015 		u8 type = STACK_MISC;
1016 
1017 		/* regular write of data into stack */
1018 		state->stack[spi].spilled_ptr = (struct bpf_reg_state) {};
1019 
1020 		/* only mark the slot as written if all 8 bytes were written
1021 		 * otherwise read propagation may incorrectly stop too soon
1022 		 * when stack slots are partially written.
1023 		 * This heuristic means that read propagation will be
1024 		 * conservative, since it will add reg_live_read marks
1025 		 * to stack slots all the way to first state when programs
1026 		 * writes+reads less than 8 bytes
1027 		 */
1028 		if (size == BPF_REG_SIZE)
1029 			state->stack[spi].spilled_ptr.live |= REG_LIVE_WRITTEN;
1030 
1031 		/* when we zero initialize stack slots mark them as such */
1032 		if (value_regno >= 0 &&
1033 		    register_is_null(&cur->regs[value_regno]))
1034 			type = STACK_ZERO;
1035 
1036 		for (i = 0; i < size; i++)
1037 			state->stack[spi].slot_type[(slot - i) % BPF_REG_SIZE] =
1038 				type;
1039 	}
1040 	return 0;
1041 }
1042 
1043 /* registers of every function are unique and mark_reg_read() propagates
1044  * the liveness in the following cases:
1045  * - from callee into caller for R1 - R5 that were used as arguments
1046  * - from caller into callee for R0 that used as result of the call
1047  * - from caller to the same caller skipping states of the callee for R6 - R9,
1048  *   since R6 - R9 are callee saved by implicit function prologue and
1049  *   caller's R6 != callee's R6, so when we propagate liveness up to
1050  *   parent states we need to skip callee states for R6 - R9.
1051  *
1052  * stack slot marking is different, since stacks of caller and callee are
1053  * accessible in both (since caller can pass a pointer to caller's stack to
1054  * callee which can pass it to another function), hence mark_stack_slot_read()
1055  * has to propagate the stack liveness to all parent states at given frame number.
1056  * Consider code:
1057  * f1() {
1058  *   ptr = fp - 8;
1059  *   *ptr = ctx;
1060  *   call f2 {
1061  *      .. = *ptr;
1062  *   }
1063  *   .. = *ptr;
1064  * }
1065  * First *ptr is reading from f1's stack and mark_stack_slot_read() has
1066  * to mark liveness at the f1's frame and not f2's frame.
1067  * Second *ptr is also reading from f1's stack and mark_stack_slot_read() has
1068  * to propagate liveness to f2 states at f1's frame level and further into
1069  * f1 states at f1's frame level until write into that stack slot
1070  */
1071 static void mark_stack_slot_read(struct bpf_verifier_env *env,
1072 				 const struct bpf_verifier_state *state,
1073 				 struct bpf_verifier_state *parent,
1074 				 int slot, int frameno)
1075 {
1076 	bool writes = parent == state->parent; /* Observe write marks */
1077 
1078 	while (parent) {
1079 		if (parent->frame[frameno]->allocated_stack <= slot * BPF_REG_SIZE)
1080 			/* since LIVE_WRITTEN mark is only done for full 8-byte
1081 			 * write the read marks are conservative and parent
1082 			 * state may not even have the stack allocated. In such case
1083 			 * end the propagation, since the loop reached beginning
1084 			 * of the function
1085 			 */
1086 			break;
1087 		/* if read wasn't screened by an earlier write ... */
1088 		if (writes && state->frame[frameno]->stack[slot].spilled_ptr.live & REG_LIVE_WRITTEN)
1089 			break;
1090 		/* ... then we depend on parent's value */
1091 		parent->frame[frameno]->stack[slot].spilled_ptr.live |= REG_LIVE_READ;
1092 		state = parent;
1093 		parent = state->parent;
1094 		writes = true;
1095 	}
1096 }
1097 
1098 static int check_stack_read(struct bpf_verifier_env *env,
1099 			    struct bpf_func_state *reg_state /* func where register points to */,
1100 			    int off, int size, int value_regno)
1101 {
1102 	struct bpf_verifier_state *vstate = env->cur_state;
1103 	struct bpf_func_state *state = vstate->frame[vstate->curframe];
1104 	int i, slot = -off - 1, spi = slot / BPF_REG_SIZE;
1105 	u8 *stype;
1106 
1107 	if (reg_state->allocated_stack <= slot) {
1108 		verbose(env, "invalid read from stack off %d+0 size %d\n",
1109 			off, size);
1110 		return -EACCES;
1111 	}
1112 	stype = reg_state->stack[spi].slot_type;
1113 
1114 	if (stype[0] == STACK_SPILL) {
1115 		if (size != BPF_REG_SIZE) {
1116 			verbose(env, "invalid size of register spill\n");
1117 			return -EACCES;
1118 		}
1119 		for (i = 1; i < BPF_REG_SIZE; i++) {
1120 			if (stype[(slot - i) % BPF_REG_SIZE] != STACK_SPILL) {
1121 				verbose(env, "corrupted spill memory\n");
1122 				return -EACCES;
1123 			}
1124 		}
1125 
1126 		if (value_regno >= 0) {
1127 			/* restore register state from stack */
1128 			state->regs[value_regno] = reg_state->stack[spi].spilled_ptr;
1129 			/* mark reg as written since spilled pointer state likely
1130 			 * has its liveness marks cleared by is_state_visited()
1131 			 * which resets stack/reg liveness for state transitions
1132 			 */
1133 			state->regs[value_regno].live |= REG_LIVE_WRITTEN;
1134 		}
1135 		mark_stack_slot_read(env, vstate, vstate->parent, spi,
1136 				     reg_state->frameno);
1137 		return 0;
1138 	} else {
1139 		int zeros = 0;
1140 
1141 		for (i = 0; i < size; i++) {
1142 			if (stype[(slot - i) % BPF_REG_SIZE] == STACK_MISC)
1143 				continue;
1144 			if (stype[(slot - i) % BPF_REG_SIZE] == STACK_ZERO) {
1145 				zeros++;
1146 				continue;
1147 			}
1148 			verbose(env, "invalid read from stack off %d+%d size %d\n",
1149 				off, i, size);
1150 			return -EACCES;
1151 		}
1152 		mark_stack_slot_read(env, vstate, vstate->parent, spi,
1153 				     reg_state->frameno);
1154 		if (value_regno >= 0) {
1155 			if (zeros == size) {
1156 				/* any size read into register is zero extended,
1157 				 * so the whole register == const_zero
1158 				 */
1159 				__mark_reg_const_zero(&state->regs[value_regno]);
1160 			} else {
1161 				/* have read misc data from the stack */
1162 				mark_reg_unknown(env, state->regs, value_regno);
1163 			}
1164 			state->regs[value_regno].live |= REG_LIVE_WRITTEN;
1165 		}
1166 		return 0;
1167 	}
1168 }
1169 
1170 /* check read/write into map element returned by bpf_map_lookup_elem() */
1171 static int __check_map_access(struct bpf_verifier_env *env, u32 regno, int off,
1172 			      int size, bool zero_size_allowed)
1173 {
1174 	struct bpf_reg_state *regs = cur_regs(env);
1175 	struct bpf_map *map = regs[regno].map_ptr;
1176 
1177 	if (off < 0 || size < 0 || (size == 0 && !zero_size_allowed) ||
1178 	    off + size > map->value_size) {
1179 		verbose(env, "invalid access to map value, value_size=%d off=%d size=%d\n",
1180 			map->value_size, off, size);
1181 		return -EACCES;
1182 	}
1183 	return 0;
1184 }
1185 
1186 /* check read/write into a map element with possible variable offset */
1187 static int check_map_access(struct bpf_verifier_env *env, u32 regno,
1188 			    int off, int size, bool zero_size_allowed)
1189 {
1190 	struct bpf_verifier_state *vstate = env->cur_state;
1191 	struct bpf_func_state *state = vstate->frame[vstate->curframe];
1192 	struct bpf_reg_state *reg = &state->regs[regno];
1193 	int err;
1194 
1195 	/* We may have adjusted the register to this map value, so we
1196 	 * need to try adding each of min_value and max_value to off
1197 	 * to make sure our theoretical access will be safe.
1198 	 */
1199 	if (env->log.level)
1200 		print_verifier_state(env, state);
1201 	/* The minimum value is only important with signed
1202 	 * comparisons where we can't assume the floor of a
1203 	 * value is 0.  If we are using signed variables for our
1204 	 * index'es we need to make sure that whatever we use
1205 	 * will have a set floor within our range.
1206 	 */
1207 	if (reg->smin_value < 0) {
1208 		verbose(env, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n",
1209 			regno);
1210 		return -EACCES;
1211 	}
1212 	err = __check_map_access(env, regno, reg->smin_value + off, size,
1213 				 zero_size_allowed);
1214 	if (err) {
1215 		verbose(env, "R%d min value is outside of the array range\n",
1216 			regno);
1217 		return err;
1218 	}
1219 
1220 	/* If we haven't set a max value then we need to bail since we can't be
1221 	 * sure we won't do bad things.
1222 	 * If reg->umax_value + off could overflow, treat that as unbounded too.
1223 	 */
1224 	if (reg->umax_value >= BPF_MAX_VAR_OFF) {
1225 		verbose(env, "R%d unbounded memory access, make sure to bounds check any array access into a map\n",
1226 			regno);
1227 		return -EACCES;
1228 	}
1229 	err = __check_map_access(env, regno, reg->umax_value + off, size,
1230 				 zero_size_allowed);
1231 	if (err)
1232 		verbose(env, "R%d max value is outside of the array range\n",
1233 			regno);
1234 	return err;
1235 }
1236 
1237 #define MAX_PACKET_OFF 0xffff
1238 
1239 static bool may_access_direct_pkt_data(struct bpf_verifier_env *env,
1240 				       const struct bpf_call_arg_meta *meta,
1241 				       enum bpf_access_type t)
1242 {
1243 	switch (env->prog->type) {
1244 	case BPF_PROG_TYPE_LWT_IN:
1245 	case BPF_PROG_TYPE_LWT_OUT:
1246 		/* dst_input() and dst_output() can't write for now */
1247 		if (t == BPF_WRITE)
1248 			return false;
1249 		/* fallthrough */
1250 	case BPF_PROG_TYPE_SCHED_CLS:
1251 	case BPF_PROG_TYPE_SCHED_ACT:
1252 	case BPF_PROG_TYPE_XDP:
1253 	case BPF_PROG_TYPE_LWT_XMIT:
1254 	case BPF_PROG_TYPE_SK_SKB:
1255 		if (meta)
1256 			return meta->pkt_access;
1257 
1258 		env->seen_direct_write = true;
1259 		return true;
1260 	default:
1261 		return false;
1262 	}
1263 }
1264 
1265 static int __check_packet_access(struct bpf_verifier_env *env, u32 regno,
1266 				 int off, int size, bool zero_size_allowed)
1267 {
1268 	struct bpf_reg_state *regs = cur_regs(env);
1269 	struct bpf_reg_state *reg = &regs[regno];
1270 
1271 	if (off < 0 || size < 0 || (size == 0 && !zero_size_allowed) ||
1272 	    (u64)off + size > reg->range) {
1273 		verbose(env, "invalid access to packet, off=%d size=%d, R%d(id=%d,off=%d,r=%d)\n",
1274 			off, size, regno, reg->id, reg->off, reg->range);
1275 		return -EACCES;
1276 	}
1277 	return 0;
1278 }
1279 
1280 static int check_packet_access(struct bpf_verifier_env *env, u32 regno, int off,
1281 			       int size, bool zero_size_allowed)
1282 {
1283 	struct bpf_reg_state *regs = cur_regs(env);
1284 	struct bpf_reg_state *reg = &regs[regno];
1285 	int err;
1286 
1287 	/* We may have added a variable offset to the packet pointer; but any
1288 	 * reg->range we have comes after that.  We are only checking the fixed
1289 	 * offset.
1290 	 */
1291 
1292 	/* We don't allow negative numbers, because we aren't tracking enough
1293 	 * detail to prove they're safe.
1294 	 */
1295 	if (reg->smin_value < 0) {
1296 		verbose(env, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n",
1297 			regno);
1298 		return -EACCES;
1299 	}
1300 	err = __check_packet_access(env, regno, off, size, zero_size_allowed);
1301 	if (err) {
1302 		verbose(env, "R%d offset is outside of the packet\n", regno);
1303 		return err;
1304 	}
1305 	return err;
1306 }
1307 
1308 /* check access to 'struct bpf_context' fields.  Supports fixed offsets only */
1309 static int check_ctx_access(struct bpf_verifier_env *env, int insn_idx, int off, int size,
1310 			    enum bpf_access_type t, enum bpf_reg_type *reg_type)
1311 {
1312 	struct bpf_insn_access_aux info = {
1313 		.reg_type = *reg_type,
1314 	};
1315 
1316 	if (env->ops->is_valid_access &&
1317 	    env->ops->is_valid_access(off, size, t, &info)) {
1318 		/* A non zero info.ctx_field_size indicates that this field is a
1319 		 * candidate for later verifier transformation to load the whole
1320 		 * field and then apply a mask when accessed with a narrower
1321 		 * access than actual ctx access size. A zero info.ctx_field_size
1322 		 * will only allow for whole field access and rejects any other
1323 		 * type of narrower access.
1324 		 */
1325 		*reg_type = info.reg_type;
1326 
1327 		env->insn_aux_data[insn_idx].ctx_field_size = info.ctx_field_size;
1328 		/* remember the offset of last byte accessed in ctx */
1329 		if (env->prog->aux->max_ctx_offset < off + size)
1330 			env->prog->aux->max_ctx_offset = off + size;
1331 		return 0;
1332 	}
1333 
1334 	verbose(env, "invalid bpf_context access off=%d size=%d\n", off, size);
1335 	return -EACCES;
1336 }
1337 
1338 static bool __is_pointer_value(bool allow_ptr_leaks,
1339 			       const struct bpf_reg_state *reg)
1340 {
1341 	if (allow_ptr_leaks)
1342 		return false;
1343 
1344 	return reg->type != SCALAR_VALUE;
1345 }
1346 
1347 static bool is_pointer_value(struct bpf_verifier_env *env, int regno)
1348 {
1349 	return __is_pointer_value(env->allow_ptr_leaks, cur_regs(env) + regno);
1350 }
1351 
1352 static bool is_ctx_reg(struct bpf_verifier_env *env, int regno)
1353 {
1354 	const struct bpf_reg_state *reg = cur_regs(env) + regno;
1355 
1356 	return reg->type == PTR_TO_CTX;
1357 }
1358 
1359 static int check_pkt_ptr_alignment(struct bpf_verifier_env *env,
1360 				   const struct bpf_reg_state *reg,
1361 				   int off, int size, bool strict)
1362 {
1363 	struct tnum reg_off;
1364 	int ip_align;
1365 
1366 	/* Byte size accesses are always allowed. */
1367 	if (!strict || size == 1)
1368 		return 0;
1369 
1370 	/* For platforms that do not have a Kconfig enabling
1371 	 * CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS the value of
1372 	 * NET_IP_ALIGN is universally set to '2'.  And on platforms
1373 	 * that do set CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS, we get
1374 	 * to this code only in strict mode where we want to emulate
1375 	 * the NET_IP_ALIGN==2 checking.  Therefore use an
1376 	 * unconditional IP align value of '2'.
1377 	 */
1378 	ip_align = 2;
1379 
1380 	reg_off = tnum_add(reg->var_off, tnum_const(ip_align + reg->off + off));
1381 	if (!tnum_is_aligned(reg_off, size)) {
1382 		char tn_buf[48];
1383 
1384 		tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off);
1385 		verbose(env,
1386 			"misaligned packet access off %d+%s+%d+%d size %d\n",
1387 			ip_align, tn_buf, reg->off, off, size);
1388 		return -EACCES;
1389 	}
1390 
1391 	return 0;
1392 }
1393 
1394 static int check_generic_ptr_alignment(struct bpf_verifier_env *env,
1395 				       const struct bpf_reg_state *reg,
1396 				       const char *pointer_desc,
1397 				       int off, int size, bool strict)
1398 {
1399 	struct tnum reg_off;
1400 
1401 	/* Byte size accesses are always allowed. */
1402 	if (!strict || size == 1)
1403 		return 0;
1404 
1405 	reg_off = tnum_add(reg->var_off, tnum_const(reg->off + off));
1406 	if (!tnum_is_aligned(reg_off, size)) {
1407 		char tn_buf[48];
1408 
1409 		tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off);
1410 		verbose(env, "misaligned %saccess off %s+%d+%d size %d\n",
1411 			pointer_desc, tn_buf, reg->off, off, size);
1412 		return -EACCES;
1413 	}
1414 
1415 	return 0;
1416 }
1417 
1418 static int check_ptr_alignment(struct bpf_verifier_env *env,
1419 			       const struct bpf_reg_state *reg,
1420 			       int off, int size)
1421 {
1422 	bool strict = env->strict_alignment;
1423 	const char *pointer_desc = "";
1424 
1425 	switch (reg->type) {
1426 	case PTR_TO_PACKET:
1427 	case PTR_TO_PACKET_META:
1428 		/* Special case, because of NET_IP_ALIGN. Given metadata sits
1429 		 * right in front, treat it the very same way.
1430 		 */
1431 		return check_pkt_ptr_alignment(env, reg, off, size, strict);
1432 	case PTR_TO_MAP_VALUE:
1433 		pointer_desc = "value ";
1434 		break;
1435 	case PTR_TO_CTX:
1436 		pointer_desc = "context ";
1437 		break;
1438 	case PTR_TO_STACK:
1439 		pointer_desc = "stack ";
1440 		/* The stack spill tracking logic in check_stack_write()
1441 		 * and check_stack_read() relies on stack accesses being
1442 		 * aligned.
1443 		 */
1444 		strict = true;
1445 		break;
1446 	default:
1447 		break;
1448 	}
1449 	return check_generic_ptr_alignment(env, reg, pointer_desc, off, size,
1450 					   strict);
1451 }
1452 
1453 static int update_stack_depth(struct bpf_verifier_env *env,
1454 			      const struct bpf_func_state *func,
1455 			      int off)
1456 {
1457 	u16 stack = env->subprog_stack_depth[func->subprogno];
1458 
1459 	if (stack >= -off)
1460 		return 0;
1461 
1462 	/* update known max for given subprogram */
1463 	env->subprog_stack_depth[func->subprogno] = -off;
1464 	return 0;
1465 }
1466 
1467 /* starting from main bpf function walk all instructions of the function
1468  * and recursively walk all callees that given function can call.
1469  * Ignore jump and exit insns.
1470  * Since recursion is prevented by check_cfg() this algorithm
1471  * only needs a local stack of MAX_CALL_FRAMES to remember callsites
1472  */
1473 static int check_max_stack_depth(struct bpf_verifier_env *env)
1474 {
1475 	int depth = 0, frame = 0, subprog = 0, i = 0, subprog_end;
1476 	struct bpf_insn *insn = env->prog->insnsi;
1477 	int insn_cnt = env->prog->len;
1478 	int ret_insn[MAX_CALL_FRAMES];
1479 	int ret_prog[MAX_CALL_FRAMES];
1480 
1481 process_func:
1482 	/* round up to 32-bytes, since this is granularity
1483 	 * of interpreter stack size
1484 	 */
1485 	depth += round_up(max_t(u32, env->subprog_stack_depth[subprog], 1), 32);
1486 	if (depth > MAX_BPF_STACK) {
1487 		verbose(env, "combined stack size of %d calls is %d. Too large\n",
1488 			frame + 1, depth);
1489 		return -EACCES;
1490 	}
1491 continue_func:
1492 	if (env->subprog_cnt == subprog)
1493 		subprog_end = insn_cnt;
1494 	else
1495 		subprog_end = env->subprog_starts[subprog];
1496 	for (; i < subprog_end; i++) {
1497 		if (insn[i].code != (BPF_JMP | BPF_CALL))
1498 			continue;
1499 		if (insn[i].src_reg != BPF_PSEUDO_CALL)
1500 			continue;
1501 		/* remember insn and function to return to */
1502 		ret_insn[frame] = i + 1;
1503 		ret_prog[frame] = subprog;
1504 
1505 		/* find the callee */
1506 		i = i + insn[i].imm + 1;
1507 		subprog = find_subprog(env, i);
1508 		if (subprog < 0) {
1509 			WARN_ONCE(1, "verifier bug. No program starts at insn %d\n",
1510 				  i);
1511 			return -EFAULT;
1512 		}
1513 		subprog++;
1514 		frame++;
1515 		if (frame >= MAX_CALL_FRAMES) {
1516 			WARN_ONCE(1, "verifier bug. Call stack is too deep\n");
1517 			return -EFAULT;
1518 		}
1519 		goto process_func;
1520 	}
1521 	/* end of for() loop means the last insn of the 'subprog'
1522 	 * was reached. Doesn't matter whether it was JA or EXIT
1523 	 */
1524 	if (frame == 0)
1525 		return 0;
1526 	depth -= round_up(max_t(u32, env->subprog_stack_depth[subprog], 1), 32);
1527 	frame--;
1528 	i = ret_insn[frame];
1529 	subprog = ret_prog[frame];
1530 	goto continue_func;
1531 }
1532 
1533 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
1534 static int get_callee_stack_depth(struct bpf_verifier_env *env,
1535 				  const struct bpf_insn *insn, int idx)
1536 {
1537 	int start = idx + insn->imm + 1, subprog;
1538 
1539 	subprog = find_subprog(env, start);
1540 	if (subprog < 0) {
1541 		WARN_ONCE(1, "verifier bug. No program starts at insn %d\n",
1542 			  start);
1543 		return -EFAULT;
1544 	}
1545 	subprog++;
1546 	return env->subprog_stack_depth[subprog];
1547 }
1548 #endif
1549 
1550 /* truncate register to smaller size (in bytes)
1551  * must be called with size < BPF_REG_SIZE
1552  */
1553 static void coerce_reg_to_size(struct bpf_reg_state *reg, int size)
1554 {
1555 	u64 mask;
1556 
1557 	/* clear high bits in bit representation */
1558 	reg->var_off = tnum_cast(reg->var_off, size);
1559 
1560 	/* fix arithmetic bounds */
1561 	mask = ((u64)1 << (size * 8)) - 1;
1562 	if ((reg->umin_value & ~mask) == (reg->umax_value & ~mask)) {
1563 		reg->umin_value &= mask;
1564 		reg->umax_value &= mask;
1565 	} else {
1566 		reg->umin_value = 0;
1567 		reg->umax_value = mask;
1568 	}
1569 	reg->smin_value = reg->umin_value;
1570 	reg->smax_value = reg->umax_value;
1571 }
1572 
1573 /* check whether memory at (regno + off) is accessible for t = (read | write)
1574  * if t==write, value_regno is a register which value is stored into memory
1575  * if t==read, value_regno is a register which will receive the value from memory
1576  * if t==write && value_regno==-1, some unknown value is stored into memory
1577  * if t==read && value_regno==-1, don't care what we read from memory
1578  */
1579 static int check_mem_access(struct bpf_verifier_env *env, int insn_idx, u32 regno, int off,
1580 			    int bpf_size, enum bpf_access_type t,
1581 			    int value_regno)
1582 {
1583 	struct bpf_reg_state *regs = cur_regs(env);
1584 	struct bpf_reg_state *reg = regs + regno;
1585 	struct bpf_func_state *state;
1586 	int size, err = 0;
1587 
1588 	size = bpf_size_to_bytes(bpf_size);
1589 	if (size < 0)
1590 		return size;
1591 
1592 	/* alignment checks will add in reg->off themselves */
1593 	err = check_ptr_alignment(env, reg, off, size);
1594 	if (err)
1595 		return err;
1596 
1597 	/* for access checks, reg->off is just part of off */
1598 	off += reg->off;
1599 
1600 	if (reg->type == PTR_TO_MAP_VALUE) {
1601 		if (t == BPF_WRITE && value_regno >= 0 &&
1602 		    is_pointer_value(env, value_regno)) {
1603 			verbose(env, "R%d leaks addr into map\n", value_regno);
1604 			return -EACCES;
1605 		}
1606 
1607 		err = check_map_access(env, regno, off, size, false);
1608 		if (!err && t == BPF_READ && value_regno >= 0)
1609 			mark_reg_unknown(env, regs, value_regno);
1610 
1611 	} else if (reg->type == PTR_TO_CTX) {
1612 		enum bpf_reg_type reg_type = SCALAR_VALUE;
1613 
1614 		if (t == BPF_WRITE && value_regno >= 0 &&
1615 		    is_pointer_value(env, value_regno)) {
1616 			verbose(env, "R%d leaks addr into ctx\n", value_regno);
1617 			return -EACCES;
1618 		}
1619 		/* ctx accesses must be at a fixed offset, so that we can
1620 		 * determine what type of data were returned.
1621 		 */
1622 		if (reg->off) {
1623 			verbose(env,
1624 				"dereference of modified ctx ptr R%d off=%d+%d, ctx+const is allowed, ctx+const+const is not\n",
1625 				regno, reg->off, off - reg->off);
1626 			return -EACCES;
1627 		}
1628 		if (!tnum_is_const(reg->var_off) || reg->var_off.value) {
1629 			char tn_buf[48];
1630 
1631 			tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off);
1632 			verbose(env,
1633 				"variable ctx access var_off=%s off=%d size=%d",
1634 				tn_buf, off, size);
1635 			return -EACCES;
1636 		}
1637 		err = check_ctx_access(env, insn_idx, off, size, t, &reg_type);
1638 		if (!err && t == BPF_READ && value_regno >= 0) {
1639 			/* ctx access returns either a scalar, or a
1640 			 * PTR_TO_PACKET[_META,_END]. In the latter
1641 			 * case, we know the offset is zero.
1642 			 */
1643 			if (reg_type == SCALAR_VALUE)
1644 				mark_reg_unknown(env, regs, value_regno);
1645 			else
1646 				mark_reg_known_zero(env, regs,
1647 						    value_regno);
1648 			regs[value_regno].id = 0;
1649 			regs[value_regno].off = 0;
1650 			regs[value_regno].range = 0;
1651 			regs[value_regno].type = reg_type;
1652 		}
1653 
1654 	} else if (reg->type == PTR_TO_STACK) {
1655 		/* stack accesses must be at a fixed offset, so that we can
1656 		 * determine what type of data were returned.
1657 		 * See check_stack_read().
1658 		 */
1659 		if (!tnum_is_const(reg->var_off)) {
1660 			char tn_buf[48];
1661 
1662 			tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off);
1663 			verbose(env, "variable stack access var_off=%s off=%d size=%d",
1664 				tn_buf, off, size);
1665 			return -EACCES;
1666 		}
1667 		off += reg->var_off.value;
1668 		if (off >= 0 || off < -MAX_BPF_STACK) {
1669 			verbose(env, "invalid stack off=%d size=%d\n", off,
1670 				size);
1671 			return -EACCES;
1672 		}
1673 
1674 		state = func(env, reg);
1675 		err = update_stack_depth(env, state, off);
1676 		if (err)
1677 			return err;
1678 
1679 		if (t == BPF_WRITE)
1680 			err = check_stack_write(env, state, off, size,
1681 						value_regno);
1682 		else
1683 			err = check_stack_read(env, state, off, size,
1684 					       value_regno);
1685 	} else if (reg_is_pkt_pointer(reg)) {
1686 		if (t == BPF_WRITE && !may_access_direct_pkt_data(env, NULL, t)) {
1687 			verbose(env, "cannot write into packet\n");
1688 			return -EACCES;
1689 		}
1690 		if (t == BPF_WRITE && value_regno >= 0 &&
1691 		    is_pointer_value(env, value_regno)) {
1692 			verbose(env, "R%d leaks addr into packet\n",
1693 				value_regno);
1694 			return -EACCES;
1695 		}
1696 		err = check_packet_access(env, regno, off, size, false);
1697 		if (!err && t == BPF_READ && value_regno >= 0)
1698 			mark_reg_unknown(env, regs, value_regno);
1699 	} else {
1700 		verbose(env, "R%d invalid mem access '%s'\n", regno,
1701 			reg_type_str[reg->type]);
1702 		return -EACCES;
1703 	}
1704 
1705 	if (!err && size < BPF_REG_SIZE && value_regno >= 0 && t == BPF_READ &&
1706 	    regs[value_regno].type == SCALAR_VALUE) {
1707 		/* b/h/w load zero-extends, mark upper bits as known 0 */
1708 		coerce_reg_to_size(&regs[value_regno], size);
1709 	}
1710 	return err;
1711 }
1712 
1713 static int check_xadd(struct bpf_verifier_env *env, int insn_idx, struct bpf_insn *insn)
1714 {
1715 	int err;
1716 
1717 	if ((BPF_SIZE(insn->code) != BPF_W && BPF_SIZE(insn->code) != BPF_DW) ||
1718 	    insn->imm != 0) {
1719 		verbose(env, "BPF_XADD uses reserved fields\n");
1720 		return -EINVAL;
1721 	}
1722 
1723 	/* check src1 operand */
1724 	err = check_reg_arg(env, insn->src_reg, SRC_OP);
1725 	if (err)
1726 		return err;
1727 
1728 	/* check src2 operand */
1729 	err = check_reg_arg(env, insn->dst_reg, SRC_OP);
1730 	if (err)
1731 		return err;
1732 
1733 	if (is_pointer_value(env, insn->src_reg)) {
1734 		verbose(env, "R%d leaks addr into mem\n", insn->src_reg);
1735 		return -EACCES;
1736 	}
1737 
1738 	if (is_ctx_reg(env, insn->dst_reg)) {
1739 		verbose(env, "BPF_XADD stores into R%d context is not allowed\n",
1740 			insn->dst_reg);
1741 		return -EACCES;
1742 	}
1743 
1744 	/* check whether atomic_add can read the memory */
1745 	err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off,
1746 			       BPF_SIZE(insn->code), BPF_READ, -1);
1747 	if (err)
1748 		return err;
1749 
1750 	/* check whether atomic_add can write into the same memory */
1751 	return check_mem_access(env, insn_idx, insn->dst_reg, insn->off,
1752 				BPF_SIZE(insn->code), BPF_WRITE, -1);
1753 }
1754 
1755 /* when register 'regno' is passed into function that will read 'access_size'
1756  * bytes from that pointer, make sure that it's within stack boundary
1757  * and all elements of stack are initialized.
1758  * Unlike most pointer bounds-checking functions, this one doesn't take an
1759  * 'off' argument, so it has to add in reg->off itself.
1760  */
1761 static int check_stack_boundary(struct bpf_verifier_env *env, int regno,
1762 				int access_size, bool zero_size_allowed,
1763 				struct bpf_call_arg_meta *meta)
1764 {
1765 	struct bpf_reg_state *reg = cur_regs(env) + regno;
1766 	struct bpf_func_state *state = func(env, reg);
1767 	int off, i, slot, spi;
1768 
1769 	if (reg->type != PTR_TO_STACK) {
1770 		/* Allow zero-byte read from NULL, regardless of pointer type */
1771 		if (zero_size_allowed && access_size == 0 &&
1772 		    register_is_null(reg))
1773 			return 0;
1774 
1775 		verbose(env, "R%d type=%s expected=%s\n", regno,
1776 			reg_type_str[reg->type],
1777 			reg_type_str[PTR_TO_STACK]);
1778 		return -EACCES;
1779 	}
1780 
1781 	/* Only allow fixed-offset stack reads */
1782 	if (!tnum_is_const(reg->var_off)) {
1783 		char tn_buf[48];
1784 
1785 		tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off);
1786 		verbose(env, "invalid variable stack read R%d var_off=%s\n",
1787 			regno, tn_buf);
1788 		return -EACCES;
1789 	}
1790 	off = reg->off + reg->var_off.value;
1791 	if (off >= 0 || off < -MAX_BPF_STACK || off + access_size > 0 ||
1792 	    access_size < 0 || (access_size == 0 && !zero_size_allowed)) {
1793 		verbose(env, "invalid stack type R%d off=%d access_size=%d\n",
1794 			regno, off, access_size);
1795 		return -EACCES;
1796 	}
1797 
1798 	if (meta && meta->raw_mode) {
1799 		meta->access_size = access_size;
1800 		meta->regno = regno;
1801 		return 0;
1802 	}
1803 
1804 	for (i = 0; i < access_size; i++) {
1805 		u8 *stype;
1806 
1807 		slot = -(off + i) - 1;
1808 		spi = slot / BPF_REG_SIZE;
1809 		if (state->allocated_stack <= slot)
1810 			goto err;
1811 		stype = &state->stack[spi].slot_type[slot % BPF_REG_SIZE];
1812 		if (*stype == STACK_MISC)
1813 			goto mark;
1814 		if (*stype == STACK_ZERO) {
1815 			/* helper can write anything into the stack */
1816 			*stype = STACK_MISC;
1817 			goto mark;
1818 		}
1819 err:
1820 		verbose(env, "invalid indirect read from stack off %d+%d size %d\n",
1821 			off, i, access_size);
1822 		return -EACCES;
1823 mark:
1824 		/* reading any byte out of 8-byte 'spill_slot' will cause
1825 		 * the whole slot to be marked as 'read'
1826 		 */
1827 		mark_stack_slot_read(env, env->cur_state, env->cur_state->parent,
1828 				     spi, state->frameno);
1829 	}
1830 	return update_stack_depth(env, state, off);
1831 }
1832 
1833 static int check_helper_mem_access(struct bpf_verifier_env *env, int regno,
1834 				   int access_size, bool zero_size_allowed,
1835 				   struct bpf_call_arg_meta *meta)
1836 {
1837 	struct bpf_reg_state *regs = cur_regs(env), *reg = &regs[regno];
1838 
1839 	switch (reg->type) {
1840 	case PTR_TO_PACKET:
1841 	case PTR_TO_PACKET_META:
1842 		return check_packet_access(env, regno, reg->off, access_size,
1843 					   zero_size_allowed);
1844 	case PTR_TO_MAP_VALUE:
1845 		return check_map_access(env, regno, reg->off, access_size,
1846 					zero_size_allowed);
1847 	default: /* scalar_value|ptr_to_stack or invalid ptr */
1848 		return check_stack_boundary(env, regno, access_size,
1849 					    zero_size_allowed, meta);
1850 	}
1851 }
1852 
1853 static bool arg_type_is_mem_ptr(enum bpf_arg_type type)
1854 {
1855 	return type == ARG_PTR_TO_MEM ||
1856 	       type == ARG_PTR_TO_MEM_OR_NULL ||
1857 	       type == ARG_PTR_TO_UNINIT_MEM;
1858 }
1859 
1860 static bool arg_type_is_mem_size(enum bpf_arg_type type)
1861 {
1862 	return type == ARG_CONST_SIZE ||
1863 	       type == ARG_CONST_SIZE_OR_ZERO;
1864 }
1865 
1866 static int check_func_arg(struct bpf_verifier_env *env, u32 regno,
1867 			  enum bpf_arg_type arg_type,
1868 			  struct bpf_call_arg_meta *meta)
1869 {
1870 	struct bpf_reg_state *regs = cur_regs(env), *reg = &regs[regno];
1871 	enum bpf_reg_type expected_type, type = reg->type;
1872 	int err = 0;
1873 
1874 	if (arg_type == ARG_DONTCARE)
1875 		return 0;
1876 
1877 	err = check_reg_arg(env, regno, SRC_OP);
1878 	if (err)
1879 		return err;
1880 
1881 	if (arg_type == ARG_ANYTHING) {
1882 		if (is_pointer_value(env, regno)) {
1883 			verbose(env, "R%d leaks addr into helper function\n",
1884 				regno);
1885 			return -EACCES;
1886 		}
1887 		return 0;
1888 	}
1889 
1890 	if (type_is_pkt_pointer(type) &&
1891 	    !may_access_direct_pkt_data(env, meta, BPF_READ)) {
1892 		verbose(env, "helper access to the packet is not allowed\n");
1893 		return -EACCES;
1894 	}
1895 
1896 	if (arg_type == ARG_PTR_TO_MAP_KEY ||
1897 	    arg_type == ARG_PTR_TO_MAP_VALUE) {
1898 		expected_type = PTR_TO_STACK;
1899 		if (!type_is_pkt_pointer(type) &&
1900 		    type != expected_type)
1901 			goto err_type;
1902 	} else if (arg_type == ARG_CONST_SIZE ||
1903 		   arg_type == ARG_CONST_SIZE_OR_ZERO) {
1904 		expected_type = SCALAR_VALUE;
1905 		if (type != expected_type)
1906 			goto err_type;
1907 	} else if (arg_type == ARG_CONST_MAP_PTR) {
1908 		expected_type = CONST_PTR_TO_MAP;
1909 		if (type != expected_type)
1910 			goto err_type;
1911 	} else if (arg_type == ARG_PTR_TO_CTX) {
1912 		expected_type = PTR_TO_CTX;
1913 		if (type != expected_type)
1914 			goto err_type;
1915 	} else if (arg_type_is_mem_ptr(arg_type)) {
1916 		expected_type = PTR_TO_STACK;
1917 		/* One exception here. In case function allows for NULL to be
1918 		 * passed in as argument, it's a SCALAR_VALUE type. Final test
1919 		 * happens during stack boundary checking.
1920 		 */
1921 		if (register_is_null(reg) &&
1922 		    arg_type == ARG_PTR_TO_MEM_OR_NULL)
1923 			/* final test in check_stack_boundary() */;
1924 		else if (!type_is_pkt_pointer(type) &&
1925 			 type != PTR_TO_MAP_VALUE &&
1926 			 type != expected_type)
1927 			goto err_type;
1928 		meta->raw_mode = arg_type == ARG_PTR_TO_UNINIT_MEM;
1929 	} else {
1930 		verbose(env, "unsupported arg_type %d\n", arg_type);
1931 		return -EFAULT;
1932 	}
1933 
1934 	if (arg_type == ARG_CONST_MAP_PTR) {
1935 		/* bpf_map_xxx(map_ptr) call: remember that map_ptr */
1936 		meta->map_ptr = reg->map_ptr;
1937 	} else if (arg_type == ARG_PTR_TO_MAP_KEY) {
1938 		/* bpf_map_xxx(..., map_ptr, ..., key) call:
1939 		 * check that [key, key + map->key_size) are within
1940 		 * stack limits and initialized
1941 		 */
1942 		if (!meta->map_ptr) {
1943 			/* in function declaration map_ptr must come before
1944 			 * map_key, so that it's verified and known before
1945 			 * we have to check map_key here. Otherwise it means
1946 			 * that kernel subsystem misconfigured verifier
1947 			 */
1948 			verbose(env, "invalid map_ptr to access map->key\n");
1949 			return -EACCES;
1950 		}
1951 		if (type_is_pkt_pointer(type))
1952 			err = check_packet_access(env, regno, reg->off,
1953 						  meta->map_ptr->key_size,
1954 						  false);
1955 		else
1956 			err = check_stack_boundary(env, regno,
1957 						   meta->map_ptr->key_size,
1958 						   false, NULL);
1959 	} else if (arg_type == ARG_PTR_TO_MAP_VALUE) {
1960 		/* bpf_map_xxx(..., map_ptr, ..., value) call:
1961 		 * check [value, value + map->value_size) validity
1962 		 */
1963 		if (!meta->map_ptr) {
1964 			/* kernel subsystem misconfigured verifier */
1965 			verbose(env, "invalid map_ptr to access map->value\n");
1966 			return -EACCES;
1967 		}
1968 		if (type_is_pkt_pointer(type))
1969 			err = check_packet_access(env, regno, reg->off,
1970 						  meta->map_ptr->value_size,
1971 						  false);
1972 		else
1973 			err = check_stack_boundary(env, regno,
1974 						   meta->map_ptr->value_size,
1975 						   false, NULL);
1976 	} else if (arg_type_is_mem_size(arg_type)) {
1977 		bool zero_size_allowed = (arg_type == ARG_CONST_SIZE_OR_ZERO);
1978 
1979 		/* The register is SCALAR_VALUE; the access check
1980 		 * happens using its boundaries.
1981 		 */
1982 		if (!tnum_is_const(reg->var_off))
1983 			/* For unprivileged variable accesses, disable raw
1984 			 * mode so that the program is required to
1985 			 * initialize all the memory that the helper could
1986 			 * just partially fill up.
1987 			 */
1988 			meta = NULL;
1989 
1990 		if (reg->smin_value < 0) {
1991 			verbose(env, "R%d min value is negative, either use unsigned or 'var &= const'\n",
1992 				regno);
1993 			return -EACCES;
1994 		}
1995 
1996 		if (reg->umin_value == 0) {
1997 			err = check_helper_mem_access(env, regno - 1, 0,
1998 						      zero_size_allowed,
1999 						      meta);
2000 			if (err)
2001 				return err;
2002 		}
2003 
2004 		if (reg->umax_value >= BPF_MAX_VAR_SIZ) {
2005 			verbose(env, "R%d unbounded memory access, use 'var &= const' or 'if (var < const)'\n",
2006 				regno);
2007 			return -EACCES;
2008 		}
2009 		err = check_helper_mem_access(env, regno - 1,
2010 					      reg->umax_value,
2011 					      zero_size_allowed, meta);
2012 	}
2013 
2014 	return err;
2015 err_type:
2016 	verbose(env, "R%d type=%s expected=%s\n", regno,
2017 		reg_type_str[type], reg_type_str[expected_type]);
2018 	return -EACCES;
2019 }
2020 
2021 static int check_map_func_compatibility(struct bpf_verifier_env *env,
2022 					struct bpf_map *map, int func_id)
2023 {
2024 	if (!map)
2025 		return 0;
2026 
2027 	/* We need a two way check, first is from map perspective ... */
2028 	switch (map->map_type) {
2029 	case BPF_MAP_TYPE_PROG_ARRAY:
2030 		if (func_id != BPF_FUNC_tail_call)
2031 			goto error;
2032 		break;
2033 	case BPF_MAP_TYPE_PERF_EVENT_ARRAY:
2034 		if (func_id != BPF_FUNC_perf_event_read &&
2035 		    func_id != BPF_FUNC_perf_event_output &&
2036 		    func_id != BPF_FUNC_perf_event_read_value)
2037 			goto error;
2038 		break;
2039 	case BPF_MAP_TYPE_STACK_TRACE:
2040 		if (func_id != BPF_FUNC_get_stackid)
2041 			goto error;
2042 		break;
2043 	case BPF_MAP_TYPE_CGROUP_ARRAY:
2044 		if (func_id != BPF_FUNC_skb_under_cgroup &&
2045 		    func_id != BPF_FUNC_current_task_under_cgroup)
2046 			goto error;
2047 		break;
2048 	/* devmap returns a pointer to a live net_device ifindex that we cannot
2049 	 * allow to be modified from bpf side. So do not allow lookup elements
2050 	 * for now.
2051 	 */
2052 	case BPF_MAP_TYPE_DEVMAP:
2053 		if (func_id != BPF_FUNC_redirect_map)
2054 			goto error;
2055 		break;
2056 	/* Restrict bpf side of cpumap, open when use-cases appear */
2057 	case BPF_MAP_TYPE_CPUMAP:
2058 		if (func_id != BPF_FUNC_redirect_map)
2059 			goto error;
2060 		break;
2061 	case BPF_MAP_TYPE_ARRAY_OF_MAPS:
2062 	case BPF_MAP_TYPE_HASH_OF_MAPS:
2063 		if (func_id != BPF_FUNC_map_lookup_elem)
2064 			goto error;
2065 		break;
2066 	case BPF_MAP_TYPE_SOCKMAP:
2067 		if (func_id != BPF_FUNC_sk_redirect_map &&
2068 		    func_id != BPF_FUNC_sock_map_update &&
2069 		    func_id != BPF_FUNC_map_delete_elem)
2070 			goto error;
2071 		break;
2072 	default:
2073 		break;
2074 	}
2075 
2076 	/* ... and second from the function itself. */
2077 	switch (func_id) {
2078 	case BPF_FUNC_tail_call:
2079 		if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY)
2080 			goto error;
2081 		if (env->subprog_cnt) {
2082 			verbose(env, "tail_calls are not allowed in programs with bpf-to-bpf calls\n");
2083 			return -EINVAL;
2084 		}
2085 		break;
2086 	case BPF_FUNC_perf_event_read:
2087 	case BPF_FUNC_perf_event_output:
2088 	case BPF_FUNC_perf_event_read_value:
2089 		if (map->map_type != BPF_MAP_TYPE_PERF_EVENT_ARRAY)
2090 			goto error;
2091 		break;
2092 	case BPF_FUNC_get_stackid:
2093 		if (map->map_type != BPF_MAP_TYPE_STACK_TRACE)
2094 			goto error;
2095 		break;
2096 	case BPF_FUNC_current_task_under_cgroup:
2097 	case BPF_FUNC_skb_under_cgroup:
2098 		if (map->map_type != BPF_MAP_TYPE_CGROUP_ARRAY)
2099 			goto error;
2100 		break;
2101 	case BPF_FUNC_redirect_map:
2102 		if (map->map_type != BPF_MAP_TYPE_DEVMAP &&
2103 		    map->map_type != BPF_MAP_TYPE_CPUMAP)
2104 			goto error;
2105 		break;
2106 	case BPF_FUNC_sk_redirect_map:
2107 		if (map->map_type != BPF_MAP_TYPE_SOCKMAP)
2108 			goto error;
2109 		break;
2110 	case BPF_FUNC_sock_map_update:
2111 		if (map->map_type != BPF_MAP_TYPE_SOCKMAP)
2112 			goto error;
2113 		break;
2114 	default:
2115 		break;
2116 	}
2117 
2118 	return 0;
2119 error:
2120 	verbose(env, "cannot pass map_type %d into func %s#%d\n",
2121 		map->map_type, func_id_name(func_id), func_id);
2122 	return -EINVAL;
2123 }
2124 
2125 static bool check_raw_mode_ok(const struct bpf_func_proto *fn)
2126 {
2127 	int count = 0;
2128 
2129 	if (fn->arg1_type == ARG_PTR_TO_UNINIT_MEM)
2130 		count++;
2131 	if (fn->arg2_type == ARG_PTR_TO_UNINIT_MEM)
2132 		count++;
2133 	if (fn->arg3_type == ARG_PTR_TO_UNINIT_MEM)
2134 		count++;
2135 	if (fn->arg4_type == ARG_PTR_TO_UNINIT_MEM)
2136 		count++;
2137 	if (fn->arg5_type == ARG_PTR_TO_UNINIT_MEM)
2138 		count++;
2139 
2140 	/* We only support one arg being in raw mode at the moment,
2141 	 * which is sufficient for the helper functions we have
2142 	 * right now.
2143 	 */
2144 	return count <= 1;
2145 }
2146 
2147 static bool check_args_pair_invalid(enum bpf_arg_type arg_curr,
2148 				    enum bpf_arg_type arg_next)
2149 {
2150 	return (arg_type_is_mem_ptr(arg_curr) &&
2151 	        !arg_type_is_mem_size(arg_next)) ||
2152 	       (!arg_type_is_mem_ptr(arg_curr) &&
2153 		arg_type_is_mem_size(arg_next));
2154 }
2155 
2156 static bool check_arg_pair_ok(const struct bpf_func_proto *fn)
2157 {
2158 	/* bpf_xxx(..., buf, len) call will access 'len'
2159 	 * bytes from memory 'buf'. Both arg types need
2160 	 * to be paired, so make sure there's no buggy
2161 	 * helper function specification.
2162 	 */
2163 	if (arg_type_is_mem_size(fn->arg1_type) ||
2164 	    arg_type_is_mem_ptr(fn->arg5_type)  ||
2165 	    check_args_pair_invalid(fn->arg1_type, fn->arg2_type) ||
2166 	    check_args_pair_invalid(fn->arg2_type, fn->arg3_type) ||
2167 	    check_args_pair_invalid(fn->arg3_type, fn->arg4_type) ||
2168 	    check_args_pair_invalid(fn->arg4_type, fn->arg5_type))
2169 		return false;
2170 
2171 	return true;
2172 }
2173 
2174 static int check_func_proto(const struct bpf_func_proto *fn)
2175 {
2176 	return check_raw_mode_ok(fn) &&
2177 	       check_arg_pair_ok(fn) ? 0 : -EINVAL;
2178 }
2179 
2180 /* Packet data might have moved, any old PTR_TO_PACKET[_META,_END]
2181  * are now invalid, so turn them into unknown SCALAR_VALUE.
2182  */
2183 static void __clear_all_pkt_pointers(struct bpf_verifier_env *env,
2184 				     struct bpf_func_state *state)
2185 {
2186 	struct bpf_reg_state *regs = state->regs, *reg;
2187 	int i;
2188 
2189 	for (i = 0; i < MAX_BPF_REG; i++)
2190 		if (reg_is_pkt_pointer_any(&regs[i]))
2191 			mark_reg_unknown(env, regs, i);
2192 
2193 	for (i = 0; i < state->allocated_stack / BPF_REG_SIZE; i++) {
2194 		if (state->stack[i].slot_type[0] != STACK_SPILL)
2195 			continue;
2196 		reg = &state->stack[i].spilled_ptr;
2197 		if (reg_is_pkt_pointer_any(reg))
2198 			__mark_reg_unknown(reg);
2199 	}
2200 }
2201 
2202 static void clear_all_pkt_pointers(struct bpf_verifier_env *env)
2203 {
2204 	struct bpf_verifier_state *vstate = env->cur_state;
2205 	int i;
2206 
2207 	for (i = 0; i <= vstate->curframe; i++)
2208 		__clear_all_pkt_pointers(env, vstate->frame[i]);
2209 }
2210 
2211 static int check_func_call(struct bpf_verifier_env *env, struct bpf_insn *insn,
2212 			   int *insn_idx)
2213 {
2214 	struct bpf_verifier_state *state = env->cur_state;
2215 	struct bpf_func_state *caller, *callee;
2216 	int i, subprog, target_insn;
2217 
2218 	if (state->curframe + 1 >= MAX_CALL_FRAMES) {
2219 		verbose(env, "the call stack of %d frames is too deep\n",
2220 			state->curframe + 2);
2221 		return -E2BIG;
2222 	}
2223 
2224 	target_insn = *insn_idx + insn->imm;
2225 	subprog = find_subprog(env, target_insn + 1);
2226 	if (subprog < 0) {
2227 		verbose(env, "verifier bug. No program starts at insn %d\n",
2228 			target_insn + 1);
2229 		return -EFAULT;
2230 	}
2231 
2232 	caller = state->frame[state->curframe];
2233 	if (state->frame[state->curframe + 1]) {
2234 		verbose(env, "verifier bug. Frame %d already allocated\n",
2235 			state->curframe + 1);
2236 		return -EFAULT;
2237 	}
2238 
2239 	callee = kzalloc(sizeof(*callee), GFP_KERNEL);
2240 	if (!callee)
2241 		return -ENOMEM;
2242 	state->frame[state->curframe + 1] = callee;
2243 
2244 	/* callee cannot access r0, r6 - r9 for reading and has to write
2245 	 * into its own stack before reading from it.
2246 	 * callee can read/write into caller's stack
2247 	 */
2248 	init_func_state(env, callee,
2249 			/* remember the callsite, it will be used by bpf_exit */
2250 			*insn_idx /* callsite */,
2251 			state->curframe + 1 /* frameno within this callchain */,
2252 			subprog + 1 /* subprog number within this prog */);
2253 
2254 	/* copy r1 - r5 args that callee can access */
2255 	for (i = BPF_REG_1; i <= BPF_REG_5; i++)
2256 		callee->regs[i] = caller->regs[i];
2257 
2258 	/* after the call regsiters r0 - r5 were scratched */
2259 	for (i = 0; i < CALLER_SAVED_REGS; i++) {
2260 		mark_reg_not_init(env, caller->regs, caller_saved[i]);
2261 		check_reg_arg(env, caller_saved[i], DST_OP_NO_MARK);
2262 	}
2263 
2264 	/* only increment it after check_reg_arg() finished */
2265 	state->curframe++;
2266 
2267 	/* and go analyze first insn of the callee */
2268 	*insn_idx = target_insn;
2269 
2270 	if (env->log.level) {
2271 		verbose(env, "caller:\n");
2272 		print_verifier_state(env, caller);
2273 		verbose(env, "callee:\n");
2274 		print_verifier_state(env, callee);
2275 	}
2276 	return 0;
2277 }
2278 
2279 static int prepare_func_exit(struct bpf_verifier_env *env, int *insn_idx)
2280 {
2281 	struct bpf_verifier_state *state = env->cur_state;
2282 	struct bpf_func_state *caller, *callee;
2283 	struct bpf_reg_state *r0;
2284 
2285 	callee = state->frame[state->curframe];
2286 	r0 = &callee->regs[BPF_REG_0];
2287 	if (r0->type == PTR_TO_STACK) {
2288 		/* technically it's ok to return caller's stack pointer
2289 		 * (or caller's caller's pointer) back to the caller,
2290 		 * since these pointers are valid. Only current stack
2291 		 * pointer will be invalid as soon as function exits,
2292 		 * but let's be conservative
2293 		 */
2294 		verbose(env, "cannot return stack pointer to the caller\n");
2295 		return -EINVAL;
2296 	}
2297 
2298 	state->curframe--;
2299 	caller = state->frame[state->curframe];
2300 	/* return to the caller whatever r0 had in the callee */
2301 	caller->regs[BPF_REG_0] = *r0;
2302 
2303 	*insn_idx = callee->callsite + 1;
2304 	if (env->log.level) {
2305 		verbose(env, "returning from callee:\n");
2306 		print_verifier_state(env, callee);
2307 		verbose(env, "to caller at %d:\n", *insn_idx);
2308 		print_verifier_state(env, caller);
2309 	}
2310 	/* clear everything in the callee */
2311 	free_func_state(callee);
2312 	state->frame[state->curframe + 1] = NULL;
2313 	return 0;
2314 }
2315 
2316 static int check_helper_call(struct bpf_verifier_env *env, int func_id, int insn_idx)
2317 {
2318 	const struct bpf_func_proto *fn = NULL;
2319 	struct bpf_reg_state *regs;
2320 	struct bpf_call_arg_meta meta;
2321 	bool changes_data;
2322 	int i, err;
2323 
2324 	/* find function prototype */
2325 	if (func_id < 0 || func_id >= __BPF_FUNC_MAX_ID) {
2326 		verbose(env, "invalid func %s#%d\n", func_id_name(func_id),
2327 			func_id);
2328 		return -EINVAL;
2329 	}
2330 
2331 	if (env->ops->get_func_proto)
2332 		fn = env->ops->get_func_proto(func_id);
2333 	if (!fn) {
2334 		verbose(env, "unknown func %s#%d\n", func_id_name(func_id),
2335 			func_id);
2336 		return -EINVAL;
2337 	}
2338 
2339 	/* eBPF programs must be GPL compatible to use GPL-ed functions */
2340 	if (!env->prog->gpl_compatible && fn->gpl_only) {
2341 		verbose(env, "cannot call GPL only function from proprietary program\n");
2342 		return -EINVAL;
2343 	}
2344 
2345 	/* With LD_ABS/IND some JITs save/restore skb from r1. */
2346 	changes_data = bpf_helper_changes_pkt_data(fn->func);
2347 	if (changes_data && fn->arg1_type != ARG_PTR_TO_CTX) {
2348 		verbose(env, "kernel subsystem misconfigured func %s#%d: r1 != ctx\n",
2349 			func_id_name(func_id), func_id);
2350 		return -EINVAL;
2351 	}
2352 
2353 	memset(&meta, 0, sizeof(meta));
2354 	meta.pkt_access = fn->pkt_access;
2355 
2356 	err = check_func_proto(fn);
2357 	if (err) {
2358 		verbose(env, "kernel subsystem misconfigured func %s#%d\n",
2359 			func_id_name(func_id), func_id);
2360 		return err;
2361 	}
2362 
2363 	/* check args */
2364 	err = check_func_arg(env, BPF_REG_1, fn->arg1_type, &meta);
2365 	if (err)
2366 		return err;
2367 	err = check_func_arg(env, BPF_REG_2, fn->arg2_type, &meta);
2368 	if (err)
2369 		return err;
2370 	if (func_id == BPF_FUNC_tail_call) {
2371 		if (meta.map_ptr == NULL) {
2372 			verbose(env, "verifier bug\n");
2373 			return -EINVAL;
2374 		}
2375 		env->insn_aux_data[insn_idx].map_ptr = meta.map_ptr;
2376 	}
2377 	err = check_func_arg(env, BPF_REG_3, fn->arg3_type, &meta);
2378 	if (err)
2379 		return err;
2380 	err = check_func_arg(env, BPF_REG_4, fn->arg4_type, &meta);
2381 	if (err)
2382 		return err;
2383 	err = check_func_arg(env, BPF_REG_5, fn->arg5_type, &meta);
2384 	if (err)
2385 		return err;
2386 
2387 	/* Mark slots with STACK_MISC in case of raw mode, stack offset
2388 	 * is inferred from register state.
2389 	 */
2390 	for (i = 0; i < meta.access_size; i++) {
2391 		err = check_mem_access(env, insn_idx, meta.regno, i, BPF_B, BPF_WRITE, -1);
2392 		if (err)
2393 			return err;
2394 	}
2395 
2396 	regs = cur_regs(env);
2397 	/* reset caller saved regs */
2398 	for (i = 0; i < CALLER_SAVED_REGS; i++) {
2399 		mark_reg_not_init(env, regs, caller_saved[i]);
2400 		check_reg_arg(env, caller_saved[i], DST_OP_NO_MARK);
2401 	}
2402 
2403 	/* update return register (already marked as written above) */
2404 	if (fn->ret_type == RET_INTEGER) {
2405 		/* sets type to SCALAR_VALUE */
2406 		mark_reg_unknown(env, regs, BPF_REG_0);
2407 	} else if (fn->ret_type == RET_VOID) {
2408 		regs[BPF_REG_0].type = NOT_INIT;
2409 	} else if (fn->ret_type == RET_PTR_TO_MAP_VALUE_OR_NULL) {
2410 		struct bpf_insn_aux_data *insn_aux;
2411 
2412 		regs[BPF_REG_0].type = PTR_TO_MAP_VALUE_OR_NULL;
2413 		/* There is no offset yet applied, variable or fixed */
2414 		mark_reg_known_zero(env, regs, BPF_REG_0);
2415 		regs[BPF_REG_0].off = 0;
2416 		/* remember map_ptr, so that check_map_access()
2417 		 * can check 'value_size' boundary of memory access
2418 		 * to map element returned from bpf_map_lookup_elem()
2419 		 */
2420 		if (meta.map_ptr == NULL) {
2421 			verbose(env,
2422 				"kernel subsystem misconfigured verifier\n");
2423 			return -EINVAL;
2424 		}
2425 		regs[BPF_REG_0].map_ptr = meta.map_ptr;
2426 		regs[BPF_REG_0].id = ++env->id_gen;
2427 		insn_aux = &env->insn_aux_data[insn_idx];
2428 		if (!insn_aux->map_ptr)
2429 			insn_aux->map_ptr = meta.map_ptr;
2430 		else if (insn_aux->map_ptr != meta.map_ptr)
2431 			insn_aux->map_ptr = BPF_MAP_PTR_POISON;
2432 	} else {
2433 		verbose(env, "unknown return type %d of func %s#%d\n",
2434 			fn->ret_type, func_id_name(func_id), func_id);
2435 		return -EINVAL;
2436 	}
2437 
2438 	err = check_map_func_compatibility(env, meta.map_ptr, func_id);
2439 	if (err)
2440 		return err;
2441 
2442 	if (changes_data)
2443 		clear_all_pkt_pointers(env);
2444 	return 0;
2445 }
2446 
2447 static bool signed_add_overflows(s64 a, s64 b)
2448 {
2449 	/* Do the add in u64, where overflow is well-defined */
2450 	s64 res = (s64)((u64)a + (u64)b);
2451 
2452 	if (b < 0)
2453 		return res > a;
2454 	return res < a;
2455 }
2456 
2457 static bool signed_sub_overflows(s64 a, s64 b)
2458 {
2459 	/* Do the sub in u64, where overflow is well-defined */
2460 	s64 res = (s64)((u64)a - (u64)b);
2461 
2462 	if (b < 0)
2463 		return res < a;
2464 	return res > a;
2465 }
2466 
2467 static bool check_reg_sane_offset(struct bpf_verifier_env *env,
2468 				  const struct bpf_reg_state *reg,
2469 				  enum bpf_reg_type type)
2470 {
2471 	bool known = tnum_is_const(reg->var_off);
2472 	s64 val = reg->var_off.value;
2473 	s64 smin = reg->smin_value;
2474 
2475 	if (known && (val >= BPF_MAX_VAR_OFF || val <= -BPF_MAX_VAR_OFF)) {
2476 		verbose(env, "math between %s pointer and %lld is not allowed\n",
2477 			reg_type_str[type], val);
2478 		return false;
2479 	}
2480 
2481 	if (reg->off >= BPF_MAX_VAR_OFF || reg->off <= -BPF_MAX_VAR_OFF) {
2482 		verbose(env, "%s pointer offset %d is not allowed\n",
2483 			reg_type_str[type], reg->off);
2484 		return false;
2485 	}
2486 
2487 	if (smin == S64_MIN) {
2488 		verbose(env, "math between %s pointer and register with unbounded min value is not allowed\n",
2489 			reg_type_str[type]);
2490 		return false;
2491 	}
2492 
2493 	if (smin >= BPF_MAX_VAR_OFF || smin <= -BPF_MAX_VAR_OFF) {
2494 		verbose(env, "value %lld makes %s pointer be out of bounds\n",
2495 			smin, reg_type_str[type]);
2496 		return false;
2497 	}
2498 
2499 	return true;
2500 }
2501 
2502 /* Handles arithmetic on a pointer and a scalar: computes new min/max and var_off.
2503  * Caller should also handle BPF_MOV case separately.
2504  * If we return -EACCES, caller may want to try again treating pointer as a
2505  * scalar.  So we only emit a diagnostic if !env->allow_ptr_leaks.
2506  */
2507 static int adjust_ptr_min_max_vals(struct bpf_verifier_env *env,
2508 				   struct bpf_insn *insn,
2509 				   const struct bpf_reg_state *ptr_reg,
2510 				   const struct bpf_reg_state *off_reg)
2511 {
2512 	struct bpf_verifier_state *vstate = env->cur_state;
2513 	struct bpf_func_state *state = vstate->frame[vstate->curframe];
2514 	struct bpf_reg_state *regs = state->regs, *dst_reg;
2515 	bool known = tnum_is_const(off_reg->var_off);
2516 	s64 smin_val = off_reg->smin_value, smax_val = off_reg->smax_value,
2517 	    smin_ptr = ptr_reg->smin_value, smax_ptr = ptr_reg->smax_value;
2518 	u64 umin_val = off_reg->umin_value, umax_val = off_reg->umax_value,
2519 	    umin_ptr = ptr_reg->umin_value, umax_ptr = ptr_reg->umax_value;
2520 	u8 opcode = BPF_OP(insn->code);
2521 	u32 dst = insn->dst_reg;
2522 
2523 	dst_reg = &regs[dst];
2524 
2525 	if ((known && (smin_val != smax_val || umin_val != umax_val)) ||
2526 	    smin_val > smax_val || umin_val > umax_val) {
2527 		/* Taint dst register if offset had invalid bounds derived from
2528 		 * e.g. dead branches.
2529 		 */
2530 		__mark_reg_unknown(dst_reg);
2531 		return 0;
2532 	}
2533 
2534 	if (BPF_CLASS(insn->code) != BPF_ALU64) {
2535 		/* 32-bit ALU ops on pointers produce (meaningless) scalars */
2536 		verbose(env,
2537 			"R%d 32-bit pointer arithmetic prohibited\n",
2538 			dst);
2539 		return -EACCES;
2540 	}
2541 
2542 	if (ptr_reg->type == PTR_TO_MAP_VALUE_OR_NULL) {
2543 		verbose(env, "R%d pointer arithmetic on PTR_TO_MAP_VALUE_OR_NULL prohibited, null-check it first\n",
2544 			dst);
2545 		return -EACCES;
2546 	}
2547 	if (ptr_reg->type == CONST_PTR_TO_MAP) {
2548 		verbose(env, "R%d pointer arithmetic on CONST_PTR_TO_MAP prohibited\n",
2549 			dst);
2550 		return -EACCES;
2551 	}
2552 	if (ptr_reg->type == PTR_TO_PACKET_END) {
2553 		verbose(env, "R%d pointer arithmetic on PTR_TO_PACKET_END prohibited\n",
2554 			dst);
2555 		return -EACCES;
2556 	}
2557 
2558 	/* In case of 'scalar += pointer', dst_reg inherits pointer type and id.
2559 	 * The id may be overwritten later if we create a new variable offset.
2560 	 */
2561 	dst_reg->type = ptr_reg->type;
2562 	dst_reg->id = ptr_reg->id;
2563 
2564 	if (!check_reg_sane_offset(env, off_reg, ptr_reg->type) ||
2565 	    !check_reg_sane_offset(env, ptr_reg, ptr_reg->type))
2566 		return -EINVAL;
2567 
2568 	switch (opcode) {
2569 	case BPF_ADD:
2570 		/* We can take a fixed offset as long as it doesn't overflow
2571 		 * the s32 'off' field
2572 		 */
2573 		if (known && (ptr_reg->off + smin_val ==
2574 			      (s64)(s32)(ptr_reg->off + smin_val))) {
2575 			/* pointer += K.  Accumulate it into fixed offset */
2576 			dst_reg->smin_value = smin_ptr;
2577 			dst_reg->smax_value = smax_ptr;
2578 			dst_reg->umin_value = umin_ptr;
2579 			dst_reg->umax_value = umax_ptr;
2580 			dst_reg->var_off = ptr_reg->var_off;
2581 			dst_reg->off = ptr_reg->off + smin_val;
2582 			dst_reg->range = ptr_reg->range;
2583 			break;
2584 		}
2585 		/* A new variable offset is created.  Note that off_reg->off
2586 		 * == 0, since it's a scalar.
2587 		 * dst_reg gets the pointer type and since some positive
2588 		 * integer value was added to the pointer, give it a new 'id'
2589 		 * if it's a PTR_TO_PACKET.
2590 		 * this creates a new 'base' pointer, off_reg (variable) gets
2591 		 * added into the variable offset, and we copy the fixed offset
2592 		 * from ptr_reg.
2593 		 */
2594 		if (signed_add_overflows(smin_ptr, smin_val) ||
2595 		    signed_add_overflows(smax_ptr, smax_val)) {
2596 			dst_reg->smin_value = S64_MIN;
2597 			dst_reg->smax_value = S64_MAX;
2598 		} else {
2599 			dst_reg->smin_value = smin_ptr + smin_val;
2600 			dst_reg->smax_value = smax_ptr + smax_val;
2601 		}
2602 		if (umin_ptr + umin_val < umin_ptr ||
2603 		    umax_ptr + umax_val < umax_ptr) {
2604 			dst_reg->umin_value = 0;
2605 			dst_reg->umax_value = U64_MAX;
2606 		} else {
2607 			dst_reg->umin_value = umin_ptr + umin_val;
2608 			dst_reg->umax_value = umax_ptr + umax_val;
2609 		}
2610 		dst_reg->var_off = tnum_add(ptr_reg->var_off, off_reg->var_off);
2611 		dst_reg->off = ptr_reg->off;
2612 		if (reg_is_pkt_pointer(ptr_reg)) {
2613 			dst_reg->id = ++env->id_gen;
2614 			/* something was added to pkt_ptr, set range to zero */
2615 			dst_reg->range = 0;
2616 		}
2617 		break;
2618 	case BPF_SUB:
2619 		if (dst_reg == off_reg) {
2620 			/* scalar -= pointer.  Creates an unknown scalar */
2621 			verbose(env, "R%d tried to subtract pointer from scalar\n",
2622 				dst);
2623 			return -EACCES;
2624 		}
2625 		/* We don't allow subtraction from FP, because (according to
2626 		 * test_verifier.c test "invalid fp arithmetic", JITs might not
2627 		 * be able to deal with it.
2628 		 */
2629 		if (ptr_reg->type == PTR_TO_STACK) {
2630 			verbose(env, "R%d subtraction from stack pointer prohibited\n",
2631 				dst);
2632 			return -EACCES;
2633 		}
2634 		if (known && (ptr_reg->off - smin_val ==
2635 			      (s64)(s32)(ptr_reg->off - smin_val))) {
2636 			/* pointer -= K.  Subtract it from fixed offset */
2637 			dst_reg->smin_value = smin_ptr;
2638 			dst_reg->smax_value = smax_ptr;
2639 			dst_reg->umin_value = umin_ptr;
2640 			dst_reg->umax_value = umax_ptr;
2641 			dst_reg->var_off = ptr_reg->var_off;
2642 			dst_reg->id = ptr_reg->id;
2643 			dst_reg->off = ptr_reg->off - smin_val;
2644 			dst_reg->range = ptr_reg->range;
2645 			break;
2646 		}
2647 		/* A new variable offset is created.  If the subtrahend is known
2648 		 * nonnegative, then any reg->range we had before is still good.
2649 		 */
2650 		if (signed_sub_overflows(smin_ptr, smax_val) ||
2651 		    signed_sub_overflows(smax_ptr, smin_val)) {
2652 			/* Overflow possible, we know nothing */
2653 			dst_reg->smin_value = S64_MIN;
2654 			dst_reg->smax_value = S64_MAX;
2655 		} else {
2656 			dst_reg->smin_value = smin_ptr - smax_val;
2657 			dst_reg->smax_value = smax_ptr - smin_val;
2658 		}
2659 		if (umin_ptr < umax_val) {
2660 			/* Overflow possible, we know nothing */
2661 			dst_reg->umin_value = 0;
2662 			dst_reg->umax_value = U64_MAX;
2663 		} else {
2664 			/* Cannot overflow (as long as bounds are consistent) */
2665 			dst_reg->umin_value = umin_ptr - umax_val;
2666 			dst_reg->umax_value = umax_ptr - umin_val;
2667 		}
2668 		dst_reg->var_off = tnum_sub(ptr_reg->var_off, off_reg->var_off);
2669 		dst_reg->off = ptr_reg->off;
2670 		if (reg_is_pkt_pointer(ptr_reg)) {
2671 			dst_reg->id = ++env->id_gen;
2672 			/* something was added to pkt_ptr, set range to zero */
2673 			if (smin_val < 0)
2674 				dst_reg->range = 0;
2675 		}
2676 		break;
2677 	case BPF_AND:
2678 	case BPF_OR:
2679 	case BPF_XOR:
2680 		/* bitwise ops on pointers are troublesome, prohibit. */
2681 		verbose(env, "R%d bitwise operator %s on pointer prohibited\n",
2682 			dst, bpf_alu_string[opcode >> 4]);
2683 		return -EACCES;
2684 	default:
2685 		/* other operators (e.g. MUL,LSH) produce non-pointer results */
2686 		verbose(env, "R%d pointer arithmetic with %s operator prohibited\n",
2687 			dst, bpf_alu_string[opcode >> 4]);
2688 		return -EACCES;
2689 	}
2690 
2691 	if (!check_reg_sane_offset(env, dst_reg, ptr_reg->type))
2692 		return -EINVAL;
2693 
2694 	__update_reg_bounds(dst_reg);
2695 	__reg_deduce_bounds(dst_reg);
2696 	__reg_bound_offset(dst_reg);
2697 	return 0;
2698 }
2699 
2700 /* WARNING: This function does calculations on 64-bit values, but the actual
2701  * execution may occur on 32-bit values. Therefore, things like bitshifts
2702  * need extra checks in the 32-bit case.
2703  */
2704 static int adjust_scalar_min_max_vals(struct bpf_verifier_env *env,
2705 				      struct bpf_insn *insn,
2706 				      struct bpf_reg_state *dst_reg,
2707 				      struct bpf_reg_state src_reg)
2708 {
2709 	struct bpf_reg_state *regs = cur_regs(env);
2710 	u8 opcode = BPF_OP(insn->code);
2711 	bool src_known, dst_known;
2712 	s64 smin_val, smax_val;
2713 	u64 umin_val, umax_val;
2714 	u64 insn_bitness = (BPF_CLASS(insn->code) == BPF_ALU64) ? 64 : 32;
2715 
2716 	smin_val = src_reg.smin_value;
2717 	smax_val = src_reg.smax_value;
2718 	umin_val = src_reg.umin_value;
2719 	umax_val = src_reg.umax_value;
2720 	src_known = tnum_is_const(src_reg.var_off);
2721 	dst_known = tnum_is_const(dst_reg->var_off);
2722 
2723 	if ((src_known && (smin_val != smax_val || umin_val != umax_val)) ||
2724 	    smin_val > smax_val || umin_val > umax_val) {
2725 		/* Taint dst register if offset had invalid bounds derived from
2726 		 * e.g. dead branches.
2727 		 */
2728 		__mark_reg_unknown(dst_reg);
2729 		return 0;
2730 	}
2731 
2732 	if (!src_known &&
2733 	    opcode != BPF_ADD && opcode != BPF_SUB && opcode != BPF_AND) {
2734 		__mark_reg_unknown(dst_reg);
2735 		return 0;
2736 	}
2737 
2738 	switch (opcode) {
2739 	case BPF_ADD:
2740 		if (signed_add_overflows(dst_reg->smin_value, smin_val) ||
2741 		    signed_add_overflows(dst_reg->smax_value, smax_val)) {
2742 			dst_reg->smin_value = S64_MIN;
2743 			dst_reg->smax_value = S64_MAX;
2744 		} else {
2745 			dst_reg->smin_value += smin_val;
2746 			dst_reg->smax_value += smax_val;
2747 		}
2748 		if (dst_reg->umin_value + umin_val < umin_val ||
2749 		    dst_reg->umax_value + umax_val < umax_val) {
2750 			dst_reg->umin_value = 0;
2751 			dst_reg->umax_value = U64_MAX;
2752 		} else {
2753 			dst_reg->umin_value += umin_val;
2754 			dst_reg->umax_value += umax_val;
2755 		}
2756 		dst_reg->var_off = tnum_add(dst_reg->var_off, src_reg.var_off);
2757 		break;
2758 	case BPF_SUB:
2759 		if (signed_sub_overflows(dst_reg->smin_value, smax_val) ||
2760 		    signed_sub_overflows(dst_reg->smax_value, smin_val)) {
2761 			/* Overflow possible, we know nothing */
2762 			dst_reg->smin_value = S64_MIN;
2763 			dst_reg->smax_value = S64_MAX;
2764 		} else {
2765 			dst_reg->smin_value -= smax_val;
2766 			dst_reg->smax_value -= smin_val;
2767 		}
2768 		if (dst_reg->umin_value < umax_val) {
2769 			/* Overflow possible, we know nothing */
2770 			dst_reg->umin_value = 0;
2771 			dst_reg->umax_value = U64_MAX;
2772 		} else {
2773 			/* Cannot overflow (as long as bounds are consistent) */
2774 			dst_reg->umin_value -= umax_val;
2775 			dst_reg->umax_value -= umin_val;
2776 		}
2777 		dst_reg->var_off = tnum_sub(dst_reg->var_off, src_reg.var_off);
2778 		break;
2779 	case BPF_MUL:
2780 		dst_reg->var_off = tnum_mul(dst_reg->var_off, src_reg.var_off);
2781 		if (smin_val < 0 || dst_reg->smin_value < 0) {
2782 			/* Ain't nobody got time to multiply that sign */
2783 			__mark_reg_unbounded(dst_reg);
2784 			__update_reg_bounds(dst_reg);
2785 			break;
2786 		}
2787 		/* Both values are positive, so we can work with unsigned and
2788 		 * copy the result to signed (unless it exceeds S64_MAX).
2789 		 */
2790 		if (umax_val > U32_MAX || dst_reg->umax_value > U32_MAX) {
2791 			/* Potential overflow, we know nothing */
2792 			__mark_reg_unbounded(dst_reg);
2793 			/* (except what we can learn from the var_off) */
2794 			__update_reg_bounds(dst_reg);
2795 			break;
2796 		}
2797 		dst_reg->umin_value *= umin_val;
2798 		dst_reg->umax_value *= umax_val;
2799 		if (dst_reg->umax_value > S64_MAX) {
2800 			/* Overflow possible, we know nothing */
2801 			dst_reg->smin_value = S64_MIN;
2802 			dst_reg->smax_value = S64_MAX;
2803 		} else {
2804 			dst_reg->smin_value = dst_reg->umin_value;
2805 			dst_reg->smax_value = dst_reg->umax_value;
2806 		}
2807 		break;
2808 	case BPF_AND:
2809 		if (src_known && dst_known) {
2810 			__mark_reg_known(dst_reg, dst_reg->var_off.value &
2811 						  src_reg.var_off.value);
2812 			break;
2813 		}
2814 		/* We get our minimum from the var_off, since that's inherently
2815 		 * bitwise.  Our maximum is the minimum of the operands' maxima.
2816 		 */
2817 		dst_reg->var_off = tnum_and(dst_reg->var_off, src_reg.var_off);
2818 		dst_reg->umin_value = dst_reg->var_off.value;
2819 		dst_reg->umax_value = min(dst_reg->umax_value, umax_val);
2820 		if (dst_reg->smin_value < 0 || smin_val < 0) {
2821 			/* Lose signed bounds when ANDing negative numbers,
2822 			 * ain't nobody got time for that.
2823 			 */
2824 			dst_reg->smin_value = S64_MIN;
2825 			dst_reg->smax_value = S64_MAX;
2826 		} else {
2827 			/* ANDing two positives gives a positive, so safe to
2828 			 * cast result into s64.
2829 			 */
2830 			dst_reg->smin_value = dst_reg->umin_value;
2831 			dst_reg->smax_value = dst_reg->umax_value;
2832 		}
2833 		/* We may learn something more from the var_off */
2834 		__update_reg_bounds(dst_reg);
2835 		break;
2836 	case BPF_OR:
2837 		if (src_known && dst_known) {
2838 			__mark_reg_known(dst_reg, dst_reg->var_off.value |
2839 						  src_reg.var_off.value);
2840 			break;
2841 		}
2842 		/* We get our maximum from the var_off, and our minimum is the
2843 		 * maximum of the operands' minima
2844 		 */
2845 		dst_reg->var_off = tnum_or(dst_reg->var_off, src_reg.var_off);
2846 		dst_reg->umin_value = max(dst_reg->umin_value, umin_val);
2847 		dst_reg->umax_value = dst_reg->var_off.value |
2848 				      dst_reg->var_off.mask;
2849 		if (dst_reg->smin_value < 0 || smin_val < 0) {
2850 			/* Lose signed bounds when ORing negative numbers,
2851 			 * ain't nobody got time for that.
2852 			 */
2853 			dst_reg->smin_value = S64_MIN;
2854 			dst_reg->smax_value = S64_MAX;
2855 		} else {
2856 			/* ORing two positives gives a positive, so safe to
2857 			 * cast result into s64.
2858 			 */
2859 			dst_reg->smin_value = dst_reg->umin_value;
2860 			dst_reg->smax_value = dst_reg->umax_value;
2861 		}
2862 		/* We may learn something more from the var_off */
2863 		__update_reg_bounds(dst_reg);
2864 		break;
2865 	case BPF_LSH:
2866 		if (umax_val >= insn_bitness) {
2867 			/* Shifts greater than 31 or 63 are undefined.
2868 			 * This includes shifts by a negative number.
2869 			 */
2870 			mark_reg_unknown(env, regs, insn->dst_reg);
2871 			break;
2872 		}
2873 		/* We lose all sign bit information (except what we can pick
2874 		 * up from var_off)
2875 		 */
2876 		dst_reg->smin_value = S64_MIN;
2877 		dst_reg->smax_value = S64_MAX;
2878 		/* If we might shift our top bit out, then we know nothing */
2879 		if (dst_reg->umax_value > 1ULL << (63 - umax_val)) {
2880 			dst_reg->umin_value = 0;
2881 			dst_reg->umax_value = U64_MAX;
2882 		} else {
2883 			dst_reg->umin_value <<= umin_val;
2884 			dst_reg->umax_value <<= umax_val;
2885 		}
2886 		if (src_known)
2887 			dst_reg->var_off = tnum_lshift(dst_reg->var_off, umin_val);
2888 		else
2889 			dst_reg->var_off = tnum_lshift(tnum_unknown, umin_val);
2890 		/* We may learn something more from the var_off */
2891 		__update_reg_bounds(dst_reg);
2892 		break;
2893 	case BPF_RSH:
2894 		if (umax_val >= insn_bitness) {
2895 			/* Shifts greater than 31 or 63 are undefined.
2896 			 * This includes shifts by a negative number.
2897 			 */
2898 			mark_reg_unknown(env, regs, insn->dst_reg);
2899 			break;
2900 		}
2901 		/* BPF_RSH is an unsigned shift.  If the value in dst_reg might
2902 		 * be negative, then either:
2903 		 * 1) src_reg might be zero, so the sign bit of the result is
2904 		 *    unknown, so we lose our signed bounds
2905 		 * 2) it's known negative, thus the unsigned bounds capture the
2906 		 *    signed bounds
2907 		 * 3) the signed bounds cross zero, so they tell us nothing
2908 		 *    about the result
2909 		 * If the value in dst_reg is known nonnegative, then again the
2910 		 * unsigned bounts capture the signed bounds.
2911 		 * Thus, in all cases it suffices to blow away our signed bounds
2912 		 * and rely on inferring new ones from the unsigned bounds and
2913 		 * var_off of the result.
2914 		 */
2915 		dst_reg->smin_value = S64_MIN;
2916 		dst_reg->smax_value = S64_MAX;
2917 		if (src_known)
2918 			dst_reg->var_off = tnum_rshift(dst_reg->var_off,
2919 						       umin_val);
2920 		else
2921 			dst_reg->var_off = tnum_rshift(tnum_unknown, umin_val);
2922 		dst_reg->umin_value >>= umax_val;
2923 		dst_reg->umax_value >>= umin_val;
2924 		/* We may learn something more from the var_off */
2925 		__update_reg_bounds(dst_reg);
2926 		break;
2927 	default:
2928 		mark_reg_unknown(env, regs, insn->dst_reg);
2929 		break;
2930 	}
2931 
2932 	if (BPF_CLASS(insn->code) != BPF_ALU64) {
2933 		/* 32-bit ALU ops are (32,32)->32 */
2934 		coerce_reg_to_size(dst_reg, 4);
2935 		coerce_reg_to_size(&src_reg, 4);
2936 	}
2937 
2938 	__reg_deduce_bounds(dst_reg);
2939 	__reg_bound_offset(dst_reg);
2940 	return 0;
2941 }
2942 
2943 /* Handles ALU ops other than BPF_END, BPF_NEG and BPF_MOV: computes new min/max
2944  * and var_off.
2945  */
2946 static int adjust_reg_min_max_vals(struct bpf_verifier_env *env,
2947 				   struct bpf_insn *insn)
2948 {
2949 	struct bpf_verifier_state *vstate = env->cur_state;
2950 	struct bpf_func_state *state = vstate->frame[vstate->curframe];
2951 	struct bpf_reg_state *regs = state->regs, *dst_reg, *src_reg;
2952 	struct bpf_reg_state *ptr_reg = NULL, off_reg = {0};
2953 	u8 opcode = BPF_OP(insn->code);
2954 
2955 	dst_reg = &regs[insn->dst_reg];
2956 	src_reg = NULL;
2957 	if (dst_reg->type != SCALAR_VALUE)
2958 		ptr_reg = dst_reg;
2959 	if (BPF_SRC(insn->code) == BPF_X) {
2960 		src_reg = &regs[insn->src_reg];
2961 		if (src_reg->type != SCALAR_VALUE) {
2962 			if (dst_reg->type != SCALAR_VALUE) {
2963 				/* Combining two pointers by any ALU op yields
2964 				 * an arbitrary scalar. Disallow all math except
2965 				 * pointer subtraction
2966 				 */
2967 				if (opcode == BPF_SUB){
2968 					mark_reg_unknown(env, regs, insn->dst_reg);
2969 					return 0;
2970 				}
2971 				verbose(env, "R%d pointer %s pointer prohibited\n",
2972 					insn->dst_reg,
2973 					bpf_alu_string[opcode >> 4]);
2974 				return -EACCES;
2975 			} else {
2976 				/* scalar += pointer
2977 				 * This is legal, but we have to reverse our
2978 				 * src/dest handling in computing the range
2979 				 */
2980 				return adjust_ptr_min_max_vals(env, insn,
2981 							       src_reg, dst_reg);
2982 			}
2983 		} else if (ptr_reg) {
2984 			/* pointer += scalar */
2985 			return adjust_ptr_min_max_vals(env, insn,
2986 						       dst_reg, src_reg);
2987 		}
2988 	} else {
2989 		/* Pretend the src is a reg with a known value, since we only
2990 		 * need to be able to read from this state.
2991 		 */
2992 		off_reg.type = SCALAR_VALUE;
2993 		__mark_reg_known(&off_reg, insn->imm);
2994 		src_reg = &off_reg;
2995 		if (ptr_reg) /* pointer += K */
2996 			return adjust_ptr_min_max_vals(env, insn,
2997 						       ptr_reg, src_reg);
2998 	}
2999 
3000 	/* Got here implies adding two SCALAR_VALUEs */
3001 	if (WARN_ON_ONCE(ptr_reg)) {
3002 		print_verifier_state(env, state);
3003 		verbose(env, "verifier internal error: unexpected ptr_reg\n");
3004 		return -EINVAL;
3005 	}
3006 	if (WARN_ON(!src_reg)) {
3007 		print_verifier_state(env, state);
3008 		verbose(env, "verifier internal error: no src_reg\n");
3009 		return -EINVAL;
3010 	}
3011 	return adjust_scalar_min_max_vals(env, insn, dst_reg, *src_reg);
3012 }
3013 
3014 /* check validity of 32-bit and 64-bit arithmetic operations */
3015 static int check_alu_op(struct bpf_verifier_env *env, struct bpf_insn *insn)
3016 {
3017 	struct bpf_reg_state *regs = cur_regs(env);
3018 	u8 opcode = BPF_OP(insn->code);
3019 	int err;
3020 
3021 	if (opcode == BPF_END || opcode == BPF_NEG) {
3022 		if (opcode == BPF_NEG) {
3023 			if (BPF_SRC(insn->code) != 0 ||
3024 			    insn->src_reg != BPF_REG_0 ||
3025 			    insn->off != 0 || insn->imm != 0) {
3026 				verbose(env, "BPF_NEG uses reserved fields\n");
3027 				return -EINVAL;
3028 			}
3029 		} else {
3030 			if (insn->src_reg != BPF_REG_0 || insn->off != 0 ||
3031 			    (insn->imm != 16 && insn->imm != 32 && insn->imm != 64) ||
3032 			    BPF_CLASS(insn->code) == BPF_ALU64) {
3033 				verbose(env, "BPF_END uses reserved fields\n");
3034 				return -EINVAL;
3035 			}
3036 		}
3037 
3038 		/* check src operand */
3039 		err = check_reg_arg(env, insn->dst_reg, SRC_OP);
3040 		if (err)
3041 			return err;
3042 
3043 		if (is_pointer_value(env, insn->dst_reg)) {
3044 			verbose(env, "R%d pointer arithmetic prohibited\n",
3045 				insn->dst_reg);
3046 			return -EACCES;
3047 		}
3048 
3049 		/* check dest operand */
3050 		err = check_reg_arg(env, insn->dst_reg, DST_OP);
3051 		if (err)
3052 			return err;
3053 
3054 	} else if (opcode == BPF_MOV) {
3055 
3056 		if (BPF_SRC(insn->code) == BPF_X) {
3057 			if (insn->imm != 0 || insn->off != 0) {
3058 				verbose(env, "BPF_MOV uses reserved fields\n");
3059 				return -EINVAL;
3060 			}
3061 
3062 			/* check src operand */
3063 			err = check_reg_arg(env, insn->src_reg, SRC_OP);
3064 			if (err)
3065 				return err;
3066 		} else {
3067 			if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
3068 				verbose(env, "BPF_MOV uses reserved fields\n");
3069 				return -EINVAL;
3070 			}
3071 		}
3072 
3073 		/* check dest operand */
3074 		err = check_reg_arg(env, insn->dst_reg, DST_OP);
3075 		if (err)
3076 			return err;
3077 
3078 		if (BPF_SRC(insn->code) == BPF_X) {
3079 			if (BPF_CLASS(insn->code) == BPF_ALU64) {
3080 				/* case: R1 = R2
3081 				 * copy register state to dest reg
3082 				 */
3083 				regs[insn->dst_reg] = regs[insn->src_reg];
3084 				regs[insn->dst_reg].live |= REG_LIVE_WRITTEN;
3085 			} else {
3086 				/* R1 = (u32) R2 */
3087 				if (is_pointer_value(env, insn->src_reg)) {
3088 					verbose(env,
3089 						"R%d partial copy of pointer\n",
3090 						insn->src_reg);
3091 					return -EACCES;
3092 				}
3093 				mark_reg_unknown(env, regs, insn->dst_reg);
3094 				coerce_reg_to_size(&regs[insn->dst_reg], 4);
3095 			}
3096 		} else {
3097 			/* case: R = imm
3098 			 * remember the value we stored into this reg
3099 			 */
3100 			regs[insn->dst_reg].type = SCALAR_VALUE;
3101 			if (BPF_CLASS(insn->code) == BPF_ALU64) {
3102 				__mark_reg_known(regs + insn->dst_reg,
3103 						 insn->imm);
3104 			} else {
3105 				__mark_reg_known(regs + insn->dst_reg,
3106 						 (u32)insn->imm);
3107 			}
3108 		}
3109 
3110 	} else if (opcode > BPF_END) {
3111 		verbose(env, "invalid BPF_ALU opcode %x\n", opcode);
3112 		return -EINVAL;
3113 
3114 	} else {	/* all other ALU ops: and, sub, xor, add, ... */
3115 
3116 		if (BPF_SRC(insn->code) == BPF_X) {
3117 			if (insn->imm != 0 || insn->off != 0) {
3118 				verbose(env, "BPF_ALU uses reserved fields\n");
3119 				return -EINVAL;
3120 			}
3121 			/* check src1 operand */
3122 			err = check_reg_arg(env, insn->src_reg, SRC_OP);
3123 			if (err)
3124 				return err;
3125 		} else {
3126 			if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
3127 				verbose(env, "BPF_ALU uses reserved fields\n");
3128 				return -EINVAL;
3129 			}
3130 		}
3131 
3132 		/* check src2 operand */
3133 		err = check_reg_arg(env, insn->dst_reg, SRC_OP);
3134 		if (err)
3135 			return err;
3136 
3137 		if ((opcode == BPF_MOD || opcode == BPF_DIV) &&
3138 		    BPF_SRC(insn->code) == BPF_K && insn->imm == 0) {
3139 			verbose(env, "div by zero\n");
3140 			return -EINVAL;
3141 		}
3142 
3143 		if (opcode == BPF_ARSH && BPF_CLASS(insn->code) != BPF_ALU64) {
3144 			verbose(env, "BPF_ARSH not supported for 32 bit ALU\n");
3145 			return -EINVAL;
3146 		}
3147 
3148 		if ((opcode == BPF_LSH || opcode == BPF_RSH ||
3149 		     opcode == BPF_ARSH) && BPF_SRC(insn->code) == BPF_K) {
3150 			int size = BPF_CLASS(insn->code) == BPF_ALU64 ? 64 : 32;
3151 
3152 			if (insn->imm < 0 || insn->imm >= size) {
3153 				verbose(env, "invalid shift %d\n", insn->imm);
3154 				return -EINVAL;
3155 			}
3156 		}
3157 
3158 		/* check dest operand */
3159 		err = check_reg_arg(env, insn->dst_reg, DST_OP_NO_MARK);
3160 		if (err)
3161 			return err;
3162 
3163 		return adjust_reg_min_max_vals(env, insn);
3164 	}
3165 
3166 	return 0;
3167 }
3168 
3169 static void find_good_pkt_pointers(struct bpf_verifier_state *vstate,
3170 				   struct bpf_reg_state *dst_reg,
3171 				   enum bpf_reg_type type,
3172 				   bool range_right_open)
3173 {
3174 	struct bpf_func_state *state = vstate->frame[vstate->curframe];
3175 	struct bpf_reg_state *regs = state->regs, *reg;
3176 	u16 new_range;
3177 	int i, j;
3178 
3179 	if (dst_reg->off < 0 ||
3180 	    (dst_reg->off == 0 && range_right_open))
3181 		/* This doesn't give us any range */
3182 		return;
3183 
3184 	if (dst_reg->umax_value > MAX_PACKET_OFF ||
3185 	    dst_reg->umax_value + dst_reg->off > MAX_PACKET_OFF)
3186 		/* Risk of overflow.  For instance, ptr + (1<<63) may be less
3187 		 * than pkt_end, but that's because it's also less than pkt.
3188 		 */
3189 		return;
3190 
3191 	new_range = dst_reg->off;
3192 	if (range_right_open)
3193 		new_range--;
3194 
3195 	/* Examples for register markings:
3196 	 *
3197 	 * pkt_data in dst register:
3198 	 *
3199 	 *   r2 = r3;
3200 	 *   r2 += 8;
3201 	 *   if (r2 > pkt_end) goto <handle exception>
3202 	 *   <access okay>
3203 	 *
3204 	 *   r2 = r3;
3205 	 *   r2 += 8;
3206 	 *   if (r2 < pkt_end) goto <access okay>
3207 	 *   <handle exception>
3208 	 *
3209 	 *   Where:
3210 	 *     r2 == dst_reg, pkt_end == src_reg
3211 	 *     r2=pkt(id=n,off=8,r=0)
3212 	 *     r3=pkt(id=n,off=0,r=0)
3213 	 *
3214 	 * pkt_data in src register:
3215 	 *
3216 	 *   r2 = r3;
3217 	 *   r2 += 8;
3218 	 *   if (pkt_end >= r2) goto <access okay>
3219 	 *   <handle exception>
3220 	 *
3221 	 *   r2 = r3;
3222 	 *   r2 += 8;
3223 	 *   if (pkt_end <= r2) goto <handle exception>
3224 	 *   <access okay>
3225 	 *
3226 	 *   Where:
3227 	 *     pkt_end == dst_reg, r2 == src_reg
3228 	 *     r2=pkt(id=n,off=8,r=0)
3229 	 *     r3=pkt(id=n,off=0,r=0)
3230 	 *
3231 	 * Find register r3 and mark its range as r3=pkt(id=n,off=0,r=8)
3232 	 * or r3=pkt(id=n,off=0,r=8-1), so that range of bytes [r3, r3 + 8)
3233 	 * and [r3, r3 + 8-1) respectively is safe to access depending on
3234 	 * the check.
3235 	 */
3236 
3237 	/* If our ids match, then we must have the same max_value.  And we
3238 	 * don't care about the other reg's fixed offset, since if it's too big
3239 	 * the range won't allow anything.
3240 	 * dst_reg->off is known < MAX_PACKET_OFF, therefore it fits in a u16.
3241 	 */
3242 	for (i = 0; i < MAX_BPF_REG; i++)
3243 		if (regs[i].type == type && regs[i].id == dst_reg->id)
3244 			/* keep the maximum range already checked */
3245 			regs[i].range = max(regs[i].range, new_range);
3246 
3247 	for (j = 0; j <= vstate->curframe; j++) {
3248 		state = vstate->frame[j];
3249 		for (i = 0; i < state->allocated_stack / BPF_REG_SIZE; i++) {
3250 			if (state->stack[i].slot_type[0] != STACK_SPILL)
3251 				continue;
3252 			reg = &state->stack[i].spilled_ptr;
3253 			if (reg->type == type && reg->id == dst_reg->id)
3254 				reg->range = max(reg->range, new_range);
3255 		}
3256 	}
3257 }
3258 
3259 /* Adjusts the register min/max values in the case that the dst_reg is the
3260  * variable register that we are working on, and src_reg is a constant or we're
3261  * simply doing a BPF_K check.
3262  * In JEQ/JNE cases we also adjust the var_off values.
3263  */
3264 static void reg_set_min_max(struct bpf_reg_state *true_reg,
3265 			    struct bpf_reg_state *false_reg, u64 val,
3266 			    u8 opcode)
3267 {
3268 	/* If the dst_reg is a pointer, we can't learn anything about its
3269 	 * variable offset from the compare (unless src_reg were a pointer into
3270 	 * the same object, but we don't bother with that.
3271 	 * Since false_reg and true_reg have the same type by construction, we
3272 	 * only need to check one of them for pointerness.
3273 	 */
3274 	if (__is_pointer_value(false, false_reg))
3275 		return;
3276 
3277 	switch (opcode) {
3278 	case BPF_JEQ:
3279 		/* If this is false then we know nothing Jon Snow, but if it is
3280 		 * true then we know for sure.
3281 		 */
3282 		__mark_reg_known(true_reg, val);
3283 		break;
3284 	case BPF_JNE:
3285 		/* If this is true we know nothing Jon Snow, but if it is false
3286 		 * we know the value for sure;
3287 		 */
3288 		__mark_reg_known(false_reg, val);
3289 		break;
3290 	case BPF_JGT:
3291 		false_reg->umax_value = min(false_reg->umax_value, val);
3292 		true_reg->umin_value = max(true_reg->umin_value, val + 1);
3293 		break;
3294 	case BPF_JSGT:
3295 		false_reg->smax_value = min_t(s64, false_reg->smax_value, val);
3296 		true_reg->smin_value = max_t(s64, true_reg->smin_value, val + 1);
3297 		break;
3298 	case BPF_JLT:
3299 		false_reg->umin_value = max(false_reg->umin_value, val);
3300 		true_reg->umax_value = min(true_reg->umax_value, val - 1);
3301 		break;
3302 	case BPF_JSLT:
3303 		false_reg->smin_value = max_t(s64, false_reg->smin_value, val);
3304 		true_reg->smax_value = min_t(s64, true_reg->smax_value, val - 1);
3305 		break;
3306 	case BPF_JGE:
3307 		false_reg->umax_value = min(false_reg->umax_value, val - 1);
3308 		true_reg->umin_value = max(true_reg->umin_value, val);
3309 		break;
3310 	case BPF_JSGE:
3311 		false_reg->smax_value = min_t(s64, false_reg->smax_value, val - 1);
3312 		true_reg->smin_value = max_t(s64, true_reg->smin_value, val);
3313 		break;
3314 	case BPF_JLE:
3315 		false_reg->umin_value = max(false_reg->umin_value, val + 1);
3316 		true_reg->umax_value = min(true_reg->umax_value, val);
3317 		break;
3318 	case BPF_JSLE:
3319 		false_reg->smin_value = max_t(s64, false_reg->smin_value, val + 1);
3320 		true_reg->smax_value = min_t(s64, true_reg->smax_value, val);
3321 		break;
3322 	default:
3323 		break;
3324 	}
3325 
3326 	__reg_deduce_bounds(false_reg);
3327 	__reg_deduce_bounds(true_reg);
3328 	/* We might have learned some bits from the bounds. */
3329 	__reg_bound_offset(false_reg);
3330 	__reg_bound_offset(true_reg);
3331 	/* Intersecting with the old var_off might have improved our bounds
3332 	 * slightly.  e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc),
3333 	 * then new var_off is (0; 0x7f...fc) which improves our umax.
3334 	 */
3335 	__update_reg_bounds(false_reg);
3336 	__update_reg_bounds(true_reg);
3337 }
3338 
3339 /* Same as above, but for the case that dst_reg holds a constant and src_reg is
3340  * the variable reg.
3341  */
3342 static void reg_set_min_max_inv(struct bpf_reg_state *true_reg,
3343 				struct bpf_reg_state *false_reg, u64 val,
3344 				u8 opcode)
3345 {
3346 	if (__is_pointer_value(false, false_reg))
3347 		return;
3348 
3349 	switch (opcode) {
3350 	case BPF_JEQ:
3351 		/* If this is false then we know nothing Jon Snow, but if it is
3352 		 * true then we know for sure.
3353 		 */
3354 		__mark_reg_known(true_reg, val);
3355 		break;
3356 	case BPF_JNE:
3357 		/* If this is true we know nothing Jon Snow, but if it is false
3358 		 * we know the value for sure;
3359 		 */
3360 		__mark_reg_known(false_reg, val);
3361 		break;
3362 	case BPF_JGT:
3363 		true_reg->umax_value = min(true_reg->umax_value, val - 1);
3364 		false_reg->umin_value = max(false_reg->umin_value, val);
3365 		break;
3366 	case BPF_JSGT:
3367 		true_reg->smax_value = min_t(s64, true_reg->smax_value, val - 1);
3368 		false_reg->smin_value = max_t(s64, false_reg->smin_value, val);
3369 		break;
3370 	case BPF_JLT:
3371 		true_reg->umin_value = max(true_reg->umin_value, val + 1);
3372 		false_reg->umax_value = min(false_reg->umax_value, val);
3373 		break;
3374 	case BPF_JSLT:
3375 		true_reg->smin_value = max_t(s64, true_reg->smin_value, val + 1);
3376 		false_reg->smax_value = min_t(s64, false_reg->smax_value, val);
3377 		break;
3378 	case BPF_JGE:
3379 		true_reg->umax_value = min(true_reg->umax_value, val);
3380 		false_reg->umin_value = max(false_reg->umin_value, val + 1);
3381 		break;
3382 	case BPF_JSGE:
3383 		true_reg->smax_value = min_t(s64, true_reg->smax_value, val);
3384 		false_reg->smin_value = max_t(s64, false_reg->smin_value, val + 1);
3385 		break;
3386 	case BPF_JLE:
3387 		true_reg->umin_value = max(true_reg->umin_value, val);
3388 		false_reg->umax_value = min(false_reg->umax_value, val - 1);
3389 		break;
3390 	case BPF_JSLE:
3391 		true_reg->smin_value = max_t(s64, true_reg->smin_value, val);
3392 		false_reg->smax_value = min_t(s64, false_reg->smax_value, val - 1);
3393 		break;
3394 	default:
3395 		break;
3396 	}
3397 
3398 	__reg_deduce_bounds(false_reg);
3399 	__reg_deduce_bounds(true_reg);
3400 	/* We might have learned some bits from the bounds. */
3401 	__reg_bound_offset(false_reg);
3402 	__reg_bound_offset(true_reg);
3403 	/* Intersecting with the old var_off might have improved our bounds
3404 	 * slightly.  e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc),
3405 	 * then new var_off is (0; 0x7f...fc) which improves our umax.
3406 	 */
3407 	__update_reg_bounds(false_reg);
3408 	__update_reg_bounds(true_reg);
3409 }
3410 
3411 /* Regs are known to be equal, so intersect their min/max/var_off */
3412 static void __reg_combine_min_max(struct bpf_reg_state *src_reg,
3413 				  struct bpf_reg_state *dst_reg)
3414 {
3415 	src_reg->umin_value = dst_reg->umin_value = max(src_reg->umin_value,
3416 							dst_reg->umin_value);
3417 	src_reg->umax_value = dst_reg->umax_value = min(src_reg->umax_value,
3418 							dst_reg->umax_value);
3419 	src_reg->smin_value = dst_reg->smin_value = max(src_reg->smin_value,
3420 							dst_reg->smin_value);
3421 	src_reg->smax_value = dst_reg->smax_value = min(src_reg->smax_value,
3422 							dst_reg->smax_value);
3423 	src_reg->var_off = dst_reg->var_off = tnum_intersect(src_reg->var_off,
3424 							     dst_reg->var_off);
3425 	/* We might have learned new bounds from the var_off. */
3426 	__update_reg_bounds(src_reg);
3427 	__update_reg_bounds(dst_reg);
3428 	/* We might have learned something about the sign bit. */
3429 	__reg_deduce_bounds(src_reg);
3430 	__reg_deduce_bounds(dst_reg);
3431 	/* We might have learned some bits from the bounds. */
3432 	__reg_bound_offset(src_reg);
3433 	__reg_bound_offset(dst_reg);
3434 	/* Intersecting with the old var_off might have improved our bounds
3435 	 * slightly.  e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc),
3436 	 * then new var_off is (0; 0x7f...fc) which improves our umax.
3437 	 */
3438 	__update_reg_bounds(src_reg);
3439 	__update_reg_bounds(dst_reg);
3440 }
3441 
3442 static void reg_combine_min_max(struct bpf_reg_state *true_src,
3443 				struct bpf_reg_state *true_dst,
3444 				struct bpf_reg_state *false_src,
3445 				struct bpf_reg_state *false_dst,
3446 				u8 opcode)
3447 {
3448 	switch (opcode) {
3449 	case BPF_JEQ:
3450 		__reg_combine_min_max(true_src, true_dst);
3451 		break;
3452 	case BPF_JNE:
3453 		__reg_combine_min_max(false_src, false_dst);
3454 		break;
3455 	}
3456 }
3457 
3458 static void mark_map_reg(struct bpf_reg_state *regs, u32 regno, u32 id,
3459 			 bool is_null)
3460 {
3461 	struct bpf_reg_state *reg = &regs[regno];
3462 
3463 	if (reg->type == PTR_TO_MAP_VALUE_OR_NULL && reg->id == id) {
3464 		/* Old offset (both fixed and variable parts) should
3465 		 * have been known-zero, because we don't allow pointer
3466 		 * arithmetic on pointers that might be NULL.
3467 		 */
3468 		if (WARN_ON_ONCE(reg->smin_value || reg->smax_value ||
3469 				 !tnum_equals_const(reg->var_off, 0) ||
3470 				 reg->off)) {
3471 			__mark_reg_known_zero(reg);
3472 			reg->off = 0;
3473 		}
3474 		if (is_null) {
3475 			reg->type = SCALAR_VALUE;
3476 		} else if (reg->map_ptr->inner_map_meta) {
3477 			reg->type = CONST_PTR_TO_MAP;
3478 			reg->map_ptr = reg->map_ptr->inner_map_meta;
3479 		} else {
3480 			reg->type = PTR_TO_MAP_VALUE;
3481 		}
3482 		/* We don't need id from this point onwards anymore, thus we
3483 		 * should better reset it, so that state pruning has chances
3484 		 * to take effect.
3485 		 */
3486 		reg->id = 0;
3487 	}
3488 }
3489 
3490 /* The logic is similar to find_good_pkt_pointers(), both could eventually
3491  * be folded together at some point.
3492  */
3493 static void mark_map_regs(struct bpf_verifier_state *vstate, u32 regno,
3494 			  bool is_null)
3495 {
3496 	struct bpf_func_state *state = vstate->frame[vstate->curframe];
3497 	struct bpf_reg_state *regs = state->regs;
3498 	u32 id = regs[regno].id;
3499 	int i, j;
3500 
3501 	for (i = 0; i < MAX_BPF_REG; i++)
3502 		mark_map_reg(regs, i, id, is_null);
3503 
3504 	for (j = 0; j <= vstate->curframe; j++) {
3505 		state = vstate->frame[j];
3506 		for (i = 0; i < state->allocated_stack / BPF_REG_SIZE; i++) {
3507 			if (state->stack[i].slot_type[0] != STACK_SPILL)
3508 				continue;
3509 			mark_map_reg(&state->stack[i].spilled_ptr, 0, id, is_null);
3510 		}
3511 	}
3512 }
3513 
3514 static bool try_match_pkt_pointers(const struct bpf_insn *insn,
3515 				   struct bpf_reg_state *dst_reg,
3516 				   struct bpf_reg_state *src_reg,
3517 				   struct bpf_verifier_state *this_branch,
3518 				   struct bpf_verifier_state *other_branch)
3519 {
3520 	if (BPF_SRC(insn->code) != BPF_X)
3521 		return false;
3522 
3523 	switch (BPF_OP(insn->code)) {
3524 	case BPF_JGT:
3525 		if ((dst_reg->type == PTR_TO_PACKET &&
3526 		     src_reg->type == PTR_TO_PACKET_END) ||
3527 		    (dst_reg->type == PTR_TO_PACKET_META &&
3528 		     reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) {
3529 			/* pkt_data' > pkt_end, pkt_meta' > pkt_data */
3530 			find_good_pkt_pointers(this_branch, dst_reg,
3531 					       dst_reg->type, false);
3532 		} else if ((dst_reg->type == PTR_TO_PACKET_END &&
3533 			    src_reg->type == PTR_TO_PACKET) ||
3534 			   (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) &&
3535 			    src_reg->type == PTR_TO_PACKET_META)) {
3536 			/* pkt_end > pkt_data', pkt_data > pkt_meta' */
3537 			find_good_pkt_pointers(other_branch, src_reg,
3538 					       src_reg->type, true);
3539 		} else {
3540 			return false;
3541 		}
3542 		break;
3543 	case BPF_JLT:
3544 		if ((dst_reg->type == PTR_TO_PACKET &&
3545 		     src_reg->type == PTR_TO_PACKET_END) ||
3546 		    (dst_reg->type == PTR_TO_PACKET_META &&
3547 		     reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) {
3548 			/* pkt_data' < pkt_end, pkt_meta' < pkt_data */
3549 			find_good_pkt_pointers(other_branch, dst_reg,
3550 					       dst_reg->type, true);
3551 		} else if ((dst_reg->type == PTR_TO_PACKET_END &&
3552 			    src_reg->type == PTR_TO_PACKET) ||
3553 			   (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) &&
3554 			    src_reg->type == PTR_TO_PACKET_META)) {
3555 			/* pkt_end < pkt_data', pkt_data > pkt_meta' */
3556 			find_good_pkt_pointers(this_branch, src_reg,
3557 					       src_reg->type, false);
3558 		} else {
3559 			return false;
3560 		}
3561 		break;
3562 	case BPF_JGE:
3563 		if ((dst_reg->type == PTR_TO_PACKET &&
3564 		     src_reg->type == PTR_TO_PACKET_END) ||
3565 		    (dst_reg->type == PTR_TO_PACKET_META &&
3566 		     reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) {
3567 			/* pkt_data' >= pkt_end, pkt_meta' >= pkt_data */
3568 			find_good_pkt_pointers(this_branch, dst_reg,
3569 					       dst_reg->type, true);
3570 		} else if ((dst_reg->type == PTR_TO_PACKET_END &&
3571 			    src_reg->type == PTR_TO_PACKET) ||
3572 			   (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) &&
3573 			    src_reg->type == PTR_TO_PACKET_META)) {
3574 			/* pkt_end >= pkt_data', pkt_data >= pkt_meta' */
3575 			find_good_pkt_pointers(other_branch, src_reg,
3576 					       src_reg->type, false);
3577 		} else {
3578 			return false;
3579 		}
3580 		break;
3581 	case BPF_JLE:
3582 		if ((dst_reg->type == PTR_TO_PACKET &&
3583 		     src_reg->type == PTR_TO_PACKET_END) ||
3584 		    (dst_reg->type == PTR_TO_PACKET_META &&
3585 		     reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) {
3586 			/* pkt_data' <= pkt_end, pkt_meta' <= pkt_data */
3587 			find_good_pkt_pointers(other_branch, dst_reg,
3588 					       dst_reg->type, false);
3589 		} else if ((dst_reg->type == PTR_TO_PACKET_END &&
3590 			    src_reg->type == PTR_TO_PACKET) ||
3591 			   (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) &&
3592 			    src_reg->type == PTR_TO_PACKET_META)) {
3593 			/* pkt_end <= pkt_data', pkt_data <= pkt_meta' */
3594 			find_good_pkt_pointers(this_branch, src_reg,
3595 					       src_reg->type, true);
3596 		} else {
3597 			return false;
3598 		}
3599 		break;
3600 	default:
3601 		return false;
3602 	}
3603 
3604 	return true;
3605 }
3606 
3607 static int check_cond_jmp_op(struct bpf_verifier_env *env,
3608 			     struct bpf_insn *insn, int *insn_idx)
3609 {
3610 	struct bpf_verifier_state *this_branch = env->cur_state;
3611 	struct bpf_verifier_state *other_branch;
3612 	struct bpf_reg_state *regs = this_branch->frame[this_branch->curframe]->regs;
3613 	struct bpf_reg_state *dst_reg, *other_branch_regs;
3614 	u8 opcode = BPF_OP(insn->code);
3615 	int err;
3616 
3617 	if (opcode > BPF_JSLE) {
3618 		verbose(env, "invalid BPF_JMP opcode %x\n", opcode);
3619 		return -EINVAL;
3620 	}
3621 
3622 	if (BPF_SRC(insn->code) == BPF_X) {
3623 		if (insn->imm != 0) {
3624 			verbose(env, "BPF_JMP uses reserved fields\n");
3625 			return -EINVAL;
3626 		}
3627 
3628 		/* check src1 operand */
3629 		err = check_reg_arg(env, insn->src_reg, SRC_OP);
3630 		if (err)
3631 			return err;
3632 
3633 		if (is_pointer_value(env, insn->src_reg)) {
3634 			verbose(env, "R%d pointer comparison prohibited\n",
3635 				insn->src_reg);
3636 			return -EACCES;
3637 		}
3638 	} else {
3639 		if (insn->src_reg != BPF_REG_0) {
3640 			verbose(env, "BPF_JMP uses reserved fields\n");
3641 			return -EINVAL;
3642 		}
3643 	}
3644 
3645 	/* check src2 operand */
3646 	err = check_reg_arg(env, insn->dst_reg, SRC_OP);
3647 	if (err)
3648 		return err;
3649 
3650 	dst_reg = &regs[insn->dst_reg];
3651 
3652 	/* detect if R == 0 where R was initialized to zero earlier */
3653 	if (BPF_SRC(insn->code) == BPF_K &&
3654 	    (opcode == BPF_JEQ || opcode == BPF_JNE) &&
3655 	    dst_reg->type == SCALAR_VALUE &&
3656 	    tnum_is_const(dst_reg->var_off)) {
3657 		if ((opcode == BPF_JEQ && dst_reg->var_off.value == insn->imm) ||
3658 		    (opcode == BPF_JNE && dst_reg->var_off.value != insn->imm)) {
3659 			/* if (imm == imm) goto pc+off;
3660 			 * only follow the goto, ignore fall-through
3661 			 */
3662 			*insn_idx += insn->off;
3663 			return 0;
3664 		} else {
3665 			/* if (imm != imm) goto pc+off;
3666 			 * only follow fall-through branch, since
3667 			 * that's where the program will go
3668 			 */
3669 			return 0;
3670 		}
3671 	}
3672 
3673 	other_branch = push_stack(env, *insn_idx + insn->off + 1, *insn_idx);
3674 	if (!other_branch)
3675 		return -EFAULT;
3676 	other_branch_regs = other_branch->frame[other_branch->curframe]->regs;
3677 
3678 	/* detect if we are comparing against a constant value so we can adjust
3679 	 * our min/max values for our dst register.
3680 	 * this is only legit if both are scalars (or pointers to the same
3681 	 * object, I suppose, but we don't support that right now), because
3682 	 * otherwise the different base pointers mean the offsets aren't
3683 	 * comparable.
3684 	 */
3685 	if (BPF_SRC(insn->code) == BPF_X) {
3686 		if (dst_reg->type == SCALAR_VALUE &&
3687 		    regs[insn->src_reg].type == SCALAR_VALUE) {
3688 			if (tnum_is_const(regs[insn->src_reg].var_off))
3689 				reg_set_min_max(&other_branch_regs[insn->dst_reg],
3690 						dst_reg, regs[insn->src_reg].var_off.value,
3691 						opcode);
3692 			else if (tnum_is_const(dst_reg->var_off))
3693 				reg_set_min_max_inv(&other_branch_regs[insn->src_reg],
3694 						    &regs[insn->src_reg],
3695 						    dst_reg->var_off.value, opcode);
3696 			else if (opcode == BPF_JEQ || opcode == BPF_JNE)
3697 				/* Comparing for equality, we can combine knowledge */
3698 				reg_combine_min_max(&other_branch_regs[insn->src_reg],
3699 						    &other_branch_regs[insn->dst_reg],
3700 						    &regs[insn->src_reg],
3701 						    &regs[insn->dst_reg], opcode);
3702 		}
3703 	} else if (dst_reg->type == SCALAR_VALUE) {
3704 		reg_set_min_max(&other_branch_regs[insn->dst_reg],
3705 					dst_reg, insn->imm, opcode);
3706 	}
3707 
3708 	/* detect if R == 0 where R is returned from bpf_map_lookup_elem() */
3709 	if (BPF_SRC(insn->code) == BPF_K &&
3710 	    insn->imm == 0 && (opcode == BPF_JEQ || opcode == BPF_JNE) &&
3711 	    dst_reg->type == PTR_TO_MAP_VALUE_OR_NULL) {
3712 		/* Mark all identical map registers in each branch as either
3713 		 * safe or unknown depending R == 0 or R != 0 conditional.
3714 		 */
3715 		mark_map_regs(this_branch, insn->dst_reg, opcode == BPF_JNE);
3716 		mark_map_regs(other_branch, insn->dst_reg, opcode == BPF_JEQ);
3717 	} else if (!try_match_pkt_pointers(insn, dst_reg, &regs[insn->src_reg],
3718 					   this_branch, other_branch) &&
3719 		   is_pointer_value(env, insn->dst_reg)) {
3720 		verbose(env, "R%d pointer comparison prohibited\n",
3721 			insn->dst_reg);
3722 		return -EACCES;
3723 	}
3724 	if (env->log.level)
3725 		print_verifier_state(env, this_branch->frame[this_branch->curframe]);
3726 	return 0;
3727 }
3728 
3729 /* return the map pointer stored inside BPF_LD_IMM64 instruction */
3730 static struct bpf_map *ld_imm64_to_map_ptr(struct bpf_insn *insn)
3731 {
3732 	u64 imm64 = ((u64) (u32) insn[0].imm) | ((u64) (u32) insn[1].imm) << 32;
3733 
3734 	return (struct bpf_map *) (unsigned long) imm64;
3735 }
3736 
3737 /* verify BPF_LD_IMM64 instruction */
3738 static int check_ld_imm(struct bpf_verifier_env *env, struct bpf_insn *insn)
3739 {
3740 	struct bpf_reg_state *regs = cur_regs(env);
3741 	int err;
3742 
3743 	if (BPF_SIZE(insn->code) != BPF_DW) {
3744 		verbose(env, "invalid BPF_LD_IMM insn\n");
3745 		return -EINVAL;
3746 	}
3747 	if (insn->off != 0) {
3748 		verbose(env, "BPF_LD_IMM64 uses reserved fields\n");
3749 		return -EINVAL;
3750 	}
3751 
3752 	err = check_reg_arg(env, insn->dst_reg, DST_OP);
3753 	if (err)
3754 		return err;
3755 
3756 	if (insn->src_reg == 0) {
3757 		u64 imm = ((u64)(insn + 1)->imm << 32) | (u32)insn->imm;
3758 
3759 		regs[insn->dst_reg].type = SCALAR_VALUE;
3760 		__mark_reg_known(&regs[insn->dst_reg], imm);
3761 		return 0;
3762 	}
3763 
3764 	/* replace_map_fd_with_map_ptr() should have caught bad ld_imm64 */
3765 	BUG_ON(insn->src_reg != BPF_PSEUDO_MAP_FD);
3766 
3767 	regs[insn->dst_reg].type = CONST_PTR_TO_MAP;
3768 	regs[insn->dst_reg].map_ptr = ld_imm64_to_map_ptr(insn);
3769 	return 0;
3770 }
3771 
3772 static bool may_access_skb(enum bpf_prog_type type)
3773 {
3774 	switch (type) {
3775 	case BPF_PROG_TYPE_SOCKET_FILTER:
3776 	case BPF_PROG_TYPE_SCHED_CLS:
3777 	case BPF_PROG_TYPE_SCHED_ACT:
3778 		return true;
3779 	default:
3780 		return false;
3781 	}
3782 }
3783 
3784 /* verify safety of LD_ABS|LD_IND instructions:
3785  * - they can only appear in the programs where ctx == skb
3786  * - since they are wrappers of function calls, they scratch R1-R5 registers,
3787  *   preserve R6-R9, and store return value into R0
3788  *
3789  * Implicit input:
3790  *   ctx == skb == R6 == CTX
3791  *
3792  * Explicit input:
3793  *   SRC == any register
3794  *   IMM == 32-bit immediate
3795  *
3796  * Output:
3797  *   R0 - 8/16/32-bit skb data converted to cpu endianness
3798  */
3799 static int check_ld_abs(struct bpf_verifier_env *env, struct bpf_insn *insn)
3800 {
3801 	struct bpf_reg_state *regs = cur_regs(env);
3802 	u8 mode = BPF_MODE(insn->code);
3803 	int i, err;
3804 
3805 	if (!may_access_skb(env->prog->type)) {
3806 		verbose(env, "BPF_LD_[ABS|IND] instructions not allowed for this program type\n");
3807 		return -EINVAL;
3808 	}
3809 
3810 	if (env->subprog_cnt) {
3811 		/* when program has LD_ABS insn JITs and interpreter assume
3812 		 * that r1 == ctx == skb which is not the case for callees
3813 		 * that can have arbitrary arguments. It's problematic
3814 		 * for main prog as well since JITs would need to analyze
3815 		 * all functions in order to make proper register save/restore
3816 		 * decisions in the main prog. Hence disallow LD_ABS with calls
3817 		 */
3818 		verbose(env, "BPF_LD_[ABS|IND] instructions cannot be mixed with bpf-to-bpf calls\n");
3819 		return -EINVAL;
3820 	}
3821 
3822 	if (insn->dst_reg != BPF_REG_0 || insn->off != 0 ||
3823 	    BPF_SIZE(insn->code) == BPF_DW ||
3824 	    (mode == BPF_ABS && insn->src_reg != BPF_REG_0)) {
3825 		verbose(env, "BPF_LD_[ABS|IND] uses reserved fields\n");
3826 		return -EINVAL;
3827 	}
3828 
3829 	/* check whether implicit source operand (register R6) is readable */
3830 	err = check_reg_arg(env, BPF_REG_6, SRC_OP);
3831 	if (err)
3832 		return err;
3833 
3834 	if (regs[BPF_REG_6].type != PTR_TO_CTX) {
3835 		verbose(env,
3836 			"at the time of BPF_LD_ABS|IND R6 != pointer to skb\n");
3837 		return -EINVAL;
3838 	}
3839 
3840 	if (mode == BPF_IND) {
3841 		/* check explicit source operand */
3842 		err = check_reg_arg(env, insn->src_reg, SRC_OP);
3843 		if (err)
3844 			return err;
3845 	}
3846 
3847 	/* reset caller saved regs to unreadable */
3848 	for (i = 0; i < CALLER_SAVED_REGS; i++) {
3849 		mark_reg_not_init(env, regs, caller_saved[i]);
3850 		check_reg_arg(env, caller_saved[i], DST_OP_NO_MARK);
3851 	}
3852 
3853 	/* mark destination R0 register as readable, since it contains
3854 	 * the value fetched from the packet.
3855 	 * Already marked as written above.
3856 	 */
3857 	mark_reg_unknown(env, regs, BPF_REG_0);
3858 	return 0;
3859 }
3860 
3861 static int check_return_code(struct bpf_verifier_env *env)
3862 {
3863 	struct bpf_reg_state *reg;
3864 	struct tnum range = tnum_range(0, 1);
3865 
3866 	switch (env->prog->type) {
3867 	case BPF_PROG_TYPE_CGROUP_SKB:
3868 	case BPF_PROG_TYPE_CGROUP_SOCK:
3869 	case BPF_PROG_TYPE_SOCK_OPS:
3870 	case BPF_PROG_TYPE_CGROUP_DEVICE:
3871 		break;
3872 	default:
3873 		return 0;
3874 	}
3875 
3876 	reg = cur_regs(env) + BPF_REG_0;
3877 	if (reg->type != SCALAR_VALUE) {
3878 		verbose(env, "At program exit the register R0 is not a known value (%s)\n",
3879 			reg_type_str[reg->type]);
3880 		return -EINVAL;
3881 	}
3882 
3883 	if (!tnum_in(range, reg->var_off)) {
3884 		verbose(env, "At program exit the register R0 ");
3885 		if (!tnum_is_unknown(reg->var_off)) {
3886 			char tn_buf[48];
3887 
3888 			tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off);
3889 			verbose(env, "has value %s", tn_buf);
3890 		} else {
3891 			verbose(env, "has unknown scalar value");
3892 		}
3893 		verbose(env, " should have been 0 or 1\n");
3894 		return -EINVAL;
3895 	}
3896 	return 0;
3897 }
3898 
3899 /* non-recursive DFS pseudo code
3900  * 1  procedure DFS-iterative(G,v):
3901  * 2      label v as discovered
3902  * 3      let S be a stack
3903  * 4      S.push(v)
3904  * 5      while S is not empty
3905  * 6            t <- S.pop()
3906  * 7            if t is what we're looking for:
3907  * 8                return t
3908  * 9            for all edges e in G.adjacentEdges(t) do
3909  * 10               if edge e is already labelled
3910  * 11                   continue with the next edge
3911  * 12               w <- G.adjacentVertex(t,e)
3912  * 13               if vertex w is not discovered and not explored
3913  * 14                   label e as tree-edge
3914  * 15                   label w as discovered
3915  * 16                   S.push(w)
3916  * 17                   continue at 5
3917  * 18               else if vertex w is discovered
3918  * 19                   label e as back-edge
3919  * 20               else
3920  * 21                   // vertex w is explored
3921  * 22                   label e as forward- or cross-edge
3922  * 23           label t as explored
3923  * 24           S.pop()
3924  *
3925  * convention:
3926  * 0x10 - discovered
3927  * 0x11 - discovered and fall-through edge labelled
3928  * 0x12 - discovered and fall-through and branch edges labelled
3929  * 0x20 - explored
3930  */
3931 
3932 enum {
3933 	DISCOVERED = 0x10,
3934 	EXPLORED = 0x20,
3935 	FALLTHROUGH = 1,
3936 	BRANCH = 2,
3937 };
3938 
3939 #define STATE_LIST_MARK ((struct bpf_verifier_state_list *) -1L)
3940 
3941 static int *insn_stack;	/* stack of insns to process */
3942 static int cur_stack;	/* current stack index */
3943 static int *insn_state;
3944 
3945 /* t, w, e - match pseudo-code above:
3946  * t - index of current instruction
3947  * w - next instruction
3948  * e - edge
3949  */
3950 static int push_insn(int t, int w, int e, struct bpf_verifier_env *env)
3951 {
3952 	if (e == FALLTHROUGH && insn_state[t] >= (DISCOVERED | FALLTHROUGH))
3953 		return 0;
3954 
3955 	if (e == BRANCH && insn_state[t] >= (DISCOVERED | BRANCH))
3956 		return 0;
3957 
3958 	if (w < 0 || w >= env->prog->len) {
3959 		verbose(env, "jump out of range from insn %d to %d\n", t, w);
3960 		return -EINVAL;
3961 	}
3962 
3963 	if (e == BRANCH)
3964 		/* mark branch target for state pruning */
3965 		env->explored_states[w] = STATE_LIST_MARK;
3966 
3967 	if (insn_state[w] == 0) {
3968 		/* tree-edge */
3969 		insn_state[t] = DISCOVERED | e;
3970 		insn_state[w] = DISCOVERED;
3971 		if (cur_stack >= env->prog->len)
3972 			return -E2BIG;
3973 		insn_stack[cur_stack++] = w;
3974 		return 1;
3975 	} else if ((insn_state[w] & 0xF0) == DISCOVERED) {
3976 		verbose(env, "back-edge from insn %d to %d\n", t, w);
3977 		return -EINVAL;
3978 	} else if (insn_state[w] == EXPLORED) {
3979 		/* forward- or cross-edge */
3980 		insn_state[t] = DISCOVERED | e;
3981 	} else {
3982 		verbose(env, "insn state internal bug\n");
3983 		return -EFAULT;
3984 	}
3985 	return 0;
3986 }
3987 
3988 /* non-recursive depth-first-search to detect loops in BPF program
3989  * loop == back-edge in directed graph
3990  */
3991 static int check_cfg(struct bpf_verifier_env *env)
3992 {
3993 	struct bpf_insn *insns = env->prog->insnsi;
3994 	int insn_cnt = env->prog->len;
3995 	int ret = 0;
3996 	int i, t;
3997 
3998 	ret = check_subprogs(env);
3999 	if (ret < 0)
4000 		return ret;
4001 
4002 	insn_state = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
4003 	if (!insn_state)
4004 		return -ENOMEM;
4005 
4006 	insn_stack = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
4007 	if (!insn_stack) {
4008 		kfree(insn_state);
4009 		return -ENOMEM;
4010 	}
4011 
4012 	insn_state[0] = DISCOVERED; /* mark 1st insn as discovered */
4013 	insn_stack[0] = 0; /* 0 is the first instruction */
4014 	cur_stack = 1;
4015 
4016 peek_stack:
4017 	if (cur_stack == 0)
4018 		goto check_state;
4019 	t = insn_stack[cur_stack - 1];
4020 
4021 	if (BPF_CLASS(insns[t].code) == BPF_JMP) {
4022 		u8 opcode = BPF_OP(insns[t].code);
4023 
4024 		if (opcode == BPF_EXIT) {
4025 			goto mark_explored;
4026 		} else if (opcode == BPF_CALL) {
4027 			ret = push_insn(t, t + 1, FALLTHROUGH, env);
4028 			if (ret == 1)
4029 				goto peek_stack;
4030 			else if (ret < 0)
4031 				goto err_free;
4032 			if (t + 1 < insn_cnt)
4033 				env->explored_states[t + 1] = STATE_LIST_MARK;
4034 			if (insns[t].src_reg == BPF_PSEUDO_CALL) {
4035 				env->explored_states[t] = STATE_LIST_MARK;
4036 				ret = push_insn(t, t + insns[t].imm + 1, BRANCH, env);
4037 				if (ret == 1)
4038 					goto peek_stack;
4039 				else if (ret < 0)
4040 					goto err_free;
4041 			}
4042 		} else if (opcode == BPF_JA) {
4043 			if (BPF_SRC(insns[t].code) != BPF_K) {
4044 				ret = -EINVAL;
4045 				goto err_free;
4046 			}
4047 			/* unconditional jump with single edge */
4048 			ret = push_insn(t, t + insns[t].off + 1,
4049 					FALLTHROUGH, env);
4050 			if (ret == 1)
4051 				goto peek_stack;
4052 			else if (ret < 0)
4053 				goto err_free;
4054 			/* tell verifier to check for equivalent states
4055 			 * after every call and jump
4056 			 */
4057 			if (t + 1 < insn_cnt)
4058 				env->explored_states[t + 1] = STATE_LIST_MARK;
4059 		} else {
4060 			/* conditional jump with two edges */
4061 			env->explored_states[t] = STATE_LIST_MARK;
4062 			ret = push_insn(t, t + 1, FALLTHROUGH, env);
4063 			if (ret == 1)
4064 				goto peek_stack;
4065 			else if (ret < 0)
4066 				goto err_free;
4067 
4068 			ret = push_insn(t, t + insns[t].off + 1, BRANCH, env);
4069 			if (ret == 1)
4070 				goto peek_stack;
4071 			else if (ret < 0)
4072 				goto err_free;
4073 		}
4074 	} else {
4075 		/* all other non-branch instructions with single
4076 		 * fall-through edge
4077 		 */
4078 		ret = push_insn(t, t + 1, FALLTHROUGH, env);
4079 		if (ret == 1)
4080 			goto peek_stack;
4081 		else if (ret < 0)
4082 			goto err_free;
4083 	}
4084 
4085 mark_explored:
4086 	insn_state[t] = EXPLORED;
4087 	if (cur_stack-- <= 0) {
4088 		verbose(env, "pop stack internal bug\n");
4089 		ret = -EFAULT;
4090 		goto err_free;
4091 	}
4092 	goto peek_stack;
4093 
4094 check_state:
4095 	for (i = 0; i < insn_cnt; i++) {
4096 		if (insn_state[i] != EXPLORED) {
4097 			verbose(env, "unreachable insn %d\n", i);
4098 			ret = -EINVAL;
4099 			goto err_free;
4100 		}
4101 	}
4102 	ret = 0; /* cfg looks good */
4103 
4104 err_free:
4105 	kfree(insn_state);
4106 	kfree(insn_stack);
4107 	return ret;
4108 }
4109 
4110 /* check %cur's range satisfies %old's */
4111 static bool range_within(struct bpf_reg_state *old,
4112 			 struct bpf_reg_state *cur)
4113 {
4114 	return old->umin_value <= cur->umin_value &&
4115 	       old->umax_value >= cur->umax_value &&
4116 	       old->smin_value <= cur->smin_value &&
4117 	       old->smax_value >= cur->smax_value;
4118 }
4119 
4120 /* Maximum number of register states that can exist at once */
4121 #define ID_MAP_SIZE	(MAX_BPF_REG + MAX_BPF_STACK / BPF_REG_SIZE)
4122 struct idpair {
4123 	u32 old;
4124 	u32 cur;
4125 };
4126 
4127 /* If in the old state two registers had the same id, then they need to have
4128  * the same id in the new state as well.  But that id could be different from
4129  * the old state, so we need to track the mapping from old to new ids.
4130  * Once we have seen that, say, a reg with old id 5 had new id 9, any subsequent
4131  * regs with old id 5 must also have new id 9 for the new state to be safe.  But
4132  * regs with a different old id could still have new id 9, we don't care about
4133  * that.
4134  * So we look through our idmap to see if this old id has been seen before.  If
4135  * so, we require the new id to match; otherwise, we add the id pair to the map.
4136  */
4137 static bool check_ids(u32 old_id, u32 cur_id, struct idpair *idmap)
4138 {
4139 	unsigned int i;
4140 
4141 	for (i = 0; i < ID_MAP_SIZE; i++) {
4142 		if (!idmap[i].old) {
4143 			/* Reached an empty slot; haven't seen this id before */
4144 			idmap[i].old = old_id;
4145 			idmap[i].cur = cur_id;
4146 			return true;
4147 		}
4148 		if (idmap[i].old == old_id)
4149 			return idmap[i].cur == cur_id;
4150 	}
4151 	/* We ran out of idmap slots, which should be impossible */
4152 	WARN_ON_ONCE(1);
4153 	return false;
4154 }
4155 
4156 /* Returns true if (rold safe implies rcur safe) */
4157 static bool regsafe(struct bpf_reg_state *rold, struct bpf_reg_state *rcur,
4158 		    struct idpair *idmap)
4159 {
4160 	bool equal;
4161 
4162 	if (!(rold->live & REG_LIVE_READ))
4163 		/* explored state didn't use this */
4164 		return true;
4165 
4166 	equal = memcmp(rold, rcur, offsetof(struct bpf_reg_state, frameno)) == 0;
4167 
4168 	if (rold->type == PTR_TO_STACK)
4169 		/* two stack pointers are equal only if they're pointing to
4170 		 * the same stack frame, since fp-8 in foo != fp-8 in bar
4171 		 */
4172 		return equal && rold->frameno == rcur->frameno;
4173 
4174 	if (equal)
4175 		return true;
4176 
4177 	if (rold->type == NOT_INIT)
4178 		/* explored state can't have used this */
4179 		return true;
4180 	if (rcur->type == NOT_INIT)
4181 		return false;
4182 	switch (rold->type) {
4183 	case SCALAR_VALUE:
4184 		if (rcur->type == SCALAR_VALUE) {
4185 			/* new val must satisfy old val knowledge */
4186 			return range_within(rold, rcur) &&
4187 			       tnum_in(rold->var_off, rcur->var_off);
4188 		} else {
4189 			/* We're trying to use a pointer in place of a scalar.
4190 			 * Even if the scalar was unbounded, this could lead to
4191 			 * pointer leaks because scalars are allowed to leak
4192 			 * while pointers are not. We could make this safe in
4193 			 * special cases if root is calling us, but it's
4194 			 * probably not worth the hassle.
4195 			 */
4196 			return false;
4197 		}
4198 	case PTR_TO_MAP_VALUE:
4199 		/* If the new min/max/var_off satisfy the old ones and
4200 		 * everything else matches, we are OK.
4201 		 * We don't care about the 'id' value, because nothing
4202 		 * uses it for PTR_TO_MAP_VALUE (only for ..._OR_NULL)
4203 		 */
4204 		return memcmp(rold, rcur, offsetof(struct bpf_reg_state, id)) == 0 &&
4205 		       range_within(rold, rcur) &&
4206 		       tnum_in(rold->var_off, rcur->var_off);
4207 	case PTR_TO_MAP_VALUE_OR_NULL:
4208 		/* a PTR_TO_MAP_VALUE could be safe to use as a
4209 		 * PTR_TO_MAP_VALUE_OR_NULL into the same map.
4210 		 * However, if the old PTR_TO_MAP_VALUE_OR_NULL then got NULL-
4211 		 * checked, doing so could have affected others with the same
4212 		 * id, and we can't check for that because we lost the id when
4213 		 * we converted to a PTR_TO_MAP_VALUE.
4214 		 */
4215 		if (rcur->type != PTR_TO_MAP_VALUE_OR_NULL)
4216 			return false;
4217 		if (memcmp(rold, rcur, offsetof(struct bpf_reg_state, id)))
4218 			return false;
4219 		/* Check our ids match any regs they're supposed to */
4220 		return check_ids(rold->id, rcur->id, idmap);
4221 	case PTR_TO_PACKET_META:
4222 	case PTR_TO_PACKET:
4223 		if (rcur->type != rold->type)
4224 			return false;
4225 		/* We must have at least as much range as the old ptr
4226 		 * did, so that any accesses which were safe before are
4227 		 * still safe.  This is true even if old range < old off,
4228 		 * since someone could have accessed through (ptr - k), or
4229 		 * even done ptr -= k in a register, to get a safe access.
4230 		 */
4231 		if (rold->range > rcur->range)
4232 			return false;
4233 		/* If the offsets don't match, we can't trust our alignment;
4234 		 * nor can we be sure that we won't fall out of range.
4235 		 */
4236 		if (rold->off != rcur->off)
4237 			return false;
4238 		/* id relations must be preserved */
4239 		if (rold->id && !check_ids(rold->id, rcur->id, idmap))
4240 			return false;
4241 		/* new val must satisfy old val knowledge */
4242 		return range_within(rold, rcur) &&
4243 		       tnum_in(rold->var_off, rcur->var_off);
4244 	case PTR_TO_CTX:
4245 	case CONST_PTR_TO_MAP:
4246 	case PTR_TO_PACKET_END:
4247 		/* Only valid matches are exact, which memcmp() above
4248 		 * would have accepted
4249 		 */
4250 	default:
4251 		/* Don't know what's going on, just say it's not safe */
4252 		return false;
4253 	}
4254 
4255 	/* Shouldn't get here; if we do, say it's not safe */
4256 	WARN_ON_ONCE(1);
4257 	return false;
4258 }
4259 
4260 static bool stacksafe(struct bpf_func_state *old,
4261 		      struct bpf_func_state *cur,
4262 		      struct idpair *idmap)
4263 {
4264 	int i, spi;
4265 
4266 	/* if explored stack has more populated slots than current stack
4267 	 * such stacks are not equivalent
4268 	 */
4269 	if (old->allocated_stack > cur->allocated_stack)
4270 		return false;
4271 
4272 	/* walk slots of the explored stack and ignore any additional
4273 	 * slots in the current stack, since explored(safe) state
4274 	 * didn't use them
4275 	 */
4276 	for (i = 0; i < old->allocated_stack; i++) {
4277 		spi = i / BPF_REG_SIZE;
4278 
4279 		if (!(old->stack[spi].spilled_ptr.live & REG_LIVE_READ))
4280 			/* explored state didn't use this */
4281 			continue;
4282 
4283 		if (old->stack[spi].slot_type[i % BPF_REG_SIZE] == STACK_INVALID)
4284 			continue;
4285 		/* if old state was safe with misc data in the stack
4286 		 * it will be safe with zero-initialized stack.
4287 		 * The opposite is not true
4288 		 */
4289 		if (old->stack[spi].slot_type[i % BPF_REG_SIZE] == STACK_MISC &&
4290 		    cur->stack[spi].slot_type[i % BPF_REG_SIZE] == STACK_ZERO)
4291 			continue;
4292 		if (old->stack[spi].slot_type[i % BPF_REG_SIZE] !=
4293 		    cur->stack[spi].slot_type[i % BPF_REG_SIZE])
4294 			/* Ex: old explored (safe) state has STACK_SPILL in
4295 			 * this stack slot, but current has has STACK_MISC ->
4296 			 * this verifier states are not equivalent,
4297 			 * return false to continue verification of this path
4298 			 */
4299 			return false;
4300 		if (i % BPF_REG_SIZE)
4301 			continue;
4302 		if (old->stack[spi].slot_type[0] != STACK_SPILL)
4303 			continue;
4304 		if (!regsafe(&old->stack[spi].spilled_ptr,
4305 			     &cur->stack[spi].spilled_ptr,
4306 			     idmap))
4307 			/* when explored and current stack slot are both storing
4308 			 * spilled registers, check that stored pointers types
4309 			 * are the same as well.
4310 			 * Ex: explored safe path could have stored
4311 			 * (bpf_reg_state) {.type = PTR_TO_STACK, .off = -8}
4312 			 * but current path has stored:
4313 			 * (bpf_reg_state) {.type = PTR_TO_STACK, .off = -16}
4314 			 * such verifier states are not equivalent.
4315 			 * return false to continue verification of this path
4316 			 */
4317 			return false;
4318 	}
4319 	return true;
4320 }
4321 
4322 /* compare two verifier states
4323  *
4324  * all states stored in state_list are known to be valid, since
4325  * verifier reached 'bpf_exit' instruction through them
4326  *
4327  * this function is called when verifier exploring different branches of
4328  * execution popped from the state stack. If it sees an old state that has
4329  * more strict register state and more strict stack state then this execution
4330  * branch doesn't need to be explored further, since verifier already
4331  * concluded that more strict state leads to valid finish.
4332  *
4333  * Therefore two states are equivalent if register state is more conservative
4334  * and explored stack state is more conservative than the current one.
4335  * Example:
4336  *       explored                   current
4337  * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC)
4338  * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC)
4339  *
4340  * In other words if current stack state (one being explored) has more
4341  * valid slots than old one that already passed validation, it means
4342  * the verifier can stop exploring and conclude that current state is valid too
4343  *
4344  * Similarly with registers. If explored state has register type as invalid
4345  * whereas register type in current state is meaningful, it means that
4346  * the current state will reach 'bpf_exit' instruction safely
4347  */
4348 static bool func_states_equal(struct bpf_func_state *old,
4349 			      struct bpf_func_state *cur)
4350 {
4351 	struct idpair *idmap;
4352 	bool ret = false;
4353 	int i;
4354 
4355 	idmap = kcalloc(ID_MAP_SIZE, sizeof(struct idpair), GFP_KERNEL);
4356 	/* If we failed to allocate the idmap, just say it's not safe */
4357 	if (!idmap)
4358 		return false;
4359 
4360 	for (i = 0; i < MAX_BPF_REG; i++) {
4361 		if (!regsafe(&old->regs[i], &cur->regs[i], idmap))
4362 			goto out_free;
4363 	}
4364 
4365 	if (!stacksafe(old, cur, idmap))
4366 		goto out_free;
4367 	ret = true;
4368 out_free:
4369 	kfree(idmap);
4370 	return ret;
4371 }
4372 
4373 static bool states_equal(struct bpf_verifier_env *env,
4374 			 struct bpf_verifier_state *old,
4375 			 struct bpf_verifier_state *cur)
4376 {
4377 	int i;
4378 
4379 	if (old->curframe != cur->curframe)
4380 		return false;
4381 
4382 	/* for states to be equal callsites have to be the same
4383 	 * and all frame states need to be equivalent
4384 	 */
4385 	for (i = 0; i <= old->curframe; i++) {
4386 		if (old->frame[i]->callsite != cur->frame[i]->callsite)
4387 			return false;
4388 		if (!func_states_equal(old->frame[i], cur->frame[i]))
4389 			return false;
4390 	}
4391 	return true;
4392 }
4393 
4394 /* A write screens off any subsequent reads; but write marks come from the
4395  * straight-line code between a state and its parent.  When we arrive at an
4396  * equivalent state (jump target or such) we didn't arrive by the straight-line
4397  * code, so read marks in the state must propagate to the parent regardless
4398  * of the state's write marks. That's what 'parent == state->parent' comparison
4399  * in mark_reg_read() and mark_stack_slot_read() is for.
4400  */
4401 static int propagate_liveness(struct bpf_verifier_env *env,
4402 			      const struct bpf_verifier_state *vstate,
4403 			      struct bpf_verifier_state *vparent)
4404 {
4405 	int i, frame, err = 0;
4406 	struct bpf_func_state *state, *parent;
4407 
4408 	if (vparent->curframe != vstate->curframe) {
4409 		WARN(1, "propagate_live: parent frame %d current frame %d\n",
4410 		     vparent->curframe, vstate->curframe);
4411 		return -EFAULT;
4412 	}
4413 	/* Propagate read liveness of registers... */
4414 	BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
4415 	/* We don't need to worry about FP liveness because it's read-only */
4416 	for (i = 0; i < BPF_REG_FP; i++) {
4417 		if (vparent->frame[vparent->curframe]->regs[i].live & REG_LIVE_READ)
4418 			continue;
4419 		if (vstate->frame[vstate->curframe]->regs[i].live & REG_LIVE_READ) {
4420 			err = mark_reg_read(env, vstate, vparent, i);
4421 			if (err)
4422 				return err;
4423 		}
4424 	}
4425 
4426 	/* ... and stack slots */
4427 	for (frame = 0; frame <= vstate->curframe; frame++) {
4428 		state = vstate->frame[frame];
4429 		parent = vparent->frame[frame];
4430 		for (i = 0; i < state->allocated_stack / BPF_REG_SIZE &&
4431 			    i < parent->allocated_stack / BPF_REG_SIZE; i++) {
4432 			if (parent->stack[i].spilled_ptr.live & REG_LIVE_READ)
4433 				continue;
4434 			if (state->stack[i].spilled_ptr.live & REG_LIVE_READ)
4435 				mark_stack_slot_read(env, vstate, vparent, i, frame);
4436 		}
4437 	}
4438 	return err;
4439 }
4440 
4441 static int is_state_visited(struct bpf_verifier_env *env, int insn_idx)
4442 {
4443 	struct bpf_verifier_state_list *new_sl;
4444 	struct bpf_verifier_state_list *sl;
4445 	struct bpf_verifier_state *cur = env->cur_state;
4446 	int i, j, err;
4447 
4448 	sl = env->explored_states[insn_idx];
4449 	if (!sl)
4450 		/* this 'insn_idx' instruction wasn't marked, so we will not
4451 		 * be doing state search here
4452 		 */
4453 		return 0;
4454 
4455 	while (sl != STATE_LIST_MARK) {
4456 		if (states_equal(env, &sl->state, cur)) {
4457 			/* reached equivalent register/stack state,
4458 			 * prune the search.
4459 			 * Registers read by the continuation are read by us.
4460 			 * If we have any write marks in env->cur_state, they
4461 			 * will prevent corresponding reads in the continuation
4462 			 * from reaching our parent (an explored_state).  Our
4463 			 * own state will get the read marks recorded, but
4464 			 * they'll be immediately forgotten as we're pruning
4465 			 * this state and will pop a new one.
4466 			 */
4467 			err = propagate_liveness(env, &sl->state, cur);
4468 			if (err)
4469 				return err;
4470 			return 1;
4471 		}
4472 		sl = sl->next;
4473 	}
4474 
4475 	/* there were no equivalent states, remember current one.
4476 	 * technically the current state is not proven to be safe yet,
4477 	 * but it will either reach outer most bpf_exit (which means it's safe)
4478 	 * or it will be rejected. Since there are no loops, we won't be
4479 	 * seeing this tuple (frame[0].callsite, frame[1].callsite, .. insn_idx)
4480 	 * again on the way to bpf_exit
4481 	 */
4482 	new_sl = kzalloc(sizeof(struct bpf_verifier_state_list), GFP_KERNEL);
4483 	if (!new_sl)
4484 		return -ENOMEM;
4485 
4486 	/* add new state to the head of linked list */
4487 	err = copy_verifier_state(&new_sl->state, cur);
4488 	if (err) {
4489 		free_verifier_state(&new_sl->state, false);
4490 		kfree(new_sl);
4491 		return err;
4492 	}
4493 	new_sl->next = env->explored_states[insn_idx];
4494 	env->explored_states[insn_idx] = new_sl;
4495 	/* connect new state to parentage chain */
4496 	cur->parent = &new_sl->state;
4497 	/* clear write marks in current state: the writes we did are not writes
4498 	 * our child did, so they don't screen off its reads from us.
4499 	 * (There are no read marks in current state, because reads always mark
4500 	 * their parent and current state never has children yet.  Only
4501 	 * explored_states can get read marks.)
4502 	 */
4503 	for (i = 0; i < BPF_REG_FP; i++)
4504 		cur->frame[cur->curframe]->regs[i].live = REG_LIVE_NONE;
4505 
4506 	/* all stack frames are accessible from callee, clear them all */
4507 	for (j = 0; j <= cur->curframe; j++) {
4508 		struct bpf_func_state *frame = cur->frame[j];
4509 
4510 		for (i = 0; i < frame->allocated_stack / BPF_REG_SIZE; i++)
4511 			frame->stack[i].spilled_ptr.live = REG_LIVE_NONE;
4512 	}
4513 	return 0;
4514 }
4515 
4516 static int do_check(struct bpf_verifier_env *env)
4517 {
4518 	struct bpf_verifier_state *state;
4519 	struct bpf_insn *insns = env->prog->insnsi;
4520 	struct bpf_reg_state *regs;
4521 	int insn_cnt = env->prog->len, i;
4522 	int insn_idx, prev_insn_idx = 0;
4523 	int insn_processed = 0;
4524 	bool do_print_state = false;
4525 
4526 	state = kzalloc(sizeof(struct bpf_verifier_state), GFP_KERNEL);
4527 	if (!state)
4528 		return -ENOMEM;
4529 	state->curframe = 0;
4530 	state->parent = NULL;
4531 	state->frame[0] = kzalloc(sizeof(struct bpf_func_state), GFP_KERNEL);
4532 	if (!state->frame[0]) {
4533 		kfree(state);
4534 		return -ENOMEM;
4535 	}
4536 	env->cur_state = state;
4537 	init_func_state(env, state->frame[0],
4538 			BPF_MAIN_FUNC /* callsite */,
4539 			0 /* frameno */,
4540 			0 /* subprogno, zero == main subprog */);
4541 	insn_idx = 0;
4542 	for (;;) {
4543 		struct bpf_insn *insn;
4544 		u8 class;
4545 		int err;
4546 
4547 		if (insn_idx >= insn_cnt) {
4548 			verbose(env, "invalid insn idx %d insn_cnt %d\n",
4549 				insn_idx, insn_cnt);
4550 			return -EFAULT;
4551 		}
4552 
4553 		insn = &insns[insn_idx];
4554 		class = BPF_CLASS(insn->code);
4555 
4556 		if (++insn_processed > BPF_COMPLEXITY_LIMIT_INSNS) {
4557 			verbose(env,
4558 				"BPF program is too large. Processed %d insn\n",
4559 				insn_processed);
4560 			return -E2BIG;
4561 		}
4562 
4563 		err = is_state_visited(env, insn_idx);
4564 		if (err < 0)
4565 			return err;
4566 		if (err == 1) {
4567 			/* found equivalent state, can prune the search */
4568 			if (env->log.level) {
4569 				if (do_print_state)
4570 					verbose(env, "\nfrom %d to %d: safe\n",
4571 						prev_insn_idx, insn_idx);
4572 				else
4573 					verbose(env, "%d: safe\n", insn_idx);
4574 			}
4575 			goto process_bpf_exit;
4576 		}
4577 
4578 		if (need_resched())
4579 			cond_resched();
4580 
4581 		if (env->log.level > 1 || (env->log.level && do_print_state)) {
4582 			if (env->log.level > 1)
4583 				verbose(env, "%d:", insn_idx);
4584 			else
4585 				verbose(env, "\nfrom %d to %d:",
4586 					prev_insn_idx, insn_idx);
4587 			print_verifier_state(env, state->frame[state->curframe]);
4588 			do_print_state = false;
4589 		}
4590 
4591 		if (env->log.level) {
4592 			const struct bpf_insn_cbs cbs = {
4593 				.cb_print	= verbose,
4594 			};
4595 
4596 			verbose(env, "%d: ", insn_idx);
4597 			print_bpf_insn(&cbs, env, insn, env->allow_ptr_leaks);
4598 		}
4599 
4600 		if (bpf_prog_is_dev_bound(env->prog->aux)) {
4601 			err = bpf_prog_offload_verify_insn(env, insn_idx,
4602 							   prev_insn_idx);
4603 			if (err)
4604 				return err;
4605 		}
4606 
4607 		regs = cur_regs(env);
4608 		env->insn_aux_data[insn_idx].seen = true;
4609 		if (class == BPF_ALU || class == BPF_ALU64) {
4610 			err = check_alu_op(env, insn);
4611 			if (err)
4612 				return err;
4613 
4614 		} else if (class == BPF_LDX) {
4615 			enum bpf_reg_type *prev_src_type, src_reg_type;
4616 
4617 			/* check for reserved fields is already done */
4618 
4619 			/* check src operand */
4620 			err = check_reg_arg(env, insn->src_reg, SRC_OP);
4621 			if (err)
4622 				return err;
4623 
4624 			err = check_reg_arg(env, insn->dst_reg, DST_OP_NO_MARK);
4625 			if (err)
4626 				return err;
4627 
4628 			src_reg_type = regs[insn->src_reg].type;
4629 
4630 			/* check that memory (src_reg + off) is readable,
4631 			 * the state of dst_reg will be updated by this func
4632 			 */
4633 			err = check_mem_access(env, insn_idx, insn->src_reg, insn->off,
4634 					       BPF_SIZE(insn->code), BPF_READ,
4635 					       insn->dst_reg);
4636 			if (err)
4637 				return err;
4638 
4639 			prev_src_type = &env->insn_aux_data[insn_idx].ptr_type;
4640 
4641 			if (*prev_src_type == NOT_INIT) {
4642 				/* saw a valid insn
4643 				 * dst_reg = *(u32 *)(src_reg + off)
4644 				 * save type to validate intersecting paths
4645 				 */
4646 				*prev_src_type = src_reg_type;
4647 
4648 			} else if (src_reg_type != *prev_src_type &&
4649 				   (src_reg_type == PTR_TO_CTX ||
4650 				    *prev_src_type == PTR_TO_CTX)) {
4651 				/* ABuser program is trying to use the same insn
4652 				 * dst_reg = *(u32*) (src_reg + off)
4653 				 * with different pointer types:
4654 				 * src_reg == ctx in one branch and
4655 				 * src_reg == stack|map in some other branch.
4656 				 * Reject it.
4657 				 */
4658 				verbose(env, "same insn cannot be used with different pointers\n");
4659 				return -EINVAL;
4660 			}
4661 
4662 		} else if (class == BPF_STX) {
4663 			enum bpf_reg_type *prev_dst_type, dst_reg_type;
4664 
4665 			if (BPF_MODE(insn->code) == BPF_XADD) {
4666 				err = check_xadd(env, insn_idx, insn);
4667 				if (err)
4668 					return err;
4669 				insn_idx++;
4670 				continue;
4671 			}
4672 
4673 			/* check src1 operand */
4674 			err = check_reg_arg(env, insn->src_reg, SRC_OP);
4675 			if (err)
4676 				return err;
4677 			/* check src2 operand */
4678 			err = check_reg_arg(env, insn->dst_reg, SRC_OP);
4679 			if (err)
4680 				return err;
4681 
4682 			dst_reg_type = regs[insn->dst_reg].type;
4683 
4684 			/* check that memory (dst_reg + off) is writeable */
4685 			err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off,
4686 					       BPF_SIZE(insn->code), BPF_WRITE,
4687 					       insn->src_reg);
4688 			if (err)
4689 				return err;
4690 
4691 			prev_dst_type = &env->insn_aux_data[insn_idx].ptr_type;
4692 
4693 			if (*prev_dst_type == NOT_INIT) {
4694 				*prev_dst_type = dst_reg_type;
4695 			} else if (dst_reg_type != *prev_dst_type &&
4696 				   (dst_reg_type == PTR_TO_CTX ||
4697 				    *prev_dst_type == PTR_TO_CTX)) {
4698 				verbose(env, "same insn cannot be used with different pointers\n");
4699 				return -EINVAL;
4700 			}
4701 
4702 		} else if (class == BPF_ST) {
4703 			if (BPF_MODE(insn->code) != BPF_MEM ||
4704 			    insn->src_reg != BPF_REG_0) {
4705 				verbose(env, "BPF_ST uses reserved fields\n");
4706 				return -EINVAL;
4707 			}
4708 			/* check src operand */
4709 			err = check_reg_arg(env, insn->dst_reg, SRC_OP);
4710 			if (err)
4711 				return err;
4712 
4713 			if (is_ctx_reg(env, insn->dst_reg)) {
4714 				verbose(env, "BPF_ST stores into R%d context is not allowed\n",
4715 					insn->dst_reg);
4716 				return -EACCES;
4717 			}
4718 
4719 			/* check that memory (dst_reg + off) is writeable */
4720 			err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off,
4721 					       BPF_SIZE(insn->code), BPF_WRITE,
4722 					       -1);
4723 			if (err)
4724 				return err;
4725 
4726 		} else if (class == BPF_JMP) {
4727 			u8 opcode = BPF_OP(insn->code);
4728 
4729 			if (opcode == BPF_CALL) {
4730 				if (BPF_SRC(insn->code) != BPF_K ||
4731 				    insn->off != 0 ||
4732 				    (insn->src_reg != BPF_REG_0 &&
4733 				     insn->src_reg != BPF_PSEUDO_CALL) ||
4734 				    insn->dst_reg != BPF_REG_0) {
4735 					verbose(env, "BPF_CALL uses reserved fields\n");
4736 					return -EINVAL;
4737 				}
4738 
4739 				if (insn->src_reg == BPF_PSEUDO_CALL)
4740 					err = check_func_call(env, insn, &insn_idx);
4741 				else
4742 					err = check_helper_call(env, insn->imm, insn_idx);
4743 				if (err)
4744 					return err;
4745 
4746 			} else if (opcode == BPF_JA) {
4747 				if (BPF_SRC(insn->code) != BPF_K ||
4748 				    insn->imm != 0 ||
4749 				    insn->src_reg != BPF_REG_0 ||
4750 				    insn->dst_reg != BPF_REG_0) {
4751 					verbose(env, "BPF_JA uses reserved fields\n");
4752 					return -EINVAL;
4753 				}
4754 
4755 				insn_idx += insn->off + 1;
4756 				continue;
4757 
4758 			} else if (opcode == BPF_EXIT) {
4759 				if (BPF_SRC(insn->code) != BPF_K ||
4760 				    insn->imm != 0 ||
4761 				    insn->src_reg != BPF_REG_0 ||
4762 				    insn->dst_reg != BPF_REG_0) {
4763 					verbose(env, "BPF_EXIT uses reserved fields\n");
4764 					return -EINVAL;
4765 				}
4766 
4767 				if (state->curframe) {
4768 					/* exit from nested function */
4769 					prev_insn_idx = insn_idx;
4770 					err = prepare_func_exit(env, &insn_idx);
4771 					if (err)
4772 						return err;
4773 					do_print_state = true;
4774 					continue;
4775 				}
4776 
4777 				/* eBPF calling convetion is such that R0 is used
4778 				 * to return the value from eBPF program.
4779 				 * Make sure that it's readable at this time
4780 				 * of bpf_exit, which means that program wrote
4781 				 * something into it earlier
4782 				 */
4783 				err = check_reg_arg(env, BPF_REG_0, SRC_OP);
4784 				if (err)
4785 					return err;
4786 
4787 				if (is_pointer_value(env, BPF_REG_0)) {
4788 					verbose(env, "R0 leaks addr as return value\n");
4789 					return -EACCES;
4790 				}
4791 
4792 				err = check_return_code(env);
4793 				if (err)
4794 					return err;
4795 process_bpf_exit:
4796 				err = pop_stack(env, &prev_insn_idx, &insn_idx);
4797 				if (err < 0) {
4798 					if (err != -ENOENT)
4799 						return err;
4800 					break;
4801 				} else {
4802 					do_print_state = true;
4803 					continue;
4804 				}
4805 			} else {
4806 				err = check_cond_jmp_op(env, insn, &insn_idx);
4807 				if (err)
4808 					return err;
4809 			}
4810 		} else if (class == BPF_LD) {
4811 			u8 mode = BPF_MODE(insn->code);
4812 
4813 			if (mode == BPF_ABS || mode == BPF_IND) {
4814 				err = check_ld_abs(env, insn);
4815 				if (err)
4816 					return err;
4817 
4818 			} else if (mode == BPF_IMM) {
4819 				err = check_ld_imm(env, insn);
4820 				if (err)
4821 					return err;
4822 
4823 				insn_idx++;
4824 				env->insn_aux_data[insn_idx].seen = true;
4825 			} else {
4826 				verbose(env, "invalid BPF_LD mode\n");
4827 				return -EINVAL;
4828 			}
4829 		} else {
4830 			verbose(env, "unknown insn class %d\n", class);
4831 			return -EINVAL;
4832 		}
4833 
4834 		insn_idx++;
4835 	}
4836 
4837 	verbose(env, "processed %d insns (limit %d), stack depth ",
4838 		insn_processed, BPF_COMPLEXITY_LIMIT_INSNS);
4839 	for (i = 0; i < env->subprog_cnt + 1; i++) {
4840 		u32 depth = env->subprog_stack_depth[i];
4841 
4842 		verbose(env, "%d", depth);
4843 		if (i + 1 < env->subprog_cnt + 1)
4844 			verbose(env, "+");
4845 	}
4846 	verbose(env, "\n");
4847 	env->prog->aux->stack_depth = env->subprog_stack_depth[0];
4848 	return 0;
4849 }
4850 
4851 static int check_map_prealloc(struct bpf_map *map)
4852 {
4853 	return (map->map_type != BPF_MAP_TYPE_HASH &&
4854 		map->map_type != BPF_MAP_TYPE_PERCPU_HASH &&
4855 		map->map_type != BPF_MAP_TYPE_HASH_OF_MAPS) ||
4856 		!(map->map_flags & BPF_F_NO_PREALLOC);
4857 }
4858 
4859 static int check_map_prog_compatibility(struct bpf_verifier_env *env,
4860 					struct bpf_map *map,
4861 					struct bpf_prog *prog)
4862 
4863 {
4864 	/* Make sure that BPF_PROG_TYPE_PERF_EVENT programs only use
4865 	 * preallocated hash maps, since doing memory allocation
4866 	 * in overflow_handler can crash depending on where nmi got
4867 	 * triggered.
4868 	 */
4869 	if (prog->type == BPF_PROG_TYPE_PERF_EVENT) {
4870 		if (!check_map_prealloc(map)) {
4871 			verbose(env, "perf_event programs can only use preallocated hash map\n");
4872 			return -EINVAL;
4873 		}
4874 		if (map->inner_map_meta &&
4875 		    !check_map_prealloc(map->inner_map_meta)) {
4876 			verbose(env, "perf_event programs can only use preallocated inner hash map\n");
4877 			return -EINVAL;
4878 		}
4879 	}
4880 
4881 	if ((bpf_prog_is_dev_bound(prog->aux) || bpf_map_is_dev_bound(map)) &&
4882 	    !bpf_offload_dev_match(prog, map)) {
4883 		verbose(env, "offload device mismatch between prog and map\n");
4884 		return -EINVAL;
4885 	}
4886 
4887 	return 0;
4888 }
4889 
4890 /* look for pseudo eBPF instructions that access map FDs and
4891  * replace them with actual map pointers
4892  */
4893 static int replace_map_fd_with_map_ptr(struct bpf_verifier_env *env)
4894 {
4895 	struct bpf_insn *insn = env->prog->insnsi;
4896 	int insn_cnt = env->prog->len;
4897 	int i, j, err;
4898 
4899 	err = bpf_prog_calc_tag(env->prog);
4900 	if (err)
4901 		return err;
4902 
4903 	for (i = 0; i < insn_cnt; i++, insn++) {
4904 		if (BPF_CLASS(insn->code) == BPF_LDX &&
4905 		    (BPF_MODE(insn->code) != BPF_MEM || insn->imm != 0)) {
4906 			verbose(env, "BPF_LDX uses reserved fields\n");
4907 			return -EINVAL;
4908 		}
4909 
4910 		if (BPF_CLASS(insn->code) == BPF_STX &&
4911 		    ((BPF_MODE(insn->code) != BPF_MEM &&
4912 		      BPF_MODE(insn->code) != BPF_XADD) || insn->imm != 0)) {
4913 			verbose(env, "BPF_STX uses reserved fields\n");
4914 			return -EINVAL;
4915 		}
4916 
4917 		if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW)) {
4918 			struct bpf_map *map;
4919 			struct fd f;
4920 
4921 			if (i == insn_cnt - 1 || insn[1].code != 0 ||
4922 			    insn[1].dst_reg != 0 || insn[1].src_reg != 0 ||
4923 			    insn[1].off != 0) {
4924 				verbose(env, "invalid bpf_ld_imm64 insn\n");
4925 				return -EINVAL;
4926 			}
4927 
4928 			if (insn->src_reg == 0)
4929 				/* valid generic load 64-bit imm */
4930 				goto next_insn;
4931 
4932 			if (insn->src_reg != BPF_PSEUDO_MAP_FD) {
4933 				verbose(env,
4934 					"unrecognized bpf_ld_imm64 insn\n");
4935 				return -EINVAL;
4936 			}
4937 
4938 			f = fdget(insn->imm);
4939 			map = __bpf_map_get(f);
4940 			if (IS_ERR(map)) {
4941 				verbose(env, "fd %d is not pointing to valid bpf_map\n",
4942 					insn->imm);
4943 				return PTR_ERR(map);
4944 			}
4945 
4946 			err = check_map_prog_compatibility(env, map, env->prog);
4947 			if (err) {
4948 				fdput(f);
4949 				return err;
4950 			}
4951 
4952 			/* store map pointer inside BPF_LD_IMM64 instruction */
4953 			insn[0].imm = (u32) (unsigned long) map;
4954 			insn[1].imm = ((u64) (unsigned long) map) >> 32;
4955 
4956 			/* check whether we recorded this map already */
4957 			for (j = 0; j < env->used_map_cnt; j++)
4958 				if (env->used_maps[j] == map) {
4959 					fdput(f);
4960 					goto next_insn;
4961 				}
4962 
4963 			if (env->used_map_cnt >= MAX_USED_MAPS) {
4964 				fdput(f);
4965 				return -E2BIG;
4966 			}
4967 
4968 			/* hold the map. If the program is rejected by verifier,
4969 			 * the map will be released by release_maps() or it
4970 			 * will be used by the valid program until it's unloaded
4971 			 * and all maps are released in free_bpf_prog_info()
4972 			 */
4973 			map = bpf_map_inc(map, false);
4974 			if (IS_ERR(map)) {
4975 				fdput(f);
4976 				return PTR_ERR(map);
4977 			}
4978 			env->used_maps[env->used_map_cnt++] = map;
4979 
4980 			fdput(f);
4981 next_insn:
4982 			insn++;
4983 			i++;
4984 			continue;
4985 		}
4986 
4987 		/* Basic sanity check before we invest more work here. */
4988 		if (!bpf_opcode_in_insntable(insn->code)) {
4989 			verbose(env, "unknown opcode %02x\n", insn->code);
4990 			return -EINVAL;
4991 		}
4992 	}
4993 
4994 	/* now all pseudo BPF_LD_IMM64 instructions load valid
4995 	 * 'struct bpf_map *' into a register instead of user map_fd.
4996 	 * These pointers will be used later by verifier to validate map access.
4997 	 */
4998 	return 0;
4999 }
5000 
5001 /* drop refcnt of maps used by the rejected program */
5002 static void release_maps(struct bpf_verifier_env *env)
5003 {
5004 	int i;
5005 
5006 	for (i = 0; i < env->used_map_cnt; i++)
5007 		bpf_map_put(env->used_maps[i]);
5008 }
5009 
5010 /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */
5011 static void convert_pseudo_ld_imm64(struct bpf_verifier_env *env)
5012 {
5013 	struct bpf_insn *insn = env->prog->insnsi;
5014 	int insn_cnt = env->prog->len;
5015 	int i;
5016 
5017 	for (i = 0; i < insn_cnt; i++, insn++)
5018 		if (insn->code == (BPF_LD | BPF_IMM | BPF_DW))
5019 			insn->src_reg = 0;
5020 }
5021 
5022 /* single env->prog->insni[off] instruction was replaced with the range
5023  * insni[off, off + cnt).  Adjust corresponding insn_aux_data by copying
5024  * [0, off) and [off, end) to new locations, so the patched range stays zero
5025  */
5026 static int adjust_insn_aux_data(struct bpf_verifier_env *env, u32 prog_len,
5027 				u32 off, u32 cnt)
5028 {
5029 	struct bpf_insn_aux_data *new_data, *old_data = env->insn_aux_data;
5030 	int i;
5031 
5032 	if (cnt == 1)
5033 		return 0;
5034 	new_data = vzalloc(sizeof(struct bpf_insn_aux_data) * prog_len);
5035 	if (!new_data)
5036 		return -ENOMEM;
5037 	memcpy(new_data, old_data, sizeof(struct bpf_insn_aux_data) * off);
5038 	memcpy(new_data + off + cnt - 1, old_data + off,
5039 	       sizeof(struct bpf_insn_aux_data) * (prog_len - off - cnt + 1));
5040 	for (i = off; i < off + cnt - 1; i++)
5041 		new_data[i].seen = true;
5042 	env->insn_aux_data = new_data;
5043 	vfree(old_data);
5044 	return 0;
5045 }
5046 
5047 static void adjust_subprog_starts(struct bpf_verifier_env *env, u32 off, u32 len)
5048 {
5049 	int i;
5050 
5051 	if (len == 1)
5052 		return;
5053 	for (i = 0; i < env->subprog_cnt; i++) {
5054 		if (env->subprog_starts[i] < off)
5055 			continue;
5056 		env->subprog_starts[i] += len - 1;
5057 	}
5058 }
5059 
5060 static struct bpf_prog *bpf_patch_insn_data(struct bpf_verifier_env *env, u32 off,
5061 					    const struct bpf_insn *patch, u32 len)
5062 {
5063 	struct bpf_prog *new_prog;
5064 
5065 	new_prog = bpf_patch_insn_single(env->prog, off, patch, len);
5066 	if (!new_prog)
5067 		return NULL;
5068 	if (adjust_insn_aux_data(env, new_prog->len, off, len))
5069 		return NULL;
5070 	adjust_subprog_starts(env, off, len);
5071 	return new_prog;
5072 }
5073 
5074 /* The verifier does more data flow analysis than llvm and will not
5075  * explore branches that are dead at run time. Malicious programs can
5076  * have dead code too. Therefore replace all dead at-run-time code
5077  * with 'ja -1'.
5078  *
5079  * Just nops are not optimal, e.g. if they would sit at the end of the
5080  * program and through another bug we would manage to jump there, then
5081  * we'd execute beyond program memory otherwise. Returning exception
5082  * code also wouldn't work since we can have subprogs where the dead
5083  * code could be located.
5084  */
5085 static void sanitize_dead_code(struct bpf_verifier_env *env)
5086 {
5087 	struct bpf_insn_aux_data *aux_data = env->insn_aux_data;
5088 	struct bpf_insn trap = BPF_JMP_IMM(BPF_JA, 0, 0, -1);
5089 	struct bpf_insn *insn = env->prog->insnsi;
5090 	const int insn_cnt = env->prog->len;
5091 	int i;
5092 
5093 	for (i = 0; i < insn_cnt; i++) {
5094 		if (aux_data[i].seen)
5095 			continue;
5096 		memcpy(insn + i, &trap, sizeof(trap));
5097 	}
5098 }
5099 
5100 /* convert load instructions that access fields of 'struct __sk_buff'
5101  * into sequence of instructions that access fields of 'struct sk_buff'
5102  */
5103 static int convert_ctx_accesses(struct bpf_verifier_env *env)
5104 {
5105 	const struct bpf_verifier_ops *ops = env->ops;
5106 	int i, cnt, size, ctx_field_size, delta = 0;
5107 	const int insn_cnt = env->prog->len;
5108 	struct bpf_insn insn_buf[16], *insn;
5109 	struct bpf_prog *new_prog;
5110 	enum bpf_access_type type;
5111 	bool is_narrower_load;
5112 	u32 target_size;
5113 
5114 	if (ops->gen_prologue) {
5115 		cnt = ops->gen_prologue(insn_buf, env->seen_direct_write,
5116 					env->prog);
5117 		if (cnt >= ARRAY_SIZE(insn_buf)) {
5118 			verbose(env, "bpf verifier is misconfigured\n");
5119 			return -EINVAL;
5120 		} else if (cnt) {
5121 			new_prog = bpf_patch_insn_data(env, 0, insn_buf, cnt);
5122 			if (!new_prog)
5123 				return -ENOMEM;
5124 
5125 			env->prog = new_prog;
5126 			delta += cnt - 1;
5127 		}
5128 	}
5129 
5130 	if (!ops->convert_ctx_access)
5131 		return 0;
5132 
5133 	insn = env->prog->insnsi + delta;
5134 
5135 	for (i = 0; i < insn_cnt; i++, insn++) {
5136 		if (insn->code == (BPF_LDX | BPF_MEM | BPF_B) ||
5137 		    insn->code == (BPF_LDX | BPF_MEM | BPF_H) ||
5138 		    insn->code == (BPF_LDX | BPF_MEM | BPF_W) ||
5139 		    insn->code == (BPF_LDX | BPF_MEM | BPF_DW))
5140 			type = BPF_READ;
5141 		else if (insn->code == (BPF_STX | BPF_MEM | BPF_B) ||
5142 			 insn->code == (BPF_STX | BPF_MEM | BPF_H) ||
5143 			 insn->code == (BPF_STX | BPF_MEM | BPF_W) ||
5144 			 insn->code == (BPF_STX | BPF_MEM | BPF_DW))
5145 			type = BPF_WRITE;
5146 		else
5147 			continue;
5148 
5149 		if (env->insn_aux_data[i + delta].ptr_type != PTR_TO_CTX)
5150 			continue;
5151 
5152 		ctx_field_size = env->insn_aux_data[i + delta].ctx_field_size;
5153 		size = BPF_LDST_BYTES(insn);
5154 
5155 		/* If the read access is a narrower load of the field,
5156 		 * convert to a 4/8-byte load, to minimum program type specific
5157 		 * convert_ctx_access changes. If conversion is successful,
5158 		 * we will apply proper mask to the result.
5159 		 */
5160 		is_narrower_load = size < ctx_field_size;
5161 		if (is_narrower_load) {
5162 			u32 off = insn->off;
5163 			u8 size_code;
5164 
5165 			if (type == BPF_WRITE) {
5166 				verbose(env, "bpf verifier narrow ctx access misconfigured\n");
5167 				return -EINVAL;
5168 			}
5169 
5170 			size_code = BPF_H;
5171 			if (ctx_field_size == 4)
5172 				size_code = BPF_W;
5173 			else if (ctx_field_size == 8)
5174 				size_code = BPF_DW;
5175 
5176 			insn->off = off & ~(ctx_field_size - 1);
5177 			insn->code = BPF_LDX | BPF_MEM | size_code;
5178 		}
5179 
5180 		target_size = 0;
5181 		cnt = ops->convert_ctx_access(type, insn, insn_buf, env->prog,
5182 					      &target_size);
5183 		if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf) ||
5184 		    (ctx_field_size && !target_size)) {
5185 			verbose(env, "bpf verifier is misconfigured\n");
5186 			return -EINVAL;
5187 		}
5188 
5189 		if (is_narrower_load && size < target_size) {
5190 			if (ctx_field_size <= 4)
5191 				insn_buf[cnt++] = BPF_ALU32_IMM(BPF_AND, insn->dst_reg,
5192 								(1 << size * 8) - 1);
5193 			else
5194 				insn_buf[cnt++] = BPF_ALU64_IMM(BPF_AND, insn->dst_reg,
5195 								(1 << size * 8) - 1);
5196 		}
5197 
5198 		new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt);
5199 		if (!new_prog)
5200 			return -ENOMEM;
5201 
5202 		delta += cnt - 1;
5203 
5204 		/* keep walking new program and skip insns we just inserted */
5205 		env->prog = new_prog;
5206 		insn      = new_prog->insnsi + i + delta;
5207 	}
5208 
5209 	return 0;
5210 }
5211 
5212 static int jit_subprogs(struct bpf_verifier_env *env)
5213 {
5214 	struct bpf_prog *prog = env->prog, **func, *tmp;
5215 	int i, j, subprog_start, subprog_end = 0, len, subprog;
5216 	struct bpf_insn *insn;
5217 	void *old_bpf_func;
5218 	int err = -ENOMEM;
5219 
5220 	if (env->subprog_cnt == 0)
5221 		return 0;
5222 
5223 	for (i = 0, insn = prog->insnsi; i < prog->len; i++, insn++) {
5224 		if (insn->code != (BPF_JMP | BPF_CALL) ||
5225 		    insn->src_reg != BPF_PSEUDO_CALL)
5226 			continue;
5227 		subprog = find_subprog(env, i + insn->imm + 1);
5228 		if (subprog < 0) {
5229 			WARN_ONCE(1, "verifier bug. No program starts at insn %d\n",
5230 				  i + insn->imm + 1);
5231 			return -EFAULT;
5232 		}
5233 		/* temporarily remember subprog id inside insn instead of
5234 		 * aux_data, since next loop will split up all insns into funcs
5235 		 */
5236 		insn->off = subprog + 1;
5237 		/* remember original imm in case JIT fails and fallback
5238 		 * to interpreter will be needed
5239 		 */
5240 		env->insn_aux_data[i].call_imm = insn->imm;
5241 		/* point imm to __bpf_call_base+1 from JITs point of view */
5242 		insn->imm = 1;
5243 	}
5244 
5245 	func = kzalloc(sizeof(prog) * (env->subprog_cnt + 1), GFP_KERNEL);
5246 	if (!func)
5247 		return -ENOMEM;
5248 
5249 	for (i = 0; i <= env->subprog_cnt; i++) {
5250 		subprog_start = subprog_end;
5251 		if (env->subprog_cnt == i)
5252 			subprog_end = prog->len;
5253 		else
5254 			subprog_end = env->subprog_starts[i];
5255 
5256 		len = subprog_end - subprog_start;
5257 		func[i] = bpf_prog_alloc(bpf_prog_size(len), GFP_USER);
5258 		if (!func[i])
5259 			goto out_free;
5260 		memcpy(func[i]->insnsi, &prog->insnsi[subprog_start],
5261 		       len * sizeof(struct bpf_insn));
5262 		func[i]->type = prog->type;
5263 		func[i]->len = len;
5264 		if (bpf_prog_calc_tag(func[i]))
5265 			goto out_free;
5266 		func[i]->is_func = 1;
5267 		/* Use bpf_prog_F_tag to indicate functions in stack traces.
5268 		 * Long term would need debug info to populate names
5269 		 */
5270 		func[i]->aux->name[0] = 'F';
5271 		func[i]->aux->stack_depth = env->subprog_stack_depth[i];
5272 		func[i]->jit_requested = 1;
5273 		func[i] = bpf_int_jit_compile(func[i]);
5274 		if (!func[i]->jited) {
5275 			err = -ENOTSUPP;
5276 			goto out_free;
5277 		}
5278 		cond_resched();
5279 	}
5280 	/* at this point all bpf functions were successfully JITed
5281 	 * now populate all bpf_calls with correct addresses and
5282 	 * run last pass of JIT
5283 	 */
5284 	for (i = 0; i <= env->subprog_cnt; i++) {
5285 		insn = func[i]->insnsi;
5286 		for (j = 0; j < func[i]->len; j++, insn++) {
5287 			if (insn->code != (BPF_JMP | BPF_CALL) ||
5288 			    insn->src_reg != BPF_PSEUDO_CALL)
5289 				continue;
5290 			subprog = insn->off;
5291 			insn->off = 0;
5292 			insn->imm = (u64 (*)(u64, u64, u64, u64, u64))
5293 				func[subprog]->bpf_func -
5294 				__bpf_call_base;
5295 		}
5296 	}
5297 	for (i = 0; i <= env->subprog_cnt; i++) {
5298 		old_bpf_func = func[i]->bpf_func;
5299 		tmp = bpf_int_jit_compile(func[i]);
5300 		if (tmp != func[i] || func[i]->bpf_func != old_bpf_func) {
5301 			verbose(env, "JIT doesn't support bpf-to-bpf calls\n");
5302 			err = -EFAULT;
5303 			goto out_free;
5304 		}
5305 		cond_resched();
5306 	}
5307 
5308 	/* finally lock prog and jit images for all functions and
5309 	 * populate kallsysm
5310 	 */
5311 	for (i = 0; i <= env->subprog_cnt; i++) {
5312 		bpf_prog_lock_ro(func[i]);
5313 		bpf_prog_kallsyms_add(func[i]);
5314 	}
5315 
5316 	/* Last step: make now unused interpreter insns from main
5317 	 * prog consistent for later dump requests, so they can
5318 	 * later look the same as if they were interpreted only.
5319 	 */
5320 	for (i = 0, insn = prog->insnsi; i < prog->len; i++, insn++) {
5321 		unsigned long addr;
5322 
5323 		if (insn->code != (BPF_JMP | BPF_CALL) ||
5324 		    insn->src_reg != BPF_PSEUDO_CALL)
5325 			continue;
5326 		insn->off = env->insn_aux_data[i].call_imm;
5327 		subprog = find_subprog(env, i + insn->off + 1);
5328 		addr  = (unsigned long)func[subprog + 1]->bpf_func;
5329 		addr &= PAGE_MASK;
5330 		insn->imm = (u64 (*)(u64, u64, u64, u64, u64))
5331 			    addr - __bpf_call_base;
5332 	}
5333 
5334 	prog->jited = 1;
5335 	prog->bpf_func = func[0]->bpf_func;
5336 	prog->aux->func = func;
5337 	prog->aux->func_cnt = env->subprog_cnt + 1;
5338 	return 0;
5339 out_free:
5340 	for (i = 0; i <= env->subprog_cnt; i++)
5341 		if (func[i])
5342 			bpf_jit_free(func[i]);
5343 	kfree(func);
5344 	/* cleanup main prog to be interpreted */
5345 	prog->jit_requested = 0;
5346 	for (i = 0, insn = prog->insnsi; i < prog->len; i++, insn++) {
5347 		if (insn->code != (BPF_JMP | BPF_CALL) ||
5348 		    insn->src_reg != BPF_PSEUDO_CALL)
5349 			continue;
5350 		insn->off = 0;
5351 		insn->imm = env->insn_aux_data[i].call_imm;
5352 	}
5353 	return err;
5354 }
5355 
5356 static int fixup_call_args(struct bpf_verifier_env *env)
5357 {
5358 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
5359 	struct bpf_prog *prog = env->prog;
5360 	struct bpf_insn *insn = prog->insnsi;
5361 	int i, depth;
5362 #endif
5363 	int err;
5364 
5365 	err = 0;
5366 	if (env->prog->jit_requested) {
5367 		err = jit_subprogs(env);
5368 		if (err == 0)
5369 			return 0;
5370 	}
5371 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
5372 	for (i = 0; i < prog->len; i++, insn++) {
5373 		if (insn->code != (BPF_JMP | BPF_CALL) ||
5374 		    insn->src_reg != BPF_PSEUDO_CALL)
5375 			continue;
5376 		depth = get_callee_stack_depth(env, insn, i);
5377 		if (depth < 0)
5378 			return depth;
5379 		bpf_patch_call_args(insn, depth);
5380 	}
5381 	err = 0;
5382 #endif
5383 	return err;
5384 }
5385 
5386 /* fixup insn->imm field of bpf_call instructions
5387  * and inline eligible helpers as explicit sequence of BPF instructions
5388  *
5389  * this function is called after eBPF program passed verification
5390  */
5391 static int fixup_bpf_calls(struct bpf_verifier_env *env)
5392 {
5393 	struct bpf_prog *prog = env->prog;
5394 	struct bpf_insn *insn = prog->insnsi;
5395 	const struct bpf_func_proto *fn;
5396 	const int insn_cnt = prog->len;
5397 	struct bpf_insn insn_buf[16];
5398 	struct bpf_prog *new_prog;
5399 	struct bpf_map *map_ptr;
5400 	int i, cnt, delta = 0;
5401 
5402 	for (i = 0; i < insn_cnt; i++, insn++) {
5403 		if (insn->code == (BPF_ALU64 | BPF_MOD | BPF_X) ||
5404 		    insn->code == (BPF_ALU64 | BPF_DIV | BPF_X) ||
5405 		    insn->code == (BPF_ALU | BPF_MOD | BPF_X) ||
5406 		    insn->code == (BPF_ALU | BPF_DIV | BPF_X)) {
5407 			bool is64 = BPF_CLASS(insn->code) == BPF_ALU64;
5408 			struct bpf_insn mask_and_div[] = {
5409 				BPF_MOV32_REG(insn->src_reg, insn->src_reg),
5410 				/* Rx div 0 -> 0 */
5411 				BPF_JMP_IMM(BPF_JNE, insn->src_reg, 0, 2),
5412 				BPF_ALU32_REG(BPF_XOR, insn->dst_reg, insn->dst_reg),
5413 				BPF_JMP_IMM(BPF_JA, 0, 0, 1),
5414 				*insn,
5415 			};
5416 			struct bpf_insn mask_and_mod[] = {
5417 				BPF_MOV32_REG(insn->src_reg, insn->src_reg),
5418 				/* Rx mod 0 -> Rx */
5419 				BPF_JMP_IMM(BPF_JEQ, insn->src_reg, 0, 1),
5420 				*insn,
5421 			};
5422 			struct bpf_insn *patchlet;
5423 
5424 			if (insn->code == (BPF_ALU64 | BPF_DIV | BPF_X) ||
5425 			    insn->code == (BPF_ALU | BPF_DIV | BPF_X)) {
5426 				patchlet = mask_and_div + (is64 ? 1 : 0);
5427 				cnt = ARRAY_SIZE(mask_and_div) - (is64 ? 1 : 0);
5428 			} else {
5429 				patchlet = mask_and_mod + (is64 ? 1 : 0);
5430 				cnt = ARRAY_SIZE(mask_and_mod) - (is64 ? 1 : 0);
5431 			}
5432 
5433 			new_prog = bpf_patch_insn_data(env, i + delta, patchlet, cnt);
5434 			if (!new_prog)
5435 				return -ENOMEM;
5436 
5437 			delta    += cnt - 1;
5438 			env->prog = prog = new_prog;
5439 			insn      = new_prog->insnsi + i + delta;
5440 			continue;
5441 		}
5442 
5443 		if (insn->code != (BPF_JMP | BPF_CALL))
5444 			continue;
5445 		if (insn->src_reg == BPF_PSEUDO_CALL)
5446 			continue;
5447 
5448 		if (insn->imm == BPF_FUNC_get_route_realm)
5449 			prog->dst_needed = 1;
5450 		if (insn->imm == BPF_FUNC_get_prandom_u32)
5451 			bpf_user_rnd_init_once();
5452 		if (insn->imm == BPF_FUNC_override_return)
5453 			prog->kprobe_override = 1;
5454 		if (insn->imm == BPF_FUNC_tail_call) {
5455 			/* If we tail call into other programs, we
5456 			 * cannot make any assumptions since they can
5457 			 * be replaced dynamically during runtime in
5458 			 * the program array.
5459 			 */
5460 			prog->cb_access = 1;
5461 			env->prog->aux->stack_depth = MAX_BPF_STACK;
5462 
5463 			/* mark bpf_tail_call as different opcode to avoid
5464 			 * conditional branch in the interpeter for every normal
5465 			 * call and to prevent accidental JITing by JIT compiler
5466 			 * that doesn't support bpf_tail_call yet
5467 			 */
5468 			insn->imm = 0;
5469 			insn->code = BPF_JMP | BPF_TAIL_CALL;
5470 
5471 			/* instead of changing every JIT dealing with tail_call
5472 			 * emit two extra insns:
5473 			 * if (index >= max_entries) goto out;
5474 			 * index &= array->index_mask;
5475 			 * to avoid out-of-bounds cpu speculation
5476 			 */
5477 			map_ptr = env->insn_aux_data[i + delta].map_ptr;
5478 			if (map_ptr == BPF_MAP_PTR_POISON) {
5479 				verbose(env, "tail_call abusing map_ptr\n");
5480 				return -EINVAL;
5481 			}
5482 			if (!map_ptr->unpriv_array)
5483 				continue;
5484 			insn_buf[0] = BPF_JMP_IMM(BPF_JGE, BPF_REG_3,
5485 						  map_ptr->max_entries, 2);
5486 			insn_buf[1] = BPF_ALU32_IMM(BPF_AND, BPF_REG_3,
5487 						    container_of(map_ptr,
5488 								 struct bpf_array,
5489 								 map)->index_mask);
5490 			insn_buf[2] = *insn;
5491 			cnt = 3;
5492 			new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt);
5493 			if (!new_prog)
5494 				return -ENOMEM;
5495 
5496 			delta    += cnt - 1;
5497 			env->prog = prog = new_prog;
5498 			insn      = new_prog->insnsi + i + delta;
5499 			continue;
5500 		}
5501 
5502 		/* BPF_EMIT_CALL() assumptions in some of the map_gen_lookup
5503 		 * handlers are currently limited to 64 bit only.
5504 		 */
5505 		if (prog->jit_requested && BITS_PER_LONG == 64 &&
5506 		    insn->imm == BPF_FUNC_map_lookup_elem) {
5507 			map_ptr = env->insn_aux_data[i + delta].map_ptr;
5508 			if (map_ptr == BPF_MAP_PTR_POISON ||
5509 			    !map_ptr->ops->map_gen_lookup)
5510 				goto patch_call_imm;
5511 
5512 			cnt = map_ptr->ops->map_gen_lookup(map_ptr, insn_buf);
5513 			if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) {
5514 				verbose(env, "bpf verifier is misconfigured\n");
5515 				return -EINVAL;
5516 			}
5517 
5518 			new_prog = bpf_patch_insn_data(env, i + delta, insn_buf,
5519 						       cnt);
5520 			if (!new_prog)
5521 				return -ENOMEM;
5522 
5523 			delta += cnt - 1;
5524 
5525 			/* keep walking new program and skip insns we just inserted */
5526 			env->prog = prog = new_prog;
5527 			insn      = new_prog->insnsi + i + delta;
5528 			continue;
5529 		}
5530 
5531 		if (insn->imm == BPF_FUNC_redirect_map) {
5532 			/* Note, we cannot use prog directly as imm as subsequent
5533 			 * rewrites would still change the prog pointer. The only
5534 			 * stable address we can use is aux, which also works with
5535 			 * prog clones during blinding.
5536 			 */
5537 			u64 addr = (unsigned long)prog->aux;
5538 			struct bpf_insn r4_ld[] = {
5539 				BPF_LD_IMM64(BPF_REG_4, addr),
5540 				*insn,
5541 			};
5542 			cnt = ARRAY_SIZE(r4_ld);
5543 
5544 			new_prog = bpf_patch_insn_data(env, i + delta, r4_ld, cnt);
5545 			if (!new_prog)
5546 				return -ENOMEM;
5547 
5548 			delta    += cnt - 1;
5549 			env->prog = prog = new_prog;
5550 			insn      = new_prog->insnsi + i + delta;
5551 		}
5552 patch_call_imm:
5553 		fn = env->ops->get_func_proto(insn->imm);
5554 		/* all functions that have prototype and verifier allowed
5555 		 * programs to call them, must be real in-kernel functions
5556 		 */
5557 		if (!fn->func) {
5558 			verbose(env,
5559 				"kernel subsystem misconfigured func %s#%d\n",
5560 				func_id_name(insn->imm), insn->imm);
5561 			return -EFAULT;
5562 		}
5563 		insn->imm = fn->func - __bpf_call_base;
5564 	}
5565 
5566 	return 0;
5567 }
5568 
5569 static void free_states(struct bpf_verifier_env *env)
5570 {
5571 	struct bpf_verifier_state_list *sl, *sln;
5572 	int i;
5573 
5574 	if (!env->explored_states)
5575 		return;
5576 
5577 	for (i = 0; i < env->prog->len; i++) {
5578 		sl = env->explored_states[i];
5579 
5580 		if (sl)
5581 			while (sl != STATE_LIST_MARK) {
5582 				sln = sl->next;
5583 				free_verifier_state(&sl->state, false);
5584 				kfree(sl);
5585 				sl = sln;
5586 			}
5587 	}
5588 
5589 	kfree(env->explored_states);
5590 }
5591 
5592 int bpf_check(struct bpf_prog **prog, union bpf_attr *attr)
5593 {
5594 	struct bpf_verifier_env *env;
5595 	struct bpf_verifer_log *log;
5596 	int ret = -EINVAL;
5597 
5598 	/* no program is valid */
5599 	if (ARRAY_SIZE(bpf_verifier_ops) == 0)
5600 		return -EINVAL;
5601 
5602 	/* 'struct bpf_verifier_env' can be global, but since it's not small,
5603 	 * allocate/free it every time bpf_check() is called
5604 	 */
5605 	env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL);
5606 	if (!env)
5607 		return -ENOMEM;
5608 	log = &env->log;
5609 
5610 	env->insn_aux_data = vzalloc(sizeof(struct bpf_insn_aux_data) *
5611 				     (*prog)->len);
5612 	ret = -ENOMEM;
5613 	if (!env->insn_aux_data)
5614 		goto err_free_env;
5615 	env->prog = *prog;
5616 	env->ops = bpf_verifier_ops[env->prog->type];
5617 
5618 	/* grab the mutex to protect few globals used by verifier */
5619 	mutex_lock(&bpf_verifier_lock);
5620 
5621 	if (attr->log_level || attr->log_buf || attr->log_size) {
5622 		/* user requested verbose verifier output
5623 		 * and supplied buffer to store the verification trace
5624 		 */
5625 		log->level = attr->log_level;
5626 		log->ubuf = (char __user *) (unsigned long) attr->log_buf;
5627 		log->len_total = attr->log_size;
5628 
5629 		ret = -EINVAL;
5630 		/* log attributes have to be sane */
5631 		if (log->len_total < 128 || log->len_total > UINT_MAX >> 8 ||
5632 		    !log->level || !log->ubuf)
5633 			goto err_unlock;
5634 	}
5635 
5636 	env->strict_alignment = !!(attr->prog_flags & BPF_F_STRICT_ALIGNMENT);
5637 	if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS))
5638 		env->strict_alignment = true;
5639 
5640 	if (bpf_prog_is_dev_bound(env->prog->aux)) {
5641 		ret = bpf_prog_offload_verifier_prep(env);
5642 		if (ret)
5643 			goto err_unlock;
5644 	}
5645 
5646 	ret = replace_map_fd_with_map_ptr(env);
5647 	if (ret < 0)
5648 		goto skip_full_check;
5649 
5650 	env->explored_states = kcalloc(env->prog->len,
5651 				       sizeof(struct bpf_verifier_state_list *),
5652 				       GFP_USER);
5653 	ret = -ENOMEM;
5654 	if (!env->explored_states)
5655 		goto skip_full_check;
5656 
5657 	env->allow_ptr_leaks = capable(CAP_SYS_ADMIN);
5658 
5659 	ret = check_cfg(env);
5660 	if (ret < 0)
5661 		goto skip_full_check;
5662 
5663 	ret = do_check(env);
5664 	if (env->cur_state) {
5665 		free_verifier_state(env->cur_state, true);
5666 		env->cur_state = NULL;
5667 	}
5668 
5669 skip_full_check:
5670 	while (!pop_stack(env, NULL, NULL));
5671 	free_states(env);
5672 
5673 	if (ret == 0)
5674 		sanitize_dead_code(env);
5675 
5676 	if (ret == 0)
5677 		ret = check_max_stack_depth(env);
5678 
5679 	if (ret == 0)
5680 		/* program is valid, convert *(u32*)(ctx + off) accesses */
5681 		ret = convert_ctx_accesses(env);
5682 
5683 	if (ret == 0)
5684 		ret = fixup_bpf_calls(env);
5685 
5686 	if (ret == 0)
5687 		ret = fixup_call_args(env);
5688 
5689 	if (log->level && bpf_verifier_log_full(log))
5690 		ret = -ENOSPC;
5691 	if (log->level && !log->ubuf) {
5692 		ret = -EFAULT;
5693 		goto err_release_maps;
5694 	}
5695 
5696 	if (ret == 0 && env->used_map_cnt) {
5697 		/* if program passed verifier, update used_maps in bpf_prog_info */
5698 		env->prog->aux->used_maps = kmalloc_array(env->used_map_cnt,
5699 							  sizeof(env->used_maps[0]),
5700 							  GFP_KERNEL);
5701 
5702 		if (!env->prog->aux->used_maps) {
5703 			ret = -ENOMEM;
5704 			goto err_release_maps;
5705 		}
5706 
5707 		memcpy(env->prog->aux->used_maps, env->used_maps,
5708 		       sizeof(env->used_maps[0]) * env->used_map_cnt);
5709 		env->prog->aux->used_map_cnt = env->used_map_cnt;
5710 
5711 		/* program is valid. Convert pseudo bpf_ld_imm64 into generic
5712 		 * bpf_ld_imm64 instructions
5713 		 */
5714 		convert_pseudo_ld_imm64(env);
5715 	}
5716 
5717 err_release_maps:
5718 	if (!env->prog->aux->used_maps)
5719 		/* if we didn't copy map pointers into bpf_prog_info, release
5720 		 * them now. Otherwise free_bpf_prog_info() will release them.
5721 		 */
5722 		release_maps(env);
5723 	*prog = env->prog;
5724 err_unlock:
5725 	mutex_unlock(&bpf_verifier_lock);
5726 	vfree(env->insn_aux_data);
5727 err_free_env:
5728 	kfree(env);
5729 	return ret;
5730 }
5731