?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
bpf: Add bpf_current_task_under_cgroup helper
[linux/fpc-iii.git] / kernel / bpf / verifier.c
blobd504722ebfa446561789750441df6c0dbc2ae1b7
1 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
2 * Copyright (c) 2016 Facebook
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of version 2 of the GNU General Public
6 * License as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 */
13 #include <linux/kernel.h>
14 #include <linux/types.h>
15 #include <linux/slab.h>
16 #include <linux/bpf.h>
17 #include <linux/filter.h>
18 #include <net/netlink.h>
19 #include <linux/file.h>
20 #include <linux/vmalloc.h>
21
22 /* bpf_check() is a static code analyzer that walks eBPF program
23 * instruction by instruction and updates register/stack state.
24 * All paths of conditional branches are analyzed until 'bpf_exit' insn.
25 *
26 * The first pass is depth-first-search to check that the program is a DAG.
27 * It rejects the following programs:
28 * - larger than BPF_MAXINSNS insns
29 * - if loop is present (detected via back-edge)
30 * - unreachable insns exist (shouldn't be a forest. program = one function)
31 * - out of bounds or malformed jumps
32 * The second pass is all possible path descent from the 1st insn.
33 * Since it's analyzing all pathes through the program, the length of the
34 * analysis is limited to 32k insn, which may be hit even if total number of
35 * insn is less then 4K, but there are too many branches that change stack/regs.
36 * Number of 'branches to be analyzed' is limited to 1k
37 *
38 * On entry to each instruction, each register has a type, and the instruction
39 * changes the types of the registers depending on instruction semantics.
40 * If instruction is BPF_MOV64_REG(BPF_REG_1, BPF_REG_5), then type of R5 is
41 * copied to R1.
42 *
43 * All registers are 64-bit.
44 * R0 - return register
45 * R1-R5 argument passing registers
46 * R6-R9 callee saved registers
47 * R10 - frame pointer read-only
48 *
49 * At the start of BPF program the register R1 contains a pointer to bpf_context
50 * and has type PTR_TO_CTX.
51 *
52 * Verifier tracks arithmetic operations on pointers in case:
53 * BPF_MOV64_REG(BPF_REG_1, BPF_REG_10),
54 * BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -20),
55 * 1st insn copies R10 (which has FRAME_PTR) type into R1
56 * and 2nd arithmetic instruction is pattern matched to recognize
57 * that it wants to construct a pointer to some element within stack.
58 * So after 2nd insn, the register R1 has type PTR_TO_STACK
59 * (and -20 constant is saved for further stack bounds checking).
60 * Meaning that this reg is a pointer to stack plus known immediate constant.
61 *
62 * Most of the time the registers have UNKNOWN_VALUE type, which
63 * means the register has some value, but it's not a valid pointer.
64 * (like pointer plus pointer becomes UNKNOWN_VALUE type)
65 *
66 * When verifier sees load or store instructions the type of base register
67 * can be: PTR_TO_MAP_VALUE, PTR_TO_CTX, FRAME_PTR. These are three pointer
68 * types recognized by check_mem_access() function.
69 *
70 * PTR_TO_MAP_VALUE means that this register is pointing to 'map element value'
71 * and the range of [ptr, ptr + map's value_size) is accessible.
72 *
73 * registers used to pass values to function calls are checked against
74 * function argument constraints.
75 *
76 * ARG_PTR_TO_MAP_KEY is one of such argument constraints.
77 * It means that the register type passed to this function must be
78 * PTR_TO_STACK and it will be used inside the function as
79 * 'pointer to map element key'
80 *
81 * For example the argument constraints for bpf_map_lookup_elem():
82 * .ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL,
83 * .arg1_type = ARG_CONST_MAP_PTR,
84 * .arg2_type = ARG_PTR_TO_MAP_KEY,
85 *
86 * ret_type says that this function returns 'pointer to map elem value or null'
87 * function expects 1st argument to be a const pointer to 'struct bpf_map' and
88 * 2nd argument should be a pointer to stack, which will be used inside
89 * the helper function as a pointer to map element key.
90 *
91 * On the kernel side the helper function looks like:
92 * u64 bpf_map_lookup_elem(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
93 * {
94 * struct bpf_map *map = (struct bpf_map *) (unsigned long) r1;
95 * void *key = (void *) (unsigned long) r2;
96 * void *value;
97 *
98 * here kernel can access 'key' and 'map' pointers safely, knowing that
99 * [key, key + map->key_size) bytes are valid and were initialized on
100 * the stack of eBPF program.
101 * }
102 *
103 * Corresponding eBPF program may look like:
104 * BPF_MOV64_REG(BPF_REG_2, BPF_REG_10), // after this insn R2 type is FRAME_PTR
105 * BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -4), // after this insn R2 type is PTR_TO_STACK
106 * BPF_LD_MAP_FD(BPF_REG_1, map_fd), // after this insn R1 type is CONST_PTR_TO_MAP
107 * BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem),
108 * here verifier looks at prototype of map_lookup_elem() and sees:
109 * .arg1_type == ARG_CONST_MAP_PTR and R1->type == CONST_PTR_TO_MAP, which is ok,
110 * Now verifier knows that this map has key of R1->map_ptr->key_size bytes
111 *
112 * Then .arg2_type == ARG_PTR_TO_MAP_KEY and R2->type == PTR_TO_STACK, ok so far,
113 * Now verifier checks that [R2, R2 + map's key_size) are within stack limits
114 * and were initialized prior to this call.
115 * If it's ok, then verifier allows this BPF_CALL insn and looks at
116 * .ret_type which is RET_PTR_TO_MAP_VALUE_OR_NULL, so it sets
117 * R0->type = PTR_TO_MAP_VALUE_OR_NULL which means bpf_map_lookup_elem() function
118 * returns ether pointer to map value or NULL.
119 *
120 * When type PTR_TO_MAP_VALUE_OR_NULL passes through 'if (reg != 0) goto +off'
121 * insn, the register holding that pointer in the true branch changes state to
122 * PTR_TO_MAP_VALUE and the same register changes state to CONST_IMM in the false
123 * branch. See check_cond_jmp_op().
124 *
125 * After the call R0 is set to return type of the function and registers R1-R5
126 * are set to NOT_INIT to indicate that they are no longer readable.
127 */
128
129 struct reg_state {
130 enum bpf_reg_type type;
131 union {
132 /* valid when type == CONST_IMM | PTR_TO_STACK | UNKNOWN_VALUE */
133 s64 imm;
134
135 /* valid when type == PTR_TO_PACKET* */
136 struct {
137 u32 id;
138 u16 off;
139 u16 range;
140 };
141
142 /* valid when type == CONST_PTR_TO_MAP | PTR_TO_MAP_VALUE |
143 * PTR_TO_MAP_VALUE_OR_NULL
144 */
145 struct bpf_map *map_ptr;
146 };
147 };
148
149 enum bpf_stack_slot_type {
150 STACK_INVALID, /* nothing was stored in this stack slot */
151 STACK_SPILL, /* register spilled into stack */
152 STACK_MISC /* BPF program wrote some data into this slot */
153 };
154
155 #define BPF_REG_SIZE 8 /* size of eBPF register in bytes */
156
157 /* state of the program:
158 * type of all registers and stack info
159 */
160 struct verifier_state {
161 struct reg_state regs[MAX_BPF_REG];
162 u8 stack_slot_type[MAX_BPF_STACK];
163 struct reg_state spilled_regs[MAX_BPF_STACK / BPF_REG_SIZE];
164 };
165
166 /* linked list of verifier states used to prune search */
167 struct verifier_state_list {
168 struct verifier_state state;
169 struct verifier_state_list *next;
170 };
171
172 /* verifier_state + insn_idx are pushed to stack when branch is encountered */
173 struct verifier_stack_elem {
174 /* verifer state is 'st'
175 * before processing instruction 'insn_idx'
176 * and after processing instruction 'prev_insn_idx'
177 */
178 struct verifier_state st;
179 int insn_idx;
180 int prev_insn_idx;
181 struct verifier_stack_elem *next;
182 };
183
184 #define MAX_USED_MAPS 64 /* max number of maps accessed by one eBPF program */
185
186 /* single container for all structs
187 * one verifier_env per bpf_check() call
188 */
189 struct verifier_env {
190 struct bpf_prog *prog; /* eBPF program being verified */
191 struct verifier_stack_elem *head; /* stack of verifier states to be processed */
192 int stack_size; /* number of states to be processed */
193 struct verifier_state cur_state; /* current verifier state */
194 struct verifier_state_list **explored_states; /* search pruning optimization */
195 struct bpf_map *used_maps[MAX_USED_MAPS]; /* array of map's used by eBPF program */
196 u32 used_map_cnt; /* number of used maps */
197 u32 id_gen; /* used to generate unique reg IDs */
198 bool allow_ptr_leaks;
199 };
200
201 #define BPF_COMPLEXITY_LIMIT_INSNS 65536
202 #define BPF_COMPLEXITY_LIMIT_STACK 1024
203
204 struct bpf_call_arg_meta {
205 struct bpf_map *map_ptr;
206 bool raw_mode;
207 int regno;
208 int access_size;
209 };
210
211 /* verbose verifier prints what it's seeing
212 * bpf_check() is called under lock, so no race to access these global vars
213 */
214 static u32 log_level, log_size, log_len;
215 static char *log_buf;
216
217 static DEFINE_MUTEX(bpf_verifier_lock);
218
219 /* log_level controls verbosity level of eBPF verifier.
220 * verbose() is used to dump the verification trace to the log, so the user
221 * can figure out what's wrong with the program
222 */
223 static __printf(1, 2) void verbose(const char *fmt, ...)
224 {
225 va_list args;
226
227 if (log_level == 0 || log_len >= log_size - 1)
228 return;
229
230 va_start(args, fmt);
231 log_len += vscnprintf(log_buf + log_len, log_size - log_len, fmt, args);
232 va_end(args);
233 }
234
235 /* string representation of 'enum bpf_reg_type' */
236 static const char * const reg_type_str[] = {
237 [NOT_INIT] = "?",
238 [UNKNOWN_VALUE] = "inv",
239 [PTR_TO_CTX] = "ctx",
240 [CONST_PTR_TO_MAP] = "map_ptr",
241 [PTR_TO_MAP_VALUE] = "map_value",
242 [PTR_TO_MAP_VALUE_OR_NULL] = "map_value_or_null",
243 [FRAME_PTR] = "fp",
244 [PTR_TO_STACK] = "fp",
245 [CONST_IMM] = "imm",
246 [PTR_TO_PACKET] = "pkt",
247 [PTR_TO_PACKET_END] = "pkt_end",
248 };
249
250 static void print_verifier_state(struct verifier_state *state)
251 {
252 struct reg_state *reg;
253 enum bpf_reg_type t;
254 int i;
255
256 for (i = 0; i < MAX_BPF_REG; i++) {
257 reg = &state->regs[i];
258 t = reg->type;
259 if (t == NOT_INIT)
260 continue;
261 verbose(" R%d=%s", i, reg_type_str[t]);
262 if (t == CONST_IMM || t == PTR_TO_STACK)
263 verbose("%lld", reg->imm);
264 else if (t == PTR_TO_PACKET)
265 verbose("(id=%d,off=%d,r=%d)",
266 reg->id, reg->off, reg->range);
267 else if (t == UNKNOWN_VALUE && reg->imm)
268 verbose("%lld", reg->imm);
269 else if (t == CONST_PTR_TO_MAP || t == PTR_TO_MAP_VALUE ||
270 t == PTR_TO_MAP_VALUE_OR_NULL)
271 verbose("(ks=%d,vs=%d)",
272 reg->map_ptr->key_size,
273 reg->map_ptr->value_size);
274 }
275 for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
276 if (state->stack_slot_type[i] == STACK_SPILL)
277 verbose(" fp%d=%s", -MAX_BPF_STACK + i,
278 reg_type_str[state->spilled_regs[i / BPF_REG_SIZE].type]);
279 }
280 verbose("\n");
281 }
282
283 static const char *const bpf_class_string[] = {
284 [BPF_LD] = "ld",
285 [BPF_LDX] = "ldx",
286 [BPF_ST] = "st",
287 [BPF_STX] = "stx",
288 [BPF_ALU] = "alu",
289 [BPF_JMP] = "jmp",
290 [BPF_RET] = "BUG",
291 [BPF_ALU64] = "alu64",
292 };
293
294 static const char *const bpf_alu_string[16] = {
295 [BPF_ADD >> 4] = "+=",
296 [BPF_SUB >> 4] = "-=",
297 [BPF_MUL >> 4] = "*=",
298 [BPF_DIV >> 4] = "/=",
299 [BPF_OR >> 4] = "|=",
300 [BPF_AND >> 4] = "&=",
301 [BPF_LSH >> 4] = "<<=",
302 [BPF_RSH >> 4] = ">>=",
303 [BPF_NEG >> 4] = "neg",
304 [BPF_MOD >> 4] = "%=",
305 [BPF_XOR >> 4] = "^=",
306 [BPF_MOV >> 4] = "=",
307 [BPF_ARSH >> 4] = "s>>=",
308 [BPF_END >> 4] = "endian",
309 };
310
311 static const char *const bpf_ldst_string[] = {
312 [BPF_W >> 3] = "u32",
313 [BPF_H >> 3] = "u16",
314 [BPF_B >> 3] = "u8",
315 [BPF_DW >> 3] = "u64",
316 };
317
318 static const char *const bpf_jmp_string[16] = {
319 [BPF_JA >> 4] = "jmp",
320 [BPF_JEQ >> 4] = "==",
321 [BPF_JGT >> 4] = ">",
322 [BPF_JGE >> 4] = ">=",
323 [BPF_JSET >> 4] = "&",
324 [BPF_JNE >> 4] = "!=",
325 [BPF_JSGT >> 4] = "s>",
326 [BPF_JSGE >> 4] = "s>=",
327 [BPF_CALL >> 4] = "call",
328 [BPF_EXIT >> 4] = "exit",
329 };
330
331 static void print_bpf_insn(struct bpf_insn *insn)
332 {
333 u8 class = BPF_CLASS(insn->code);
334
335 if (class == BPF_ALU || class == BPF_ALU64) {
336 if (BPF_SRC(insn->code) == BPF_X)
337 verbose("(%02x) %sr%d %s %sr%d\n",
338 insn->code, class == BPF_ALU ? "(u32) " : "",
339 insn->dst_reg,
340 bpf_alu_string[BPF_OP(insn->code) >> 4],
341 class == BPF_ALU ? "(u32) " : "",
342 insn->src_reg);
343 else
344 verbose("(%02x) %sr%d %s %s%d\n",
345 insn->code, class == BPF_ALU ? "(u32) " : "",
346 insn->dst_reg,
347 bpf_alu_string[BPF_OP(insn->code) >> 4],
348 class == BPF_ALU ? "(u32) " : "",
349 insn->imm);
350 } else if (class == BPF_STX) {
351 if (BPF_MODE(insn->code) == BPF_MEM)
352 verbose("(%02x) *(%s *)(r%d %+d) = r%d\n",
353 insn->code,
354 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
355 insn->dst_reg,
356 insn->off, insn->src_reg);
357 else if (BPF_MODE(insn->code) == BPF_XADD)
358 verbose("(%02x) lock *(%s *)(r%d %+d) += r%d\n",
359 insn->code,
360 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
361 insn->dst_reg, insn->off,
362 insn->src_reg);
363 else
364 verbose("BUG_%02x\n", insn->code);
365 } else if (class == BPF_ST) {
366 if (BPF_MODE(insn->code) != BPF_MEM) {
367 verbose("BUG_st_%02x\n", insn->code);
368 return;
369 }
370 verbose("(%02x) *(%s *)(r%d %+d) = %d\n",
371 insn->code,
372 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
373 insn->dst_reg,
374 insn->off, insn->imm);
375 } else if (class == BPF_LDX) {
376 if (BPF_MODE(insn->code) != BPF_MEM) {
377 verbose("BUG_ldx_%02x\n", insn->code);
378 return;
379 }
380 verbose("(%02x) r%d = *(%s *)(r%d %+d)\n",
381 insn->code, insn->dst_reg,
382 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
383 insn->src_reg, insn->off);
384 } else if (class == BPF_LD) {
385 if (BPF_MODE(insn->code) == BPF_ABS) {
386 verbose("(%02x) r0 = *(%s *)skb[%d]\n",
387 insn->code,
388 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
389 insn->imm);
390 } else if (BPF_MODE(insn->code) == BPF_IND) {
391 verbose("(%02x) r0 = *(%s *)skb[r%d + %d]\n",
392 insn->code,
393 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
394 insn->src_reg, insn->imm);
395 } else if (BPF_MODE(insn->code) == BPF_IMM) {
396 verbose("(%02x) r%d = 0x%x\n",
397 insn->code, insn->dst_reg, insn->imm);
398 } else {
399 verbose("BUG_ld_%02x\n", insn->code);
400 return;
401 }
402 } else if (class == BPF_JMP) {
403 u8 opcode = BPF_OP(insn->code);
404
405 if (opcode == BPF_CALL) {
406 verbose("(%02x) call %d\n", insn->code, insn->imm);
407 } else if (insn->code == (BPF_JMP | BPF_JA)) {
408 verbose("(%02x) goto pc%+d\n",
409 insn->code, insn->off);
410 } else if (insn->code == (BPF_JMP | BPF_EXIT)) {
411 verbose("(%02x) exit\n", insn->code);
412 } else if (BPF_SRC(insn->code) == BPF_X) {
413 verbose("(%02x) if r%d %s r%d goto pc%+d\n",
414 insn->code, insn->dst_reg,
415 bpf_jmp_string[BPF_OP(insn->code) >> 4],
416 insn->src_reg, insn->off);
417 } else {
418 verbose("(%02x) if r%d %s 0x%x goto pc%+d\n",
419 insn->code, insn->dst_reg,
420 bpf_jmp_string[BPF_OP(insn->code) >> 4],
421 insn->imm, insn->off);
422 }
423 } else {
424 verbose("(%02x) %s\n", insn->code, bpf_class_string[class]);
425 }
426 }
427
428 static int pop_stack(struct verifier_env *env, int *prev_insn_idx)
429 {
430 struct verifier_stack_elem *elem;
431 int insn_idx;
432
433 if (env->head == NULL)
434 return -1;
435
436 memcpy(&env->cur_state, &env->head->st, sizeof(env->cur_state));
437 insn_idx = env->head->insn_idx;
438 if (prev_insn_idx)
439 *prev_insn_idx = env->head->prev_insn_idx;
440 elem = env->head->next;
441 kfree(env->head);
442 env->head = elem;
443 env->stack_size--;
444 return insn_idx;
445 }
446
447 static struct verifier_state *push_stack(struct verifier_env *env, int insn_idx,
448 int prev_insn_idx)
449 {
450 struct verifier_stack_elem *elem;
451
452 elem = kmalloc(sizeof(struct verifier_stack_elem), GFP_KERNEL);
453 if (!elem)
454 goto err;
455
456 memcpy(&elem->st, &env->cur_state, sizeof(env->cur_state));
457 elem->insn_idx = insn_idx;
458 elem->prev_insn_idx = prev_insn_idx;
459 elem->next = env->head;
460 env->head = elem;
461 env->stack_size++;
462 if (env->stack_size > BPF_COMPLEXITY_LIMIT_STACK) {
463 verbose("BPF program is too complex\n");
464 goto err;
465 }
466 return &elem->st;
467 err:
468 /* pop all elements and return */
469 while (pop_stack(env, NULL) >= 0);
470 return NULL;
471 }
472
473 #define CALLER_SAVED_REGS 6
474 static const int caller_saved[CALLER_SAVED_REGS] = {
475 BPF_REG_0, BPF_REG_1, BPF_REG_2, BPF_REG_3, BPF_REG_4, BPF_REG_5
476 };
477
478 static void init_reg_state(struct reg_state *regs)
479 {
480 int i;
481
482 for (i = 0; i < MAX_BPF_REG; i++) {
483 regs[i].type = NOT_INIT;
484 regs[i].imm = 0;
485 }
486
487 /* frame pointer */
488 regs[BPF_REG_FP].type = FRAME_PTR;
489
490 /* 1st arg to a function */
491 regs[BPF_REG_1].type = PTR_TO_CTX;
492 }
493
494 static void mark_reg_unknown_value(struct reg_state *regs, u32 regno)
495 {
496 BUG_ON(regno >= MAX_BPF_REG);
497 regs[regno].type = UNKNOWN_VALUE;
498 regs[regno].imm = 0;
499 }
500
501 enum reg_arg_type {
502 SRC_OP, /* register is used as source operand */
503 DST_OP, /* register is used as destination operand */
504 DST_OP_NO_MARK /* same as above, check only, don't mark */
505 };
506
507 static int check_reg_arg(struct reg_state *regs, u32 regno,
508 enum reg_arg_type t)
509 {
510 if (regno >= MAX_BPF_REG) {
511 verbose("R%d is invalid\n", regno);
512 return -EINVAL;
513 }
514
515 if (t == SRC_OP) {
516 /* check whether register used as source operand can be read */
517 if (regs[regno].type == NOT_INIT) {
518 verbose("R%d !read_ok\n", regno);
519 return -EACCES;
520 }
521 } else {
522 /* check whether register used as dest operand can be written to */
523 if (regno == BPF_REG_FP) {
524 verbose("frame pointer is read only\n");
525 return -EACCES;
526 }
527 if (t == DST_OP)
528 mark_reg_unknown_value(regs, regno);
529 }
530 return 0;
531 }
532
533 static int bpf_size_to_bytes(int bpf_size)
534 {
535 if (bpf_size == BPF_W)
536 return 4;
537 else if (bpf_size == BPF_H)
538 return 2;
539 else if (bpf_size == BPF_B)
540 return 1;
541 else if (bpf_size == BPF_DW)
542 return 8;
543 else
544 return -EINVAL;
545 }
546
547 static bool is_spillable_regtype(enum bpf_reg_type type)
548 {
549 switch (type) {
550 case PTR_TO_MAP_VALUE:
551 case PTR_TO_MAP_VALUE_OR_NULL:
552 case PTR_TO_STACK:
553 case PTR_TO_CTX:
554 case PTR_TO_PACKET:
555 case PTR_TO_PACKET_END:
556 case FRAME_PTR:
557 case CONST_PTR_TO_MAP:
558 return true;
559 default:
560 return false;
561 }
562 }
563
564 /* check_stack_read/write functions track spill/fill of registers,
565 * stack boundary and alignment are checked in check_mem_access()
566 */
567 static int check_stack_write(struct verifier_state *state, int off, int size,
568 int value_regno)
569 {
570 int i;
571 /* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0,
572 * so it's aligned access and [off, off + size) are within stack limits
573 */
574
575 if (value_regno >= 0 &&
576 is_spillable_regtype(state->regs[value_regno].type)) {
577
578 /* register containing pointer is being spilled into stack */
579 if (size != BPF_REG_SIZE) {
580 verbose("invalid size of register spill\n");
581 return -EACCES;
582 }
583
584 /* save register state */
585 state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] =
586 state->regs[value_regno];
587
588 for (i = 0; i < BPF_REG_SIZE; i++)
589 state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_SPILL;
590 } else {
591 /* regular write of data into stack */
592 state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] =
593 (struct reg_state) {};
594
595 for (i = 0; i < size; i++)
596 state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_MISC;
597 }
598 return 0;
599 }
600
601 static int check_stack_read(struct verifier_state *state, int off, int size,
602 int value_regno)
603 {
604 u8 *slot_type;
605 int i;
606
607 slot_type = &state->stack_slot_type[MAX_BPF_STACK + off];
608
609 if (slot_type[0] == STACK_SPILL) {
610 if (size != BPF_REG_SIZE) {
611 verbose("invalid size of register spill\n");
612 return -EACCES;
613 }
614 for (i = 1; i < BPF_REG_SIZE; i++) {
615 if (slot_type[i] != STACK_SPILL) {
616 verbose("corrupted spill memory\n");
617 return -EACCES;
618 }
619 }
620
621 if (value_regno >= 0)
622 /* restore register state from stack */
623 state->regs[value_regno] =
624 state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE];
625 return 0;
626 } else {
627 for (i = 0; i < size; i++) {
628 if (slot_type[i] != STACK_MISC) {
629 verbose("invalid read from stack off %d+%d size %d\n",
630 off, i, size);
631 return -EACCES;
632 }
633 }
634 if (value_regno >= 0)
635 /* have read misc data from the stack */
636 mark_reg_unknown_value(state->regs, value_regno);
637 return 0;
638 }
639 }
640
641 /* check read/write into map element returned by bpf_map_lookup_elem() */
642 static int check_map_access(struct verifier_env *env, u32 regno, int off,
643 int size)
644 {
645 struct bpf_map *map = env->cur_state.regs[regno].map_ptr;
646
647 if (off < 0 || off + size > map->value_size) {
648 verbose("invalid access to map value, value_size=%d off=%d size=%d\n",
649 map->value_size, off, size);
650 return -EACCES;
651 }
652 return 0;
653 }
654
655 #define MAX_PACKET_OFF 0xffff
656
657 static bool may_write_pkt_data(enum bpf_prog_type type)
658 {
659 switch (type) {
660 case BPF_PROG_TYPE_XDP:
661 return true;
662 default:
663 return false;
664 }
665 }
666
667 static int check_packet_access(struct verifier_env *env, u32 regno, int off,
668 int size)
669 {
670 struct reg_state *regs = env->cur_state.regs;
671 struct reg_state *reg = &regs[regno];
672
673 off += reg->off;
674 if (off < 0 || off + size > reg->range) {
675 verbose("invalid access to packet, off=%d size=%d, R%d(id=%d,off=%d,r=%d)\n",
676 off, size, regno, reg->id, reg->off, reg->range);
677 return -EACCES;
678 }
679 return 0;
680 }
681
682 /* check access to 'struct bpf_context' fields */
683 static int check_ctx_access(struct verifier_env *env, int off, int size,
684 enum bpf_access_type t, enum bpf_reg_type *reg_type)
685 {
686 if (env->prog->aux->ops->is_valid_access &&
687 env->prog->aux->ops->is_valid_access(off, size, t, reg_type)) {
688 /* remember the offset of last byte accessed in ctx */
689 if (env->prog->aux->max_ctx_offset < off + size)
690 env->prog->aux->max_ctx_offset = off + size;
691 return 0;
692 }
693
694 verbose("invalid bpf_context access off=%d size=%d\n", off, size);
695 return -EACCES;
696 }
697
698 static bool is_pointer_value(struct verifier_env *env, int regno)
699 {
700 if (env->allow_ptr_leaks)
701 return false;
702
703 switch (env->cur_state.regs[regno].type) {
704 case UNKNOWN_VALUE:
705 case CONST_IMM:
706 return false;
707 default:
708 return true;
709 }
710 }
711
712 static int check_ptr_alignment(struct verifier_env *env, struct reg_state *reg,
713 int off, int size)
714 {
715 if (reg->type != PTR_TO_PACKET) {
716 if (off % size != 0) {
717 verbose("misaligned access off %d size %d\n", off, size);
718 return -EACCES;
719 } else {
720 return 0;
721 }
722 }
723
724 switch (env->prog->type) {
725 case BPF_PROG_TYPE_SCHED_CLS:
726 case BPF_PROG_TYPE_SCHED_ACT:
727 case BPF_PROG_TYPE_XDP:
728 break;
729 default:
730 verbose("verifier is misconfigured\n");
731 return -EACCES;
732 }
733
734 if (IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS))
735 /* misaligned access to packet is ok on x86,arm,arm64 */
736 return 0;
737
738 if (reg->id && size != 1) {
739 verbose("Unknown packet alignment. Only byte-sized access allowed\n");
740 return -EACCES;
741 }
742
743 /* skb->data is NET_IP_ALIGN-ed */
744 if ((NET_IP_ALIGN + reg->off + off) % size != 0) {
745 verbose("misaligned packet access off %d+%d+%d size %d\n",
746 NET_IP_ALIGN, reg->off, off, size);
747 return -EACCES;
748 }
749 return 0;
750 }
751
752 /* check whether memory at (regno + off) is accessible for t = (read | write)
753 * if t==write, value_regno is a register which value is stored into memory
754 * if t==read, value_regno is a register which will receive the value from memory
755 * if t==write && value_regno==-1, some unknown value is stored into memory
756 * if t==read && value_regno==-1, don't care what we read from memory
757 */
758 static int check_mem_access(struct verifier_env *env, u32 regno, int off,
759 int bpf_size, enum bpf_access_type t,
760 int value_regno)
761 {
762 struct verifier_state *state = &env->cur_state;
763 struct reg_state *reg = &state->regs[regno];
764 int size, err = 0;
765
766 if (reg->type == PTR_TO_STACK)
767 off += reg->imm;
768
769 size = bpf_size_to_bytes(bpf_size);
770 if (size < 0)
771 return size;
772
773 err = check_ptr_alignment(env, reg, off, size);
774 if (err)
775 return err;
776
777 if (reg->type == PTR_TO_MAP_VALUE) {
778 if (t == BPF_WRITE && value_regno >= 0 &&
779 is_pointer_value(env, value_regno)) {
780 verbose("R%d leaks addr into map\n", value_regno);
781 return -EACCES;
782 }
783 err = check_map_access(env, regno, off, size);
784 if (!err && t == BPF_READ && value_regno >= 0)
785 mark_reg_unknown_value(state->regs, value_regno);
786
787 } else if (reg->type == PTR_TO_CTX) {
788 enum bpf_reg_type reg_type = UNKNOWN_VALUE;
789
790 if (t == BPF_WRITE && value_regno >= 0 &&
791 is_pointer_value(env, value_regno)) {
792 verbose("R%d leaks addr into ctx\n", value_regno);
793 return -EACCES;
794 }
795 err = check_ctx_access(env, off, size, t, &reg_type);
796 if (!err && t == BPF_READ && value_regno >= 0) {
797 mark_reg_unknown_value(state->regs, value_regno);
798 if (env->allow_ptr_leaks)
799 /* note that reg.[id|off|range] == 0 */
800 state->regs[value_regno].type = reg_type;
801 }
802
803 } else if (reg->type == FRAME_PTR || reg->type == PTR_TO_STACK) {
804 if (off >= 0 || off < -MAX_BPF_STACK) {
805 verbose("invalid stack off=%d size=%d\n", off, size);
806 return -EACCES;
807 }
808 if (t == BPF_WRITE) {
809 if (!env->allow_ptr_leaks &&
810 state->stack_slot_type[MAX_BPF_STACK + off] == STACK_SPILL &&
811 size != BPF_REG_SIZE) {
812 verbose("attempt to corrupt spilled pointer on stack\n");
813 return -EACCES;
814 }
815 err = check_stack_write(state, off, size, value_regno);
816 } else {
817 err = check_stack_read(state, off, size, value_regno);
818 }
819 } else if (state->regs[regno].type == PTR_TO_PACKET) {
820 if (t == BPF_WRITE && !may_write_pkt_data(env->prog->type)) {
821 verbose("cannot write into packet\n");
822 return -EACCES;
823 }
824 if (t == BPF_WRITE && value_regno >= 0 &&
825 is_pointer_value(env, value_regno)) {
826 verbose("R%d leaks addr into packet\n", value_regno);
827 return -EACCES;
828 }
829 err = check_packet_access(env, regno, off, size);
830 if (!err && t == BPF_READ && value_regno >= 0)
831 mark_reg_unknown_value(state->regs, value_regno);
832 } else {
833 verbose("R%d invalid mem access '%s'\n",
834 regno, reg_type_str[reg->type]);
835 return -EACCES;
836 }
837
838 if (!err && size <= 2 && value_regno >= 0 && env->allow_ptr_leaks &&
839 state->regs[value_regno].type == UNKNOWN_VALUE) {
840 /* 1 or 2 byte load zero-extends, determine the number of
841 * zero upper bits. Not doing it fo 4 byte load, since
842 * such values cannot be added to ptr_to_packet anyway.
843 */
844 state->regs[value_regno].imm = 64 - size * 8;
845 }
846 return err;
847 }
848
849 static int check_xadd(struct verifier_env *env, struct bpf_insn *insn)
850 {
851 struct reg_state *regs = env->cur_state.regs;
852 int err;
853
854 if ((BPF_SIZE(insn->code) != BPF_W && BPF_SIZE(insn->code) != BPF_DW) ||
855 insn->imm != 0) {
856 verbose("BPF_XADD uses reserved fields\n");
857 return -EINVAL;
858 }
859
860 /* check src1 operand */
861 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
862 if (err)
863 return err;
864
865 /* check src2 operand */
866 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
867 if (err)
868 return err;
869
870 /* check whether atomic_add can read the memory */
871 err = check_mem_access(env, insn->dst_reg, insn->off,
872 BPF_SIZE(insn->code), BPF_READ, -1);
873 if (err)
874 return err;
875
876 /* check whether atomic_add can write into the same memory */
877 return check_mem_access(env, insn->dst_reg, insn->off,
878 BPF_SIZE(insn->code), BPF_WRITE, -1);
879 }
880
881 /* when register 'regno' is passed into function that will read 'access_size'
882 * bytes from that pointer, make sure that it's within stack boundary
883 * and all elements of stack are initialized
884 */
885 static int check_stack_boundary(struct verifier_env *env, int regno,
886 int access_size, bool zero_size_allowed,
887 struct bpf_call_arg_meta *meta)
888 {
889 struct verifier_state *state = &env->cur_state;
890 struct reg_state *regs = state->regs;
891 int off, i;
892
893 if (regs[regno].type != PTR_TO_STACK) {
894 if (zero_size_allowed && access_size == 0 &&
895 regs[regno].type == CONST_IMM &&
896 regs[regno].imm == 0)
897 return 0;
898
899 verbose("R%d type=%s expected=%s\n", regno,
900 reg_type_str[regs[regno].type],
901 reg_type_str[PTR_TO_STACK]);
902 return -EACCES;
903 }
904
905 off = regs[regno].imm;
906 if (off >= 0 || off < -MAX_BPF_STACK || off + access_size > 0 ||
907 access_size <= 0) {
908 verbose("invalid stack type R%d off=%d access_size=%d\n",
909 regno, off, access_size);
910 return -EACCES;
911 }
912
913 if (meta && meta->raw_mode) {
914 meta->access_size = access_size;
915 meta->regno = regno;
916 return 0;
917 }
918
919 for (i = 0; i < access_size; i++) {
920 if (state->stack_slot_type[MAX_BPF_STACK + off + i] != STACK_MISC) {
921 verbose("invalid indirect read from stack off %d+%d size %d\n",
922 off, i, access_size);
923 return -EACCES;
924 }
925 }
926 return 0;
927 }
928
929 static int check_func_arg(struct verifier_env *env, u32 regno,
930 enum bpf_arg_type arg_type,
931 struct bpf_call_arg_meta *meta)
932 {
933 struct reg_state *reg = env->cur_state.regs + regno;
934 enum bpf_reg_type expected_type;
935 int err = 0;
936
937 if (arg_type == ARG_DONTCARE)
938 return 0;
939
940 if (reg->type == NOT_INIT) {
941 verbose("R%d !read_ok\n", regno);
942 return -EACCES;
943 }
944
945 if (arg_type == ARG_ANYTHING) {
946 if (is_pointer_value(env, regno)) {
947 verbose("R%d leaks addr into helper function\n", regno);
948 return -EACCES;
949 }
950 return 0;
951 }
952
953 if (arg_type == ARG_PTR_TO_MAP_KEY ||
954 arg_type == ARG_PTR_TO_MAP_VALUE) {
955 expected_type = PTR_TO_STACK;
956 } else if (arg_type == ARG_CONST_STACK_SIZE ||
957 arg_type == ARG_CONST_STACK_SIZE_OR_ZERO) {
958 expected_type = CONST_IMM;
959 } else if (arg_type == ARG_CONST_MAP_PTR) {
960 expected_type = CONST_PTR_TO_MAP;
961 } else if (arg_type == ARG_PTR_TO_CTX) {
962 expected_type = PTR_TO_CTX;
963 } else if (arg_type == ARG_PTR_TO_STACK ||
964 arg_type == ARG_PTR_TO_RAW_STACK) {
965 expected_type = PTR_TO_STACK;
966 /* One exception here. In case function allows for NULL to be
967 * passed in as argument, it's a CONST_IMM type. Final test
968 * happens during stack boundary checking.
969 */
970 if (reg->type == CONST_IMM && reg->imm == 0)
971 expected_type = CONST_IMM;
972 meta->raw_mode = arg_type == ARG_PTR_TO_RAW_STACK;
973 } else {
974 verbose("unsupported arg_type %d\n", arg_type);
975 return -EFAULT;
976 }
977
978 if (reg->type != expected_type) {
979 verbose("R%d type=%s expected=%s\n", regno,
980 reg_type_str[reg->type], reg_type_str[expected_type]);
981 return -EACCES;
982 }
983
984 if (arg_type == ARG_CONST_MAP_PTR) {
985 /* bpf_map_xxx(map_ptr) call: remember that map_ptr */
986 meta->map_ptr = reg->map_ptr;
987 } else if (arg_type == ARG_PTR_TO_MAP_KEY) {
988 /* bpf_map_xxx(..., map_ptr, ..., key) call:
989 * check that [key, key + map->key_size) are within
990 * stack limits and initialized
991 */
992 if (!meta->map_ptr) {
993 /* in function declaration map_ptr must come before
994 * map_key, so that it's verified and known before
995 * we have to check map_key here. Otherwise it means
996 * that kernel subsystem misconfigured verifier
997 */
998 verbose("invalid map_ptr to access map->key\n");
999 return -EACCES;
1000 }
1001 err = check_stack_boundary(env, regno, meta->map_ptr->key_size,
1002 false, NULL);
1003 } else if (arg_type == ARG_PTR_TO_MAP_VALUE) {
1004 /* bpf_map_xxx(..., map_ptr, ..., value) call:
1005 * check [value, value + map->value_size) validity
1006 */
1007 if (!meta->map_ptr) {
1008 /* kernel subsystem misconfigured verifier */
1009 verbose("invalid map_ptr to access map->value\n");
1010 return -EACCES;
1011 }
1012 err = check_stack_boundary(env, regno,
1013 meta->map_ptr->value_size,
1014 false, NULL);
1015 } else if (arg_type == ARG_CONST_STACK_SIZE ||
1016 arg_type == ARG_CONST_STACK_SIZE_OR_ZERO) {
1017 bool zero_size_allowed = (arg_type == ARG_CONST_STACK_SIZE_OR_ZERO);
1018
1019 /* bpf_xxx(..., buf, len) call will access 'len' bytes
1020 * from stack pointer 'buf'. Check it
1021 * note: regno == len, regno - 1 == buf
1022 */
1023 if (regno == 0) {
1024 /* kernel subsystem misconfigured verifier */
1025 verbose("ARG_CONST_STACK_SIZE cannot be first argument\n");
1026 return -EACCES;
1027 }
1028 err = check_stack_boundary(env, regno - 1, reg->imm,
1029 zero_size_allowed, meta);
1030 }
1031
1032 return err;
1033 }
1034
1035 static int check_map_func_compatibility(struct bpf_map *map, int func_id)
1036 {
1037 if (!map)
1038 return 0;
1039
1040 /* We need a two way check, first is from map perspective ... */
1041 switch (map->map_type) {
1042 case BPF_MAP_TYPE_PROG_ARRAY:
1043 if (func_id != BPF_FUNC_tail_call)
1044 goto error;
1045 break;
1046 case BPF_MAP_TYPE_PERF_EVENT_ARRAY:
1047 if (func_id != BPF_FUNC_perf_event_read &&
1048 func_id != BPF_FUNC_perf_event_output)
1049 goto error;
1050 break;
1051 case BPF_MAP_TYPE_STACK_TRACE:
1052 if (func_id != BPF_FUNC_get_stackid)
1053 goto error;
1054 break;
1055 case BPF_MAP_TYPE_CGROUP_ARRAY:
1056 if (func_id != BPF_FUNC_skb_in_cgroup &&
1057 func_id != BPF_FUNC_current_task_under_cgroup)
1058 goto error;
1059 break;
1060 default:
1061 break;
1062 }
1063
1064 /* ... and second from the function itself. */
1065 switch (func_id) {
1066 case BPF_FUNC_tail_call:
1067 if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY)
1068 goto error;
1069 break;
1070 case BPF_FUNC_perf_event_read:
1071 case BPF_FUNC_perf_event_output:
1072 if (map->map_type != BPF_MAP_TYPE_PERF_EVENT_ARRAY)
1073 goto error;
1074 break;
1075 case BPF_FUNC_get_stackid:
1076 if (map->map_type != BPF_MAP_TYPE_STACK_TRACE)
1077 goto error;
1078 break;
1079 case BPF_FUNC_current_task_under_cgroup:
1080 case BPF_FUNC_skb_in_cgroup:
1081 if (map->map_type != BPF_MAP_TYPE_CGROUP_ARRAY)
1082 goto error;
1083 break;
1084 default:
1085 break;
1086 }
1087
1088 return 0;
1089 error:
1090 verbose("cannot pass map_type %d into func %d\n",
1091 map->map_type, func_id);
1092 return -EINVAL;
1093 }
1094
1095 static int check_raw_mode(const struct bpf_func_proto *fn)
1096 {
1097 int count = 0;
1098
1099 if (fn->arg1_type == ARG_PTR_TO_RAW_STACK)
1100 count++;
1101 if (fn->arg2_type == ARG_PTR_TO_RAW_STACK)
1102 count++;
1103 if (fn->arg3_type == ARG_PTR_TO_RAW_STACK)
1104 count++;
1105 if (fn->arg4_type == ARG_PTR_TO_RAW_STACK)
1106 count++;
1107 if (fn->arg5_type == ARG_PTR_TO_RAW_STACK)
1108 count++;
1109
1110 return count > 1 ? -EINVAL : 0;
1111 }
1112
1113 static void clear_all_pkt_pointers(struct verifier_env *env)
1114 {
1115 struct verifier_state *state = &env->cur_state;
1116 struct reg_state *regs = state->regs, *reg;
1117 int i;
1118
1119 for (i = 0; i < MAX_BPF_REG; i++)
1120 if (regs[i].type == PTR_TO_PACKET ||
1121 regs[i].type == PTR_TO_PACKET_END)
1122 mark_reg_unknown_value(regs, i);
1123
1124 for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
1125 if (state->stack_slot_type[i] != STACK_SPILL)
1126 continue;
1127 reg = &state->spilled_regs[i / BPF_REG_SIZE];
1128 if (reg->type != PTR_TO_PACKET &&
1129 reg->type != PTR_TO_PACKET_END)
1130 continue;
1131 reg->type = UNKNOWN_VALUE;
1132 reg->imm = 0;
1133 }
1134 }
1135
1136 static int check_call(struct verifier_env *env, int func_id)
1137 {
1138 struct verifier_state *state = &env->cur_state;
1139 const struct bpf_func_proto *fn = NULL;
1140 struct reg_state *regs = state->regs;
1141 struct reg_state *reg;
1142 struct bpf_call_arg_meta meta;
1143 bool changes_data;
1144 int i, err;
1145
1146 /* find function prototype */
1147 if (func_id < 0 || func_id >= __BPF_FUNC_MAX_ID) {
1148 verbose("invalid func %d\n", func_id);
1149 return -EINVAL;
1150 }
1151
1152 if (env->prog->aux->ops->get_func_proto)
1153 fn = env->prog->aux->ops->get_func_proto(func_id);
1154
1155 if (!fn) {
1156 verbose("unknown func %d\n", func_id);
1157 return -EINVAL;
1158 }
1159
1160 /* eBPF programs must be GPL compatible to use GPL-ed functions */
1161 if (!env->prog->gpl_compatible && fn->gpl_only) {
1162 verbose("cannot call GPL only function from proprietary program\n");
1163 return -EINVAL;
1164 }
1165
1166 changes_data = bpf_helper_changes_skb_data(fn->func);
1167
1168 memset(&meta, 0, sizeof(meta));
1169
1170 /* We only support one arg being in raw mode at the moment, which
1171 * is sufficient for the helper functions we have right now.
1172 */
1173 err = check_raw_mode(fn);
1174 if (err) {
1175 verbose("kernel subsystem misconfigured func %d\n", func_id);
1176 return err;
1177 }
1178
1179 /* check args */
1180 err = check_func_arg(env, BPF_REG_1, fn->arg1_type, &meta);
1181 if (err)
1182 return err;
1183 err = check_func_arg(env, BPF_REG_2, fn->arg2_type, &meta);
1184 if (err)
1185 return err;
1186 err = check_func_arg(env, BPF_REG_3, fn->arg3_type, &meta);
1187 if (err)
1188 return err;
1189 err = check_func_arg(env, BPF_REG_4, fn->arg4_type, &meta);
1190 if (err)
1191 return err;
1192 err = check_func_arg(env, BPF_REG_5, fn->arg5_type, &meta);
1193 if (err)
1194 return err;
1195
1196 /* Mark slots with STACK_MISC in case of raw mode, stack offset
1197 * is inferred from register state.
1198 */
1199 for (i = 0; i < meta.access_size; i++) {
1200 err = check_mem_access(env, meta.regno, i, BPF_B, BPF_WRITE, -1);
1201 if (err)
1202 return err;
1203 }
1204
1205 /* reset caller saved regs */
1206 for (i = 0; i < CALLER_SAVED_REGS; i++) {
1207 reg = regs + caller_saved[i];
1208 reg->type = NOT_INIT;
1209 reg->imm = 0;
1210 }
1211
1212 /* update return register */
1213 if (fn->ret_type == RET_INTEGER) {
1214 regs[BPF_REG_0].type = UNKNOWN_VALUE;
1215 } else if (fn->ret_type == RET_VOID) {
1216 regs[BPF_REG_0].type = NOT_INIT;
1217 } else if (fn->ret_type == RET_PTR_TO_MAP_VALUE_OR_NULL) {
1218 regs[BPF_REG_0].type = PTR_TO_MAP_VALUE_OR_NULL;
1219 /* remember map_ptr, so that check_map_access()
1220 * can check 'value_size' boundary of memory access
1221 * to map element returned from bpf_map_lookup_elem()
1222 */
1223 if (meta.map_ptr == NULL) {
1224 verbose("kernel subsystem misconfigured verifier\n");
1225 return -EINVAL;
1226 }
1227 regs[BPF_REG_0].map_ptr = meta.map_ptr;
1228 } else {
1229 verbose("unknown return type %d of func %d\n",
1230 fn->ret_type, func_id);
1231 return -EINVAL;
1232 }
1233
1234 err = check_map_func_compatibility(meta.map_ptr, func_id);
1235 if (err)
1236 return err;
1237
1238 if (changes_data)
1239 clear_all_pkt_pointers(env);
1240 return 0;
1241 }
1242
1243 static int check_packet_ptr_add(struct verifier_env *env, struct bpf_insn *insn)
1244 {
1245 struct reg_state *regs = env->cur_state.regs;
1246 struct reg_state *dst_reg = &regs[insn->dst_reg];
1247 struct reg_state *src_reg = &regs[insn->src_reg];
1248 struct reg_state tmp_reg;
1249 s32 imm;
1250
1251 if (BPF_SRC(insn->code) == BPF_K) {
1252 /* pkt_ptr += imm */
1253 imm = insn->imm;
1254
1255 add_imm:
1256 if (imm <= 0) {
1257 verbose("addition of negative constant to packet pointer is not allowed\n");
1258 return -EACCES;
1259 }
1260 if (imm >= MAX_PACKET_OFF ||
1261 imm + dst_reg->off >= MAX_PACKET_OFF) {
1262 verbose("constant %d is too large to add to packet pointer\n",
1263 imm);
1264 return -EACCES;
1265 }
1266 /* a constant was added to pkt_ptr.
1267 * Remember it while keeping the same 'id'
1268 */
1269 dst_reg->off += imm;
1270 } else {
1271 if (src_reg->type == PTR_TO_PACKET) {
1272 /* R6=pkt(id=0,off=0,r=62) R7=imm22; r7 += r6 */
1273 tmp_reg = *dst_reg; /* save r7 state */
1274 *dst_reg = *src_reg; /* copy pkt_ptr state r6 into r7 */
1275 src_reg = &tmp_reg; /* pretend it's src_reg state */
1276 /* if the checks below reject it, the copy won't matter,
1277 * since we're rejecting the whole program. If all ok,
1278 * then imm22 state will be added to r7
1279 * and r7 will be pkt(id=0,off=22,r=62) while
1280 * r6 will stay as pkt(id=0,off=0,r=62)
1281 */
1282 }
1283
1284 if (src_reg->type == CONST_IMM) {
1285 /* pkt_ptr += reg where reg is known constant */
1286 imm = src_reg->imm;
1287 goto add_imm;
1288 }
1289 /* disallow pkt_ptr += reg
1290 * if reg is not uknown_value with guaranteed zero upper bits
1291 * otherwise pkt_ptr may overflow and addition will become
1292 * subtraction which is not allowed
1293 */
1294 if (src_reg->type != UNKNOWN_VALUE) {
1295 verbose("cannot add '%s' to ptr_to_packet\n",
1296 reg_type_str[src_reg->type]);
1297 return -EACCES;
1298 }
1299 if (src_reg->imm < 48) {
1300 verbose("cannot add integer value with %lld upper zero bits to ptr_to_packet\n",
1301 src_reg->imm);
1302 return -EACCES;
1303 }
1304 /* dst_reg stays as pkt_ptr type and since some positive
1305 * integer value was added to the pointer, increment its 'id'
1306 */
1307 dst_reg->id = ++env->id_gen;
1308
1309 /* something was added to pkt_ptr, set range and off to zero */
1310 dst_reg->off = 0;
1311 dst_reg->range = 0;
1312 }
1313 return 0;
1314 }
1315
1316 static int evaluate_reg_alu(struct verifier_env *env, struct bpf_insn *insn)
1317 {
1318 struct reg_state *regs = env->cur_state.regs;
1319 struct reg_state *dst_reg = &regs[insn->dst_reg];
1320 u8 opcode = BPF_OP(insn->code);
1321 s64 imm_log2;
1322
1323 /* for type == UNKNOWN_VALUE:
1324 * imm > 0 -> number of zero upper bits
1325 * imm == 0 -> don't track which is the same as all bits can be non-zero
1326 */
1327
1328 if (BPF_SRC(insn->code) == BPF_X) {
1329 struct reg_state *src_reg = &regs[insn->src_reg];
1330
1331 if (src_reg->type == UNKNOWN_VALUE && src_reg->imm > 0 &&
1332 dst_reg->imm && opcode == BPF_ADD) {
1333 /* dreg += sreg
1334 * where both have zero upper bits. Adding them
1335 * can only result making one more bit non-zero
1336 * in the larger value.
1337 * Ex. 0xffff (imm=48) + 1 (imm=63) = 0x10000 (imm=47)
1338 * 0xffff (imm=48) + 0xffff = 0x1fffe (imm=47)
1339 */
1340 dst_reg->imm = min(dst_reg->imm, src_reg->imm);
1341 dst_reg->imm--;
1342 return 0;
1343 }
1344 if (src_reg->type == CONST_IMM && src_reg->imm > 0 &&
1345 dst_reg->imm && opcode == BPF_ADD) {
1346 /* dreg += sreg
1347 * where dreg has zero upper bits and sreg is const.
1348 * Adding them can only result making one more bit
1349 * non-zero in the larger value.
1350 */
1351 imm_log2 = __ilog2_u64((long long)src_reg->imm);
1352 dst_reg->imm = min(dst_reg->imm, 63 - imm_log2);
1353 dst_reg->imm--;
1354 return 0;
1355 }
1356 /* all other cases non supported yet, just mark dst_reg */
1357 dst_reg->imm = 0;
1358 return 0;
1359 }
1360
1361 /* sign extend 32-bit imm into 64-bit to make sure that
1362 * negative values occupy bit 63. Note ilog2() would have
1363 * been incorrect, since sizeof(insn->imm) == 4
1364 */
1365 imm_log2 = __ilog2_u64((long long)insn->imm);
1366
1367 if (dst_reg->imm && opcode == BPF_LSH) {
1368 /* reg <<= imm
1369 * if reg was a result of 2 byte load, then its imm == 48
1370 * which means that upper 48 bits are zero and shifting this reg
1371 * left by 4 would mean that upper 44 bits are still zero
1372 */
1373 dst_reg->imm -= insn->imm;
1374 } else if (dst_reg->imm && opcode == BPF_MUL) {
1375 /* reg *= imm
1376 * if multiplying by 14 subtract 4
1377 * This is conservative calculation of upper zero bits.
1378 * It's not trying to special case insn->imm == 1 or 0 cases
1379 */
1380 dst_reg->imm -= imm_log2 + 1;
1381 } else if (opcode == BPF_AND) {
1382 /* reg &= imm */
1383 dst_reg->imm = 63 - imm_log2;
1384 } else if (dst_reg->imm && opcode == BPF_ADD) {
1385 /* reg += imm */
1386 dst_reg->imm = min(dst_reg->imm, 63 - imm_log2);
1387 dst_reg->imm--;
1388 } else if (opcode == BPF_RSH) {
1389 /* reg >>= imm
1390 * which means that after right shift, upper bits will be zero
1391 * note that verifier already checked that
1392 * 0 <= imm < 64 for shift insn
1393 */
1394 dst_reg->imm += insn->imm;
1395 if (unlikely(dst_reg->imm > 64))
1396 /* some dumb code did:
1397 * r2 = *(u32 *)mem;
1398 * r2 >>= 32;
1399 * and all bits are zero now */
1400 dst_reg->imm = 64;
1401 } else {
1402 /* all other alu ops, means that we don't know what will
1403 * happen to the value, mark it with unknown number of zero bits
1404 */
1405 dst_reg->imm = 0;
1406 }
1407
1408 if (dst_reg->imm < 0) {
1409 /* all 64 bits of the register can contain non-zero bits
1410 * and such value cannot be added to ptr_to_packet, since it
1411 * may overflow, mark it as unknown to avoid further eval
1412 */
1413 dst_reg->imm = 0;
1414 }
1415 return 0;
1416 }
1417
1418 static int evaluate_reg_imm_alu(struct verifier_env *env, struct bpf_insn *insn)
1419 {
1420 struct reg_state *regs = env->cur_state.regs;
1421 struct reg_state *dst_reg = &regs[insn->dst_reg];
1422 struct reg_state *src_reg = &regs[insn->src_reg];
1423 u8 opcode = BPF_OP(insn->code);
1424
1425 /* dst_reg->type == CONST_IMM here, simulate execution of 'add' insn.
1426 * Don't care about overflow or negative values, just add them
1427 */
1428 if (opcode == BPF_ADD && BPF_SRC(insn->code) == BPF_K)
1429 dst_reg->imm += insn->imm;
1430 else if (opcode == BPF_ADD && BPF_SRC(insn->code) == BPF_X &&
1431 src_reg->type == CONST_IMM)
1432 dst_reg->imm += src_reg->imm;
1433 else
1434 mark_reg_unknown_value(regs, insn->dst_reg);
1435 return 0;
1436 }
1437
1438 /* check validity of 32-bit and 64-bit arithmetic operations */
1439 static int check_alu_op(struct verifier_env *env, struct bpf_insn *insn)
1440 {
1441 struct reg_state *regs = env->cur_state.regs, *dst_reg;
1442 u8 opcode = BPF_OP(insn->code);
1443 int err;
1444
1445 if (opcode == BPF_END || opcode == BPF_NEG) {
1446 if (opcode == BPF_NEG) {
1447 if (BPF_SRC(insn->code) != 0 ||
1448 insn->src_reg != BPF_REG_0 ||
1449 insn->off != 0 || insn->imm != 0) {
1450 verbose("BPF_NEG uses reserved fields\n");
1451 return -EINVAL;
1452 }
1453 } else {
1454 if (insn->src_reg != BPF_REG_0 || insn->off != 0 ||
1455 (insn->imm != 16 && insn->imm != 32 && insn->imm != 64)) {
1456 verbose("BPF_END uses reserved fields\n");
1457 return -EINVAL;
1458 }
1459 }
1460
1461 /* check src operand */
1462 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1463 if (err)
1464 return err;
1465
1466 if (is_pointer_value(env, insn->dst_reg)) {
1467 verbose("R%d pointer arithmetic prohibited\n",
1468 insn->dst_reg);
1469 return -EACCES;
1470 }
1471
1472 /* check dest operand */
1473 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1474 if (err)
1475 return err;
1476
1477 } else if (opcode == BPF_MOV) {
1478
1479 if (BPF_SRC(insn->code) == BPF_X) {
1480 if (insn->imm != 0 || insn->off != 0) {
1481 verbose("BPF_MOV uses reserved fields\n");
1482 return -EINVAL;
1483 }
1484
1485 /* check src operand */
1486 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1487 if (err)
1488 return err;
1489 } else {
1490 if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
1491 verbose("BPF_MOV uses reserved fields\n");
1492 return -EINVAL;
1493 }
1494 }
1495
1496 /* check dest operand */
1497 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1498 if (err)
1499 return err;
1500
1501 if (BPF_SRC(insn->code) == BPF_X) {
1502 if (BPF_CLASS(insn->code) == BPF_ALU64) {
1503 /* case: R1 = R2
1504 * copy register state to dest reg
1505 */
1506 regs[insn->dst_reg] = regs[insn->src_reg];
1507 } else {
1508 if (is_pointer_value(env, insn->src_reg)) {
1509 verbose("R%d partial copy of pointer\n",
1510 insn->src_reg);
1511 return -EACCES;
1512 }
1513 regs[insn->dst_reg].type = UNKNOWN_VALUE;
1514 regs[insn->dst_reg].map_ptr = NULL;
1515 }
1516 } else {
1517 /* case: R = imm
1518 * remember the value we stored into this reg
1519 */
1520 regs[insn->dst_reg].type = CONST_IMM;
1521 regs[insn->dst_reg].imm = insn->imm;
1522 }
1523
1524 } else if (opcode > BPF_END) {
1525 verbose("invalid BPF_ALU opcode %x\n", opcode);
1526 return -EINVAL;
1527
1528 } else { /* all other ALU ops: and, sub, xor, add, ... */
1529
1530 if (BPF_SRC(insn->code) == BPF_X) {
1531 if (insn->imm != 0 || insn->off != 0) {
1532 verbose("BPF_ALU uses reserved fields\n");
1533 return -EINVAL;
1534 }
1535 /* check src1 operand */
1536 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1537 if (err)
1538 return err;
1539 } else {
1540 if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
1541 verbose("BPF_ALU uses reserved fields\n");
1542 return -EINVAL;
1543 }
1544 }
1545
1546 /* check src2 operand */
1547 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1548 if (err)
1549 return err;
1550
1551 if ((opcode == BPF_MOD || opcode == BPF_DIV) &&
1552 BPF_SRC(insn->code) == BPF_K && insn->imm == 0) {
1553 verbose("div by zero\n");
1554 return -EINVAL;
1555 }
1556
1557 if ((opcode == BPF_LSH || opcode == BPF_RSH ||
1558 opcode == BPF_ARSH) && BPF_SRC(insn->code) == BPF_K) {
1559 int size = BPF_CLASS(insn->code) == BPF_ALU64 ? 64 : 32;
1560
1561 if (insn->imm < 0 || insn->imm >= size) {
1562 verbose("invalid shift %d\n", insn->imm);
1563 return -EINVAL;
1564 }
1565 }
1566
1567 /* check dest operand */
1568 err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK);
1569 if (err)
1570 return err;
1571
1572 dst_reg = &regs[insn->dst_reg];
1573
1574 /* pattern match 'bpf_add Rx, imm' instruction */
1575 if (opcode == BPF_ADD && BPF_CLASS(insn->code) == BPF_ALU64 &&
1576 dst_reg->type == FRAME_PTR && BPF_SRC(insn->code) == BPF_K) {
1577 dst_reg->type = PTR_TO_STACK;
1578 dst_reg->imm = insn->imm;
1579 return 0;
1580 } else if (opcode == BPF_ADD &&
1581 BPF_CLASS(insn->code) == BPF_ALU64 &&
1582 (dst_reg->type == PTR_TO_PACKET ||
1583 (BPF_SRC(insn->code) == BPF_X &&
1584 regs[insn->src_reg].type == PTR_TO_PACKET))) {
1585 /* ptr_to_packet += K|X */
1586 return check_packet_ptr_add(env, insn);
1587 } else if (BPF_CLASS(insn->code) == BPF_ALU64 &&
1588 dst_reg->type == UNKNOWN_VALUE &&
1589 env->allow_ptr_leaks) {
1590 /* unknown += K|X */
1591 return evaluate_reg_alu(env, insn);
1592 } else if (BPF_CLASS(insn->code) == BPF_ALU64 &&
1593 dst_reg->type == CONST_IMM &&
1594 env->allow_ptr_leaks) {
1595 /* reg_imm += K|X */
1596 return evaluate_reg_imm_alu(env, insn);
1597 } else if (is_pointer_value(env, insn->dst_reg)) {
1598 verbose("R%d pointer arithmetic prohibited\n",
1599 insn->dst_reg);
1600 return -EACCES;
1601 } else if (BPF_SRC(insn->code) == BPF_X &&
1602 is_pointer_value(env, insn->src_reg)) {
1603 verbose("R%d pointer arithmetic prohibited\n",
1604 insn->src_reg);
1605 return -EACCES;
1606 }
1607
1608 /* mark dest operand */
1609 mark_reg_unknown_value(regs, insn->dst_reg);
1610 }
1611
1612 return 0;
1613 }
1614
1615 static void find_good_pkt_pointers(struct verifier_env *env,
1616 struct reg_state *dst_reg)
1617 {
1618 struct verifier_state *state = &env->cur_state;
1619 struct reg_state *regs = state->regs, *reg;
1620 int i;
1621 /* r2 = r3;
1622 * r2 += 8
1623 * if (r2 > pkt_end) goto somewhere
1624 * r2 == dst_reg, pkt_end == src_reg,
1625 * r2=pkt(id=n,off=8,r=0)
1626 * r3=pkt(id=n,off=0,r=0)
1627 * find register r3 and mark its range as r3=pkt(id=n,off=0,r=8)
1628 * so that range of bytes [r3, r3 + 8) is safe to access
1629 */
1630 for (i = 0; i < MAX_BPF_REG; i++)
1631 if (regs[i].type == PTR_TO_PACKET && regs[i].id == dst_reg->id)
1632 regs[i].range = dst_reg->off;
1633
1634 for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
1635 if (state->stack_slot_type[i] != STACK_SPILL)
1636 continue;
1637 reg = &state->spilled_regs[i / BPF_REG_SIZE];
1638 if (reg->type == PTR_TO_PACKET && reg->id == dst_reg->id)
1639 reg->range = dst_reg->off;
1640 }
1641 }
1642
1643 static int check_cond_jmp_op(struct verifier_env *env,
1644 struct bpf_insn *insn, int *insn_idx)
1645 {
1646 struct reg_state *regs = env->cur_state.regs, *dst_reg;
1647 struct verifier_state *other_branch;
1648 u8 opcode = BPF_OP(insn->code);
1649 int err;
1650
1651 if (opcode > BPF_EXIT) {
1652 verbose("invalid BPF_JMP opcode %x\n", opcode);
1653 return -EINVAL;
1654 }
1655
1656 if (BPF_SRC(insn->code) == BPF_X) {
1657 if (insn->imm != 0) {
1658 verbose("BPF_JMP uses reserved fields\n");
1659 return -EINVAL;
1660 }
1661
1662 /* check src1 operand */
1663 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1664 if (err)
1665 return err;
1666
1667 if (is_pointer_value(env, insn->src_reg)) {
1668 verbose("R%d pointer comparison prohibited\n",
1669 insn->src_reg);
1670 return -EACCES;
1671 }
1672 } else {
1673 if (insn->src_reg != BPF_REG_0) {
1674 verbose("BPF_JMP uses reserved fields\n");
1675 return -EINVAL;
1676 }
1677 }
1678
1679 /* check src2 operand */
1680 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1681 if (err)
1682 return err;
1683
1684 dst_reg = &regs[insn->dst_reg];
1685
1686 /* detect if R == 0 where R was initialized to zero earlier */
1687 if (BPF_SRC(insn->code) == BPF_K &&
1688 (opcode == BPF_JEQ || opcode == BPF_JNE) &&
1689 dst_reg->type == CONST_IMM && dst_reg->imm == insn->imm) {
1690 if (opcode == BPF_JEQ) {
1691 /* if (imm == imm) goto pc+off;
1692 * only follow the goto, ignore fall-through
1693 */
1694 *insn_idx += insn->off;
1695 return 0;
1696 } else {
1697 /* if (imm != imm) goto pc+off;
1698 * only follow fall-through branch, since
1699 * that's where the program will go
1700 */
1701 return 0;
1702 }
1703 }
1704
1705 other_branch = push_stack(env, *insn_idx + insn->off + 1, *insn_idx);
1706 if (!other_branch)
1707 return -EFAULT;
1708
1709 /* detect if R == 0 where R is returned value from bpf_map_lookup_elem() */
1710 if (BPF_SRC(insn->code) == BPF_K &&
1711 insn->imm == 0 && (opcode == BPF_JEQ || opcode == BPF_JNE) &&
1712 dst_reg->type == PTR_TO_MAP_VALUE_OR_NULL) {
1713 if (opcode == BPF_JEQ) {
1714 /* next fallthrough insn can access memory via
1715 * this register
1716 */
1717 regs[insn->dst_reg].type = PTR_TO_MAP_VALUE;
1718 /* branch targer cannot access it, since reg == 0 */
1719 mark_reg_unknown_value(other_branch->regs,
1720 insn->dst_reg);
1721 } else {
1722 other_branch->regs[insn->dst_reg].type = PTR_TO_MAP_VALUE;
1723 mark_reg_unknown_value(regs, insn->dst_reg);
1724 }
1725 } else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JGT &&
1726 dst_reg->type == PTR_TO_PACKET &&
1727 regs[insn->src_reg].type == PTR_TO_PACKET_END) {
1728 find_good_pkt_pointers(env, dst_reg);
1729 } else if (is_pointer_value(env, insn->dst_reg)) {
1730 verbose("R%d pointer comparison prohibited\n", insn->dst_reg);
1731 return -EACCES;
1732 }
1733 if (log_level)
1734 print_verifier_state(&env->cur_state);
1735 return 0;
1736 }
1737
1738 /* return the map pointer stored inside BPF_LD_IMM64 instruction */
1739 static struct bpf_map *ld_imm64_to_map_ptr(struct bpf_insn *insn)
1740 {
1741 u64 imm64 = ((u64) (u32) insn[0].imm) | ((u64) (u32) insn[1].imm) << 32;
1742
1743 return (struct bpf_map *) (unsigned long) imm64;
1744 }
1745
1746 /* verify BPF_LD_IMM64 instruction */
1747 static int check_ld_imm(struct verifier_env *env, struct bpf_insn *insn)
1748 {
1749 struct reg_state *regs = env->cur_state.regs;
1750 int err;
1751
1752 if (BPF_SIZE(insn->code) != BPF_DW) {
1753 verbose("invalid BPF_LD_IMM insn\n");
1754 return -EINVAL;
1755 }
1756 if (insn->off != 0) {
1757 verbose("BPF_LD_IMM64 uses reserved fields\n");
1758 return -EINVAL;
1759 }
1760
1761 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1762 if (err)
1763 return err;
1764
1765 if (insn->src_reg == 0)
1766 /* generic move 64-bit immediate into a register */
1767 return 0;
1768
1769 /* replace_map_fd_with_map_ptr() should have caught bad ld_imm64 */
1770 BUG_ON(insn->src_reg != BPF_PSEUDO_MAP_FD);
1771
1772 regs[insn->dst_reg].type = CONST_PTR_TO_MAP;
1773 regs[insn->dst_reg].map_ptr = ld_imm64_to_map_ptr(insn);
1774 return 0;
1775 }
1776
1777 static bool may_access_skb(enum bpf_prog_type type)
1778 {
1779 switch (type) {
1780 case BPF_PROG_TYPE_SOCKET_FILTER:
1781 case BPF_PROG_TYPE_SCHED_CLS:
1782 case BPF_PROG_TYPE_SCHED_ACT:
1783 return true;
1784 default:
1785 return false;
1786 }
1787 }
1788
1789 /* verify safety of LD_ABS|LD_IND instructions:
1790 * - they can only appear in the programs where ctx == skb
1791 * - since they are wrappers of function calls, they scratch R1-R5 registers,
1792 * preserve R6-R9, and store return value into R0
1793 *
1794 * Implicit input:
1795 * ctx == skb == R6 == CTX
1796 *
1797 * Explicit input:
1798 * SRC == any register
1799 * IMM == 32-bit immediate
1800 *
1801 * Output:
1802 * R0 - 8/16/32-bit skb data converted to cpu endianness
1803 */
1804 static int check_ld_abs(struct verifier_env *env, struct bpf_insn *insn)
1805 {
1806 struct reg_state *regs = env->cur_state.regs;
1807 u8 mode = BPF_MODE(insn->code);
1808 struct reg_state *reg;
1809 int i, err;
1810
1811 if (!may_access_skb(env->prog->type)) {
1812 verbose("BPF_LD_[ABS|IND] instructions not allowed for this program type\n");
1813 return -EINVAL;
1814 }
1815
1816 if (insn->dst_reg != BPF_REG_0 || insn->off != 0 ||
1817 BPF_SIZE(insn->code) == BPF_DW ||
1818 (mode == BPF_ABS && insn->src_reg != BPF_REG_0)) {
1819 verbose("BPF_LD_[ABS|IND] uses reserved fields\n");
1820 return -EINVAL;
1821 }
1822
1823 /* check whether implicit source operand (register R6) is readable */
1824 err = check_reg_arg(regs, BPF_REG_6, SRC_OP);
1825 if (err)
1826 return err;
1827
1828 if (regs[BPF_REG_6].type != PTR_TO_CTX) {
1829 verbose("at the time of BPF_LD_ABS|IND R6 != pointer to skb\n");
1830 return -EINVAL;
1831 }
1832
1833 if (mode == BPF_IND) {
1834 /* check explicit source operand */
1835 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1836 if (err)
1837 return err;
1838 }
1839
1840 /* reset caller saved regs to unreadable */
1841 for (i = 0; i < CALLER_SAVED_REGS; i++) {
1842 reg = regs + caller_saved[i];
1843 reg->type = NOT_INIT;
1844 reg->imm = 0;
1845 }
1846
1847 /* mark destination R0 register as readable, since it contains
1848 * the value fetched from the packet
1849 */
1850 regs[BPF_REG_0].type = UNKNOWN_VALUE;
1851 return 0;
1852 }
1853
1854 /* non-recursive DFS pseudo code
1855 * 1 procedure DFS-iterative(G,v):
1856 * 2 label v as discovered
1857 * 3 let S be a stack
1858 * 4 S.push(v)
1859 * 5 while S is not empty
1860 * 6 t <- S.pop()
1861 * 7 if t is what we're looking for:
1862 * 8 return t
1863 * 9 for all edges e in G.adjacentEdges(t) do
1864 * 10 if edge e is already labelled
1865 * 11 continue with the next edge
1866 * 12 w <- G.adjacentVertex(t,e)
1867 * 13 if vertex w is not discovered and not explored
1868 * 14 label e as tree-edge
1869 * 15 label w as discovered
1870 * 16 S.push(w)
1871 * 17 continue at 5
1872 * 18 else if vertex w is discovered
1873 * 19 label e as back-edge
1874 * 20 else
1875 * 21 // vertex w is explored
1876 * 22 label e as forward- or cross-edge
1877 * 23 label t as explored
1878 * 24 S.pop()
1879 *
1880 * convention:
1881 * 0x10 - discovered
1882 * 0x11 - discovered and fall-through edge labelled
1883 * 0x12 - discovered and fall-through and branch edges labelled
1884 * 0x20 - explored
1885 */
1886
1887 enum {
1888 DISCOVERED = 0x10,
1889 EXPLORED = 0x20,
1890 FALLTHROUGH = 1,
1891 BRANCH = 2,
1892 };
1893
1894 #define STATE_LIST_MARK ((struct verifier_state_list *) -1L)
1895
1896 static int *insn_stack; /* stack of insns to process */
1897 static int cur_stack; /* current stack index */
1898 static int *insn_state;
1899
1900 /* t, w, e - match pseudo-code above:
1901 * t - index of current instruction
1902 * w - next instruction
1903 * e - edge
1904 */
1905 static int push_insn(int t, int w, int e, struct verifier_env *env)
1906 {
1907 if (e == FALLTHROUGH && insn_state[t] >= (DISCOVERED | FALLTHROUGH))
1908 return 0;
1909
1910 if (e == BRANCH && insn_state[t] >= (DISCOVERED | BRANCH))
1911 return 0;
1912
1913 if (w < 0 || w >= env->prog->len) {
1914 verbose("jump out of range from insn %d to %d\n", t, w);
1915 return -EINVAL;
1916 }
1917
1918 if (e == BRANCH)
1919 /* mark branch target for state pruning */
1920 env->explored_states[w] = STATE_LIST_MARK;
1921
1922 if (insn_state[w] == 0) {
1923 /* tree-edge */
1924 insn_state[t] = DISCOVERED | e;
1925 insn_state[w] = DISCOVERED;
1926 if (cur_stack >= env->prog->len)
1927 return -E2BIG;
1928 insn_stack[cur_stack++] = w;
1929 return 1;
1930 } else if ((insn_state[w] & 0xF0) == DISCOVERED) {
1931 verbose("back-edge from insn %d to %d\n", t, w);
1932 return -EINVAL;
1933 } else if (insn_state[w] == EXPLORED) {
1934 /* forward- or cross-edge */
1935 insn_state[t] = DISCOVERED | e;
1936 } else {
1937 verbose("insn state internal bug\n");
1938 return -EFAULT;
1939 }
1940 return 0;
1941 }
1942
1943 /* non-recursive depth-first-search to detect loops in BPF program
1944 * loop == back-edge in directed graph
1945 */
1946 static int check_cfg(struct verifier_env *env)
1947 {
1948 struct bpf_insn *insns = env->prog->insnsi;
1949 int insn_cnt = env->prog->len;
1950 int ret = 0;
1951 int i, t;
1952
1953 insn_state = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
1954 if (!insn_state)
1955 return -ENOMEM;
1956
1957 insn_stack = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
1958 if (!insn_stack) {
1959 kfree(insn_state);
1960 return -ENOMEM;
1961 }
1962
1963 insn_state[0] = DISCOVERED; /* mark 1st insn as discovered */
1964 insn_stack[0] = 0; /* 0 is the first instruction */
1965 cur_stack = 1;
1966
1967 peek_stack:
1968 if (cur_stack == 0)
1969 goto check_state;
1970 t = insn_stack[cur_stack - 1];
1971
1972 if (BPF_CLASS(insns[t].code) == BPF_JMP) {
1973 u8 opcode = BPF_OP(insns[t].code);
1974
1975 if (opcode == BPF_EXIT) {
1976 goto mark_explored;
1977 } else if (opcode == BPF_CALL) {
1978 ret = push_insn(t, t + 1, FALLTHROUGH, env);
1979 if (ret == 1)
1980 goto peek_stack;
1981 else if (ret < 0)
1982 goto err_free;
1983 if (t + 1 < insn_cnt)
1984 env->explored_states[t + 1] = STATE_LIST_MARK;
1985 } else if (opcode == BPF_JA) {
1986 if (BPF_SRC(insns[t].code) != BPF_K) {
1987 ret = -EINVAL;
1988 goto err_free;
1989 }
1990 /* unconditional jump with single edge */
1991 ret = push_insn(t, t + insns[t].off + 1,
1992 FALLTHROUGH, env);
1993 if (ret == 1)
1994 goto peek_stack;
1995 else if (ret < 0)
1996 goto err_free;
1997 /* tell verifier to check for equivalent states
1998 * after every call and jump
1999 */
2000 if (t + 1 < insn_cnt)
2001 env->explored_states[t + 1] = STATE_LIST_MARK;
2002 } else {
2003 /* conditional jump with two edges */
2004 ret = push_insn(t, t + 1, FALLTHROUGH, env);
2005 if (ret == 1)
2006 goto peek_stack;
2007 else if (ret < 0)
2008 goto err_free;
2009
2010 ret = push_insn(t, t + insns[t].off + 1, BRANCH, env);
2011 if (ret == 1)
2012 goto peek_stack;
2013 else if (ret < 0)
2014 goto err_free;
2015 }
2016 } else {
2017 /* all other non-branch instructions with single
2018 * fall-through edge
2019 */
2020 ret = push_insn(t, t + 1, FALLTHROUGH, env);
2021 if (ret == 1)
2022 goto peek_stack;
2023 else if (ret < 0)
2024 goto err_free;
2025 }
2026
2027 mark_explored:
2028 insn_state[t] = EXPLORED;
2029 if (cur_stack-- <= 0) {
2030 verbose("pop stack internal bug\n");
2031 ret = -EFAULT;
2032 goto err_free;
2033 }
2034 goto peek_stack;
2035
2036 check_state:
2037 for (i = 0; i < insn_cnt; i++) {
2038 if (insn_state[i] != EXPLORED) {
2039 verbose("unreachable insn %d\n", i);
2040 ret = -EINVAL;
2041 goto err_free;
2042 }
2043 }
2044 ret = 0; /* cfg looks good */
2045
2046 err_free:
2047 kfree(insn_state);
2048 kfree(insn_stack);
2049 return ret;
2050 }
2051
2052 /* the following conditions reduce the number of explored insns
2053 * from ~140k to ~80k for ultra large programs that use a lot of ptr_to_packet
2054 */
2055 static bool compare_ptrs_to_packet(struct reg_state *old, struct reg_state *cur)
2056 {
2057 if (old->id != cur->id)
2058 return false;
2059
2060 /* old ptr_to_packet is more conservative, since it allows smaller
2061 * range. Ex:
2062 * old(off=0,r=10) is equal to cur(off=0,r=20), because
2063 * old(off=0,r=10) means that with range=10 the verifier proceeded
2064 * further and found no issues with the program. Now we're in the same
2065 * spot with cur(off=0,r=20), so we're safe too, since anything further
2066 * will only be looking at most 10 bytes after this pointer.
2067 */
2068 if (old->off == cur->off && old->range < cur->range)
2069 return true;
2070
2071 /* old(off=20,r=10) is equal to cur(off=22,re=22 or 5 or 0)
2072 * since both cannot be used for packet access and safe(old)
2073 * pointer has smaller off that could be used for further
2074 * 'if (ptr > data_end)' check
2075 * Ex:
2076 * old(off=20,r=10) and cur(off=22,r=22) and cur(off=22,r=0) mean
2077 * that we cannot access the packet.
2078 * The safe range is:
2079 * [ptr, ptr + range - off)
2080 * so whenever off >=range, it means no safe bytes from this pointer.
2081 * When comparing old->off <= cur->off, it means that older code
2082 * went with smaller offset and that offset was later
2083 * used to figure out the safe range after 'if (ptr > data_end)' check
2084 * Say, 'old' state was explored like:
2085 * ... R3(off=0, r=0)
2086 * R4 = R3 + 20
2087 * ... now R4(off=20,r=0) <-- here
2088 * if (R4 > data_end)
2089 * ... R4(off=20,r=20), R3(off=0,r=20) and R3 can be used to access.
2090 * ... the code further went all the way to bpf_exit.
2091 * Now the 'cur' state at the mark 'here' has R4(off=30,r=0).
2092 * old_R4(off=20,r=0) equal to cur_R4(off=30,r=0), since if the verifier
2093 * goes further, such cur_R4 will give larger safe packet range after
2094 * 'if (R4 > data_end)' and all further insn were already good with r=20,
2095 * so they will be good with r=30 and we can prune the search.
2096 */
2097 if (old->off <= cur->off &&
2098 old->off >= old->range && cur->off >= cur->range)
2099 return true;
2100
2101 return false;
2102 }
2103
2104 /* compare two verifier states
2105 *
2106 * all states stored in state_list are known to be valid, since
2107 * verifier reached 'bpf_exit' instruction through them
2108 *
2109 * this function is called when verifier exploring different branches of
2110 * execution popped from the state stack. If it sees an old state that has
2111 * more strict register state and more strict stack state then this execution
2112 * branch doesn't need to be explored further, since verifier already
2113 * concluded that more strict state leads to valid finish.
2114 *
2115 * Therefore two states are equivalent if register state is more conservative
2116 * and explored stack state is more conservative than the current one.
2117 * Example:
2118 * explored current
2119 * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC)
2120 * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC)
2121 *
2122 * In other words if current stack state (one being explored) has more
2123 * valid slots than old one that already passed validation, it means
2124 * the verifier can stop exploring and conclude that current state is valid too
2125 *
2126 * Similarly with registers. If explored state has register type as invalid
2127 * whereas register type in current state is meaningful, it means that
2128 * the current state will reach 'bpf_exit' instruction safely
2129 */
2130 static bool states_equal(struct verifier_state *old, struct verifier_state *cur)
2131 {
2132 struct reg_state *rold, *rcur;
2133 int i;
2134
2135 for (i = 0; i < MAX_BPF_REG; i++) {
2136 rold = &old->regs[i];
2137 rcur = &cur->regs[i];
2138
2139 if (memcmp(rold, rcur, sizeof(*rold)) == 0)
2140 continue;
2141
2142 if (rold->type == NOT_INIT ||
2143 (rold->type == UNKNOWN_VALUE && rcur->type != NOT_INIT))
2144 continue;
2145
2146 if (rold->type == PTR_TO_PACKET && rcur->type == PTR_TO_PACKET &&
2147 compare_ptrs_to_packet(rold, rcur))
2148 continue;
2149
2150 return false;
2151 }
2152
2153 for (i = 0; i < MAX_BPF_STACK; i++) {
2154 if (old->stack_slot_type[i] == STACK_INVALID)
2155 continue;
2156 if (old->stack_slot_type[i] != cur->stack_slot_type[i])
2157 /* Ex: old explored (safe) state has STACK_SPILL in
2158 * this stack slot, but current has has STACK_MISC ->
2159 * this verifier states are not equivalent,
2160 * return false to continue verification of this path
2161 */
2162 return false;
2163 if (i % BPF_REG_SIZE)
2164 continue;
2165 if (memcmp(&old->spilled_regs[i / BPF_REG_SIZE],
2166 &cur->spilled_regs[i / BPF_REG_SIZE],
2167 sizeof(old->spilled_regs[0])))
2168 /* when explored and current stack slot types are
2169 * the same, check that stored pointers types
2170 * are the same as well.
2171 * Ex: explored safe path could have stored
2172 * (struct reg_state) {.type = PTR_TO_STACK, .imm = -8}
2173 * but current path has stored:
2174 * (struct reg_state) {.type = PTR_TO_STACK, .imm = -16}
2175 * such verifier states are not equivalent.
2176 * return false to continue verification of this path
2177 */
2178 return false;
2179 else
2180 continue;
2181 }
2182 return true;
2183 }
2184
2185 static int is_state_visited(struct verifier_env *env, int insn_idx)
2186 {
2187 struct verifier_state_list *new_sl;
2188 struct verifier_state_list *sl;
2189
2190 sl = env->explored_states[insn_idx];
2191 if (!sl)
2192 /* this 'insn_idx' instruction wasn't marked, so we will not
2193 * be doing state search here
2194 */
2195 return 0;
2196
2197 while (sl != STATE_LIST_MARK) {
2198 if (states_equal(&sl->state, &env->cur_state))
2199 /* reached equivalent register/stack state,
2200 * prune the search
2201 */
2202 return 1;
2203 sl = sl->next;
2204 }
2205
2206 /* there were no equivalent states, remember current one.
2207 * technically the current state is not proven to be safe yet,
2208 * but it will either reach bpf_exit (which means it's safe) or
2209 * it will be rejected. Since there are no loops, we won't be
2210 * seeing this 'insn_idx' instruction again on the way to bpf_exit
2211 */
2212 new_sl = kmalloc(sizeof(struct verifier_state_list), GFP_USER);
2213 if (!new_sl)
2214 return -ENOMEM;
2215
2216 /* add new state to the head of linked list */
2217 memcpy(&new_sl->state, &env->cur_state, sizeof(env->cur_state));
2218 new_sl->next = env->explored_states[insn_idx];
2219 env->explored_states[insn_idx] = new_sl;
2220 return 0;
2221 }
2222
2223 static int do_check(struct verifier_env *env)
2224 {
2225 struct verifier_state *state = &env->cur_state;
2226 struct bpf_insn *insns = env->prog->insnsi;
2227 struct reg_state *regs = state->regs;
2228 int insn_cnt = env->prog->len;
2229 int insn_idx, prev_insn_idx = 0;
2230 int insn_processed = 0;
2231 bool do_print_state = false;
2232
2233 init_reg_state(regs);
2234 insn_idx = 0;
2235 for (;;) {
2236 struct bpf_insn *insn;
2237 u8 class;
2238 int err;
2239
2240 if (insn_idx >= insn_cnt) {
2241 verbose("invalid insn idx %d insn_cnt %d\n",
2242 insn_idx, insn_cnt);
2243 return -EFAULT;
2244 }
2245
2246 insn = &insns[insn_idx];
2247 class = BPF_CLASS(insn->code);
2248
2249 if (++insn_processed > BPF_COMPLEXITY_LIMIT_INSNS) {
2250 verbose("BPF program is too large. Proccessed %d insn\n",
2251 insn_processed);
2252 return -E2BIG;
2253 }
2254
2255 err = is_state_visited(env, insn_idx);
2256 if (err < 0)
2257 return err;
2258 if (err == 1) {
2259 /* found equivalent state, can prune the search */
2260 if (log_level) {
2261 if (do_print_state)
2262 verbose("\nfrom %d to %d: safe\n",
2263 prev_insn_idx, insn_idx);
2264 else
2265 verbose("%d: safe\n", insn_idx);
2266 }
2267 goto process_bpf_exit;
2268 }
2269
2270 if (log_level && do_print_state) {
2271 verbose("\nfrom %d to %d:", prev_insn_idx, insn_idx);
2272 print_verifier_state(&env->cur_state);
2273 do_print_state = false;
2274 }
2275
2276 if (log_level) {
2277 verbose("%d: ", insn_idx);
2278 print_bpf_insn(insn);
2279 }
2280
2281 if (class == BPF_ALU || class == BPF_ALU64) {
2282 err = check_alu_op(env, insn);
2283 if (err)
2284 return err;
2285
2286 } else if (class == BPF_LDX) {
2287 enum bpf_reg_type src_reg_type;
2288
2289 /* check for reserved fields is already done */
2290
2291 /* check src operand */
2292 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
2293 if (err)
2294 return err;
2295
2296 err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK);
2297 if (err)
2298 return err;
2299
2300 src_reg_type = regs[insn->src_reg].type;
2301
2302 /* check that memory (src_reg + off) is readable,
2303 * the state of dst_reg will be updated by this func
2304 */
2305 err = check_mem_access(env, insn->src_reg, insn->off,
2306 BPF_SIZE(insn->code), BPF_READ,
2307 insn->dst_reg);
2308 if (err)
2309 return err;
2310
2311 if (BPF_SIZE(insn->code) != BPF_W) {
2312 insn_idx++;
2313 continue;
2314 }
2315
2316 if (insn->imm == 0) {
2317 /* saw a valid insn
2318 * dst_reg = *(u32 *)(src_reg + off)
2319 * use reserved 'imm' field to mark this insn
2320 */
2321 insn->imm = src_reg_type;
2322
2323 } else if (src_reg_type != insn->imm &&
2324 (src_reg_type == PTR_TO_CTX ||
2325 insn->imm == PTR_TO_CTX)) {
2326 /* ABuser program is trying to use the same insn
2327 * dst_reg = *(u32*) (src_reg + off)
2328 * with different pointer types:
2329 * src_reg == ctx in one branch and
2330 * src_reg == stack|map in some other branch.
2331 * Reject it.
2332 */
2333 verbose("same insn cannot be used with different pointers\n");
2334 return -EINVAL;
2335 }
2336
2337 } else if (class == BPF_STX) {
2338 enum bpf_reg_type dst_reg_type;
2339
2340 if (BPF_MODE(insn->code) == BPF_XADD) {
2341 err = check_xadd(env, insn);
2342 if (err)
2343 return err;
2344 insn_idx++;
2345 continue;
2346 }
2347
2348 /* check src1 operand */
2349 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
2350 if (err)
2351 return err;
2352 /* check src2 operand */
2353 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
2354 if (err)
2355 return err;
2356
2357 dst_reg_type = regs[insn->dst_reg].type;
2358
2359 /* check that memory (dst_reg + off) is writeable */
2360 err = check_mem_access(env, insn->dst_reg, insn->off,
2361 BPF_SIZE(insn->code), BPF_WRITE,
2362 insn->src_reg);
2363 if (err)
2364 return err;
2365
2366 if (insn->imm == 0) {
2367 insn->imm = dst_reg_type;
2368 } else if (dst_reg_type != insn->imm &&
2369 (dst_reg_type == PTR_TO_CTX ||
2370 insn->imm == PTR_TO_CTX)) {
2371 verbose("same insn cannot be used with different pointers\n");
2372 return -EINVAL;
2373 }
2374
2375 } else if (class == BPF_ST) {
2376 if (BPF_MODE(insn->code) != BPF_MEM ||
2377 insn->src_reg != BPF_REG_0) {
2378 verbose("BPF_ST uses reserved fields\n");
2379 return -EINVAL;
2380 }
2381 /* check src operand */
2382 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
2383 if (err)
2384 return err;
2385
2386 /* check that memory (dst_reg + off) is writeable */
2387 err = check_mem_access(env, insn->dst_reg, insn->off,
2388 BPF_SIZE(insn->code), BPF_WRITE,
2389 -1);
2390 if (err)
2391 return err;
2392
2393 } else if (class == BPF_JMP) {
2394 u8 opcode = BPF_OP(insn->code);
2395
2396 if (opcode == BPF_CALL) {
2397 if (BPF_SRC(insn->code) != BPF_K ||
2398 insn->off != 0 ||
2399 insn->src_reg != BPF_REG_0 ||
2400 insn->dst_reg != BPF_REG_0) {
2401 verbose("BPF_CALL uses reserved fields\n");
2402 return -EINVAL;
2403 }
2404
2405 err = check_call(env, insn->imm);
2406 if (err)
2407 return err;
2408
2409 } else if (opcode == BPF_JA) {
2410 if (BPF_SRC(insn->code) != BPF_K ||
2411 insn->imm != 0 ||
2412 insn->src_reg != BPF_REG_0 ||
2413 insn->dst_reg != BPF_REG_0) {
2414 verbose("BPF_JA uses reserved fields\n");
2415 return -EINVAL;
2416 }
2417
2418 insn_idx += insn->off + 1;
2419 continue;
2420
2421 } else if (opcode == BPF_EXIT) {
2422 if (BPF_SRC(insn->code) != BPF_K ||
2423 insn->imm != 0 ||
2424 insn->src_reg != BPF_REG_0 ||
2425 insn->dst_reg != BPF_REG_0) {
2426 verbose("BPF_EXIT uses reserved fields\n");
2427 return -EINVAL;
2428 }
2429
2430 /* eBPF calling convetion is such that R0 is used
2431 * to return the value from eBPF program.
2432 * Make sure that it's readable at this time
2433 * of bpf_exit, which means that program wrote
2434 * something into it earlier
2435 */
2436 err = check_reg_arg(regs, BPF_REG_0, SRC_OP);
2437 if (err)
2438 return err;
2439
2440 if (is_pointer_value(env, BPF_REG_0)) {
2441 verbose("R0 leaks addr as return value\n");
2442 return -EACCES;
2443 }
2444
2445 process_bpf_exit:
2446 insn_idx = pop_stack(env, &prev_insn_idx);
2447 if (insn_idx < 0) {
2448 break;
2449 } else {
2450 do_print_state = true;
2451 continue;
2452 }
2453 } else {
2454 err = check_cond_jmp_op(env, insn, &insn_idx);
2455 if (err)
2456 return err;
2457 }
2458 } else if (class == BPF_LD) {
2459 u8 mode = BPF_MODE(insn->code);
2460
2461 if (mode == BPF_ABS || mode == BPF_IND) {
2462 err = check_ld_abs(env, insn);
2463 if (err)
2464 return err;
2465
2466 } else if (mode == BPF_IMM) {
2467 err = check_ld_imm(env, insn);
2468 if (err)
2469 return err;
2470
2471 insn_idx++;
2472 } else {
2473 verbose("invalid BPF_LD mode\n");
2474 return -EINVAL;
2475 }
2476 } else {
2477 verbose("unknown insn class %d\n", class);
2478 return -EINVAL;
2479 }
2480
2481 insn_idx++;
2482 }
2483
2484 verbose("processed %d insns\n", insn_processed);
2485 return 0;
2486 }
2487
2488 /* look for pseudo eBPF instructions that access map FDs and
2489 * replace them with actual map pointers
2490 */
2491 static int replace_map_fd_with_map_ptr(struct verifier_env *env)
2492 {
2493 struct bpf_insn *insn = env->prog->insnsi;
2494 int insn_cnt = env->prog->len;
2495 int i, j;
2496
2497 for (i = 0; i < insn_cnt; i++, insn++) {
2498 if (BPF_CLASS(insn->code) == BPF_LDX &&
2499 (BPF_MODE(insn->code) != BPF_MEM || insn->imm != 0)) {
2500 verbose("BPF_LDX uses reserved fields\n");
2501 return -EINVAL;
2502 }
2503
2504 if (BPF_CLASS(insn->code) == BPF_STX &&
2505 ((BPF_MODE(insn->code) != BPF_MEM &&
2506 BPF_MODE(insn->code) != BPF_XADD) || insn->imm != 0)) {
2507 verbose("BPF_STX uses reserved fields\n");
2508 return -EINVAL;
2509 }
2510
2511 if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW)) {
2512 struct bpf_map *map;
2513 struct fd f;
2514
2515 if (i == insn_cnt - 1 || insn[1].code != 0 ||
2516 insn[1].dst_reg != 0 || insn[1].src_reg != 0 ||
2517 insn[1].off != 0) {
2518 verbose("invalid bpf_ld_imm64 insn\n");
2519 return -EINVAL;
2520 }
2521
2522 if (insn->src_reg == 0)
2523 /* valid generic load 64-bit imm */
2524 goto next_insn;
2525
2526 if (insn->src_reg != BPF_PSEUDO_MAP_FD) {
2527 verbose("unrecognized bpf_ld_imm64 insn\n");
2528 return -EINVAL;
2529 }
2530
2531 f = fdget(insn->imm);
2532 map = __bpf_map_get(f);
2533 if (IS_ERR(map)) {
2534 verbose("fd %d is not pointing to valid bpf_map\n",
2535 insn->imm);
2536 return PTR_ERR(map);
2537 }
2538
2539 /* store map pointer inside BPF_LD_IMM64 instruction */
2540 insn[0].imm = (u32) (unsigned long) map;
2541 insn[1].imm = ((u64) (unsigned long) map) >> 32;
2542
2543 /* check whether we recorded this map already */
2544 for (j = 0; j < env->used_map_cnt; j++)
2545 if (env->used_maps[j] == map) {
2546 fdput(f);
2547 goto next_insn;
2548 }
2549
2550 if (env->used_map_cnt >= MAX_USED_MAPS) {
2551 fdput(f);
2552 return -E2BIG;
2553 }
2554
2555 /* hold the map. If the program is rejected by verifier,
2556 * the map will be released by release_maps() or it
2557 * will be used by the valid program until it's unloaded
2558 * and all maps are released in free_bpf_prog_info()
2559 */
2560 map = bpf_map_inc(map, false);
2561 if (IS_ERR(map)) {
2562 fdput(f);
2563 return PTR_ERR(map);
2564 }
2565 env->used_maps[env->used_map_cnt++] = map;
2566
2567 fdput(f);
2568 next_insn:
2569 insn++;
2570 i++;
2571 }
2572 }
2573
2574 /* now all pseudo BPF_LD_IMM64 instructions load valid
2575 * 'struct bpf_map *' into a register instead of user map_fd.
2576 * These pointers will be used later by verifier to validate map access.
2577 */
2578 return 0;
2579 }
2580
2581 /* drop refcnt of maps used by the rejected program */
2582 static void release_maps(struct verifier_env *env)
2583 {
2584 int i;
2585
2586 for (i = 0; i < env->used_map_cnt; i++)
2587 bpf_map_put(env->used_maps[i]);
2588 }
2589
2590 /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */
2591 static void convert_pseudo_ld_imm64(struct verifier_env *env)
2592 {
2593 struct bpf_insn *insn = env->prog->insnsi;
2594 int insn_cnt = env->prog->len;
2595 int i;
2596
2597 for (i = 0; i < insn_cnt; i++, insn++)
2598 if (insn->code == (BPF_LD | BPF_IMM | BPF_DW))
2599 insn->src_reg = 0;
2600 }
2601
2602 /* convert load instructions that access fields of 'struct __sk_buff'
2603 * into sequence of instructions that access fields of 'struct sk_buff'
2604 */
2605 static int convert_ctx_accesses(struct verifier_env *env)
2606 {
2607 struct bpf_insn *insn = env->prog->insnsi;
2608 int insn_cnt = env->prog->len;
2609 struct bpf_insn insn_buf[16];
2610 struct bpf_prog *new_prog;
2611 enum bpf_access_type type;
2612 int i;
2613
2614 if (!env->prog->aux->ops->convert_ctx_access)
2615 return 0;
2616
2617 for (i = 0; i < insn_cnt; i++, insn++) {
2618 u32 insn_delta, cnt;
2619
2620 if (insn->code == (BPF_LDX | BPF_MEM | BPF_W))
2621 type = BPF_READ;
2622 else if (insn->code == (BPF_STX | BPF_MEM | BPF_W))
2623 type = BPF_WRITE;
2624 else
2625 continue;
2626
2627 if (insn->imm != PTR_TO_CTX) {
2628 /* clear internal mark */
2629 insn->imm = 0;
2630 continue;
2631 }
2632
2633 cnt = env->prog->aux->ops->
2634 convert_ctx_access(type, insn->dst_reg, insn->src_reg,
2635 insn->off, insn_buf, env->prog);
2636 if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) {
2637 verbose("bpf verifier is misconfigured\n");
2638 return -EINVAL;
2639 }
2640
2641 new_prog = bpf_patch_insn_single(env->prog, i, insn_buf, cnt);
2642 if (!new_prog)
2643 return -ENOMEM;
2644
2645 insn_delta = cnt - 1;
2646
2647 /* keep walking new program and skip insns we just inserted */
2648 env->prog = new_prog;
2649 insn = new_prog->insnsi + i + insn_delta;
2650
2651 insn_cnt += insn_delta;
2652 i += insn_delta;
2653 }
2654
2655 return 0;
2656 }
2657
2658 static void free_states(struct verifier_env *env)
2659 {
2660 struct verifier_state_list *sl, *sln;
2661 int i;
2662
2663 if (!env->explored_states)
2664 return;
2665
2666 for (i = 0; i < env->prog->len; i++) {
2667 sl = env->explored_states[i];
2668
2669 if (sl)
2670 while (sl != STATE_LIST_MARK) {
2671 sln = sl->next;
2672 kfree(sl);
2673 sl = sln;
2674 }
2675 }
2676
2677 kfree(env->explored_states);
2678 }
2679
2680 int bpf_check(struct bpf_prog **prog, union bpf_attr *attr)
2681 {
2682 char __user *log_ubuf = NULL;
2683 struct verifier_env *env;
2684 int ret = -EINVAL;
2685
2686 if ((*prog)->len <= 0 || (*prog)->len > BPF_MAXINSNS)
2687 return -E2BIG;
2688
2689 /* 'struct verifier_env' can be global, but since it's not small,
2690 * allocate/free it every time bpf_check() is called
2691 */
2692 env = kzalloc(sizeof(struct verifier_env), GFP_KERNEL);
2693 if (!env)
2694 return -ENOMEM;
2695
2696 env->prog = *prog;
2697
2698 /* grab the mutex to protect few globals used by verifier */
2699 mutex_lock(&bpf_verifier_lock);
2700
2701 if (attr->log_level || attr->log_buf || attr->log_size) {
2702 /* user requested verbose verifier output
2703 * and supplied buffer to store the verification trace
2704 */
2705 log_level = attr->log_level;
2706 log_ubuf = (char __user *) (unsigned long) attr->log_buf;
2707 log_size = attr->log_size;
2708 log_len = 0;
2709
2710 ret = -EINVAL;
2711 /* log_* values have to be sane */
2712 if (log_size < 128 || log_size > UINT_MAX >> 8 ||
2713 log_level == 0 || log_ubuf == NULL)
2714 goto free_env;
2715
2716 ret = -ENOMEM;
2717 log_buf = vmalloc(log_size);
2718 if (!log_buf)
2719 goto free_env;
2720 } else {
2721 log_level = 0;
2722 }
2723
2724 ret = replace_map_fd_with_map_ptr(env);
2725 if (ret < 0)
2726 goto skip_full_check;
2727
2728 env->explored_states = kcalloc(env->prog->len,
2729 sizeof(struct verifier_state_list *),
2730 GFP_USER);
2731 ret = -ENOMEM;
2732 if (!env->explored_states)
2733 goto skip_full_check;
2734
2735 ret = check_cfg(env);
2736 if (ret < 0)
2737 goto skip_full_check;
2738
2739 env->allow_ptr_leaks = capable(CAP_SYS_ADMIN);
2740
2741 ret = do_check(env);
2742
2743 skip_full_check:
2744 while (pop_stack(env, NULL) >= 0);
2745 free_states(env);
2746
2747 if (ret == 0)
2748 /* program is valid, convert *(u32*)(ctx + off) accesses */
2749 ret = convert_ctx_accesses(env);
2750
2751 if (log_level && log_len >= log_size - 1) {
2752 BUG_ON(log_len >= log_size);
2753 /* verifier log exceeded user supplied buffer */
2754 ret = -ENOSPC;
2755 /* fall through to return what was recorded */
2756 }
2757
2758 /* copy verifier log back to user space including trailing zero */
2759 if (log_level && copy_to_user(log_ubuf, log_buf, log_len + 1) != 0) {
2760 ret = -EFAULT;
2761 goto free_log_buf;
2762 }
2763
2764 if (ret == 0 && env->used_map_cnt) {
2765 /* if program passed verifier, update used_maps in bpf_prog_info */
2766 env->prog->aux->used_maps = kmalloc_array(env->used_map_cnt,
2767 sizeof(env->used_maps[0]),
2768 GFP_KERNEL);
2769
2770 if (!env->prog->aux->used_maps) {
2771 ret = -ENOMEM;
2772 goto free_log_buf;
2773 }
2774
2775 memcpy(env->prog->aux->used_maps, env->used_maps,
2776 sizeof(env->used_maps[0]) * env->used_map_cnt);
2777 env->prog->aux->used_map_cnt = env->used_map_cnt;
2778
2779 /* program is valid. Convert pseudo bpf_ld_imm64 into generic
2780 * bpf_ld_imm64 instructions
2781 */
2782 convert_pseudo_ld_imm64(env);
2783 }
2784
2785 free_log_buf:
2786 if (log_level)
2787 vfree(log_buf);
2788 free_env:
2789 if (!env->prog->aux->used_maps)
2790 /* if we didn't copy map pointers into bpf_prog_info, release
2791 * them now. Otherwise free_bpf_prog_info() will release them.
2792 */
2793 release_maps(env);
2794 *prog = env->prog;
2795 kfree(env);
2796 mutex_unlock(&bpf_verifier_lock);
2797 return ret;
2798 }
Linux with added FPC-III support