?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
hyperv: Add processing of MTU reduced by the host
[linux/fpc-iii.git] / kernel / bpf / verifier.c
blobb6a1f7c14a67394ab895a2110a7896741aef0064
1 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
2 *
3 * This program is free software; you can redistribute it and/or
4 * modify it under the terms of version 2 of the GNU General Public
5 * License as published by the Free Software Foundation.
6 *
7 * This program is distributed in the hope that it will be useful, but
8 * WITHOUT ANY WARRANTY; without even the implied warranty of
9 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
10 * General Public License for more details.
11 */
12 #include <linux/kernel.h>
13 #include <linux/types.h>
14 #include <linux/slab.h>
15 #include <linux/bpf.h>
16 #include <linux/filter.h>
17 #include <net/netlink.h>
18 #include <linux/file.h>
19 #include <linux/vmalloc.h>
20
21 /* bpf_check() is a static code analyzer that walks eBPF program
22 * instruction by instruction and updates register/stack state.
23 * All paths of conditional branches are analyzed until 'bpf_exit' insn.
24 *
25 * The first pass is depth-first-search to check that the program is a DAG.
26 * It rejects the following programs:
27 * - larger than BPF_MAXINSNS insns
28 * - if loop is present (detected via back-edge)
29 * - unreachable insns exist (shouldn't be a forest. program = one function)
30 * - out of bounds or malformed jumps
31 * The second pass is all possible path descent from the 1st insn.
32 * Since it's analyzing all pathes through the program, the length of the
33 * analysis is limited to 32k insn, which may be hit even if total number of
34 * insn is less then 4K, but there are too many branches that change stack/regs.
35 * Number of 'branches to be analyzed' is limited to 1k
36 *
37 * On entry to each instruction, each register has a type, and the instruction
38 * changes the types of the registers depending on instruction semantics.
39 * If instruction is BPF_MOV64_REG(BPF_REG_1, BPF_REG_5), then type of R5 is
40 * copied to R1.
41 *
42 * All registers are 64-bit.
43 * R0 - return register
44 * R1-R5 argument passing registers
45 * R6-R9 callee saved registers
46 * R10 - frame pointer read-only
47 *
48 * At the start of BPF program the register R1 contains a pointer to bpf_context
49 * and has type PTR_TO_CTX.
50 *
51 * Verifier tracks arithmetic operations on pointers in case:
52 * BPF_MOV64_REG(BPF_REG_1, BPF_REG_10),
53 * BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -20),
54 * 1st insn copies R10 (which has FRAME_PTR) type into R1
55 * and 2nd arithmetic instruction is pattern matched to recognize
56 * that it wants to construct a pointer to some element within stack.
57 * So after 2nd insn, the register R1 has type PTR_TO_STACK
58 * (and -20 constant is saved for further stack bounds checking).
59 * Meaning that this reg is a pointer to stack plus known immediate constant.
60 *
61 * Most of the time the registers have UNKNOWN_VALUE type, which
62 * means the register has some value, but it's not a valid pointer.
63 * (like pointer plus pointer becomes UNKNOWN_VALUE type)
64 *
65 * When verifier sees load or store instructions the type of base register
66 * can be: PTR_TO_MAP_VALUE, PTR_TO_CTX, FRAME_PTR. These are three pointer
67 * types recognized by check_mem_access() function.
68 *
69 * PTR_TO_MAP_VALUE means that this register is pointing to 'map element value'
70 * and the range of [ptr, ptr + map's value_size) is accessible.
71 *
72 * registers used to pass values to function calls are checked against
73 * function argument constraints.
74 *
75 * ARG_PTR_TO_MAP_KEY is one of such argument constraints.
76 * It means that the register type passed to this function must be
77 * PTR_TO_STACK and it will be used inside the function as
78 * 'pointer to map element key'
79 *
80 * For example the argument constraints for bpf_map_lookup_elem():
81 * .ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL,
82 * .arg1_type = ARG_CONST_MAP_PTR,
83 * .arg2_type = ARG_PTR_TO_MAP_KEY,
84 *
85 * ret_type says that this function returns 'pointer to map elem value or null'
86 * function expects 1st argument to be a const pointer to 'struct bpf_map' and
87 * 2nd argument should be a pointer to stack, which will be used inside
88 * the helper function as a pointer to map element key.
89 *
90 * On the kernel side the helper function looks like:
91 * u64 bpf_map_lookup_elem(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
92 * {
93 * struct bpf_map *map = (struct bpf_map *) (unsigned long) r1;
94 * void *key = (void *) (unsigned long) r2;
95 * void *value;
96 *
97 * here kernel can access 'key' and 'map' pointers safely, knowing that
98 * [key, key + map->key_size) bytes are valid and were initialized on
99 * the stack of eBPF program.
100 * }
101 *
102 * Corresponding eBPF program may look like:
103 * BPF_MOV64_REG(BPF_REG_2, BPF_REG_10), // after this insn R2 type is FRAME_PTR
104 * BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -4), // after this insn R2 type is PTR_TO_STACK
105 * BPF_LD_MAP_FD(BPF_REG_1, map_fd), // after this insn R1 type is CONST_PTR_TO_MAP
106 * BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem),
107 * here verifier looks at prototype of map_lookup_elem() and sees:
108 * .arg1_type == ARG_CONST_MAP_PTR and R1->type == CONST_PTR_TO_MAP, which is ok,
109 * Now verifier knows that this map has key of R1->map_ptr->key_size bytes
110 *
111 * Then .arg2_type == ARG_PTR_TO_MAP_KEY and R2->type == PTR_TO_STACK, ok so far,
112 * Now verifier checks that [R2, R2 + map's key_size) are within stack limits
113 * and were initialized prior to this call.
114 * If it's ok, then verifier allows this BPF_CALL insn and looks at
115 * .ret_type which is RET_PTR_TO_MAP_VALUE_OR_NULL, so it sets
116 * R0->type = PTR_TO_MAP_VALUE_OR_NULL which means bpf_map_lookup_elem() function
117 * returns ether pointer to map value or NULL.
118 *
119 * When type PTR_TO_MAP_VALUE_OR_NULL passes through 'if (reg != 0) goto +off'
120 * insn, the register holding that pointer in the true branch changes state to
121 * PTR_TO_MAP_VALUE and the same register changes state to CONST_IMM in the false
122 * branch. See check_cond_jmp_op().
123 *
124 * After the call R0 is set to return type of the function and registers R1-R5
125 * are set to NOT_INIT to indicate that they are no longer readable.
126 */
127
128 /* types of values stored in eBPF registers */
129 enum bpf_reg_type {
130 NOT_INIT = 0, /* nothing was written into register */
131 UNKNOWN_VALUE, /* reg doesn't contain a valid pointer */
132 PTR_TO_CTX, /* reg points to bpf_context */
133 CONST_PTR_TO_MAP, /* reg points to struct bpf_map */
134 PTR_TO_MAP_VALUE, /* reg points to map element value */
135 PTR_TO_MAP_VALUE_OR_NULL,/* points to map elem value or NULL */
136 FRAME_PTR, /* reg == frame_pointer */
137 PTR_TO_STACK, /* reg == frame_pointer + imm */
138 CONST_IMM, /* constant integer value */
139 };
140
141 struct reg_state {
142 enum bpf_reg_type type;
143 union {
144 /* valid when type == CONST_IMM | PTR_TO_STACK */
145 int imm;
146
147 /* valid when type == CONST_PTR_TO_MAP | PTR_TO_MAP_VALUE |
148 * PTR_TO_MAP_VALUE_OR_NULL
149 */
150 struct bpf_map *map_ptr;
151 };
152 };
153
154 enum bpf_stack_slot_type {
155 STACK_INVALID, /* nothing was stored in this stack slot */
156 STACK_SPILL, /* register spilled into stack */
157 STACK_MISC /* BPF program wrote some data into this slot */
158 };
159
160 #define BPF_REG_SIZE 8 /* size of eBPF register in bytes */
161
162 /* state of the program:
163 * type of all registers and stack info
164 */
165 struct verifier_state {
166 struct reg_state regs[MAX_BPF_REG];
167 u8 stack_slot_type[MAX_BPF_STACK];
168 struct reg_state spilled_regs[MAX_BPF_STACK / BPF_REG_SIZE];
169 };
170
171 /* linked list of verifier states used to prune search */
172 struct verifier_state_list {
173 struct verifier_state state;
174 struct verifier_state_list *next;
175 };
176
177 /* verifier_state + insn_idx are pushed to stack when branch is encountered */
178 struct verifier_stack_elem {
179 /* verifer state is 'st'
180 * before processing instruction 'insn_idx'
181 * and after processing instruction 'prev_insn_idx'
182 */
183 struct verifier_state st;
184 int insn_idx;
185 int prev_insn_idx;
186 struct verifier_stack_elem *next;
187 };
188
189 #define MAX_USED_MAPS 64 /* max number of maps accessed by one eBPF program */
190
191 /* single container for all structs
192 * one verifier_env per bpf_check() call
193 */
194 struct verifier_env {
195 struct bpf_prog *prog; /* eBPF program being verified */
196 struct verifier_stack_elem *head; /* stack of verifier states to be processed */
197 int stack_size; /* number of states to be processed */
198 struct verifier_state cur_state; /* current verifier state */
199 struct verifier_state_list **explored_states; /* search pruning optimization */
200 struct bpf_map *used_maps[MAX_USED_MAPS]; /* array of map's used by eBPF program */
201 u32 used_map_cnt; /* number of used maps */
202 };
203
204 /* verbose verifier prints what it's seeing
205 * bpf_check() is called under lock, so no race to access these global vars
206 */
207 static u32 log_level, log_size, log_len;
208 static char *log_buf;
209
210 static DEFINE_MUTEX(bpf_verifier_lock);
211
212 /* log_level controls verbosity level of eBPF verifier.
213 * verbose() is used to dump the verification trace to the log, so the user
214 * can figure out what's wrong with the program
215 */
216 static void verbose(const char *fmt, ...)
217 {
218 va_list args;
219
220 if (log_level == 0 || log_len >= log_size - 1)
221 return;
222
223 va_start(args, fmt);
224 log_len += vscnprintf(log_buf + log_len, log_size - log_len, fmt, args);
225 va_end(args);
226 }
227
228 /* string representation of 'enum bpf_reg_type' */
229 static const char * const reg_type_str[] = {
230 [NOT_INIT] = "?",
231 [UNKNOWN_VALUE] = "inv",
232 [PTR_TO_CTX] = "ctx",
233 [CONST_PTR_TO_MAP] = "map_ptr",
234 [PTR_TO_MAP_VALUE] = "map_value",
235 [PTR_TO_MAP_VALUE_OR_NULL] = "map_value_or_null",
236 [FRAME_PTR] = "fp",
237 [PTR_TO_STACK] = "fp",
238 [CONST_IMM] = "imm",
239 };
240
241 static void print_verifier_state(struct verifier_env *env)
242 {
243 enum bpf_reg_type t;
244 int i;
245
246 for (i = 0; i < MAX_BPF_REG; i++) {
247 t = env->cur_state.regs[i].type;
248 if (t == NOT_INIT)
249 continue;
250 verbose(" R%d=%s", i, reg_type_str[t]);
251 if (t == CONST_IMM || t == PTR_TO_STACK)
252 verbose("%d", env->cur_state.regs[i].imm);
253 else if (t == CONST_PTR_TO_MAP || t == PTR_TO_MAP_VALUE ||
254 t == PTR_TO_MAP_VALUE_OR_NULL)
255 verbose("(ks=%d,vs=%d)",
256 env->cur_state.regs[i].map_ptr->key_size,
257 env->cur_state.regs[i].map_ptr->value_size);
258 }
259 for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
260 if (env->cur_state.stack_slot_type[i] == STACK_SPILL)
261 verbose(" fp%d=%s", -MAX_BPF_STACK + i,
262 reg_type_str[env->cur_state.spilled_regs[i / BPF_REG_SIZE].type]);
263 }
264 verbose("\n");
265 }
266
267 static const char *const bpf_class_string[] = {
268 [BPF_LD] = "ld",
269 [BPF_LDX] = "ldx",
270 [BPF_ST] = "st",
271 [BPF_STX] = "stx",
272 [BPF_ALU] = "alu",
273 [BPF_JMP] = "jmp",
274 [BPF_RET] = "BUG",
275 [BPF_ALU64] = "alu64",
276 };
277
278 static const char *const bpf_alu_string[] = {
279 [BPF_ADD >> 4] = "+=",
280 [BPF_SUB >> 4] = "-=",
281 [BPF_MUL >> 4] = "*=",
282 [BPF_DIV >> 4] = "/=",
283 [BPF_OR >> 4] = "|=",
284 [BPF_AND >> 4] = "&=",
285 [BPF_LSH >> 4] = "<<=",
286 [BPF_RSH >> 4] = ">>=",
287 [BPF_NEG >> 4] = "neg",
288 [BPF_MOD >> 4] = "%=",
289 [BPF_XOR >> 4] = "^=",
290 [BPF_MOV >> 4] = "=",
291 [BPF_ARSH >> 4] = "s>>=",
292 [BPF_END >> 4] = "endian",
293 };
294
295 static const char *const bpf_ldst_string[] = {
296 [BPF_W >> 3] = "u32",
297 [BPF_H >> 3] = "u16",
298 [BPF_B >> 3] = "u8",
299 [BPF_DW >> 3] = "u64",
300 };
301
302 static const char *const bpf_jmp_string[] = {
303 [BPF_JA >> 4] = "jmp",
304 [BPF_JEQ >> 4] = "==",
305 [BPF_JGT >> 4] = ">",
306 [BPF_JGE >> 4] = ">=",
307 [BPF_JSET >> 4] = "&",
308 [BPF_JNE >> 4] = "!=",
309 [BPF_JSGT >> 4] = "s>",
310 [BPF_JSGE >> 4] = "s>=",
311 [BPF_CALL >> 4] = "call",
312 [BPF_EXIT >> 4] = "exit",
313 };
314
315 static void print_bpf_insn(struct bpf_insn *insn)
316 {
317 u8 class = BPF_CLASS(insn->code);
318
319 if (class == BPF_ALU || class == BPF_ALU64) {
320 if (BPF_SRC(insn->code) == BPF_X)
321 verbose("(%02x) %sr%d %s %sr%d\n",
322 insn->code, class == BPF_ALU ? "(u32) " : "",
323 insn->dst_reg,
324 bpf_alu_string[BPF_OP(insn->code) >> 4],
325 class == BPF_ALU ? "(u32) " : "",
326 insn->src_reg);
327 else
328 verbose("(%02x) %sr%d %s %s%d\n",
329 insn->code, class == BPF_ALU ? "(u32) " : "",
330 insn->dst_reg,
331 bpf_alu_string[BPF_OP(insn->code) >> 4],
332 class == BPF_ALU ? "(u32) " : "",
333 insn->imm);
334 } else if (class == BPF_STX) {
335 if (BPF_MODE(insn->code) == BPF_MEM)
336 verbose("(%02x) *(%s *)(r%d %+d) = r%d\n",
337 insn->code,
338 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
339 insn->dst_reg,
340 insn->off, insn->src_reg);
341 else if (BPF_MODE(insn->code) == BPF_XADD)
342 verbose("(%02x) lock *(%s *)(r%d %+d) += r%d\n",
343 insn->code,
344 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
345 insn->dst_reg, insn->off,
346 insn->src_reg);
347 else
348 verbose("BUG_%02x\n", insn->code);
349 } else if (class == BPF_ST) {
350 if (BPF_MODE(insn->code) != BPF_MEM) {
351 verbose("BUG_st_%02x\n", insn->code);
352 return;
353 }
354 verbose("(%02x) *(%s *)(r%d %+d) = %d\n",
355 insn->code,
356 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
357 insn->dst_reg,
358 insn->off, insn->imm);
359 } else if (class == BPF_LDX) {
360 if (BPF_MODE(insn->code) != BPF_MEM) {
361 verbose("BUG_ldx_%02x\n", insn->code);
362 return;
363 }
364 verbose("(%02x) r%d = *(%s *)(r%d %+d)\n",
365 insn->code, insn->dst_reg,
366 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
367 insn->src_reg, insn->off);
368 } else if (class == BPF_LD) {
369 if (BPF_MODE(insn->code) == BPF_ABS) {
370 verbose("(%02x) r0 = *(%s *)skb[%d]\n",
371 insn->code,
372 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
373 insn->imm);
374 } else if (BPF_MODE(insn->code) == BPF_IND) {
375 verbose("(%02x) r0 = *(%s *)skb[r%d + %d]\n",
376 insn->code,
377 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
378 insn->src_reg, insn->imm);
379 } else if (BPF_MODE(insn->code) == BPF_IMM) {
380 verbose("(%02x) r%d = 0x%x\n",
381 insn->code, insn->dst_reg, insn->imm);
382 } else {
383 verbose("BUG_ld_%02x\n", insn->code);
384 return;
385 }
386 } else if (class == BPF_JMP) {
387 u8 opcode = BPF_OP(insn->code);
388
389 if (opcode == BPF_CALL) {
390 verbose("(%02x) call %d\n", insn->code, insn->imm);
391 } else if (insn->code == (BPF_JMP | BPF_JA)) {
392 verbose("(%02x) goto pc%+d\n",
393 insn->code, insn->off);
394 } else if (insn->code == (BPF_JMP | BPF_EXIT)) {
395 verbose("(%02x) exit\n", insn->code);
396 } else if (BPF_SRC(insn->code) == BPF_X) {
397 verbose("(%02x) if r%d %s r%d goto pc%+d\n",
398 insn->code, insn->dst_reg,
399 bpf_jmp_string[BPF_OP(insn->code) >> 4],
400 insn->src_reg, insn->off);
401 } else {
402 verbose("(%02x) if r%d %s 0x%x goto pc%+d\n",
403 insn->code, insn->dst_reg,
404 bpf_jmp_string[BPF_OP(insn->code) >> 4],
405 insn->imm, insn->off);
406 }
407 } else {
408 verbose("(%02x) %s\n", insn->code, bpf_class_string[class]);
409 }
410 }
411
412 static int pop_stack(struct verifier_env *env, int *prev_insn_idx)
413 {
414 struct verifier_stack_elem *elem;
415 int insn_idx;
416
417 if (env->head == NULL)
418 return -1;
419
420 memcpy(&env->cur_state, &env->head->st, sizeof(env->cur_state));
421 insn_idx = env->head->insn_idx;
422 if (prev_insn_idx)
423 *prev_insn_idx = env->head->prev_insn_idx;
424 elem = env->head->next;
425 kfree(env->head);
426 env->head = elem;
427 env->stack_size--;
428 return insn_idx;
429 }
430
431 static struct verifier_state *push_stack(struct verifier_env *env, int insn_idx,
432 int prev_insn_idx)
433 {
434 struct verifier_stack_elem *elem;
435
436 elem = kmalloc(sizeof(struct verifier_stack_elem), GFP_KERNEL);
437 if (!elem)
438 goto err;
439
440 memcpy(&elem->st, &env->cur_state, sizeof(env->cur_state));
441 elem->insn_idx = insn_idx;
442 elem->prev_insn_idx = prev_insn_idx;
443 elem->next = env->head;
444 env->head = elem;
445 env->stack_size++;
446 if (env->stack_size > 1024) {
447 verbose("BPF program is too complex\n");
448 goto err;
449 }
450 return &elem->st;
451 err:
452 /* pop all elements and return */
453 while (pop_stack(env, NULL) >= 0);
454 return NULL;
455 }
456
457 #define CALLER_SAVED_REGS 6
458 static const int caller_saved[CALLER_SAVED_REGS] = {
459 BPF_REG_0, BPF_REG_1, BPF_REG_2, BPF_REG_3, BPF_REG_4, BPF_REG_5
460 };
461
462 static void init_reg_state(struct reg_state *regs)
463 {
464 int i;
465
466 for (i = 0; i < MAX_BPF_REG; i++) {
467 regs[i].type = NOT_INIT;
468 regs[i].imm = 0;
469 regs[i].map_ptr = NULL;
470 }
471
472 /* frame pointer */
473 regs[BPF_REG_FP].type = FRAME_PTR;
474
475 /* 1st arg to a function */
476 regs[BPF_REG_1].type = PTR_TO_CTX;
477 }
478
479 static void mark_reg_unknown_value(struct reg_state *regs, u32 regno)
480 {
481 BUG_ON(regno >= MAX_BPF_REG);
482 regs[regno].type = UNKNOWN_VALUE;
483 regs[regno].imm = 0;
484 regs[regno].map_ptr = NULL;
485 }
486
487 enum reg_arg_type {
488 SRC_OP, /* register is used as source operand */
489 DST_OP, /* register is used as destination operand */
490 DST_OP_NO_MARK /* same as above, check only, don't mark */
491 };
492
493 static int check_reg_arg(struct reg_state *regs, u32 regno,
494 enum reg_arg_type t)
495 {
496 if (regno >= MAX_BPF_REG) {
497 verbose("R%d is invalid\n", regno);
498 return -EINVAL;
499 }
500
501 if (t == SRC_OP) {
502 /* check whether register used as source operand can be read */
503 if (regs[regno].type == NOT_INIT) {
504 verbose("R%d !read_ok\n", regno);
505 return -EACCES;
506 }
507 } else {
508 /* check whether register used as dest operand can be written to */
509 if (regno == BPF_REG_FP) {
510 verbose("frame pointer is read only\n");
511 return -EACCES;
512 }
513 if (t == DST_OP)
514 mark_reg_unknown_value(regs, regno);
515 }
516 return 0;
517 }
518
519 static int bpf_size_to_bytes(int bpf_size)
520 {
521 if (bpf_size == BPF_W)
522 return 4;
523 else if (bpf_size == BPF_H)
524 return 2;
525 else if (bpf_size == BPF_B)
526 return 1;
527 else if (bpf_size == BPF_DW)
528 return 8;
529 else
530 return -EINVAL;
531 }
532
533 /* check_stack_read/write functions track spill/fill of registers,
534 * stack boundary and alignment are checked in check_mem_access()
535 */
536 static int check_stack_write(struct verifier_state *state, int off, int size,
537 int value_regno)
538 {
539 int i;
540 /* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0,
541 * so it's aligned access and [off, off + size) are within stack limits
542 */
543
544 if (value_regno >= 0 &&
545 (state->regs[value_regno].type == PTR_TO_MAP_VALUE ||
546 state->regs[value_regno].type == PTR_TO_STACK ||
547 state->regs[value_regno].type == PTR_TO_CTX)) {
548
549 /* register containing pointer is being spilled into stack */
550 if (size != BPF_REG_SIZE) {
551 verbose("invalid size of register spill\n");
552 return -EACCES;
553 }
554
555 /* save register state */
556 state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] =
557 state->regs[value_regno];
558
559 for (i = 0; i < BPF_REG_SIZE; i++)
560 state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_SPILL;
561 } else {
562 /* regular write of data into stack */
563 state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] =
564 (struct reg_state) {};
565
566 for (i = 0; i < size; i++)
567 state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_MISC;
568 }
569 return 0;
570 }
571
572 static int check_stack_read(struct verifier_state *state, int off, int size,
573 int value_regno)
574 {
575 u8 *slot_type;
576 int i;
577
578 slot_type = &state->stack_slot_type[MAX_BPF_STACK + off];
579
580 if (slot_type[0] == STACK_SPILL) {
581 if (size != BPF_REG_SIZE) {
582 verbose("invalid size of register spill\n");
583 return -EACCES;
584 }
585 for (i = 1; i < BPF_REG_SIZE; i++) {
586 if (slot_type[i] != STACK_SPILL) {
587 verbose("corrupted spill memory\n");
588 return -EACCES;
589 }
590 }
591
592 if (value_regno >= 0)
593 /* restore register state from stack */
594 state->regs[value_regno] =
595 state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE];
596 return 0;
597 } else {
598 for (i = 0; i < size; i++) {
599 if (slot_type[i] != STACK_MISC) {
600 verbose("invalid read from stack off %d+%d size %d\n",
601 off, i, size);
602 return -EACCES;
603 }
604 }
605 if (value_regno >= 0)
606 /* have read misc data from the stack */
607 mark_reg_unknown_value(state->regs, value_regno);
608 return 0;
609 }
610 }
611
612 /* check read/write into map element returned by bpf_map_lookup_elem() */
613 static int check_map_access(struct verifier_env *env, u32 regno, int off,
614 int size)
615 {
616 struct bpf_map *map = env->cur_state.regs[regno].map_ptr;
617
618 if (off < 0 || off + size > map->value_size) {
619 verbose("invalid access to map value, value_size=%d off=%d size=%d\n",
620 map->value_size, off, size);
621 return -EACCES;
622 }
623 return 0;
624 }
625
626 /* check access to 'struct bpf_context' fields */
627 static int check_ctx_access(struct verifier_env *env, int off, int size,
628 enum bpf_access_type t)
629 {
630 if (env->prog->aux->ops->is_valid_access &&
631 env->prog->aux->ops->is_valid_access(off, size, t))
632 return 0;
633
634 verbose("invalid bpf_context access off=%d size=%d\n", off, size);
635 return -EACCES;
636 }
637
638 /* check whether memory at (regno + off) is accessible for t = (read | write)
639 * if t==write, value_regno is a register which value is stored into memory
640 * if t==read, value_regno is a register which will receive the value from memory
641 * if t==write && value_regno==-1, some unknown value is stored into memory
642 * if t==read && value_regno==-1, don't care what we read from memory
643 */
644 static int check_mem_access(struct verifier_env *env, u32 regno, int off,
645 int bpf_size, enum bpf_access_type t,
646 int value_regno)
647 {
648 struct verifier_state *state = &env->cur_state;
649 int size, err = 0;
650
651 size = bpf_size_to_bytes(bpf_size);
652 if (size < 0)
653 return size;
654
655 if (off % size != 0) {
656 verbose("misaligned access off %d size %d\n", off, size);
657 return -EACCES;
658 }
659
660 if (state->regs[regno].type == PTR_TO_MAP_VALUE) {
661 err = check_map_access(env, regno, off, size);
662 if (!err && t == BPF_READ && value_regno >= 0)
663 mark_reg_unknown_value(state->regs, value_regno);
664
665 } else if (state->regs[regno].type == PTR_TO_CTX) {
666 err = check_ctx_access(env, off, size, t);
667 if (!err && t == BPF_READ && value_regno >= 0)
668 mark_reg_unknown_value(state->regs, value_regno);
669
670 } else if (state->regs[regno].type == FRAME_PTR) {
671 if (off >= 0 || off < -MAX_BPF_STACK) {
672 verbose("invalid stack off=%d size=%d\n", off, size);
673 return -EACCES;
674 }
675 if (t == BPF_WRITE)
676 err = check_stack_write(state, off, size, value_regno);
677 else
678 err = check_stack_read(state, off, size, value_regno);
679 } else {
680 verbose("R%d invalid mem access '%s'\n",
681 regno, reg_type_str[state->regs[regno].type]);
682 return -EACCES;
683 }
684 return err;
685 }
686
687 static int check_xadd(struct verifier_env *env, struct bpf_insn *insn)
688 {
689 struct reg_state *regs = env->cur_state.regs;
690 int err;
691
692 if ((BPF_SIZE(insn->code) != BPF_W && BPF_SIZE(insn->code) != BPF_DW) ||
693 insn->imm != 0) {
694 verbose("BPF_XADD uses reserved fields\n");
695 return -EINVAL;
696 }
697
698 /* check src1 operand */
699 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
700 if (err)
701 return err;
702
703 /* check src2 operand */
704 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
705 if (err)
706 return err;
707
708 /* check whether atomic_add can read the memory */
709 err = check_mem_access(env, insn->dst_reg, insn->off,
710 BPF_SIZE(insn->code), BPF_READ, -1);
711 if (err)
712 return err;
713
714 /* check whether atomic_add can write into the same memory */
715 return check_mem_access(env, insn->dst_reg, insn->off,
716 BPF_SIZE(insn->code), BPF_WRITE, -1);
717 }
718
719 /* when register 'regno' is passed into function that will read 'access_size'
720 * bytes from that pointer, make sure that it's within stack boundary
721 * and all elements of stack are initialized
722 */
723 static int check_stack_boundary(struct verifier_env *env,
724 int regno, int access_size)
725 {
726 struct verifier_state *state = &env->cur_state;
727 struct reg_state *regs = state->regs;
728 int off, i;
729
730 if (regs[regno].type != PTR_TO_STACK)
731 return -EACCES;
732
733 off = regs[regno].imm;
734 if (off >= 0 || off < -MAX_BPF_STACK || off + access_size > 0 ||
735 access_size <= 0) {
736 verbose("invalid stack type R%d off=%d access_size=%d\n",
737 regno, off, access_size);
738 return -EACCES;
739 }
740
741 for (i = 0; i < access_size; i++) {
742 if (state->stack_slot_type[MAX_BPF_STACK + off + i] != STACK_MISC) {
743 verbose("invalid indirect read from stack off %d+%d size %d\n",
744 off, i, access_size);
745 return -EACCES;
746 }
747 }
748 return 0;
749 }
750
751 static int check_func_arg(struct verifier_env *env, u32 regno,
752 enum bpf_arg_type arg_type, struct bpf_map **mapp)
753 {
754 struct reg_state *reg = env->cur_state.regs + regno;
755 enum bpf_reg_type expected_type;
756 int err = 0;
757
758 if (arg_type == ARG_ANYTHING)
759 return 0;
760
761 if (reg->type == NOT_INIT) {
762 verbose("R%d !read_ok\n", regno);
763 return -EACCES;
764 }
765
766 if (arg_type == ARG_PTR_TO_STACK || arg_type == ARG_PTR_TO_MAP_KEY ||
767 arg_type == ARG_PTR_TO_MAP_VALUE) {
768 expected_type = PTR_TO_STACK;
769 } else if (arg_type == ARG_CONST_STACK_SIZE) {
770 expected_type = CONST_IMM;
771 } else if (arg_type == ARG_CONST_MAP_PTR) {
772 expected_type = CONST_PTR_TO_MAP;
773 } else {
774 verbose("unsupported arg_type %d\n", arg_type);
775 return -EFAULT;
776 }
777
778 if (reg->type != expected_type) {
779 verbose("R%d type=%s expected=%s\n", regno,
780 reg_type_str[reg->type], reg_type_str[expected_type]);
781 return -EACCES;
782 }
783
784 if (arg_type == ARG_CONST_MAP_PTR) {
785 /* bpf_map_xxx(map_ptr) call: remember that map_ptr */
786 *mapp = reg->map_ptr;
787
788 } else if (arg_type == ARG_PTR_TO_MAP_KEY) {
789 /* bpf_map_xxx(..., map_ptr, ..., key) call:
790 * check that [key, key + map->key_size) are within
791 * stack limits and initialized
792 */
793 if (!*mapp) {
794 /* in function declaration map_ptr must come before
795 * map_key, so that it's verified and known before
796 * we have to check map_key here. Otherwise it means
797 * that kernel subsystem misconfigured verifier
798 */
799 verbose("invalid map_ptr to access map->key\n");
800 return -EACCES;
801 }
802 err = check_stack_boundary(env, regno, (*mapp)->key_size);
803
804 } else if (arg_type == ARG_PTR_TO_MAP_VALUE) {
805 /* bpf_map_xxx(..., map_ptr, ..., value) call:
806 * check [value, value + map->value_size) validity
807 */
808 if (!*mapp) {
809 /* kernel subsystem misconfigured verifier */
810 verbose("invalid map_ptr to access map->value\n");
811 return -EACCES;
812 }
813 err = check_stack_boundary(env, regno, (*mapp)->value_size);
814
815 } else if (arg_type == ARG_CONST_STACK_SIZE) {
816 /* bpf_xxx(..., buf, len) call will access 'len' bytes
817 * from stack pointer 'buf'. Check it
818 * note: regno == len, regno - 1 == buf
819 */
820 if (regno == 0) {
821 /* kernel subsystem misconfigured verifier */
822 verbose("ARG_CONST_STACK_SIZE cannot be first argument\n");
823 return -EACCES;
824 }
825 err = check_stack_boundary(env, regno - 1, reg->imm);
826 }
827
828 return err;
829 }
830
831 static int check_call(struct verifier_env *env, int func_id)
832 {
833 struct verifier_state *state = &env->cur_state;
834 const struct bpf_func_proto *fn = NULL;
835 struct reg_state *regs = state->regs;
836 struct bpf_map *map = NULL;
837 struct reg_state *reg;
838 int i, err;
839
840 /* find function prototype */
841 if (func_id < 0 || func_id >= __BPF_FUNC_MAX_ID) {
842 verbose("invalid func %d\n", func_id);
843 return -EINVAL;
844 }
845
846 if (env->prog->aux->ops->get_func_proto)
847 fn = env->prog->aux->ops->get_func_proto(func_id);
848
849 if (!fn) {
850 verbose("unknown func %d\n", func_id);
851 return -EINVAL;
852 }
853
854 /* eBPF programs must be GPL compatible to use GPL-ed functions */
855 if (!env->prog->aux->is_gpl_compatible && fn->gpl_only) {
856 verbose("cannot call GPL only function from proprietary program\n");
857 return -EINVAL;
858 }
859
860 /* check args */
861 err = check_func_arg(env, BPF_REG_1, fn->arg1_type, &map);
862 if (err)
863 return err;
864 err = check_func_arg(env, BPF_REG_2, fn->arg2_type, &map);
865 if (err)
866 return err;
867 err = check_func_arg(env, BPF_REG_3, fn->arg3_type, &map);
868 if (err)
869 return err;
870 err = check_func_arg(env, BPF_REG_4, fn->arg4_type, &map);
871 if (err)
872 return err;
873 err = check_func_arg(env, BPF_REG_5, fn->arg5_type, &map);
874 if (err)
875 return err;
876
877 /* reset caller saved regs */
878 for (i = 0; i < CALLER_SAVED_REGS; i++) {
879 reg = regs + caller_saved[i];
880 reg->type = NOT_INIT;
881 reg->imm = 0;
882 }
883
884 /* update return register */
885 if (fn->ret_type == RET_INTEGER) {
886 regs[BPF_REG_0].type = UNKNOWN_VALUE;
887 } else if (fn->ret_type == RET_VOID) {
888 regs[BPF_REG_0].type = NOT_INIT;
889 } else if (fn->ret_type == RET_PTR_TO_MAP_VALUE_OR_NULL) {
890 regs[BPF_REG_0].type = PTR_TO_MAP_VALUE_OR_NULL;
891 /* remember map_ptr, so that check_map_access()
892 * can check 'value_size' boundary of memory access
893 * to map element returned from bpf_map_lookup_elem()
894 */
895 if (map == NULL) {
896 verbose("kernel subsystem misconfigured verifier\n");
897 return -EINVAL;
898 }
899 regs[BPF_REG_0].map_ptr = map;
900 } else {
901 verbose("unknown return type %d of func %d\n",
902 fn->ret_type, func_id);
903 return -EINVAL;
904 }
905 return 0;
906 }
907
908 /* check validity of 32-bit and 64-bit arithmetic operations */
909 static int check_alu_op(struct reg_state *regs, struct bpf_insn *insn)
910 {
911 u8 opcode = BPF_OP(insn->code);
912 int err;
913
914 if (opcode == BPF_END || opcode == BPF_NEG) {
915 if (opcode == BPF_NEG) {
916 if (BPF_SRC(insn->code) != 0 ||
917 insn->src_reg != BPF_REG_0 ||
918 insn->off != 0 || insn->imm != 0) {
919 verbose("BPF_NEG uses reserved fields\n");
920 return -EINVAL;
921 }
922 } else {
923 if (insn->src_reg != BPF_REG_0 || insn->off != 0 ||
924 (insn->imm != 16 && insn->imm != 32 && insn->imm != 64)) {
925 verbose("BPF_END uses reserved fields\n");
926 return -EINVAL;
927 }
928 }
929
930 /* check src operand */
931 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
932 if (err)
933 return err;
934
935 /* check dest operand */
936 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
937 if (err)
938 return err;
939
940 } else if (opcode == BPF_MOV) {
941
942 if (BPF_SRC(insn->code) == BPF_X) {
943 if (insn->imm != 0 || insn->off != 0) {
944 verbose("BPF_MOV uses reserved fields\n");
945 return -EINVAL;
946 }
947
948 /* check src operand */
949 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
950 if (err)
951 return err;
952 } else {
953 if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
954 verbose("BPF_MOV uses reserved fields\n");
955 return -EINVAL;
956 }
957 }
958
959 /* check dest operand */
960 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
961 if (err)
962 return err;
963
964 if (BPF_SRC(insn->code) == BPF_X) {
965 if (BPF_CLASS(insn->code) == BPF_ALU64) {
966 /* case: R1 = R2
967 * copy register state to dest reg
968 */
969 regs[insn->dst_reg] = regs[insn->src_reg];
970 } else {
971 regs[insn->dst_reg].type = UNKNOWN_VALUE;
972 regs[insn->dst_reg].map_ptr = NULL;
973 }
974 } else {
975 /* case: R = imm
976 * remember the value we stored into this reg
977 */
978 regs[insn->dst_reg].type = CONST_IMM;
979 regs[insn->dst_reg].imm = insn->imm;
980 }
981
982 } else if (opcode > BPF_END) {
983 verbose("invalid BPF_ALU opcode %x\n", opcode);
984 return -EINVAL;
985
986 } else { /* all other ALU ops: and, sub, xor, add, ... */
987
988 bool stack_relative = false;
989
990 if (BPF_SRC(insn->code) == BPF_X) {
991 if (insn->imm != 0 || insn->off != 0) {
992 verbose("BPF_ALU uses reserved fields\n");
993 return -EINVAL;
994 }
995 /* check src1 operand */
996 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
997 if (err)
998 return err;
999 } else {
1000 if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
1001 verbose("BPF_ALU uses reserved fields\n");
1002 return -EINVAL;
1003 }
1004 }
1005
1006 /* check src2 operand */
1007 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1008 if (err)
1009 return err;
1010
1011 if ((opcode == BPF_MOD || opcode == BPF_DIV) &&
1012 BPF_SRC(insn->code) == BPF_K && insn->imm == 0) {
1013 verbose("div by zero\n");
1014 return -EINVAL;
1015 }
1016
1017 /* pattern match 'bpf_add Rx, imm' instruction */
1018 if (opcode == BPF_ADD && BPF_CLASS(insn->code) == BPF_ALU64 &&
1019 regs[insn->dst_reg].type == FRAME_PTR &&
1020 BPF_SRC(insn->code) == BPF_K)
1021 stack_relative = true;
1022
1023 /* check dest operand */
1024 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1025 if (err)
1026 return err;
1027
1028 if (stack_relative) {
1029 regs[insn->dst_reg].type = PTR_TO_STACK;
1030 regs[insn->dst_reg].imm = insn->imm;
1031 }
1032 }
1033
1034 return 0;
1035 }
1036
1037 static int check_cond_jmp_op(struct verifier_env *env,
1038 struct bpf_insn *insn, int *insn_idx)
1039 {
1040 struct reg_state *regs = env->cur_state.regs;
1041 struct verifier_state *other_branch;
1042 u8 opcode = BPF_OP(insn->code);
1043 int err;
1044
1045 if (opcode > BPF_EXIT) {
1046 verbose("invalid BPF_JMP opcode %x\n", opcode);
1047 return -EINVAL;
1048 }
1049
1050 if (BPF_SRC(insn->code) == BPF_X) {
1051 if (insn->imm != 0) {
1052 verbose("BPF_JMP uses reserved fields\n");
1053 return -EINVAL;
1054 }
1055
1056 /* check src1 operand */
1057 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1058 if (err)
1059 return err;
1060 } else {
1061 if (insn->src_reg != BPF_REG_0) {
1062 verbose("BPF_JMP uses reserved fields\n");
1063 return -EINVAL;
1064 }
1065 }
1066
1067 /* check src2 operand */
1068 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1069 if (err)
1070 return err;
1071
1072 /* detect if R == 0 where R was initialized to zero earlier */
1073 if (BPF_SRC(insn->code) == BPF_K &&
1074 (opcode == BPF_JEQ || opcode == BPF_JNE) &&
1075 regs[insn->dst_reg].type == CONST_IMM &&
1076 regs[insn->dst_reg].imm == insn->imm) {
1077 if (opcode == BPF_JEQ) {
1078 /* if (imm == imm) goto pc+off;
1079 * only follow the goto, ignore fall-through
1080 */
1081 *insn_idx += insn->off;
1082 return 0;
1083 } else {
1084 /* if (imm != imm) goto pc+off;
1085 * only follow fall-through branch, since
1086 * that's where the program will go
1087 */
1088 return 0;
1089 }
1090 }
1091
1092 other_branch = push_stack(env, *insn_idx + insn->off + 1, *insn_idx);
1093 if (!other_branch)
1094 return -EFAULT;
1095
1096 /* detect if R == 0 where R is returned value from bpf_map_lookup_elem() */
1097 if (BPF_SRC(insn->code) == BPF_K &&
1098 insn->imm == 0 && (opcode == BPF_JEQ ||
1099 opcode == BPF_JNE) &&
1100 regs[insn->dst_reg].type == PTR_TO_MAP_VALUE_OR_NULL) {
1101 if (opcode == BPF_JEQ) {
1102 /* next fallthrough insn can access memory via
1103 * this register
1104 */
1105 regs[insn->dst_reg].type = PTR_TO_MAP_VALUE;
1106 /* branch targer cannot access it, since reg == 0 */
1107 other_branch->regs[insn->dst_reg].type = CONST_IMM;
1108 other_branch->regs[insn->dst_reg].imm = 0;
1109 } else {
1110 other_branch->regs[insn->dst_reg].type = PTR_TO_MAP_VALUE;
1111 regs[insn->dst_reg].type = CONST_IMM;
1112 regs[insn->dst_reg].imm = 0;
1113 }
1114 } else if (BPF_SRC(insn->code) == BPF_K &&
1115 (opcode == BPF_JEQ || opcode == BPF_JNE)) {
1116
1117 if (opcode == BPF_JEQ) {
1118 /* detect if (R == imm) goto
1119 * and in the target state recognize that R = imm
1120 */
1121 other_branch->regs[insn->dst_reg].type = CONST_IMM;
1122 other_branch->regs[insn->dst_reg].imm = insn->imm;
1123 } else {
1124 /* detect if (R != imm) goto
1125 * and in the fall-through state recognize that R = imm
1126 */
1127 regs[insn->dst_reg].type = CONST_IMM;
1128 regs[insn->dst_reg].imm = insn->imm;
1129 }
1130 }
1131 if (log_level)
1132 print_verifier_state(env);
1133 return 0;
1134 }
1135
1136 /* return the map pointer stored inside BPF_LD_IMM64 instruction */
1137 static struct bpf_map *ld_imm64_to_map_ptr(struct bpf_insn *insn)
1138 {
1139 u64 imm64 = ((u64) (u32) insn[0].imm) | ((u64) (u32) insn[1].imm) << 32;
1140
1141 return (struct bpf_map *) (unsigned long) imm64;
1142 }
1143
1144 /* verify BPF_LD_IMM64 instruction */
1145 static int check_ld_imm(struct verifier_env *env, struct bpf_insn *insn)
1146 {
1147 struct reg_state *regs = env->cur_state.regs;
1148 int err;
1149
1150 if (BPF_SIZE(insn->code) != BPF_DW) {
1151 verbose("invalid BPF_LD_IMM insn\n");
1152 return -EINVAL;
1153 }
1154 if (insn->off != 0) {
1155 verbose("BPF_LD_IMM64 uses reserved fields\n");
1156 return -EINVAL;
1157 }
1158
1159 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1160 if (err)
1161 return err;
1162
1163 if (insn->src_reg == 0)
1164 /* generic move 64-bit immediate into a register */
1165 return 0;
1166
1167 /* replace_map_fd_with_map_ptr() should have caught bad ld_imm64 */
1168 BUG_ON(insn->src_reg != BPF_PSEUDO_MAP_FD);
1169
1170 regs[insn->dst_reg].type = CONST_PTR_TO_MAP;
1171 regs[insn->dst_reg].map_ptr = ld_imm64_to_map_ptr(insn);
1172 return 0;
1173 }
1174
1175 /* non-recursive DFS pseudo code
1176 * 1 procedure DFS-iterative(G,v):
1177 * 2 label v as discovered
1178 * 3 let S be a stack
1179 * 4 S.push(v)
1180 * 5 while S is not empty
1181 * 6 t <- S.pop()
1182 * 7 if t is what we're looking for:
1183 * 8 return t
1184 * 9 for all edges e in G.adjacentEdges(t) do
1185 * 10 if edge e is already labelled
1186 * 11 continue with the next edge
1187 * 12 w <- G.adjacentVertex(t,e)
1188 * 13 if vertex w is not discovered and not explored
1189 * 14 label e as tree-edge
1190 * 15 label w as discovered
1191 * 16 S.push(w)
1192 * 17 continue at 5
1193 * 18 else if vertex w is discovered
1194 * 19 label e as back-edge
1195 * 20 else
1196 * 21 // vertex w is explored
1197 * 22 label e as forward- or cross-edge
1198 * 23 label t as explored
1199 * 24 S.pop()
1200 *
1201 * convention:
1202 * 0x10 - discovered
1203 * 0x11 - discovered and fall-through edge labelled
1204 * 0x12 - discovered and fall-through and branch edges labelled
1205 * 0x20 - explored
1206 */
1207
1208 enum {
1209 DISCOVERED = 0x10,
1210 EXPLORED = 0x20,
1211 FALLTHROUGH = 1,
1212 BRANCH = 2,
1213 };
1214
1215 #define STATE_LIST_MARK ((struct verifier_state_list *) -1L)
1216
1217 static int *insn_stack; /* stack of insns to process */
1218 static int cur_stack; /* current stack index */
1219 static int *insn_state;
1220
1221 /* t, w, e - match pseudo-code above:
1222 * t - index of current instruction
1223 * w - next instruction
1224 * e - edge
1225 */
1226 static int push_insn(int t, int w, int e, struct verifier_env *env)
1227 {
1228 if (e == FALLTHROUGH && insn_state[t] >= (DISCOVERED | FALLTHROUGH))
1229 return 0;
1230
1231 if (e == BRANCH && insn_state[t] >= (DISCOVERED | BRANCH))
1232 return 0;
1233
1234 if (w < 0 || w >= env->prog->len) {
1235 verbose("jump out of range from insn %d to %d\n", t, w);
1236 return -EINVAL;
1237 }
1238
1239 if (e == BRANCH)
1240 /* mark branch target for state pruning */
1241 env->explored_states[w] = STATE_LIST_MARK;
1242
1243 if (insn_state[w] == 0) {
1244 /* tree-edge */
1245 insn_state[t] = DISCOVERED | e;
1246 insn_state[w] = DISCOVERED;
1247 if (cur_stack >= env->prog->len)
1248 return -E2BIG;
1249 insn_stack[cur_stack++] = w;
1250 return 1;
1251 } else if ((insn_state[w] & 0xF0) == DISCOVERED) {
1252 verbose("back-edge from insn %d to %d\n", t, w);
1253 return -EINVAL;
1254 } else if (insn_state[w] == EXPLORED) {
1255 /* forward- or cross-edge */
1256 insn_state[t] = DISCOVERED | e;
1257 } else {
1258 verbose("insn state internal bug\n");
1259 return -EFAULT;
1260 }
1261 return 0;
1262 }
1263
1264 /* non-recursive depth-first-search to detect loops in BPF program
1265 * loop == back-edge in directed graph
1266 */
1267 static int check_cfg(struct verifier_env *env)
1268 {
1269 struct bpf_insn *insns = env->prog->insnsi;
1270 int insn_cnt = env->prog->len;
1271 int ret = 0;
1272 int i, t;
1273
1274 insn_state = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
1275 if (!insn_state)
1276 return -ENOMEM;
1277
1278 insn_stack = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
1279 if (!insn_stack) {
1280 kfree(insn_state);
1281 return -ENOMEM;
1282 }
1283
1284 insn_state[0] = DISCOVERED; /* mark 1st insn as discovered */
1285 insn_stack[0] = 0; /* 0 is the first instruction */
1286 cur_stack = 1;
1287
1288 peek_stack:
1289 if (cur_stack == 0)
1290 goto check_state;
1291 t = insn_stack[cur_stack - 1];
1292
1293 if (BPF_CLASS(insns[t].code) == BPF_JMP) {
1294 u8 opcode = BPF_OP(insns[t].code);
1295
1296 if (opcode == BPF_EXIT) {
1297 goto mark_explored;
1298 } else if (opcode == BPF_CALL) {
1299 ret = push_insn(t, t + 1, FALLTHROUGH, env);
1300 if (ret == 1)
1301 goto peek_stack;
1302 else if (ret < 0)
1303 goto err_free;
1304 } else if (opcode == BPF_JA) {
1305 if (BPF_SRC(insns[t].code) != BPF_K) {
1306 ret = -EINVAL;
1307 goto err_free;
1308 }
1309 /* unconditional jump with single edge */
1310 ret = push_insn(t, t + insns[t].off + 1,
1311 FALLTHROUGH, env);
1312 if (ret == 1)
1313 goto peek_stack;
1314 else if (ret < 0)
1315 goto err_free;
1316 /* tell verifier to check for equivalent states
1317 * after every call and jump
1318 */
1319 env->explored_states[t + 1] = STATE_LIST_MARK;
1320 } else {
1321 /* conditional jump with two edges */
1322 ret = push_insn(t, t + 1, FALLTHROUGH, env);
1323 if (ret == 1)
1324 goto peek_stack;
1325 else if (ret < 0)
1326 goto err_free;
1327
1328 ret = push_insn(t, t + insns[t].off + 1, BRANCH, env);
1329 if (ret == 1)
1330 goto peek_stack;
1331 else if (ret < 0)
1332 goto err_free;
1333 }
1334 } else {
1335 /* all other non-branch instructions with single
1336 * fall-through edge
1337 */
1338 ret = push_insn(t, t + 1, FALLTHROUGH, env);
1339 if (ret == 1)
1340 goto peek_stack;
1341 else if (ret < 0)
1342 goto err_free;
1343 }
1344
1345 mark_explored:
1346 insn_state[t] = EXPLORED;
1347 if (cur_stack-- <= 0) {
1348 verbose("pop stack internal bug\n");
1349 ret = -EFAULT;
1350 goto err_free;
1351 }
1352 goto peek_stack;
1353
1354 check_state:
1355 for (i = 0; i < insn_cnt; i++) {
1356 if (insn_state[i] != EXPLORED) {
1357 verbose("unreachable insn %d\n", i);
1358 ret = -EINVAL;
1359 goto err_free;
1360 }
1361 }
1362 ret = 0; /* cfg looks good */
1363
1364 err_free:
1365 kfree(insn_state);
1366 kfree(insn_stack);
1367 return ret;
1368 }
1369
1370 /* compare two verifier states
1371 *
1372 * all states stored in state_list are known to be valid, since
1373 * verifier reached 'bpf_exit' instruction through them
1374 *
1375 * this function is called when verifier exploring different branches of
1376 * execution popped from the state stack. If it sees an old state that has
1377 * more strict register state and more strict stack state then this execution
1378 * branch doesn't need to be explored further, since verifier already
1379 * concluded that more strict state leads to valid finish.
1380 *
1381 * Therefore two states are equivalent if register state is more conservative
1382 * and explored stack state is more conservative than the current one.
1383 * Example:
1384 * explored current
1385 * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC)
1386 * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC)
1387 *
1388 * In other words if current stack state (one being explored) has more
1389 * valid slots than old one that already passed validation, it means
1390 * the verifier can stop exploring and conclude that current state is valid too
1391 *
1392 * Similarly with registers. If explored state has register type as invalid
1393 * whereas register type in current state is meaningful, it means that
1394 * the current state will reach 'bpf_exit' instruction safely
1395 */
1396 static bool states_equal(struct verifier_state *old, struct verifier_state *cur)
1397 {
1398 int i;
1399
1400 for (i = 0; i < MAX_BPF_REG; i++) {
1401 if (memcmp(&old->regs[i], &cur->regs[i],
1402 sizeof(old->regs[0])) != 0) {
1403 if (old->regs[i].type == NOT_INIT ||
1404 (old->regs[i].type == UNKNOWN_VALUE &&
1405 cur->regs[i].type != NOT_INIT))
1406 continue;
1407 return false;
1408 }
1409 }
1410
1411 for (i = 0; i < MAX_BPF_STACK; i++) {
1412 if (old->stack_slot_type[i] == STACK_INVALID)
1413 continue;
1414 if (old->stack_slot_type[i] != cur->stack_slot_type[i])
1415 /* Ex: old explored (safe) state has STACK_SPILL in
1416 * this stack slot, but current has has STACK_MISC ->
1417 * this verifier states are not equivalent,
1418 * return false to continue verification of this path
1419 */
1420 return false;
1421 if (i % BPF_REG_SIZE)
1422 continue;
1423 if (memcmp(&old->spilled_regs[i / BPF_REG_SIZE],
1424 &cur->spilled_regs[i / BPF_REG_SIZE],
1425 sizeof(old->spilled_regs[0])))
1426 /* when explored and current stack slot types are
1427 * the same, check that stored pointers types
1428 * are the same as well.
1429 * Ex: explored safe path could have stored
1430 * (struct reg_state) {.type = PTR_TO_STACK, .imm = -8}
1431 * but current path has stored:
1432 * (struct reg_state) {.type = PTR_TO_STACK, .imm = -16}
1433 * such verifier states are not equivalent.
1434 * return false to continue verification of this path
1435 */
1436 return false;
1437 else
1438 continue;
1439 }
1440 return true;
1441 }
1442
1443 static int is_state_visited(struct verifier_env *env, int insn_idx)
1444 {
1445 struct verifier_state_list *new_sl;
1446 struct verifier_state_list *sl;
1447
1448 sl = env->explored_states[insn_idx];
1449 if (!sl)
1450 /* this 'insn_idx' instruction wasn't marked, so we will not
1451 * be doing state search here
1452 */
1453 return 0;
1454
1455 while (sl != STATE_LIST_MARK) {
1456 if (states_equal(&sl->state, &env->cur_state))
1457 /* reached equivalent register/stack state,
1458 * prune the search
1459 */
1460 return 1;
1461 sl = sl->next;
1462 }
1463
1464 /* there were no equivalent states, remember current one.
1465 * technically the current state is not proven to be safe yet,
1466 * but it will either reach bpf_exit (which means it's safe) or
1467 * it will be rejected. Since there are no loops, we won't be
1468 * seeing this 'insn_idx' instruction again on the way to bpf_exit
1469 */
1470 new_sl = kmalloc(sizeof(struct verifier_state_list), GFP_USER);
1471 if (!new_sl)
1472 return -ENOMEM;
1473
1474 /* add new state to the head of linked list */
1475 memcpy(&new_sl->state, &env->cur_state, sizeof(env->cur_state));
1476 new_sl->next = env->explored_states[insn_idx];
1477 env->explored_states[insn_idx] = new_sl;
1478 return 0;
1479 }
1480
1481 static int do_check(struct verifier_env *env)
1482 {
1483 struct verifier_state *state = &env->cur_state;
1484 struct bpf_insn *insns = env->prog->insnsi;
1485 struct reg_state *regs = state->regs;
1486 int insn_cnt = env->prog->len;
1487 int insn_idx, prev_insn_idx = 0;
1488 int insn_processed = 0;
1489 bool do_print_state = false;
1490
1491 init_reg_state(regs);
1492 insn_idx = 0;
1493 for (;;) {
1494 struct bpf_insn *insn;
1495 u8 class;
1496 int err;
1497
1498 if (insn_idx >= insn_cnt) {
1499 verbose("invalid insn idx %d insn_cnt %d\n",
1500 insn_idx, insn_cnt);
1501 return -EFAULT;
1502 }
1503
1504 insn = &insns[insn_idx];
1505 class = BPF_CLASS(insn->code);
1506
1507 if (++insn_processed > 32768) {
1508 verbose("BPF program is too large. Proccessed %d insn\n",
1509 insn_processed);
1510 return -E2BIG;
1511 }
1512
1513 err = is_state_visited(env, insn_idx);
1514 if (err < 0)
1515 return err;
1516 if (err == 1) {
1517 /* found equivalent state, can prune the search */
1518 if (log_level) {
1519 if (do_print_state)
1520 verbose("\nfrom %d to %d: safe\n",
1521 prev_insn_idx, insn_idx);
1522 else
1523 verbose("%d: safe\n", insn_idx);
1524 }
1525 goto process_bpf_exit;
1526 }
1527
1528 if (log_level && do_print_state) {
1529 verbose("\nfrom %d to %d:", prev_insn_idx, insn_idx);
1530 print_verifier_state(env);
1531 do_print_state = false;
1532 }
1533
1534 if (log_level) {
1535 verbose("%d: ", insn_idx);
1536 print_bpf_insn(insn);
1537 }
1538
1539 if (class == BPF_ALU || class == BPF_ALU64) {
1540 err = check_alu_op(regs, insn);
1541 if (err)
1542 return err;
1543
1544 } else if (class == BPF_LDX) {
1545 if (BPF_MODE(insn->code) != BPF_MEM ||
1546 insn->imm != 0) {
1547 verbose("BPF_LDX uses reserved fields\n");
1548 return -EINVAL;
1549 }
1550 /* check src operand */
1551 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1552 if (err)
1553 return err;
1554
1555 err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK);
1556 if (err)
1557 return err;
1558
1559 /* check that memory (src_reg + off) is readable,
1560 * the state of dst_reg will be updated by this func
1561 */
1562 err = check_mem_access(env, insn->src_reg, insn->off,
1563 BPF_SIZE(insn->code), BPF_READ,
1564 insn->dst_reg);
1565 if (err)
1566 return err;
1567
1568 } else if (class == BPF_STX) {
1569 if (BPF_MODE(insn->code) == BPF_XADD) {
1570 err = check_xadd(env, insn);
1571 if (err)
1572 return err;
1573 insn_idx++;
1574 continue;
1575 }
1576
1577 if (BPF_MODE(insn->code) != BPF_MEM ||
1578 insn->imm != 0) {
1579 verbose("BPF_STX uses reserved fields\n");
1580 return -EINVAL;
1581 }
1582 /* check src1 operand */
1583 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1584 if (err)
1585 return err;
1586 /* check src2 operand */
1587 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1588 if (err)
1589 return err;
1590
1591 /* check that memory (dst_reg + off) is writeable */
1592 err = check_mem_access(env, insn->dst_reg, insn->off,
1593 BPF_SIZE(insn->code), BPF_WRITE,
1594 insn->src_reg);
1595 if (err)
1596 return err;
1597
1598 } else if (class == BPF_ST) {
1599 if (BPF_MODE(insn->code) != BPF_MEM ||
1600 insn->src_reg != BPF_REG_0) {
1601 verbose("BPF_ST uses reserved fields\n");
1602 return -EINVAL;
1603 }
1604 /* check src operand */
1605 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1606 if (err)
1607 return err;
1608
1609 /* check that memory (dst_reg + off) is writeable */
1610 err = check_mem_access(env, insn->dst_reg, insn->off,
1611 BPF_SIZE(insn->code), BPF_WRITE,
1612 -1);
1613 if (err)
1614 return err;
1615
1616 } else if (class == BPF_JMP) {
1617 u8 opcode = BPF_OP(insn->code);
1618
1619 if (opcode == BPF_CALL) {
1620 if (BPF_SRC(insn->code) != BPF_K ||
1621 insn->off != 0 ||
1622 insn->src_reg != BPF_REG_0 ||
1623 insn->dst_reg != BPF_REG_0) {
1624 verbose("BPF_CALL uses reserved fields\n");
1625 return -EINVAL;
1626 }
1627
1628 err = check_call(env, insn->imm);
1629 if (err)
1630 return err;
1631
1632 } else if (opcode == BPF_JA) {
1633 if (BPF_SRC(insn->code) != BPF_K ||
1634 insn->imm != 0 ||
1635 insn->src_reg != BPF_REG_0 ||
1636 insn->dst_reg != BPF_REG_0) {
1637 verbose("BPF_JA uses reserved fields\n");
1638 return -EINVAL;
1639 }
1640
1641 insn_idx += insn->off + 1;
1642 continue;
1643
1644 } else if (opcode == BPF_EXIT) {
1645 if (BPF_SRC(insn->code) != BPF_K ||
1646 insn->imm != 0 ||
1647 insn->src_reg != BPF_REG_0 ||
1648 insn->dst_reg != BPF_REG_0) {
1649 verbose("BPF_EXIT uses reserved fields\n");
1650 return -EINVAL;
1651 }
1652
1653 /* eBPF calling convetion is such that R0 is used
1654 * to return the value from eBPF program.
1655 * Make sure that it's readable at this time
1656 * of bpf_exit, which means that program wrote
1657 * something into it earlier
1658 */
1659 err = check_reg_arg(regs, BPF_REG_0, SRC_OP);
1660 if (err)
1661 return err;
1662
1663 process_bpf_exit:
1664 insn_idx = pop_stack(env, &prev_insn_idx);
1665 if (insn_idx < 0) {
1666 break;
1667 } else {
1668 do_print_state = true;
1669 continue;
1670 }
1671 } else {
1672 err = check_cond_jmp_op(env, insn, &insn_idx);
1673 if (err)
1674 return err;
1675 }
1676 } else if (class == BPF_LD) {
1677 u8 mode = BPF_MODE(insn->code);
1678
1679 if (mode == BPF_ABS || mode == BPF_IND) {
1680 verbose("LD_ABS is not supported yet\n");
1681 return -EINVAL;
1682 } else if (mode == BPF_IMM) {
1683 err = check_ld_imm(env, insn);
1684 if (err)
1685 return err;
1686
1687 insn_idx++;
1688 } else {
1689 verbose("invalid BPF_LD mode\n");
1690 return -EINVAL;
1691 }
1692 } else {
1693 verbose("unknown insn class %d\n", class);
1694 return -EINVAL;
1695 }
1696
1697 insn_idx++;
1698 }
1699
1700 return 0;
1701 }
1702
1703 /* look for pseudo eBPF instructions that access map FDs and
1704 * replace them with actual map pointers
1705 */
1706 static int replace_map_fd_with_map_ptr(struct verifier_env *env)
1707 {
1708 struct bpf_insn *insn = env->prog->insnsi;
1709 int insn_cnt = env->prog->len;
1710 int i, j;
1711
1712 for (i = 0; i < insn_cnt; i++, insn++) {
1713 if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW)) {
1714 struct bpf_map *map;
1715 struct fd f;
1716
1717 if (i == insn_cnt - 1 || insn[1].code != 0 ||
1718 insn[1].dst_reg != 0 || insn[1].src_reg != 0 ||
1719 insn[1].off != 0) {
1720 verbose("invalid bpf_ld_imm64 insn\n");
1721 return -EINVAL;
1722 }
1723
1724 if (insn->src_reg == 0)
1725 /* valid generic load 64-bit imm */
1726 goto next_insn;
1727
1728 if (insn->src_reg != BPF_PSEUDO_MAP_FD) {
1729 verbose("unrecognized bpf_ld_imm64 insn\n");
1730 return -EINVAL;
1731 }
1732
1733 f = fdget(insn->imm);
1734
1735 map = bpf_map_get(f);
1736 if (IS_ERR(map)) {
1737 verbose("fd %d is not pointing to valid bpf_map\n",
1738 insn->imm);
1739 fdput(f);
1740 return PTR_ERR(map);
1741 }
1742
1743 /* store map pointer inside BPF_LD_IMM64 instruction */
1744 insn[0].imm = (u32) (unsigned long) map;
1745 insn[1].imm = ((u64) (unsigned long) map) >> 32;
1746
1747 /* check whether we recorded this map already */
1748 for (j = 0; j < env->used_map_cnt; j++)
1749 if (env->used_maps[j] == map) {
1750 fdput(f);
1751 goto next_insn;
1752 }
1753
1754 if (env->used_map_cnt >= MAX_USED_MAPS) {
1755 fdput(f);
1756 return -E2BIG;
1757 }
1758
1759 /* remember this map */
1760 env->used_maps[env->used_map_cnt++] = map;
1761
1762 /* hold the map. If the program is rejected by verifier,
1763 * the map will be released by release_maps() or it
1764 * will be used by the valid program until it's unloaded
1765 * and all maps are released in free_bpf_prog_info()
1766 */
1767 atomic_inc(&map->refcnt);
1768
1769 fdput(f);
1770 next_insn:
1771 insn++;
1772 i++;
1773 }
1774 }
1775
1776 /* now all pseudo BPF_LD_IMM64 instructions load valid
1777 * 'struct bpf_map *' into a register instead of user map_fd.
1778 * These pointers will be used later by verifier to validate map access.
1779 */
1780 return 0;
1781 }
1782
1783 /* drop refcnt of maps used by the rejected program */
1784 static void release_maps(struct verifier_env *env)
1785 {
1786 int i;
1787
1788 for (i = 0; i < env->used_map_cnt; i++)
1789 bpf_map_put(env->used_maps[i]);
1790 }
1791
1792 /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */
1793 static void convert_pseudo_ld_imm64(struct verifier_env *env)
1794 {
1795 struct bpf_insn *insn = env->prog->insnsi;
1796 int insn_cnt = env->prog->len;
1797 int i;
1798
1799 for (i = 0; i < insn_cnt; i++, insn++)
1800 if (insn->code == (BPF_LD | BPF_IMM | BPF_DW))
1801 insn->src_reg = 0;
1802 }
1803
1804 static void free_states(struct verifier_env *env)
1805 {
1806 struct verifier_state_list *sl, *sln;
1807 int i;
1808
1809 if (!env->explored_states)
1810 return;
1811
1812 for (i = 0; i < env->prog->len; i++) {
1813 sl = env->explored_states[i];
1814
1815 if (sl)
1816 while (sl != STATE_LIST_MARK) {
1817 sln = sl->next;
1818 kfree(sl);
1819 sl = sln;
1820 }
1821 }
1822
1823 kfree(env->explored_states);
1824 }
1825
1826 int bpf_check(struct bpf_prog *prog, union bpf_attr *attr)
1827 {
1828 char __user *log_ubuf = NULL;
1829 struct verifier_env *env;
1830 int ret = -EINVAL;
1831
1832 if (prog->len <= 0 || prog->len > BPF_MAXINSNS)
1833 return -E2BIG;
1834
1835 /* 'struct verifier_env' can be global, but since it's not small,
1836 * allocate/free it every time bpf_check() is called
1837 */
1838 env = kzalloc(sizeof(struct verifier_env), GFP_KERNEL);
1839 if (!env)
1840 return -ENOMEM;
1841
1842 env->prog = prog;
1843
1844 /* grab the mutex to protect few globals used by verifier */
1845 mutex_lock(&bpf_verifier_lock);
1846
1847 if (attr->log_level || attr->log_buf || attr->log_size) {
1848 /* user requested verbose verifier output
1849 * and supplied buffer to store the verification trace
1850 */
1851 log_level = attr->log_level;
1852 log_ubuf = (char __user *) (unsigned long) attr->log_buf;
1853 log_size = attr->log_size;
1854 log_len = 0;
1855
1856 ret = -EINVAL;
1857 /* log_* values have to be sane */
1858 if (log_size < 128 || log_size > UINT_MAX >> 8 ||
1859 log_level == 0 || log_ubuf == NULL)
1860 goto free_env;
1861
1862 ret = -ENOMEM;
1863 log_buf = vmalloc(log_size);
1864 if (!log_buf)
1865 goto free_env;
1866 } else {
1867 log_level = 0;
1868 }
1869
1870 ret = replace_map_fd_with_map_ptr(env);
1871 if (ret < 0)
1872 goto skip_full_check;
1873
1874 env->explored_states = kcalloc(prog->len,
1875 sizeof(struct verifier_state_list *),
1876 GFP_USER);
1877 ret = -ENOMEM;
1878 if (!env->explored_states)
1879 goto skip_full_check;
1880
1881 ret = check_cfg(env);
1882 if (ret < 0)
1883 goto skip_full_check;
1884
1885 ret = do_check(env);
1886
1887 skip_full_check:
1888 while (pop_stack(env, NULL) >= 0);
1889 free_states(env);
1890
1891 if (log_level && log_len >= log_size - 1) {
1892 BUG_ON(log_len >= log_size);
1893 /* verifier log exceeded user supplied buffer */
1894 ret = -ENOSPC;
1895 /* fall through to return what was recorded */
1896 }
1897
1898 /* copy verifier log back to user space including trailing zero */
1899 if (log_level && copy_to_user(log_ubuf, log_buf, log_len + 1) != 0) {
1900 ret = -EFAULT;
1901 goto free_log_buf;
1902 }
1903
1904 if (ret == 0 && env->used_map_cnt) {
1905 /* if program passed verifier, update used_maps in bpf_prog_info */
1906 prog->aux->used_maps = kmalloc_array(env->used_map_cnt,
1907 sizeof(env->used_maps[0]),
1908 GFP_KERNEL);
1909
1910 if (!prog->aux->used_maps) {
1911 ret = -ENOMEM;
1912 goto free_log_buf;
1913 }
1914
1915 memcpy(prog->aux->used_maps, env->used_maps,
1916 sizeof(env->used_maps[0]) * env->used_map_cnt);
1917 prog->aux->used_map_cnt = env->used_map_cnt;
1918
1919 /* program is valid. Convert pseudo bpf_ld_imm64 into generic
1920 * bpf_ld_imm64 instructions
1921 */
1922 convert_pseudo_ld_imm64(env);
1923 }
1924
1925 free_log_buf:
1926 if (log_level)
1927 vfree(log_buf);
1928 free_env:
1929 if (!prog->aux->used_maps)
1930 /* if we didn't copy map pointers into bpf_prog_info, release
1931 * them now. Otherwise free_bpf_prog_info() will release them.
1932 */
1933 release_maps(env);
1934 kfree(env);
1935 mutex_unlock(&bpf_verifier_lock);
1936 return ret;
1937 }
Linux with added FPC-III support