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 */
verbose(const char * fmt,...)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
print_verifier_state(struct verifier_env * env)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
print_bpf_insn(struct bpf_insn * insn)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
pop_stack(struct verifier_env * env,int * prev_insn_idx)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
push_stack(struct verifier_env * env,int insn_idx,int prev_insn_idx)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
init_reg_state(struct reg_state * regs)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
mark_reg_unknown_value(struct reg_state * regs,u32 regno)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
check_reg_arg(struct reg_state * regs,u32 regno,enum reg_arg_type t)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
bpf_size_to_bytes(int bpf_size)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 */
check_stack_write(struct verifier_state * state,int off,int size,int value_regno)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
check_stack_read(struct verifier_state * state,int off,int size,int value_regno)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() */
check_map_access(struct verifier_env * env,u32 regno,int off,int size)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 */
check_ctx_access(struct verifier_env * env,int off,int size,enum bpf_access_type t)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 */
check_mem_access(struct verifier_env * env,u32 regno,int off,int bpf_size,enum bpf_access_type t,int value_regno)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
check_xadd(struct verifier_env * env,struct bpf_insn * insn)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 */
check_stack_boundary(struct verifier_env * env,int regno,int access_size)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
check_func_arg(struct verifier_env * env,u32 regno,enum bpf_arg_type arg_type,struct bpf_map ** mapp)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_DONTCARE)
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_ANYTHING)
767 return 0;
768
769 if (arg_type == ARG_PTR_TO_STACK || arg_type == ARG_PTR_TO_MAP_KEY ||
770 arg_type == ARG_PTR_TO_MAP_VALUE) {
771 expected_type = PTR_TO_STACK;
772 } else if (arg_type == ARG_CONST_STACK_SIZE) {
773 expected_type = CONST_IMM;
774 } else if (arg_type == ARG_CONST_MAP_PTR) {
775 expected_type = CONST_PTR_TO_MAP;
776 } else if (arg_type == ARG_PTR_TO_CTX) {
777 expected_type = PTR_TO_CTX;
778 } else {
779 verbose("unsupported arg_type %d\n", arg_type);
780 return -EFAULT;
781 }
782
783 if (reg->type != expected_type) {
784 verbose("R%d type=%s expected=%s\n", regno,
785 reg_type_str[reg->type], reg_type_str[expected_type]);
786 return -EACCES;
787 }
788
789 if (arg_type == ARG_CONST_MAP_PTR) {
790 /* bpf_map_xxx(map_ptr) call: remember that map_ptr */
791 *mapp = reg->map_ptr;
792
793 } else if (arg_type == ARG_PTR_TO_MAP_KEY) {
794 /* bpf_map_xxx(..., map_ptr, ..., key) call:
795 * check that [key, key + map->key_size) are within
796 * stack limits and initialized
797 */
798 if (!*mapp) {
799 /* in function declaration map_ptr must come before
800 * map_key, so that it's verified and known before
801 * we have to check map_key here. Otherwise it means
802 * that kernel subsystem misconfigured verifier
803 */
804 verbose("invalid map_ptr to access map->key\n");
805 return -EACCES;
806 }
807 err = check_stack_boundary(env, regno, (*mapp)->key_size);
808
809 } else if (arg_type == ARG_PTR_TO_MAP_VALUE) {
810 /* bpf_map_xxx(..., map_ptr, ..., value) call:
811 * check [value, value + map->value_size) validity
812 */
813 if (!*mapp) {
814 /* kernel subsystem misconfigured verifier */
815 verbose("invalid map_ptr to access map->value\n");
816 return -EACCES;
817 }
818 err = check_stack_boundary(env, regno, (*mapp)->value_size);
819
820 } else if (arg_type == ARG_CONST_STACK_SIZE) {
821 /* bpf_xxx(..., buf, len) call will access 'len' bytes
822 * from stack pointer 'buf'. Check it
823 * note: regno == len, regno - 1 == buf
824 */
825 if (regno == 0) {
826 /* kernel subsystem misconfigured verifier */
827 verbose("ARG_CONST_STACK_SIZE cannot be first argument\n");
828 return -EACCES;
829 }
830 err = check_stack_boundary(env, regno - 1, reg->imm);
831 }
832
833 return err;
834 }
835
check_call(struct verifier_env * env,int func_id)836 static int check_call(struct verifier_env *env, int func_id)
837 {
838 struct verifier_state *state = &env->cur_state;
839 const struct bpf_func_proto *fn = NULL;
840 struct reg_state *regs = state->regs;
841 struct bpf_map *map = NULL;
842 struct reg_state *reg;
843 int i, err;
844
845 /* find function prototype */
846 if (func_id < 0 || func_id >= __BPF_FUNC_MAX_ID) {
847 verbose("invalid func %d\n", func_id);
848 return -EINVAL;
849 }
850
851 if (env->prog->aux->ops->get_func_proto)
852 fn = env->prog->aux->ops->get_func_proto(func_id);
853
854 if (!fn) {
855 verbose("unknown func %d\n", func_id);
856 return -EINVAL;
857 }
858
859 /* eBPF programs must be GPL compatible to use GPL-ed functions */
860 if (!env->prog->gpl_compatible && fn->gpl_only) {
861 verbose("cannot call GPL only function from proprietary program\n");
862 return -EINVAL;
863 }
864
865 /* check args */
866 err = check_func_arg(env, BPF_REG_1, fn->arg1_type, &map);
867 if (err)
868 return err;
869 err = check_func_arg(env, BPF_REG_2, fn->arg2_type, &map);
870 if (err)
871 return err;
872 err = check_func_arg(env, BPF_REG_3, fn->arg3_type, &map);
873 if (err)
874 return err;
875 err = check_func_arg(env, BPF_REG_4, fn->arg4_type, &map);
876 if (err)
877 return err;
878 err = check_func_arg(env, BPF_REG_5, fn->arg5_type, &map);
879 if (err)
880 return err;
881
882 /* reset caller saved regs */
883 for (i = 0; i < CALLER_SAVED_REGS; i++) {
884 reg = regs + caller_saved[i];
885 reg->type = NOT_INIT;
886 reg->imm = 0;
887 }
888
889 /* update return register */
890 if (fn->ret_type == RET_INTEGER) {
891 regs[BPF_REG_0].type = UNKNOWN_VALUE;
892 } else if (fn->ret_type == RET_VOID) {
893 regs[BPF_REG_0].type = NOT_INIT;
894 } else if (fn->ret_type == RET_PTR_TO_MAP_VALUE_OR_NULL) {
895 regs[BPF_REG_0].type = PTR_TO_MAP_VALUE_OR_NULL;
896 /* remember map_ptr, so that check_map_access()
897 * can check 'value_size' boundary of memory access
898 * to map element returned from bpf_map_lookup_elem()
899 */
900 if (map == NULL) {
901 verbose("kernel subsystem misconfigured verifier\n");
902 return -EINVAL;
903 }
904 regs[BPF_REG_0].map_ptr = map;
905 } else {
906 verbose("unknown return type %d of func %d\n",
907 fn->ret_type, func_id);
908 return -EINVAL;
909 }
910 return 0;
911 }
912
913 /* check validity of 32-bit and 64-bit arithmetic operations */
check_alu_op(struct reg_state * regs,struct bpf_insn * insn)914 static int check_alu_op(struct reg_state *regs, struct bpf_insn *insn)
915 {
916 u8 opcode = BPF_OP(insn->code);
917 int err;
918
919 if (opcode == BPF_END || opcode == BPF_NEG) {
920 if (opcode == BPF_NEG) {
921 if (BPF_SRC(insn->code) != 0 ||
922 insn->src_reg != BPF_REG_0 ||
923 insn->off != 0 || insn->imm != 0) {
924 verbose("BPF_NEG uses reserved fields\n");
925 return -EINVAL;
926 }
927 } else {
928 if (insn->src_reg != BPF_REG_0 || insn->off != 0 ||
929 (insn->imm != 16 && insn->imm != 32 && insn->imm != 64)) {
930 verbose("BPF_END uses reserved fields\n");
931 return -EINVAL;
932 }
933 }
934
935 /* check src operand */
936 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
937 if (err)
938 return err;
939
940 /* check dest operand */
941 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
942 if (err)
943 return err;
944
945 } else if (opcode == BPF_MOV) {
946
947 if (BPF_SRC(insn->code) == BPF_X) {
948 if (insn->imm != 0 || insn->off != 0) {
949 verbose("BPF_MOV uses reserved fields\n");
950 return -EINVAL;
951 }
952
953 /* check src operand */
954 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
955 if (err)
956 return err;
957 } else {
958 if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
959 verbose("BPF_MOV uses reserved fields\n");
960 return -EINVAL;
961 }
962 }
963
964 /* check dest operand */
965 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
966 if (err)
967 return err;
968
969 if (BPF_SRC(insn->code) == BPF_X) {
970 if (BPF_CLASS(insn->code) == BPF_ALU64) {
971 /* case: R1 = R2
972 * copy register state to dest reg
973 */
974 regs[insn->dst_reg] = regs[insn->src_reg];
975 } else {
976 regs[insn->dst_reg].type = UNKNOWN_VALUE;
977 regs[insn->dst_reg].map_ptr = NULL;
978 }
979 } else {
980 /* case: R = imm
981 * remember the value we stored into this reg
982 */
983 regs[insn->dst_reg].type = CONST_IMM;
984 regs[insn->dst_reg].imm = insn->imm;
985 }
986
987 } else if (opcode > BPF_END) {
988 verbose("invalid BPF_ALU opcode %x\n", opcode);
989 return -EINVAL;
990
991 } else { /* all other ALU ops: and, sub, xor, add, ... */
992
993 bool stack_relative = false;
994
995 if (BPF_SRC(insn->code) == BPF_X) {
996 if (insn->imm != 0 || insn->off != 0) {
997 verbose("BPF_ALU uses reserved fields\n");
998 return -EINVAL;
999 }
1000 /* check src1 operand */
1001 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1002 if (err)
1003 return err;
1004 } else {
1005 if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
1006 verbose("BPF_ALU uses reserved fields\n");
1007 return -EINVAL;
1008 }
1009 }
1010
1011 /* check src2 operand */
1012 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1013 if (err)
1014 return err;
1015
1016 if ((opcode == BPF_MOD || opcode == BPF_DIV) &&
1017 BPF_SRC(insn->code) == BPF_K && insn->imm == 0) {
1018 verbose("div by zero\n");
1019 return -EINVAL;
1020 }
1021
1022 if ((opcode == BPF_LSH || opcode == BPF_RSH ||
1023 opcode == BPF_ARSH) && BPF_SRC(insn->code) == BPF_K) {
1024 int size = BPF_CLASS(insn->code) == BPF_ALU64 ? 64 : 32;
1025
1026 if (insn->imm < 0 || insn->imm >= size) {
1027 verbose("invalid shift %d\n", insn->imm);
1028 return -EINVAL;
1029 }
1030 }
1031
1032 /* pattern match 'bpf_add Rx, imm' instruction */
1033 if (opcode == BPF_ADD && BPF_CLASS(insn->code) == BPF_ALU64 &&
1034 regs[insn->dst_reg].type == FRAME_PTR &&
1035 BPF_SRC(insn->code) == BPF_K)
1036 stack_relative = true;
1037
1038 /* check dest operand */
1039 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1040 if (err)
1041 return err;
1042
1043 if (stack_relative) {
1044 regs[insn->dst_reg].type = PTR_TO_STACK;
1045 regs[insn->dst_reg].imm = insn->imm;
1046 }
1047 }
1048
1049 return 0;
1050 }
1051
check_cond_jmp_op(struct verifier_env * env,struct bpf_insn * insn,int * insn_idx)1052 static int check_cond_jmp_op(struct verifier_env *env,
1053 struct bpf_insn *insn, int *insn_idx)
1054 {
1055 struct reg_state *regs = env->cur_state.regs;
1056 struct verifier_state *other_branch;
1057 u8 opcode = BPF_OP(insn->code);
1058 int err;
1059
1060 if (opcode > BPF_EXIT) {
1061 verbose("invalid BPF_JMP opcode %x\n", opcode);
1062 return -EINVAL;
1063 }
1064
1065 if (BPF_SRC(insn->code) == BPF_X) {
1066 if (insn->imm != 0) {
1067 verbose("BPF_JMP uses reserved fields\n");
1068 return -EINVAL;
1069 }
1070
1071 /* check src1 operand */
1072 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1073 if (err)
1074 return err;
1075 } else {
1076 if (insn->src_reg != BPF_REG_0) {
1077 verbose("BPF_JMP uses reserved fields\n");
1078 return -EINVAL;
1079 }
1080 }
1081
1082 /* check src2 operand */
1083 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1084 if (err)
1085 return err;
1086
1087 /* detect if R == 0 where R was initialized to zero earlier */
1088 if (BPF_SRC(insn->code) == BPF_K &&
1089 (opcode == BPF_JEQ || opcode == BPF_JNE) &&
1090 regs[insn->dst_reg].type == CONST_IMM &&
1091 regs[insn->dst_reg].imm == insn->imm) {
1092 if (opcode == BPF_JEQ) {
1093 /* if (imm == imm) goto pc+off;
1094 * only follow the goto, ignore fall-through
1095 */
1096 *insn_idx += insn->off;
1097 return 0;
1098 } else {
1099 /* if (imm != imm) goto pc+off;
1100 * only follow fall-through branch, since
1101 * that's where the program will go
1102 */
1103 return 0;
1104 }
1105 }
1106
1107 other_branch = push_stack(env, *insn_idx + insn->off + 1, *insn_idx);
1108 if (!other_branch)
1109 return -EFAULT;
1110
1111 /* detect if R == 0 where R is returned value from bpf_map_lookup_elem() */
1112 if (BPF_SRC(insn->code) == BPF_K &&
1113 insn->imm == 0 && (opcode == BPF_JEQ ||
1114 opcode == BPF_JNE) &&
1115 regs[insn->dst_reg].type == PTR_TO_MAP_VALUE_OR_NULL) {
1116 if (opcode == BPF_JEQ) {
1117 /* next fallthrough insn can access memory via
1118 * this register
1119 */
1120 regs[insn->dst_reg].type = PTR_TO_MAP_VALUE;
1121 /* branch targer cannot access it, since reg == 0 */
1122 other_branch->regs[insn->dst_reg].type = CONST_IMM;
1123 other_branch->regs[insn->dst_reg].imm = 0;
1124 } else {
1125 other_branch->regs[insn->dst_reg].type = PTR_TO_MAP_VALUE;
1126 regs[insn->dst_reg].type = CONST_IMM;
1127 regs[insn->dst_reg].imm = 0;
1128 }
1129 } else if (BPF_SRC(insn->code) == BPF_K &&
1130 (opcode == BPF_JEQ || opcode == BPF_JNE)) {
1131
1132 if (opcode == BPF_JEQ) {
1133 /* detect if (R == imm) goto
1134 * and in the target state recognize that R = imm
1135 */
1136 other_branch->regs[insn->dst_reg].type = CONST_IMM;
1137 other_branch->regs[insn->dst_reg].imm = insn->imm;
1138 } else {
1139 /* detect if (R != imm) goto
1140 * and in the fall-through state recognize that R = imm
1141 */
1142 regs[insn->dst_reg].type = CONST_IMM;
1143 regs[insn->dst_reg].imm = insn->imm;
1144 }
1145 }
1146 if (log_level)
1147 print_verifier_state(env);
1148 return 0;
1149 }
1150
1151 /* return the map pointer stored inside BPF_LD_IMM64 instruction */
ld_imm64_to_map_ptr(struct bpf_insn * insn)1152 static struct bpf_map *ld_imm64_to_map_ptr(struct bpf_insn *insn)
1153 {
1154 u64 imm64 = ((u64) (u32) insn[0].imm) | ((u64) (u32) insn[1].imm) << 32;
1155
1156 return (struct bpf_map *) (unsigned long) imm64;
1157 }
1158
1159 /* verify BPF_LD_IMM64 instruction */
check_ld_imm(struct verifier_env * env,struct bpf_insn * insn)1160 static int check_ld_imm(struct verifier_env *env, struct bpf_insn *insn)
1161 {
1162 struct reg_state *regs = env->cur_state.regs;
1163 int err;
1164
1165 if (BPF_SIZE(insn->code) != BPF_DW) {
1166 verbose("invalid BPF_LD_IMM insn\n");
1167 return -EINVAL;
1168 }
1169 if (insn->off != 0) {
1170 verbose("BPF_LD_IMM64 uses reserved fields\n");
1171 return -EINVAL;
1172 }
1173
1174 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1175 if (err)
1176 return err;
1177
1178 if (insn->src_reg == 0)
1179 /* generic move 64-bit immediate into a register */
1180 return 0;
1181
1182 /* replace_map_fd_with_map_ptr() should have caught bad ld_imm64 */
1183 BUG_ON(insn->src_reg != BPF_PSEUDO_MAP_FD);
1184
1185 regs[insn->dst_reg].type = CONST_PTR_TO_MAP;
1186 regs[insn->dst_reg].map_ptr = ld_imm64_to_map_ptr(insn);
1187 return 0;
1188 }
1189
may_access_skb(enum bpf_prog_type type)1190 static bool may_access_skb(enum bpf_prog_type type)
1191 {
1192 switch (type) {
1193 case BPF_PROG_TYPE_SOCKET_FILTER:
1194 case BPF_PROG_TYPE_SCHED_CLS:
1195 case BPF_PROG_TYPE_SCHED_ACT:
1196 return true;
1197 default:
1198 return false;
1199 }
1200 }
1201
1202 /* verify safety of LD_ABS|LD_IND instructions:
1203 * - they can only appear in the programs where ctx == skb
1204 * - since they are wrappers of function calls, they scratch R1-R5 registers,
1205 * preserve R6-R9, and store return value into R0
1206 *
1207 * Implicit input:
1208 * ctx == skb == R6 == CTX
1209 *
1210 * Explicit input:
1211 * SRC == any register
1212 * IMM == 32-bit immediate
1213 *
1214 * Output:
1215 * R0 - 8/16/32-bit skb data converted to cpu endianness
1216 */
check_ld_abs(struct verifier_env * env,struct bpf_insn * insn)1217 static int check_ld_abs(struct verifier_env *env, struct bpf_insn *insn)
1218 {
1219 struct reg_state *regs = env->cur_state.regs;
1220 u8 mode = BPF_MODE(insn->code);
1221 struct reg_state *reg;
1222 int i, err;
1223
1224 if (!may_access_skb(env->prog->type)) {
1225 verbose("BPF_LD_ABS|IND instructions not allowed for this program type\n");
1226 return -EINVAL;
1227 }
1228
1229 if (insn->dst_reg != BPF_REG_0 || insn->off != 0 ||
1230 (mode == BPF_ABS && insn->src_reg != BPF_REG_0)) {
1231 verbose("BPF_LD_ABS uses reserved fields\n");
1232 return -EINVAL;
1233 }
1234
1235 /* check whether implicit source operand (register R6) is readable */
1236 err = check_reg_arg(regs, BPF_REG_6, SRC_OP);
1237 if (err)
1238 return err;
1239
1240 if (regs[BPF_REG_6].type != PTR_TO_CTX) {
1241 verbose("at the time of BPF_LD_ABS|IND R6 != pointer to skb\n");
1242 return -EINVAL;
1243 }
1244
1245 if (mode == BPF_IND) {
1246 /* check explicit source operand */
1247 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1248 if (err)
1249 return err;
1250 }
1251
1252 /* reset caller saved regs to unreadable */
1253 for (i = 0; i < CALLER_SAVED_REGS; i++) {
1254 reg = regs + caller_saved[i];
1255 reg->type = NOT_INIT;
1256 reg->imm = 0;
1257 }
1258
1259 /* mark destination R0 register as readable, since it contains
1260 * the value fetched from the packet
1261 */
1262 regs[BPF_REG_0].type = UNKNOWN_VALUE;
1263 return 0;
1264 }
1265
1266 /* non-recursive DFS pseudo code
1267 * 1 procedure DFS-iterative(G,v):
1268 * 2 label v as discovered
1269 * 3 let S be a stack
1270 * 4 S.push(v)
1271 * 5 while S is not empty
1272 * 6 t <- S.pop()
1273 * 7 if t is what we're looking for:
1274 * 8 return t
1275 * 9 for all edges e in G.adjacentEdges(t) do
1276 * 10 if edge e is already labelled
1277 * 11 continue with the next edge
1278 * 12 w <- G.adjacentVertex(t,e)
1279 * 13 if vertex w is not discovered and not explored
1280 * 14 label e as tree-edge
1281 * 15 label w as discovered
1282 * 16 S.push(w)
1283 * 17 continue at 5
1284 * 18 else if vertex w is discovered
1285 * 19 label e as back-edge
1286 * 20 else
1287 * 21 // vertex w is explored
1288 * 22 label e as forward- or cross-edge
1289 * 23 label t as explored
1290 * 24 S.pop()
1291 *
1292 * convention:
1293 * 0x10 - discovered
1294 * 0x11 - discovered and fall-through edge labelled
1295 * 0x12 - discovered and fall-through and branch edges labelled
1296 * 0x20 - explored
1297 */
1298
1299 enum {
1300 DISCOVERED = 0x10,
1301 EXPLORED = 0x20,
1302 FALLTHROUGH = 1,
1303 BRANCH = 2,
1304 };
1305
1306 #define STATE_LIST_MARK ((struct verifier_state_list *) -1L)
1307
1308 static int *insn_stack; /* stack of insns to process */
1309 static int cur_stack; /* current stack index */
1310 static int *insn_state;
1311
1312 /* t, w, e - match pseudo-code above:
1313 * t - index of current instruction
1314 * w - next instruction
1315 * e - edge
1316 */
push_insn(int t,int w,int e,struct verifier_env * env)1317 static int push_insn(int t, int w, int e, struct verifier_env *env)
1318 {
1319 if (e == FALLTHROUGH && insn_state[t] >= (DISCOVERED | FALLTHROUGH))
1320 return 0;
1321
1322 if (e == BRANCH && insn_state[t] >= (DISCOVERED | BRANCH))
1323 return 0;
1324
1325 if (w < 0 || w >= env->prog->len) {
1326 verbose("jump out of range from insn %d to %d\n", t, w);
1327 return -EINVAL;
1328 }
1329
1330 if (e == BRANCH)
1331 /* mark branch target for state pruning */
1332 env->explored_states[w] = STATE_LIST_MARK;
1333
1334 if (insn_state[w] == 0) {
1335 /* tree-edge */
1336 insn_state[t] = DISCOVERED | e;
1337 insn_state[w] = DISCOVERED;
1338 if (cur_stack >= env->prog->len)
1339 return -E2BIG;
1340 insn_stack[cur_stack++] = w;
1341 return 1;
1342 } else if ((insn_state[w] & 0xF0) == DISCOVERED) {
1343 verbose("back-edge from insn %d to %d\n", t, w);
1344 return -EINVAL;
1345 } else if (insn_state[w] == EXPLORED) {
1346 /* forward- or cross-edge */
1347 insn_state[t] = DISCOVERED | e;
1348 } else {
1349 verbose("insn state internal bug\n");
1350 return -EFAULT;
1351 }
1352 return 0;
1353 }
1354
1355 /* non-recursive depth-first-search to detect loops in BPF program
1356 * loop == back-edge in directed graph
1357 */
check_cfg(struct verifier_env * env)1358 static int check_cfg(struct verifier_env *env)
1359 {
1360 struct bpf_insn *insns = env->prog->insnsi;
1361 int insn_cnt = env->prog->len;
1362 int ret = 0;
1363 int i, t;
1364
1365 insn_state = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
1366 if (!insn_state)
1367 return -ENOMEM;
1368
1369 insn_stack = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
1370 if (!insn_stack) {
1371 kfree(insn_state);
1372 return -ENOMEM;
1373 }
1374
1375 insn_state[0] = DISCOVERED; /* mark 1st insn as discovered */
1376 insn_stack[0] = 0; /* 0 is the first instruction */
1377 cur_stack = 1;
1378
1379 peek_stack:
1380 if (cur_stack == 0)
1381 goto check_state;
1382 t = insn_stack[cur_stack - 1];
1383
1384 if (BPF_CLASS(insns[t].code) == BPF_JMP) {
1385 u8 opcode = BPF_OP(insns[t].code);
1386
1387 if (opcode == BPF_EXIT) {
1388 goto mark_explored;
1389 } else if (opcode == BPF_CALL) {
1390 ret = push_insn(t, t + 1, FALLTHROUGH, env);
1391 if (ret == 1)
1392 goto peek_stack;
1393 else if (ret < 0)
1394 goto err_free;
1395 } else if (opcode == BPF_JA) {
1396 if (BPF_SRC(insns[t].code) != BPF_K) {
1397 ret = -EINVAL;
1398 goto err_free;
1399 }
1400 /* unconditional jump with single edge */
1401 ret = push_insn(t, t + insns[t].off + 1,
1402 FALLTHROUGH, env);
1403 if (ret == 1)
1404 goto peek_stack;
1405 else if (ret < 0)
1406 goto err_free;
1407 /* tell verifier to check for equivalent states
1408 * after every call and jump
1409 */
1410 if (t + 1 < insn_cnt)
1411 env->explored_states[t + 1] = STATE_LIST_MARK;
1412 } else {
1413 /* conditional jump with two edges */
1414 ret = push_insn(t, t + 1, FALLTHROUGH, env);
1415 if (ret == 1)
1416 goto peek_stack;
1417 else if (ret < 0)
1418 goto err_free;
1419
1420 ret = push_insn(t, t + insns[t].off + 1, BRANCH, env);
1421 if (ret == 1)
1422 goto peek_stack;
1423 else if (ret < 0)
1424 goto err_free;
1425 }
1426 } else {
1427 /* all other non-branch instructions with single
1428 * fall-through edge
1429 */
1430 ret = push_insn(t, t + 1, FALLTHROUGH, env);
1431 if (ret == 1)
1432 goto peek_stack;
1433 else if (ret < 0)
1434 goto err_free;
1435 }
1436
1437 mark_explored:
1438 insn_state[t] = EXPLORED;
1439 if (cur_stack-- <= 0) {
1440 verbose("pop stack internal bug\n");
1441 ret = -EFAULT;
1442 goto err_free;
1443 }
1444 goto peek_stack;
1445
1446 check_state:
1447 for (i = 0; i < insn_cnt; i++) {
1448 if (insn_state[i] != EXPLORED) {
1449 verbose("unreachable insn %d\n", i);
1450 ret = -EINVAL;
1451 goto err_free;
1452 }
1453 }
1454 ret = 0; /* cfg looks good */
1455
1456 err_free:
1457 kfree(insn_state);
1458 kfree(insn_stack);
1459 return ret;
1460 }
1461
1462 /* compare two verifier states
1463 *
1464 * all states stored in state_list are known to be valid, since
1465 * verifier reached 'bpf_exit' instruction through them
1466 *
1467 * this function is called when verifier exploring different branches of
1468 * execution popped from the state stack. If it sees an old state that has
1469 * more strict register state and more strict stack state then this execution
1470 * branch doesn't need to be explored further, since verifier already
1471 * concluded that more strict state leads to valid finish.
1472 *
1473 * Therefore two states are equivalent if register state is more conservative
1474 * and explored stack state is more conservative than the current one.
1475 * Example:
1476 * explored current
1477 * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC)
1478 * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC)
1479 *
1480 * In other words if current stack state (one being explored) has more
1481 * valid slots than old one that already passed validation, it means
1482 * the verifier can stop exploring and conclude that current state is valid too
1483 *
1484 * Similarly with registers. If explored state has register type as invalid
1485 * whereas register type in current state is meaningful, it means that
1486 * the current state will reach 'bpf_exit' instruction safely
1487 */
states_equal(struct verifier_state * old,struct verifier_state * cur)1488 static bool states_equal(struct verifier_state *old, struct verifier_state *cur)
1489 {
1490 int i;
1491
1492 for (i = 0; i < MAX_BPF_REG; i++) {
1493 if (memcmp(&old->regs[i], &cur->regs[i],
1494 sizeof(old->regs[0])) != 0) {
1495 if (old->regs[i].type == NOT_INIT ||
1496 (old->regs[i].type == UNKNOWN_VALUE &&
1497 cur->regs[i].type != NOT_INIT))
1498 continue;
1499 return false;
1500 }
1501 }
1502
1503 for (i = 0; i < MAX_BPF_STACK; i++) {
1504 if (old->stack_slot_type[i] == STACK_INVALID)
1505 continue;
1506 if (old->stack_slot_type[i] != cur->stack_slot_type[i])
1507 /* Ex: old explored (safe) state has STACK_SPILL in
1508 * this stack slot, but current has has STACK_MISC ->
1509 * this verifier states are not equivalent,
1510 * return false to continue verification of this path
1511 */
1512 return false;
1513 if (i % BPF_REG_SIZE)
1514 continue;
1515 if (memcmp(&old->spilled_regs[i / BPF_REG_SIZE],
1516 &cur->spilled_regs[i / BPF_REG_SIZE],
1517 sizeof(old->spilled_regs[0])))
1518 /* when explored and current stack slot types are
1519 * the same, check that stored pointers types
1520 * are the same as well.
1521 * Ex: explored safe path could have stored
1522 * (struct reg_state) {.type = PTR_TO_STACK, .imm = -8}
1523 * but current path has stored:
1524 * (struct reg_state) {.type = PTR_TO_STACK, .imm = -16}
1525 * such verifier states are not equivalent.
1526 * return false to continue verification of this path
1527 */
1528 return false;
1529 else
1530 continue;
1531 }
1532 return true;
1533 }
1534
is_state_visited(struct verifier_env * env,int insn_idx)1535 static int is_state_visited(struct verifier_env *env, int insn_idx)
1536 {
1537 struct verifier_state_list *new_sl;
1538 struct verifier_state_list *sl;
1539
1540 sl = env->explored_states[insn_idx];
1541 if (!sl)
1542 /* this 'insn_idx' instruction wasn't marked, so we will not
1543 * be doing state search here
1544 */
1545 return 0;
1546
1547 while (sl != STATE_LIST_MARK) {
1548 if (states_equal(&sl->state, &env->cur_state))
1549 /* reached equivalent register/stack state,
1550 * prune the search
1551 */
1552 return 1;
1553 sl = sl->next;
1554 }
1555
1556 /* there were no equivalent states, remember current one.
1557 * technically the current state is not proven to be safe yet,
1558 * but it will either reach bpf_exit (which means it's safe) or
1559 * it will be rejected. Since there are no loops, we won't be
1560 * seeing this 'insn_idx' instruction again on the way to bpf_exit
1561 */
1562 new_sl = kmalloc(sizeof(struct verifier_state_list), GFP_USER);
1563 if (!new_sl)
1564 return -ENOMEM;
1565
1566 /* add new state to the head of linked list */
1567 memcpy(&new_sl->state, &env->cur_state, sizeof(env->cur_state));
1568 new_sl->next = env->explored_states[insn_idx];
1569 env->explored_states[insn_idx] = new_sl;
1570 return 0;
1571 }
1572
do_check(struct verifier_env * env)1573 static int do_check(struct verifier_env *env)
1574 {
1575 struct verifier_state *state = &env->cur_state;
1576 struct bpf_insn *insns = env->prog->insnsi;
1577 struct reg_state *regs = state->regs;
1578 int insn_cnt = env->prog->len;
1579 int insn_idx, prev_insn_idx = 0;
1580 int insn_processed = 0;
1581 bool do_print_state = false;
1582
1583 init_reg_state(regs);
1584 insn_idx = 0;
1585 for (;;) {
1586 struct bpf_insn *insn;
1587 u8 class;
1588 int err;
1589
1590 if (insn_idx >= insn_cnt) {
1591 verbose("invalid insn idx %d insn_cnt %d\n",
1592 insn_idx, insn_cnt);
1593 return -EFAULT;
1594 }
1595
1596 insn = &insns[insn_idx];
1597 class = BPF_CLASS(insn->code);
1598
1599 if (++insn_processed > 32768) {
1600 verbose("BPF program is too large. Proccessed %d insn\n",
1601 insn_processed);
1602 return -E2BIG;
1603 }
1604
1605 err = is_state_visited(env, insn_idx);
1606 if (err < 0)
1607 return err;
1608 if (err == 1) {
1609 /* found equivalent state, can prune the search */
1610 if (log_level) {
1611 if (do_print_state)
1612 verbose("\nfrom %d to %d: safe\n",
1613 prev_insn_idx, insn_idx);
1614 else
1615 verbose("%d: safe\n", insn_idx);
1616 }
1617 goto process_bpf_exit;
1618 }
1619
1620 if (log_level && do_print_state) {
1621 verbose("\nfrom %d to %d:", prev_insn_idx, insn_idx);
1622 print_verifier_state(env);
1623 do_print_state = false;
1624 }
1625
1626 if (log_level) {
1627 verbose("%d: ", insn_idx);
1628 print_bpf_insn(insn);
1629 }
1630
1631 if (class == BPF_ALU || class == BPF_ALU64) {
1632 err = check_alu_op(regs, insn);
1633 if (err)
1634 return err;
1635
1636 } else if (class == BPF_LDX) {
1637 enum bpf_reg_type src_reg_type;
1638
1639 /* check for reserved fields is already done */
1640
1641 /* check src operand */
1642 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1643 if (err)
1644 return err;
1645
1646 err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK);
1647 if (err)
1648 return err;
1649
1650 src_reg_type = regs[insn->src_reg].type;
1651
1652 /* check that memory (src_reg + off) is readable,
1653 * the state of dst_reg will be updated by this func
1654 */
1655 err = check_mem_access(env, insn->src_reg, insn->off,
1656 BPF_SIZE(insn->code), BPF_READ,
1657 insn->dst_reg);
1658 if (err)
1659 return err;
1660
1661 if (BPF_SIZE(insn->code) != BPF_W) {
1662 insn_idx++;
1663 continue;
1664 }
1665
1666 if (insn->imm == 0) {
1667 /* saw a valid insn
1668 * dst_reg = *(u32 *)(src_reg + off)
1669 * use reserved 'imm' field to mark this insn
1670 */
1671 insn->imm = src_reg_type;
1672
1673 } else if (src_reg_type != insn->imm &&
1674 (src_reg_type == PTR_TO_CTX ||
1675 insn->imm == PTR_TO_CTX)) {
1676 /* ABuser program is trying to use the same insn
1677 * dst_reg = *(u32*) (src_reg + off)
1678 * with different pointer types:
1679 * src_reg == ctx in one branch and
1680 * src_reg == stack|map in some other branch.
1681 * Reject it.
1682 */
1683 verbose("same insn cannot be used with different pointers\n");
1684 return -EINVAL;
1685 }
1686
1687 } else if (class == BPF_STX) {
1688 if (BPF_MODE(insn->code) == BPF_XADD) {
1689 err = check_xadd(env, insn);
1690 if (err)
1691 return err;
1692 insn_idx++;
1693 continue;
1694 }
1695
1696 if (BPF_MODE(insn->code) != BPF_MEM ||
1697 insn->imm != 0) {
1698 verbose("BPF_STX uses reserved fields\n");
1699 return -EINVAL;
1700 }
1701 /* check src1 operand */
1702 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1703 if (err)
1704 return err;
1705 /* check src2 operand */
1706 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1707 if (err)
1708 return err;
1709
1710 /* check that memory (dst_reg + off) is writeable */
1711 err = check_mem_access(env, insn->dst_reg, insn->off,
1712 BPF_SIZE(insn->code), BPF_WRITE,
1713 insn->src_reg);
1714 if (err)
1715 return err;
1716
1717 } else if (class == BPF_ST) {
1718 if (BPF_MODE(insn->code) != BPF_MEM ||
1719 insn->src_reg != BPF_REG_0) {
1720 verbose("BPF_ST uses reserved fields\n");
1721 return -EINVAL;
1722 }
1723 /* check src operand */
1724 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1725 if (err)
1726 return err;
1727
1728 /* check that memory (dst_reg + off) is writeable */
1729 err = check_mem_access(env, insn->dst_reg, insn->off,
1730 BPF_SIZE(insn->code), BPF_WRITE,
1731 -1);
1732 if (err)
1733 return err;
1734
1735 } else if (class == BPF_JMP) {
1736 u8 opcode = BPF_OP(insn->code);
1737
1738 if (opcode == BPF_CALL) {
1739 if (BPF_SRC(insn->code) != BPF_K ||
1740 insn->off != 0 ||
1741 insn->src_reg != BPF_REG_0 ||
1742 insn->dst_reg != BPF_REG_0) {
1743 verbose("BPF_CALL uses reserved fields\n");
1744 return -EINVAL;
1745 }
1746
1747 err = check_call(env, insn->imm);
1748 if (err)
1749 return err;
1750
1751 } else if (opcode == BPF_JA) {
1752 if (BPF_SRC(insn->code) != BPF_K ||
1753 insn->imm != 0 ||
1754 insn->src_reg != BPF_REG_0 ||
1755 insn->dst_reg != BPF_REG_0) {
1756 verbose("BPF_JA uses reserved fields\n");
1757 return -EINVAL;
1758 }
1759
1760 insn_idx += insn->off + 1;
1761 continue;
1762
1763 } else if (opcode == BPF_EXIT) {
1764 if (BPF_SRC(insn->code) != BPF_K ||
1765 insn->imm != 0 ||
1766 insn->src_reg != BPF_REG_0 ||
1767 insn->dst_reg != BPF_REG_0) {
1768 verbose("BPF_EXIT uses reserved fields\n");
1769 return -EINVAL;
1770 }
1771
1772 /* eBPF calling convetion is such that R0 is used
1773 * to return the value from eBPF program.
1774 * Make sure that it's readable at this time
1775 * of bpf_exit, which means that program wrote
1776 * something into it earlier
1777 */
1778 err = check_reg_arg(regs, BPF_REG_0, SRC_OP);
1779 if (err)
1780 return err;
1781
1782 process_bpf_exit:
1783 insn_idx = pop_stack(env, &prev_insn_idx);
1784 if (insn_idx < 0) {
1785 break;
1786 } else {
1787 do_print_state = true;
1788 continue;
1789 }
1790 } else {
1791 err = check_cond_jmp_op(env, insn, &insn_idx);
1792 if (err)
1793 return err;
1794 }
1795 } else if (class == BPF_LD) {
1796 u8 mode = BPF_MODE(insn->code);
1797
1798 if (mode == BPF_ABS || mode == BPF_IND) {
1799 err = check_ld_abs(env, insn);
1800 if (err)
1801 return err;
1802
1803 } else if (mode == BPF_IMM) {
1804 err = check_ld_imm(env, insn);
1805 if (err)
1806 return err;
1807
1808 insn_idx++;
1809 } else {
1810 verbose("invalid BPF_LD mode\n");
1811 return -EINVAL;
1812 }
1813 } else {
1814 verbose("unknown insn class %d\n", class);
1815 return -EINVAL;
1816 }
1817
1818 insn_idx++;
1819 }
1820
1821 return 0;
1822 }
1823
1824 /* look for pseudo eBPF instructions that access map FDs and
1825 * replace them with actual map pointers
1826 */
replace_map_fd_with_map_ptr(struct verifier_env * env)1827 static int replace_map_fd_with_map_ptr(struct verifier_env *env)
1828 {
1829 struct bpf_insn *insn = env->prog->insnsi;
1830 int insn_cnt = env->prog->len;
1831 int i, j;
1832
1833 for (i = 0; i < insn_cnt; i++, insn++) {
1834 if (BPF_CLASS(insn->code) == BPF_LDX &&
1835 (BPF_MODE(insn->code) != BPF_MEM ||
1836 insn->imm != 0)) {
1837 verbose("BPF_LDX uses reserved fields\n");
1838 return -EINVAL;
1839 }
1840
1841 if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW)) {
1842 struct bpf_map *map;
1843 struct fd f;
1844
1845 if (i == insn_cnt - 1 || insn[1].code != 0 ||
1846 insn[1].dst_reg != 0 || insn[1].src_reg != 0 ||
1847 insn[1].off != 0) {
1848 verbose("invalid bpf_ld_imm64 insn\n");
1849 return -EINVAL;
1850 }
1851
1852 if (insn->src_reg == 0)
1853 /* valid generic load 64-bit imm */
1854 goto next_insn;
1855
1856 if (insn->src_reg != BPF_PSEUDO_MAP_FD) {
1857 verbose("unrecognized bpf_ld_imm64 insn\n");
1858 return -EINVAL;
1859 }
1860
1861 f = fdget(insn->imm);
1862
1863 map = bpf_map_get(f);
1864 if (IS_ERR(map)) {
1865 verbose("fd %d is not pointing to valid bpf_map\n",
1866 insn->imm);
1867 fdput(f);
1868 return PTR_ERR(map);
1869 }
1870
1871 /* store map pointer inside BPF_LD_IMM64 instruction */
1872 insn[0].imm = (u32) (unsigned long) map;
1873 insn[1].imm = ((u64) (unsigned long) map) >> 32;
1874
1875 /* check whether we recorded this map already */
1876 for (j = 0; j < env->used_map_cnt; j++)
1877 if (env->used_maps[j] == map) {
1878 fdput(f);
1879 goto next_insn;
1880 }
1881
1882 if (env->used_map_cnt >= MAX_USED_MAPS) {
1883 fdput(f);
1884 return -E2BIG;
1885 }
1886
1887 /* remember this map */
1888 env->used_maps[env->used_map_cnt++] = map;
1889
1890 /* hold the map. If the program is rejected by verifier,
1891 * the map will be released by release_maps() or it
1892 * will be used by the valid program until it's unloaded
1893 * and all maps are released in free_bpf_prog_info()
1894 */
1895 atomic_inc(&map->refcnt);
1896
1897 fdput(f);
1898 next_insn:
1899 insn++;
1900 i++;
1901 }
1902 }
1903
1904 /* now all pseudo BPF_LD_IMM64 instructions load valid
1905 * 'struct bpf_map *' into a register instead of user map_fd.
1906 * These pointers will be used later by verifier to validate map access.
1907 */
1908 return 0;
1909 }
1910
1911 /* drop refcnt of maps used by the rejected program */
release_maps(struct verifier_env * env)1912 static void release_maps(struct verifier_env *env)
1913 {
1914 int i;
1915
1916 for (i = 0; i < env->used_map_cnt; i++)
1917 bpf_map_put(env->used_maps[i]);
1918 }
1919
1920 /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */
convert_pseudo_ld_imm64(struct verifier_env * env)1921 static void convert_pseudo_ld_imm64(struct verifier_env *env)
1922 {
1923 struct bpf_insn *insn = env->prog->insnsi;
1924 int insn_cnt = env->prog->len;
1925 int i;
1926
1927 for (i = 0; i < insn_cnt; i++, insn++)
1928 if (insn->code == (BPF_LD | BPF_IMM | BPF_DW))
1929 insn->src_reg = 0;
1930 }
1931
adjust_branches(struct bpf_prog * prog,int pos,int delta)1932 static void adjust_branches(struct bpf_prog *prog, int pos, int delta)
1933 {
1934 struct bpf_insn *insn = prog->insnsi;
1935 int insn_cnt = prog->len;
1936 int i;
1937
1938 for (i = 0; i < insn_cnt; i++, insn++) {
1939 if (BPF_CLASS(insn->code) != BPF_JMP ||
1940 BPF_OP(insn->code) == BPF_CALL ||
1941 BPF_OP(insn->code) == BPF_EXIT)
1942 continue;
1943
1944 /* adjust offset of jmps if necessary */
1945 if (i < pos && i + insn->off + 1 > pos)
1946 insn->off += delta;
1947 else if (i > pos + delta && i + insn->off + 1 <= pos + delta)
1948 insn->off -= delta;
1949 }
1950 }
1951
1952 /* convert load instructions that access fields of 'struct __sk_buff'
1953 * into sequence of instructions that access fields of 'struct sk_buff'
1954 */
convert_ctx_accesses(struct verifier_env * env)1955 static int convert_ctx_accesses(struct verifier_env *env)
1956 {
1957 struct bpf_insn *insn = env->prog->insnsi;
1958 int insn_cnt = env->prog->len;
1959 struct bpf_insn insn_buf[16];
1960 struct bpf_prog *new_prog;
1961 u32 cnt;
1962 int i;
1963
1964 if (!env->prog->aux->ops->convert_ctx_access)
1965 return 0;
1966
1967 for (i = 0; i < insn_cnt; i++, insn++) {
1968 if (insn->code != (BPF_LDX | BPF_MEM | BPF_W))
1969 continue;
1970
1971 if (insn->imm != PTR_TO_CTX) {
1972 /* clear internal mark */
1973 insn->imm = 0;
1974 continue;
1975 }
1976
1977 cnt = env->prog->aux->ops->
1978 convert_ctx_access(insn->dst_reg, insn->src_reg,
1979 insn->off, insn_buf);
1980 if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) {
1981 verbose("bpf verifier is misconfigured\n");
1982 return -EINVAL;
1983 }
1984
1985 if (cnt == 1) {
1986 memcpy(insn, insn_buf, sizeof(*insn));
1987 continue;
1988 }
1989
1990 /* several new insns need to be inserted. Make room for them */
1991 insn_cnt += cnt - 1;
1992 new_prog = bpf_prog_realloc(env->prog,
1993 bpf_prog_size(insn_cnt),
1994 GFP_USER);
1995 if (!new_prog)
1996 return -ENOMEM;
1997
1998 new_prog->len = insn_cnt;
1999
2000 memmove(new_prog->insnsi + i + cnt, new_prog->insns + i + 1,
2001 sizeof(*insn) * (insn_cnt - i - cnt));
2002
2003 /* copy substitute insns in place of load instruction */
2004 memcpy(new_prog->insnsi + i, insn_buf, sizeof(*insn) * cnt);
2005
2006 /* adjust branches in the whole program */
2007 adjust_branches(new_prog, i, cnt - 1);
2008
2009 /* keep walking new program and skip insns we just inserted */
2010 env->prog = new_prog;
2011 insn = new_prog->insnsi + i + cnt - 1;
2012 i += cnt - 1;
2013 }
2014
2015 return 0;
2016 }
2017
free_states(struct verifier_env * env)2018 static void free_states(struct verifier_env *env)
2019 {
2020 struct verifier_state_list *sl, *sln;
2021 int i;
2022
2023 if (!env->explored_states)
2024 return;
2025
2026 for (i = 0; i < env->prog->len; i++) {
2027 sl = env->explored_states[i];
2028
2029 if (sl)
2030 while (sl != STATE_LIST_MARK) {
2031 sln = sl->next;
2032 kfree(sl);
2033 sl = sln;
2034 }
2035 }
2036
2037 kfree(env->explored_states);
2038 }
2039
bpf_check(struct bpf_prog ** prog,union bpf_attr * attr)2040 int bpf_check(struct bpf_prog **prog, union bpf_attr *attr)
2041 {
2042 char __user *log_ubuf = NULL;
2043 struct verifier_env *env;
2044 int ret = -EINVAL;
2045
2046 if ((*prog)->len <= 0 || (*prog)->len > BPF_MAXINSNS)
2047 return -E2BIG;
2048
2049 /* 'struct verifier_env' can be global, but since it's not small,
2050 * allocate/free it every time bpf_check() is called
2051 */
2052 env = kzalloc(sizeof(struct verifier_env), GFP_KERNEL);
2053 if (!env)
2054 return -ENOMEM;
2055
2056 env->prog = *prog;
2057
2058 /* grab the mutex to protect few globals used by verifier */
2059 mutex_lock(&bpf_verifier_lock);
2060
2061 if (attr->log_level || attr->log_buf || attr->log_size) {
2062 /* user requested verbose verifier output
2063 * and supplied buffer to store the verification trace
2064 */
2065 log_level = attr->log_level;
2066 log_ubuf = (char __user *) (unsigned long) attr->log_buf;
2067 log_size = attr->log_size;
2068 log_len = 0;
2069
2070 ret = -EINVAL;
2071 /* log_* values have to be sane */
2072 if (log_size < 128 || log_size > UINT_MAX >> 8 ||
2073 log_level == 0 || log_ubuf == NULL)
2074 goto free_env;
2075
2076 ret = -ENOMEM;
2077 log_buf = vmalloc(log_size);
2078 if (!log_buf)
2079 goto free_env;
2080 } else {
2081 log_level = 0;
2082 }
2083
2084 ret = replace_map_fd_with_map_ptr(env);
2085 if (ret < 0)
2086 goto skip_full_check;
2087
2088 env->explored_states = kcalloc(env->prog->len,
2089 sizeof(struct verifier_state_list *),
2090 GFP_USER);
2091 ret = -ENOMEM;
2092 if (!env->explored_states)
2093 goto skip_full_check;
2094
2095 ret = check_cfg(env);
2096 if (ret < 0)
2097 goto skip_full_check;
2098
2099 ret = do_check(env);
2100
2101 skip_full_check:
2102 while (pop_stack(env, NULL) >= 0);
2103 free_states(env);
2104
2105 if (ret == 0)
2106 /* program is valid, convert *(u32*)(ctx + off) accesses */
2107 ret = convert_ctx_accesses(env);
2108
2109 if (log_level && log_len >= log_size - 1) {
2110 BUG_ON(log_len >= log_size);
2111 /* verifier log exceeded user supplied buffer */
2112 ret = -ENOSPC;
2113 /* fall through to return what was recorded */
2114 }
2115
2116 /* copy verifier log back to user space including trailing zero */
2117 if (log_level && copy_to_user(log_ubuf, log_buf, log_len + 1) != 0) {
2118 ret = -EFAULT;
2119 goto free_log_buf;
2120 }
2121
2122 if (ret == 0 && env->used_map_cnt) {
2123 /* if program passed verifier, update used_maps in bpf_prog_info */
2124 env->prog->aux->used_maps = kmalloc_array(env->used_map_cnt,
2125 sizeof(env->used_maps[0]),
2126 GFP_KERNEL);
2127
2128 if (!env->prog->aux->used_maps) {
2129 ret = -ENOMEM;
2130 goto free_log_buf;
2131 }
2132
2133 memcpy(env->prog->aux->used_maps, env->used_maps,
2134 sizeof(env->used_maps[0]) * env->used_map_cnt);
2135 env->prog->aux->used_map_cnt = env->used_map_cnt;
2136
2137 /* program is valid. Convert pseudo bpf_ld_imm64 into generic
2138 * bpf_ld_imm64 instructions
2139 */
2140 convert_pseudo_ld_imm64(env);
2141 }
2142
2143 free_log_buf:
2144 if (log_level)
2145 vfree(log_buf);
2146 free_env:
2147 if (!env->prog->aux->used_maps)
2148 /* if we didn't copy map pointers into bpf_prog_info, release
2149 * them now. Otherwise free_bpf_prog_info() will release them.
2150 */
2151 release_maps(env);
2152 *prog = env->prog;
2153 kfree(env);
2154 mutex_unlock(&bpf_verifier_lock);
2155 return ret;
2156 }
2157