]> git.proxmox.com Git - mirror_ubuntu-eoan-kernel.git/blob - kernel/bpf/verifier.c
projects / mirror_ubuntu-eoan-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
[mirror_ubuntu-eoan-kernel.git] / kernel / bpf / verifier.c
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 long 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 bool allow_ptr_leaks;
203 };
204
205 #define BPF_COMPLEXITY_LIMIT_INSNS 65536
206 #define BPF_COMPLEXITY_LIMIT_STACK 1024
207
208 struct bpf_call_arg_meta {
209 struct bpf_map *map_ptr;
210 bool raw_mode;
211 int regno;
212 int access_size;
213 };
214
215 /* verbose verifier prints what it's seeing
216 * bpf_check() is called under lock, so no race to access these global vars
217 */
218 static u32 log_level, log_size, log_len;
219 static char *log_buf;
220
221 static DEFINE_MUTEX(bpf_verifier_lock);
222
223 /* log_level controls verbosity level of eBPF verifier.
224 * verbose() is used to dump the verification trace to the log, so the user
225 * can figure out what's wrong with the program
226 */
227 static __printf(1, 2) void verbose(const char *fmt, ...)
228 {
229 va_list args;
230
231 if (log_level == 0 || log_len >= log_size - 1)
232 return;
233
234 va_start(args, fmt);
235 log_len += vscnprintf(log_buf + log_len, log_size - log_len, fmt, args);
236 va_end(args);
237 }
238
239 /* string representation of 'enum bpf_reg_type' */
240 static const char * const reg_type_str[] = {
241 [NOT_INIT] = "?",
242 [UNKNOWN_VALUE] = "inv",
243 [PTR_TO_CTX] = "ctx",
244 [CONST_PTR_TO_MAP] = "map_ptr",
245 [PTR_TO_MAP_VALUE] = "map_value",
246 [PTR_TO_MAP_VALUE_OR_NULL] = "map_value_or_null",
247 [FRAME_PTR] = "fp",
248 [PTR_TO_STACK] = "fp",
249 [CONST_IMM] = "imm",
250 };
251
252 static const struct {
253 int map_type;
254 int func_id;
255 } func_limit[] = {
256 {BPF_MAP_TYPE_PROG_ARRAY, BPF_FUNC_tail_call},
257 {BPF_MAP_TYPE_PERF_EVENT_ARRAY, BPF_FUNC_perf_event_read},
258 {BPF_MAP_TYPE_PERF_EVENT_ARRAY, BPF_FUNC_perf_event_output},
259 {BPF_MAP_TYPE_STACK_TRACE, BPF_FUNC_get_stackid},
260 };
261
262 static void print_verifier_state(struct verifier_env *env)
263 {
264 enum bpf_reg_type t;
265 int i;
266
267 for (i = 0; i < MAX_BPF_REG; i++) {
268 t = env->cur_state.regs[i].type;
269 if (t == NOT_INIT)
270 continue;
271 verbose(" R%d=%s", i, reg_type_str[t]);
272 if (t == CONST_IMM || t == PTR_TO_STACK)
273 verbose("%ld", env->cur_state.regs[i].imm);
274 else if (t == CONST_PTR_TO_MAP || t == PTR_TO_MAP_VALUE ||
275 t == PTR_TO_MAP_VALUE_OR_NULL)
276 verbose("(ks=%d,vs=%d)",
277 env->cur_state.regs[i].map_ptr->key_size,
278 env->cur_state.regs[i].map_ptr->value_size);
279 }
280 for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
281 if (env->cur_state.stack_slot_type[i] == STACK_SPILL)
282 verbose(" fp%d=%s", -MAX_BPF_STACK + i,
283 reg_type_str[env->cur_state.spilled_regs[i / BPF_REG_SIZE].type]);
284 }
285 verbose("\n");
286 }
287
288 static const char *const bpf_class_string[] = {
289 [BPF_LD] = "ld",
290 [BPF_LDX] = "ldx",
291 [BPF_ST] = "st",
292 [BPF_STX] = "stx",
293 [BPF_ALU] = "alu",
294 [BPF_JMP] = "jmp",
295 [BPF_RET] = "BUG",
296 [BPF_ALU64] = "alu64",
297 };
298
299 static const char *const bpf_alu_string[16] = {
300 [BPF_ADD >> 4] = "+=",
301 [BPF_SUB >> 4] = "-=",
302 [BPF_MUL >> 4] = "*=",
303 [BPF_DIV >> 4] = "/=",
304 [BPF_OR >> 4] = "|=",
305 [BPF_AND >> 4] = "&=",
306 [BPF_LSH >> 4] = "<<=",
307 [BPF_RSH >> 4] = ">>=",
308 [BPF_NEG >> 4] = "neg",
309 [BPF_MOD >> 4] = "%=",
310 [BPF_XOR >> 4] = "^=",
311 [BPF_MOV >> 4] = "=",
312 [BPF_ARSH >> 4] = "s>>=",
313 [BPF_END >> 4] = "endian",
314 };
315
316 static const char *const bpf_ldst_string[] = {
317 [BPF_W >> 3] = "u32",
318 [BPF_H >> 3] = "u16",
319 [BPF_B >> 3] = "u8",
320 [BPF_DW >> 3] = "u64",
321 };
322
323 static const char *const bpf_jmp_string[16] = {
324 [BPF_JA >> 4] = "jmp",
325 [BPF_JEQ >> 4] = "==",
326 [BPF_JGT >> 4] = ">",
327 [BPF_JGE >> 4] = ">=",
328 [BPF_JSET >> 4] = "&",
329 [BPF_JNE >> 4] = "!=",
330 [BPF_JSGT >> 4] = "s>",
331 [BPF_JSGE >> 4] = "s>=",
332 [BPF_CALL >> 4] = "call",
333 [BPF_EXIT >> 4] = "exit",
334 };
335
336 static void print_bpf_insn(struct bpf_insn *insn)
337 {
338 u8 class = BPF_CLASS(insn->code);
339
340 if (class == BPF_ALU || class == BPF_ALU64) {
341 if (BPF_SRC(insn->code) == BPF_X)
342 verbose("(%02x) %sr%d %s %sr%d\n",
343 insn->code, class == BPF_ALU ? "(u32) " : "",
344 insn->dst_reg,
345 bpf_alu_string[BPF_OP(insn->code) >> 4],
346 class == BPF_ALU ? "(u32) " : "",
347 insn->src_reg);
348 else
349 verbose("(%02x) %sr%d %s %s%d\n",
350 insn->code, class == BPF_ALU ? "(u32) " : "",
351 insn->dst_reg,
352 bpf_alu_string[BPF_OP(insn->code) >> 4],
353 class == BPF_ALU ? "(u32) " : "",
354 insn->imm);
355 } else if (class == BPF_STX) {
356 if (BPF_MODE(insn->code) == BPF_MEM)
357 verbose("(%02x) *(%s *)(r%d %+d) = r%d\n",
358 insn->code,
359 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
360 insn->dst_reg,
361 insn->off, insn->src_reg);
362 else if (BPF_MODE(insn->code) == BPF_XADD)
363 verbose("(%02x) lock *(%s *)(r%d %+d) += r%d\n",
364 insn->code,
365 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
366 insn->dst_reg, insn->off,
367 insn->src_reg);
368 else
369 verbose("BUG_%02x\n", insn->code);
370 } else if (class == BPF_ST) {
371 if (BPF_MODE(insn->code) != BPF_MEM) {
372 verbose("BUG_st_%02x\n", insn->code);
373 return;
374 }
375 verbose("(%02x) *(%s *)(r%d %+d) = %d\n",
376 insn->code,
377 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
378 insn->dst_reg,
379 insn->off, insn->imm);
380 } else if (class == BPF_LDX) {
381 if (BPF_MODE(insn->code) != BPF_MEM) {
382 verbose("BUG_ldx_%02x\n", insn->code);
383 return;
384 }
385 verbose("(%02x) r%d = *(%s *)(r%d %+d)\n",
386 insn->code, insn->dst_reg,
387 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
388 insn->src_reg, insn->off);
389 } else if (class == BPF_LD) {
390 if (BPF_MODE(insn->code) == BPF_ABS) {
391 verbose("(%02x) r0 = *(%s *)skb[%d]\n",
392 insn->code,
393 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
394 insn->imm);
395 } else if (BPF_MODE(insn->code) == BPF_IND) {
396 verbose("(%02x) r0 = *(%s *)skb[r%d + %d]\n",
397 insn->code,
398 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
399 insn->src_reg, insn->imm);
400 } else if (BPF_MODE(insn->code) == BPF_IMM) {
401 verbose("(%02x) r%d = 0x%x\n",
402 insn->code, insn->dst_reg, insn->imm);
403 } else {
404 verbose("BUG_ld_%02x\n", insn->code);
405 return;
406 }
407 } else if (class == BPF_JMP) {
408 u8 opcode = BPF_OP(insn->code);
409
410 if (opcode == BPF_CALL) {
411 verbose("(%02x) call %d\n", insn->code, insn->imm);
412 } else if (insn->code == (BPF_JMP | BPF_JA)) {
413 verbose("(%02x) goto pc%+d\n",
414 insn->code, insn->off);
415 } else if (insn->code == (BPF_JMP | BPF_EXIT)) {
416 verbose("(%02x) exit\n", insn->code);
417 } else if (BPF_SRC(insn->code) == BPF_X) {
418 verbose("(%02x) if r%d %s r%d goto pc%+d\n",
419 insn->code, insn->dst_reg,
420 bpf_jmp_string[BPF_OP(insn->code) >> 4],
421 insn->src_reg, insn->off);
422 } else {
423 verbose("(%02x) if r%d %s 0x%x goto pc%+d\n",
424 insn->code, insn->dst_reg,
425 bpf_jmp_string[BPF_OP(insn->code) >> 4],
426 insn->imm, insn->off);
427 }
428 } else {
429 verbose("(%02x) %s\n", insn->code, bpf_class_string[class]);
430 }
431 }
432
433 static int pop_stack(struct verifier_env *env, int *prev_insn_idx)
434 {
435 struct verifier_stack_elem *elem;
436 int insn_idx;
437
438 if (env->head == NULL)
439 return -1;
440
441 memcpy(&env->cur_state, &env->head->st, sizeof(env->cur_state));
442 insn_idx = env->head->insn_idx;
443 if (prev_insn_idx)
444 *prev_insn_idx = env->head->prev_insn_idx;
445 elem = env->head->next;
446 kfree(env->head);
447 env->head = elem;
448 env->stack_size--;
449 return insn_idx;
450 }
451
452 static struct verifier_state *push_stack(struct verifier_env *env, int insn_idx,
453 int prev_insn_idx)
454 {
455 struct verifier_stack_elem *elem;
456
457 elem = kmalloc(sizeof(struct verifier_stack_elem), GFP_KERNEL);
458 if (!elem)
459 goto err;
460
461 memcpy(&elem->st, &env->cur_state, sizeof(env->cur_state));
462 elem->insn_idx = insn_idx;
463 elem->prev_insn_idx = prev_insn_idx;
464 elem->next = env->head;
465 env->head = elem;
466 env->stack_size++;
467 if (env->stack_size > BPF_COMPLEXITY_LIMIT_STACK) {
468 verbose("BPF program is too complex\n");
469 goto err;
470 }
471 return &elem->st;
472 err:
473 /* pop all elements and return */
474 while (pop_stack(env, NULL) >= 0);
475 return NULL;
476 }
477
478 #define CALLER_SAVED_REGS 6
479 static const int caller_saved[CALLER_SAVED_REGS] = {
480 BPF_REG_0, BPF_REG_1, BPF_REG_2, BPF_REG_3, BPF_REG_4, BPF_REG_5
481 };
482
483 static void init_reg_state(struct reg_state *regs)
484 {
485 int i;
486
487 for (i = 0; i < MAX_BPF_REG; i++) {
488 regs[i].type = NOT_INIT;
489 regs[i].imm = 0;
490 }
491
492 /* frame pointer */
493 regs[BPF_REG_FP].type = FRAME_PTR;
494
495 /* 1st arg to a function */
496 regs[BPF_REG_1].type = PTR_TO_CTX;
497 }
498
499 static void mark_reg_unknown_value(struct reg_state *regs, u32 regno)
500 {
501 BUG_ON(regno >= MAX_BPF_REG);
502 regs[regno].type = UNKNOWN_VALUE;
503 regs[regno].imm = 0;
504 }
505
506 enum reg_arg_type {
507 SRC_OP, /* register is used as source operand */
508 DST_OP, /* register is used as destination operand */
509 DST_OP_NO_MARK /* same as above, check only, don't mark */
510 };
511
512 static int check_reg_arg(struct reg_state *regs, u32 regno,
513 enum reg_arg_type t)
514 {
515 if (regno >= MAX_BPF_REG) {
516 verbose("R%d is invalid\n", regno);
517 return -EINVAL;
518 }
519
520 if (t == SRC_OP) {
521 /* check whether register used as source operand can be read */
522 if (regs[regno].type == NOT_INIT) {
523 verbose("R%d !read_ok\n", regno);
524 return -EACCES;
525 }
526 } else {
527 /* check whether register used as dest operand can be written to */
528 if (regno == BPF_REG_FP) {
529 verbose("frame pointer is read only\n");
530 return -EACCES;
531 }
532 if (t == DST_OP)
533 mark_reg_unknown_value(regs, regno);
534 }
535 return 0;
536 }
537
538 static int bpf_size_to_bytes(int bpf_size)
539 {
540 if (bpf_size == BPF_W)
541 return 4;
542 else if (bpf_size == BPF_H)
543 return 2;
544 else if (bpf_size == BPF_B)
545 return 1;
546 else if (bpf_size == BPF_DW)
547 return 8;
548 else
549 return -EINVAL;
550 }
551
552 static bool is_spillable_regtype(enum bpf_reg_type type)
553 {
554 switch (type) {
555 case PTR_TO_MAP_VALUE:
556 case PTR_TO_MAP_VALUE_OR_NULL:
557 case PTR_TO_STACK:
558 case PTR_TO_CTX:
559 case FRAME_PTR:
560 case CONST_PTR_TO_MAP:
561 return true;
562 default:
563 return false;
564 }
565 }
566
567 /* check_stack_read/write functions track spill/fill of registers,
568 * stack boundary and alignment are checked in check_mem_access()
569 */
570 static int check_stack_write(struct verifier_state *state, int off, int size,
571 int value_regno)
572 {
573 int i;
574 /* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0,
575 * so it's aligned access and [off, off + size) are within stack limits
576 */
577
578 if (value_regno >= 0 &&
579 is_spillable_regtype(state->regs[value_regno].type)) {
580
581 /* register containing pointer is being spilled into stack */
582 if (size != BPF_REG_SIZE) {
583 verbose("invalid size of register spill\n");
584 return -EACCES;
585 }
586
587 /* save register state */
588 state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] =
589 state->regs[value_regno];
590
591 for (i = 0; i < BPF_REG_SIZE; i++)
592 state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_SPILL;
593 } else {
594 /* regular write of data into stack */
595 state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] =
596 (struct reg_state) {};
597
598 for (i = 0; i < size; i++)
599 state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_MISC;
600 }
601 return 0;
602 }
603
604 static int check_stack_read(struct verifier_state *state, int off, int size,
605 int value_regno)
606 {
607 u8 *slot_type;
608 int i;
609
610 slot_type = &state->stack_slot_type[MAX_BPF_STACK + off];
611
612 if (slot_type[0] == STACK_SPILL) {
613 if (size != BPF_REG_SIZE) {
614 verbose("invalid size of register spill\n");
615 return -EACCES;
616 }
617 for (i = 1; i < BPF_REG_SIZE; i++) {
618 if (slot_type[i] != STACK_SPILL) {
619 verbose("corrupted spill memory\n");
620 return -EACCES;
621 }
622 }
623
624 if (value_regno >= 0)
625 /* restore register state from stack */
626 state->regs[value_regno] =
627 state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE];
628 return 0;
629 } else {
630 for (i = 0; i < size; i++) {
631 if (slot_type[i] != STACK_MISC) {
632 verbose("invalid read from stack off %d+%d size %d\n",
633 off, i, size);
634 return -EACCES;
635 }
636 }
637 if (value_regno >= 0)
638 /* have read misc data from the stack */
639 mark_reg_unknown_value(state->regs, value_regno);
640 return 0;
641 }
642 }
643
644 /* check read/write into map element returned by bpf_map_lookup_elem() */
645 static int check_map_access(struct verifier_env *env, u32 regno, int off,
646 int size)
647 {
648 struct bpf_map *map = env->cur_state.regs[regno].map_ptr;
649
650 if (off < 0 || off + size > map->value_size) {
651 verbose("invalid access to map value, value_size=%d off=%d size=%d\n",
652 map->value_size, off, size);
653 return -EACCES;
654 }
655 return 0;
656 }
657
658 /* check access to 'struct bpf_context' fields */
659 static int check_ctx_access(struct verifier_env *env, int off, int size,
660 enum bpf_access_type t)
661 {
662 if (env->prog->aux->ops->is_valid_access &&
663 env->prog->aux->ops->is_valid_access(off, size, t)) {
664 /* remember the offset of last byte accessed in ctx */
665 if (env->prog->aux->max_ctx_offset < off + size)
666 env->prog->aux->max_ctx_offset = off + size;
667 return 0;
668 }
669
670 verbose("invalid bpf_context access off=%d size=%d\n", off, size);
671 return -EACCES;
672 }
673
674 static bool is_pointer_value(struct verifier_env *env, int regno)
675 {
676 if (env->allow_ptr_leaks)
677 return false;
678
679 switch (env->cur_state.regs[regno].type) {
680 case UNKNOWN_VALUE:
681 case CONST_IMM:
682 return false;
683 default:
684 return true;
685 }
686 }
687
688 /* check whether memory at (regno + off) is accessible for t = (read | write)
689 * if t==write, value_regno is a register which value is stored into memory
690 * if t==read, value_regno is a register which will receive the value from memory
691 * if t==write && value_regno==-1, some unknown value is stored into memory
692 * if t==read && value_regno==-1, don't care what we read from memory
693 */
694 static int check_mem_access(struct verifier_env *env, u32 regno, int off,
695 int bpf_size, enum bpf_access_type t,
696 int value_regno)
697 {
698 struct verifier_state *state = &env->cur_state;
699 int size, err = 0;
700
701 if (state->regs[regno].type == PTR_TO_STACK)
702 off += state->regs[regno].imm;
703
704 size = bpf_size_to_bytes(bpf_size);
705 if (size < 0)
706 return size;
707
708 if (off % size != 0) {
709 verbose("misaligned access off %d size %d\n", off, size);
710 return -EACCES;
711 }
712
713 if (state->regs[regno].type == PTR_TO_MAP_VALUE) {
714 if (t == BPF_WRITE && value_regno >= 0 &&
715 is_pointer_value(env, value_regno)) {
716 verbose("R%d leaks addr into map\n", value_regno);
717 return -EACCES;
718 }
719 err = check_map_access(env, regno, off, size);
720 if (!err && t == BPF_READ && value_regno >= 0)
721 mark_reg_unknown_value(state->regs, value_regno);
722
723 } else if (state->regs[regno].type == PTR_TO_CTX) {
724 if (t == BPF_WRITE && value_regno >= 0 &&
725 is_pointer_value(env, value_regno)) {
726 verbose("R%d leaks addr into ctx\n", value_regno);
727 return -EACCES;
728 }
729 err = check_ctx_access(env, off, size, t);
730 if (!err && t == BPF_READ && value_regno >= 0)
731 mark_reg_unknown_value(state->regs, value_regno);
732
733 } else if (state->regs[regno].type == FRAME_PTR ||
734 state->regs[regno].type == PTR_TO_STACK) {
735 if (off >= 0 || off < -MAX_BPF_STACK) {
736 verbose("invalid stack off=%d size=%d\n", off, size);
737 return -EACCES;
738 }
739 if (t == BPF_WRITE) {
740 if (!env->allow_ptr_leaks &&
741 state->stack_slot_type[MAX_BPF_STACK + off] == STACK_SPILL &&
742 size != BPF_REG_SIZE) {
743 verbose("attempt to corrupt spilled pointer on stack\n");
744 return -EACCES;
745 }
746 err = check_stack_write(state, off, size, value_regno);
747 } else {
748 err = check_stack_read(state, off, size, value_regno);
749 }
750 } else {
751 verbose("R%d invalid mem access '%s'\n",
752 regno, reg_type_str[state->regs[regno].type]);
753 return -EACCES;
754 }
755 return err;
756 }
757
758 static int check_xadd(struct verifier_env *env, struct bpf_insn *insn)
759 {
760 struct reg_state *regs = env->cur_state.regs;
761 int err;
762
763 if ((BPF_SIZE(insn->code) != BPF_W && BPF_SIZE(insn->code) != BPF_DW) ||
764 insn->imm != 0) {
765 verbose("BPF_XADD uses reserved fields\n");
766 return -EINVAL;
767 }
768
769 /* check src1 operand */
770 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
771 if (err)
772 return err;
773
774 /* check src2 operand */
775 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
776 if (err)
777 return err;
778
779 /* check whether atomic_add can read the memory */
780 err = check_mem_access(env, insn->dst_reg, insn->off,
781 BPF_SIZE(insn->code), BPF_READ, -1);
782 if (err)
783 return err;
784
785 /* check whether atomic_add can write into the same memory */
786 return check_mem_access(env, insn->dst_reg, insn->off,
787 BPF_SIZE(insn->code), BPF_WRITE, -1);
788 }
789
790 /* when register 'regno' is passed into function that will read 'access_size'
791 * bytes from that pointer, make sure that it's within stack boundary
792 * and all elements of stack are initialized
793 */
794 static int check_stack_boundary(struct verifier_env *env, int regno,
795 int access_size, bool zero_size_allowed,
796 struct bpf_call_arg_meta *meta)
797 {
798 struct verifier_state *state = &env->cur_state;
799 struct reg_state *regs = state->regs;
800 int off, i;
801
802 if (regs[regno].type != PTR_TO_STACK) {
803 if (zero_size_allowed && access_size == 0 &&
804 regs[regno].type == CONST_IMM &&
805 regs[regno].imm == 0)
806 return 0;
807
808 verbose("R%d type=%s expected=%s\n", regno,
809 reg_type_str[regs[regno].type],
810 reg_type_str[PTR_TO_STACK]);
811 return -EACCES;
812 }
813
814 off = regs[regno].imm;
815 if (off >= 0 || off < -MAX_BPF_STACK || off + access_size > 0 ||
816 access_size <= 0) {
817 verbose("invalid stack type R%d off=%d access_size=%d\n",
818 regno, off, access_size);
819 return -EACCES;
820 }
821
822 if (meta && meta->raw_mode) {
823 meta->access_size = access_size;
824 meta->regno = regno;
825 return 0;
826 }
827
828 for (i = 0; i < access_size; i++) {
829 if (state->stack_slot_type[MAX_BPF_STACK + off + i] != STACK_MISC) {
830 verbose("invalid indirect read from stack off %d+%d size %d\n",
831 off, i, access_size);
832 return -EACCES;
833 }
834 }
835 return 0;
836 }
837
838 static int check_func_arg(struct verifier_env *env, u32 regno,
839 enum bpf_arg_type arg_type,
840 struct bpf_call_arg_meta *meta)
841 {
842 struct reg_state *reg = env->cur_state.regs + regno;
843 enum bpf_reg_type expected_type;
844 int err = 0;
845
846 if (arg_type == ARG_DONTCARE)
847 return 0;
848
849 if (reg->type == NOT_INIT) {
850 verbose("R%d !read_ok\n", regno);
851 return -EACCES;
852 }
853
854 if (arg_type == ARG_ANYTHING) {
855 if (is_pointer_value(env, regno)) {
856 verbose("R%d leaks addr into helper function\n", regno);
857 return -EACCES;
858 }
859 return 0;
860 }
861
862 if (arg_type == ARG_PTR_TO_MAP_KEY ||
863 arg_type == ARG_PTR_TO_MAP_VALUE) {
864 expected_type = PTR_TO_STACK;
865 } else if (arg_type == ARG_CONST_STACK_SIZE ||
866 arg_type == ARG_CONST_STACK_SIZE_OR_ZERO) {
867 expected_type = CONST_IMM;
868 } else if (arg_type == ARG_CONST_MAP_PTR) {
869 expected_type = CONST_PTR_TO_MAP;
870 } else if (arg_type == ARG_PTR_TO_CTX) {
871 expected_type = PTR_TO_CTX;
872 } else if (arg_type == ARG_PTR_TO_STACK ||
873 arg_type == ARG_PTR_TO_RAW_STACK) {
874 expected_type = PTR_TO_STACK;
875 /* One exception here. In case function allows for NULL to be
876 * passed in as argument, it's a CONST_IMM type. Final test
877 * happens during stack boundary checking.
878 */
879 if (reg->type == CONST_IMM && reg->imm == 0)
880 expected_type = CONST_IMM;
881 meta->raw_mode = arg_type == ARG_PTR_TO_RAW_STACK;
882 } else {
883 verbose("unsupported arg_type %d\n", arg_type);
884 return -EFAULT;
885 }
886
887 if (reg->type != expected_type) {
888 verbose("R%d type=%s expected=%s\n", regno,
889 reg_type_str[reg->type], reg_type_str[expected_type]);
890 return -EACCES;
891 }
892
893 if (arg_type == ARG_CONST_MAP_PTR) {
894 /* bpf_map_xxx(map_ptr) call: remember that map_ptr */
895 meta->map_ptr = reg->map_ptr;
896 } else if (arg_type == ARG_PTR_TO_MAP_KEY) {
897 /* bpf_map_xxx(..., map_ptr, ..., key) call:
898 * check that [key, key + map->key_size) are within
899 * stack limits and initialized
900 */
901 if (!meta->map_ptr) {
902 /* in function declaration map_ptr must come before
903 * map_key, so that it's verified and known before
904 * we have to check map_key here. Otherwise it means
905 * that kernel subsystem misconfigured verifier
906 */
907 verbose("invalid map_ptr to access map->key\n");
908 return -EACCES;
909 }
910 err = check_stack_boundary(env, regno, meta->map_ptr->key_size,
911 false, NULL);
912 } else if (arg_type == ARG_PTR_TO_MAP_VALUE) {
913 /* bpf_map_xxx(..., map_ptr, ..., value) call:
914 * check [value, value + map->value_size) validity
915 */
916 if (!meta->map_ptr) {
917 /* kernel subsystem misconfigured verifier */
918 verbose("invalid map_ptr to access map->value\n");
919 return -EACCES;
920 }
921 err = check_stack_boundary(env, regno,
922 meta->map_ptr->value_size,
923 false, NULL);
924 } else if (arg_type == ARG_CONST_STACK_SIZE ||
925 arg_type == ARG_CONST_STACK_SIZE_OR_ZERO) {
926 bool zero_size_allowed = (arg_type == ARG_CONST_STACK_SIZE_OR_ZERO);
927
928 /* bpf_xxx(..., buf, len) call will access 'len' bytes
929 * from stack pointer 'buf'. Check it
930 * note: regno == len, regno - 1 == buf
931 */
932 if (regno == 0) {
933 /* kernel subsystem misconfigured verifier */
934 verbose("ARG_CONST_STACK_SIZE cannot be first argument\n");
935 return -EACCES;
936 }
937 err = check_stack_boundary(env, regno - 1, reg->imm,
938 zero_size_allowed, meta);
939 }
940
941 return err;
942 }
943
944 static int check_map_func_compatibility(struct bpf_map *map, int func_id)
945 {
946 bool bool_map, bool_func;
947 int i;
948
949 if (!map)
950 return 0;
951
952 for (i = 0; i < ARRAY_SIZE(func_limit); i++) {
953 bool_map = (map->map_type == func_limit[i].map_type);
954 bool_func = (func_id == func_limit[i].func_id);
955 /* only when map & func pair match it can continue.
956 * don't allow any other map type to be passed into
957 * the special func;
958 */
959 if (bool_func && bool_map != bool_func) {
960 verbose("cannot pass map_type %d into func %d\n",
961 map->map_type, func_id);
962 return -EINVAL;
963 }
964 }
965
966 return 0;
967 }
968
969 static int check_raw_mode(const struct bpf_func_proto *fn)
970 {
971 int count = 0;
972
973 if (fn->arg1_type == ARG_PTR_TO_RAW_STACK)
974 count++;
975 if (fn->arg2_type == ARG_PTR_TO_RAW_STACK)
976 count++;
977 if (fn->arg3_type == ARG_PTR_TO_RAW_STACK)
978 count++;
979 if (fn->arg4_type == ARG_PTR_TO_RAW_STACK)
980 count++;
981 if (fn->arg5_type == ARG_PTR_TO_RAW_STACK)
982 count++;
983
984 return count > 1 ? -EINVAL : 0;
985 }
986
987 static int check_call(struct verifier_env *env, int func_id)
988 {
989 struct verifier_state *state = &env->cur_state;
990 const struct bpf_func_proto *fn = NULL;
991 struct reg_state *regs = state->regs;
992 struct reg_state *reg;
993 struct bpf_call_arg_meta meta;
994 int i, err;
995
996 /* find function prototype */
997 if (func_id < 0 || func_id >= __BPF_FUNC_MAX_ID) {
998 verbose("invalid func %d\n", func_id);
999 return -EINVAL;
1000 }
1001
1002 if (env->prog->aux->ops->get_func_proto)
1003 fn = env->prog->aux->ops->get_func_proto(func_id);
1004
1005 if (!fn) {
1006 verbose("unknown func %d\n", func_id);
1007 return -EINVAL;
1008 }
1009
1010 /* eBPF programs must be GPL compatible to use GPL-ed functions */
1011 if (!env->prog->gpl_compatible && fn->gpl_only) {
1012 verbose("cannot call GPL only function from proprietary program\n");
1013 return -EINVAL;
1014 }
1015
1016 memset(&meta, 0, sizeof(meta));
1017
1018 /* We only support one arg being in raw mode at the moment, which
1019 * is sufficient for the helper functions we have right now.
1020 */
1021 err = check_raw_mode(fn);
1022 if (err) {
1023 verbose("kernel subsystem misconfigured func %d\n", func_id);
1024 return err;
1025 }
1026
1027 /* check args */
1028 err = check_func_arg(env, BPF_REG_1, fn->arg1_type, &meta);
1029 if (err)
1030 return err;
1031 err = check_func_arg(env, BPF_REG_2, fn->arg2_type, &meta);
1032 if (err)
1033 return err;
1034 err = check_func_arg(env, BPF_REG_3, fn->arg3_type, &meta);
1035 if (err)
1036 return err;
1037 err = check_func_arg(env, BPF_REG_4, fn->arg4_type, &meta);
1038 if (err)
1039 return err;
1040 err = check_func_arg(env, BPF_REG_5, fn->arg5_type, &meta);
1041 if (err)
1042 return err;
1043
1044 /* Mark slots with STACK_MISC in case of raw mode, stack offset
1045 * is inferred from register state.
1046 */
1047 for (i = 0; i < meta.access_size; i++) {
1048 err = check_mem_access(env, meta.regno, i, BPF_B, BPF_WRITE, -1);
1049 if (err)
1050 return err;
1051 }
1052
1053 /* reset caller saved regs */
1054 for (i = 0; i < CALLER_SAVED_REGS; i++) {
1055 reg = regs + caller_saved[i];
1056 reg->type = NOT_INIT;
1057 reg->imm = 0;
1058 }
1059
1060 /* update return register */
1061 if (fn->ret_type == RET_INTEGER) {
1062 regs[BPF_REG_0].type = UNKNOWN_VALUE;
1063 } else if (fn->ret_type == RET_VOID) {
1064 regs[BPF_REG_0].type = NOT_INIT;
1065 } else if (fn->ret_type == RET_PTR_TO_MAP_VALUE_OR_NULL) {
1066 regs[BPF_REG_0].type = PTR_TO_MAP_VALUE_OR_NULL;
1067 /* remember map_ptr, so that check_map_access()
1068 * can check 'value_size' boundary of memory access
1069 * to map element returned from bpf_map_lookup_elem()
1070 */
1071 if (meta.map_ptr == NULL) {
1072 verbose("kernel subsystem misconfigured verifier\n");
1073 return -EINVAL;
1074 }
1075 regs[BPF_REG_0].map_ptr = meta.map_ptr;
1076 } else {
1077 verbose("unknown return type %d of func %d\n",
1078 fn->ret_type, func_id);
1079 return -EINVAL;
1080 }
1081
1082 err = check_map_func_compatibility(meta.map_ptr, func_id);
1083 if (err)
1084 return err;
1085
1086 return 0;
1087 }
1088
1089 /* check validity of 32-bit and 64-bit arithmetic operations */
1090 static int check_alu_op(struct verifier_env *env, struct bpf_insn *insn)
1091 {
1092 struct reg_state *regs = env->cur_state.regs;
1093 u8 opcode = BPF_OP(insn->code);
1094 int err;
1095
1096 if (opcode == BPF_END || opcode == BPF_NEG) {
1097 if (opcode == BPF_NEG) {
1098 if (BPF_SRC(insn->code) != 0 ||
1099 insn->src_reg != BPF_REG_0 ||
1100 insn->off != 0 || insn->imm != 0) {
1101 verbose("BPF_NEG uses reserved fields\n");
1102 return -EINVAL;
1103 }
1104 } else {
1105 if (insn->src_reg != BPF_REG_0 || insn->off != 0 ||
1106 (insn->imm != 16 && insn->imm != 32 && insn->imm != 64)) {
1107 verbose("BPF_END uses reserved fields\n");
1108 return -EINVAL;
1109 }
1110 }
1111
1112 /* check src operand */
1113 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1114 if (err)
1115 return err;
1116
1117 if (is_pointer_value(env, insn->dst_reg)) {
1118 verbose("R%d pointer arithmetic prohibited\n",
1119 insn->dst_reg);
1120 return -EACCES;
1121 }
1122
1123 /* check dest operand */
1124 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1125 if (err)
1126 return err;
1127
1128 } else if (opcode == BPF_MOV) {
1129
1130 if (BPF_SRC(insn->code) == BPF_X) {
1131 if (insn->imm != 0 || insn->off != 0) {
1132 verbose("BPF_MOV uses reserved fields\n");
1133 return -EINVAL;
1134 }
1135
1136 /* check src operand */
1137 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1138 if (err)
1139 return err;
1140 } else {
1141 if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
1142 verbose("BPF_MOV uses reserved fields\n");
1143 return -EINVAL;
1144 }
1145 }
1146
1147 /* check dest operand */
1148 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1149 if (err)
1150 return err;
1151
1152 if (BPF_SRC(insn->code) == BPF_X) {
1153 if (BPF_CLASS(insn->code) == BPF_ALU64) {
1154 /* case: R1 = R2
1155 * copy register state to dest reg
1156 */
1157 regs[insn->dst_reg] = regs[insn->src_reg];
1158 } else {
1159 if (is_pointer_value(env, insn->src_reg)) {
1160 verbose("R%d partial copy of pointer\n",
1161 insn->src_reg);
1162 return -EACCES;
1163 }
1164 regs[insn->dst_reg].type = UNKNOWN_VALUE;
1165 regs[insn->dst_reg].map_ptr = NULL;
1166 }
1167 } else {
1168 /* case: R = imm
1169 * remember the value we stored into this reg
1170 */
1171 regs[insn->dst_reg].type = CONST_IMM;
1172 regs[insn->dst_reg].imm = insn->imm;
1173 }
1174
1175 } else if (opcode > BPF_END) {
1176 verbose("invalid BPF_ALU opcode %x\n", opcode);
1177 return -EINVAL;
1178
1179 } else { /* all other ALU ops: and, sub, xor, add, ... */
1180
1181 bool stack_relative = false;
1182
1183 if (BPF_SRC(insn->code) == BPF_X) {
1184 if (insn->imm != 0 || insn->off != 0) {
1185 verbose("BPF_ALU uses reserved fields\n");
1186 return -EINVAL;
1187 }
1188 /* check src1 operand */
1189 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1190 if (err)
1191 return err;
1192 } else {
1193 if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
1194 verbose("BPF_ALU uses reserved fields\n");
1195 return -EINVAL;
1196 }
1197 }
1198
1199 /* check src2 operand */
1200 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1201 if (err)
1202 return err;
1203
1204 if ((opcode == BPF_MOD || opcode == BPF_DIV) &&
1205 BPF_SRC(insn->code) == BPF_K && insn->imm == 0) {
1206 verbose("div by zero\n");
1207 return -EINVAL;
1208 }
1209
1210 if ((opcode == BPF_LSH || opcode == BPF_RSH ||
1211 opcode == BPF_ARSH) && BPF_SRC(insn->code) == BPF_K) {
1212 int size = BPF_CLASS(insn->code) == BPF_ALU64 ? 64 : 32;
1213
1214 if (insn->imm < 0 || insn->imm >= size) {
1215 verbose("invalid shift %d\n", insn->imm);
1216 return -EINVAL;
1217 }
1218 }
1219
1220 /* pattern match 'bpf_add Rx, imm' instruction */
1221 if (opcode == BPF_ADD && BPF_CLASS(insn->code) == BPF_ALU64 &&
1222 regs[insn->dst_reg].type == FRAME_PTR &&
1223 BPF_SRC(insn->code) == BPF_K) {
1224 stack_relative = true;
1225 } else if (is_pointer_value(env, insn->dst_reg)) {
1226 verbose("R%d pointer arithmetic prohibited\n",
1227 insn->dst_reg);
1228 return -EACCES;
1229 } else if (BPF_SRC(insn->code) == BPF_X &&
1230 is_pointer_value(env, insn->src_reg)) {
1231 verbose("R%d pointer arithmetic prohibited\n",
1232 insn->src_reg);
1233 return -EACCES;
1234 }
1235
1236 /* check dest operand */
1237 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1238 if (err)
1239 return err;
1240
1241 if (stack_relative) {
1242 regs[insn->dst_reg].type = PTR_TO_STACK;
1243 regs[insn->dst_reg].imm = insn->imm;
1244 }
1245 }
1246
1247 return 0;
1248 }
1249
1250 static int check_cond_jmp_op(struct verifier_env *env,
1251 struct bpf_insn *insn, int *insn_idx)
1252 {
1253 struct reg_state *regs = env->cur_state.regs;
1254 struct verifier_state *other_branch;
1255 u8 opcode = BPF_OP(insn->code);
1256 int err;
1257
1258 if (opcode > BPF_EXIT) {
1259 verbose("invalid BPF_JMP opcode %x\n", opcode);
1260 return -EINVAL;
1261 }
1262
1263 if (BPF_SRC(insn->code) == BPF_X) {
1264 if (insn->imm != 0) {
1265 verbose("BPF_JMP uses reserved fields\n");
1266 return -EINVAL;
1267 }
1268
1269 /* check src1 operand */
1270 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1271 if (err)
1272 return err;
1273
1274 if (is_pointer_value(env, insn->src_reg)) {
1275 verbose("R%d pointer comparison prohibited\n",
1276 insn->src_reg);
1277 return -EACCES;
1278 }
1279 } else {
1280 if (insn->src_reg != BPF_REG_0) {
1281 verbose("BPF_JMP uses reserved fields\n");
1282 return -EINVAL;
1283 }
1284 }
1285
1286 /* check src2 operand */
1287 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1288 if (err)
1289 return err;
1290
1291 /* detect if R == 0 where R was initialized to zero earlier */
1292 if (BPF_SRC(insn->code) == BPF_K &&
1293 (opcode == BPF_JEQ || opcode == BPF_JNE) &&
1294 regs[insn->dst_reg].type == CONST_IMM &&
1295 regs[insn->dst_reg].imm == insn->imm) {
1296 if (opcode == BPF_JEQ) {
1297 /* if (imm == imm) goto pc+off;
1298 * only follow the goto, ignore fall-through
1299 */
1300 *insn_idx += insn->off;
1301 return 0;
1302 } else {
1303 /* if (imm != imm) goto pc+off;
1304 * only follow fall-through branch, since
1305 * that's where the program will go
1306 */
1307 return 0;
1308 }
1309 }
1310
1311 other_branch = push_stack(env, *insn_idx + insn->off + 1, *insn_idx);
1312 if (!other_branch)
1313 return -EFAULT;
1314
1315 /* detect if R == 0 where R is returned value from bpf_map_lookup_elem() */
1316 if (BPF_SRC(insn->code) == BPF_K &&
1317 insn->imm == 0 && (opcode == BPF_JEQ ||
1318 opcode == BPF_JNE) &&
1319 regs[insn->dst_reg].type == PTR_TO_MAP_VALUE_OR_NULL) {
1320 if (opcode == BPF_JEQ) {
1321 /* next fallthrough insn can access memory via
1322 * this register
1323 */
1324 regs[insn->dst_reg].type = PTR_TO_MAP_VALUE;
1325 /* branch targer cannot access it, since reg == 0 */
1326 other_branch->regs[insn->dst_reg].type = CONST_IMM;
1327 other_branch->regs[insn->dst_reg].imm = 0;
1328 } else {
1329 other_branch->regs[insn->dst_reg].type = PTR_TO_MAP_VALUE;
1330 regs[insn->dst_reg].type = CONST_IMM;
1331 regs[insn->dst_reg].imm = 0;
1332 }
1333 } else if (is_pointer_value(env, insn->dst_reg)) {
1334 verbose("R%d pointer comparison prohibited\n", insn->dst_reg);
1335 return -EACCES;
1336 } else if (BPF_SRC(insn->code) == BPF_K &&
1337 (opcode == BPF_JEQ || opcode == BPF_JNE)) {
1338
1339 if (opcode == BPF_JEQ) {
1340 /* detect if (R == imm) goto
1341 * and in the target state recognize that R = imm
1342 */
1343 other_branch->regs[insn->dst_reg].type = CONST_IMM;
1344 other_branch->regs[insn->dst_reg].imm = insn->imm;
1345 } else {
1346 /* detect if (R != imm) goto
1347 * and in the fall-through state recognize that R = imm
1348 */
1349 regs[insn->dst_reg].type = CONST_IMM;
1350 regs[insn->dst_reg].imm = insn->imm;
1351 }
1352 }
1353 if (log_level)
1354 print_verifier_state(env);
1355 return 0;
1356 }
1357
1358 /* return the map pointer stored inside BPF_LD_IMM64 instruction */
1359 static struct bpf_map *ld_imm64_to_map_ptr(struct bpf_insn *insn)
1360 {
1361 u64 imm64 = ((u64) (u32) insn[0].imm) | ((u64) (u32) insn[1].imm) << 32;
1362
1363 return (struct bpf_map *) (unsigned long) imm64;
1364 }
1365
1366 /* verify BPF_LD_IMM64 instruction */
1367 static int check_ld_imm(struct verifier_env *env, struct bpf_insn *insn)
1368 {
1369 struct reg_state *regs = env->cur_state.regs;
1370 int err;
1371
1372 if (BPF_SIZE(insn->code) != BPF_DW) {
1373 verbose("invalid BPF_LD_IMM insn\n");
1374 return -EINVAL;
1375 }
1376 if (insn->off != 0) {
1377 verbose("BPF_LD_IMM64 uses reserved fields\n");
1378 return -EINVAL;
1379 }
1380
1381 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1382 if (err)
1383 return err;
1384
1385 if (insn->src_reg == 0)
1386 /* generic move 64-bit immediate into a register */
1387 return 0;
1388
1389 /* replace_map_fd_with_map_ptr() should have caught bad ld_imm64 */
1390 BUG_ON(insn->src_reg != BPF_PSEUDO_MAP_FD);
1391
1392 regs[insn->dst_reg].type = CONST_PTR_TO_MAP;
1393 regs[insn->dst_reg].map_ptr = ld_imm64_to_map_ptr(insn);
1394 return 0;
1395 }
1396
1397 static bool may_access_skb(enum bpf_prog_type type)
1398 {
1399 switch (type) {
1400 case BPF_PROG_TYPE_SOCKET_FILTER:
1401 case BPF_PROG_TYPE_SCHED_CLS:
1402 case BPF_PROG_TYPE_SCHED_ACT:
1403 return true;
1404 default:
1405 return false;
1406 }
1407 }
1408
1409 /* verify safety of LD_ABS|LD_IND instructions:
1410 * - they can only appear in the programs where ctx == skb
1411 * - since they are wrappers of function calls, they scratch R1-R5 registers,
1412 * preserve R6-R9, and store return value into R0
1413 *
1414 * Implicit input:
1415 * ctx == skb == R6 == CTX
1416 *
1417 * Explicit input:
1418 * SRC == any register
1419 * IMM == 32-bit immediate
1420 *
1421 * Output:
1422 * R0 - 8/16/32-bit skb data converted to cpu endianness
1423 */
1424 static int check_ld_abs(struct verifier_env *env, struct bpf_insn *insn)
1425 {
1426 struct reg_state *regs = env->cur_state.regs;
1427 u8 mode = BPF_MODE(insn->code);
1428 struct reg_state *reg;
1429 int i, err;
1430
1431 if (!may_access_skb(env->prog->type)) {
1432 verbose("BPF_LD_ABS|IND instructions not allowed for this program type\n");
1433 return -EINVAL;
1434 }
1435
1436 if (insn->dst_reg != BPF_REG_0 || insn->off != 0 ||
1437 BPF_SIZE(insn->code) == BPF_DW ||
1438 (mode == BPF_ABS && insn->src_reg != BPF_REG_0)) {
1439 verbose("BPF_LD_ABS uses reserved fields\n");
1440 return -EINVAL;
1441 }
1442
1443 /* check whether implicit source operand (register R6) is readable */
1444 err = check_reg_arg(regs, BPF_REG_6, SRC_OP);
1445 if (err)
1446 return err;
1447
1448 if (regs[BPF_REG_6].type != PTR_TO_CTX) {
1449 verbose("at the time of BPF_LD_ABS|IND R6 != pointer to skb\n");
1450 return -EINVAL;
1451 }
1452
1453 if (mode == BPF_IND) {
1454 /* check explicit source operand */
1455 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1456 if (err)
1457 return err;
1458 }
1459
1460 /* reset caller saved regs to unreadable */
1461 for (i = 0; i < CALLER_SAVED_REGS; i++) {
1462 reg = regs + caller_saved[i];
1463 reg->type = NOT_INIT;
1464 reg->imm = 0;
1465 }
1466
1467 /* mark destination R0 register as readable, since it contains
1468 * the value fetched from the packet
1469 */
1470 regs[BPF_REG_0].type = UNKNOWN_VALUE;
1471 return 0;
1472 }
1473
1474 /* non-recursive DFS pseudo code
1475 * 1 procedure DFS-iterative(G,v):
1476 * 2 label v as discovered
1477 * 3 let S be a stack
1478 * 4 S.push(v)
1479 * 5 while S is not empty
1480 * 6 t <- S.pop()
1481 * 7 if t is what we're looking for:
1482 * 8 return t
1483 * 9 for all edges e in G.adjacentEdges(t) do
1484 * 10 if edge e is already labelled
1485 * 11 continue with the next edge
1486 * 12 w <- G.adjacentVertex(t,e)
1487 * 13 if vertex w is not discovered and not explored
1488 * 14 label e as tree-edge
1489 * 15 label w as discovered
1490 * 16 S.push(w)
1491 * 17 continue at 5
1492 * 18 else if vertex w is discovered
1493 * 19 label e as back-edge
1494 * 20 else
1495 * 21 // vertex w is explored
1496 * 22 label e as forward- or cross-edge
1497 * 23 label t as explored
1498 * 24 S.pop()
1499 *
1500 * convention:
1501 * 0x10 - discovered
1502 * 0x11 - discovered and fall-through edge labelled
1503 * 0x12 - discovered and fall-through and branch edges labelled
1504 * 0x20 - explored
1505 */
1506
1507 enum {
1508 DISCOVERED = 0x10,
1509 EXPLORED = 0x20,
1510 FALLTHROUGH = 1,
1511 BRANCH = 2,
1512 };
1513
1514 #define STATE_LIST_MARK ((struct verifier_state_list *) -1L)
1515
1516 static int *insn_stack; /* stack of insns to process */
1517 static int cur_stack; /* current stack index */
1518 static int *insn_state;
1519
1520 /* t, w, e - match pseudo-code above:
1521 * t - index of current instruction
1522 * w - next instruction
1523 * e - edge
1524 */
1525 static int push_insn(int t, int w, int e, struct verifier_env *env)
1526 {
1527 if (e == FALLTHROUGH && insn_state[t] >= (DISCOVERED | FALLTHROUGH))
1528 return 0;
1529
1530 if (e == BRANCH && insn_state[t] >= (DISCOVERED | BRANCH))
1531 return 0;
1532
1533 if (w < 0 || w >= env->prog->len) {
1534 verbose("jump out of range from insn %d to %d\n", t, w);
1535 return -EINVAL;
1536 }
1537
1538 if (e == BRANCH)
1539 /* mark branch target for state pruning */
1540 env->explored_states[w] = STATE_LIST_MARK;
1541
1542 if (insn_state[w] == 0) {
1543 /* tree-edge */
1544 insn_state[t] = DISCOVERED | e;
1545 insn_state[w] = DISCOVERED;
1546 if (cur_stack >= env->prog->len)
1547 return -E2BIG;
1548 insn_stack[cur_stack++] = w;
1549 return 1;
1550 } else if ((insn_state[w] & 0xF0) == DISCOVERED) {
1551 verbose("back-edge from insn %d to %d\n", t, w);
1552 return -EINVAL;
1553 } else if (insn_state[w] == EXPLORED) {
1554 /* forward- or cross-edge */
1555 insn_state[t] = DISCOVERED | e;
1556 } else {
1557 verbose("insn state internal bug\n");
1558 return -EFAULT;
1559 }
1560 return 0;
1561 }
1562
1563 /* non-recursive depth-first-search to detect loops in BPF program
1564 * loop == back-edge in directed graph
1565 */
1566 static int check_cfg(struct verifier_env *env)
1567 {
1568 struct bpf_insn *insns = env->prog->insnsi;
1569 int insn_cnt = env->prog->len;
1570 int ret = 0;
1571 int i, t;
1572
1573 insn_state = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
1574 if (!insn_state)
1575 return -ENOMEM;
1576
1577 insn_stack = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
1578 if (!insn_stack) {
1579 kfree(insn_state);
1580 return -ENOMEM;
1581 }
1582
1583 insn_state[0] = DISCOVERED; /* mark 1st insn as discovered */
1584 insn_stack[0] = 0; /* 0 is the first instruction */
1585 cur_stack = 1;
1586
1587 peek_stack:
1588 if (cur_stack == 0)
1589 goto check_state;
1590 t = insn_stack[cur_stack - 1];
1591
1592 if (BPF_CLASS(insns[t].code) == BPF_JMP) {
1593 u8 opcode = BPF_OP(insns[t].code);
1594
1595 if (opcode == BPF_EXIT) {
1596 goto mark_explored;
1597 } else if (opcode == BPF_CALL) {
1598 ret = push_insn(t, t + 1, FALLTHROUGH, env);
1599 if (ret == 1)
1600 goto peek_stack;
1601 else if (ret < 0)
1602 goto err_free;
1603 if (t + 1 < insn_cnt)
1604 env->explored_states[t + 1] = STATE_LIST_MARK;
1605 } else if (opcode == BPF_JA) {
1606 if (BPF_SRC(insns[t].code) != BPF_K) {
1607 ret = -EINVAL;
1608 goto err_free;
1609 }
1610 /* unconditional jump with single edge */
1611 ret = push_insn(t, t + insns[t].off + 1,
1612 FALLTHROUGH, env);
1613 if (ret == 1)
1614 goto peek_stack;
1615 else if (ret < 0)
1616 goto err_free;
1617 /* tell verifier to check for equivalent states
1618 * after every call and jump
1619 */
1620 if (t + 1 < insn_cnt)
1621 env->explored_states[t + 1] = STATE_LIST_MARK;
1622 } else {
1623 /* conditional jump with two edges */
1624 ret = push_insn(t, t + 1, FALLTHROUGH, env);
1625 if (ret == 1)
1626 goto peek_stack;
1627 else if (ret < 0)
1628 goto err_free;
1629
1630 ret = push_insn(t, t + insns[t].off + 1, BRANCH, env);
1631 if (ret == 1)
1632 goto peek_stack;
1633 else if (ret < 0)
1634 goto err_free;
1635 }
1636 } else {
1637 /* all other non-branch instructions with single
1638 * fall-through edge
1639 */
1640 ret = push_insn(t, t + 1, FALLTHROUGH, env);
1641 if (ret == 1)
1642 goto peek_stack;
1643 else if (ret < 0)
1644 goto err_free;
1645 }
1646
1647 mark_explored:
1648 insn_state[t] = EXPLORED;
1649 if (cur_stack-- <= 0) {
1650 verbose("pop stack internal bug\n");
1651 ret = -EFAULT;
1652 goto err_free;
1653 }
1654 goto peek_stack;
1655
1656 check_state:
1657 for (i = 0; i < insn_cnt; i++) {
1658 if (insn_state[i] != EXPLORED) {
1659 verbose("unreachable insn %d\n", i);
1660 ret = -EINVAL;
1661 goto err_free;
1662 }
1663 }
1664 ret = 0; /* cfg looks good */
1665
1666 err_free:
1667 kfree(insn_state);
1668 kfree(insn_stack);
1669 return ret;
1670 }
1671
1672 /* compare two verifier states
1673 *
1674 * all states stored in state_list are known to be valid, since
1675 * verifier reached 'bpf_exit' instruction through them
1676 *
1677 * this function is called when verifier exploring different branches of
1678 * execution popped from the state stack. If it sees an old state that has
1679 * more strict register state and more strict stack state then this execution
1680 * branch doesn't need to be explored further, since verifier already
1681 * concluded that more strict state leads to valid finish.
1682 *
1683 * Therefore two states are equivalent if register state is more conservative
1684 * and explored stack state is more conservative than the current one.
1685 * Example:
1686 * explored current
1687 * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC)
1688 * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC)
1689 *
1690 * In other words if current stack state (one being explored) has more
1691 * valid slots than old one that already passed validation, it means
1692 * the verifier can stop exploring and conclude that current state is valid too
1693 *
1694 * Similarly with registers. If explored state has register type as invalid
1695 * whereas register type in current state is meaningful, it means that
1696 * the current state will reach 'bpf_exit' instruction safely
1697 */
1698 static bool states_equal(struct verifier_state *old, struct verifier_state *cur)
1699 {
1700 int i;
1701
1702 for (i = 0; i < MAX_BPF_REG; i++) {
1703 if (memcmp(&old->regs[i], &cur->regs[i],
1704 sizeof(old->regs[0])) != 0) {
1705 if (old->regs[i].type == NOT_INIT ||
1706 (old->regs[i].type == UNKNOWN_VALUE &&
1707 cur->regs[i].type != NOT_INIT))
1708 continue;
1709 return false;
1710 }
1711 }
1712
1713 for (i = 0; i < MAX_BPF_STACK; i++) {
1714 if (old->stack_slot_type[i] == STACK_INVALID)
1715 continue;
1716 if (old->stack_slot_type[i] != cur->stack_slot_type[i])
1717 /* Ex: old explored (safe) state has STACK_SPILL in
1718 * this stack slot, but current has has STACK_MISC ->
1719 * this verifier states are not equivalent,
1720 * return false to continue verification of this path
1721 */
1722 return false;
1723 if (i % BPF_REG_SIZE)
1724 continue;
1725 if (memcmp(&old->spilled_regs[i / BPF_REG_SIZE],
1726 &cur->spilled_regs[i / BPF_REG_SIZE],
1727 sizeof(old->spilled_regs[0])))
1728 /* when explored and current stack slot types are
1729 * the same, check that stored pointers types
1730 * are the same as well.
1731 * Ex: explored safe path could have stored
1732 * (struct reg_state) {.type = PTR_TO_STACK, .imm = -8}
1733 * but current path has stored:
1734 * (struct reg_state) {.type = PTR_TO_STACK, .imm = -16}
1735 * such verifier states are not equivalent.
1736 * return false to continue verification of this path
1737 */
1738 return false;
1739 else
1740 continue;
1741 }
1742 return true;
1743 }
1744
1745 static int is_state_visited(struct verifier_env *env, int insn_idx)
1746 {
1747 struct verifier_state_list *new_sl;
1748 struct verifier_state_list *sl;
1749
1750 sl = env->explored_states[insn_idx];
1751 if (!sl)
1752 /* this 'insn_idx' instruction wasn't marked, so we will not
1753 * be doing state search here
1754 */
1755 return 0;
1756
1757 while (sl != STATE_LIST_MARK) {
1758 if (states_equal(&sl->state, &env->cur_state))
1759 /* reached equivalent register/stack state,
1760 * prune the search
1761 */
1762 return 1;
1763 sl = sl->next;
1764 }
1765
1766 /* there were no equivalent states, remember current one.
1767 * technically the current state is not proven to be safe yet,
1768 * but it will either reach bpf_exit (which means it's safe) or
1769 * it will be rejected. Since there are no loops, we won't be
1770 * seeing this 'insn_idx' instruction again on the way to bpf_exit
1771 */
1772 new_sl = kmalloc(sizeof(struct verifier_state_list), GFP_USER);
1773 if (!new_sl)
1774 return -ENOMEM;
1775
1776 /* add new state to the head of linked list */
1777 memcpy(&new_sl->state, &env->cur_state, sizeof(env->cur_state));
1778 new_sl->next = env->explored_states[insn_idx];
1779 env->explored_states[insn_idx] = new_sl;
1780 return 0;
1781 }
1782
1783 static int do_check(struct verifier_env *env)
1784 {
1785 struct verifier_state *state = &env->cur_state;
1786 struct bpf_insn *insns = env->prog->insnsi;
1787 struct reg_state *regs = state->regs;
1788 int insn_cnt = env->prog->len;
1789 int insn_idx, prev_insn_idx = 0;
1790 int insn_processed = 0;
1791 bool do_print_state = false;
1792
1793 init_reg_state(regs);
1794 insn_idx = 0;
1795 for (;;) {
1796 struct bpf_insn *insn;
1797 u8 class;
1798 int err;
1799
1800 if (insn_idx >= insn_cnt) {
1801 verbose("invalid insn idx %d insn_cnt %d\n",
1802 insn_idx, insn_cnt);
1803 return -EFAULT;
1804 }
1805
1806 insn = &insns[insn_idx];
1807 class = BPF_CLASS(insn->code);
1808
1809 if (++insn_processed > BPF_COMPLEXITY_LIMIT_INSNS) {
1810 verbose("BPF program is too large. Proccessed %d insn\n",
1811 insn_processed);
1812 return -E2BIG;
1813 }
1814
1815 err = is_state_visited(env, insn_idx);
1816 if (err < 0)
1817 return err;
1818 if (err == 1) {
1819 /* found equivalent state, can prune the search */
1820 if (log_level) {
1821 if (do_print_state)
1822 verbose("\nfrom %d to %d: safe\n",
1823 prev_insn_idx, insn_idx);
1824 else
1825 verbose("%d: safe\n", insn_idx);
1826 }
1827 goto process_bpf_exit;
1828 }
1829
1830 if (log_level && do_print_state) {
1831 verbose("\nfrom %d to %d:", prev_insn_idx, insn_idx);
1832 print_verifier_state(env);
1833 do_print_state = false;
1834 }
1835
1836 if (log_level) {
1837 verbose("%d: ", insn_idx);
1838 print_bpf_insn(insn);
1839 }
1840
1841 if (class == BPF_ALU || class == BPF_ALU64) {
1842 err = check_alu_op(env, insn);
1843 if (err)
1844 return err;
1845
1846 } else if (class == BPF_LDX) {
1847 enum bpf_reg_type src_reg_type;
1848
1849 /* check for reserved fields is already done */
1850
1851 /* check src operand */
1852 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1853 if (err)
1854 return err;
1855
1856 err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK);
1857 if (err)
1858 return err;
1859
1860 src_reg_type = regs[insn->src_reg].type;
1861
1862 /* check that memory (src_reg + off) is readable,
1863 * the state of dst_reg will be updated by this func
1864 */
1865 err = check_mem_access(env, insn->src_reg, insn->off,
1866 BPF_SIZE(insn->code), BPF_READ,
1867 insn->dst_reg);
1868 if (err)
1869 return err;
1870
1871 if (BPF_SIZE(insn->code) != BPF_W) {
1872 insn_idx++;
1873 continue;
1874 }
1875
1876 if (insn->imm == 0) {
1877 /* saw a valid insn
1878 * dst_reg = *(u32 *)(src_reg + off)
1879 * use reserved 'imm' field to mark this insn
1880 */
1881 insn->imm = src_reg_type;
1882
1883 } else if (src_reg_type != insn->imm &&
1884 (src_reg_type == PTR_TO_CTX ||
1885 insn->imm == PTR_TO_CTX)) {
1886 /* ABuser program is trying to use the same insn
1887 * dst_reg = *(u32*) (src_reg + off)
1888 * with different pointer types:
1889 * src_reg == ctx in one branch and
1890 * src_reg == stack|map in some other branch.
1891 * Reject it.
1892 */
1893 verbose("same insn cannot be used with different pointers\n");
1894 return -EINVAL;
1895 }
1896
1897 } else if (class == BPF_STX) {
1898 enum bpf_reg_type dst_reg_type;
1899
1900 if (BPF_MODE(insn->code) == BPF_XADD) {
1901 err = check_xadd(env, insn);
1902 if (err)
1903 return err;
1904 insn_idx++;
1905 continue;
1906 }
1907
1908 /* check src1 operand */
1909 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1910 if (err)
1911 return err;
1912 /* check src2 operand */
1913 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1914 if (err)
1915 return err;
1916
1917 dst_reg_type = regs[insn->dst_reg].type;
1918
1919 /* check that memory (dst_reg + off) is writeable */
1920 err = check_mem_access(env, insn->dst_reg, insn->off,
1921 BPF_SIZE(insn->code), BPF_WRITE,
1922 insn->src_reg);
1923 if (err)
1924 return err;
1925
1926 if (insn->imm == 0) {
1927 insn->imm = dst_reg_type;
1928 } else if (dst_reg_type != insn->imm &&
1929 (dst_reg_type == PTR_TO_CTX ||
1930 insn->imm == PTR_TO_CTX)) {
1931 verbose("same insn cannot be used with different pointers\n");
1932 return -EINVAL;
1933 }
1934
1935 } else if (class == BPF_ST) {
1936 if (BPF_MODE(insn->code) != BPF_MEM ||
1937 insn->src_reg != BPF_REG_0) {
1938 verbose("BPF_ST uses reserved fields\n");
1939 return -EINVAL;
1940 }
1941 /* check src operand */
1942 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1943 if (err)
1944 return err;
1945
1946 /* check that memory (dst_reg + off) is writeable */
1947 err = check_mem_access(env, insn->dst_reg, insn->off,
1948 BPF_SIZE(insn->code), BPF_WRITE,
1949 -1);
1950 if (err)
1951 return err;
1952
1953 } else if (class == BPF_JMP) {
1954 u8 opcode = BPF_OP(insn->code);
1955
1956 if (opcode == BPF_CALL) {
1957 if (BPF_SRC(insn->code) != BPF_K ||
1958 insn->off != 0 ||
1959 insn->src_reg != BPF_REG_0 ||
1960 insn->dst_reg != BPF_REG_0) {
1961 verbose("BPF_CALL uses reserved fields\n");
1962 return -EINVAL;
1963 }
1964
1965 err = check_call(env, insn->imm);
1966 if (err)
1967 return err;
1968
1969 } else if (opcode == BPF_JA) {
1970 if (BPF_SRC(insn->code) != BPF_K ||
1971 insn->imm != 0 ||
1972 insn->src_reg != BPF_REG_0 ||
1973 insn->dst_reg != BPF_REG_0) {
1974 verbose("BPF_JA uses reserved fields\n");
1975 return -EINVAL;
1976 }
1977
1978 insn_idx += insn->off + 1;
1979 continue;
1980
1981 } else if (opcode == BPF_EXIT) {
1982 if (BPF_SRC(insn->code) != BPF_K ||
1983 insn->imm != 0 ||
1984 insn->src_reg != BPF_REG_0 ||
1985 insn->dst_reg != BPF_REG_0) {
1986 verbose("BPF_EXIT uses reserved fields\n");
1987 return -EINVAL;
1988 }
1989
1990 /* eBPF calling convetion is such that R0 is used
1991 * to return the value from eBPF program.
1992 * Make sure that it's readable at this time
1993 * of bpf_exit, which means that program wrote
1994 * something into it earlier
1995 */
1996 err = check_reg_arg(regs, BPF_REG_0, SRC_OP);
1997 if (err)
1998 return err;
1999
2000 if (is_pointer_value(env, BPF_REG_0)) {
2001 verbose("R0 leaks addr as return value\n");
2002 return -EACCES;
2003 }
2004
2005 process_bpf_exit:
2006 insn_idx = pop_stack(env, &prev_insn_idx);
2007 if (insn_idx < 0) {
2008 break;
2009 } else {
2010 do_print_state = true;
2011 continue;
2012 }
2013 } else {
2014 err = check_cond_jmp_op(env, insn, &insn_idx);
2015 if (err)
2016 return err;
2017 }
2018 } else if (class == BPF_LD) {
2019 u8 mode = BPF_MODE(insn->code);
2020
2021 if (mode == BPF_ABS || mode == BPF_IND) {
2022 err = check_ld_abs(env, insn);
2023 if (err)
2024 return err;
2025
2026 } else if (mode == BPF_IMM) {
2027 err = check_ld_imm(env, insn);
2028 if (err)
2029 return err;
2030
2031 insn_idx++;
2032 } else {
2033 verbose("invalid BPF_LD mode\n");
2034 return -EINVAL;
2035 }
2036 } else {
2037 verbose("unknown insn class %d\n", class);
2038 return -EINVAL;
2039 }
2040
2041 insn_idx++;
2042 }
2043
2044 return 0;
2045 }
2046
2047 /* look for pseudo eBPF instructions that access map FDs and
2048 * replace them with actual map pointers
2049 */
2050 static int replace_map_fd_with_map_ptr(struct verifier_env *env)
2051 {
2052 struct bpf_insn *insn = env->prog->insnsi;
2053 int insn_cnt = env->prog->len;
2054 int i, j;
2055
2056 for (i = 0; i < insn_cnt; i++, insn++) {
2057 if (BPF_CLASS(insn->code) == BPF_LDX &&
2058 (BPF_MODE(insn->code) != BPF_MEM || insn->imm != 0)) {
2059 verbose("BPF_LDX uses reserved fields\n");
2060 return -EINVAL;
2061 }
2062
2063 if (BPF_CLASS(insn->code) == BPF_STX &&
2064 ((BPF_MODE(insn->code) != BPF_MEM &&
2065 BPF_MODE(insn->code) != BPF_XADD) || insn->imm != 0)) {
2066 verbose("BPF_STX uses reserved fields\n");
2067 return -EINVAL;
2068 }
2069
2070 if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW)) {
2071 struct bpf_map *map;
2072 struct fd f;
2073
2074 if (i == insn_cnt - 1 || insn[1].code != 0 ||
2075 insn[1].dst_reg != 0 || insn[1].src_reg != 0 ||
2076 insn[1].off != 0) {
2077 verbose("invalid bpf_ld_imm64 insn\n");
2078 return -EINVAL;
2079 }
2080
2081 if (insn->src_reg == 0)
2082 /* valid generic load 64-bit imm */
2083 goto next_insn;
2084
2085 if (insn->src_reg != BPF_PSEUDO_MAP_FD) {
2086 verbose("unrecognized bpf_ld_imm64 insn\n");
2087 return -EINVAL;
2088 }
2089
2090 f = fdget(insn->imm);
2091 map = __bpf_map_get(f);
2092 if (IS_ERR(map)) {
2093 verbose("fd %d is not pointing to valid bpf_map\n",
2094 insn->imm);
2095 fdput(f);
2096 return PTR_ERR(map);
2097 }
2098
2099 /* store map pointer inside BPF_LD_IMM64 instruction */
2100 insn[0].imm = (u32) (unsigned long) map;
2101 insn[1].imm = ((u64) (unsigned long) map) >> 32;
2102
2103 /* check whether we recorded this map already */
2104 for (j = 0; j < env->used_map_cnt; j++)
2105 if (env->used_maps[j] == map) {
2106 fdput(f);
2107 goto next_insn;
2108 }
2109
2110 if (env->used_map_cnt >= MAX_USED_MAPS) {
2111 fdput(f);
2112 return -E2BIG;
2113 }
2114
2115 /* remember this map */
2116 env->used_maps[env->used_map_cnt++] = map;
2117
2118 /* hold the map. If the program is rejected by verifier,
2119 * the map will be released by release_maps() or it
2120 * will be used by the valid program until it's unloaded
2121 * and all maps are released in free_bpf_prog_info()
2122 */
2123 bpf_map_inc(map, false);
2124 fdput(f);
2125 next_insn:
2126 insn++;
2127 i++;
2128 }
2129 }
2130
2131 /* now all pseudo BPF_LD_IMM64 instructions load valid
2132 * 'struct bpf_map *' into a register instead of user map_fd.
2133 * These pointers will be used later by verifier to validate map access.
2134 */
2135 return 0;
2136 }
2137
2138 /* drop refcnt of maps used by the rejected program */
2139 static void release_maps(struct verifier_env *env)
2140 {
2141 int i;
2142
2143 for (i = 0; i < env->used_map_cnt; i++)
2144 bpf_map_put(env->used_maps[i]);
2145 }
2146
2147 /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */
2148 static void convert_pseudo_ld_imm64(struct verifier_env *env)
2149 {
2150 struct bpf_insn *insn = env->prog->insnsi;
2151 int insn_cnt = env->prog->len;
2152 int i;
2153
2154 for (i = 0; i < insn_cnt; i++, insn++)
2155 if (insn->code == (BPF_LD | BPF_IMM | BPF_DW))
2156 insn->src_reg = 0;
2157 }
2158
2159 static void adjust_branches(struct bpf_prog *prog, int pos, int delta)
2160 {
2161 struct bpf_insn *insn = prog->insnsi;
2162 int insn_cnt = prog->len;
2163 int i;
2164
2165 for (i = 0; i < insn_cnt; i++, insn++) {
2166 if (BPF_CLASS(insn->code) != BPF_JMP ||
2167 BPF_OP(insn->code) == BPF_CALL ||
2168 BPF_OP(insn->code) == BPF_EXIT)
2169 continue;
2170
2171 /* adjust offset of jmps if necessary */
2172 if (i < pos && i + insn->off + 1 > pos)
2173 insn->off += delta;
2174 else if (i > pos + delta && i + insn->off + 1 <= pos + delta)
2175 insn->off -= delta;
2176 }
2177 }
2178
2179 /* convert load instructions that access fields of 'struct __sk_buff'
2180 * into sequence of instructions that access fields of 'struct sk_buff'
2181 */
2182 static int convert_ctx_accesses(struct verifier_env *env)
2183 {
2184 struct bpf_insn *insn = env->prog->insnsi;
2185 int insn_cnt = env->prog->len;
2186 struct bpf_insn insn_buf[16];
2187 struct bpf_prog *new_prog;
2188 u32 cnt;
2189 int i;
2190 enum bpf_access_type type;
2191
2192 if (!env->prog->aux->ops->convert_ctx_access)
2193 return 0;
2194
2195 for (i = 0; i < insn_cnt; i++, insn++) {
2196 if (insn->code == (BPF_LDX | BPF_MEM | BPF_W))
2197 type = BPF_READ;
2198 else if (insn->code == (BPF_STX | BPF_MEM | BPF_W))
2199 type = BPF_WRITE;
2200 else
2201 continue;
2202
2203 if (insn->imm != PTR_TO_CTX) {
2204 /* clear internal mark */
2205 insn->imm = 0;
2206 continue;
2207 }
2208
2209 cnt = env->prog->aux->ops->
2210 convert_ctx_access(type, insn->dst_reg, insn->src_reg,
2211 insn->off, insn_buf, env->prog);
2212 if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) {
2213 verbose("bpf verifier is misconfigured\n");
2214 return -EINVAL;
2215 }
2216
2217 if (cnt == 1) {
2218 memcpy(insn, insn_buf, sizeof(*insn));
2219 continue;
2220 }
2221
2222 /* several new insns need to be inserted. Make room for them */
2223 insn_cnt += cnt - 1;
2224 new_prog = bpf_prog_realloc(env->prog,
2225 bpf_prog_size(insn_cnt),
2226 GFP_USER);
2227 if (!new_prog)
2228 return -ENOMEM;
2229
2230 new_prog->len = insn_cnt;
2231
2232 memmove(new_prog->insnsi + i + cnt, new_prog->insns + i + 1,
2233 sizeof(*insn) * (insn_cnt - i - cnt));
2234
2235 /* copy substitute insns in place of load instruction */
2236 memcpy(new_prog->insnsi + i, insn_buf, sizeof(*insn) * cnt);
2237
2238 /* adjust branches in the whole program */
2239 adjust_branches(new_prog, i, cnt - 1);
2240
2241 /* keep walking new program and skip insns we just inserted */
2242 env->prog = new_prog;
2243 insn = new_prog->insnsi + i + cnt - 1;
2244 i += cnt - 1;
2245 }
2246
2247 return 0;
2248 }
2249
2250 static void free_states(struct verifier_env *env)
2251 {
2252 struct verifier_state_list *sl, *sln;
2253 int i;
2254
2255 if (!env->explored_states)
2256 return;
2257
2258 for (i = 0; i < env->prog->len; i++) {
2259 sl = env->explored_states[i];
2260
2261 if (sl)
2262 while (sl != STATE_LIST_MARK) {
2263 sln = sl->next;
2264 kfree(sl);
2265 sl = sln;
2266 }
2267 }
2268
2269 kfree(env->explored_states);
2270 }
2271
2272 int bpf_check(struct bpf_prog **prog, union bpf_attr *attr)
2273 {
2274 char __user *log_ubuf = NULL;
2275 struct verifier_env *env;
2276 int ret = -EINVAL;
2277
2278 if ((*prog)->len <= 0 || (*prog)->len > BPF_MAXINSNS)
2279 return -E2BIG;
2280
2281 /* 'struct verifier_env' can be global, but since it's not small,
2282 * allocate/free it every time bpf_check() is called
2283 */
2284 env = kzalloc(sizeof(struct verifier_env), GFP_KERNEL);
2285 if (!env)
2286 return -ENOMEM;
2287
2288 env->prog = *prog;
2289
2290 /* grab the mutex to protect few globals used by verifier */
2291 mutex_lock(&bpf_verifier_lock);
2292
2293 if (attr->log_level || attr->log_buf || attr->log_size) {
2294 /* user requested verbose verifier output
2295 * and supplied buffer to store the verification trace
2296 */
2297 log_level = attr->log_level;
2298 log_ubuf = (char __user *) (unsigned long) attr->log_buf;
2299 log_size = attr->log_size;
2300 log_len = 0;
2301
2302 ret = -EINVAL;
2303 /* log_* values have to be sane */
2304 if (log_size < 128 || log_size > UINT_MAX >> 8 ||
2305 log_level == 0 || log_ubuf == NULL)
2306 goto free_env;
2307
2308 ret = -ENOMEM;
2309 log_buf = vmalloc(log_size);
2310 if (!log_buf)
2311 goto free_env;
2312 } else {
2313 log_level = 0;
2314 }
2315
2316 ret = replace_map_fd_with_map_ptr(env);
2317 if (ret < 0)
2318 goto skip_full_check;
2319
2320 env->explored_states = kcalloc(env->prog->len,
2321 sizeof(struct verifier_state_list *),
2322 GFP_USER);
2323 ret = -ENOMEM;
2324 if (!env->explored_states)
2325 goto skip_full_check;
2326
2327 ret = check_cfg(env);
2328 if (ret < 0)
2329 goto skip_full_check;
2330
2331 env->allow_ptr_leaks = capable(CAP_SYS_ADMIN);
2332
2333 ret = do_check(env);
2334
2335 skip_full_check:
2336 while (pop_stack(env, NULL) >= 0);
2337 free_states(env);
2338
2339 if (ret == 0)
2340 /* program is valid, convert *(u32*)(ctx + off) accesses */
2341 ret = convert_ctx_accesses(env);
2342
2343 if (log_level && log_len >= log_size - 1) {
2344 BUG_ON(log_len >= log_size);
2345 /* verifier log exceeded user supplied buffer */
2346 ret = -ENOSPC;
2347 /* fall through to return what was recorded */
2348 }
2349
2350 /* copy verifier log back to user space including trailing zero */
2351 if (log_level && copy_to_user(log_ubuf, log_buf, log_len + 1) != 0) {
2352 ret = -EFAULT;
2353 goto free_log_buf;
2354 }
2355
2356 if (ret == 0 && env->used_map_cnt) {
2357 /* if program passed verifier, update used_maps in bpf_prog_info */
2358 env->prog->aux->used_maps = kmalloc_array(env->used_map_cnt,
2359 sizeof(env->used_maps[0]),
2360 GFP_KERNEL);
2361
2362 if (!env->prog->aux->used_maps) {
2363 ret = -ENOMEM;
2364 goto free_log_buf;
2365 }
2366
2367 memcpy(env->prog->aux->used_maps, env->used_maps,
2368 sizeof(env->used_maps[0]) * env->used_map_cnt);
2369 env->prog->aux->used_map_cnt = env->used_map_cnt;
2370
2371 /* program is valid. Convert pseudo bpf_ld_imm64 into generic
2372 * bpf_ld_imm64 instructions
2373 */
2374 convert_pseudo_ld_imm64(env);
2375 }
2376
2377 free_log_buf:
2378 if (log_level)
2379 vfree(log_buf);
2380 free_env:
2381 if (!env->prog->aux->used_maps)
2382 /* if we didn't copy map pointers into bpf_prog_info, release
2383 * them now. Otherwise free_bpf_prog_info() will release them.
2384 */
2385 release_maps(env);
2386 *prog = env->prog;
2387 kfree(env);
2388 mutex_unlock(&bpf_verifier_lock);
2389 return ret;
2390 }
Ubuntu Eoan Linux kernel mirror