aboutsummaryrefslogtreecommitdiffstats
path: root/kernel/bpf/verifier.c
diff options
context:
space:
mode:
Diffstat (limited to 'kernel/bpf/verifier.c')
-rw-r--r--kernel/bpf/verifier.c1777
1 files changed, 1777 insertions, 0 deletions
diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
new file mode 100644
index 000000000000..a086dd3210a8
--- /dev/null
+++ b/kernel/bpf/verifier.c
@@ -0,0 +1,1777 @@
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 */
129enum 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
141struct 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
154enum bpf_stack_slot_type {
155 STACK_INVALID, /* nothing was stored in this stack slot */
156 STACK_SPILL, /* 1st byte of register spilled into stack */
157 STACK_SPILL_PART, /* other 7 bytes of register spill */
158 STACK_MISC /* BPF program wrote some data into this slot */
159};
160
161struct bpf_stack_slot {
162 enum bpf_stack_slot_type stype;
163 struct reg_state reg_st;
164};
165
166/* state of the program:
167 * type of all registers and stack info
168 */
169struct verifier_state {
170 struct reg_state regs[MAX_BPF_REG];
171 struct bpf_stack_slot stack[MAX_BPF_STACK];
172};
173
174/* linked list of verifier states used to prune search */
175struct verifier_state_list {
176 struct verifier_state state;
177 struct verifier_state_list *next;
178};
179
180/* verifier_state + insn_idx are pushed to stack when branch is encountered */
181struct verifier_stack_elem {
182 /* verifer state is 'st'
183 * before processing instruction 'insn_idx'
184 * and after processing instruction 'prev_insn_idx'
185 */
186 struct verifier_state st;
187 int insn_idx;
188 int prev_insn_idx;
189 struct verifier_stack_elem *next;
190};
191
192#define MAX_USED_MAPS 64 /* max number of maps accessed by one eBPF program */
193
194/* single container for all structs
195 * one verifier_env per bpf_check() call
196 */
197struct verifier_env {
198 struct bpf_prog *prog; /* eBPF program being verified */
199 struct verifier_stack_elem *head; /* stack of verifier states to be processed */
200 int stack_size; /* number of states to be processed */
201 struct verifier_state cur_state; /* current verifier state */
202 struct bpf_map *used_maps[MAX_USED_MAPS]; /* array of map's used by eBPF program */
203 u32 used_map_cnt; /* number of used maps */
204};
205
206/* verbose verifier prints what it's seeing
207 * bpf_check() is called under lock, so no race to access these global vars
208 */
209static u32 log_level, log_size, log_len;
210static char *log_buf;
211
212static DEFINE_MUTEX(bpf_verifier_lock);
213
214/* log_level controls verbosity level of eBPF verifier.
215 * verbose() is used to dump the verification trace to the log, so the user
216 * can figure out what's wrong with the program
217 */
218static void verbose(const char *fmt, ...)
219{
220 va_list args;
221
222 if (log_level == 0 || log_len >= log_size - 1)
223 return;
224
225 va_start(args, fmt);
226 log_len += vscnprintf(log_buf + log_len, log_size - log_len, fmt, args);
227 va_end(args);
228}
229
230/* string representation of 'enum bpf_reg_type' */
231static const char * const reg_type_str[] = {
232 [NOT_INIT] = "?",
233 [UNKNOWN_VALUE] = "inv",
234 [PTR_TO_CTX] = "ctx",
235 [CONST_PTR_TO_MAP] = "map_ptr",
236 [PTR_TO_MAP_VALUE] = "map_value",
237 [PTR_TO_MAP_VALUE_OR_NULL] = "map_value_or_null",
238 [FRAME_PTR] = "fp",
239 [PTR_TO_STACK] = "fp",
240 [CONST_IMM] = "imm",
241};
242
243static void print_verifier_state(struct verifier_env *env)
244{
245 enum bpf_reg_type t;
246 int i;
247
248 for (i = 0; i < MAX_BPF_REG; i++) {
249 t = env->cur_state.regs[i].type;
250 if (t == NOT_INIT)
251 continue;
252 verbose(" R%d=%s", i, reg_type_str[t]);
253 if (t == CONST_IMM || t == PTR_TO_STACK)
254 verbose("%d", env->cur_state.regs[i].imm);
255 else if (t == CONST_PTR_TO_MAP || t == PTR_TO_MAP_VALUE ||
256 t == PTR_TO_MAP_VALUE_OR_NULL)
257 verbose("(ks=%d,vs=%d)",
258 env->cur_state.regs[i].map_ptr->key_size,
259 env->cur_state.regs[i].map_ptr->value_size);
260 }
261 for (i = 0; i < MAX_BPF_STACK; i++) {
262 if (env->cur_state.stack[i].stype == STACK_SPILL)
263 verbose(" fp%d=%s", -MAX_BPF_STACK + i,
264 reg_type_str[env->cur_state.stack[i].reg_st.type]);
265 }
266 verbose("\n");
267}
268
269static const char *const bpf_class_string[] = {
270 [BPF_LD] = "ld",
271 [BPF_LDX] = "ldx",
272 [BPF_ST] = "st",
273 [BPF_STX] = "stx",
274 [BPF_ALU] = "alu",
275 [BPF_JMP] = "jmp",
276 [BPF_RET] = "BUG",
277 [BPF_ALU64] = "alu64",
278};
279
280static const char *const bpf_alu_string[] = {
281 [BPF_ADD >> 4] = "+=",
282 [BPF_SUB >> 4] = "-=",
283 [BPF_MUL >> 4] = "*=",
284 [BPF_DIV >> 4] = "/=",
285 [BPF_OR >> 4] = "|=",
286 [BPF_AND >> 4] = "&=",
287 [BPF_LSH >> 4] = "<<=",
288 [BPF_RSH >> 4] = ">>=",
289 [BPF_NEG >> 4] = "neg",
290 [BPF_MOD >> 4] = "%=",
291 [BPF_XOR >> 4] = "^=",
292 [BPF_MOV >> 4] = "=",
293 [BPF_ARSH >> 4] = "s>>=",
294 [BPF_END >> 4] = "endian",
295};
296
297static const char *const bpf_ldst_string[] = {
298 [BPF_W >> 3] = "u32",
299 [BPF_H >> 3] = "u16",
300 [BPF_B >> 3] = "u8",
301 [BPF_DW >> 3] = "u64",
302};
303
304static const char *const bpf_jmp_string[] = {
305 [BPF_JA >> 4] = "jmp",
306 [BPF_JEQ >> 4] = "==",
307 [BPF_JGT >> 4] = ">",
308 [BPF_JGE >> 4] = ">=",
309 [BPF_JSET >> 4] = "&",
310 [BPF_JNE >> 4] = "!=",
311 [BPF_JSGT >> 4] = "s>",
312 [BPF_JSGE >> 4] = "s>=",
313 [BPF_CALL >> 4] = "call",
314 [BPF_EXIT >> 4] = "exit",
315};
316
317static void print_bpf_insn(struct bpf_insn *insn)
318{
319 u8 class = BPF_CLASS(insn->code);
320
321 if (class == BPF_ALU || class == BPF_ALU64) {
322 if (BPF_SRC(insn->code) == BPF_X)
323 verbose("(%02x) %sr%d %s %sr%d\n",
324 insn->code, class == BPF_ALU ? "(u32) " : "",
325 insn->dst_reg,
326 bpf_alu_string[BPF_OP(insn->code) >> 4],
327 class == BPF_ALU ? "(u32) " : "",
328 insn->src_reg);
329 else
330 verbose("(%02x) %sr%d %s %s%d\n",
331 insn->code, class == BPF_ALU ? "(u32) " : "",
332 insn->dst_reg,
333 bpf_alu_string[BPF_OP(insn->code) >> 4],
334 class == BPF_ALU ? "(u32) " : "",
335 insn->imm);
336 } else if (class == BPF_STX) {
337 if (BPF_MODE(insn->code) == BPF_MEM)
338 verbose("(%02x) *(%s *)(r%d %+d) = r%d\n",
339 insn->code,
340 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
341 insn->dst_reg,
342 insn->off, insn->src_reg);
343 else if (BPF_MODE(insn->code) == BPF_XADD)
344 verbose("(%02x) lock *(%s *)(r%d %+d) += r%d\n",
345 insn->code,
346 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
347 insn->dst_reg, insn->off,
348 insn->src_reg);
349 else
350 verbose("BUG_%02x\n", insn->code);
351 } else if (class == BPF_ST) {
352 if (BPF_MODE(insn->code) != BPF_MEM) {
353 verbose("BUG_st_%02x\n", insn->code);
354 return;
355 }
356 verbose("(%02x) *(%s *)(r%d %+d) = %d\n",
357 insn->code,
358 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
359 insn->dst_reg,
360 insn->off, insn->imm);
361 } else if (class == BPF_LDX) {
362 if (BPF_MODE(insn->code) != BPF_MEM) {
363 verbose("BUG_ldx_%02x\n", insn->code);
364 return;
365 }
366 verbose("(%02x) r%d = *(%s *)(r%d %+d)\n",
367 insn->code, insn->dst_reg,
368 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
369 insn->src_reg, insn->off);
370 } else if (class == BPF_LD) {
371 if (BPF_MODE(insn->code) == BPF_ABS) {
372 verbose("(%02x) r0 = *(%s *)skb[%d]\n",
373 insn->code,
374 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
375 insn->imm);
376 } else if (BPF_MODE(insn->code) == BPF_IND) {
377 verbose("(%02x) r0 = *(%s *)skb[r%d + %d]\n",
378 insn->code,
379 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
380 insn->src_reg, insn->imm);
381 } else if (BPF_MODE(insn->code) == BPF_IMM) {
382 verbose("(%02x) r%d = 0x%x\n",
383 insn->code, insn->dst_reg, insn->imm);
384 } else {
385 verbose("BUG_ld_%02x\n", insn->code);
386 return;
387 }
388 } else if (class == BPF_JMP) {
389 u8 opcode = BPF_OP(insn->code);
390
391 if (opcode == BPF_CALL) {
392 verbose("(%02x) call %d\n", insn->code, insn->imm);
393 } else if (insn->code == (BPF_JMP | BPF_JA)) {
394 verbose("(%02x) goto pc%+d\n",
395 insn->code, insn->off);
396 } else if (insn->code == (BPF_JMP | BPF_EXIT)) {
397 verbose("(%02x) exit\n", insn->code);
398 } else if (BPF_SRC(insn->code) == BPF_X) {
399 verbose("(%02x) if r%d %s r%d goto pc%+d\n",
400 insn->code, insn->dst_reg,
401 bpf_jmp_string[BPF_OP(insn->code) >> 4],
402 insn->src_reg, insn->off);
403 } else {
404 verbose("(%02x) if r%d %s 0x%x goto pc%+d\n",
405 insn->code, insn->dst_reg,
406 bpf_jmp_string[BPF_OP(insn->code) >> 4],
407 insn->imm, insn->off);
408 }
409 } else {
410 verbose("(%02x) %s\n", insn->code, bpf_class_string[class]);
411 }
412}
413
414static int pop_stack(struct verifier_env *env, int *prev_insn_idx)
415{
416 struct verifier_stack_elem *elem;
417 int insn_idx;
418
419 if (env->head == NULL)
420 return -1;
421
422 memcpy(&env->cur_state, &env->head->st, sizeof(env->cur_state));
423 insn_idx = env->head->insn_idx;
424 if (prev_insn_idx)
425 *prev_insn_idx = env->head->prev_insn_idx;
426 elem = env->head->next;
427 kfree(env->head);
428 env->head = elem;
429 env->stack_size--;
430 return insn_idx;
431}
432
433static struct verifier_state *push_stack(struct verifier_env *env, int insn_idx,
434 int prev_insn_idx)
435{
436 struct verifier_stack_elem *elem;
437
438 elem = kmalloc(sizeof(struct verifier_stack_elem), GFP_KERNEL);
439 if (!elem)
440 goto err;
441
442 memcpy(&elem->st, &env->cur_state, sizeof(env->cur_state));
443 elem->insn_idx = insn_idx;
444 elem->prev_insn_idx = prev_insn_idx;
445 elem->next = env->head;
446 env->head = elem;
447 env->stack_size++;
448 if (env->stack_size > 1024) {
449 verbose("BPF program is too complex\n");
450 goto err;
451 }
452 return &elem->st;
453err:
454 /* pop all elements and return */
455 while (pop_stack(env, NULL) >= 0);
456 return NULL;
457}
458
459#define CALLER_SAVED_REGS 6
460static const int caller_saved[CALLER_SAVED_REGS] = {
461 BPF_REG_0, BPF_REG_1, BPF_REG_2, BPF_REG_3, BPF_REG_4, BPF_REG_5
462};
463
464static void init_reg_state(struct reg_state *regs)
465{
466 int i;
467
468 for (i = 0; i < MAX_BPF_REG; i++) {
469 regs[i].type = NOT_INIT;
470 regs[i].imm = 0;
471 regs[i].map_ptr = NULL;
472 }
473
474 /* frame pointer */
475 regs[BPF_REG_FP].type = FRAME_PTR;
476
477 /* 1st arg to a function */
478 regs[BPF_REG_1].type = PTR_TO_CTX;
479}
480
481static void mark_reg_unknown_value(struct reg_state *regs, u32 regno)
482{
483 BUG_ON(regno >= MAX_BPF_REG);
484 regs[regno].type = UNKNOWN_VALUE;
485 regs[regno].imm = 0;
486 regs[regno].map_ptr = NULL;
487}
488
489enum reg_arg_type {
490 SRC_OP, /* register is used as source operand */
491 DST_OP, /* register is used as destination operand */
492 DST_OP_NO_MARK /* same as above, check only, don't mark */
493};
494
495static int check_reg_arg(struct reg_state *regs, u32 regno,
496 enum reg_arg_type t)
497{
498 if (regno >= MAX_BPF_REG) {
499 verbose("R%d is invalid\n", regno);
500 return -EINVAL;
501 }
502
503 if (t == SRC_OP) {
504 /* check whether register used as source operand can be read */
505 if (regs[regno].type == NOT_INIT) {
506 verbose("R%d !read_ok\n", regno);
507 return -EACCES;
508 }
509 } else {
510 /* check whether register used as dest operand can be written to */
511 if (regno == BPF_REG_FP) {
512 verbose("frame pointer is read only\n");
513 return -EACCES;
514 }
515 if (t == DST_OP)
516 mark_reg_unknown_value(regs, regno);
517 }
518 return 0;
519}
520
521static int bpf_size_to_bytes(int bpf_size)
522{
523 if (bpf_size == BPF_W)
524 return 4;
525 else if (bpf_size == BPF_H)
526 return 2;
527 else if (bpf_size == BPF_B)
528 return 1;
529 else if (bpf_size == BPF_DW)
530 return 8;
531 else
532 return -EINVAL;
533}
534
535/* check_stack_read/write functions track spill/fill of registers,
536 * stack boundary and alignment are checked in check_mem_access()
537 */
538static int check_stack_write(struct verifier_state *state, int off, int size,
539 int value_regno)
540{
541 struct bpf_stack_slot *slot;
542 int i;
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 != 8) {
551 verbose("invalid size of register spill\n");
552 return -EACCES;
553 }
554
555 slot = &state->stack[MAX_BPF_STACK + off];
556 slot->stype = STACK_SPILL;
557 /* save register state */
558 slot->reg_st = state->regs[value_regno];
559 for (i = 1; i < 8; i++) {
560 slot = &state->stack[MAX_BPF_STACK + off + i];
561 slot->stype = STACK_SPILL_PART;
562 slot->reg_st.type = UNKNOWN_VALUE;
563 slot->reg_st.map_ptr = NULL;
564 }
565 } else {
566
567 /* regular write of data into stack */
568 for (i = 0; i < size; i++) {
569 slot = &state->stack[MAX_BPF_STACK + off + i];
570 slot->stype = STACK_MISC;
571 slot->reg_st.type = UNKNOWN_VALUE;
572 slot->reg_st.map_ptr = NULL;
573 }
574 }
575 return 0;
576}
577
578static int check_stack_read(struct verifier_state *state, int off, int size,
579 int value_regno)
580{
581 int i;
582 struct bpf_stack_slot *slot;
583
584 slot = &state->stack[MAX_BPF_STACK + off];
585
586 if (slot->stype == STACK_SPILL) {
587 if (size != 8) {
588 verbose("invalid size of register spill\n");
589 return -EACCES;
590 }
591 for (i = 1; i < 8; i++) {
592 if (state->stack[MAX_BPF_STACK + off + i].stype !=
593 STACK_SPILL_PART) {
594 verbose("corrupted spill memory\n");
595 return -EACCES;
596 }
597 }
598
599 if (value_regno >= 0)
600 /* restore register state from stack */
601 state->regs[value_regno] = slot->reg_st;
602 return 0;
603 } else {
604 for (i = 0; i < size; i++) {
605 if (state->stack[MAX_BPF_STACK + off + i].stype !=
606 STACK_MISC) {
607 verbose("invalid read from stack off %d+%d size %d\n",
608 off, i, size);
609 return -EACCES;
610 }
611 }
612 if (value_regno >= 0)
613 /* have read misc data from the stack */
614 mark_reg_unknown_value(state->regs, value_regno);
615 return 0;
616 }
617}
618
619/* check read/write into map element returned by bpf_map_lookup_elem() */
620static int check_map_access(struct verifier_env *env, u32 regno, int off,
621 int size)
622{
623 struct bpf_map *map = env->cur_state.regs[regno].map_ptr;
624
625 if (off < 0 || off + size > map->value_size) {
626 verbose("invalid access to map value, value_size=%d off=%d size=%d\n",
627 map->value_size, off, size);
628 return -EACCES;
629 }
630 return 0;
631}
632
633/* check access to 'struct bpf_context' fields */
634static int check_ctx_access(struct verifier_env *env, int off, int size,
635 enum bpf_access_type t)
636{
637 if (env->prog->aux->ops->is_valid_access &&
638 env->prog->aux->ops->is_valid_access(off, size, t))
639 return 0;
640
641 verbose("invalid bpf_context access off=%d size=%d\n", off, size);
642 return -EACCES;
643}
644
645/* check whether memory at (regno + off) is accessible for t = (read | write)
646 * if t==write, value_regno is a register which value is stored into memory
647 * if t==read, value_regno is a register which will receive the value from memory
648 * if t==write && value_regno==-1, some unknown value is stored into memory
649 * if t==read && value_regno==-1, don't care what we read from memory
650 */
651static int check_mem_access(struct verifier_env *env, u32 regno, int off,
652 int bpf_size, enum bpf_access_type t,
653 int value_regno)
654{
655 struct verifier_state *state = &env->cur_state;
656 int size, err = 0;
657
658 size = bpf_size_to_bytes(bpf_size);
659 if (size < 0)
660 return size;
661
662 if (off % size != 0) {
663 verbose("misaligned access off %d size %d\n", off, size);
664 return -EACCES;
665 }
666
667 if (state->regs[regno].type == PTR_TO_MAP_VALUE) {
668 err = check_map_access(env, regno, off, size);
669 if (!err && t == BPF_READ && value_regno >= 0)
670 mark_reg_unknown_value(state->regs, value_regno);
671
672 } else if (state->regs[regno].type == PTR_TO_CTX) {
673 err = check_ctx_access(env, off, size, t);
674 if (!err && t == BPF_READ && value_regno >= 0)
675 mark_reg_unknown_value(state->regs, value_regno);
676
677 } else if (state->regs[regno].type == FRAME_PTR) {
678 if (off >= 0 || off < -MAX_BPF_STACK) {
679 verbose("invalid stack off=%d size=%d\n", off, size);
680 return -EACCES;
681 }
682 if (t == BPF_WRITE)
683 err = check_stack_write(state, off, size, value_regno);
684 else
685 err = check_stack_read(state, off, size, value_regno);
686 } else {
687 verbose("R%d invalid mem access '%s'\n",
688 regno, reg_type_str[state->regs[regno].type]);
689 return -EACCES;
690 }
691 return err;
692}
693
694static int check_xadd(struct verifier_env *env, struct bpf_insn *insn)
695{
696 struct reg_state *regs = env->cur_state.regs;
697 int err;
698
699 if ((BPF_SIZE(insn->code) != BPF_W && BPF_SIZE(insn->code) != BPF_DW) ||
700 insn->imm != 0) {
701 verbose("BPF_XADD uses reserved fields\n");
702 return -EINVAL;
703 }
704
705 /* check src1 operand */
706 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
707 if (err)
708 return err;
709
710 /* check src2 operand */
711 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
712 if (err)
713 return err;
714
715 /* check whether atomic_add can read the memory */
716 err = check_mem_access(env, insn->dst_reg, insn->off,
717 BPF_SIZE(insn->code), BPF_READ, -1);
718 if (err)
719 return err;
720
721 /* check whether atomic_add can write into the same memory */
722 return check_mem_access(env, insn->dst_reg, insn->off,
723 BPF_SIZE(insn->code), BPF_WRITE, -1);
724}
725
726/* when register 'regno' is passed into function that will read 'access_size'
727 * bytes from that pointer, make sure that it's within stack boundary
728 * and all elements of stack are initialized
729 */
730static int check_stack_boundary(struct verifier_env *env,
731 int regno, int access_size)
732{
733 struct verifier_state *state = &env->cur_state;
734 struct reg_state *regs = state->regs;
735 int off, i;
736
737 if (regs[regno].type != PTR_TO_STACK)
738 return -EACCES;
739
740 off = regs[regno].imm;
741 if (off >= 0 || off < -MAX_BPF_STACK || off + access_size > 0 ||
742 access_size <= 0) {
743 verbose("invalid stack type R%d off=%d access_size=%d\n",
744 regno, off, access_size);
745 return -EACCES;
746 }
747
748 for (i = 0; i < access_size; i++) {
749 if (state->stack[MAX_BPF_STACK + off + i].stype != STACK_MISC) {
750 verbose("invalid indirect read from stack off %d+%d size %d\n",
751 off, i, access_size);
752 return -EACCES;
753 }
754 }
755 return 0;
756}
757
758static int check_func_arg(struct verifier_env *env, u32 regno,
759 enum bpf_arg_type arg_type, struct bpf_map **mapp)
760{
761 struct reg_state *reg = env->cur_state.regs + regno;
762 enum bpf_reg_type expected_type;
763 int err = 0;
764
765 if (arg_type == ARG_ANYTHING)
766 return 0;
767
768 if (reg->type == NOT_INIT) {
769 verbose("R%d !read_ok\n", regno);
770 return -EACCES;
771 }
772
773 if (arg_type == ARG_PTR_TO_STACK || arg_type == ARG_PTR_TO_MAP_KEY ||
774 arg_type == ARG_PTR_TO_MAP_VALUE) {
775 expected_type = PTR_TO_STACK;
776 } else if (arg_type == ARG_CONST_STACK_SIZE) {
777 expected_type = CONST_IMM;
778 } else if (arg_type == ARG_CONST_MAP_PTR) {
779 expected_type = CONST_PTR_TO_MAP;
780 } else {
781 verbose("unsupported arg_type %d\n", arg_type);
782 return -EFAULT;
783 }
784
785 if (reg->type != expected_type) {
786 verbose("R%d type=%s expected=%s\n", regno,
787 reg_type_str[reg->type], reg_type_str[expected_type]);
788 return -EACCES;
789 }
790
791 if (arg_type == ARG_CONST_MAP_PTR) {
792 /* bpf_map_xxx(map_ptr) call: remember that map_ptr */
793 *mapp = reg->map_ptr;
794
795 } else if (arg_type == ARG_PTR_TO_MAP_KEY) {
796 /* bpf_map_xxx(..., map_ptr, ..., key) call:
797 * check that [key, key + map->key_size) are within
798 * stack limits and initialized
799 */
800 if (!*mapp) {
801 /* in function declaration map_ptr must come before
802 * map_key, so that it's verified and known before
803 * we have to check map_key here. Otherwise it means
804 * that kernel subsystem misconfigured verifier
805 */
806 verbose("invalid map_ptr to access map->key\n");
807 return -EACCES;
808 }
809 err = check_stack_boundary(env, regno, (*mapp)->key_size);
810
811 } else if (arg_type == ARG_PTR_TO_MAP_VALUE) {
812 /* bpf_map_xxx(..., map_ptr, ..., value) call:
813 * check [value, value + map->value_size) validity
814 */
815 if (!*mapp) {
816 /* kernel subsystem misconfigured verifier */
817 verbose("invalid map_ptr to access map->value\n");
818 return -EACCES;
819 }
820 err = check_stack_boundary(env, regno, (*mapp)->value_size);
821
822 } else if (arg_type == ARG_CONST_STACK_SIZE) {
823 /* bpf_xxx(..., buf, len) call will access 'len' bytes
824 * from stack pointer 'buf'. Check it
825 * note: regno == len, regno - 1 == buf
826 */
827 if (regno == 0) {
828 /* kernel subsystem misconfigured verifier */
829 verbose("ARG_CONST_STACK_SIZE cannot be first argument\n");
830 return -EACCES;
831 }
832 err = check_stack_boundary(env, regno - 1, reg->imm);
833 }
834
835 return err;
836}
837
838static int check_call(struct verifier_env *env, int func_id)
839{
840 struct verifier_state *state = &env->cur_state;
841 const struct bpf_func_proto *fn = NULL;
842 struct reg_state *regs = state->regs;
843 struct bpf_map *map = NULL;
844 struct reg_state *reg;
845 int i, err;
846
847 /* find function prototype */
848 if (func_id < 0 || func_id >= __BPF_FUNC_MAX_ID) {
849 verbose("invalid func %d\n", func_id);
850 return -EINVAL;
851 }
852
853 if (env->prog->aux->ops->get_func_proto)
854 fn = env->prog->aux->ops->get_func_proto(func_id);
855
856 if (!fn) {
857 verbose("unknown func %d\n", func_id);
858 return -EINVAL;
859 }
860
861 /* eBPF programs must be GPL compatible to use GPL-ed functions */
862 if (!env->prog->aux->is_gpl_compatible && fn->gpl_only) {
863 verbose("cannot call GPL only function from proprietary program\n");
864 return -EINVAL;
865 }
866
867 /* check args */
868 err = check_func_arg(env, BPF_REG_1, fn->arg1_type, &map);
869 if (err)
870 return err;
871 err = check_func_arg(env, BPF_REG_2, fn->arg2_type, &map);
872 if (err)
873 return err;
874 err = check_func_arg(env, BPF_REG_3, fn->arg3_type, &map);
875 if (err)
876 return err;
877 err = check_func_arg(env, BPF_REG_4, fn->arg4_type, &map);
878 if (err)
879 return err;
880 err = check_func_arg(env, BPF_REG_5, fn->arg5_type, &map);
881 if (err)
882 return err;
883
884 /* reset caller saved regs */
885 for (i = 0; i < CALLER_SAVED_REGS; i++) {
886 reg = regs + caller_saved[i];
887 reg->type = NOT_INIT;
888 reg->imm = 0;
889 }
890
891 /* update return register */
892 if (fn->ret_type == RET_INTEGER) {
893 regs[BPF_REG_0].type = UNKNOWN_VALUE;
894 } else if (fn->ret_type == RET_VOID) {
895 regs[BPF_REG_0].type = NOT_INIT;
896 } else if (fn->ret_type == RET_PTR_TO_MAP_VALUE_OR_NULL) {
897 regs[BPF_REG_0].type = PTR_TO_MAP_VALUE_OR_NULL;
898 /* remember map_ptr, so that check_map_access()
899 * can check 'value_size' boundary of memory access
900 * to map element returned from bpf_map_lookup_elem()
901 */
902 if (map == NULL) {
903 verbose("kernel subsystem misconfigured verifier\n");
904 return -EINVAL;
905 }
906 regs[BPF_REG_0].map_ptr = map;
907 } else {
908 verbose("unknown return type %d of func %d\n",
909 fn->ret_type, func_id);
910 return -EINVAL;
911 }
912 return 0;
913}
914
915/* check validity of 32-bit and 64-bit arithmetic operations */
916static int check_alu_op(struct reg_state *regs, struct bpf_insn *insn)
917{
918 u8 opcode = BPF_OP(insn->code);
919 int err;
920
921 if (opcode == BPF_END || opcode == BPF_NEG) {
922 if (opcode == BPF_NEG) {
923 if (BPF_SRC(insn->code) != 0 ||
924 insn->src_reg != BPF_REG_0 ||
925 insn->off != 0 || insn->imm != 0) {
926 verbose("BPF_NEG uses reserved fields\n");
927 return -EINVAL;
928 }
929 } else {
930 if (insn->src_reg != BPF_REG_0 || insn->off != 0 ||
931 (insn->imm != 16 && insn->imm != 32 && insn->imm != 64)) {
932 verbose("BPF_END uses reserved fields\n");
933 return -EINVAL;
934 }
935 }
936
937 /* check src operand */
938 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
939 if (err)
940 return err;
941
942 /* check dest operand */
943 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
944 if (err)
945 return err;
946
947 } else if (opcode == BPF_MOV) {
948
949 if (BPF_SRC(insn->code) == BPF_X) {
950 if (insn->imm != 0 || insn->off != 0) {
951 verbose("BPF_MOV uses reserved fields\n");
952 return -EINVAL;
953 }
954
955 /* check src operand */
956 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
957 if (err)
958 return err;
959 } else {
960 if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
961 verbose("BPF_MOV uses reserved fields\n");
962 return -EINVAL;
963 }
964 }
965
966 /* check dest operand */
967 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
968 if (err)
969 return err;
970
971 if (BPF_SRC(insn->code) == BPF_X) {
972 if (BPF_CLASS(insn->code) == BPF_ALU64) {
973 /* case: R1 = R2
974 * copy register state to dest reg
975 */
976 regs[insn->dst_reg] = regs[insn->src_reg];
977 } else {
978 regs[insn->dst_reg].type = UNKNOWN_VALUE;
979 regs[insn->dst_reg].map_ptr = NULL;
980 }
981 } else {
982 /* case: R = imm
983 * remember the value we stored into this reg
984 */
985 regs[insn->dst_reg].type = CONST_IMM;
986 regs[insn->dst_reg].imm = insn->imm;
987 }
988
989 } else if (opcode > BPF_END) {
990 verbose("invalid BPF_ALU opcode %x\n", opcode);
991 return -EINVAL;
992
993 } else { /* all other ALU ops: and, sub, xor, add, ... */
994
995 bool stack_relative = false;
996
997 if (BPF_SRC(insn->code) == BPF_X) {
998 if (insn->imm != 0 || insn->off != 0) {
999 verbose("BPF_ALU uses reserved fields\n");
1000 return -EINVAL;
1001 }
1002 /* check src1 operand */
1003 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1004 if (err)
1005 return err;
1006 } else {
1007 if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
1008 verbose("BPF_ALU uses reserved fields\n");
1009 return -EINVAL;
1010 }
1011 }
1012
1013 /* check src2 operand */
1014 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1015 if (err)
1016 return err;
1017
1018 if ((opcode == BPF_MOD || opcode == BPF_DIV) &&
1019 BPF_SRC(insn->code) == BPF_K && insn->imm == 0) {
1020 verbose("div by zero\n");
1021 return -EINVAL;
1022 }
1023
1024 /* pattern match 'bpf_add Rx, imm' instruction */
1025 if (opcode == BPF_ADD && BPF_CLASS(insn->code) == BPF_ALU64 &&
1026 regs[insn->dst_reg].type == FRAME_PTR &&
1027 BPF_SRC(insn->code) == BPF_K)
1028 stack_relative = true;
1029
1030 /* check dest operand */
1031 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1032 if (err)
1033 return err;
1034
1035 if (stack_relative) {
1036 regs[insn->dst_reg].type = PTR_TO_STACK;
1037 regs[insn->dst_reg].imm = insn->imm;
1038 }
1039 }
1040
1041 return 0;
1042}
1043
1044static int check_cond_jmp_op(struct verifier_env *env,
1045 struct bpf_insn *insn, int *insn_idx)
1046{
1047 struct reg_state *regs = env->cur_state.regs;
1048 struct verifier_state *other_branch;
1049 u8 opcode = BPF_OP(insn->code);
1050 int err;
1051
1052 if (opcode > BPF_EXIT) {
1053 verbose("invalid BPF_JMP opcode %x\n", opcode);
1054 return -EINVAL;
1055 }
1056
1057 if (BPF_SRC(insn->code) == BPF_X) {
1058 if (insn->imm != 0) {
1059 verbose("BPF_JMP uses reserved fields\n");
1060 return -EINVAL;
1061 }
1062
1063 /* check src1 operand */
1064 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1065 if (err)
1066 return err;
1067 } else {
1068 if (insn->src_reg != BPF_REG_0) {
1069 verbose("BPF_JMP uses reserved fields\n");
1070 return -EINVAL;
1071 }
1072 }
1073
1074 /* check src2 operand */
1075 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1076 if (err)
1077 return err;
1078
1079 /* detect if R == 0 where R was initialized to zero earlier */
1080 if (BPF_SRC(insn->code) == BPF_K &&
1081 (opcode == BPF_JEQ || opcode == BPF_JNE) &&
1082 regs[insn->dst_reg].type == CONST_IMM &&
1083 regs[insn->dst_reg].imm == insn->imm) {
1084 if (opcode == BPF_JEQ) {
1085 /* if (imm == imm) goto pc+off;
1086 * only follow the goto, ignore fall-through
1087 */
1088 *insn_idx += insn->off;
1089 return 0;
1090 } else {
1091 /* if (imm != imm) goto pc+off;
1092 * only follow fall-through branch, since
1093 * that's where the program will go
1094 */
1095 return 0;
1096 }
1097 }
1098
1099 other_branch = push_stack(env, *insn_idx + insn->off + 1, *insn_idx);
1100 if (!other_branch)
1101 return -EFAULT;
1102
1103 /* detect if R == 0 where R is returned value from bpf_map_lookup_elem() */
1104 if (BPF_SRC(insn->code) == BPF_K &&
1105 insn->imm == 0 && (opcode == BPF_JEQ ||
1106 opcode == BPF_JNE) &&
1107 regs[insn->dst_reg].type == PTR_TO_MAP_VALUE_OR_NULL) {
1108 if (opcode == BPF_JEQ) {
1109 /* next fallthrough insn can access memory via
1110 * this register
1111 */
1112 regs[insn->dst_reg].type = PTR_TO_MAP_VALUE;
1113 /* branch targer cannot access it, since reg == 0 */
1114 other_branch->regs[insn->dst_reg].type = CONST_IMM;
1115 other_branch->regs[insn->dst_reg].imm = 0;
1116 } else {
1117 other_branch->regs[insn->dst_reg].type = PTR_TO_MAP_VALUE;
1118 regs[insn->dst_reg].type = CONST_IMM;
1119 regs[insn->dst_reg].imm = 0;
1120 }
1121 } else if (BPF_SRC(insn->code) == BPF_K &&
1122 (opcode == BPF_JEQ || opcode == BPF_JNE)) {
1123
1124 if (opcode == BPF_JEQ) {
1125 /* detect if (R == imm) goto
1126 * and in the target state recognize that R = imm
1127 */
1128 other_branch->regs[insn->dst_reg].type = CONST_IMM;
1129 other_branch->regs[insn->dst_reg].imm = insn->imm;
1130 } else {
1131 /* detect if (R != imm) goto
1132 * and in the fall-through state recognize that R = imm
1133 */
1134 regs[insn->dst_reg].type = CONST_IMM;
1135 regs[insn->dst_reg].imm = insn->imm;
1136 }
1137 }
1138 if (log_level)
1139 print_verifier_state(env);
1140 return 0;
1141}
1142
1143/* return the map pointer stored inside BPF_LD_IMM64 instruction */
1144static struct bpf_map *ld_imm64_to_map_ptr(struct bpf_insn *insn)
1145{
1146 u64 imm64 = ((u64) (u32) insn[0].imm) | ((u64) (u32) insn[1].imm) << 32;
1147
1148 return (struct bpf_map *) (unsigned long) imm64;
1149}
1150
1151/* verify BPF_LD_IMM64 instruction */
1152static int check_ld_imm(struct verifier_env *env, struct bpf_insn *insn)
1153{
1154 struct reg_state *regs = env->cur_state.regs;
1155 int err;
1156
1157 if (BPF_SIZE(insn->code) != BPF_DW) {
1158 verbose("invalid BPF_LD_IMM insn\n");
1159 return -EINVAL;
1160 }
1161 if (insn->off != 0) {
1162 verbose("BPF_LD_IMM64 uses reserved fields\n");
1163 return -EINVAL;
1164 }
1165
1166 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1167 if (err)
1168 return err;
1169
1170 if (insn->src_reg == 0)
1171 /* generic move 64-bit immediate into a register */
1172 return 0;
1173
1174 /* replace_map_fd_with_map_ptr() should have caught bad ld_imm64 */
1175 BUG_ON(insn->src_reg != BPF_PSEUDO_MAP_FD);
1176
1177 regs[insn->dst_reg].type = CONST_PTR_TO_MAP;
1178 regs[insn->dst_reg].map_ptr = ld_imm64_to_map_ptr(insn);
1179 return 0;
1180}
1181
1182/* non-recursive DFS pseudo code
1183 * 1 procedure DFS-iterative(G,v):
1184 * 2 label v as discovered
1185 * 3 let S be a stack
1186 * 4 S.push(v)
1187 * 5 while S is not empty
1188 * 6 t <- S.pop()
1189 * 7 if t is what we're looking for:
1190 * 8 return t
1191 * 9 for all edges e in G.adjacentEdges(t) do
1192 * 10 if edge e is already labelled
1193 * 11 continue with the next edge
1194 * 12 w <- G.adjacentVertex(t,e)
1195 * 13 if vertex w is not discovered and not explored
1196 * 14 label e as tree-edge
1197 * 15 label w as discovered
1198 * 16 S.push(w)
1199 * 17 continue at 5
1200 * 18 else if vertex w is discovered
1201 * 19 label e as back-edge
1202 * 20 else
1203 * 21 // vertex w is explored
1204 * 22 label e as forward- or cross-edge
1205 * 23 label t as explored
1206 * 24 S.pop()
1207 *
1208 * convention:
1209 * 0x10 - discovered
1210 * 0x11 - discovered and fall-through edge labelled
1211 * 0x12 - discovered and fall-through and branch edges labelled
1212 * 0x20 - explored
1213 */
1214
1215enum {
1216 DISCOVERED = 0x10,
1217 EXPLORED = 0x20,
1218 FALLTHROUGH = 1,
1219 BRANCH = 2,
1220};
1221
1222static int *insn_stack; /* stack of insns to process */
1223static int cur_stack; /* current stack index */
1224static int *insn_state;
1225
1226/* t, w, e - match pseudo-code above:
1227 * t - index of current instruction
1228 * w - next instruction
1229 * e - edge
1230 */
1231static int push_insn(int t, int w, int e, struct verifier_env *env)
1232{
1233 if (e == FALLTHROUGH && insn_state[t] >= (DISCOVERED | FALLTHROUGH))
1234 return 0;
1235
1236 if (e == BRANCH && insn_state[t] >= (DISCOVERED | BRANCH))
1237 return 0;
1238
1239 if (w < 0 || w >= env->prog->len) {
1240 verbose("jump out of range from insn %d to %d\n", t, w);
1241 return -EINVAL;
1242 }
1243
1244 if (insn_state[w] == 0) {
1245 /* tree-edge */
1246 insn_state[t] = DISCOVERED | e;
1247 insn_state[w] = DISCOVERED;
1248 if (cur_stack >= env->prog->len)
1249 return -E2BIG;
1250 insn_stack[cur_stack++] = w;
1251 return 1;
1252 } else if ((insn_state[w] & 0xF0) == DISCOVERED) {
1253 verbose("back-edge from insn %d to %d\n", t, w);
1254 return -EINVAL;
1255 } else if (insn_state[w] == EXPLORED) {
1256 /* forward- or cross-edge */
1257 insn_state[t] = DISCOVERED | e;
1258 } else {
1259 verbose("insn state internal bug\n");
1260 return -EFAULT;
1261 }
1262 return 0;
1263}
1264
1265/* non-recursive depth-first-search to detect loops in BPF program
1266 * loop == back-edge in directed graph
1267 */
1268static int check_cfg(struct verifier_env *env)
1269{
1270 struct bpf_insn *insns = env->prog->insnsi;
1271 int insn_cnt = env->prog->len;
1272 int ret = 0;
1273 int i, t;
1274
1275 insn_state = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
1276 if (!insn_state)
1277 return -ENOMEM;
1278
1279 insn_stack = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
1280 if (!insn_stack) {
1281 kfree(insn_state);
1282 return -ENOMEM;
1283 }
1284
1285 insn_state[0] = DISCOVERED; /* mark 1st insn as discovered */
1286 insn_stack[0] = 0; /* 0 is the first instruction */
1287 cur_stack = 1;
1288
1289peek_stack:
1290 if (cur_stack == 0)
1291 goto check_state;
1292 t = insn_stack[cur_stack - 1];
1293
1294 if (BPF_CLASS(insns[t].code) == BPF_JMP) {
1295 u8 opcode = BPF_OP(insns[t].code);
1296
1297 if (opcode == BPF_EXIT) {
1298 goto mark_explored;
1299 } else if (opcode == BPF_CALL) {
1300 ret = push_insn(t, t + 1, FALLTHROUGH, env);
1301 if (ret == 1)
1302 goto peek_stack;
1303 else if (ret < 0)
1304 goto err_free;
1305 } else if (opcode == BPF_JA) {
1306 if (BPF_SRC(insns[t].code) != BPF_K) {
1307 ret = -EINVAL;
1308 goto err_free;
1309 }
1310 /* unconditional jump with single edge */
1311 ret = push_insn(t, t + insns[t].off + 1,
1312 FALLTHROUGH, env);
1313 if (ret == 1)
1314 goto peek_stack;
1315 else if (ret < 0)
1316 goto err_free;
1317 } else {
1318 /* conditional jump with two edges */
1319 ret = push_insn(t, t + 1, FALLTHROUGH, env);
1320 if (ret == 1)
1321 goto peek_stack;
1322 else if (ret < 0)
1323 goto err_free;
1324
1325 ret = push_insn(t, t + insns[t].off + 1, BRANCH, env);
1326 if (ret == 1)
1327 goto peek_stack;
1328 else if (ret < 0)
1329 goto err_free;
1330 }
1331 } else {
1332 /* all other non-branch instructions with single
1333 * fall-through edge
1334 */
1335 ret = push_insn(t, t + 1, FALLTHROUGH, env);
1336 if (ret == 1)
1337 goto peek_stack;
1338 else if (ret < 0)
1339 goto err_free;
1340 }
1341
1342mark_explored:
1343 insn_state[t] = EXPLORED;
1344 if (cur_stack-- <= 0) {
1345 verbose("pop stack internal bug\n");
1346 ret = -EFAULT;
1347 goto err_free;
1348 }
1349 goto peek_stack;
1350
1351check_state:
1352 for (i = 0; i < insn_cnt; i++) {
1353 if (insn_state[i] != EXPLORED) {
1354 verbose("unreachable insn %d\n", i);
1355 ret = -EINVAL;
1356 goto err_free;
1357 }
1358 }
1359 ret = 0; /* cfg looks good */
1360
1361err_free:
1362 kfree(insn_state);
1363 kfree(insn_stack);
1364 return ret;
1365}
1366
1367static int do_check(struct verifier_env *env)
1368{
1369 struct verifier_state *state = &env->cur_state;
1370 struct bpf_insn *insns = env->prog->insnsi;
1371 struct reg_state *regs = state->regs;
1372 int insn_cnt = env->prog->len;
1373 int insn_idx, prev_insn_idx = 0;
1374 int insn_processed = 0;
1375 bool do_print_state = false;
1376
1377 init_reg_state(regs);
1378 insn_idx = 0;
1379 for (;;) {
1380 struct bpf_insn *insn;
1381 u8 class;
1382 int err;
1383
1384 if (insn_idx >= insn_cnt) {
1385 verbose("invalid insn idx %d insn_cnt %d\n",
1386 insn_idx, insn_cnt);
1387 return -EFAULT;
1388 }
1389
1390 insn = &insns[insn_idx];
1391 class = BPF_CLASS(insn->code);
1392
1393 if (++insn_processed > 32768) {
1394 verbose("BPF program is too large. Proccessed %d insn\n",
1395 insn_processed);
1396 return -E2BIG;
1397 }
1398
1399 if (log_level && do_print_state) {
1400 verbose("\nfrom %d to %d:", prev_insn_idx, insn_idx);
1401 print_verifier_state(env);
1402 do_print_state = false;
1403 }
1404
1405 if (log_level) {
1406 verbose("%d: ", insn_idx);
1407 print_bpf_insn(insn);
1408 }
1409
1410 if (class == BPF_ALU || class == BPF_ALU64) {
1411 err = check_alu_op(regs, insn);
1412 if (err)
1413 return err;
1414
1415 } else if (class == BPF_LDX) {
1416 if (BPF_MODE(insn->code) != BPF_MEM ||
1417 insn->imm != 0) {
1418 verbose("BPF_LDX uses reserved fields\n");
1419 return -EINVAL;
1420 }
1421 /* check src operand */
1422 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1423 if (err)
1424 return err;
1425
1426 err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK);
1427 if (err)
1428 return err;
1429
1430 /* check that memory (src_reg + off) is readable,
1431 * the state of dst_reg will be updated by this func
1432 */
1433 err = check_mem_access(env, insn->src_reg, insn->off,
1434 BPF_SIZE(insn->code), BPF_READ,
1435 insn->dst_reg);
1436 if (err)
1437 return err;
1438
1439 } else if (class == BPF_STX) {
1440 if (BPF_MODE(insn->code) == BPF_XADD) {
1441 err = check_xadd(env, insn);
1442 if (err)
1443 return err;
1444 insn_idx++;
1445 continue;
1446 }
1447
1448 if (BPF_MODE(insn->code) != BPF_MEM ||
1449 insn->imm != 0) {
1450 verbose("BPF_STX uses reserved fields\n");
1451 return -EINVAL;
1452 }
1453 /* check src1 operand */
1454 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1455 if (err)
1456 return err;
1457 /* check src2 operand */
1458 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1459 if (err)
1460 return err;
1461
1462 /* check that memory (dst_reg + off) is writeable */
1463 err = check_mem_access(env, insn->dst_reg, insn->off,
1464 BPF_SIZE(insn->code), BPF_WRITE,
1465 insn->src_reg);
1466 if (err)
1467 return err;
1468
1469 } else if (class == BPF_ST) {
1470 if (BPF_MODE(insn->code) != BPF_MEM ||
1471 insn->src_reg != BPF_REG_0) {
1472 verbose("BPF_ST uses reserved fields\n");
1473 return -EINVAL;
1474 }
1475 /* check src operand */
1476 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1477 if (err)
1478 return err;
1479
1480 /* check that memory (dst_reg + off) is writeable */
1481 err = check_mem_access(env, insn->dst_reg, insn->off,
1482 BPF_SIZE(insn->code), BPF_WRITE,
1483 -1);
1484 if (err)
1485 return err;
1486
1487 } else if (class == BPF_JMP) {
1488 u8 opcode = BPF_OP(insn->code);
1489
1490 if (opcode == BPF_CALL) {
1491 if (BPF_SRC(insn->code) != BPF_K ||
1492 insn->off != 0 ||
1493 insn->src_reg != BPF_REG_0 ||
1494 insn->dst_reg != BPF_REG_0) {
1495 verbose("BPF_CALL uses reserved fields\n");
1496 return -EINVAL;
1497 }
1498
1499 err = check_call(env, insn->imm);
1500 if (err)
1501 return err;
1502
1503 } else if (opcode == BPF_JA) {
1504 if (BPF_SRC(insn->code) != BPF_K ||
1505 insn->imm != 0 ||
1506 insn->src_reg != BPF_REG_0 ||
1507 insn->dst_reg != BPF_REG_0) {
1508 verbose("BPF_JA uses reserved fields\n");
1509 return -EINVAL;
1510 }
1511
1512 insn_idx += insn->off + 1;
1513 continue;
1514
1515 } else if (opcode == BPF_EXIT) {
1516 if (BPF_SRC(insn->code) != BPF_K ||
1517 insn->imm != 0 ||
1518 insn->src_reg != BPF_REG_0 ||
1519 insn->dst_reg != BPF_REG_0) {
1520 verbose("BPF_EXIT uses reserved fields\n");
1521 return -EINVAL;
1522 }
1523
1524 /* eBPF calling convetion is such that R0 is used
1525 * to return the value from eBPF program.
1526 * Make sure that it's readable at this time
1527 * of bpf_exit, which means that program wrote
1528 * something into it earlier
1529 */
1530 err = check_reg_arg(regs, BPF_REG_0, SRC_OP);
1531 if (err)
1532 return err;
1533
1534 insn_idx = pop_stack(env, &prev_insn_idx);
1535 if (insn_idx < 0) {
1536 break;
1537 } else {
1538 do_print_state = true;
1539 continue;
1540 }
1541 } else {
1542 err = check_cond_jmp_op(env, insn, &insn_idx);
1543 if (err)
1544 return err;
1545 }
1546 } else if (class == BPF_LD) {
1547 u8 mode = BPF_MODE(insn->code);
1548
1549 if (mode == BPF_ABS || mode == BPF_IND) {
1550 verbose("LD_ABS is not supported yet\n");
1551 return -EINVAL;
1552 } else if (mode == BPF_IMM) {
1553 err = check_ld_imm(env, insn);
1554 if (err)
1555 return err;
1556
1557 insn_idx++;
1558 } else {
1559 verbose("invalid BPF_LD mode\n");
1560 return -EINVAL;
1561 }
1562 } else {
1563 verbose("unknown insn class %d\n", class);
1564 return -EINVAL;
1565 }
1566
1567 insn_idx++;
1568 }
1569
1570 return 0;
1571}
1572
1573/* look for pseudo eBPF instructions that access map FDs and
1574 * replace them with actual map pointers
1575 */
1576static int replace_map_fd_with_map_ptr(struct verifier_env *env)
1577{
1578 struct bpf_insn *insn = env->prog->insnsi;
1579 int insn_cnt = env->prog->len;
1580 int i, j;
1581
1582 for (i = 0; i < insn_cnt; i++, insn++) {
1583 if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW)) {
1584 struct bpf_map *map;
1585 struct fd f;
1586
1587 if (i == insn_cnt - 1 || insn[1].code != 0 ||
1588 insn[1].dst_reg != 0 || insn[1].src_reg != 0 ||
1589 insn[1].off != 0) {
1590 verbose("invalid bpf_ld_imm64 insn\n");
1591 return -EINVAL;
1592 }
1593
1594 if (insn->src_reg == 0)
1595 /* valid generic load 64-bit imm */
1596 goto next_insn;
1597
1598 if (insn->src_reg != BPF_PSEUDO_MAP_FD) {
1599 verbose("unrecognized bpf_ld_imm64 insn\n");
1600 return -EINVAL;
1601 }
1602
1603 f = fdget(insn->imm);
1604
1605 map = bpf_map_get(f);
1606 if (IS_ERR(map)) {
1607 verbose("fd %d is not pointing to valid bpf_map\n",
1608 insn->imm);
1609 fdput(f);
1610 return PTR_ERR(map);
1611 }
1612
1613 /* store map pointer inside BPF_LD_IMM64 instruction */
1614 insn[0].imm = (u32) (unsigned long) map;
1615 insn[1].imm = ((u64) (unsigned long) map) >> 32;
1616
1617 /* check whether we recorded this map already */
1618 for (j = 0; j < env->used_map_cnt; j++)
1619 if (env->used_maps[j] == map) {
1620 fdput(f);
1621 goto next_insn;
1622 }
1623
1624 if (env->used_map_cnt >= MAX_USED_MAPS) {
1625 fdput(f);
1626 return -E2BIG;
1627 }
1628
1629 /* remember this map */
1630 env->used_maps[env->used_map_cnt++] = map;
1631
1632 /* hold the map. If the program is rejected by verifier,
1633 * the map will be released by release_maps() or it
1634 * will be used by the valid program until it's unloaded
1635 * and all maps are released in free_bpf_prog_info()
1636 */
1637 atomic_inc(&map->refcnt);
1638
1639 fdput(f);
1640next_insn:
1641 insn++;
1642 i++;
1643 }
1644 }
1645
1646 /* now all pseudo BPF_LD_IMM64 instructions load valid
1647 * 'struct bpf_map *' into a register instead of user map_fd.
1648 * These pointers will be used later by verifier to validate map access.
1649 */
1650 return 0;
1651}
1652
1653/* drop refcnt of maps used by the rejected program */
1654static void release_maps(struct verifier_env *env)
1655{
1656 int i;
1657
1658 for (i = 0; i < env->used_map_cnt; i++)
1659 bpf_map_put(env->used_maps[i]);
1660}
1661
1662/* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */
1663static void convert_pseudo_ld_imm64(struct verifier_env *env)
1664{
1665 struct bpf_insn *insn = env->prog->insnsi;
1666 int insn_cnt = env->prog->len;
1667 int i;
1668
1669 for (i = 0; i < insn_cnt; i++, insn++)
1670 if (insn->code == (BPF_LD | BPF_IMM | BPF_DW))
1671 insn->src_reg = 0;
1672}
1673
1674int bpf_check(struct bpf_prog *prog, union bpf_attr *attr)
1675{
1676 char __user *log_ubuf = NULL;
1677 struct verifier_env *env;
1678 int ret = -EINVAL;
1679
1680 if (prog->len <= 0 || prog->len > BPF_MAXINSNS)
1681 return -E2BIG;
1682
1683 /* 'struct verifier_env' can be global, but since it's not small,
1684 * allocate/free it every time bpf_check() is called
1685 */
1686 env = kzalloc(sizeof(struct verifier_env), GFP_KERNEL);
1687 if (!env)
1688 return -ENOMEM;
1689
1690 env->prog = prog;
1691
1692 /* grab the mutex to protect few globals used by verifier */
1693 mutex_lock(&bpf_verifier_lock);
1694
1695 if (attr->log_level || attr->log_buf || attr->log_size) {
1696 /* user requested verbose verifier output
1697 * and supplied buffer to store the verification trace
1698 */
1699 log_level = attr->log_level;
1700 log_ubuf = (char __user *) (unsigned long) attr->log_buf;
1701 log_size = attr->log_size;
1702 log_len = 0;
1703
1704 ret = -EINVAL;
1705 /* log_* values have to be sane */
1706 if (log_size < 128 || log_size > UINT_MAX >> 8 ||
1707 log_level == 0 || log_ubuf == NULL)
1708 goto free_env;
1709
1710 ret = -ENOMEM;
1711 log_buf = vmalloc(log_size);
1712 if (!log_buf)
1713 goto free_env;
1714 } else {
1715 log_level = 0;
1716 }
1717
1718 ret = replace_map_fd_with_map_ptr(env);
1719 if (ret < 0)
1720 goto skip_full_check;
1721
1722 ret = check_cfg(env);
1723 if (ret < 0)
1724 goto skip_full_check;
1725
1726 ret = do_check(env);
1727
1728skip_full_check:
1729 while (pop_stack(env, NULL) >= 0);
1730
1731 if (log_level && log_len >= log_size - 1) {
1732 BUG_ON(log_len >= log_size);
1733 /* verifier log exceeded user supplied buffer */
1734 ret = -ENOSPC;
1735 /* fall through to return what was recorded */
1736 }
1737
1738 /* copy verifier log back to user space including trailing zero */
1739 if (log_level && copy_to_user(log_ubuf, log_buf, log_len + 1) != 0) {
1740 ret = -EFAULT;
1741 goto free_log_buf;
1742 }
1743
1744 if (ret == 0 && env->used_map_cnt) {
1745 /* if program passed verifier, update used_maps in bpf_prog_info */
1746 prog->aux->used_maps = kmalloc_array(env->used_map_cnt,
1747 sizeof(env->used_maps[0]),
1748 GFP_KERNEL);
1749
1750 if (!prog->aux->used_maps) {
1751 ret = -ENOMEM;
1752 goto free_log_buf;
1753 }
1754
1755 memcpy(prog->aux->used_maps, env->used_maps,
1756 sizeof(env->used_maps[0]) * env->used_map_cnt);
1757 prog->aux->used_map_cnt = env->used_map_cnt;
1758
1759 /* program is valid. Convert pseudo bpf_ld_imm64 into generic
1760 * bpf_ld_imm64 instructions
1761 */
1762 convert_pseudo_ld_imm64(env);
1763 }
1764
1765free_log_buf:
1766 if (log_level)
1767 vfree(log_buf);
1768free_env:
1769 if (!prog->aux->used_maps)
1770 /* if we didn't copy map pointers into bpf_prog_info, release
1771 * them now. Otherwise free_bpf_prog_info() will release them.
1772 */
1773 release_maps(env);
1774 kfree(env);
1775 mutex_unlock(&bpf_verifier_lock);
1776 return ret;
1777}
generated by cgit v1.2.2 (git 2.25.0) at 2025-09-19 19:02:14 -0400