]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - kernel/bpf/verifier.c
projects / mirror_ubuntu-jammy-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
bpf, verifier: further improve search pruning
[mirror_ubuntu-jammy-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 int imm;
146
147 /* valid when type == CONST_PTR_TO_MAP | PTR_TO_MAP_VALUE |
148 * PTR_TO_MAP_VALUE_OR_NULL
149 */
150 struct bpf_map *map_ptr;
151 };
152 };
153
154 enum bpf_stack_slot_type {
155 STACK_INVALID, /* nothing was stored in this stack slot */
156 STACK_SPILL, /* register spilled into stack */
157 STACK_MISC /* BPF program wrote some data into this slot */
158 };
159
160 #define BPF_REG_SIZE 8 /* size of eBPF register in bytes */
161
162 /* state of the program:
163 * type of all registers and stack info
164 */
165 struct verifier_state {
166 struct reg_state regs[MAX_BPF_REG];
167 u8 stack_slot_type[MAX_BPF_STACK];
168 struct reg_state spilled_regs[MAX_BPF_STACK / BPF_REG_SIZE];
169 };
170
171 /* linked list of verifier states used to prune search */
172 struct verifier_state_list {
173 struct verifier_state state;
174 struct verifier_state_list *next;
175 };
176
177 /* verifier_state + insn_idx are pushed to stack when branch is encountered */
178 struct verifier_stack_elem {
179 /* verifer state is 'st'
180 * before processing instruction 'insn_idx'
181 * and after processing instruction 'prev_insn_idx'
182 */
183 struct verifier_state st;
184 int insn_idx;
185 int prev_insn_idx;
186 struct verifier_stack_elem *next;
187 };
188
189 #define MAX_USED_MAPS 64 /* max number of maps accessed by one eBPF program */
190
191 /* single container for all structs
192 * one verifier_env per bpf_check() call
193 */
194 struct verifier_env {
195 struct bpf_prog *prog; /* eBPF program being verified */
196 struct verifier_stack_elem *head; /* stack of verifier states to be processed */
197 int stack_size; /* number of states to be processed */
198 struct verifier_state cur_state; /* current verifier state */
199 struct verifier_state_list **explored_states; /* search pruning optimization */
200 struct bpf_map *used_maps[MAX_USED_MAPS]; /* array of map's used by eBPF program */
201 u32 used_map_cnt; /* number of used maps */
202 bool allow_ptr_leaks;
203 };
204
205 #define BPF_COMPLEXITY_LIMIT_INSNS 65536
206 #define BPF_COMPLEXITY_LIMIT_STACK 1024
207
208 /* verbose verifier prints what it's seeing
209 * bpf_check() is called under lock, so no race to access these global vars
210 */
211 static u32 log_level, log_size, log_len;
212 static char *log_buf;
213
214 static DEFINE_MUTEX(bpf_verifier_lock);
215
216 /* log_level controls verbosity level of eBPF verifier.
217 * verbose() is used to dump the verification trace to the log, so the user
218 * can figure out what's wrong with the program
219 */
220 static __printf(1, 2) void verbose(const char *fmt, ...)
221 {
222 va_list args;
223
224 if (log_level == 0 || log_len >= log_size - 1)
225 return;
226
227 va_start(args, fmt);
228 log_len += vscnprintf(log_buf + log_len, log_size - log_len, fmt, args);
229 va_end(args);
230 }
231
232 /* string representation of 'enum bpf_reg_type' */
233 static const char * const reg_type_str[] = {
234 [NOT_INIT] = "?",
235 [UNKNOWN_VALUE] = "inv",
236 [PTR_TO_CTX] = "ctx",
237 [CONST_PTR_TO_MAP] = "map_ptr",
238 [PTR_TO_MAP_VALUE] = "map_value",
239 [PTR_TO_MAP_VALUE_OR_NULL] = "map_value_or_null",
240 [FRAME_PTR] = "fp",
241 [PTR_TO_STACK] = "fp",
242 [CONST_IMM] = "imm",
243 };
244
245 static const struct {
246 int map_type;
247 int func_id;
248 } func_limit[] = {
249 {BPF_MAP_TYPE_PROG_ARRAY, BPF_FUNC_tail_call},
250 {BPF_MAP_TYPE_PERF_EVENT_ARRAY, BPF_FUNC_perf_event_read},
251 {BPF_MAP_TYPE_PERF_EVENT_ARRAY, BPF_FUNC_perf_event_output},
252 {BPF_MAP_TYPE_STACK_TRACE, BPF_FUNC_get_stackid},
253 };
254
255 static void print_verifier_state(struct verifier_env *env)
256 {
257 enum bpf_reg_type t;
258 int i;
259
260 for (i = 0; i < MAX_BPF_REG; i++) {
261 t = env->cur_state.regs[i].type;
262 if (t == NOT_INIT)
263 continue;
264 verbose(" R%d=%s", i, reg_type_str[t]);
265 if (t == CONST_IMM || t == PTR_TO_STACK)
266 verbose("%d", env->cur_state.regs[i].imm);
267 else if (t == CONST_PTR_TO_MAP || t == PTR_TO_MAP_VALUE ||
268 t == PTR_TO_MAP_VALUE_OR_NULL)
269 verbose("(ks=%d,vs=%d)",
270 env->cur_state.regs[i].map_ptr->key_size,
271 env->cur_state.regs[i].map_ptr->value_size);
272 }
273 for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
274 if (env->cur_state.stack_slot_type[i] == STACK_SPILL)
275 verbose(" fp%d=%s", -MAX_BPF_STACK + i,
276 reg_type_str[env->cur_state.spilled_regs[i / BPF_REG_SIZE].type]);
277 }
278 verbose("\n");
279 }
280
281 static const char *const bpf_class_string[] = {
282 [BPF_LD] = "ld",
283 [BPF_LDX] = "ldx",
284 [BPF_ST] = "st",
285 [BPF_STX] = "stx",
286 [BPF_ALU] = "alu",
287 [BPF_JMP] = "jmp",
288 [BPF_RET] = "BUG",
289 [BPF_ALU64] = "alu64",
290 };
291
292 static const char *const bpf_alu_string[16] = {
293 [BPF_ADD >> 4] = "+=",
294 [BPF_SUB >> 4] = "-=",
295 [BPF_MUL >> 4] = "*=",
296 [BPF_DIV >> 4] = "/=",
297 [BPF_OR >> 4] = "|=",
298 [BPF_AND >> 4] = "&=",
299 [BPF_LSH >> 4] = "<<=",
300 [BPF_RSH >> 4] = ">>=",
301 [BPF_NEG >> 4] = "neg",
302 [BPF_MOD >> 4] = "%=",
303 [BPF_XOR >> 4] = "^=",
304 [BPF_MOV >> 4] = "=",
305 [BPF_ARSH >> 4] = "s>>=",
306 [BPF_END >> 4] = "endian",
307 };
308
309 static const char *const bpf_ldst_string[] = {
310 [BPF_W >> 3] = "u32",
311 [BPF_H >> 3] = "u16",
312 [BPF_B >> 3] = "u8",
313 [BPF_DW >> 3] = "u64",
314 };
315
316 static const char *const bpf_jmp_string[16] = {
317 [BPF_JA >> 4] = "jmp",
318 [BPF_JEQ >> 4] = "==",
319 [BPF_JGT >> 4] = ">",
320 [BPF_JGE >> 4] = ">=",
321 [BPF_JSET >> 4] = "&",
322 [BPF_JNE >> 4] = "!=",
323 [BPF_JSGT >> 4] = "s>",
324 [BPF_JSGE >> 4] = "s>=",
325 [BPF_CALL >> 4] = "call",
326 [BPF_EXIT >> 4] = "exit",
327 };
328
329 static void print_bpf_insn(struct bpf_insn *insn)
330 {
331 u8 class = BPF_CLASS(insn->code);
332
333 if (class == BPF_ALU || class == BPF_ALU64) {
334 if (BPF_SRC(insn->code) == BPF_X)
335 verbose("(%02x) %sr%d %s %sr%d\n",
336 insn->code, class == BPF_ALU ? "(u32) " : "",
337 insn->dst_reg,
338 bpf_alu_string[BPF_OP(insn->code) >> 4],
339 class == BPF_ALU ? "(u32) " : "",
340 insn->src_reg);
341 else
342 verbose("(%02x) %sr%d %s %s%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->imm);
348 } else if (class == BPF_STX) {
349 if (BPF_MODE(insn->code) == BPF_MEM)
350 verbose("(%02x) *(%s *)(r%d %+d) = r%d\n",
351 insn->code,
352 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
353 insn->dst_reg,
354 insn->off, insn->src_reg);
355 else if (BPF_MODE(insn->code) == BPF_XADD)
356 verbose("(%02x) lock *(%s *)(r%d %+d) += r%d\n",
357 insn->code,
358 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
359 insn->dst_reg, insn->off,
360 insn->src_reg);
361 else
362 verbose("BUG_%02x\n", insn->code);
363 } else if (class == BPF_ST) {
364 if (BPF_MODE(insn->code) != BPF_MEM) {
365 verbose("BUG_st_%02x\n", insn->code);
366 return;
367 }
368 verbose("(%02x) *(%s *)(r%d %+d) = %d\n",
369 insn->code,
370 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
371 insn->dst_reg,
372 insn->off, insn->imm);
373 } else if (class == BPF_LDX) {
374 if (BPF_MODE(insn->code) != BPF_MEM) {
375 verbose("BUG_ldx_%02x\n", insn->code);
376 return;
377 }
378 verbose("(%02x) r%d = *(%s *)(r%d %+d)\n",
379 insn->code, insn->dst_reg,
380 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
381 insn->src_reg, insn->off);
382 } else if (class == BPF_LD) {
383 if (BPF_MODE(insn->code) == BPF_ABS) {
384 verbose("(%02x) r0 = *(%s *)skb[%d]\n",
385 insn->code,
386 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
387 insn->imm);
388 } else if (BPF_MODE(insn->code) == BPF_IND) {
389 verbose("(%02x) r0 = *(%s *)skb[r%d + %d]\n",
390 insn->code,
391 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
392 insn->src_reg, insn->imm);
393 } else if (BPF_MODE(insn->code) == BPF_IMM) {
394 verbose("(%02x) r%d = 0x%x\n",
395 insn->code, insn->dst_reg, insn->imm);
396 } else {
397 verbose("BUG_ld_%02x\n", insn->code);
398 return;
399 }
400 } else if (class == BPF_JMP) {
401 u8 opcode = BPF_OP(insn->code);
402
403 if (opcode == BPF_CALL) {
404 verbose("(%02x) call %d\n", insn->code, insn->imm);
405 } else if (insn->code == (BPF_JMP | BPF_JA)) {
406 verbose("(%02x) goto pc%+d\n",
407 insn->code, insn->off);
408 } else if (insn->code == (BPF_JMP | BPF_EXIT)) {
409 verbose("(%02x) exit\n", insn->code);
410 } else if (BPF_SRC(insn->code) == BPF_X) {
411 verbose("(%02x) if r%d %s r%d goto pc%+d\n",
412 insn->code, insn->dst_reg,
413 bpf_jmp_string[BPF_OP(insn->code) >> 4],
414 insn->src_reg, insn->off);
415 } else {
416 verbose("(%02x) if r%d %s 0x%x goto pc%+d\n",
417 insn->code, insn->dst_reg,
418 bpf_jmp_string[BPF_OP(insn->code) >> 4],
419 insn->imm, insn->off);
420 }
421 } else {
422 verbose("(%02x) %s\n", insn->code, bpf_class_string[class]);
423 }
424 }
425
426 static int pop_stack(struct verifier_env *env, int *prev_insn_idx)
427 {
428 struct verifier_stack_elem *elem;
429 int insn_idx;
430
431 if (env->head == NULL)
432 return -1;
433
434 memcpy(&env->cur_state, &env->head->st, sizeof(env->cur_state));
435 insn_idx = env->head->insn_idx;
436 if (prev_insn_idx)
437 *prev_insn_idx = env->head->prev_insn_idx;
438 elem = env->head->next;
439 kfree(env->head);
440 env->head = elem;
441 env->stack_size--;
442 return insn_idx;
443 }
444
445 static struct verifier_state *push_stack(struct verifier_env *env, int insn_idx,
446 int prev_insn_idx)
447 {
448 struct verifier_stack_elem *elem;
449
450 elem = kmalloc(sizeof(struct verifier_stack_elem), GFP_KERNEL);
451 if (!elem)
452 goto err;
453
454 memcpy(&elem->st, &env->cur_state, sizeof(env->cur_state));
455 elem->insn_idx = insn_idx;
456 elem->prev_insn_idx = prev_insn_idx;
457 elem->next = env->head;
458 env->head = elem;
459 env->stack_size++;
460 if (env->stack_size > BPF_COMPLEXITY_LIMIT_STACK) {
461 verbose("BPF program is too complex\n");
462 goto err;
463 }
464 return &elem->st;
465 err:
466 /* pop all elements and return */
467 while (pop_stack(env, NULL) >= 0);
468 return NULL;
469 }
470
471 #define CALLER_SAVED_REGS 6
472 static const int caller_saved[CALLER_SAVED_REGS] = {
473 BPF_REG_0, BPF_REG_1, BPF_REG_2, BPF_REG_3, BPF_REG_4, BPF_REG_5
474 };
475
476 static void init_reg_state(struct reg_state *regs)
477 {
478 int i;
479
480 for (i = 0; i < MAX_BPF_REG; i++) {
481 regs[i].type = NOT_INIT;
482 regs[i].imm = 0;
483 regs[i].map_ptr = NULL;
484 }
485
486 /* frame pointer */
487 regs[BPF_REG_FP].type = FRAME_PTR;
488
489 /* 1st arg to a function */
490 regs[BPF_REG_1].type = PTR_TO_CTX;
491 }
492
493 static void mark_reg_unknown_value(struct reg_state *regs, u32 regno)
494 {
495 BUG_ON(regno >= MAX_BPF_REG);
496 regs[regno].type = UNKNOWN_VALUE;
497 regs[regno].imm = 0;
498 regs[regno].map_ptr = NULL;
499 }
500
501 enum reg_arg_type {
502 SRC_OP, /* register is used as source operand */
503 DST_OP, /* register is used as destination operand */
504 DST_OP_NO_MARK /* same as above, check only, don't mark */
505 };
506
507 static int check_reg_arg(struct reg_state *regs, u32 regno,
508 enum reg_arg_type t)
509 {
510 if (regno >= MAX_BPF_REG) {
511 verbose("R%d is invalid\n", regno);
512 return -EINVAL;
513 }
514
515 if (t == SRC_OP) {
516 /* check whether register used as source operand can be read */
517 if (regs[regno].type == NOT_INIT) {
518 verbose("R%d !read_ok\n", regno);
519 return -EACCES;
520 }
521 } else {
522 /* check whether register used as dest operand can be written to */
523 if (regno == BPF_REG_FP) {
524 verbose("frame pointer is read only\n");
525 return -EACCES;
526 }
527 if (t == DST_OP)
528 mark_reg_unknown_value(regs, regno);
529 }
530 return 0;
531 }
532
533 static int bpf_size_to_bytes(int bpf_size)
534 {
535 if (bpf_size == BPF_W)
536 return 4;
537 else if (bpf_size == BPF_H)
538 return 2;
539 else if (bpf_size == BPF_B)
540 return 1;
541 else if (bpf_size == BPF_DW)
542 return 8;
543 else
544 return -EINVAL;
545 }
546
547 static bool is_spillable_regtype(enum bpf_reg_type type)
548 {
549 switch (type) {
550 case PTR_TO_MAP_VALUE:
551 case PTR_TO_MAP_VALUE_OR_NULL:
552 case PTR_TO_STACK:
553 case PTR_TO_CTX:
554 case FRAME_PTR:
555 case CONST_PTR_TO_MAP:
556 return true;
557 default:
558 return false;
559 }
560 }
561
562 /* check_stack_read/write functions track spill/fill of registers,
563 * stack boundary and alignment are checked in check_mem_access()
564 */
565 static int check_stack_write(struct verifier_state *state, int off, int size,
566 int value_regno)
567 {
568 int i;
569 /* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0,
570 * so it's aligned access and [off, off + size) are within stack limits
571 */
572
573 if (value_regno >= 0 &&
574 is_spillable_regtype(state->regs[value_regno].type)) {
575
576 /* register containing pointer is being spilled into stack */
577 if (size != BPF_REG_SIZE) {
578 verbose("invalid size of register spill\n");
579 return -EACCES;
580 }
581
582 /* save register state */
583 state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] =
584 state->regs[value_regno];
585
586 for (i = 0; i < BPF_REG_SIZE; i++)
587 state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_SPILL;
588 } else {
589 /* regular write of data into stack */
590 state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] =
591 (struct reg_state) {};
592
593 for (i = 0; i < size; i++)
594 state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_MISC;
595 }
596 return 0;
597 }
598
599 static int check_stack_read(struct verifier_state *state, int off, int size,
600 int value_regno)
601 {
602 u8 *slot_type;
603 int i;
604
605 slot_type = &state->stack_slot_type[MAX_BPF_STACK + off];
606
607 if (slot_type[0] == STACK_SPILL) {
608 if (size != BPF_REG_SIZE) {
609 verbose("invalid size of register spill\n");
610 return -EACCES;
611 }
612 for (i = 1; i < BPF_REG_SIZE; i++) {
613 if (slot_type[i] != STACK_SPILL) {
614 verbose("corrupted spill memory\n");
615 return -EACCES;
616 }
617 }
618
619 if (value_regno >= 0)
620 /* restore register state from stack */
621 state->regs[value_regno] =
622 state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE];
623 return 0;
624 } else {
625 for (i = 0; i < size; i++) {
626 if (slot_type[i] != STACK_MISC) {
627 verbose("invalid read from stack off %d+%d size %d\n",
628 off, i, size);
629 return -EACCES;
630 }
631 }
632 if (value_regno >= 0)
633 /* have read misc data from the stack */
634 mark_reg_unknown_value(state->regs, value_regno);
635 return 0;
636 }
637 }
638
639 /* check read/write into map element returned by bpf_map_lookup_elem() */
640 static int check_map_access(struct verifier_env *env, u32 regno, int off,
641 int size)
642 {
643 struct bpf_map *map = env->cur_state.regs[regno].map_ptr;
644
645 if (off < 0 || off + size > map->value_size) {
646 verbose("invalid access to map value, value_size=%d off=%d size=%d\n",
647 map->value_size, off, size);
648 return -EACCES;
649 }
650 return 0;
651 }
652
653 /* check access to 'struct bpf_context' fields */
654 static int check_ctx_access(struct verifier_env *env, int off, int size,
655 enum bpf_access_type t)
656 {
657 if (env->prog->aux->ops->is_valid_access &&
658 env->prog->aux->ops->is_valid_access(off, size, t)) {
659 /* remember the offset of last byte accessed in ctx */
660 if (env->prog->aux->max_ctx_offset < off + size)
661 env->prog->aux->max_ctx_offset = off + size;
662 return 0;
663 }
664
665 verbose("invalid bpf_context access off=%d size=%d\n", off, size);
666 return -EACCES;
667 }
668
669 static bool is_pointer_value(struct verifier_env *env, int regno)
670 {
671 if (env->allow_ptr_leaks)
672 return false;
673
674 switch (env->cur_state.regs[regno].type) {
675 case UNKNOWN_VALUE:
676 case CONST_IMM:
677 return false;
678 default:
679 return true;
680 }
681 }
682
683 /* check whether memory at (regno + off) is accessible for t = (read | write)
684 * if t==write, value_regno is a register which value is stored into memory
685 * if t==read, value_regno is a register which will receive the value from memory
686 * if t==write && value_regno==-1, some unknown value is stored into memory
687 * if t==read && value_regno==-1, don't care what we read from memory
688 */
689 static int check_mem_access(struct verifier_env *env, u32 regno, int off,
690 int bpf_size, enum bpf_access_type t,
691 int value_regno)
692 {
693 struct verifier_state *state = &env->cur_state;
694 int size, err = 0;
695
696 if (state->regs[regno].type == PTR_TO_STACK)
697 off += state->regs[regno].imm;
698
699 size = bpf_size_to_bytes(bpf_size);
700 if (size < 0)
701 return size;
702
703 if (off % size != 0) {
704 verbose("misaligned access off %d size %d\n", off, size);
705 return -EACCES;
706 }
707
708 if (state->regs[regno].type == PTR_TO_MAP_VALUE) {
709 if (t == BPF_WRITE && value_regno >= 0 &&
710 is_pointer_value(env, value_regno)) {
711 verbose("R%d leaks addr into map\n", value_regno);
712 return -EACCES;
713 }
714 err = check_map_access(env, regno, off, size);
715 if (!err && t == BPF_READ && value_regno >= 0)
716 mark_reg_unknown_value(state->regs, value_regno);
717
718 } else if (state->regs[regno].type == PTR_TO_CTX) {
719 if (t == BPF_WRITE && value_regno >= 0 &&
720 is_pointer_value(env, value_regno)) {
721 verbose("R%d leaks addr into ctx\n", value_regno);
722 return -EACCES;
723 }
724 err = check_ctx_access(env, off, size, t);
725 if (!err && t == BPF_READ && value_regno >= 0)
726 mark_reg_unknown_value(state->regs, value_regno);
727
728 } else if (state->regs[regno].type == FRAME_PTR ||
729 state->regs[regno].type == PTR_TO_STACK) {
730 if (off >= 0 || off < -MAX_BPF_STACK) {
731 verbose("invalid stack off=%d size=%d\n", off, size);
732 return -EACCES;
733 }
734 if (t == BPF_WRITE) {
735 if (!env->allow_ptr_leaks &&
736 state->stack_slot_type[MAX_BPF_STACK + off] == STACK_SPILL &&
737 size != BPF_REG_SIZE) {
738 verbose("attempt to corrupt spilled pointer on stack\n");
739 return -EACCES;
740 }
741 err = check_stack_write(state, off, size, value_regno);
742 } else {
743 err = check_stack_read(state, off, size, value_regno);
744 }
745 } else {
746 verbose("R%d invalid mem access '%s'\n",
747 regno, reg_type_str[state->regs[regno].type]);
748 return -EACCES;
749 }
750 return err;
751 }
752
753 static int check_xadd(struct verifier_env *env, struct bpf_insn *insn)
754 {
755 struct reg_state *regs = env->cur_state.regs;
756 int err;
757
758 if ((BPF_SIZE(insn->code) != BPF_W && BPF_SIZE(insn->code) != BPF_DW) ||
759 insn->imm != 0) {
760 verbose("BPF_XADD uses reserved fields\n");
761 return -EINVAL;
762 }
763
764 /* check src1 operand */
765 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
766 if (err)
767 return err;
768
769 /* check src2 operand */
770 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
771 if (err)
772 return err;
773
774 /* check whether atomic_add can read the memory */
775 err = check_mem_access(env, insn->dst_reg, insn->off,
776 BPF_SIZE(insn->code), BPF_READ, -1);
777 if (err)
778 return err;
779
780 /* check whether atomic_add can write into the same memory */
781 return check_mem_access(env, insn->dst_reg, insn->off,
782 BPF_SIZE(insn->code), BPF_WRITE, -1);
783 }
784
785 /* when register 'regno' is passed into function that will read 'access_size'
786 * bytes from that pointer, make sure that it's within stack boundary
787 * and all elements of stack are initialized
788 */
789 static int check_stack_boundary(struct verifier_env *env, int regno,
790 int access_size, bool zero_size_allowed)
791 {
792 struct verifier_state *state = &env->cur_state;
793 struct reg_state *regs = state->regs;
794 int off, i;
795
796 if (regs[regno].type != PTR_TO_STACK) {
797 if (zero_size_allowed && access_size == 0 &&
798 regs[regno].type == CONST_IMM &&
799 regs[regno].imm == 0)
800 return 0;
801
802 verbose("R%d type=%s expected=%s\n", regno,
803 reg_type_str[regs[regno].type],
804 reg_type_str[PTR_TO_STACK]);
805 return -EACCES;
806 }
807
808 off = regs[regno].imm;
809 if (off >= 0 || off < -MAX_BPF_STACK || off + access_size > 0 ||
810 access_size <= 0) {
811 verbose("invalid stack type R%d off=%d access_size=%d\n",
812 regno, off, access_size);
813 return -EACCES;
814 }
815
816 for (i = 0; i < access_size; i++) {
817 if (state->stack_slot_type[MAX_BPF_STACK + off + i] != STACK_MISC) {
818 verbose("invalid indirect read from stack off %d+%d size %d\n",
819 off, i, access_size);
820 return -EACCES;
821 }
822 }
823 return 0;
824 }
825
826 static int check_func_arg(struct verifier_env *env, u32 regno,
827 enum bpf_arg_type arg_type, struct bpf_map **mapp)
828 {
829 struct reg_state *reg = env->cur_state.regs + regno;
830 enum bpf_reg_type expected_type;
831 int err = 0;
832
833 if (arg_type == ARG_DONTCARE)
834 return 0;
835
836 if (reg->type == NOT_INIT) {
837 verbose("R%d !read_ok\n", regno);
838 return -EACCES;
839 }
840
841 if (arg_type == ARG_ANYTHING) {
842 if (is_pointer_value(env, regno)) {
843 verbose("R%d leaks addr into helper function\n", regno);
844 return -EACCES;
845 }
846 return 0;
847 }
848
849 if (arg_type == ARG_PTR_TO_MAP_KEY ||
850 arg_type == ARG_PTR_TO_MAP_VALUE) {
851 expected_type = PTR_TO_STACK;
852 } else if (arg_type == ARG_CONST_STACK_SIZE ||
853 arg_type == ARG_CONST_STACK_SIZE_OR_ZERO) {
854 expected_type = CONST_IMM;
855 } else if (arg_type == ARG_CONST_MAP_PTR) {
856 expected_type = CONST_PTR_TO_MAP;
857 } else if (arg_type == ARG_PTR_TO_CTX) {
858 expected_type = PTR_TO_CTX;
859 } else if (arg_type == ARG_PTR_TO_STACK) {
860 expected_type = PTR_TO_STACK;
861 /* One exception here. In case function allows for NULL to be
862 * passed in as argument, it's a CONST_IMM type. Final test
863 * happens during stack boundary checking.
864 */
865 if (reg->type == CONST_IMM && reg->imm == 0)
866 expected_type = CONST_IMM;
867 } else {
868 verbose("unsupported arg_type %d\n", arg_type);
869 return -EFAULT;
870 }
871
872 if (reg->type != expected_type) {
873 verbose("R%d type=%s expected=%s\n", regno,
874 reg_type_str[reg->type], reg_type_str[expected_type]);
875 return -EACCES;
876 }
877
878 if (arg_type == ARG_CONST_MAP_PTR) {
879 /* bpf_map_xxx(map_ptr) call: remember that map_ptr */
880 *mapp = reg->map_ptr;
881
882 } else if (arg_type == ARG_PTR_TO_MAP_KEY) {
883 /* bpf_map_xxx(..., map_ptr, ..., key) call:
884 * check that [key, key + map->key_size) are within
885 * stack limits and initialized
886 */
887 if (!*mapp) {
888 /* in function declaration map_ptr must come before
889 * map_key, so that it's verified and known before
890 * we have to check map_key here. Otherwise it means
891 * that kernel subsystem misconfigured verifier
892 */
893 verbose("invalid map_ptr to access map->key\n");
894 return -EACCES;
895 }
896 err = check_stack_boundary(env, regno, (*mapp)->key_size,
897 false);
898 } else if (arg_type == ARG_PTR_TO_MAP_VALUE) {
899 /* bpf_map_xxx(..., map_ptr, ..., value) call:
900 * check [value, value + map->value_size) validity
901 */
902 if (!*mapp) {
903 /* kernel subsystem misconfigured verifier */
904 verbose("invalid map_ptr to access map->value\n");
905 return -EACCES;
906 }
907 err = check_stack_boundary(env, regno, (*mapp)->value_size,
908 false);
909 } else if (arg_type == ARG_CONST_STACK_SIZE ||
910 arg_type == ARG_CONST_STACK_SIZE_OR_ZERO) {
911 bool zero_size_allowed = (arg_type == ARG_CONST_STACK_SIZE_OR_ZERO);
912
913 /* bpf_xxx(..., buf, len) call will access 'len' bytes
914 * from stack pointer 'buf'. Check it
915 * note: regno == len, regno - 1 == buf
916 */
917 if (regno == 0) {
918 /* kernel subsystem misconfigured verifier */
919 verbose("ARG_CONST_STACK_SIZE cannot be first argument\n");
920 return -EACCES;
921 }
922 err = check_stack_boundary(env, regno - 1, reg->imm,
923 zero_size_allowed);
924 }
925
926 return err;
927 }
928
929 static int check_map_func_compatibility(struct bpf_map *map, int func_id)
930 {
931 bool bool_map, bool_func;
932 int i;
933
934 if (!map)
935 return 0;
936
937 for (i = 0; i < ARRAY_SIZE(func_limit); i++) {
938 bool_map = (map->map_type == func_limit[i].map_type);
939 bool_func = (func_id == func_limit[i].func_id);
940 /* only when map & func pair match it can continue.
941 * don't allow any other map type to be passed into
942 * the special func;
943 */
944 if (bool_func && bool_map != bool_func) {
945 verbose("cannot pass map_type %d into func %d\n",
946 map->map_type, func_id);
947 return -EINVAL;
948 }
949 }
950
951 return 0;
952 }
953
954 static int check_call(struct verifier_env *env, int func_id)
955 {
956 struct verifier_state *state = &env->cur_state;
957 const struct bpf_func_proto *fn = NULL;
958 struct reg_state *regs = state->regs;
959 struct bpf_map *map = NULL;
960 struct reg_state *reg;
961 int i, err;
962
963 /* find function prototype */
964 if (func_id < 0 || func_id >= __BPF_FUNC_MAX_ID) {
965 verbose("invalid func %d\n", func_id);
966 return -EINVAL;
967 }
968
969 if (env->prog->aux->ops->get_func_proto)
970 fn = env->prog->aux->ops->get_func_proto(func_id);
971
972 if (!fn) {
973 verbose("unknown func %d\n", func_id);
974 return -EINVAL;
975 }
976
977 /* eBPF programs must be GPL compatible to use GPL-ed functions */
978 if (!env->prog->gpl_compatible && fn->gpl_only) {
979 verbose("cannot call GPL only function from proprietary program\n");
980 return -EINVAL;
981 }
982
983 /* check args */
984 err = check_func_arg(env, BPF_REG_1, fn->arg1_type, &map);
985 if (err)
986 return err;
987 err = check_func_arg(env, BPF_REG_2, fn->arg2_type, &map);
988 if (err)
989 return err;
990 err = check_func_arg(env, BPF_REG_3, fn->arg3_type, &map);
991 if (err)
992 return err;
993 err = check_func_arg(env, BPF_REG_4, fn->arg4_type, &map);
994 if (err)
995 return err;
996 err = check_func_arg(env, BPF_REG_5, fn->arg5_type, &map);
997 if (err)
998 return err;
999
1000 /* reset caller saved regs */
1001 for (i = 0; i < CALLER_SAVED_REGS; i++) {
1002 reg = regs + caller_saved[i];
1003 reg->type = NOT_INIT;
1004 reg->imm = 0;
1005 }
1006
1007 /* update return register */
1008 if (fn->ret_type == RET_INTEGER) {
1009 regs[BPF_REG_0].type = UNKNOWN_VALUE;
1010 } else if (fn->ret_type == RET_VOID) {
1011 regs[BPF_REG_0].type = NOT_INIT;
1012 } else if (fn->ret_type == RET_PTR_TO_MAP_VALUE_OR_NULL) {
1013 regs[BPF_REG_0].type = PTR_TO_MAP_VALUE_OR_NULL;
1014 /* remember map_ptr, so that check_map_access()
1015 * can check 'value_size' boundary of memory access
1016 * to map element returned from bpf_map_lookup_elem()
1017 */
1018 if (map == NULL) {
1019 verbose("kernel subsystem misconfigured verifier\n");
1020 return -EINVAL;
1021 }
1022 regs[BPF_REG_0].map_ptr = map;
1023 } else {
1024 verbose("unknown return type %d of func %d\n",
1025 fn->ret_type, func_id);
1026 return -EINVAL;
1027 }
1028
1029 err = check_map_func_compatibility(map, func_id);
1030 if (err)
1031 return err;
1032
1033 return 0;
1034 }
1035
1036 /* check validity of 32-bit and 64-bit arithmetic operations */
1037 static int check_alu_op(struct verifier_env *env, struct bpf_insn *insn)
1038 {
1039 struct reg_state *regs = env->cur_state.regs;
1040 u8 opcode = BPF_OP(insn->code);
1041 int err;
1042
1043 if (opcode == BPF_END || opcode == BPF_NEG) {
1044 if (opcode == BPF_NEG) {
1045 if (BPF_SRC(insn->code) != 0 ||
1046 insn->src_reg != BPF_REG_0 ||
1047 insn->off != 0 || insn->imm != 0) {
1048 verbose("BPF_NEG uses reserved fields\n");
1049 return -EINVAL;
1050 }
1051 } else {
1052 if (insn->src_reg != BPF_REG_0 || insn->off != 0 ||
1053 (insn->imm != 16 && insn->imm != 32 && insn->imm != 64)) {
1054 verbose("BPF_END uses reserved fields\n");
1055 return -EINVAL;
1056 }
1057 }
1058
1059 /* check src operand */
1060 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1061 if (err)
1062 return err;
1063
1064 if (is_pointer_value(env, insn->dst_reg)) {
1065 verbose("R%d pointer arithmetic prohibited\n",
1066 insn->dst_reg);
1067 return -EACCES;
1068 }
1069
1070 /* check dest operand */
1071 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1072 if (err)
1073 return err;
1074
1075 } else if (opcode == BPF_MOV) {
1076
1077 if (BPF_SRC(insn->code) == BPF_X) {
1078 if (insn->imm != 0 || insn->off != 0) {
1079 verbose("BPF_MOV uses reserved fields\n");
1080 return -EINVAL;
1081 }
1082
1083 /* check src operand */
1084 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1085 if (err)
1086 return err;
1087 } else {
1088 if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
1089 verbose("BPF_MOV uses reserved fields\n");
1090 return -EINVAL;
1091 }
1092 }
1093
1094 /* check dest operand */
1095 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1096 if (err)
1097 return err;
1098
1099 if (BPF_SRC(insn->code) == BPF_X) {
1100 if (BPF_CLASS(insn->code) == BPF_ALU64) {
1101 /* case: R1 = R2
1102 * copy register state to dest reg
1103 */
1104 regs[insn->dst_reg] = regs[insn->src_reg];
1105 } else {
1106 if (is_pointer_value(env, insn->src_reg)) {
1107 verbose("R%d partial copy of pointer\n",
1108 insn->src_reg);
1109 return -EACCES;
1110 }
1111 regs[insn->dst_reg].type = UNKNOWN_VALUE;
1112 regs[insn->dst_reg].map_ptr = NULL;
1113 }
1114 } else {
1115 /* case: R = imm
1116 * remember the value we stored into this reg
1117 */
1118 regs[insn->dst_reg].type = CONST_IMM;
1119 regs[insn->dst_reg].imm = insn->imm;
1120 }
1121
1122 } else if (opcode > BPF_END) {
1123 verbose("invalid BPF_ALU opcode %x\n", opcode);
1124 return -EINVAL;
1125
1126 } else { /* all other ALU ops: and, sub, xor, add, ... */
1127
1128 bool stack_relative = false;
1129
1130 if (BPF_SRC(insn->code) == BPF_X) {
1131 if (insn->imm != 0 || insn->off != 0) {
1132 verbose("BPF_ALU uses reserved fields\n");
1133 return -EINVAL;
1134 }
1135 /* check src1 operand */
1136 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1137 if (err)
1138 return err;
1139 } else {
1140 if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
1141 verbose("BPF_ALU uses reserved fields\n");
1142 return -EINVAL;
1143 }
1144 }
1145
1146 /* check src2 operand */
1147 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1148 if (err)
1149 return err;
1150
1151 if ((opcode == BPF_MOD || opcode == BPF_DIV) &&
1152 BPF_SRC(insn->code) == BPF_K && insn->imm == 0) {
1153 verbose("div by zero\n");
1154 return -EINVAL;
1155 }
1156
1157 if ((opcode == BPF_LSH || opcode == BPF_RSH ||
1158 opcode == BPF_ARSH) && BPF_SRC(insn->code) == BPF_K) {
1159 int size = BPF_CLASS(insn->code) == BPF_ALU64 ? 64 : 32;
1160
1161 if (insn->imm < 0 || insn->imm >= size) {
1162 verbose("invalid shift %d\n", insn->imm);
1163 return -EINVAL;
1164 }
1165 }
1166
1167 /* pattern match 'bpf_add Rx, imm' instruction */
1168 if (opcode == BPF_ADD && BPF_CLASS(insn->code) == BPF_ALU64 &&
1169 regs[insn->dst_reg].type == FRAME_PTR &&
1170 BPF_SRC(insn->code) == BPF_K) {
1171 stack_relative = true;
1172 } else if (is_pointer_value(env, insn->dst_reg)) {
1173 verbose("R%d pointer arithmetic prohibited\n",
1174 insn->dst_reg);
1175 return -EACCES;
1176 } else if (BPF_SRC(insn->code) == BPF_X &&
1177 is_pointer_value(env, insn->src_reg)) {
1178 verbose("R%d pointer arithmetic prohibited\n",
1179 insn->src_reg);
1180 return -EACCES;
1181 }
1182
1183 /* check dest operand */
1184 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1185 if (err)
1186 return err;
1187
1188 if (stack_relative) {
1189 regs[insn->dst_reg].type = PTR_TO_STACK;
1190 regs[insn->dst_reg].imm = insn->imm;
1191 }
1192 }
1193
1194 return 0;
1195 }
1196
1197 static int check_cond_jmp_op(struct verifier_env *env,
1198 struct bpf_insn *insn, int *insn_idx)
1199 {
1200 struct reg_state *regs = env->cur_state.regs;
1201 struct verifier_state *other_branch;
1202 u8 opcode = BPF_OP(insn->code);
1203 int err;
1204
1205 if (opcode > BPF_EXIT) {
1206 verbose("invalid BPF_JMP opcode %x\n", opcode);
1207 return -EINVAL;
1208 }
1209
1210 if (BPF_SRC(insn->code) == BPF_X) {
1211 if (insn->imm != 0) {
1212 verbose("BPF_JMP uses reserved fields\n");
1213 return -EINVAL;
1214 }
1215
1216 /* check src1 operand */
1217 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1218 if (err)
1219 return err;
1220
1221 if (is_pointer_value(env, insn->src_reg)) {
1222 verbose("R%d pointer comparison prohibited\n",
1223 insn->src_reg);
1224 return -EACCES;
1225 }
1226 } else {
1227 if (insn->src_reg != BPF_REG_0) {
1228 verbose("BPF_JMP uses reserved fields\n");
1229 return -EINVAL;
1230 }
1231 }
1232
1233 /* check src2 operand */
1234 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1235 if (err)
1236 return err;
1237
1238 /* detect if R == 0 where R was initialized to zero earlier */
1239 if (BPF_SRC(insn->code) == BPF_K &&
1240 (opcode == BPF_JEQ || opcode == BPF_JNE) &&
1241 regs[insn->dst_reg].type == CONST_IMM &&
1242 regs[insn->dst_reg].imm == insn->imm) {
1243 if (opcode == BPF_JEQ) {
1244 /* if (imm == imm) goto pc+off;
1245 * only follow the goto, ignore fall-through
1246 */
1247 *insn_idx += insn->off;
1248 return 0;
1249 } else {
1250 /* if (imm != imm) goto pc+off;
1251 * only follow fall-through branch, since
1252 * that's where the program will go
1253 */
1254 return 0;
1255 }
1256 }
1257
1258 other_branch = push_stack(env, *insn_idx + insn->off + 1, *insn_idx);
1259 if (!other_branch)
1260 return -EFAULT;
1261
1262 /* detect if R == 0 where R is returned value from bpf_map_lookup_elem() */
1263 if (BPF_SRC(insn->code) == BPF_K &&
1264 insn->imm == 0 && (opcode == BPF_JEQ ||
1265 opcode == BPF_JNE) &&
1266 regs[insn->dst_reg].type == PTR_TO_MAP_VALUE_OR_NULL) {
1267 if (opcode == BPF_JEQ) {
1268 /* next fallthrough insn can access memory via
1269 * this register
1270 */
1271 regs[insn->dst_reg].type = PTR_TO_MAP_VALUE;
1272 /* branch targer cannot access it, since reg == 0 */
1273 other_branch->regs[insn->dst_reg].type = CONST_IMM;
1274 other_branch->regs[insn->dst_reg].imm = 0;
1275 } else {
1276 other_branch->regs[insn->dst_reg].type = PTR_TO_MAP_VALUE;
1277 regs[insn->dst_reg].type = CONST_IMM;
1278 regs[insn->dst_reg].imm = 0;
1279 }
1280 } else if (is_pointer_value(env, insn->dst_reg)) {
1281 verbose("R%d pointer comparison prohibited\n", insn->dst_reg);
1282 return -EACCES;
1283 } else if (BPF_SRC(insn->code) == BPF_K &&
1284 (opcode == BPF_JEQ || opcode == BPF_JNE)) {
1285
1286 if (opcode == BPF_JEQ) {
1287 /* detect if (R == imm) goto
1288 * and in the target state recognize that R = imm
1289 */
1290 other_branch->regs[insn->dst_reg].type = CONST_IMM;
1291 other_branch->regs[insn->dst_reg].imm = insn->imm;
1292 } else {
1293 /* detect if (R != imm) goto
1294 * and in the fall-through state recognize that R = imm
1295 */
1296 regs[insn->dst_reg].type = CONST_IMM;
1297 regs[insn->dst_reg].imm = insn->imm;
1298 }
1299 }
1300 if (log_level)
1301 print_verifier_state(env);
1302 return 0;
1303 }
1304
1305 /* return the map pointer stored inside BPF_LD_IMM64 instruction */
1306 static struct bpf_map *ld_imm64_to_map_ptr(struct bpf_insn *insn)
1307 {
1308 u64 imm64 = ((u64) (u32) insn[0].imm) | ((u64) (u32) insn[1].imm) << 32;
1309
1310 return (struct bpf_map *) (unsigned long) imm64;
1311 }
1312
1313 /* verify BPF_LD_IMM64 instruction */
1314 static int check_ld_imm(struct verifier_env *env, struct bpf_insn *insn)
1315 {
1316 struct reg_state *regs = env->cur_state.regs;
1317 int err;
1318
1319 if (BPF_SIZE(insn->code) != BPF_DW) {
1320 verbose("invalid BPF_LD_IMM insn\n");
1321 return -EINVAL;
1322 }
1323 if (insn->off != 0) {
1324 verbose("BPF_LD_IMM64 uses reserved fields\n");
1325 return -EINVAL;
1326 }
1327
1328 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1329 if (err)
1330 return err;
1331
1332 if (insn->src_reg == 0)
1333 /* generic move 64-bit immediate into a register */
1334 return 0;
1335
1336 /* replace_map_fd_with_map_ptr() should have caught bad ld_imm64 */
1337 BUG_ON(insn->src_reg != BPF_PSEUDO_MAP_FD);
1338
1339 regs[insn->dst_reg].type = CONST_PTR_TO_MAP;
1340 regs[insn->dst_reg].map_ptr = ld_imm64_to_map_ptr(insn);
1341 return 0;
1342 }
1343
1344 static bool may_access_skb(enum bpf_prog_type type)
1345 {
1346 switch (type) {
1347 case BPF_PROG_TYPE_SOCKET_FILTER:
1348 case BPF_PROG_TYPE_SCHED_CLS:
1349 case BPF_PROG_TYPE_SCHED_ACT:
1350 return true;
1351 default:
1352 return false;
1353 }
1354 }
1355
1356 /* verify safety of LD_ABS|LD_IND instructions:
1357 * - they can only appear in the programs where ctx == skb
1358 * - since they are wrappers of function calls, they scratch R1-R5 registers,
1359 * preserve R6-R9, and store return value into R0
1360 *
1361 * Implicit input:
1362 * ctx == skb == R6 == CTX
1363 *
1364 * Explicit input:
1365 * SRC == any register
1366 * IMM == 32-bit immediate
1367 *
1368 * Output:
1369 * R0 - 8/16/32-bit skb data converted to cpu endianness
1370 */
1371 static int check_ld_abs(struct verifier_env *env, struct bpf_insn *insn)
1372 {
1373 struct reg_state *regs = env->cur_state.regs;
1374 u8 mode = BPF_MODE(insn->code);
1375 struct reg_state *reg;
1376 int i, err;
1377
1378 if (!may_access_skb(env->prog->type)) {
1379 verbose("BPF_LD_ABS|IND instructions not allowed for this program type\n");
1380 return -EINVAL;
1381 }
1382
1383 if (insn->dst_reg != BPF_REG_0 || insn->off != 0 ||
1384 (mode == BPF_ABS && insn->src_reg != BPF_REG_0)) {
1385 verbose("BPF_LD_ABS uses reserved fields\n");
1386 return -EINVAL;
1387 }
1388
1389 /* check whether implicit source operand (register R6) is readable */
1390 err = check_reg_arg(regs, BPF_REG_6, SRC_OP);
1391 if (err)
1392 return err;
1393
1394 if (regs[BPF_REG_6].type != PTR_TO_CTX) {
1395 verbose("at the time of BPF_LD_ABS|IND R6 != pointer to skb\n");
1396 return -EINVAL;
1397 }
1398
1399 if (mode == BPF_IND) {
1400 /* check explicit source operand */
1401 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1402 if (err)
1403 return err;
1404 }
1405
1406 /* reset caller saved regs to unreadable */
1407 for (i = 0; i < CALLER_SAVED_REGS; i++) {
1408 reg = regs + caller_saved[i];
1409 reg->type = NOT_INIT;
1410 reg->imm = 0;
1411 }
1412
1413 /* mark destination R0 register as readable, since it contains
1414 * the value fetched from the packet
1415 */
1416 regs[BPF_REG_0].type = UNKNOWN_VALUE;
1417 return 0;
1418 }
1419
1420 /* non-recursive DFS pseudo code
1421 * 1 procedure DFS-iterative(G,v):
1422 * 2 label v as discovered
1423 * 3 let S be a stack
1424 * 4 S.push(v)
1425 * 5 while S is not empty
1426 * 6 t <- S.pop()
1427 * 7 if t is what we're looking for:
1428 * 8 return t
1429 * 9 for all edges e in G.adjacentEdges(t) do
1430 * 10 if edge e is already labelled
1431 * 11 continue with the next edge
1432 * 12 w <- G.adjacentVertex(t,e)
1433 * 13 if vertex w is not discovered and not explored
1434 * 14 label e as tree-edge
1435 * 15 label w as discovered
1436 * 16 S.push(w)
1437 * 17 continue at 5
1438 * 18 else if vertex w is discovered
1439 * 19 label e as back-edge
1440 * 20 else
1441 * 21 // vertex w is explored
1442 * 22 label e as forward- or cross-edge
1443 * 23 label t as explored
1444 * 24 S.pop()
1445 *
1446 * convention:
1447 * 0x10 - discovered
1448 * 0x11 - discovered and fall-through edge labelled
1449 * 0x12 - discovered and fall-through and branch edges labelled
1450 * 0x20 - explored
1451 */
1452
1453 enum {
1454 DISCOVERED = 0x10,
1455 EXPLORED = 0x20,
1456 FALLTHROUGH = 1,
1457 BRANCH = 2,
1458 };
1459
1460 #define STATE_LIST_MARK ((struct verifier_state_list *) -1L)
1461
1462 static int *insn_stack; /* stack of insns to process */
1463 static int cur_stack; /* current stack index */
1464 static int *insn_state;
1465
1466 /* t, w, e - match pseudo-code above:
1467 * t - index of current instruction
1468 * w - next instruction
1469 * e - edge
1470 */
1471 static int push_insn(int t, int w, int e, struct verifier_env *env)
1472 {
1473 if (e == FALLTHROUGH && insn_state[t] >= (DISCOVERED | FALLTHROUGH))
1474 return 0;
1475
1476 if (e == BRANCH && insn_state[t] >= (DISCOVERED | BRANCH))
1477 return 0;
1478
1479 if (w < 0 || w >= env->prog->len) {
1480 verbose("jump out of range from insn %d to %d\n", t, w);
1481 return -EINVAL;
1482 }
1483
1484 if (e == BRANCH)
1485 /* mark branch target for state pruning */
1486 env->explored_states[w] = STATE_LIST_MARK;
1487
1488 if (insn_state[w] == 0) {
1489 /* tree-edge */
1490 insn_state[t] = DISCOVERED | e;
1491 insn_state[w] = DISCOVERED;
1492 if (cur_stack >= env->prog->len)
1493 return -E2BIG;
1494 insn_stack[cur_stack++] = w;
1495 return 1;
1496 } else if ((insn_state[w] & 0xF0) == DISCOVERED) {
1497 verbose("back-edge from insn %d to %d\n", t, w);
1498 return -EINVAL;
1499 } else if (insn_state[w] == EXPLORED) {
1500 /* forward- or cross-edge */
1501 insn_state[t] = DISCOVERED | e;
1502 } else {
1503 verbose("insn state internal bug\n");
1504 return -EFAULT;
1505 }
1506 return 0;
1507 }
1508
1509 /* non-recursive depth-first-search to detect loops in BPF program
1510 * loop == back-edge in directed graph
1511 */
1512 static int check_cfg(struct verifier_env *env)
1513 {
1514 struct bpf_insn *insns = env->prog->insnsi;
1515 int insn_cnt = env->prog->len;
1516 int ret = 0;
1517 int i, t;
1518
1519 insn_state = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
1520 if (!insn_state)
1521 return -ENOMEM;
1522
1523 insn_stack = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
1524 if (!insn_stack) {
1525 kfree(insn_state);
1526 return -ENOMEM;
1527 }
1528
1529 insn_state[0] = DISCOVERED; /* mark 1st insn as discovered */
1530 insn_stack[0] = 0; /* 0 is the first instruction */
1531 cur_stack = 1;
1532
1533 peek_stack:
1534 if (cur_stack == 0)
1535 goto check_state;
1536 t = insn_stack[cur_stack - 1];
1537
1538 if (BPF_CLASS(insns[t].code) == BPF_JMP) {
1539 u8 opcode = BPF_OP(insns[t].code);
1540
1541 if (opcode == BPF_EXIT) {
1542 goto mark_explored;
1543 } else if (opcode == BPF_CALL) {
1544 ret = push_insn(t, t + 1, FALLTHROUGH, env);
1545 if (ret == 1)
1546 goto peek_stack;
1547 else if (ret < 0)
1548 goto err_free;
1549 if (t + 1 < insn_cnt)
1550 env->explored_states[t + 1] = STATE_LIST_MARK;
1551 } else if (opcode == BPF_JA) {
1552 if (BPF_SRC(insns[t].code) != BPF_K) {
1553 ret = -EINVAL;
1554 goto err_free;
1555 }
1556 /* unconditional jump with single edge */
1557 ret = push_insn(t, t + insns[t].off + 1,
1558 FALLTHROUGH, env);
1559 if (ret == 1)
1560 goto peek_stack;
1561 else if (ret < 0)
1562 goto err_free;
1563 /* tell verifier to check for equivalent states
1564 * after every call and jump
1565 */
1566 if (t + 1 < insn_cnt)
1567 env->explored_states[t + 1] = STATE_LIST_MARK;
1568 } else {
1569 /* conditional jump with two edges */
1570 ret = push_insn(t, t + 1, FALLTHROUGH, env);
1571 if (ret == 1)
1572 goto peek_stack;
1573 else if (ret < 0)
1574 goto err_free;
1575
1576 ret = push_insn(t, t + insns[t].off + 1, BRANCH, env);
1577 if (ret == 1)
1578 goto peek_stack;
1579 else if (ret < 0)
1580 goto err_free;
1581 }
1582 } else {
1583 /* all other non-branch instructions with single
1584 * fall-through edge
1585 */
1586 ret = push_insn(t, t + 1, FALLTHROUGH, env);
1587 if (ret == 1)
1588 goto peek_stack;
1589 else if (ret < 0)
1590 goto err_free;
1591 }
1592
1593 mark_explored:
1594 insn_state[t] = EXPLORED;
1595 if (cur_stack-- <= 0) {
1596 verbose("pop stack internal bug\n");
1597 ret = -EFAULT;
1598 goto err_free;
1599 }
1600 goto peek_stack;
1601
1602 check_state:
1603 for (i = 0; i < insn_cnt; i++) {
1604 if (insn_state[i] != EXPLORED) {
1605 verbose("unreachable insn %d\n", i);
1606 ret = -EINVAL;
1607 goto err_free;
1608 }
1609 }
1610 ret = 0; /* cfg looks good */
1611
1612 err_free:
1613 kfree(insn_state);
1614 kfree(insn_stack);
1615 return ret;
1616 }
1617
1618 /* compare two verifier states
1619 *
1620 * all states stored in state_list are known to be valid, since
1621 * verifier reached 'bpf_exit' instruction through them
1622 *
1623 * this function is called when verifier exploring different branches of
1624 * execution popped from the state stack. If it sees an old state that has
1625 * more strict register state and more strict stack state then this execution
1626 * branch doesn't need to be explored further, since verifier already
1627 * concluded that more strict state leads to valid finish.
1628 *
1629 * Therefore two states are equivalent if register state is more conservative
1630 * and explored stack state is more conservative than the current one.
1631 * Example:
1632 * explored current
1633 * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC)
1634 * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC)
1635 *
1636 * In other words if current stack state (one being explored) has more
1637 * valid slots than old one that already passed validation, it means
1638 * the verifier can stop exploring and conclude that current state is valid too
1639 *
1640 * Similarly with registers. If explored state has register type as invalid
1641 * whereas register type in current state is meaningful, it means that
1642 * the current state will reach 'bpf_exit' instruction safely
1643 */
1644 static bool states_equal(struct verifier_state *old, struct verifier_state *cur)
1645 {
1646 int i;
1647
1648 for (i = 0; i < MAX_BPF_REG; i++) {
1649 if (memcmp(&old->regs[i], &cur->regs[i],
1650 sizeof(old->regs[0])) != 0) {
1651 if (old->regs[i].type == NOT_INIT ||
1652 (old->regs[i].type == UNKNOWN_VALUE &&
1653 cur->regs[i].type != NOT_INIT))
1654 continue;
1655 return false;
1656 }
1657 }
1658
1659 for (i = 0; i < MAX_BPF_STACK; i++) {
1660 if (old->stack_slot_type[i] == STACK_INVALID)
1661 continue;
1662 if (old->stack_slot_type[i] != cur->stack_slot_type[i])
1663 /* Ex: old explored (safe) state has STACK_SPILL in
1664 * this stack slot, but current has has STACK_MISC ->
1665 * this verifier states are not equivalent,
1666 * return false to continue verification of this path
1667 */
1668 return false;
1669 if (i % BPF_REG_SIZE)
1670 continue;
1671 if (memcmp(&old->spilled_regs[i / BPF_REG_SIZE],
1672 &cur->spilled_regs[i / BPF_REG_SIZE],
1673 sizeof(old->spilled_regs[0])))
1674 /* when explored and current stack slot types are
1675 * the same, check that stored pointers types
1676 * are the same as well.
1677 * Ex: explored safe path could have stored
1678 * (struct reg_state) {.type = PTR_TO_STACK, .imm = -8}
1679 * but current path has stored:
1680 * (struct reg_state) {.type = PTR_TO_STACK, .imm = -16}
1681 * such verifier states are not equivalent.
1682 * return false to continue verification of this path
1683 */
1684 return false;
1685 else
1686 continue;
1687 }
1688 return true;
1689 }
1690
1691 static int is_state_visited(struct verifier_env *env, int insn_idx)
1692 {
1693 struct verifier_state_list *new_sl;
1694 struct verifier_state_list *sl;
1695
1696 sl = env->explored_states[insn_idx];
1697 if (!sl)
1698 /* this 'insn_idx' instruction wasn't marked, so we will not
1699 * be doing state search here
1700 */
1701 return 0;
1702
1703 while (sl != STATE_LIST_MARK) {
1704 if (states_equal(&sl->state, &env->cur_state))
1705 /* reached equivalent register/stack state,
1706 * prune the search
1707 */
1708 return 1;
1709 sl = sl->next;
1710 }
1711
1712 /* there were no equivalent states, remember current one.
1713 * technically the current state is not proven to be safe yet,
1714 * but it will either reach bpf_exit (which means it's safe) or
1715 * it will be rejected. Since there are no loops, we won't be
1716 * seeing this 'insn_idx' instruction again on the way to bpf_exit
1717 */
1718 new_sl = kmalloc(sizeof(struct verifier_state_list), GFP_USER);
1719 if (!new_sl)
1720 return -ENOMEM;
1721
1722 /* add new state to the head of linked list */
1723 memcpy(&new_sl->state, &env->cur_state, sizeof(env->cur_state));
1724 new_sl->next = env->explored_states[insn_idx];
1725 env->explored_states[insn_idx] = new_sl;
1726 return 0;
1727 }
1728
1729 static int do_check(struct verifier_env *env)
1730 {
1731 struct verifier_state *state = &env->cur_state;
1732 struct bpf_insn *insns = env->prog->insnsi;
1733 struct reg_state *regs = state->regs;
1734 int insn_cnt = env->prog->len;
1735 int insn_idx, prev_insn_idx = 0;
1736 int insn_processed = 0;
1737 bool do_print_state = false;
1738
1739 init_reg_state(regs);
1740 insn_idx = 0;
1741 for (;;) {
1742 struct bpf_insn *insn;
1743 u8 class;
1744 int err;
1745
1746 if (insn_idx >= insn_cnt) {
1747 verbose("invalid insn idx %d insn_cnt %d\n",
1748 insn_idx, insn_cnt);
1749 return -EFAULT;
1750 }
1751
1752 insn = &insns[insn_idx];
1753 class = BPF_CLASS(insn->code);
1754
1755 if (++insn_processed > BPF_COMPLEXITY_LIMIT_INSNS) {
1756 verbose("BPF program is too large. Proccessed %d insn\n",
1757 insn_processed);
1758 return -E2BIG;
1759 }
1760
1761 err = is_state_visited(env, insn_idx);
1762 if (err < 0)
1763 return err;
1764 if (err == 1) {
1765 /* found equivalent state, can prune the search */
1766 if (log_level) {
1767 if (do_print_state)
1768 verbose("\nfrom %d to %d: safe\n",
1769 prev_insn_idx, insn_idx);
1770 else
1771 verbose("%d: safe\n", insn_idx);
1772 }
1773 goto process_bpf_exit;
1774 }
1775
1776 if (log_level && do_print_state) {
1777 verbose("\nfrom %d to %d:", prev_insn_idx, insn_idx);
1778 print_verifier_state(env);
1779 do_print_state = false;
1780 }
1781
1782 if (log_level) {
1783 verbose("%d: ", insn_idx);
1784 print_bpf_insn(insn);
1785 }
1786
1787 if (class == BPF_ALU || class == BPF_ALU64) {
1788 err = check_alu_op(env, insn);
1789 if (err)
1790 return err;
1791
1792 } else if (class == BPF_LDX) {
1793 enum bpf_reg_type src_reg_type;
1794
1795 /* check for reserved fields is already done */
1796
1797 /* check src operand */
1798 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1799 if (err)
1800 return err;
1801
1802 err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK);
1803 if (err)
1804 return err;
1805
1806 src_reg_type = regs[insn->src_reg].type;
1807
1808 /* check that memory (src_reg + off) is readable,
1809 * the state of dst_reg will be updated by this func
1810 */
1811 err = check_mem_access(env, insn->src_reg, insn->off,
1812 BPF_SIZE(insn->code), BPF_READ,
1813 insn->dst_reg);
1814 if (err)
1815 return err;
1816
1817 if (BPF_SIZE(insn->code) != BPF_W) {
1818 insn_idx++;
1819 continue;
1820 }
1821
1822 if (insn->imm == 0) {
1823 /* saw a valid insn
1824 * dst_reg = *(u32 *)(src_reg + off)
1825 * use reserved 'imm' field to mark this insn
1826 */
1827 insn->imm = src_reg_type;
1828
1829 } else if (src_reg_type != insn->imm &&
1830 (src_reg_type == PTR_TO_CTX ||
1831 insn->imm == PTR_TO_CTX)) {
1832 /* ABuser program is trying to use the same insn
1833 * dst_reg = *(u32*) (src_reg + off)
1834 * with different pointer types:
1835 * src_reg == ctx in one branch and
1836 * src_reg == stack|map in some other branch.
1837 * Reject it.
1838 */
1839 verbose("same insn cannot be used with different pointers\n");
1840 return -EINVAL;
1841 }
1842
1843 } else if (class == BPF_STX) {
1844 enum bpf_reg_type dst_reg_type;
1845
1846 if (BPF_MODE(insn->code) == BPF_XADD) {
1847 err = check_xadd(env, insn);
1848 if (err)
1849 return err;
1850 insn_idx++;
1851 continue;
1852 }
1853
1854 /* check src1 operand */
1855 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1856 if (err)
1857 return err;
1858 /* check src2 operand */
1859 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1860 if (err)
1861 return err;
1862
1863 dst_reg_type = regs[insn->dst_reg].type;
1864
1865 /* check that memory (dst_reg + off) is writeable */
1866 err = check_mem_access(env, insn->dst_reg, insn->off,
1867 BPF_SIZE(insn->code), BPF_WRITE,
1868 insn->src_reg);
1869 if (err)
1870 return err;
1871
1872 if (insn->imm == 0) {
1873 insn->imm = dst_reg_type;
1874 } else if (dst_reg_type != insn->imm &&
1875 (dst_reg_type == PTR_TO_CTX ||
1876 insn->imm == PTR_TO_CTX)) {
1877 verbose("same insn cannot be used with different pointers\n");
1878 return -EINVAL;
1879 }
1880
1881 } else if (class == BPF_ST) {
1882 if (BPF_MODE(insn->code) != BPF_MEM ||
1883 insn->src_reg != BPF_REG_0) {
1884 verbose("BPF_ST uses reserved fields\n");
1885 return -EINVAL;
1886 }
1887 /* check src operand */
1888 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1889 if (err)
1890 return err;
1891
1892 /* check that memory (dst_reg + off) is writeable */
1893 err = check_mem_access(env, insn->dst_reg, insn->off,
1894 BPF_SIZE(insn->code), BPF_WRITE,
1895 -1);
1896 if (err)
1897 return err;
1898
1899 } else if (class == BPF_JMP) {
1900 u8 opcode = BPF_OP(insn->code);
1901
1902 if (opcode == BPF_CALL) {
1903 if (BPF_SRC(insn->code) != BPF_K ||
1904 insn->off != 0 ||
1905 insn->src_reg != BPF_REG_0 ||
1906 insn->dst_reg != BPF_REG_0) {
1907 verbose("BPF_CALL uses reserved fields\n");
1908 return -EINVAL;
1909 }
1910
1911 err = check_call(env, insn->imm);
1912 if (err)
1913 return err;
1914
1915 } else if (opcode == BPF_JA) {
1916 if (BPF_SRC(insn->code) != BPF_K ||
1917 insn->imm != 0 ||
1918 insn->src_reg != BPF_REG_0 ||
1919 insn->dst_reg != BPF_REG_0) {
1920 verbose("BPF_JA uses reserved fields\n");
1921 return -EINVAL;
1922 }
1923
1924 insn_idx += insn->off + 1;
1925 continue;
1926
1927 } else if (opcode == BPF_EXIT) {
1928 if (BPF_SRC(insn->code) != BPF_K ||
1929 insn->imm != 0 ||
1930 insn->src_reg != BPF_REG_0 ||
1931 insn->dst_reg != BPF_REG_0) {
1932 verbose("BPF_EXIT uses reserved fields\n");
1933 return -EINVAL;
1934 }
1935
1936 /* eBPF calling convetion is such that R0 is used
1937 * to return the value from eBPF program.
1938 * Make sure that it's readable at this time
1939 * of bpf_exit, which means that program wrote
1940 * something into it earlier
1941 */
1942 err = check_reg_arg(regs, BPF_REG_0, SRC_OP);
1943 if (err)
1944 return err;
1945
1946 if (is_pointer_value(env, BPF_REG_0)) {
1947 verbose("R0 leaks addr as return value\n");
1948 return -EACCES;
1949 }
1950
1951 process_bpf_exit:
1952 insn_idx = pop_stack(env, &prev_insn_idx);
1953 if (insn_idx < 0) {
1954 break;
1955 } else {
1956 do_print_state = true;
1957 continue;
1958 }
1959 } else {
1960 err = check_cond_jmp_op(env, insn, &insn_idx);
1961 if (err)
1962 return err;
1963 }
1964 } else if (class == BPF_LD) {
1965 u8 mode = BPF_MODE(insn->code);
1966
1967 if (mode == BPF_ABS || mode == BPF_IND) {
1968 err = check_ld_abs(env, insn);
1969 if (err)
1970 return err;
1971
1972 } else if (mode == BPF_IMM) {
1973 err = check_ld_imm(env, insn);
1974 if (err)
1975 return err;
1976
1977 insn_idx++;
1978 } else {
1979 verbose("invalid BPF_LD mode\n");
1980 return -EINVAL;
1981 }
1982 } else {
1983 verbose("unknown insn class %d\n", class);
1984 return -EINVAL;
1985 }
1986
1987 insn_idx++;
1988 }
1989
1990 return 0;
1991 }
1992
1993 /* look for pseudo eBPF instructions that access map FDs and
1994 * replace them with actual map pointers
1995 */
1996 static int replace_map_fd_with_map_ptr(struct verifier_env *env)
1997 {
1998 struct bpf_insn *insn = env->prog->insnsi;
1999 int insn_cnt = env->prog->len;
2000 int i, j;
2001
2002 for (i = 0; i < insn_cnt; i++, insn++) {
2003 if (BPF_CLASS(insn->code) == BPF_LDX &&
2004 (BPF_MODE(insn->code) != BPF_MEM || insn->imm != 0)) {
2005 verbose("BPF_LDX uses reserved fields\n");
2006 return -EINVAL;
2007 }
2008
2009 if (BPF_CLASS(insn->code) == BPF_STX &&
2010 ((BPF_MODE(insn->code) != BPF_MEM &&
2011 BPF_MODE(insn->code) != BPF_XADD) || insn->imm != 0)) {
2012 verbose("BPF_STX uses reserved fields\n");
2013 return -EINVAL;
2014 }
2015
2016 if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW)) {
2017 struct bpf_map *map;
2018 struct fd f;
2019
2020 if (i == insn_cnt - 1 || insn[1].code != 0 ||
2021 insn[1].dst_reg != 0 || insn[1].src_reg != 0 ||
2022 insn[1].off != 0) {
2023 verbose("invalid bpf_ld_imm64 insn\n");
2024 return -EINVAL;
2025 }
2026
2027 if (insn->src_reg == 0)
2028 /* valid generic load 64-bit imm */
2029 goto next_insn;
2030
2031 if (insn->src_reg != BPF_PSEUDO_MAP_FD) {
2032 verbose("unrecognized bpf_ld_imm64 insn\n");
2033 return -EINVAL;
2034 }
2035
2036 f = fdget(insn->imm);
2037 map = __bpf_map_get(f);
2038 if (IS_ERR(map)) {
2039 verbose("fd %d is not pointing to valid bpf_map\n",
2040 insn->imm);
2041 fdput(f);
2042 return PTR_ERR(map);
2043 }
2044
2045 /* store map pointer inside BPF_LD_IMM64 instruction */
2046 insn[0].imm = (u32) (unsigned long) map;
2047 insn[1].imm = ((u64) (unsigned long) map) >> 32;
2048
2049 /* check whether we recorded this map already */
2050 for (j = 0; j < env->used_map_cnt; j++)
2051 if (env->used_maps[j] == map) {
2052 fdput(f);
2053 goto next_insn;
2054 }
2055
2056 if (env->used_map_cnt >= MAX_USED_MAPS) {
2057 fdput(f);
2058 return -E2BIG;
2059 }
2060
2061 /* remember this map */
2062 env->used_maps[env->used_map_cnt++] = map;
2063
2064 /* hold the map. If the program is rejected by verifier,
2065 * the map will be released by release_maps() or it
2066 * will be used by the valid program until it's unloaded
2067 * and all maps are released in free_bpf_prog_info()
2068 */
2069 bpf_map_inc(map, false);
2070 fdput(f);
2071 next_insn:
2072 insn++;
2073 i++;
2074 }
2075 }
2076
2077 /* now all pseudo BPF_LD_IMM64 instructions load valid
2078 * 'struct bpf_map *' into a register instead of user map_fd.
2079 * These pointers will be used later by verifier to validate map access.
2080 */
2081 return 0;
2082 }
2083
2084 /* drop refcnt of maps used by the rejected program */
2085 static void release_maps(struct verifier_env *env)
2086 {
2087 int i;
2088
2089 for (i = 0; i < env->used_map_cnt; i++)
2090 bpf_map_put(env->used_maps[i]);
2091 }
2092
2093 /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */
2094 static void convert_pseudo_ld_imm64(struct verifier_env *env)
2095 {
2096 struct bpf_insn *insn = env->prog->insnsi;
2097 int insn_cnt = env->prog->len;
2098 int i;
2099
2100 for (i = 0; i < insn_cnt; i++, insn++)
2101 if (insn->code == (BPF_LD | BPF_IMM | BPF_DW))
2102 insn->src_reg = 0;
2103 }
2104
2105 static void adjust_branches(struct bpf_prog *prog, int pos, int delta)
2106 {
2107 struct bpf_insn *insn = prog->insnsi;
2108 int insn_cnt = prog->len;
2109 int i;
2110
2111 for (i = 0; i < insn_cnt; i++, insn++) {
2112 if (BPF_CLASS(insn->code) != BPF_JMP ||
2113 BPF_OP(insn->code) == BPF_CALL ||
2114 BPF_OP(insn->code) == BPF_EXIT)
2115 continue;
2116
2117 /* adjust offset of jmps if necessary */
2118 if (i < pos && i + insn->off + 1 > pos)
2119 insn->off += delta;
2120 else if (i > pos + delta && i + insn->off + 1 <= pos + delta)
2121 insn->off -= delta;
2122 }
2123 }
2124
2125 /* convert load instructions that access fields of 'struct __sk_buff'
2126 * into sequence of instructions that access fields of 'struct sk_buff'
2127 */
2128 static int convert_ctx_accesses(struct verifier_env *env)
2129 {
2130 struct bpf_insn *insn = env->prog->insnsi;
2131 int insn_cnt = env->prog->len;
2132 struct bpf_insn insn_buf[16];
2133 struct bpf_prog *new_prog;
2134 u32 cnt;
2135 int i;
2136 enum bpf_access_type type;
2137
2138 if (!env->prog->aux->ops->convert_ctx_access)
2139 return 0;
2140
2141 for (i = 0; i < insn_cnt; i++, insn++) {
2142 if (insn->code == (BPF_LDX | BPF_MEM | BPF_W))
2143 type = BPF_READ;
2144 else if (insn->code == (BPF_STX | BPF_MEM | BPF_W))
2145 type = BPF_WRITE;
2146 else
2147 continue;
2148
2149 if (insn->imm != PTR_TO_CTX) {
2150 /* clear internal mark */
2151 insn->imm = 0;
2152 continue;
2153 }
2154
2155 cnt = env->prog->aux->ops->
2156 convert_ctx_access(type, insn->dst_reg, insn->src_reg,
2157 insn->off, insn_buf, env->prog);
2158 if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) {
2159 verbose("bpf verifier is misconfigured\n");
2160 return -EINVAL;
2161 }
2162
2163 if (cnt == 1) {
2164 memcpy(insn, insn_buf, sizeof(*insn));
2165 continue;
2166 }
2167
2168 /* several new insns need to be inserted. Make room for them */
2169 insn_cnt += cnt - 1;
2170 new_prog = bpf_prog_realloc(env->prog,
2171 bpf_prog_size(insn_cnt),
2172 GFP_USER);
2173 if (!new_prog)
2174 return -ENOMEM;
2175
2176 new_prog->len = insn_cnt;
2177
2178 memmove(new_prog->insnsi + i + cnt, new_prog->insns + i + 1,
2179 sizeof(*insn) * (insn_cnt - i - cnt));
2180
2181 /* copy substitute insns in place of load instruction */
2182 memcpy(new_prog->insnsi + i, insn_buf, sizeof(*insn) * cnt);
2183
2184 /* adjust branches in the whole program */
2185 adjust_branches(new_prog, i, cnt - 1);
2186
2187 /* keep walking new program and skip insns we just inserted */
2188 env->prog = new_prog;
2189 insn = new_prog->insnsi + i + cnt - 1;
2190 i += cnt - 1;
2191 }
2192
2193 return 0;
2194 }
2195
2196 static void free_states(struct verifier_env *env)
2197 {
2198 struct verifier_state_list *sl, *sln;
2199 int i;
2200
2201 if (!env->explored_states)
2202 return;
2203
2204 for (i = 0; i < env->prog->len; i++) {
2205 sl = env->explored_states[i];
2206
2207 if (sl)
2208 while (sl != STATE_LIST_MARK) {
2209 sln = sl->next;
2210 kfree(sl);
2211 sl = sln;
2212 }
2213 }
2214
2215 kfree(env->explored_states);
2216 }
2217
2218 int bpf_check(struct bpf_prog **prog, union bpf_attr *attr)
2219 {
2220 char __user *log_ubuf = NULL;
2221 struct verifier_env *env;
2222 int ret = -EINVAL;
2223
2224 if ((*prog)->len <= 0 || (*prog)->len > BPF_MAXINSNS)
2225 return -E2BIG;
2226
2227 /* 'struct verifier_env' can be global, but since it's not small,
2228 * allocate/free it every time bpf_check() is called
2229 */
2230 env = kzalloc(sizeof(struct verifier_env), GFP_KERNEL);
2231 if (!env)
2232 return -ENOMEM;
2233
2234 env->prog = *prog;
2235
2236 /* grab the mutex to protect few globals used by verifier */
2237 mutex_lock(&bpf_verifier_lock);
2238
2239 if (attr->log_level || attr->log_buf || attr->log_size) {
2240 /* user requested verbose verifier output
2241 * and supplied buffer to store the verification trace
2242 */
2243 log_level = attr->log_level;
2244 log_ubuf = (char __user *) (unsigned long) attr->log_buf;
2245 log_size = attr->log_size;
2246 log_len = 0;
2247
2248 ret = -EINVAL;
2249 /* log_* values have to be sane */
2250 if (log_size < 128 || log_size > UINT_MAX >> 8 ||
2251 log_level == 0 || log_ubuf == NULL)
2252 goto free_env;
2253
2254 ret = -ENOMEM;
2255 log_buf = vmalloc(log_size);
2256 if (!log_buf)
2257 goto free_env;
2258 } else {
2259 log_level = 0;
2260 }
2261
2262 ret = replace_map_fd_with_map_ptr(env);
2263 if (ret < 0)
2264 goto skip_full_check;
2265
2266 env->explored_states = kcalloc(env->prog->len,
2267 sizeof(struct verifier_state_list *),
2268 GFP_USER);
2269 ret = -ENOMEM;
2270 if (!env->explored_states)
2271 goto skip_full_check;
2272
2273 ret = check_cfg(env);
2274 if (ret < 0)
2275 goto skip_full_check;
2276
2277 env->allow_ptr_leaks = capable(CAP_SYS_ADMIN);
2278
2279 ret = do_check(env);
2280
2281 skip_full_check:
2282 while (pop_stack(env, NULL) >= 0);
2283 free_states(env);
2284
2285 if (ret == 0)
2286 /* program is valid, convert *(u32*)(ctx + off) accesses */
2287 ret = convert_ctx_accesses(env);
2288
2289 if (log_level && log_len >= log_size - 1) {
2290 BUG_ON(log_len >= log_size);
2291 /* verifier log exceeded user supplied buffer */
2292 ret = -ENOSPC;
2293 /* fall through to return what was recorded */
2294 }
2295
2296 /* copy verifier log back to user space including trailing zero */
2297 if (log_level && copy_to_user(log_ubuf, log_buf, log_len + 1) != 0) {
2298 ret = -EFAULT;
2299 goto free_log_buf;
2300 }
2301
2302 if (ret == 0 && env->used_map_cnt) {
2303 /* if program passed verifier, update used_maps in bpf_prog_info */
2304 env->prog->aux->used_maps = kmalloc_array(env->used_map_cnt,
2305 sizeof(env->used_maps[0]),
2306 GFP_KERNEL);
2307
2308 if (!env->prog->aux->used_maps) {
2309 ret = -ENOMEM;
2310 goto free_log_buf;
2311 }
2312
2313 memcpy(env->prog->aux->used_maps, env->used_maps,
2314 sizeof(env->used_maps[0]) * env->used_map_cnt);
2315 env->prog->aux->used_map_cnt = env->used_map_cnt;
2316
2317 /* program is valid. Convert pseudo bpf_ld_imm64 into generic
2318 * bpf_ld_imm64 instructions
2319 */
2320 convert_pseudo_ld_imm64(env);
2321 }
2322
2323 free_log_buf:
2324 if (log_level)
2325 vfree(log_buf);
2326 free_env:
2327 if (!env->prog->aux->used_maps)
2328 /* if we didn't copy map pointers into bpf_prog_info, release
2329 * them now. Otherwise free_bpf_prog_info() will release them.
2330 */
2331 release_maps(env);
2332 *prog = env->prog;
2333 kfree(env);
2334 mutex_unlock(&bpf_verifier_lock);
2335 return ret;
2336 }
Ubuntu Jammy Linux kernel mirror