projects / GitHub / exynos8895 / android_kernel_samsung_universal8895.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
bpf: prevent leaking pointer via xadd on unpriviledged
[GitHub/exynos8895/android_kernel_samsung_universal8895.git] / kernel / bpf / verifier.c
CommitLineData
51580e79
AS		 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
17a52670
AS		 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 */
9c399760 156 STACK_SPILL, /* register spilled into stack */
17a52670
AS		 157 STACK_MISC /* BPF program wrote some data into this slot */
158};
159
9c399760 160#define BPF_REG_SIZE 8 /* size of eBPF register in bytes */
17a52670
AS		 161
162/* state of the program:
163 * type of all registers and stack info
164 */
165struct verifier_state {
166 struct reg_state regs[MAX_BPF_REG];
9c399760
AS		 167 u8 stack_slot_type[MAX_BPF_STACK];
168 struct reg_state spilled_regs[MAX_BPF_STACK / BPF_REG_SIZE];
17a52670
AS		 169};
170
171/* linked list of verifier states used to prune search */
172struct 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 */
178struct 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
0246e64d
AS		 189#define MAX_USED_MAPS 64 /* max number of maps accessed by one eBPF program */
190
cbd35700
AS		 191/* single container for all structs
192 * one verifier_env per bpf_check() call
193 */
194struct verifier_env {
0246e64d 195 struct bpf_prog *prog; /* eBPF program being verified */
17a52670
AS		 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 */
f1bca824 199 struct verifier_state_list **explored_states; /* search pruning optimization */
0246e64d
AS		 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 */
1be7f75d 202 bool allow_ptr_leaks;
cbd35700
AS		 203};
204
205/* verbose verifier prints what it's seeing
206 * bpf_check() is called under lock, so no race to access these global vars
207 */
208static u32 log_level, log_size, log_len;
209static char *log_buf;
210
211static DEFINE_MUTEX(bpf_verifier_lock);
212
213/* log_level controls verbosity level of eBPF verifier.
214 * verbose() is used to dump the verification trace to the log, so the user
215 * can figure out what's wrong with the program
216 */
1d056d9c 217static __printf(1, 2) void verbose(const char *fmt, ...)
cbd35700
AS		 218{
219 va_list args;
220
221 if (log_level == 0 || log_len >= log_size - 1)
222 return;
223
224 va_start(args, fmt);
225 log_len += vscnprintf(log_buf + log_len, log_size - log_len, fmt, args);
226 va_end(args);
227}
228
17a52670
AS		 229/* string representation of 'enum bpf_reg_type' */
230static const char * const reg_type_str[] = {
231 [NOT_INIT] = "?",
232 [UNKNOWN_VALUE] = "inv",
233 [PTR_TO_CTX] = "ctx",
234 [CONST_PTR_TO_MAP] = "map_ptr",
235 [PTR_TO_MAP_VALUE] = "map_value",
236 [PTR_TO_MAP_VALUE_OR_NULL] = "map_value_or_null",
237 [FRAME_PTR] = "fp",
238 [PTR_TO_STACK] = "fp",
239 [CONST_IMM] = "imm",
240};
241
242static void print_verifier_state(struct verifier_env *env)
243{
244 enum bpf_reg_type t;
245 int i;
246
247 for (i = 0; i < MAX_BPF_REG; i++) {
248 t = env->cur_state.regs[i].type;
249 if (t == NOT_INIT)
250 continue;
251 verbose(" R%d=%s", i, reg_type_str[t]);
252 if (t == CONST_IMM || t == PTR_TO_STACK)
253 verbose("%d", env->cur_state.regs[i].imm);
254 else if (t == CONST_PTR_TO_MAP || t == PTR_TO_MAP_VALUE ||
255 t == PTR_TO_MAP_VALUE_OR_NULL)
256 verbose("(ks=%d,vs=%d)",
257 env->cur_state.regs[i].map_ptr->key_size,
258 env->cur_state.regs[i].map_ptr->value_size);
259 }
9c399760
AS		 260 for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
261 if (env->cur_state.stack_slot_type[i] == STACK_SPILL)
17a52670 262 verbose(" fp%d=%s", -MAX_BPF_STACK + i,
9c399760 263 reg_type_str[env->cur_state.spilled_regs[i / BPF_REG_SIZE].type]);
17a52670
AS		 264 }
265 verbose("\n");
266}
267
cbd35700
AS		 268static const char *const bpf_class_string[] = {
269 [BPF_LD] = "ld",
270 [BPF_LDX] = "ldx",
271 [BPF_ST] = "st",
272 [BPF_STX] = "stx",
273 [BPF_ALU] = "alu",
274 [BPF_JMP] = "jmp",
275 [BPF_RET] = "BUG",
276 [BPF_ALU64] = "alu64",
277};
278
687f0715 279static const char *const bpf_alu_string[16] = {
cbd35700
AS		 280 [BPF_ADD >> 4] = "+=",
281 [BPF_SUB >> 4] = "-=",
282 [BPF_MUL >> 4] = "*=",
283 [BPF_DIV >> 4] = "/=",
284 [BPF_OR >> 4] = "|=",
285 [BPF_AND >> 4] = "&=",
286 [BPF_LSH >> 4] = "<<=",
287 [BPF_RSH >> 4] = ">>=",
288 [BPF_NEG >> 4] = "neg",
289 [BPF_MOD >> 4] = "%=",
290 [BPF_XOR >> 4] = "^=",
291 [BPF_MOV >> 4] = "=",
292 [BPF_ARSH >> 4] = "s>>=",
293 [BPF_END >> 4] = "endian",
294};
295
296static const char *const bpf_ldst_string[] = {
297 [BPF_W >> 3] = "u32",
298 [BPF_H >> 3] = "u16",
299 [BPF_B >> 3] = "u8",
300 [BPF_DW >> 3] = "u64",
301};
302
687f0715 303static const char *const bpf_jmp_string[16] = {
cbd35700
AS		 304 [BPF_JA >> 4] = "jmp",
305 [BPF_JEQ >> 4] = "==",
306 [BPF_JGT >> 4] = ">",
307 [BPF_JGE >> 4] = ">=",
308 [BPF_JSET >> 4] = "&",
309 [BPF_JNE >> 4] = "!=",
310 [BPF_JSGT >> 4] = "s>",
311 [BPF_JSGE >> 4] = "s>=",
312 [BPF_CALL >> 4] = "call",
313 [BPF_EXIT >> 4] = "exit",
314};
315
316static void print_bpf_insn(struct bpf_insn *insn)
317{
318 u8 class = BPF_CLASS(insn->code);
319
320 if (class == BPF_ALU || class == BPF_ALU64) {
321 if (BPF_SRC(insn->code) == BPF_X)
322 verbose("(%02x) %sr%d %s %sr%d\n",
323 insn->code, class == BPF_ALU ? "(u32) " : "",
324 insn->dst_reg,
325 bpf_alu_string[BPF_OP(insn->code) >> 4],
326 class == BPF_ALU ? "(u32) " : "",
327 insn->src_reg);
328 else
329 verbose("(%02x) %sr%d %s %s%d\n",
330 insn->code, class == BPF_ALU ? "(u32) " : "",
331 insn->dst_reg,
332 bpf_alu_string[BPF_OP(insn->code) >> 4],
333 class == BPF_ALU ? "(u32) " : "",
334 insn->imm);
335 } else if (class == BPF_STX) {
336 if (BPF_MODE(insn->code) == BPF_MEM)
337 verbose("(%02x) *(%s *)(r%d %+d) = r%d\n",
338 insn->code,
339 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
340 insn->dst_reg,
341 insn->off, insn->src_reg);
342 else if (BPF_MODE(insn->code) == BPF_XADD)
343 verbose("(%02x) lock *(%s *)(r%d %+d) += r%d\n",
344 insn->code,
345 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
346 insn->dst_reg, insn->off,
347 insn->src_reg);
348 else
349 verbose("BUG_%02x\n", insn->code);
350 } else if (class == BPF_ST) {
351 if (BPF_MODE(insn->code) != BPF_MEM) {
352 verbose("BUG_st_%02x\n", insn->code);
353 return;
354 }
355 verbose("(%02x) *(%s *)(r%d %+d) = %d\n",
356 insn->code,
357 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
358 insn->dst_reg,
359 insn->off, insn->imm);
360 } else if (class == BPF_LDX) {
361 if (BPF_MODE(insn->code) != BPF_MEM) {
362 verbose("BUG_ldx_%02x\n", insn->code);
363 return;
364 }
365 verbose("(%02x) r%d = *(%s *)(r%d %+d)\n",
366 insn->code, insn->dst_reg,
367 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
368 insn->src_reg, insn->off);
369 } else if (class == BPF_LD) {
370 if (BPF_MODE(insn->code) == BPF_ABS) {
371 verbose("(%02x) r0 = *(%s *)skb[%d]\n",
372 insn->code,
373 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
374 insn->imm);
375 } else if (BPF_MODE(insn->code) == BPF_IND) {
376 verbose("(%02x) r0 = *(%s *)skb[r%d + %d]\n",
377 insn->code,
378 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
379 insn->src_reg, insn->imm);
380 } else if (BPF_MODE(insn->code) == BPF_IMM) {
381 verbose("(%02x) r%d = 0x%x\n",
382 insn->code, insn->dst_reg, insn->imm);
383 } else {
384 verbose("BUG_ld_%02x\n", insn->code);
385 return;
386 }
387 } else if (class == BPF_JMP) {
388 u8 opcode = BPF_OP(insn->code);
389
390 if (opcode == BPF_CALL) {
391 verbose("(%02x) call %d\n", insn->code, insn->imm);
392 } else if (insn->code == (BPF_JMP | BPF_JA)) {
393 verbose("(%02x) goto pc%+d\n",
394 insn->code, insn->off);
395 } else if (insn->code == (BPF_JMP | BPF_EXIT)) {
396 verbose("(%02x) exit\n", insn->code);
397 } else if (BPF_SRC(insn->code) == BPF_X) {
398 verbose("(%02x) if r%d %s r%d goto pc%+d\n",
399 insn->code, insn->dst_reg,
400 bpf_jmp_string[BPF_OP(insn->code) >> 4],
401 insn->src_reg, insn->off);
402 } else {
403 verbose("(%02x) if r%d %s 0x%x goto pc%+d\n",
404 insn->code, insn->dst_reg,
405 bpf_jmp_string[BPF_OP(insn->code) >> 4],
406 insn->imm, insn->off);
407 }
408 } else {
409 verbose("(%02x) %s\n", insn->code, bpf_class_string[class]);
410 }
411}
412
17a52670
AS		 413static int pop_stack(struct verifier_env *env, int *prev_insn_idx)
414{
415 struct verifier_stack_elem *elem;
416 int insn_idx;
417
418 if (env->head == NULL)
419 return -1;
420
421 memcpy(&env->cur_state, &env->head->st, sizeof(env->cur_state));
422 insn_idx = env->head->insn_idx;
423 if (prev_insn_idx)
424 *prev_insn_idx = env->head->prev_insn_idx;
425 elem = env->head->next;
426 kfree(env->head);
427 env->head = elem;
428 env->stack_size--;
429 return insn_idx;
430}
431
432static struct verifier_state *push_stack(struct verifier_env *env, int insn_idx,
433 int prev_insn_idx)
434{
435 struct verifier_stack_elem *elem;
436
437 elem = kmalloc(sizeof(struct verifier_stack_elem), GFP_KERNEL);
438 if (!elem)
439 goto err;
440
441 memcpy(&elem->st, &env->cur_state, sizeof(env->cur_state));
442 elem->insn_idx = insn_idx;
443 elem->prev_insn_idx = prev_insn_idx;
444 elem->next = env->head;
445 env->head = elem;
446 env->stack_size++;
447 if (env->stack_size > 1024) {
448 verbose("BPF program is too complex\n");
449 goto err;
450 }
451 return &elem->st;
452err:
453 /* pop all elements and return */
454 while (pop_stack(env, NULL) >= 0);
455 return NULL;
456}
457
458#define CALLER_SAVED_REGS 6
459static const int caller_saved[CALLER_SAVED_REGS] = {
460 BPF_REG_0, BPF_REG_1, BPF_REG_2, BPF_REG_3, BPF_REG_4, BPF_REG_5
461};
462
463static void init_reg_state(struct reg_state *regs)
464{
465 int i;
466
467 for (i = 0; i < MAX_BPF_REG; i++) {
468 regs[i].type = NOT_INIT;
469 regs[i].imm = 0;
470 regs[i].map_ptr = NULL;
471 }
472
473 /* frame pointer */
474 regs[BPF_REG_FP].type = FRAME_PTR;
475
476 /* 1st arg to a function */
477 regs[BPF_REG_1].type = PTR_TO_CTX;
478}
479
480static void mark_reg_unknown_value(struct reg_state *regs, u32 regno)
481{
482 BUG_ON(regno >= MAX_BPF_REG);
483 regs[regno].type = UNKNOWN_VALUE;
484 regs[regno].imm = 0;
485 regs[regno].map_ptr = NULL;
486}
487
488enum reg_arg_type {
489 SRC_OP, /* register is used as source operand */
490 DST_OP, /* register is used as destination operand */
491 DST_OP_NO_MARK /* same as above, check only, don't mark */
492};
493
494static int check_reg_arg(struct reg_state *regs, u32 regno,
495 enum reg_arg_type t)
496{
497 if (regno >= MAX_BPF_REG) {
498 verbose("R%d is invalid\n", regno);
499 return -EINVAL;
500 }
501
502 if (t == SRC_OP) {
503 /* check whether register used as source operand can be read */
504 if (regs[regno].type == NOT_INIT) {
505 verbose("R%d !read_ok\n", regno);
506 return -EACCES;
507 }
508 } else {
509 /* check whether register used as dest operand can be written to */
510 if (regno == BPF_REG_FP) {
511 verbose("frame pointer is read only\n");
512 return -EACCES;
513 }
514 if (t == DST_OP)
515 mark_reg_unknown_value(regs, regno);
516 }
517 return 0;
518}
519
520static int bpf_size_to_bytes(int bpf_size)
521{
522 if (bpf_size == BPF_W)
523 return 4;
524 else if (bpf_size == BPF_H)
525 return 2;
526 else if (bpf_size == BPF_B)
527 return 1;
528 else if (bpf_size == BPF_DW)
529 return 8;
530 else
531 return -EINVAL;
532}
533
1be7f75d
AS		 534static bool is_spillable_regtype(enum bpf_reg_type type)
535{
536 switch (type) {
537 case PTR_TO_MAP_VALUE:
538 case PTR_TO_MAP_VALUE_OR_NULL:
539 case PTR_TO_STACK:
540 case PTR_TO_CTX:
541 case FRAME_PTR:
542 case CONST_PTR_TO_MAP:
543 return true;
544 default:
545 return false;
546 }
547}
548
17a52670
AS		 549/* check_stack_read/write functions track spill/fill of registers,
550 * stack boundary and alignment are checked in check_mem_access()
551 */
552static int check_stack_write(struct verifier_state *state, int off, int size,
553 int value_regno)
554{
17a52670 555 int i;
9c399760
AS		 556 /* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0,
557 * so it's aligned access and [off, off + size) are within stack limits
558 */
17a52670
AS		 559
560 if (value_regno >= 0 &&
1be7f75d 561 is_spillable_regtype(state->regs[value_regno].type)) {
17a52670
AS		 562
563 /* register containing pointer is being spilled into stack */
9c399760 564 if (size != BPF_REG_SIZE) {
17a52670
AS		 565 verbose("invalid size of register spill\n");
566 return -EACCES;
567 }
568
17a52670 569 /* save register state */
9c399760
AS		 570 state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] =
571 state->regs[value_regno];
17a52670 572
9c399760
AS		 573 for (i = 0; i < BPF_REG_SIZE; i++)
574 state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_SPILL;
575 } else {
17a52670 576 /* regular write of data into stack */
9c399760
AS		 577 state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] =
578 (struct reg_state) {};
579
580 for (i = 0; i < size; i++)
581 state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_MISC;
17a52670
AS		 582 }
583 return 0;
584}
585
586static int check_stack_read(struct verifier_state *state, int off, int size,
587 int value_regno)
588{
9c399760 589 u8 *slot_type;
17a52670 590 int i;
17a52670 591
9c399760 592 slot_type = &state->stack_slot_type[MAX_BPF_STACK + off];
17a52670 593
9c399760
AS		 594 if (slot_type[0] == STACK_SPILL) {
595 if (size != BPF_REG_SIZE) {
17a52670
AS		 596 verbose("invalid size of register spill\n");
597 return -EACCES;
598 }
9c399760
AS		 599 for (i = 1; i < BPF_REG_SIZE; i++) {
600 if (slot_type[i] != STACK_SPILL) {
17a52670
AS		 601 verbose("corrupted spill memory\n");
602 return -EACCES;
603 }
604 }
605
606 if (value_regno >= 0)
607 /* restore register state from stack */
9c399760
AS		 608 state->regs[value_regno] =
609 state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE];
17a52670
AS		 610 return 0;
611 } else {
612 for (i = 0; i < size; i++) {
9c399760 613 if (slot_type[i] != STACK_MISC) {
17a52670
AS		 614 verbose("invalid read from stack off %d+%d size %d\n",
615 off, i, size);
616 return -EACCES;
617 }
618 }
619 if (value_regno >= 0)
620 /* have read misc data from the stack */
621 mark_reg_unknown_value(state->regs, value_regno);
622 return 0;
623 }
624}
625
626/* check read/write into map element returned by bpf_map_lookup_elem() */
627static int check_map_access(struct verifier_env *env, u32 regno, int off,
628 int size)
629{
630 struct bpf_map *map = env->cur_state.regs[regno].map_ptr;
631
632 if (off < 0 || off + size > map->value_size) {
633 verbose("invalid access to map value, value_size=%d off=%d size=%d\n",
634 map->value_size, off, size);
635 return -EACCES;
636 }
637 return 0;
638}
639
640/* check access to 'struct bpf_context' fields */
641static int check_ctx_access(struct verifier_env *env, int off, int size,
642 enum bpf_access_type t)
643{
644 if (env->prog->aux->ops->is_valid_access &&
645 env->prog->aux->ops->is_valid_access(off, size, t))
646 return 0;
647
648 verbose("invalid bpf_context access off=%d size=%d\n", off, size);
649 return -EACCES;
650}
651
1be7f75d
AS		 652static bool is_pointer_value(struct verifier_env *env, int regno)
653{
654 if (env->allow_ptr_leaks)
655 return false;
656
657 switch (env->cur_state.regs[regno].type) {
658 case UNKNOWN_VALUE:
659 case CONST_IMM:
660 return false;
661 default:
662 return true;
663 }
664}
665
17a52670
AS		 666/* check whether memory at (regno + off) is accessible for t = (read | write)
667 * if t==write, value_regno is a register which value is stored into memory
668 * if t==read, value_regno is a register which will receive the value from memory
669 * if t==write && value_regno==-1, some unknown value is stored into memory
670 * if t==read && value_regno==-1, don't care what we read from memory
671 */
672static int check_mem_access(struct verifier_env *env, u32 regno, int off,
673 int bpf_size, enum bpf_access_type t,
674 int value_regno)
675{
676 struct verifier_state *state = &env->cur_state;
677 int size, err = 0;
678
24b4d2ab
AG		 679 if (state->regs[regno].type == PTR_TO_STACK)
680 off += state->regs[regno].imm;
681
17a52670
AS		 682 size = bpf_size_to_bytes(bpf_size);
683 if (size < 0)
684 return size;
685
686 if (off % size != 0) {
687 verbose("misaligned access off %d size %d\n", off, size);
688 return -EACCES;
689 }
690
691 if (state->regs[regno].type == PTR_TO_MAP_VALUE) {
1be7f75d
AS		 692 if (t == BPF_WRITE && value_regno >= 0 &&
693 is_pointer_value(env, value_regno)) {
694 verbose("R%d leaks addr into map\n", value_regno);
695 return -EACCES;
696 }
17a52670
AS		 697 err = check_map_access(env, regno, off, size);
698 if (!err && t == BPF_READ && value_regno >= 0)
699 mark_reg_unknown_value(state->regs, value_regno);
700
701 } else if (state->regs[regno].type == PTR_TO_CTX) {
1be7f75d
AS		 702 if (t == BPF_WRITE && value_regno >= 0 &&
703 is_pointer_value(env, value_regno)) {
704 verbose("R%d leaks addr into ctx\n", value_regno);
705 return -EACCES;
706 }
17a52670
AS		 707 err = check_ctx_access(env, off, size, t);
708 if (!err && t == BPF_READ && value_regno >= 0)
709 mark_reg_unknown_value(state->regs, value_regno);
710
24b4d2ab
AG		 711 } else if (state->regs[regno].type == FRAME_PTR ||
712 state->regs[regno].type == PTR_TO_STACK) {
17a52670
AS		 713 if (off >= 0 || off < -MAX_BPF_STACK) {
714 verbose("invalid stack off=%d size=%d\n", off, size);
715 return -EACCES;
716 }
1be7f75d
AS		 717 if (t == BPF_WRITE) {
718 if (!env->allow_ptr_leaks &&
719 state->stack_slot_type[MAX_BPF_STACK + off] == STACK_SPILL &&
720 size != BPF_REG_SIZE) {
721 verbose("attempt to corrupt spilled pointer on stack\n");
722 return -EACCES;
723 }
17a52670 724 err = check_stack_write(state, off, size, value_regno);
1be7f75d 725 } else {
17a52670 726 err = check_stack_read(state, off, size, value_regno);
1be7f75d 727 }
17a52670
AS		 728 } else {
729 verbose("R%d invalid mem access '%s'\n",
730 regno, reg_type_str[state->regs[regno].type]);
731 return -EACCES;
732 }
733 return err;
734}
735
736static int check_xadd(struct verifier_env *env, struct bpf_insn *insn)
737{
738 struct reg_state *regs = env->cur_state.regs;
739 int err;
740
741 if ((BPF_SIZE(insn->code) != BPF_W && BPF_SIZE(insn->code) != BPF_DW) ||
742 insn->imm != 0) {
743 verbose("BPF_XADD uses reserved fields\n");
744 return -EINVAL;
745 }
746
747 /* check src1 operand */
748 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
749 if (err)
750 return err;
751
752 /* check src2 operand */
753 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
754 if (err)
755 return err;
756
1a4f13e0
DB		 757 if (is_pointer_value(env, insn->src_reg)) {
758 verbose("R%d leaks addr into mem\n", insn->src_reg);
759 return -EACCES;
760 }
761
17a52670
AS		 762 /* check whether atomic_add can read the memory */
763 err = check_mem_access(env, insn->dst_reg, insn->off,
764 BPF_SIZE(insn->code), BPF_READ, -1);
765 if (err)
766 return err;
767
768 /* check whether atomic_add can write into the same memory */
769 return check_mem_access(env, insn->dst_reg, insn->off,
770 BPF_SIZE(insn->code), BPF_WRITE, -1);
771}
772
773/* when register 'regno' is passed into function that will read 'access_size'
774 * bytes from that pointer, make sure that it's within stack boundary
775 * and all elements of stack are initialized
776 */
777static int check_stack_boundary(struct verifier_env *env,
778 int regno, int access_size)
779{
780 struct verifier_state *state = &env->cur_state;
781 struct reg_state *regs = state->regs;
782 int off, i;
783
784 if (regs[regno].type != PTR_TO_STACK)
785 return -EACCES;
786
787 off = regs[regno].imm;
788 if (off >= 0 || off < -MAX_BPF_STACK || off + access_size > 0 ||
789 access_size <= 0) {
790 verbose("invalid stack type R%d off=%d access_size=%d\n",
791 regno, off, access_size);
792 return -EACCES;
793 }
794
795 for (i = 0; i < access_size; i++) {
9c399760 796 if (state->stack_slot_type[MAX_BPF_STACK + off + i] != STACK_MISC) {
17a52670
AS		 797 verbose("invalid indirect read from stack off %d+%d size %d\n",
798 off, i, access_size);
799 return -EACCES;
800 }
801 }
802 return 0;
803}
804
805static int check_func_arg(struct verifier_env *env, u32 regno,
806 enum bpf_arg_type arg_type, struct bpf_map **mapp)
807{
808 struct reg_state *reg = env->cur_state.regs + regno;
809 enum bpf_reg_type expected_type;
810 int err = 0;
811
80f1d68c 812 if (arg_type == ARG_DONTCARE)
17a52670
AS		 813 return 0;
814
815 if (reg->type == NOT_INIT) {
816 verbose("R%d !read_ok\n", regno);
817 return -EACCES;
818 }
819
1be7f75d
AS		 820 if (arg_type == ARG_ANYTHING) {
821 if (is_pointer_value(env, regno)) {
822 verbose("R%d leaks addr into helper function\n", regno);
823 return -EACCES;
824 }
80f1d68c 825 return 0;
1be7f75d 826 }
80f1d68c 827
17a52670
AS		 828 if (arg_type == ARG_PTR_TO_STACK || arg_type == ARG_PTR_TO_MAP_KEY ||
829 arg_type == ARG_PTR_TO_MAP_VALUE) {
830 expected_type = PTR_TO_STACK;
831 } else if (arg_type == ARG_CONST_STACK_SIZE) {
832 expected_type = CONST_IMM;
833 } else if (arg_type == ARG_CONST_MAP_PTR) {
834 expected_type = CONST_PTR_TO_MAP;
608cd71a
AS		 835 } else if (arg_type == ARG_PTR_TO_CTX) {
836 expected_type = PTR_TO_CTX;
17a52670
AS		 837 } else {
838 verbose("unsupported arg_type %d\n", arg_type);
839 return -EFAULT;
840 }
841
842 if (reg->type != expected_type) {
843 verbose("R%d type=%s expected=%s\n", regno,
844 reg_type_str[reg->type], reg_type_str[expected_type]);
845 return -EACCES;
846 }
847
848 if (arg_type == ARG_CONST_MAP_PTR) {
849 /* bpf_map_xxx(map_ptr) call: remember that map_ptr */
850 *mapp = reg->map_ptr;
851
852 } else if (arg_type == ARG_PTR_TO_MAP_KEY) {
853 /* bpf_map_xxx(..., map_ptr, ..., key) call:
854 * check that [key, key + map->key_size) are within
855 * stack limits and initialized
856 */
857 if (!*mapp) {
858 /* in function declaration map_ptr must come before
859 * map_key, so that it's verified and known before
860 * we have to check map_key here. Otherwise it means
861 * that kernel subsystem misconfigured verifier
862 */
863 verbose("invalid map_ptr to access map->key\n");
864 return -EACCES;
865 }
866 err = check_stack_boundary(env, regno, (*mapp)->key_size);
867
868 } else if (arg_type == ARG_PTR_TO_MAP_VALUE) {
869 /* bpf_map_xxx(..., map_ptr, ..., value) call:
870 * check [value, value + map->value_size) validity
871 */
872 if (!*mapp) {
873 /* kernel subsystem misconfigured verifier */
874 verbose("invalid map_ptr to access map->value\n");
875 return -EACCES;
876 }
877 err = check_stack_boundary(env, regno, (*mapp)->value_size);
878
879 } else if (arg_type == ARG_CONST_STACK_SIZE) {
880 /* bpf_xxx(..., buf, len) call will access 'len' bytes
881 * from stack pointer 'buf'. Check it
882 * note: regno == len, regno - 1 == buf
883 */
884 if (regno == 0) {
885 /* kernel subsystem misconfigured verifier */
886 verbose("ARG_CONST_STACK_SIZE cannot be first argument\n");
887 return -EACCES;
888 }
889 err = check_stack_boundary(env, regno - 1, reg->imm);
890 }
891
892 return err;
893}
894
35578d79
KX		 895static int check_map_func_compatibility(struct bpf_map *map, int func_id)
896{
35578d79
KX		 897 if (!map)
898 return 0;
899
bb10156f
AS		 900 /* We need a two way check, first is from map perspective ... */
901 switch (map->map_type) {
902 case BPF_MAP_TYPE_PROG_ARRAY:
903 if (func_id != BPF_FUNC_tail_call)
904 goto error;
905 break;
906 case BPF_MAP_TYPE_PERF_EVENT_ARRAY:
907 if (func_id != BPF_FUNC_perf_event_read &&
908 func_id != BPF_FUNC_perf_event_output)
909 goto error;
910 break;
911 default:
912 break;
913 }
914
915 /* ... and second from the function itself. */
916 switch (func_id) {
917 case BPF_FUNC_tail_call:
918 if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY)
919 goto error;
920 break;
921 case BPF_FUNC_perf_event_read:
922 case BPF_FUNC_perf_event_output:
923 if (map->map_type != BPF_MAP_TYPE_PERF_EVENT_ARRAY)
924 goto error;
925 break;
926 default:
927 break;
35578d79
KX		 928 }
929
930 return 0;
bb10156f
AS		 931error:
932 verbose("cannot pass map_type %d into func %d\n",
933 map->map_type, func_id);
934 return -EINVAL;
35578d79
KX		 935}
936
17a52670
AS		 937static int check_call(struct verifier_env *env, int func_id)
938{
939 struct verifier_state *state = &env->cur_state;
940 const struct bpf_func_proto *fn = NULL;
941 struct reg_state *regs = state->regs;
942 struct bpf_map *map = NULL;
943 struct reg_state *reg;
944 int i, err;
945
946 /* find function prototype */
947 if (func_id < 0 || func_id >= __BPF_FUNC_MAX_ID) {
948 verbose("invalid func %d\n", func_id);
949 return -EINVAL;
950 }
951
952 if (env->prog->aux->ops->get_func_proto)
953 fn = env->prog->aux->ops->get_func_proto(func_id);
954
955 if (!fn) {
956 verbose("unknown func %d\n", func_id);
957 return -EINVAL;
958 }
959
960 /* eBPF programs must be GPL compatible to use GPL-ed functions */
24701ece 961 if (!env->prog->gpl_compatible && fn->gpl_only) {
17a52670
AS		 962 verbose("cannot call GPL only function from proprietary program\n");
963 return -EINVAL;
964 }
965
966 /* check args */
967 err = check_func_arg(env, BPF_REG_1, fn->arg1_type, &map);
968 if (err)
969 return err;
970 err = check_func_arg(env, BPF_REG_2, fn->arg2_type, &map);
971 if (err)
972 return err;
973 err = check_func_arg(env, BPF_REG_3, fn->arg3_type, &map);
974 if (err)
975 return err;
976 err = check_func_arg(env, BPF_REG_4, fn->arg4_type, &map);
977 if (err)
978 return err;
979 err = check_func_arg(env, BPF_REG_5, fn->arg5_type, &map);
980 if (err)
981 return err;
982
983 /* reset caller saved regs */
984 for (i = 0; i < CALLER_SAVED_REGS; i++) {
985 reg = regs + caller_saved[i];
986 reg->type = NOT_INIT;
987 reg->imm = 0;
988 }
989
990 /* update return register */
991 if (fn->ret_type == RET_INTEGER) {
992 regs[BPF_REG_0].type = UNKNOWN_VALUE;
993 } else if (fn->ret_type == RET_VOID) {
994 regs[BPF_REG_0].type = NOT_INIT;
995 } else if (fn->ret_type == RET_PTR_TO_MAP_VALUE_OR_NULL) {
996 regs[BPF_REG_0].type = PTR_TO_MAP_VALUE_OR_NULL;
997 /* remember map_ptr, so that check_map_access()
998 * can check 'value_size' boundary of memory access
999 * to map element returned from bpf_map_lookup_elem()
1000 */
1001 if (map == NULL) {
1002 verbose("kernel subsystem misconfigured verifier\n");
1003 return -EINVAL;
1004 }
1005 regs[BPF_REG_0].map_ptr = map;
1006 } else {
1007 verbose("unknown return type %d of func %d\n",
1008 fn->ret_type, func_id);
1009 return -EINVAL;
1010 }
04fd61ab 1011
35578d79
KX		 1012 err = check_map_func_compatibility(map, func_id);
1013 if (err)
1014 return err;
04fd61ab 1015
17a52670
AS		 1016 return 0;
1017}
1018
1019/* check validity of 32-bit and 64-bit arithmetic operations */
1be7f75d 1020static int check_alu_op(struct verifier_env *env, struct bpf_insn *insn)
17a52670 1021{
1be7f75d 1022 struct reg_state *regs = env->cur_state.regs;
17a52670
AS		 1023 u8 opcode = BPF_OP(insn->code);
1024 int err;
1025
1026 if (opcode == BPF_END || opcode == BPF_NEG) {
1027 if (opcode == BPF_NEG) {
1028 if (BPF_SRC(insn->code) != 0 ||
1029 insn->src_reg != BPF_REG_0 ||
1030 insn->off != 0 || insn->imm != 0) {
1031 verbose("BPF_NEG uses reserved fields\n");
1032 return -EINVAL;
1033 }
1034 } else {
1035 if (insn->src_reg != BPF_REG_0 || insn->off != 0 ||
1036 (insn->imm != 16 && insn->imm != 32 && insn->imm != 64)) {
1037 verbose("BPF_END uses reserved fields\n");
1038 return -EINVAL;
1039 }
1040 }
1041
1042 /* check src operand */
1043 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1044 if (err)
1045 return err;
1046
1be7f75d
AS		 1047 if (is_pointer_value(env, insn->dst_reg)) {
1048 verbose("R%d pointer arithmetic prohibited\n",
1049 insn->dst_reg);
1050 return -EACCES;
1051 }
1052
17a52670
AS		 1053 /* check dest operand */
1054 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1055 if (err)
1056 return err;
1057
1058 } else if (opcode == BPF_MOV) {
1059
1060 if (BPF_SRC(insn->code) == BPF_X) {
1061 if (insn->imm != 0 || insn->off != 0) {
1062 verbose("BPF_MOV uses reserved fields\n");
1063 return -EINVAL;
1064 }
1065
1066 /* check src operand */
1067 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1068 if (err)
1069 return err;
1070 } else {
1071 if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
1072 verbose("BPF_MOV uses reserved fields\n");
1073 return -EINVAL;
1074 }
1075 }
1076
1077 /* check dest operand */
1078 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1079 if (err)
1080 return err;
1081
1082 if (BPF_SRC(insn->code) == BPF_X) {
1083 if (BPF_CLASS(insn->code) == BPF_ALU64) {
1084 /* case: R1 = R2
1085 * copy register state to dest reg
1086 */
1087 regs[insn->dst_reg] = regs[insn->src_reg];
1088 } else {
1be7f75d
AS		 1089 if (is_pointer_value(env, insn->src_reg)) {
1090 verbose("R%d partial copy of pointer\n",
1091 insn->src_reg);
1092 return -EACCES;
1093 }
17a52670
AS		 1094 regs[insn->dst_reg].type = UNKNOWN_VALUE;
1095 regs[insn->dst_reg].map_ptr = NULL;
1096 }
1097 } else {
1098 /* case: R = imm
1099 * remember the value we stored into this reg
1100 */
1101 regs[insn->dst_reg].type = CONST_IMM;
1102 regs[insn->dst_reg].imm = insn->imm;
1103 }
1104
1105 } else if (opcode > BPF_END) {
1106 verbose("invalid BPF_ALU opcode %x\n", opcode);
1107 return -EINVAL;
1108
1109 } else { /* all other ALU ops: and, sub, xor, add, ... */
1110
1111 bool stack_relative = false;
1112
1113 if (BPF_SRC(insn->code) == BPF_X) {
1114 if (insn->imm != 0 || insn->off != 0) {
1115 verbose("BPF_ALU uses reserved fields\n");
1116 return -EINVAL;
1117 }
1118 /* check src1 operand */
1119 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1120 if (err)
1121 return err;
1122 } else {
1123 if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
1124 verbose("BPF_ALU uses reserved fields\n");
1125 return -EINVAL;
1126 }
1127 }
1128
1129 /* check src2 operand */
1130 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1131 if (err)
1132 return err;
1133
1134 if ((opcode == BPF_MOD || opcode == BPF_DIV) &&
1135 BPF_SRC(insn->code) == BPF_K && insn->imm == 0) {
1136 verbose("div by zero\n");
1137 return -EINVAL;
1138 }
1139
35987ff2
RV		 1140 if ((opcode == BPF_LSH || opcode == BPF_RSH ||
1141 opcode == BPF_ARSH) && BPF_SRC(insn->code) == BPF_K) {
1142 int size = BPF_CLASS(insn->code) == BPF_ALU64 ? 64 : 32;
1143
1144 if (insn->imm < 0 || insn->imm >= size) {
1145 verbose("invalid shift %d\n", insn->imm);
1146 return -EINVAL;
1147 }
1148 }
1149
17a52670
AS		 1150 /* pattern match 'bpf_add Rx, imm' instruction */
1151 if (opcode == BPF_ADD && BPF_CLASS(insn->code) == BPF_ALU64 &&
1152 regs[insn->dst_reg].type == FRAME_PTR &&
1be7f75d 1153 BPF_SRC(insn->code) == BPF_K) {
17a52670 1154 stack_relative = true;
1be7f75d
AS		 1155 } else if (is_pointer_value(env, insn->dst_reg)) {
1156 verbose("R%d pointer arithmetic prohibited\n",
1157 insn->dst_reg);
1158 return -EACCES;
1159 } else if (BPF_SRC(insn->code) == BPF_X &&
1160 is_pointer_value(env, insn->src_reg)) {
1161 verbose("R%d pointer arithmetic prohibited\n",
1162 insn->src_reg);
1163 return -EACCES;
1164 }
17a52670
AS		 1165
1166 /* check dest operand */
1167 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1168 if (err)
1169 return err;
1170
1171 if (stack_relative) {
1172 regs[insn->dst_reg].type = PTR_TO_STACK;
1173 regs[insn->dst_reg].imm = insn->imm;
1174 }
1175 }
1176
1177 return 0;
1178}
1179
1180static int check_cond_jmp_op(struct verifier_env *env,
1181 struct bpf_insn *insn, int *insn_idx)
1182{
1183 struct reg_state *regs = env->cur_state.regs;
1184 struct verifier_state *other_branch;
1185 u8 opcode = BPF_OP(insn->code);
1186 int err;
1187
1188 if (opcode > BPF_EXIT) {
1189 verbose("invalid BPF_JMP opcode %x\n", opcode);
1190 return -EINVAL;
1191 }
1192
1193 if (BPF_SRC(insn->code) == BPF_X) {
1194 if (insn->imm != 0) {
1195 verbose("BPF_JMP uses reserved fields\n");
1196 return -EINVAL;
1197 }
1198
1199 /* check src1 operand */
1200 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1201 if (err)
1202 return err;
1be7f75d
AS		 1203
1204 if (is_pointer_value(env, insn->src_reg)) {
1205 verbose("R%d pointer comparison prohibited\n",
1206 insn->src_reg);
1207 return -EACCES;
1208 }
17a52670
AS		 1209 } else {
1210 if (insn->src_reg != BPF_REG_0) {
1211 verbose("BPF_JMP uses reserved fields\n");
1212 return -EINVAL;
1213 }
1214 }
1215
1216 /* check src2 operand */
1217 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1218 if (err)
1219 return err;
1220
1221 /* detect if R == 0 where R was initialized to zero earlier */
1222 if (BPF_SRC(insn->code) == BPF_K &&
1223 (opcode == BPF_JEQ || opcode == BPF_JNE) &&
1224 regs[insn->dst_reg].type == CONST_IMM &&
1225 regs[insn->dst_reg].imm == insn->imm) {
1226 if (opcode == BPF_JEQ) {
1227 /* if (imm == imm) goto pc+off;
1228 * only follow the goto, ignore fall-through
1229 */
1230 *insn_idx += insn->off;
1231 return 0;
1232 } else {
1233 /* if (imm != imm) goto pc+off;
1234 * only follow fall-through branch, since
1235 * that's where the program will go
1236 */
1237 return 0;
1238 }
1239 }
1240
1241 other_branch = push_stack(env, *insn_idx + insn->off + 1, *insn_idx);
1242 if (!other_branch)
1243 return -EFAULT;
1244
1245 /* detect if R == 0 where R is returned value from bpf_map_lookup_elem() */
1246 if (BPF_SRC(insn->code) == BPF_K &&
1247 insn->imm == 0 && (opcode == BPF_JEQ ||
1248 opcode == BPF_JNE) &&
1249 regs[insn->dst_reg].type == PTR_TO_MAP_VALUE_OR_NULL) {
1250 if (opcode == BPF_JEQ) {
1251 /* next fallthrough insn can access memory via
1252 * this register
1253 */
1254 regs[insn->dst_reg].type = PTR_TO_MAP_VALUE;
1255 /* branch targer cannot access it, since reg == 0 */
1256 other_branch->regs[insn->dst_reg].type = CONST_IMM;
1257 other_branch->regs[insn->dst_reg].imm = 0;
1258 } else {
1259 other_branch->regs[insn->dst_reg].type = PTR_TO_MAP_VALUE;
1260 regs[insn->dst_reg].type = CONST_IMM;
1261 regs[insn->dst_reg].imm = 0;
1262 }
1be7f75d
AS		 1263 } else if (is_pointer_value(env, insn->dst_reg)) {
1264 verbose("R%d pointer comparison prohibited\n", insn->dst_reg);
1265 return -EACCES;
17a52670
AS		 1266 } else if (BPF_SRC(insn->code) == BPF_K &&
1267 (opcode == BPF_JEQ || opcode == BPF_JNE)) {
1268
1269 if (opcode == BPF_JEQ) {
1270 /* detect if (R == imm) goto
1271 * and in the target state recognize that R = imm
1272 */
1273 other_branch->regs[insn->dst_reg].type = CONST_IMM;
1274 other_branch->regs[insn->dst_reg].imm = insn->imm;
1275 } else {
1276 /* detect if (R != imm) goto
1277 * and in the fall-through state recognize that R = imm
1278 */
1279 regs[insn->dst_reg].type = CONST_IMM;
1280 regs[insn->dst_reg].imm = insn->imm;
1281 }
1282 }
1283 if (log_level)
1284 print_verifier_state(env);
1285 return 0;
1286}
1287
0246e64d
AS		 1288/* return the map pointer stored inside BPF_LD_IMM64 instruction */
1289static struct bpf_map *ld_imm64_to_map_ptr(struct bpf_insn *insn)
1290{
1291 u64 imm64 = ((u64) (u32) insn[0].imm) | ((u64) (u32) insn[1].imm) << 32;
1292
1293 return (struct bpf_map *) (unsigned long) imm64;
1294}
1295
17a52670
AS		 1296/* verify BPF_LD_IMM64 instruction */
1297static int check_ld_imm(struct verifier_env *env, struct bpf_insn *insn)
1298{
1299 struct reg_state *regs = env->cur_state.regs;
1300 int err;
1301
1302 if (BPF_SIZE(insn->code) != BPF_DW) {
1303 verbose("invalid BPF_LD_IMM insn\n");
1304 return -EINVAL;
1305 }
1306 if (insn->off != 0) {
1307 verbose("BPF_LD_IMM64 uses reserved fields\n");
1308 return -EINVAL;
1309 }
1310
1311 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1312 if (err)
1313 return err;
1314
1315 if (insn->src_reg == 0)
1316 /* generic move 64-bit immediate into a register */
1317 return 0;
1318
1319 /* replace_map_fd_with_map_ptr() should have caught bad ld_imm64 */
1320 BUG_ON(insn->src_reg != BPF_PSEUDO_MAP_FD);
1321
1322 regs[insn->dst_reg].type = CONST_PTR_TO_MAP;
1323 regs[insn->dst_reg].map_ptr = ld_imm64_to_map_ptr(insn);
1324 return 0;
1325}
1326
96be4325
DB		 1327static bool may_access_skb(enum bpf_prog_type type)
1328{
1329 switch (type) {
1330 case BPF_PROG_TYPE_SOCKET_FILTER:
1331 case BPF_PROG_TYPE_SCHED_CLS:
94caee8c 1332 case BPF_PROG_TYPE_SCHED_ACT:
96be4325
DB		 1333 return true;
1334 default:
1335 return false;
1336 }
1337}
1338
ddd872bc
AS		 1339/* verify safety of LD_ABS|LD_IND instructions:
1340 * - they can only appear in the programs where ctx == skb
1341 * - since they are wrappers of function calls, they scratch R1-R5 registers,
1342 * preserve R6-R9, and store return value into R0
1343 *
1344 * Implicit input:
1345 * ctx == skb == R6 == CTX
1346 *
1347 * Explicit input:
1348 * SRC == any register
1349 * IMM == 32-bit immediate
1350 *
1351 * Output:
1352 * R0 - 8/16/32-bit skb data converted to cpu endianness
1353 */
1354static int check_ld_abs(struct verifier_env *env, struct bpf_insn *insn)
1355{
1356 struct reg_state *regs = env->cur_state.regs;
1357 u8 mode = BPF_MODE(insn->code);
1358 struct reg_state *reg;
1359 int i, err;
1360
24701ece 1361 if (!may_access_skb(env->prog->type)) {
96be4325 1362 verbose("BPF_LD_ABS|IND instructions not allowed for this program type\n");
ddd872bc
AS		 1363 return -EINVAL;
1364 }
1365
1366 if (insn->dst_reg != BPF_REG_0 || insn->off != 0 ||
8427d554 1367 BPF_SIZE(insn->code) == BPF_DW ||
ddd872bc
AS		 1368 (mode == BPF_ABS && insn->src_reg != BPF_REG_0)) {
1369 verbose("BPF_LD_ABS uses reserved fields\n");
1370 return -EINVAL;
1371 }
1372
1373 /* check whether implicit source operand (register R6) is readable */
1374 err = check_reg_arg(regs, BPF_REG_6, SRC_OP);
1375 if (err)
1376 return err;
1377
1378 if (regs[BPF_REG_6].type != PTR_TO_CTX) {
1379 verbose("at the time of BPF_LD_ABS|IND R6 != pointer to skb\n");
1380 return -EINVAL;
1381 }
1382
1383 if (mode == BPF_IND) {
1384 /* check explicit source operand */
1385 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1386 if (err)
1387 return err;
1388 }
1389
1390 /* reset caller saved regs to unreadable */
1391 for (i = 0; i < CALLER_SAVED_REGS; i++) {
1392 reg = regs + caller_saved[i];
1393 reg->type = NOT_INIT;
1394 reg->imm = 0;
1395 }
1396
1397 /* mark destination R0 register as readable, since it contains
1398 * the value fetched from the packet
1399 */
1400 regs[BPF_REG_0].type = UNKNOWN_VALUE;
1401 return 0;
1402}
1403
475fb78f
AS		 1404/* non-recursive DFS pseudo code
1405 * 1 procedure DFS-iterative(G,v):
1406 * 2 label v as discovered
1407 * 3 let S be a stack
1408 * 4 S.push(v)
1409 * 5 while S is not empty
1410 * 6 t <- S.pop()
1411 * 7 if t is what we're looking for:
1412 * 8 return t
1413 * 9 for all edges e in G.adjacentEdges(t) do
1414 * 10 if edge e is already labelled
1415 * 11 continue with the next edge
1416 * 12 w <- G.adjacentVertex(t,e)
1417 * 13 if vertex w is not discovered and not explored
1418 * 14 label e as tree-edge
1419 * 15 label w as discovered
1420 * 16 S.push(w)
1421 * 17 continue at 5
1422 * 18 else if vertex w is discovered
1423 * 19 label e as back-edge
1424 * 20 else
1425 * 21 // vertex w is explored
1426 * 22 label e as forward- or cross-edge
1427 * 23 label t as explored
1428 * 24 S.pop()
1429 *
1430 * convention:
1431 * 0x10 - discovered
1432 * 0x11 - discovered and fall-through edge labelled
1433 * 0x12 - discovered and fall-through and branch edges labelled
1434 * 0x20 - explored
1435 */
1436
1437enum {
1438 DISCOVERED = 0x10,
1439 EXPLORED = 0x20,
1440 FALLTHROUGH = 1,
1441 BRANCH = 2,
1442};
1443
f1bca824
AS		 1444#define STATE_LIST_MARK ((struct verifier_state_list *) -1L)
1445
475fb78f
AS		 1446static int *insn_stack; /* stack of insns to process */
1447static int cur_stack; /* current stack index */
1448static int *insn_state;
1449
1450/* t, w, e - match pseudo-code above:
1451 * t - index of current instruction
1452 * w - next instruction
1453 * e - edge
1454 */
1455static int push_insn(int t, int w, int e, struct verifier_env *env)
1456{
1457 if (e == FALLTHROUGH && insn_state[t] >= (DISCOVERED | FALLTHROUGH))
1458 return 0;
1459
1460 if (e == BRANCH && insn_state[t] >= (DISCOVERED | BRANCH))
1461 return 0;
1462
1463 if (w < 0 || w >= env->prog->len) {
1464 verbose("jump out of range from insn %d to %d\n", t, w);
1465 return -EINVAL;
1466 }
1467
f1bca824
AS		 1468 if (e == BRANCH)
1469 /* mark branch target for state pruning */
1470 env->explored_states[w] = STATE_LIST_MARK;
1471
475fb78f
AS		 1472 if (insn_state[w] == 0) {
1473 /* tree-edge */
1474 insn_state[t] = DISCOVERED | e;
1475 insn_state[w] = DISCOVERED;
1476 if (cur_stack >= env->prog->len)
1477 return -E2BIG;
1478 insn_stack[cur_stack++] = w;
1479 return 1;
1480 } else if ((insn_state[w] & 0xF0) == DISCOVERED) {
1481 verbose("back-edge from insn %d to %d\n", t, w);
1482 return -EINVAL;
1483 } else if (insn_state[w] == EXPLORED) {
1484 /* forward- or cross-edge */
1485 insn_state[t] = DISCOVERED | e;
1486 } else {
1487 verbose("insn state internal bug\n");
1488 return -EFAULT;
1489 }
1490 return 0;
1491}
1492
1493/* non-recursive depth-first-search to detect loops in BPF program
1494 * loop == back-edge in directed graph
1495 */
1496static int check_cfg(struct verifier_env *env)
1497{
1498 struct bpf_insn *insns = env->prog->insnsi;
1499 int insn_cnt = env->prog->len;
1500 int ret = 0;
1501 int i, t;
1502
1503 insn_state = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
1504 if (!insn_state)
1505 return -ENOMEM;
1506
1507 insn_stack = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
1508 if (!insn_stack) {
1509 kfree(insn_state);
1510 return -ENOMEM;
1511 }
1512
1513 insn_state[0] = DISCOVERED; /* mark 1st insn as discovered */
1514 insn_stack[0] = 0; /* 0 is the first instruction */
1515 cur_stack = 1;
1516
1517peek_stack:
1518 if (cur_stack == 0)
1519 goto check_state;
1520 t = insn_stack[cur_stack - 1];
1521
1522 if (BPF_CLASS(insns[t].code) == BPF_JMP) {
1523 u8 opcode = BPF_OP(insns[t].code);
1524
1525 if (opcode == BPF_EXIT) {
1526 goto mark_explored;
1527 } else if (opcode == BPF_CALL) {
1528 ret = push_insn(t, t + 1, FALLTHROUGH, env);
1529 if (ret == 1)
1530 goto peek_stack;
1531 else if (ret < 0)
1532 goto err_free;
1533 } else if (opcode == BPF_JA) {
1534 if (BPF_SRC(insns[t].code) != BPF_K) {
1535 ret = -EINVAL;
1536 goto err_free;
1537 }
1538 /* unconditional jump with single edge */
1539 ret = push_insn(t, t + insns[t].off + 1,
1540 FALLTHROUGH, env);
1541 if (ret == 1)
1542 goto peek_stack;
1543 else if (ret < 0)
1544 goto err_free;
f1bca824
AS		 1545 /* tell verifier to check for equivalent states
1546 * after every call and jump
1547 */
c3de6317
AS		 1548 if (t + 1 < insn_cnt)
1549 env->explored_states[t + 1] = STATE_LIST_MARK;
475fb78f
AS		 1550 } else {
1551 /* conditional jump with two edges */
1552 ret = push_insn(t, t + 1, FALLTHROUGH, env);
1553 if (ret == 1)
1554 goto peek_stack;
1555 else if (ret < 0)
1556 goto err_free;
1557
1558 ret = push_insn(t, t + insns[t].off + 1, BRANCH, env);
1559 if (ret == 1)
1560 goto peek_stack;
1561 else if (ret < 0)
1562 goto err_free;
1563 }
1564 } else {
1565 /* all other non-branch instructions with single
1566 * fall-through edge
1567 */
1568 ret = push_insn(t, t + 1, FALLTHROUGH, env);
1569 if (ret == 1)
1570 goto peek_stack;
1571 else if (ret < 0)
1572 goto err_free;
1573 }
1574
1575mark_explored:
1576 insn_state[t] = EXPLORED;
1577 if (cur_stack-- <= 0) {
1578 verbose("pop stack internal bug\n");
1579 ret = -EFAULT;
1580 goto err_free;
1581 }
1582 goto peek_stack;
1583
1584check_state:
1585 for (i = 0; i < insn_cnt; i++) {
1586 if (insn_state[i] != EXPLORED) {
1587 verbose("unreachable insn %d\n", i);
1588 ret = -EINVAL;
1589 goto err_free;
1590 }
1591 }
1592 ret = 0; /* cfg looks good */
1593
1594err_free:
1595 kfree(insn_state);
1596 kfree(insn_stack);
1597 return ret;
1598}
1599
f1bca824
AS		 1600/* compare two verifier states
1601 *
1602 * all states stored in state_list are known to be valid, since
1603 * verifier reached 'bpf_exit' instruction through them
1604 *
1605 * this function is called when verifier exploring different branches of
1606 * execution popped from the state stack. If it sees an old state that has
1607 * more strict register state and more strict stack state then this execution
1608 * branch doesn't need to be explored further, since verifier already
1609 * concluded that more strict state leads to valid finish.
1610 *
1611 * Therefore two states are equivalent if register state is more conservative
1612 * and explored stack state is more conservative than the current one.
1613 * Example:
1614 * explored current
1615 * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC)
1616 * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC)
1617 *
1618 * In other words if current stack state (one being explored) has more
1619 * valid slots than old one that already passed validation, it means
1620 * the verifier can stop exploring and conclude that current state is valid too
1621 *
1622 * Similarly with registers. If explored state has register type as invalid
1623 * whereas register type in current state is meaningful, it means that
1624 * the current state will reach 'bpf_exit' instruction safely
1625 */
1626static bool states_equal(struct verifier_state *old, struct verifier_state *cur)
1627{
1628 int i;
1629
1630 for (i = 0; i < MAX_BPF_REG; i++) {
1631 if (memcmp(&old->regs[i], &cur->regs[i],
1632 sizeof(old->regs[0])) != 0) {
1633 if (old->regs[i].type == NOT_INIT ||
32bf08a6
AS		 1634 (old->regs[i].type == UNKNOWN_VALUE &&
1635 cur->regs[i].type != NOT_INIT))
f1bca824
AS		 1636 continue;
1637 return false;
1638 }
1639 }
1640
1641 for (i = 0; i < MAX_BPF_STACK; i++) {
9c399760
AS		 1642 if (old->stack_slot_type[i] == STACK_INVALID)
1643 continue;
1644 if (old->stack_slot_type[i] != cur->stack_slot_type[i])
1645 /* Ex: old explored (safe) state has STACK_SPILL in
1646 * this stack slot, but current has has STACK_MISC ->
1647 * this verifier states are not equivalent,
1648 * return false to continue verification of this path
1649 */
f1bca824 1650 return false;
9c399760
AS		 1651 if (i % BPF_REG_SIZE)
1652 continue;
1653 if (memcmp(&old->spilled_regs[i / BPF_REG_SIZE],
1654 &cur->spilled_regs[i / BPF_REG_SIZE],
1655 sizeof(old->spilled_regs[0])))
1656 /* when explored and current stack slot types are
1657 * the same, check that stored pointers types
1658 * are the same as well.
1659 * Ex: explored safe path could have stored
1660 * (struct reg_state) {.type = PTR_TO_STACK, .imm = -8}
1661 * but current path has stored:
1662 * (struct reg_state) {.type = PTR_TO_STACK, .imm = -16}
1663 * such verifier states are not equivalent.
1664 * return false to continue verification of this path
1665 */
1666 return false;
1667 else
1668 continue;
f1bca824
AS		 1669 }
1670 return true;
1671}
1672
1673static int is_state_visited(struct verifier_env *env, int insn_idx)
1674{
1675 struct verifier_state_list *new_sl;
1676 struct verifier_state_list *sl;
1677
1678 sl = env->explored_states[insn_idx];
1679 if (!sl)
1680 /* this 'insn_idx' instruction wasn't marked, so we will not
1681 * be doing state search here
1682 */
1683 return 0;
1684
1685 while (sl != STATE_LIST_MARK) {
1686 if (states_equal(&sl->state, &env->cur_state))
1687 /* reached equivalent register/stack state,
1688 * prune the search
1689 */
1690 return 1;
1691 sl = sl->next;
1692 }
1693
1694 /* there were no equivalent states, remember current one.
1695 * technically the current state is not proven to be safe yet,
1696 * but it will either reach bpf_exit (which means it's safe) or
1697 * it will be rejected. Since there are no loops, we won't be
1698 * seeing this 'insn_idx' instruction again on the way to bpf_exit
1699 */
1700 new_sl = kmalloc(sizeof(struct verifier_state_list), GFP_USER);
1701 if (!new_sl)
1702 return -ENOMEM;
1703
1704 /* add new state to the head of linked list */
1705 memcpy(&new_sl->state, &env->cur_state, sizeof(env->cur_state));
1706 new_sl->next = env->explored_states[insn_idx];
1707 env->explored_states[insn_idx] = new_sl;
1708 return 0;
1709}
1710
17a52670
AS		 1711static int do_check(struct verifier_env *env)
1712{
1713 struct verifier_state *state = &env->cur_state;
1714 struct bpf_insn *insns = env->prog->insnsi;
1715 struct reg_state *regs = state->regs;
1716 int insn_cnt = env->prog->len;
1717 int insn_idx, prev_insn_idx = 0;
1718 int insn_processed = 0;
1719 bool do_print_state = false;
1720
1721 init_reg_state(regs);
1722 insn_idx = 0;
1723 for (;;) {
1724 struct bpf_insn *insn;
1725 u8 class;
1726 int err;
1727
1728 if (insn_idx >= insn_cnt) {
1729 verbose("invalid insn idx %d insn_cnt %d\n",
1730 insn_idx, insn_cnt);
1731 return -EFAULT;
1732 }
1733
1734 insn = &insns[insn_idx];
1735 class = BPF_CLASS(insn->code);
1736
1737 if (++insn_processed > 32768) {
1738 verbose("BPF program is too large. Proccessed %d insn\n",
1739 insn_processed);
1740 return -E2BIG;
1741 }
1742
f1bca824
AS		 1743 err = is_state_visited(env, insn_idx);
1744 if (err < 0)
1745 return err;
1746 if (err == 1) {
1747 /* found equivalent state, can prune the search */
1748 if (log_level) {
1749 if (do_print_state)
1750 verbose("\nfrom %d to %d: safe\n",
1751 prev_insn_idx, insn_idx);
1752 else
1753 verbose("%d: safe\n", insn_idx);
1754 }
1755 goto process_bpf_exit;
1756 }
1757
17a52670
AS		 1758 if (log_level && do_print_state) {
1759 verbose("\nfrom %d to %d:", prev_insn_idx, insn_idx);
1760 print_verifier_state(env);
1761 do_print_state = false;
1762 }
1763
1764 if (log_level) {
1765 verbose("%d: ", insn_idx);
1766 print_bpf_insn(insn);
1767 }
1768
1769 if (class == BPF_ALU || class == BPF_ALU64) {
1be7f75d 1770 err = check_alu_op(env, insn);
17a52670
AS		 1771 if (err)
1772 return err;
1773
1774 } else if (class == BPF_LDX) {
9bac3d6d
AS		 1775 enum bpf_reg_type src_reg_type;
1776
1777 /* check for reserved fields is already done */
1778
17a52670
AS		 1779 /* check src operand */
1780 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1781 if (err)
1782 return err;
1783
1784 err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK);
1785 if (err)
1786 return err;
1787
725f9dcd
AS		 1788 src_reg_type = regs[insn->src_reg].type;
1789
17a52670
AS		 1790 /* check that memory (src_reg + off) is readable,
1791 * the state of dst_reg will be updated by this func
1792 */
1793 err = check_mem_access(env, insn->src_reg, insn->off,
1794 BPF_SIZE(insn->code), BPF_READ,
1795 insn->dst_reg);
1796 if (err)
1797 return err;
1798
725f9dcd
AS		 1799 if (BPF_SIZE(insn->code) != BPF_W) {
1800 insn_idx++;
1801 continue;
1802 }
9bac3d6d 1803
725f9dcd 1804 if (insn->imm == 0) {
9bac3d6d
AS		 1805 /* saw a valid insn
1806 * dst_reg = *(u32 *)(src_reg + off)
1807 * use reserved 'imm' field to mark this insn
1808 */
1809 insn->imm = src_reg_type;
1810
1811 } else if (src_reg_type != insn->imm &&
1812 (src_reg_type == PTR_TO_CTX ||
1813 insn->imm == PTR_TO_CTX)) {
1814 /* ABuser program is trying to use the same insn
1815 * dst_reg = *(u32*) (src_reg + off)
1816 * with different pointer types:
1817 * src_reg == ctx in one branch and
1818 * src_reg == stack|map in some other branch.
1819 * Reject it.
1820 */
1821 verbose("same insn cannot be used with different pointers\n");
1822 return -EINVAL;
1823 }
1824
17a52670 1825 } else if (class == BPF_STX) {
d691f9e8
AS		 1826 enum bpf_reg_type dst_reg_type;
1827
17a52670
AS		 1828 if (BPF_MODE(insn->code) == BPF_XADD) {
1829 err = check_xadd(env, insn);
1830 if (err)
1831 return err;
1832 insn_idx++;
1833 continue;
1834 }
1835
17a52670
AS		 1836 /* check src1 operand */
1837 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1838 if (err)
1839 return err;
1840 /* check src2 operand */
1841 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1842 if (err)
1843 return err;
1844
d691f9e8
AS		 1845 dst_reg_type = regs[insn->dst_reg].type;
1846
17a52670
AS		 1847 /* check that memory (dst_reg + off) is writeable */
1848 err = check_mem_access(env, insn->dst_reg, insn->off,
1849 BPF_SIZE(insn->code), BPF_WRITE,
1850 insn->src_reg);
1851 if (err)
1852 return err;
1853
d691f9e8
AS		 1854 if (insn->imm == 0) {
1855 insn->imm = dst_reg_type;
1856 } else if (dst_reg_type != insn->imm &&
1857 (dst_reg_type == PTR_TO_CTX ||
1858 insn->imm == PTR_TO_CTX)) {
1859 verbose("same insn cannot be used with different pointers\n");
1860 return -EINVAL;
1861 }
1862
17a52670
AS		 1863 } else if (class == BPF_ST) {
1864 if (BPF_MODE(insn->code) != BPF_MEM ||
1865 insn->src_reg != BPF_REG_0) {
1866 verbose("BPF_ST uses reserved fields\n");
1867 return -EINVAL;
1868 }
1869 /* check src operand */
1870 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1871 if (err)
1872 return err;
1873
1874 /* check that memory (dst_reg + off) is writeable */
1875 err = check_mem_access(env, insn->dst_reg, insn->off,
1876 BPF_SIZE(insn->code), BPF_WRITE,
1877 -1);
1878 if (err)
1879 return err;
1880
1881 } else if (class == BPF_JMP) {
1882 u8 opcode = BPF_OP(insn->code);
1883
1884 if (opcode == BPF_CALL) {
1885 if (BPF_SRC(insn->code) != BPF_K ||
1886 insn->off != 0 ||
1887 insn->src_reg != BPF_REG_0 ||
1888 insn->dst_reg != BPF_REG_0) {
1889 verbose("BPF_CALL uses reserved fields\n");
1890 return -EINVAL;
1891 }
1892
1893 err = check_call(env, insn->imm);
1894 if (err)
1895 return err;
1896
1897 } else if (opcode == BPF_JA) {
1898 if (BPF_SRC(insn->code) != BPF_K ||
1899 insn->imm != 0 ||
1900 insn->src_reg != BPF_REG_0 ||
1901 insn->dst_reg != BPF_REG_0) {
1902 verbose("BPF_JA uses reserved fields\n");
1903 return -EINVAL;
1904 }
1905
1906 insn_idx += insn->off + 1;
1907 continue;
1908
1909 } else if (opcode == BPF_EXIT) {
1910 if (BPF_SRC(insn->code) != BPF_K ||
1911 insn->imm != 0 ||
1912 insn->src_reg != BPF_REG_0 ||
1913 insn->dst_reg != BPF_REG_0) {
1914 verbose("BPF_EXIT uses reserved fields\n");
1915 return -EINVAL;
1916 }
1917
1918 /* eBPF calling convetion is such that R0 is used
1919 * to return the value from eBPF program.
1920 * Make sure that it's readable at this time
1921 * of bpf_exit, which means that program wrote
1922 * something into it earlier
1923 */
1924 err = check_reg_arg(regs, BPF_REG_0, SRC_OP);
1925 if (err)
1926 return err;
1927
1be7f75d
AS		 1928 if (is_pointer_value(env, BPF_REG_0)) {
1929 verbose("R0 leaks addr as return value\n");
1930 return -EACCES;
1931 }
1932
f1bca824 1933process_bpf_exit:
17a52670
AS		 1934 insn_idx = pop_stack(env, &prev_insn_idx);
1935 if (insn_idx < 0) {
1936 break;
1937 } else {
1938 do_print_state = true;
1939 continue;
1940 }
1941 } else {
1942 err = check_cond_jmp_op(env, insn, &insn_idx);
1943 if (err)
1944 return err;
1945 }
1946 } else if (class == BPF_LD) {
1947 u8 mode = BPF_MODE(insn->code);
1948
1949 if (mode == BPF_ABS || mode == BPF_IND) {
ddd872bc
AS		 1950 err = check_ld_abs(env, insn);
1951 if (err)
1952 return err;
1953
17a52670
AS		 1954 } else if (mode == BPF_IMM) {
1955 err = check_ld_imm(env, insn);
1956 if (err)
1957 return err;
1958
1959 insn_idx++;
1960 } else {
1961 verbose("invalid BPF_LD mode\n");
1962 return -EINVAL;
1963 }
1964 } else {
1965 verbose("unknown insn class %d\n", class);
1966 return -EINVAL;
1967 }
1968
1969 insn_idx++;
1970 }
1971
1972 return 0;
1973}
1974
0246e64d
AS		 1975/* look for pseudo eBPF instructions that access map FDs and
1976 * replace them with actual map pointers
1977 */
1978static int replace_map_fd_with_map_ptr(struct verifier_env *env)
1979{
1980 struct bpf_insn *insn = env->prog->insnsi;
1981 int insn_cnt = env->prog->len;
1982 int i, j;
1983
1984 for (i = 0; i < insn_cnt; i++, insn++) {
9bac3d6d 1985 if (BPF_CLASS(insn->code) == BPF_LDX &&
d691f9e8 1986 (BPF_MODE(insn->code) != BPF_MEM || insn->imm != 0)) {
9bac3d6d
AS		 1987 verbose("BPF_LDX uses reserved fields\n");
1988 return -EINVAL;
1989 }
1990
d691f9e8
AS		 1991 if (BPF_CLASS(insn->code) == BPF_STX &&
1992 ((BPF_MODE(insn->code) != BPF_MEM &&
1993 BPF_MODE(insn->code) != BPF_XADD) || insn->imm != 0)) {
1994 verbose("BPF_STX uses reserved fields\n");
1995 return -EINVAL;
1996 }
1997
0246e64d
AS		 1998 if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW)) {
1999 struct bpf_map *map;
2000 struct fd f;
2001
2002 if (i == insn_cnt - 1 || insn[1].code != 0 ||
2003 insn[1].dst_reg != 0 || insn[1].src_reg != 0 ||
2004 insn[1].off != 0) {
2005 verbose("invalid bpf_ld_imm64 insn\n");
2006 return -EINVAL;
2007 }
2008
2009 if (insn->src_reg == 0)
2010 /* valid generic load 64-bit imm */
2011 goto next_insn;
2012
2013 if (insn->src_reg != BPF_PSEUDO_MAP_FD) {
2014 verbose("unrecognized bpf_ld_imm64 insn\n");
2015 return -EINVAL;
2016 }
2017
2018 f = fdget(insn->imm);
c2101297 2019 map = __bpf_map_get(f);
0246e64d
AS		 2020 if (IS_ERR(map)) {
2021 verbose("fd %d is not pointing to valid bpf_map\n",
2022 insn->imm);
0246e64d
AS		 2023 return PTR_ERR(map);
2024 }
2025
2026 /* store map pointer inside BPF_LD_IMM64 instruction */
2027 insn[0].imm = (u32) (unsigned long) map;
2028 insn[1].imm = ((u64) (unsigned long) map) >> 32;
2029
2030 /* check whether we recorded this map already */
2031 for (j = 0; j < env->used_map_cnt; j++)
2032 if (env->used_maps[j] == map) {
2033 fdput(f);
2034 goto next_insn;
2035 }
2036
2037 if (env->used_map_cnt >= MAX_USED_MAPS) {
2038 fdput(f);
2039 return -E2BIG;
2040 }
2041
0246e64d
AS		 2042 /* hold the map. If the program is rejected by verifier,
2043 * the map will be released by release_maps() or it
2044 * will be used by the valid program until it's unloaded
2045 * and all maps are released in free_bpf_prog_info()
2046 */
3899251b
AS		 2047 map = bpf_map_inc(map, false);
2048 if (IS_ERR(map)) {
2049 fdput(f);
2050 return PTR_ERR(map);
2051 }
2052 env->used_maps[env->used_map_cnt++] = map;
2053
0246e64d
AS		 2054 fdput(f);
2055next_insn:
2056 insn++;
2057 i++;
2058 }
2059 }
2060
2061 /* now all pseudo BPF_LD_IMM64 instructions load valid
2062 * 'struct bpf_map *' into a register instead of user map_fd.
2063 * These pointers will be used later by verifier to validate map access.
2064 */
2065 return 0;
2066}
2067
2068/* drop refcnt of maps used by the rejected program */
2069static void release_maps(struct verifier_env *env)
2070{
2071 int i;
2072
2073 for (i = 0; i < env->used_map_cnt; i++)
2074 bpf_map_put(env->used_maps[i]);
2075}
2076
2077/* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */
2078static void convert_pseudo_ld_imm64(struct verifier_env *env)
2079{
2080 struct bpf_insn *insn = env->prog->insnsi;
2081 int insn_cnt = env->prog->len;
2082 int i;
2083
2084 for (i = 0; i < insn_cnt; i++, insn++)
2085 if (insn->code == (BPF_LD | BPF_IMM | BPF_DW))
2086 insn->src_reg = 0;
2087}
2088
9bac3d6d
AS		 2089static void adjust_branches(struct bpf_prog *prog, int pos, int delta)
2090{
2091 struct bpf_insn *insn = prog->insnsi;
2092 int insn_cnt = prog->len;
2093 int i;
2094
2095 for (i = 0; i < insn_cnt; i++, insn++) {
2096 if (BPF_CLASS(insn->code) != BPF_JMP ||
2097 BPF_OP(insn->code) == BPF_CALL ||
2098 BPF_OP(insn->code) == BPF_EXIT)
2099 continue;
2100
2101 /* adjust offset of jmps if necessary */
2102 if (i < pos && i + insn->off + 1 > pos)
2103 insn->off += delta;
a34f2f9f 2104 else if (i > pos + delta && i + insn->off + 1 <= pos + delta)
9bac3d6d
AS		 2105 insn->off -= delta;
2106 }
2107}
2108
2109/* convert load instructions that access fields of 'struct __sk_buff'
2110 * into sequence of instructions that access fields of 'struct sk_buff'
2111 */
2112static int convert_ctx_accesses(struct verifier_env *env)
2113{
2114 struct bpf_insn *insn = env->prog->insnsi;
2115 int insn_cnt = env->prog->len;
2116 struct bpf_insn insn_buf[16];
2117 struct bpf_prog *new_prog;
2118 u32 cnt;
2119 int i;
d691f9e8 2120 enum bpf_access_type type;
9bac3d6d
AS		 2121
2122 if (!env->prog->aux->ops->convert_ctx_access)
2123 return 0;
2124
2125 for (i = 0; i < insn_cnt; i++, insn++) {
d691f9e8
AS		 2126 if (insn->code == (BPF_LDX | BPF_MEM | BPF_W))
2127 type = BPF_READ;
2128 else if (insn->code == (BPF_STX | BPF_MEM | BPF_W))
2129 type = BPF_WRITE;
2130 else
9bac3d6d
AS		 2131 continue;
2132
2133 if (insn->imm != PTR_TO_CTX) {
2134 /* clear internal mark */
2135 insn->imm = 0;
2136 continue;
2137 }
2138
2139 cnt = env->prog->aux->ops->
d691f9e8 2140 convert_ctx_access(type, insn->dst_reg, insn->src_reg,
ff936a04 2141 insn->off, insn_buf, env->prog);
9bac3d6d
AS		 2142 if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) {
2143 verbose("bpf verifier is misconfigured\n");
2144 return -EINVAL;
2145 }
2146
2147 if (cnt == 1) {
2148 memcpy(insn, insn_buf, sizeof(*insn));
2149 continue;
2150 }
2151
2152 /* several new insns need to be inserted. Make room for them */
2153 insn_cnt += cnt - 1;
2154 new_prog = bpf_prog_realloc(env->prog,
2155 bpf_prog_size(insn_cnt),
2156 GFP_USER);
2157 if (!new_prog)
2158 return -ENOMEM;
2159
2160 new_prog->len = insn_cnt;
2161
2162 memmove(new_prog->insnsi + i + cnt, new_prog->insns + i + 1,
2163 sizeof(*insn) * (insn_cnt - i - cnt));
2164
2165 /* copy substitute insns in place of load instruction */
2166 memcpy(new_prog->insnsi + i, insn_buf, sizeof(*insn) * cnt);
2167
2168 /* adjust branches in the whole program */
2169 adjust_branches(new_prog, i, cnt - 1);
2170
2171 /* keep walking new program and skip insns we just inserted */
2172 env->prog = new_prog;
2173 insn = new_prog->insnsi + i + cnt - 1;
2174 i += cnt - 1;
2175 }
2176
2177 return 0;
2178}
2179
f1bca824
AS		 2180static void free_states(struct verifier_env *env)
2181{
2182 struct verifier_state_list *sl, *sln;
2183 int i;
2184
2185 if (!env->explored_states)
2186 return;
2187
2188 for (i = 0; i < env->prog->len; i++) {
2189 sl = env->explored_states[i];
2190
2191 if (sl)
2192 while (sl != STATE_LIST_MARK) {
2193 sln = sl->next;
2194 kfree(sl);
2195 sl = sln;
2196 }
2197 }
2198
2199 kfree(env->explored_states);
2200}
2201
9bac3d6d 2202int bpf_check(struct bpf_prog **prog, union bpf_attr *attr)
51580e79 2203{
cbd35700
AS		 2204 char __user *log_ubuf = NULL;
2205 struct verifier_env *env;
51580e79
AS		 2206 int ret = -EINVAL;
2207
9bac3d6d 2208 if ((*prog)->len <= 0 || (*prog)->len > BPF_MAXINSNS)
cbd35700
AS		 2209 return -E2BIG;
2210
2211 /* 'struct verifier_env' can be global, but since it's not small,
2212 * allocate/free it every time bpf_check() is called
2213 */
2214 env = kzalloc(sizeof(struct verifier_env), GFP_KERNEL);
2215 if (!env)
2216 return -ENOMEM;
2217
9bac3d6d 2218 env->prog = *prog;
0246e64d 2219
cbd35700
AS		 2220 /* grab the mutex to protect few globals used by verifier */
2221 mutex_lock(&bpf_verifier_lock);
2222
2223 if (attr->log_level || attr->log_buf || attr->log_size) {
2224 /* user requested verbose verifier output
2225 * and supplied buffer to store the verification trace
2226 */
2227 log_level = attr->log_level;
2228 log_ubuf = (char __user *) (unsigned long) attr->log_buf;
2229 log_size = attr->log_size;
2230 log_len = 0;
2231
2232 ret = -EINVAL;
2233 /* log_* values have to be sane */
2234 if (log_size < 128 || log_size > UINT_MAX >> 8 ||
2235 log_level == 0 || log_ubuf == NULL)
2236 goto free_env;
2237
2238 ret = -ENOMEM;
2239 log_buf = vmalloc(log_size);
2240 if (!log_buf)
2241 goto free_env;
2242 } else {
2243 log_level = 0;
2244 }
2245
0246e64d
AS		 2246 ret = replace_map_fd_with_map_ptr(env);
2247 if (ret < 0)
2248 goto skip_full_check;
2249
9bac3d6d 2250 env->explored_states = kcalloc(env->prog->len,
f1bca824
AS		 2251 sizeof(struct verifier_state_list *),
2252 GFP_USER);
2253 ret = -ENOMEM;
2254 if (!env->explored_states)
2255 goto skip_full_check;
2256
475fb78f
AS		 2257 ret = check_cfg(env);
2258 if (ret < 0)
2259 goto skip_full_check;
2260
1be7f75d
AS		 2261 env->allow_ptr_leaks = capable(CAP_SYS_ADMIN);
2262
17a52670 2263 ret = do_check(env);
cbd35700 2264
0246e64d 2265skip_full_check:
17a52670 2266 while (pop_stack(env, NULL) >= 0);
f1bca824 2267 free_states(env);
0246e64d 2268
9bac3d6d
AS		 2269 if (ret == 0)
2270 /* program is valid, convert *(u32*)(ctx + off) accesses */
2271 ret = convert_ctx_accesses(env);
2272
cbd35700
AS		 2273 if (log_level && log_len >= log_size - 1) {
2274 BUG_ON(log_len >= log_size);
2275 /* verifier log exceeded user supplied buffer */
2276 ret = -ENOSPC;
2277 /* fall through to return what was recorded */
2278 }
2279
2280 /* copy verifier log back to user space including trailing zero */
2281 if (log_level && copy_to_user(log_ubuf, log_buf, log_len + 1) != 0) {
2282 ret = -EFAULT;
2283 goto free_log_buf;
2284 }
2285
0246e64d
AS		 2286 if (ret == 0 && env->used_map_cnt) {
2287 /* if program passed verifier, update used_maps in bpf_prog_info */
9bac3d6d
AS		 2288 env->prog->aux->used_maps = kmalloc_array(env->used_map_cnt,
2289 sizeof(env->used_maps[0]),
2290 GFP_KERNEL);
0246e64d 2291
9bac3d6d 2292 if (!env->prog->aux->used_maps) {
0246e64d
AS		 2293 ret = -ENOMEM;
2294 goto free_log_buf;
2295 }
2296
9bac3d6d 2297 memcpy(env->prog->aux->used_maps, env->used_maps,
0246e64d 2298 sizeof(env->used_maps[0]) * env->used_map_cnt);
9bac3d6d 2299 env->prog->aux->used_map_cnt = env->used_map_cnt;
0246e64d
AS		 2300
2301 /* program is valid. Convert pseudo bpf_ld_imm64 into generic
2302 * bpf_ld_imm64 instructions
2303 */
2304 convert_pseudo_ld_imm64(env);
2305 }
cbd35700
AS		 2306
2307free_log_buf:
2308 if (log_level)
2309 vfree(log_buf);
2310free_env:
9bac3d6d 2311 if (!env->prog->aux->used_maps)
0246e64d
AS		 2312 /* if we didn't copy map pointers into bpf_prog_info, release
2313 * them now. Otherwise free_bpf_prog_info() will release them.
2314 */
2315 release_maps(env);
9bac3d6d 2316 *prog = env->prog;
cbd35700
AS		 2317 kfree(env);
2318 mutex_unlock(&bpf_verifier_lock);
51580e79
AS		 2319 return ret;
2320}