• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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,      /* 1st byte of register spilled into stack */
157 	STACK_SPILL_PART, /* other 7 bytes of register spill */
158 	STACK_MISC	  /* BPF program wrote some data into this slot */
159 };
160 
161 struct bpf_stack_slot {
162 	enum bpf_stack_slot_type stype;
163 	struct reg_state reg_st;
164 };
165 
166 /* state of the program:
167  * type of all registers and stack info
168  */
169 struct verifier_state {
170 	struct reg_state regs[MAX_BPF_REG];
171 	struct bpf_stack_slot stack[MAX_BPF_STACK];
172 };
173 
174 /* linked list of verifier states used to prune search */
175 struct verifier_state_list {
176 	struct verifier_state state;
177 	struct verifier_state_list *next;
178 };
179 
180 /* verifier_state + insn_idx are pushed to stack when branch is encountered */
181 struct verifier_stack_elem {
182 	/* verifer state is 'st'
183 	 * before processing instruction 'insn_idx'
184 	 * and after processing instruction 'prev_insn_idx'
185 	 */
186 	struct verifier_state st;
187 	int insn_idx;
188 	int prev_insn_idx;
189 	struct verifier_stack_elem *next;
190 };
191 
192 #define MAX_USED_MAPS 64 /* max number of maps accessed by one eBPF program */
193 
194 /* single container for all structs
195  * one verifier_env per bpf_check() call
196  */
197 struct verifier_env {
198 	struct bpf_prog *prog;		/* eBPF program being verified */
199 	struct verifier_stack_elem *head; /* stack of verifier states to be processed */
200 	int stack_size;			/* number of states to be processed */
201 	struct verifier_state cur_state; /* current verifier state */
202 	struct verifier_state_list **explored_states; /* search pruning optimization */
203 	struct bpf_map *used_maps[MAX_USED_MAPS]; /* array of map's used by eBPF program */
204 	u32 used_map_cnt;		/* number of used maps */
205 };
206 
207 /* verbose verifier prints what it's seeing
208  * bpf_check() is called under lock, so no race to access these global vars
209  */
210 static u32 log_level, log_size, log_len;
211 static char *log_buf;
212 
213 static DEFINE_MUTEX(bpf_verifier_lock);
214 
215 /* log_level controls verbosity level of eBPF verifier.
216  * verbose() is used to dump the verification trace to the log, so the user
217  * can figure out what's wrong with the program
218  */
verbose(const char * fmt,...)219 static void verbose(const char *fmt, ...)
220 {
221 	va_list args;
222 
223 	if (log_level == 0 || log_len >= log_size - 1)
224 		return;
225 
226 	va_start(args, fmt);
227 	log_len += vscnprintf(log_buf + log_len, log_size - log_len, fmt, args);
228 	va_end(args);
229 }
230 
231 /* string representation of 'enum bpf_reg_type' */
232 static const char * const reg_type_str[] = {
233 	[NOT_INIT]		= "?",
234 	[UNKNOWN_VALUE]		= "inv",
235 	[PTR_TO_CTX]		= "ctx",
236 	[CONST_PTR_TO_MAP]	= "map_ptr",
237 	[PTR_TO_MAP_VALUE]	= "map_value",
238 	[PTR_TO_MAP_VALUE_OR_NULL] = "map_value_or_null",
239 	[FRAME_PTR]		= "fp",
240 	[PTR_TO_STACK]		= "fp",
241 	[CONST_IMM]		= "imm",
242 };
243 
print_verifier_state(struct verifier_env * env)244 static void print_verifier_state(struct verifier_env *env)
245 {
246 	enum bpf_reg_type t;
247 	int i;
248 
249 	for (i = 0; i < MAX_BPF_REG; i++) {
250 		t = env->cur_state.regs[i].type;
251 		if (t == NOT_INIT)
252 			continue;
253 		verbose(" R%d=%s", i, reg_type_str[t]);
254 		if (t == CONST_IMM || t == PTR_TO_STACK)
255 			verbose("%d", env->cur_state.regs[i].imm);
256 		else if (t == CONST_PTR_TO_MAP || t == PTR_TO_MAP_VALUE ||
257 			 t == PTR_TO_MAP_VALUE_OR_NULL)
258 			verbose("(ks=%d,vs=%d)",
259 				env->cur_state.regs[i].map_ptr->key_size,
260 				env->cur_state.regs[i].map_ptr->value_size);
261 	}
262 	for (i = 0; i < MAX_BPF_STACK; i++) {
263 		if (env->cur_state.stack[i].stype == STACK_SPILL)
264 			verbose(" fp%d=%s", -MAX_BPF_STACK + i,
265 				reg_type_str[env->cur_state.stack[i].reg_st.type]);
266 	}
267 	verbose("\n");
268 }
269 
270 static const char *const bpf_class_string[] = {
271 	[BPF_LD]    = "ld",
272 	[BPF_LDX]   = "ldx",
273 	[BPF_ST]    = "st",
274 	[BPF_STX]   = "stx",
275 	[BPF_ALU]   = "alu",
276 	[BPF_JMP]   = "jmp",
277 	[BPF_RET]   = "BUG",
278 	[BPF_ALU64] = "alu64",
279 };
280 
281 static const char *const bpf_alu_string[] = {
282 	[BPF_ADD >> 4]  = "+=",
283 	[BPF_SUB >> 4]  = "-=",
284 	[BPF_MUL >> 4]  = "*=",
285 	[BPF_DIV >> 4]  = "/=",
286 	[BPF_OR  >> 4]  = "|=",
287 	[BPF_AND >> 4]  = "&=",
288 	[BPF_LSH >> 4]  = "<<=",
289 	[BPF_RSH >> 4]  = ">>=",
290 	[BPF_NEG >> 4]  = "neg",
291 	[BPF_MOD >> 4]  = "%=",
292 	[BPF_XOR >> 4]  = "^=",
293 	[BPF_MOV >> 4]  = "=",
294 	[BPF_ARSH >> 4] = "s>>=",
295 	[BPF_END >> 4]  = "endian",
296 };
297 
298 static const char *const bpf_ldst_string[] = {
299 	[BPF_W >> 3]  = "u32",
300 	[BPF_H >> 3]  = "u16",
301 	[BPF_B >> 3]  = "u8",
302 	[BPF_DW >> 3] = "u64",
303 };
304 
305 static const char *const bpf_jmp_string[] = {
306 	[BPF_JA >> 4]   = "jmp",
307 	[BPF_JEQ >> 4]  = "==",
308 	[BPF_JGT >> 4]  = ">",
309 	[BPF_JGE >> 4]  = ">=",
310 	[BPF_JSET >> 4] = "&",
311 	[BPF_JNE >> 4]  = "!=",
312 	[BPF_JSGT >> 4] = "s>",
313 	[BPF_JSGE >> 4] = "s>=",
314 	[BPF_CALL >> 4] = "call",
315 	[BPF_EXIT >> 4] = "exit",
316 };
317 
print_bpf_insn(struct bpf_insn * insn)318 static void print_bpf_insn(struct bpf_insn *insn)
319 {
320 	u8 class = BPF_CLASS(insn->code);
321 
322 	if (class == BPF_ALU || class == BPF_ALU64) {
323 		if (BPF_SRC(insn->code) == BPF_X)
324 			verbose("(%02x) %sr%d %s %sr%d\n",
325 				insn->code, class == BPF_ALU ? "(u32) " : "",
326 				insn->dst_reg,
327 				bpf_alu_string[BPF_OP(insn->code) >> 4],
328 				class == BPF_ALU ? "(u32) " : "",
329 				insn->src_reg);
330 		else
331 			verbose("(%02x) %sr%d %s %s%d\n",
332 				insn->code, class == BPF_ALU ? "(u32) " : "",
333 				insn->dst_reg,
334 				bpf_alu_string[BPF_OP(insn->code) >> 4],
335 				class == BPF_ALU ? "(u32) " : "",
336 				insn->imm);
337 	} else if (class == BPF_STX) {
338 		if (BPF_MODE(insn->code) == BPF_MEM)
339 			verbose("(%02x) *(%s *)(r%d %+d) = r%d\n",
340 				insn->code,
341 				bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
342 				insn->dst_reg,
343 				insn->off, insn->src_reg);
344 		else if (BPF_MODE(insn->code) == BPF_XADD)
345 			verbose("(%02x) lock *(%s *)(r%d %+d) += r%d\n",
346 				insn->code,
347 				bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
348 				insn->dst_reg, insn->off,
349 				insn->src_reg);
350 		else
351 			verbose("BUG_%02x\n", insn->code);
352 	} else if (class == BPF_ST) {
353 		if (BPF_MODE(insn->code) != BPF_MEM) {
354 			verbose("BUG_st_%02x\n", insn->code);
355 			return;
356 		}
357 		verbose("(%02x) *(%s *)(r%d %+d) = %d\n",
358 			insn->code,
359 			bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
360 			insn->dst_reg,
361 			insn->off, insn->imm);
362 	} else if (class == BPF_LDX) {
363 		if (BPF_MODE(insn->code) != BPF_MEM) {
364 			verbose("BUG_ldx_%02x\n", insn->code);
365 			return;
366 		}
367 		verbose("(%02x) r%d = *(%s *)(r%d %+d)\n",
368 			insn->code, insn->dst_reg,
369 			bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
370 			insn->src_reg, insn->off);
371 	} else if (class == BPF_LD) {
372 		if (BPF_MODE(insn->code) == BPF_ABS) {
373 			verbose("(%02x) r0 = *(%s *)skb[%d]\n",
374 				insn->code,
375 				bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
376 				insn->imm);
377 		} else if (BPF_MODE(insn->code) == BPF_IND) {
378 			verbose("(%02x) r0 = *(%s *)skb[r%d + %d]\n",
379 				insn->code,
380 				bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
381 				insn->src_reg, insn->imm);
382 		} else if (BPF_MODE(insn->code) == BPF_IMM) {
383 			verbose("(%02x) r%d = 0x%x\n",
384 				insn->code, insn->dst_reg, insn->imm);
385 		} else {
386 			verbose("BUG_ld_%02x\n", insn->code);
387 			return;
388 		}
389 	} else if (class == BPF_JMP) {
390 		u8 opcode = BPF_OP(insn->code);
391 
392 		if (opcode == BPF_CALL) {
393 			verbose("(%02x) call %d\n", insn->code, insn->imm);
394 		} else if (insn->code == (BPF_JMP | BPF_JA)) {
395 			verbose("(%02x) goto pc%+d\n",
396 				insn->code, insn->off);
397 		} else if (insn->code == (BPF_JMP | BPF_EXIT)) {
398 			verbose("(%02x) exit\n", insn->code);
399 		} else if (BPF_SRC(insn->code) == BPF_X) {
400 			verbose("(%02x) if r%d %s r%d goto pc%+d\n",
401 				insn->code, insn->dst_reg,
402 				bpf_jmp_string[BPF_OP(insn->code) >> 4],
403 				insn->src_reg, insn->off);
404 		} else {
405 			verbose("(%02x) if r%d %s 0x%x goto pc%+d\n",
406 				insn->code, insn->dst_reg,
407 				bpf_jmp_string[BPF_OP(insn->code) >> 4],
408 				insn->imm, insn->off);
409 		}
410 	} else {
411 		verbose("(%02x) %s\n", insn->code, bpf_class_string[class]);
412 	}
413 }
414 
pop_stack(struct verifier_env * env,int * prev_insn_idx)415 static int pop_stack(struct verifier_env *env, int *prev_insn_idx)
416 {
417 	struct verifier_stack_elem *elem;
418 	int insn_idx;
419 
420 	if (env->head == NULL)
421 		return -1;
422 
423 	memcpy(&env->cur_state, &env->head->st, sizeof(env->cur_state));
424 	insn_idx = env->head->insn_idx;
425 	if (prev_insn_idx)
426 		*prev_insn_idx = env->head->prev_insn_idx;
427 	elem = env->head->next;
428 	kfree(env->head);
429 	env->head = elem;
430 	env->stack_size--;
431 	return insn_idx;
432 }
433 
push_stack(struct verifier_env * env,int insn_idx,int prev_insn_idx)434 static struct verifier_state *push_stack(struct verifier_env *env, int insn_idx,
435 					 int prev_insn_idx)
436 {
437 	struct verifier_stack_elem *elem;
438 
439 	elem = kmalloc(sizeof(struct verifier_stack_elem), GFP_KERNEL);
440 	if (!elem)
441 		goto err;
442 
443 	memcpy(&elem->st, &env->cur_state, sizeof(env->cur_state));
444 	elem->insn_idx = insn_idx;
445 	elem->prev_insn_idx = prev_insn_idx;
446 	elem->next = env->head;
447 	env->head = elem;
448 	env->stack_size++;
449 	if (env->stack_size > 1024) {
450 		verbose("BPF program is too complex\n");
451 		goto err;
452 	}
453 	return &elem->st;
454 err:
455 	/* pop all elements and return */
456 	while (pop_stack(env, NULL) >= 0);
457 	return NULL;
458 }
459 
460 #define CALLER_SAVED_REGS 6
461 static const int caller_saved[CALLER_SAVED_REGS] = {
462 	BPF_REG_0, BPF_REG_1, BPF_REG_2, BPF_REG_3, BPF_REG_4, BPF_REG_5
463 };
464 
init_reg_state(struct reg_state * regs)465 static void init_reg_state(struct reg_state *regs)
466 {
467 	int i;
468 
469 	for (i = 0; i < MAX_BPF_REG; i++) {
470 		regs[i].type = NOT_INIT;
471 		regs[i].imm = 0;
472 		regs[i].map_ptr = NULL;
473 	}
474 
475 	/* frame pointer */
476 	regs[BPF_REG_FP].type = FRAME_PTR;
477 
478 	/* 1st arg to a function */
479 	regs[BPF_REG_1].type = PTR_TO_CTX;
480 }
481 
mark_reg_unknown_value(struct reg_state * regs,u32 regno)482 static void mark_reg_unknown_value(struct reg_state *regs, u32 regno)
483 {
484 	BUG_ON(regno >= MAX_BPF_REG);
485 	regs[regno].type = UNKNOWN_VALUE;
486 	regs[regno].imm = 0;
487 	regs[regno].map_ptr = NULL;
488 }
489 
490 enum reg_arg_type {
491 	SRC_OP,		/* register is used as source operand */
492 	DST_OP,		/* register is used as destination operand */
493 	DST_OP_NO_MARK	/* same as above, check only, don't mark */
494 };
495 
check_reg_arg(struct reg_state * regs,u32 regno,enum reg_arg_type t)496 static int check_reg_arg(struct reg_state *regs, u32 regno,
497 			 enum reg_arg_type t)
498 {
499 	if (regno >= MAX_BPF_REG) {
500 		verbose("R%d is invalid\n", regno);
501 		return -EINVAL;
502 	}
503 
504 	if (t == SRC_OP) {
505 		/* check whether register used as source operand can be read */
506 		if (regs[regno].type == NOT_INIT) {
507 			verbose("R%d !read_ok\n", regno);
508 			return -EACCES;
509 		}
510 	} else {
511 		/* check whether register used as dest operand can be written to */
512 		if (regno == BPF_REG_FP) {
513 			verbose("frame pointer is read only\n");
514 			return -EACCES;
515 		}
516 		if (t == DST_OP)
517 			mark_reg_unknown_value(regs, regno);
518 	}
519 	return 0;
520 }
521 
bpf_size_to_bytes(int bpf_size)522 static int bpf_size_to_bytes(int bpf_size)
523 {
524 	if (bpf_size == BPF_W)
525 		return 4;
526 	else if (bpf_size == BPF_H)
527 		return 2;
528 	else if (bpf_size == BPF_B)
529 		return 1;
530 	else if (bpf_size == BPF_DW)
531 		return 8;
532 	else
533 		return -EINVAL;
534 }
535 
536 /* check_stack_read/write functions track spill/fill of registers,
537  * stack boundary and alignment are checked in check_mem_access()
538  */
check_stack_write(struct verifier_state * state,int off,int size,int value_regno)539 static int check_stack_write(struct verifier_state *state, int off, int size,
540 			     int value_regno)
541 {
542 	struct bpf_stack_slot *slot;
543 	int i;
544 
545 	if (value_regno >= 0 &&
546 	    (state->regs[value_regno].type == PTR_TO_MAP_VALUE ||
547 	     state->regs[value_regno].type == PTR_TO_STACK ||
548 	     state->regs[value_regno].type == PTR_TO_CTX)) {
549 
550 		/* register containing pointer is being spilled into stack */
551 		if (size != 8) {
552 			verbose("invalid size of register spill\n");
553 			return -EACCES;
554 		}
555 
556 		slot = &state->stack[MAX_BPF_STACK + off];
557 		slot->stype = STACK_SPILL;
558 		/* save register state */
559 		slot->reg_st = state->regs[value_regno];
560 		for (i = 1; i < 8; i++) {
561 			slot = &state->stack[MAX_BPF_STACK + off + i];
562 			slot->stype = STACK_SPILL_PART;
563 			slot->reg_st.type = UNKNOWN_VALUE;
564 			slot->reg_st.map_ptr = NULL;
565 		}
566 	} else {
567 
568 		/* regular write of data into stack */
569 		for (i = 0; i < size; i++) {
570 			slot = &state->stack[MAX_BPF_STACK + off + i];
571 			slot->stype = STACK_MISC;
572 			slot->reg_st.type = UNKNOWN_VALUE;
573 			slot->reg_st.map_ptr = NULL;
574 		}
575 	}
576 	return 0;
577 }
578 
check_stack_read(struct verifier_state * state,int off,int size,int value_regno)579 static int check_stack_read(struct verifier_state *state, int off, int size,
580 			    int value_regno)
581 {
582 	int i;
583 	struct bpf_stack_slot *slot;
584 
585 	slot = &state->stack[MAX_BPF_STACK + off];
586 
587 	if (slot->stype == STACK_SPILL) {
588 		if (size != 8) {
589 			verbose("invalid size of register spill\n");
590 			return -EACCES;
591 		}
592 		for (i = 1; i < 8; i++) {
593 			if (state->stack[MAX_BPF_STACK + off + i].stype !=
594 			    STACK_SPILL_PART) {
595 				verbose("corrupted spill memory\n");
596 				return -EACCES;
597 			}
598 		}
599 
600 		if (value_regno >= 0)
601 			/* restore register state from stack */
602 			state->regs[value_regno] = slot->reg_st;
603 		return 0;
604 	} else {
605 		for (i = 0; i < size; i++) {
606 			if (state->stack[MAX_BPF_STACK + off + i].stype !=
607 			    STACK_MISC) {
608 				verbose("invalid read from stack off %d+%d size %d\n",
609 					off, i, size);
610 				return -EACCES;
611 			}
612 		}
613 		if (value_regno >= 0)
614 			/* have read misc data from the stack */
615 			mark_reg_unknown_value(state->regs, value_regno);
616 		return 0;
617 	}
618 }
619 
620 /* check read/write into map element returned by bpf_map_lookup_elem() */
check_map_access(struct verifier_env * env,u32 regno,int off,int size)621 static int check_map_access(struct verifier_env *env, u32 regno, int off,
622 			    int size)
623 {
624 	struct bpf_map *map = env->cur_state.regs[regno].map_ptr;
625 
626 	if (off < 0 || off + size > map->value_size) {
627 		verbose("invalid access to map value, value_size=%d off=%d size=%d\n",
628 			map->value_size, off, size);
629 		return -EACCES;
630 	}
631 	return 0;
632 }
633 
634 /* check access to 'struct bpf_context' fields */
check_ctx_access(struct verifier_env * env,int off,int size,enum bpf_access_type t)635 static int check_ctx_access(struct verifier_env *env, int off, int size,
636 			    enum bpf_access_type t)
637 {
638 	if (env->prog->aux->ops->is_valid_access &&
639 	    env->prog->aux->ops->is_valid_access(off, size, t))
640 		return 0;
641 
642 	verbose("invalid bpf_context access off=%d size=%d\n", off, size);
643 	return -EACCES;
644 }
645 
646 /* check whether memory at (regno + off) is accessible for t = (read | write)
647  * if t==write, value_regno is a register which value is stored into memory
648  * if t==read, value_regno is a register which will receive the value from memory
649  * if t==write && value_regno==-1, some unknown value is stored into memory
650  * if t==read && value_regno==-1, don't care what we read from memory
651  */
check_mem_access(struct verifier_env * env,u32 regno,int off,int bpf_size,enum bpf_access_type t,int value_regno)652 static int check_mem_access(struct verifier_env *env, u32 regno, int off,
653 			    int bpf_size, enum bpf_access_type t,
654 			    int value_regno)
655 {
656 	struct verifier_state *state = &env->cur_state;
657 	int size, err = 0;
658 
659 	size = bpf_size_to_bytes(bpf_size);
660 	if (size < 0)
661 		return size;
662 
663 	if (off % size != 0) {
664 		verbose("misaligned access off %d size %d\n", off, size);
665 		return -EACCES;
666 	}
667 
668 	if (state->regs[regno].type == PTR_TO_MAP_VALUE) {
669 		err = check_map_access(env, regno, off, size);
670 		if (!err && t == BPF_READ && value_regno >= 0)
671 			mark_reg_unknown_value(state->regs, value_regno);
672 
673 	} else if (state->regs[regno].type == PTR_TO_CTX) {
674 		err = check_ctx_access(env, off, size, t);
675 		if (!err && t == BPF_READ && value_regno >= 0)
676 			mark_reg_unknown_value(state->regs, value_regno);
677 
678 	} else if (state->regs[regno].type == FRAME_PTR) {
679 		if (off >= 0 || off < -MAX_BPF_STACK) {
680 			verbose("invalid stack off=%d size=%d\n", off, size);
681 			return -EACCES;
682 		}
683 		if (t == BPF_WRITE)
684 			err = check_stack_write(state, off, size, value_regno);
685 		else
686 			err = check_stack_read(state, off, size, value_regno);
687 	} else {
688 		verbose("R%d invalid mem access '%s'\n",
689 			regno, reg_type_str[state->regs[regno].type]);
690 		return -EACCES;
691 	}
692 	return err;
693 }
694 
check_xadd(struct verifier_env * env,struct bpf_insn * insn)695 static int check_xadd(struct verifier_env *env, struct bpf_insn *insn)
696 {
697 	struct reg_state *regs = env->cur_state.regs;
698 	int err;
699 
700 	if ((BPF_SIZE(insn->code) != BPF_W && BPF_SIZE(insn->code) != BPF_DW) ||
701 	    insn->imm != 0) {
702 		verbose("BPF_XADD uses reserved fields\n");
703 		return -EINVAL;
704 	}
705 
706 	/* check src1 operand */
707 	err = check_reg_arg(regs, insn->src_reg, SRC_OP);
708 	if (err)
709 		return err;
710 
711 	/* check src2 operand */
712 	err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
713 	if (err)
714 		return err;
715 
716 	/* check whether atomic_add can read the memory */
717 	err = check_mem_access(env, insn->dst_reg, insn->off,
718 			       BPF_SIZE(insn->code), BPF_READ, -1);
719 	if (err)
720 		return err;
721 
722 	/* check whether atomic_add can write into the same memory */
723 	return check_mem_access(env, insn->dst_reg, insn->off,
724 				BPF_SIZE(insn->code), BPF_WRITE, -1);
725 }
726 
727 /* when register 'regno' is passed into function that will read 'access_size'
728  * bytes from that pointer, make sure that it's within stack boundary
729  * and all elements of stack are initialized
730  */
check_stack_boundary(struct verifier_env * env,int regno,int access_size)731 static int check_stack_boundary(struct verifier_env *env,
732 				int regno, int access_size)
733 {
734 	struct verifier_state *state = &env->cur_state;
735 	struct reg_state *regs = state->regs;
736 	int off, i;
737 
738 	if (regs[regno].type != PTR_TO_STACK)
739 		return -EACCES;
740 
741 	off = regs[regno].imm;
742 	if (off >= 0 || off < -MAX_BPF_STACK || off + access_size > 0 ||
743 	    access_size <= 0) {
744 		verbose("invalid stack type R%d off=%d access_size=%d\n",
745 			regno, off, access_size);
746 		return -EACCES;
747 	}
748 
749 	for (i = 0; i < access_size; i++) {
750 		if (state->stack[MAX_BPF_STACK + off + i].stype != STACK_MISC) {
751 			verbose("invalid indirect read from stack off %d+%d size %d\n",
752 				off, i, access_size);
753 			return -EACCES;
754 		}
755 	}
756 	return 0;
757 }
758 
check_func_arg(struct verifier_env * env,u32 regno,enum bpf_arg_type arg_type,struct bpf_map ** mapp)759 static int check_func_arg(struct verifier_env *env, u32 regno,
760 			  enum bpf_arg_type arg_type, struct bpf_map **mapp)
761 {
762 	struct reg_state *reg = env->cur_state.regs + regno;
763 	enum bpf_reg_type expected_type;
764 	int err = 0;
765 
766 	if (arg_type == ARG_DONTCARE)
767 		return 0;
768 
769 	if (reg->type == NOT_INIT) {
770 		verbose("R%d !read_ok\n", regno);
771 		return -EACCES;
772 	}
773 
774 	if (arg_type == ARG_ANYTHING)
775 		return 0;
776 
777 	if (arg_type == ARG_PTR_TO_STACK || arg_type == ARG_PTR_TO_MAP_KEY ||
778 	    arg_type == ARG_PTR_TO_MAP_VALUE) {
779 		expected_type = PTR_TO_STACK;
780 	} else if (arg_type == ARG_CONST_STACK_SIZE) {
781 		expected_type = CONST_IMM;
782 	} else if (arg_type == ARG_CONST_MAP_PTR) {
783 		expected_type = CONST_PTR_TO_MAP;
784 	} else {
785 		verbose("unsupported arg_type %d\n", arg_type);
786 		return -EFAULT;
787 	}
788 
789 	if (reg->type != expected_type) {
790 		verbose("R%d type=%s expected=%s\n", regno,
791 			reg_type_str[reg->type], reg_type_str[expected_type]);
792 		return -EACCES;
793 	}
794 
795 	if (arg_type == ARG_CONST_MAP_PTR) {
796 		/* bpf_map_xxx(map_ptr) call: remember that map_ptr */
797 		*mapp = reg->map_ptr;
798 
799 	} else if (arg_type == ARG_PTR_TO_MAP_KEY) {
800 		/* bpf_map_xxx(..., map_ptr, ..., key) call:
801 		 * check that [key, key + map->key_size) are within
802 		 * stack limits and initialized
803 		 */
804 		if (!*mapp) {
805 			/* in function declaration map_ptr must come before
806 			 * map_key, so that it's verified and known before
807 			 * we have to check map_key here. Otherwise it means
808 			 * that kernel subsystem misconfigured verifier
809 			 */
810 			verbose("invalid map_ptr to access map->key\n");
811 			return -EACCES;
812 		}
813 		err = check_stack_boundary(env, regno, (*mapp)->key_size);
814 
815 	} else if (arg_type == ARG_PTR_TO_MAP_VALUE) {
816 		/* bpf_map_xxx(..., map_ptr, ..., value) call:
817 		 * check [value, value + map->value_size) validity
818 		 */
819 		if (!*mapp) {
820 			/* kernel subsystem misconfigured verifier */
821 			verbose("invalid map_ptr to access map->value\n");
822 			return -EACCES;
823 		}
824 		err = check_stack_boundary(env, regno, (*mapp)->value_size);
825 
826 	} else if (arg_type == ARG_CONST_STACK_SIZE) {
827 		/* bpf_xxx(..., buf, len) call will access 'len' bytes
828 		 * from stack pointer 'buf'. Check it
829 		 * note: regno == len, regno - 1 == buf
830 		 */
831 		if (regno == 0) {
832 			/* kernel subsystem misconfigured verifier */
833 			verbose("ARG_CONST_STACK_SIZE cannot be first argument\n");
834 			return -EACCES;
835 		}
836 		err = check_stack_boundary(env, regno - 1, reg->imm);
837 	}
838 
839 	return err;
840 }
841 
check_call(struct verifier_env * env,int func_id)842 static int check_call(struct verifier_env *env, int func_id)
843 {
844 	struct verifier_state *state = &env->cur_state;
845 	const struct bpf_func_proto *fn = NULL;
846 	struct reg_state *regs = state->regs;
847 	struct bpf_map *map = NULL;
848 	struct reg_state *reg;
849 	int i, err;
850 
851 	/* find function prototype */
852 	if (func_id < 0 || func_id >= __BPF_FUNC_MAX_ID) {
853 		verbose("invalid func %d\n", func_id);
854 		return -EINVAL;
855 	}
856 
857 	if (env->prog->aux->ops->get_func_proto)
858 		fn = env->prog->aux->ops->get_func_proto(func_id);
859 
860 	if (!fn) {
861 		verbose("unknown func %d\n", func_id);
862 		return -EINVAL;
863 	}
864 
865 	/* eBPF programs must be GPL compatible to use GPL-ed functions */
866 	if (!env->prog->aux->is_gpl_compatible && fn->gpl_only) {
867 		verbose("cannot call GPL only function from proprietary program\n");
868 		return -EINVAL;
869 	}
870 
871 	/* check args */
872 	err = check_func_arg(env, BPF_REG_1, fn->arg1_type, &map);
873 	if (err)
874 		return err;
875 	err = check_func_arg(env, BPF_REG_2, fn->arg2_type, &map);
876 	if (err)
877 		return err;
878 	err = check_func_arg(env, BPF_REG_3, fn->arg3_type, &map);
879 	if (err)
880 		return err;
881 	err = check_func_arg(env, BPF_REG_4, fn->arg4_type, &map);
882 	if (err)
883 		return err;
884 	err = check_func_arg(env, BPF_REG_5, fn->arg5_type, &map);
885 	if (err)
886 		return err;
887 
888 	/* reset caller saved regs */
889 	for (i = 0; i < CALLER_SAVED_REGS; i++) {
890 		reg = regs + caller_saved[i];
891 		reg->type = NOT_INIT;
892 		reg->imm = 0;
893 	}
894 
895 	/* update return register */
896 	if (fn->ret_type == RET_INTEGER) {
897 		regs[BPF_REG_0].type = UNKNOWN_VALUE;
898 	} else if (fn->ret_type == RET_VOID) {
899 		regs[BPF_REG_0].type = NOT_INIT;
900 	} else if (fn->ret_type == RET_PTR_TO_MAP_VALUE_OR_NULL) {
901 		regs[BPF_REG_0].type = PTR_TO_MAP_VALUE_OR_NULL;
902 		/* remember map_ptr, so that check_map_access()
903 		 * can check 'value_size' boundary of memory access
904 		 * to map element returned from bpf_map_lookup_elem()
905 		 */
906 		if (map == NULL) {
907 			verbose("kernel subsystem misconfigured verifier\n");
908 			return -EINVAL;
909 		}
910 		regs[BPF_REG_0].map_ptr = map;
911 	} else {
912 		verbose("unknown return type %d of func %d\n",
913 			fn->ret_type, func_id);
914 		return -EINVAL;
915 	}
916 	return 0;
917 }
918 
919 /* check validity of 32-bit and 64-bit arithmetic operations */
check_alu_op(struct reg_state * regs,struct bpf_insn * insn)920 static int check_alu_op(struct reg_state *regs, struct bpf_insn *insn)
921 {
922 	u8 opcode = BPF_OP(insn->code);
923 	int err;
924 
925 	if (opcode == BPF_END || opcode == BPF_NEG) {
926 		if (opcode == BPF_NEG) {
927 			if (BPF_SRC(insn->code) != 0 ||
928 			    insn->src_reg != BPF_REG_0 ||
929 			    insn->off != 0 || insn->imm != 0) {
930 				verbose("BPF_NEG uses reserved fields\n");
931 				return -EINVAL;
932 			}
933 		} else {
934 			if (insn->src_reg != BPF_REG_0 || insn->off != 0 ||
935 			    (insn->imm != 16 && insn->imm != 32 && insn->imm != 64) ||
936 			    BPF_CLASS(insn->code) == BPF_ALU64) {
937 				verbose("BPF_END uses reserved fields\n");
938 				return -EINVAL;
939 			}
940 		}
941 
942 		/* check src operand */
943 		err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
944 		if (err)
945 			return err;
946 
947 		/* check dest operand */
948 		err = check_reg_arg(regs, insn->dst_reg, DST_OP);
949 		if (err)
950 			return err;
951 
952 	} else if (opcode == BPF_MOV) {
953 
954 		if (BPF_SRC(insn->code) == BPF_X) {
955 			if (insn->imm != 0 || insn->off != 0) {
956 				verbose("BPF_MOV uses reserved fields\n");
957 				return -EINVAL;
958 			}
959 
960 			/* check src operand */
961 			err = check_reg_arg(regs, insn->src_reg, SRC_OP);
962 			if (err)
963 				return err;
964 		} else {
965 			if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
966 				verbose("BPF_MOV uses reserved fields\n");
967 				return -EINVAL;
968 			}
969 		}
970 
971 		/* check dest operand */
972 		err = check_reg_arg(regs, insn->dst_reg, DST_OP);
973 		if (err)
974 			return err;
975 
976 		if (BPF_SRC(insn->code) == BPF_X) {
977 			if (BPF_CLASS(insn->code) == BPF_ALU64) {
978 				/* case: R1 = R2
979 				 * copy register state to dest reg
980 				 */
981 				regs[insn->dst_reg] = regs[insn->src_reg];
982 			} else {
983 				regs[insn->dst_reg].type = UNKNOWN_VALUE;
984 				regs[insn->dst_reg].map_ptr = NULL;
985 			}
986 		} else {
987 			/* case: R = imm
988 			 * remember the value we stored into this reg
989 			 */
990 			regs[insn->dst_reg].type = CONST_IMM;
991 			regs[insn->dst_reg].imm = insn->imm;
992 		}
993 
994 	} else if (opcode > BPF_END) {
995 		verbose("invalid BPF_ALU opcode %x\n", opcode);
996 		return -EINVAL;
997 
998 	} else {	/* all other ALU ops: and, sub, xor, add, ... */
999 
1000 		bool stack_relative = false;
1001 
1002 		if (BPF_SRC(insn->code) == BPF_X) {
1003 			if (insn->imm != 0 || insn->off != 0) {
1004 				verbose("BPF_ALU uses reserved fields\n");
1005 				return -EINVAL;
1006 			}
1007 			/* check src1 operand */
1008 			err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1009 			if (err)
1010 				return err;
1011 		} else {
1012 			if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
1013 				verbose("BPF_ALU uses reserved fields\n");
1014 				return -EINVAL;
1015 			}
1016 		}
1017 
1018 		/* check src2 operand */
1019 		err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1020 		if (err)
1021 			return err;
1022 
1023 		if ((opcode == BPF_MOD || opcode == BPF_DIV) &&
1024 		    BPF_SRC(insn->code) == BPF_K && insn->imm == 0) {
1025 			verbose("div by zero\n");
1026 			return -EINVAL;
1027 		}
1028 
1029 		if ((opcode == BPF_LSH || opcode == BPF_RSH ||
1030 		     opcode == BPF_ARSH) && BPF_SRC(insn->code) == BPF_K) {
1031 			int size = BPF_CLASS(insn->code) == BPF_ALU64 ? 64 : 32;
1032 
1033 			if (insn->imm < 0 || insn->imm >= size) {
1034 				verbose("invalid shift %d\n", insn->imm);
1035 				return -EINVAL;
1036 			}
1037 		}
1038 
1039 		/* pattern match 'bpf_add Rx, imm' instruction */
1040 		if (opcode == BPF_ADD && BPF_CLASS(insn->code) == BPF_ALU64 &&
1041 		    regs[insn->dst_reg].type == FRAME_PTR &&
1042 		    BPF_SRC(insn->code) == BPF_K)
1043 			stack_relative = true;
1044 
1045 		/* check dest operand */
1046 		err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1047 		if (err)
1048 			return err;
1049 
1050 		if (stack_relative) {
1051 			regs[insn->dst_reg].type = PTR_TO_STACK;
1052 			regs[insn->dst_reg].imm = insn->imm;
1053 		}
1054 	}
1055 
1056 	return 0;
1057 }
1058 
check_cond_jmp_op(struct verifier_env * env,struct bpf_insn * insn,int * insn_idx)1059 static int check_cond_jmp_op(struct verifier_env *env,
1060 			     struct bpf_insn *insn, int *insn_idx)
1061 {
1062 	struct reg_state *regs = env->cur_state.regs;
1063 	struct verifier_state *other_branch;
1064 	u8 opcode = BPF_OP(insn->code);
1065 	int err;
1066 
1067 	if (opcode > BPF_EXIT) {
1068 		verbose("invalid BPF_JMP opcode %x\n", opcode);
1069 		return -EINVAL;
1070 	}
1071 
1072 	if (BPF_SRC(insn->code) == BPF_X) {
1073 		if (insn->imm != 0) {
1074 			verbose("BPF_JMP uses reserved fields\n");
1075 			return -EINVAL;
1076 		}
1077 
1078 		/* check src1 operand */
1079 		err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1080 		if (err)
1081 			return err;
1082 	} else {
1083 		if (insn->src_reg != BPF_REG_0) {
1084 			verbose("BPF_JMP uses reserved fields\n");
1085 			return -EINVAL;
1086 		}
1087 	}
1088 
1089 	/* check src2 operand */
1090 	err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1091 	if (err)
1092 		return err;
1093 
1094 	/* detect if R == 0 where R was initialized to zero earlier */
1095 	if (BPF_SRC(insn->code) == BPF_K &&
1096 	    (opcode == BPF_JEQ || opcode == BPF_JNE) &&
1097 	    regs[insn->dst_reg].type == CONST_IMM &&
1098 	    regs[insn->dst_reg].imm == insn->imm) {
1099 		if (opcode == BPF_JEQ) {
1100 			/* if (imm == imm) goto pc+off;
1101 			 * only follow the goto, ignore fall-through
1102 			 */
1103 			*insn_idx += insn->off;
1104 			return 0;
1105 		} else {
1106 			/* if (imm != imm) goto pc+off;
1107 			 * only follow fall-through branch, since
1108 			 * that's where the program will go
1109 			 */
1110 			return 0;
1111 		}
1112 	}
1113 
1114 	other_branch = push_stack(env, *insn_idx + insn->off + 1, *insn_idx);
1115 	if (!other_branch)
1116 		return -EFAULT;
1117 
1118 	/* detect if R == 0 where R is returned value from bpf_map_lookup_elem() */
1119 	if (BPF_SRC(insn->code) == BPF_K &&
1120 	    insn->imm == 0 && (opcode == BPF_JEQ ||
1121 			       opcode == BPF_JNE) &&
1122 	    regs[insn->dst_reg].type == PTR_TO_MAP_VALUE_OR_NULL) {
1123 		if (opcode == BPF_JEQ) {
1124 			/* next fallthrough insn can access memory via
1125 			 * this register
1126 			 */
1127 			regs[insn->dst_reg].type = PTR_TO_MAP_VALUE;
1128 			/* branch targer cannot access it, since reg == 0 */
1129 			other_branch->regs[insn->dst_reg].type = CONST_IMM;
1130 			other_branch->regs[insn->dst_reg].imm = 0;
1131 		} else {
1132 			other_branch->regs[insn->dst_reg].type = PTR_TO_MAP_VALUE;
1133 			regs[insn->dst_reg].type = CONST_IMM;
1134 			regs[insn->dst_reg].imm = 0;
1135 		}
1136 	} else if (BPF_SRC(insn->code) == BPF_K &&
1137 		   (opcode == BPF_JEQ || opcode == BPF_JNE)) {
1138 
1139 		if (opcode == BPF_JEQ) {
1140 			/* detect if (R == imm) goto
1141 			 * and in the target state recognize that R = imm
1142 			 */
1143 			other_branch->regs[insn->dst_reg].type = CONST_IMM;
1144 			other_branch->regs[insn->dst_reg].imm = insn->imm;
1145 		} else {
1146 			/* detect if (R != imm) goto
1147 			 * and in the fall-through state recognize that R = imm
1148 			 */
1149 			regs[insn->dst_reg].type = CONST_IMM;
1150 			regs[insn->dst_reg].imm = insn->imm;
1151 		}
1152 	}
1153 	if (log_level)
1154 		print_verifier_state(env);
1155 	return 0;
1156 }
1157 
1158 /* return the map pointer stored inside BPF_LD_IMM64 instruction */
ld_imm64_to_map_ptr(struct bpf_insn * insn)1159 static struct bpf_map *ld_imm64_to_map_ptr(struct bpf_insn *insn)
1160 {
1161 	u64 imm64 = ((u64) (u32) insn[0].imm) | ((u64) (u32) insn[1].imm) << 32;
1162 
1163 	return (struct bpf_map *) (unsigned long) imm64;
1164 }
1165 
1166 /* verify BPF_LD_IMM64 instruction */
check_ld_imm(struct verifier_env * env,struct bpf_insn * insn)1167 static int check_ld_imm(struct verifier_env *env, struct bpf_insn *insn)
1168 {
1169 	struct reg_state *regs = env->cur_state.regs;
1170 	int err;
1171 
1172 	if (BPF_SIZE(insn->code) != BPF_DW) {
1173 		verbose("invalid BPF_LD_IMM insn\n");
1174 		return -EINVAL;
1175 	}
1176 	if (insn->off != 0) {
1177 		verbose("BPF_LD_IMM64 uses reserved fields\n");
1178 		return -EINVAL;
1179 	}
1180 
1181 	err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1182 	if (err)
1183 		return err;
1184 
1185 	if (insn->src_reg == 0)
1186 		/* generic move 64-bit immediate into a register */
1187 		return 0;
1188 
1189 	/* replace_map_fd_with_map_ptr() should have caught bad ld_imm64 */
1190 	BUG_ON(insn->src_reg != BPF_PSEUDO_MAP_FD);
1191 
1192 	regs[insn->dst_reg].type = CONST_PTR_TO_MAP;
1193 	regs[insn->dst_reg].map_ptr = ld_imm64_to_map_ptr(insn);
1194 	return 0;
1195 }
1196 
1197 /* non-recursive DFS pseudo code
1198  * 1  procedure DFS-iterative(G,v):
1199  * 2      label v as discovered
1200  * 3      let S be a stack
1201  * 4      S.push(v)
1202  * 5      while S is not empty
1203  * 6            t <- S.pop()
1204  * 7            if t is what we're looking for:
1205  * 8                return t
1206  * 9            for all edges e in G.adjacentEdges(t) do
1207  * 10               if edge e is already labelled
1208  * 11                   continue with the next edge
1209  * 12               w <- G.adjacentVertex(t,e)
1210  * 13               if vertex w is not discovered and not explored
1211  * 14                   label e as tree-edge
1212  * 15                   label w as discovered
1213  * 16                   S.push(w)
1214  * 17                   continue at 5
1215  * 18               else if vertex w is discovered
1216  * 19                   label e as back-edge
1217  * 20               else
1218  * 21                   // vertex w is explored
1219  * 22                   label e as forward- or cross-edge
1220  * 23           label t as explored
1221  * 24           S.pop()
1222  *
1223  * convention:
1224  * 0x10 - discovered
1225  * 0x11 - discovered and fall-through edge labelled
1226  * 0x12 - discovered and fall-through and branch edges labelled
1227  * 0x20 - explored
1228  */
1229 
1230 enum {
1231 	DISCOVERED = 0x10,
1232 	EXPLORED = 0x20,
1233 	FALLTHROUGH = 1,
1234 	BRANCH = 2,
1235 };
1236 
1237 #define STATE_LIST_MARK ((struct verifier_state_list *) -1L)
1238 
1239 static int *insn_stack;	/* stack of insns to process */
1240 static int cur_stack;	/* current stack index */
1241 static int *insn_state;
1242 
1243 /* t, w, e - match pseudo-code above:
1244  * t - index of current instruction
1245  * w - next instruction
1246  * e - edge
1247  */
push_insn(int t,int w,int e,struct verifier_env * env)1248 static int push_insn(int t, int w, int e, struct verifier_env *env)
1249 {
1250 	if (e == FALLTHROUGH && insn_state[t] >= (DISCOVERED | FALLTHROUGH))
1251 		return 0;
1252 
1253 	if (e == BRANCH && insn_state[t] >= (DISCOVERED | BRANCH))
1254 		return 0;
1255 
1256 	if (w < 0 || w >= env->prog->len) {
1257 		verbose("jump out of range from insn %d to %d\n", t, w);
1258 		return -EINVAL;
1259 	}
1260 
1261 	if (e == BRANCH)
1262 		/* mark branch target for state pruning */
1263 		env->explored_states[w] = STATE_LIST_MARK;
1264 
1265 	if (insn_state[w] == 0) {
1266 		/* tree-edge */
1267 		insn_state[t] = DISCOVERED | e;
1268 		insn_state[w] = DISCOVERED;
1269 		if (cur_stack >= env->prog->len)
1270 			return -E2BIG;
1271 		insn_stack[cur_stack++] = w;
1272 		return 1;
1273 	} else if ((insn_state[w] & 0xF0) == DISCOVERED) {
1274 		verbose("back-edge from insn %d to %d\n", t, w);
1275 		return -EINVAL;
1276 	} else if (insn_state[w] == EXPLORED) {
1277 		/* forward- or cross-edge */
1278 		insn_state[t] = DISCOVERED | e;
1279 	} else {
1280 		verbose("insn state internal bug\n");
1281 		return -EFAULT;
1282 	}
1283 	return 0;
1284 }
1285 
1286 /* non-recursive depth-first-search to detect loops in BPF program
1287  * loop == back-edge in directed graph
1288  */
check_cfg(struct verifier_env * env)1289 static int check_cfg(struct verifier_env *env)
1290 {
1291 	struct bpf_insn *insns = env->prog->insnsi;
1292 	int insn_cnt = env->prog->len;
1293 	int ret = 0;
1294 	int i, t;
1295 
1296 	insn_state = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
1297 	if (!insn_state)
1298 		return -ENOMEM;
1299 
1300 	insn_stack = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
1301 	if (!insn_stack) {
1302 		kfree(insn_state);
1303 		return -ENOMEM;
1304 	}
1305 
1306 	insn_state[0] = DISCOVERED; /* mark 1st insn as discovered */
1307 	insn_stack[0] = 0; /* 0 is the first instruction */
1308 	cur_stack = 1;
1309 
1310 peek_stack:
1311 	if (cur_stack == 0)
1312 		goto check_state;
1313 	t = insn_stack[cur_stack - 1];
1314 
1315 	if (BPF_CLASS(insns[t].code) == BPF_JMP) {
1316 		u8 opcode = BPF_OP(insns[t].code);
1317 
1318 		if (opcode == BPF_EXIT) {
1319 			goto mark_explored;
1320 		} else if (opcode == BPF_CALL) {
1321 			ret = push_insn(t, t + 1, FALLTHROUGH, env);
1322 			if (ret == 1)
1323 				goto peek_stack;
1324 			else if (ret < 0)
1325 				goto err_free;
1326 		} else if (opcode == BPF_JA) {
1327 			if (BPF_SRC(insns[t].code) != BPF_K) {
1328 				ret = -EINVAL;
1329 				goto err_free;
1330 			}
1331 			/* unconditional jump with single edge */
1332 			ret = push_insn(t, t + insns[t].off + 1,
1333 					FALLTHROUGH, env);
1334 			if (ret == 1)
1335 				goto peek_stack;
1336 			else if (ret < 0)
1337 				goto err_free;
1338 			/* tell verifier to check for equivalent states
1339 			 * after every call and jump
1340 			 */
1341 			if (t + 1 < insn_cnt)
1342 				env->explored_states[t + 1] = STATE_LIST_MARK;
1343 		} else {
1344 			/* conditional jump with two edges */
1345 			ret = push_insn(t, t + 1, FALLTHROUGH, env);
1346 			if (ret == 1)
1347 				goto peek_stack;
1348 			else if (ret < 0)
1349 				goto err_free;
1350 
1351 			ret = push_insn(t, t + insns[t].off + 1, BRANCH, env);
1352 			if (ret == 1)
1353 				goto peek_stack;
1354 			else if (ret < 0)
1355 				goto err_free;
1356 		}
1357 	} else {
1358 		/* all other non-branch instructions with single
1359 		 * fall-through edge
1360 		 */
1361 		ret = push_insn(t, t + 1, FALLTHROUGH, env);
1362 		if (ret == 1)
1363 			goto peek_stack;
1364 		else if (ret < 0)
1365 			goto err_free;
1366 	}
1367 
1368 mark_explored:
1369 	insn_state[t] = EXPLORED;
1370 	if (cur_stack-- <= 0) {
1371 		verbose("pop stack internal bug\n");
1372 		ret = -EFAULT;
1373 		goto err_free;
1374 	}
1375 	goto peek_stack;
1376 
1377 check_state:
1378 	for (i = 0; i < insn_cnt; i++) {
1379 		if (insn_state[i] != EXPLORED) {
1380 			verbose("unreachable insn %d\n", i);
1381 			ret = -EINVAL;
1382 			goto err_free;
1383 		}
1384 	}
1385 	ret = 0; /* cfg looks good */
1386 
1387 err_free:
1388 	kfree(insn_state);
1389 	kfree(insn_stack);
1390 	return ret;
1391 }
1392 
1393 /* compare two verifier states
1394  *
1395  * all states stored in state_list are known to be valid, since
1396  * verifier reached 'bpf_exit' instruction through them
1397  *
1398  * this function is called when verifier exploring different branches of
1399  * execution popped from the state stack. If it sees an old state that has
1400  * more strict register state and more strict stack state then this execution
1401  * branch doesn't need to be explored further, since verifier already
1402  * concluded that more strict state leads to valid finish.
1403  *
1404  * Therefore two states are equivalent if register state is more conservative
1405  * and explored stack state is more conservative than the current one.
1406  * Example:
1407  *       explored                   current
1408  * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC)
1409  * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC)
1410  *
1411  * In other words if current stack state (one being explored) has more
1412  * valid slots than old one that already passed validation, it means
1413  * the verifier can stop exploring and conclude that current state is valid too
1414  *
1415  * Similarly with registers. If explored state has register type as invalid
1416  * whereas register type in current state is meaningful, it means that
1417  * the current state will reach 'bpf_exit' instruction safely
1418  */
states_equal(struct verifier_state * old,struct verifier_state * cur)1419 static bool states_equal(struct verifier_state *old, struct verifier_state *cur)
1420 {
1421 	int i;
1422 
1423 	for (i = 0; i < MAX_BPF_REG; i++) {
1424 		if (memcmp(&old->regs[i], &cur->regs[i],
1425 			   sizeof(old->regs[0])) != 0) {
1426 			if (old->regs[i].type == NOT_INIT ||
1427 			    (old->regs[i].type == UNKNOWN_VALUE &&
1428 			     cur->regs[i].type != NOT_INIT))
1429 				continue;
1430 			return false;
1431 		}
1432 	}
1433 
1434 	for (i = 0; i < MAX_BPF_STACK; i++) {
1435 		if (memcmp(&old->stack[i], &cur->stack[i],
1436 			   sizeof(old->stack[0])) != 0) {
1437 			if (old->stack[i].stype == STACK_INVALID)
1438 				continue;
1439 			return false;
1440 		}
1441 	}
1442 	return true;
1443 }
1444 
is_state_visited(struct verifier_env * env,int insn_idx)1445 static int is_state_visited(struct verifier_env *env, int insn_idx)
1446 {
1447 	struct verifier_state_list *new_sl;
1448 	struct verifier_state_list *sl;
1449 
1450 	sl = env->explored_states[insn_idx];
1451 	if (!sl)
1452 		/* this 'insn_idx' instruction wasn't marked, so we will not
1453 		 * be doing state search here
1454 		 */
1455 		return 0;
1456 
1457 	while (sl != STATE_LIST_MARK) {
1458 		if (states_equal(&sl->state, &env->cur_state))
1459 			/* reached equivalent register/stack state,
1460 			 * prune the search
1461 			 */
1462 			return 1;
1463 		sl = sl->next;
1464 	}
1465 
1466 	/* there were no equivalent states, remember current one.
1467 	 * technically the current state is not proven to be safe yet,
1468 	 * but it will either reach bpf_exit (which means it's safe) or
1469 	 * it will be rejected. Since there are no loops, we won't be
1470 	 * seeing this 'insn_idx' instruction again on the way to bpf_exit
1471 	 */
1472 	new_sl = kmalloc(sizeof(struct verifier_state_list), GFP_USER);
1473 	if (!new_sl)
1474 		return -ENOMEM;
1475 
1476 	/* add new state to the head of linked list */
1477 	memcpy(&new_sl->state, &env->cur_state, sizeof(env->cur_state));
1478 	new_sl->next = env->explored_states[insn_idx];
1479 	env->explored_states[insn_idx] = new_sl;
1480 	return 0;
1481 }
1482 
do_check(struct verifier_env * env)1483 static int do_check(struct verifier_env *env)
1484 {
1485 	struct verifier_state *state = &env->cur_state;
1486 	struct bpf_insn *insns = env->prog->insnsi;
1487 	struct reg_state *regs = state->regs;
1488 	int insn_cnt = env->prog->len;
1489 	int insn_idx, prev_insn_idx = 0;
1490 	int insn_processed = 0;
1491 	bool do_print_state = false;
1492 
1493 	init_reg_state(regs);
1494 	insn_idx = 0;
1495 	for (;;) {
1496 		struct bpf_insn *insn;
1497 		u8 class;
1498 		int err;
1499 
1500 		if (insn_idx >= insn_cnt) {
1501 			verbose("invalid insn idx %d insn_cnt %d\n",
1502 				insn_idx, insn_cnt);
1503 			return -EFAULT;
1504 		}
1505 
1506 		insn = &insns[insn_idx];
1507 		class = BPF_CLASS(insn->code);
1508 
1509 		if (++insn_processed > 32768) {
1510 			verbose("BPF program is too large. Proccessed %d insn\n",
1511 				insn_processed);
1512 			return -E2BIG;
1513 		}
1514 
1515 		err = is_state_visited(env, insn_idx);
1516 		if (err < 0)
1517 			return err;
1518 		if (err == 1) {
1519 			/* found equivalent state, can prune the search */
1520 			if (log_level) {
1521 				if (do_print_state)
1522 					verbose("\nfrom %d to %d: safe\n",
1523 						prev_insn_idx, insn_idx);
1524 				else
1525 					verbose("%d: safe\n", insn_idx);
1526 			}
1527 			goto process_bpf_exit;
1528 		}
1529 
1530 		if (log_level && do_print_state) {
1531 			verbose("\nfrom %d to %d:", prev_insn_idx, insn_idx);
1532 			print_verifier_state(env);
1533 			do_print_state = false;
1534 		}
1535 
1536 		if (log_level) {
1537 			verbose("%d: ", insn_idx);
1538 			print_bpf_insn(insn);
1539 		}
1540 
1541 		if (class == BPF_ALU || class == BPF_ALU64) {
1542 			err = check_alu_op(regs, insn);
1543 			if (err)
1544 				return err;
1545 
1546 		} else if (class == BPF_LDX) {
1547 			if (BPF_MODE(insn->code) != BPF_MEM ||
1548 			    insn->imm != 0) {
1549 				verbose("BPF_LDX uses reserved fields\n");
1550 				return -EINVAL;
1551 			}
1552 			/* check src operand */
1553 			err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1554 			if (err)
1555 				return err;
1556 
1557 			err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK);
1558 			if (err)
1559 				return err;
1560 
1561 			/* check that memory (src_reg + off) is readable,
1562 			 * the state of dst_reg will be updated by this func
1563 			 */
1564 			err = check_mem_access(env, insn->src_reg, insn->off,
1565 					       BPF_SIZE(insn->code), BPF_READ,
1566 					       insn->dst_reg);
1567 			if (err)
1568 				return err;
1569 
1570 		} else if (class == BPF_STX) {
1571 			if (BPF_MODE(insn->code) == BPF_XADD) {
1572 				err = check_xadd(env, insn);
1573 				if (err)
1574 					return err;
1575 				insn_idx++;
1576 				continue;
1577 			}
1578 
1579 			if (BPF_MODE(insn->code) != BPF_MEM ||
1580 			    insn->imm != 0) {
1581 				verbose("BPF_STX uses reserved fields\n");
1582 				return -EINVAL;
1583 			}
1584 			/* check src1 operand */
1585 			err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1586 			if (err)
1587 				return err;
1588 			/* check src2 operand */
1589 			err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1590 			if (err)
1591 				return err;
1592 
1593 			/* check that memory (dst_reg + off) is writeable */
1594 			err = check_mem_access(env, insn->dst_reg, insn->off,
1595 					       BPF_SIZE(insn->code), BPF_WRITE,
1596 					       insn->src_reg);
1597 			if (err)
1598 				return err;
1599 
1600 		} else if (class == BPF_ST) {
1601 			if (BPF_MODE(insn->code) != BPF_MEM ||
1602 			    insn->src_reg != BPF_REG_0) {
1603 				verbose("BPF_ST uses reserved fields\n");
1604 				return -EINVAL;
1605 			}
1606 			/* check src operand */
1607 			err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1608 			if (err)
1609 				return err;
1610 
1611 			/* check that memory (dst_reg + off) is writeable */
1612 			err = check_mem_access(env, insn->dst_reg, insn->off,
1613 					       BPF_SIZE(insn->code), BPF_WRITE,
1614 					       -1);
1615 			if (err)
1616 				return err;
1617 
1618 		} else if (class == BPF_JMP) {
1619 			u8 opcode = BPF_OP(insn->code);
1620 
1621 			if (opcode == BPF_CALL) {
1622 				if (BPF_SRC(insn->code) != BPF_K ||
1623 				    insn->off != 0 ||
1624 				    insn->src_reg != BPF_REG_0 ||
1625 				    insn->dst_reg != BPF_REG_0) {
1626 					verbose("BPF_CALL uses reserved fields\n");
1627 					return -EINVAL;
1628 				}
1629 
1630 				err = check_call(env, insn->imm);
1631 				if (err)
1632 					return err;
1633 
1634 			} else if (opcode == BPF_JA) {
1635 				if (BPF_SRC(insn->code) != BPF_K ||
1636 				    insn->imm != 0 ||
1637 				    insn->src_reg != BPF_REG_0 ||
1638 				    insn->dst_reg != BPF_REG_0) {
1639 					verbose("BPF_JA uses reserved fields\n");
1640 					return -EINVAL;
1641 				}
1642 
1643 				insn_idx += insn->off + 1;
1644 				continue;
1645 
1646 			} else if (opcode == BPF_EXIT) {
1647 				if (BPF_SRC(insn->code) != BPF_K ||
1648 				    insn->imm != 0 ||
1649 				    insn->src_reg != BPF_REG_0 ||
1650 				    insn->dst_reg != BPF_REG_0) {
1651 					verbose("BPF_EXIT uses reserved fields\n");
1652 					return -EINVAL;
1653 				}
1654 
1655 				/* eBPF calling convetion is such that R0 is used
1656 				 * to return the value from eBPF program.
1657 				 * Make sure that it's readable at this time
1658 				 * of bpf_exit, which means that program wrote
1659 				 * something into it earlier
1660 				 */
1661 				err = check_reg_arg(regs, BPF_REG_0, SRC_OP);
1662 				if (err)
1663 					return err;
1664 
1665 process_bpf_exit:
1666 				insn_idx = pop_stack(env, &prev_insn_idx);
1667 				if (insn_idx < 0) {
1668 					break;
1669 				} else {
1670 					do_print_state = true;
1671 					continue;
1672 				}
1673 			} else {
1674 				err = check_cond_jmp_op(env, insn, &insn_idx);
1675 				if (err)
1676 					return err;
1677 			}
1678 		} else if (class == BPF_LD) {
1679 			u8 mode = BPF_MODE(insn->code);
1680 
1681 			if (mode == BPF_ABS || mode == BPF_IND) {
1682 				verbose("LD_ABS is not supported yet\n");
1683 				return -EINVAL;
1684 			} else if (mode == BPF_IMM) {
1685 				err = check_ld_imm(env, insn);
1686 				if (err)
1687 					return err;
1688 
1689 				insn_idx++;
1690 			} else {
1691 				verbose("invalid BPF_LD mode\n");
1692 				return -EINVAL;
1693 			}
1694 		} else {
1695 			verbose("unknown insn class %d\n", class);
1696 			return -EINVAL;
1697 		}
1698 
1699 		insn_idx++;
1700 	}
1701 
1702 	return 0;
1703 }
1704 
1705 /* look for pseudo eBPF instructions that access map FDs and
1706  * replace them with actual map pointers
1707  */
replace_map_fd_with_map_ptr(struct verifier_env * env)1708 static int replace_map_fd_with_map_ptr(struct verifier_env *env)
1709 {
1710 	struct bpf_insn *insn = env->prog->insnsi;
1711 	int insn_cnt = env->prog->len;
1712 	int i, j;
1713 
1714 	for (i = 0; i < insn_cnt; i++, insn++) {
1715 		if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW)) {
1716 			struct bpf_map *map;
1717 			struct fd f;
1718 
1719 			if (i == insn_cnt - 1 || insn[1].code != 0 ||
1720 			    insn[1].dst_reg != 0 || insn[1].src_reg != 0 ||
1721 			    insn[1].off != 0) {
1722 				verbose("invalid bpf_ld_imm64 insn\n");
1723 				return -EINVAL;
1724 			}
1725 
1726 			if (insn->src_reg == 0)
1727 				/* valid generic load 64-bit imm */
1728 				goto next_insn;
1729 
1730 			if (insn->src_reg != BPF_PSEUDO_MAP_FD) {
1731 				verbose("unrecognized bpf_ld_imm64 insn\n");
1732 				return -EINVAL;
1733 			}
1734 
1735 			f = fdget(insn->imm);
1736 
1737 			map = bpf_map_get(f);
1738 			if (IS_ERR(map)) {
1739 				verbose("fd %d is not pointing to valid bpf_map\n",
1740 					insn->imm);
1741 				return PTR_ERR(map);
1742 			}
1743 
1744 			/* store map pointer inside BPF_LD_IMM64 instruction */
1745 			insn[0].imm = (u32) (unsigned long) map;
1746 			insn[1].imm = ((u64) (unsigned long) map) >> 32;
1747 
1748 			/* check whether we recorded this map already */
1749 			for (j = 0; j < env->used_map_cnt; j++)
1750 				if (env->used_maps[j] == map) {
1751 					fdput(f);
1752 					goto next_insn;
1753 				}
1754 
1755 			if (env->used_map_cnt >= MAX_USED_MAPS) {
1756 				fdput(f);
1757 				return -E2BIG;
1758 			}
1759 
1760 			/* remember this map */
1761 			env->used_maps[env->used_map_cnt++] = map;
1762 
1763 			/* hold the map. If the program is rejected by verifier,
1764 			 * the map will be released by release_maps() or it
1765 			 * will be used by the valid program until it's unloaded
1766 			 * and all maps are released in free_bpf_prog_info()
1767 			 */
1768 			atomic_inc(&map->refcnt);
1769 
1770 			fdput(f);
1771 next_insn:
1772 			insn++;
1773 			i++;
1774 		}
1775 	}
1776 
1777 	/* now all pseudo BPF_LD_IMM64 instructions load valid
1778 	 * 'struct bpf_map *' into a register instead of user map_fd.
1779 	 * These pointers will be used later by verifier to validate map access.
1780 	 */
1781 	return 0;
1782 }
1783 
1784 /* drop refcnt of maps used by the rejected program */
release_maps(struct verifier_env * env)1785 static void release_maps(struct verifier_env *env)
1786 {
1787 	int i;
1788 
1789 	for (i = 0; i < env->used_map_cnt; i++)
1790 		bpf_map_put(env->used_maps[i]);
1791 }
1792 
1793 /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */
convert_pseudo_ld_imm64(struct verifier_env * env)1794 static void convert_pseudo_ld_imm64(struct verifier_env *env)
1795 {
1796 	struct bpf_insn *insn = env->prog->insnsi;
1797 	int insn_cnt = env->prog->len;
1798 	int i;
1799 
1800 	for (i = 0; i < insn_cnt; i++, insn++)
1801 		if (insn->code == (BPF_LD | BPF_IMM | BPF_DW))
1802 			insn->src_reg = 0;
1803 }
1804 
free_states(struct verifier_env * env)1805 static void free_states(struct verifier_env *env)
1806 {
1807 	struct verifier_state_list *sl, *sln;
1808 	int i;
1809 
1810 	if (!env->explored_states)
1811 		return;
1812 
1813 	for (i = 0; i < env->prog->len; i++) {
1814 		sl = env->explored_states[i];
1815 
1816 		if (sl)
1817 			while (sl != STATE_LIST_MARK) {
1818 				sln = sl->next;
1819 				kfree(sl);
1820 				sl = sln;
1821 			}
1822 	}
1823 
1824 	kfree(env->explored_states);
1825 }
1826 
bpf_check(struct bpf_prog * prog,union bpf_attr * attr)1827 int bpf_check(struct bpf_prog *prog, union bpf_attr *attr)
1828 {
1829 	char __user *log_ubuf = NULL;
1830 	struct verifier_env *env;
1831 	int ret = -EINVAL;
1832 
1833 	if (prog->len <= 0 || prog->len > BPF_MAXINSNS)
1834 		return -E2BIG;
1835 
1836 	/* 'struct verifier_env' can be global, but since it's not small,
1837 	 * allocate/free it every time bpf_check() is called
1838 	 */
1839 	env = kzalloc(sizeof(struct verifier_env), GFP_KERNEL);
1840 	if (!env)
1841 		return -ENOMEM;
1842 
1843 	env->prog = prog;
1844 
1845 	/* grab the mutex to protect few globals used by verifier */
1846 	mutex_lock(&bpf_verifier_lock);
1847 
1848 	if (attr->log_level || attr->log_buf || attr->log_size) {
1849 		/* user requested verbose verifier output
1850 		 * and supplied buffer to store the verification trace
1851 		 */
1852 		log_level = attr->log_level;
1853 		log_ubuf = (char __user *) (unsigned long) attr->log_buf;
1854 		log_size = attr->log_size;
1855 		log_len = 0;
1856 
1857 		ret = -EINVAL;
1858 		/* log_* values have to be sane */
1859 		if (log_size < 128 || log_size > UINT_MAX >> 8 ||
1860 		    log_level == 0 || log_ubuf == NULL)
1861 			goto free_env;
1862 
1863 		ret = -ENOMEM;
1864 		log_buf = vmalloc(log_size);
1865 		if (!log_buf)
1866 			goto free_env;
1867 	} else {
1868 		log_level = 0;
1869 	}
1870 
1871 	ret = replace_map_fd_with_map_ptr(env);
1872 	if (ret < 0)
1873 		goto skip_full_check;
1874 
1875 	env->explored_states = kcalloc(prog->len,
1876 				       sizeof(struct verifier_state_list *),
1877 				       GFP_USER);
1878 	ret = -ENOMEM;
1879 	if (!env->explored_states)
1880 		goto skip_full_check;
1881 
1882 	ret = check_cfg(env);
1883 	if (ret < 0)
1884 		goto skip_full_check;
1885 
1886 	ret = do_check(env);
1887 
1888 skip_full_check:
1889 	while (pop_stack(env, NULL) >= 0);
1890 	free_states(env);
1891 
1892 	if (log_level && log_len >= log_size - 1) {
1893 		BUG_ON(log_len >= log_size);
1894 		/* verifier log exceeded user supplied buffer */
1895 		ret = -ENOSPC;
1896 		/* fall through to return what was recorded */
1897 	}
1898 
1899 	/* copy verifier log back to user space including trailing zero */
1900 	if (log_level && copy_to_user(log_ubuf, log_buf, log_len + 1) != 0) {
1901 		ret = -EFAULT;
1902 		goto free_log_buf;
1903 	}
1904 
1905 	if (ret == 0 && env->used_map_cnt) {
1906 		/* if program passed verifier, update used_maps in bpf_prog_info */
1907 		prog->aux->used_maps = kmalloc_array(env->used_map_cnt,
1908 						     sizeof(env->used_maps[0]),
1909 						     GFP_KERNEL);
1910 
1911 		if (!prog->aux->used_maps) {
1912 			ret = -ENOMEM;
1913 			goto free_log_buf;
1914 		}
1915 
1916 		memcpy(prog->aux->used_maps, env->used_maps,
1917 		       sizeof(env->used_maps[0]) * env->used_map_cnt);
1918 		prog->aux->used_map_cnt = env->used_map_cnt;
1919 
1920 		/* program is valid. Convert pseudo bpf_ld_imm64 into generic
1921 		 * bpf_ld_imm64 instructions
1922 		 */
1923 		convert_pseudo_ld_imm64(env);
1924 	}
1925 
1926 free_log_buf:
1927 	if (log_level)
1928 		vfree(log_buf);
1929 free_env:
1930 	if (!prog->aux->used_maps)
1931 		/* if we didn't copy map pointers into bpf_prog_info, release
1932 		 * them now. Otherwise free_bpf_prog_info() will release them.
1933 		 */
1934 		release_maps(env);
1935 	kfree(env);
1936 	mutex_unlock(&bpf_verifier_lock);
1937 	return ret;
1938 }
1939