1 /* SPDX-License-Identifier: GPL-2.0-only */
2 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
3 */
4 #ifndef _LINUX_BPF_VERIFIER_H
5 #define _LINUX_BPF_VERIFIER_H 1
6
7 #include <linux/bpf.h> /* for enum bpf_reg_type */
8 #include <linux/filter.h> /* for MAX_BPF_STACK */
9 #include <linux/tnum.h>
10 #include <linux/android_kabi.h>
11
12 /* Maximum variable offset umax_value permitted when resolving memory accesses.
13 * In practice this is far bigger than any realistic pointer offset; this limit
14 * ensures that umax_value + (int)off + (int)size cannot overflow a u64.
15 */
16 #define BPF_MAX_VAR_OFF (1 << 29)
17 /* Maximum variable size permitted for ARG_CONST_SIZE[_OR_ZERO]. This ensures
18 * that converting umax_value to int cannot overflow.
19 */
20 #define BPF_MAX_VAR_SIZ (1 << 29)
21
22 /* Liveness marks, used for registers and spilled-regs (in stack slots).
23 * Read marks propagate upwards until they find a write mark; they record that
24 * "one of this state's descendants read this reg" (and therefore the reg is
25 * relevant for states_equal() checks).
26 * Write marks collect downwards and do not propagate; they record that "the
27 * straight-line code that reached this state (from its parent) wrote this reg"
28 * (and therefore that reads propagated from this state or its descendants
29 * should not propagate to its parent).
30 * A state with a write mark can receive read marks; it just won't propagate
31 * them to its parent, since the write mark is a property, not of the state,
32 * but of the link between it and its parent. See mark_reg_read() and
33 * mark_stack_slot_read() in kernel/bpf/verifier.c.
34 */
35 enum bpf_reg_liveness {
36 REG_LIVE_NONE = 0, /* reg hasn't been read or written this branch */
37 REG_LIVE_READ32 = 0x1, /* reg was read, so we're sensitive to initial value */
38 REG_LIVE_READ64 = 0x2, /* likewise, but full 64-bit content matters */
39 REG_LIVE_READ = REG_LIVE_READ32 | REG_LIVE_READ64,
40 REG_LIVE_WRITTEN = 0x4, /* reg was written first, screening off later reads */
41 REG_LIVE_DONE = 0x8, /* liveness won't be updating this register anymore */
42 };
43
44 struct bpf_reg_state {
45 /* Ordering of fields matters. See states_equal() */
46 enum bpf_reg_type type;
47 union {
48 /* valid when type == PTR_TO_PACKET */
49 int range;
50
51 /* valid when type == CONST_PTR_TO_MAP | PTR_TO_MAP_VALUE |
52 * PTR_TO_MAP_VALUE_OR_NULL
53 */
54 struct bpf_map *map_ptr;
55
56 u32 btf_id; /* for PTR_TO_BTF_ID */
57
58 u32 mem_size; /* for PTR_TO_MEM | PTR_TO_MEM_OR_NULL */
59
60 /* Max size from any of the above. */
61 unsigned long raw;
62 };
63 /* Fixed part of pointer offset, pointer types only */
64 s32 off;
65 /* For PTR_TO_PACKET, used to find other pointers with the same variable
66 * offset, so they can share range knowledge.
67 * For PTR_TO_MAP_VALUE_OR_NULL this is used to share which map value we
68 * came from, when one is tested for != NULL.
69 * For PTR_TO_MEM_OR_NULL this is used to identify memory allocation
70 * for the purpose of tracking that it's freed.
71 * For PTR_TO_SOCKET this is used to share which pointers retain the
72 * same reference to the socket, to determine proper reference freeing.
73 */
74 u32 id;
75 /* PTR_TO_SOCKET and PTR_TO_TCP_SOCK could be a ptr returned
76 * from a pointer-cast helper, bpf_sk_fullsock() and
77 * bpf_tcp_sock().
78 *
79 * Consider the following where "sk" is a reference counted
80 * pointer returned from "sk = bpf_sk_lookup_tcp();":
81 *
82 * 1: sk = bpf_sk_lookup_tcp();
83 * 2: if (!sk) { return 0; }
84 * 3: fullsock = bpf_sk_fullsock(sk);
85 * 4: if (!fullsock) { bpf_sk_release(sk); return 0; }
86 * 5: tp = bpf_tcp_sock(fullsock);
87 * 6: if (!tp) { bpf_sk_release(sk); return 0; }
88 * 7: bpf_sk_release(sk);
89 * 8: snd_cwnd = tp->snd_cwnd; // verifier will complain
90 *
91 * After bpf_sk_release(sk) at line 7, both "fullsock" ptr and
92 * "tp" ptr should be invalidated also. In order to do that,
93 * the reg holding "fullsock" and "sk" need to remember
94 * the original refcounted ptr id (i.e. sk_reg->id) in ref_obj_id
95 * such that the verifier can reset all regs which have
96 * ref_obj_id matching the sk_reg->id.
97 *
98 * sk_reg->ref_obj_id is set to sk_reg->id at line 1.
99 * sk_reg->id will stay as NULL-marking purpose only.
100 * After NULL-marking is done, sk_reg->id can be reset to 0.
101 *
102 * After "fullsock = bpf_sk_fullsock(sk);" at line 3,
103 * fullsock_reg->ref_obj_id is set to sk_reg->ref_obj_id.
104 *
105 * After "tp = bpf_tcp_sock(fullsock);" at line 5,
106 * tp_reg->ref_obj_id is set to fullsock_reg->ref_obj_id
107 * which is the same as sk_reg->ref_obj_id.
108 *
109 * From the verifier perspective, if sk, fullsock and tp
110 * are not NULL, they are the same ptr with different
111 * reg->type. In particular, bpf_sk_release(tp) is also
112 * allowed and has the same effect as bpf_sk_release(sk).
113 */
114 u32 ref_obj_id;
115 /* For scalar types (SCALAR_VALUE), this represents our knowledge of
116 * the actual value.
117 * For pointer types, this represents the variable part of the offset
118 * from the pointed-to object, and is shared with all bpf_reg_states
119 * with the same id as us.
120 */
121 struct tnum var_off;
122 /* Used to determine if any memory access using this register will
123 * result in a bad access.
124 * These refer to the same value as var_off, not necessarily the actual
125 * contents of the register.
126 */
127 s64 smin_value; /* minimum possible (s64)value */
128 s64 smax_value; /* maximum possible (s64)value */
129 u64 umin_value; /* minimum possible (u64)value */
130 u64 umax_value; /* maximum possible (u64)value */
131 s32 s32_min_value; /* minimum possible (s32)value */
132 s32 s32_max_value; /* maximum possible (s32)value */
133 u32 u32_min_value; /* minimum possible (u32)value */
134 u32 u32_max_value; /* maximum possible (u32)value */
135 /* parentage chain for liveness checking */
136 struct bpf_reg_state *parent;
137 /* Inside the callee two registers can be both PTR_TO_STACK like
138 * R1=fp-8 and R2=fp-8, but one of them points to this function stack
139 * while another to the caller's stack. To differentiate them 'frameno'
140 * is used which is an index in bpf_verifier_state->frame[] array
141 * pointing to bpf_func_state.
142 */
143 u32 frameno;
144 /* Tracks subreg definition. The stored value is the insn_idx of the
145 * writing insn. This is safe because subreg_def is used before any insn
146 * patching which only happens after main verification finished.
147 */
148 s32 subreg_def;
149 enum bpf_reg_liveness live;
150 /* if (!precise && SCALAR_VALUE) min/max/tnum don't affect safety */
151 bool precise;
152 };
153
154 enum bpf_stack_slot_type {
155 STACK_INVALID, /* nothing was stored in this stack slot */
156 STACK_SPILL, /* register spilled into stack */
157 STACK_MISC, /* BPF program wrote some data into this slot */
158 STACK_ZERO, /* BPF program wrote constant zero */
159 };
160
161 #define BPF_REG_SIZE 8 /* size of eBPF register in bytes */
162
163 struct bpf_stack_state {
164 struct bpf_reg_state spilled_ptr;
165 u8 slot_type[BPF_REG_SIZE];
166 };
167
168 struct bpf_reference_state {
169 /* Track each reference created with a unique id, even if the same
170 * instruction creates the reference multiple times (eg, via CALL).
171 */
172 int id;
173 /* Instruction where the allocation of this reference occurred. This
174 * is used purely to inform the user of a reference leak.
175 */
176 int insn_idx;
177 };
178
179 /* state of the program:
180 * type of all registers and stack info
181 */
182 struct bpf_func_state {
183 struct bpf_reg_state regs[MAX_BPF_REG];
184 /* index of call instruction that called into this func */
185 int callsite;
186 /* stack frame number of this function state from pov of
187 * enclosing bpf_verifier_state.
188 * 0 = main function, 1 = first callee.
189 */
190 u32 frameno;
191 /* subprog number == index within subprog_info
192 * zero == main subprog
193 */
194 u32 subprogno;
195
196 /* The following fields should be last. See copy_func_state() */
197 int acquired_refs;
198 struct bpf_reference_state *refs;
199 int allocated_stack;
200 struct bpf_stack_state *stack;
201 };
202
203 struct bpf_idx_pair {
204 u32 prev_idx;
205 u32 idx;
206 };
207
208 struct bpf_id_pair {
209 u32 old;
210 u32 cur;
211 };
212
213 /* Maximum number of register states that can exist at once */
214 #define BPF_ID_MAP_SIZE (MAX_BPF_REG + MAX_BPF_STACK / BPF_REG_SIZE)
215 #define MAX_CALL_FRAMES 8
216 struct bpf_verifier_state {
217 /* call stack tracking */
218 struct bpf_func_state *frame[MAX_CALL_FRAMES];
219 struct bpf_verifier_state *parent;
220 /*
221 * 'branches' field is the number of branches left to explore:
222 * 0 - all possible paths from this state reached bpf_exit or
223 * were safely pruned
224 * 1 - at least one path is being explored.
225 * This state hasn't reached bpf_exit
226 * 2 - at least two paths are being explored.
227 * This state is an immediate parent of two children.
228 * One is fallthrough branch with branches==1 and another
229 * state is pushed into stack (to be explored later) also with
230 * branches==1. The parent of this state has branches==1.
231 * The verifier state tree connected via 'parent' pointer looks like:
232 * 1
233 * 1
234 * 2 -> 1 (first 'if' pushed into stack)
235 * 1
236 * 2 -> 1 (second 'if' pushed into stack)
237 * 1
238 * 1
239 * 1 bpf_exit.
240 *
241 * Once do_check() reaches bpf_exit, it calls update_branch_counts()
242 * and the verifier state tree will look:
243 * 1
244 * 1
245 * 2 -> 1 (first 'if' pushed into stack)
246 * 1
247 * 1 -> 1 (second 'if' pushed into stack)
248 * 0
249 * 0
250 * 0 bpf_exit.
251 * After pop_stack() the do_check() will resume at second 'if'.
252 *
253 * If is_state_visited() sees a state with branches > 0 it means
254 * there is a loop. If such state is exactly equal to the current state
255 * it's an infinite loop. Note states_equal() checks for states
256 * equvalency, so two states being 'states_equal' does not mean
257 * infinite loop. The exact comparison is provided by
258 * states_maybe_looping() function. It's a stronger pre-check and
259 * much faster than states_equal().
260 *
261 * This algorithm may not find all possible infinite loops or
262 * loop iteration count may be too high.
263 * In such cases BPF_COMPLEXITY_LIMIT_INSNS limit kicks in.
264 */
265 u32 branches;
266 u32 insn_idx;
267 u32 curframe;
268 u32 active_spin_lock;
269 bool speculative;
270
271 /* first and last insn idx of this verifier state */
272 u32 first_insn_idx;
273 u32 last_insn_idx;
274 /* jmp history recorded from first to last.
275 * backtracking is using it to go from last to first.
276 * For most states jmp_history_cnt is [0-3].
277 * For loops can go up to ~40.
278 */
279 struct bpf_idx_pair *jmp_history;
280 u32 jmp_history_cnt;
281 };
282
283 #define bpf_get_spilled_reg(slot, frame) \
284 (((slot < frame->allocated_stack / BPF_REG_SIZE) && \
285 (frame->stack[slot].slot_type[0] == STACK_SPILL)) \
286 ? &frame->stack[slot].spilled_ptr : NULL)
287
288 /* Iterate over 'frame', setting 'reg' to either NULL or a spilled register. */
289 #define bpf_for_each_spilled_reg(iter, frame, reg) \
290 for (iter = 0, reg = bpf_get_spilled_reg(iter, frame); \
291 iter < frame->allocated_stack / BPF_REG_SIZE; \
292 iter++, reg = bpf_get_spilled_reg(iter, frame))
293
294 /* Invoke __expr over regsiters in __vst, setting __state and __reg */
295 #define bpf_for_each_reg_in_vstate(__vst, __state, __reg, __expr) \
296 ({ \
297 struct bpf_verifier_state *___vstate = __vst; \
298 int ___i, ___j; \
299 for (___i = 0; ___i <= ___vstate->curframe; ___i++) { \
300 struct bpf_reg_state *___regs; \
301 __state = ___vstate->frame[___i]; \
302 ___regs = __state->regs; \
303 for (___j = 0; ___j < MAX_BPF_REG; ___j++) { \
304 __reg = &___regs[___j]; \
305 (void)(__expr); \
306 } \
307 bpf_for_each_spilled_reg(___j, __state, __reg) { \
308 if (!__reg) \
309 continue; \
310 (void)(__expr); \
311 } \
312 } \
313 })
314
315 /* linked list of verifier states used to prune search */
316 struct bpf_verifier_state_list {
317 struct bpf_verifier_state state;
318 struct bpf_verifier_state_list *next;
319 int miss_cnt, hit_cnt;
320 };
321
322 /* Possible states for alu_state member. */
323 #define BPF_ALU_SANITIZE_SRC (1U << 0)
324 #define BPF_ALU_SANITIZE_DST (1U << 1)
325 #define BPF_ALU_NEG_VALUE (1U << 2)
326 #define BPF_ALU_NON_POINTER (1U << 3)
327 #define BPF_ALU_IMMEDIATE (1U << 4)
328 #define BPF_ALU_SANITIZE (BPF_ALU_SANITIZE_SRC | \
329 BPF_ALU_SANITIZE_DST)
330
331 struct bpf_insn_aux_data {
332 union {
333 enum bpf_reg_type ptr_type; /* pointer type for load/store insns */
334 unsigned long map_ptr_state; /* pointer/poison value for maps */
335 s32 call_imm; /* saved imm field of call insn */
336 u32 alu_limit; /* limit for add/sub register with pointer */
337 struct {
338 u32 map_index; /* index into used_maps[] */
339 u32 map_off; /* offset from value base address */
340 };
341 struct {
342 enum bpf_reg_type reg_type; /* type of pseudo_btf_id */
343 union {
344 u32 btf_id; /* btf_id for struct typed var */
345 u32 mem_size; /* mem_size for non-struct typed var */
346 };
347 } btf_var;
348 };
349 u64 map_key_state; /* constant (32 bit) key tracking for maps */
350 int ctx_field_size; /* the ctx field size for load insn, maybe 0 */
351 u32 seen; /* this insn was processed by the verifier at env->pass_cnt */
352 bool sanitize_stack_spill; /* subject to Spectre v4 sanitation */
353 bool zext_dst; /* this insn zero extends dst reg */
354 u8 alu_state; /* used in combination with alu_limit */
355
356 /* below fields are initialized once */
357 unsigned int orig_idx; /* original instruction index */
358 bool prune_point;
359 };
360
361 #define MAX_USED_MAPS 64 /* max number of maps accessed by one eBPF program */
362
363 #define BPF_VERIFIER_TMP_LOG_SIZE 1024
364
365 struct bpf_verifier_log {
366 u32 level;
367 char kbuf[BPF_VERIFIER_TMP_LOG_SIZE];
368 char __user *ubuf;
369 u32 len_used;
370 u32 len_total;
371 };
372
bpf_verifier_log_full(const struct bpf_verifier_log * log)373 static inline bool bpf_verifier_log_full(const struct bpf_verifier_log *log)
374 {
375 return log->len_used >= log->len_total - 1;
376 }
377
378 #define BPF_LOG_LEVEL1 1
379 #define BPF_LOG_LEVEL2 2
380 #define BPF_LOG_STATS 4
381 #define BPF_LOG_LEVEL (BPF_LOG_LEVEL1 | BPF_LOG_LEVEL2)
382 #define BPF_LOG_MASK (BPF_LOG_LEVEL | BPF_LOG_STATS)
383 #define BPF_LOG_KERNEL (BPF_LOG_MASK + 1) /* kernel internal flag */
384
bpf_verifier_log_needed(const struct bpf_verifier_log * log)385 static inline bool bpf_verifier_log_needed(const struct bpf_verifier_log *log)
386 {
387 return log &&
388 ((log->level && log->ubuf && !bpf_verifier_log_full(log)) ||
389 log->level == BPF_LOG_KERNEL);
390 }
391
392 static inline bool
bpf_verifier_log_attr_valid(const struct bpf_verifier_log * log)393 bpf_verifier_log_attr_valid(const struct bpf_verifier_log *log)
394 {
395 return log->len_total >= 128 && log->len_total <= UINT_MAX >> 2 &&
396 log->level && log->ubuf && !(log->level & ~BPF_LOG_MASK);
397 }
398
399 #define BPF_MAX_SUBPROGS 256
400
401 struct bpf_subprog_info {
402 /* 'start' has to be the first field otherwise find_subprog() won't work */
403 u32 start; /* insn idx of function entry point */
404 u32 linfo_idx; /* The idx to the main_prog->aux->linfo */
405 u16 stack_depth; /* max. stack depth used by this function */
406 bool has_tail_call;
407 bool tail_call_reachable;
408 bool has_ld_abs;
409
410 ANDROID_KABI_RESERVE(1);
411 };
412
413 /* single container for all structs
414 * one verifier_env per bpf_check() call
415 */
416 struct bpf_verifier_env {
417 u32 insn_idx;
418 u32 prev_insn_idx;
419 struct bpf_prog *prog; /* eBPF program being verified */
420 const struct bpf_verifier_ops *ops;
421 struct bpf_verifier_stack_elem *head; /* stack of verifier states to be processed */
422 int stack_size; /* number of states to be processed */
423 bool strict_alignment; /* perform strict pointer alignment checks */
424 bool test_state_freq; /* test verifier with different pruning frequency */
425 struct bpf_verifier_state *cur_state; /* current verifier state */
426 struct bpf_verifier_state_list **explored_states; /* search pruning optimization */
427 struct bpf_verifier_state_list *free_list;
428 struct bpf_map *used_maps[MAX_USED_MAPS]; /* array of map's used by eBPF program */
429 u32 used_map_cnt; /* number of used maps */
430 u32 id_gen; /* used to generate unique reg IDs */
431 bool explore_alu_limits;
432 bool allow_ptr_leaks;
433 bool allow_uninit_stack;
434 bool allow_ptr_to_map_access;
435 bool bpf_capable;
436 bool bypass_spec_v1;
437 bool bypass_spec_v4;
438 bool seen_direct_write;
439 struct bpf_insn_aux_data *insn_aux_data; /* array of per-insn state */
440 const struct bpf_line_info *prev_linfo;
441 struct bpf_verifier_log log;
442 struct bpf_subprog_info subprog_info[BPF_MAX_SUBPROGS + 1];
443 struct bpf_id_pair idmap_scratch[BPF_ID_MAP_SIZE];
444 struct {
445 int *insn_state;
446 int *insn_stack;
447 int cur_stack;
448 } cfg;
449 u32 pass_cnt; /* number of times do_check() was called */
450 u32 subprog_cnt;
451 /* number of instructions analyzed by the verifier */
452 u32 prev_insn_processed, insn_processed;
453 /* number of jmps, calls, exits analyzed so far */
454 u32 prev_jmps_processed, jmps_processed;
455 /* total verification time */
456 u64 verification_time;
457 /* maximum number of verifier states kept in 'branching' instructions */
458 u32 max_states_per_insn;
459 /* total number of allocated verifier states */
460 u32 total_states;
461 /* some states are freed during program analysis.
462 * this is peak number of states. this number dominates kernel
463 * memory consumption during verification
464 */
465 u32 peak_states;
466 /* longest register parentage chain walked for liveness marking */
467 u32 longest_mark_read_walk;
468
469 ANDROID_KABI_RESERVE(1);
470 ANDROID_KABI_RESERVE(2);
471 };
472
473 __printf(2, 0) void bpf_verifier_vlog(struct bpf_verifier_log *log,
474 const char *fmt, va_list args);
475 __printf(2, 3) void bpf_verifier_log_write(struct bpf_verifier_env *env,
476 const char *fmt, ...);
477 __printf(2, 3) void bpf_log(struct bpf_verifier_log *log,
478 const char *fmt, ...);
479
cur_func(struct bpf_verifier_env * env)480 static inline struct bpf_func_state *cur_func(struct bpf_verifier_env *env)
481 {
482 struct bpf_verifier_state *cur = env->cur_state;
483
484 return cur->frame[cur->curframe];
485 }
486
cur_regs(struct bpf_verifier_env * env)487 static inline struct bpf_reg_state *cur_regs(struct bpf_verifier_env *env)
488 {
489 return cur_func(env)->regs;
490 }
491
492 int bpf_prog_offload_verifier_prep(struct bpf_prog *prog);
493 int bpf_prog_offload_verify_insn(struct bpf_verifier_env *env,
494 int insn_idx, int prev_insn_idx);
495 int bpf_prog_offload_finalize(struct bpf_verifier_env *env);
496 void
497 bpf_prog_offload_replace_insn(struct bpf_verifier_env *env, u32 off,
498 struct bpf_insn *insn);
499 void
500 bpf_prog_offload_remove_insns(struct bpf_verifier_env *env, u32 off, u32 cnt);
501
502 int check_ctx_reg(struct bpf_verifier_env *env,
503 const struct bpf_reg_state *reg, int regno);
504
505 /* this lives here instead of in bpf.h because it needs to dereference tgt_prog */
bpf_trampoline_compute_key(const struct bpf_prog * tgt_prog,u32 btf_id)506 static inline u64 bpf_trampoline_compute_key(const struct bpf_prog *tgt_prog,
507 u32 btf_id)
508 {
509 return tgt_prog ? (((u64)tgt_prog->aux->id) << 32 | btf_id) : btf_id;
510 }
511
512 int bpf_check_attach_target(struct bpf_verifier_log *log,
513 const struct bpf_prog *prog,
514 const struct bpf_prog *tgt_prog,
515 u32 btf_id,
516 struct bpf_attach_target_info *tgt_info);
517
518 #endif /* _LINUX_BPF_VERIFIER_H */
519