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