1 /* SPDX-License-Identifier: GPL-2.0-only */
2 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
4 #ifndef _LINUX_BPF_VERIFIER_H
5 #define _LINUX_BPF_VERIFIER_H 1
7 #include <linux/bpf.h> /* for enum bpf_reg_type */
8 #include <linux/filter.h> /* for MAX_BPF_STACK */
9 #include <linux/tnum.h>
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.
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.
19 #define BPF_MAX_VAR_SIZ (1 << 29)
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.
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 */
43 struct bpf_reg_state {
44 /* Ordering of fields matters. See states_equal() */
45 enum bpf_reg_type type;
47 /* valid when type == PTR_TO_PACKET */
50 /* valid when type == CONST_PTR_TO_MAP | PTR_TO_MAP_VALUE |
51 * PTR_TO_MAP_VALUE_OR_NULL
53 struct bpf_map *map_ptr;
55 u32 btf_id; /* for PTR_TO_BTF_ID */
57 /* Max size from any of the above. */
60 /* Fixed part of pointer offset, pointer types only */
62 /* For PTR_TO_PACKET, used to find other pointers with the same variable
63 * offset, so they can share range knowledge.
64 * For PTR_TO_MAP_VALUE_OR_NULL this is used to share which map value we
65 * came from, when one is tested for != NULL.
66 * For PTR_TO_SOCKET this is used to share which pointers retain the
67 * same reference to the socket, to determine proper reference freeing.
70 /* PTR_TO_SOCKET and PTR_TO_TCP_SOCK could be a ptr returned
71 * from a pointer-cast helper, bpf_sk_fullsock() and
74 * Consider the following where "sk" is a reference counted
75 * pointer returned from "sk = bpf_sk_lookup_tcp();":
77 * 1: sk = bpf_sk_lookup_tcp();
78 * 2: if (!sk) { return 0; }
79 * 3: fullsock = bpf_sk_fullsock(sk);
80 * 4: if (!fullsock) { bpf_sk_release(sk); return 0; }
81 * 5: tp = bpf_tcp_sock(fullsock);
82 * 6: if (!tp) { bpf_sk_release(sk); return 0; }
83 * 7: bpf_sk_release(sk);
84 * 8: snd_cwnd = tp->snd_cwnd; // verifier will complain
86 * After bpf_sk_release(sk) at line 7, both "fullsock" ptr and
87 * "tp" ptr should be invalidated also. In order to do that,
88 * the reg holding "fullsock" and "sk" need to remember
89 * the original refcounted ptr id (i.e. sk_reg->id) in ref_obj_id
90 * such that the verifier can reset all regs which have
91 * ref_obj_id matching the sk_reg->id.
93 * sk_reg->ref_obj_id is set to sk_reg->id at line 1.
94 * sk_reg->id will stay as NULL-marking purpose only.
95 * After NULL-marking is done, sk_reg->id can be reset to 0.
97 * After "fullsock = bpf_sk_fullsock(sk);" at line 3,
98 * fullsock_reg->ref_obj_id is set to sk_reg->ref_obj_id.
100 * After "tp = bpf_tcp_sock(fullsock);" at line 5,
101 * tp_reg->ref_obj_id is set to fullsock_reg->ref_obj_id
102 * which is the same as sk_reg->ref_obj_id.
104 * From the verifier perspective, if sk, fullsock and tp
105 * are not NULL, they are the same ptr with different
106 * reg->type. In particular, bpf_sk_release(tp) is also
107 * allowed and has the same effect as bpf_sk_release(sk).
110 /* For scalar types (SCALAR_VALUE), this represents our knowledge of
112 * For pointer types, this represents the variable part of the offset
113 * from the pointed-to object, and is shared with all bpf_reg_states
114 * with the same id as us.
117 /* Used to determine if any memory access using this register will
118 * result in a bad access.
119 * These refer to the same value as var_off, not necessarily the actual
120 * contents of the register.
122 s64 smin_value; /* minimum possible (s64)value */
123 s64 smax_value; /* maximum possible (s64)value */
124 u64 umin_value; /* minimum possible (u64)value */
125 u64 umax_value; /* maximum possible (u64)value */
126 /* parentage chain for liveness checking */
127 struct bpf_reg_state *parent;
128 /* Inside the callee two registers can be both PTR_TO_STACK like
129 * R1=fp-8 and R2=fp-8, but one of them points to this function stack
130 * while another to the caller's stack. To differentiate them 'frameno'
131 * is used which is an index in bpf_verifier_state->frame[] array
132 * pointing to bpf_func_state.
135 /* Tracks subreg definition. The stored value is the insn_idx of the
136 * writing insn. This is safe because subreg_def is used before any insn
137 * patching which only happens after main verification finished.
140 enum bpf_reg_liveness live;
141 /* if (!precise && SCALAR_VALUE) min/max/tnum don't affect safety */
145 enum bpf_stack_slot_type {
146 STACK_INVALID, /* nothing was stored in this stack slot */
147 STACK_SPILL, /* register spilled into stack */
148 STACK_MISC, /* BPF program wrote some data into this slot */
149 STACK_ZERO, /* BPF program wrote constant zero */
152 #define BPF_REG_SIZE 8 /* size of eBPF register in bytes */
154 struct bpf_stack_state {
155 struct bpf_reg_state spilled_ptr;
156 u8 slot_type[BPF_REG_SIZE];
159 struct bpf_reference_state {
160 /* Track each reference created with a unique id, even if the same
161 * instruction creates the reference multiple times (eg, via CALL).
164 /* Instruction where the allocation of this reference occurred. This
165 * is used purely to inform the user of a reference leak.
170 /* state of the program:
171 * type of all registers and stack info
173 struct bpf_func_state {
174 struct bpf_reg_state regs[MAX_BPF_REG];
175 /* index of call instruction that called into this func */
177 /* stack frame number of this function state from pov of
178 * enclosing bpf_verifier_state.
179 * 0 = main function, 1 = first callee.
182 /* subprog number == index within subprog_stack_depth
183 * zero == main subprog
187 /* The following fields should be last. See copy_func_state() */
189 struct bpf_reference_state *refs;
191 struct bpf_stack_state *stack;
194 struct bpf_idx_pair {
199 #define MAX_CALL_FRAMES 8
200 struct bpf_verifier_state {
201 /* call stack tracking */
202 struct bpf_func_state *frame[MAX_CALL_FRAMES];
203 struct bpf_verifier_state *parent;
205 * 'branches' field is the number of branches left to explore:
206 * 0 - all possible paths from this state reached bpf_exit or
208 * 1 - at least one path is being explored.
209 * This state hasn't reached bpf_exit
210 * 2 - at least two paths are being explored.
211 * This state is an immediate parent of two children.
212 * One is fallthrough branch with branches==1 and another
213 * state is pushed into stack (to be explored later) also with
214 * branches==1. The parent of this state has branches==1.
215 * The verifier state tree connected via 'parent' pointer looks like:
218 * 2 -> 1 (first 'if' pushed into stack)
220 * 2 -> 1 (second 'if' pushed into stack)
225 * Once do_check() reaches bpf_exit, it calls update_branch_counts()
226 * and the verifier state tree will look:
229 * 2 -> 1 (first 'if' pushed into stack)
231 * 1 -> 1 (second 'if' pushed into stack)
235 * After pop_stack() the do_check() will resume at second 'if'.
237 * If is_state_visited() sees a state with branches > 0 it means
238 * there is a loop. If such state is exactly equal to the current state
239 * it's an infinite loop. Note states_equal() checks for states
240 * equvalency, so two states being 'states_equal' does not mean
241 * infinite loop. The exact comparison is provided by
242 * states_maybe_looping() function. It's a stronger pre-check and
243 * much faster than states_equal().
245 * This algorithm may not find all possible infinite loops or
246 * loop iteration count may be too high.
247 * In such cases BPF_COMPLEXITY_LIMIT_INSNS limit kicks in.
252 u32 active_spin_lock;
255 /* first and last insn idx of this verifier state */
258 /* jmp history recorded from first to last.
259 * backtracking is using it to go from last to first.
260 * For most states jmp_history_cnt is [0-3].
261 * For loops can go up to ~40.
263 struct bpf_idx_pair *jmp_history;
267 #define bpf_get_spilled_reg(slot, frame) \
268 (((slot < frame->allocated_stack / BPF_REG_SIZE) && \
269 (frame->stack[slot].slot_type[0] == STACK_SPILL)) \
270 ? &frame->stack[slot].spilled_ptr : NULL)
272 /* Iterate over 'frame', setting 'reg' to either NULL or a spilled register. */
273 #define bpf_for_each_spilled_reg(iter, frame, reg) \
274 for (iter = 0, reg = bpf_get_spilled_reg(iter, frame); \
275 iter < frame->allocated_stack / BPF_REG_SIZE; \
276 iter++, reg = bpf_get_spilled_reg(iter, frame))
278 /* linked list of verifier states used to prune search */
279 struct bpf_verifier_state_list {
280 struct bpf_verifier_state state;
281 struct bpf_verifier_state_list *next;
282 int miss_cnt, hit_cnt;
285 /* Possible states for alu_state member. */
286 #define BPF_ALU_SANITIZE_SRC 1U
287 #define BPF_ALU_SANITIZE_DST 2U
288 #define BPF_ALU_NEG_VALUE (1U << 2)
289 #define BPF_ALU_NON_POINTER (1U << 3)
290 #define BPF_ALU_SANITIZE (BPF_ALU_SANITIZE_SRC | \
291 BPF_ALU_SANITIZE_DST)
293 struct bpf_insn_aux_data {
295 enum bpf_reg_type ptr_type; /* pointer type for load/store insns */
296 unsigned long map_state; /* pointer/poison value for maps */
297 s32 call_imm; /* saved imm field of call insn */
298 u32 alu_limit; /* limit for add/sub register with pointer */
300 u32 map_index; /* index into used_maps[] */
301 u32 map_off; /* offset from value base address */
304 int ctx_field_size; /* the ctx field size for load insn, maybe 0 */
305 int sanitize_stack_off; /* stack slot to be cleared */
306 bool seen; /* this insn was processed by the verifier */
307 bool zext_dst; /* this insn zero extends dst reg */
308 u8 alu_state; /* used in combination with alu_limit */
310 unsigned int orig_idx; /* original instruction index */
313 #define MAX_USED_MAPS 64 /* max number of maps accessed by one eBPF program */
315 #define BPF_VERIFIER_TMP_LOG_SIZE 1024
317 struct bpf_verifier_log {
319 char kbuf[BPF_VERIFIER_TMP_LOG_SIZE];
325 static inline bool bpf_verifier_log_full(const struct bpf_verifier_log *log)
327 return log->len_used >= log->len_total - 1;
330 #define BPF_LOG_LEVEL1 1
331 #define BPF_LOG_LEVEL2 2
332 #define BPF_LOG_STATS 4
333 #define BPF_LOG_LEVEL (BPF_LOG_LEVEL1 | BPF_LOG_LEVEL2)
334 #define BPF_LOG_MASK (BPF_LOG_LEVEL | BPF_LOG_STATS)
335 #define BPF_LOG_KERNEL (BPF_LOG_MASK + 1) /* kernel internal flag */
337 static inline bool bpf_verifier_log_needed(const struct bpf_verifier_log *log)
339 return (log->level && log->ubuf && !bpf_verifier_log_full(log)) ||
340 log->level == BPF_LOG_KERNEL;
343 #define BPF_MAX_SUBPROGS 256
345 struct bpf_subprog_info {
346 u32 start; /* insn idx of function entry point */
347 u32 linfo_idx; /* The idx to the main_prog->aux->linfo */
348 u16 stack_depth; /* max. stack depth used by this function */
351 /* single container for all structs
352 * one verifier_env per bpf_check() call
354 struct bpf_verifier_env {
357 struct bpf_prog *prog; /* eBPF program being verified */
358 const struct bpf_verifier_ops *ops;
359 struct bpf_verifier_stack_elem *head; /* stack of verifier states to be processed */
360 int stack_size; /* number of states to be processed */
361 bool strict_alignment; /* perform strict pointer alignment checks */
362 bool test_state_freq; /* test verifier with different pruning frequency */
363 struct bpf_verifier_state *cur_state; /* current verifier state */
364 struct bpf_verifier_state_list **explored_states; /* search pruning optimization */
365 struct bpf_verifier_state_list *free_list;
366 struct bpf_map *used_maps[MAX_USED_MAPS]; /* array of map's used by eBPF program */
367 u32 used_map_cnt; /* number of used maps */
368 u32 id_gen; /* used to generate unique reg IDs */
369 bool allow_ptr_leaks;
370 bool seen_direct_write;
371 struct bpf_insn_aux_data *insn_aux_data; /* array of per-insn state */
372 const struct bpf_line_info *prev_linfo;
373 struct bpf_verifier_log log;
374 struct bpf_subprog_info subprog_info[BPF_MAX_SUBPROGS + 1];
381 /* number of instructions analyzed by the verifier */
382 u32 prev_insn_processed, insn_processed;
383 /* number of jmps, calls, exits analyzed so far */
384 u32 prev_jmps_processed, jmps_processed;
385 /* total verification time */
386 u64 verification_time;
387 /* maximum number of verifier states kept in 'branching' instructions */
388 u32 max_states_per_insn;
389 /* total number of allocated verifier states */
391 /* some states are freed during program analysis.
392 * this is peak number of states. this number dominates kernel
393 * memory consumption during verification
396 /* longest register parentage chain walked for liveness marking */
397 u32 longest_mark_read_walk;
400 __printf(2, 0) void bpf_verifier_vlog(struct bpf_verifier_log *log,
401 const char *fmt, va_list args);
402 __printf(2, 3) void bpf_verifier_log_write(struct bpf_verifier_env *env,
403 const char *fmt, ...);
404 __printf(2, 3) void bpf_log(struct bpf_verifier_log *log,
405 const char *fmt, ...);
407 static inline struct bpf_func_state *cur_func(struct bpf_verifier_env *env)
409 struct bpf_verifier_state *cur = env->cur_state;
411 return cur->frame[cur->curframe];
414 static inline struct bpf_reg_state *cur_regs(struct bpf_verifier_env *env)
416 return cur_func(env)->regs;
419 int bpf_prog_offload_verifier_prep(struct bpf_prog *prog);
420 int bpf_prog_offload_verify_insn(struct bpf_verifier_env *env,
421 int insn_idx, int prev_insn_idx);
422 int bpf_prog_offload_finalize(struct bpf_verifier_env *env);
424 bpf_prog_offload_replace_insn(struct bpf_verifier_env *env, u32 off,
425 struct bpf_insn *insn);
427 bpf_prog_offload_remove_insns(struct bpf_verifier_env *env, u32 off, u32 cnt);
429 #endif /* _LINUX_BPF_VERIFIER_H */