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 u32 mem_size; /* for PTR_TO_MEM | PTR_TO_MEM_OR_NULL */
59 /* Max size from any of the above. */
62 /* Fixed part of pointer offset, pointer types only */
64 /* For PTR_TO_PACKET, used to find other pointers with the same variable
65 * offset, so they can share range knowledge.
66 * For PTR_TO_MAP_VALUE_OR_NULL this is used to share which map value we
67 * came from, when one is tested for != NULL.
68 * For PTR_TO_MEM_OR_NULL this is used to identify memory allocation
69 * for the purpose of tracking that it's freed.
70 * For PTR_TO_SOCKET this is used to share which pointers retain the
71 * same reference to the socket, to determine proper reference freeing.
74 /* PTR_TO_SOCKET and PTR_TO_TCP_SOCK could be a ptr returned
75 * from a pointer-cast helper, bpf_sk_fullsock() and
78 * Consider the following where "sk" is a reference counted
79 * pointer returned from "sk = bpf_sk_lookup_tcp();":
81 * 1: sk = bpf_sk_lookup_tcp();
82 * 2: if (!sk) { return 0; }
83 * 3: fullsock = bpf_sk_fullsock(sk);
84 * 4: if (!fullsock) { bpf_sk_release(sk); return 0; }
85 * 5: tp = bpf_tcp_sock(fullsock);
86 * 6: if (!tp) { bpf_sk_release(sk); return 0; }
87 * 7: bpf_sk_release(sk);
88 * 8: snd_cwnd = tp->snd_cwnd; // verifier will complain
90 * After bpf_sk_release(sk) at line 7, both "fullsock" ptr and
91 * "tp" ptr should be invalidated also. In order to do that,
92 * the reg holding "fullsock" and "sk" need to remember
93 * the original refcounted ptr id (i.e. sk_reg->id) in ref_obj_id
94 * such that the verifier can reset all regs which have
95 * ref_obj_id matching the sk_reg->id.
97 * sk_reg->ref_obj_id is set to sk_reg->id at line 1.
98 * sk_reg->id will stay as NULL-marking purpose only.
99 * After NULL-marking is done, sk_reg->id can be reset to 0.
101 * After "fullsock = bpf_sk_fullsock(sk);" at line 3,
102 * fullsock_reg->ref_obj_id is set to sk_reg->ref_obj_id.
104 * After "tp = bpf_tcp_sock(fullsock);" at line 5,
105 * tp_reg->ref_obj_id is set to fullsock_reg->ref_obj_id
106 * which is the same as sk_reg->ref_obj_id.
108 * From the verifier perspective, if sk, fullsock and tp
109 * are not NULL, they are the same ptr with different
110 * reg->type. In particular, bpf_sk_release(tp) is also
111 * allowed and has the same effect as bpf_sk_release(sk).
114 /* For scalar types (SCALAR_VALUE), this represents our knowledge of
116 * For pointer types, this represents the variable part of the offset
117 * from the pointed-to object, and is shared with all bpf_reg_states
118 * with the same id as us.
121 /* Used to determine if any memory access using this register will
122 * result in a bad access.
123 * These refer to the same value as var_off, not necessarily the actual
124 * contents of the register.
126 s64 smin_value; /* minimum possible (s64)value */
127 s64 smax_value; /* maximum possible (s64)value */
128 u64 umin_value; /* minimum possible (u64)value */
129 u64 umax_value; /* maximum possible (u64)value */
130 s32 s32_min_value; /* minimum possible (s32)value */
131 s32 s32_max_value; /* maximum possible (s32)value */
132 u32 u32_min_value; /* minimum possible (u32)value */
133 u32 u32_max_value; /* maximum possible (u32)value */
134 /* parentage chain for liveness checking */
135 struct bpf_reg_state *parent;
136 /* Inside the callee two registers can be both PTR_TO_STACK like
137 * R1=fp-8 and R2=fp-8, but one of them points to this function stack
138 * while another to the caller's stack. To differentiate them 'frameno'
139 * is used which is an index in bpf_verifier_state->frame[] array
140 * pointing to bpf_func_state.
143 /* Tracks subreg definition. The stored value is the insn_idx of the
144 * writing insn. This is safe because subreg_def is used before any insn
145 * patching which only happens after main verification finished.
148 enum bpf_reg_liveness live;
149 /* if (!precise && SCALAR_VALUE) min/max/tnum don't affect safety */
153 enum bpf_stack_slot_type {
154 STACK_INVALID, /* nothing was stored in this stack slot */
155 STACK_SPILL, /* register spilled into stack */
156 STACK_MISC, /* BPF program wrote some data into this slot */
157 STACK_ZERO, /* BPF program wrote constant zero */
160 #define BPF_REG_SIZE 8 /* size of eBPF register in bytes */
162 struct bpf_stack_state {
163 struct bpf_reg_state spilled_ptr;
164 u8 slot_type[BPF_REG_SIZE];
167 struct bpf_reference_state {
168 /* Track each reference created with a unique id, even if the same
169 * instruction creates the reference multiple times (eg, via CALL).
172 /* Instruction where the allocation of this reference occurred. This
173 * is used purely to inform the user of a reference leak.
178 /* state of the program:
179 * type of all registers and stack info
181 struct bpf_func_state {
182 struct bpf_reg_state regs[MAX_BPF_REG];
183 /* index of call instruction that called into this func */
185 /* stack frame number of this function state from pov of
186 * enclosing bpf_verifier_state.
187 * 0 = main function, 1 = first callee.
190 /* subprog number == index within subprog_stack_depth
191 * zero == main subprog
195 /* The following fields should be last. See copy_func_state() */
197 struct bpf_reference_state *refs;
199 struct bpf_stack_state *stack;
202 struct bpf_idx_pair {
207 #define MAX_CALL_FRAMES 8
208 struct bpf_verifier_state {
209 /* call stack tracking */
210 struct bpf_func_state *frame[MAX_CALL_FRAMES];
211 struct bpf_verifier_state *parent;
213 * 'branches' field is the number of branches left to explore:
214 * 0 - all possible paths from this state reached bpf_exit or
216 * 1 - at least one path is being explored.
217 * This state hasn't reached bpf_exit
218 * 2 - at least two paths are being explored.
219 * This state is an immediate parent of two children.
220 * One is fallthrough branch with branches==1 and another
221 * state is pushed into stack (to be explored later) also with
222 * branches==1. The parent of this state has branches==1.
223 * The verifier state tree connected via 'parent' pointer looks like:
226 * 2 -> 1 (first 'if' pushed into stack)
228 * 2 -> 1 (second 'if' pushed into stack)
233 * Once do_check() reaches bpf_exit, it calls update_branch_counts()
234 * and the verifier state tree will look:
237 * 2 -> 1 (first 'if' pushed into stack)
239 * 1 -> 1 (second 'if' pushed into stack)
243 * After pop_stack() the do_check() will resume at second 'if'.
245 * If is_state_visited() sees a state with branches > 0 it means
246 * there is a loop. If such state is exactly equal to the current state
247 * it's an infinite loop. Note states_equal() checks for states
248 * equvalency, so two states being 'states_equal' does not mean
249 * infinite loop. The exact comparison is provided by
250 * states_maybe_looping() function. It's a stronger pre-check and
251 * much faster than states_equal().
253 * This algorithm may not find all possible infinite loops or
254 * loop iteration count may be too high.
255 * In such cases BPF_COMPLEXITY_LIMIT_INSNS limit kicks in.
260 u32 active_spin_lock;
263 /* first and last insn idx of this verifier state */
266 /* jmp history recorded from first to last.
267 * backtracking is using it to go from last to first.
268 * For most states jmp_history_cnt is [0-3].
269 * For loops can go up to ~40.
271 struct bpf_idx_pair *jmp_history;
275 #define bpf_get_spilled_reg(slot, frame) \
276 (((slot < frame->allocated_stack / BPF_REG_SIZE) && \
277 (frame->stack[slot].slot_type[0] == STACK_SPILL)) \
278 ? &frame->stack[slot].spilled_ptr : NULL)
280 /* Iterate over 'frame', setting 'reg' to either NULL or a spilled register. */
281 #define bpf_for_each_spilled_reg(iter, frame, reg) \
282 for (iter = 0, reg = bpf_get_spilled_reg(iter, frame); \
283 iter < frame->allocated_stack / BPF_REG_SIZE; \
284 iter++, reg = bpf_get_spilled_reg(iter, frame))
286 /* linked list of verifier states used to prune search */
287 struct bpf_verifier_state_list {
288 struct bpf_verifier_state state;
289 struct bpf_verifier_state_list *next;
290 int miss_cnt, hit_cnt;
293 /* Possible states for alu_state member. */
294 #define BPF_ALU_SANITIZE_SRC 1U
295 #define BPF_ALU_SANITIZE_DST 2U
296 #define BPF_ALU_NEG_VALUE (1U << 2)
297 #define BPF_ALU_NON_POINTER (1U << 3)
298 #define BPF_ALU_SANITIZE (BPF_ALU_SANITIZE_SRC | \
299 BPF_ALU_SANITIZE_DST)
301 struct bpf_insn_aux_data {
303 enum bpf_reg_type ptr_type; /* pointer type for load/store insns */
304 unsigned long map_ptr_state; /* pointer/poison value for maps */
305 s32 call_imm; /* saved imm field of call insn */
306 u32 alu_limit; /* limit for add/sub register with pointer */
308 u32 map_index; /* index into used_maps[] */
309 u32 map_off; /* offset from value base address */
312 u64 map_key_state; /* constant (32 bit) key tracking for maps */
313 int ctx_field_size; /* the ctx field size for load insn, maybe 0 */
314 int sanitize_stack_off; /* stack slot to be cleared */
315 u32 seen; /* this insn was processed by the verifier at env->pass_cnt */
316 bool zext_dst; /* this insn zero extends dst reg */
317 u8 alu_state; /* used in combination with alu_limit */
319 /* below fields are initialized once */
320 unsigned int orig_idx; /* original instruction index */
324 #define MAX_USED_MAPS 64 /* max number of maps accessed by one eBPF program */
326 #define BPF_VERIFIER_TMP_LOG_SIZE 1024
328 struct bpf_verifier_log {
330 char kbuf[BPF_VERIFIER_TMP_LOG_SIZE];
336 static inline bool bpf_verifier_log_full(const struct bpf_verifier_log *log)
338 return log->len_used >= log->len_total - 1;
341 #define BPF_LOG_LEVEL1 1
342 #define BPF_LOG_LEVEL2 2
343 #define BPF_LOG_STATS 4
344 #define BPF_LOG_LEVEL (BPF_LOG_LEVEL1 | BPF_LOG_LEVEL2)
345 #define BPF_LOG_MASK (BPF_LOG_LEVEL | BPF_LOG_STATS)
346 #define BPF_LOG_KERNEL (BPF_LOG_MASK + 1) /* kernel internal flag */
348 static inline bool bpf_verifier_log_needed(const struct bpf_verifier_log *log)
350 return (log->level && log->ubuf && !bpf_verifier_log_full(log)) ||
351 log->level == BPF_LOG_KERNEL;
354 #define BPF_MAX_SUBPROGS 256
356 struct bpf_subprog_info {
357 /* 'start' has to be the first field otherwise find_subprog() won't work */
358 u32 start; /* insn idx of function entry point */
359 u32 linfo_idx; /* The idx to the main_prog->aux->linfo */
360 u16 stack_depth; /* max. stack depth used by this function */
363 /* single container for all structs
364 * one verifier_env per bpf_check() call
366 struct bpf_verifier_env {
369 struct bpf_prog *prog; /* eBPF program being verified */
370 const struct bpf_verifier_ops *ops;
371 struct bpf_verifier_stack_elem *head; /* stack of verifier states to be processed */
372 int stack_size; /* number of states to be processed */
373 bool strict_alignment; /* perform strict pointer alignment checks */
374 bool test_state_freq; /* test verifier with different pruning frequency */
375 struct bpf_verifier_state *cur_state; /* current verifier state */
376 struct bpf_verifier_state_list **explored_states; /* search pruning optimization */
377 struct bpf_verifier_state_list *free_list;
378 struct bpf_map *used_maps[MAX_USED_MAPS]; /* array of map's used by eBPF program */
379 u32 used_map_cnt; /* number of used maps */
380 u32 id_gen; /* used to generate unique reg IDs */
381 bool allow_ptr_leaks;
382 bool allow_ptr_to_map_access;
386 bool seen_direct_write;
387 struct bpf_insn_aux_data *insn_aux_data; /* array of per-insn state */
388 const struct bpf_line_info *prev_linfo;
389 struct bpf_verifier_log log;
390 struct bpf_subprog_info subprog_info[BPF_MAX_SUBPROGS + 1];
396 u32 pass_cnt; /* number of times do_check() was called */
398 /* number of instructions analyzed by the verifier */
399 u32 prev_insn_processed, insn_processed;
400 /* number of jmps, calls, exits analyzed so far */
401 u32 prev_jmps_processed, jmps_processed;
402 /* total verification time */
403 u64 verification_time;
404 /* maximum number of verifier states kept in 'branching' instructions */
405 u32 max_states_per_insn;
406 /* total number of allocated verifier states */
408 /* some states are freed during program analysis.
409 * this is peak number of states. this number dominates kernel
410 * memory consumption during verification
413 /* longest register parentage chain walked for liveness marking */
414 u32 longest_mark_read_walk;
417 __printf(2, 0) void bpf_verifier_vlog(struct bpf_verifier_log *log,
418 const char *fmt, va_list args);
419 __printf(2, 3) void bpf_verifier_log_write(struct bpf_verifier_env *env,
420 const char *fmt, ...);
421 __printf(2, 3) void bpf_log(struct bpf_verifier_log *log,
422 const char *fmt, ...);
424 static inline struct bpf_func_state *cur_func(struct bpf_verifier_env *env)
426 struct bpf_verifier_state *cur = env->cur_state;
428 return cur->frame[cur->curframe];
431 static inline struct bpf_reg_state *cur_regs(struct bpf_verifier_env *env)
433 return cur_func(env)->regs;
436 int bpf_prog_offload_verifier_prep(struct bpf_prog *prog);
437 int bpf_prog_offload_verify_insn(struct bpf_verifier_env *env,
438 int insn_idx, int prev_insn_idx);
439 int bpf_prog_offload_finalize(struct bpf_verifier_env *env);
441 bpf_prog_offload_replace_insn(struct bpf_verifier_env *env, u32 off,
442 struct bpf_insn *insn);
444 bpf_prog_offload_remove_insns(struct bpf_verifier_env *env, u32 off, u32 cnt);
446 int check_ctx_reg(struct bpf_verifier_env *env,
447 const struct bpf_reg_state *reg, int regno);
449 #endif /* _LINUX_BPF_VERIFIER_H */