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/btf.h> /* for struct btf and btf_id() */
9 #include <linux/filter.h> /* for MAX_BPF_STACK */
10 #include <linux/tnum.h>
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.
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.
20 #define BPF_MAX_VAR_SIZ (1 << 29)
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.
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 */
44 struct bpf_reg_state
{
45 /* Ordering of fields matters. See states_equal() */
46 enum bpf_reg_type type
;
47 /* Fixed part of pointer offset, pointer types only */
50 /* valid when type == PTR_TO_PACKET */
53 /* valid when type == CONST_PTR_TO_MAP | PTR_TO_MAP_VALUE |
54 * PTR_TO_MAP_VALUE_OR_NULL
57 struct bpf_map
*map_ptr
;
58 /* To distinguish map lookups from outer map
59 * the map_uid is non-zero for registers
60 * pointing to inner maps.
65 /* for PTR_TO_BTF_ID */
71 u32 mem_size
; /* for PTR_TO_MEM | PTR_TO_MEM_OR_NULL */
73 /* Max size from any of the above. */
79 u32 subprogno
; /* for PTR_TO_FUNC */
81 /* For PTR_TO_PACKET, used to find other pointers with the same variable
82 * offset, so they can share range knowledge.
83 * For PTR_TO_MAP_VALUE_OR_NULL this is used to share which map value we
84 * came from, when one is tested for != NULL.
85 * For PTR_TO_MEM_OR_NULL this is used to identify memory allocation
86 * for the purpose of tracking that it's freed.
87 * For PTR_TO_SOCKET this is used to share which pointers retain the
88 * same reference to the socket, to determine proper reference freeing.
91 /* PTR_TO_SOCKET and PTR_TO_TCP_SOCK could be a ptr returned
92 * from a pointer-cast helper, bpf_sk_fullsock() and
95 * Consider the following where "sk" is a reference counted
96 * pointer returned from "sk = bpf_sk_lookup_tcp();":
98 * 1: sk = bpf_sk_lookup_tcp();
99 * 2: if (!sk) { return 0; }
100 * 3: fullsock = bpf_sk_fullsock(sk);
101 * 4: if (!fullsock) { bpf_sk_release(sk); return 0; }
102 * 5: tp = bpf_tcp_sock(fullsock);
103 * 6: if (!tp) { bpf_sk_release(sk); return 0; }
104 * 7: bpf_sk_release(sk);
105 * 8: snd_cwnd = tp->snd_cwnd; // verifier will complain
107 * After bpf_sk_release(sk) at line 7, both "fullsock" ptr and
108 * "tp" ptr should be invalidated also. In order to do that,
109 * the reg holding "fullsock" and "sk" need to remember
110 * the original refcounted ptr id (i.e. sk_reg->id) in ref_obj_id
111 * such that the verifier can reset all regs which have
112 * ref_obj_id matching the sk_reg->id.
114 * sk_reg->ref_obj_id is set to sk_reg->id at line 1.
115 * sk_reg->id will stay as NULL-marking purpose only.
116 * After NULL-marking is done, sk_reg->id can be reset to 0.
118 * After "fullsock = bpf_sk_fullsock(sk);" at line 3,
119 * fullsock_reg->ref_obj_id is set to sk_reg->ref_obj_id.
121 * After "tp = bpf_tcp_sock(fullsock);" at line 5,
122 * tp_reg->ref_obj_id is set to fullsock_reg->ref_obj_id
123 * which is the same as sk_reg->ref_obj_id.
125 * From the verifier perspective, if sk, fullsock and tp
126 * are not NULL, they are the same ptr with different
127 * reg->type. In particular, bpf_sk_release(tp) is also
128 * allowed and has the same effect as bpf_sk_release(sk).
131 /* For scalar types (SCALAR_VALUE), this represents our knowledge of
133 * For pointer types, this represents the variable part of the offset
134 * from the pointed-to object, and is shared with all bpf_reg_states
135 * with the same id as us.
138 /* Used to determine if any memory access using this register will
139 * result in a bad access.
140 * These refer to the same value as var_off, not necessarily the actual
141 * contents of the register.
143 s64 smin_value
; /* minimum possible (s64)value */
144 s64 smax_value
; /* maximum possible (s64)value */
145 u64 umin_value
; /* minimum possible (u64)value */
146 u64 umax_value
; /* maximum possible (u64)value */
147 s32 s32_min_value
; /* minimum possible (s32)value */
148 s32 s32_max_value
; /* maximum possible (s32)value */
149 u32 u32_min_value
; /* minimum possible (u32)value */
150 u32 u32_max_value
; /* maximum possible (u32)value */
151 /* parentage chain for liveness checking */
152 struct bpf_reg_state
*parent
;
153 /* Inside the callee two registers can be both PTR_TO_STACK like
154 * R1=fp-8 and R2=fp-8, but one of them points to this function stack
155 * while another to the caller's stack. To differentiate them 'frameno'
156 * is used which is an index in bpf_verifier_state->frame[] array
157 * pointing to bpf_func_state.
160 /* Tracks subreg definition. The stored value is the insn_idx of the
161 * writing insn. This is safe because subreg_def is used before any insn
162 * patching which only happens after main verification finished.
165 enum bpf_reg_liveness live
;
166 /* if (!precise && SCALAR_VALUE) min/max/tnum don't affect safety */
170 enum bpf_stack_slot_type
{
171 STACK_INVALID
, /* nothing was stored in this stack slot */
172 STACK_SPILL
, /* register spilled into stack */
173 STACK_MISC
, /* BPF program wrote some data into this slot */
174 STACK_ZERO
, /* BPF program wrote constant zero */
177 #define BPF_REG_SIZE 8 /* size of eBPF register in bytes */
179 struct bpf_stack_state
{
180 struct bpf_reg_state spilled_ptr
;
181 u8 slot_type
[BPF_REG_SIZE
];
184 struct bpf_reference_state
{
185 /* Track each reference created with a unique id, even if the same
186 * instruction creates the reference multiple times (eg, via CALL).
189 /* Instruction where the allocation of this reference occurred. This
190 * is used purely to inform the user of a reference leak.
195 /* state of the program:
196 * type of all registers and stack info
198 struct bpf_func_state
{
199 struct bpf_reg_state regs
[MAX_BPF_REG
];
200 /* index of call instruction that called into this func */
202 /* stack frame number of this function state from pov of
203 * enclosing bpf_verifier_state.
204 * 0 = main function, 1 = first callee.
207 /* subprog number == index within subprog_info
208 * zero == main subprog
211 /* Every bpf_timer_start will increment async_entry_cnt.
212 * It's used to distinguish:
213 * void foo(void) { for(;;); }
214 * void foo(void) { bpf_timer_set_callback(,foo); }
218 bool in_async_callback_fn
;
220 /* The following fields should be last. See copy_func_state() */
222 struct bpf_reference_state
*refs
;
224 struct bpf_stack_state
*stack
;
227 struct bpf_idx_pair
{
237 /* Maximum number of register states that can exist at once */
238 #define BPF_ID_MAP_SIZE (MAX_BPF_REG + MAX_BPF_STACK / BPF_REG_SIZE)
239 #define MAX_CALL_FRAMES 8
240 struct bpf_verifier_state
{
241 /* call stack tracking */
242 struct bpf_func_state
*frame
[MAX_CALL_FRAMES
];
243 struct bpf_verifier_state
*parent
;
245 * 'branches' field is the number of branches left to explore:
246 * 0 - all possible paths from this state reached bpf_exit or
248 * 1 - at least one path is being explored.
249 * This state hasn't reached bpf_exit
250 * 2 - at least two paths are being explored.
251 * This state is an immediate parent of two children.
252 * One is fallthrough branch with branches==1 and another
253 * state is pushed into stack (to be explored later) also with
254 * branches==1. The parent of this state has branches==1.
255 * The verifier state tree connected via 'parent' pointer looks like:
258 * 2 -> 1 (first 'if' pushed into stack)
260 * 2 -> 1 (second 'if' pushed into stack)
265 * Once do_check() reaches bpf_exit, it calls update_branch_counts()
266 * and the verifier state tree will look:
269 * 2 -> 1 (first 'if' pushed into stack)
271 * 1 -> 1 (second 'if' pushed into stack)
275 * After pop_stack() the do_check() will resume at second 'if'.
277 * If is_state_visited() sees a state with branches > 0 it means
278 * there is a loop. If such state is exactly equal to the current state
279 * it's an infinite loop. Note states_equal() checks for states
280 * equvalency, so two states being 'states_equal' does not mean
281 * infinite loop. The exact comparison is provided by
282 * states_maybe_looping() function. It's a stronger pre-check and
283 * much faster than states_equal().
285 * This algorithm may not find all possible infinite loops or
286 * loop iteration count may be too high.
287 * In such cases BPF_COMPLEXITY_LIMIT_INSNS limit kicks in.
292 u32 active_spin_lock
;
295 /* first and last insn idx of this verifier state */
298 /* jmp history recorded from first to last.
299 * backtracking is using it to go from last to first.
300 * For most states jmp_history_cnt is [0-3].
301 * For loops can go up to ~40.
303 struct bpf_idx_pair
*jmp_history
;
307 #define bpf_get_spilled_reg(slot, frame) \
308 (((slot < frame->allocated_stack / BPF_REG_SIZE) && \
309 (frame->stack[slot].slot_type[0] == STACK_SPILL)) \
310 ? &frame->stack[slot].spilled_ptr : NULL)
312 /* Iterate over 'frame', setting 'reg' to either NULL or a spilled register. */
313 #define bpf_for_each_spilled_reg(iter, frame, reg) \
314 for (iter = 0, reg = bpf_get_spilled_reg(iter, frame); \
315 iter < frame->allocated_stack / BPF_REG_SIZE; \
316 iter++, reg = bpf_get_spilled_reg(iter, frame))
318 /* linked list of verifier states used to prune search */
319 struct bpf_verifier_state_list
{
320 struct bpf_verifier_state state
;
321 struct bpf_verifier_state_list
*next
;
322 int miss_cnt
, hit_cnt
;
325 /* Possible states for alu_state member. */
326 #define BPF_ALU_SANITIZE_SRC (1U << 0)
327 #define BPF_ALU_SANITIZE_DST (1U << 1)
328 #define BPF_ALU_NEG_VALUE (1U << 2)
329 #define BPF_ALU_NON_POINTER (1U << 3)
330 #define BPF_ALU_IMMEDIATE (1U << 4)
331 #define BPF_ALU_SANITIZE (BPF_ALU_SANITIZE_SRC | \
332 BPF_ALU_SANITIZE_DST)
334 struct bpf_insn_aux_data
{
336 enum bpf_reg_type ptr_type
; /* pointer type for load/store insns */
337 unsigned long map_ptr_state
; /* pointer/poison value for maps */
338 s32 call_imm
; /* saved imm field of call insn */
339 u32 alu_limit
; /* limit for add/sub register with pointer */
341 u32 map_index
; /* index into used_maps[] */
342 u32 map_off
; /* offset from value base address */
345 enum bpf_reg_type reg_type
; /* type of pseudo_btf_id */
349 u32 btf_id
; /* btf_id for struct typed var */
351 u32 mem_size
; /* mem_size for non-struct typed var */
355 u64 map_key_state
; /* constant (32 bit) key tracking for maps */
356 int ctx_field_size
; /* the ctx field size for load insn, maybe 0 */
357 u32 seen
; /* this insn was processed by the verifier at env->pass_cnt */
358 bool sanitize_stack_spill
; /* subject to Spectre v4 sanitation */
359 bool zext_dst
; /* this insn zero extends dst reg */
360 u8 alu_state
; /* used in combination with alu_limit */
362 /* below fields are initialized once */
363 unsigned int orig_idx
; /* original instruction index */
367 #define MAX_USED_MAPS 64 /* max number of maps accessed by one eBPF program */
368 #define MAX_USED_BTFS 64 /* max number of BTFs accessed by one BPF program */
370 #define BPF_VERIFIER_TMP_LOG_SIZE 1024
372 struct bpf_verifier_log
{
374 char kbuf
[BPF_VERIFIER_TMP_LOG_SIZE
];
380 static inline bool bpf_verifier_log_full(const struct bpf_verifier_log
*log
)
382 return log
->len_used
>= log
->len_total
- 1;
385 #define BPF_LOG_LEVEL1 1
386 #define BPF_LOG_LEVEL2 2
387 #define BPF_LOG_STATS 4
388 #define BPF_LOG_LEVEL (BPF_LOG_LEVEL1 | BPF_LOG_LEVEL2)
389 #define BPF_LOG_MASK (BPF_LOG_LEVEL | BPF_LOG_STATS)
390 #define BPF_LOG_KERNEL (BPF_LOG_MASK + 1) /* kernel internal flag */
392 static inline bool bpf_verifier_log_needed(const struct bpf_verifier_log
*log
)
395 ((log
->level
&& log
->ubuf
&& !bpf_verifier_log_full(log
)) ||
396 log
->level
== BPF_LOG_KERNEL
);
399 #define BPF_MAX_SUBPROGS 256
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 */
407 bool tail_call_reachable
;
412 /* single container for all structs
413 * one verifier_env per bpf_check() call
415 struct bpf_verifier_env
{
418 struct bpf_prog
*prog
; /* eBPF program being verified */
419 const struct bpf_verifier_ops
*ops
;
420 struct bpf_verifier_stack_elem
*head
; /* stack of verifier states to be processed */
421 int stack_size
; /* number of states to be processed */
422 bool strict_alignment
; /* perform strict pointer alignment checks */
423 bool test_state_freq
; /* test verifier with different pruning frequency */
424 struct bpf_verifier_state
*cur_state
; /* current verifier state */
425 struct bpf_verifier_state_list
**explored_states
; /* search pruning optimization */
426 struct bpf_verifier_state_list
*free_list
;
427 struct bpf_map
*used_maps
[MAX_USED_MAPS
]; /* array of map's used by eBPF program */
428 struct btf_mod_pair used_btfs
[MAX_USED_BTFS
]; /* array of BTF's used by BPF program */
429 u32 used_map_cnt
; /* number of used maps */
430 u32 used_btf_cnt
; /* number of used BTF objects */
431 u32 id_gen
; /* used to generate unique reg IDs */
432 bool explore_alu_limits
;
433 bool allow_ptr_leaks
;
434 bool allow_uninit_stack
;
435 bool allow_ptr_to_map_access
;
439 bool seen_direct_write
;
440 struct bpf_insn_aux_data
*insn_aux_data
; /* array of per-insn state */
441 const struct bpf_line_info
*prev_linfo
;
442 struct bpf_verifier_log log
;
443 struct bpf_subprog_info subprog_info
[BPF_MAX_SUBPROGS
+ 1];
444 struct bpf_id_pair idmap_scratch
[BPF_ID_MAP_SIZE
];
450 u32 pass_cnt
; /* number of times do_check() was called */
452 /* number of instructions analyzed by the verifier */
453 u32 prev_insn_processed
, insn_processed
;
454 /* number of jmps, calls, exits analyzed so far */
455 u32 prev_jmps_processed
, jmps_processed
;
456 /* total verification time */
457 u64 verification_time
;
458 /* maximum number of verifier states kept in 'branching' instructions */
459 u32 max_states_per_insn
;
460 /* total number of allocated verifier states */
462 /* some states are freed during program analysis.
463 * this is peak number of states. this number dominates kernel
464 * memory consumption during verification
467 /* longest register parentage chain walked for liveness marking */
468 u32 longest_mark_read_walk
;
472 __printf(2, 0) void bpf_verifier_vlog(struct bpf_verifier_log
*log
,
473 const char *fmt
, va_list args
);
474 __printf(2, 3) void bpf_verifier_log_write(struct bpf_verifier_env
*env
,
475 const char *fmt
, ...);
476 __printf(2, 3) void bpf_log(struct bpf_verifier_log
*log
,
477 const char *fmt
, ...);
479 static inline struct bpf_func_state
*cur_func(struct bpf_verifier_env
*env
)
481 struct bpf_verifier_state
*cur
= env
->cur_state
;
483 return cur
->frame
[cur
->curframe
];
486 static inline struct bpf_reg_state
*cur_regs(struct bpf_verifier_env
*env
)
488 return cur_func(env
)->regs
;
491 int bpf_prog_offload_verifier_prep(struct bpf_prog
*prog
);
492 int bpf_prog_offload_verify_insn(struct bpf_verifier_env
*env
,
493 int insn_idx
, int prev_insn_idx
);
494 int bpf_prog_offload_finalize(struct bpf_verifier_env
*env
);
496 bpf_prog_offload_replace_insn(struct bpf_verifier_env
*env
, u32 off
,
497 struct bpf_insn
*insn
);
499 bpf_prog_offload_remove_insns(struct bpf_verifier_env
*env
, u32 off
, u32 cnt
);
501 int check_ctx_reg(struct bpf_verifier_env
*env
,
502 const struct bpf_reg_state
*reg
, int regno
);
503 int check_mem_reg(struct bpf_verifier_env
*env
, struct bpf_reg_state
*reg
,
504 u32 regno
, u32 mem_size
);
506 /* this lives here instead of in bpf.h because it needs to dereference tgt_prog */
507 static inline u64
bpf_trampoline_compute_key(const struct bpf_prog
*tgt_prog
,
508 struct btf
*btf
, u32 btf_id
)
511 return ((u64
)tgt_prog
->aux
->id
<< 32) | btf_id
;
513 return ((u64
)btf_obj_id(btf
) << 32) | 0x80000000 | btf_id
;
516 /* unpack the IDs from the key as constructed above */
517 static inline void bpf_trampoline_unpack_key(u64 key
, u32
*obj_id
, u32
*btf_id
)
522 *btf_id
= key
& 0x7FFFFFFF;
525 int bpf_check_attach_target(struct bpf_verifier_log
*log
,
526 const struct bpf_prog
*prog
,
527 const struct bpf_prog
*tgt_prog
,
529 struct bpf_attach_target_info
*tgt_info
);
531 #endif /* _LINUX_BPF_VERIFIER_H */