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/btf.h> /* for struct btf and btf_id() */
9 #include <linux/filter.h> /* for MAX_BPF_STACK */
10 #include <linux/tnum.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 /* size of tmp_str_buf in bpf_verifier.
22 * we need at least 306 bytes to fit full stack mask representation
23 * (in the "-8,-16,...,-512" form)
24 */
25 #define TMP_STR_BUF_LEN 320
26 /* Patch buffer size */
27 #define INSN_BUF_SIZE 32
28
29 /* Liveness marks, used for registers and spilled-regs (in stack slots).
30 * Read marks propagate upwards until they find a write mark; they record that
31 * "one of this state's descendants read this reg" (and therefore the reg is
32 * relevant for states_equal() checks).
33 * Write marks collect downwards and do not propagate; they record that "the
34 * straight-line code that reached this state (from its parent) wrote this reg"
35 * (and therefore that reads propagated from this state or its descendants
36 * should not propagate to its parent).
37 * A state with a write mark can receive read marks; it just won't propagate
38 * them to its parent, since the write mark is a property, not of the state,
39 * but of the link between it and its parent. See mark_reg_read() and
40 * mark_stack_slot_read() in kernel/bpf/verifier.c.
41 */
42 enum bpf_reg_liveness {
43 REG_LIVE_NONE = 0, /* reg hasn't been read or written this branch */
44 REG_LIVE_READ32 = 0x1, /* reg was read, so we're sensitive to initial value */
45 REG_LIVE_READ64 = 0x2, /* likewise, but full 64-bit content matters */
46 REG_LIVE_READ = REG_LIVE_READ32 | REG_LIVE_READ64,
47 REG_LIVE_WRITTEN = 0x4, /* reg was written first, screening off later reads */
48 REG_LIVE_DONE = 0x8, /* liveness won't be updating this register anymore */
49 };
50
51 #define ITER_PREFIX "bpf_iter_"
52
53 enum bpf_iter_state {
54 BPF_ITER_STATE_INVALID, /* for non-first slot */
55 BPF_ITER_STATE_ACTIVE,
56 BPF_ITER_STATE_DRAINED,
57 };
58
59 struct bpf_reg_state {
60 /* Ordering of fields matters. See states_equal() */
61 enum bpf_reg_type type;
62 /*
63 * Fixed part of pointer offset, pointer types only.
64 * Or constant delta between "linked" scalars with the same ID.
65 */
66 s32 off;
67 union {
68 /* valid when type == PTR_TO_PACKET */
69 int range;
70
71 /* valid when type == CONST_PTR_TO_MAP | PTR_TO_MAP_VALUE |
72 * PTR_TO_MAP_VALUE_OR_NULL
73 */
74 struct {
75 struct bpf_map *map_ptr;
76 /* To distinguish map lookups from outer map
77 * the map_uid is non-zero for registers
78 * pointing to inner maps.
79 */
80 u32 map_uid;
81 };
82
83 /* for PTR_TO_BTF_ID */
84 struct {
85 struct btf *btf;
86 u32 btf_id;
87 };
88
89 struct { /* for PTR_TO_MEM | PTR_TO_MEM_OR_NULL */
90 u32 mem_size;
91 u32 dynptr_id; /* for dynptr slices */
92 };
93
94 /* For dynptr stack slots */
95 struct {
96 enum bpf_dynptr_type type;
97 /* A dynptr is 16 bytes so it takes up 2 stack slots.
98 * We need to track which slot is the first slot
99 * to protect against cases where the user may try to
100 * pass in an address starting at the second slot of the
101 * dynptr.
102 */
103 bool first_slot;
104 } dynptr;
105
106 /* For bpf_iter stack slots */
107 struct {
108 /* BTF container and BTF type ID describing
109 * struct bpf_iter_<type> of an iterator state
110 */
111 struct btf *btf;
112 u32 btf_id;
113 /* packing following two fields to fit iter state into 16 bytes */
114 enum bpf_iter_state state:2;
115 int depth:30;
116 } iter;
117
118 /* Max size from any of the above. */
119 struct {
120 unsigned long raw1;
121 unsigned long raw2;
122 } raw;
123
124 u32 subprogno; /* for PTR_TO_FUNC */
125 };
126 /* For scalar types (SCALAR_VALUE), this represents our knowledge of
127 * the actual value.
128 * For pointer types, this represents the variable part of the offset
129 * from the pointed-to object, and is shared with all bpf_reg_states
130 * with the same id as us.
131 */
132 struct tnum var_off;
133 /* Used to determine if any memory access using this register will
134 * result in a bad access.
135 * These refer to the same value as var_off, not necessarily the actual
136 * contents of the register.
137 */
138 s64 smin_value; /* minimum possible (s64)value */
139 s64 smax_value; /* maximum possible (s64)value */
140 u64 umin_value; /* minimum possible (u64)value */
141 u64 umax_value; /* maximum possible (u64)value */
142 s32 s32_min_value; /* minimum possible (s32)value */
143 s32 s32_max_value; /* maximum possible (s32)value */
144 u32 u32_min_value; /* minimum possible (u32)value */
145 u32 u32_max_value; /* maximum possible (u32)value */
146 /* For PTR_TO_PACKET, used to find other pointers with the same variable
147 * offset, so they can share range knowledge.
148 * For PTR_TO_MAP_VALUE_OR_NULL this is used to share which map value we
149 * came from, when one is tested for != NULL.
150 * For PTR_TO_MEM_OR_NULL this is used to identify memory allocation
151 * for the purpose of tracking that it's freed.
152 * For PTR_TO_SOCKET this is used to share which pointers retain the
153 * same reference to the socket, to determine proper reference freeing.
154 * For stack slots that are dynptrs, this is used to track references to
155 * the dynptr to determine proper reference freeing.
156 * Similarly to dynptrs, we use ID to track "belonging" of a reference
157 * to a specific instance of bpf_iter.
158 */
159 /*
160 * Upper bit of ID is used to remember relationship between "linked"
161 * registers. Example:
162 * r1 = r2; both will have r1->id == r2->id == N
163 * r1 += 10; r1->id == N | BPF_ADD_CONST and r1->off == 10
164 */
165 #define BPF_ADD_CONST (1U << 31)
166 u32 id;
167 /* PTR_TO_SOCKET and PTR_TO_TCP_SOCK could be a ptr returned
168 * from a pointer-cast helper, bpf_sk_fullsock() and
169 * bpf_tcp_sock().
170 *
171 * Consider the following where "sk" is a reference counted
172 * pointer returned from "sk = bpf_sk_lookup_tcp();":
173 *
174 * 1: sk = bpf_sk_lookup_tcp();
175 * 2: if (!sk) { return 0; }
176 * 3: fullsock = bpf_sk_fullsock(sk);
177 * 4: if (!fullsock) { bpf_sk_release(sk); return 0; }
178 * 5: tp = bpf_tcp_sock(fullsock);
179 * 6: if (!tp) { bpf_sk_release(sk); return 0; }
180 * 7: bpf_sk_release(sk);
181 * 8: snd_cwnd = tp->snd_cwnd; // verifier will complain
182 *
183 * After bpf_sk_release(sk) at line 7, both "fullsock" ptr and
184 * "tp" ptr should be invalidated also. In order to do that,
185 * the reg holding "fullsock" and "sk" need to remember
186 * the original refcounted ptr id (i.e. sk_reg->id) in ref_obj_id
187 * such that the verifier can reset all regs which have
188 * ref_obj_id matching the sk_reg->id.
189 *
190 * sk_reg->ref_obj_id is set to sk_reg->id at line 1.
191 * sk_reg->id will stay as NULL-marking purpose only.
192 * After NULL-marking is done, sk_reg->id can be reset to 0.
193 *
194 * After "fullsock = bpf_sk_fullsock(sk);" at line 3,
195 * fullsock_reg->ref_obj_id is set to sk_reg->ref_obj_id.
196 *
197 * After "tp = bpf_tcp_sock(fullsock);" at line 5,
198 * tp_reg->ref_obj_id is set to fullsock_reg->ref_obj_id
199 * which is the same as sk_reg->ref_obj_id.
200 *
201 * From the verifier perspective, if sk, fullsock and tp
202 * are not NULL, they are the same ptr with different
203 * reg->type. In particular, bpf_sk_release(tp) is also
204 * allowed and has the same effect as bpf_sk_release(sk).
205 */
206 u32 ref_obj_id;
207 /* parentage chain for liveness checking */
208 struct bpf_reg_state *parent;
209 /* Inside the callee two registers can be both PTR_TO_STACK like
210 * R1=fp-8 and R2=fp-8, but one of them points to this function stack
211 * while another to the caller's stack. To differentiate them 'frameno'
212 * is used which is an index in bpf_verifier_state->frame[] array
213 * pointing to bpf_func_state.
214 */
215 u32 frameno;
216 /* Tracks subreg definition. The stored value is the insn_idx of the
217 * writing insn. This is safe because subreg_def is used before any insn
218 * patching which only happens after main verification finished.
219 */
220 s32 subreg_def;
221 enum bpf_reg_liveness live;
222 /* if (!precise && SCALAR_VALUE) min/max/tnum don't affect safety */
223 bool precise;
224 };
225
226 enum bpf_stack_slot_type {
227 STACK_INVALID, /* nothing was stored in this stack slot */
228 STACK_SPILL, /* register spilled into stack */
229 STACK_MISC, /* BPF program wrote some data into this slot */
230 STACK_ZERO, /* BPF program wrote constant zero */
231 /* A dynptr is stored in this stack slot. The type of dynptr
232 * is stored in bpf_stack_state->spilled_ptr.dynptr.type
233 */
234 STACK_DYNPTR,
235 STACK_ITER,
236 STACK_IRQ_FLAG,
237 };
238
239 #define BPF_REG_SIZE 8 /* size of eBPF register in bytes */
240
241 #define BPF_REGMASK_ARGS ((1 << BPF_REG_1) | (1 << BPF_REG_2) | \
242 (1 << BPF_REG_3) | (1 << BPF_REG_4) | \
243 (1 << BPF_REG_5))
244
245 #define BPF_DYNPTR_SIZE sizeof(struct bpf_dynptr_kern)
246 #define BPF_DYNPTR_NR_SLOTS (BPF_DYNPTR_SIZE / BPF_REG_SIZE)
247
248 struct bpf_stack_state {
249 struct bpf_reg_state spilled_ptr;
250 u8 slot_type[BPF_REG_SIZE];
251 };
252
253 struct bpf_reference_state {
254 /* Each reference object has a type. Ensure REF_TYPE_PTR is zero to
255 * default to pointer reference on zero initialization of a state.
256 */
257 enum ref_state_type {
258 REF_TYPE_PTR = 1,
259 REF_TYPE_IRQ = 2,
260 REF_TYPE_LOCK = 3,
261 } type;
262 /* Track each reference created with a unique id, even if the same
263 * instruction creates the reference multiple times (eg, via CALL).
264 */
265 int id;
266 /* Instruction where the allocation of this reference occurred. This
267 * is used purely to inform the user of a reference leak.
268 */
269 int insn_idx;
270 /* Use to keep track of the source object of a lock, to ensure
271 * it matches on unlock.
272 */
273 void *ptr;
274 };
275
276 struct bpf_retval_range {
277 s32 minval;
278 s32 maxval;
279 };
280
281 /* state of the program:
282 * type of all registers and stack info
283 */
284 struct bpf_func_state {
285 struct bpf_reg_state regs[MAX_BPF_REG];
286 /* index of call instruction that called into this func */
287 int callsite;
288 /* stack frame number of this function state from pov of
289 * enclosing bpf_verifier_state.
290 * 0 = main function, 1 = first callee.
291 */
292 u32 frameno;
293 /* subprog number == index within subprog_info
294 * zero == main subprog
295 */
296 u32 subprogno;
297 /* Every bpf_timer_start will increment async_entry_cnt.
298 * It's used to distinguish:
299 * void foo(void) { for(;;); }
300 * void foo(void) { bpf_timer_set_callback(,foo); }
301 */
302 u32 async_entry_cnt;
303 struct bpf_retval_range callback_ret_range;
304 bool in_callback_fn;
305 bool in_async_callback_fn;
306 bool in_exception_callback_fn;
307 /* For callback calling functions that limit number of possible
308 * callback executions (e.g. bpf_loop) keeps track of current
309 * simulated iteration number.
310 * Value in frame N refers to number of times callback with frame
311 * N+1 was simulated, e.g. for the following call:
312 *
313 * bpf_loop(..., fn, ...); | suppose current frame is N
314 * | fn would be simulated in frame N+1
315 * | number of simulations is tracked in frame N
316 */
317 u32 callback_depth;
318
319 /* The following fields should be last. See copy_func_state() */
320 /* The state of the stack. Each element of the array describes BPF_REG_SIZE
321 * (i.e. 8) bytes worth of stack memory.
322 * stack[0] represents bytes [*(r10-8)..*(r10-1)]
323 * stack[1] represents bytes [*(r10-16)..*(r10-9)]
324 * ...
325 * stack[allocated_stack/8 - 1] represents [*(r10-allocated_stack)..*(r10-allocated_stack+7)]
326 */
327 struct bpf_stack_state *stack;
328 /* Size of the current stack, in bytes. The stack state is tracked below, in
329 * `stack`. allocated_stack is always a multiple of BPF_REG_SIZE.
330 */
331 int allocated_stack;
332 };
333
334 #define MAX_CALL_FRAMES 8
335
336 /* instruction history flags, used in bpf_insn_hist_entry.flags field */
337 enum {
338 /* instruction references stack slot through PTR_TO_STACK register;
339 * we also store stack's frame number in lower 3 bits (MAX_CALL_FRAMES is 8)
340 * and accessed stack slot's index in next 6 bits (MAX_BPF_STACK is 512,
341 * 8 bytes per slot, so slot index (spi) is [0, 63])
342 */
343 INSN_F_FRAMENO_MASK = 0x7, /* 3 bits */
344
345 INSN_F_SPI_MASK = 0x3f, /* 6 bits */
346 INSN_F_SPI_SHIFT = 3, /* shifted 3 bits to the left */
347
348 INSN_F_STACK_ACCESS = BIT(9), /* we need 10 bits total */
349 };
350
351 static_assert(INSN_F_FRAMENO_MASK + 1 >= MAX_CALL_FRAMES);
352 static_assert(INSN_F_SPI_MASK + 1 >= MAX_BPF_STACK / 8);
353
354 struct bpf_insn_hist_entry {
355 u32 idx;
356 /* insn idx can't be bigger than 1 million */
357 u32 prev_idx : 22;
358 /* special flags, e.g., whether insn is doing register stack spill/load */
359 u32 flags : 10;
360 /* additional registers that need precision tracking when this
361 * jump is backtracked, vector of six 10-bit records
362 */
363 u64 linked_regs;
364 };
365
366 /* Maximum number of register states that can exist at once */
367 #define BPF_ID_MAP_SIZE ((MAX_BPF_REG + MAX_BPF_STACK / BPF_REG_SIZE) * MAX_CALL_FRAMES)
368 struct bpf_verifier_state {
369 /* call stack tracking */
370 struct bpf_func_state *frame[MAX_CALL_FRAMES];
371 struct bpf_verifier_state *parent;
372 /* Acquired reference states */
373 struct bpf_reference_state *refs;
374 /*
375 * 'branches' field is the number of branches left to explore:
376 * 0 - all possible paths from this state reached bpf_exit or
377 * were safely pruned
378 * 1 - at least one path is being explored.
379 * This state hasn't reached bpf_exit
380 * 2 - at least two paths are being explored.
381 * This state is an immediate parent of two children.
382 * One is fallthrough branch with branches==1 and another
383 * state is pushed into stack (to be explored later) also with
384 * branches==1. The parent of this state has branches==1.
385 * The verifier state tree connected via 'parent' pointer looks like:
386 * 1
387 * 1
388 * 2 -> 1 (first 'if' pushed into stack)
389 * 1
390 * 2 -> 1 (second 'if' pushed into stack)
391 * 1
392 * 1
393 * 1 bpf_exit.
394 *
395 * Once do_check() reaches bpf_exit, it calls update_branch_counts()
396 * and the verifier state tree will look:
397 * 1
398 * 1
399 * 2 -> 1 (first 'if' pushed into stack)
400 * 1
401 * 1 -> 1 (second 'if' pushed into stack)
402 * 0
403 * 0
404 * 0 bpf_exit.
405 * After pop_stack() the do_check() will resume at second 'if'.
406 *
407 * If is_state_visited() sees a state with branches > 0 it means
408 * there is a loop. If such state is exactly equal to the current state
409 * it's an infinite loop. Note states_equal() checks for states
410 * equivalency, so two states being 'states_equal' does not mean
411 * infinite loop. The exact comparison is provided by
412 * states_maybe_looping() function. It's a stronger pre-check and
413 * much faster than states_equal().
414 *
415 * This algorithm may not find all possible infinite loops or
416 * loop iteration count may be too high.
417 * In such cases BPF_COMPLEXITY_LIMIT_INSNS limit kicks in.
418 */
419 u32 branches;
420 u32 insn_idx;
421 u32 curframe;
422
423 u32 acquired_refs;
424 u32 active_locks;
425 u32 active_preempt_locks;
426 u32 active_irq_id;
427 bool active_rcu_lock;
428
429 bool speculative;
430 /* If this state was ever pointed-to by other state's loop_entry field
431 * this flag would be set to true. Used to avoid freeing such states
432 * while they are still in use.
433 */
434 bool used_as_loop_entry;
435 bool in_sleepable;
436
437 /* first and last insn idx of this verifier state */
438 u32 first_insn_idx;
439 u32 last_insn_idx;
440 /* If this state is a part of states loop this field points to some
441 * parent of this state such that:
442 * - it is also a member of the same states loop;
443 * - DFS states traversal starting from initial state visits loop_entry
444 * state before this state.
445 * Used to compute topmost loop entry for state loops.
446 * State loops might appear because of open coded iterators logic.
447 * See get_loop_entry() for more information.
448 */
449 struct bpf_verifier_state *loop_entry;
450 /* Sub-range of env->insn_hist[] corresponding to this state's
451 * instruction history.
452 * Backtracking is using it to go from last to first.
453 * For most states instruction history is short, 0-3 instructions.
454 * For loops can go up to ~40.
455 */
456 u32 insn_hist_start;
457 u32 insn_hist_end;
458 u32 dfs_depth;
459 u32 callback_unroll_depth;
460 u32 may_goto_depth;
461 };
462
463 #define bpf_get_spilled_reg(slot, frame, mask) \
464 (((slot < frame->allocated_stack / BPF_REG_SIZE) && \
465 ((1 << frame->stack[slot].slot_type[BPF_REG_SIZE - 1]) & (mask))) \
466 ? &frame->stack[slot].spilled_ptr : NULL)
467
468 /* Iterate over 'frame', setting 'reg' to either NULL or a spilled register. */
469 #define bpf_for_each_spilled_reg(iter, frame, reg, mask) \
470 for (iter = 0, reg = bpf_get_spilled_reg(iter, frame, mask); \
471 iter < frame->allocated_stack / BPF_REG_SIZE; \
472 iter++, reg = bpf_get_spilled_reg(iter, frame, mask))
473
474 #define bpf_for_each_reg_in_vstate_mask(__vst, __state, __reg, __mask, __expr) \
475 ({ \
476 struct bpf_verifier_state *___vstate = __vst; \
477 int ___i, ___j; \
478 for (___i = 0; ___i <= ___vstate->curframe; ___i++) { \
479 struct bpf_reg_state *___regs; \
480 __state = ___vstate->frame[___i]; \
481 ___regs = __state->regs; \
482 for (___j = 0; ___j < MAX_BPF_REG; ___j++) { \
483 __reg = &___regs[___j]; \
484 (void)(__expr); \
485 } \
486 bpf_for_each_spilled_reg(___j, __state, __reg, __mask) { \
487 if (!__reg) \
488 continue; \
489 (void)(__expr); \
490 } \
491 } \
492 })
493
494 /* Invoke __expr over regsiters in __vst, setting __state and __reg */
495 #define bpf_for_each_reg_in_vstate(__vst, __state, __reg, __expr) \
496 bpf_for_each_reg_in_vstate_mask(__vst, __state, __reg, 1 << STACK_SPILL, __expr)
497
498 /* linked list of verifier states used to prune search */
499 struct bpf_verifier_state_list {
500 struct bpf_verifier_state state;
501 struct bpf_verifier_state_list *next;
502 int miss_cnt, hit_cnt;
503 };
504
505 struct bpf_loop_inline_state {
506 unsigned int initialized:1; /* set to true upon first entry */
507 unsigned int fit_for_inline:1; /* true if callback function is the same
508 * at each call and flags are always zero
509 */
510 u32 callback_subprogno; /* valid when fit_for_inline is true */
511 };
512
513 /* pointer and state for maps */
514 struct bpf_map_ptr_state {
515 struct bpf_map *map_ptr;
516 bool poison;
517 bool unpriv;
518 };
519
520 /* Possible states for alu_state member. */
521 #define BPF_ALU_SANITIZE_SRC (1U << 0)
522 #define BPF_ALU_SANITIZE_DST (1U << 1)
523 #define BPF_ALU_NEG_VALUE (1U << 2)
524 #define BPF_ALU_NON_POINTER (1U << 3)
525 #define BPF_ALU_IMMEDIATE (1U << 4)
526 #define BPF_ALU_SANITIZE (BPF_ALU_SANITIZE_SRC | \
527 BPF_ALU_SANITIZE_DST)
528
529 struct bpf_insn_aux_data {
530 union {
531 enum bpf_reg_type ptr_type; /* pointer type for load/store insns */
532 struct bpf_map_ptr_state map_ptr_state;
533 s32 call_imm; /* saved imm field of call insn */
534 u32 alu_limit; /* limit for add/sub register with pointer */
535 struct {
536 u32 map_index; /* index into used_maps[] */
537 u32 map_off; /* offset from value base address */
538 };
539 struct {
540 enum bpf_reg_type reg_type; /* type of pseudo_btf_id */
541 union {
542 struct {
543 struct btf *btf;
544 u32 btf_id; /* btf_id for struct typed var */
545 };
546 u32 mem_size; /* mem_size for non-struct typed var */
547 };
548 } btf_var;
549 /* if instruction is a call to bpf_loop this field tracks
550 * the state of the relevant registers to make decision about inlining
551 */
552 struct bpf_loop_inline_state loop_inline_state;
553 };
554 union {
555 /* remember the size of type passed to bpf_obj_new to rewrite R1 */
556 u64 obj_new_size;
557 /* remember the offset of node field within type to rewrite */
558 u64 insert_off;
559 };
560 struct btf_struct_meta *kptr_struct_meta;
561 u64 map_key_state; /* constant (32 bit) key tracking for maps */
562 int ctx_field_size; /* the ctx field size for load insn, maybe 0 */
563 u32 seen; /* this insn was processed by the verifier at env->pass_cnt */
564 bool sanitize_stack_spill; /* subject to Spectre v4 sanitation */
565 bool zext_dst; /* this insn zero extends dst reg */
566 bool needs_zext; /* alu op needs to clear upper bits */
567 bool storage_get_func_atomic; /* bpf_*_storage_get() with atomic memory alloc */
568 bool is_iter_next; /* bpf_iter_<type>_next() kfunc call */
569 bool call_with_percpu_alloc_ptr; /* {this,per}_cpu_ptr() with prog percpu alloc */
570 u8 alu_state; /* used in combination with alu_limit */
571 /* true if STX or LDX instruction is a part of a spill/fill
572 * pattern for a bpf_fastcall call.
573 */
574 u8 fastcall_pattern:1;
575 /* for CALL instructions, a number of spill/fill pairs in the
576 * bpf_fastcall pattern.
577 */
578 u8 fastcall_spills_num:3;
579
580 /* below fields are initialized once */
581 unsigned int orig_idx; /* original instruction index */
582 bool jmp_point;
583 bool prune_point;
584 /* ensure we check state equivalence and save state checkpoint and
585 * this instruction, regardless of any heuristics
586 */
587 bool force_checkpoint;
588 /* true if instruction is a call to a helper function that
589 * accepts callback function as a parameter.
590 */
591 bool calls_callback;
592 };
593
594 #define MAX_USED_MAPS 64 /* max number of maps accessed by one eBPF program */
595 #define MAX_USED_BTFS 64 /* max number of BTFs accessed by one BPF program */
596
597 #define BPF_VERIFIER_TMP_LOG_SIZE 1024
598
599 struct bpf_verifier_log {
600 /* Logical start and end positions of a "log window" of the verifier log.
601 * start_pos == 0 means we haven't truncated anything.
602 * Once truncation starts to happen, start_pos + len_total == end_pos,
603 * except during log reset situations, in which (end_pos - start_pos)
604 * might get smaller than len_total (see bpf_vlog_reset()).
605 * Generally, (end_pos - start_pos) gives number of useful data in
606 * user log buffer.
607 */
608 u64 start_pos;
609 u64 end_pos;
610 char __user *ubuf;
611 u32 level;
612 u32 len_total;
613 u32 len_max;
614 char kbuf[BPF_VERIFIER_TMP_LOG_SIZE];
615 };
616
617 #define BPF_LOG_LEVEL1 1
618 #define BPF_LOG_LEVEL2 2
619 #define BPF_LOG_STATS 4
620 #define BPF_LOG_FIXED 8
621 #define BPF_LOG_LEVEL (BPF_LOG_LEVEL1 | BPF_LOG_LEVEL2)
622 #define BPF_LOG_MASK (BPF_LOG_LEVEL | BPF_LOG_STATS | BPF_LOG_FIXED)
623 #define BPF_LOG_KERNEL (BPF_LOG_MASK + 1) /* kernel internal flag */
624 #define BPF_LOG_MIN_ALIGNMENT 8U
625 #define BPF_LOG_ALIGNMENT 40U
626
bpf_verifier_log_needed(const struct bpf_verifier_log * log)627 static inline bool bpf_verifier_log_needed(const struct bpf_verifier_log *log)
628 {
629 return log && log->level;
630 }
631
632 #define BPF_MAX_SUBPROGS 256
633
634 struct bpf_subprog_arg_info {
635 enum bpf_arg_type arg_type;
636 union {
637 u32 mem_size;
638 u32 btf_id;
639 };
640 };
641
642 enum priv_stack_mode {
643 PRIV_STACK_UNKNOWN,
644 NO_PRIV_STACK,
645 PRIV_STACK_ADAPTIVE,
646 };
647
648 struct bpf_subprog_info {
649 /* 'start' has to be the first field otherwise find_subprog() won't work */
650 u32 start; /* insn idx of function entry point */
651 u32 linfo_idx; /* The idx to the main_prog->aux->linfo */
652 u16 stack_depth; /* max. stack depth used by this function */
653 u16 stack_extra;
654 /* offsets in range [stack_depth .. fastcall_stack_off)
655 * are used for bpf_fastcall spills and fills.
656 */
657 s16 fastcall_stack_off;
658 bool has_tail_call: 1;
659 bool tail_call_reachable: 1;
660 bool has_ld_abs: 1;
661 bool is_cb: 1;
662 bool is_async_cb: 1;
663 bool is_exception_cb: 1;
664 bool args_cached: 1;
665 /* true if bpf_fastcall stack region is used by functions that can't be inlined */
666 bool keep_fastcall_stack: 1;
667 bool changes_pkt_data: 1;
668
669 enum priv_stack_mode priv_stack_mode;
670 u8 arg_cnt;
671 struct bpf_subprog_arg_info args[MAX_BPF_FUNC_REG_ARGS];
672 };
673
674 struct bpf_verifier_env;
675
676 struct backtrack_state {
677 struct bpf_verifier_env *env;
678 u32 frame;
679 u32 reg_masks[MAX_CALL_FRAMES];
680 u64 stack_masks[MAX_CALL_FRAMES];
681 };
682
683 struct bpf_id_pair {
684 u32 old;
685 u32 cur;
686 };
687
688 struct bpf_idmap {
689 u32 tmp_id_gen;
690 struct bpf_id_pair map[BPF_ID_MAP_SIZE];
691 };
692
693 struct bpf_idset {
694 u32 count;
695 u32 ids[BPF_ID_MAP_SIZE];
696 };
697
698 /* single container for all structs
699 * one verifier_env per bpf_check() call
700 */
701 struct bpf_verifier_env {
702 u32 insn_idx;
703 u32 prev_insn_idx;
704 struct bpf_prog *prog; /* eBPF program being verified */
705 const struct bpf_verifier_ops *ops;
706 struct module *attach_btf_mod; /* The owner module of prog->aux->attach_btf */
707 struct bpf_verifier_stack_elem *head; /* stack of verifier states to be processed */
708 int stack_size; /* number of states to be processed */
709 bool strict_alignment; /* perform strict pointer alignment checks */
710 bool test_state_freq; /* test verifier with different pruning frequency */
711 bool test_reg_invariants; /* fail verification on register invariants violations */
712 struct bpf_verifier_state *cur_state; /* current verifier state */
713 struct bpf_verifier_state_list **explored_states; /* search pruning optimization */
714 struct bpf_verifier_state_list *free_list;
715 struct bpf_map *used_maps[MAX_USED_MAPS]; /* array of map's used by eBPF program */
716 struct btf_mod_pair used_btfs[MAX_USED_BTFS]; /* array of BTF's used by BPF program */
717 u32 used_map_cnt; /* number of used maps */
718 u32 used_btf_cnt; /* number of used BTF objects */
719 u32 id_gen; /* used to generate unique reg IDs */
720 u32 hidden_subprog_cnt; /* number of hidden subprogs */
721 int exception_callback_subprog;
722 bool explore_alu_limits;
723 bool allow_ptr_leaks;
724 /* Allow access to uninitialized stack memory. Writes with fixed offset are
725 * always allowed, so this refers to reads (with fixed or variable offset),
726 * to writes with variable offset and to indirect (helper) accesses.
727 */
728 bool allow_uninit_stack;
729 bool bpf_capable;
730 bool bypass_spec_v1;
731 bool bypass_spec_v4;
732 bool seen_direct_write;
733 bool seen_exception;
734 struct bpf_insn_aux_data *insn_aux_data; /* array of per-insn state */
735 const struct bpf_line_info *prev_linfo;
736 struct bpf_verifier_log log;
737 struct bpf_subprog_info subprog_info[BPF_MAX_SUBPROGS + 2]; /* max + 2 for the fake and exception subprogs */
738 union {
739 struct bpf_idmap idmap_scratch;
740 struct bpf_idset idset_scratch;
741 };
742 struct {
743 int *insn_state;
744 int *insn_stack;
745 int cur_stack;
746 } cfg;
747 struct backtrack_state bt;
748 struct bpf_insn_hist_entry *insn_hist;
749 struct bpf_insn_hist_entry *cur_hist_ent;
750 u32 insn_hist_cap;
751 u32 pass_cnt; /* number of times do_check() was called */
752 u32 subprog_cnt;
753 /* number of instructions analyzed by the verifier */
754 u32 prev_insn_processed, insn_processed;
755 /* number of jmps, calls, exits analyzed so far */
756 u32 prev_jmps_processed, jmps_processed;
757 /* total verification time */
758 u64 verification_time;
759 /* maximum number of verifier states kept in 'branching' instructions */
760 u32 max_states_per_insn;
761 /* total number of allocated verifier states */
762 u32 total_states;
763 /* some states are freed during program analysis.
764 * this is peak number of states. this number dominates kernel
765 * memory consumption during verification
766 */
767 u32 peak_states;
768 /* longest register parentage chain walked for liveness marking */
769 u32 longest_mark_read_walk;
770 bpfptr_t fd_array;
771
772 /* bit mask to keep track of whether a register has been accessed
773 * since the last time the function state was printed
774 */
775 u32 scratched_regs;
776 /* Same as scratched_regs but for stack slots */
777 u64 scratched_stack_slots;
778 u64 prev_log_pos, prev_insn_print_pos;
779 /* buffer used to temporary hold constants as scalar registers */
780 struct bpf_reg_state fake_reg[2];
781 /* buffer used to generate temporary string representations,
782 * e.g., in reg_type_str() to generate reg_type string
783 */
784 char tmp_str_buf[TMP_STR_BUF_LEN];
785 struct bpf_insn insn_buf[INSN_BUF_SIZE];
786 struct bpf_insn epilogue_buf[INSN_BUF_SIZE];
787 };
788
subprog_aux(struct bpf_verifier_env * env,int subprog)789 static inline struct bpf_func_info_aux *subprog_aux(struct bpf_verifier_env *env, int subprog)
790 {
791 return &env->prog->aux->func_info_aux[subprog];
792 }
793
subprog_info(struct bpf_verifier_env * env,int subprog)794 static inline struct bpf_subprog_info *subprog_info(struct bpf_verifier_env *env, int subprog)
795 {
796 return &env->subprog_info[subprog];
797 }
798
799 __printf(2, 0) void bpf_verifier_vlog(struct bpf_verifier_log *log,
800 const char *fmt, va_list args);
801 __printf(2, 3) void bpf_verifier_log_write(struct bpf_verifier_env *env,
802 const char *fmt, ...);
803 __printf(2, 3) void bpf_log(struct bpf_verifier_log *log,
804 const char *fmt, ...);
805 int bpf_vlog_init(struct bpf_verifier_log *log, u32 log_level,
806 char __user *log_buf, u32 log_size);
807 void bpf_vlog_reset(struct bpf_verifier_log *log, u64 new_pos);
808 int bpf_vlog_finalize(struct bpf_verifier_log *log, u32 *log_size_actual);
809
810 __printf(3, 4) void verbose_linfo(struct bpf_verifier_env *env,
811 u32 insn_off,
812 const char *prefix_fmt, ...);
813
cur_func(struct bpf_verifier_env * env)814 static inline struct bpf_func_state *cur_func(struct bpf_verifier_env *env)
815 {
816 struct bpf_verifier_state *cur = env->cur_state;
817
818 return cur->frame[cur->curframe];
819 }
820
cur_regs(struct bpf_verifier_env * env)821 static inline struct bpf_reg_state *cur_regs(struct bpf_verifier_env *env)
822 {
823 return cur_func(env)->regs;
824 }
825
826 int bpf_prog_offload_verifier_prep(struct bpf_prog *prog);
827 int bpf_prog_offload_verify_insn(struct bpf_verifier_env *env,
828 int insn_idx, int prev_insn_idx);
829 int bpf_prog_offload_finalize(struct bpf_verifier_env *env);
830 void
831 bpf_prog_offload_replace_insn(struct bpf_verifier_env *env, u32 off,
832 struct bpf_insn *insn);
833 void
834 bpf_prog_offload_remove_insns(struct bpf_verifier_env *env, u32 off, u32 cnt);
835
836 /* 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,struct btf * btf,u32 btf_id)837 static inline u64 bpf_trampoline_compute_key(const struct bpf_prog *tgt_prog,
838 struct btf *btf, u32 btf_id)
839 {
840 if (tgt_prog)
841 return ((u64)tgt_prog->aux->id << 32) | btf_id;
842 else
843 return ((u64)btf_obj_id(btf) << 32) | 0x80000000 | btf_id;
844 }
845
846 /* unpack the IDs from the key as constructed above */
bpf_trampoline_unpack_key(u64 key,u32 * obj_id,u32 * btf_id)847 static inline void bpf_trampoline_unpack_key(u64 key, u32 *obj_id, u32 *btf_id)
848 {
849 if (obj_id)
850 *obj_id = key >> 32;
851 if (btf_id)
852 *btf_id = key & 0x7FFFFFFF;
853 }
854
855 int bpf_check_attach_target(struct bpf_verifier_log *log,
856 const struct bpf_prog *prog,
857 const struct bpf_prog *tgt_prog,
858 u32 btf_id,
859 struct bpf_attach_target_info *tgt_info);
860 void bpf_free_kfunc_btf_tab(struct bpf_kfunc_btf_tab *tab);
861
862 int mark_chain_precision(struct bpf_verifier_env *env, int regno);
863
864 #define BPF_BASE_TYPE_MASK GENMASK(BPF_BASE_TYPE_BITS - 1, 0)
865
866 /* extract base type from bpf_{arg, return, reg}_type. */
base_type(u32 type)867 static inline u32 base_type(u32 type)
868 {
869 return type & BPF_BASE_TYPE_MASK;
870 }
871
872 /* extract flags from an extended type. See bpf_type_flag in bpf.h. */
type_flag(u32 type)873 static inline u32 type_flag(u32 type)
874 {
875 return type & ~BPF_BASE_TYPE_MASK;
876 }
877
878 /* only use after check_attach_btf_id() */
resolve_prog_type(const struct bpf_prog * prog)879 static inline enum bpf_prog_type resolve_prog_type(const struct bpf_prog *prog)
880 {
881 return (prog->type == BPF_PROG_TYPE_EXT && prog->aux->saved_dst_prog_type) ?
882 prog->aux->saved_dst_prog_type : prog->type;
883 }
884
bpf_prog_check_recur(const struct bpf_prog * prog)885 static inline bool bpf_prog_check_recur(const struct bpf_prog *prog)
886 {
887 switch (resolve_prog_type(prog)) {
888 case BPF_PROG_TYPE_TRACING:
889 return prog->expected_attach_type != BPF_TRACE_ITER;
890 case BPF_PROG_TYPE_STRUCT_OPS:
891 return prog->aux->jits_use_priv_stack;
892 case BPF_PROG_TYPE_LSM:
893 return false;
894 default:
895 return true;
896 }
897 }
898
899 #define BPF_REG_TRUSTED_MODIFIERS (MEM_ALLOC | PTR_TRUSTED | NON_OWN_REF)
900
bpf_type_has_unsafe_modifiers(u32 type)901 static inline bool bpf_type_has_unsafe_modifiers(u32 type)
902 {
903 return type_flag(type) & ~BPF_REG_TRUSTED_MODIFIERS;
904 }
905
type_is_ptr_alloc_obj(u32 type)906 static inline bool type_is_ptr_alloc_obj(u32 type)
907 {
908 return base_type(type) == PTR_TO_BTF_ID && type_flag(type) & MEM_ALLOC;
909 }
910
type_is_non_owning_ref(u32 type)911 static inline bool type_is_non_owning_ref(u32 type)
912 {
913 return type_is_ptr_alloc_obj(type) && type_flag(type) & NON_OWN_REF;
914 }
915
type_is_pkt_pointer(enum bpf_reg_type type)916 static inline bool type_is_pkt_pointer(enum bpf_reg_type type)
917 {
918 type = base_type(type);
919 return type == PTR_TO_PACKET ||
920 type == PTR_TO_PACKET_META;
921 }
922
type_is_sk_pointer(enum bpf_reg_type type)923 static inline bool type_is_sk_pointer(enum bpf_reg_type type)
924 {
925 return type == PTR_TO_SOCKET ||
926 type == PTR_TO_SOCK_COMMON ||
927 type == PTR_TO_TCP_SOCK ||
928 type == PTR_TO_XDP_SOCK;
929 }
930
type_may_be_null(u32 type)931 static inline bool type_may_be_null(u32 type)
932 {
933 return type & PTR_MAYBE_NULL;
934 }
935
mark_reg_scratched(struct bpf_verifier_env * env,u32 regno)936 static inline void mark_reg_scratched(struct bpf_verifier_env *env, u32 regno)
937 {
938 env->scratched_regs |= 1U << regno;
939 }
940
mark_stack_slot_scratched(struct bpf_verifier_env * env,u32 spi)941 static inline void mark_stack_slot_scratched(struct bpf_verifier_env *env, u32 spi)
942 {
943 env->scratched_stack_slots |= 1ULL << spi;
944 }
945
reg_scratched(const struct bpf_verifier_env * env,u32 regno)946 static inline bool reg_scratched(const struct bpf_verifier_env *env, u32 regno)
947 {
948 return (env->scratched_regs >> regno) & 1;
949 }
950
stack_slot_scratched(const struct bpf_verifier_env * env,u64 regno)951 static inline bool stack_slot_scratched(const struct bpf_verifier_env *env, u64 regno)
952 {
953 return (env->scratched_stack_slots >> regno) & 1;
954 }
955
verifier_state_scratched(const struct bpf_verifier_env * env)956 static inline bool verifier_state_scratched(const struct bpf_verifier_env *env)
957 {
958 return env->scratched_regs || env->scratched_stack_slots;
959 }
960
mark_verifier_state_clean(struct bpf_verifier_env * env)961 static inline void mark_verifier_state_clean(struct bpf_verifier_env *env)
962 {
963 env->scratched_regs = 0U;
964 env->scratched_stack_slots = 0ULL;
965 }
966
967 /* Used for printing the entire verifier state. */
mark_verifier_state_scratched(struct bpf_verifier_env * env)968 static inline void mark_verifier_state_scratched(struct bpf_verifier_env *env)
969 {
970 env->scratched_regs = ~0U;
971 env->scratched_stack_slots = ~0ULL;
972 }
973
bpf_stack_narrow_access_ok(int off,int fill_size,int spill_size)974 static inline bool bpf_stack_narrow_access_ok(int off, int fill_size, int spill_size)
975 {
976 #ifdef __BIG_ENDIAN
977 off -= spill_size - fill_size;
978 #endif
979
980 return !(off % BPF_REG_SIZE);
981 }
982
983 const char *reg_type_str(struct bpf_verifier_env *env, enum bpf_reg_type type);
984 const char *dynptr_type_str(enum bpf_dynptr_type type);
985 const char *iter_type_str(const struct btf *btf, u32 btf_id);
986 const char *iter_state_str(enum bpf_iter_state state);
987
988 void print_verifier_state(struct bpf_verifier_env *env, const struct bpf_verifier_state *vstate,
989 u32 frameno, bool print_all);
990 void print_insn_state(struct bpf_verifier_env *env, const struct bpf_verifier_state *vstate,
991 u32 frameno);
992
993 #endif /* _LINUX_BPF_VERIFIER_H */
994