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