Commit | Line | Data |
---|---|---|
25763b3c | 1 | /* SPDX-License-Identifier: GPL-2.0-only */ |
58e2af8b | 2 | /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com |
58e2af8b JK |
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 */ | |
22dc4a0f | 8 | #include <linux/btf.h> /* for struct btf and btf_id() */ |
58e2af8b | 9 | #include <linux/filter.h> /* for MAX_BPF_STACK */ |
f1174f77 | 10 | #include <linux/tnum.h> |
58e2af8b | 11 | |
b03c9f9f EC |
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 | */ | |
bb7f0f98 | 16 | #define BPF_MAX_VAR_OFF (1 << 29) |
b03c9f9f EC |
17 | /* Maximum variable size permitted for ARG_CONST_SIZE[_OR_ZERO]. This ensures |
18 | * that converting umax_value to int cannot overflow. | |
19 | */ | |
bb7f0f98 | 20 | #define BPF_MAX_VAR_SIZ (1 << 29) |
c25b2ae1 | 21 | /* size of type_str_buf in bpf_verifier. */ |
ef66c547 | 22 | #define TYPE_STR_BUF_LEN 128 |
48461135 | 23 | |
8e9cd9ce EC |
24 | /* Liveness marks, used for registers and spilled-regs (in stack slots). |
25 | * Read marks propagate upwards until they find a write mark; they record that | |
26 | * "one of this state's descendants read this reg" (and therefore the reg is | |
27 | * relevant for states_equal() checks). | |
28 | * Write marks collect downwards and do not propagate; they record that "the | |
29 | * straight-line code that reached this state (from its parent) wrote this reg" | |
30 | * (and therefore that reads propagated from this state or its descendants | |
31 | * should not propagate to its parent). | |
32 | * A state with a write mark can receive read marks; it just won't propagate | |
33 | * them to its parent, since the write mark is a property, not of the state, | |
34 | * but of the link between it and its parent. See mark_reg_read() and | |
35 | * mark_stack_slot_read() in kernel/bpf/verifier.c. | |
36 | */ | |
dc503a8a EC |
37 | enum bpf_reg_liveness { |
38 | REG_LIVE_NONE = 0, /* reg hasn't been read or written this branch */ | |
5327ed3d JW |
39 | REG_LIVE_READ32 = 0x1, /* reg was read, so we're sensitive to initial value */ |
40 | REG_LIVE_READ64 = 0x2, /* likewise, but full 64-bit content matters */ | |
41 | REG_LIVE_READ = REG_LIVE_READ32 | REG_LIVE_READ64, | |
42 | REG_LIVE_WRITTEN = 0x4, /* reg was written first, screening off later reads */ | |
43 | REG_LIVE_DONE = 0x8, /* liveness won't be updating this register anymore */ | |
dc503a8a EC |
44 | }; |
45 | ||
6a3cd331 DM |
46 | /* For every reg representing a map value or allocated object pointer, |
47 | * we consider the tuple of (ptr, id) for them to be unique in verifier | |
48 | * context and conside them to not alias each other for the purposes of | |
49 | * tracking lock state. | |
50 | */ | |
51 | struct bpf_active_lock { | |
52 | /* This can either be reg->map_ptr or reg->btf. If ptr is NULL, | |
53 | * there's no active lock held, and other fields have no | |
54 | * meaning. If non-NULL, it indicates that a lock is held and | |
55 | * id member has the reg->id of the register which can be >= 0. | |
56 | */ | |
57 | void *ptr; | |
58 | /* This will be reg->id */ | |
59 | u32 id; | |
60 | }; | |
61 | ||
215bf496 AN |
62 | #define ITER_PREFIX "bpf_iter_" |
63 | ||
06accc87 AN |
64 | enum bpf_iter_state { |
65 | BPF_ITER_STATE_INVALID, /* for non-first slot */ | |
66 | BPF_ITER_STATE_ACTIVE, | |
67 | BPF_ITER_STATE_DRAINED, | |
68 | }; | |
69 | ||
58e2af8b | 70 | struct bpf_reg_state { |
679c782d | 71 | /* Ordering of fields matters. See states_equal() */ |
58e2af8b | 72 | enum bpf_reg_type type; |
22dc4a0f AN |
73 | /* Fixed part of pointer offset, pointer types only */ |
74 | s32 off; | |
58e2af8b | 75 | union { |
f1174f77 | 76 | /* valid when type == PTR_TO_PACKET */ |
6d94e741 | 77 | int range; |
58e2af8b JK |
78 | |
79 | /* valid when type == CONST_PTR_TO_MAP | PTR_TO_MAP_VALUE | | |
80 | * PTR_TO_MAP_VALUE_OR_NULL | |
81 | */ | |
3e8ce298 AS |
82 | struct { |
83 | struct bpf_map *map_ptr; | |
84 | /* To distinguish map lookups from outer map | |
85 | * the map_uid is non-zero for registers | |
86 | * pointing to inner maps. | |
87 | */ | |
88 | u32 map_uid; | |
89 | }; | |
0962590e | 90 | |
22dc4a0f AN |
91 | /* for PTR_TO_BTF_ID */ |
92 | struct { | |
93 | struct btf *btf; | |
94 | u32 btf_id; | |
95 | }; | |
9e15db66 | 96 | |
f8064ab9 KKD |
97 | struct { /* for PTR_TO_MEM | PTR_TO_MEM_OR_NULL */ |
98 | u32 mem_size; | |
99 | u32 dynptr_id; /* for dynptr slices */ | |
100 | }; | |
457f4436 | 101 | |
97e03f52 JK |
102 | /* For dynptr stack slots */ |
103 | struct { | |
104 | enum bpf_dynptr_type type; | |
105 | /* A dynptr is 16 bytes so it takes up 2 stack slots. | |
106 | * We need to track which slot is the first slot | |
107 | * to protect against cases where the user may try to | |
108 | * pass in an address starting at the second slot of the | |
109 | * dynptr. | |
110 | */ | |
111 | bool first_slot; | |
112 | } dynptr; | |
113 | ||
06accc87 AN |
114 | /* For bpf_iter stack slots */ |
115 | struct { | |
116 | /* BTF container and BTF type ID describing | |
117 | * struct bpf_iter_<type> of an iterator state | |
118 | */ | |
119 | struct btf *btf; | |
120 | u32 btf_id; | |
121 | /* packing following two fields to fit iter state into 16 bytes */ | |
122 | enum bpf_iter_state state:2; | |
123 | int depth:30; | |
124 | } iter; | |
125 | ||
0962590e | 126 | /* Max size from any of the above. */ |
22dc4a0f AN |
127 | struct { |
128 | unsigned long raw1; | |
129 | unsigned long raw2; | |
130 | } raw; | |
69c087ba YS |
131 | |
132 | u32 subprogno; /* for PTR_TO_FUNC */ | |
58e2af8b | 133 | }; |
a73bf9f2 AN |
134 | /* For scalar types (SCALAR_VALUE), this represents our knowledge of |
135 | * the actual value. | |
136 | * For pointer types, this represents the variable part of the offset | |
137 | * from the pointed-to object, and is shared with all bpf_reg_states | |
138 | * with the same id as us. | |
139 | */ | |
140 | struct tnum var_off; | |
141 | /* Used to determine if any memory access using this register will | |
142 | * result in a bad access. | |
143 | * These refer to the same value as var_off, not necessarily the actual | |
144 | * contents of the register. | |
145 | */ | |
146 | s64 smin_value; /* minimum possible (s64)value */ | |
147 | s64 smax_value; /* maximum possible (s64)value */ | |
148 | u64 umin_value; /* minimum possible (u64)value */ | |
149 | u64 umax_value; /* maximum possible (u64)value */ | |
150 | s32 s32_min_value; /* minimum possible (s32)value */ | |
151 | s32 s32_max_value; /* maximum possible (s32)value */ | |
152 | u32 u32_min_value; /* minimum possible (u32)value */ | |
153 | u32 u32_max_value; /* maximum possible (u32)value */ | |
f1174f77 EC |
154 | /* For PTR_TO_PACKET, used to find other pointers with the same variable |
155 | * offset, so they can share range knowledge. | |
156 | * For PTR_TO_MAP_VALUE_OR_NULL this is used to share which map value we | |
157 | * came from, when one is tested for != NULL. | |
457f4436 AN |
158 | * For PTR_TO_MEM_OR_NULL this is used to identify memory allocation |
159 | * for the purpose of tracking that it's freed. | |
c64b7983 JS |
160 | * For PTR_TO_SOCKET this is used to share which pointers retain the |
161 | * same reference to the socket, to determine proper reference freeing. | |
bc34dee6 JK |
162 | * For stack slots that are dynptrs, this is used to track references to |
163 | * the dynptr to determine proper reference freeing. | |
06accc87 AN |
164 | * Similarly to dynptrs, we use ID to track "belonging" of a reference |
165 | * to a specific instance of bpf_iter. | |
f1174f77 | 166 | */ |
d2a4dd37 | 167 | u32 id; |
1b986589 MKL |
168 | /* PTR_TO_SOCKET and PTR_TO_TCP_SOCK could be a ptr returned |
169 | * from a pointer-cast helper, bpf_sk_fullsock() and | |
170 | * bpf_tcp_sock(). | |
171 | * | |
172 | * Consider the following where "sk" is a reference counted | |
173 | * pointer returned from "sk = bpf_sk_lookup_tcp();": | |
174 | * | |
175 | * 1: sk = bpf_sk_lookup_tcp(); | |
176 | * 2: if (!sk) { return 0; } | |
177 | * 3: fullsock = bpf_sk_fullsock(sk); | |
178 | * 4: if (!fullsock) { bpf_sk_release(sk); return 0; } | |
179 | * 5: tp = bpf_tcp_sock(fullsock); | |
180 | * 6: if (!tp) { bpf_sk_release(sk); return 0; } | |
181 | * 7: bpf_sk_release(sk); | |
182 | * 8: snd_cwnd = tp->snd_cwnd; // verifier will complain | |
183 | * | |
184 | * After bpf_sk_release(sk) at line 7, both "fullsock" ptr and | |
185 | * "tp" ptr should be invalidated also. In order to do that, | |
186 | * the reg holding "fullsock" and "sk" need to remember | |
187 | * the original refcounted ptr id (i.e. sk_reg->id) in ref_obj_id | |
188 | * such that the verifier can reset all regs which have | |
189 | * ref_obj_id matching the sk_reg->id. | |
190 | * | |
191 | * sk_reg->ref_obj_id is set to sk_reg->id at line 1. | |
192 | * sk_reg->id will stay as NULL-marking purpose only. | |
193 | * After NULL-marking is done, sk_reg->id can be reset to 0. | |
194 | * | |
195 | * After "fullsock = bpf_sk_fullsock(sk);" at line 3, | |
196 | * fullsock_reg->ref_obj_id is set to sk_reg->ref_obj_id. | |
197 | * | |
198 | * After "tp = bpf_tcp_sock(fullsock);" at line 5, | |
199 | * tp_reg->ref_obj_id is set to fullsock_reg->ref_obj_id | |
200 | * which is the same as sk_reg->ref_obj_id. | |
201 | * | |
202 | * From the verifier perspective, if sk, fullsock and tp | |
203 | * are not NULL, they are the same ptr with different | |
204 | * reg->type. In particular, bpf_sk_release(tp) is also | |
205 | * allowed and has the same effect as bpf_sk_release(sk). | |
206 | */ | |
207 | u32 ref_obj_id; | |
679c782d EC |
208 | /* parentage chain for liveness checking */ |
209 | struct bpf_reg_state *parent; | |
f4d7e40a AS |
210 | /* Inside the callee two registers can be both PTR_TO_STACK like |
211 | * R1=fp-8 and R2=fp-8, but one of them points to this function stack | |
212 | * while another to the caller's stack. To differentiate them 'frameno' | |
213 | * is used which is an index in bpf_verifier_state->frame[] array | |
214 | * pointing to bpf_func_state. | |
f4d7e40a AS |
215 | */ |
216 | u32 frameno; | |
5327ed3d JW |
217 | /* Tracks subreg definition. The stored value is the insn_idx of the |
218 | * writing insn. This is safe because subreg_def is used before any insn | |
219 | * patching which only happens after main verification finished. | |
220 | */ | |
221 | s32 subreg_def; | |
dc503a8a | 222 | enum bpf_reg_liveness live; |
b5dc0163 AS |
223 | /* if (!precise && SCALAR_VALUE) min/max/tnum don't affect safety */ |
224 | bool precise; | |
58e2af8b JK |
225 | }; |
226 | ||
227 | enum bpf_stack_slot_type { | |
228 | STACK_INVALID, /* nothing was stored in this stack slot */ | |
229 | STACK_SPILL, /* register spilled into stack */ | |
cc2b14d5 AS |
230 | STACK_MISC, /* BPF program wrote some data into this slot */ |
231 | STACK_ZERO, /* BPF program wrote constant zero */ | |
97e03f52 JK |
232 | /* A dynptr is stored in this stack slot. The type of dynptr |
233 | * is stored in bpf_stack_state->spilled_ptr.dynptr.type | |
234 | */ | |
235 | STACK_DYNPTR, | |
06accc87 | 236 | STACK_ITER, |
58e2af8b JK |
237 | }; |
238 | ||
239 | #define BPF_REG_SIZE 8 /* size of eBPF register in bytes */ | |
06accc87 | 240 | |
97e03f52 JK |
241 | #define BPF_DYNPTR_SIZE sizeof(struct bpf_dynptr_kern) |
242 | #define BPF_DYNPTR_NR_SLOTS (BPF_DYNPTR_SIZE / BPF_REG_SIZE) | |
58e2af8b | 243 | |
638f5b90 AS |
244 | struct bpf_stack_state { |
245 | struct bpf_reg_state spilled_ptr; | |
246 | u8 slot_type[BPF_REG_SIZE]; | |
247 | }; | |
248 | ||
fd978bf7 JS |
249 | struct bpf_reference_state { |
250 | /* Track each reference created with a unique id, even if the same | |
251 | * instruction creates the reference multiple times (eg, via CALL). | |
252 | */ | |
253 | int id; | |
254 | /* Instruction where the allocation of this reference occurred. This | |
255 | * is used purely to inform the user of a reference leak. | |
256 | */ | |
257 | int insn_idx; | |
9d9d00ac KKD |
258 | /* There can be a case like: |
259 | * main (frame 0) | |
260 | * cb (frame 1) | |
261 | * func (frame 3) | |
262 | * cb (frame 4) | |
263 | * Hence for frame 4, if callback_ref just stored boolean, it would be | |
264 | * impossible to distinguish nested callback refs. Hence store the | |
265 | * frameno and compare that to callback_ref in check_reference_leak when | |
266 | * exiting a callback function. | |
267 | */ | |
268 | int callback_ref; | |
fd978bf7 JS |
269 | }; |
270 | ||
58e2af8b JK |
271 | /* state of the program: |
272 | * type of all registers and stack info | |
273 | */ | |
f4d7e40a | 274 | struct bpf_func_state { |
58e2af8b | 275 | struct bpf_reg_state regs[MAX_BPF_REG]; |
f4d7e40a AS |
276 | /* index of call instruction that called into this func */ |
277 | int callsite; | |
278 | /* stack frame number of this function state from pov of | |
279 | * enclosing bpf_verifier_state. | |
280 | * 0 = main function, 1 = first callee. | |
281 | */ | |
282 | u32 frameno; | |
01f810ac | 283 | /* subprog number == index within subprog_info |
f4d7e40a AS |
284 | * zero == main subprog |
285 | */ | |
286 | u32 subprogno; | |
bfc6bb74 AS |
287 | /* Every bpf_timer_start will increment async_entry_cnt. |
288 | * It's used to distinguish: | |
289 | * void foo(void) { for(;;); } | |
290 | * void foo(void) { bpf_timer_set_callback(,foo); } | |
291 | */ | |
292 | u32 async_entry_cnt; | |
293 | bool in_callback_fn; | |
1bfe26fb | 294 | struct tnum callback_ret_range; |
bfc6bb74 | 295 | bool in_async_callback_fn; |
f4d7e40a | 296 | |
fd978bf7 JS |
297 | /* The following fields should be last. See copy_func_state() */ |
298 | int acquired_refs; | |
299 | struct bpf_reference_state *refs; | |
638f5b90 AS |
300 | int allocated_stack; |
301 | struct bpf_stack_state *stack; | |
58e2af8b JK |
302 | }; |
303 | ||
b5dc0163 AS |
304 | struct bpf_idx_pair { |
305 | u32 prev_idx; | |
306 | u32 idx; | |
307 | }; | |
308 | ||
c9e73e3d LB |
309 | struct bpf_id_pair { |
310 | u32 old; | |
311 | u32 cur; | |
312 | }; | |
313 | ||
f4d7e40a | 314 | #define MAX_CALL_FRAMES 8 |
5dd9cdbc EZ |
315 | /* Maximum number of register states that can exist at once */ |
316 | #define BPF_ID_MAP_SIZE ((MAX_BPF_REG + MAX_BPF_STACK / BPF_REG_SIZE) * MAX_CALL_FRAMES) | |
f4d7e40a AS |
317 | struct bpf_verifier_state { |
318 | /* call stack tracking */ | |
319 | struct bpf_func_state *frame[MAX_CALL_FRAMES]; | |
2589726d AS |
320 | struct bpf_verifier_state *parent; |
321 | /* | |
322 | * 'branches' field is the number of branches left to explore: | |
323 | * 0 - all possible paths from this state reached bpf_exit or | |
324 | * were safely pruned | |
325 | * 1 - at least one path is being explored. | |
326 | * This state hasn't reached bpf_exit | |
327 | * 2 - at least two paths are being explored. | |
328 | * This state is an immediate parent of two children. | |
329 | * One is fallthrough branch with branches==1 and another | |
330 | * state is pushed into stack (to be explored later) also with | |
331 | * branches==1. The parent of this state has branches==1. | |
332 | * The verifier state tree connected via 'parent' pointer looks like: | |
333 | * 1 | |
334 | * 1 | |
335 | * 2 -> 1 (first 'if' pushed into stack) | |
336 | * 1 | |
337 | * 2 -> 1 (second 'if' pushed into stack) | |
338 | * 1 | |
339 | * 1 | |
340 | * 1 bpf_exit. | |
341 | * | |
342 | * Once do_check() reaches bpf_exit, it calls update_branch_counts() | |
343 | * and the verifier state tree will look: | |
344 | * 1 | |
345 | * 1 | |
346 | * 2 -> 1 (first 'if' pushed into stack) | |
347 | * 1 | |
348 | * 1 -> 1 (second 'if' pushed into stack) | |
349 | * 0 | |
350 | * 0 | |
351 | * 0 bpf_exit. | |
352 | * After pop_stack() the do_check() will resume at second 'if'. | |
353 | * | |
354 | * If is_state_visited() sees a state with branches > 0 it means | |
355 | * there is a loop. If such state is exactly equal to the current state | |
356 | * it's an infinite loop. Note states_equal() checks for states | |
6dbdc9f3 | 357 | * equivalency, so two states being 'states_equal' does not mean |
2589726d AS |
358 | * infinite loop. The exact comparison is provided by |
359 | * states_maybe_looping() function. It's a stronger pre-check and | |
360 | * much faster than states_equal(). | |
361 | * | |
362 | * This algorithm may not find all possible infinite loops or | |
363 | * loop iteration count may be too high. | |
364 | * In such cases BPF_COMPLEXITY_LIMIT_INSNS limit kicks in. | |
365 | */ | |
366 | u32 branches; | |
dc2a4ebc | 367 | u32 insn_idx; |
f4d7e40a | 368 | u32 curframe; |
6a3cd331 DM |
369 | |
370 | struct bpf_active_lock active_lock; | |
979d63d5 | 371 | bool speculative; |
9bb00b28 | 372 | bool active_rcu_lock; |
b5dc0163 AS |
373 | |
374 | /* first and last insn idx of this verifier state */ | |
375 | u32 first_insn_idx; | |
376 | u32 last_insn_idx; | |
377 | /* jmp history recorded from first to last. | |
378 | * backtracking is using it to go from last to first. | |
379 | * For most states jmp_history_cnt is [0-3]. | |
380 | * For loops can go up to ~40. | |
381 | */ | |
382 | struct bpf_idx_pair *jmp_history; | |
383 | u32 jmp_history_cnt; | |
f4d7e40a AS |
384 | }; |
385 | ||
f3709f69 JS |
386 | #define bpf_get_spilled_reg(slot, frame) \ |
387 | (((slot < frame->allocated_stack / BPF_REG_SIZE) && \ | |
388 | (frame->stack[slot].slot_type[0] == STACK_SPILL)) \ | |
389 | ? &frame->stack[slot].spilled_ptr : NULL) | |
390 | ||
391 | /* Iterate over 'frame', setting 'reg' to either NULL or a spilled register. */ | |
392 | #define bpf_for_each_spilled_reg(iter, frame, reg) \ | |
393 | for (iter = 0, reg = bpf_get_spilled_reg(iter, frame); \ | |
394 | iter < frame->allocated_stack / BPF_REG_SIZE; \ | |
395 | iter++, reg = bpf_get_spilled_reg(iter, frame)) | |
396 | ||
b239da34 KKD |
397 | /* Invoke __expr over regsiters in __vst, setting __state and __reg */ |
398 | #define bpf_for_each_reg_in_vstate(__vst, __state, __reg, __expr) \ | |
399 | ({ \ | |
400 | struct bpf_verifier_state *___vstate = __vst; \ | |
401 | int ___i, ___j; \ | |
402 | for (___i = 0; ___i <= ___vstate->curframe; ___i++) { \ | |
403 | struct bpf_reg_state *___regs; \ | |
404 | __state = ___vstate->frame[___i]; \ | |
405 | ___regs = __state->regs; \ | |
406 | for (___j = 0; ___j < MAX_BPF_REG; ___j++) { \ | |
407 | __reg = &___regs[___j]; \ | |
408 | (void)(__expr); \ | |
409 | } \ | |
410 | bpf_for_each_spilled_reg(___j, __state, __reg) { \ | |
411 | if (!__reg) \ | |
412 | continue; \ | |
413 | (void)(__expr); \ | |
414 | } \ | |
415 | } \ | |
416 | }) | |
417 | ||
58e2af8b JK |
418 | /* linked list of verifier states used to prune search */ |
419 | struct bpf_verifier_state_list { | |
420 | struct bpf_verifier_state state; | |
421 | struct bpf_verifier_state_list *next; | |
9f4686c4 | 422 | int miss_cnt, hit_cnt; |
58e2af8b JK |
423 | }; |
424 | ||
1ade2371 | 425 | struct bpf_loop_inline_state { |
f16214c1 MB |
426 | unsigned int initialized:1; /* set to true upon first entry */ |
427 | unsigned int fit_for_inline:1; /* true if callback function is the same | |
428 | * at each call and flags are always zero | |
429 | */ | |
1ade2371 EZ |
430 | u32 callback_subprogno; /* valid when fit_for_inline is true */ |
431 | }; | |
432 | ||
979d63d5 | 433 | /* Possible states for alu_state member. */ |
801c6058 DB |
434 | #define BPF_ALU_SANITIZE_SRC (1U << 0) |
435 | #define BPF_ALU_SANITIZE_DST (1U << 1) | |
979d63d5 | 436 | #define BPF_ALU_NEG_VALUE (1U << 2) |
d3bd7413 | 437 | #define BPF_ALU_NON_POINTER (1U << 3) |
801c6058 | 438 | #define BPF_ALU_IMMEDIATE (1U << 4) |
979d63d5 DB |
439 | #define BPF_ALU_SANITIZE (BPF_ALU_SANITIZE_SRC | \ |
440 | BPF_ALU_SANITIZE_DST) | |
441 | ||
58e2af8b | 442 | struct bpf_insn_aux_data { |
81ed18ab AS |
443 | union { |
444 | enum bpf_reg_type ptr_type; /* pointer type for load/store insns */ | |
d2e4c1e6 | 445 | unsigned long map_ptr_state; /* pointer/poison value for maps */ |
1c2a088a | 446 | s32 call_imm; /* saved imm field of call insn */ |
979d63d5 | 447 | u32 alu_limit; /* limit for add/sub register with pointer */ |
d8eca5bb DB |
448 | struct { |
449 | u32 map_index; /* index into used_maps[] */ | |
450 | u32 map_off; /* offset from value base address */ | |
451 | }; | |
4976b718 HL |
452 | struct { |
453 | enum bpf_reg_type reg_type; /* type of pseudo_btf_id */ | |
454 | union { | |
22dc4a0f AN |
455 | struct { |
456 | struct btf *btf; | |
457 | u32 btf_id; /* btf_id for struct typed var */ | |
458 | }; | |
4976b718 HL |
459 | u32 mem_size; /* mem_size for non-struct typed var */ |
460 | }; | |
461 | } btf_var; | |
1ade2371 EZ |
462 | /* if instruction is a call to bpf_loop this field tracks |
463 | * the state of the relevant registers to make decision about inlining | |
464 | */ | |
465 | struct bpf_loop_inline_state loop_inline_state; | |
81ed18ab | 466 | }; |
958cf2e2 KKD |
467 | u64 obj_new_size; /* remember the size of type passed to bpf_obj_new to rewrite R1 */ |
468 | struct btf_struct_meta *kptr_struct_meta; | |
d2e4c1e6 | 469 | u64 map_key_state; /* constant (32 bit) key tracking for maps */ |
23994631 | 470 | int ctx_field_size; /* the ctx field size for load insn, maybe 0 */ |
51c39bb1 | 471 | u32 seen; /* this insn was processed by the verifier at env->pass_cnt */ |
2039f26f | 472 | bool sanitize_stack_spill; /* subject to Spectre v4 sanitation */ |
5327ed3d | 473 | bool zext_dst; /* this insn zero extends dst reg */ |
9bb00b28 | 474 | bool storage_get_func_atomic; /* bpf_*_storage_get() with atomic memory alloc */ |
06accc87 | 475 | bool is_iter_next; /* bpf_iter_<type>_next() kfunc call */ |
979d63d5 | 476 | u8 alu_state; /* used in combination with alu_limit */ |
51c39bb1 AS |
477 | |
478 | /* below fields are initialized once */ | |
9e4c24e7 | 479 | unsigned int orig_idx; /* original instruction index */ |
51c39bb1 | 480 | bool prune_point; |
bffdeaa8 | 481 | bool jmp_point; |
58e2af8b JK |
482 | }; |
483 | ||
484 | #define MAX_USED_MAPS 64 /* max number of maps accessed by one eBPF program */ | |
541c3bad | 485 | #define MAX_USED_BTFS 64 /* max number of BTFs accessed by one BPF program */ |
58e2af8b | 486 | |
a2a7d570 JK |
487 | #define BPF_VERIFIER_TMP_LOG_SIZE 1024 |
488 | ||
b9193c1b | 489 | struct bpf_verifier_log { |
e7bf8249 | 490 | u32 level; |
a2a7d570 | 491 | char kbuf[BPF_VERIFIER_TMP_LOG_SIZE]; |
e7bf8249 JK |
492 | char __user *ubuf; |
493 | u32 len_used; | |
494 | u32 len_total; | |
495 | }; | |
496 | ||
b9193c1b | 497 | static inline bool bpf_verifier_log_full(const struct bpf_verifier_log *log) |
e7bf8249 JK |
498 | { |
499 | return log->len_used >= log->len_total - 1; | |
500 | } | |
501 | ||
06ee7115 AS |
502 | #define BPF_LOG_LEVEL1 1 |
503 | #define BPF_LOG_LEVEL2 2 | |
504 | #define BPF_LOG_STATS 4 | |
505 | #define BPF_LOG_LEVEL (BPF_LOG_LEVEL1 | BPF_LOG_LEVEL2) | |
506 | #define BPF_LOG_MASK (BPF_LOG_LEVEL | BPF_LOG_STATS) | |
8580ac94 | 507 | #define BPF_LOG_KERNEL (BPF_LOG_MASK + 1) /* kernel internal flag */ |
2e576648 CL |
508 | #define BPF_LOG_MIN_ALIGNMENT 8U |
509 | #define BPF_LOG_ALIGNMENT 40U | |
06ee7115 | 510 | |
77d2e05a MKL |
511 | static inline bool bpf_verifier_log_needed(const struct bpf_verifier_log *log) |
512 | { | |
efc68158 THJ |
513 | return log && |
514 | ((log->level && log->ubuf && !bpf_verifier_log_full(log)) || | |
515 | log->level == BPF_LOG_KERNEL); | |
77d2e05a MKL |
516 | } |
517 | ||
866de407 HT |
518 | static inline bool |
519 | bpf_verifier_log_attr_valid(const struct bpf_verifier_log *log) | |
520 | { | |
521 | return log->len_total >= 128 && log->len_total <= UINT_MAX >> 2 && | |
522 | log->level && log->ubuf && !(log->level & ~BPF_LOG_MASK); | |
523 | } | |
524 | ||
cc8b0b92 AS |
525 | #define BPF_MAX_SUBPROGS 256 |
526 | ||
9c8105bd | 527 | struct bpf_subprog_info { |
8c1b6e69 | 528 | /* 'start' has to be the first field otherwise find_subprog() won't work */ |
9c8105bd | 529 | u32 start; /* insn idx of function entry point */ |
c454a46b | 530 | u32 linfo_idx; /* The idx to the main_prog->aux->linfo */ |
9c8105bd | 531 | u16 stack_depth; /* max. stack depth used by this function */ |
7f6e4312 | 532 | bool has_tail_call; |
ebf7d1f5 | 533 | bool tail_call_reachable; |
09b28d76 | 534 | bool has_ld_abs; |
7ddc80a4 | 535 | bool is_async_cb; |
9c8105bd JW |
536 | }; |
537 | ||
58e2af8b JK |
538 | /* single container for all structs |
539 | * one verifier_env per bpf_check() call | |
540 | */ | |
541 | struct bpf_verifier_env { | |
c08435ec DB |
542 | u32 insn_idx; |
543 | u32 prev_insn_idx; | |
58e2af8b | 544 | struct bpf_prog *prog; /* eBPF program being verified */ |
00176a34 | 545 | const struct bpf_verifier_ops *ops; |
58e2af8b JK |
546 | struct bpf_verifier_stack_elem *head; /* stack of verifier states to be processed */ |
547 | int stack_size; /* number of states to be processed */ | |
e07b98d9 | 548 | bool strict_alignment; /* perform strict pointer alignment checks */ |
10d274e8 | 549 | bool test_state_freq; /* test verifier with different pruning frequency */ |
638f5b90 | 550 | struct bpf_verifier_state *cur_state; /* current verifier state */ |
58e2af8b | 551 | struct bpf_verifier_state_list **explored_states; /* search pruning optimization */ |
9f4686c4 | 552 | struct bpf_verifier_state_list *free_list; |
58e2af8b | 553 | struct bpf_map *used_maps[MAX_USED_MAPS]; /* array of map's used by eBPF program */ |
541c3bad | 554 | struct btf_mod_pair used_btfs[MAX_USED_BTFS]; /* array of BTF's used by BPF program */ |
58e2af8b | 555 | u32 used_map_cnt; /* number of used maps */ |
541c3bad | 556 | u32 used_btf_cnt; /* number of used BTF objects */ |
58e2af8b | 557 | u32 id_gen; /* used to generate unique reg IDs */ |
e042aa53 | 558 | bool explore_alu_limits; |
58e2af8b | 559 | bool allow_ptr_leaks; |
01f810ac | 560 | bool allow_uninit_stack; |
2c78ee89 AS |
561 | bool bpf_capable; |
562 | bool bypass_spec_v1; | |
563 | bool bypass_spec_v4; | |
58e2af8b JK |
564 | bool seen_direct_write; |
565 | struct bpf_insn_aux_data *insn_aux_data; /* array of per-insn state */ | |
d9762e84 | 566 | const struct bpf_line_info *prev_linfo; |
b9193c1b | 567 | struct bpf_verifier_log log; |
9c8105bd | 568 | struct bpf_subprog_info subprog_info[BPF_MAX_SUBPROGS + 1]; |
c9e73e3d | 569 | struct bpf_id_pair idmap_scratch[BPF_ID_MAP_SIZE]; |
7df737e9 AS |
570 | struct { |
571 | int *insn_state; | |
572 | int *insn_stack; | |
573 | int cur_stack; | |
574 | } cfg; | |
51c39bb1 | 575 | u32 pass_cnt; /* number of times do_check() was called */ |
cc8b0b92 | 576 | u32 subprog_cnt; |
06ee7115 | 577 | /* number of instructions analyzed by the verifier */ |
2589726d AS |
578 | u32 prev_insn_processed, insn_processed; |
579 | /* number of jmps, calls, exits analyzed so far */ | |
580 | u32 prev_jmps_processed, jmps_processed; | |
06ee7115 AS |
581 | /* total verification time */ |
582 | u64 verification_time; | |
583 | /* maximum number of verifier states kept in 'branching' instructions */ | |
584 | u32 max_states_per_insn; | |
585 | /* total number of allocated verifier states */ | |
586 | u32 total_states; | |
587 | /* some states are freed during program analysis. | |
588 | * this is peak number of states. this number dominates kernel | |
589 | * memory consumption during verification | |
590 | */ | |
591 | u32 peak_states; | |
592 | /* longest register parentage chain walked for liveness marking */ | |
593 | u32 longest_mark_read_walk; | |
387544bf | 594 | bpfptr_t fd_array; |
0f55f9ed CL |
595 | |
596 | /* bit mask to keep track of whether a register has been accessed | |
597 | * since the last time the function state was printed | |
598 | */ | |
599 | u32 scratched_regs; | |
600 | /* Same as scratched_regs but for stack slots */ | |
601 | u64 scratched_stack_slots; | |
2e576648 | 602 | u32 prev_log_len, prev_insn_print_len; |
c25b2ae1 HL |
603 | /* buffer used in reg_type_str() to generate reg_type string */ |
604 | char type_str_buf[TYPE_STR_BUF_LEN]; | |
58e2af8b JK |
605 | }; |
606 | ||
be2d04d1 MM |
607 | __printf(2, 0) void bpf_verifier_vlog(struct bpf_verifier_log *log, |
608 | const char *fmt, va_list args); | |
430e68d1 QM |
609 | __printf(2, 3) void bpf_verifier_log_write(struct bpf_verifier_env *env, |
610 | const char *fmt, ...); | |
9e15db66 AS |
611 | __printf(2, 3) void bpf_log(struct bpf_verifier_log *log, |
612 | const char *fmt, ...); | |
430e68d1 | 613 | |
fd978bf7 | 614 | static inline struct bpf_func_state *cur_func(struct bpf_verifier_env *env) |
638f5b90 | 615 | { |
f4d7e40a AS |
616 | struct bpf_verifier_state *cur = env->cur_state; |
617 | ||
fd978bf7 JS |
618 | return cur->frame[cur->curframe]; |
619 | } | |
620 | ||
621 | static inline struct bpf_reg_state *cur_regs(struct bpf_verifier_env *env) | |
622 | { | |
623 | return cur_func(env)->regs; | |
638f5b90 AS |
624 | } |
625 | ||
a40a2632 | 626 | int bpf_prog_offload_verifier_prep(struct bpf_prog *prog); |
cae1927c JK |
627 | int bpf_prog_offload_verify_insn(struct bpf_verifier_env *env, |
628 | int insn_idx, int prev_insn_idx); | |
c941ce9c | 629 | int bpf_prog_offload_finalize(struct bpf_verifier_env *env); |
08ca90af JK |
630 | void |
631 | bpf_prog_offload_replace_insn(struct bpf_verifier_env *env, u32 off, | |
632 | struct bpf_insn *insn); | |
633 | void | |
634 | bpf_prog_offload_remove_insns(struct bpf_verifier_env *env, u32 off, u32 cnt); | |
ab3f0063 | 635 | |
be80a1d3 DB |
636 | int check_ptr_off_reg(struct bpf_verifier_env *env, |
637 | const struct bpf_reg_state *reg, int regno); | |
25b35dd2 KKD |
638 | int check_func_arg_reg_off(struct bpf_verifier_env *env, |
639 | const struct bpf_reg_state *reg, int regno, | |
8f14852e | 640 | enum bpf_arg_type arg_type); |
e5069b9c DB |
641 | int check_mem_reg(struct bpf_verifier_env *env, struct bpf_reg_state *reg, |
642 | u32 regno, u32 mem_size); | |
51c39bb1 | 643 | |
f7b12b6f THJ |
644 | /* this lives here instead of in bpf.h because it needs to dereference tgt_prog */ |
645 | static inline u64 bpf_trampoline_compute_key(const struct bpf_prog *tgt_prog, | |
22dc4a0f | 646 | struct btf *btf, u32 btf_id) |
f7b12b6f | 647 | { |
22dc4a0f AN |
648 | if (tgt_prog) |
649 | return ((u64)tgt_prog->aux->id << 32) | btf_id; | |
650 | else | |
651 | return ((u64)btf_obj_id(btf) << 32) | 0x80000000 | btf_id; | |
f7b12b6f THJ |
652 | } |
653 | ||
441e8c66 THJ |
654 | /* unpack the IDs from the key as constructed above */ |
655 | static inline void bpf_trampoline_unpack_key(u64 key, u32 *obj_id, u32 *btf_id) | |
656 | { | |
657 | if (obj_id) | |
658 | *obj_id = key >> 32; | |
659 | if (btf_id) | |
660 | *btf_id = key & 0x7FFFFFFF; | |
661 | } | |
662 | ||
f7b12b6f THJ |
663 | int bpf_check_attach_target(struct bpf_verifier_log *log, |
664 | const struct bpf_prog *prog, | |
665 | const struct bpf_prog *tgt_prog, | |
666 | u32 btf_id, | |
667 | struct bpf_attach_target_info *tgt_info); | |
2357672c KKD |
668 | void bpf_free_kfunc_btf_tab(struct bpf_kfunc_btf_tab *tab); |
669 | ||
eb1f7f71 BT |
670 | int mark_chain_precision(struct bpf_verifier_env *env, int regno); |
671 | ||
d639b9d1 HL |
672 | #define BPF_BASE_TYPE_MASK GENMASK(BPF_BASE_TYPE_BITS - 1, 0) |
673 | ||
674 | /* extract base type from bpf_{arg, return, reg}_type. */ | |
675 | static inline u32 base_type(u32 type) | |
676 | { | |
677 | return type & BPF_BASE_TYPE_MASK; | |
678 | } | |
679 | ||
680 | /* extract flags from an extended type. See bpf_type_flag in bpf.h. */ | |
681 | static inline u32 type_flag(u32 type) | |
682 | { | |
683 | return type & ~BPF_BASE_TYPE_MASK; | |
684 | } | |
f7b12b6f | 685 | |
4a9c7bbe | 686 | /* only use after check_attach_btf_id() */ |
271de525 | 687 | static inline enum bpf_prog_type resolve_prog_type(const struct bpf_prog *prog) |
5c073f26 | 688 | { |
4a9c7bbe MKL |
689 | return prog->type == BPF_PROG_TYPE_EXT ? |
690 | prog->aux->dst_prog->type : prog->type; | |
5c073f26 KKD |
691 | } |
692 | ||
271de525 MKL |
693 | static inline bool bpf_prog_check_recur(const struct bpf_prog *prog) |
694 | { | |
695 | switch (resolve_prog_type(prog)) { | |
696 | case BPF_PROG_TYPE_TRACING: | |
697 | return prog->expected_attach_type != BPF_TRACE_ITER; | |
698 | case BPF_PROG_TYPE_STRUCT_OPS: | |
699 | case BPF_PROG_TYPE_LSM: | |
700 | return false; | |
701 | default: | |
702 | return true; | |
703 | } | |
704 | } | |
705 | ||
fca1aa75 | 706 | #define BPF_REG_TRUSTED_MODIFIERS (MEM_ALLOC | PTR_TRUSTED) |
3f00c523 DV |
707 | |
708 | static inline bool bpf_type_has_unsafe_modifiers(u32 type) | |
709 | { | |
710 | return type_flag(type) & ~BPF_REG_TRUSTED_MODIFIERS; | |
711 | } | |
712 | ||
58e2af8b | 713 | #endif /* _LINUX_BPF_VERIFIER_H */ |