xref: /linux/kernel/bpf/log.c (revision 6e7fd890f1d6ac83805409e9c346240de2705584)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
3  * Copyright (c) 2016 Facebook
4  * Copyright (c) 2018 Covalent IO, Inc. http://covalent.io
5  */
6 #include <uapi/linux/btf.h>
7 #include <linux/kernel.h>
8 #include <linux/types.h>
9 #include <linux/bpf.h>
10 #include <linux/bpf_verifier.h>
11 #include <linux/math64.h>
12 #include <linux/string.h>
13 
14 #define verbose(env, fmt, args...) bpf_verifier_log_write(env, fmt, ##args)
15 
16 static bool bpf_verifier_log_attr_valid(const struct bpf_verifier_log *log)
17 {
18 	/* ubuf and len_total should both be specified (or not) together */
19 	if (!!log->ubuf != !!log->len_total)
20 		return false;
21 	/* log buf without log_level is meaningless */
22 	if (log->ubuf && log->level == 0)
23 		return false;
24 	if (log->level & ~BPF_LOG_MASK)
25 		return false;
26 	if (log->len_total > UINT_MAX >> 2)
27 		return false;
28 	return true;
29 }
30 
31 int bpf_vlog_init(struct bpf_verifier_log *log, u32 log_level,
32 		  char __user *log_buf, u32 log_size)
33 {
34 	log->level = log_level;
35 	log->ubuf = log_buf;
36 	log->len_total = log_size;
37 
38 	/* log attributes have to be sane */
39 	if (!bpf_verifier_log_attr_valid(log))
40 		return -EINVAL;
41 
42 	return 0;
43 }
44 
45 static void bpf_vlog_update_len_max(struct bpf_verifier_log *log, u32 add_len)
46 {
47 	/* add_len includes terminal \0, so no need for +1. */
48 	u64 len = log->end_pos + add_len;
49 
50 	/* log->len_max could be larger than our current len due to
51 	 * bpf_vlog_reset() calls, so we maintain the max of any length at any
52 	 * previous point
53 	 */
54 	if (len > UINT_MAX)
55 		log->len_max = UINT_MAX;
56 	else if (len > log->len_max)
57 		log->len_max = len;
58 }
59 
60 void bpf_verifier_vlog(struct bpf_verifier_log *log, const char *fmt,
61 		       va_list args)
62 {
63 	u64 cur_pos;
64 	u32 new_n, n;
65 
66 	n = vscnprintf(log->kbuf, BPF_VERIFIER_TMP_LOG_SIZE, fmt, args);
67 
68 	if (log->level == BPF_LOG_KERNEL) {
69 		bool newline = n > 0 && log->kbuf[n - 1] == '\n';
70 
71 		pr_err("BPF: %s%s", log->kbuf, newline ? "" : "\n");
72 		return;
73 	}
74 
75 	n += 1; /* include terminating zero */
76 	bpf_vlog_update_len_max(log, n);
77 
78 	if (log->level & BPF_LOG_FIXED) {
79 		/* check if we have at least something to put into user buf */
80 		new_n = 0;
81 		if (log->end_pos < log->len_total) {
82 			new_n = min_t(u32, log->len_total - log->end_pos, n);
83 			log->kbuf[new_n - 1] = '\0';
84 		}
85 
86 		cur_pos = log->end_pos;
87 		log->end_pos += n - 1; /* don't count terminating '\0' */
88 
89 		if (log->ubuf && new_n &&
90 		    copy_to_user(log->ubuf + cur_pos, log->kbuf, new_n))
91 			goto fail;
92 	} else {
93 		u64 new_end, new_start;
94 		u32 buf_start, buf_end;
95 
96 		new_end = log->end_pos + n;
97 		if (new_end - log->start_pos >= log->len_total)
98 			new_start = new_end - log->len_total;
99 		else
100 			new_start = log->start_pos;
101 
102 		log->start_pos = new_start;
103 		log->end_pos = new_end - 1; /* don't count terminating '\0' */
104 
105 		if (!log->ubuf)
106 			return;
107 
108 		new_n = min(n, log->len_total);
109 		cur_pos = new_end - new_n;
110 		div_u64_rem(cur_pos, log->len_total, &buf_start);
111 		div_u64_rem(new_end, log->len_total, &buf_end);
112 		/* new_end and buf_end are exclusive indices, so if buf_end is
113 		 * exactly zero, then it actually points right to the end of
114 		 * ubuf and there is no wrap around
115 		 */
116 		if (buf_end == 0)
117 			buf_end = log->len_total;
118 
119 		/* if buf_start > buf_end, we wrapped around;
120 		 * if buf_start == buf_end, then we fill ubuf completely; we
121 		 * can't have buf_start == buf_end to mean that there is
122 		 * nothing to write, because we always write at least
123 		 * something, even if terminal '\0'
124 		 */
125 		if (buf_start < buf_end) {
126 			/* message fits within contiguous chunk of ubuf */
127 			if (copy_to_user(log->ubuf + buf_start,
128 					 log->kbuf + n - new_n,
129 					 buf_end - buf_start))
130 				goto fail;
131 		} else {
132 			/* message wraps around the end of ubuf, copy in two chunks */
133 			if (copy_to_user(log->ubuf + buf_start,
134 					 log->kbuf + n - new_n,
135 					 log->len_total - buf_start))
136 				goto fail;
137 			if (copy_to_user(log->ubuf,
138 					 log->kbuf + n - buf_end,
139 					 buf_end))
140 				goto fail;
141 		}
142 	}
143 
144 	return;
145 fail:
146 	log->ubuf = NULL;
147 }
148 
149 void bpf_vlog_reset(struct bpf_verifier_log *log, u64 new_pos)
150 {
151 	char zero = 0;
152 	u32 pos;
153 
154 	if (WARN_ON_ONCE(new_pos > log->end_pos))
155 		return;
156 
157 	if (!bpf_verifier_log_needed(log) || log->level == BPF_LOG_KERNEL)
158 		return;
159 
160 	/* if position to which we reset is beyond current log window,
161 	 * then we didn't preserve any useful content and should adjust
162 	 * start_pos to end up with an empty log (start_pos == end_pos)
163 	 */
164 	log->end_pos = new_pos;
165 	if (log->end_pos < log->start_pos)
166 		log->start_pos = log->end_pos;
167 
168 	if (!log->ubuf)
169 		return;
170 
171 	if (log->level & BPF_LOG_FIXED)
172 		pos = log->end_pos + 1;
173 	else
174 		div_u64_rem(new_pos, log->len_total, &pos);
175 
176 	if (pos < log->len_total && put_user(zero, log->ubuf + pos))
177 		log->ubuf = NULL;
178 }
179 
180 static void bpf_vlog_reverse_kbuf(char *buf, int len)
181 {
182 	int i, j;
183 
184 	for (i = 0, j = len - 1; i < j; i++, j--)
185 		swap(buf[i], buf[j]);
186 }
187 
188 static int bpf_vlog_reverse_ubuf(struct bpf_verifier_log *log, int start, int end)
189 {
190 	/* we split log->kbuf into two equal parts for both ends of array */
191 	int n = sizeof(log->kbuf) / 2, nn;
192 	char *lbuf = log->kbuf, *rbuf = log->kbuf + n;
193 
194 	/* Read ubuf's section [start, end) two chunks at a time, from left
195 	 * and right side; within each chunk, swap all the bytes; after that
196 	 * reverse the order of lbuf and rbuf and write result back to ubuf.
197 	 * This way we'll end up with swapped contents of specified
198 	 * [start, end) ubuf segment.
199 	 */
200 	while (end - start > 1) {
201 		nn = min(n, (end - start ) / 2);
202 
203 		if (copy_from_user(lbuf, log->ubuf + start, nn))
204 			return -EFAULT;
205 		if (copy_from_user(rbuf, log->ubuf + end - nn, nn))
206 			return -EFAULT;
207 
208 		bpf_vlog_reverse_kbuf(lbuf, nn);
209 		bpf_vlog_reverse_kbuf(rbuf, nn);
210 
211 		/* we write lbuf to the right end of ubuf, while rbuf to the
212 		 * left one to end up with properly reversed overall ubuf
213 		 */
214 		if (copy_to_user(log->ubuf + start, rbuf, nn))
215 			return -EFAULT;
216 		if (copy_to_user(log->ubuf + end - nn, lbuf, nn))
217 			return -EFAULT;
218 
219 		start += nn;
220 		end -= nn;
221 	}
222 
223 	return 0;
224 }
225 
226 int bpf_vlog_finalize(struct bpf_verifier_log *log, u32 *log_size_actual)
227 {
228 	u32 sublen;
229 	int err;
230 
231 	*log_size_actual = 0;
232 	if (!log || log->level == 0 || log->level == BPF_LOG_KERNEL)
233 		return 0;
234 
235 	if (!log->ubuf)
236 		goto skip_log_rotate;
237 	/* If we never truncated log, there is nothing to move around. */
238 	if (log->start_pos == 0)
239 		goto skip_log_rotate;
240 
241 	/* Otherwise we need to rotate log contents to make it start from the
242 	 * buffer beginning and be a continuous zero-terminated string. Note
243 	 * that if log->start_pos != 0 then we definitely filled up entire log
244 	 * buffer with no gaps, and we just need to shift buffer contents to
245 	 * the left by (log->start_pos % log->len_total) bytes.
246 	 *
247 	 * Unfortunately, user buffer could be huge and we don't want to
248 	 * allocate temporary kernel memory of the same size just to shift
249 	 * contents in a straightforward fashion. Instead, we'll be clever and
250 	 * do in-place array rotation. This is a leetcode-style problem, which
251 	 * could be solved by three rotations.
252 	 *
253 	 * Let's say we have log buffer that has to be shifted left by 7 bytes
254 	 * (spaces and vertical bar is just for demonstrative purposes):
255 	 *   E F G H I J K | A B C D
256 	 *
257 	 * First, we reverse entire array:
258 	 *   D C B A | K J I H G F E
259 	 *
260 	 * Then we rotate first 4 bytes (DCBA) and separately last 7 bytes
261 	 * (KJIHGFE), resulting in a properly rotated array:
262 	 *   A B C D | E F G H I J K
263 	 *
264 	 * We'll utilize log->kbuf to read user memory chunk by chunk, swap
265 	 * bytes, and write them back. Doing it byte-by-byte would be
266 	 * unnecessarily inefficient. Altogether we are going to read and
267 	 * write each byte twice, for total 4 memory copies between kernel and
268 	 * user space.
269 	 */
270 
271 	/* length of the chopped off part that will be the beginning;
272 	 * len(ABCD) in the example above
273 	 */
274 	div_u64_rem(log->start_pos, log->len_total, &sublen);
275 	sublen = log->len_total - sublen;
276 
277 	err = bpf_vlog_reverse_ubuf(log, 0, log->len_total);
278 	err = err ?: bpf_vlog_reverse_ubuf(log, 0, sublen);
279 	err = err ?: bpf_vlog_reverse_ubuf(log, sublen, log->len_total);
280 	if (err)
281 		log->ubuf = NULL;
282 
283 skip_log_rotate:
284 	*log_size_actual = log->len_max;
285 
286 	/* properly initialized log has either both ubuf!=NULL and len_total>0
287 	 * or ubuf==NULL and len_total==0, so if this condition doesn't hold,
288 	 * we got a fault somewhere along the way, so report it back
289 	 */
290 	if (!!log->ubuf != !!log->len_total)
291 		return -EFAULT;
292 
293 	/* did truncation actually happen? */
294 	if (log->ubuf && log->len_max > log->len_total)
295 		return -ENOSPC;
296 
297 	return 0;
298 }
299 
300 /* log_level controls verbosity level of eBPF verifier.
301  * bpf_verifier_log_write() is used to dump the verification trace to the log,
302  * so the user can figure out what's wrong with the program
303  */
304 __printf(2, 3) void bpf_verifier_log_write(struct bpf_verifier_env *env,
305 					   const char *fmt, ...)
306 {
307 	va_list args;
308 
309 	if (!bpf_verifier_log_needed(&env->log))
310 		return;
311 
312 	va_start(args, fmt);
313 	bpf_verifier_vlog(&env->log, fmt, args);
314 	va_end(args);
315 }
316 EXPORT_SYMBOL_GPL(bpf_verifier_log_write);
317 
318 __printf(2, 3) void bpf_log(struct bpf_verifier_log *log,
319 			    const char *fmt, ...)
320 {
321 	va_list args;
322 
323 	if (!bpf_verifier_log_needed(log))
324 		return;
325 
326 	va_start(args, fmt);
327 	bpf_verifier_vlog(log, fmt, args);
328 	va_end(args);
329 }
330 EXPORT_SYMBOL_GPL(bpf_log);
331 
332 static const struct bpf_line_info *
333 find_linfo(const struct bpf_verifier_env *env, u32 insn_off)
334 {
335 	const struct bpf_line_info *linfo;
336 	const struct bpf_prog *prog;
337 	u32 nr_linfo;
338 	int l, r, m;
339 
340 	prog = env->prog;
341 	nr_linfo = prog->aux->nr_linfo;
342 
343 	if (!nr_linfo || insn_off >= prog->len)
344 		return NULL;
345 
346 	linfo = prog->aux->linfo;
347 	/* Loop invariant: linfo[l].insn_off <= insns_off.
348 	 * linfo[0].insn_off == 0 which always satisfies above condition.
349 	 * Binary search is searching for rightmost linfo entry that satisfies
350 	 * the above invariant, giving us the desired record that covers given
351 	 * instruction offset.
352 	 */
353 	l = 0;
354 	r = nr_linfo - 1;
355 	while (l < r) {
356 		/* (r - l + 1) / 2 means we break a tie to the right, so if:
357 		 * l=1, r=2, linfo[l].insn_off <= insn_off, linfo[r].insn_off > insn_off,
358 		 * then m=2, we see that linfo[m].insn_off > insn_off, and so
359 		 * r becomes 1 and we exit the loop with correct l==1.
360 		 * If the tie was broken to the left, m=1 would end us up in
361 		 * an endless loop where l and m stay at 1 and r stays at 2.
362 		 */
363 		m = l + (r - l + 1) / 2;
364 		if (linfo[m].insn_off <= insn_off)
365 			l = m;
366 		else
367 			r = m - 1;
368 	}
369 
370 	return &linfo[l];
371 }
372 
373 static const char *ltrim(const char *s)
374 {
375 	while (isspace(*s))
376 		s++;
377 
378 	return s;
379 }
380 
381 __printf(3, 4) void verbose_linfo(struct bpf_verifier_env *env,
382 				  u32 insn_off,
383 				  const char *prefix_fmt, ...)
384 {
385 	const struct bpf_line_info *linfo, *prev_linfo;
386 	const struct btf *btf;
387 	const char *s, *fname;
388 
389 	if (!bpf_verifier_log_needed(&env->log))
390 		return;
391 
392 	prev_linfo = env->prev_linfo;
393 	linfo = find_linfo(env, insn_off);
394 	if (!linfo || linfo == prev_linfo)
395 		return;
396 
397 	/* It often happens that two separate linfo records point to the same
398 	 * source code line, but have differing column numbers. Given verifier
399 	 * log doesn't emit column information, from user perspective we just
400 	 * end up emitting the same source code line twice unnecessarily.
401 	 * So instead check that previous and current linfo record point to
402 	 * the same file (file_name_offs match) and the same line number, and
403 	 * avoid emitting duplicated source code line in such case.
404 	 */
405 	if (prev_linfo && linfo->file_name_off == prev_linfo->file_name_off &&
406 	    BPF_LINE_INFO_LINE_NUM(linfo->line_col) == BPF_LINE_INFO_LINE_NUM(prev_linfo->line_col))
407 		return;
408 
409 	if (prefix_fmt) {
410 		va_list args;
411 
412 		va_start(args, prefix_fmt);
413 		bpf_verifier_vlog(&env->log, prefix_fmt, args);
414 		va_end(args);
415 	}
416 
417 	btf = env->prog->aux->btf;
418 	s = ltrim(btf_name_by_offset(btf, linfo->line_off));
419 	verbose(env, "%s", s); /* source code line */
420 
421 	s = btf_name_by_offset(btf, linfo->file_name_off);
422 	/* leave only file name */
423 	fname = strrchr(s, '/');
424 	fname = fname ? fname + 1 : s;
425 	verbose(env, " @ %s:%u\n", fname, BPF_LINE_INFO_LINE_NUM(linfo->line_col));
426 
427 	env->prev_linfo = linfo;
428 }
429 
430 static const char *btf_type_name(const struct btf *btf, u32 id)
431 {
432 	return btf_name_by_offset(btf, btf_type_by_id(btf, id)->name_off);
433 }
434 
435 /* string representation of 'enum bpf_reg_type'
436  *
437  * Note that reg_type_str() can not appear more than once in a single verbose()
438  * statement.
439  */
440 const char *reg_type_str(struct bpf_verifier_env *env, enum bpf_reg_type type)
441 {
442 	char postfix[16] = {0}, prefix[64] = {0};
443 	static const char * const str[] = {
444 		[NOT_INIT]		= "?",
445 		[SCALAR_VALUE]		= "scalar",
446 		[PTR_TO_CTX]		= "ctx",
447 		[CONST_PTR_TO_MAP]	= "map_ptr",
448 		[PTR_TO_MAP_VALUE]	= "map_value",
449 		[PTR_TO_STACK]		= "fp",
450 		[PTR_TO_PACKET]		= "pkt",
451 		[PTR_TO_PACKET_META]	= "pkt_meta",
452 		[PTR_TO_PACKET_END]	= "pkt_end",
453 		[PTR_TO_FLOW_KEYS]	= "flow_keys",
454 		[PTR_TO_SOCKET]		= "sock",
455 		[PTR_TO_SOCK_COMMON]	= "sock_common",
456 		[PTR_TO_TCP_SOCK]	= "tcp_sock",
457 		[PTR_TO_TP_BUFFER]	= "tp_buffer",
458 		[PTR_TO_XDP_SOCK]	= "xdp_sock",
459 		[PTR_TO_BTF_ID]		= "ptr_",
460 		[PTR_TO_MEM]		= "mem",
461 		[PTR_TO_ARENA]		= "arena",
462 		[PTR_TO_BUF]		= "buf",
463 		[PTR_TO_FUNC]		= "func",
464 		[PTR_TO_MAP_KEY]	= "map_key",
465 		[CONST_PTR_TO_DYNPTR]	= "dynptr_ptr",
466 	};
467 
468 	if (type & PTR_MAYBE_NULL) {
469 		if (base_type(type) == PTR_TO_BTF_ID)
470 			strscpy(postfix, "or_null_");
471 		else
472 			strscpy(postfix, "_or_null");
473 	}
474 
475 	snprintf(prefix, sizeof(prefix), "%s%s%s%s%s%s%s",
476 		 type & MEM_RDONLY ? "rdonly_" : "",
477 		 type & MEM_RINGBUF ? "ringbuf_" : "",
478 		 type & MEM_USER ? "user_" : "",
479 		 type & MEM_PERCPU ? "percpu_" : "",
480 		 type & MEM_RCU ? "rcu_" : "",
481 		 type & PTR_UNTRUSTED ? "untrusted_" : "",
482 		 type & PTR_TRUSTED ? "trusted_" : ""
483 	);
484 
485 	snprintf(env->tmp_str_buf, TMP_STR_BUF_LEN, "%s%s%s",
486 		 prefix, str[base_type(type)], postfix);
487 	return env->tmp_str_buf;
488 }
489 
490 const char *dynptr_type_str(enum bpf_dynptr_type type)
491 {
492 	switch (type) {
493 	case BPF_DYNPTR_TYPE_LOCAL:
494 		return "local";
495 	case BPF_DYNPTR_TYPE_RINGBUF:
496 		return "ringbuf";
497 	case BPF_DYNPTR_TYPE_SKB:
498 		return "skb";
499 	case BPF_DYNPTR_TYPE_XDP:
500 		return "xdp";
501 	case BPF_DYNPTR_TYPE_INVALID:
502 		return "<invalid>";
503 	default:
504 		WARN_ONCE(1, "unknown dynptr type %d\n", type);
505 		return "<unknown>";
506 	}
507 }
508 
509 const char *iter_type_str(const struct btf *btf, u32 btf_id)
510 {
511 	if (!btf || btf_id == 0)
512 		return "<invalid>";
513 
514 	/* we already validated that type is valid and has conforming name */
515 	return btf_type_name(btf, btf_id) + sizeof(ITER_PREFIX) - 1;
516 }
517 
518 const char *iter_state_str(enum bpf_iter_state state)
519 {
520 	switch (state) {
521 	case BPF_ITER_STATE_ACTIVE:
522 		return "active";
523 	case BPF_ITER_STATE_DRAINED:
524 		return "drained";
525 	case BPF_ITER_STATE_INVALID:
526 		return "<invalid>";
527 	default:
528 		WARN_ONCE(1, "unknown iter state %d\n", state);
529 		return "<unknown>";
530 	}
531 }
532 
533 static char slot_type_char[] = {
534 	[STACK_INVALID]	= '?',
535 	[STACK_SPILL]	= 'r',
536 	[STACK_MISC]	= 'm',
537 	[STACK_ZERO]	= '0',
538 	[STACK_DYNPTR]	= 'd',
539 	[STACK_ITER]	= 'i',
540 };
541 
542 static void print_liveness(struct bpf_verifier_env *env,
543 			   enum bpf_reg_liveness live)
544 {
545 	if (live & (REG_LIVE_READ | REG_LIVE_WRITTEN | REG_LIVE_DONE))
546 	    verbose(env, "_");
547 	if (live & REG_LIVE_READ)
548 		verbose(env, "r");
549 	if (live & REG_LIVE_WRITTEN)
550 		verbose(env, "w");
551 	if (live & REG_LIVE_DONE)
552 		verbose(env, "D");
553 }
554 
555 #define UNUM_MAX_DECIMAL U16_MAX
556 #define SNUM_MAX_DECIMAL S16_MAX
557 #define SNUM_MIN_DECIMAL S16_MIN
558 
559 static bool is_unum_decimal(u64 num)
560 {
561 	return num <= UNUM_MAX_DECIMAL;
562 }
563 
564 static bool is_snum_decimal(s64 num)
565 {
566 	return num >= SNUM_MIN_DECIMAL && num <= SNUM_MAX_DECIMAL;
567 }
568 
569 static void verbose_unum(struct bpf_verifier_env *env, u64 num)
570 {
571 	if (is_unum_decimal(num))
572 		verbose(env, "%llu", num);
573 	else
574 		verbose(env, "%#llx", num);
575 }
576 
577 static void verbose_snum(struct bpf_verifier_env *env, s64 num)
578 {
579 	if (is_snum_decimal(num))
580 		verbose(env, "%lld", num);
581 	else
582 		verbose(env, "%#llx", num);
583 }
584 
585 int tnum_strn(char *str, size_t size, struct tnum a)
586 {
587 	/* print as a constant, if tnum is fully known */
588 	if (a.mask == 0) {
589 		if (is_unum_decimal(a.value))
590 			return snprintf(str, size, "%llu", a.value);
591 		else
592 			return snprintf(str, size, "%#llx", a.value);
593 	}
594 	return snprintf(str, size, "(%#llx; %#llx)", a.value, a.mask);
595 }
596 EXPORT_SYMBOL_GPL(tnum_strn);
597 
598 static void print_scalar_ranges(struct bpf_verifier_env *env,
599 				const struct bpf_reg_state *reg,
600 				const char **sep)
601 {
602 	/* For signed ranges, we want to unify 64-bit and 32-bit values in the
603 	 * output as much as possible, but there is a bit of a complication.
604 	 * If we choose to print values as decimals, this is natural to do,
605 	 * because negative 64-bit and 32-bit values >= -S32_MIN have the same
606 	 * representation due to sign extension. But if we choose to print
607 	 * them in hex format (see is_snum_decimal()), then sign extension is
608 	 * misleading.
609 	 * E.g., smin=-2 and smin32=-2 are exactly the same in decimal, but in
610 	 * hex they will be smin=0xfffffffffffffffe and smin32=0xfffffffe, two
611 	 * very different numbers.
612 	 * So we avoid sign extension if we choose to print values in hex.
613 	 */
614 	struct {
615 		const char *name;
616 		u64 val;
617 		bool omit;
618 	} minmaxs[] = {
619 		{"smin",   reg->smin_value,         reg->smin_value == S64_MIN},
620 		{"smax",   reg->smax_value,         reg->smax_value == S64_MAX},
621 		{"umin",   reg->umin_value,         reg->umin_value == 0},
622 		{"umax",   reg->umax_value,         reg->umax_value == U64_MAX},
623 		{"smin32",
624 		 is_snum_decimal((s64)reg->s32_min_value)
625 			 ? (s64)reg->s32_min_value
626 			 : (u32)reg->s32_min_value, reg->s32_min_value == S32_MIN},
627 		{"smax32",
628 		 is_snum_decimal((s64)reg->s32_max_value)
629 			 ? (s64)reg->s32_max_value
630 			 : (u32)reg->s32_max_value, reg->s32_max_value == S32_MAX},
631 		{"umin32", reg->u32_min_value,      reg->u32_min_value == 0},
632 		{"umax32", reg->u32_max_value,      reg->u32_max_value == U32_MAX},
633 	}, *m1, *m2, *mend = &minmaxs[ARRAY_SIZE(minmaxs)];
634 	bool neg1, neg2;
635 
636 	for (m1 = &minmaxs[0]; m1 < mend; m1++) {
637 		if (m1->omit)
638 			continue;
639 
640 		neg1 = m1->name[0] == 's' && (s64)m1->val < 0;
641 
642 		verbose(env, "%s%s=", *sep, m1->name);
643 		*sep = ",";
644 
645 		for (m2 = m1 + 2; m2 < mend; m2 += 2) {
646 			if (m2->omit || m2->val != m1->val)
647 				continue;
648 			/* don't mix negatives with positives */
649 			neg2 = m2->name[0] == 's' && (s64)m2->val < 0;
650 			if (neg2 != neg1)
651 				continue;
652 			m2->omit = true;
653 			verbose(env, "%s=", m2->name);
654 		}
655 
656 		if (m1->name[0] == 's')
657 			verbose_snum(env, m1->val);
658 		else
659 			verbose_unum(env, m1->val);
660 	}
661 }
662 
663 static bool type_is_map_ptr(enum bpf_reg_type t) {
664 	switch (base_type(t)) {
665 	case CONST_PTR_TO_MAP:
666 	case PTR_TO_MAP_KEY:
667 	case PTR_TO_MAP_VALUE:
668 		return true;
669 	default:
670 		return false;
671 	}
672 }
673 
674 /*
675  * _a stands for append, was shortened to avoid multiline statements below.
676  * This macro is used to output a comma separated list of attributes.
677  */
678 #define verbose_a(fmt, ...) ({ verbose(env, "%s" fmt, sep, ##__VA_ARGS__); sep = ","; })
679 
680 static void print_reg_state(struct bpf_verifier_env *env,
681 			    const struct bpf_func_state *state,
682 			    const struct bpf_reg_state *reg)
683 {
684 	enum bpf_reg_type t;
685 	const char *sep = "";
686 
687 	t = reg->type;
688 	if (t == SCALAR_VALUE && reg->precise)
689 		verbose(env, "P");
690 	if (t == SCALAR_VALUE && tnum_is_const(reg->var_off)) {
691 		/* reg->off should be 0 for SCALAR_VALUE */
692 		verbose_snum(env, reg->var_off.value + reg->off);
693 		return;
694 	}
695 
696 	verbose(env, "%s", reg_type_str(env, t));
697 	if (t == PTR_TO_ARENA)
698 		return;
699 	if (t == PTR_TO_STACK) {
700 		if (state->frameno != reg->frameno)
701 			verbose(env, "[%d]", reg->frameno);
702 		if (tnum_is_const(reg->var_off)) {
703 			verbose_snum(env, reg->var_off.value + reg->off);
704 			return;
705 		}
706 	}
707 	if (base_type(t) == PTR_TO_BTF_ID)
708 		verbose(env, "%s", btf_type_name(reg->btf, reg->btf_id));
709 	verbose(env, "(");
710 	if (reg->id)
711 		verbose_a("id=%d", reg->id & ~BPF_ADD_CONST);
712 	if (reg->id & BPF_ADD_CONST)
713 		verbose(env, "%+d", reg->off);
714 	if (reg->ref_obj_id)
715 		verbose_a("ref_obj_id=%d", reg->ref_obj_id);
716 	if (type_is_non_owning_ref(reg->type))
717 		verbose_a("%s", "non_own_ref");
718 	if (type_is_map_ptr(t)) {
719 		if (reg->map_ptr->name[0])
720 			verbose_a("map=%s", reg->map_ptr->name);
721 		verbose_a("ks=%d,vs=%d",
722 			  reg->map_ptr->key_size,
723 			  reg->map_ptr->value_size);
724 	}
725 	if (t != SCALAR_VALUE && reg->off) {
726 		verbose_a("off=");
727 		verbose_snum(env, reg->off);
728 	}
729 	if (type_is_pkt_pointer(t)) {
730 		verbose_a("r=");
731 		verbose_unum(env, reg->range);
732 	}
733 	if (base_type(t) == PTR_TO_MEM) {
734 		verbose_a("sz=");
735 		verbose_unum(env, reg->mem_size);
736 	}
737 	if (t == CONST_PTR_TO_DYNPTR)
738 		verbose_a("type=%s",  dynptr_type_str(reg->dynptr.type));
739 	if (tnum_is_const(reg->var_off)) {
740 		/* a pointer register with fixed offset */
741 		if (reg->var_off.value) {
742 			verbose_a("imm=");
743 			verbose_snum(env, reg->var_off.value);
744 		}
745 	} else {
746 		print_scalar_ranges(env, reg, &sep);
747 		if (!tnum_is_unknown(reg->var_off)) {
748 			char tn_buf[48];
749 
750 			tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off);
751 			verbose_a("var_off=%s", tn_buf);
752 		}
753 	}
754 	verbose(env, ")");
755 }
756 
757 void print_verifier_state(struct bpf_verifier_env *env, const struct bpf_func_state *state,
758 			  bool print_all)
759 {
760 	const struct bpf_reg_state *reg;
761 	int i;
762 
763 	if (state->frameno)
764 		verbose(env, " frame%d:", state->frameno);
765 	for (i = 0; i < MAX_BPF_REG; i++) {
766 		reg = &state->regs[i];
767 		if (reg->type == NOT_INIT)
768 			continue;
769 		if (!print_all && !reg_scratched(env, i))
770 			continue;
771 		verbose(env, " R%d", i);
772 		print_liveness(env, reg->live);
773 		verbose(env, "=");
774 		print_reg_state(env, state, reg);
775 	}
776 	for (i = 0; i < state->allocated_stack / BPF_REG_SIZE; i++) {
777 		char types_buf[BPF_REG_SIZE + 1];
778 		const char *sep = "";
779 		bool valid = false;
780 		u8 slot_type;
781 		int j;
782 
783 		if (!print_all && !stack_slot_scratched(env, i))
784 			continue;
785 
786 		for (j = 0; j < BPF_REG_SIZE; j++) {
787 			slot_type = state->stack[i].slot_type[j];
788 			if (slot_type != STACK_INVALID)
789 				valid = true;
790 			types_buf[j] = slot_type_char[slot_type];
791 		}
792 		types_buf[BPF_REG_SIZE] = 0;
793 		if (!valid)
794 			continue;
795 
796 		reg = &state->stack[i].spilled_ptr;
797 		switch (state->stack[i].slot_type[BPF_REG_SIZE - 1]) {
798 		case STACK_SPILL:
799 			/* print MISC/ZERO/INVALID slots above subreg spill */
800 			for (j = 0; j < BPF_REG_SIZE; j++)
801 				if (state->stack[i].slot_type[j] == STACK_SPILL)
802 					break;
803 			types_buf[j] = '\0';
804 
805 			verbose(env, " fp%d", (-i - 1) * BPF_REG_SIZE);
806 			print_liveness(env, reg->live);
807 			verbose(env, "=%s", types_buf);
808 			print_reg_state(env, state, reg);
809 			break;
810 		case STACK_DYNPTR:
811 			/* skip to main dynptr slot */
812 			i += BPF_DYNPTR_NR_SLOTS - 1;
813 			reg = &state->stack[i].spilled_ptr;
814 
815 			verbose(env, " fp%d", (-i - 1) * BPF_REG_SIZE);
816 			print_liveness(env, reg->live);
817 			verbose(env, "=dynptr_%s(", dynptr_type_str(reg->dynptr.type));
818 			if (reg->id)
819 				verbose_a("id=%d", reg->id);
820 			if (reg->ref_obj_id)
821 				verbose_a("ref_id=%d", reg->ref_obj_id);
822 			if (reg->dynptr_id)
823 				verbose_a("dynptr_id=%d", reg->dynptr_id);
824 			verbose(env, ")");
825 			break;
826 		case STACK_ITER:
827 			/* only main slot has ref_obj_id set; skip others */
828 			if (!reg->ref_obj_id)
829 				continue;
830 
831 			verbose(env, " fp%d", (-i - 1) * BPF_REG_SIZE);
832 			print_liveness(env, reg->live);
833 			verbose(env, "=iter_%s(ref_id=%d,state=%s,depth=%u)",
834 				iter_type_str(reg->iter.btf, reg->iter.btf_id),
835 				reg->ref_obj_id, iter_state_str(reg->iter.state),
836 				reg->iter.depth);
837 			break;
838 		case STACK_MISC:
839 		case STACK_ZERO:
840 		default:
841 			verbose(env, " fp%d", (-i - 1) * BPF_REG_SIZE);
842 			print_liveness(env, reg->live);
843 			verbose(env, "=%s", types_buf);
844 			break;
845 		}
846 	}
847 	if (state->acquired_refs && state->refs[0].id) {
848 		verbose(env, " refs=%d", state->refs[0].id);
849 		for (i = 1; i < state->acquired_refs; i++)
850 			if (state->refs[i].id)
851 				verbose(env, ",%d", state->refs[i].id);
852 	}
853 	if (state->in_callback_fn)
854 		verbose(env, " cb");
855 	if (state->in_async_callback_fn)
856 		verbose(env, " async_cb");
857 	verbose(env, "\n");
858 	if (!print_all)
859 		mark_verifier_state_clean(env);
860 }
861 
862 static inline u32 vlog_alignment(u32 pos)
863 {
864 	return round_up(max(pos + BPF_LOG_MIN_ALIGNMENT / 2, BPF_LOG_ALIGNMENT),
865 			BPF_LOG_MIN_ALIGNMENT) - pos - 1;
866 }
867 
868 void print_insn_state(struct bpf_verifier_env *env, const struct bpf_func_state *state)
869 {
870 	if (env->prev_log_pos && env->prev_log_pos == env->log.end_pos) {
871 		/* remove new line character */
872 		bpf_vlog_reset(&env->log, env->prev_log_pos - 1);
873 		verbose(env, "%*c;", vlog_alignment(env->prev_insn_print_pos), ' ');
874 	} else {
875 		verbose(env, "%d:", env->insn_idx);
876 	}
877 	print_verifier_state(env, state, false);
878 }
879