1 // SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause)
2 /* Copyright (c) 2022 Meta Platforms, Inc. and affiliates. */
3 #define _GNU_SOURCE
4 #include <argp.h>
5 #include <string.h>
6 #include <stdlib.h>
7 #include <sched.h>
8 #include <pthread.h>
9 #include <dirent.h>
10 #include <signal.h>
11 #include <fcntl.h>
12 #include <unistd.h>
13 #include <sys/time.h>
14 #include <sys/sysinfo.h>
15 #include <sys/stat.h>
16 #include <bpf/libbpf.h>
17 #include <bpf/btf.h>
18 #include <libelf.h>
19 #include <gelf.h>
20 #include <float.h>
21 
22 #ifndef ARRAY_SIZE
23 #define ARRAY_SIZE(arr) (sizeof(arr) / sizeof((arr)[0]))
24 #endif
25 
26 enum stat_id {
27 	VERDICT,
28 	DURATION,
29 	TOTAL_INSNS,
30 	TOTAL_STATES,
31 	PEAK_STATES,
32 	MAX_STATES_PER_INSN,
33 	MARK_READ_MAX_LEN,
34 
35 	FILE_NAME,
36 	PROG_NAME,
37 
38 	ALL_STATS_CNT,
39 	NUM_STATS_CNT = FILE_NAME - VERDICT,
40 };
41 
42 /* In comparison mode each stat can specify up to four different values:
43  *   - A side value;
44  *   - B side value;
45  *   - absolute diff value;
46  *   - relative (percentage) diff value.
47  *
48  * When specifying stat specs in comparison mode, user can use one of the
49  * following variant suffixes to specify which exact variant should be used for
50  * ordering or filtering:
51  *   - `_a` for A side value;
52  *   - `_b` for B side value;
53  *   - `_diff` for absolute diff value;
54  *   - `_pct` for relative (percentage) diff value.
55  *
56  * If no variant suffix is provided, then `_b` (control data) is assumed.
57  *
58  * As an example, let's say instructions stat has the following output:
59  *
60  * Insns (A)  Insns (B)  Insns   (DIFF)
61  * ---------  ---------  --------------
62  * 21547      20920       -627 (-2.91%)
63  *
64  * Then:
65  *   - 21547 is A side value (insns_a);
66  *   - 20920 is B side value (insns_b);
67  *   - -627 is absolute diff value (insns_diff);
68  *   - -2.91% is relative diff value (insns_pct).
69  *
70  * For verdict there is no verdict_pct variant.
71  * For file and program name, _a and _b variants are equivalent and there are
72  * no _diff or _pct variants.
73  */
74 enum stat_variant {
75 	VARIANT_A,
76 	VARIANT_B,
77 	VARIANT_DIFF,
78 	VARIANT_PCT,
79 };
80 
81 struct verif_stats {
82 	char *file_name;
83 	char *prog_name;
84 
85 	long stats[NUM_STATS_CNT];
86 };
87 
88 /* joined comparison mode stats */
89 struct verif_stats_join {
90 	char *file_name;
91 	char *prog_name;
92 
93 	const struct verif_stats *stats_a;
94 	const struct verif_stats *stats_b;
95 };
96 
97 struct stat_specs {
98 	int spec_cnt;
99 	enum stat_id ids[ALL_STATS_CNT];
100 	enum stat_variant variants[ALL_STATS_CNT];
101 	bool asc[ALL_STATS_CNT];
102 	int lens[ALL_STATS_CNT * 3]; /* 3x for comparison mode */
103 };
104 
105 enum resfmt {
106 	RESFMT_TABLE,
107 	RESFMT_TABLE_CALCLEN, /* fake format to pre-calculate table's column widths */
108 	RESFMT_CSV,
109 };
110 
111 enum filter_kind {
112 	FILTER_NAME,
113 	FILTER_STAT,
114 };
115 
116 enum operator_kind {
117 	OP_EQ,		/* == or = */
118 	OP_NEQ,		/* != or <> */
119 	OP_LT,		/* < */
120 	OP_LE,		/* <= */
121 	OP_GT,		/* > */
122 	OP_GE,		/* >= */
123 };
124 
125 struct filter {
126 	enum filter_kind kind;
127 	/* FILTER_NAME */
128 	char *any_glob;
129 	char *file_glob;
130 	char *prog_glob;
131 	/* FILTER_STAT */
132 	enum operator_kind op;
133 	int stat_id;
134 	enum stat_variant stat_var;
135 	long value;
136 };
137 
138 static struct env {
139 	char **filenames;
140 	int filename_cnt;
141 	bool verbose;
142 	bool debug;
143 	bool quiet;
144 	bool force_checkpoints;
145 	enum resfmt out_fmt;
146 	bool show_version;
147 	bool comparison_mode;
148 	bool replay_mode;
149 
150 	int log_level;
151 	int log_size;
152 	bool log_fixed;
153 
154 	struct verif_stats *prog_stats;
155 	int prog_stat_cnt;
156 
157 	/* baseline_stats is allocated and used only in comparison mode */
158 	struct verif_stats *baseline_stats;
159 	int baseline_stat_cnt;
160 
161 	struct verif_stats_join *join_stats;
162 	int join_stat_cnt;
163 
164 	struct stat_specs output_spec;
165 	struct stat_specs sort_spec;
166 
167 	struct filter *allow_filters;
168 	struct filter *deny_filters;
169 	int allow_filter_cnt;
170 	int deny_filter_cnt;
171 
172 	int files_processed;
173 	int files_skipped;
174 	int progs_processed;
175 	int progs_skipped;
176 } env;
177 
libbpf_print_fn(enum libbpf_print_level level,const char * format,va_list args)178 static int libbpf_print_fn(enum libbpf_print_level level, const char *format, va_list args)
179 {
180 	if (!env.verbose)
181 		return 0;
182 	if (level == LIBBPF_DEBUG  && !env.debug)
183 		return 0;
184 	return vfprintf(stderr, format, args);
185 }
186 
187 #ifndef VERISTAT_VERSION
188 #define VERISTAT_VERSION "<kernel>"
189 #endif
190 
191 const char *argp_program_version = "veristat v" VERISTAT_VERSION;
192 const char *argp_program_bug_address = "<bpf@vger.kernel.org>";
193 const char argp_program_doc[] =
194 "veristat    BPF verifier stats collection and comparison tool.\n"
195 "\n"
196 "USAGE: veristat <obj-file> [<obj-file>...]\n"
197 "   OR: veristat -C <baseline.csv> <comparison.csv>\n"
198 "   OR: veristat -R <results.csv>\n";
199 
200 enum {
201 	OPT_LOG_FIXED = 1000,
202 	OPT_LOG_SIZE = 1001,
203 };
204 
205 static const struct argp_option opts[] = {
206 	{ NULL, 'h', NULL, OPTION_HIDDEN, "Show the full help" },
207 	{ "version", 'V', NULL, 0, "Print version" },
208 	{ "verbose", 'v', NULL, 0, "Verbose mode" },
209 	{ "debug", 'd', NULL, 0, "Debug mode (turns on libbpf debug logging)" },
210 	{ "log-level", 'l', "LEVEL", 0, "Verifier log level (default 0 for normal mode, 1 for verbose mode)" },
211 	{ "log-fixed", OPT_LOG_FIXED, NULL, 0, "Disable verifier log rotation" },
212 	{ "log-size", OPT_LOG_SIZE, "BYTES", 0, "Customize verifier log size (default to 16MB)" },
213 	{ "test-states", 't', NULL, 0,
214 	  "Force frequent BPF verifier state checkpointing (set BPF_F_TEST_STATE_FREQ program flag)" },
215 	{ "quiet", 'q', NULL, 0, "Quiet mode" },
216 	{ "emit", 'e', "SPEC", 0, "Specify stats to be emitted" },
217 	{ "sort", 's', "SPEC", 0, "Specify sort order" },
218 	{ "output-format", 'o', "FMT", 0, "Result output format (table, csv), default is table." },
219 	{ "compare", 'C', NULL, 0, "Comparison mode" },
220 	{ "replay", 'R', NULL, 0, "Replay mode" },
221 	{ "filter", 'f', "FILTER", 0, "Filter expressions (or @filename for file with expressions)." },
222 	{},
223 };
224 
225 static int parse_stats(const char *stats_str, struct stat_specs *specs);
226 static int append_filter(struct filter **filters, int *cnt, const char *str);
227 static int append_filter_file(const char *path);
228 
parse_arg(int key,char * arg,struct argp_state * state)229 static error_t parse_arg(int key, char *arg, struct argp_state *state)
230 {
231 	void *tmp;
232 	int err;
233 
234 	switch (key) {
235 	case 'h':
236 		argp_state_help(state, stderr, ARGP_HELP_STD_HELP);
237 		break;
238 	case 'V':
239 		env.show_version = true;
240 		break;
241 	case 'v':
242 		env.verbose = true;
243 		break;
244 	case 'd':
245 		env.debug = true;
246 		env.verbose = true;
247 		break;
248 	case 'q':
249 		env.quiet = true;
250 		break;
251 	case 'e':
252 		err = parse_stats(arg, &env.output_spec);
253 		if (err)
254 			return err;
255 		break;
256 	case 's':
257 		err = parse_stats(arg, &env.sort_spec);
258 		if (err)
259 			return err;
260 		break;
261 	case 'o':
262 		if (strcmp(arg, "table") == 0) {
263 			env.out_fmt = RESFMT_TABLE;
264 		} else if (strcmp(arg, "csv") == 0) {
265 			env.out_fmt = RESFMT_CSV;
266 		} else {
267 			fprintf(stderr, "Unrecognized output format '%s'\n", arg);
268 			return -EINVAL;
269 		}
270 		break;
271 	case 'l':
272 		errno = 0;
273 		env.log_level = strtol(arg, NULL, 10);
274 		if (errno) {
275 			fprintf(stderr, "invalid log level: %s\n", arg);
276 			argp_usage(state);
277 		}
278 		break;
279 	case OPT_LOG_FIXED:
280 		env.log_fixed = true;
281 		break;
282 	case OPT_LOG_SIZE:
283 		errno = 0;
284 		env.log_size = strtol(arg, NULL, 10);
285 		if (errno) {
286 			fprintf(stderr, "invalid log size: %s\n", arg);
287 			argp_usage(state);
288 		}
289 		break;
290 	case 't':
291 		env.force_checkpoints = true;
292 		break;
293 	case 'C':
294 		env.comparison_mode = true;
295 		break;
296 	case 'R':
297 		env.replay_mode = true;
298 		break;
299 	case 'f':
300 		if (arg[0] == '@')
301 			err = append_filter_file(arg + 1);
302 		else if (arg[0] == '!')
303 			err = append_filter(&env.deny_filters, &env.deny_filter_cnt, arg + 1);
304 		else
305 			err = append_filter(&env.allow_filters, &env.allow_filter_cnt, arg);
306 		if (err) {
307 			fprintf(stderr, "Failed to collect program filter expressions: %d\n", err);
308 			return err;
309 		}
310 		break;
311 	case ARGP_KEY_ARG:
312 		tmp = realloc(env.filenames, (env.filename_cnt + 1) * sizeof(*env.filenames));
313 		if (!tmp)
314 			return -ENOMEM;
315 		env.filenames = tmp;
316 		env.filenames[env.filename_cnt] = strdup(arg);
317 		if (!env.filenames[env.filename_cnt])
318 			return -ENOMEM;
319 		env.filename_cnt++;
320 		break;
321 	default:
322 		return ARGP_ERR_UNKNOWN;
323 	}
324 	return 0;
325 }
326 
327 static const struct argp argp = {
328 	.options = opts,
329 	.parser = parse_arg,
330 	.doc = argp_program_doc,
331 };
332 
333 
334 /* Adapted from perf/util/string.c */
glob_matches(const char * str,const char * pat)335 static bool glob_matches(const char *str, const char *pat)
336 {
337 	while (*str && *pat && *pat != '*') {
338 		if (*str != *pat)
339 			return false;
340 		str++;
341 		pat++;
342 	}
343 	/* Check wild card */
344 	if (*pat == '*') {
345 		while (*pat == '*')
346 			pat++;
347 		if (!*pat) /* Tail wild card matches all */
348 			return true;
349 		while (*str)
350 			if (glob_matches(str++, pat))
351 				return true;
352 	}
353 	return !*str && !*pat;
354 }
355 
is_bpf_obj_file(const char * path)356 static bool is_bpf_obj_file(const char *path) {
357 	Elf64_Ehdr *ehdr;
358 	int fd, err = -EINVAL;
359 	Elf *elf = NULL;
360 
361 	fd = open(path, O_RDONLY | O_CLOEXEC);
362 	if (fd < 0)
363 		return true; /* we'll fail later and propagate error */
364 
365 	/* ensure libelf is initialized */
366 	(void)elf_version(EV_CURRENT);
367 
368 	elf = elf_begin(fd, ELF_C_READ, NULL);
369 	if (!elf)
370 		goto cleanup;
371 
372 	if (elf_kind(elf) != ELF_K_ELF || gelf_getclass(elf) != ELFCLASS64)
373 		goto cleanup;
374 
375 	ehdr = elf64_getehdr(elf);
376 	/* Old LLVM set e_machine to EM_NONE */
377 	if (!ehdr || ehdr->e_type != ET_REL || (ehdr->e_machine && ehdr->e_machine != EM_BPF))
378 		goto cleanup;
379 
380 	err = 0;
381 cleanup:
382 	if (elf)
383 		elf_end(elf);
384 	close(fd);
385 	return err == 0;
386 }
387 
should_process_file_prog(const char * filename,const char * prog_name)388 static bool should_process_file_prog(const char *filename, const char *prog_name)
389 {
390 	struct filter *f;
391 	int i, allow_cnt = 0;
392 
393 	for (i = 0; i < env.deny_filter_cnt; i++) {
394 		f = &env.deny_filters[i];
395 		if (f->kind != FILTER_NAME)
396 			continue;
397 
398 		if (f->any_glob && glob_matches(filename, f->any_glob))
399 			return false;
400 		if (f->any_glob && prog_name && glob_matches(prog_name, f->any_glob))
401 			return false;
402 		if (f->file_glob && glob_matches(filename, f->file_glob))
403 			return false;
404 		if (f->prog_glob && prog_name && glob_matches(prog_name, f->prog_glob))
405 			return false;
406 	}
407 
408 	for (i = 0; i < env.allow_filter_cnt; i++) {
409 		f = &env.allow_filters[i];
410 		if (f->kind != FILTER_NAME)
411 			continue;
412 
413 		allow_cnt++;
414 		if (f->any_glob) {
415 			if (glob_matches(filename, f->any_glob))
416 				return true;
417 			/* If we don't know program name yet, any_glob filter
418 			 * has to assume that current BPF object file might be
419 			 * relevant; we'll check again later on after opening
420 			 * BPF object file, at which point program name will
421 			 * be known finally.
422 			 */
423 			if (!prog_name || glob_matches(prog_name, f->any_glob))
424 				return true;
425 		} else {
426 			if (f->file_glob && !glob_matches(filename, f->file_glob))
427 				continue;
428 			if (f->prog_glob && prog_name && !glob_matches(prog_name, f->prog_glob))
429 				continue;
430 			return true;
431 		}
432 	}
433 
434 	/* if there are no file/prog name allow filters, allow all progs,
435 	 * unless they are denied earlier explicitly
436 	 */
437 	return allow_cnt == 0;
438 }
439 
440 static struct {
441 	enum operator_kind op_kind;
442 	const char *op_str;
443 } operators[] = {
444 	/* Order of these definitions matter to avoid situations like '<'
445 	 * matching part of what is actually a '<>' operator. That is,
446 	 * substrings should go last.
447 	 */
448 	{ OP_EQ, "==" },
449 	{ OP_NEQ, "!=" },
450 	{ OP_NEQ, "<>" },
451 	{ OP_LE, "<=" },
452 	{ OP_LT, "<" },
453 	{ OP_GE, ">=" },
454 	{ OP_GT, ">" },
455 	{ OP_EQ, "=" },
456 };
457 
458 static bool parse_stat_id_var(const char *name, size_t len, int *id, enum stat_variant *var);
459 
append_filter(struct filter ** filters,int * cnt,const char * str)460 static int append_filter(struct filter **filters, int *cnt, const char *str)
461 {
462 	struct filter *f;
463 	void *tmp;
464 	const char *p;
465 	int i;
466 
467 	tmp = realloc(*filters, (*cnt + 1) * sizeof(**filters));
468 	if (!tmp)
469 		return -ENOMEM;
470 	*filters = tmp;
471 
472 	f = &(*filters)[*cnt];
473 	memset(f, 0, sizeof(*f));
474 
475 	/* First, let's check if it's a stats filter of the following form:
476 	 * <stat><op><value, where:
477 	 *   - <stat> is one of supported numerical stats (verdict is also
478 	 *     considered numerical, failure == 0, success == 1);
479 	 *   - <op> is comparison operator (see `operators` definitions);
480 	 *   - <value> is an integer (or failure/success, or false/true as
481 	 *     special aliases for 0 and 1, respectively).
482 	 * If the form doesn't match what user provided, we assume file/prog
483 	 * glob filter.
484 	 */
485 	for (i = 0; i < ARRAY_SIZE(operators); i++) {
486 		enum stat_variant var;
487 		int id;
488 		long val;
489 		const char *end = str;
490 		const char *op_str;
491 
492 		op_str = operators[i].op_str;
493 		p = strstr(str, op_str);
494 		if (!p)
495 			continue;
496 
497 		if (!parse_stat_id_var(str, p - str, &id, &var)) {
498 			fprintf(stderr, "Unrecognized stat name in '%s'!\n", str);
499 			return -EINVAL;
500 		}
501 		if (id >= FILE_NAME) {
502 			fprintf(stderr, "Non-integer stat is specified in '%s'!\n", str);
503 			return -EINVAL;
504 		}
505 
506 		p += strlen(op_str);
507 
508 		if (strcasecmp(p, "true") == 0 ||
509 		    strcasecmp(p, "t") == 0 ||
510 		    strcasecmp(p, "success") == 0 ||
511 		    strcasecmp(p, "succ") == 0 ||
512 		    strcasecmp(p, "s") == 0 ||
513 		    strcasecmp(p, "match") == 0 ||
514 		    strcasecmp(p, "m") == 0) {
515 			val = 1;
516 		} else if (strcasecmp(p, "false") == 0 ||
517 			   strcasecmp(p, "f") == 0 ||
518 			   strcasecmp(p, "failure") == 0 ||
519 			   strcasecmp(p, "fail") == 0 ||
520 			   strcasecmp(p, "mismatch") == 0 ||
521 			   strcasecmp(p, "mis") == 0) {
522 			val = 0;
523 		} else {
524 			errno = 0;
525 			val = strtol(p, (char **)&end, 10);
526 			if (errno || end == p || *end != '\0' ) {
527 				fprintf(stderr, "Invalid integer value in '%s'!\n", str);
528 				return -EINVAL;
529 			}
530 		}
531 
532 		f->kind = FILTER_STAT;
533 		f->stat_id = id;
534 		f->stat_var = var;
535 		f->op = operators[i].op_kind;
536 		f->value = val;
537 
538 		*cnt += 1;
539 		return 0;
540 	}
541 
542 	/* File/prog filter can be specified either as '<glob>' or
543 	 * '<file-glob>/<prog-glob>'. In the former case <glob> is applied to
544 	 * both file and program names. This seems to be way more useful in
545 	 * practice. If user needs full control, they can use '/<prog-glob>'
546 	 * form to glob just program name, or '<file-glob>/' to glob only file
547 	 * name. But usually common <glob> seems to be the most useful and
548 	 * ergonomic way.
549 	 */
550 	f->kind = FILTER_NAME;
551 	p = strchr(str, '/');
552 	if (!p) {
553 		f->any_glob = strdup(str);
554 		if (!f->any_glob)
555 			return -ENOMEM;
556 	} else {
557 		if (str != p) {
558 			/* non-empty file glob */
559 			f->file_glob = strndup(str, p - str);
560 			if (!f->file_glob)
561 				return -ENOMEM;
562 		}
563 		if (strlen(p + 1) > 0) {
564 			/* non-empty prog glob */
565 			f->prog_glob = strdup(p + 1);
566 			if (!f->prog_glob) {
567 				free(f->file_glob);
568 				f->file_glob = NULL;
569 				return -ENOMEM;
570 			}
571 		}
572 	}
573 
574 	*cnt += 1;
575 	return 0;
576 }
577 
append_filter_file(const char * path)578 static int append_filter_file(const char *path)
579 {
580 	char buf[1024];
581 	FILE *f;
582 	int err = 0;
583 
584 	f = fopen(path, "r");
585 	if (!f) {
586 		err = -errno;
587 		fprintf(stderr, "Failed to open filters in '%s': %d\n", path, err);
588 		return err;
589 	}
590 
591 	while (fscanf(f, " %1023[^\n]\n", buf) == 1) {
592 		/* lines starting with # are comments, skip them */
593 		if (buf[0] == '\0' || buf[0] == '#')
594 			continue;
595 		/* lines starting with ! are negative match filters */
596 		if (buf[0] == '!')
597 			err = append_filter(&env.deny_filters, &env.deny_filter_cnt, buf + 1);
598 		else
599 			err = append_filter(&env.allow_filters, &env.allow_filter_cnt, buf);
600 		if (err)
601 			goto cleanup;
602 	}
603 
604 cleanup:
605 	fclose(f);
606 	return err;
607 }
608 
609 static const struct stat_specs default_output_spec = {
610 	.spec_cnt = 7,
611 	.ids = {
612 		FILE_NAME, PROG_NAME, VERDICT, DURATION,
613 		TOTAL_INSNS, TOTAL_STATES, PEAK_STATES,
614 	},
615 };
616 
617 static const struct stat_specs default_csv_output_spec = {
618 	.spec_cnt = 9,
619 	.ids = {
620 		FILE_NAME, PROG_NAME, VERDICT, DURATION,
621 		TOTAL_INSNS, TOTAL_STATES, PEAK_STATES,
622 		MAX_STATES_PER_INSN, MARK_READ_MAX_LEN,
623 	},
624 };
625 
626 static const struct stat_specs default_sort_spec = {
627 	.spec_cnt = 2,
628 	.ids = {
629 		FILE_NAME, PROG_NAME,
630 	},
631 	.asc = { true, true, },
632 };
633 
634 /* sorting for comparison mode to join two data sets */
635 static const struct stat_specs join_sort_spec = {
636 	.spec_cnt = 2,
637 	.ids = {
638 		FILE_NAME, PROG_NAME,
639 	},
640 	.asc = { true, true, },
641 };
642 
643 static struct stat_def {
644 	const char *header;
645 	const char *names[4];
646 	bool asc_by_default;
647 	bool left_aligned;
648 } stat_defs[] = {
649 	[FILE_NAME] = { "File", {"file_name", "filename", "file"}, true /* asc */, true /* left */ },
650 	[PROG_NAME] = { "Program", {"prog_name", "progname", "prog"}, true /* asc */, true /* left */ },
651 	[VERDICT] = { "Verdict", {"verdict"}, true /* asc: failure, success */, true /* left */ },
652 	[DURATION] = { "Duration (us)", {"duration", "dur"}, },
653 	[TOTAL_INSNS] = { "Insns", {"total_insns", "insns"}, },
654 	[TOTAL_STATES] = { "States", {"total_states", "states"}, },
655 	[PEAK_STATES] = { "Peak states", {"peak_states"}, },
656 	[MAX_STATES_PER_INSN] = { "Max states per insn", {"max_states_per_insn"}, },
657 	[MARK_READ_MAX_LEN] = { "Max mark read length", {"max_mark_read_len", "mark_read"}, },
658 };
659 
parse_stat_id_var(const char * name,size_t len,int * id,enum stat_variant * var)660 static bool parse_stat_id_var(const char *name, size_t len, int *id, enum stat_variant *var)
661 {
662 	static const char *var_sfxs[] = {
663 		[VARIANT_A] = "_a",
664 		[VARIANT_B] = "_b",
665 		[VARIANT_DIFF] = "_diff",
666 		[VARIANT_PCT] = "_pct",
667 	};
668 	int i, j, k;
669 
670 	for (i = 0; i < ARRAY_SIZE(stat_defs); i++) {
671 		struct stat_def *def = &stat_defs[i];
672 		size_t alias_len, sfx_len;
673 		const char *alias;
674 
675 		for (j = 0; j < ARRAY_SIZE(stat_defs[i].names); j++) {
676 			alias = def->names[j];
677 			if (!alias)
678 				continue;
679 
680 			alias_len = strlen(alias);
681 			if (strncmp(name, alias, alias_len) != 0)
682 				continue;
683 
684 			if (alias_len == len) {
685 				/* If no variant suffix is specified, we
686 				 * assume control group (just in case we are
687 				 * in comparison mode. Variant is ignored in
688 				 * non-comparison mode.
689 				 */
690 				*var = VARIANT_B;
691 				*id = i;
692 				return true;
693 			}
694 
695 			for (k = 0; k < ARRAY_SIZE(var_sfxs); k++) {
696 				sfx_len = strlen(var_sfxs[k]);
697 				if (alias_len + sfx_len != len)
698 					continue;
699 
700 				if (strncmp(name + alias_len, var_sfxs[k], sfx_len) == 0) {
701 					*var = (enum stat_variant)k;
702 					*id = i;
703 					return true;
704 				}
705 			}
706 		}
707 	}
708 
709 	return false;
710 }
711 
is_asc_sym(char c)712 static bool is_asc_sym(char c)
713 {
714 	return c == '^';
715 }
716 
is_desc_sym(char c)717 static bool is_desc_sym(char c)
718 {
719 	return c == 'v' || c == 'V' || c == '.' || c == '!' || c == '_';
720 }
721 
parse_stat(const char * stat_name,struct stat_specs * specs)722 static int parse_stat(const char *stat_name, struct stat_specs *specs)
723 {
724 	int id;
725 	bool has_order = false, is_asc = false;
726 	size_t len = strlen(stat_name);
727 	enum stat_variant var;
728 
729 	if (specs->spec_cnt >= ARRAY_SIZE(specs->ids)) {
730 		fprintf(stderr, "Can't specify more than %zd stats\n", ARRAY_SIZE(specs->ids));
731 		return -E2BIG;
732 	}
733 
734 	if (len > 1 && (is_asc_sym(stat_name[len - 1]) || is_desc_sym(stat_name[len - 1]))) {
735 		has_order = true;
736 		is_asc = is_asc_sym(stat_name[len - 1]);
737 		len -= 1;
738 	}
739 
740 	if (!parse_stat_id_var(stat_name, len, &id, &var)) {
741 		fprintf(stderr, "Unrecognized stat name '%s'\n", stat_name);
742 		return -ESRCH;
743 	}
744 
745 	specs->ids[specs->spec_cnt] = id;
746 	specs->variants[specs->spec_cnt] = var;
747 	specs->asc[specs->spec_cnt] = has_order ? is_asc : stat_defs[id].asc_by_default;
748 	specs->spec_cnt++;
749 
750 	return 0;
751 }
752 
parse_stats(const char * stats_str,struct stat_specs * specs)753 static int parse_stats(const char *stats_str, struct stat_specs *specs)
754 {
755 	char *input, *state = NULL, *next;
756 	int err, cnt = 0;
757 
758 	input = strdup(stats_str);
759 	if (!input)
760 		return -ENOMEM;
761 
762 	while ((next = strtok_r(cnt++ ? NULL : input, ",", &state))) {
763 		err = parse_stat(next, specs);
764 		if (err)
765 			return err;
766 	}
767 
768 	return 0;
769 }
770 
free_verif_stats(struct verif_stats * stats,size_t stat_cnt)771 static void free_verif_stats(struct verif_stats *stats, size_t stat_cnt)
772 {
773 	int i;
774 
775 	if (!stats)
776 		return;
777 
778 	for (i = 0; i < stat_cnt; i++) {
779 		free(stats[i].file_name);
780 		free(stats[i].prog_name);
781 	}
782 	free(stats);
783 }
784 
785 static char verif_log_buf[64 * 1024];
786 
787 #define MAX_PARSED_LOG_LINES 100
788 
parse_verif_log(char * const buf,size_t buf_sz,struct verif_stats * s)789 static int parse_verif_log(char * const buf, size_t buf_sz, struct verif_stats *s)
790 {
791 	const char *cur;
792 	int pos, lines;
793 
794 	buf[buf_sz - 1] = '\0';
795 
796 	for (pos = strlen(buf) - 1, lines = 0; pos >= 0 && lines < MAX_PARSED_LOG_LINES; lines++) {
797 		/* find previous endline or otherwise take the start of log buf */
798 		for (cur = &buf[pos]; cur > buf && cur[0] != '\n'; cur--, pos--) {
799 		}
800 		/* next time start from end of previous line (or pos goes to <0) */
801 		pos--;
802 		/* if we found endline, point right after endline symbol;
803 		 * otherwise, stay at the beginning of log buf
804 		 */
805 		if (cur[0] == '\n')
806 			cur++;
807 
808 		if (1 == sscanf(cur, "verification time %ld usec\n", &s->stats[DURATION]))
809 			continue;
810 		if (6 == sscanf(cur, "processed %ld insns (limit %*d) max_states_per_insn %ld total_states %ld peak_states %ld mark_read %ld",
811 				&s->stats[TOTAL_INSNS],
812 				&s->stats[MAX_STATES_PER_INSN],
813 				&s->stats[TOTAL_STATES],
814 				&s->stats[PEAK_STATES],
815 				&s->stats[MARK_READ_MAX_LEN]))
816 			continue;
817 	}
818 
819 	return 0;
820 }
821 
guess_prog_type_by_ctx_name(const char * ctx_name,enum bpf_prog_type * prog_type,enum bpf_attach_type * attach_type)822 static int guess_prog_type_by_ctx_name(const char *ctx_name,
823 				       enum bpf_prog_type *prog_type,
824 				       enum bpf_attach_type *attach_type)
825 {
826 	/* We need to guess program type based on its declared context type.
827 	 * This guess can't be perfect as many different program types might
828 	 * share the same context type.  So we can only hope to reasonably
829 	 * well guess this and get lucky.
830 	 *
831 	 * Just in case, we support both UAPI-side type names and
832 	 * kernel-internal names.
833 	 */
834 	static struct {
835 		const char *uapi_name;
836 		const char *kern_name;
837 		enum bpf_prog_type prog_type;
838 		enum bpf_attach_type attach_type;
839 	} ctx_map[] = {
840 		/* __sk_buff is most ambiguous, we assume TC program */
841 		{ "__sk_buff", "sk_buff", BPF_PROG_TYPE_SCHED_CLS },
842 		{ "bpf_sock", "sock", BPF_PROG_TYPE_CGROUP_SOCK, BPF_CGROUP_INET4_POST_BIND },
843 		{ "bpf_sock_addr", "bpf_sock_addr_kern",  BPF_PROG_TYPE_CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_BIND },
844 		{ "bpf_sock_ops", "bpf_sock_ops_kern", BPF_PROG_TYPE_SOCK_OPS, BPF_CGROUP_SOCK_OPS },
845 		{ "sk_msg_md", "sk_msg", BPF_PROG_TYPE_SK_MSG, BPF_SK_MSG_VERDICT },
846 		{ "bpf_cgroup_dev_ctx", "bpf_cgroup_dev_ctx", BPF_PROG_TYPE_CGROUP_DEVICE, BPF_CGROUP_DEVICE },
847 		{ "bpf_sysctl", "bpf_sysctl_kern", BPF_PROG_TYPE_CGROUP_SYSCTL, BPF_CGROUP_SYSCTL },
848 		{ "bpf_sockopt", "bpf_sockopt_kern", BPF_PROG_TYPE_CGROUP_SOCKOPT, BPF_CGROUP_SETSOCKOPT },
849 		{ "sk_reuseport_md", "sk_reuseport_kern", BPF_PROG_TYPE_SK_REUSEPORT, BPF_SK_REUSEPORT_SELECT_OR_MIGRATE },
850 		{ "bpf_sk_lookup", "bpf_sk_lookup_kern", BPF_PROG_TYPE_SK_LOOKUP, BPF_SK_LOOKUP },
851 		{ "xdp_md", "xdp_buff", BPF_PROG_TYPE_XDP, BPF_XDP },
852 		/* tracing types with no expected attach type */
853 		{ "bpf_user_pt_regs_t", "pt_regs", BPF_PROG_TYPE_KPROBE },
854 		{ "bpf_perf_event_data", "bpf_perf_event_data_kern", BPF_PROG_TYPE_PERF_EVENT },
855 		/* raw_tp programs use u64[] from kernel side, we don't want
856 		 * to match on that, probably; so NULL for kern-side type
857 		 */
858 		{ "bpf_raw_tracepoint_args", NULL, BPF_PROG_TYPE_RAW_TRACEPOINT },
859 	};
860 	int i;
861 
862 	if (!ctx_name)
863 		return -EINVAL;
864 
865 	for (i = 0; i < ARRAY_SIZE(ctx_map); i++) {
866 		if (strcmp(ctx_map[i].uapi_name, ctx_name) == 0 ||
867 		    (ctx_map[i].kern_name && strcmp(ctx_map[i].kern_name, ctx_name) == 0)) {
868 			*prog_type = ctx_map[i].prog_type;
869 			*attach_type = ctx_map[i].attach_type;
870 			return 0;
871 		}
872 	}
873 
874 	return -ESRCH;
875 }
876 
fixup_obj(struct bpf_object * obj,struct bpf_program * prog,const char * filename)877 static void fixup_obj(struct bpf_object *obj, struct bpf_program *prog, const char *filename)
878 {
879 	struct bpf_map *map;
880 
881 	bpf_object__for_each_map(map, obj) {
882 		/* disable pinning */
883 		bpf_map__set_pin_path(map, NULL);
884 
885 		/* fix up map size, if necessary */
886 		switch (bpf_map__type(map)) {
887 		case BPF_MAP_TYPE_SK_STORAGE:
888 		case BPF_MAP_TYPE_TASK_STORAGE:
889 		case BPF_MAP_TYPE_INODE_STORAGE:
890 		case BPF_MAP_TYPE_CGROUP_STORAGE:
891 			break;
892 		default:
893 			if (bpf_map__max_entries(map) == 0)
894 				bpf_map__set_max_entries(map, 1);
895 		}
896 	}
897 
898 	/* SEC(freplace) programs can't be loaded with veristat as is,
899 	 * but we can try guessing their target program's expected type by
900 	 * looking at the type of program's first argument and substituting
901 	 * corresponding program type
902 	 */
903 	if (bpf_program__type(prog) == BPF_PROG_TYPE_EXT) {
904 		const struct btf *btf = bpf_object__btf(obj);
905 		const char *prog_name = bpf_program__name(prog);
906 		enum bpf_prog_type prog_type;
907 		enum bpf_attach_type attach_type;
908 		const struct btf_type *t;
909 		const char *ctx_name;
910 		int id;
911 
912 		if (!btf)
913 			goto skip_freplace_fixup;
914 
915 		id = btf__find_by_name_kind(btf, prog_name, BTF_KIND_FUNC);
916 		t = btf__type_by_id(btf, id);
917 		t = btf__type_by_id(btf, t->type);
918 		if (!btf_is_func_proto(t) || btf_vlen(t) != 1)
919 			goto skip_freplace_fixup;
920 
921 		/* context argument is a pointer to a struct/typedef */
922 		t = btf__type_by_id(btf, btf_params(t)[0].type);
923 		while (t && btf_is_mod(t))
924 			t = btf__type_by_id(btf, t->type);
925 		if (!t || !btf_is_ptr(t))
926 			goto skip_freplace_fixup;
927 		t = btf__type_by_id(btf, t->type);
928 		while (t && btf_is_mod(t))
929 			t = btf__type_by_id(btf, t->type);
930 		if (!t)
931 			goto skip_freplace_fixup;
932 
933 		ctx_name = btf__name_by_offset(btf, t->name_off);
934 
935 		if (guess_prog_type_by_ctx_name(ctx_name, &prog_type, &attach_type) == 0) {
936 			bpf_program__set_type(prog, prog_type);
937 			bpf_program__set_expected_attach_type(prog, attach_type);
938 
939 			if (!env.quiet) {
940 				printf("Using guessed program type '%s' for %s/%s...\n",
941 					libbpf_bpf_prog_type_str(prog_type),
942 					filename, prog_name);
943 			}
944 		} else {
945 			if (!env.quiet) {
946 				printf("Failed to guess program type for freplace program with context type name '%s' for %s/%s. Consider using canonical type names to help veristat...\n",
947 					ctx_name, filename, prog_name);
948 			}
949 		}
950 	}
951 skip_freplace_fixup:
952 	return;
953 }
954 
process_prog(const char * filename,struct bpf_object * obj,struct bpf_program * prog)955 static int process_prog(const char *filename, struct bpf_object *obj, struct bpf_program *prog)
956 {
957 	const char *prog_name = bpf_program__name(prog);
958 	const char *base_filename = basename(filename);
959 	char *buf;
960 	int buf_sz, log_level;
961 	struct verif_stats *stats;
962 	int err = 0;
963 	void *tmp;
964 
965 	if (!should_process_file_prog(base_filename, bpf_program__name(prog))) {
966 		env.progs_skipped++;
967 		return 0;
968 	}
969 
970 	tmp = realloc(env.prog_stats, (env.prog_stat_cnt + 1) * sizeof(*env.prog_stats));
971 	if (!tmp)
972 		return -ENOMEM;
973 	env.prog_stats = tmp;
974 	stats = &env.prog_stats[env.prog_stat_cnt++];
975 	memset(stats, 0, sizeof(*stats));
976 
977 	if (env.verbose) {
978 		buf_sz = env.log_size ? env.log_size : 16 * 1024 * 1024;
979 		buf = malloc(buf_sz);
980 		if (!buf)
981 			return -ENOMEM;
982 		/* ensure we always request stats */
983 		log_level = env.log_level | 4 | (env.log_fixed ? 8 : 0);
984 	} else {
985 		buf = verif_log_buf;
986 		buf_sz = sizeof(verif_log_buf);
987 		/* request only verifier stats */
988 		log_level = 4 | (env.log_fixed ? 8 : 0);
989 	}
990 	verif_log_buf[0] = '\0';
991 
992 	bpf_program__set_log_buf(prog, buf, buf_sz);
993 	bpf_program__set_log_level(prog, log_level);
994 
995 	/* increase chances of successful BPF object loading */
996 	fixup_obj(obj, prog, base_filename);
997 
998 	if (env.force_checkpoints)
999 		bpf_program__set_flags(prog, bpf_program__flags(prog) | BPF_F_TEST_STATE_FREQ);
1000 
1001 	err = bpf_object__load(obj);
1002 	env.progs_processed++;
1003 
1004 	stats->file_name = strdup(base_filename);
1005 	stats->prog_name = strdup(bpf_program__name(prog));
1006 	stats->stats[VERDICT] = err == 0; /* 1 - success, 0 - failure */
1007 	parse_verif_log(buf, buf_sz, stats);
1008 
1009 	if (env.verbose) {
1010 		printf("PROCESSING %s/%s, DURATION US: %ld, VERDICT: %s, VERIFIER LOG:\n%s\n",
1011 		       filename, prog_name, stats->stats[DURATION],
1012 		       err ? "failure" : "success", buf);
1013 	}
1014 
1015 	if (verif_log_buf != buf)
1016 		free(buf);
1017 
1018 	return 0;
1019 };
1020 
process_obj(const char * filename)1021 static int process_obj(const char *filename)
1022 {
1023 	struct bpf_object *obj = NULL, *tobj;
1024 	struct bpf_program *prog, *tprog, *lprog;
1025 	libbpf_print_fn_t old_libbpf_print_fn;
1026 	LIBBPF_OPTS(bpf_object_open_opts, opts);
1027 	int err = 0, prog_cnt = 0;
1028 
1029 	if (!should_process_file_prog(basename(filename), NULL)) {
1030 		if (env.verbose)
1031 			printf("Skipping '%s' due to filters...\n", filename);
1032 		env.files_skipped++;
1033 		return 0;
1034 	}
1035 	if (!is_bpf_obj_file(filename)) {
1036 		if (env.verbose)
1037 			printf("Skipping '%s' as it's not a BPF object file...\n", filename);
1038 		env.files_skipped++;
1039 		return 0;
1040 	}
1041 
1042 	if (!env.quiet && env.out_fmt == RESFMT_TABLE)
1043 		printf("Processing '%s'...\n", basename(filename));
1044 
1045 	old_libbpf_print_fn = libbpf_set_print(libbpf_print_fn);
1046 	obj = bpf_object__open_file(filename, &opts);
1047 	if (!obj) {
1048 		/* if libbpf can't open BPF object file, it could be because
1049 		 * that BPF object file is incomplete and has to be statically
1050 		 * linked into a final BPF object file; instead of bailing
1051 		 * out, report it into stderr, mark it as skipped, and
1052 		 * proceed
1053 		 */
1054 		fprintf(stderr, "Failed to open '%s': %d\n", filename, -errno);
1055 		env.files_skipped++;
1056 		err = 0;
1057 		goto cleanup;
1058 	}
1059 
1060 	env.files_processed++;
1061 
1062 	bpf_object__for_each_program(prog, obj) {
1063 		prog_cnt++;
1064 	}
1065 
1066 	if (prog_cnt == 1) {
1067 		prog = bpf_object__next_program(obj, NULL);
1068 		bpf_program__set_autoload(prog, true);
1069 		process_prog(filename, obj, prog);
1070 		goto cleanup;
1071 	}
1072 
1073 	bpf_object__for_each_program(prog, obj) {
1074 		const char *prog_name = bpf_program__name(prog);
1075 
1076 		tobj = bpf_object__open_file(filename, &opts);
1077 		if (!tobj) {
1078 			err = -errno;
1079 			fprintf(stderr, "Failed to open '%s': %d\n", filename, err);
1080 			goto cleanup;
1081 		}
1082 
1083 		lprog = NULL;
1084 		bpf_object__for_each_program(tprog, tobj) {
1085 			const char *tprog_name = bpf_program__name(tprog);
1086 
1087 			if (strcmp(prog_name, tprog_name) == 0) {
1088 				bpf_program__set_autoload(tprog, true);
1089 				lprog = tprog;
1090 			} else {
1091 				bpf_program__set_autoload(tprog, false);
1092 			}
1093 		}
1094 
1095 		process_prog(filename, tobj, lprog);
1096 		bpf_object__close(tobj);
1097 	}
1098 
1099 cleanup:
1100 	bpf_object__close(obj);
1101 	libbpf_set_print(old_libbpf_print_fn);
1102 	return err;
1103 }
1104 
cmp_stat(const struct verif_stats * s1,const struct verif_stats * s2,enum stat_id id,bool asc)1105 static int cmp_stat(const struct verif_stats *s1, const struct verif_stats *s2,
1106 		    enum stat_id id, bool asc)
1107 {
1108 	int cmp = 0;
1109 
1110 	switch (id) {
1111 	case FILE_NAME:
1112 		cmp = strcmp(s1->file_name, s2->file_name);
1113 		break;
1114 	case PROG_NAME:
1115 		cmp = strcmp(s1->prog_name, s2->prog_name);
1116 		break;
1117 	case VERDICT:
1118 	case DURATION:
1119 	case TOTAL_INSNS:
1120 	case TOTAL_STATES:
1121 	case PEAK_STATES:
1122 	case MAX_STATES_PER_INSN:
1123 	case MARK_READ_MAX_LEN: {
1124 		long v1 = s1->stats[id];
1125 		long v2 = s2->stats[id];
1126 
1127 		if (v1 != v2)
1128 			cmp = v1 < v2 ? -1 : 1;
1129 		break;
1130 	}
1131 	default:
1132 		fprintf(stderr, "Unrecognized stat #%d\n", id);
1133 		exit(1);
1134 	}
1135 
1136 	return asc ? cmp : -cmp;
1137 }
1138 
cmp_prog_stats(const void * v1,const void * v2)1139 static int cmp_prog_stats(const void *v1, const void *v2)
1140 {
1141 	const struct verif_stats *s1 = v1, *s2 = v2;
1142 	int i, cmp;
1143 
1144 	for (i = 0; i < env.sort_spec.spec_cnt; i++) {
1145 		cmp = cmp_stat(s1, s2, env.sort_spec.ids[i], env.sort_spec.asc[i]);
1146 		if (cmp != 0)
1147 			return cmp;
1148 	}
1149 
1150 	/* always disambiguate with file+prog, which are unique */
1151 	cmp = strcmp(s1->file_name, s2->file_name);
1152 	if (cmp != 0)
1153 		return cmp;
1154 	return strcmp(s1->prog_name, s2->prog_name);
1155 }
1156 
fetch_join_stat_value(const struct verif_stats_join * s,enum stat_id id,enum stat_variant var,const char ** str_val,double * num_val)1157 static void fetch_join_stat_value(const struct verif_stats_join *s,
1158 				  enum stat_id id, enum stat_variant var,
1159 				  const char **str_val,
1160 				  double *num_val)
1161 {
1162 	long v1, v2;
1163 
1164 	if (id == FILE_NAME) {
1165 		*str_val = s->file_name;
1166 		return;
1167 	}
1168 	if (id == PROG_NAME) {
1169 		*str_val = s->prog_name;
1170 		return;
1171 	}
1172 
1173 	v1 = s->stats_a ? s->stats_a->stats[id] : 0;
1174 	v2 = s->stats_b ? s->stats_b->stats[id] : 0;
1175 
1176 	switch (var) {
1177 	case VARIANT_A:
1178 		if (!s->stats_a)
1179 			*num_val = -DBL_MAX;
1180 		else
1181 			*num_val = s->stats_a->stats[id];
1182 		return;
1183 	case VARIANT_B:
1184 		if (!s->stats_b)
1185 			*num_val = -DBL_MAX;
1186 		else
1187 			*num_val = s->stats_b->stats[id];
1188 		return;
1189 	case VARIANT_DIFF:
1190 		if (!s->stats_a || !s->stats_b)
1191 			*num_val = -DBL_MAX;
1192 		else if (id == VERDICT)
1193 			*num_val = v1 == v2 ? 1.0 /* MATCH */ : 0.0 /* MISMATCH */;
1194 		else
1195 			*num_val = (double)(v2 - v1);
1196 		return;
1197 	case VARIANT_PCT:
1198 		if (!s->stats_a || !s->stats_b) {
1199 			*num_val = -DBL_MAX;
1200 		} else if (v1 == 0) {
1201 			if (v1 == v2)
1202 				*num_val = 0.0;
1203 			else
1204 				*num_val = v2 < v1 ? -100.0 : 100.0;
1205 		} else {
1206 			 *num_val = (v2 - v1) * 100.0 / v1;
1207 		}
1208 		return;
1209 	}
1210 }
1211 
cmp_join_stat(const struct verif_stats_join * s1,const struct verif_stats_join * s2,enum stat_id id,enum stat_variant var,bool asc)1212 static int cmp_join_stat(const struct verif_stats_join *s1,
1213 			 const struct verif_stats_join *s2,
1214 			 enum stat_id id, enum stat_variant var, bool asc)
1215 {
1216 	const char *str1 = NULL, *str2 = NULL;
1217 	double v1 = 0.0, v2 = 0.0;
1218 	int cmp = 0;
1219 
1220 	fetch_join_stat_value(s1, id, var, &str1, &v1);
1221 	fetch_join_stat_value(s2, id, var, &str2, &v2);
1222 
1223 	if (str1)
1224 		cmp = strcmp(str1, str2);
1225 	else if (v1 != v2)
1226 		cmp = v1 < v2 ? -1 : 1;
1227 
1228 	return asc ? cmp : -cmp;
1229 }
1230 
cmp_join_stats(const void * v1,const void * v2)1231 static int cmp_join_stats(const void *v1, const void *v2)
1232 {
1233 	const struct verif_stats_join *s1 = v1, *s2 = v2;
1234 	int i, cmp;
1235 
1236 	for (i = 0; i < env.sort_spec.spec_cnt; i++) {
1237 		cmp = cmp_join_stat(s1, s2,
1238 				    env.sort_spec.ids[i],
1239 				    env.sort_spec.variants[i],
1240 				    env.sort_spec.asc[i]);
1241 		if (cmp != 0)
1242 			return cmp;
1243 	}
1244 
1245 	/* always disambiguate with file+prog, which are unique */
1246 	cmp = strcmp(s1->file_name, s2->file_name);
1247 	if (cmp != 0)
1248 		return cmp;
1249 	return strcmp(s1->prog_name, s2->prog_name);
1250 }
1251 
1252 #define HEADER_CHAR '-'
1253 #define COLUMN_SEP "  "
1254 
output_header_underlines(void)1255 static void output_header_underlines(void)
1256 {
1257 	int i, j, len;
1258 
1259 	for (i = 0; i < env.output_spec.spec_cnt; i++) {
1260 		len = env.output_spec.lens[i];
1261 
1262 		printf("%s", i == 0 ? "" : COLUMN_SEP);
1263 		for (j = 0; j < len; j++)
1264 			printf("%c", HEADER_CHAR);
1265 	}
1266 	printf("\n");
1267 }
1268 
output_headers(enum resfmt fmt)1269 static void output_headers(enum resfmt fmt)
1270 {
1271 	const char *fmt_str;
1272 	int i, len;
1273 
1274 	for (i = 0; i < env.output_spec.spec_cnt; i++) {
1275 		int id = env.output_spec.ids[i];
1276 		int *max_len = &env.output_spec.lens[i];
1277 
1278 		switch (fmt) {
1279 		case RESFMT_TABLE_CALCLEN:
1280 			len = snprintf(NULL, 0, "%s", stat_defs[id].header);
1281 			if (len > *max_len)
1282 				*max_len = len;
1283 			break;
1284 		case RESFMT_TABLE:
1285 			fmt_str = stat_defs[id].left_aligned ? "%s%-*s" : "%s%*s";
1286 			printf(fmt_str, i == 0 ? "" : COLUMN_SEP,  *max_len, stat_defs[id].header);
1287 			if (i == env.output_spec.spec_cnt - 1)
1288 				printf("\n");
1289 			break;
1290 		case RESFMT_CSV:
1291 			printf("%s%s", i == 0 ? "" : ",", stat_defs[id].names[0]);
1292 			if (i == env.output_spec.spec_cnt - 1)
1293 				printf("\n");
1294 			break;
1295 		}
1296 	}
1297 
1298 	if (fmt == RESFMT_TABLE)
1299 		output_header_underlines();
1300 }
1301 
prepare_value(const struct verif_stats * s,enum stat_id id,const char ** str,long * val)1302 static void prepare_value(const struct verif_stats *s, enum stat_id id,
1303 			  const char **str, long *val)
1304 {
1305 	switch (id) {
1306 	case FILE_NAME:
1307 		*str = s ? s->file_name : "N/A";
1308 		break;
1309 	case PROG_NAME:
1310 		*str = s ? s->prog_name : "N/A";
1311 		break;
1312 	case VERDICT:
1313 		if (!s)
1314 			*str = "N/A";
1315 		else
1316 			*str = s->stats[VERDICT] ? "success" : "failure";
1317 		break;
1318 	case DURATION:
1319 	case TOTAL_INSNS:
1320 	case TOTAL_STATES:
1321 	case PEAK_STATES:
1322 	case MAX_STATES_PER_INSN:
1323 	case MARK_READ_MAX_LEN:
1324 		*val = s ? s->stats[id] : 0;
1325 		break;
1326 	default:
1327 		fprintf(stderr, "Unrecognized stat #%d\n", id);
1328 		exit(1);
1329 	}
1330 }
1331 
output_stats(const struct verif_stats * s,enum resfmt fmt,bool last)1332 static void output_stats(const struct verif_stats *s, enum resfmt fmt, bool last)
1333 {
1334 	int i;
1335 
1336 	for (i = 0; i < env.output_spec.spec_cnt; i++) {
1337 		int id = env.output_spec.ids[i];
1338 		int *max_len = &env.output_spec.lens[i], len;
1339 		const char *str = NULL;
1340 		long val = 0;
1341 
1342 		prepare_value(s, id, &str, &val);
1343 
1344 		switch (fmt) {
1345 		case RESFMT_TABLE_CALCLEN:
1346 			if (str)
1347 				len = snprintf(NULL, 0, "%s", str);
1348 			else
1349 				len = snprintf(NULL, 0, "%ld", val);
1350 			if (len > *max_len)
1351 				*max_len = len;
1352 			break;
1353 		case RESFMT_TABLE:
1354 			if (str)
1355 				printf("%s%-*s", i == 0 ? "" : COLUMN_SEP, *max_len, str);
1356 			else
1357 				printf("%s%*ld", i == 0 ? "" : COLUMN_SEP,  *max_len, val);
1358 			if (i == env.output_spec.spec_cnt - 1)
1359 				printf("\n");
1360 			break;
1361 		case RESFMT_CSV:
1362 			if (str)
1363 				printf("%s%s", i == 0 ? "" : ",", str);
1364 			else
1365 				printf("%s%ld", i == 0 ? "" : ",", val);
1366 			if (i == env.output_spec.spec_cnt - 1)
1367 				printf("\n");
1368 			break;
1369 		}
1370 	}
1371 
1372 	if (last && fmt == RESFMT_TABLE) {
1373 		output_header_underlines();
1374 		printf("Done. Processed %d files, %d programs. Skipped %d files, %d programs.\n",
1375 		       env.files_processed, env.files_skipped, env.progs_processed, env.progs_skipped);
1376 	}
1377 }
1378 
parse_stat_value(const char * str,enum stat_id id,struct verif_stats * st)1379 static int parse_stat_value(const char *str, enum stat_id id, struct verif_stats *st)
1380 {
1381 	switch (id) {
1382 	case FILE_NAME:
1383 		st->file_name = strdup(str);
1384 		if (!st->file_name)
1385 			return -ENOMEM;
1386 		break;
1387 	case PROG_NAME:
1388 		st->prog_name = strdup(str);
1389 		if (!st->prog_name)
1390 			return -ENOMEM;
1391 		break;
1392 	case VERDICT:
1393 		if (strcmp(str, "success") == 0) {
1394 			st->stats[VERDICT] = true;
1395 		} else if (strcmp(str, "failure") == 0) {
1396 			st->stats[VERDICT] = false;
1397 		} else {
1398 			fprintf(stderr, "Unrecognized verification verdict '%s'\n", str);
1399 			return -EINVAL;
1400 		}
1401 		break;
1402 	case DURATION:
1403 	case TOTAL_INSNS:
1404 	case TOTAL_STATES:
1405 	case PEAK_STATES:
1406 	case MAX_STATES_PER_INSN:
1407 	case MARK_READ_MAX_LEN: {
1408 		long val;
1409 		int err, n;
1410 
1411 		if (sscanf(str, "%ld %n", &val, &n) != 1 || n != strlen(str)) {
1412 			err = -errno;
1413 			fprintf(stderr, "Failed to parse '%s' as integer\n", str);
1414 			return err;
1415 		}
1416 
1417 		st->stats[id] = val;
1418 		break;
1419 	}
1420 	default:
1421 		fprintf(stderr, "Unrecognized stat #%d\n", id);
1422 		return -EINVAL;
1423 	}
1424 	return 0;
1425 }
1426 
parse_stats_csv(const char * filename,struct stat_specs * specs,struct verif_stats ** statsp,int * stat_cntp)1427 static int parse_stats_csv(const char *filename, struct stat_specs *specs,
1428 			   struct verif_stats **statsp, int *stat_cntp)
1429 {
1430 	char line[4096];
1431 	FILE *f;
1432 	int err = 0;
1433 	bool header = true;
1434 
1435 	f = fopen(filename, "r");
1436 	if (!f) {
1437 		err = -errno;
1438 		fprintf(stderr, "Failed to open '%s': %d\n", filename, err);
1439 		return err;
1440 	}
1441 
1442 	*stat_cntp = 0;
1443 
1444 	while (fgets(line, sizeof(line), f)) {
1445 		char *input = line, *state = NULL, *next;
1446 		struct verif_stats *st = NULL;
1447 		int col = 0, cnt = 0;
1448 
1449 		if (!header) {
1450 			void *tmp;
1451 
1452 			tmp = realloc(*statsp, (*stat_cntp + 1) * sizeof(**statsp));
1453 			if (!tmp) {
1454 				err = -ENOMEM;
1455 				goto cleanup;
1456 			}
1457 			*statsp = tmp;
1458 
1459 			st = &(*statsp)[*stat_cntp];
1460 			memset(st, 0, sizeof(*st));
1461 
1462 			*stat_cntp += 1;
1463 		}
1464 
1465 		while ((next = strtok_r(cnt++ ? NULL : input, ",\n", &state))) {
1466 			if (header) {
1467 				/* for the first line, set up spec stats */
1468 				err = parse_stat(next, specs);
1469 				if (err)
1470 					goto cleanup;
1471 				continue;
1472 			}
1473 
1474 			/* for all other lines, parse values based on spec */
1475 			if (col >= specs->spec_cnt) {
1476 				fprintf(stderr, "Found extraneous column #%d in row #%d of '%s'\n",
1477 					col, *stat_cntp, filename);
1478 				err = -EINVAL;
1479 				goto cleanup;
1480 			}
1481 			err = parse_stat_value(next, specs->ids[col], st);
1482 			if (err)
1483 				goto cleanup;
1484 			col++;
1485 		}
1486 
1487 		if (header) {
1488 			header = false;
1489 			continue;
1490 		}
1491 
1492 		if (col < specs->spec_cnt) {
1493 			fprintf(stderr, "Not enough columns in row #%d in '%s'\n",
1494 				*stat_cntp, filename);
1495 			err = -EINVAL;
1496 			goto cleanup;
1497 		}
1498 
1499 		if (!st->file_name || !st->prog_name) {
1500 			fprintf(stderr, "Row #%d in '%s' is missing file and/or program name\n",
1501 				*stat_cntp, filename);
1502 			err = -EINVAL;
1503 			goto cleanup;
1504 		}
1505 
1506 		/* in comparison mode we can only check filters after we
1507 		 * parsed entire line; if row should be ignored we pretend we
1508 		 * never parsed it
1509 		 */
1510 		if (!should_process_file_prog(st->file_name, st->prog_name)) {
1511 			free(st->file_name);
1512 			free(st->prog_name);
1513 			*stat_cntp -= 1;
1514 		}
1515 	}
1516 
1517 	if (!feof(f)) {
1518 		err = -errno;
1519 		fprintf(stderr, "Failed I/O for '%s': %d\n", filename, err);
1520 	}
1521 
1522 cleanup:
1523 	fclose(f);
1524 	return err;
1525 }
1526 
1527 /* empty/zero stats for mismatched rows */
1528 static const struct verif_stats fallback_stats = { .file_name = "", .prog_name = "" };
1529 
is_key_stat(enum stat_id id)1530 static bool is_key_stat(enum stat_id id)
1531 {
1532 	return id == FILE_NAME || id == PROG_NAME;
1533 }
1534 
output_comp_header_underlines(void)1535 static void output_comp_header_underlines(void)
1536 {
1537 	int i, j, k;
1538 
1539 	for (i = 0; i < env.output_spec.spec_cnt; i++) {
1540 		int id = env.output_spec.ids[i];
1541 		int max_j = is_key_stat(id) ? 1 : 3;
1542 
1543 		for (j = 0; j < max_j; j++) {
1544 			int len = env.output_spec.lens[3 * i + j];
1545 
1546 			printf("%s", i + j == 0 ? "" : COLUMN_SEP);
1547 
1548 			for (k = 0; k < len; k++)
1549 				printf("%c", HEADER_CHAR);
1550 		}
1551 	}
1552 	printf("\n");
1553 }
1554 
output_comp_headers(enum resfmt fmt)1555 static void output_comp_headers(enum resfmt fmt)
1556 {
1557 	static const char *table_sfxs[3] = {" (A)", " (B)", " (DIFF)"};
1558 	static const char *name_sfxs[3] = {"_base", "_comp", "_diff"};
1559 	int i, j, len;
1560 
1561 	for (i = 0; i < env.output_spec.spec_cnt; i++) {
1562 		int id = env.output_spec.ids[i];
1563 		/* key stats don't have A/B/DIFF columns, they are common for both data sets */
1564 		int max_j = is_key_stat(id) ? 1 : 3;
1565 
1566 		for (j = 0; j < max_j; j++) {
1567 			int *max_len = &env.output_spec.lens[3 * i + j];
1568 			bool last = (i == env.output_spec.spec_cnt - 1) && (j == max_j - 1);
1569 			const char *sfx;
1570 
1571 			switch (fmt) {
1572 			case RESFMT_TABLE_CALCLEN:
1573 				sfx = is_key_stat(id) ? "" : table_sfxs[j];
1574 				len = snprintf(NULL, 0, "%s%s", stat_defs[id].header, sfx);
1575 				if (len > *max_len)
1576 					*max_len = len;
1577 				break;
1578 			case RESFMT_TABLE:
1579 				sfx = is_key_stat(id) ? "" : table_sfxs[j];
1580 				printf("%s%-*s%s", i + j == 0 ? "" : COLUMN_SEP,
1581 				       *max_len - (int)strlen(sfx), stat_defs[id].header, sfx);
1582 				if (last)
1583 					printf("\n");
1584 				break;
1585 			case RESFMT_CSV:
1586 				sfx = is_key_stat(id) ? "" : name_sfxs[j];
1587 				printf("%s%s%s", i + j == 0 ? "" : ",", stat_defs[id].names[0], sfx);
1588 				if (last)
1589 					printf("\n");
1590 				break;
1591 			}
1592 		}
1593 	}
1594 
1595 	if (fmt == RESFMT_TABLE)
1596 		output_comp_header_underlines();
1597 }
1598 
output_comp_stats(const struct verif_stats_join * join_stats,enum resfmt fmt,bool last)1599 static void output_comp_stats(const struct verif_stats_join *join_stats,
1600 			      enum resfmt fmt, bool last)
1601 {
1602 	const struct verif_stats *base = join_stats->stats_a;
1603 	const struct verif_stats *comp = join_stats->stats_b;
1604 	char base_buf[1024] = {}, comp_buf[1024] = {}, diff_buf[1024] = {};
1605 	int i;
1606 
1607 	for (i = 0; i < env.output_spec.spec_cnt; i++) {
1608 		int id = env.output_spec.ids[i], len;
1609 		int *max_len_base = &env.output_spec.lens[3 * i + 0];
1610 		int *max_len_comp = &env.output_spec.lens[3 * i + 1];
1611 		int *max_len_diff = &env.output_spec.lens[3 * i + 2];
1612 		const char *base_str = NULL, *comp_str = NULL;
1613 		long base_val = 0, comp_val = 0, diff_val = 0;
1614 
1615 		prepare_value(base, id, &base_str, &base_val);
1616 		prepare_value(comp, id, &comp_str, &comp_val);
1617 
1618 		/* normalize all the outputs to be in string buffers for simplicity */
1619 		if (is_key_stat(id)) {
1620 			/* key stats (file and program name) are always strings */
1621 			if (base)
1622 				snprintf(base_buf, sizeof(base_buf), "%s", base_str);
1623 			else
1624 				snprintf(base_buf, sizeof(base_buf), "%s", comp_str);
1625 		} else if (base_str) {
1626 			snprintf(base_buf, sizeof(base_buf), "%s", base_str);
1627 			snprintf(comp_buf, sizeof(comp_buf), "%s", comp_str);
1628 			if (!base || !comp)
1629 				snprintf(diff_buf, sizeof(diff_buf), "%s", "N/A");
1630 			else if (strcmp(base_str, comp_str) == 0)
1631 				snprintf(diff_buf, sizeof(diff_buf), "%s", "MATCH");
1632 			else
1633 				snprintf(diff_buf, sizeof(diff_buf), "%s", "MISMATCH");
1634 		} else {
1635 			double p = 0.0;
1636 
1637 			if (base)
1638 				snprintf(base_buf, sizeof(base_buf), "%ld", base_val);
1639 			else
1640 				snprintf(base_buf, sizeof(base_buf), "%s", "N/A");
1641 			if (comp)
1642 				snprintf(comp_buf, sizeof(comp_buf), "%ld", comp_val);
1643 			else
1644 				snprintf(comp_buf, sizeof(comp_buf), "%s", "N/A");
1645 
1646 			diff_val = comp_val - base_val;
1647 			if (!base || !comp) {
1648 				snprintf(diff_buf, sizeof(diff_buf), "%s", "N/A");
1649 			} else {
1650 				if (base_val == 0) {
1651 					if (comp_val == base_val)
1652 						p = 0.0; /* avoid +0 (+100%) case */
1653 					else
1654 						p = comp_val < base_val ? -100.0 : 100.0;
1655 				} else {
1656 					 p = diff_val * 100.0 / base_val;
1657 				}
1658 				snprintf(diff_buf, sizeof(diff_buf), "%+ld (%+.2lf%%)", diff_val, p);
1659 			}
1660 		}
1661 
1662 		switch (fmt) {
1663 		case RESFMT_TABLE_CALCLEN:
1664 			len = strlen(base_buf);
1665 			if (len > *max_len_base)
1666 				*max_len_base = len;
1667 			if (!is_key_stat(id)) {
1668 				len = strlen(comp_buf);
1669 				if (len > *max_len_comp)
1670 					*max_len_comp = len;
1671 				len = strlen(diff_buf);
1672 				if (len > *max_len_diff)
1673 					*max_len_diff = len;
1674 			}
1675 			break;
1676 		case RESFMT_TABLE: {
1677 			/* string outputs are left-aligned, number outputs are right-aligned */
1678 			const char *fmt = base_str ? "%s%-*s" : "%s%*s";
1679 
1680 			printf(fmt, i == 0 ? "" : COLUMN_SEP, *max_len_base, base_buf);
1681 			if (!is_key_stat(id)) {
1682 				printf(fmt, COLUMN_SEP, *max_len_comp, comp_buf);
1683 				printf(fmt, COLUMN_SEP, *max_len_diff, diff_buf);
1684 			}
1685 			if (i == env.output_spec.spec_cnt - 1)
1686 				printf("\n");
1687 			break;
1688 		}
1689 		case RESFMT_CSV:
1690 			printf("%s%s", i == 0 ? "" : ",", base_buf);
1691 			if (!is_key_stat(id)) {
1692 				printf("%s%s", i == 0 ? "" : ",", comp_buf);
1693 				printf("%s%s", i == 0 ? "" : ",", diff_buf);
1694 			}
1695 			if (i == env.output_spec.spec_cnt - 1)
1696 				printf("\n");
1697 			break;
1698 		}
1699 	}
1700 
1701 	if (last && fmt == RESFMT_TABLE)
1702 		output_comp_header_underlines();
1703 }
1704 
cmp_stats_key(const struct verif_stats * base,const struct verif_stats * comp)1705 static int cmp_stats_key(const struct verif_stats *base, const struct verif_stats *comp)
1706 {
1707 	int r;
1708 
1709 	r = strcmp(base->file_name, comp->file_name);
1710 	if (r != 0)
1711 		return r;
1712 	return strcmp(base->prog_name, comp->prog_name);
1713 }
1714 
is_join_stat_filter_matched(struct filter * f,const struct verif_stats_join * stats)1715 static bool is_join_stat_filter_matched(struct filter *f, const struct verif_stats_join *stats)
1716 {
1717 	static const double eps = 1e-9;
1718 	const char *str = NULL;
1719 	double value = 0.0;
1720 
1721 	fetch_join_stat_value(stats, f->stat_id, f->stat_var, &str, &value);
1722 
1723 	switch (f->op) {
1724 	case OP_EQ: return value > f->value - eps && value < f->value + eps;
1725 	case OP_NEQ: return value < f->value - eps || value > f->value + eps;
1726 	case OP_LT: return value < f->value - eps;
1727 	case OP_LE: return value <= f->value + eps;
1728 	case OP_GT: return value > f->value + eps;
1729 	case OP_GE: return value >= f->value - eps;
1730 	}
1731 
1732 	fprintf(stderr, "BUG: unknown filter op %d!\n", f->op);
1733 	return false;
1734 }
1735 
should_output_join_stats(const struct verif_stats_join * stats)1736 static bool should_output_join_stats(const struct verif_stats_join *stats)
1737 {
1738 	struct filter *f;
1739 	int i, allow_cnt = 0;
1740 
1741 	for (i = 0; i < env.deny_filter_cnt; i++) {
1742 		f = &env.deny_filters[i];
1743 		if (f->kind != FILTER_STAT)
1744 			continue;
1745 
1746 		if (is_join_stat_filter_matched(f, stats))
1747 			return false;
1748 	}
1749 
1750 	for (i = 0; i < env.allow_filter_cnt; i++) {
1751 		f = &env.allow_filters[i];
1752 		if (f->kind != FILTER_STAT)
1753 			continue;
1754 		allow_cnt++;
1755 
1756 		if (is_join_stat_filter_matched(f, stats))
1757 			return true;
1758 	}
1759 
1760 	/* if there are no stat allowed filters, pass everything through */
1761 	return allow_cnt == 0;
1762 }
1763 
handle_comparison_mode(void)1764 static int handle_comparison_mode(void)
1765 {
1766 	struct stat_specs base_specs = {}, comp_specs = {};
1767 	struct stat_specs tmp_sort_spec;
1768 	enum resfmt cur_fmt;
1769 	int err, i, j, last_idx;
1770 
1771 	if (env.filename_cnt != 2) {
1772 		fprintf(stderr, "Comparison mode expects exactly two input CSV files!\n\n");
1773 		argp_help(&argp, stderr, ARGP_HELP_USAGE, "veristat");
1774 		return -EINVAL;
1775 	}
1776 
1777 	err = parse_stats_csv(env.filenames[0], &base_specs,
1778 			      &env.baseline_stats, &env.baseline_stat_cnt);
1779 	if (err) {
1780 		fprintf(stderr, "Failed to parse stats from '%s': %d\n", env.filenames[0], err);
1781 		return err;
1782 	}
1783 	err = parse_stats_csv(env.filenames[1], &comp_specs,
1784 			      &env.prog_stats, &env.prog_stat_cnt);
1785 	if (err) {
1786 		fprintf(stderr, "Failed to parse stats from '%s': %d\n", env.filenames[1], err);
1787 		return err;
1788 	}
1789 
1790 	/* To keep it simple we validate that the set and order of stats in
1791 	 * both CSVs are exactly the same. This can be lifted with a bit more
1792 	 * pre-processing later.
1793 	 */
1794 	if (base_specs.spec_cnt != comp_specs.spec_cnt) {
1795 		fprintf(stderr, "Number of stats in '%s' and '%s' differs (%d != %d)!\n",
1796 			env.filenames[0], env.filenames[1],
1797 			base_specs.spec_cnt, comp_specs.spec_cnt);
1798 		return -EINVAL;
1799 	}
1800 	for (i = 0; i < base_specs.spec_cnt; i++) {
1801 		if (base_specs.ids[i] != comp_specs.ids[i]) {
1802 			fprintf(stderr, "Stats composition differs between '%s' and '%s' (%s != %s)!\n",
1803 				env.filenames[0], env.filenames[1],
1804 				stat_defs[base_specs.ids[i]].names[0],
1805 				stat_defs[comp_specs.ids[i]].names[0]);
1806 			return -EINVAL;
1807 		}
1808 	}
1809 
1810 	/* Replace user-specified sorting spec with file+prog sorting rule to
1811 	 * be able to join two datasets correctly. Once we are done, we will
1812 	 * restore the original sort spec.
1813 	 */
1814 	tmp_sort_spec = env.sort_spec;
1815 	env.sort_spec = join_sort_spec;
1816 	qsort(env.prog_stats, env.prog_stat_cnt, sizeof(*env.prog_stats), cmp_prog_stats);
1817 	qsort(env.baseline_stats, env.baseline_stat_cnt, sizeof(*env.baseline_stats), cmp_prog_stats);
1818 	env.sort_spec = tmp_sort_spec;
1819 
1820 	/* Join two datasets together. If baseline and comparison datasets
1821 	 * have different subset of rows (we match by 'object + prog' as
1822 	 * a unique key) then assume empty/missing/zero value for rows that
1823 	 * are missing in the opposite data set.
1824 	 */
1825 	i = j = 0;
1826 	while (i < env.baseline_stat_cnt || j < env.prog_stat_cnt) {
1827 		const struct verif_stats *base, *comp;
1828 		struct verif_stats_join *join;
1829 		void *tmp;
1830 		int r;
1831 
1832 		base = i < env.baseline_stat_cnt ? &env.baseline_stats[i] : &fallback_stats;
1833 		comp = j < env.prog_stat_cnt ? &env.prog_stats[j] : &fallback_stats;
1834 
1835 		if (!base->file_name || !base->prog_name) {
1836 			fprintf(stderr, "Entry #%d in '%s' doesn't have file and/or program name specified!\n",
1837 				i, env.filenames[0]);
1838 			return -EINVAL;
1839 		}
1840 		if (!comp->file_name || !comp->prog_name) {
1841 			fprintf(stderr, "Entry #%d in '%s' doesn't have file and/or program name specified!\n",
1842 				j, env.filenames[1]);
1843 			return -EINVAL;
1844 		}
1845 
1846 		tmp = realloc(env.join_stats, (env.join_stat_cnt + 1) * sizeof(*env.join_stats));
1847 		if (!tmp)
1848 			return -ENOMEM;
1849 		env.join_stats = tmp;
1850 
1851 		join = &env.join_stats[env.join_stat_cnt];
1852 		memset(join, 0, sizeof(*join));
1853 
1854 		r = cmp_stats_key(base, comp);
1855 		if (r == 0) {
1856 			join->file_name = base->file_name;
1857 			join->prog_name = base->prog_name;
1858 			join->stats_a = base;
1859 			join->stats_b = comp;
1860 			i++;
1861 			j++;
1862 		} else if (base != &fallback_stats && (comp == &fallback_stats || r < 0)) {
1863 			join->file_name = base->file_name;
1864 			join->prog_name = base->prog_name;
1865 			join->stats_a = base;
1866 			join->stats_b = NULL;
1867 			i++;
1868 		} else if (comp != &fallback_stats && (base == &fallback_stats || r > 0)) {
1869 			join->file_name = comp->file_name;
1870 			join->prog_name = comp->prog_name;
1871 			join->stats_a = NULL;
1872 			join->stats_b = comp;
1873 			j++;
1874 		} else {
1875 			fprintf(stderr, "%s:%d: should never reach here i=%i, j=%i",
1876 				__FILE__, __LINE__, i, j);
1877 			return -EINVAL;
1878 		}
1879 		env.join_stat_cnt += 1;
1880 	}
1881 
1882 	/* now sort joined results accorsing to sort spec */
1883 	qsort(env.join_stats, env.join_stat_cnt, sizeof(*env.join_stats), cmp_join_stats);
1884 
1885 	/* for human-readable table output we need to do extra pass to
1886 	 * calculate column widths, so we substitute current output format
1887 	 * with RESFMT_TABLE_CALCLEN and later revert it back to RESFMT_TABLE
1888 	 * and do everything again.
1889 	 */
1890 	if (env.out_fmt == RESFMT_TABLE)
1891 		cur_fmt = RESFMT_TABLE_CALCLEN;
1892 	else
1893 		cur_fmt = env.out_fmt;
1894 
1895 one_more_time:
1896 	output_comp_headers(cur_fmt);
1897 
1898 	last_idx = -1;
1899 	for (i = 0; i < env.join_stat_cnt; i++) {
1900 		const struct verif_stats_join *join = &env.join_stats[i];
1901 
1902 		if (!should_output_join_stats(join))
1903 			continue;
1904 
1905 		if (cur_fmt == RESFMT_TABLE_CALCLEN)
1906 			last_idx = i;
1907 
1908 		output_comp_stats(join, cur_fmt, i == last_idx);
1909 	}
1910 
1911 	if (cur_fmt == RESFMT_TABLE_CALCLEN) {
1912 		cur_fmt = RESFMT_TABLE;
1913 		goto one_more_time; /* ... this time with feeling */
1914 	}
1915 
1916 	return 0;
1917 }
1918 
is_stat_filter_matched(struct filter * f,const struct verif_stats * stats)1919 static bool is_stat_filter_matched(struct filter *f, const struct verif_stats *stats)
1920 {
1921 	long value = stats->stats[f->stat_id];
1922 
1923 	switch (f->op) {
1924 	case OP_EQ: return value == f->value;
1925 	case OP_NEQ: return value != f->value;
1926 	case OP_LT: return value < f->value;
1927 	case OP_LE: return value <= f->value;
1928 	case OP_GT: return value > f->value;
1929 	case OP_GE: return value >= f->value;
1930 	}
1931 
1932 	fprintf(stderr, "BUG: unknown filter op %d!\n", f->op);
1933 	return false;
1934 }
1935 
should_output_stats(const struct verif_stats * stats)1936 static bool should_output_stats(const struct verif_stats *stats)
1937 {
1938 	struct filter *f;
1939 	int i, allow_cnt = 0;
1940 
1941 	for (i = 0; i < env.deny_filter_cnt; i++) {
1942 		f = &env.deny_filters[i];
1943 		if (f->kind != FILTER_STAT)
1944 			continue;
1945 
1946 		if (is_stat_filter_matched(f, stats))
1947 			return false;
1948 	}
1949 
1950 	for (i = 0; i < env.allow_filter_cnt; i++) {
1951 		f = &env.allow_filters[i];
1952 		if (f->kind != FILTER_STAT)
1953 			continue;
1954 		allow_cnt++;
1955 
1956 		if (is_stat_filter_matched(f, stats))
1957 			return true;
1958 	}
1959 
1960 	/* if there are no stat allowed filters, pass everything through */
1961 	return allow_cnt == 0;
1962 }
1963 
output_prog_stats(void)1964 static void output_prog_stats(void)
1965 {
1966 	const struct verif_stats *stats;
1967 	int i, last_stat_idx = 0;
1968 
1969 	if (env.out_fmt == RESFMT_TABLE) {
1970 		/* calculate column widths */
1971 		output_headers(RESFMT_TABLE_CALCLEN);
1972 		for (i = 0; i < env.prog_stat_cnt; i++) {
1973 			stats = &env.prog_stats[i];
1974 			if (!should_output_stats(stats))
1975 				continue;
1976 			output_stats(stats, RESFMT_TABLE_CALCLEN, false);
1977 			last_stat_idx = i;
1978 		}
1979 	}
1980 
1981 	/* actually output the table */
1982 	output_headers(env.out_fmt);
1983 	for (i = 0; i < env.prog_stat_cnt; i++) {
1984 		stats = &env.prog_stats[i];
1985 		if (!should_output_stats(stats))
1986 			continue;
1987 		output_stats(stats, env.out_fmt, i == last_stat_idx);
1988 	}
1989 }
1990 
handle_verif_mode(void)1991 static int handle_verif_mode(void)
1992 {
1993 	int i, err;
1994 
1995 	if (env.filename_cnt == 0) {
1996 		fprintf(stderr, "Please provide path to BPF object file!\n\n");
1997 		argp_help(&argp, stderr, ARGP_HELP_USAGE, "veristat");
1998 		return -EINVAL;
1999 	}
2000 
2001 	for (i = 0; i < env.filename_cnt; i++) {
2002 		err = process_obj(env.filenames[i]);
2003 		if (err) {
2004 			fprintf(stderr, "Failed to process '%s': %d\n", env.filenames[i], err);
2005 			return err;
2006 		}
2007 	}
2008 
2009 	qsort(env.prog_stats, env.prog_stat_cnt, sizeof(*env.prog_stats), cmp_prog_stats);
2010 
2011 	output_prog_stats();
2012 
2013 	return 0;
2014 }
2015 
handle_replay_mode(void)2016 static int handle_replay_mode(void)
2017 {
2018 	struct stat_specs specs = {};
2019 	int err;
2020 
2021 	if (env.filename_cnt != 1) {
2022 		fprintf(stderr, "Replay mode expects exactly one input CSV file!\n\n");
2023 		argp_help(&argp, stderr, ARGP_HELP_USAGE, "veristat");
2024 		return -EINVAL;
2025 	}
2026 
2027 	err = parse_stats_csv(env.filenames[0], &specs,
2028 			      &env.prog_stats, &env.prog_stat_cnt);
2029 	if (err) {
2030 		fprintf(stderr, "Failed to parse stats from '%s': %d\n", env.filenames[0], err);
2031 		return err;
2032 	}
2033 
2034 	qsort(env.prog_stats, env.prog_stat_cnt, sizeof(*env.prog_stats), cmp_prog_stats);
2035 
2036 	output_prog_stats();
2037 
2038 	return 0;
2039 }
2040 
main(int argc,char ** argv)2041 int main(int argc, char **argv)
2042 {
2043 	int err = 0, i;
2044 
2045 	if (argp_parse(&argp, argc, argv, 0, NULL, NULL))
2046 		return 1;
2047 
2048 	if (env.show_version) {
2049 		printf("%s\n", argp_program_version);
2050 		return 0;
2051 	}
2052 
2053 	if (env.verbose && env.quiet) {
2054 		fprintf(stderr, "Verbose and quiet modes are incompatible, please specify just one or neither!\n\n");
2055 		argp_help(&argp, stderr, ARGP_HELP_USAGE, "veristat");
2056 		return 1;
2057 	}
2058 	if (env.verbose && env.log_level == 0)
2059 		env.log_level = 1;
2060 
2061 	if (env.output_spec.spec_cnt == 0) {
2062 		if (env.out_fmt == RESFMT_CSV)
2063 			env.output_spec = default_csv_output_spec;
2064 		else
2065 			env.output_spec = default_output_spec;
2066 	}
2067 	if (env.sort_spec.spec_cnt == 0)
2068 		env.sort_spec = default_sort_spec;
2069 
2070 	if (env.comparison_mode && env.replay_mode) {
2071 		fprintf(stderr, "Can't specify replay and comparison mode at the same time!\n\n");
2072 		argp_help(&argp, stderr, ARGP_HELP_USAGE, "veristat");
2073 		return 1;
2074 	}
2075 
2076 	if (env.comparison_mode)
2077 		err = handle_comparison_mode();
2078 	else if (env.replay_mode)
2079 		err = handle_replay_mode();
2080 	else
2081 		err = handle_verif_mode();
2082 
2083 	free_verif_stats(env.prog_stats, env.prog_stat_cnt);
2084 	free_verif_stats(env.baseline_stats, env.baseline_stat_cnt);
2085 	free(env.join_stats);
2086 	for (i = 0; i < env.filename_cnt; i++)
2087 		free(env.filenames[i]);
2088 	free(env.filenames);
2089 	for (i = 0; i < env.allow_filter_cnt; i++) {
2090 		free(env.allow_filters[i].any_glob);
2091 		free(env.allow_filters[i].file_glob);
2092 		free(env.allow_filters[i].prog_glob);
2093 	}
2094 	free(env.allow_filters);
2095 	for (i = 0; i < env.deny_filter_cnt; i++) {
2096 		free(env.deny_filters[i].any_glob);
2097 		free(env.deny_filters[i].file_glob);
2098 		free(env.deny_filters[i].prog_glob);
2099 	}
2100 	free(env.deny_filters);
2101 	return -err;
2102 }
2103