• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 <libgen.h>
6 #include <string.h>
7 #include <stdlib.h>
8 #include <sched.h>
9 #include <pthread.h>
10 #include <dirent.h>
11 #include <signal.h>
12 #include <fcntl.h>
13 #include <unistd.h>
14 #include <sys/time.h>
15 #include <sys/sysinfo.h>
16 #include <sys/stat.h>
17 #include <bpf/libbpf.h>
18 #include <bpf/btf.h>
19 #include <libelf.h>
20 #include <gelf.h>
21 #include <float.h>
22 #include <math.h>
23 
24 #ifndef ARRAY_SIZE
25 #define ARRAY_SIZE(arr) (sizeof(arr) / sizeof((arr)[0]))
26 #endif
27 
28 enum stat_id {
29 	VERDICT,
30 	DURATION,
31 	TOTAL_INSNS,
32 	TOTAL_STATES,
33 	PEAK_STATES,
34 	MAX_STATES_PER_INSN,
35 	MARK_READ_MAX_LEN,
36 
37 	FILE_NAME,
38 	PROG_NAME,
39 
40 	ALL_STATS_CNT,
41 	NUM_STATS_CNT = FILE_NAME - VERDICT,
42 };
43 
44 /* In comparison mode each stat can specify up to four different values:
45  *   - A side value;
46  *   - B side value;
47  *   - absolute diff value;
48  *   - relative (percentage) diff value.
49  *
50  * When specifying stat specs in comparison mode, user can use one of the
51  * following variant suffixes to specify which exact variant should be used for
52  * ordering or filtering:
53  *   - `_a` for A side value;
54  *   - `_b` for B side value;
55  *   - `_diff` for absolute diff value;
56  *   - `_pct` for relative (percentage) diff value.
57  *
58  * If no variant suffix is provided, then `_b` (control data) is assumed.
59  *
60  * As an example, let's say instructions stat has the following output:
61  *
62  * Insns (A)  Insns (B)  Insns   (DIFF)
63  * ---------  ---------  --------------
64  * 21547      20920       -627 (-2.91%)
65  *
66  * Then:
67  *   - 21547 is A side value (insns_a);
68  *   - 20920 is B side value (insns_b);
69  *   - -627 is absolute diff value (insns_diff);
70  *   - -2.91% is relative diff value (insns_pct).
71  *
72  * For verdict there is no verdict_pct variant.
73  * For file and program name, _a and _b variants are equivalent and there are
74  * no _diff or _pct variants.
75  */
76 enum stat_variant {
77 	VARIANT_A,
78 	VARIANT_B,
79 	VARIANT_DIFF,
80 	VARIANT_PCT,
81 };
82 
83 struct verif_stats {
84 	char *file_name;
85 	char *prog_name;
86 
87 	long stats[NUM_STATS_CNT];
88 };
89 
90 /* joined comparison mode stats */
91 struct verif_stats_join {
92 	char *file_name;
93 	char *prog_name;
94 
95 	const struct verif_stats *stats_a;
96 	const struct verif_stats *stats_b;
97 };
98 
99 struct stat_specs {
100 	int spec_cnt;
101 	enum stat_id ids[ALL_STATS_CNT];
102 	enum stat_variant variants[ALL_STATS_CNT];
103 	bool asc[ALL_STATS_CNT];
104 	bool abs[ALL_STATS_CNT];
105 	int lens[ALL_STATS_CNT * 3]; /* 3x for comparison mode */
106 };
107 
108 enum resfmt {
109 	RESFMT_TABLE,
110 	RESFMT_TABLE_CALCLEN, /* fake format to pre-calculate table's column widths */
111 	RESFMT_CSV,
112 };
113 
114 enum filter_kind {
115 	FILTER_NAME,
116 	FILTER_STAT,
117 };
118 
119 enum operator_kind {
120 	OP_EQ,		/* == or = */
121 	OP_NEQ,		/* != or <> */
122 	OP_LT,		/* < */
123 	OP_LE,		/* <= */
124 	OP_GT,		/* > */
125 	OP_GE,		/* >= */
126 };
127 
128 struct filter {
129 	enum filter_kind kind;
130 	/* FILTER_NAME */
131 	char *any_glob;
132 	char *file_glob;
133 	char *prog_glob;
134 	/* FILTER_STAT */
135 	enum operator_kind op;
136 	int stat_id;
137 	enum stat_variant stat_var;
138 	long value;
139 	bool abs;
140 };
141 
142 static struct env {
143 	char **filenames;
144 	int filename_cnt;
145 	bool verbose;
146 	bool debug;
147 	bool quiet;
148 	bool force_checkpoints;
149 	bool force_reg_invariants;
150 	enum resfmt out_fmt;
151 	bool show_version;
152 	bool comparison_mode;
153 	bool replay_mode;
154 	int top_n;
155 
156 	int log_level;
157 	int log_size;
158 	bool log_fixed;
159 
160 	struct verif_stats *prog_stats;
161 	int prog_stat_cnt;
162 
163 	/* baseline_stats is allocated and used only in comparison mode */
164 	struct verif_stats *baseline_stats;
165 	int baseline_stat_cnt;
166 
167 	struct verif_stats_join *join_stats;
168 	int join_stat_cnt;
169 
170 	struct stat_specs output_spec;
171 	struct stat_specs sort_spec;
172 
173 	struct filter *allow_filters;
174 	struct filter *deny_filters;
175 	int allow_filter_cnt;
176 	int deny_filter_cnt;
177 
178 	int files_processed;
179 	int files_skipped;
180 	int progs_processed;
181 	int progs_skipped;
182 } env;
183 
libbpf_print_fn(enum libbpf_print_level level,const char * format,va_list args)184 static int libbpf_print_fn(enum libbpf_print_level level, const char *format, va_list args)
185 {
186 	if (!env.verbose)
187 		return 0;
188 	if (level == LIBBPF_DEBUG  && !env.debug)
189 		return 0;
190 	return vfprintf(stderr, format, args);
191 }
192 
193 #ifndef VERISTAT_VERSION
194 #define VERISTAT_VERSION "<kernel>"
195 #endif
196 
197 const char *argp_program_version = "veristat v" VERISTAT_VERSION;
198 const char *argp_program_bug_address = "<bpf@vger.kernel.org>";
199 const char argp_program_doc[] =
200 "veristat    BPF verifier stats collection and comparison tool.\n"
201 "\n"
202 "USAGE: veristat <obj-file> [<obj-file>...]\n"
203 "   OR: veristat -C <baseline.csv> <comparison.csv>\n"
204 "   OR: veristat -R <results.csv>\n";
205 
206 enum {
207 	OPT_LOG_FIXED = 1000,
208 	OPT_LOG_SIZE = 1001,
209 };
210 
211 static const struct argp_option opts[] = {
212 	{ NULL, 'h', NULL, OPTION_HIDDEN, "Show the full help" },
213 	{ "version", 'V', NULL, 0, "Print version" },
214 	{ "verbose", 'v', NULL, 0, "Verbose mode" },
215 	{ "debug", 'd', NULL, 0, "Debug mode (turns on libbpf debug logging)" },
216 	{ "log-level", 'l', "LEVEL", 0, "Verifier log level (default 0 for normal mode, 1 for verbose mode)" },
217 	{ "log-fixed", OPT_LOG_FIXED, NULL, 0, "Disable verifier log rotation" },
218 	{ "log-size", OPT_LOG_SIZE, "BYTES", 0, "Customize verifier log size (default to 16MB)" },
219 	{ "top-n", 'n', "N", 0, "Emit only up to first N results." },
220 	{ "quiet", 'q', NULL, 0, "Quiet mode" },
221 	{ "emit", 'e', "SPEC", 0, "Specify stats to be emitted" },
222 	{ "sort", 's', "SPEC", 0, "Specify sort order" },
223 	{ "output-format", 'o', "FMT", 0, "Result output format (table, csv), default is table." },
224 	{ "compare", 'C', NULL, 0, "Comparison mode" },
225 	{ "replay", 'R', NULL, 0, "Replay mode" },
226 	{ "filter", 'f', "FILTER", 0, "Filter expressions (or @filename for file with expressions)." },
227 	{ "test-states", 't', NULL, 0,
228 	  "Force frequent BPF verifier state checkpointing (set BPF_F_TEST_STATE_FREQ program flag)" },
229 	{ "test-reg-invariants", 'r', NULL, 0,
230 	  "Force BPF verifier failure on register invariant violation (BPF_F_TEST_REG_INVARIANTS program flag)" },
231 	{},
232 };
233 
234 static int parse_stats(const char *stats_str, struct stat_specs *specs);
235 static int append_filter(struct filter **filters, int *cnt, const char *str);
236 static int append_filter_file(const char *path);
237 
parse_arg(int key,char * arg,struct argp_state * state)238 static error_t parse_arg(int key, char *arg, struct argp_state *state)
239 {
240 	void *tmp;
241 	int err;
242 
243 	switch (key) {
244 	case 'h':
245 		argp_state_help(state, stderr, ARGP_HELP_STD_HELP);
246 		break;
247 	case 'V':
248 		env.show_version = true;
249 		break;
250 	case 'v':
251 		env.verbose = true;
252 		break;
253 	case 'd':
254 		env.debug = true;
255 		env.verbose = true;
256 		break;
257 	case 'q':
258 		env.quiet = true;
259 		break;
260 	case 'e':
261 		err = parse_stats(arg, &env.output_spec);
262 		if (err)
263 			return err;
264 		break;
265 	case 's':
266 		err = parse_stats(arg, &env.sort_spec);
267 		if (err)
268 			return err;
269 		break;
270 	case 'o':
271 		if (strcmp(arg, "table") == 0) {
272 			env.out_fmt = RESFMT_TABLE;
273 		} else if (strcmp(arg, "csv") == 0) {
274 			env.out_fmt = RESFMT_CSV;
275 		} else {
276 			fprintf(stderr, "Unrecognized output format '%s'\n", arg);
277 			return -EINVAL;
278 		}
279 		break;
280 	case 'l':
281 		errno = 0;
282 		env.log_level = strtol(arg, NULL, 10);
283 		if (errno) {
284 			fprintf(stderr, "invalid log level: %s\n", arg);
285 			argp_usage(state);
286 		}
287 		break;
288 	case OPT_LOG_FIXED:
289 		env.log_fixed = true;
290 		break;
291 	case OPT_LOG_SIZE:
292 		errno = 0;
293 		env.log_size = strtol(arg, NULL, 10);
294 		if (errno) {
295 			fprintf(stderr, "invalid log size: %s\n", arg);
296 			argp_usage(state);
297 		}
298 		break;
299 	case 't':
300 		env.force_checkpoints = true;
301 		break;
302 	case 'r':
303 		env.force_reg_invariants = true;
304 		break;
305 	case 'n':
306 		errno = 0;
307 		env.top_n = strtol(arg, NULL, 10);
308 		if (errno) {
309 			fprintf(stderr, "invalid top N specifier: %s\n", arg);
310 			argp_usage(state);
311 		}
312 		break;
313 	case 'C':
314 		env.comparison_mode = true;
315 		break;
316 	case 'R':
317 		env.replay_mode = true;
318 		break;
319 	case 'f':
320 		if (arg[0] == '@')
321 			err = append_filter_file(arg + 1);
322 		else if (arg[0] == '!')
323 			err = append_filter(&env.deny_filters, &env.deny_filter_cnt, arg + 1);
324 		else
325 			err = append_filter(&env.allow_filters, &env.allow_filter_cnt, arg);
326 		if (err) {
327 			fprintf(stderr, "Failed to collect program filter expressions: %d\n", err);
328 			return err;
329 		}
330 		break;
331 	case ARGP_KEY_ARG:
332 		tmp = realloc(env.filenames, (env.filename_cnt + 1) * sizeof(*env.filenames));
333 		if (!tmp)
334 			return -ENOMEM;
335 		env.filenames = tmp;
336 		env.filenames[env.filename_cnt] = strdup(arg);
337 		if (!env.filenames[env.filename_cnt])
338 			return -ENOMEM;
339 		env.filename_cnt++;
340 		break;
341 	default:
342 		return ARGP_ERR_UNKNOWN;
343 	}
344 	return 0;
345 }
346 
347 static const struct argp argp = {
348 	.options = opts,
349 	.parser = parse_arg,
350 	.doc = argp_program_doc,
351 };
352 
353 
354 /* Adapted from perf/util/string.c */
glob_matches(const char * str,const char * pat)355 static bool glob_matches(const char *str, const char *pat)
356 {
357 	while (*str && *pat && *pat != '*') {
358 		if (*str != *pat)
359 			return false;
360 		str++;
361 		pat++;
362 	}
363 	/* Check wild card */
364 	if (*pat == '*') {
365 		while (*pat == '*')
366 			pat++;
367 		if (!*pat) /* Tail wild card matches all */
368 			return true;
369 		while (*str)
370 			if (glob_matches(str++, pat))
371 				return true;
372 	}
373 	return !*str && !*pat;
374 }
375 
is_bpf_obj_file(const char * path)376 static bool is_bpf_obj_file(const char *path) {
377 	Elf64_Ehdr *ehdr;
378 	int fd, err = -EINVAL;
379 	Elf *elf = NULL;
380 
381 	fd = open(path, O_RDONLY | O_CLOEXEC);
382 	if (fd < 0)
383 		return true; /* we'll fail later and propagate error */
384 
385 	/* ensure libelf is initialized */
386 	(void)elf_version(EV_CURRENT);
387 
388 	elf = elf_begin(fd, ELF_C_READ, NULL);
389 	if (!elf)
390 		goto cleanup;
391 
392 	if (elf_kind(elf) != ELF_K_ELF || gelf_getclass(elf) != ELFCLASS64)
393 		goto cleanup;
394 
395 	ehdr = elf64_getehdr(elf);
396 	/* Old LLVM set e_machine to EM_NONE */
397 	if (!ehdr || ehdr->e_type != ET_REL || (ehdr->e_machine && ehdr->e_machine != EM_BPF))
398 		goto cleanup;
399 
400 	err = 0;
401 cleanup:
402 	if (elf)
403 		elf_end(elf);
404 	close(fd);
405 	return err == 0;
406 }
407 
should_process_file_prog(const char * filename,const char * prog_name)408 static bool should_process_file_prog(const char *filename, const char *prog_name)
409 {
410 	struct filter *f;
411 	int i, allow_cnt = 0;
412 
413 	for (i = 0; i < env.deny_filter_cnt; i++) {
414 		f = &env.deny_filters[i];
415 		if (f->kind != FILTER_NAME)
416 			continue;
417 
418 		if (f->any_glob && glob_matches(filename, f->any_glob))
419 			return false;
420 		if (f->any_glob && prog_name && glob_matches(prog_name, f->any_glob))
421 			return false;
422 		if (f->file_glob && glob_matches(filename, f->file_glob))
423 			return false;
424 		if (f->prog_glob && prog_name && glob_matches(prog_name, f->prog_glob))
425 			return false;
426 	}
427 
428 	for (i = 0; i < env.allow_filter_cnt; i++) {
429 		f = &env.allow_filters[i];
430 		if (f->kind != FILTER_NAME)
431 			continue;
432 
433 		allow_cnt++;
434 		if (f->any_glob) {
435 			if (glob_matches(filename, f->any_glob))
436 				return true;
437 			/* If we don't know program name yet, any_glob filter
438 			 * has to assume that current BPF object file might be
439 			 * relevant; we'll check again later on after opening
440 			 * BPF object file, at which point program name will
441 			 * be known finally.
442 			 */
443 			if (!prog_name || glob_matches(prog_name, f->any_glob))
444 				return true;
445 		} else {
446 			if (f->file_glob && !glob_matches(filename, f->file_glob))
447 				continue;
448 			if (f->prog_glob && prog_name && !glob_matches(prog_name, f->prog_glob))
449 				continue;
450 			return true;
451 		}
452 	}
453 
454 	/* if there are no file/prog name allow filters, allow all progs,
455 	 * unless they are denied earlier explicitly
456 	 */
457 	return allow_cnt == 0;
458 }
459 
460 static struct {
461 	enum operator_kind op_kind;
462 	const char *op_str;
463 } operators[] = {
464 	/* Order of these definitions matter to avoid situations like '<'
465 	 * matching part of what is actually a '<>' operator. That is,
466 	 * substrings should go last.
467 	 */
468 	{ OP_EQ, "==" },
469 	{ OP_NEQ, "!=" },
470 	{ OP_NEQ, "<>" },
471 	{ OP_LE, "<=" },
472 	{ OP_LT, "<" },
473 	{ OP_GE, ">=" },
474 	{ OP_GT, ">" },
475 	{ OP_EQ, "=" },
476 };
477 
478 static bool parse_stat_id_var(const char *name, size_t len, int *id,
479 			      enum stat_variant *var, bool *is_abs);
480 
append_filter(struct filter ** filters,int * cnt,const char * str)481 static int append_filter(struct filter **filters, int *cnt, const char *str)
482 {
483 	struct filter *f;
484 	void *tmp;
485 	const char *p;
486 	int i;
487 
488 	tmp = realloc(*filters, (*cnt + 1) * sizeof(**filters));
489 	if (!tmp)
490 		return -ENOMEM;
491 	*filters = tmp;
492 
493 	f = &(*filters)[*cnt];
494 	memset(f, 0, sizeof(*f));
495 
496 	/* First, let's check if it's a stats filter of the following form:
497 	 * <stat><op><value, where:
498 	 *   - <stat> is one of supported numerical stats (verdict is also
499 	 *     considered numerical, failure == 0, success == 1);
500 	 *   - <op> is comparison operator (see `operators` definitions);
501 	 *   - <value> is an integer (or failure/success, or false/true as
502 	 *     special aliases for 0 and 1, respectively).
503 	 * If the form doesn't match what user provided, we assume file/prog
504 	 * glob filter.
505 	 */
506 	for (i = 0; i < ARRAY_SIZE(operators); i++) {
507 		enum stat_variant var;
508 		int id;
509 		long val;
510 		const char *end = str;
511 		const char *op_str;
512 		bool is_abs;
513 
514 		op_str = operators[i].op_str;
515 		p = strstr(str, op_str);
516 		if (!p)
517 			continue;
518 
519 		if (!parse_stat_id_var(str, p - str, &id, &var, &is_abs)) {
520 			fprintf(stderr, "Unrecognized stat name in '%s'!\n", str);
521 			return -EINVAL;
522 		}
523 		if (id >= FILE_NAME) {
524 			fprintf(stderr, "Non-integer stat is specified in '%s'!\n", str);
525 			return -EINVAL;
526 		}
527 
528 		p += strlen(op_str);
529 
530 		if (strcasecmp(p, "true") == 0 ||
531 		    strcasecmp(p, "t") == 0 ||
532 		    strcasecmp(p, "success") == 0 ||
533 		    strcasecmp(p, "succ") == 0 ||
534 		    strcasecmp(p, "s") == 0 ||
535 		    strcasecmp(p, "match") == 0 ||
536 		    strcasecmp(p, "m") == 0) {
537 			val = 1;
538 		} else if (strcasecmp(p, "false") == 0 ||
539 			   strcasecmp(p, "f") == 0 ||
540 			   strcasecmp(p, "failure") == 0 ||
541 			   strcasecmp(p, "fail") == 0 ||
542 			   strcasecmp(p, "mismatch") == 0 ||
543 			   strcasecmp(p, "mis") == 0) {
544 			val = 0;
545 		} else {
546 			errno = 0;
547 			val = strtol(p, (char **)&end, 10);
548 			if (errno || end == p || *end != '\0' ) {
549 				fprintf(stderr, "Invalid integer value in '%s'!\n", str);
550 				return -EINVAL;
551 			}
552 		}
553 
554 		f->kind = FILTER_STAT;
555 		f->stat_id = id;
556 		f->stat_var = var;
557 		f->op = operators[i].op_kind;
558 		f->abs = true;
559 		f->value = val;
560 
561 		*cnt += 1;
562 		return 0;
563 	}
564 
565 	/* File/prog filter can be specified either as '<glob>' or
566 	 * '<file-glob>/<prog-glob>'. In the former case <glob> is applied to
567 	 * both file and program names. This seems to be way more useful in
568 	 * practice. If user needs full control, they can use '/<prog-glob>'
569 	 * form to glob just program name, or '<file-glob>/' to glob only file
570 	 * name. But usually common <glob> seems to be the most useful and
571 	 * ergonomic way.
572 	 */
573 	f->kind = FILTER_NAME;
574 	p = strchr(str, '/');
575 	if (!p) {
576 		f->any_glob = strdup(str);
577 		if (!f->any_glob)
578 			return -ENOMEM;
579 	} else {
580 		if (str != p) {
581 			/* non-empty file glob */
582 			f->file_glob = strndup(str, p - str);
583 			if (!f->file_glob)
584 				return -ENOMEM;
585 		}
586 		if (strlen(p + 1) > 0) {
587 			/* non-empty prog glob */
588 			f->prog_glob = strdup(p + 1);
589 			if (!f->prog_glob) {
590 				free(f->file_glob);
591 				f->file_glob = NULL;
592 				return -ENOMEM;
593 			}
594 		}
595 	}
596 
597 	*cnt += 1;
598 	return 0;
599 }
600 
append_filter_file(const char * path)601 static int append_filter_file(const char *path)
602 {
603 	char buf[1024];
604 	FILE *f;
605 	int err = 0;
606 
607 	f = fopen(path, "r");
608 	if (!f) {
609 		err = -errno;
610 		fprintf(stderr, "Failed to open filters in '%s': %d\n", path, err);
611 		return err;
612 	}
613 
614 	while (fscanf(f, " %1023[^\n]\n", buf) == 1) {
615 		/* lines starting with # are comments, skip them */
616 		if (buf[0] == '\0' || buf[0] == '#')
617 			continue;
618 		/* lines starting with ! are negative match filters */
619 		if (buf[0] == '!')
620 			err = append_filter(&env.deny_filters, &env.deny_filter_cnt, buf + 1);
621 		else
622 			err = append_filter(&env.allow_filters, &env.allow_filter_cnt, buf);
623 		if (err)
624 			goto cleanup;
625 	}
626 
627 cleanup:
628 	fclose(f);
629 	return err;
630 }
631 
632 static const struct stat_specs default_output_spec = {
633 	.spec_cnt = 7,
634 	.ids = {
635 		FILE_NAME, PROG_NAME, VERDICT, DURATION,
636 		TOTAL_INSNS, TOTAL_STATES, PEAK_STATES,
637 	},
638 };
639 
640 static const struct stat_specs default_csv_output_spec = {
641 	.spec_cnt = 9,
642 	.ids = {
643 		FILE_NAME, PROG_NAME, VERDICT, DURATION,
644 		TOTAL_INSNS, TOTAL_STATES, PEAK_STATES,
645 		MAX_STATES_PER_INSN, MARK_READ_MAX_LEN,
646 	},
647 };
648 
649 static const struct stat_specs default_sort_spec = {
650 	.spec_cnt = 2,
651 	.ids = {
652 		FILE_NAME, PROG_NAME,
653 	},
654 	.asc = { true, true, },
655 };
656 
657 /* sorting for comparison mode to join two data sets */
658 static const struct stat_specs join_sort_spec = {
659 	.spec_cnt = 2,
660 	.ids = {
661 		FILE_NAME, PROG_NAME,
662 	},
663 	.asc = { true, true, },
664 };
665 
666 static struct stat_def {
667 	const char *header;
668 	const char *names[4];
669 	bool asc_by_default;
670 	bool left_aligned;
671 } stat_defs[] = {
672 	[FILE_NAME] = { "File", {"file_name", "filename", "file"}, true /* asc */, true /* left */ },
673 	[PROG_NAME] = { "Program", {"prog_name", "progname", "prog"}, true /* asc */, true /* left */ },
674 	[VERDICT] = { "Verdict", {"verdict"}, true /* asc: failure, success */, true /* left */ },
675 	[DURATION] = { "Duration (us)", {"duration", "dur"}, },
676 	[TOTAL_INSNS] = { "Insns", {"total_insns", "insns"}, },
677 	[TOTAL_STATES] = { "States", {"total_states", "states"}, },
678 	[PEAK_STATES] = { "Peak states", {"peak_states"}, },
679 	[MAX_STATES_PER_INSN] = { "Max states per insn", {"max_states_per_insn"}, },
680 	[MARK_READ_MAX_LEN] = { "Max mark read length", {"max_mark_read_len", "mark_read"}, },
681 };
682 
parse_stat_id_var(const char * name,size_t len,int * id,enum stat_variant * var,bool * is_abs)683 static bool parse_stat_id_var(const char *name, size_t len, int *id,
684 			      enum stat_variant *var, bool *is_abs)
685 {
686 	static const char *var_sfxs[] = {
687 		[VARIANT_A] = "_a",
688 		[VARIANT_B] = "_b",
689 		[VARIANT_DIFF] = "_diff",
690 		[VARIANT_PCT] = "_pct",
691 	};
692 	int i, j, k;
693 
694 	/* |<stat>| means we take absolute value of given stat */
695 	*is_abs = false;
696 	if (len > 2 && name[0] == '|' && name[len - 1] == '|') {
697 		*is_abs = true;
698 		name += 1;
699 		len -= 2;
700 	}
701 
702 	for (i = 0; i < ARRAY_SIZE(stat_defs); i++) {
703 		struct stat_def *def = &stat_defs[i];
704 		size_t alias_len, sfx_len;
705 		const char *alias;
706 
707 		for (j = 0; j < ARRAY_SIZE(stat_defs[i].names); j++) {
708 			alias = def->names[j];
709 			if (!alias)
710 				continue;
711 
712 			alias_len = strlen(alias);
713 			if (strncmp(name, alias, alias_len) != 0)
714 				continue;
715 
716 			if (alias_len == len) {
717 				/* If no variant suffix is specified, we
718 				 * assume control group (just in case we are
719 				 * in comparison mode. Variant is ignored in
720 				 * non-comparison mode.
721 				 */
722 				*var = VARIANT_B;
723 				*id = i;
724 				return true;
725 			}
726 
727 			for (k = 0; k < ARRAY_SIZE(var_sfxs); k++) {
728 				sfx_len = strlen(var_sfxs[k]);
729 				if (alias_len + sfx_len != len)
730 					continue;
731 
732 				if (strncmp(name + alias_len, var_sfxs[k], sfx_len) == 0) {
733 					*var = (enum stat_variant)k;
734 					*id = i;
735 					return true;
736 				}
737 			}
738 		}
739 	}
740 
741 	return false;
742 }
743 
is_asc_sym(char c)744 static bool is_asc_sym(char c)
745 {
746 	return c == '^';
747 }
748 
is_desc_sym(char c)749 static bool is_desc_sym(char c)
750 {
751 	return c == 'v' || c == 'V' || c == '.' || c == '!' || c == '_';
752 }
753 
parse_stat(const char * stat_name,struct stat_specs * specs)754 static int parse_stat(const char *stat_name, struct stat_specs *specs)
755 {
756 	int id;
757 	bool has_order = false, is_asc = false, is_abs = false;
758 	size_t len = strlen(stat_name);
759 	enum stat_variant var;
760 
761 	if (specs->spec_cnt >= ARRAY_SIZE(specs->ids)) {
762 		fprintf(stderr, "Can't specify more than %zd stats\n", ARRAY_SIZE(specs->ids));
763 		return -E2BIG;
764 	}
765 
766 	if (len > 1 && (is_asc_sym(stat_name[len - 1]) || is_desc_sym(stat_name[len - 1]))) {
767 		has_order = true;
768 		is_asc = is_asc_sym(stat_name[len - 1]);
769 		len -= 1;
770 	}
771 
772 	if (!parse_stat_id_var(stat_name, len, &id, &var, &is_abs)) {
773 		fprintf(stderr, "Unrecognized stat name '%s'\n", stat_name);
774 		return -ESRCH;
775 	}
776 
777 	specs->ids[specs->spec_cnt] = id;
778 	specs->variants[specs->spec_cnt] = var;
779 	specs->asc[specs->spec_cnt] = has_order ? is_asc : stat_defs[id].asc_by_default;
780 	specs->abs[specs->spec_cnt] = is_abs;
781 	specs->spec_cnt++;
782 
783 	return 0;
784 }
785 
parse_stats(const char * stats_str,struct stat_specs * specs)786 static int parse_stats(const char *stats_str, struct stat_specs *specs)
787 {
788 	char *input, *state = NULL, *next;
789 	int err, cnt = 0;
790 
791 	input = strdup(stats_str);
792 	if (!input)
793 		return -ENOMEM;
794 
795 	while ((next = strtok_r(cnt++ ? NULL : input, ",", &state))) {
796 		err = parse_stat(next, specs);
797 		if (err) {
798 			free(input);
799 			return err;
800 		}
801 	}
802 
803 	free(input);
804 	return 0;
805 }
806 
free_verif_stats(struct verif_stats * stats,size_t stat_cnt)807 static void free_verif_stats(struct verif_stats *stats, size_t stat_cnt)
808 {
809 	int i;
810 
811 	if (!stats)
812 		return;
813 
814 	for (i = 0; i < stat_cnt; i++) {
815 		free(stats[i].file_name);
816 		free(stats[i].prog_name);
817 	}
818 	free(stats);
819 }
820 
821 static char verif_log_buf[64 * 1024];
822 
823 #define MAX_PARSED_LOG_LINES 100
824 
parse_verif_log(char * const buf,size_t buf_sz,struct verif_stats * s)825 static int parse_verif_log(char * const buf, size_t buf_sz, struct verif_stats *s)
826 {
827 	const char *cur;
828 	int pos, lines;
829 
830 	buf[buf_sz - 1] = '\0';
831 
832 	for (pos = strlen(buf) - 1, lines = 0; pos >= 0 && lines < MAX_PARSED_LOG_LINES; lines++) {
833 		/* find previous endline or otherwise take the start of log buf */
834 		for (cur = &buf[pos]; cur > buf && cur[0] != '\n'; cur--, pos--) {
835 		}
836 		/* next time start from end of previous line (or pos goes to <0) */
837 		pos--;
838 		/* if we found endline, point right after endline symbol;
839 		 * otherwise, stay at the beginning of log buf
840 		 */
841 		if (cur[0] == '\n')
842 			cur++;
843 
844 		if (1 == sscanf(cur, "verification time %ld usec\n", &s->stats[DURATION]))
845 			continue;
846 		if (6 == sscanf(cur, "processed %ld insns (limit %*d) max_states_per_insn %ld total_states %ld peak_states %ld mark_read %ld",
847 				&s->stats[TOTAL_INSNS],
848 				&s->stats[MAX_STATES_PER_INSN],
849 				&s->stats[TOTAL_STATES],
850 				&s->stats[PEAK_STATES],
851 				&s->stats[MARK_READ_MAX_LEN]))
852 			continue;
853 	}
854 
855 	return 0;
856 }
857 
guess_prog_type_by_ctx_name(const char * ctx_name,enum bpf_prog_type * prog_type,enum bpf_attach_type * attach_type)858 static int guess_prog_type_by_ctx_name(const char *ctx_name,
859 				       enum bpf_prog_type *prog_type,
860 				       enum bpf_attach_type *attach_type)
861 {
862 	/* We need to guess program type based on its declared context type.
863 	 * This guess can't be perfect as many different program types might
864 	 * share the same context type.  So we can only hope to reasonably
865 	 * well guess this and get lucky.
866 	 *
867 	 * Just in case, we support both UAPI-side type names and
868 	 * kernel-internal names.
869 	 */
870 	static struct {
871 		const char *uapi_name;
872 		const char *kern_name;
873 		enum bpf_prog_type prog_type;
874 		enum bpf_attach_type attach_type;
875 	} ctx_map[] = {
876 		/* __sk_buff is most ambiguous, we assume TC program */
877 		{ "__sk_buff", "sk_buff", BPF_PROG_TYPE_SCHED_CLS },
878 		{ "bpf_sock", "sock", BPF_PROG_TYPE_CGROUP_SOCK, BPF_CGROUP_INET4_POST_BIND },
879 		{ "bpf_sock_addr", "bpf_sock_addr_kern",  BPF_PROG_TYPE_CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_BIND },
880 		{ "bpf_sock_ops", "bpf_sock_ops_kern", BPF_PROG_TYPE_SOCK_OPS, BPF_CGROUP_SOCK_OPS },
881 		{ "sk_msg_md", "sk_msg", BPF_PROG_TYPE_SK_MSG, BPF_SK_MSG_VERDICT },
882 		{ "bpf_cgroup_dev_ctx", "bpf_cgroup_dev_ctx", BPF_PROG_TYPE_CGROUP_DEVICE, BPF_CGROUP_DEVICE },
883 		{ "bpf_sysctl", "bpf_sysctl_kern", BPF_PROG_TYPE_CGROUP_SYSCTL, BPF_CGROUP_SYSCTL },
884 		{ "bpf_sockopt", "bpf_sockopt_kern", BPF_PROG_TYPE_CGROUP_SOCKOPT, BPF_CGROUP_SETSOCKOPT },
885 		{ "sk_reuseport_md", "sk_reuseport_kern", BPF_PROG_TYPE_SK_REUSEPORT, BPF_SK_REUSEPORT_SELECT_OR_MIGRATE },
886 		{ "bpf_sk_lookup", "bpf_sk_lookup_kern", BPF_PROG_TYPE_SK_LOOKUP, BPF_SK_LOOKUP },
887 		{ "xdp_md", "xdp_buff", BPF_PROG_TYPE_XDP, BPF_XDP },
888 		/* tracing types with no expected attach type */
889 		{ "bpf_user_pt_regs_t", "pt_regs", BPF_PROG_TYPE_KPROBE },
890 		{ "bpf_perf_event_data", "bpf_perf_event_data_kern", BPF_PROG_TYPE_PERF_EVENT },
891 		/* raw_tp programs use u64[] from kernel side, we don't want
892 		 * to match on that, probably; so NULL for kern-side type
893 		 */
894 		{ "bpf_raw_tracepoint_args", NULL, BPF_PROG_TYPE_RAW_TRACEPOINT },
895 	};
896 	int i;
897 
898 	if (!ctx_name)
899 		return -EINVAL;
900 
901 	for (i = 0; i < ARRAY_SIZE(ctx_map); i++) {
902 		if (strcmp(ctx_map[i].uapi_name, ctx_name) == 0 ||
903 		    (ctx_map[i].kern_name && strcmp(ctx_map[i].kern_name, ctx_name) == 0)) {
904 			*prog_type = ctx_map[i].prog_type;
905 			*attach_type = ctx_map[i].attach_type;
906 			return 0;
907 		}
908 	}
909 
910 	return -ESRCH;
911 }
912 
fixup_obj(struct bpf_object * obj,struct bpf_program * prog,const char * filename)913 static void fixup_obj(struct bpf_object *obj, struct bpf_program *prog, const char *filename)
914 {
915 	struct bpf_map *map;
916 
917 	bpf_object__for_each_map(map, obj) {
918 		/* disable pinning */
919 		bpf_map__set_pin_path(map, NULL);
920 
921 		/* fix up map size, if necessary */
922 		switch (bpf_map__type(map)) {
923 		case BPF_MAP_TYPE_SK_STORAGE:
924 		case BPF_MAP_TYPE_TASK_STORAGE:
925 		case BPF_MAP_TYPE_INODE_STORAGE:
926 		case BPF_MAP_TYPE_CGROUP_STORAGE:
927 			break;
928 		default:
929 			if (bpf_map__max_entries(map) == 0)
930 				bpf_map__set_max_entries(map, 1);
931 		}
932 	}
933 
934 	/* SEC(freplace) programs can't be loaded with veristat as is,
935 	 * but we can try guessing their target program's expected type by
936 	 * looking at the type of program's first argument and substituting
937 	 * corresponding program type
938 	 */
939 	if (bpf_program__type(prog) == BPF_PROG_TYPE_EXT) {
940 		const struct btf *btf = bpf_object__btf(obj);
941 		const char *prog_name = bpf_program__name(prog);
942 		enum bpf_prog_type prog_type;
943 		enum bpf_attach_type attach_type;
944 		const struct btf_type *t;
945 		const char *ctx_name;
946 		int id;
947 
948 		if (!btf)
949 			goto skip_freplace_fixup;
950 
951 		id = btf__find_by_name_kind(btf, prog_name, BTF_KIND_FUNC);
952 		t = btf__type_by_id(btf, id);
953 		t = btf__type_by_id(btf, t->type);
954 		if (!btf_is_func_proto(t) || btf_vlen(t) != 1)
955 			goto skip_freplace_fixup;
956 
957 		/* context argument is a pointer to a struct/typedef */
958 		t = btf__type_by_id(btf, btf_params(t)[0].type);
959 		while (t && btf_is_mod(t))
960 			t = btf__type_by_id(btf, t->type);
961 		if (!t || !btf_is_ptr(t))
962 			goto skip_freplace_fixup;
963 		t = btf__type_by_id(btf, t->type);
964 		while (t && btf_is_mod(t))
965 			t = btf__type_by_id(btf, t->type);
966 		if (!t)
967 			goto skip_freplace_fixup;
968 
969 		ctx_name = btf__name_by_offset(btf, t->name_off);
970 
971 		if (guess_prog_type_by_ctx_name(ctx_name, &prog_type, &attach_type) == 0) {
972 			bpf_program__set_type(prog, prog_type);
973 			bpf_program__set_expected_attach_type(prog, attach_type);
974 
975 			if (!env.quiet) {
976 				printf("Using guessed program type '%s' for %s/%s...\n",
977 					libbpf_bpf_prog_type_str(prog_type),
978 					filename, prog_name);
979 			}
980 		} else {
981 			if (!env.quiet) {
982 				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",
983 					ctx_name, filename, prog_name);
984 			}
985 		}
986 	}
987 skip_freplace_fixup:
988 	return;
989 }
990 
process_prog(const char * filename,struct bpf_object * obj,struct bpf_program * prog)991 static int process_prog(const char *filename, struct bpf_object *obj, struct bpf_program *prog)
992 {
993 	const char *base_filename = basename(strdupa(filename));
994 	const char *prog_name = bpf_program__name(prog);
995 	char *buf;
996 	int buf_sz, log_level;
997 	struct verif_stats *stats;
998 	int err = 0;
999 	void *tmp;
1000 
1001 	if (!should_process_file_prog(base_filename, bpf_program__name(prog))) {
1002 		env.progs_skipped++;
1003 		return 0;
1004 	}
1005 
1006 	tmp = realloc(env.prog_stats, (env.prog_stat_cnt + 1) * sizeof(*env.prog_stats));
1007 	if (!tmp)
1008 		return -ENOMEM;
1009 	env.prog_stats = tmp;
1010 	stats = &env.prog_stats[env.prog_stat_cnt++];
1011 	memset(stats, 0, sizeof(*stats));
1012 
1013 	if (env.verbose) {
1014 		buf_sz = env.log_size ? env.log_size : 16 * 1024 * 1024;
1015 		buf = malloc(buf_sz);
1016 		if (!buf)
1017 			return -ENOMEM;
1018 		/* ensure we always request stats */
1019 		log_level = env.log_level | 4 | (env.log_fixed ? 8 : 0);
1020 	} else {
1021 		buf = verif_log_buf;
1022 		buf_sz = sizeof(verif_log_buf);
1023 		/* request only verifier stats */
1024 		log_level = 4 | (env.log_fixed ? 8 : 0);
1025 	}
1026 	verif_log_buf[0] = '\0';
1027 
1028 	bpf_program__set_log_buf(prog, buf, buf_sz);
1029 	bpf_program__set_log_level(prog, log_level);
1030 
1031 	/* increase chances of successful BPF object loading */
1032 	fixup_obj(obj, prog, base_filename);
1033 
1034 	if (env.force_checkpoints)
1035 		bpf_program__set_flags(prog, bpf_program__flags(prog) | BPF_F_TEST_STATE_FREQ);
1036 	if (env.force_reg_invariants)
1037 		bpf_program__set_flags(prog, bpf_program__flags(prog) | BPF_F_TEST_REG_INVARIANTS);
1038 
1039 	err = bpf_object__load(obj);
1040 	env.progs_processed++;
1041 
1042 	stats->file_name = strdup(base_filename);
1043 	stats->prog_name = strdup(bpf_program__name(prog));
1044 	stats->stats[VERDICT] = err == 0; /* 1 - success, 0 - failure */
1045 	parse_verif_log(buf, buf_sz, stats);
1046 
1047 	if (env.verbose) {
1048 		printf("PROCESSING %s/%s, DURATION US: %ld, VERDICT: %s, VERIFIER LOG:\n%s\n",
1049 		       filename, prog_name, stats->stats[DURATION],
1050 		       err ? "failure" : "success", buf);
1051 	}
1052 
1053 	if (verif_log_buf != buf)
1054 		free(buf);
1055 
1056 	return 0;
1057 };
1058 
process_obj(const char * filename)1059 static int process_obj(const char *filename)
1060 {
1061 	const char *base_filename = basename(strdupa(filename));
1062 	struct bpf_object *obj = NULL, *tobj;
1063 	struct bpf_program *prog, *tprog, *lprog;
1064 	libbpf_print_fn_t old_libbpf_print_fn;
1065 	LIBBPF_OPTS(bpf_object_open_opts, opts);
1066 	int err = 0, prog_cnt = 0;
1067 
1068 	if (!should_process_file_prog(base_filename, NULL)) {
1069 		if (env.verbose)
1070 			printf("Skipping '%s' due to filters...\n", filename);
1071 		env.files_skipped++;
1072 		return 0;
1073 	}
1074 	if (!is_bpf_obj_file(filename)) {
1075 		if (env.verbose)
1076 			printf("Skipping '%s' as it's not a BPF object file...\n", filename);
1077 		env.files_skipped++;
1078 		return 0;
1079 	}
1080 
1081 	if (!env.quiet && env.out_fmt == RESFMT_TABLE)
1082 		printf("Processing '%s'...\n", base_filename);
1083 
1084 	old_libbpf_print_fn = libbpf_set_print(libbpf_print_fn);
1085 	obj = bpf_object__open_file(filename, &opts);
1086 	if (!obj) {
1087 		/* if libbpf can't open BPF object file, it could be because
1088 		 * that BPF object file is incomplete and has to be statically
1089 		 * linked into a final BPF object file; instead of bailing
1090 		 * out, report it into stderr, mark it as skipped, and
1091 		 * proceed
1092 		 */
1093 		fprintf(stderr, "Failed to open '%s': %d\n", filename, -errno);
1094 		env.files_skipped++;
1095 		err = 0;
1096 		goto cleanup;
1097 	}
1098 
1099 	env.files_processed++;
1100 
1101 	bpf_object__for_each_program(prog, obj) {
1102 		prog_cnt++;
1103 	}
1104 
1105 	if (prog_cnt == 1) {
1106 		prog = bpf_object__next_program(obj, NULL);
1107 		bpf_program__set_autoload(prog, true);
1108 		process_prog(filename, obj, prog);
1109 		goto cleanup;
1110 	}
1111 
1112 	bpf_object__for_each_program(prog, obj) {
1113 		const char *prog_name = bpf_program__name(prog);
1114 
1115 		tobj = bpf_object__open_file(filename, &opts);
1116 		if (!tobj) {
1117 			err = -errno;
1118 			fprintf(stderr, "Failed to open '%s': %d\n", filename, err);
1119 			goto cleanup;
1120 		}
1121 
1122 		lprog = NULL;
1123 		bpf_object__for_each_program(tprog, tobj) {
1124 			const char *tprog_name = bpf_program__name(tprog);
1125 
1126 			if (strcmp(prog_name, tprog_name) == 0) {
1127 				bpf_program__set_autoload(tprog, true);
1128 				lprog = tprog;
1129 			} else {
1130 				bpf_program__set_autoload(tprog, false);
1131 			}
1132 		}
1133 
1134 		process_prog(filename, tobj, lprog);
1135 		bpf_object__close(tobj);
1136 	}
1137 
1138 cleanup:
1139 	bpf_object__close(obj);
1140 	libbpf_set_print(old_libbpf_print_fn);
1141 	return err;
1142 }
1143 
cmp_stat(const struct verif_stats * s1,const struct verif_stats * s2,enum stat_id id,bool asc,bool abs)1144 static int cmp_stat(const struct verif_stats *s1, const struct verif_stats *s2,
1145 		    enum stat_id id, bool asc, bool abs)
1146 {
1147 	int cmp = 0;
1148 
1149 	switch (id) {
1150 	case FILE_NAME:
1151 		cmp = strcmp(s1->file_name, s2->file_name);
1152 		break;
1153 	case PROG_NAME:
1154 		cmp = strcmp(s1->prog_name, s2->prog_name);
1155 		break;
1156 	case VERDICT:
1157 	case DURATION:
1158 	case TOTAL_INSNS:
1159 	case TOTAL_STATES:
1160 	case PEAK_STATES:
1161 	case MAX_STATES_PER_INSN:
1162 	case MARK_READ_MAX_LEN: {
1163 		long v1 = s1->stats[id];
1164 		long v2 = s2->stats[id];
1165 
1166 		if (abs) {
1167 			v1 = v1 < 0 ? -v1 : v1;
1168 			v2 = v2 < 0 ? -v2 : v2;
1169 		}
1170 
1171 		if (v1 != v2)
1172 			cmp = v1 < v2 ? -1 : 1;
1173 		break;
1174 	}
1175 	default:
1176 		fprintf(stderr, "Unrecognized stat #%d\n", id);
1177 		exit(1);
1178 	}
1179 
1180 	return asc ? cmp : -cmp;
1181 }
1182 
cmp_prog_stats(const void * v1,const void * v2)1183 static int cmp_prog_stats(const void *v1, const void *v2)
1184 {
1185 	const struct verif_stats *s1 = v1, *s2 = v2;
1186 	int i, cmp;
1187 
1188 	for (i = 0; i < env.sort_spec.spec_cnt; i++) {
1189 		cmp = cmp_stat(s1, s2, env.sort_spec.ids[i],
1190 			       env.sort_spec.asc[i], env.sort_spec.abs[i]);
1191 		if (cmp != 0)
1192 			return cmp;
1193 	}
1194 
1195 	/* always disambiguate with file+prog, which are unique */
1196 	cmp = strcmp(s1->file_name, s2->file_name);
1197 	if (cmp != 0)
1198 		return cmp;
1199 	return strcmp(s1->prog_name, s2->prog_name);
1200 }
1201 
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)1202 static void fetch_join_stat_value(const struct verif_stats_join *s,
1203 				  enum stat_id id, enum stat_variant var,
1204 				  const char **str_val,
1205 				  double *num_val)
1206 {
1207 	long v1, v2;
1208 
1209 	if (id == FILE_NAME) {
1210 		*str_val = s->file_name;
1211 		return;
1212 	}
1213 	if (id == PROG_NAME) {
1214 		*str_val = s->prog_name;
1215 		return;
1216 	}
1217 
1218 	v1 = s->stats_a ? s->stats_a->stats[id] : 0;
1219 	v2 = s->stats_b ? s->stats_b->stats[id] : 0;
1220 
1221 	switch (var) {
1222 	case VARIANT_A:
1223 		if (!s->stats_a)
1224 			*num_val = -DBL_MAX;
1225 		else
1226 			*num_val = s->stats_a->stats[id];
1227 		return;
1228 	case VARIANT_B:
1229 		if (!s->stats_b)
1230 			*num_val = -DBL_MAX;
1231 		else
1232 			*num_val = s->stats_b->stats[id];
1233 		return;
1234 	case VARIANT_DIFF:
1235 		if (!s->stats_a || !s->stats_b)
1236 			*num_val = -DBL_MAX;
1237 		else if (id == VERDICT)
1238 			*num_val = v1 == v2 ? 1.0 /* MATCH */ : 0.0 /* MISMATCH */;
1239 		else
1240 			*num_val = (double)(v2 - v1);
1241 		return;
1242 	case VARIANT_PCT:
1243 		if (!s->stats_a || !s->stats_b) {
1244 			*num_val = -DBL_MAX;
1245 		} else if (v1 == 0) {
1246 			if (v1 == v2)
1247 				*num_val = 0.0;
1248 			else
1249 				*num_val = v2 < v1 ? -100.0 : 100.0;
1250 		} else {
1251 			 *num_val = (v2 - v1) * 100.0 / v1;
1252 		}
1253 		return;
1254 	}
1255 }
1256 
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,bool abs)1257 static int cmp_join_stat(const struct verif_stats_join *s1,
1258 			 const struct verif_stats_join *s2,
1259 			 enum stat_id id, enum stat_variant var,
1260 			 bool asc, bool abs)
1261 {
1262 	const char *str1 = NULL, *str2 = NULL;
1263 	double v1 = 0.0, v2 = 0.0;
1264 	int cmp = 0;
1265 
1266 	fetch_join_stat_value(s1, id, var, &str1, &v1);
1267 	fetch_join_stat_value(s2, id, var, &str2, &v2);
1268 
1269 	if (abs) {
1270 		v1 = fabs(v1);
1271 		v2 = fabs(v2);
1272 	}
1273 
1274 	if (str1)
1275 		cmp = strcmp(str1, str2);
1276 	else if (v1 != v2)
1277 		cmp = v1 < v2 ? -1 : 1;
1278 
1279 	return asc ? cmp : -cmp;
1280 }
1281 
cmp_join_stats(const void * v1,const void * v2)1282 static int cmp_join_stats(const void *v1, const void *v2)
1283 {
1284 	const struct verif_stats_join *s1 = v1, *s2 = v2;
1285 	int i, cmp;
1286 
1287 	for (i = 0; i < env.sort_spec.spec_cnt; i++) {
1288 		cmp = cmp_join_stat(s1, s2,
1289 				    env.sort_spec.ids[i],
1290 				    env.sort_spec.variants[i],
1291 				    env.sort_spec.asc[i],
1292 				    env.sort_spec.abs[i]);
1293 		if (cmp != 0)
1294 			return cmp;
1295 	}
1296 
1297 	/* always disambiguate with file+prog, which are unique */
1298 	cmp = strcmp(s1->file_name, s2->file_name);
1299 	if (cmp != 0)
1300 		return cmp;
1301 	return strcmp(s1->prog_name, s2->prog_name);
1302 }
1303 
1304 #define HEADER_CHAR '-'
1305 #define COLUMN_SEP "  "
1306 
output_header_underlines(void)1307 static void output_header_underlines(void)
1308 {
1309 	int i, j, len;
1310 
1311 	for (i = 0; i < env.output_spec.spec_cnt; i++) {
1312 		len = env.output_spec.lens[i];
1313 
1314 		printf("%s", i == 0 ? "" : COLUMN_SEP);
1315 		for (j = 0; j < len; j++)
1316 			printf("%c", HEADER_CHAR);
1317 	}
1318 	printf("\n");
1319 }
1320 
output_headers(enum resfmt fmt)1321 static void output_headers(enum resfmt fmt)
1322 {
1323 	const char *fmt_str;
1324 	int i, len;
1325 
1326 	for (i = 0; i < env.output_spec.spec_cnt; i++) {
1327 		int id = env.output_spec.ids[i];
1328 		int *max_len = &env.output_spec.lens[i];
1329 
1330 		switch (fmt) {
1331 		case RESFMT_TABLE_CALCLEN:
1332 			len = snprintf(NULL, 0, "%s", stat_defs[id].header);
1333 			if (len > *max_len)
1334 				*max_len = len;
1335 			break;
1336 		case RESFMT_TABLE:
1337 			fmt_str = stat_defs[id].left_aligned ? "%s%-*s" : "%s%*s";
1338 			printf(fmt_str, i == 0 ? "" : COLUMN_SEP,  *max_len, stat_defs[id].header);
1339 			if (i == env.output_spec.spec_cnt - 1)
1340 				printf("\n");
1341 			break;
1342 		case RESFMT_CSV:
1343 			printf("%s%s", i == 0 ? "" : ",", stat_defs[id].names[0]);
1344 			if (i == env.output_spec.spec_cnt - 1)
1345 				printf("\n");
1346 			break;
1347 		}
1348 	}
1349 
1350 	if (fmt == RESFMT_TABLE)
1351 		output_header_underlines();
1352 }
1353 
prepare_value(const struct verif_stats * s,enum stat_id id,const char ** str,long * val)1354 static void prepare_value(const struct verif_stats *s, enum stat_id id,
1355 			  const char **str, long *val)
1356 {
1357 	switch (id) {
1358 	case FILE_NAME:
1359 		*str = s ? s->file_name : "N/A";
1360 		break;
1361 	case PROG_NAME:
1362 		*str = s ? s->prog_name : "N/A";
1363 		break;
1364 	case VERDICT:
1365 		if (!s)
1366 			*str = "N/A";
1367 		else
1368 			*str = s->stats[VERDICT] ? "success" : "failure";
1369 		break;
1370 	case DURATION:
1371 	case TOTAL_INSNS:
1372 	case TOTAL_STATES:
1373 	case PEAK_STATES:
1374 	case MAX_STATES_PER_INSN:
1375 	case MARK_READ_MAX_LEN:
1376 		*val = s ? s->stats[id] : 0;
1377 		break;
1378 	default:
1379 		fprintf(stderr, "Unrecognized stat #%d\n", id);
1380 		exit(1);
1381 	}
1382 }
1383 
output_stats(const struct verif_stats * s,enum resfmt fmt,bool last)1384 static void output_stats(const struct verif_stats *s, enum resfmt fmt, bool last)
1385 {
1386 	int i;
1387 
1388 	for (i = 0; i < env.output_spec.spec_cnt; i++) {
1389 		int id = env.output_spec.ids[i];
1390 		int *max_len = &env.output_spec.lens[i], len;
1391 		const char *str = NULL;
1392 		long val = 0;
1393 
1394 		prepare_value(s, id, &str, &val);
1395 
1396 		switch (fmt) {
1397 		case RESFMT_TABLE_CALCLEN:
1398 			if (str)
1399 				len = snprintf(NULL, 0, "%s", str);
1400 			else
1401 				len = snprintf(NULL, 0, "%ld", val);
1402 			if (len > *max_len)
1403 				*max_len = len;
1404 			break;
1405 		case RESFMT_TABLE:
1406 			if (str)
1407 				printf("%s%-*s", i == 0 ? "" : COLUMN_SEP, *max_len, str);
1408 			else
1409 				printf("%s%*ld", i == 0 ? "" : COLUMN_SEP,  *max_len, val);
1410 			if (i == env.output_spec.spec_cnt - 1)
1411 				printf("\n");
1412 			break;
1413 		case RESFMT_CSV:
1414 			if (str)
1415 				printf("%s%s", i == 0 ? "" : ",", str);
1416 			else
1417 				printf("%s%ld", i == 0 ? "" : ",", val);
1418 			if (i == env.output_spec.spec_cnt - 1)
1419 				printf("\n");
1420 			break;
1421 		}
1422 	}
1423 
1424 	if (last && fmt == RESFMT_TABLE) {
1425 		output_header_underlines();
1426 		printf("Done. Processed %d files, %d programs. Skipped %d files, %d programs.\n",
1427 		       env.files_processed, env.files_skipped, env.progs_processed, env.progs_skipped);
1428 	}
1429 }
1430 
parse_stat_value(const char * str,enum stat_id id,struct verif_stats * st)1431 static int parse_stat_value(const char *str, enum stat_id id, struct verif_stats *st)
1432 {
1433 	switch (id) {
1434 	case FILE_NAME:
1435 		st->file_name = strdup(str);
1436 		if (!st->file_name)
1437 			return -ENOMEM;
1438 		break;
1439 	case PROG_NAME:
1440 		st->prog_name = strdup(str);
1441 		if (!st->prog_name)
1442 			return -ENOMEM;
1443 		break;
1444 	case VERDICT:
1445 		if (strcmp(str, "success") == 0) {
1446 			st->stats[VERDICT] = true;
1447 		} else if (strcmp(str, "failure") == 0) {
1448 			st->stats[VERDICT] = false;
1449 		} else {
1450 			fprintf(stderr, "Unrecognized verification verdict '%s'\n", str);
1451 			return -EINVAL;
1452 		}
1453 		break;
1454 	case DURATION:
1455 	case TOTAL_INSNS:
1456 	case TOTAL_STATES:
1457 	case PEAK_STATES:
1458 	case MAX_STATES_PER_INSN:
1459 	case MARK_READ_MAX_LEN: {
1460 		long val;
1461 		int err, n;
1462 
1463 		if (sscanf(str, "%ld %n", &val, &n) != 1 || n != strlen(str)) {
1464 			err = -errno;
1465 			fprintf(stderr, "Failed to parse '%s' as integer\n", str);
1466 			return err;
1467 		}
1468 
1469 		st->stats[id] = val;
1470 		break;
1471 	}
1472 	default:
1473 		fprintf(stderr, "Unrecognized stat #%d\n", id);
1474 		return -EINVAL;
1475 	}
1476 	return 0;
1477 }
1478 
parse_stats_csv(const char * filename,struct stat_specs * specs,struct verif_stats ** statsp,int * stat_cntp)1479 static int parse_stats_csv(const char *filename, struct stat_specs *specs,
1480 			   struct verif_stats **statsp, int *stat_cntp)
1481 {
1482 	char line[4096];
1483 	FILE *f;
1484 	int err = 0;
1485 	bool header = true;
1486 
1487 	f = fopen(filename, "r");
1488 	if (!f) {
1489 		err = -errno;
1490 		fprintf(stderr, "Failed to open '%s': %d\n", filename, err);
1491 		return err;
1492 	}
1493 
1494 	*stat_cntp = 0;
1495 
1496 	while (fgets(line, sizeof(line), f)) {
1497 		char *input = line, *state = NULL, *next;
1498 		struct verif_stats *st = NULL;
1499 		int col = 0, cnt = 0;
1500 
1501 		if (!header) {
1502 			void *tmp;
1503 
1504 			tmp = realloc(*statsp, (*stat_cntp + 1) * sizeof(**statsp));
1505 			if (!tmp) {
1506 				err = -ENOMEM;
1507 				goto cleanup;
1508 			}
1509 			*statsp = tmp;
1510 
1511 			st = &(*statsp)[*stat_cntp];
1512 			memset(st, 0, sizeof(*st));
1513 
1514 			*stat_cntp += 1;
1515 		}
1516 
1517 		while ((next = strtok_r(cnt++ ? NULL : input, ",\n", &state))) {
1518 			if (header) {
1519 				/* for the first line, set up spec stats */
1520 				err = parse_stat(next, specs);
1521 				if (err)
1522 					goto cleanup;
1523 				continue;
1524 			}
1525 
1526 			/* for all other lines, parse values based on spec */
1527 			if (col >= specs->spec_cnt) {
1528 				fprintf(stderr, "Found extraneous column #%d in row #%d of '%s'\n",
1529 					col, *stat_cntp, filename);
1530 				err = -EINVAL;
1531 				goto cleanup;
1532 			}
1533 			err = parse_stat_value(next, specs->ids[col], st);
1534 			if (err)
1535 				goto cleanup;
1536 			col++;
1537 		}
1538 
1539 		if (header) {
1540 			header = false;
1541 			continue;
1542 		}
1543 
1544 		if (col < specs->spec_cnt) {
1545 			fprintf(stderr, "Not enough columns in row #%d in '%s'\n",
1546 				*stat_cntp, filename);
1547 			err = -EINVAL;
1548 			goto cleanup;
1549 		}
1550 
1551 		if (!st->file_name || !st->prog_name) {
1552 			fprintf(stderr, "Row #%d in '%s' is missing file and/or program name\n",
1553 				*stat_cntp, filename);
1554 			err = -EINVAL;
1555 			goto cleanup;
1556 		}
1557 
1558 		/* in comparison mode we can only check filters after we
1559 		 * parsed entire line; if row should be ignored we pretend we
1560 		 * never parsed it
1561 		 */
1562 		if (!should_process_file_prog(st->file_name, st->prog_name)) {
1563 			free(st->file_name);
1564 			free(st->prog_name);
1565 			*stat_cntp -= 1;
1566 		}
1567 	}
1568 
1569 	if (!feof(f)) {
1570 		err = -errno;
1571 		fprintf(stderr, "Failed I/O for '%s': %d\n", filename, err);
1572 	}
1573 
1574 cleanup:
1575 	fclose(f);
1576 	return err;
1577 }
1578 
1579 /* empty/zero stats for mismatched rows */
1580 static const struct verif_stats fallback_stats = { .file_name = "", .prog_name = "" };
1581 
is_key_stat(enum stat_id id)1582 static bool is_key_stat(enum stat_id id)
1583 {
1584 	return id == FILE_NAME || id == PROG_NAME;
1585 }
1586 
output_comp_header_underlines(void)1587 static void output_comp_header_underlines(void)
1588 {
1589 	int i, j, k;
1590 
1591 	for (i = 0; i < env.output_spec.spec_cnt; i++) {
1592 		int id = env.output_spec.ids[i];
1593 		int max_j = is_key_stat(id) ? 1 : 3;
1594 
1595 		for (j = 0; j < max_j; j++) {
1596 			int len = env.output_spec.lens[3 * i + j];
1597 
1598 			printf("%s", i + j == 0 ? "" : COLUMN_SEP);
1599 
1600 			for (k = 0; k < len; k++)
1601 				printf("%c", HEADER_CHAR);
1602 		}
1603 	}
1604 	printf("\n");
1605 }
1606 
output_comp_headers(enum resfmt fmt)1607 static void output_comp_headers(enum resfmt fmt)
1608 {
1609 	static const char *table_sfxs[3] = {" (A)", " (B)", " (DIFF)"};
1610 	static const char *name_sfxs[3] = {"_base", "_comp", "_diff"};
1611 	int i, j, len;
1612 
1613 	for (i = 0; i < env.output_spec.spec_cnt; i++) {
1614 		int id = env.output_spec.ids[i];
1615 		/* key stats don't have A/B/DIFF columns, they are common for both data sets */
1616 		int max_j = is_key_stat(id) ? 1 : 3;
1617 
1618 		for (j = 0; j < max_j; j++) {
1619 			int *max_len = &env.output_spec.lens[3 * i + j];
1620 			bool last = (i == env.output_spec.spec_cnt - 1) && (j == max_j - 1);
1621 			const char *sfx;
1622 
1623 			switch (fmt) {
1624 			case RESFMT_TABLE_CALCLEN:
1625 				sfx = is_key_stat(id) ? "" : table_sfxs[j];
1626 				len = snprintf(NULL, 0, "%s%s", stat_defs[id].header, sfx);
1627 				if (len > *max_len)
1628 					*max_len = len;
1629 				break;
1630 			case RESFMT_TABLE:
1631 				sfx = is_key_stat(id) ? "" : table_sfxs[j];
1632 				printf("%s%-*s%s", i + j == 0 ? "" : COLUMN_SEP,
1633 				       *max_len - (int)strlen(sfx), stat_defs[id].header, sfx);
1634 				if (last)
1635 					printf("\n");
1636 				break;
1637 			case RESFMT_CSV:
1638 				sfx = is_key_stat(id) ? "" : name_sfxs[j];
1639 				printf("%s%s%s", i + j == 0 ? "" : ",", stat_defs[id].names[0], sfx);
1640 				if (last)
1641 					printf("\n");
1642 				break;
1643 			}
1644 		}
1645 	}
1646 
1647 	if (fmt == RESFMT_TABLE)
1648 		output_comp_header_underlines();
1649 }
1650 
output_comp_stats(const struct verif_stats_join * join_stats,enum resfmt fmt,bool last)1651 static void output_comp_stats(const struct verif_stats_join *join_stats,
1652 			      enum resfmt fmt, bool last)
1653 {
1654 	const struct verif_stats *base = join_stats->stats_a;
1655 	const struct verif_stats *comp = join_stats->stats_b;
1656 	char base_buf[1024] = {}, comp_buf[1024] = {}, diff_buf[1024] = {};
1657 	int i;
1658 
1659 	for (i = 0; i < env.output_spec.spec_cnt; i++) {
1660 		int id = env.output_spec.ids[i], len;
1661 		int *max_len_base = &env.output_spec.lens[3 * i + 0];
1662 		int *max_len_comp = &env.output_spec.lens[3 * i + 1];
1663 		int *max_len_diff = &env.output_spec.lens[3 * i + 2];
1664 		const char *base_str = NULL, *comp_str = NULL;
1665 		long base_val = 0, comp_val = 0, diff_val = 0;
1666 
1667 		prepare_value(base, id, &base_str, &base_val);
1668 		prepare_value(comp, id, &comp_str, &comp_val);
1669 
1670 		/* normalize all the outputs to be in string buffers for simplicity */
1671 		if (is_key_stat(id)) {
1672 			/* key stats (file and program name) are always strings */
1673 			if (base)
1674 				snprintf(base_buf, sizeof(base_buf), "%s", base_str);
1675 			else
1676 				snprintf(base_buf, sizeof(base_buf), "%s", comp_str);
1677 		} else if (base_str) {
1678 			snprintf(base_buf, sizeof(base_buf), "%s", base_str);
1679 			snprintf(comp_buf, sizeof(comp_buf), "%s", comp_str);
1680 			if (!base || !comp)
1681 				snprintf(diff_buf, sizeof(diff_buf), "%s", "N/A");
1682 			else if (strcmp(base_str, comp_str) == 0)
1683 				snprintf(diff_buf, sizeof(diff_buf), "%s", "MATCH");
1684 			else
1685 				snprintf(diff_buf, sizeof(diff_buf), "%s", "MISMATCH");
1686 		} else {
1687 			double p = 0.0;
1688 
1689 			if (base)
1690 				snprintf(base_buf, sizeof(base_buf), "%ld", base_val);
1691 			else
1692 				snprintf(base_buf, sizeof(base_buf), "%s", "N/A");
1693 			if (comp)
1694 				snprintf(comp_buf, sizeof(comp_buf), "%ld", comp_val);
1695 			else
1696 				snprintf(comp_buf, sizeof(comp_buf), "%s", "N/A");
1697 
1698 			diff_val = comp_val - base_val;
1699 			if (!base || !comp) {
1700 				snprintf(diff_buf, sizeof(diff_buf), "%s", "N/A");
1701 			} else {
1702 				if (base_val == 0) {
1703 					if (comp_val == base_val)
1704 						p = 0.0; /* avoid +0 (+100%) case */
1705 					else
1706 						p = comp_val < base_val ? -100.0 : 100.0;
1707 				} else {
1708 					 p = diff_val * 100.0 / base_val;
1709 				}
1710 				snprintf(diff_buf, sizeof(diff_buf), "%+ld (%+.2lf%%)", diff_val, p);
1711 			}
1712 		}
1713 
1714 		switch (fmt) {
1715 		case RESFMT_TABLE_CALCLEN:
1716 			len = strlen(base_buf);
1717 			if (len > *max_len_base)
1718 				*max_len_base = len;
1719 			if (!is_key_stat(id)) {
1720 				len = strlen(comp_buf);
1721 				if (len > *max_len_comp)
1722 					*max_len_comp = len;
1723 				len = strlen(diff_buf);
1724 				if (len > *max_len_diff)
1725 					*max_len_diff = len;
1726 			}
1727 			break;
1728 		case RESFMT_TABLE: {
1729 			/* string outputs are left-aligned, number outputs are right-aligned */
1730 			const char *fmt = base_str ? "%s%-*s" : "%s%*s";
1731 
1732 			printf(fmt, i == 0 ? "" : COLUMN_SEP, *max_len_base, base_buf);
1733 			if (!is_key_stat(id)) {
1734 				printf(fmt, COLUMN_SEP, *max_len_comp, comp_buf);
1735 				printf(fmt, COLUMN_SEP, *max_len_diff, diff_buf);
1736 			}
1737 			if (i == env.output_spec.spec_cnt - 1)
1738 				printf("\n");
1739 			break;
1740 		}
1741 		case RESFMT_CSV:
1742 			printf("%s%s", i == 0 ? "" : ",", base_buf);
1743 			if (!is_key_stat(id)) {
1744 				printf("%s%s", i == 0 ? "" : ",", comp_buf);
1745 				printf("%s%s", i == 0 ? "" : ",", diff_buf);
1746 			}
1747 			if (i == env.output_spec.spec_cnt - 1)
1748 				printf("\n");
1749 			break;
1750 		}
1751 	}
1752 
1753 	if (last && fmt == RESFMT_TABLE)
1754 		output_comp_header_underlines();
1755 }
1756 
cmp_stats_key(const struct verif_stats * base,const struct verif_stats * comp)1757 static int cmp_stats_key(const struct verif_stats *base, const struct verif_stats *comp)
1758 {
1759 	int r;
1760 
1761 	r = strcmp(base->file_name, comp->file_name);
1762 	if (r != 0)
1763 		return r;
1764 	return strcmp(base->prog_name, comp->prog_name);
1765 }
1766 
is_join_stat_filter_matched(struct filter * f,const struct verif_stats_join * stats)1767 static bool is_join_stat_filter_matched(struct filter *f, const struct verif_stats_join *stats)
1768 {
1769 	static const double eps = 1e-9;
1770 	const char *str = NULL;
1771 	double value = 0.0;
1772 
1773 	fetch_join_stat_value(stats, f->stat_id, f->stat_var, &str, &value);
1774 
1775 	if (f->abs)
1776 		value = fabs(value);
1777 
1778 	switch (f->op) {
1779 	case OP_EQ: return value > f->value - eps && value < f->value + eps;
1780 	case OP_NEQ: return value < f->value - eps || value > f->value + eps;
1781 	case OP_LT: return value < f->value - eps;
1782 	case OP_LE: return value <= f->value + eps;
1783 	case OP_GT: return value > f->value + eps;
1784 	case OP_GE: return value >= f->value - eps;
1785 	}
1786 
1787 	fprintf(stderr, "BUG: unknown filter op %d!\n", f->op);
1788 	return false;
1789 }
1790 
should_output_join_stats(const struct verif_stats_join * stats)1791 static bool should_output_join_stats(const struct verif_stats_join *stats)
1792 {
1793 	struct filter *f;
1794 	int i, allow_cnt = 0;
1795 
1796 	for (i = 0; i < env.deny_filter_cnt; i++) {
1797 		f = &env.deny_filters[i];
1798 		if (f->kind != FILTER_STAT)
1799 			continue;
1800 
1801 		if (is_join_stat_filter_matched(f, stats))
1802 			return false;
1803 	}
1804 
1805 	for (i = 0; i < env.allow_filter_cnt; i++) {
1806 		f = &env.allow_filters[i];
1807 		if (f->kind != FILTER_STAT)
1808 			continue;
1809 		allow_cnt++;
1810 
1811 		if (is_join_stat_filter_matched(f, stats))
1812 			return true;
1813 	}
1814 
1815 	/* if there are no stat allowed filters, pass everything through */
1816 	return allow_cnt == 0;
1817 }
1818 
handle_comparison_mode(void)1819 static int handle_comparison_mode(void)
1820 {
1821 	struct stat_specs base_specs = {}, comp_specs = {};
1822 	struct stat_specs tmp_sort_spec;
1823 	enum resfmt cur_fmt;
1824 	int err, i, j, last_idx, cnt;
1825 
1826 	if (env.filename_cnt != 2) {
1827 		fprintf(stderr, "Comparison mode expects exactly two input CSV files!\n\n");
1828 		argp_help(&argp, stderr, ARGP_HELP_USAGE, "veristat");
1829 		return -EINVAL;
1830 	}
1831 
1832 	err = parse_stats_csv(env.filenames[0], &base_specs,
1833 			      &env.baseline_stats, &env.baseline_stat_cnt);
1834 	if (err) {
1835 		fprintf(stderr, "Failed to parse stats from '%s': %d\n", env.filenames[0], err);
1836 		return err;
1837 	}
1838 	err = parse_stats_csv(env.filenames[1], &comp_specs,
1839 			      &env.prog_stats, &env.prog_stat_cnt);
1840 	if (err) {
1841 		fprintf(stderr, "Failed to parse stats from '%s': %d\n", env.filenames[1], err);
1842 		return err;
1843 	}
1844 
1845 	/* To keep it simple we validate that the set and order of stats in
1846 	 * both CSVs are exactly the same. This can be lifted with a bit more
1847 	 * pre-processing later.
1848 	 */
1849 	if (base_specs.spec_cnt != comp_specs.spec_cnt) {
1850 		fprintf(stderr, "Number of stats in '%s' and '%s' differs (%d != %d)!\n",
1851 			env.filenames[0], env.filenames[1],
1852 			base_specs.spec_cnt, comp_specs.spec_cnt);
1853 		return -EINVAL;
1854 	}
1855 	for (i = 0; i < base_specs.spec_cnt; i++) {
1856 		if (base_specs.ids[i] != comp_specs.ids[i]) {
1857 			fprintf(stderr, "Stats composition differs between '%s' and '%s' (%s != %s)!\n",
1858 				env.filenames[0], env.filenames[1],
1859 				stat_defs[base_specs.ids[i]].names[0],
1860 				stat_defs[comp_specs.ids[i]].names[0]);
1861 			return -EINVAL;
1862 		}
1863 	}
1864 
1865 	/* Replace user-specified sorting spec with file+prog sorting rule to
1866 	 * be able to join two datasets correctly. Once we are done, we will
1867 	 * restore the original sort spec.
1868 	 */
1869 	tmp_sort_spec = env.sort_spec;
1870 	env.sort_spec = join_sort_spec;
1871 	qsort(env.prog_stats, env.prog_stat_cnt, sizeof(*env.prog_stats), cmp_prog_stats);
1872 	qsort(env.baseline_stats, env.baseline_stat_cnt, sizeof(*env.baseline_stats), cmp_prog_stats);
1873 	env.sort_spec = tmp_sort_spec;
1874 
1875 	/* Join two datasets together. If baseline and comparison datasets
1876 	 * have different subset of rows (we match by 'object + prog' as
1877 	 * a unique key) then assume empty/missing/zero value for rows that
1878 	 * are missing in the opposite data set.
1879 	 */
1880 	i = j = 0;
1881 	while (i < env.baseline_stat_cnt || j < env.prog_stat_cnt) {
1882 		const struct verif_stats *base, *comp;
1883 		struct verif_stats_join *join;
1884 		void *tmp;
1885 		int r;
1886 
1887 		base = i < env.baseline_stat_cnt ? &env.baseline_stats[i] : &fallback_stats;
1888 		comp = j < env.prog_stat_cnt ? &env.prog_stats[j] : &fallback_stats;
1889 
1890 		if (!base->file_name || !base->prog_name) {
1891 			fprintf(stderr, "Entry #%d in '%s' doesn't have file and/or program name specified!\n",
1892 				i, env.filenames[0]);
1893 			return -EINVAL;
1894 		}
1895 		if (!comp->file_name || !comp->prog_name) {
1896 			fprintf(stderr, "Entry #%d in '%s' doesn't have file and/or program name specified!\n",
1897 				j, env.filenames[1]);
1898 			return -EINVAL;
1899 		}
1900 
1901 		tmp = realloc(env.join_stats, (env.join_stat_cnt + 1) * sizeof(*env.join_stats));
1902 		if (!tmp)
1903 			return -ENOMEM;
1904 		env.join_stats = tmp;
1905 
1906 		join = &env.join_stats[env.join_stat_cnt];
1907 		memset(join, 0, sizeof(*join));
1908 
1909 		r = cmp_stats_key(base, comp);
1910 		if (r == 0) {
1911 			join->file_name = base->file_name;
1912 			join->prog_name = base->prog_name;
1913 			join->stats_a = base;
1914 			join->stats_b = comp;
1915 			i++;
1916 			j++;
1917 		} else if (base != &fallback_stats && (comp == &fallback_stats || r < 0)) {
1918 			join->file_name = base->file_name;
1919 			join->prog_name = base->prog_name;
1920 			join->stats_a = base;
1921 			join->stats_b = NULL;
1922 			i++;
1923 		} else if (comp != &fallback_stats && (base == &fallback_stats || r > 0)) {
1924 			join->file_name = comp->file_name;
1925 			join->prog_name = comp->prog_name;
1926 			join->stats_a = NULL;
1927 			join->stats_b = comp;
1928 			j++;
1929 		} else {
1930 			fprintf(stderr, "%s:%d: should never reach here i=%i, j=%i",
1931 				__FILE__, __LINE__, i, j);
1932 			return -EINVAL;
1933 		}
1934 		env.join_stat_cnt += 1;
1935 	}
1936 
1937 	/* now sort joined results according to sort spec */
1938 	qsort(env.join_stats, env.join_stat_cnt, sizeof(*env.join_stats), cmp_join_stats);
1939 
1940 	/* for human-readable table output we need to do extra pass to
1941 	 * calculate column widths, so we substitute current output format
1942 	 * with RESFMT_TABLE_CALCLEN and later revert it back to RESFMT_TABLE
1943 	 * and do everything again.
1944 	 */
1945 	if (env.out_fmt == RESFMT_TABLE)
1946 		cur_fmt = RESFMT_TABLE_CALCLEN;
1947 	else
1948 		cur_fmt = env.out_fmt;
1949 
1950 one_more_time:
1951 	output_comp_headers(cur_fmt);
1952 
1953 	last_idx = -1;
1954 	cnt = 0;
1955 	for (i = 0; i < env.join_stat_cnt; i++) {
1956 		const struct verif_stats_join *join = &env.join_stats[i];
1957 
1958 		if (!should_output_join_stats(join))
1959 			continue;
1960 
1961 		if (env.top_n && cnt >= env.top_n)
1962 			break;
1963 
1964 		if (cur_fmt == RESFMT_TABLE_CALCLEN)
1965 			last_idx = i;
1966 
1967 		output_comp_stats(join, cur_fmt, i == last_idx);
1968 
1969 		cnt++;
1970 	}
1971 
1972 	if (cur_fmt == RESFMT_TABLE_CALCLEN) {
1973 		cur_fmt = RESFMT_TABLE;
1974 		goto one_more_time; /* ... this time with feeling */
1975 	}
1976 
1977 	return 0;
1978 }
1979 
is_stat_filter_matched(struct filter * f,const struct verif_stats * stats)1980 static bool is_stat_filter_matched(struct filter *f, const struct verif_stats *stats)
1981 {
1982 	long value = stats->stats[f->stat_id];
1983 
1984 	if (f->abs)
1985 		value = value < 0 ? -value : value;
1986 
1987 	switch (f->op) {
1988 	case OP_EQ: return value == f->value;
1989 	case OP_NEQ: return value != f->value;
1990 	case OP_LT: return value < f->value;
1991 	case OP_LE: return value <= f->value;
1992 	case OP_GT: return value > f->value;
1993 	case OP_GE: return value >= f->value;
1994 	}
1995 
1996 	fprintf(stderr, "BUG: unknown filter op %d!\n", f->op);
1997 	return false;
1998 }
1999 
should_output_stats(const struct verif_stats * stats)2000 static bool should_output_stats(const struct verif_stats *stats)
2001 {
2002 	struct filter *f;
2003 	int i, allow_cnt = 0;
2004 
2005 	for (i = 0; i < env.deny_filter_cnt; i++) {
2006 		f = &env.deny_filters[i];
2007 		if (f->kind != FILTER_STAT)
2008 			continue;
2009 
2010 		if (is_stat_filter_matched(f, stats))
2011 			return false;
2012 	}
2013 
2014 	for (i = 0; i < env.allow_filter_cnt; i++) {
2015 		f = &env.allow_filters[i];
2016 		if (f->kind != FILTER_STAT)
2017 			continue;
2018 		allow_cnt++;
2019 
2020 		if (is_stat_filter_matched(f, stats))
2021 			return true;
2022 	}
2023 
2024 	/* if there are no stat allowed filters, pass everything through */
2025 	return allow_cnt == 0;
2026 }
2027 
output_prog_stats(void)2028 static void output_prog_stats(void)
2029 {
2030 	const struct verif_stats *stats;
2031 	int i, last_stat_idx = 0, cnt = 0;
2032 
2033 	if (env.out_fmt == RESFMT_TABLE) {
2034 		/* calculate column widths */
2035 		output_headers(RESFMT_TABLE_CALCLEN);
2036 		for (i = 0; i < env.prog_stat_cnt; i++) {
2037 			stats = &env.prog_stats[i];
2038 			if (!should_output_stats(stats))
2039 				continue;
2040 			output_stats(stats, RESFMT_TABLE_CALCLEN, false);
2041 			last_stat_idx = i;
2042 		}
2043 	}
2044 
2045 	/* actually output the table */
2046 	output_headers(env.out_fmt);
2047 	for (i = 0; i < env.prog_stat_cnt; i++) {
2048 		stats = &env.prog_stats[i];
2049 		if (!should_output_stats(stats))
2050 			continue;
2051 		if (env.top_n && cnt >= env.top_n)
2052 			break;
2053 		output_stats(stats, env.out_fmt, i == last_stat_idx);
2054 		cnt++;
2055 	}
2056 }
2057 
handle_verif_mode(void)2058 static int handle_verif_mode(void)
2059 {
2060 	int i, err;
2061 
2062 	if (env.filename_cnt == 0) {
2063 		fprintf(stderr, "Please provide path to BPF object file!\n\n");
2064 		argp_help(&argp, stderr, ARGP_HELP_USAGE, "veristat");
2065 		return -EINVAL;
2066 	}
2067 
2068 	for (i = 0; i < env.filename_cnt; i++) {
2069 		err = process_obj(env.filenames[i]);
2070 		if (err) {
2071 			fprintf(stderr, "Failed to process '%s': %d\n", env.filenames[i], err);
2072 			return err;
2073 		}
2074 	}
2075 
2076 	qsort(env.prog_stats, env.prog_stat_cnt, sizeof(*env.prog_stats), cmp_prog_stats);
2077 
2078 	output_prog_stats();
2079 
2080 	return 0;
2081 }
2082 
handle_replay_mode(void)2083 static int handle_replay_mode(void)
2084 {
2085 	struct stat_specs specs = {};
2086 	int err;
2087 
2088 	if (env.filename_cnt != 1) {
2089 		fprintf(stderr, "Replay mode expects exactly one input CSV file!\n\n");
2090 		argp_help(&argp, stderr, ARGP_HELP_USAGE, "veristat");
2091 		return -EINVAL;
2092 	}
2093 
2094 	err = parse_stats_csv(env.filenames[0], &specs,
2095 			      &env.prog_stats, &env.prog_stat_cnt);
2096 	if (err) {
2097 		fprintf(stderr, "Failed to parse stats from '%s': %d\n", env.filenames[0], err);
2098 		return err;
2099 	}
2100 
2101 	qsort(env.prog_stats, env.prog_stat_cnt, sizeof(*env.prog_stats), cmp_prog_stats);
2102 
2103 	output_prog_stats();
2104 
2105 	return 0;
2106 }
2107 
main(int argc,char ** argv)2108 int main(int argc, char **argv)
2109 {
2110 	int err = 0, i;
2111 
2112 	if (argp_parse(&argp, argc, argv, 0, NULL, NULL))
2113 		return 1;
2114 
2115 	if (env.show_version) {
2116 		printf("%s\n", argp_program_version);
2117 		return 0;
2118 	}
2119 
2120 	if (env.verbose && env.quiet) {
2121 		fprintf(stderr, "Verbose and quiet modes are incompatible, please specify just one or neither!\n\n");
2122 		argp_help(&argp, stderr, ARGP_HELP_USAGE, "veristat");
2123 		return 1;
2124 	}
2125 	if (env.verbose && env.log_level == 0)
2126 		env.log_level = 1;
2127 
2128 	if (env.output_spec.spec_cnt == 0) {
2129 		if (env.out_fmt == RESFMT_CSV)
2130 			env.output_spec = default_csv_output_spec;
2131 		else
2132 			env.output_spec = default_output_spec;
2133 	}
2134 	if (env.sort_spec.spec_cnt == 0)
2135 		env.sort_spec = default_sort_spec;
2136 
2137 	if (env.comparison_mode && env.replay_mode) {
2138 		fprintf(stderr, "Can't specify replay and comparison mode at the same time!\n\n");
2139 		argp_help(&argp, stderr, ARGP_HELP_USAGE, "veristat");
2140 		return 1;
2141 	}
2142 
2143 	if (env.comparison_mode)
2144 		err = handle_comparison_mode();
2145 	else if (env.replay_mode)
2146 		err = handle_replay_mode();
2147 	else
2148 		err = handle_verif_mode();
2149 
2150 	free_verif_stats(env.prog_stats, env.prog_stat_cnt);
2151 	free_verif_stats(env.baseline_stats, env.baseline_stat_cnt);
2152 	free(env.join_stats);
2153 	for (i = 0; i < env.filename_cnt; i++)
2154 		free(env.filenames[i]);
2155 	free(env.filenames);
2156 	for (i = 0; i < env.allow_filter_cnt; i++) {
2157 		free(env.allow_filters[i].any_glob);
2158 		free(env.allow_filters[i].file_glob);
2159 		free(env.allow_filters[i].prog_glob);
2160 	}
2161 	free(env.allow_filters);
2162 	for (i = 0; i < env.deny_filter_cnt; i++) {
2163 		free(env.deny_filters[i].any_glob);
2164 		free(env.deny_filters[i].file_glob);
2165 		free(env.deny_filters[i].prog_glob);
2166 	}
2167 	free(env.deny_filters);
2168 	return -err;
2169 }
2170