~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/tools/testing/selftests/bpf/veristat.c

Version: ~ [ linux-6.11.5 ] ~ [ linux-6.10.14 ] ~ [ linux-6.9.12 ] ~ [ linux-6.8.12 ] ~ [ linux-6.7.12 ] ~ [ linux-6.6.58 ] ~ [ linux-6.5.13 ] ~ [ linux-6.4.16 ] ~ [ linux-6.3.13 ] ~ [ linux-6.2.16 ] ~ [ linux-6.1.114 ] ~ [ linux-6.0.19 ] ~ [ linux-5.19.17 ] ~ [ linux-5.18.19 ] ~ [ linux-5.17.15 ] ~ [ linux-5.16.20 ] ~ [ linux-5.15.169 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.228 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.284 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.322 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.336 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.337 ] ~ [ linux-4.4.302 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.9 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

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

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

sflogo.php