1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * JSON export. 4 * 5 * Copyright (C) 2021, CodeWeavers Inc. <nfraser@codeweavers.com> 6 */ 7 8 #include "data-convert.h" 9 10 #include <fcntl.h> 11 #include <inttypes.h> 12 #include <sys/stat.h> 13 #include <unistd.h> 14 15 #include "linux/compiler.h" 16 #include "linux/err.h" 17 #include "util/auxtrace.h" 18 #include "util/debug.h" 19 #include "util/dso.h" 20 #include "util/event.h" 21 #include "util/evsel.h" 22 #include "util/evlist.h" 23 #include "util/header.h" 24 #include "util/map.h" 25 #include "util/session.h" 26 #include "util/symbol.h" 27 #include "util/thread.h" 28 #include "util/tool.h" 29 30 #ifdef HAVE_LIBTRACEEVENT 31 #include <traceevent/event-parse.h> 32 #endif 33 34 struct convert_json { 35 struct perf_tool tool; 36 FILE *out; 37 bool first; 38 u64 events_count; 39 }; 40 41 // Outputs a JSON-encoded string surrounded by quotes with characters escaped. 42 static void output_json_string(FILE *out, const char *s) 43 { 44 fputc('"', out); 45 while (*s) { 46 switch (*s) { 47 48 // required escapes with special forms as per RFC 8259 49 case '"': fputs("\\\"", out); break; 50 case '\\': fputs("\\\\", out); break; 51 case '\b': fputs("\\b", out); break; 52 case '\f': fputs("\\f", out); break; 53 case '\n': fputs("\\n", out); break; 54 case '\r': fputs("\\r", out); break; 55 case '\t': fputs("\\t", out); break; 56 57 default: 58 // all other control characters must be escaped by hex code 59 if (*s <= 0x1f) 60 fprintf(out, "\\u%04x", *s); 61 else 62 fputc(*s, out); 63 break; 64 } 65 66 ++s; 67 } 68 fputc('"', out); 69 } 70 71 // Outputs an optional comma, newline and indentation to delimit a new value 72 // from the previous one in a JSON object or array. 73 static void output_json_delimiters(FILE *out, bool comma, int depth) 74 { 75 int i; 76 77 if (comma) 78 fputc(',', out); 79 fputc('\n', out); 80 for (i = 0; i < depth; ++i) 81 fputc('\t', out); 82 } 83 84 // Outputs a printf format string (with delimiter) as a JSON value. 85 __printf(4, 5) 86 static void output_json_format(FILE *out, bool comma, int depth, const char *format, ...) 87 { 88 va_list args; 89 90 output_json_delimiters(out, comma, depth); 91 va_start(args, format); 92 vfprintf(out, format, args); 93 va_end(args); 94 } 95 96 // Outputs a JSON key-value pair where the value is a string. 97 static void output_json_key_string(FILE *out, bool comma, int depth, 98 const char *key, const char *value) 99 { 100 output_json_delimiters(out, comma, depth); 101 output_json_string(out, key); 102 fputs(": ", out); 103 output_json_string(out, value); 104 } 105 106 // Outputs a JSON key-value pair where the value is a printf format string. 107 __printf(5, 6) 108 static void output_json_key_format(FILE *out, bool comma, int depth, 109 const char *key, const char *format, ...) 110 { 111 va_list args; 112 113 output_json_delimiters(out, comma, depth); 114 output_json_string(out, key); 115 fputs(": ", out); 116 va_start(args, format); 117 vfprintf(out, format, args); 118 va_end(args); 119 } 120 121 static void output_sample_callchain_entry(struct perf_tool *tool, 122 u64 ip, struct addr_location *al) 123 { 124 struct convert_json *c = container_of(tool, struct convert_json, tool); 125 FILE *out = c->out; 126 127 output_json_format(out, false, 4, "{"); 128 output_json_key_format(out, false, 5, "ip", "\"0x%" PRIx64 "\"", ip); 129 130 if (al && al->sym && al->sym->namelen) { 131 struct dso *dso = al->map ? map__dso(al->map) : NULL; 132 133 fputc(',', out); 134 output_json_key_string(out, false, 5, "symbol", al->sym->name); 135 136 if (dso) { 137 const char *dso_name = dso->short_name; 138 139 if (dso_name && strlen(dso_name) > 0) { 140 fputc(',', out); 141 output_json_key_string(out, false, 5, "dso", dso_name); 142 } 143 } 144 } 145 146 output_json_format(out, false, 4, "}"); 147 } 148 149 static int process_sample_event(struct perf_tool *tool, 150 union perf_event *event __maybe_unused, 151 struct perf_sample *sample, 152 struct evsel *evsel __maybe_unused, 153 struct machine *machine) 154 { 155 struct convert_json *c = container_of(tool, struct convert_json, tool); 156 FILE *out = c->out; 157 struct addr_location al, tal; 158 u64 sample_type = __evlist__combined_sample_type(evsel->evlist); 159 u8 cpumode = PERF_RECORD_MISC_USER; 160 161 if (machine__resolve(machine, &al, sample) < 0) { 162 pr_err("Sample resolution failed!\n"); 163 return -1; 164 } 165 166 ++c->events_count; 167 168 if (c->first) 169 c->first = false; 170 else 171 fputc(',', out); 172 output_json_format(out, false, 2, "{"); 173 174 output_json_key_format(out, false, 3, "timestamp", "%" PRIi64, sample->time); 175 output_json_key_format(out, true, 3, "pid", "%i", al.thread->pid_); 176 output_json_key_format(out, true, 3, "tid", "%i", al.thread->tid); 177 178 if ((sample_type & PERF_SAMPLE_CPU)) 179 output_json_key_format(out, true, 3, "cpu", "%i", sample->cpu); 180 else if (al.thread->cpu >= 0) 181 output_json_key_format(out, true, 3, "cpu", "%i", al.thread->cpu); 182 183 output_json_key_string(out, true, 3, "comm", thread__comm_str(al.thread)); 184 185 output_json_key_format(out, true, 3, "callchain", "["); 186 if (sample->callchain) { 187 unsigned int i; 188 bool ok; 189 bool first_callchain = true; 190 191 for (i = 0; i < sample->callchain->nr; ++i) { 192 u64 ip = sample->callchain->ips[i]; 193 194 if (ip >= PERF_CONTEXT_MAX) { 195 switch (ip) { 196 case PERF_CONTEXT_HV: 197 cpumode = PERF_RECORD_MISC_HYPERVISOR; 198 break; 199 case PERF_CONTEXT_KERNEL: 200 cpumode = PERF_RECORD_MISC_KERNEL; 201 break; 202 case PERF_CONTEXT_USER: 203 cpumode = PERF_RECORD_MISC_USER; 204 break; 205 default: 206 pr_debug("invalid callchain context: %" 207 PRId64 "\n", (s64) ip); 208 break; 209 } 210 continue; 211 } 212 213 if (first_callchain) 214 first_callchain = false; 215 else 216 fputc(',', out); 217 218 ok = thread__find_symbol(al.thread, cpumode, ip, &tal); 219 output_sample_callchain_entry(tool, ip, ok ? &tal : NULL); 220 } 221 } else { 222 output_sample_callchain_entry(tool, sample->ip, &al); 223 } 224 output_json_format(out, false, 3, "]"); 225 226 #ifdef HAVE_LIBTRACEEVENT 227 if (sample->raw_data) { 228 int i; 229 struct tep_format_field **fields; 230 231 fields = tep_event_fields(evsel->tp_format); 232 if (fields) { 233 i = 0; 234 while (fields[i]) { 235 struct trace_seq s; 236 237 trace_seq_init(&s); 238 tep_print_field(&s, sample->raw_data, fields[i]); 239 output_json_key_string(out, true, 3, fields[i]->name, s.buffer); 240 241 i++; 242 } 243 free(fields); 244 } 245 } 246 #endif 247 output_json_format(out, false, 2, "}"); 248 return 0; 249 } 250 251 static void output_headers(struct perf_session *session, struct convert_json *c) 252 { 253 struct stat st; 254 struct perf_header *header = &session->header; 255 int ret; 256 int fd = perf_data__fd(session->data); 257 int i; 258 FILE *out = c->out; 259 260 output_json_key_format(out, false, 2, "header-version", "%u", header->version); 261 262 ret = fstat(fd, &st); 263 if (ret >= 0) { 264 time_t stctime = st.st_mtime; 265 char buf[256]; 266 267 strftime(buf, sizeof(buf), "%FT%TZ", gmtime(&stctime)); 268 output_json_key_string(out, true, 2, "captured-on", buf); 269 } else { 270 pr_debug("Failed to get mtime of source file, not writing captured-on"); 271 } 272 273 output_json_key_format(out, true, 2, "data-offset", "%" PRIu64, header->data_offset); 274 output_json_key_format(out, true, 2, "data-size", "%" PRIu64, header->data_size); 275 output_json_key_format(out, true, 2, "feat-offset", "%" PRIu64, header->feat_offset); 276 277 output_json_key_string(out, true, 2, "hostname", header->env.hostname); 278 output_json_key_string(out, true, 2, "os-release", header->env.os_release); 279 output_json_key_string(out, true, 2, "arch", header->env.arch); 280 281 output_json_key_string(out, true, 2, "cpu-desc", header->env.cpu_desc); 282 output_json_key_string(out, true, 2, "cpuid", header->env.cpuid); 283 output_json_key_format(out, true, 2, "nrcpus-online", "%u", header->env.nr_cpus_online); 284 output_json_key_format(out, true, 2, "nrcpus-avail", "%u", header->env.nr_cpus_avail); 285 286 if (header->env.clock.enabled) { 287 output_json_key_format(out, true, 2, "clockid", 288 "%u", header->env.clock.clockid); 289 output_json_key_format(out, true, 2, "clock-time", 290 "%" PRIu64, header->env.clock.clockid_ns); 291 output_json_key_format(out, true, 2, "real-time", 292 "%" PRIu64, header->env.clock.tod_ns); 293 } 294 295 output_json_key_string(out, true, 2, "perf-version", header->env.version); 296 297 output_json_key_format(out, true, 2, "cmdline", "["); 298 for (i = 0; i < header->env.nr_cmdline; i++) { 299 output_json_delimiters(out, i != 0, 3); 300 output_json_string(c->out, header->env.cmdline_argv[i]); 301 } 302 output_json_format(out, false, 2, "]"); 303 } 304 305 int bt_convert__perf2json(const char *input_name, const char *output_name, 306 struct perf_data_convert_opts *opts __maybe_unused) 307 { 308 struct perf_session *session; 309 int fd; 310 int ret = -1; 311 312 struct convert_json c = { 313 .tool = { 314 .sample = process_sample_event, 315 .mmap = perf_event__process_mmap, 316 .mmap2 = perf_event__process_mmap2, 317 .comm = perf_event__process_comm, 318 .namespaces = perf_event__process_namespaces, 319 .cgroup = perf_event__process_cgroup, 320 .exit = perf_event__process_exit, 321 .fork = perf_event__process_fork, 322 .lost = perf_event__process_lost, 323 #ifdef HAVE_LIBTRACEEVENT 324 .tracing_data = perf_event__process_tracing_data, 325 #endif 326 .build_id = perf_event__process_build_id, 327 .id_index = perf_event__process_id_index, 328 .auxtrace_info = perf_event__process_auxtrace_info, 329 .auxtrace = perf_event__process_auxtrace, 330 .event_update = perf_event__process_event_update, 331 .ordered_events = true, 332 .ordering_requires_timestamps = true, 333 }, 334 .first = true, 335 .events_count = 0, 336 }; 337 338 struct perf_data data = { 339 .mode = PERF_DATA_MODE_READ, 340 .path = input_name, 341 .force = opts->force, 342 }; 343 344 if (opts->all) { 345 pr_err("--all is currently unsupported for JSON output.\n"); 346 goto err; 347 } 348 if (opts->tod) { 349 pr_err("--tod is currently unsupported for JSON output.\n"); 350 goto err; 351 } 352 353 fd = open(output_name, O_CREAT | O_WRONLY | (opts->force ? O_TRUNC : O_EXCL), 0666); 354 if (fd == -1) { 355 if (errno == EEXIST) 356 pr_err("Output file exists. Use --force to overwrite it.\n"); 357 else 358 pr_err("Error opening output file!\n"); 359 goto err; 360 } 361 362 c.out = fdopen(fd, "w"); 363 if (!c.out) { 364 fprintf(stderr, "Error opening output file!\n"); 365 close(fd); 366 goto err; 367 } 368 369 session = perf_session__new(&data, &c.tool); 370 if (IS_ERR(session)) { 371 fprintf(stderr, "Error creating perf session!\n"); 372 goto err_fclose; 373 } 374 375 if (symbol__init(&session->header.env) < 0) { 376 fprintf(stderr, "Symbol init error!\n"); 377 goto err_session_delete; 378 } 379 380 // The opening brace is printed manually because it isn't delimited from a 381 // previous value (i.e. we don't want a leading newline) 382 fputc('{', c.out); 383 384 // Version number for future-proofing. Most additions should be able to be 385 // done in a backwards-compatible way so this should only need to be bumped 386 // if some major breaking change must be made. 387 output_json_format(c.out, false, 1, "\"linux-perf-json-version\": 1"); 388 389 // Output headers 390 output_json_format(c.out, true, 1, "\"headers\": {"); 391 output_headers(session, &c); 392 output_json_format(c.out, false, 1, "}"); 393 394 // Output samples 395 output_json_format(c.out, true, 1, "\"samples\": ["); 396 perf_session__process_events(session); 397 output_json_format(c.out, false, 1, "]"); 398 output_json_format(c.out, false, 0, "}"); 399 fputc('\n', c.out); 400 401 fprintf(stderr, 402 "[ perf data convert: Converted '%s' into JSON data '%s' ]\n", 403 data.path, output_name); 404 405 fprintf(stderr, 406 "[ perf data convert: Converted and wrote %.3f MB (%" PRIu64 " samples) ]\n", 407 (ftell(c.out)) / 1024.0 / 1024.0, c.events_count); 408 409 ret = 0; 410 err_session_delete: 411 perf_session__delete(session); 412 err_fclose: 413 fclose(c.out); 414 err: 415 return ret; 416 } 417