src/bootchart/bootchart.c

   1 /*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
   2
   3 /***
   4   This file is part of systemd.
   5
   6   Copyright (C) 2009-2013 Intel Corporation
   7
   8   Authors:
   9     Auke Kok <auke-jan.h.kok@intel.com>
  10
  11   systemd is free software; you can redistribute it and/or modify it
  12   under the terms of the GNU Lesser General Public License as published by
  13   the Free Software Foundation; either version 2.1 of the License, or
  14   (at your option) any later version.
  15
  16   systemd is distributed in the hope that it will be useful, but
  17   WITHOUT ANY WARRANTY; without even the implied warranty of
  18   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  19   Lesser General Public License for more details.
  20
  21   You should have received a copy of the GNU Lesser General Public License
  22   along with systemd; If not, see <http://www.gnu.org/licenses/>.
  23  ***/
  24
  25 /***
  26
  27   Many thanks to those who contributed ideas and code:
  28   - Ziga Mahkovec - Original bootchart author
  29   - Anders Norgaard - PyBootchartgui
  30   - Michael Meeks - bootchart2
  31   - Scott James Remnant - Ubuntu C-based logger
  32   - Arjan van der Ven - for the idea to merge bootgraph.pl functionality
  33
  34  ***/
  35
  36 #include <sys/resource.h>
  37 #include <stdio.h>
  38 #include <signal.h>
  39 #include <stdlib.h>
  40 #include <string.h>
  41 #include <unistd.h>
  42 #include <time.h>
  43 #include <getopt.h>
  44 #include <limits.h>
  45 #include <errno.h>
  46 #include <fcntl.h>
  47 #include <stdbool.h>
  48 #include "systemd/sd-journal.h"
  49
  50 #include "util.h"
  51 #include "fileio.h"
  52 #include "macro.h"
  53 #include "conf-parser.h"
  54 #include "strxcpyx.h"
  55 #include "path-util.h"
  56 #include "store.h"
  57 #include "svg.h"
  58 #include "bootchart.h"
  59 #include "list.h"
  60
  61 static int exiting = 0;
  62
  63 #define DEFAULT_SAMPLES_LEN 500
  64 #define DEFAULT_HZ 25.0
  65 #define DEFAULT_SCALE_X 100.0 /* 100px = 1sec */
  66 #define DEFAULT_SCALE_Y 20.0  /* 16px = 1 process bar */
  67 #define DEFAULT_INIT ROOTLIBEXECDIR "/systemd"
  68 #define DEFAULT_OUTPUT "/run/log"
  69
  70 /* graph defaults */
  71 bool arg_entropy = false;
  72 bool arg_initcall = true;
  73 bool arg_relative = false;
  74 bool arg_filter = true;
  75 bool arg_show_cmdline = false;
  76 bool arg_show_cgroup = false;
  77 bool arg_pss = false;
  78 bool arg_percpu = false;
  79 int arg_samples_len = DEFAULT_SAMPLES_LEN; /* we record len+1 (1 start sample) */
  80 double arg_hz = DEFAULT_HZ;
  81 double arg_scale_x = DEFAULT_SCALE_X;
  82 double arg_scale_y = DEFAULT_SCALE_Y;
  83
  84 char arg_init_path[PATH_MAX] = DEFAULT_INIT;
  85 char arg_output_path[PATH_MAX] = DEFAULT_OUTPUT;
  86
  87 static void signal_handler(int sig) {
  88         exiting = 1;
  89 }
  90
  91 #define BOOTCHART_CONF "/etc/systemd/bootchart.conf"
  92
  93 #define BOOTCHART_MAX (16*1024*1024)
  94
  95 static void parse_conf(void) {
  96         char *init = NULL, *output = NULL;
  97         const ConfigTableItem items[] = {
  98                 { "Bootchart", "Samples",          config_parse_int,    0, &arg_samples_len },
  99                 { "Bootchart", "Frequency",        config_parse_double, 0, &arg_hz          },
 100                 { "Bootchart", "Relative",         config_parse_bool,   0, &arg_relative    },
 101                 { "Bootchart", "Filter",           config_parse_bool,   0, &arg_filter      },
 102                 { "Bootchart", "Output",           config_parse_path,   0, &output          },
 103                 { "Bootchart", "Init",             config_parse_path,   0, &init            },
 104                 { "Bootchart", "PlotMemoryUsage",  config_parse_bool,   0, &arg_pss         },
 105                 { "Bootchart", "PlotEntropyGraph", config_parse_bool,   0, &arg_entropy     },
 106                 { "Bootchart", "ScaleX",           config_parse_double, 0, &arg_scale_x     },
 107                 { "Bootchart", "ScaleY",           config_parse_double, 0, &arg_scale_y     },
 108                 { "Bootchart", "ControlGroup",     config_parse_bool,   0, &arg_show_cgroup },
 109                 { "Bootchart", "PerCPU",           config_parse_bool,   0, &arg_percpu      },
 110                 { NULL, NULL, NULL, 0, NULL }
 111         };
 112
 113         config_parse_many(BOOTCHART_CONF,
 114                           CONF_DIRS_NULSTR("systemd/bootchart.conf"),
 115                           NULL, config_item_table_lookup, items, true, NULL);
 116
 117         if (init != NULL)
 118                 strscpy(arg_init_path, sizeof(arg_init_path), init);
 119         if (output != NULL)
 120                 strscpy(arg_output_path, sizeof(arg_output_path), output);
 121 }
 122
 123 static void help(void) {
 124         printf("Usage: %s [OPTIONS]\n\n"
 125                "Options:\n"
 126                "  -r --rel             Record time relative to recording\n"
 127                "  -f --freq=FREQ       Sample frequency [%g]\n"
 128                "  -n --samples=N       Stop sampling at [%d] samples\n"
 129                "  -x --scale-x=N       Scale the graph horizontally [%g] \n"
 130                "  -y --scale-y=N       Scale the graph vertically [%g] \n"
 131                "  -p --pss             Enable PSS graph (CPU intensive)\n"
 132                "  -e --entropy         Enable the entropy_avail graph\n"
 133                "  -o --output=PATH     Path to output files [%s]\n"
 134                "  -i --init=PATH       Path to init executable [%s]\n"
 135                "  -F --no-filter       Disable filtering of unimportant or ephemeral processes\n"
 136                "  -C --cmdline         Display full command lines with arguments\n"
 137                "  -c --control-group   Display process control group\n"
 138                "     --per-cpu         Draw each CPU utilization and wait bar also\n"
 139                "  -h --help            Display this message\n\n"
 140                "See bootchart.conf for more information.\n",
 141                program_invocation_short_name,
 142                DEFAULT_HZ,
 143                DEFAULT_SAMPLES_LEN,
 144                DEFAULT_SCALE_X,
 145                DEFAULT_SCALE_Y,
 146                DEFAULT_OUTPUT,
 147                DEFAULT_INIT);
 148 }
 149
 150 static int parse_argv(int argc, char *argv[]) {
 151
 152         enum {
 153                 ARG_PERCPU = 0x100,
 154         };
 155
 156         static const struct option options[] = {
 157                 {"rel",           no_argument,        NULL,  'r'       },
 158                 {"freq",          required_argument,  NULL,  'f'       },
 159                 {"samples",       required_argument,  NULL,  'n'       },
 160                 {"pss",           no_argument,        NULL,  'p'       },
 161                 {"output",        required_argument,  NULL,  'o'       },
 162                 {"init",          required_argument,  NULL,  'i'       },
 163                 {"no-filter",     no_argument,        NULL,  'F'       },
 164                 {"cmdline",       no_argument,        NULL,  'C'       },
 165                 {"control-group", no_argument,        NULL,  'c'       },
 166                 {"help",          no_argument,        NULL,  'h'       },
 167                 {"scale-x",       required_argument,  NULL,  'x'       },
 168                 {"scale-y",       required_argument,  NULL,  'y'       },
 169                 {"entropy",       no_argument,        NULL,  'e'       },
 170                 {"per-cpu",       no_argument,        NULL,  ARG_PERCPU},
 171                 {}
 172         };
 173         int c, r;
 174
 175         if (getpid() == 1)
 176                 opterr = 0;
 177
 178         while ((c = getopt_long(argc, argv, "erpf:n:o:i:FCchx:y:", options, NULL)) >= 0)
 179                 switch (c) {
 180
 181                 case 'r':
 182                         arg_relative = true;
 183                         break;
 184                 case 'f':
 185                         r = safe_atod(optarg, &arg_hz);
 186                         if (r < 0)
 187                                 log_warning_errno(r, "failed to parse --freq/-f argument '%s': %m",
 188                                                   optarg);
 189                         break;
 190                 case 'F':
 191                         arg_filter = false;
 192                         break;
 193                 case 'C':
 194                         arg_show_cmdline = true;
 195                         break;
 196                 case 'c':
 197                         arg_show_cgroup = true;
 198                         break;
 199                 case 'n':
 200                         r = safe_atoi(optarg, &arg_samples_len);
 201                         if (r < 0)
 202                                 log_warning_errno(r, "failed to parse --samples/-n argument '%s': %m",
 203                                                   optarg);
 204                         break;
 205                 case 'o':
 206                         path_kill_slashes(optarg);
 207                         strscpy(arg_output_path, sizeof(arg_output_path), optarg);
 208                         break;
 209                 case 'i':
 210                         path_kill_slashes(optarg);
 211                         strscpy(arg_init_path, sizeof(arg_init_path), optarg);
 212                         break;
 213                 case 'p':
 214                         arg_pss = true;
 215                         break;
 216                 case 'x':
 217                         r = safe_atod(optarg, &arg_scale_x);
 218                         if (r < 0)
 219                                 log_warning_errno(r, "failed to parse --scale-x/-x argument '%s': %m",
 220                                                   optarg);
 221                         break;
 222                 case 'y':
 223                         r = safe_atod(optarg, &arg_scale_y);
 224                         if (r < 0)
 225                                 log_warning_errno(r, "failed to parse --scale-y/-y argument '%s': %m",
 226                                                   optarg);
 227                         break;
 228                 case 'e':
 229                         arg_entropy = true;
 230                         break;
 231                 case ARG_PERCPU:
 232                         arg_percpu = true;
 233                         break;
 234                 case 'h':
 235                         help();
 236                         return 0;
 237                 case '?':
 238                         if (getpid() != 1)
 239                                 return -EINVAL;
 240                         else
 241                                 return 0;
 242                 default:
 243                         assert_not_reached("Unhandled option code.");
 244                 }
 245
 246         if (arg_hz <= 0) {
 247                 log_error("Frequency needs to be > 0");
 248                 return -EINVAL;
 249         }
 250
 251         return 1;
 252 }
 253
 254 static int do_journal_append(char *file) {
 255         _cleanup_free_ char *bootchart_message = NULL;
 256         _cleanup_free_ char *bootchart_file = NULL;
 257         _cleanup_free_ char *p = NULL;
 258         _cleanup_close_ int fd = -1;
 259         struct iovec iovec[5];
 260         int r, j = 0;
 261         ssize_t n;
 262
 263         bootchart_file = strappend("BOOTCHART_FILE=", file);
 264         if (!bootchart_file)
 265                 return log_oom();
 266
 267         IOVEC_SET_STRING(iovec[j++], bootchart_file);
 268         IOVEC_SET_STRING(iovec[j++], "MESSAGE_ID=9f26aa562cf440c2b16c773d0479b518");
 269         IOVEC_SET_STRING(iovec[j++], "PRIORITY=7");
 270         bootchart_message = strjoin("MESSAGE=Bootchart created: ", file, NULL);
 271         if (!bootchart_message)
 272                 return log_oom();
 273
 274         IOVEC_SET_STRING(iovec[j++], bootchart_message);
 275
 276         p = malloc(10 + BOOTCHART_MAX);
 277         if (!p)
 278                 return log_oom();
 279
 280         memcpy(p, "BOOTCHART=", 10);
 281
 282         fd = open(file, O_RDONLY|O_CLOEXEC);
 283         if (fd < 0)
 284                 return log_error_errno(errno, "Failed to open bootchart data \"%s\": %m", file);
 285
 286         n = loop_read(fd, p + 10, BOOTCHART_MAX, false);
 287         if (n < 0)
 288                 return log_error_errno(n, "Failed to read bootchart data: %m");
 289
 290         iovec[j].iov_base = p;
 291         iovec[j].iov_len = 10 + n;
 292         j++;
 293
 294         r = sd_journal_sendv(iovec, j);
 295         if (r < 0)
 296                 log_error_errno(r, "Failed to send bootchart: %m");
 297
 298         return 0;
 299 }
 300
 301 int main(int argc, char *argv[]) {
 302         static struct list_sample_data *sampledata;
 303         _cleanup_closedir_ DIR *proc = NULL;
 304         _cleanup_free_ char *build = NULL;
 305         _cleanup_fclose_ FILE *of = NULL;
 306         _cleanup_close_ int sysfd = -1;
 307         struct ps_struct *ps_first;
 308         double graph_start;
 309         double log_start;
 310         double interval;
 311         char output_file[PATH_MAX];
 312         char datestr[200];
 313         int pscount = 0;
 314         int n_cpus = 0;
 315         int overrun = 0;
 316         time_t t = 0;
 317         int r, samples;
 318         struct ps_struct *ps;
 319         struct rlimit rlim;
 320         struct list_sample_data *head;
 321         struct sigaction sig = {
 322                 .sa_handler = signal_handler,
 323         };
 324
 325         parse_conf();
 326
 327         r = parse_argv(argc, argv);
 328         if (r < 0)
 329                 return EXIT_FAILURE;
 330
 331         if (r == 0)
 332                 return EXIT_SUCCESS;
 333
 334         /*
 335          * If the kernel executed us through init=/usr/lib/systemd/systemd-bootchart, then
 336          * fork:
 337          * - parent execs executable specified via init_path[] (/usr/lib/systemd/systemd by default) as pid=1
 338          * - child logs data
 339          */
 340         if (getpid() == 1) {
 341                 if (fork())
 342                         /* parent */
 343                         execl(arg_init_path, arg_init_path, NULL);
 344         }
 345         argv[0][0] = '@';
 346
 347         rlim.rlim_cur = 4096;
 348         rlim.rlim_max = 4096;
 349         (void) setrlimit(RLIMIT_NOFILE, &rlim);
 350
 351         /* start with empty ps LL */
 352         ps_first = new0(struct ps_struct, 1);
 353         if (!ps_first) {
 354                 log_oom();
 355                 return EXIT_FAILURE;
 356         }
 357
 358         /* handle TERM/INT nicely */
 359         sigaction(SIGHUP, &sig, NULL);
 360
 361         interval = (1.0 / arg_hz) * 1000000000.0;
 362
 363         if (arg_relative)
 364                 graph_start = log_start = gettime_ns();
 365         else {
 366                 struct timespec n;
 367                 double uptime;
 368
 369                 clock_gettime(clock_boottime_or_monotonic(), &n);
 370                 uptime = (n.tv_sec + (n.tv_nsec / (double) NSEC_PER_SEC));
 371
 372                 log_start = gettime_ns();
 373                 graph_start = log_start - uptime;
 374         }
 375
 376         if (graph_start < 0.0) {
 377                 log_error("Failed to setup graph start time.\n\n"
 378                           "The system uptime probably includes time that the system was suspended. "
 379                           "Use --rel to bypass this issue.");
 380                 return EXIT_FAILURE;
 381         }
 382
 383         LIST_HEAD_INIT(head);
 384
 385         /* main program loop */
 386         for (samples = 0; !exiting && samples < arg_samples_len; samples++) {
 387                 int res;
 388                 double sample_stop;
 389                 double elapsed;
 390                 double timeleft;
 391
 392                 sampledata = new0(struct list_sample_data, 1);
 393                 if (sampledata == NULL) {
 394                         log_oom();
 395                         return EXIT_FAILURE;
 396                 }
 397
 398                 sampledata->sampletime = gettime_ns();
 399                 sampledata->counter = samples;
 400
 401                 if (sysfd < 0)
 402                         sysfd = open("/sys", O_RDONLY|O_CLOEXEC);
 403
 404                 if (!build) {
 405                         if (parse_env_file("/etc/os-release", NEWLINE, "PRETTY_NAME", &build, NULL) == -ENOENT)
 406                                 parse_env_file("/usr/lib/os-release", NEWLINE, "PRETTY_NAME", &build, NULL);
 407                 }
 408
 409                 if (proc)
 410                         rewinddir(proc);
 411                 else
 412                         proc = opendir("/proc");
 413
 414                 /* wait for /proc to become available, discarding samples */
 415                 if (proc) {
 416                         r = log_sample(proc, samples, ps_first, &sampledata, &pscount, &n_cpus);
 417                         if (r < 0)
 418                                 return EXIT_FAILURE;
 419                 }
 420
 421                 sample_stop = gettime_ns();
 422
 423                 elapsed = (sample_stop - sampledata->sampletime) * 1000000000.0;
 424                 timeleft = interval - elapsed;
 425
 426                 /*
 427                  * check if we have not consumed our entire timeslice. If we
 428                  * do, don't sleep and take a new sample right away.
 429                  * we'll lose all the missed samples and overrun our total
 430                  * time
 431                  */
 432                 if (timeleft > 0) {
 433                         struct timespec req;
 434
 435                         req.tv_sec = (time_t)(timeleft / 1000000000.0);
 436                         req.tv_nsec = (long)(timeleft - (req.tv_sec * 1000000000.0));
 437
 438                         res = nanosleep(&req, NULL);
 439                         if (res) {
 440                                 if (errno == EINTR)
 441                                         /* caught signal, probably HUP! */
 442                                         break;
 443                                 log_error_errno(errno, "nanosleep() failed: %m");
 444                                 return EXIT_FAILURE;
 445                         }
 446                 } else {
 447                         overrun++;
 448                         /* calculate how many samples we lost and scrap them */
 449                         arg_samples_len -= (int)(-timeleft / interval);
 450                 }
 451                 LIST_PREPEND(link, head, sampledata);
 452         }
 453
 454         /* do some cleanup, close fd's */
 455         ps = ps_first;
 456         while (ps->next_ps) {
 457                 ps = ps->next_ps;
 458                 ps->schedstat = safe_close(ps->schedstat);
 459                 ps->sched = safe_close(ps->sched);
 460                 ps->smaps = safe_fclose(ps->smaps);
 461         }
 462
 463         if (!of) {
 464                 t = time(NULL);
 465                 r = strftime(datestr, sizeof(datestr), "%Y%m%d-%H%M", localtime(&t));
 466                 assert_se(r > 0);
 467
 468                 snprintf(output_file, PATH_MAX, "%s/bootchart-%s.svg", arg_output_path, datestr);
 469                 of = fopen(output_file, "we");
 470         }
 471
 472         if (!of) {
 473                 log_error("Error opening output file '%s': %m\n", output_file);
 474                 return EXIT_FAILURE;
 475         }
 476
 477         r = svg_do(of, strna(build), head, ps_first,
 478                    samples, pscount, n_cpus, graph_start,
 479                    log_start, interval, overrun);
 480
 481         if (r < 0) {
 482                 log_error_errno(r, "Error generating svg file: %m");
 483                 return EXIT_FAILURE;
 484         }
 485
 486         log_info("systemd-bootchart wrote %s\n", output_file);
 487
 488         r = do_journal_append(output_file);
 489         if (r < 0)
 490                 return EXIT_FAILURE;
 491
 492         /* nitpic cleanups */
 493         ps = ps_first->next_ps;
 494         while (ps->next_ps) {
 495                 struct ps_struct *old;
 496
 497                 old = ps;
 498                 old->sample = ps->first;
 499                 ps = ps->next_ps;
 500                 while (old->sample->next) {
 501                         struct ps_sched_struct *oldsample = old->sample;
 502
 503                         old->sample = old->sample->next;
 504                         free(oldsample);
 505                 }
 506                 free(old->cgroup);
 507                 free(old->sample);
 508                 free(old);
 509         }
 510
 511         free(ps->cgroup);
 512         free(ps->sample);
 513         free(ps);
 514
 515         sampledata = head;
 516         while (sampledata->link_prev) {
 517                 struct list_sample_data *old_sampledata = sampledata;
 518                 sampledata = sampledata->link_prev;
 519                 free(old_sampledata);
 520         }
 521         free(sampledata);
 522
 523         /* don't complain when overrun once, happens most commonly on 1st sample */
 524         if (overrun > 1)
 525                 log_warning("systemd-bootchart: sample time overrun %i times\n", overrun);
 526
 527         return 0;
 528 }