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