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