src/processes.c

   1 /**
   2  * collectd - src/processes.c
   3  * Copyright (C) 2005  Lyonel Vincent
   4  * Copyright (C) 2006  Florian Forster (Mach code)
   5  *
   6  * This program is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License as published by the
   8  * Free Software Foundation; either version 2 of the License, or (at your
   9  * option) any later version.
  10  *
  11  * This program is distributed in the hope that it will be useful, but
  12  * WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  14  * General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU General Public License along
  17  * with this program; if not, write to the Free Software Foundation, Inc.,
  18  * 51 Franklin St, Fifth Floor, Boston, MA  02110-1301 USA
  19  *
  20  * Authors:
  21  *   Lyonel Vincent <lyonel at ezix.org>
  22  *   Florian octo Forster <octo at verplant.org>
  23  **/
  24
  25 #include "collectd.h"
  26 #include "common.h"
  27 #include "plugin.h"
  28 #include "utils_debug.h"
  29 #include "configfile.h"
  30
  31 /* Include header files for the mach system, if they exist.. */
  32 #if HAVE_THREAD_INFO
  33 #  if HAVE_MACH_MACH_INIT_H
  34 #    include <mach/mach_init.h>
  35 #  endif
  36 #  if HAVE_MACH_HOST_PRIV_H
  37 #    include <mach/host_priv.h>
  38 #  endif
  39 #  if HAVE_MACH_MACH_ERROR_H
  40 #    include <mach/mach_error.h>
  41 #  endif
  42 #  if HAVE_MACH_MACH_HOST_H
  43 #    include <mach/mach_host.h>
  44 #  endif
  45 #  if HAVE_MACH_MACH_PORT_H
  46 #    include <mach/mach_port.h>
  47 #  endif
  48 #  if HAVE_MACH_MACH_TYPES_H
  49 #    include <mach/mach_types.h>
  50 #  endif
  51 #  if HAVE_MACH_MESSAGE_H
  52 #    include <mach/message.h>
  53 #  endif
  54 #  if HAVE_MACH_PROCESSOR_SET_H
  55 #    include <mach/processor_set.h>
  56 #  endif
  57 #  if HAVE_MACH_TASK_H
  58 #    include <mach/task.h>
  59 #  endif
  60 #  if HAVE_MACH_THREAD_ACT_H
  61 #    include <mach/thread_act.h>
  62 #  endif
  63 #  if HAVE_MACH_VM_REGION_H
  64 #    include <mach/vm_region.h>
  65 #  endif
  66 #  if HAVE_MACH_VM_MAP_H
  67 #    include <mach/vm_map.h>
  68 #  endif
  69 #  if HAVE_MACH_VM_PROT_H
  70 #    include <mach/vm_prot.h>
  71 #  endif
  72 #  if HAVE_SYS_SYSCTL_H
  73 #    include <sys/sysctl.h>
  74 #  endif
  75 /* #endif HAVE_THREAD_INFO */
  76
  77 #elif KERNEL_LINUX
  78 #  if HAVE_LINUX_CONFIG_H
  79 #    include <linux/config.h>
  80 #  endif
  81 #  ifndef CONFIG_HZ
  82 #    define CONFIG_HZ 100
  83 #  endif
  84 #endif /* KERNEL_LINUX */
  85
  86 #define MODULE_NAME "processes"
  87
  88 #if HAVE_THREAD_INFO || KERNEL_LINUX
  89 # define PROCESSES_HAVE_READ 1
  90 #else
  91 # define PROCESSES_HAVE_READ 0
  92 #endif
  93
  94 #define BUFSIZE 256
  95
  96 static char *processes_file = "processes.rrd";
  97 static char *processes_ds_def[] =
  98 {
  99         "DS:running:GAUGE:"COLLECTD_HEARTBEAT":0:65535",
 100         "DS:sleeping:GAUGE:"COLLECTD_HEARTBEAT":0:65535",
 101         "DS:zombies:GAUGE:"COLLECTD_HEARTBEAT":0:65535",
 102         "DS:stopped:GAUGE:"COLLECTD_HEARTBEAT":0:65535",
 103         "DS:paging:GAUGE:"COLLECTD_HEARTBEAT":0:65535",
 104         "DS:blocked:GAUGE:"COLLECTD_HEARTBEAT":0:65535",
 105         NULL
 106 };
 107 static int processes_ds_num = 6;
 108
 109 static char *ps_rss_file = "processes/ps_rss-%s.rrd";
 110 static char *ps_rss_ds_def[] =
 111 {
 112         /* max = 2^63 - 1 */
 113         "DS:byte:GAUGE:"COLLECTD_HEARTBEAT":0:9223372036854775807",
 114         NULL
 115 };
 116 static int ps_rss_ds_num = 1;
 117
 118 static char *ps_cputime_file = "processes/ps_cputime-%s.rrd";
 119 static char *ps_cputime_ds_def[] =
 120 {
 121         /* 1 second in user-mode per second ought to be enough.. */
 122         "DS:user:COUNTER:"COLLECTD_HEARTBEAT":0:1000000",
 123         "DS:syst:COUNTER:"COLLECTD_HEARTBEAT":0:1000000",
 124         NULL
 125 };
 126 static int ps_cputime_ds_num = 2;
 127
 128 static char *ps_count_file = "processes/ps_count-%s.rrd";
 129 static char *ps_count_ds_def[] =
 130 {
 131         "DS:processes:GAUGE:"COLLECTD_HEARTBEAT":0:65535",
 132         "DS:threads:GAUGE:"COLLECTD_HEARTBEAT":0:65535",
 133         NULL
 134 };
 135 static int ps_count_ds_num = 2;
 136
 137 static char *ps_pagefaults_file = "processes/ps_pagefaults-%s.rrd";
 138 static char *ps_pagefaults_ds_def[] =
 139 {
 140         /* max = 2^63 - 1 */
 141         "DS:minflt:COUNTER:"COLLECTD_HEARTBEAT":0:9223372036854775807",
 142         "DS:majflt:COUNTER:"COLLECTD_HEARTBEAT":0:9223372036854775807",
 143         NULL
 144 };
 145 static int ps_pagefaults_ds_num = 2;
 146
 147 #if HAVE_THREAD_INFO | KERNEL_LINUX
 148 static char *config_keys[] =
 149 {
 150         "Process",
 151         NULL
 152 };
 153 static int config_keys_num = 1;
 154 #endif
 155
 156 typedef struct procstat_entry_s
 157 {
 158         unsigned long id;
 159         unsigned long age;
 160
 161         unsigned long num_proc;
 162         unsigned long num_lwp;
 163         unsigned long vmem_rss;
 164
 165         unsigned long vmem_minflt;
 166         unsigned long vmem_majflt;
 167         unsigned long vmem_minflt_counter;
 168         unsigned long vmem_majflt_counter;
 169
 170         unsigned long cpu_user;
 171         unsigned long cpu_system;
 172         unsigned long cpu_user_counter;
 173         unsigned long cpu_system_counter;
 174
 175         struct procstat_entry_s *next;
 176 } procstat_entry_t;
 177
 178 #define PROCSTAT_NAME_LEN 256
 179 typedef struct procstat
 180 {
 181         char          name[PROCSTAT_NAME_LEN];
 182
 183         unsigned long num_proc;
 184         unsigned long num_lwp;
 185         unsigned long vmem_rss;
 186
 187         unsigned long vmem_minflt_counter;
 188         unsigned long vmem_majflt_counter;
 189
 190         unsigned long cpu_user_counter;
 191         unsigned long cpu_system_counter;
 192
 193         struct procstat   *next;
 194         struct procstat_entry_s *instances;
 195 } procstat_t;
 196
 197 #if HAVE_THREAD_INFO | KERNEL_LINUX
 198 static procstat_t *list_head_g = NULL;
 199 #endif
 200
 201 #if HAVE_THREAD_INFO
 202 static mach_port_t port_host_self;
 203 static mach_port_t port_task_self;
 204
 205 static processor_set_name_array_t pset_list;
 206 static mach_msg_type_number_t     pset_list_len;
 207 /* #endif HAVE_THREAD_INFO */
 208
 209 #elif KERNEL_LINUX
 210 static long pagesize_g;
 211 #endif /* KERNEL_LINUX */
 212
 213 #if HAVE_THREAD_INFO | KERNEL_LINUX
 214 static void ps_list_register (const char *name)
 215 {
 216         procstat_t *new;
 217         procstat_t *ptr;
 218
 219         if ((new = (procstat_t *) malloc (sizeof (procstat_t))) == NULL)
 220                 return;
 221         memset (new, 0, sizeof (procstat_t));
 222         strncpy (new->name, name, PROCSTAT_NAME_LEN);
 223
 224         for (ptr = list_head_g; ptr != NULL; ptr = ptr->next)
 225         {
 226                 if (strcmp (ptr->name, name) == 0)
 227                         return;
 228                 if (ptr->next == NULL)
 229                         break;
 230         }
 231
 232         if (ptr == NULL)
 233                 list_head_g = new;
 234         else
 235                 ptr->next = new;
 236 }
 237
 238 static procstat_t *ps_list_search (const char *name)
 239 {
 240         procstat_t *ptr;
 241
 242         for (ptr = list_head_g; ptr != NULL; ptr = ptr->next)
 243                 if (strcmp (ptr->name, name) == 0)
 244                         break;
 245
 246         return (ptr);
 247 }
 248
 249 static void ps_list_add (const char *name, procstat_entry_t *entry)
 250 {
 251         procstat_t *ps;
 252         procstat_entry_t *pse;
 253
 254         if (entry->id == 0)
 255                 return;
 256
 257         if ((ps = ps_list_search (name)) == NULL)
 258                 return;
 259
 260         for (pse = ps->instances; pse != NULL; pse = pse->next)
 261                 if ((pse->id == entry->id) || (pse->next == NULL))
 262                         break;
 263
 264         if ((pse == NULL) || (pse->id != entry->id))
 265         {
 266                 procstat_entry_t *new;
 267
 268                 new = (procstat_entry_t *) malloc (sizeof (procstat_entry_t));
 269                 if (new == NULL)
 270                         return;
 271                 memset (new, 0, sizeof (procstat_entry_t));
 272                 new->id = entry->id;
 273
 274                 if (pse == NULL)
 275                         ps->instances = new;
 276                 else
 277                         pse->next = new;
 278
 279                 pse = new;
 280         }
 281
 282         pse->age = 0;
 283         pse->num_proc = entry->num_proc;
 284         pse->num_lwp  = entry->num_lwp;
 285         pse->vmem_rss = entry->vmem_rss;
 286
 287         ps->num_proc += pse->num_proc;
 288         ps->num_lwp  += pse->num_lwp;
 289         ps->vmem_rss += pse->vmem_rss;
 290
 291         if ((entry->vmem_minflt_counter == 0)
 292                         && (entry->vmem_majflt_counter == 0))
 293         {
 294                 pse->vmem_minflt_counter += entry->vmem_minflt;
 295                 pse->vmem_minflt = entry->vmem_minflt;
 296
 297                 pse->vmem_majflt_counter += entry->vmem_majflt;
 298                 pse->vmem_majflt = entry->vmem_majflt;
 299         }
 300         else
 301         {
 302                 if (entry->vmem_minflt_counter < pse->vmem_minflt_counter)
 303                 {
 304                         pse->vmem_minflt = entry->vmem_minflt_counter
 305                                 + (ULONG_MAX - pse->vmem_minflt_counter);
 306                 }
 307                 else
 308                 {
 309                         pse->vmem_minflt = entry->vmem_minflt_counter - pse->vmem_minflt_counter;
 310                 }
 311                 pse->vmem_minflt_counter = entry->vmem_minflt_counter;
 312
 313                 if (entry->vmem_majflt_counter < pse->vmem_majflt_counter)
 314                 {
 315                         pse->vmem_majflt = entry->vmem_majflt_counter
 316                                 + (ULONG_MAX - pse->vmem_majflt_counter);
 317                 }
 318                 else
 319                 {
 320                         pse->vmem_majflt = entry->vmem_majflt_counter - pse->vmem_majflt_counter;
 321                 }
 322                 pse->vmem_majflt_counter = entry->vmem_majflt_counter;
 323         }
 324
 325         ps->vmem_minflt_counter += pse->vmem_minflt;
 326         ps->vmem_majflt_counter += pse->vmem_majflt;
 327
 328         if ((entry->cpu_user_counter == 0)
 329                         && (entry->cpu_system_counter == 0))
 330         {
 331                 pse->cpu_user_counter += entry->cpu_user;
 332                 pse->cpu_user = entry->cpu_user;
 333
 334                 pse->cpu_system_counter += entry->cpu_system;
 335                 pse->cpu_system = entry->cpu_system;
 336         }
 337         else
 338         {
 339                 if (entry->cpu_user_counter < pse->cpu_user_counter)
 340                 {
 341                         pse->cpu_user = entry->cpu_user_counter
 342                                 + (ULONG_MAX - pse->cpu_user_counter);
 343                 }
 344                 else
 345                 {
 346                         pse->cpu_user = entry->cpu_user_counter - pse->cpu_user_counter;
 347                 }
 348                 pse->cpu_user_counter = entry->cpu_user_counter;
 349
 350                 if (entry->cpu_system_counter < pse->cpu_system_counter)
 351                 {
 352                         pse->cpu_system = entry->cpu_system_counter
 353                                 + (ULONG_MAX - pse->cpu_system_counter);
 354                 }
 355                 else
 356                 {
 357                         pse->cpu_system = entry->cpu_system_counter - pse->cpu_system_counter;
 358                 }
 359                 pse->cpu_system_counter = entry->cpu_system_counter;
 360         }
 361
 362         ps->cpu_user_counter   += pse->cpu_user;
 363         ps->cpu_system_counter += pse->cpu_system;
 364 }
 365
 366 static void ps_list_reset (void)
 367 {
 368         procstat_t *ps;
 369         procstat_entry_t *pse;
 370         procstat_entry_t *pse_prev;
 371
 372         for (ps = list_head_g; ps != NULL; ps = ps->next)
 373         {
 374                 ps->num_proc    = 0;
 375                 ps->num_lwp     = 0;
 376                 ps->vmem_rss    = 0;
 377
 378                 pse_prev = NULL;
 379                 pse = ps->instances;
 380                 while (pse != NULL)
 381                 {
 382                         if (pse->age > 10)
 383                         {
 384                                 DBG ("Removing this procstat entry cause it's too old: "
 385                                                 "id = %lu; name = %s;",
 386                                                 pse->id, ps->name);
 387
 388                                 if (pse_prev == NULL)
 389                                 {
 390                                         ps->instances = pse->next;
 391                                         free (pse);
 392                                         pse = ps->instances;
 393                                 }
 394                                 else
 395                                 {
 396                                         pse_prev->next = pse->next;
 397                                         free (pse);
 398                                         pse = pse_prev->next;
 399                                 }
 400                         }
 401                         else
 402                         {
 403                                 pse->age++;
 404                                 pse_prev = pse;
 405                                 pse = pse->next;
 406                         }
 407                 } /* while (pse != NULL) */
 408         } /* for (ps = list_head_g; ps != NULL; ps = ps->next) */
 409 }
 410
 411 static int ps_config (char *key, char *value)
 412 {
 413         if (strcasecmp (key, "Process") == 0)
 414         {
 415                 ps_list_register (value);
 416         }
 417         else
 418         {
 419                 return (-1);
 420         }
 421
 422         return (0);
 423 }
 424 #endif /* HAVE_THREAD_INFO | KERNEL_LINUX */
 425
 426 static void ps_init (void)
 427 {
 428 #if HAVE_THREAD_INFO
 429         kern_return_t status;
 430
 431         port_host_self = mach_host_self ();
 432         port_task_self = mach_task_self ();
 433
 434         if (pset_list != NULL)
 435         {
 436                 vm_deallocate (port_task_self,
 437                                 (vm_address_t) pset_list,
 438                                 pset_list_len * sizeof (processor_set_t));
 439                 pset_list = NULL;
 440                 pset_list_len = 0;
 441         }
 442
 443         if ((status = host_processor_sets (port_host_self,
 444                                         &pset_list,
 445                                         &pset_list_len)) != KERN_SUCCESS)
 446         {
 447                 syslog (LOG_ERR, "host_processor_sets failed: %s\n",
 448                                 mach_error_string (status));
 449                 pset_list = NULL;
 450                 pset_list_len = 0;
 451                 return;
 452         }
 453 /* #endif HAVE_THREAD_INFO */
 454
 455 #elif KERNEL_LINUX
 456         pagesize_g = sysconf(_SC_PAGESIZE);
 457         DBG ("pagesize_g = %li; CONFIG_HZ = %i;",
 458                         pagesize_g, CONFIG_HZ);
 459 #endif /* KERNEL_LINUX */
 460
 461         return;
 462 }
 463
 464 static void ps_write (char *host, char *inst, char *val)
 465 {
 466         rrd_update_file (host, processes_file, val,
 467                         processes_ds_def, processes_ds_num);
 468 }
 469
 470 static void ps_rss_write (char *host, char *inst, char *val)
 471 {
 472         char filename[256];
 473         int status;
 474
 475         status = snprintf (filename, 256, ps_rss_file, inst);
 476         if ((status < 1) || (status >= 256))
 477                 return;
 478
 479         rrd_update_file (host, filename, val, ps_rss_ds_def, ps_rss_ds_num);
 480 }
 481
 482 static void ps_cputime_write (char *host, char *inst, char *val)
 483 {
 484         char filename[256];
 485         int status;
 486
 487         status = snprintf (filename, 256, ps_cputime_file, inst);
 488         if ((status < 1) || (status >= 256))
 489                 return;
 490
 491         DBG ("host = %s; filename = %s; val = %s;",
 492                         host, filename, val);
 493         rrd_update_file (host, filename, val,
 494                         ps_cputime_ds_def, ps_cputime_ds_num);
 495 }
 496
 497 static void ps_count_write (char *host, char *inst, char *val)
 498 {
 499         char filename[256];
 500         int status;
 501
 502         status = snprintf (filename, 256, ps_count_file, inst);
 503         if ((status < 1) || (status >= 256))
 504                 return;
 505
 506         DBG ("host = %s; filename = %s; val = %s;",
 507                         host, filename, val);
 508         rrd_update_file (host, filename, val,
 509                         ps_count_ds_def, ps_count_ds_num);
 510 }
 511
 512 static void ps_pagefaults_write (char *host, char *inst, char *val)
 513 {
 514         char filename[256];
 515         int status;
 516
 517         status = snprintf (filename, 256, ps_pagefaults_file, inst);
 518         if ((status < 1) || (status >= 256))
 519                 return;
 520
 521         DBG ("host = %s; filename = %s; val = %s;",
 522                         host, filename, val);
 523         rrd_update_file (host, filename, val,
 524                         ps_pagefaults_ds_def, ps_pagefaults_ds_num);
 525 }
 526
 527 #if PROCESSES_HAVE_READ
 528 static void ps_submit (int running,
 529                 int sleeping,
 530                 int zombies,
 531                 int stopped,
 532                 int paging,
 533                 int blocked)
 534 {
 535         char buf[BUFSIZE];
 536
 537         if (snprintf (buf, BUFSIZE, "%u:%i:%i:%i:%i:%i:%i",
 538                                 (unsigned int) curtime,
 539                                 running, sleeping, zombies, stopped, paging,
 540                                 blocked) >= BUFSIZE)
 541                 return;
 542
 543         DBG ("running = %i; sleeping = %i; zombies = %i; stopped = %i; paging = %i; blocked = %i;",
 544                         running, sleeping, zombies, stopped, paging, blocked);
 545
 546         plugin_submit (MODULE_NAME, "-", buf);
 547 }
 548
 549 static void ps_submit_proc_list (procstat_t *ps)
 550 {
 551         char buffer[64];
 552
 553         if (ps == NULL)
 554                 return;
 555
 556         snprintf (buffer, 64, "%u:%lu",
 557                         (unsigned int) curtime,
 558                         ps->vmem_rss);
 559         buffer[63] = '\0';
 560         plugin_submit ("ps_rss", ps->name, buffer);
 561
 562         snprintf (buffer, 64, "%u:%u:%u",
 563                         (unsigned int) curtime,
 564                         /* Make the counter overflow */
 565                         (unsigned int) (ps->cpu_user_counter   & 0xFFFFFFFF),
 566                         (unsigned int) (ps->cpu_system_counter & 0xFFFFFFFF));
 567         buffer[63] = '\0';
 568         plugin_submit ("ps_cputime", ps->name, buffer);
 569
 570         snprintf (buffer, 64, "%u:%lu:%lu",
 571                         (unsigned int) curtime,
 572                         ps->num_proc, ps->num_lwp);
 573         buffer[63] = '\0';
 574         plugin_submit ("ps_count", ps->name, buffer);
 575
 576         snprintf (buffer, 64, "%u:%lu:%lu",
 577                         (unsigned int) curtime,
 578                         ps->vmem_minflt_counter, ps->vmem_majflt_counter);
 579         buffer[63] = '\0';
 580         plugin_submit ("ps_pagefaults", ps->name, buffer);
 581
 582         DBG ("name = %s; num_proc = %lu; num_lwp = %lu; vmem_rss = %lu; "
 583                         "vmem_minflt_counter = %lu; vmem_majflt_counter = %lu; "
 584                         "cpu_user_counter = %lu; cpu_system_counter = %lu;",
 585                         ps->name, ps->num_proc, ps->num_lwp, ps->vmem_rss,
 586                         ps->vmem_minflt_counter, ps->vmem_majflt_counter, ps->cpu_user_counter,
 587                         ps->cpu_system_counter);
 588
 589 }
 590
 591 #if KERNEL_LINUX
 592 static int *ps_read_tasks (int pid)
 593 {
 594         int *list = NULL;
 595         int  list_size = 1; /* size of allocated space, in elements */
 596         int  list_len = 0;  /* number of currently used elements */
 597
 598         char           dirname[64];
 599         DIR           *dh;
 600         struct dirent *ent;
 601
 602         snprintf (dirname, 64, "/proc/%i/task", pid);
 603         dirname[63] = '\0';
 604
 605         if ((dh = opendir (dirname)) == NULL)
 606         {
 607                 DBG ("Failed to open directory `%s'", dirname);
 608                 return (NULL);
 609         }
 610
 611         while ((ent = readdir (dh)) != NULL)
 612         {
 613                 if (!isdigit (ent->d_name[0]))
 614                         continue;
 615
 616                 if ((list_len + 1) >= list_size)
 617                 {
 618                         int *new_ptr;
 619                         int  new_size = 2 * list_size;
 620                         /* Comes in sizes: 2, 4, 8, 16, ... */
 621
 622                         new_ptr = (int *) realloc (list, (size_t) (sizeof (int) * new_size));
 623                         if (new_ptr == NULL)
 624                         {
 625                                 if (list != NULL)
 626                                         free (list);
 627                                 syslog (LOG_ERR, "processes plugin: "
 628                                                 "Failed to allocate more memory.");
 629                                 return (NULL);
 630                         }
 631
 632                         list = new_ptr;
 633                         list_size = new_size;
 634
 635                         memset (list + list_len, 0, sizeof (int) * (list_size - list_len));
 636                 }
 637
 638                 list[list_len] = atoi (ent->d_name);
 639                 if (list[list_len] != 0)
 640                         list_len++;
 641         }
 642
 643         closedir (dh);
 644
 645         if (list_len == 0)
 646                 return (NULL);
 647
 648         assert (list_len < list_size);
 649         assert (list[list_len] == 0);
 650
 651         return (list);
 652 } /* int *ps_read_tasks */
 653
 654 int ps_read_process (int pid, procstat_t *ps, char *state)
 655 {
 656         char  filename[64];
 657         char  buffer[1024];
 658         FILE *fh;
 659
 660         char *fields[64];
 661         char  fields_len;
 662
 663         int  *tasks;
 664         int   i;
 665
 666         int   ppid;
 667         int   name_len;
 668
 669         long long unsigned cpu_user_counter;
 670         long long unsigned cpu_system_counter;
 671         long long unsigned vmem_rss;
 672
 673         memset (ps, 0, sizeof (procstat_t));
 674
 675         snprintf (filename, 64, "/proc/%i/stat", pid);
 676         filename[63] = '\0';
 677
 678         if ((fh = fopen (filename, "r")) == NULL)
 679                 return (-1);
 680
 681         if (fgets (buffer, 1024, fh) == NULL)
 682         {
 683                 fclose (fh);
 684                 return (-1);
 685         }
 686
 687         fclose (fh);
 688
 689         fields_len = strsplit (buffer, fields, 64);
 690         if (fields_len < 24)
 691         {
 692                 DBG ("`%s' has only %i fields..",
 693                                 filename, fields_len);
 694                 return (-1);
 695         }
 696         else if (fields_len != 41)
 697         {
 698                 DBG ("WARNING: (fields_len = %i) != 41", fields_len);
 699         }
 700
 701         /* copy the name, strip brackets in the process */
 702         name_len = strlen (fields[1]) - 2;
 703         if ((fields[1][0] != '(') || (fields[1][name_len + 1] != ')'))
 704         {
 705                 DBG ("No brackets found in process name: `%s'", fields[1]);
 706                 return (-1);
 707         }
 708         fields[1] = fields[1] + 1;
 709         fields[1][name_len] = '\0';
 710         strncpy (ps->name, fields[1], PROCSTAT_NAME_LEN);
 711
 712         ppid = atoi (fields[3]);
 713
 714         *state = fields[2][0];
 715
 716         if (*state == 'Z')
 717         {
 718                 ps->num_lwp  = 0;
 719                 ps->num_proc = 0;
 720         }
 721         else if ((tasks = ps_read_tasks (pid)) == NULL)
 722         {
 723                 /* Kernel 2.4 or so */
 724                 ps->num_lwp  = 1;
 725                 ps->num_proc = 1;
 726         }
 727         else
 728         {
 729                 ps->num_lwp  = 0;
 730                 ps->num_proc = 1;
 731                 for (i = 0; tasks[i] != 0; i++)
 732                         ps->num_lwp++;
 733
 734                 free (tasks);
 735                 tasks = NULL;
 736         }
 737
 738         /* Leave the rest at zero if this is only a zombi */
 739         if (ps->num_proc == 0)
 740         {
 741                 DBG ("This is only a zombi: pid = %i; name = %s;",
 742                                 pid, ps->name);
 743                 return (0);
 744         }
 745
 746         cpu_user_counter   = atoll (fields[13]);
 747         cpu_system_counter = atoll (fields[14]);
 748         vmem_rss = atoll (fields[23]);
 749         ps->vmem_minflt_counter = atol (fields[9]);
 750         ps->vmem_majflt_counter = atol (fields[11]);
 751
 752         /* Convert jiffies to useconds */
 753         cpu_user_counter   = cpu_user_counter   * 1000000 / CONFIG_HZ;
 754         cpu_system_counter = cpu_system_counter * 1000000 / CONFIG_HZ;
 755         vmem_rss = vmem_rss * pagesize_g;
 756
 757         ps->cpu_user_counter = (unsigned long) cpu_user_counter;
 758         ps->cpu_system_counter = (unsigned long) cpu_system_counter;
 759         ps->vmem_rss = (unsigned long) vmem_rss;
 760
 761         /* success */
 762         return (0);
 763 } /* int ps_read_process (...) */
 764 #endif /* KERNEL_LINUX */
 765
 766 #if HAVE_THREAD_INFO
 767 static int mach_get_task_name (task_t t, int *pid, char *name, size_t name_max_len)
 768 {
 769         int mib[4];
 770
 771         struct kinfo_proc kp;
 772         size_t            kp_size;
 773
 774         mib[0] = CTL_KERN;
 775         mib[1] = KERN_PROC;
 776         mib[2] = KERN_PROC_PID;
 777
 778         if (pid_for_task (t, pid) != KERN_SUCCESS)
 779                 return (-1);
 780         mib[3] = *pid;
 781
 782         kp_size = sizeof (kp);
 783         if (sysctl (mib, 4, &kp, &kp_size, NULL, 0) != 0)
 784                 return (-1);
 785
 786         if (name_max_len > (MAXCOMLEN + 1))
 787                 name_max_len = MAXCOMLEN + 1;
 788
 789         strncpy (name, kp.kp_proc.p_comm, name_max_len - 1);
 790         name[name_max_len - 1] = '\0';
 791
 792         DBG ("pid = %i; name = %s;", *pid, name);
 793
 794         /* We don't do the special handling for `p_comm == "LaunchCFMApp"' as
 795          * `top' does it, because it is a lot of work and only used when
 796          * debugging. -octo */
 797
 798         return (0);
 799 }
 800 #endif /* HAVE_THREAD_INFO */
 801
 802 static void ps_read (void)
 803 {
 804 #if HAVE_THREAD_INFO
 805         kern_return_t            status;
 806
 807         int                      pset;
 808         processor_set_t          port_pset_priv;
 809
 810         int                      task;
 811         task_array_t             task_list;
 812         mach_msg_type_number_t   task_list_len;
 813
 814         int                      task_pid;
 815         char                     task_name[MAXCOMLEN + 1];
 816
 817         int                      thread;
 818         thread_act_array_t       thread_list;
 819         mach_msg_type_number_t   thread_list_len;
 820         thread_basic_info_data_t thread_data;
 821         mach_msg_type_number_t   thread_data_len;
 822
 823         int running  = 0;
 824         int sleeping = 0;
 825         int zombies  = 0;
 826         int stopped  = 0;
 827         int blocked  = 0;
 828
 829         procstat_t *ps;
 830         procstat_entry_t pse;
 831
 832         ps_list_reset ();
 833
 834         /*
 835          * The Mach-concept is a little different from the traditional UNIX
 836          * concept: All the work is done in threads. Threads are contained in
 837          * `tasks'. Therefore, `task status' doesn't make much sense, since
 838          * it's actually a `thread status'.
 839          * Tasks are assigned to sets of processors, so that's where you go to
 840          * get a list.
 841          */
 842         for (pset = 0; pset < pset_list_len; pset++)
 843         {
 844                 if ((status = host_processor_set_priv (port_host_self,
 845                                                 pset_list[pset],
 846                                                 &port_pset_priv)) != KERN_SUCCESS)
 847                 {
 848                         syslog (LOG_ERR, "host_processor_set_priv failed: %s\n",
 849                                         mach_error_string (status));
 850                         continue;
 851                 }
 852
 853                 if ((status = processor_set_tasks (port_pset_priv,
 854                                                 &task_list,
 855                                                 &task_list_len)) != KERN_SUCCESS)
 856                 {
 857                         syslog (LOG_ERR, "processor_set_tasks failed: %s\n",
 858                                         mach_error_string (status));
 859                         mach_port_deallocate (port_task_self, port_pset_priv);
 860                         continue;
 861                 }
 862
 863                 for (task = 0; task < task_list_len; task++)
 864                 {
 865                         ps = NULL;
 866                         if (mach_get_task_name (task_list[task],
 867                                                 &task_pid,
 868                                                 task_name, PROCSTAT_NAME_LEN) == 0)
 869                                 ps = ps_list_search (task_name);
 870
 871                         /* Collect more detailed statistics for this process */
 872                         if (ps != NULL)
 873                         {
 874                                 task_basic_info_data_t        task_basic_info;
 875                                 mach_msg_type_number_t        task_basic_info_len;
 876                                 task_events_info_data_t       task_events_info;
 877                                 mach_msg_type_number_t        task_events_info_len;
 878                                 task_absolutetime_info_data_t task_absolutetime_info;
 879                                 mach_msg_type_number_t        task_absolutetime_info_len;
 880
 881                                 memset (&pse, '\0', sizeof (pse));
 882                                 pse.id = task_pid;
 883
 884                                 task_basic_info_len = TASK_BASIC_INFO_COUNT;
 885                                 status = task_info (task_list[task],
 886                                                 TASK_BASIC_INFO,
 887                                                 (task_info_t) &task_basic_info,
 888                                                 &task_basic_info_len);
 889                                 if (status != KERN_SUCCESS)
 890                                 {
 891                                         syslog (LOG_ERR, "task_info failed: %s",
 892                                                         mach_error_string (status));
 893                                         continue; /* with next thread_list */
 894                                 }
 895
 896                                 task_events_info_len = TASK_EVENTS_INFO_COUNT;
 897                                 status = task_info (task_list[task],
 898                                                 TASK_EVENTS_INFO,
 899                                                 (task_info_t) &task_events_info,
 900                                                 &task_events_info_len);
 901                                 if (status != KERN_SUCCESS)
 902                                 {
 903                                         syslog (LOG_ERR, "task_info failed: %s",
 904                                                         mach_error_string (status));
 905                                         continue; /* with next thread_list */
 906                                 }
 907
 908                                 task_absolutetime_info_len = TASK_ABSOLUTETIME_INFO_COUNT;
 909                                 status = task_info (task_list[task],
 910                                                 TASK_ABSOLUTETIME_INFO,
 911                                                 (task_info_t) &task_absolutetime_info,
 912                                                 &task_absolutetime_info_len);
 913                                 if (status != KERN_SUCCESS)
 914                                 {
 915                                         syslog (LOG_ERR, "task_info failed: %s",
 916                                                         mach_error_string (status));
 917                                         continue; /* with next thread_list */
 918                                 }
 919
 920                                 pse.num_proc++;
 921                                 pse.vmem_rss = task_basic_info.resident_size;
 922
 923                                 pse.vmem_minflt_counter = task_events_info.cow_faults;
 924                                 pse.vmem_majflt_counter = task_events_info.faults;
 925
 926                                 pse.cpu_user_counter = task_absolutetime_info.total_user;
 927                                 pse.cpu_system_counter = task_absolutetime_info.total_system;
 928                         }
 929
 930                         status = task_threads (task_list[task], &thread_list,
 931                                         &thread_list_len);
 932                         if (status != KERN_SUCCESS)
 933                         {
 934                                 /* Apple's `top' treats this case a zombie. It
 935                                  * makes sense to some extend: A `zombie'
 936                                  * thread is nonsense, since the task/process
 937                                  * is dead. */
 938                                 zombies++;
 939                                 DBG ("task_threads failed: %s",
 940                                                 mach_error_string (status));
 941                                 if (task_list[task] != port_task_self)
 942                                         mach_port_deallocate (port_task_self,
 943                                                         task_list[task]);
 944                                 continue; /* with next task_list */
 945                         }
 946
 947                         for (thread = 0; thread < thread_list_len; thread++)
 948                         {
 949                                 thread_data_len = THREAD_BASIC_INFO_COUNT;
 950                                 status = thread_info (thread_list[thread],
 951                                                 THREAD_BASIC_INFO,
 952                                                 (thread_info_t) &thread_data,
 953                                                 &thread_data_len);
 954                                 if (status != KERN_SUCCESS)
 955                                 {
 956                                         syslog (LOG_ERR, "thread_info failed: %s",
 957                                                         mach_error_string (status));
 958                                         if (task_list[task] != port_task_self)
 959                                                 mach_port_deallocate (port_task_self,
 960                                                                 thread_list[thread]);
 961                                         continue; /* with next thread_list */
 962                                 }
 963
 964                                 if (ps != NULL)
 965                                         pse.num_lwp++;
 966
 967                                 switch (thread_data.run_state)
 968                                 {
 969                                         case TH_STATE_RUNNING:
 970                                                 running++;
 971                                                 break;
 972                                         case TH_STATE_STOPPED:
 973                                         /* What exactly is `halted'? */
 974                                         case TH_STATE_HALTED:
 975                                                 stopped++;
 976                                                 break;
 977                                         case TH_STATE_WAITING:
 978                                                 sleeping++;
 979                                                 break;
 980                                         case TH_STATE_UNINTERRUPTIBLE:
 981                                                 blocked++;
 982                                                 break;
 983                                         /* There is no `zombie' case here,
 984                                          * since there are no zombie-threads.
 985                                          * There's only zombie tasks, which are
 986                                          * handled above. */
 987                                         default:
 988                                                 syslog (LOG_WARNING,
 989                                                                 "Unknown thread status: %s",
 990                                                                 thread_data.run_state);
 991                                                 break;
 992                                 } /* switch (thread_data.run_state) */
 993
 994                                 if (task_list[task] != port_task_self)
 995                                 {
 996                                         status = mach_port_deallocate (port_task_self,
 997                                                         thread_list[thread]);
 998                                         if (status != KERN_SUCCESS)
 999                                                 syslog (LOG_ERR, "mach_port_deallocate failed: %s",
1000                                                                 mach_error_string (status));
1001                                 }
1002                         } /* for (thread_list) */
1003
1004                         if ((status = vm_deallocate (port_task_self,
1005                                                         (vm_address_t) thread_list,
1006                                                         thread_list_len * sizeof (thread_act_t)))
1007                                         != KERN_SUCCESS)
1008                         {
1009                                 syslog (LOG_ERR, "vm_deallocate failed: %s",
1010                                                 mach_error_string (status));
1011                         }
1012                         thread_list = NULL;
1013                         thread_list_len = 0;
1014
1015                         /* Only deallocate the task port, if it isn't our own.
1016                          * Don't know what would happen in that case, but this
1017                          * is what Apple's top does.. ;) */
1018                         if (task_list[task] != port_task_self)
1019                         {
1020                                 status = mach_port_deallocate (port_task_self,
1021                                                 task_list[task]);
1022                                 if (status != KERN_SUCCESS)
1023                                         syslog (LOG_ERR, "mach_port_deallocate failed: %s",
1024                                                         mach_error_string (status));
1025                         }
1026
1027                         if (ps != NULL)
1028                                 ps_list_add (task_name, &pse);
1029                 } /* for (task_list) */
1030
1031                 if ((status = vm_deallocate (port_task_self,
1032                                 (vm_address_t) task_list,
1033                                 task_list_len * sizeof (task_t))) != KERN_SUCCESS)
1034                 {
1035                         syslog (LOG_ERR, "vm_deallocate failed: %s",
1036                                         mach_error_string (status));
1037                 }
1038                 task_list = NULL;
1039                 task_list_len = 0;
1040
1041                 if ((status = mach_port_deallocate (port_task_self, port_pset_priv))
1042                                 != KERN_SUCCESS)
1043                 {
1044                         syslog (LOG_ERR, "mach_port_deallocate failed: %s",
1045                                         mach_error_string (status));
1046                 }
1047         } /* for (pset_list) */
1048
1049         ps_submit (running, sleeping, zombies, stopped, -1, blocked);
1050
1051         for (ps = list_head_g; ps != NULL; ps = ps->next)
1052                 ps_submit_proc_list (ps);
1053 /* #endif HAVE_THREAD_INFO */
1054
1055 #elif KERNEL_LINUX
1056         int running  = 0;
1057         int sleeping = 0;
1058         int zombies  = 0;
1059         int stopped  = 0;
1060         int paging   = 0;
1061         int blocked  = 0;
1062
1063         struct dirent *ent;
1064         DIR           *proc;
1065         int            pid;
1066
1067         int        status;
1068         procstat_t ps;
1069         procstat_entry_t pse;
1070         char       state;
1071
1072         procstat_t *ps_ptr;
1073
1074         running = sleeping = zombies = stopped = paging = blocked = 0;
1075         ps_list_reset ();
1076
1077         if ((proc = opendir ("/proc")) == NULL)
1078         {
1079                 syslog (LOG_ERR, "Cannot open `/proc': %s", strerror (errno));
1080                 return;
1081         }
1082
1083         while ((ent = readdir (proc)) != NULL)
1084         {
1085                 if (!isdigit (ent->d_name[0]))
1086                         continue;
1087
1088                 if ((pid = atoi (ent->d_name)) < 1)
1089                         continue;
1090
1091                 status = ps_read_process (pid, &ps, &state);
1092                 if (status != 0)
1093                 {
1094                         DBG ("ps_read_process failed: %i", status);
1095                         continue;
1096                 }
1097
1098                 pse.id       = pid;
1099                 pse.age      = 0;
1100
1101                 pse.num_proc = ps.num_proc;
1102                 pse.num_lwp  = ps.num_lwp;
1103                 pse.vmem_rss = ps.vmem_rss;
1104
1105                 pse.vmem_minflt = 0;
1106                 pse.vmem_minflt_counter = ps.vmem_minflt_counter;
1107                 pse.vmem_majflt = 0;
1108                 pse.vmem_majflt_counter = ps.vmem_majflt_counter;
1109
1110                 pse.cpu_user = 0;
1111                 pse.cpu_user_counter = ps.cpu_user_counter;
1112                 pse.cpu_system = 0;
1113                 pse.cpu_system_counter = ps.cpu_system_counter;
1114
1115                 switch (state)
1116                 {
1117                         case 'R': running++;  break;
1118                         case 'S': sleeping++; break;
1119                         case 'D': blocked++;  break;
1120                         case 'Z': zombies++;  break;
1121                         case 'T': stopped++;  break;
1122                         case 'W': paging++;   break;
1123                 }
1124
1125                 ps_list_add (ps.name, &pse);
1126         }
1127
1128         closedir (proc);
1129
1130         ps_submit (running, sleeping, zombies, stopped, paging, blocked);
1131
1132         for (ps_ptr = list_head_g; ps_ptr != NULL; ps_ptr = ps_ptr->next)
1133                 ps_submit_proc_list (ps_ptr);
1134 #endif /* KERNEL_LINUX */
1135 }
1136 #else
1137 # define ps_read NULL
1138 #endif /* PROCESSES_HAVE_READ */
1139
1140 void module_register (void)
1141 {
1142         plugin_register (MODULE_NAME, ps_init, ps_read, ps_write);
1143         plugin_register ("ps_rss", NULL, NULL, ps_rss_write);
1144         plugin_register ("ps_cputime", NULL, NULL, ps_cputime_write);
1145         plugin_register ("ps_count", NULL, NULL, ps_count_write);
1146         plugin_register ("ps_pagefaults", NULL, NULL, ps_pagefaults_write);
1147 #if HAVE_THREAD_INFO | KERNEL_LINUX
1148         cf_register (MODULE_NAME, ps_config, config_keys, config_keys_num);
1149 #endif
1150 }
1151
1152 #undef BUFSIZE
1153 #undef MODULE_NAME