src/processes.c

   1 /**
   2  * collectd - src/processes.c
   3  * Copyright (C) 2005  Lyonel Vincent
   4  * Copyright (C) 2006-2008  Florian Forster (Mach code)
   5  * Copyright (C) 2008  Oleg King
   6  *
   7  * This program is free software; you can redistribute it and/or modify it
   8  * under the terms of the GNU General Public License as published by the
   9  * Free Software Foundation; either version 2 of the License, or (at your
  10  * option) any later version.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU General Public License along
  18  * with this program; if not, write to the Free Software Foundation, Inc.,
  19  * 51 Franklin St, Fifth Floor, Boston, MA  02110-1301 USA
  20  *
  21  * Authors:
  22  *   Lyonel Vincent <lyonel at ezix.org>
  23  *   Florian octo Forster <octo at verplant.org>
  24  *   Oleg King <king2 at kaluga.ru>
  25  **/
  26
  27 #include "collectd.h"
  28 #include "common.h"
  29 #include "plugin.h"
  30 #include "configfile.h"
  31
  32 /* Include header files for the mach system, if they exist.. */
  33 #if HAVE_THREAD_INFO
  34 #  if HAVE_MACH_MACH_INIT_H
  35 #    include <mach/mach_init.h>
  36 #  endif
  37 #  if HAVE_MACH_HOST_PRIV_H
  38 #    include <mach/host_priv.h>
  39 #  endif
  40 #  if HAVE_MACH_MACH_ERROR_H
  41 #    include <mach/mach_error.h>
  42 #  endif
  43 #  if HAVE_MACH_MACH_HOST_H
  44 #    include <mach/mach_host.h>
  45 #  endif
  46 #  if HAVE_MACH_MACH_PORT_H
  47 #    include <mach/mach_port.h>
  48 #  endif
  49 #  if HAVE_MACH_MACH_TYPES_H
  50 #    include <mach/mach_types.h>
  51 #  endif
  52 #  if HAVE_MACH_MESSAGE_H
  53 #    include <mach/message.h>
  54 #  endif
  55 #  if HAVE_MACH_PROCESSOR_SET_H
  56 #    include <mach/processor_set.h>
  57 #  endif
  58 #  if HAVE_MACH_TASK_H
  59 #    include <mach/task.h>
  60 #  endif
  61 #  if HAVE_MACH_THREAD_ACT_H
  62 #    include <mach/thread_act.h>
  63 #  endif
  64 #  if HAVE_MACH_VM_REGION_H
  65 #    include <mach/vm_region.h>
  66 #  endif
  67 #  if HAVE_MACH_VM_MAP_H
  68 #    include <mach/vm_map.h>
  69 #  endif
  70 #  if HAVE_MACH_VM_PROT_H
  71 #    include <mach/vm_prot.h>
  72 #  endif
  73 #  if HAVE_SYS_SYSCTL_H
  74 #    include <sys/sysctl.h>
  75 #  endif
  76 /* #endif HAVE_THREAD_INFO */
  77
  78 #elif KERNEL_LINUX
  79 #  if HAVE_LINUX_CONFIG_H
  80 #    include <linux/config.h>
  81 #  endif
  82 #  ifndef CONFIG_HZ
  83 #    define CONFIG_HZ 100
  84 #  endif
  85 /* #endif KERNEL_LINUX */
  86
  87 #elif HAVE_LIBKVM_GETPROCS
  88 #  include <kvm.h>
  89 #  include <sys/user.h>
  90 #  include <sys/proc.h>
  91 #  if HAVE_SYS_SYSCTL_H
  92 #    include <sys/sysctl.h>
  93 #  endif
  94 /* #endif HAVE_LIBKVM_GETPROCS */
  95
  96 #else
  97 # error "No applicable input method."
  98 #endif
  99
 100 #if HAVE_REGEX_H
 101 # include <regex.h>
 102 #endif
 103
 104 #define BUFSIZE 256
 105
 106 static const char *config_keys[] =
 107 {
 108         "Process",
 109         "ProcessMatch",
 110         NULL
 111 };
 112 static int config_keys_num = STATIC_ARRAY_SIZE (config_keys);
 113
 114 typedef struct procstat_entry_s
 115 {
 116         unsigned long id;
 117         unsigned long age;
 118
 119         unsigned long num_proc;
 120         unsigned long num_lwp;
 121         unsigned long vmem_rss;
 122
 123         unsigned long vmem_minflt;
 124         unsigned long vmem_majflt;
 125         unsigned long vmem_minflt_counter;
 126         unsigned long vmem_majflt_counter;
 127
 128         unsigned long cpu_user;
 129         unsigned long cpu_system;
 130         unsigned long cpu_user_counter;
 131         unsigned long cpu_system_counter;
 132
 133         struct procstat_entry_s *next;
 134 } procstat_entry_t;
 135
 136 #define PROCSTAT_NAME_LEN 256
 137 typedef struct procstat
 138 {
 139         char          name[PROCSTAT_NAME_LEN];
 140 #if HAVE_REGEX_H
 141         regex_t *re;
 142 #endif
 143
 144         unsigned long num_proc;
 145         unsigned long num_lwp;
 146         unsigned long vmem_rss;
 147
 148         unsigned long vmem_minflt_counter;
 149         unsigned long vmem_majflt_counter;
 150
 151         unsigned long cpu_user_counter;
 152         unsigned long cpu_system_counter;
 153
 154         struct procstat   *next;
 155         struct procstat_entry_s *instances;
 156 } procstat_t;
 157
 158 static procstat_t *list_head_g = NULL;
 159
 160 #if HAVE_THREAD_INFO
 161 static mach_port_t port_host_self;
 162 static mach_port_t port_task_self;
 163
 164 static processor_set_name_array_t pset_list;
 165 static mach_msg_type_number_t     pset_list_len;
 166 /* #endif HAVE_THREAD_INFO */
 167
 168 #elif KERNEL_LINUX
 169 static long pagesize_g;
 170 /* #endif KERNEL_LINUX */
 171
 172 #elif HAVE_LIBKVM_GETPROCS
 173 /* no global variables */
 174 #endif /* HAVE_LIBKVM_GETPROCS */
 175
 176 /* put name of process from config to list_head_g tree
 177    list_head_g is a list of 'procstat_t' structs with
 178    processes names we want to watch */
 179 static void ps_list_register (const char *name, const char *regexp)
 180 {
 181         procstat_t *new;
 182         procstat_t *ptr;
 183         int status;
 184
 185         new = (procstat_t *) malloc (sizeof (procstat_t));
 186         if (new == NULL)
 187         {
 188                 ERROR ("processes plugin: ps_list_register: malloc failed.");
 189                 return;
 190         }
 191         memset (new, 0, sizeof (procstat_t));
 192         sstrncpy (new->name, name, sizeof (new->name));
 193
 194 #if HAVE_REGEX_H
 195         if (regexp != NULL)
 196         {
 197                 DEBUG ("ProcessMatch: adding \"%s\" as criteria to process %s.", regexp, name);
 198                 new->re = (regex_t *) malloc (sizeof (regex_t));
 199                 if (new->re == NULL)
 200                 {
 201                         ERROR ("processes plugin: ps_list_register: malloc failed.");
 202                         sfree (new);
 203                         return;
 204                 }
 205
 206                 status = regcomp (new->re, regexp, REG_EXTENDED | REG_NOSUB);
 207                 if (status != 0)
 208                 {
 209                         DEBUG ("ProcessMatch: compiling the regular expression \"%s\" failed.", regexp);
 210                         sfree(new->re);
 211                         return;
 212                 }
 213         }
 214 #else
 215         if (regexp != NULL)
 216         {
 217                 ERROR ("processes plugin: ps_list_register: "
 218                                 "Regular expression \"%s\" found in config "
 219                                 "file, but support for regular expressions "
 220                                 "has been dispabled at compile time.",
 221                                 regexp);
 222                 sfree (new);
 223                 return;
 224         }
 225 #endif
 226
 227         for (ptr = list_head_g; ptr != NULL; ptr = ptr->next)
 228         {
 229                 if (strcmp (ptr->name, name) == 0)
 230                 {
 231                         WARNING ("processes plugin: You have configured more "
 232                                         "than one `Process' or "
 233                                         "`ProcessMatch' with the same name. "
 234                                         "All but the first setting will be "
 235                                         "ignored.");
 236                         sfree (new->re);
 237                         sfree (new);
 238                         return;
 239                 }
 240
 241                 if (ptr->next == NULL)
 242                         break;
 243         }
 244
 245         if (ptr == NULL)
 246                 list_head_g = new;
 247         else
 248                 ptr->next = new;
 249 } /* void ps_list_register */
 250
 251 /* try to match name against entry, returns 1 if success */
 252 static int ps_list_match (const char *name, const char *cmdline, procstat_t *ps)
 253 {
 254 #if HAVE_REGEX_H
 255         if (ps->re != NULL)
 256         {
 257                 int status;
 258                 const char *str;
 259
 260                 str = cmdline;
 261                 if ((str == NULL) || (str[0] == 0))
 262                         str = name;
 263
 264                 assert (str != NULL);
 265
 266                 status = regexec (ps->re, str,
 267                                 /* nmatch = */ 0,
 268                                 /* pmatch = */ NULL,
 269                                 /* eflags = */ 0);
 270                 if (status == 0)
 271                         return (1);
 272         }
 273         else
 274 #endif
 275         if (strcmp (ps->name, name) == 0)
 276                 return (1);
 277
 278         return (0);
 279 } /* int ps_list_match */
 280
 281 /* add process entry to 'instances' of process 'name' (or refresh it) */
 282 static void ps_list_add (const char *name, const char *cmdline, procstat_entry_t *entry)
 283 {
 284         procstat_t *ps;
 285         procstat_entry_t *pse;
 286
 287         if (entry->id == 0)
 288                 return;
 289
 290         for (ps = list_head_g; ps != NULL; ps = ps->next)
 291         {
 292                 if ((ps_list_match (name, cmdline, ps)) == 0)
 293                         continue;
 294
 295                 for (pse = ps->instances; pse != NULL; pse = pse->next)
 296                         if ((pse->id == entry->id) || (pse->next == NULL))
 297                                 break;
 298
 299                 if ((pse == NULL) || (pse->id != entry->id))
 300                 {
 301                         procstat_entry_t *new;
 302
 303                         new = (procstat_entry_t *) malloc (sizeof (procstat_entry_t));
 304                         if (new == NULL)
 305                                 return;
 306                         memset (new, 0, sizeof (procstat_entry_t));
 307                         new->id = entry->id;
 308
 309                         if (pse == NULL)
 310                                 ps->instances = new;
 311                         else
 312                                 pse->next = new;
 313
 314                         pse = new;
 315                 }
 316
 317                 pse->age = 0;
 318                 pse->num_proc = entry->num_proc;
 319                 pse->num_lwp  = entry->num_lwp;
 320                 pse->vmem_rss = entry->vmem_rss;
 321
 322                 ps->num_proc += pse->num_proc;
 323                 ps->num_lwp  += pse->num_lwp;
 324                 ps->vmem_rss += pse->vmem_rss;
 325
 326                 if ((entry->vmem_minflt_counter == 0)
 327                                 && (entry->vmem_majflt_counter == 0))
 328                 {
 329                         pse->vmem_minflt_counter += entry->vmem_minflt;
 330                         pse->vmem_minflt = entry->vmem_minflt;
 331
 332                         pse->vmem_majflt_counter += entry->vmem_majflt;
 333                         pse->vmem_majflt = entry->vmem_majflt;
 334                 }
 335                 else
 336                 {
 337                         if (entry->vmem_minflt_counter < pse->vmem_minflt_counter)
 338                         {
 339                                 pse->vmem_minflt = entry->vmem_minflt_counter
 340                                         + (ULONG_MAX - pse->vmem_minflt_counter);
 341                         }
 342                         else
 343                         {
 344                                 pse->vmem_minflt = entry->vmem_minflt_counter - pse->vmem_minflt_counter;
 345                         }
 346                         pse->vmem_minflt_counter = entry->vmem_minflt_counter;
 347
 348                         if (entry->vmem_majflt_counter < pse->vmem_majflt_counter)
 349                         {
 350                                 pse->vmem_majflt = entry->vmem_majflt_counter
 351                                         + (ULONG_MAX - pse->vmem_majflt_counter);
 352                         }
 353                         else
 354                         {
 355                                 pse->vmem_majflt = entry->vmem_majflt_counter - pse->vmem_majflt_counter;
 356                         }
 357                         pse->vmem_majflt_counter = entry->vmem_majflt_counter;
 358                 }
 359
 360                 ps->vmem_minflt_counter += pse->vmem_minflt;
 361                 ps->vmem_majflt_counter += pse->vmem_majflt;
 362
 363                 if ((entry->cpu_user_counter == 0)
 364                                 && (entry->cpu_system_counter == 0))
 365                 {
 366                         pse->cpu_user_counter += entry->cpu_user;
 367                         pse->cpu_user = entry->cpu_user;
 368
 369                         pse->cpu_system_counter += entry->cpu_system;
 370                         pse->cpu_system = entry->cpu_system;
 371                 }
 372                 else
 373                 {
 374                         if (entry->cpu_user_counter < pse->cpu_user_counter)
 375                         {
 376                                 pse->cpu_user = entry->cpu_user_counter
 377                                         + (ULONG_MAX - pse->cpu_user_counter);
 378                         }
 379                         else
 380                         {
 381                                 pse->cpu_user = entry->cpu_user_counter - pse->cpu_user_counter;
 382                         }
 383                         pse->cpu_user_counter = entry->cpu_user_counter;
 384
 385                         if (entry->cpu_system_counter < pse->cpu_system_counter)
 386                         {
 387                                 pse->cpu_system = entry->cpu_system_counter
 388                                         + (ULONG_MAX - pse->cpu_system_counter);
 389                         }
 390                         else
 391                         {
 392                                 pse->cpu_system = entry->cpu_system_counter - pse->cpu_system_counter;
 393                         }
 394                         pse->cpu_system_counter = entry->cpu_system_counter;
 395                 }
 396
 397                 ps->cpu_user_counter   += pse->cpu_user;
 398                 ps->cpu_system_counter += pse->cpu_system;
 399         }
 400 }
 401
 402 /* remove old entries from instances of processes in list_head_g */
 403 static void ps_list_reset (void)
 404 {
 405         procstat_t *ps;
 406         procstat_entry_t *pse;
 407         procstat_entry_t *pse_prev;
 408
 409         for (ps = list_head_g; ps != NULL; ps = ps->next)
 410         {
 411                 ps->num_proc    = 0;
 412                 ps->num_lwp     = 0;
 413                 ps->vmem_rss    = 0;
 414
 415                 pse_prev = NULL;
 416                 pse = ps->instances;
 417                 while (pse != NULL)
 418                 {
 419                         if (pse->age > 10)
 420                         {
 421                                 DEBUG ("Removing this procstat entry cause it's too old: "
 422                                                 "id = %lu; name = %s;",
 423                                                 pse->id, ps->name);
 424
 425                                 if (pse_prev == NULL)
 426                                 {
 427                                         ps->instances = pse->next;
 428                                         free (pse);
 429                                         pse = ps->instances;
 430                                 }
 431                                 else
 432                                 {
 433                                         pse_prev->next = pse->next;
 434                                         free (pse);
 435                                         pse = pse_prev->next;
 436                                 }
 437                         }
 438                         else
 439                         {
 440                                 pse->age++;
 441                                 pse_prev = pse;
 442                                 pse = pse->next;
 443                         }
 444                 } /* while (pse != NULL) */
 445         } /* for (ps = list_head_g; ps != NULL; ps = ps->next) */
 446 }
 447
 448 /* put all pre-defined 'Process' names from config to list_head_g tree */
 449 static int ps_config (const char *key, const char *value)
 450 {
 451         if (strcasecmp (key, "Process") == 0)
 452         {
 453                 ps_list_register (value, NULL);
 454         }
 455         else if (strcasecmp (key, "ProcessMatch") == 0)
 456         {
 457                 char *new_val;
 458                 char *fields[3];
 459                 int fields_num;
 460
 461                 new_val = strdup (value);
 462                 if (new_val == NULL) {
 463                         ERROR ("processes plugin: strdup failed when processing "
 464                                         "`ProcessMatch %s'.", value);
 465                         return (1);
 466                 }
 467
 468                 fields_num = strsplit (new_val, fields,
 469                                 STATIC_ARRAY_SIZE (fields));
 470                 if (fields_num != 2)
 471                 {
 472                         ERROR ("processes plugin: `ProcessMatch' needs exactly "
 473                                         "two string arguments.");
 474                         sfree (new_val);
 475                         return (1);
 476                 }
 477                 ps_list_register (fields[0], fields[1]);
 478                 sfree (new_val);
 479         }
 480         else
 481         {
 482                 ERROR ("processes plugin: The `%s' configuration option is not "
 483                                 "understood and will be ignored.", key);
 484                 return (-1);
 485         }
 486
 487         return (0);
 488 }
 489
 490 static int ps_init (void)
 491 {
 492 #if HAVE_THREAD_INFO
 493         kern_return_t status;
 494
 495         port_host_self = mach_host_self ();
 496         port_task_self = mach_task_self ();
 497
 498         if (pset_list != NULL)
 499         {
 500                 vm_deallocate (port_task_self,
 501                                 (vm_address_t) pset_list,
 502                                 pset_list_len * sizeof (processor_set_t));
 503                 pset_list = NULL;
 504                 pset_list_len = 0;
 505         }
 506
 507         if ((status = host_processor_sets (port_host_self,
 508                                         &pset_list,
 509                                         &pset_list_len)) != KERN_SUCCESS)
 510         {
 511                 ERROR ("host_processor_sets failed: %s\n",
 512                                 mach_error_string (status));
 513                 pset_list = NULL;
 514                 pset_list_len = 0;
 515                 return (-1);
 516         }
 517 /* #endif HAVE_THREAD_INFO */
 518
 519 #elif KERNEL_LINUX
 520         pagesize_g = sysconf(_SC_PAGESIZE);
 521         DEBUG ("pagesize_g = %li; CONFIG_HZ = %i;",
 522                         pagesize_g, CONFIG_HZ);
 523 /* #endif KERNEL_LINUX */
 524
 525 #elif HAVE_LIBKVM_GETPROCS
 526 /* no initialization */
 527 #endif /* HAVE_LIBKVM_GETPROCS */
 528
 529         return (0);
 530 } /* int ps_init */
 531
 532 /* submit global state (e.g.: qty of zombies, running, etc..) */
 533 static void ps_submit_state (const char *state, double value)
 534 {
 535         value_t values[1];
 536         value_list_t vl = VALUE_LIST_INIT;
 537
 538         values[0].gauge = value;
 539
 540         vl.values = values;
 541         vl.values_len = 1;
 542         vl.time = time (NULL);
 543         sstrncpy (vl.host, hostname_g, sizeof (vl.host));
 544         sstrncpy (vl.plugin, "processes", sizeof (vl.plugin));
 545         sstrncpy (vl.plugin_instance, "", sizeof (vl.plugin_instance));
 546         sstrncpy (vl.type, "ps_state", sizeof (vl.type));
 547         sstrncpy (vl.type_instance, state, sizeof (vl.type_instance));
 548
 549         plugin_dispatch_values (&vl);
 550 }
 551
 552 /* submit info about specific process (e.g.: memory taken, cpu usage, etc..) */
 553 static void ps_submit_proc_list (procstat_t *ps)
 554 {
 555         value_t values[2];
 556         value_list_t vl = VALUE_LIST_INIT;
 557
 558         vl.values = values;
 559         vl.values_len = 2;
 560         vl.time = time (NULL);
 561         sstrncpy (vl.host, hostname_g, sizeof (vl.host));
 562         sstrncpy (vl.plugin, "processes", sizeof (vl.plugin));
 563         sstrncpy (vl.plugin_instance, ps->name, sizeof (vl.plugin_instance));
 564
 565         sstrncpy (vl.type, "ps_rss", sizeof (vl.type));
 566         vl.values[0].gauge = ps->vmem_rss;
 567         vl.values_len = 1;
 568         plugin_dispatch_values (&vl);
 569
 570         sstrncpy (vl.type, "ps_cputime", sizeof (vl.type));
 571         vl.values[0].counter = ps->cpu_user_counter;
 572         vl.values[1].counter = ps->cpu_system_counter;
 573         vl.values_len = 2;
 574         plugin_dispatch_values (&vl);
 575
 576         sstrncpy (vl.type, "ps_count", sizeof (vl.type));
 577         vl.values[0].gauge = ps->num_proc;
 578         vl.values[1].gauge = ps->num_lwp;
 579         vl.values_len = 2;
 580         plugin_dispatch_values (&vl);
 581
 582         sstrncpy (vl.type, "ps_pagefaults", sizeof (vl.type));
 583         vl.values[0].counter = ps->vmem_minflt_counter;
 584         vl.values[1].counter = ps->vmem_majflt_counter;
 585         vl.values_len = 2;
 586         plugin_dispatch_values (&vl);
 587
 588         DEBUG ("name = %s; num_proc = %lu; num_lwp = %lu; vmem_rss = %lu; "
 589                         "vmem_minflt_counter = %lu; vmem_majflt_counter = %lu; "
 590                         "cpu_user_counter = %lu; cpu_system_counter = %lu;",
 591                         ps->name, ps->num_proc, ps->num_lwp, ps->vmem_rss,
 592                         ps->vmem_minflt_counter, ps->vmem_majflt_counter,
 593                         ps->cpu_user_counter, ps->cpu_system_counter);
 594 } /* void ps_submit_proc_list */
 595
 596 /* ------- additional functions for KERNEL_LINUX/HAVE_THREAD_INFO ------- */
 597 #if KERNEL_LINUX
 598 static int *ps_read_tasks (int pid)
 599 {
 600         int *list = NULL;
 601         int  list_size = 1; /* size of allocated space, in elements */
 602         int  list_len = 0;  /* number of currently used elements */
 603
 604         char           dirname[64];
 605         DIR           *dh;
 606         struct dirent *ent;
 607
 608         ssnprintf (dirname, sizeof (dirname), "/proc/%i/task", pid);
 609
 610         if ((dh = opendir (dirname)) == NULL)
 611         {
 612                 DEBUG ("Failed to open directory `%s'", dirname);
 613                 return (NULL);
 614         }
 615
 616         while ((ent = readdir (dh)) != NULL)
 617         {
 618                 if (!isdigit (ent->d_name[0]))
 619                         continue;
 620
 621                 if ((list_len + 1) >= list_size)
 622                 {
 623                         int *new_ptr;
 624                         int  new_size = 2 * list_size;
 625                         /* Comes in sizes: 2, 4, 8, 16, ... */
 626
 627                         new_ptr = (int *) realloc (list, (size_t) (sizeof (int) * new_size));
 628                         if (new_ptr == NULL)
 629                         {
 630                                 if (list != NULL)
 631                                         free (list);
 632                                 ERROR ("processes plugin: "
 633                                                 "Failed to allocate more memory.");
 634                                 return (NULL);
 635                         }
 636
 637                         list = new_ptr;
 638                         list_size = new_size;
 639
 640                         memset (list + list_len, 0, sizeof (int) * (list_size - list_len));
 641                 }
 642
 643                 list[list_len] = atoi (ent->d_name);
 644                 if (list[list_len] != 0)
 645                         list_len++;
 646         }
 647
 648         closedir (dh);
 649
 650         if (list_len == 0)
 651                 return (NULL);
 652
 653         assert (list_len < list_size);
 654         assert (list[list_len] == 0);
 655
 656         return (list);
 657 } /* int *ps_read_tasks */
 658
 659 int ps_read_process (int pid, procstat_t *ps, char *state)
 660 {
 661         char  filename[64];
 662         char  buffer[1024];
 663
 664         char *fields[64];
 665         char  fields_len;
 666
 667         int  *tasks;
 668         int   i;
 669
 670         int   ppid;
 671         int   name_len;
 672
 673         long long unsigned cpu_user_counter;
 674         long long unsigned cpu_system_counter;
 675         long long unsigned vmem_rss;
 676
 677         memset (ps, 0, sizeof (procstat_t));
 678
 679         ssnprintf (filename, sizeof (filename), "/proc/%i/stat", pid);
 680
 681         i = read_file_contents (filename, buffer, sizeof(buffer) - 1);
 682         if (i <= 0)
 683                 return (-1);
 684         buffer[i] = 0;
 685
 686         fields_len = strsplit (buffer, fields, 64);
 687         if (fields_len < 24)
 688         {
 689                 DEBUG ("processes plugin: ps_read_process (pid = %i):"
 690                                 " `%s' has only %i fields..",
 691                                 (int) pid, filename, fields_len);
 692                 return (-1);
 693         }
 694
 695         /* copy the name, strip brackets in the process */
 696         name_len = strlen (fields[1]) - 2;
 697         if ((fields[1][0] != '(') || (fields[1][name_len + 1] != ')'))
 698         {
 699                 DEBUG ("No brackets found in process name: `%s'", fields[1]);
 700                 return (-1);
 701         }
 702         fields[1] = fields[1] + 1;
 703         fields[1][name_len] = '\0';
 704         strncpy (ps->name, fields[1], PROCSTAT_NAME_LEN);
 705
 706         ppid = atoi (fields[3]);
 707
 708         *state = fields[2][0];
 709
 710         if (*state == 'Z')
 711         {
 712                 ps->num_lwp  = 0;
 713                 ps->num_proc = 0;
 714         }
 715         else if ((tasks = ps_read_tasks (pid)) == NULL)
 716         {
 717                 /* Kernel 2.4 or so */
 718                 ps->num_lwp  = 1;
 719                 ps->num_proc = 1;
 720         }
 721         else
 722         {
 723                 ps->num_lwp  = 0;
 724                 ps->num_proc = 1;
 725                 for (i = 0; tasks[i] != 0; i++)
 726                         ps->num_lwp++;
 727
 728                 free (tasks);
 729                 tasks = NULL;
 730         }
 731
 732         /* Leave the rest at zero if this is only a zombi */
 733         if (ps->num_proc == 0)
 734         {
 735                 DEBUG ("processes plugin: This is only a zombi: pid = %i; "
 736                                 "name = %s;", pid, ps->name);
 737                 return (0);
 738         }
 739
 740         cpu_user_counter   = atoll (fields[13]);
 741         cpu_system_counter = atoll (fields[14]);
 742         vmem_rss = atoll (fields[23]);
 743         ps->vmem_minflt_counter = atol (fields[9]);
 744         ps->vmem_majflt_counter = atol (fields[11]);
 745
 746         /* Convert jiffies to useconds */
 747         cpu_user_counter   = cpu_user_counter   * 1000000 / CONFIG_HZ;
 748         cpu_system_counter = cpu_system_counter * 1000000 / CONFIG_HZ;
 749         vmem_rss = vmem_rss * pagesize_g;
 750
 751         ps->cpu_user_counter = (unsigned long) cpu_user_counter;
 752         ps->cpu_system_counter = (unsigned long) cpu_system_counter;
 753         ps->vmem_rss = (unsigned long) vmem_rss;
 754
 755         /* success */
 756         return (0);
 757 } /* int ps_read_process (...) */
 758 #endif /* KERNEL_LINUX */
 759
 760 #if HAVE_THREAD_INFO
 761 static int mach_get_task_name (task_t t, int *pid, char *name, size_t name_max_len)
 762 {
 763         int mib[4];
 764
 765         struct kinfo_proc kp;
 766         size_t            kp_size;
 767
 768         mib[0] = CTL_KERN;
 769         mib[1] = KERN_PROC;
 770         mib[2] = KERN_PROC_PID;
 771
 772         if (pid_for_task (t, pid) != KERN_SUCCESS)
 773                 return (-1);
 774         mib[3] = *pid;
 775
 776         kp_size = sizeof (kp);
 777         if (sysctl (mib, 4, &kp, &kp_size, NULL, 0) != 0)
 778                 return (-1);
 779
 780         if (name_max_len > (MAXCOMLEN + 1))
 781                 name_max_len = MAXCOMLEN + 1;
 782
 783         strncpy (name, kp.kp_proc.p_comm, name_max_len - 1);
 784         name[name_max_len - 1] = '\0';
 785
 786         DEBUG ("pid = %i; name = %s;", *pid, name);
 787
 788         /* We don't do the special handling for `p_comm == "LaunchCFMApp"' as
 789          * `top' does it, because it is a lot of work and only used when
 790          * debugging. -octo */
 791
 792         return (0);
 793 }
 794 #endif /* HAVE_THREAD_INFO */
 795 /* ------- end of additional functions for KERNEL_LINUX/HAVE_THREAD_INFO ------- */
 796
 797 /* do actual readings from kernel */
 798 static int ps_read (void)
 799 {
 800 #if HAVE_THREAD_INFO
 801         kern_return_t            status;
 802
 803         int                      pset;
 804         processor_set_t          port_pset_priv;
 805
 806         int                      task;
 807         task_array_t             task_list;
 808         mach_msg_type_number_t   task_list_len;
 809
 810         int                      task_pid;
 811         char                     task_name[MAXCOMLEN + 1];
 812
 813         int                      thread;
 814         thread_act_array_t       thread_list;
 815         mach_msg_type_number_t   thread_list_len;
 816         thread_basic_info_data_t thread_data;
 817         mach_msg_type_number_t   thread_data_len;
 818
 819         int running  = 0;
 820         int sleeping = 0;
 821         int zombies  = 0;
 822         int stopped  = 0;
 823         int blocked  = 0;
 824
 825         procstat_t *ps;
 826         procstat_entry_t pse;
 827
 828         ps_list_reset ();
 829
 830         /*
 831          * The Mach-concept is a little different from the traditional UNIX
 832          * concept: All the work is done in threads. Threads are contained in
 833          * `tasks'. Therefore, `task status' doesn't make much sense, since
 834          * it's actually a `thread status'.
 835          * Tasks are assigned to sets of processors, so that's where you go to
 836          * get a list.
 837          */
 838         for (pset = 0; pset < pset_list_len; pset++)
 839         {
 840                 if ((status = host_processor_set_priv (port_host_self,
 841                                                 pset_list[pset],
 842                                                 &port_pset_priv)) != KERN_SUCCESS)
 843                 {
 844                         ERROR ("host_processor_set_priv failed: %s\n",
 845                                         mach_error_string (status));
 846                         continue;
 847                 }
 848
 849                 if ((status = processor_set_tasks (port_pset_priv,
 850                                                 &task_list,
 851                                                 &task_list_len)) != KERN_SUCCESS)
 852                 {
 853                         ERROR ("processor_set_tasks failed: %s\n",
 854                                         mach_error_string (status));
 855                         mach_port_deallocate (port_task_self, port_pset_priv);
 856                         continue;
 857                 }
 858
 859                 for (task = 0; task < task_list_len; task++)
 860                 {
 861                         ps = NULL;
 862                         if (mach_get_task_name (task_list[task],
 863                                                 &task_pid,
 864                                                 task_name, PROCSTAT_NAME_LEN) == 0)
 865                         {
 866                                 /* search for at least one match */
 867                                 for (ps = list_head_g; ps != NULL; ps = ps->next)
 868                                         /* FIXME: cmdline should be here instead of NULL */
 869                                         if (ps_list_match (task_name, NULL, ps) == 1)
 870                                                 break;
 871                         }
 872
 873                         /* Collect more detailed statistics for this process */
 874                         if (ps != NULL)
 875                         {
 876                                 task_basic_info_data_t        task_basic_info;
 877                                 mach_msg_type_number_t        task_basic_info_len;
 878                                 task_events_info_data_t       task_events_info;
 879                                 mach_msg_type_number_t        task_events_info_len;
 880                                 task_absolutetime_info_data_t task_absolutetime_info;
 881                                 mach_msg_type_number_t        task_absolutetime_info_len;
 882
 883                                 memset (&pse, '\0', sizeof (pse));
 884                                 pse.id = task_pid;
 885
 886                                 task_basic_info_len = TASK_BASIC_INFO_COUNT;
 887                                 status = task_info (task_list[task],
 888                                                 TASK_BASIC_INFO,
 889                                                 (task_info_t) &task_basic_info,
 890                                                 &task_basic_info_len);
 891                                 if (status != KERN_SUCCESS)
 892                                 {
 893                                         ERROR ("task_info failed: %s",
 894                                                         mach_error_string (status));
 895                                         continue; /* with next thread_list */
 896                                 }
 897
 898                                 task_events_info_len = TASK_EVENTS_INFO_COUNT;
 899                                 status = task_info (task_list[task],
 900                                                 TASK_EVENTS_INFO,
 901                                                 (task_info_t) &task_events_info,
 902                                                 &task_events_info_len);
 903                                 if (status != KERN_SUCCESS)
 904                                 {
 905                                         ERROR ("task_info failed: %s",
 906                                                         mach_error_string (status));
 907                                         continue; /* with next thread_list */
 908                                 }
 909
 910                                 task_absolutetime_info_len = TASK_ABSOLUTETIME_INFO_COUNT;
 911                                 status = task_info (task_list[task],
 912                                                 TASK_ABSOLUTETIME_INFO,
 913                                                 (task_info_t) &task_absolutetime_info,
 914                                                 &task_absolutetime_info_len);
 915                                 if (status != KERN_SUCCESS)
 916                                 {
 917                                         ERROR ("task_info failed: %s",
 918                                                         mach_error_string (status));
 919                                         continue; /* with next thread_list */
 920                                 }
 921
 922                                 pse.num_proc++;
 923                                 pse.vmem_rss = task_basic_info.resident_size;
 924
 925                                 pse.vmem_minflt_counter = task_events_info.cow_faults;
 926                                 pse.vmem_majflt_counter = task_events_info.faults;
 927
 928                                 pse.cpu_user_counter = task_absolutetime_info.total_user;
 929                                 pse.cpu_system_counter = task_absolutetime_info.total_system;
 930                         }
 931
 932                         status = task_threads (task_list[task], &thread_list,
 933                                         &thread_list_len);
 934                         if (status != KERN_SUCCESS)
 935                         {
 936                                 /* Apple's `top' treats this case a zombie. It
 937                                  * makes sense to some extend: A `zombie'
 938                                  * thread is nonsense, since the task/process
 939                                  * is dead. */
 940                                 zombies++;
 941                                 DEBUG ("task_threads failed: %s",
 942                                                 mach_error_string (status));
 943                                 if (task_list[task] != port_task_self)
 944                                         mach_port_deallocate (port_task_self,
 945                                                         task_list[task]);
 946                                 continue; /* with next task_list */
 947                         }
 948
 949                         for (thread = 0; thread < thread_list_len; thread++)
 950                         {
 951                                 thread_data_len = THREAD_BASIC_INFO_COUNT;
 952                                 status = thread_info (thread_list[thread],
 953                                                 THREAD_BASIC_INFO,
 954                                                 (thread_info_t) &thread_data,
 955                                                 &thread_data_len);
 956                                 if (status != KERN_SUCCESS)
 957                                 {
 958                                         ERROR ("thread_info failed: %s",
 959                                                         mach_error_string (status));
 960                                         if (task_list[task] != port_task_self)
 961                                                 mach_port_deallocate (port_task_self,
 962                                                                 thread_list[thread]);
 963                                         continue; /* with next thread_list */
 964                                 }
 965
 966                                 if (ps != NULL)
 967                                         pse.num_lwp++;
 968
 969                                 switch (thread_data.run_state)
 970                                 {
 971                                         case TH_STATE_RUNNING:
 972                                                 running++;
 973                                                 break;
 974                                         case TH_STATE_STOPPED:
 975                                         /* What exactly is `halted'? */
 976                                         case TH_STATE_HALTED:
 977                                                 stopped++;
 978                                                 break;
 979                                         case TH_STATE_WAITING:
 980                                                 sleeping++;
 981                                                 break;
 982                                         case TH_STATE_UNINTERRUPTIBLE:
 983                                                 blocked++;
 984                                                 break;
 985                                         /* There is no `zombie' case here,
 986                                          * since there are no zombie-threads.
 987                                          * There's only zombie tasks, which are
 988                                          * handled above. */
 989                                         default:
 990                                                 WARNING ("Unknown thread status: %i",
 991                                                                 thread_data.run_state);
 992                                                 break;
 993                                 } /* switch (thread_data.run_state) */
 994
 995                                 if (task_list[task] != port_task_self)
 996                                 {
 997                                         status = mach_port_deallocate (port_task_self,
 998                                                         thread_list[thread]);
 999                                         if (status != KERN_SUCCESS)
1000                                                 ERROR ("mach_port_deallocate failed: %s",
1001                                                                 mach_error_string (status));
1002                                 }
1003                         } /* for (thread_list) */
1004
1005                         if ((status = vm_deallocate (port_task_self,
1006                                                         (vm_address_t) thread_list,
1007                                                         thread_list_len * sizeof (thread_act_t)))
1008                                         != KERN_SUCCESS)
1009                         {
1010                                 ERROR ("vm_deallocate failed: %s",
1011                                                 mach_error_string (status));
1012                         }
1013                         thread_list = NULL;
1014                         thread_list_len = 0;
1015
1016                         /* Only deallocate the task port, if it isn't our own.
1017                          * Don't know what would happen in that case, but this
1018                          * is what Apple's top does.. ;) */
1019                         if (task_list[task] != port_task_self)
1020                         {
1021                                 status = mach_port_deallocate (port_task_self,
1022                                                 task_list[task]);
1023                                 if (status != KERN_SUCCESS)
1024                                         ERROR ("mach_port_deallocate failed: %s",
1025                                                         mach_error_string (status));
1026                         }
1027
1028                         if (ps != NULL)
1029                                 /* FIXME: cmdline should be here instead of NULL */
1030                                 ps_list_add (task_name, NULL, &pse);
1031                 } /* for (task_list) */
1032
1033                 if ((status = vm_deallocate (port_task_self,
1034                                 (vm_address_t) task_list,
1035                                 task_list_len * sizeof (task_t))) != KERN_SUCCESS)
1036                 {
1037                         ERROR ("vm_deallocate failed: %s",
1038                                         mach_error_string (status));
1039                 }
1040                 task_list = NULL;
1041                 task_list_len = 0;
1042
1043                 if ((status = mach_port_deallocate (port_task_self, port_pset_priv))
1044                                 != KERN_SUCCESS)
1045                 {
1046                         ERROR ("mach_port_deallocate failed: %s",
1047                                         mach_error_string (status));
1048                 }
1049         } /* for (pset_list) */
1050
1051         ps_submit_state ("running", running);
1052         ps_submit_state ("sleeping", sleeping);
1053         ps_submit_state ("zombies", zombies);
1054         ps_submit_state ("stopped", stopped);
1055         ps_submit_state ("blocked", blocked);
1056
1057         for (ps = list_head_g; ps != NULL; ps = ps->next)
1058                 ps_submit_proc_list (ps);
1059 /* #endif HAVE_THREAD_INFO */
1060
1061 #elif KERNEL_LINUX
1062         int running  = 0;
1063         int sleeping = 0;
1064         int zombies  = 0;
1065         int stopped  = 0;
1066         int paging   = 0;
1067         int blocked  = 0;
1068
1069         struct dirent *ent;
1070         DIR           *proc;
1071         int            pid;
1072
1073         int        status;
1074         procstat_t ps;
1075         procstat_entry_t pse;
1076         char       state;
1077
1078         procstat_t *ps_ptr;
1079
1080         running = sleeping = zombies = stopped = paging = blocked = 0;
1081         ps_list_reset ();
1082
1083         if ((proc = opendir ("/proc")) == NULL)
1084         {
1085                 char errbuf[1024];
1086                 ERROR ("Cannot open `/proc': %s",
1087                                 sstrerror (errno, errbuf, sizeof (errbuf)));
1088                 return (-1);
1089         }
1090
1091         while ((ent = readdir (proc)) != NULL)
1092         {
1093                 if (!isdigit (ent->d_name[0]))
1094                         continue;
1095
1096                 if ((pid = atoi (ent->d_name)) < 1)
1097                         continue;
1098
1099                 status = ps_read_process (pid, &ps, &state);
1100                 if (status != 0)
1101                 {
1102                         DEBUG ("ps_read_process failed: %i", status);
1103                         continue;
1104                 }
1105
1106                 pse.id       = pid;
1107                 pse.age      = 0;
1108
1109                 pse.num_proc = ps.num_proc;
1110                 pse.num_lwp  = ps.num_lwp;
1111                 pse.vmem_rss = ps.vmem_rss;
1112
1113                 pse.vmem_minflt = 0;
1114                 pse.vmem_minflt_counter = ps.vmem_minflt_counter;
1115                 pse.vmem_majflt = 0;
1116                 pse.vmem_majflt_counter = ps.vmem_majflt_counter;
1117
1118                 pse.cpu_user = 0;
1119                 pse.cpu_user_counter = ps.cpu_user_counter;
1120                 pse.cpu_system = 0;
1121                 pse.cpu_system_counter = ps.cpu_system_counter;
1122
1123                 switch (state)
1124                 {
1125                         case 'R': running++;  break;
1126                         case 'S': sleeping++; break;
1127                         case 'D': blocked++;  break;
1128                         case 'Z': zombies++;  break;
1129                         case 'T': stopped++;  break;
1130                         case 'W': paging++;   break;
1131                 }
1132
1133                 /* FIXME: cmdline should be here instead of NULL */
1134                 ps_list_add (ps.name, NULL, &pse);
1135         }
1136
1137         closedir (proc);
1138
1139         ps_submit_state ("running",  running);
1140         ps_submit_state ("sleeping", sleeping);
1141         ps_submit_state ("zombies",  zombies);
1142         ps_submit_state ("stopped",  stopped);
1143         ps_submit_state ("paging",   paging);
1144         ps_submit_state ("blocked",  blocked);
1145
1146         for (ps_ptr = list_head_g; ps_ptr != NULL; ps_ptr = ps_ptr->next)
1147                 ps_submit_proc_list (ps_ptr);
1148 /* #endif KERNEL_LINUX */
1149
1150 #elif HAVE_LIBKVM_GETPROCS
1151         int running  = 0;
1152         int sleeping = 0;
1153         int zombies  = 0;
1154         int stopped  = 0;
1155         int blocked  = 0;
1156         int idle     = 0;
1157         int wait     = 0;
1158
1159         kvm_t *kd;
1160         char errbuf[1024];
1161         char cmdline[ARG_MAX];
1162         char *cmdline_ptr;
1163         struct kinfo_proc *procs;          /* array of processes */
1164         char **argv;
1165         int count;                         /* returns number of processes */
1166         int i;
1167
1168         procstat_t *ps_ptr;
1169         procstat_entry_t pse;
1170
1171         ps_list_reset ();
1172
1173         /* Open the kvm interface, get a descriptor */
1174         kd = kvm_open (NULL, NULL, NULL, 0, errbuf);
1175         if (kd == NULL)
1176         {
1177                 ERROR ("processes plugin: Cannot open kvm interface: %s",
1178                                 errbuf);
1179                 return (0);
1180         }
1181
1182         /* Get the list of processes. */
1183         procs = kvm_getprocs(kd, KERN_PROC_ALL, 0, &count);
1184         if (procs == NULL)
1185         {
1186                 kvm_close (kd);
1187                 ERROR ("processes plugin: Cannot get kvm processes list: %s",
1188                                 kvm_geterr(kd));
1189                 return (0);
1190         }
1191
1192         /* Iterate through the processes in kinfo_proc */
1193         for (i = 0; i < count; i++)
1194         {
1195                 /* retrieve the arguments */
1196                 cmdline[0] = 0;
1197                 cmdline_ptr = NULL;
1198
1199                 argv = kvm_getargv (kd, (const struct kinfo_proc *) &(procs[i]), 0);
1200                 if (argv != NULL)
1201                 {
1202                         int status;
1203                         int argc;
1204
1205                         argc = 0;
1206                         while (argv[argc] != NULL)
1207                                 argc++;
1208
1209                         status = strjoin (cmdline, sizeof (cmdline),
1210                                         argv, argc, " ");
1211
1212                         if (status < 0)
1213                         {
1214                                 WARNING ("processes plugin: Command line did "
1215                                                 "not fit into buffer.");
1216                         }
1217                         else
1218                         {
1219                                 cmdline_ptr = &cmdline[0];
1220                         }
1221                 }
1222
1223                 pse.id       = procs[i].ki_pid;
1224                 pse.age      = 0;
1225
1226                 pse.num_proc = 1;
1227                 pse.num_lwp  = procs[i].ki_numthreads;
1228
1229                 pse.vmem_rss = procs[i].ki_rssize * getpagesize();
1230                 pse.vmem_minflt = 0;
1231                 pse.vmem_minflt_counter = procs[i].ki_rusage.ru_minflt;
1232                 pse.vmem_majflt = 0;
1233                 pse.vmem_majflt_counter = procs[i].ki_rusage.ru_majflt;
1234
1235                 pse.cpu_user = 0;
1236                 pse.cpu_user_counter = procs[i].ki_rusage.ru_utime.tv_sec
1237                         * 1000
1238                         + procs[i].ki_rusage.ru_utime.tv_usec;
1239                 pse.cpu_system = 0;
1240                 pse.cpu_system_counter = procs[i].ki_rusage.ru_stime.tv_sec
1241                         * 1000
1242                         + procs[i].ki_rusage.ru_stime.tv_usec;
1243
1244                 switch (procs[i].ki_stat)
1245                 {
1246                         case SSTOP:     stopped++;      break;
1247                         case SSLEEP:    sleeping++;     break;
1248                         case SRUN:      running++;      break;
1249                         case SIDL:      idle++;         break;
1250                         case SWAIT:     wait++;         break;
1251                         case SLOCK:     blocked++;      break;
1252                         case SZOMB:     zombies++;      break;
1253                 }
1254
1255                 ps_list_add (procs[i].ki_comm, cmdline_ptr, &pse);
1256         }
1257
1258         kvm_close(kd);
1259
1260         ps_submit_state ("running",  running);
1261         ps_submit_state ("sleeping", sleeping);
1262         ps_submit_state ("zombies",  zombies);
1263         ps_submit_state ("stopped",  stopped);
1264         ps_submit_state ("blocked",  blocked);
1265         ps_submit_state ("idle",     idle);
1266         ps_submit_state ("wait",     wait);
1267
1268         for (ps_ptr = list_head_g; ps_ptr != NULL; ps_ptr = ps_ptr->next)
1269                 ps_submit_proc_list (ps_ptr);
1270 #endif /* HAVE_LIBKVM_GETPROCS */
1271
1272         return (0);
1273 } /* int ps_read */
1274
1275 void module_register (void)
1276 {
1277         plugin_register_config ("processes", ps_config,
1278                         config_keys, config_keys_num);
1279         plugin_register_init ("processes", ps_init);
1280         plugin_register_read ("processes", ps_read);
1281 } /* void module_register */