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_MACH_MACH_INIT_H
  33 #  include <mach/mach_init.h>
  34 #endif
  35 #if HAVE_MACH_HOST_PRIV_H
  36 #  include <mach/host_priv.h>
  37 #endif
  38 #if HAVE_MACH_MACH_ERROR_H
  39 #  include <mach/mach_error.h>
  40 #endif
  41 #if HAVE_MACH_MACH_HOST_H
  42 #  include <mach/mach_host.h>
  43 #endif
  44 #if HAVE_MACH_MACH_PORT_H
  45 #  include <mach/mach_port.h>
  46 #endif
  47 #if HAVE_MACH_MACH_TYPES_H
  48 #  include <mach/mach_types.h>
  49 #endif
  50 #if HAVE_MACH_MESSAGE_H
  51 #  include <mach/message.h>
  52 #endif
  53 #if HAVE_MACH_PROCESSOR_SET_H
  54 #  include <mach/processor_set.h>
  55 #endif
  56 #if HAVE_MACH_TASK_H
  57 #  include <mach/task.h>
  58 #endif
  59 #if HAVE_MACH_THREAD_ACT_H
  60 #  include <mach/thread_act.h>
  61 #endif
  62 #if HAVE_MACH_VM_REGION_H
  63 #  include <mach/vm_region.h>
  64 #endif
  65 #if HAVE_MACH_VM_MAP_H
  66 #  include <mach/vm_map.h>
  67 #endif
  68 #if HAVE_MACH_VM_PROT_H
  69 #  include <mach/vm_prot.h>
  70 #endif
  71
  72 #define MODULE_NAME "processes"
  73
  74 #if HAVE_THREAD_INFO || KERNEL_LINUX
  75 # define PROCESSES_HAVE_READ 1
  76 #else
  77 # define PROCESSES_HAVE_READ 0
  78 #endif
  79
  80 #define BUFSIZE 256
  81
  82 static char *ps_file = "processes.rrd";
  83
  84 static char *ds_def[] =
  85 {
  86         "DS:running:GAUGE:"COLLECTD_HEARTBEAT":0:65535",
  87         "DS:sleeping:GAUGE:"COLLECTD_HEARTBEAT":0:65535",
  88         "DS:zombies:GAUGE:"COLLECTD_HEARTBEAT":0:65535",
  89         "DS:stopped:GAUGE:"COLLECTD_HEARTBEAT":0:65535",
  90         "DS:paging:GAUGE:"COLLECTD_HEARTBEAT":0:65535",
  91         "DS:blocked:GAUGE:"COLLECTD_HEARTBEAT":0:65535",
  92         NULL
  93 };
  94 static int ds_num = 6;
  95
  96 static char *config_keys[] =
  97 {
  98         "CollectName",
  99         NULL
 100 };
 101 static int config_keys_num = 1;
 102
 103 typedef struct procstat
 104 {
 105 #define PROCSTAT_NAME_LEN 256
 106         char         name[PROCSTAT_NAME_LEN];
 107         unsigned int num_proc;
 108         unsigned int num_lwp;
 109         unsigned int vmem_rss;
 110         unsigned int vmem_minflt;
 111         unsigned int vmem_majflt;
 112         unsigned int cpu_user;
 113         unsigned int cpu_system;
 114         struct procstat *next;
 115 } procstat_t;
 116
 117 static procstat_t *list_head_g = NULL;
 118
 119 #if HAVE_THREAD_INFO
 120 static mach_port_t port_host_self;
 121 static mach_port_t port_task_self;
 122
 123 static processor_set_name_array_t pset_list;
 124 static mach_msg_type_number_t     pset_list_len;
 125 /* #endif HAVE_THREAD_INFO */
 126
 127 #elif KERNEL_LINUX
 128 /* No global variables */
 129 #endif /* KERNEL_LINUX */
 130
 131 static procstat_t *ps_list_append (procstat_t *list, const char *name)
 132 {
 133         procstat_t *new;
 134         procstat_t *ptr;
 135
 136         if ((new = (procstat_t *) malloc (sizeof (procstat_t))) == NULL)
 137                 return (NULL);
 138         memset (new, 0, sizeof (procstat_t));
 139         strncpy (new->name, name, PROCSTAT_NAME_LEN);
 140
 141         for (ptr = list; ptr != NULL; ptr = ptr->next)
 142                 if (ptr->next == NULL)
 143                         break;
 144
 145         if (ptr != NULL)
 146                 ptr->next = new;
 147
 148         return (new);
 149 }
 150
 151 static void ps_list_add (procstat_t *list, procstat_t *entry)
 152 {
 153         procstat_t *ptr;
 154
 155         ptr = list;
 156         while ((ptr != NULL) && (strcmp (ptr->name, entry->name) != 0))
 157                 ptr = ptr->next;
 158
 159         if (ptr == NULL)
 160                 return;
 161
 162         ptr->num_proc    += entry->num_proc;
 163         ptr->num_lwp     += entry->num_lwp;
 164         ptr->vmem_rss    += entry->vmem_rss;
 165         ptr->vmem_minflt += entry->vmem_minflt;
 166         ptr->vmem_majflt += entry->vmem_majflt;
 167         ptr->cpu_user    += entry->cpu_user;
 168         ptr->cpu_system  += entry->cpu_system;
 169 }
 170
 171 static void ps_list_reset (procstat_t *ps)
 172 {
 173         while (ps != NULL)
 174         {
 175                 ps->num_proc    = 0;
 176                 ps->num_lwp     = 0;
 177                 ps->vmem_rss    = 0;
 178                 ps->vmem_minflt = 0;
 179                 ps->vmem_majflt = 0;
 180                 ps->cpu_user    = 0;
 181                 ps->cpu_system  = 0;
 182                 ps = ps->next;
 183         }
 184 }
 185
 186 static int ps_config (char *key, char *value)
 187 {
 188         if (strcasecmp (key, "CollectName") == 0)
 189         {
 190                 procstat_t *entry;
 191
 192                 entry = ps_list_append (list_head_g, value);
 193                 if (entry == NULL)
 194                 {
 195                         syslog (LOG_ERR, "processes plugin: ps_list_append failed.");
 196                         return (1);
 197                 }
 198                 if (list_head_g != NULL)
 199                         list_head_g = entry;
 200         }
 201         else
 202         {
 203                 return (-1);
 204         }
 205
 206         return (0);
 207 }
 208
 209 static void ps_init (void)
 210 {
 211 #if HAVE_THREAD_INFO
 212         kern_return_t status;
 213
 214         port_host_self = mach_host_self ();
 215         port_task_self = mach_task_self ();
 216
 217         if (pset_list != NULL)
 218         {
 219                 vm_deallocate (port_task_self,
 220                                 (vm_address_t) pset_list,
 221                                 pset_list_len * sizeof (processor_set_t));
 222                 pset_list = NULL;
 223                 pset_list_len = 0;
 224         }
 225
 226         if ((status = host_processor_sets (port_host_self,
 227                                         &pset_list,
 228                                         &pset_list_len)) != KERN_SUCCESS)
 229         {
 230                 syslog (LOG_ERR, "host_processor_sets failed: %s\n",
 231                                 mach_error_string (status));
 232                 pset_list = NULL;
 233                 pset_list_len = 0;
 234                 return;
 235         }
 236 /* #endif HAVE_THREAD_INFO */
 237
 238 #elif KERNEL_LINUX
 239         /* No init */
 240 #endif /* KERNEL_LINUX */
 241
 242         return;
 243 }
 244
 245 static void ps_write (char *host, char *inst, char *val)
 246 {
 247         rrd_update_file (host, ps_file, val, ds_def, ds_num);
 248 }
 249
 250 #if PROCESSES_HAVE_READ
 251 static void ps_submit (int running,
 252                 int sleeping,
 253                 int zombies,
 254                 int stopped,
 255                 int paging,
 256                 int blocked)
 257 {
 258         char buf[BUFSIZE];
 259
 260         if (snprintf (buf, BUFSIZE, "%u:%i:%i:%i:%i:%i:%i",
 261                                 (unsigned int) curtime,
 262                                 running, sleeping, zombies, stopped, paging,
 263                                 blocked) >= BUFSIZE)
 264                 return;
 265
 266         DBG ("running = %i; sleeping = %i; zombies = %i; stopped = %i; paging = %i; blocked = %i;",
 267                         running, sleeping, zombies, stopped, paging, blocked);
 268
 269         plugin_submit (MODULE_NAME, "-", buf);
 270 }
 271
 272 static int *ps_read_tasks (int pid)
 273 {
 274         int *list;
 275         /* TODO */
 276 }
 277
 278 static void ps_read (void)
 279 {
 280 #if HAVE_THREAD_INFO
 281         kern_return_t            status;
 282
 283         int                      pset;
 284         processor_set_t          port_pset_priv;
 285
 286         int                      task;
 287         task_array_t             task_list;
 288         mach_msg_type_number_t   task_list_len;
 289
 290         int                      thread;
 291         thread_act_array_t       thread_list;
 292         mach_msg_type_number_t   thread_list_len;
 293         thread_basic_info_data_t thread_data;
 294         mach_msg_type_number_t   thread_data_len;
 295
 296         int running  = 0;
 297         int sleeping = 0;
 298         int zombies  = 0;
 299         int stopped  = 0;
 300         int blocked  = 0;
 301
 302         /*
 303          * The Mach-concept is a little different from the traditional UNIX
 304          * concept: All the work is done in threads. Threads are contained in
 305          * `tasks'. Therefore, `task status' doesn't make much sense, since
 306          * it's actually a `thread status'.
 307          * Tasks are assigned to sets of processors, so that's where you go to
 308          * get a list.
 309          */
 310         for (pset = 0; pset < pset_list_len; pset++)
 311         {
 312                 if ((status = host_processor_set_priv (port_host_self,
 313                                                 pset_list[pset],
 314                                                 &port_pset_priv)) != KERN_SUCCESS)
 315                 {
 316                         syslog (LOG_ERR, "host_processor_set_priv failed: %s\n",
 317                                         mach_error_string (status));
 318                         continue;
 319                 }
 320
 321                 if ((status = processor_set_tasks (port_pset_priv,
 322                                                 &task_list,
 323                                                 &task_list_len)) != KERN_SUCCESS)
 324                 {
 325                         syslog (LOG_ERR, "processor_set_tasks failed: %s\n",
 326                                         mach_error_string (status));
 327                         mach_port_deallocate (port_task_self, port_pset_priv);
 328                         continue;
 329                 }
 330
 331                 for (task = 0; task < task_list_len; task++)
 332                 {
 333                         status = task_threads (task_list[task], &thread_list,
 334                                         &thread_list_len);
 335                         if (status != KERN_SUCCESS)
 336                         {
 337                                 /* Apple's `top' treats this case a zombie. It
 338                                  * makes sense to some extend: A `zombie'
 339                                  * thread is nonsense, since the task/process
 340                                  * is dead. */
 341                                 zombies++;
 342                                 DBG ("task_threads failed: %s",
 343                                                 mach_error_string (status));
 344                                 if (task_list[task] != port_task_self)
 345                                         mach_port_deallocate (port_task_self,
 346                                                         task_list[task]);
 347                                 continue; /* with next task_list */
 348                         }
 349
 350                         for (thread = 0; thread < thread_list_len; thread++)
 351                         {
 352                                 thread_data_len = THREAD_BASIC_INFO_COUNT;
 353                                 status = thread_info (thread_list[thread],
 354                                                 THREAD_BASIC_INFO,
 355                                                 (thread_info_t) &thread_data,
 356                                                 &thread_data_len);
 357                                 if (status != KERN_SUCCESS)
 358                                 {
 359                                         syslog (LOG_ERR, "thread_info failed: %s\n",
 360                                                         mach_error_string (status));
 361                                         if (task_list[task] != port_task_self)
 362                                                 mach_port_deallocate (port_task_self,
 363                                                                 thread_list[thread]);
 364                                         continue; /* with next thread_list */
 365                                 }
 366
 367                                 switch (thread_data.run_state)
 368                                 {
 369                                         case TH_STATE_RUNNING:
 370                                                 running++;
 371                                                 break;
 372                                         case TH_STATE_STOPPED:
 373                                         /* What exactly is `halted'? */
 374                                         case TH_STATE_HALTED:
 375                                                 stopped++;
 376                                                 break;
 377                                         case TH_STATE_WAITING:
 378                                                 sleeping++;
 379                                                 break;
 380                                         case TH_STATE_UNINTERRUPTIBLE:
 381                                                 blocked++;
 382                                                 break;
 383                                         /* There is no `zombie' case here,
 384                                          * since there are no zombie-threads.
 385                                          * There's only zombie tasks, which are
 386                                          * handled above. */
 387                                         default:
 388                                                 syslog (LOG_WARNING,
 389                                                                 "Unknown thread status: %s",
 390                                                                 thread_data.run_state);
 391                                                 break;
 392                                 } /* switch (thread_data.run_state) */
 393
 394                                 if (task_list[task] != port_task_self)
 395                                 {
 396                                         status = mach_port_deallocate (port_task_self,
 397                                                         thread_list[thread]);
 398                                         if (status != KERN_SUCCESS)
 399                                                 syslog (LOG_ERR, "mach_port_deallocate failed: %s",
 400                                                                 mach_error_string (status));
 401                                 }
 402                         } /* for (thread_list) */
 403
 404                         if ((status = vm_deallocate (port_task_self,
 405                                                         (vm_address_t) thread_list,
 406                                                         thread_list_len * sizeof (thread_act_t)))
 407                                         != KERN_SUCCESS)
 408                         {
 409                                 syslog (LOG_ERR, "vm_deallocate failed: %s",
 410                                                 mach_error_string (status));
 411                         }
 412                         thread_list = NULL;
 413                         thread_list_len = 0;
 414
 415                         /* Only deallocate the task port, if it isn't our own.
 416                          * Don't know what would happen in that case, but this
 417                          * is what Apple's top does.. ;) */
 418                         if (task_list[task] != port_task_self)
 419                         {
 420                                 status = mach_port_deallocate (port_task_self,
 421                                                 task_list[task]);
 422                                 if (status != KERN_SUCCESS)
 423                                         syslog (LOG_ERR, "mach_port_deallocate failed: %s",
 424                                                         mach_error_string (status));
 425                         }
 426                 } /* for (task_list) */
 427
 428                 if ((status = vm_deallocate (port_task_self,
 429                                 (vm_address_t) task_list,
 430                                 task_list_len * sizeof (task_t))) != KERN_SUCCESS)
 431                 {
 432                         syslog (LOG_ERR, "vm_deallocate failed: %s",
 433                                         mach_error_string (status));
 434                 }
 435                 task_list = NULL;
 436                 task_list_len = 0;
 437
 438                 if ((status = mach_port_deallocate (port_task_self, port_pset_priv))
 439                                 != KERN_SUCCESS)
 440                 {
 441                         syslog (LOG_ERR, "mach_port_deallocate failed: %s",
 442                                         mach_error_string (status));
 443                 }
 444         } /* for (pset_list) */
 445
 446         ps_submit (running, sleeping, zombies, stopped, -1, blocked);
 447 /* #endif HAVE_THREAD_INFO */
 448
 449 #elif KERNEL_LINUX
 450         int running  = 0;
 451         int sleeping = 0;
 452         int zombies  = 0;
 453         int stopped  = 0;
 454         int paging   = 0;
 455         int blocked  = 0;
 456
 457         char buf[BUFSIZE];
 458         char filename[20]; /* need 17 bytes */
 459         char *fields[BUFSIZE];
 460
 461         struct dirent *ent;
 462         DIR *proc;
 463         FILE *fh;
 464
 465         running = sleeping = zombies = stopped = paging = blocked = 0;
 466
 467         if ((proc = opendir ("/proc")) == NULL)
 468         {
 469                 syslog (LOG_ERR, "Cannot open `/proc': %s", strerror (errno));
 470                 return;
 471         }
 472
 473         while ((ent = readdir (proc)) != NULL)
 474         {
 475                 if (!isdigit (ent->d_name[0]))
 476                         continue;
 477
 478                 if (snprintf (filename, 20, "/proc/%s/stat", ent->d_name) >= 20)
 479                         continue;
 480
 481                 if ((fh = fopen (filename, "r")) == NULL)
 482                 {
 483                         syslog (LOG_NOTICE, "Cannot open `%s': %s", filename,
 484                                         strerror (errno));
 485                         continue;
 486                 }
 487
 488                 if (fgets (buf, BUFSIZE, fh) == NULL)
 489                 {
 490                         syslog (LOG_NOTICE, "Unable to read from `%s': %s",
 491                                         filename, strerror (errno));
 492                         fclose (fh);
 493                         continue;
 494                 }
 495
 496                 fclose (fh);
 497
 498                 if (strsplit (buf, fields, BUFSIZE) < 3)
 499                 {
 500                         DBG ("Line has less than three fields.");
 501                         continue;
 502                 }
 503
 504                 switch (fields[2][0])
 505                 {
 506                         case 'R': running++;  break;
 507                         case 'S': sleeping++; break;
 508                         case 'D': blocked++;  break;
 509                         case 'Z': zombies++;  break;
 510                         case 'T': stopped++;  break;
 511                         case 'W': paging++;   break;
 512                 }
 513         }
 514
 515         closedir (proc);
 516
 517         ps_submit (running, sleeping, zombies, stopped, paging, blocked);
 518 #endif /* KERNEL_LINUX */
 519 }
 520 #else
 521 # define ps_read NULL
 522 #endif /* PROCESSES_HAVE_READ */
 523
 524 void module_register (void)
 525 {
 526         plugin_register (MODULE_NAME, ps_init, ps_read, ps_write);
 527         cf_register (MODULE_NAME, ps_config, config_keys, config_keys_num);
 528 }
 529
 530 #undef BUFSIZE
 531 #undef MODULE_NAME