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