src/processes.c

   1 /**
   2  * collectd - src/processes.c
   3  * Copyright (C) 2005  Lyonel Vincent
   4  *
   5  * This program is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License as published by the
   7  * Free Software Foundation; either version 2 of the License, or (at your
   8  * option) any later version.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License for more details.
  14  *
  15  * You should have received a copy of the GNU General Public License along
  16  * with this program; if not, write to the Free Software Foundation, Inc.,
  17  * 51 Franklin St, Fifth Floor, Boston, MA  02110-1301 USA
  18  *
  19  * Authors:
  20  *   Lyonel Vincent <lyonel at ezix.org>
  21  *   Florian octo Forster <octo at verplant.org>
  22  **/
  23
  24 #include "collectd.h"
  25 #include "common.h"
  26 #include "plugin.h"
  27 #include "utils_debug.h"
  28
  29 /* Include header files for the mach system, if they exist.. */
  30 #if HAVE_MACH_MACH_INIT_H
  31 #  include <mach/mach_init.h>
  32 #endif
  33 #if HAVE_MACH_HOST_PRIV_H
  34 #  include <mach/host_priv.h>
  35 #endif
  36 #if HAVE_MACH_MACH_ERROR_H
  37 #  include <mach/mach_error.h>
  38 #endif
  39 #if HAVE_MACH_MACH_HOST_H
  40 #  include <mach/mach_host.h>
  41 #endif
  42 #if HAVE_MACH_MACH_PORT_H
  43 #  include <mach/mach_port.h>
  44 #endif
  45 #if HAVE_MACH_MACH_TYPES_H
  46 #  include <mach/mach_types.h>
  47 #endif
  48 #if HAVE_MACH_MESSAGE_H
  49 #  include <mach/message.h>
  50 #endif
  51 #if HAVE_MACH_PROCESSOR_SET_H
  52 #  include <mach/processor_set.h>
  53 #endif
  54 #if HAVE_MACH_TASK_H
  55 #  include <mach/task.h>
  56 #endif
  57 #if HAVE_MACH_THREAD_ACT_H
  58 #  include <mach/thread_act.h>
  59 #endif
  60 #if HAVE_MACH_VM_REGION_H
  61 #  include <mach/vm_region.h>
  62 #endif
  63 #if HAVE_MACH_VM_MAP_H
  64 #  include <mach/vm_map.h>
  65 #endif
  66 #if HAVE_MACH_VM_PROT_H
  67 #  include <mach/vm_prot.h>
  68 #endif
  69
  70 #define MODULE_NAME "processes"
  71
  72 #if HAVE_THREAD_INFO || defined(KERNEL_LINUX)
  73 # define PROCESSES_HAVE_READ 1
  74 #else
  75 # define PROCESSES_HAVE_READ 0
  76 #endif
  77
  78 #define BUFSIZE 256
  79
  80 static char *ps_file = "processes.rrd";
  81
  82 static char *ds_def[] =
  83 {
  84         "DS:running:GAUGE:"COLLECTD_HEARTBEAT":0:65535",
  85         "DS:sleeping:GAUGE:"COLLECTD_HEARTBEAT":0:65535",
  86         "DS:zombies:GAUGE:"COLLECTD_HEARTBEAT":0:65535",
  87         "DS:stopped:GAUGE:"COLLECTD_HEARTBEAT":0:65535",
  88         "DS:paging:GAUGE:"COLLECTD_HEARTBEAT":0:65535",
  89         "DS:blocked:GAUGE:"COLLECTD_HEARTBEAT":0:65535",
  90         NULL
  91 };
  92 static int ds_num = 6;
  93
  94 #if HAVE_THREAD_INFO
  95 static mach_port_t port_host_self;
  96 static mach_port_t port_task_self;
  97
  98 static processor_set_name_array_t pset_list;
  99 static mach_msg_type_number_t     pset_list_len;
 100 /* #endif HAVE_THREAD_INFO */
 101
 102 #elif KERNEL_LINUX
 103 /* No global variables */
 104 #endif /* KERNEL_LINUX */
 105
 106 static void ps_init (void)
 107 {
 108 #if HAVE_THREAD_INFO
 109         kern_return_t status;
 110
 111         port_host_self = mach_host_self ();
 112         port_task_self = mach_task_self ();
 113
 114         if (pset_list != NULL)
 115         {
 116                 vm_deallocate (port_task_self,
 117                                 (vm_address_t) pset_list,
 118                                 pset_list_len * sizeof (processor_set_t));
 119                 pset_list = NULL;
 120                 pset_list_len = 0;
 121         }
 122
 123         if ((status = host_processor_sets (port_host_self,
 124                                         &pset_list,
 125                                         &pset_list_len)) != KERN_SUCCESS)
 126         {
 127                 syslog (LOG_ERR, "host_processor_sets failed: %s\n",
 128                                 mach_error_string (status));
 129                 pset_list = NULL;
 130                 pset_list_len = 0;
 131                 return;
 132         }
 133 /* #endif HAVE_THREAD_INFO */
 134
 135 #elif KERNEL_LINUX
 136         /* No init */
 137 #endif /* KERNEL_LINUX */
 138
 139         return;
 140 }
 141
 142 static void ps_write (char *host, char *inst, char *val)
 143 {
 144         rrd_update_file (host, ps_file, val, ds_def, ds_num);
 145 }
 146
 147 #if PROCESSES_HAVE_READ
 148 static void ps_submit (int running,
 149                 int sleeping,
 150                 int zombies,
 151                 int stopped,
 152                 int paging,
 153                 int blocked)
 154 {
 155         char buf[BUFSIZE];
 156
 157         if (snprintf (buf, BUFSIZE, "%u:%i:%i:%i:%i:%i:%i",
 158                                 (unsigned int) curtime,
 159                                 running, sleeping, zombies, stopped, paging,
 160                                 blocked) >= BUFSIZE)
 161                 return;
 162
 163         plugin_submit (MODULE_NAME, "-", buf);
 164 }
 165
 166 static void ps_read (void)
 167 {
 168 #if HAVE_THREAD_INFO
 169         kern_return_t            status;
 170
 171         int                      pset;
 172         processor_set_t          port_pset_priv;
 173
 174         int                      task;
 175         task_array_t             task_list;
 176         mach_msg_type_number_t   task_list_len;
 177
 178         int                      thread;
 179         thread_act_array_t       thread_list;
 180         mach_msg_type_number_t   thread_list_len;
 181         thread_basic_info_data_t thread_data;
 182         mach_msg_type_number_t   thread_data_len;
 183
 184         int running  = 0;
 185         int sleeping = 0;
 186         int zombies  = 0;
 187         int stopped  = 0;
 188         int blocked  = 0;
 189
 190         /*
 191          * The Mach-concept is a little different from the traditional UNIX
 192          * concept: All the work is done in threads. Threads are contained in
 193          * `tasks'. Therefore, `task status' doesn't make much sense, since
 194          * it's actually a `thread status'.
 195          * Tasks are assigned to sets of processors, so that's where you go to
 196          * get a list.
 197          */
 198         for (pset = 0; pset < pset_list_len; pset++)
 199         {
 200                 if ((status = host_processor_set_priv (port_host_self,
 201                                                 pset_list[pset],
 202                                                 &port_pset_priv)) != KERN_SUCCESS)
 203                 {
 204                         syslog (LOG_ERR, "host_processor_set_priv failed: %s\n",
 205                                         mach_error_string (status));
 206                         continue;
 207                 }
 208
 209                 if ((status = processor_set_tasks (port_pset_priv,
 210                                                 &task_list,
 211                                                 &task_list_len)) != KERN_SUCCESS)
 212                 {
 213                         syslog (LOG_ERR, "processor_set_tasks failed: %s\n",
 214                                         mach_error_string (status));
 215                         mach_port_deallocate (port_task_self, port_pset_priv);
 216                         continue;
 217                 }
 218
 219                 for (task = 0; task < task_list_len; task++)
 220                 {
 221                         status = task_threads (task_list[task], &thread_list,
 222                                         &thread_list_len);
 223                         if (status != KERN_SUCCESS)
 224                         {
 225                                 /* Apple's `top' treats this case a zombie. It
 226                                  * makes sense to some extend: A `zombie'
 227                                  * thread is nonsense, since the task/process
 228                                  * is dead. */
 229                                 zombies++;
 230                                 DBG ("task_threads failed: %s",
 231                                                 mach_error_string (status));
 232                                 if (task_list[task] != port_task_self)
 233                                         mach_port_deallocate (port_task_self,
 234                                                         task_list[task]);
 235                                 continue; /* with next task_list */
 236                         }
 237
 238                         for (thread = 0; thread < thread_list_len; thread++)
 239                         {
 240                                 thread_data_len = THREAD_BASIC_INFO_COUNT;
 241                                 status = thread_info (thread_list[thread],
 242                                                 THREAD_BASIC_INFO,
 243                                                 (thread_info_t) &thread_data,
 244                                                 &thread_data_len);
 245                                 if (status != KERN_SUCCESS)
 246                                 {
 247                                         syslog (LOG_ERR, "thread_info failed: %s\n",
 248                                                         mach_error_string (status));
 249                                         if (task_list[task] != port_task_self)
 250                                                 mach_port_deallocate (port_task_self,
 251                                                                 thread_list[thread]);
 252                                         continue; /* with next thread_list */
 253                                 }
 254
 255                                 switch (thread_data.run_state)
 256                                 {
 257                                         case TH_STATE_RUNNING:
 258                                                 running++;
 259                                                 break;
 260                                         case TH_STATE_STOPPED:
 261                                         /* What exactly is `halted'? */
 262                                         case TH_STATE_HALTED:
 263                                                 stopped++;
 264                                                 break;
 265                                         case TH_STATE_WAITING:
 266                                                 sleeping++;
 267                                                 break;
 268                                         case TH_STATE_UNINTERRUPTIBLE:
 269                                                 blocked++;
 270                                                 break;
 271                                         /* There is no `zombie' case here,
 272                                          * since there are no zombie-threads.
 273                                          * There's only zombie tasks, which are
 274                                          * handled above. */
 275                                         default:
 276                                                 syslog (LOG_WARNING,
 277                                                                 "Unknown thread status: %s",
 278                                                                 thread_data.run_state);
 279                                                 break;
 280                                 } /* switch (thread_data.run_state) */
 281
 282                                 if (task_list[task] != port_task_self)
 283                                 {
 284                                         status = mach_port_deallocate (port_task_self,
 285                                                         thread_list[thread]);
 286                                         if (status != KERN_SUCCESS)
 287                                                 syslog (LOG_ERR, "mach_port_deallocate failed: %s",
 288                                                                 mach_error_string (status));
 289                                 }
 290                         } /* for (thread_list) */
 291
 292                         if ((status = vm_deallocate (port_task_self,
 293                                                         (vm_address_t) thread_list,
 294                                                         thread_list_len * sizeof (thread_act_t)))
 295                                         != KERN_SUCCESS)
 296                         {
 297                                 syslog (LOG_ERR, "vm_deallocate failed: %s",
 298                                                 mach_error_string (status));
 299                         }
 300                         thread_list = NULL;
 301                         thread_list_len = 0;
 302
 303                         /* Only deallocate the task port, if it isn't our own.
 304                          * Don't know what would happen in that case, but this
 305                          * is what Apple's top does.. ;) */
 306                         if (task_list[task] != port_task_self)
 307                         {
 308                                 status = mach_port_deallocate (port_task_self,
 309                                                 task_list[task]);
 310                                 if (status != KERN_SUCCESS)
 311                                         syslog (LOG_ERR, "mach_port_deallocate failed: %s",
 312                                                         mach_error_string (status));
 313                         }
 314                 } /* for (task_list) */
 315
 316                 if ((status = vm_deallocate (port_task_self,
 317                                 (vm_address_t) task_list,
 318                                 task_list_len * sizeof (task_t))) != KERN_SUCCESS)
 319                 {
 320                         syslog (LOG_ERR, "vm_deallocate failed: %s",
 321                                         mach_error_string (status));
 322                 }
 323                 task_list = NULL;
 324                 task_list_len = 0;
 325
 326                 if ((status = mach_port_deallocate (port_task_self, port_pset_priv))
 327                                 != KERN_SUCCESS)
 328                 {
 329                         syslog (LOG_ERR, "mach_port_deallocate failed: %s",
 330                                         mach_error_string (status));
 331                 }
 332         } /* for (pset_list) */
 333 /* #endif HAVE_THREAD_INFO */
 334
 335 #elif KERNEL_LINUX
 336         int running  = 0;
 337         int sleeping = 0;
 338         int zombies  = 0;
 339         int stopped  = 0;
 340         int paging   = 0;
 341         int blocked  = 0;
 342
 343         char buf[BUFSIZE];
 344         char filename[20]; /* need 17 bytes */
 345         char *fields[BUFSIZE];
 346
 347         struct dirent *ent;
 348         DIR *proc;
 349         FILE *fh;
 350
 351         running = sleeping = zombies = stopped = paging = blocked = 0;
 352
 353         if ((proc = opendir ("/proc")) == NULL)
 354         {
 355                 syslog (LOG_ERR, "Cannot open `/proc': %s", strerror (errno));
 356                 return;
 357         }
 358
 359         while ((ent = readdir (proc)) != NULL)
 360         {
 361                 if (!isdigit (ent->d_name[0]))
 362                         continue;
 363
 364                 if (snprintf (filename, 20, "/proc/%s/stat", ent->d_name) >= 20)
 365                         continue;
 366
 367                 if ((fh = fopen (filename, "r")) == NULL)
 368                 {
 369                         syslog (LOG_ERR, "Cannot open `%s': %s", filename, strerror (errno));
 370                         continue;
 371                 }
 372
 373                 if (fgets (buf, BUFSIZE, fh) == NULL)
 374                 {
 375                         fclose (fh);
 376                         continue;
 377                 }
 378
 379                 fclose (fh);
 380
 381                 if (strsplit (buf, fields, BUFSIZE) < 3)
 382                         continue;
 383
 384                 switch (fields[2][0])
 385                 {
 386                         case 'R': running++;  break;
 387                         case 'S': sleeping++; break;
 388                         case 'D': blocked++;  break;
 389                         case 'Z': zombies++;  break;
 390                         case 'T': stopped++;  break;
 391                         case 'W': paging++;   break;
 392                 }
 393         }
 394
 395         closedir(proc);
 396
 397         ps_submit (running, sleeping, zombies, stopped, paging, blocked);
 398 #endif /* KERNEL_LINUX */
 399 }
 400 #else
 401 # define ps_read NULL
 402 #endif /* PROCESSES_HAVE_READ */
 403
 404 void module_register (void)
 405 {
 406         plugin_register (MODULE_NAME, ps_init, ps_read, ps_write);
 407 }
 408
 409 #undef BUFSIZE
 410 #undef MODULE_NAME