2 * collectd - src/processes.c
3 * Copyright (C) 2005 Lyonel Vincent
4 * Copyright (C) 2006-2017 Florian octo Forster
5 * Copyright (C) 2008 Oleg King
6 * Copyright (C) 2009 Sebastian Harl
7 * Copyright (C) 2009 Andrés J. Díaz
8 * Copyright (C) 2009 Manuel Sanmartin
9 * Copyright (C) 2010 Clément Stenac
10 * Copyright (C) 2012 Cosmin Ioiart
12 * This program is free software; you can redistribute it and/or modify it
13 * under the terms of the GNU General Public License as published by the
14 * Free Software Foundation; either version 2 of the License, or (at your
15 * option) any later version.
17 * This program is distributed in the hope that it will be useful, but
18 * WITHOUT ANY WARRANTY; without even the implied warranty of
19 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
20 * General Public License for more details.
22 * You should have received a copy of the GNU General Public License along
23 * with this program; if not, write to the Free Software Foundation, Inc.,
24 * 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
27 * Lyonel Vincent <lyonel at ezix.org>
28 * Florian octo Forster <octo at collectd.org>
29 * Oleg King <king2 at kaluga.ru>
30 * Sebastian Harl <sh at tokkee.org>
31 * Andrés J. Díaz <ajdiaz at connectical.com>
33 * Clément Stenac <clement.stenac at diwi.org>
34 * Cosmin Ioiart <cioiart at gmail.com>
35 * Pavel Rochnyack <pavel2000 at ngs.ru>
36 * Wilfried Goesgens <dothebart at citadel.org>
45 #include "utils_complain.h"
46 #include "utils_taskstats.h"
49 /* Include header files for the mach system, if they exist.. */
51 #if HAVE_MACH_MACH_INIT_H
52 #include <mach/mach_init.h>
54 #if HAVE_MACH_HOST_PRIV_H
55 #include <mach/host_priv.h>
57 #if HAVE_MACH_MACH_ERROR_H
58 #include <mach/mach_error.h>
60 #if HAVE_MACH_MACH_HOST_H
61 #include <mach/mach_host.h>
63 #if HAVE_MACH_MACH_PORT_H
64 #include <mach/mach_port.h>
66 #if HAVE_MACH_MACH_TYPES_H
67 #include <mach/mach_types.h>
69 #if HAVE_MACH_MESSAGE_H
70 #include <mach/message.h>
72 #if HAVE_MACH_PROCESSOR_SET_H
73 #include <mach/processor_set.h>
76 #include <mach/task.h>
78 #if HAVE_MACH_THREAD_ACT_H
79 #include <mach/thread_act.h>
81 #if HAVE_MACH_VM_REGION_H
82 #include <mach/vm_region.h>
84 #if HAVE_MACH_VM_MAP_H
85 #include <mach/vm_map.h>
87 #if HAVE_MACH_VM_PROT_H
88 #include <mach/vm_prot.h>
91 #include <sys/sysctl.h>
93 /* #endif HAVE_THREAD_INFO */
96 #if HAVE_LINUX_CONFIG_H
97 #include <linux/config.h>
100 #define CONFIG_HZ 100
102 /* #endif KERNEL_LINUX */
104 #elif HAVE_LIBKVM_GETPROCS && \
105 (HAVE_STRUCT_KINFO_PROC_FREEBSD || HAVE_STRUCT_KINFO_PROC_OPENBSD)
107 #include <sys/param.h>
108 #include <sys/proc.h>
109 #include <sys/sysctl.h>
110 #include <sys/user.h>
111 /* #endif HAVE_LIBKVM_GETPROCS && (HAVE_STRUCT_KINFO_PROC_FREEBSD ||
112 * HAVE_STRUCT_KINFO_PROC_OPENBSD) */
114 #elif HAVE_PROCINFO_H
115 #include <procinfo.h>
116 #include <sys/types.h>
118 #define MAXPROCENTRY 32
119 #define MAXTHRDENTRY 16
120 #define MAXARGLN 1024
121 /* #endif HAVE_PROCINFO_H */
124 /* Hack: Avoid #error when building a 32-bit binary with
125 * _FILE_OFFSET_BITS=64. There is a reason for this #error, as one
126 * of the structures in <sys/procfs.h> uses an off_t, but that
127 * isn't relevant to our usage of procfs. */
128 #if !defined(_LP64) && _FILE_OFFSET_BITS == 64
130 #undef _FILE_OFFSET_BITS
136 #define _FILE_OFFSET_BITS 64
141 #include <sys/user.h>
147 /* #endif KERNEL_SOLARIS */
150 #error "No applicable input method."
161 #ifdef HAVE_SYS_CAPABILITY_H
162 #include <sys/capability.h>
165 #ifndef CMDLINE_BUFFER_SIZE
166 #if defined(ARG_MAX) && (ARG_MAX < 4096)
167 #define CMDLINE_BUFFER_SIZE ARG_MAX
169 #define CMDLINE_BUFFER_SIZE 4096
173 #define PROCSTAT_NAME_LEN 256
174 typedef struct process_entry_s {
176 char name[PROCSTAT_NAME_LEN];
178 unsigned long num_proc;
179 unsigned long num_lwp;
180 unsigned long num_fd;
181 unsigned long num_maps;
182 unsigned long vmem_size;
183 unsigned long vmem_rss;
184 unsigned long vmem_data;
185 unsigned long vmem_code;
186 unsigned long stack_size;
188 derive_t vmem_minflt_counter;
189 derive_t vmem_majflt_counter;
191 derive_t cpu_user_counter;
192 derive_t cpu_system_counter;
203 derive_t cswitch_vol;
204 derive_t cswitch_invol;
207 #if HAVE_LIBTASKSTATS
217 typedef struct procstat_entry_s {
221 derive_t vmem_minflt_counter;
222 derive_t vmem_majflt_counter;
224 derive_t cpu_user_counter;
225 derive_t cpu_system_counter;
235 derive_t cswitch_vol;
236 derive_t cswitch_invol;
238 #if HAVE_LIBTASKSTATS
239 value_to_rate_state_t delay_cpu;
240 value_to_rate_state_t delay_blkio;
241 value_to_rate_state_t delay_swapin;
242 value_to_rate_state_t delay_freepages;
245 struct procstat_entry_s *next;
248 typedef struct procstat {
249 char name[PROCSTAT_NAME_LEN];
254 unsigned long num_proc;
255 unsigned long num_lwp;
256 unsigned long num_fd;
257 unsigned long num_maps;
258 unsigned long vmem_size;
259 unsigned long vmem_rss;
260 unsigned long vmem_data;
261 unsigned long vmem_code;
262 unsigned long stack_size;
264 derive_t vmem_minflt_counter;
265 derive_t vmem_majflt_counter;
267 derive_t cpu_user_counter;
268 derive_t cpu_system_counter;
278 derive_t cswitch_vol;
279 derive_t cswitch_invol;
281 /* Linux Delay Accounting. Unit is ns/s. */
284 gauge_t delay_swapin;
285 gauge_t delay_freepages;
288 _Bool report_maps_num;
289 _Bool report_ctx_switch;
292 struct procstat *next;
293 struct procstat_entry_s *instances;
296 static procstat_t *list_head_g = NULL;
298 static _Bool want_init = 1;
299 static _Bool report_ctx_switch = 0;
300 static _Bool report_fd_num = 0;
301 static _Bool report_maps_num = 0;
302 static _Bool report_delay = 0;
305 static mach_port_t port_host_self;
306 static mach_port_t port_task_self;
308 static processor_set_name_array_t pset_list;
309 static mach_msg_type_number_t pset_list_len;
310 /* #endif HAVE_THREAD_INFO */
313 static long pagesize_g;
314 static void ps_fill_details(const procstat_t *ps, process_entry_t *entry);
315 /* #endif KERNEL_LINUX */
317 #elif HAVE_LIBKVM_GETPROCS && \
318 (HAVE_STRUCT_KINFO_PROC_FREEBSD || HAVE_STRUCT_KINFO_PROC_OPENBSD)
320 /* #endif HAVE_LIBKVM_GETPROCS && (HAVE_STRUCT_KINFO_PROC_FREEBSD ||
321 * HAVE_STRUCT_KINFO_PROC_OPENBSD) */
323 #elif HAVE_PROCINFO_H
324 static struct procentry64 procentry[MAXPROCENTRY];
325 static struct thrdentry64 thrdentry[MAXTHRDENTRY];
328 #ifndef _AIXVERSION_610
329 int getprocs64(void *procsinfo, int sizproc, void *fdsinfo, int sizfd,
330 pid_t *index, int count);
331 int getthrds64(pid_t, void *, int, tid64_t *, int);
333 int getargs(void *processBuffer, int bufferLen, char *argsBuffer, int argsLen);
334 #endif /* HAVE_PROCINFO_H */
336 #if HAVE_LIBTASKSTATS
337 static ts_t *taskstats_handle = NULL;
340 /* put name of process from config to list_head_g tree
341 * list_head_g is a list of 'procstat_t' structs with
342 * processes names we want to watch */
343 static procstat_t *ps_list_register(const char *name, const char *regexp) {
348 new = calloc(1, sizeof(*new));
350 ERROR("processes plugin: ps_list_register: calloc failed.");
353 sstrncpy(new->name, name, sizeof(new->name));
361 new->cswitch_vol = -1;
362 new->cswitch_invol = -1;
364 new->report_fd_num = report_fd_num;
365 new->report_maps_num = report_maps_num;
366 new->report_ctx_switch = report_ctx_switch;
367 new->report_delay = report_delay;
370 if (regexp != NULL) {
371 DEBUG("ProcessMatch: adding \"%s\" as criteria to process %s.", regexp,
373 new->re = malloc(sizeof(*new->re));
374 if (new->re == NULL) {
375 ERROR("processes plugin: ps_list_register: malloc failed.");
380 status = regcomp(new->re, regexp, REG_EXTENDED | REG_NOSUB);
382 DEBUG("ProcessMatch: compiling the regular expression \"%s\" failed.",
390 if (regexp != NULL) {
391 ERROR("processes plugin: ps_list_register: "
392 "Regular expression \"%s\" found in config "
393 "file, but support for regular expressions "
394 "has been disabled at compile time.",
401 for (ptr = list_head_g; ptr != NULL; ptr = ptr->next) {
402 if (strcmp(ptr->name, name) == 0) {
403 WARNING("processes plugin: You have configured more "
404 "than one `Process' or "
405 "`ProcessMatch' with the same name. "
406 "All but the first setting will be "
415 if (ptr->next == NULL)
425 } /* void ps_list_register */
427 /* try to match name against entry, returns 1 if success */
428 static int ps_list_match(const char *name, const char *cmdline,
431 if (ps->re != NULL) {
436 if ((str == NULL) || (str[0] == 0))
441 status = regexec(ps->re, str,
449 if (strcmp(ps->name, name) == 0)
453 } /* int ps_list_match */
455 static void ps_update_counter(derive_t *group_counter, derive_t *curr_counter,
456 derive_t new_counter) {
457 unsigned long curr_value;
460 *curr_counter = new_counter;
464 if (new_counter < *curr_counter)
465 curr_value = new_counter + (ULONG_MAX - *curr_counter);
467 curr_value = new_counter - *curr_counter;
469 if (*group_counter == -1)
472 *curr_counter = new_counter;
473 *group_counter += curr_value;
476 #if HAVE_LIBTASKSTATS
477 static void ps_update_delay_one(gauge_t *out_rate_sum,
478 value_to_rate_state_t *state, uint64_t cnt,
481 int status = value_to_rate(&rate, (value_t){.counter = (counter_t)cnt},
482 DS_TYPE_COUNTER, t, state);
483 if ((status != 0) || isnan(rate)) {
487 if (isnan(*out_rate_sum)) {
488 *out_rate_sum = rate;
490 *out_rate_sum += rate;
494 static void ps_update_delay(procstat_t *out, procstat_entry_t *prev,
495 process_entry_t *curr) {
496 cdtime_t now = cdtime();
498 ps_update_delay_one(&out->delay_cpu, &prev->delay_cpu, curr->delay.cpu_ns,
500 ps_update_delay_one(&out->delay_blkio, &prev->delay_blkio,
501 curr->delay.blkio_ns, now);
502 ps_update_delay_one(&out->delay_swapin, &prev->delay_swapin,
503 curr->delay.swapin_ns, now);
504 ps_update_delay_one(&out->delay_freepages, &prev->delay_freepages,
505 curr->delay.freepages_ns, now);
509 /* add process entry to 'instances' of process 'name' (or refresh it) */
510 static void ps_list_add(const char *name, const char *cmdline,
511 process_entry_t *entry) {
512 procstat_entry_t *pse;
517 for (procstat_t *ps = list_head_g; ps != NULL; ps = ps->next) {
518 if ((ps_list_match(name, cmdline, ps)) == 0)
522 ps_fill_details(ps, entry);
525 for (pse = ps->instances; pse != NULL; pse = pse->next)
526 if ((pse->id == entry->id) || (pse->next == NULL))
529 if ((pse == NULL) || (pse->id != entry->id)) {
530 procstat_entry_t *new;
532 new = calloc(1, sizeof(*new));
547 ps->num_proc += entry->num_proc;
548 ps->num_lwp += entry->num_lwp;
549 ps->num_fd += entry->num_fd;
550 ps->num_maps += entry->num_maps;
551 ps->vmem_size += entry->vmem_size;
552 ps->vmem_rss += entry->vmem_rss;
553 ps->vmem_data += entry->vmem_data;
554 ps->vmem_code += entry->vmem_code;
555 ps->stack_size += entry->stack_size;
557 if ((entry->io_rchar != -1) && (entry->io_wchar != -1)) {
558 ps_update_counter(&ps->io_rchar, &pse->io_rchar, entry->io_rchar);
559 ps_update_counter(&ps->io_wchar, &pse->io_wchar, entry->io_wchar);
562 if ((entry->io_syscr != -1) && (entry->io_syscw != -1)) {
563 ps_update_counter(&ps->io_syscr, &pse->io_syscr, entry->io_syscr);
564 ps_update_counter(&ps->io_syscw, &pse->io_syscw, entry->io_syscw);
567 if ((entry->io_diskr != -1) && (entry->io_diskw != -1)) {
568 ps_update_counter(&ps->io_diskr, &pse->io_diskr, entry->io_diskr);
569 ps_update_counter(&ps->io_diskw, &pse->io_diskw, entry->io_diskw);
572 if ((entry->cswitch_vol != -1) && (entry->cswitch_invol != -1)) {
573 ps_update_counter(&ps->cswitch_vol, &pse->cswitch_vol,
575 ps_update_counter(&ps->cswitch_invol, &pse->cswitch_invol,
576 entry->cswitch_invol);
579 ps_update_counter(&ps->vmem_minflt_counter, &pse->vmem_minflt_counter,
580 entry->vmem_minflt_counter);
581 ps_update_counter(&ps->vmem_majflt_counter, &pse->vmem_majflt_counter,
582 entry->vmem_majflt_counter);
584 ps_update_counter(&ps->cpu_user_counter, &pse->cpu_user_counter,
585 entry->cpu_user_counter);
586 ps_update_counter(&ps->cpu_system_counter, &pse->cpu_system_counter,
587 entry->cpu_system_counter);
589 #if HAVE_LIBTASKSTATS
590 ps_update_delay(ps, pse, entry);
595 /* remove old entries from instances of processes in list_head_g */
596 static void ps_list_reset(void) {
597 procstat_entry_t *pse;
598 procstat_entry_t *pse_prev;
600 for (procstat_t *ps = list_head_g; ps != NULL; ps = ps->next) {
612 ps->delay_blkio = NAN;
613 ps->delay_swapin = NAN;
614 ps->delay_freepages = NAN;
618 while (pse != NULL) {
620 DEBUG("Removing this procstat entry cause it's too old: "
621 "id = %lu; name = %s;",
624 if (pse_prev == NULL) {
625 ps->instances = pse->next;
629 pse_prev->next = pse->next;
631 pse = pse_prev->next;
638 } /* while (pse != NULL) */
639 } /* for (ps = list_head_g; ps != NULL; ps = ps->next) */
642 static void ps_tune_instance(oconfig_item_t *ci, procstat_t *ps) {
643 for (int i = 0; i < ci->children_num; i++) {
644 oconfig_item_t *c = ci->children + i;
646 if (strcasecmp(c->key, "CollectContextSwitch") == 0)
647 cf_util_get_boolean(c, &ps->report_ctx_switch);
648 else if (strcasecmp(c->key, "CollectFileDescriptor") == 0)
649 cf_util_get_boolean(c, &ps->report_fd_num);
650 else if (strcasecmp(c->key, "CollectMemoryMaps") == 0)
651 cf_util_get_boolean(c, &ps->report_maps_num);
652 else if (strcasecmp(c->key, "CollectDelayAccounting") == 0) {
653 #if HAVE_LIBTASKSTATS
654 cf_util_get_boolean(c, &ps->report_delay);
656 WARNING("processes plugin: The plugin has been compiled without support "
657 "for the \"CollectDelayAccounting\" option.");
660 ERROR("processes plugin: Option `%s' not allowed heeere.", c->key);
662 } /* for (ci->children) */
663 } /* void ps_tune_instance */
665 /* put all pre-defined 'Process' names from config to list_head_g tree */
666 static int ps_config(oconfig_item_t *ci) {
668 const size_t max_procname_len = 15;
669 #elif KERNEL_SOLARIS || KERNEL_FREEBSD
670 const size_t max_procname_len = MAXCOMLEN - 1;
675 for (int i = 0; i < ci->children_num; ++i) {
676 oconfig_item_t *c = ci->children + i;
678 if (strcasecmp(c->key, "Process") == 0) {
679 if ((c->values_num != 1) || (OCONFIG_TYPE_STRING != c->values[0].type)) {
680 ERROR("processes plugin: `Process' expects exactly "
681 "one string argument (got %i).",
686 #if KERNEL_LINUX || KERNEL_SOLARIS || KERNEL_FREEBSD
687 if (strlen(c->values[0].value.string) > max_procname_len) {
688 WARNING("processes plugin: this platform has a %zu character limit "
689 "to process names. The `Process \"%s\"' option will "
690 "not work as expected.",
691 max_procname_len, c->values[0].value.string);
695 ps = ps_list_register(c->values[0].value.string, NULL);
697 if (c->children_num != 0 && ps != NULL)
698 ps_tune_instance(c, ps);
699 } else if (strcasecmp(c->key, "ProcessMatch") == 0) {
700 if ((c->values_num != 2) || (OCONFIG_TYPE_STRING != c->values[0].type) ||
701 (OCONFIG_TYPE_STRING != c->values[1].type)) {
702 ERROR("processes plugin: `ProcessMatch' needs exactly "
703 "two string arguments (got %i).",
708 ps = ps_list_register(c->values[0].value.string,
709 c->values[1].value.string);
711 if (c->children_num != 0 && ps != NULL)
712 ps_tune_instance(c, ps);
713 } else if (strcasecmp(c->key, "CollectContextSwitch") == 0) {
714 cf_util_get_boolean(c, &report_ctx_switch);
715 } else if (strcasecmp(c->key, "CollectFileDescriptor") == 0) {
716 cf_util_get_boolean(c, &report_fd_num);
717 } else if (strcasecmp(c->key, "CollectMemoryMaps") == 0) {
718 cf_util_get_boolean(c, &report_maps_num);
719 } else if (strcasecmp(c->key, "CollectDelayAccounting") == 0) {
720 #if HAVE_LIBTASKSTATS
721 cf_util_get_boolean(c, &report_delay);
723 WARNING("processes plugin: The plugin has been compiled without support "
724 "for the \"CollectDelayAccounting\" option.");
727 ERROR("processes plugin: The `%s' configuration option is not "
728 "understood and will be ignored.",
737 static int ps_init(void) {
739 kern_return_t status;
741 port_host_self = mach_host_self();
742 port_task_self = mach_task_self();
744 if (pset_list != NULL) {
745 vm_deallocate(port_task_self, (vm_address_t)pset_list,
746 pset_list_len * sizeof(processor_set_t));
751 if ((status = host_processor_sets(port_host_self, &pset_list,
752 &pset_list_len)) != KERN_SUCCESS) {
753 ERROR("host_processor_sets failed: %s\n", mach_error_string(status));
758 /* #endif HAVE_THREAD_INFO */
761 pagesize_g = sysconf(_SC_PAGESIZE);
762 DEBUG("pagesize_g = %li; CONFIG_HZ = %i;", pagesize_g, CONFIG_HZ);
764 #if HAVE_LIBTASKSTATS
765 if (taskstats_handle == NULL) {
766 taskstats_handle = ts_create();
767 if (taskstats_handle == NULL) {
768 WARNING("processes plugin: Creating taskstats handle failed.");
772 /* #endif KERNEL_LINUX */
774 #elif HAVE_LIBKVM_GETPROCS && \
775 (HAVE_STRUCT_KINFO_PROC_FREEBSD || HAVE_STRUCT_KINFO_PROC_OPENBSD)
776 pagesize = getpagesize();
777 /* #endif HAVE_LIBKVM_GETPROCS && (HAVE_STRUCT_KINFO_PROC_FREEBSD ||
778 * HAVE_STRUCT_KINFO_PROC_OPENBSD) */
780 #elif HAVE_PROCINFO_H
781 pagesize = getpagesize();
782 #endif /* HAVE_PROCINFO_H */
787 /* submit global state (e.g.: qty of zombies, running, etc..) */
788 static void ps_submit_state(const char *state, double value) {
789 value_list_t vl = VALUE_LIST_INIT;
791 vl.values = &(value_t){.gauge = value};
793 sstrncpy(vl.plugin, "processes", sizeof(vl.plugin));
794 sstrncpy(vl.plugin_instance, "", sizeof(vl.plugin_instance));
795 sstrncpy(vl.type, "ps_state", sizeof(vl.type));
796 sstrncpy(vl.type_instance, state, sizeof(vl.type_instance));
798 plugin_dispatch_values(&vl);
801 /* submit info about specific process (e.g.: memory taken, cpu usage, etc..) */
802 static void ps_submit_proc_list(procstat_t *ps) {
803 value_list_t vl = VALUE_LIST_INIT;
807 sstrncpy(vl.plugin, "processes", sizeof(vl.plugin));
808 sstrncpy(vl.plugin_instance, ps->name, sizeof(vl.plugin_instance));
810 sstrncpy(vl.type, "ps_vm", sizeof(vl.type));
811 vl.values[0].gauge = ps->vmem_size;
813 plugin_dispatch_values(&vl);
815 sstrncpy(vl.type, "ps_rss", sizeof(vl.type));
816 vl.values[0].gauge = ps->vmem_rss;
818 plugin_dispatch_values(&vl);
820 sstrncpy(vl.type, "ps_data", sizeof(vl.type));
821 vl.values[0].gauge = ps->vmem_data;
823 plugin_dispatch_values(&vl);
825 sstrncpy(vl.type, "ps_code", sizeof(vl.type));
826 vl.values[0].gauge = ps->vmem_code;
828 plugin_dispatch_values(&vl);
830 sstrncpy(vl.type, "ps_stacksize", sizeof(vl.type));
831 vl.values[0].gauge = ps->stack_size;
833 plugin_dispatch_values(&vl);
835 sstrncpy(vl.type, "ps_cputime", sizeof(vl.type));
836 vl.values[0].derive = ps->cpu_user_counter;
837 vl.values[1].derive = ps->cpu_system_counter;
839 plugin_dispatch_values(&vl);
841 sstrncpy(vl.type, "ps_count", sizeof(vl.type));
842 vl.values[0].gauge = ps->num_proc;
843 vl.values[1].gauge = ps->num_lwp;
845 plugin_dispatch_values(&vl);
847 sstrncpy(vl.type, "ps_pagefaults", sizeof(vl.type));
848 vl.values[0].derive = ps->vmem_minflt_counter;
849 vl.values[1].derive = ps->vmem_majflt_counter;
851 plugin_dispatch_values(&vl);
853 if ((ps->io_rchar != -1) && (ps->io_wchar != -1)) {
854 sstrncpy(vl.type, "io_octets", sizeof(vl.type));
855 vl.values[0].derive = ps->io_rchar;
856 vl.values[1].derive = ps->io_wchar;
858 plugin_dispatch_values(&vl);
861 if ((ps->io_syscr != -1) && (ps->io_syscw != -1)) {
862 sstrncpy(vl.type, "io_ops", sizeof(vl.type));
863 vl.values[0].derive = ps->io_syscr;
864 vl.values[1].derive = ps->io_syscw;
866 plugin_dispatch_values(&vl);
869 if ((ps->io_diskr != -1) && (ps->io_diskw != -1)) {
870 sstrncpy(vl.type, "disk_octets", sizeof(vl.type));
871 vl.values[0].derive = ps->io_diskr;
872 vl.values[1].derive = ps->io_diskw;
874 plugin_dispatch_values(&vl);
877 if (ps->num_fd > 0) {
878 sstrncpy(vl.type, "file_handles", sizeof(vl.type));
879 vl.values[0].gauge = ps->num_fd;
881 plugin_dispatch_values(&vl);
884 if (ps->num_maps > 0) {
885 sstrncpy(vl.type, "file_handles", sizeof(vl.type));
886 sstrncpy(vl.type_instance, "mapped", sizeof(vl.type_instance));
887 vl.values[0].gauge = ps->num_maps;
889 plugin_dispatch_values(&vl);
892 if ((ps->cswitch_vol != -1) && (ps->cswitch_invol != -1)) {
893 sstrncpy(vl.type, "contextswitch", sizeof(vl.type));
894 sstrncpy(vl.type_instance, "voluntary", sizeof(vl.type_instance));
895 vl.values[0].derive = ps->cswitch_vol;
897 plugin_dispatch_values(&vl);
899 sstrncpy(vl.type, "contextswitch", sizeof(vl.type));
900 sstrncpy(vl.type_instance, "involuntary", sizeof(vl.type_instance));
901 vl.values[0].derive = ps->cswitch_invol;
903 plugin_dispatch_values(&vl);
906 /* The ps->delay_* metrics are in nanoseconds per second. This factor converts
907 * them to a percentage. */
908 gauge_t const delay_factor = 100.0 / 1000000000.0;
910 if (!isnan(ps->delay_cpu)) {
911 sstrncpy(vl.type, "percent", sizeof(vl.type));
912 sstrncpy(vl.type_instance, "delay-cpu", sizeof(vl.type_instance));
913 vl.values[0].gauge = ps->delay_cpu * delay_factor;
915 plugin_dispatch_values(&vl);
918 if (!isnan(ps->delay_blkio)) {
919 sstrncpy(vl.type, "percent", sizeof(vl.type));
920 sstrncpy(vl.type_instance, "delay-blkio", sizeof(vl.type_instance));
921 vl.values[0].gauge = ps->delay_blkio * delay_factor;
923 plugin_dispatch_values(&vl);
926 if (!isnan(ps->delay_swapin)) {
927 sstrncpy(vl.type, "percent", sizeof(vl.type));
928 sstrncpy(vl.type_instance, "delay-swapin", sizeof(vl.type_instance));
929 vl.values[0].gauge = ps->delay_swapin * delay_factor;
931 plugin_dispatch_values(&vl);
934 if (!isnan(ps->delay_freepages)) {
935 sstrncpy(vl.type, "percent", sizeof(vl.type));
936 sstrncpy(vl.type_instance, "delay-freepages", sizeof(vl.type_instance));
937 vl.values[0].gauge = ps->delay_freepages * delay_factor;
939 plugin_dispatch_values(&vl);
943 "name = %s; num_proc = %lu; num_lwp = %lu; num_fd = %lu; num_maps = %lu; "
944 "vmem_size = %lu; vmem_rss = %lu; vmem_data = %lu; "
946 "vmem_minflt_counter = %" PRIi64 "; vmem_majflt_counter = %" PRIi64 "; "
947 "cpu_user_counter = %" PRIi64 "; cpu_system_counter = %" PRIi64 "; "
948 "io_rchar = %" PRIi64 "; io_wchar = %" PRIi64 "; "
949 "io_syscr = %" PRIi64 "; io_syscw = %" PRIi64 "; "
950 "io_diskr = %" PRIi64 "; io_diskw = %" PRIi64 "; "
951 "cswitch_vol = %" PRIi64 "; cswitch_invol = %" PRIi64 "; "
952 "delay_cpu = %g; delay_blkio = %g; "
953 "delay_swapin = %g; delay_freepages = %g;",
954 ps->name, ps->num_proc, ps->num_lwp, ps->num_fd, ps->num_maps,
955 ps->vmem_size, ps->vmem_rss, ps->vmem_data, ps->vmem_code,
956 ps->vmem_minflt_counter, ps->vmem_majflt_counter, ps->cpu_user_counter,
957 ps->cpu_system_counter, ps->io_rchar, ps->io_wchar, ps->io_syscr,
958 ps->io_syscw, ps->io_diskr, ps->io_diskw, ps->cswitch_vol,
959 ps->cswitch_invol, ps->delay_cpu, ps->delay_blkio, ps->delay_swapin,
960 ps->delay_freepages);
962 } /* void ps_submit_proc_list */
964 #if KERNEL_LINUX || KERNEL_SOLARIS
965 static void ps_submit_fork_rate(derive_t value) {
966 value_list_t vl = VALUE_LIST_INIT;
968 vl.values = &(value_t){.derive = value};
970 sstrncpy(vl.plugin, "processes", sizeof(vl.plugin));
971 sstrncpy(vl.plugin_instance, "", sizeof(vl.plugin_instance));
972 sstrncpy(vl.type, "fork_rate", sizeof(vl.type));
973 sstrncpy(vl.type_instance, "", sizeof(vl.type_instance));
975 plugin_dispatch_values(&vl);
977 #endif /* KERNEL_LINUX || KERNEL_SOLARIS*/
979 /* ------- additional functions for KERNEL_LINUX/HAVE_THREAD_INFO ------- */
981 static int ps_read_tasks_status(process_entry_t *ps) {
987 derive_t cswitch_vol = 0;
988 derive_t cswitch_invol = 0;
993 snprintf(dirname, sizeof(dirname), "/proc/%li/task", ps->id);
995 if ((dh = opendir(dirname)) == NULL) {
996 DEBUG("Failed to open directory `%s'", dirname);
1000 while ((ent = readdir(dh)) != NULL) {
1003 if (!isdigit((int)ent->d_name[0]))
1008 if (snprintf(filename, sizeof(filename), "/proc/%li/task/%s/status", ps->id,
1009 tpid) >= sizeof(filename)) {
1010 DEBUG("Filename too long: `%s'", filename);
1014 if ((fh = fopen(filename, "r")) == NULL) {
1015 DEBUG("Failed to open file `%s'", filename);
1019 while (fgets(buffer, sizeof(buffer), fh) != NULL) {
1023 if (strncmp(buffer, "voluntary_ctxt_switches", 23) != 0 &&
1024 strncmp(buffer, "nonvoluntary_ctxt_switches", 26) != 0)
1027 numfields = strsplit(buffer, fields, STATIC_ARRAY_SIZE(fields));
1034 tmp = (derive_t)strtoll(fields[1], &endptr, /* base = */ 10);
1035 if ((errno == 0) && (endptr != fields[1])) {
1036 if (strncmp(buffer, "voluntary_ctxt_switches", 23) == 0) {
1038 } else if (strncmp(buffer, "nonvoluntary_ctxt_switches", 26) == 0) {
1039 cswitch_invol += tmp;
1042 } /* while (fgets) */
1045 WARNING("processes: fclose: %s", STRERRNO);
1050 ps->cswitch_vol = cswitch_vol;
1051 ps->cswitch_invol = cswitch_invol;
1054 } /* int *ps_read_tasks_status */
1056 /* Read data from /proc/pid/status */
1057 static int ps_read_status(long pid, process_entry_t *ps) {
1061 unsigned long lib = 0;
1062 unsigned long exe = 0;
1063 unsigned long data = 0;
1064 unsigned long threads = 0;
1068 snprintf(filename, sizeof(filename), "/proc/%li/status", pid);
1069 if ((fh = fopen(filename, "r")) == NULL)
1072 while (fgets(buffer, sizeof(buffer), fh) != NULL) {
1076 if (strncmp(buffer, "Vm", 2) != 0 && strncmp(buffer, "Threads", 7) != 0)
1079 numfields = strsplit(buffer, fields, STATIC_ARRAY_SIZE(fields));
1086 tmp = strtoul(fields[1], &endptr, /* base = */ 10);
1087 if ((errno == 0) && (endptr != fields[1])) {
1088 if (strncmp(buffer, "VmData", 6) == 0) {
1090 } else if (strncmp(buffer, "VmLib", 5) == 0) {
1092 } else if (strncmp(buffer, "VmExe", 5) == 0) {
1094 } else if (strncmp(buffer, "Threads", 7) == 0) {
1098 } /* while (fgets) */
1101 WARNING("processes: fclose: %s", STRERRNO);
1104 ps->vmem_data = data * 1024;
1105 ps->vmem_code = (exe + lib) * 1024;
1107 ps->num_lwp = threads;
1110 } /* int *ps_read_status */
1112 static int ps_read_io(process_entry_t *ps) {
1120 snprintf(filename, sizeof(filename), "/proc/%li/io", ps->id);
1121 if ((fh = fopen(filename, "r")) == NULL) {
1122 DEBUG("ps_read_io: Failed to open file `%s'", filename);
1126 while (fgets(buffer, sizeof(buffer), fh) != NULL) {
1127 derive_t *val = NULL;
1131 if (strncasecmp(buffer, "rchar:", 6) == 0)
1132 val = &(ps->io_rchar);
1133 else if (strncasecmp(buffer, "wchar:", 6) == 0)
1134 val = &(ps->io_wchar);
1135 else if (strncasecmp(buffer, "syscr:", 6) == 0)
1136 val = &(ps->io_syscr);
1137 else if (strncasecmp(buffer, "syscw:", 6) == 0)
1138 val = &(ps->io_syscw);
1139 else if (strncasecmp(buffer, "read_bytes:", 11) == 0)
1140 val = &(ps->io_diskr);
1141 else if (strncasecmp(buffer, "write_bytes:", 12) == 0)
1142 val = &(ps->io_diskw);
1146 numfields = strsplit(buffer, fields, STATIC_ARRAY_SIZE(fields));
1153 tmp = strtoll(fields[1], &endptr, /* base = */ 10);
1154 if ((errno != 0) || (endptr == fields[1]))
1157 *val = (derive_t)tmp;
1158 } /* while (fgets) */
1161 WARNING("processes: fclose: %s", STRERRNO);
1164 } /* int ps_read_io (...) */
1166 static int ps_count_maps(pid_t pid) {
1172 snprintf(filename, sizeof(filename), "/proc/%d/maps", pid);
1173 if ((fh = fopen(filename, "r")) == NULL) {
1174 DEBUG("ps_count_maps: Failed to open file `%s'", filename);
1178 while (fgets(buffer, sizeof(buffer), fh) != NULL) {
1179 if (strchr(buffer, '\n')) {
1182 } /* while (fgets) */
1185 WARNING("processes: fclose: %s", STRERRNO);
1188 } /* int ps_count_maps (...) */
1190 static int ps_count_fd(int pid) {
1196 snprintf(dirname, sizeof(dirname), "/proc/%i/fd", pid);
1198 if ((dh = opendir(dirname)) == NULL) {
1199 DEBUG("Failed to open directory `%s'", dirname);
1202 while ((ent = readdir(dh)) != NULL) {
1203 if (!isdigit((int)ent->d_name[0]))
1210 return (count >= 1) ? count : 1;
1211 } /* int ps_count_fd (pid) */
1213 #if HAVE_LIBTASKSTATS
1214 static int ps_delay(process_entry_t *ps) {
1215 if (taskstats_handle == NULL) {
1219 int status = ts_delay_by_tgid(taskstats_handle, (uint32_t)ps->id, &ps->delay);
1220 if (status == EPERM) {
1221 static c_complain_t c;
1222 #if defined(HAVE_SYS_CAPABILITY_H) && defined(CAP_NET_ADMIN)
1223 if (check_capability(CAP_NET_ADMIN) != 0) {
1224 if (getuid() == 0) {
1227 "processes plugin: Reading Delay Accounting metric failed: %s. "
1228 "collectd is running as root, but missing the CAP_NET_ADMIN "
1229 "capability. The most common cause for this is that the init "
1230 "system is dropping capabilities.",
1235 "processes plugin: Reading Delay Accounting metric failed: %s. "
1236 "collectd is not running as root and missing the CAP_NET_ADMIN "
1237 "capability. Either run collectd as root or grant it the "
1238 "CAP_NET_ADMIN capability using \"setcap cap_net_admin=ep " PREFIX
1239 "/sbin/collectd\".",
1243 ERROR("processes plugin: ts_delay_by_tgid failed: %s. The CAP_NET_ADMIN "
1244 "capability is available (I checked), so this error is utterly "
1249 c_complain(LOG_ERR, &c,
1250 "processes plugin: Reading Delay Accounting metric failed: %s. "
1251 "Reading Delay Accounting metrics requires root privileges.",
1255 } else if (status != 0) {
1256 ERROR("processes plugin: ts_delay_by_tgid failed: %s", STRERROR(status));
1263 static int ps_delay(__attribute__((unused)) process_entry_t *unused) {
1268 static void ps_fill_details(const procstat_t *ps, process_entry_t *entry) {
1269 if (entry->has_io == 0) {
1274 if (ps->report_ctx_switch) {
1275 if (entry->has_cswitch == 0) {
1276 ps_read_tasks_status(entry);
1277 entry->has_cswitch = 1;
1281 if (ps->report_maps_num) {
1283 if (entry->has_maps == 0 && (num_maps = ps_count_maps(entry->id)) > 0) {
1284 entry->num_maps = num_maps;
1286 entry->has_maps = 1;
1289 if (ps->report_fd_num) {
1291 if (entry->has_fd == 0 && (num_fd = ps_count_fd(entry->id)) > 0) {
1292 entry->num_fd = num_fd;
1297 #if HAVE_LIBTASKSTATS
1298 if (ps->report_delay && !entry->has_delay) {
1299 if (ps_delay(entry) == 0) {
1300 entry->has_delay = 1;
1304 } /* void ps_fill_details (...) */
1306 /* ps_read_process reads process counters on Linux. */
1307 static int ps_read_process(long pid, process_entry_t *ps, char *state) {
1317 size_t name_start_pos;
1318 size_t name_end_pos;
1321 derive_t cpu_user_counter;
1322 derive_t cpu_system_counter;
1323 long long unsigned vmem_size;
1324 long long unsigned vmem_rss;
1325 long long unsigned stack_size;
1329 snprintf(filename, sizeof(filename), "/proc/%li/stat", pid);
1331 status = read_file_contents(filename, buffer, sizeof(buffer) - 1);
1334 buffer_len = (size_t)status;
1335 buffer[buffer_len] = 0;
1337 /* The name of the process is enclosed in parens. Since the name can
1338 * contain parens itself, spaces, numbers and pretty much everything
1339 * else, use these to determine the process name. We don't use
1340 * strchr(3) and strrchr(3) to avoid pointer arithmetic which would
1341 * otherwise be required to determine name_len. */
1343 while (name_start_pos < buffer_len && buffer[name_start_pos] != '(')
1346 name_end_pos = buffer_len;
1347 while (name_end_pos > 0 && buffer[name_end_pos] != ')')
1350 /* Either '(' or ')' is not found or they are in the wrong order.
1351 * Anyway, something weird that shouldn't happen ever. */
1352 if (name_start_pos >= name_end_pos) {
1353 ERROR("processes plugin: name_start_pos = %zu >= name_end_pos = %zu",
1354 name_start_pos, name_end_pos);
1358 name_len = (name_end_pos - name_start_pos) - 1;
1359 if (name_len >= sizeof(ps->name))
1360 name_len = sizeof(ps->name) - 1;
1362 sstrncpy(ps->name, &buffer[name_start_pos + 1], name_len + 1);
1364 if ((buffer_len - name_end_pos) < 2)
1366 buffer_ptr = &buffer[name_end_pos + 2];
1368 fields_len = strsplit(buffer_ptr, fields, STATIC_ARRAY_SIZE(fields));
1369 if (fields_len < 22) {
1370 DEBUG("processes plugin: ps_read_process (pid = %li):"
1371 " `%s' has only %i fields..",
1372 pid, filename, fields_len);
1376 *state = fields[0][0];
1378 if (*state == 'Z') {
1382 ps->num_lwp = strtoul(fields[17], /* endptr = */ NULL, /* base = */ 10);
1383 if ((ps_read_status(pid, ps)) != 0) {
1387 DEBUG("ps_read_process: did not get vmem data for pid %li", pid);
1389 if (ps->num_lwp == 0)
1394 /* Leave the rest at zero if this is only a zombi */
1395 if (ps->num_proc == 0) {
1396 DEBUG("processes plugin: This is only a zombie: pid = %li; "
1402 cpu_user_counter = atoll(fields[11]);
1403 cpu_system_counter = atoll(fields[12]);
1404 vmem_size = atoll(fields[20]);
1405 vmem_rss = atoll(fields[21]);
1406 ps->vmem_minflt_counter = atol(fields[7]);
1407 ps->vmem_majflt_counter = atol(fields[9]);
1410 unsigned long long stack_start = atoll(fields[25]);
1411 unsigned long long stack_ptr = atoll(fields[26]);
1413 stack_size = (stack_start > stack_ptr) ? stack_start - stack_ptr
1414 : stack_ptr - stack_start;
1417 /* Convert jiffies to useconds */
1418 cpu_user_counter = cpu_user_counter * 1000000 / CONFIG_HZ;
1419 cpu_system_counter = cpu_system_counter * 1000000 / CONFIG_HZ;
1420 vmem_rss = vmem_rss * pagesize_g;
1422 ps->cpu_user_counter = cpu_user_counter;
1423 ps->cpu_system_counter = cpu_system_counter;
1424 ps->vmem_size = (unsigned long)vmem_size;
1425 ps->vmem_rss = (unsigned long)vmem_rss;
1426 ps->stack_size = (unsigned long)stack_size;
1428 /* no data by default. May be filled by ps_fill_details () */
1436 ps->cswitch_vol = -1;
1437 ps->cswitch_invol = -1;
1441 } /* int ps_read_process (...) */
1443 static char *ps_get_cmdline(long pid, char *name, char *buf, size_t buf_len) {
1447 char file[PATH_MAX];
1452 if ((pid < 1) || (NULL == buf) || (buf_len < 2))
1455 snprintf(file, sizeof(file), "/proc/%li/cmdline", pid);
1458 fd = open(file, O_RDONLY);
1460 /* ENOENT means the process exited while we were handling it.
1461 * Don't complain about this, it only fills the logs. */
1462 if (errno != ENOENT)
1463 WARNING("processes plugin: Failed to open `%s': %s.", file, STRERRNO);
1475 status = read(fd, (void *)buf_ptr, len);
1479 if ((EAGAIN == errno) || (EINTR == errno))
1482 WARNING("processes plugin: Failed to read from `%s': %s.", file,
1503 /* cmdline not available; e.g. kernel thread, zombie */
1507 snprintf(buf, buf_len, "[%s]", name);
1511 assert(n <= buf_len);
1518 /* remove trailing whitespace */
1519 while ((n > 0) && (isspace(buf[n]) || ('\0' == buf[n]))) {
1524 /* arguments are separated by '\0' in /proc/<pid>/cmdline */
1531 } /* char *ps_get_cmdline (...) */
1533 static int read_fork_rate(void) {
1537 _Bool value_valid = 0;
1539 proc_stat = fopen("/proc/stat", "r");
1540 if (proc_stat == NULL) {
1541 ERROR("processes plugin: fopen (/proc/stat) failed: %s", STRERRNO);
1545 while (fgets(buffer, sizeof(buffer), proc_stat) != NULL) {
1550 fields_num = strsplit(buffer, fields, STATIC_ARRAY_SIZE(fields));
1551 if (fields_num != 2)
1554 if (strcmp("processes", fields[0]) != 0)
1557 status = parse_value(fields[1], &value, DS_TYPE_DERIVE);
1568 ps_submit_fork_rate(value.derive);
1571 #endif /*KERNEL_LINUX */
1574 static char *ps_get_cmdline(long pid,
1575 char *name __attribute__((unused)), /* {{{ */
1576 char *buffer, size_t buffer_size) {
1577 char path[PATH_MAX];
1581 snprintf(path, sizeof(path), "/proc/%li/psinfo", pid);
1583 status = read_file_contents(path, (void *)&info, sizeof(info));
1584 if ((status < 0) || (((size_t)status) != sizeof(info))) {
1585 ERROR("processes plugin: Unexpected return value "
1586 "while reading \"%s\": "
1587 "Returned %zd but expected %zu.",
1588 path, status, buffer_size);
1592 info.pr_psargs[sizeof(info.pr_psargs) - 1] = 0;
1593 sstrncpy(buffer, info.pr_psargs, buffer_size);
1596 } /* }}} int ps_get_cmdline */
1599 * Reads process information on the Solaris OS. The information comes mainly
1601 * /proc/PID/status, /proc/PID/psinfo and /proc/PID/usage
1602 * The values for input and ouput chars are calculated "by hand"
1603 * Added a few "solaris" specific process states as well
1605 static int ps_read_process(long pid, process_entry_t *ps, char *state) {
1607 char f_psinfo[64], f_usage[64];
1610 pstatus_t *myStatus;
1614 snprintf(filename, sizeof(filename), "/proc/%li/status", pid);
1615 snprintf(f_psinfo, sizeof(f_psinfo), "/proc/%li/psinfo", pid);
1616 snprintf(f_usage, sizeof(f_usage), "/proc/%li/usage", pid);
1618 buffer = calloc(1, sizeof(pstatus_t));
1619 read_file_contents(filename, buffer, sizeof(pstatus_t));
1620 myStatus = (pstatus_t *)buffer;
1622 buffer = calloc(1, sizeof(psinfo_t));
1623 read_file_contents(f_psinfo, buffer, sizeof(psinfo_t));
1624 myInfo = (psinfo_t *)buffer;
1626 buffer = calloc(1, sizeof(prusage_t));
1627 read_file_contents(f_usage, buffer, sizeof(prusage_t));
1628 myUsage = (prusage_t *)buffer;
1630 sstrncpy(ps->name, myInfo->pr_fname, sizeof(myInfo->pr_fname));
1631 ps->num_lwp = myStatus->pr_nlwp;
1632 if (myInfo->pr_wstat != 0) {
1643 ps->num_lwp = myInfo->pr_nlwp;
1647 * Convert system time and user time from nanoseconds to microseconds
1648 * for compatibility with the linux module
1650 ps->cpu_system_counter = myStatus->pr_stime.tv_nsec / 1000;
1651 ps->cpu_user_counter = myStatus->pr_utime.tv_nsec / 1000;
1654 * Convert rssize from KB to bytes to be consistent w/ the linux module
1656 ps->vmem_rss = myInfo->pr_rssize * 1024;
1657 ps->vmem_size = myInfo->pr_size * 1024;
1658 ps->vmem_minflt_counter = myUsage->pr_minf;
1659 ps->vmem_majflt_counter = myUsage->pr_majf;
1662 * TODO: Data and code segment calculations for Solaris
1667 ps->stack_size = myStatus->pr_stksize;
1670 * TODO: File descriptor count for Solaris
1674 /* Number of memory mappings */
1678 * Calculating input/ouput chars
1679 * Formula used is total chars / total blocks => chars/block
1680 * then convert input/output blocks to chars
1682 ulong_t tot_chars = myUsage->pr_ioch;
1683 ulong_t tot_blocks = myUsage->pr_inblk + myUsage->pr_oublk;
1684 ulong_t chars_per_block = 1;
1685 if (tot_blocks != 0)
1686 chars_per_block = tot_chars / tot_blocks;
1687 ps->io_rchar = myUsage->pr_inblk * chars_per_block;
1688 ps->io_wchar = myUsage->pr_oublk * chars_per_block;
1689 ps->io_syscr = myUsage->pr_sysc;
1690 ps->io_syscw = myUsage->pr_sysc;
1695 * TODO: context switch counters for Solaris
1697 ps->cswitch_vol = -1;
1698 ps->cswitch_invol = -1;
1701 * TODO: Find way of setting BLOCKED and PAGING status
1705 if (myStatus->pr_flags & PR_ASLEEP)
1707 else if (myStatus->pr_flags & PR_STOPPED)
1709 else if (myStatus->pr_flags & PR_DETACH)
1711 else if (myStatus->pr_flags & PR_DAEMON)
1713 else if (myStatus->pr_flags & PR_ISSYS)
1715 else if (myStatus->pr_flags & PR_ORPHAN)
1726 * Reads the number of threads created since the last reboot. On Solaris these
1727 * are retrieved from kstat (module cpu, name sys, class misc, stat nthreads).
1728 * The result is the sum for all the threads created on each cpu
1730 static int read_fork_rate(void) {
1731 extern kstat_ctl_t *kc;
1732 derive_t result = 0;
1737 for (kstat_t *ksp_chain = kc->kc_chain; ksp_chain != NULL;
1738 ksp_chain = ksp_chain->ks_next) {
1739 if ((strcmp(ksp_chain->ks_module, "cpu") == 0) &&
1740 (strcmp(ksp_chain->ks_name, "sys") == 0) &&
1741 (strcmp(ksp_chain->ks_class, "misc") == 0)) {
1744 kstat_read(kc, ksp_chain, NULL);
1746 tmp = get_kstat_value(ksp_chain, "nthreads");
1752 ps_submit_fork_rate(result);
1755 #endif /* KERNEL_SOLARIS */
1757 #if HAVE_THREAD_INFO
1758 static int mach_get_task_name(task_t t, int *pid, char *name,
1759 size_t name_max_len) {
1762 struct kinfo_proc kp;
1767 mib[2] = KERN_PROC_PID;
1769 if (pid_for_task(t, pid) != KERN_SUCCESS)
1773 kp_size = sizeof(kp);
1774 if (sysctl(mib, 4, &kp, &kp_size, NULL, 0) != 0)
1777 if (name_max_len > (MAXCOMLEN + 1))
1778 name_max_len = MAXCOMLEN + 1;
1780 strncpy(name, kp.kp_proc.p_comm, name_max_len - 1);
1781 name[name_max_len - 1] = '\0';
1783 DEBUG("pid = %i; name = %s;", *pid, name);
1785 /* We don't do the special handling for `p_comm == "LaunchCFMApp"' as
1786 * `top' does it, because it is a lot of work and only used when
1787 * debugging. -octo */
1791 #endif /* HAVE_THREAD_INFO */
1792 /* end of additional functions for KERNEL_LINUX/HAVE_THREAD_INFO */
1794 /* do actual readings from kernel */
1795 static int ps_read(void) {
1796 #if HAVE_THREAD_INFO
1797 kern_return_t status;
1799 processor_set_t port_pset_priv;
1801 task_array_t task_list;
1802 mach_msg_type_number_t task_list_len;
1805 char task_name[MAXCOMLEN + 1];
1807 thread_act_array_t thread_list;
1808 mach_msg_type_number_t thread_list_len;
1809 thread_basic_info_data_t thread_data;
1810 mach_msg_type_number_t thread_data_len;
1819 process_entry_t pse;
1824 * The Mach-concept is a little different from the traditional UNIX
1825 * concept: All the work is done in threads. Threads are contained in
1826 * `tasks'. Therefore, `task status' doesn't make much sense, since
1827 * it's actually a `thread status'.
1828 * Tasks are assigned to sets of processors, so that's where you go to
1831 for (mach_msg_type_number_t pset = 0; pset < pset_list_len; pset++) {
1832 if ((status = host_processor_set_priv(port_host_self, pset_list[pset],
1833 &port_pset_priv)) != KERN_SUCCESS) {
1834 ERROR("host_processor_set_priv failed: %s\n", mach_error_string(status));
1838 if ((status = processor_set_tasks(port_pset_priv, &task_list,
1839 &task_list_len)) != KERN_SUCCESS) {
1840 ERROR("processor_set_tasks failed: %s\n", mach_error_string(status));
1841 mach_port_deallocate(port_task_self, port_pset_priv);
1845 for (mach_msg_type_number_t task = 0; task < task_list_len; task++) {
1847 if (mach_get_task_name(task_list[task], &task_pid, task_name,
1848 PROCSTAT_NAME_LEN) == 0) {
1849 /* search for at least one match */
1850 for (ps = list_head_g; ps != NULL; ps = ps->next)
1851 /* FIXME: cmdline should be here instead of NULL */
1852 if (ps_list_match(task_name, NULL, ps) == 1)
1856 /* Collect more detailed statistics for this process */
1858 task_basic_info_data_t task_basic_info;
1859 mach_msg_type_number_t task_basic_info_len;
1860 task_events_info_data_t task_events_info;
1861 mach_msg_type_number_t task_events_info_len;
1862 task_absolutetime_info_data_t task_absolutetime_info;
1863 mach_msg_type_number_t task_absolutetime_info_len;
1865 memset(&pse, '\0', sizeof(pse));
1868 task_basic_info_len = TASK_BASIC_INFO_COUNT;
1869 status = task_info(task_list[task], TASK_BASIC_INFO,
1870 (task_info_t)&task_basic_info, &task_basic_info_len);
1871 if (status != KERN_SUCCESS) {
1872 ERROR("task_info failed: %s", mach_error_string(status));
1873 continue; /* with next thread_list */
1876 task_events_info_len = TASK_EVENTS_INFO_COUNT;
1878 task_info(task_list[task], TASK_EVENTS_INFO,
1879 (task_info_t)&task_events_info, &task_events_info_len);
1880 if (status != KERN_SUCCESS) {
1881 ERROR("task_info failed: %s", mach_error_string(status));
1882 continue; /* with next thread_list */
1885 task_absolutetime_info_len = TASK_ABSOLUTETIME_INFO_COUNT;
1886 status = task_info(task_list[task], TASK_ABSOLUTETIME_INFO,
1887 (task_info_t)&task_absolutetime_info,
1888 &task_absolutetime_info_len);
1889 if (status != KERN_SUCCESS) {
1890 ERROR("task_info failed: %s", mach_error_string(status));
1891 continue; /* with next thread_list */
1895 pse.vmem_size = task_basic_info.virtual_size;
1896 pse.vmem_rss = task_basic_info.resident_size;
1897 /* Does not seem to be easily exposed */
1908 /* File descriptor count not implemented */
1911 /* Number of memory mappings */
1914 pse.vmem_minflt_counter = task_events_info.cow_faults;
1915 pse.vmem_majflt_counter = task_events_info.faults;
1917 pse.cpu_user_counter = task_absolutetime_info.total_user;
1918 pse.cpu_system_counter = task_absolutetime_info.total_system;
1920 /* context switch counters not implemented */
1921 pse.cswitch_vol = -1;
1922 pse.cswitch_invol = -1;
1925 status = task_threads(task_list[task], &thread_list, &thread_list_len);
1926 if (status != KERN_SUCCESS) {
1927 /* Apple's `top' treats this case a zombie. It
1928 * makes sense to some extend: A `zombie'
1929 * thread is nonsense, since the task/process
1932 DEBUG("task_threads failed: %s", mach_error_string(status));
1933 if (task_list[task] != port_task_self)
1934 mach_port_deallocate(port_task_self, task_list[task]);
1935 continue; /* with next task_list */
1938 for (mach_msg_type_number_t thread = 0; thread < thread_list_len;
1940 thread_data_len = THREAD_BASIC_INFO_COUNT;
1941 status = thread_info(thread_list[thread], THREAD_BASIC_INFO,
1942 (thread_info_t)&thread_data, &thread_data_len);
1943 if (status != KERN_SUCCESS) {
1944 ERROR("thread_info failed: %s", mach_error_string(status));
1945 if (task_list[task] != port_task_self)
1946 mach_port_deallocate(port_task_self, thread_list[thread]);
1947 continue; /* with next thread_list */
1953 switch (thread_data.run_state) {
1954 case TH_STATE_RUNNING:
1957 case TH_STATE_STOPPED:
1958 /* What exactly is `halted'? */
1959 case TH_STATE_HALTED:
1962 case TH_STATE_WAITING:
1965 case TH_STATE_UNINTERRUPTIBLE:
1968 /* There is no `zombie' case here,
1969 * since there are no zombie-threads.
1970 * There's only zombie tasks, which are
1973 WARNING("Unknown thread status: %i", thread_data.run_state);
1975 } /* switch (thread_data.run_state) */
1977 if (task_list[task] != port_task_self) {
1978 status = mach_port_deallocate(port_task_self, thread_list[thread]);
1979 if (status != KERN_SUCCESS)
1980 ERROR("mach_port_deallocate failed: %s", mach_error_string(status));
1982 } /* for (thread_list) */
1984 if ((status = vm_deallocate(port_task_self, (vm_address_t)thread_list,
1985 thread_list_len * sizeof(thread_act_t))) !=
1987 ERROR("vm_deallocate failed: %s", mach_error_string(status));
1990 thread_list_len = 0;
1992 /* Only deallocate the task port, if it isn't our own.
1993 * Don't know what would happen in that case, but this
1994 * is what Apple's top does.. ;) */
1995 if (task_list[task] != port_task_self) {
1996 status = mach_port_deallocate(port_task_self, task_list[task]);
1997 if (status != KERN_SUCCESS)
1998 ERROR("mach_port_deallocate failed: %s", mach_error_string(status));
2002 /* FIXME: cmdline should be here instead of NULL */
2003 ps_list_add(task_name, NULL, &pse);
2004 } /* for (task_list) */
2006 if ((status = vm_deallocate(port_task_self, (vm_address_t)task_list,
2007 task_list_len * sizeof(task_t))) !=
2009 ERROR("vm_deallocate failed: %s", mach_error_string(status));
2014 if ((status = mach_port_deallocate(port_task_self, port_pset_priv)) !=
2016 ERROR("mach_port_deallocate failed: %s", mach_error_string(status));
2018 } /* for (pset_list) */
2020 ps_submit_state("running", running);
2021 ps_submit_state("sleeping", sleeping);
2022 ps_submit_state("zombies", zombies);
2023 ps_submit_state("stopped", stopped);
2024 ps_submit_state("blocked", blocked);
2026 for (ps = list_head_g; ps != NULL; ps = ps->next)
2027 ps_submit_proc_list(ps);
2028 /* #endif HAVE_THREAD_INFO */
2042 char cmdline[CMDLINE_BUFFER_SIZE];
2045 process_entry_t pse;
2048 running = sleeping = zombies = stopped = paging = blocked = 0;
2051 if ((proc = opendir("/proc")) == NULL) {
2052 ERROR("Cannot open `/proc': %s", STRERRNO);
2056 while ((ent = readdir(proc)) != NULL) {
2057 if (!isdigit(ent->d_name[0]))
2060 if ((pid = atol(ent->d_name)) < 1)
2063 memset(&pse, 0, sizeof(pse));
2066 status = ps_read_process(pid, &pse, &state);
2068 DEBUG("ps_read_process failed: %i", status);
2093 ps_list_add(pse.name,
2094 ps_get_cmdline(pid, pse.name, cmdline, sizeof(cmdline)), &pse);
2099 ps_submit_state("running", running);
2100 ps_submit_state("sleeping", sleeping);
2101 ps_submit_state("zombies", zombies);
2102 ps_submit_state("stopped", stopped);
2103 ps_submit_state("paging", paging);
2104 ps_submit_state("blocked", blocked);
2106 for (procstat_t *ps_ptr = list_head_g; ps_ptr != NULL; ps_ptr = ps_ptr->next)
2107 ps_submit_proc_list(ps_ptr);
2110 /* #endif KERNEL_LINUX */
2112 #elif HAVE_LIBKVM_GETPROCS && HAVE_STRUCT_KINFO_PROC_FREEBSD
2122 char errbuf[_POSIX2_LINE_MAX];
2123 struct kinfo_proc *procs; /* array of processes */
2124 struct kinfo_proc *proc_ptr = NULL;
2125 int count; /* returns number of processes */
2127 process_entry_t pse;
2131 /* Open the kvm interface, get a descriptor */
2132 kd = kvm_openfiles(NULL, "/dev/null", NULL, 0, errbuf);
2134 ERROR("processes plugin: Cannot open kvm interface: %s", errbuf);
2138 /* Get the list of processes. */
2139 procs = kvm_getprocs(kd, KERN_PROC_ALL, 0, &count);
2140 if (procs == NULL) {
2141 ERROR("processes plugin: Cannot get kvm processes list: %s",
2147 /* Iterate through the processes in kinfo_proc */
2148 for (int i = 0; i < count; i++) {
2149 /* Create only one process list entry per _process_, i.e.
2150 * filter out threads (duplicate PID entries). */
2151 if ((proc_ptr == NULL) || (proc_ptr->ki_pid != procs[i].ki_pid)) {
2152 char cmdline[CMDLINE_BUFFER_SIZE] = "";
2153 _Bool have_cmdline = 0;
2155 proc_ptr = &(procs[i]);
2156 /* Don't probe system processes and processes without arguments */
2157 if (((procs[i].ki_flag & P_SYSTEM) == 0) && (procs[i].ki_args != NULL)) {
2162 /* retrieve the arguments */
2163 argv = kvm_getargv(kd, proc_ptr, /* nchr = */ 0);
2165 if ((argv != NULL) && (argv[0] != NULL)) {
2166 while (argv[argc] != NULL)
2169 status = strjoin(cmdline, sizeof(cmdline), argv, argc, " ");
2171 WARNING("processes plugin: Command line did not fit into buffer.");
2175 } /* if (process has argument list) */
2177 memset(&pse, 0, sizeof(pse));
2178 pse.id = procs[i].ki_pid;
2181 pse.num_lwp = procs[i].ki_numthreads;
2183 pse.vmem_size = procs[i].ki_size;
2184 pse.vmem_rss = procs[i].ki_rssize * pagesize;
2185 pse.vmem_data = procs[i].ki_dsize * pagesize;
2186 pse.vmem_code = procs[i].ki_tsize * pagesize;
2187 pse.stack_size = procs[i].ki_ssize * pagesize;
2188 pse.vmem_minflt_counter = procs[i].ki_rusage.ru_minflt;
2189 pse.vmem_majflt_counter = procs[i].ki_rusage.ru_majflt;
2191 pse.cpu_user_counter = 0;
2192 pse.cpu_system_counter = 0;
2194 * The u-area might be swapped out, and we can't get
2195 * at it because we have a crashdump and no swap.
2196 * If it's here fill in these fields, otherwise, just
2199 if (procs[i].ki_flag & P_INMEM) {
2200 pse.cpu_user_counter = procs[i].ki_rusage.ru_utime.tv_usec +
2201 (1000000lu * procs[i].ki_rusage.ru_utime.tv_sec);
2202 pse.cpu_system_counter =
2203 procs[i].ki_rusage.ru_stime.tv_usec +
2204 (1000000lu * procs[i].ki_rusage.ru_stime.tv_sec);
2215 /* file descriptor count not implemented */
2218 /* Number of memory mappings */
2221 /* context switch counters not implemented */
2222 pse.cswitch_vol = -1;
2223 pse.cswitch_invol = -1;
2225 ps_list_add(procs[i].ki_comm, have_cmdline ? cmdline : NULL, &pse);
2227 switch (procs[i].ki_stat) {
2250 } /* if ((proc_ptr == NULL) || (proc_ptr->ki_pid != procs[i].ki_pid)) */
2255 ps_submit_state("running", running);
2256 ps_submit_state("sleeping", sleeping);
2257 ps_submit_state("zombies", zombies);
2258 ps_submit_state("stopped", stopped);
2259 ps_submit_state("blocked", blocked);
2260 ps_submit_state("idle", idle);
2261 ps_submit_state("wait", wait);
2263 for (procstat_t *ps_ptr = list_head_g; ps_ptr != NULL; ps_ptr = ps_ptr->next)
2264 ps_submit_proc_list(ps_ptr);
2265 /* #endif HAVE_LIBKVM_GETPROCS && HAVE_STRUCT_KINFO_PROC_FREEBSD */
2267 #elif HAVE_LIBKVM_GETPROCS && HAVE_STRUCT_KINFO_PROC_OPENBSD
2278 struct kinfo_proc *procs; /* array of processes */
2279 struct kinfo_proc *proc_ptr = NULL;
2280 int count; /* returns number of processes */
2282 process_entry_t pse;
2286 /* Open the kvm interface, get a descriptor */
2287 kd = kvm_openfiles(NULL, NULL, NULL, KVM_NO_FILES, errbuf);
2289 ERROR("processes plugin: Cannot open kvm interface: %s", errbuf);
2293 /* Get the list of processes. */
2294 procs = kvm_getprocs(kd, KERN_PROC_ALL, 0, sizeof(struct kinfo_proc), &count);
2295 if (procs == NULL) {
2296 ERROR("processes plugin: Cannot get kvm processes list: %s",
2302 /* Iterate through the processes in kinfo_proc */
2303 for (int i = 0; i < count; i++) {
2304 /* Create only one process list entry per _process_, i.e.
2305 * filter out threads (duplicate PID entries). */
2306 if ((proc_ptr == NULL) || (proc_ptr->p_pid != procs[i].p_pid)) {
2307 char cmdline[CMDLINE_BUFFER_SIZE] = "";
2308 _Bool have_cmdline = 0;
2310 proc_ptr = &(procs[i]);
2311 /* Don't probe zombie processes */
2312 if (!P_ZOMBIE(proc_ptr)) {
2317 /* retrieve the arguments */
2318 argv = kvm_getargv(kd, proc_ptr, /* nchr = */ 0);
2320 if ((argv != NULL) && (argv[0] != NULL)) {
2321 while (argv[argc] != NULL)
2324 status = strjoin(cmdline, sizeof(cmdline), argv, argc, " ");
2326 WARNING("processes plugin: Command line did not fit into buffer.");
2330 } /* if (process has argument list) */
2332 memset(&pse, 0, sizeof(pse));
2333 pse.id = procs[i].p_pid;
2336 pse.num_lwp = 1; /* XXX: accumulate p_tid values for a single p_pid ? */
2338 pse.vmem_rss = procs[i].p_vm_rssize * pagesize;
2339 pse.vmem_data = procs[i].p_vm_dsize * pagesize;
2340 pse.vmem_code = procs[i].p_vm_tsize * pagesize;
2341 pse.stack_size = procs[i].p_vm_ssize * pagesize;
2342 pse.vmem_size = pse.stack_size + pse.vmem_code + pse.vmem_data;
2343 pse.vmem_minflt_counter = procs[i].p_uru_minflt;
2344 pse.vmem_majflt_counter = procs[i].p_uru_majflt;
2346 pse.cpu_user_counter =
2347 procs[i].p_uutime_usec + (1000000lu * procs[i].p_uutime_sec);
2348 pse.cpu_system_counter =
2349 procs[i].p_ustime_usec + (1000000lu * procs[i].p_ustime_sec);
2359 /* file descriptor count not implemented */
2362 /* Number of memory mappings */
2365 /* context switch counters not implemented */
2366 pse.cswitch_vol = -1;
2367 pse.cswitch_invol = -1;
2369 ps_list_add(procs[i].p_comm, have_cmdline ? cmdline : NULL, &pse);
2371 switch (procs[i].p_stat) {
2394 } /* if ((proc_ptr == NULL) || (proc_ptr->p_pid != procs[i].p_pid)) */
2399 ps_submit_state("running", running);
2400 ps_submit_state("sleeping", sleeping);
2401 ps_submit_state("zombies", zombies);
2402 ps_submit_state("stopped", stopped);
2403 ps_submit_state("onproc", onproc);
2404 ps_submit_state("idle", idle);
2405 ps_submit_state("dead", dead);
2407 for (procstat_t *ps_ptr = list_head_g; ps_ptr != NULL; ps_ptr = ps_ptr->next)
2408 ps_submit_proc_list(ps_ptr);
2409 /* #endif HAVE_LIBKVM_GETPROCS && HAVE_STRUCT_KINFO_PROC_OPENBSD */
2411 #elif HAVE_PROCINFO_H
2423 process_entry_t pse;
2426 while ((nprocs = getprocs64(procentry, sizeof(struct procentry64),
2427 /* fdsinfo = */ NULL, sizeof(struct fdsinfo64),
2428 &pindex, MAXPROCENTRY)) > 0) {
2429 for (int i = 0; i < nprocs; i++) {
2432 char arglist[MAXARGLN + 1];
2436 if (procentry[i].pi_state == SNONE)
2438 /* if (procentry[i].pi_state == SZOMB) FIXME */
2440 cmdline = procentry[i].pi_comm;
2441 cargs = procentry[i].pi_comm;
2442 if (procentry[i].pi_flags & SKPROC) {
2443 if (procentry[i].pi_pid == 0)
2444 cmdline = "swapper";
2447 if (getargs(&procentry[i], sizeof(struct procentry64), arglist,
2452 while (++n < MAXARGLN) {
2453 if (arglist[n] == '\0') {
2454 if (arglist[n + 1] == '\0')
2463 memset(&pse, 0, sizeof(pse));
2465 pse.id = procentry[i].pi_pid;
2466 pse.num_lwp = procentry[i].pi_thcount;
2470 while ((nthreads = getthrds64(procentry[i].pi_pid, thrdentry,
2471 sizeof(struct thrdentry64), &thindex,
2472 MAXTHRDENTRY)) > 0) {
2475 for (j = 0; j < nthreads; j++) {
2476 switch (thrdentry[j].ti_state) {
2477 /* case TSNONE: break; */
2480 break; /* FIXME is really blocked */
2498 if (nthreads < MAXTHRDENTRY)
2502 /* tv_usec is nanosec ??? */
2503 pse.cpu_user_counter = procentry[i].pi_ru.ru_utime.tv_sec * 1000000 +
2504 procentry[i].pi_ru.ru_utime.tv_usec / 1000;
2506 /* tv_usec is nanosec ??? */
2507 pse.cpu_system_counter = procentry[i].pi_ru.ru_stime.tv_sec * 1000000 +
2508 procentry[i].pi_ru.ru_stime.tv_usec / 1000;
2510 pse.vmem_minflt_counter = procentry[i].pi_minflt;
2511 pse.vmem_majflt_counter = procentry[i].pi_majflt;
2513 pse.vmem_size = procentry[i].pi_tsize + procentry[i].pi_dvm * pagesize;
2514 pse.vmem_rss = (procentry[i].pi_drss + procentry[i].pi_trss) * pagesize;
2515 /* Not supported/implemented */
2530 pse.cswitch_vol = -1;
2531 pse.cswitch_invol = -1;
2533 ps_list_add(cmdline, cargs, &pse);
2534 } /* for (i = 0 .. nprocs) */
2536 if (nprocs < MAXPROCENTRY)
2538 } /* while (getprocs64() > 0) */
2539 ps_submit_state("running", running);
2540 ps_submit_state("sleeping", sleeping);
2541 ps_submit_state("zombies", zombies);
2542 ps_submit_state("stopped", stopped);
2543 ps_submit_state("paging", paging);
2544 ps_submit_state("blocked", blocked);
2546 for (procstat_t *ps = list_head_g; ps != NULL; ps = ps->next)
2547 ps_submit_proc_list(ps);
2548 /* #endif HAVE_PROCINFO_H */
2550 #elif KERNEL_SOLARIS
2552 * The Solaris section adds a few more process states and removes some
2553 * process states compared to linux. Most notably there is no "PAGING"
2554 * and "BLOCKED" state for a process. The rest is similar to the linux
2572 char cmdline[PRARGSZ];
2576 proc = opendir("/proc");
2580 while ((ent = readdir(proc)) != NULL) {
2582 process_entry_t pse;
2585 if (!isdigit((int)ent->d_name[0]))
2588 pid = strtol(ent->d_name, &endptr, 10);
2589 if (*endptr != 0) /* value didn't completely parse as a number */
2592 memset(&pse, 0, sizeof(pse));
2595 status = ps_read_process(pid, &pse, &state);
2597 DEBUG("ps_read_process failed: %i", status);
2628 ps_list_add(pse.name,
2629 ps_get_cmdline(pid, pse.name, cmdline, sizeof(cmdline)), &pse);
2630 } /* while(readdir) */
2633 ps_submit_state("running", running);
2634 ps_submit_state("sleeping", sleeping);
2635 ps_submit_state("zombies", zombies);
2636 ps_submit_state("stopped", stopped);
2637 ps_submit_state("detached", detached);
2638 ps_submit_state("daemon", daemon);
2639 ps_submit_state("system", system);
2640 ps_submit_state("orphan", orphan);
2642 for (procstat_t *ps_ptr = list_head_g; ps_ptr != NULL; ps_ptr = ps_ptr->next)
2643 ps_submit_proc_list(ps_ptr);
2646 #endif /* KERNEL_SOLARIS */
2653 void module_register(void) {
2654 plugin_register_complex_config("processes", ps_config);
2655 plugin_register_init("processes", ps_init);
2656 plugin_register_read("processes", ps_read);
2657 } /* void module_register */