* Krzysztof Matczak <krzysztofx.matczak@intel.com>
*/
-#include "common.h"
#include "collectd.h"
+#include "common.h"
+#include "utils_llist.h"
+
#include <poll.h>
#include <sys/socket.h>
#include <sys/un.h>
#define MCELOG_DIMM_NAME "DMI_NAME"
#define MCELOG_CORRECTED_ERR "corrected memory errors"
#define MCELOG_UNCORRECTED_ERR "uncorrected memory errors"
+#define MCELOG_CORRECTED_ERR_TYPE_INS "corrected_memory_errors"
+#define MCELOG_UNCORRECTED_ERR_TYPE_INS "uncorrected_memory_errors"
typedef struct mcelog_config_s {
char logfile[PATH_MAX]; /* mcelog logfile */
pthread_t tid; /* poll thread id */
+ llist_t *dimms_list; /* DIMMs list */
+ /* lock for dimms cache */
+ pthread_mutex_t dimms_lock;
} mcelog_config_t;
typedef struct socket_adapter_s socket_adapter_t;
static int socket_reinit(socket_adapter_t *self);
static int socket_receive(socket_adapter_t *self, FILE **p_file);
-static mcelog_config_t g_mcelog_config = {.logfile = "/var/log/mcelog"};
+static mcelog_config_t g_mcelog_config = {
+ .logfile = "/var/log/mcelog",
+};
static socket_adapter_t socket_adapter = {
.sock_fd = -1,
static _Bool mcelog_thread_running;
+static void mcelog_free_dimms_list_records(llist_t *dimms_list) {
+
+ for (llentry_t *e = llist_head(dimms_list); e != NULL; e = e->next) {
+ sfree(e->key);
+ sfree(e->value);
+ }
+
+}
+
+/* Create or get dimm by dimm name/location */
+static llentry_t *mcelog_dimm(const mcelog_memory_rec_t *rec,
+ llist_t *dimms_list) {
+
+ char dimm_name[DATA_MAX_NAME_LEN];
+
+ if (strlen(rec->dimm_name) > 0) {
+ ssnprintf(dimm_name, sizeof(dimm_name), "%s_%s", rec->location,
+ rec->dimm_name);
+ } else
+ sstrncpy(dimm_name, rec->location, sizeof(dimm_name));
+
+ llentry_t *dimm_le = llist_search(g_mcelog_config.dimms_list, dimm_name);
+
+ if (dimm_le == NULL) {
+ mcelog_memory_rec_t *dimm_mr = calloc(1, sizeof(*dimm_mr));
+ if (dimm_mr == NULL) {
+ ERROR(MCELOG_PLUGIN ": Error allocating dimm memory item");
+ return NULL;
+ }
+ char *p_name = strdup(dimm_name);
+ if (p_name == NULL) {
+ ERROR(MCELOG_PLUGIN ": strdup: error");
+ return NULL;
+ }
+
+ /* add new dimm */
+ dimm_le = llentry_create(p_name, dimm_mr);
+ if (dimm_le == NULL) {
+ ERROR(MCELOG_PLUGIN ": llentry_create(): error");
+ free(dimm_mr);
+ return NULL;
+ }
+ pthread_mutex_lock(&g_mcelog_config.dimms_lock);
+ llist_append(g_mcelog_config.dimms_list, dimm_le);
+ pthread_mutex_unlock(&g_mcelog_config.dimms_lock);
+ }
+
+ return dimm_le;
+}
+
+static void mcelog_update_dimm_stats(llentry_t *dimm,
+ const mcelog_memory_rec_t *rec) {
+ pthread_mutex_lock(&g_mcelog_config.dimms_lock);
+ memcpy(dimm->value, rec, sizeof(mcelog_memory_rec_t));
+ pthread_mutex_unlock(&g_mcelog_config.dimms_lock);
+
+}
+
static int mcelog_config(oconfig_item_t *ci) {
for (int i = 0; i < ci->children_num; i++) {
oconfig_item_t *child = ci->children + i;
0) {
ERROR(MCELOG_PLUGIN ": Invalid configuration option: \"%s\".",
child->key);
- return -1;
+ return (-1);
}
} else if (strcasecmp("McelogLogfile", child->key) == 0) {
if (cf_util_get_string_buffer(child, g_mcelog_config.logfile,
sizeof(g_mcelog_config.logfile)) < 0) {
ERROR(MCELOG_PLUGIN ": Invalid configuration option: \"%s\".",
child->key);
- return -1;
+ return (-1);
}
} else {
ERROR(MCELOG_PLUGIN ": Invalid configuration option: \"%s\".",
child->key);
- return -1;
+ return (-1);
}
}
return (0);
}
}
pthread_rwlock_unlock(&self->lock);
- return ret;
+ return (ret);
}
static int socket_write(socket_adapter_t *self, const char *msg,
if (swrite(self->sock_fd, msg, len) < 0)
ret = -1;
pthread_rwlock_unlock(&self->lock);
- return ret;
+ return (ret);
}
static void mcelog_dispatch_notification(notification_t *n) {
ERROR(MCELOG_PLUGIN ": Could not create a socket. %s",
sstrerror(errno, errbuff, sizeof(errbuff)));
pthread_rwlock_unlock(&self->lock);
- return ret;
+ return (ret);
}
/* Set socket timeout option */
.type_instance = "mcelog_status"});
}
pthread_rwlock_unlock(&self->lock);
- return ret;
+ return (ret);
}
-static int mcelog_prepare_notification(notification_t *n,
- const mcelog_memory_rec_t *mr) {
- if (n == NULL || mr == NULL)
- return (-1);
+static int mcelog_dispatch_mem_notifications(const mcelog_memory_rec_t *mr) {
+ notification_t n = {.severity = NOTIF_WARNING,
+ .time = cdtime(),
+ .plugin = MCELOG_PLUGIN,
+ .type = "errors"};
- if (mr->location[0] != '\0')
- if (plugin_notification_meta_add_string(n, MCELOG_SOCKET_STR,
- mr->location) < 0) {
- ERROR(MCELOG_PLUGIN ": add memory location meta data failed");
- return (-1);
- }
- if (mr->dimm_name[0] != '\0')
- if (plugin_notification_meta_add_string(n, MCELOG_DIMM_NAME,
- mr->dimm_name) < 0) {
- ERROR(MCELOG_PLUGIN ": add DIMM name meta data failed");
- plugin_notification_meta_free(n->meta);
- return (-1);
- }
- if (plugin_notification_meta_add_signed_int(n, MCELOG_CORRECTED_ERR,
- mr->corrected_err_total) < 0) {
- ERROR(MCELOG_PLUGIN ": add corrected errors meta data failed");
- plugin_notification_meta_free(n->meta);
+ int dispatch_corrected_notifs = 0, dispatch_uncorrected_notifs = 0;
+
+ if (mr == NULL)
return (-1);
+
+ llentry_t *dimm = mcelog_dimm(mr, g_mcelog_config.dimms_list);
+ if (dimm == NULL) {
+ ERROR(MCELOG_PLUGIN ": Error adding/getting dimm memory item to/from cache");
+ return -1;
}
- if (plugin_notification_meta_add_signed_int(
- n, "corrected memory timed errors", mr->corrected_err_timed) < 0) {
- ERROR(MCELOG_PLUGIN ": add corrected timed errors meta data failed");
- plugin_notification_meta_free(n->meta);
- return (-1);
+
+ mcelog_memory_rec_t *mr_old = dimm->value;
+
+ if (mr_old->corrected_err_total != mr->corrected_err_total ||
+ mr_old->corrected_err_timed != mr->corrected_err_timed)
+ dispatch_corrected_notifs = 1;
+
+ if (mr_old->uncorrected_err_total != mr->uncorrected_err_total ||
+ mr_old->uncorrected_err_timed != mr->uncorrected_err_timed)
+ dispatch_uncorrected_notifs = 1;
+
+ if (!dispatch_corrected_notifs && !dispatch_uncorrected_notifs) {
+ DEBUG("%s: No new notifications to dispatch", MCELOG_PLUGIN);
+ return (0);
}
- if (mr->corrected_err_timed_period[0] != '\0')
- if (plugin_notification_meta_add_string(n, "corrected errors time period",
- mr->corrected_err_timed_period) <
- 0) {
- ERROR(MCELOG_PLUGIN ": add corrected errors period meta data failed");
- plugin_notification_meta_free(n->meta);
- return (-1);
+
+ sstrncpy(n.host, hostname_g, sizeof(n.host));
+
+ if (mr->dimm_name[0] != '\0')
+ ssnprintf(n.plugin_instance, sizeof(n.plugin_instance), "%s_%s",
+ mr->location, mr->dimm_name);
+ else
+ sstrncpy(n.plugin_instance, mr->location, sizeof(n.plugin_instance));
+
+ if (dispatch_corrected_notifs) {
+ /* Corrected Error Notifications */
+ if (mr->corrected_err_total > 0 || mr->corrected_err_timed > 0) {
+ if (plugin_notification_meta_add_signed_int(
+ &n, MCELOG_CORRECTED_ERR, mr->corrected_err_total) < 0) {
+ ERROR(MCELOG_PLUGIN ": add corrected errors meta data failed");
+ plugin_notification_meta_free(n.meta);
+ return (-1);
+ }
+ if (plugin_notification_meta_add_signed_int(
+ &n, "corrected memory timed errors", mr->corrected_err_timed) <
+ 0) {
+ ERROR(MCELOG_PLUGIN ": add corrected timed errors meta data failed");
+ plugin_notification_meta_free(n.meta);
+ return (-1);
+ }
+ ssnprintf(n.message, sizeof(n.message), "Corrected Memory Errors");
+ sstrncpy(n.type_instance, MCELOG_CORRECTED_ERR_TYPE_INS,
+ sizeof(n.type_instance));
+ plugin_dispatch_notification(&n);
+
+ if (n.meta)
+ plugin_notification_meta_free(n.meta);
}
- if (plugin_notification_meta_add_signed_int(n, MCELOG_UNCORRECTED_ERR,
- mr->uncorrected_err_total) < 0) {
- ERROR(MCELOG_PLUGIN ": add corrected errors meta data failed");
- plugin_notification_meta_free(n->meta);
- return (-1);
- }
- if (plugin_notification_meta_add_signed_int(n,
- "uncorrected memory timed errors",
- mr->uncorrected_err_timed) < 0) {
- ERROR(MCELOG_PLUGIN ": add corrected timed errors meta data failed");
- plugin_notification_meta_free(n->meta);
- return (-1);
}
- if (mr->uncorrected_err_timed_period[0] != '\0')
- if (plugin_notification_meta_add_string(n, "uncorrected errors time period",
- mr->uncorrected_err_timed_period) <
- 0) {
- ERROR(MCELOG_PLUGIN ": add corrected errors period meta data failed");
- plugin_notification_meta_free(n->meta);
- return (-1);
+
+ if (dispatch_uncorrected_notifs) {
+ /* Uncorrected Error Notifications */
+ if (mr->uncorrected_err_total > 0 || mr->uncorrected_err_timed > 0) {
+ if (plugin_notification_meta_add_signed_int(
+ &n, MCELOG_UNCORRECTED_ERR, mr->uncorrected_err_total) < 0) {
+ ERROR(MCELOG_PLUGIN ": add uncorrected errors meta data failed");
+ plugin_notification_meta_free(n.meta);
+ return (-1);
+ }
+ if (plugin_notification_meta_add_signed_int(
+ &n, "uncorrected memory timed errors",
+ mr->uncorrected_err_timed) < 0) {
+ ERROR(MCELOG_PLUGIN ": add uncorrected timed errors meta data failed");
+ plugin_notification_meta_free(n.meta);
+ return (-1);
+ }
+ ssnprintf(n.message, sizeof(n.message), "Uncorrected Memory Errors");
+ sstrncpy(n.type_instance, MCELOG_UNCORRECTED_ERR_TYPE_INS,
+ sizeof(n.type_instance));
+ n.severity = NOTIF_FAILURE;
+ plugin_dispatch_notification(&n);
+
+ if (n.meta)
+ plugin_notification_meta_free(n.meta);
}
+ }
return (0);
}
if (!mr) {
ERROR(MCELOG_PLUGIN ": %s: NULL pointer", __FUNCTION__);
- return -1;
+ return (-1);
+ }
+
+ llentry_t *dimm = mcelog_dimm(mr, g_mcelog_config.dimms_list);
+ if (dimm == NULL) {
+ ERROR(MCELOG_PLUGIN ": Error adding/getting dimm memory item to/from cache");
+ return -1;
}
value_list_t vl = {
.time = cdtime(),
.plugin = MCELOG_PLUGIN,
.type = "errors",
- .type_instance = "corrected_memory_errors"};
+ .type_instance = MCELOG_CORRECTED_ERR_TYPE_INS};
+
+ mcelog_update_dimm_stats(dimm, mr);
if (mr->dimm_name[0] != '\0')
ssnprintf(vl.plugin_instance, sizeof(vl.plugin_instance), "%s_%s",
vl.values = &(value_t){.derive = (derive_t)mr->corrected_err_timed};
plugin_dispatch_values(&vl);
- sstrncpy(vl.type_instance, "uncorrected_memory_errors",
+ sstrncpy(vl.type_instance, MCELOG_UNCORRECTED_ERR_TYPE_INS,
sizeof(vl.type_instance));
vl.values = &(value_t){.derive = (derive_t)mr->uncorrected_err_total};
plugin_dispatch_values(&vl);
vl.values = &(value_t){.derive = (derive_t)mr->uncorrected_err_timed};
plugin_dispatch_values(&vl);
- return 0;
+ return (0);
}
static int parse_memory_info(FILE *p_file, mcelog_memory_rec_t *memory_record) {
/* Got empty line or "done" */
if ((!strncmp("\n", buf, strlen(buf))) ||
(!strncmp(buf, "done\n", strlen(buf))))
- return 1;
+ return (1);
if (strlen(buf) < 5)
continue;
if (!strncmp(buf, MCELOG_SOCKET_STR, strlen(MCELOG_SOCKET_STR))) {
memset(buf, 0, sizeof(buf));
}
/* parsing definitely finished */
- return 0;
+ return (0);
}
static void poll_worker_cleanup(void *arg) {
sstrerror(errno, errbuf, sizeof(errbuf)));
}
pthread_rwlock_unlock(&self->lock);
- return res;
+ return (res);
}
if (poll_fd.revents & (POLLERR | POLLHUP | POLLNVAL)) {
.type_instance = "mcelog_status"});
}
pthread_rwlock_unlock(&self->lock);
- return -1;
+ return (-1);
}
if (!(poll_fd.revents & (POLLIN | POLLPRI))) {
INFO(MCELOG_PLUGIN ": No data to read");
pthread_rwlock_unlock(&self->lock);
- return 0;
+ return (0);
}
if ((*pp_file = fdopen(dup(self->sock_fd), "r")) == NULL)
res = -1;
pthread_rwlock_unlock(&self->lock);
- return res;
+ return (res);
}
static void *poll_worker(__attribute__((unused)) void *arg) {
continue;
}
- notification_t n = {.severity = NOTIF_OKAY,
- .time = cdtime(),
- .message = "Got memory errors info.",
- .plugin = MCELOG_PLUGIN,
- .type_instance = "memory_erros"};
-
- if (mcelog_prepare_notification(&n, &memory_record) == 0)
- mcelog_dispatch_notification(&n);
+ if (mcelog_dispatch_mem_notifications(&memory_record) != 0)
+ ERROR(MCELOG_PLUGIN ": Failed to submit memory errors notification");
if (mcelog_submit(&memory_record) != 0)
ERROR(MCELOG_PLUGIN ": Failed to submit memory errors");
memset(&memory_record, 0, sizeof(memory_record));
mcelog_thread_running = 0;
pthread_cleanup_pop(1);
- return NULL;
+ return (NULL);
}
static int mcelog_init(void) {
+ g_mcelog_config.dimms_list = llist_create();
+ int err = pthread_mutex_init(&g_mcelog_config.dimms_lock, NULL);
+ if (err < 0) {
+ ERROR(MCELOG_PLUGIN ": plugin: failed to initialize cache lock");
+ return (-1);
+ }
+
if (socket_adapter.reinit(&socket_adapter) != 0) {
ERROR(MCELOG_PLUGIN ": Cannot connect to client socket");
- return -1;
+ return (-1);
}
if (plugin_thread_create(&g_mcelog_config.tid, NULL, poll_worker, NULL,
NULL) != 0) {
ERROR(MCELOG_PLUGIN ": Error creating poll thread.");
- return -1;
+ return (-1);
}
- return 0;
+ return (0);
}
static int get_memory_machine_checks(void) {
ERROR(MCELOG_PLUGIN ": SENT DUMP REQUEST FAILED");
else
DEBUG(MCELOG_PLUGIN ": SENT DUMP REQUEST OK");
- return ret;
+ return (ret);
}
static int mcelog_read(__attribute__((unused)) user_data_t *ud) {
if (get_memory_machine_checks() != 0)
ERROR(MCELOG_PLUGIN ": MACHINE CHECK INFO NOT AVAILABLE");
- return 0;
+ return (0);
}
static int mcelog_shutdown(void) {
ret = -1;
}
}
-
+ pthread_mutex_lock(&g_mcelog_config.dimms_lock);
+ mcelog_free_dimms_list_records(g_mcelog_config.dimms_list);
+ llist_destroy(g_mcelog_config.dimms_list);
+ pthread_mutex_unlock(&g_mcelog_config.dimms_lock);
+ pthread_mutex_destroy(&g_mcelog_config.dimms_lock);
+ g_mcelog_config.dimms_list = NULL;
ret = socket_adapter.close(&socket_adapter) || ret;
pthread_rwlock_destroy(&(socket_adapter.lock));
- return -ret;
+ return (-ret);
}
void module_register(void) {
projects / collectd.git / blobdiff