23040c81bea765db8a9ac1c197419b674e47c5c5
1 /*-
2 * collectd - src/mcelog.c
3 * MIT License
4 *
5 * Copyright(c) 2016-2017 Intel Corporation. All rights reserved.
6 *
7 * Permission is hereby granted, free of charge, to any person obtaining a
8 * copy of this software and associated documentation files (the "Software"),
9 * to deal in the Software without restriction, including without limitation
10 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
11 * and/or sell copies of the Software, and to permit persons to whom the
12 * Software is furnished to do so, subject to the following conditions:
13 *
14 * The above copyright notice and this permission notice shall be included in
15 * all copies or substantial portions of the Software.
16 *
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
20 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
22 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
23 * DEALINGS IN THE SOFTWARE.
25 * Authors:
26 * Maryam Tahhan <maryam.tahhan@intel.com>
27 * Volodymyr Mytnyk <volodymyrx.mytnyk@intel.com>
28 * Taras Chornyi <tarasx.chornyi@intel.com>
29 * Krzysztof Matczak <krzysztofx.matczak@intel.com>
30 */
32 #include "collectd.h"
34 #include "common.h"
35 #include "utils_llist.h"
37 #include <poll.h>
38 #include <sys/socket.h>
39 #include <sys/un.h>
40 #include <unistd.h>
42 #define MCELOG_PLUGIN "mcelog"
43 #define MCELOG_BUFF_SIZE 1024
44 #define MCELOG_POLL_TIMEOUT 1000 /* ms */
45 #define MCELOG_SOCKET_STR "SOCKET"
46 #define MCELOG_DIMM_NAME "DMI_NAME"
47 #define MCELOG_CORRECTED_ERR "corrected memory errors"
48 #define MCELOG_UNCORRECTED_ERR "uncorrected memory errors"
49 #define MCELOG_CORRECTED_ERR_TYPE_INS "corrected_memory_errors"
50 #define MCELOG_UNCORRECTED_ERR_TYPE_INS "uncorrected_memory_errors"
52 typedef struct mcelog_config_s {
53 char logfile[PATH_MAX]; /* mcelog logfile */
54 pthread_t tid; /* poll thread id */
55 llist_t *dimms_list; /* DIMMs list */
56 /* lock for dimms cache */
57 _Bool persist;
58 pthread_mutex_t dimms_lock;
59 } mcelog_config_t;
61 typedef struct socket_adapter_s socket_adapter_t;
63 struct socket_adapter_s {
64 int sock_fd; /* mcelog server socket fd */
65 struct sockaddr_un unix_sock; /* mcelog client socket */
66 pthread_rwlock_t lock;
67 /* function pointers for socket operations */
68 int (*write)(socket_adapter_t *self, const char *msg, const size_t len);
69 int (*reinit)(socket_adapter_t *self);
70 int (*receive)(socket_adapter_t *self, FILE **p_file);
71 int (*close)(socket_adapter_t *self);
72 };
74 typedef struct mcelog_memory_rec_s {
75 int corrected_err_total; /* x total*/
76 int corrected_err_timed; /* x in 24h*/
77 char corrected_err_timed_period[DATA_MAX_NAME_LEN];
78 int uncorrected_err_total; /* x total*/
79 int uncorrected_err_timed; /* x in 24h*/
80 char uncorrected_err_timed_period[DATA_MAX_NAME_LEN];
81 char location[DATA_MAX_NAME_LEN]; /* SOCKET x CHANNEL x DIMM x*/
82 char dimm_name[DATA_MAX_NAME_LEN]; /* DMI_NAME "DIMM_F1" */
83 } mcelog_memory_rec_t;
85 static int socket_close(socket_adapter_t *self);
86 static int socket_write(socket_adapter_t *self, const char *msg,
87 const size_t len);
88 static int socket_reinit(socket_adapter_t *self);
89 static int socket_receive(socket_adapter_t *self, FILE **p_file);
91 static mcelog_config_t g_mcelog_config = {
92 .logfile = "/var/log/mcelog", .persist = 0,
93 };
95 static socket_adapter_t socket_adapter = {
96 .sock_fd = -1,
97 .unix_sock =
98 {
99 .sun_family = AF_UNIX, .sun_path = "/var/run/mcelog-client",
100 },
101 .lock = PTHREAD_RWLOCK_INITIALIZER,
102 .close = socket_close,
103 .write = socket_write,
104 .reinit = socket_reinit,
105 .receive = socket_receive,
106 };
108 static _Bool mcelog_thread_running;
110 static void mcelog_free_dimms_list_records(llist_t *dimms_list) {
112 for (llentry_t *e = llist_head(dimms_list); e != NULL; e = e->next) {
113 sfree(e->key);
114 sfree(e->value);
115 }
116 }
118 /* Create or get dimm by dimm name/location */
119 static llentry_t *mcelog_dimm(const mcelog_memory_rec_t *rec,
120 llist_t *dimms_list) {
122 char dimm_name[DATA_MAX_NAME_LEN];
124 if (strlen(rec->dimm_name) > 0) {
125 ssnprintf(dimm_name, sizeof(dimm_name), "%s_%s", rec->location,
126 rec->dimm_name);
127 } else
128 sstrncpy(dimm_name, rec->location, sizeof(dimm_name));
130 llentry_t *dimm_le = llist_search(g_mcelog_config.dimms_list, dimm_name);
132 if (dimm_le == NULL) {
133 mcelog_memory_rec_t *dimm_mr = calloc(1, sizeof(*dimm_mr));
134 if (dimm_mr == NULL) {
135 ERROR(MCELOG_PLUGIN ": Error allocating dimm memory item");
136 return NULL;
137 }
138 char *p_name = strdup(dimm_name);
139 if (p_name == NULL) {
140 ERROR(MCELOG_PLUGIN ": strdup: error");
141 return NULL;
142 }
144 /* add new dimm */
145 dimm_le = llentry_create(p_name, dimm_mr);
146 if (dimm_le == NULL) {
147 ERROR(MCELOG_PLUGIN ": llentry_create(): error");
148 free(dimm_mr);
149 return NULL;
150 }
151 pthread_mutex_lock(&g_mcelog_config.dimms_lock);
152 llist_append(g_mcelog_config.dimms_list, dimm_le);
153 pthread_mutex_unlock(&g_mcelog_config.dimms_lock);
154 }
156 return dimm_le;
157 }
159 static void mcelog_update_dimm_stats(llentry_t *dimm,
160 const mcelog_memory_rec_t *rec) {
161 pthread_mutex_lock(&g_mcelog_config.dimms_lock);
162 memcpy(dimm->value, rec, sizeof(mcelog_memory_rec_t));
163 pthread_mutex_unlock(&g_mcelog_config.dimms_lock);
164 }
166 static int mcelog_config(oconfig_item_t *ci) {
167 for (int i = 0; i < ci->children_num; i++) {
168 oconfig_item_t *child = ci->children + i;
169 if (strcasecmp("McelogLogfile", child->key) == 0) {
170 if (cf_util_get_string_buffer(child, g_mcelog_config.logfile,
171 sizeof(g_mcelog_config.logfile)) < 0) {
172 ERROR(MCELOG_PLUGIN ": Invalid configuration option: \"%s\".",
173 child->key);
174 return (-1);
175 }
176 } else if (strcasecmp("Memory", child->key) == 0) {
177 oconfig_item_t *mem_child = child->children;
178 for (int j = 0; j < child->children_num; j++) {
179 mem_child += j;
180 if (strcasecmp("McelogClientSocket", mem_child->key) == 0) {
181 if (cf_util_get_string_buffer(
182 mem_child, socket_adapter.unix_sock.sun_path,
183 sizeof(socket_adapter.unix_sock.sun_path)) < 0) {
184 ERROR(MCELOG_PLUGIN ": Invalid configuration option: \"%s\".",
185 mem_child->key);
186 return (-1);
187 }
188 } else if (strcasecmp("PersistentNotification", mem_child->key) == 0) {
189 if (cf_util_get_boolean(mem_child, &g_mcelog_config.persist) < 0) {
190 ERROR(MCELOG_PLUGIN ": Invalid configuration option: \"%s\".",
191 mem_child->key);
192 return (-1);
193 }
194 } else {
195 ERROR(MCELOG_PLUGIN ": Invalid Memory configuration option: \"%s\".",
196 mem_child->key);
197 return (-1);
198 }
199 }
200 } else {
201 ERROR(MCELOG_PLUGIN ": Invalid configuration option: \"%s\".",
202 child->key);
203 return (-1);
204 }
205 }
206 return (0);
207 }
209 static int socket_close(socket_adapter_t *self) {
210 int ret = 0;
211 pthread_rwlock_rdlock(&self->lock);
212 if (fcntl(self->sock_fd, F_GETFL) != -1) {
213 char errbuf[MCELOG_BUFF_SIZE];
214 if (shutdown(self->sock_fd, SHUT_RDWR) != 0) {
215 ERROR(MCELOG_PLUGIN ": Socket shutdown failed: %s",
216 sstrerror(errno, errbuf, sizeof(errbuf)));
217 ret = -1;
218 }
219 if (close(self->sock_fd) != 0) {
220 ERROR(MCELOG_PLUGIN ": Socket close failed: %s",
221 sstrerror(errno, errbuf, sizeof(errbuf)));
222 ret = -1;
223 }
224 }
225 pthread_rwlock_unlock(&self->lock);
226 return (ret);
227 }
229 static int socket_write(socket_adapter_t *self, const char *msg,
230 const size_t len) {
231 int ret = 0;
232 pthread_rwlock_rdlock(&self->lock);
233 if (swrite(self->sock_fd, msg, len) < 0)
234 ret = -1;
235 pthread_rwlock_unlock(&self->lock);
236 return (ret);
237 }
239 static void mcelog_dispatch_notification(notification_t *n) {
240 if (!n) {
241 ERROR(MCELOG_PLUGIN ": %s: NULL pointer", __FUNCTION__);
242 return;
243 }
245 sstrncpy(n->host, hostname_g, sizeof(n->host));
246 sstrncpy(n->type, "gauge", sizeof(n->type));
247 plugin_dispatch_notification(n);
248 if (n->meta)
249 plugin_notification_meta_free(n->meta);
250 }
252 static int socket_reinit(socket_adapter_t *self) {
253 char errbuff[MCELOG_BUFF_SIZE];
254 int ret = -1;
255 cdtime_t interval = plugin_get_interval();
256 struct timeval socket_timeout = CDTIME_T_TO_TIMEVAL(interval);
258 /* synchronization via write lock since sock_fd may be changed here */
259 pthread_rwlock_wrlock(&self->lock);
260 self->sock_fd =
261 socket(PF_UNIX, SOCK_STREAM | SOCK_CLOEXEC | SOCK_NONBLOCK, 0);
262 if (self->sock_fd < 0) {
263 ERROR(MCELOG_PLUGIN ": Could not create a socket. %s",
264 sstrerror(errno, errbuff, sizeof(errbuff)));
265 pthread_rwlock_unlock(&self->lock);
266 return (ret);
267 }
269 /* Set socket timeout option */
270 if (setsockopt(self->sock_fd, SOL_SOCKET, SO_SNDTIMEO, &socket_timeout,
271 sizeof(socket_timeout)) < 0)
272 ERROR(MCELOG_PLUGIN ": Failed to set the socket timeout option.");
274 /* downgrading to read lock due to possible recursive read locks
275 * in self->close(self) call */
276 pthread_rwlock_unlock(&self->lock);
277 pthread_rwlock_rdlock(&self->lock);
278 if (connect(self->sock_fd, (struct sockaddr *)&(self->unix_sock),
279 sizeof(self->unix_sock)) < 0) {
280 ERROR(MCELOG_PLUGIN ": Failed to connect to mcelog server. %s",
281 sstrerror(errno, errbuff, sizeof(errbuff)));
282 self->close(self);
283 ret = -1;
284 } else {
285 ret = 0;
286 mcelog_dispatch_notification(
287 &(notification_t){.severity = NOTIF_OKAY,
288 .time = cdtime(),
289 .message = "Connected to mcelog server",
290 .plugin = MCELOG_PLUGIN,
291 .type_instance = "mcelog_status"});
292 }
293 pthread_rwlock_unlock(&self->lock);
294 return (ret);
295 }
297 static int mcelog_dispatch_mem_notifications(const mcelog_memory_rec_t *mr) {
298 notification_t n = {.severity = NOTIF_WARNING,
299 .time = cdtime(),
300 .plugin = MCELOG_PLUGIN,
301 .type = "errors"};
303 int dispatch_corrected_notifs = 0, dispatch_uncorrected_notifs = 0;
305 if (mr == NULL)
306 return (-1);
308 llentry_t *dimm = mcelog_dimm(mr, g_mcelog_config.dimms_list);
309 if (dimm == NULL) {
310 ERROR(MCELOG_PLUGIN
311 ": Error adding/getting dimm memory item to/from cache");
312 return (-1);
313 }
314 mcelog_memory_rec_t *mr_old = dimm->value;
315 if (!g_mcelog_config.persist) {
317 if (mr_old->corrected_err_total != mr->corrected_err_total ||
318 mr_old->corrected_err_timed != mr->corrected_err_timed)
319 dispatch_corrected_notifs = 1;
321 if (mr_old->uncorrected_err_total != mr->uncorrected_err_total ||
322 mr_old->uncorrected_err_timed != mr->uncorrected_err_timed)
323 dispatch_uncorrected_notifs = 1;
325 if (!dispatch_corrected_notifs && !dispatch_uncorrected_notifs) {
326 DEBUG("%s: No new notifications to dispatch", MCELOG_PLUGIN);
327 return (0);
328 }
329 } else {
330 dispatch_corrected_notifs = 1;
331 dispatch_uncorrected_notifs = 1;
332 }
334 sstrncpy(n.host, hostname_g, sizeof(n.host));
336 if (mr->dimm_name[0] != '\0')
337 ssnprintf(n.plugin_instance, sizeof(n.plugin_instance), "%s_%s",
338 mr->location, mr->dimm_name);
339 else
340 sstrncpy(n.plugin_instance, mr->location, sizeof(n.plugin_instance));
342 if (dispatch_corrected_notifs) {
343 /* Corrected Error Notifications */
344 if (mr->corrected_err_total > 0 || mr->corrected_err_timed > 0) {
345 if (plugin_notification_meta_add_signed_int(
346 &n, MCELOG_CORRECTED_ERR, mr->corrected_err_total) < 0) {
347 ERROR(MCELOG_PLUGIN ": add corrected errors meta data failed");
348 plugin_notification_meta_free(n.meta);
349 return (-1);
350 }
351 if (plugin_notification_meta_add_signed_int(
352 &n, "corrected memory timed errors", mr->corrected_err_timed) <
353 0) {
354 ERROR(MCELOG_PLUGIN ": add corrected timed errors meta data failed");
355 plugin_notification_meta_free(n.meta);
356 return (-1);
357 }
358 ssnprintf(n.message, sizeof(n.message), "Corrected Memory Errors");
359 sstrncpy(n.type_instance, MCELOG_CORRECTED_ERR_TYPE_INS,
360 sizeof(n.type_instance));
361 plugin_dispatch_notification(&n);
363 if (n.meta)
364 plugin_notification_meta_free(n.meta);
365 }
366 }
368 if (dispatch_uncorrected_notifs) {
369 /* Uncorrected Error Notifications */
370 if (mr->uncorrected_err_total > 0 || mr->uncorrected_err_timed > 0) {
371 if (plugin_notification_meta_add_signed_int(
372 &n, MCELOG_UNCORRECTED_ERR, mr->uncorrected_err_total) < 0) {
373 ERROR(MCELOG_PLUGIN ": add uncorrected errors meta data failed");
374 plugin_notification_meta_free(n.meta);
375 return (-1);
376 }
377 if (plugin_notification_meta_add_signed_int(
378 &n, "uncorrected memory timed errors",
379 mr->uncorrected_err_timed) < 0) {
380 ERROR(MCELOG_PLUGIN ": add uncorrected timed errors meta data failed");
381 plugin_notification_meta_free(n.meta);
382 return (-1);
383 }
384 ssnprintf(n.message, sizeof(n.message), "Uncorrected Memory Errors");
385 sstrncpy(n.type_instance, MCELOG_UNCORRECTED_ERR_TYPE_INS,
386 sizeof(n.type_instance));
387 n.severity = NOTIF_FAILURE;
388 plugin_dispatch_notification(&n);
390 if (n.meta)
391 plugin_notification_meta_free(n.meta);
392 }
393 }
395 return (0);
396 }
398 static int mcelog_submit(const mcelog_memory_rec_t *mr) {
400 if (!mr) {
401 ERROR(MCELOG_PLUGIN ": %s: NULL pointer", __FUNCTION__);
402 return (-1);
403 }
405 llentry_t *dimm = mcelog_dimm(mr, g_mcelog_config.dimms_list);
406 if (dimm == NULL) {
407 ERROR(MCELOG_PLUGIN
408 ": Error adding/getting dimm memory item to/from cache");
409 return (-1);
410 }
412 value_list_t vl = {
413 .values_len = 1,
414 .values = &(value_t){.derive = (derive_t)mr->corrected_err_total},
415 .time = cdtime(),
416 .plugin = MCELOG_PLUGIN,
417 .type = "errors",
418 .type_instance = MCELOG_CORRECTED_ERR_TYPE_INS};
420 mcelog_update_dimm_stats(dimm, mr);
422 if (mr->dimm_name[0] != '\0')
423 ssnprintf(vl.plugin_instance, sizeof(vl.plugin_instance), "%s_%s",
424 mr->location, mr->dimm_name);
425 else
426 sstrncpy(vl.plugin_instance, mr->location, sizeof(vl.plugin_instance));
428 plugin_dispatch_values(&vl);
430 ssnprintf(vl.type_instance, sizeof(vl.type_instance),
431 "corrected_memory_errors_in_%s", mr->corrected_err_timed_period);
432 vl.values = &(value_t){.derive = (derive_t)mr->corrected_err_timed};
433 plugin_dispatch_values(&vl);
435 sstrncpy(vl.type_instance, MCELOG_UNCORRECTED_ERR_TYPE_INS,
436 sizeof(vl.type_instance));
437 vl.values = &(value_t){.derive = (derive_t)mr->uncorrected_err_total};
438 plugin_dispatch_values(&vl);
440 ssnprintf(vl.type_instance, sizeof(vl.type_instance),
441 "uncorrected_memory_errors_in_%s",
442 mr->uncorrected_err_timed_period);
443 vl.values = &(value_t){.derive = (derive_t)mr->uncorrected_err_timed};
444 plugin_dispatch_values(&vl);
446 return (0);
447 }
449 static int parse_memory_info(FILE *p_file, mcelog_memory_rec_t *memory_record) {
450 char buf[DATA_MAX_NAME_LEN] = {0};
451 while (fgets(buf, sizeof(buf), p_file)) {
452 /* Got empty line or "done" */
453 if ((!strncmp("\n", buf, strlen(buf))) ||
454 (!strncmp(buf, "done\n", strlen(buf))))
455 return (1);
456 if (strlen(buf) < 5)
457 continue;
458 if (!strncmp(buf, MCELOG_SOCKET_STR, strlen(MCELOG_SOCKET_STR))) {
459 sstrncpy(memory_record->location, buf, strlen(buf));
460 /* replace spaces with '_' */
461 for (size_t i = 0; i < strlen(memory_record->location); i++)
462 if (memory_record->location[i] == ' ')
463 memory_record->location[i] = '_';
464 DEBUG(MCELOG_PLUGIN ": Got SOCKET INFO %s", memory_record->location);
465 }
466 if (!strncmp(buf, MCELOG_DIMM_NAME, strlen(MCELOG_DIMM_NAME))) {
467 char *name = NULL;
468 char *saveptr = NULL;
469 name = strtok_r(buf, "\"", &saveptr);
470 if (name != NULL && saveptr != NULL) {
471 name = strtok_r(NULL, "\"", &saveptr);
472 if (name != NULL) {
473 sstrncpy(memory_record->dimm_name, name,
474 sizeof(memory_record->dimm_name));
475 DEBUG(MCELOG_PLUGIN ": Got DIMM NAME %s", memory_record->dimm_name);
476 }
477 }
478 }
479 if (!strncmp(buf, MCELOG_CORRECTED_ERR, strlen(MCELOG_CORRECTED_ERR))) {
480 /* Get next line*/
481 if (fgets(buf, sizeof(buf), p_file) != NULL) {
482 sscanf(buf, "\t%d total", &(memory_record->corrected_err_total));
483 DEBUG(MCELOG_PLUGIN ": Got corrected error total %d",
484 memory_record->corrected_err_total);
485 }
486 if (fgets(buf, sizeof(buf), p_file) != NULL) {
487 sscanf(buf, "\t%d in %s", &(memory_record->corrected_err_timed),
488 memory_record->corrected_err_timed_period);
489 DEBUG(MCELOG_PLUGIN ": Got timed corrected errors %d in %s",
490 memory_record->corrected_err_total,
491 memory_record->corrected_err_timed_period);
492 }
493 }
494 if (!strncmp(buf, MCELOG_UNCORRECTED_ERR, strlen(MCELOG_UNCORRECTED_ERR))) {
495 if (fgets(buf, sizeof(buf), p_file) != NULL) {
496 sscanf(buf, "\t%d total", &(memory_record->uncorrected_err_total));
497 DEBUG(MCELOG_PLUGIN ": Got uncorrected error total %d",
498 memory_record->uncorrected_err_total);
499 }
500 if (fgets(buf, sizeof(buf), p_file) != NULL) {
501 sscanf(buf, "\t%d in %s", &(memory_record->uncorrected_err_timed),
502 memory_record->uncorrected_err_timed_period);
503 DEBUG(MCELOG_PLUGIN ": Got timed uncorrected errors %d in %s",
504 memory_record->uncorrected_err_total,
505 memory_record->uncorrected_err_timed_period);
506 }
507 }
508 memset(buf, 0, sizeof(buf));
509 }
510 /* parsing definitely finished */
511 return (0);
512 }
514 static void poll_worker_cleanup(void *arg) {
515 mcelog_thread_running = 0;
516 FILE *p_file = *((FILE **)arg);
517 if (p_file != NULL)
518 fclose(p_file);
519 free(arg);
520 }
522 static int socket_receive(socket_adapter_t *self, FILE **pp_file) {
523 int res = -1;
524 pthread_rwlock_rdlock(&self->lock);
525 struct pollfd poll_fd = {
526 .fd = self->sock_fd, .events = POLLIN | POLLPRI,
527 };
529 if ((res = poll(&poll_fd, 1, MCELOG_POLL_TIMEOUT)) <= 0) {
530 if (res != 0 && errno != EINTR) {
531 char errbuf[MCELOG_BUFF_SIZE];
532 ERROR("mcelog: poll failed: %s",
533 sstrerror(errno, errbuf, sizeof(errbuf)));
534 }
535 pthread_rwlock_unlock(&self->lock);
536 return (res);
537 }
539 if (poll_fd.revents & (POLLERR | POLLHUP | POLLNVAL)) {
540 /* connection is broken */
541 ERROR(MCELOG_PLUGIN ": Connection to socket is broken");
542 if (poll_fd.revents & (POLLERR | POLLHUP)) {
543 mcelog_dispatch_notification(
544 &(notification_t){.severity = NOTIF_FAILURE,
545 .time = cdtime(),
546 .message = "Connection to mcelog socket is broken.",
547 .plugin = MCELOG_PLUGIN,
548 .type_instance = "mcelog_status"});
549 }
550 pthread_rwlock_unlock(&self->lock);
551 return (-1);
552 }
554 if (!(poll_fd.revents & (POLLIN | POLLPRI))) {
555 INFO(MCELOG_PLUGIN ": No data to read");
556 pthread_rwlock_unlock(&self->lock);
557 return (0);
558 }
560 if ((*pp_file = fdopen(dup(self->sock_fd), "r")) == NULL)
561 res = -1;
563 pthread_rwlock_unlock(&self->lock);
564 return (res);
565 }
567 static void *poll_worker(__attribute__((unused)) void *arg) {
568 char errbuf[MCELOG_BUFF_SIZE];
569 mcelog_thread_running = 1;
570 FILE **pp_file = calloc(1, sizeof(*pp_file));
571 if (pp_file == NULL) {
572 ERROR("mcelog: memory allocation failed: %s",
573 sstrerror(errno, errbuf, sizeof(errbuf)));
574 pthread_exit((void *)1);
575 }
577 pthread_cleanup_push(poll_worker_cleanup, pp_file);
579 while (1) {
580 /* blocking call */
581 int res = socket_adapter.receive(&socket_adapter, pp_file);
582 if (res < 0) {
583 socket_adapter.close(&socket_adapter);
584 while (socket_adapter.reinit(&socket_adapter) != 0) {
585 nanosleep(&CDTIME_T_TO_TIMESPEC(MS_TO_CDTIME_T(MCELOG_POLL_TIMEOUT)),
586 NULL);
587 }
588 continue;
589 }
590 /* timeout or no data to read */
591 else if (res == 0)
592 continue;
594 if (*pp_file == NULL)
595 continue;
597 mcelog_memory_rec_t memory_record = {0};
598 while (parse_memory_info(*pp_file, &memory_record)) {
599 /* Check if location was successfully parsed */
600 if (memory_record.location[0] == '\0') {
601 memset(&memory_record, 0, sizeof(memory_record));
602 continue;
603 }
605 if (mcelog_dispatch_mem_notifications(&memory_record) != 0)
606 ERROR(MCELOG_PLUGIN ": Failed to submit memory errors notification");
607 if (mcelog_submit(&memory_record) != 0)
608 ERROR(MCELOG_PLUGIN ": Failed to submit memory errors");
609 memset(&memory_record, 0, sizeof(memory_record));
610 }
612 fclose(*pp_file);
613 *pp_file = NULL;
614 }
616 mcelog_thread_running = 0;
617 pthread_cleanup_pop(1);
618 return (NULL);
619 }
621 static int mcelog_init(void) {
622 g_mcelog_config.dimms_list = llist_create();
623 int err = pthread_mutex_init(&g_mcelog_config.dimms_lock, NULL);
624 if (err < 0) {
625 ERROR(MCELOG_PLUGIN ": plugin: failed to initialize cache lock");
626 return (-1);
627 }
629 if (socket_adapter.reinit(&socket_adapter) != 0) {
630 ERROR(MCELOG_PLUGIN ": Cannot connect to client socket");
631 return (-1);
632 }
634 if (plugin_thread_create(&g_mcelog_config.tid, NULL, poll_worker, NULL,
635 NULL) != 0) {
636 ERROR(MCELOG_PLUGIN ": Error creating poll thread.");
637 return (-1);
638 }
639 return (0);
640 }
642 static int get_memory_machine_checks(void) {
643 static const char dump[] = "dump all bios\n";
644 int ret = socket_adapter.write(&socket_adapter, dump, sizeof(dump));
645 if (ret != 0)
646 ERROR(MCELOG_PLUGIN ": SENT DUMP REQUEST FAILED");
647 else
648 DEBUG(MCELOG_PLUGIN ": SENT DUMP REQUEST OK");
649 return (ret);
650 }
652 static int mcelog_read(__attribute__((unused)) user_data_t *ud) {
653 DEBUG(MCELOG_PLUGIN ": %s", __FUNCTION__);
655 if (get_memory_machine_checks() != 0)
656 ERROR(MCELOG_PLUGIN ": MACHINE CHECK INFO NOT AVAILABLE");
658 return (0);
659 }
661 static int mcelog_shutdown(void) {
662 int ret = 0;
663 if (mcelog_thread_running) {
664 pthread_cancel(g_mcelog_config.tid);
665 if (pthread_join(g_mcelog_config.tid, NULL) != 0) {
666 ERROR(MCELOG_PLUGIN ": Stopping thread failed.");
667 ret = -1;
668 }
669 }
670 pthread_mutex_lock(&g_mcelog_config.dimms_lock);
671 mcelog_free_dimms_list_records(g_mcelog_config.dimms_list);
672 llist_destroy(g_mcelog_config.dimms_list);
673 pthread_mutex_unlock(&g_mcelog_config.dimms_lock);
674 pthread_mutex_destroy(&g_mcelog_config.dimms_lock);
675 g_mcelog_config.dimms_list = NULL;
676 ret = socket_adapter.close(&socket_adapter) || ret;
677 pthread_rwlock_destroy(&(socket_adapter.lock));
678 return (-ret);
679 }
681 void module_register(void) {
682 plugin_register_complex_config(MCELOG_PLUGIN, mcelog_config);
683 plugin_register_init(MCELOG_PLUGIN, mcelog_init);
684 plugin_register_complex_read(NULL, MCELOG_PLUGIN, mcelog_read, 0, NULL);
685 plugin_register_shutdown(MCELOG_PLUGIN, mcelog_shutdown);
686 }