1 /*-
2 * collectd - src/mcelog.c
3 * MIT License
4 *
5 * Copyright(c) 2016 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 "common.h"
33 #include "collectd.h"
35 #include <poll.h>
36 #include <sys/socket.h>
37 #include <sys/un.h>
38 #include <unistd.h>
40 #define MCELOG_PLUGIN "mcelog"
41 #define MCELOG_BUFF_SIZE 1024
42 #define MCELOG_POLL_TIMEOUT 1000 /* ms */
43 #define MCELOG_SOCKET_STR "SOCKET"
44 #define MCELOG_DIMM_NAME "DMI_NAME"
45 #define MCELOG_CORRECTED_ERR "corrected memory errors:"
46 #define MCELOG_UNCORRECTED_ERR "uncorrected memory errors:"
48 typedef struct mcelog_config_s {
49 char logfile[PATH_MAX]; /* mcelog logfile */
50 pthread_t tid; /* poll thread id */
51 } mcelog_config_t;
53 typedef struct socket_adapter_s socket_adapter_t;
55 struct socket_adapter_s {
56 int sock_fd; /* mcelog server socket fd */
57 struct sockaddr_un unix_sock; /* mcelog client socket */
58 pthread_rwlock_t lock;
59 /* function pointers for socket operations */
60 int (*write)(socket_adapter_t *self, const char *msg, const size_t len);
61 int (*reinit)(socket_adapter_t *self);
62 int (*receive)(socket_adapter_t *self, FILE **p_file);
63 int (*close)(socket_adapter_t *self);
64 };
66 typedef struct mcelog_memory_rec_s {
67 int corrected_err_total; /* x total*/
68 int corrected_err_timed; /* x in 24h*/
69 char corrected_err_timed_period[DATA_MAX_NAME_LEN];
70 int uncorrected_err_total; /* x total*/
71 int uncorrected_err_timed; /* x in 24h*/
72 char uncorrected_err_timed_period[DATA_MAX_NAME_LEN];
73 char location[DATA_MAX_NAME_LEN]; /* SOCKET x CHANNEL x DIMM x*/
74 char dimm_name[DATA_MAX_NAME_LEN]; /* DMI_NAME "DIMM_F1" */
75 } mcelog_memory_rec_t;
77 static int socket_close(socket_adapter_t *self);
78 static int socket_write(socket_adapter_t *self, const char *msg,
79 const size_t len);
80 static int socket_reinit(socket_adapter_t *self);
81 static int socket_receive(socket_adapter_t *self, FILE **p_file);
83 static mcelog_config_t g_mcelog_config = {
84 .logfile = "/var/log/mcelog", .tid = 0,
85 };
87 static socket_adapter_t socket_adapter = {
88 .sock_fd = -1,
89 .unix_sock =
90 {
91 .sun_family = AF_UNIX, .sun_path = "/var/run/mcelog-client",
92 },
93 .lock = PTHREAD_RWLOCK_INITIALIZER,
94 .close = socket_close,
95 .write = socket_write,
96 .reinit = socket_reinit,
97 .receive = socket_receive,
98 };
100 static _Bool mcelog_thread_running = 0;
102 static int mcelog_config(oconfig_item_t *ci) {
103 for (int i = 0; i < ci->children_num; i++) {
104 oconfig_item_t *child = ci->children + i;
105 if (strcasecmp("McelogClientSocket", child->key) == 0) {
106 if (cf_util_get_string_buffer(child, socket_adapter.unix_sock.sun_path,
107 sizeof(socket_adapter.unix_sock.sun_path)) <
108 0) {
109 ERROR("%s: Invalid configuration option: \"%s\".", MCELOG_PLUGIN,
110 child->key);
111 return -1;
112 }
113 } else if (strcasecmp("McelogLogfile", child->key) == 0) {
114 if (cf_util_get_string_buffer(child, g_mcelog_config.logfile,
115 sizeof(g_mcelog_config.logfile)) < 0) {
116 ERROR("%s: Invalid configuration option: \"%s\".", MCELOG_PLUGIN,
117 child->key);
118 return -1;
119 }
120 } else {
121 ERROR("%s: Invalid configuration option: \"%s\".", MCELOG_PLUGIN,
122 child->key);
123 return -1;
124 }
125 }
126 return (0);
127 }
129 static int socket_close(socket_adapter_t *self) {
130 int ret = 0;
131 pthread_rwlock_rdlock(&self->lock);
132 if (fcntl(self->sock_fd, F_GETFL) != -1) {
133 if (shutdown(self->sock_fd, SHUT_RDWR) != 0) {
134 char errbuf[MCELOG_BUFF_SIZE];
135 ERROR("%s: Socket shutdown failed: %s", MCELOG_PLUGIN,
136 sstrerror(errno, errbuf, sizeof(errbuf)));
137 ret = -1;
138 }
139 close(self->sock_fd);
140 }
141 pthread_rwlock_unlock(&self->lock);
142 return ret;
143 }
145 static int socket_write(socket_adapter_t *self, const char *msg,
146 const size_t len) {
147 int ret = 0;
148 pthread_rwlock_rdlock(&self->lock);
149 if (swrite(self->sock_fd, msg, len) < 0)
150 ret = -1;
151 pthread_rwlock_unlock(&self->lock);
152 return ret;
153 }
155 static int socket_reinit(socket_adapter_t *self) {
156 char errbuff[MCELOG_BUFF_SIZE];
157 int ret = -1;
158 cdtime_t interval = plugin_get_interval();
159 struct timeval socket_timeout = CDTIME_T_TO_TIMEVAL(interval);
161 /* synchronization via write lock since sock_fd may be changed here */
162 pthread_rwlock_wrlock(&self->lock);
163 self->sock_fd = socket(PF_UNIX, SOCK_STREAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
164 if (self->sock_fd < 0) {
165 ERROR("%s: Could not create a socket. %s", MCELOG_PLUGIN,
166 sstrerror(errno, errbuff, sizeof(errbuff)));
167 pthread_rwlock_unlock(&self->lock);
168 return ret;
169 }
171 /* Set socket timeout option */
172 if (setsockopt(self->sock_fd, SOL_SOCKET, SO_SNDTIMEO,
173 &socket_timeout, sizeof(socket_timeout)) < 0)
174 ERROR("%s: Failed to set the socket timeout option.", MCELOG_PLUGIN);
176 /* downgrading to read lock due to possible recursive read locks
177 * in self->close(self) call */
178 pthread_rwlock_unlock(&self->lock);
179 pthread_rwlock_rdlock(&self->lock);
180 if (connect(self->sock_fd, (struct sockaddr *)&(self->unix_sock),
181 sizeof(self->unix_sock)) < 0) {
182 ERROR("%s: Failed to connect to mcelog server. %s", MCELOG_PLUGIN,
183 sstrerror(errno, errbuff, sizeof(errbuff)));
184 self->close(self);
185 ret = -1;
186 } else
187 ret = 0;
189 pthread_rwlock_unlock(&self->lock);
190 return ret;
191 }
193 static void mcelog_dispatch_notification(notification_t n) {
194 sstrncpy(n.host, hostname_g, sizeof(n.host));
195 sstrncpy(n.type, "gauge", sizeof(n.type));
196 plugin_dispatch_notification(&n);
197 }
199 static int mcelog_prepare_notification(notification_t *n,
200 mcelog_memory_rec_t mr) {
201 if (n == NULL)
202 return (-1);
204 if (plugin_notification_meta_add_string(n, MCELOG_SOCKET_STR, mr.location) <
205 0) {
206 ERROR("%s: add memory location meta data failed", MCELOG_PLUGIN);
207 return (-1);
208 }
209 if (strlen(mr.dimm_name) > 0)
210 if (plugin_notification_meta_add_string(n, MCELOG_DIMM_NAME, mr.dimm_name) <
211 0) {
212 ERROR("%s: add DIMM name meta data failed", MCELOG_PLUGIN);
213 return (-1);
214 }
215 if (plugin_notification_meta_add_signed_int(n, MCELOG_CORRECTED_ERR,
216 mr.corrected_err_total) < 0) {
217 ERROR("%s: add corrected errors meta data failed", MCELOG_PLUGIN);
218 return (-1);
219 }
220 if (plugin_notification_meta_add_signed_int(
221 n, "corrected memory timed errors", mr.corrected_err_timed) < 0) {
222 ERROR("%s: add corrected timed errors meta data failed", MCELOG_PLUGIN);
223 return (-1);
224 }
225 if (plugin_notification_meta_add_string(n, "corrected errors time period",
226 mr.corrected_err_timed_period) < 0) {
227 ERROR("%s: add corrected errors period meta data failed", MCELOG_PLUGIN);
228 return (-1);
229 }
230 if (plugin_notification_meta_add_signed_int(n, MCELOG_UNCORRECTED_ERR,
231 mr.uncorrected_err_total) < 0) {
232 ERROR("%s: add corrected errors meta data failed", MCELOG_PLUGIN);
233 return (-1);
234 }
235 if (plugin_notification_meta_add_signed_int(
236 n, "uncorrected memory timed errors", mr.uncorrected_err_timed) < 0) {
237 ERROR("%s: add corrected timed errors meta data failed", MCELOG_PLUGIN);
238 return (-1);
239 }
240 if (plugin_notification_meta_add_string(n, "uncorrected errors time period",
241 mr.uncorrected_err_timed_period) <
242 0) {
243 ERROR("%s: add corrected errors period meta data failed", MCELOG_PLUGIN);
244 return (-1);
245 }
247 return (0);
248 }
250 static int mcelog_submit(mcelog_memory_rec_t mr) {
252 value_list_t vl = VALUE_LIST_INIT;
253 vl.values_len = 1;
254 vl.time = cdtime();
256 sstrncpy(vl.plugin, MCELOG_PLUGIN, sizeof(vl.plugin));
257 sstrncpy(vl.type, "errors", sizeof(vl.type));
258 if (strlen(mr.dimm_name) > 0) {
259 ssnprintf(vl.plugin_instance, sizeof(vl.plugin_instance), "%s_%s",
260 mr.location, mr.dimm_name);
261 } else
262 sstrncpy(vl.plugin_instance, mr.location, sizeof(vl.plugin_instance));
264 sstrncpy(vl.type_instance, "corrected_memory_errors",
265 sizeof(vl.type_instance));
266 vl.values = &(value_t){.derive = (derive_t)mr.corrected_err_total};
267 plugin_dispatch_values(&vl);
269 ssnprintf(vl.type_instance, sizeof(vl.type_instance),
270 "corrected_memory_errors_in_%s", mr.corrected_err_timed_period);
271 vl.values = &(value_t){.derive = (derive_t)mr.corrected_err_timed};
272 plugin_dispatch_values(&vl);
274 sstrncpy(vl.type_instance, "uncorrected_memory_errors",
275 sizeof(vl.type_instance));
276 vl.values = &(value_t){.derive = (derive_t)mr.uncorrected_err_total};
277 plugin_dispatch_values(&vl);
279 ssnprintf(vl.type_instance, sizeof(vl.type_instance),
280 "uncorrected_memory_errors_in_%s", mr.uncorrected_err_timed_period);
281 vl.values = &(value_t){.derive = (derive_t)mr.uncorrected_err_timed};
282 plugin_dispatch_values(&vl);
284 return 0;
285 }
287 static int parse_memory_info(FILE *p_file, mcelog_memory_rec_t *memory_record) {
288 char buf[DATA_MAX_NAME_LEN] = {0};
289 while (fgets(buf, sizeof(buf), p_file)) {
290 /* Got empty line or "done" */
291 if ((!strncmp("\n", buf, strlen(buf))) ||
292 (!strncmp(buf, "done\n", strlen(buf))))
293 return 1;
294 if (strlen(buf) < 5)
295 continue;
296 if (!strncmp(buf, MCELOG_SOCKET_STR, strlen(MCELOG_SOCKET_STR))) {
297 sstrncpy(memory_record->location, buf, strlen(buf));
298 /* replace spaces with '_' */
299 for (size_t i = 0; i < strlen(memory_record->location); i++)
300 if (memory_record->location[i] == ' ')
301 memory_record->location[i] = '_';
302 DEBUG("%s: Got SOCKET INFO %s", MCELOG_PLUGIN, memory_record->location);
303 }
304 if (!strncmp(buf, MCELOG_DIMM_NAME, strlen(MCELOG_DIMM_NAME))) {
305 char *name = NULL;
306 char *saveptr = NULL;
307 name = strtok_r(buf, "\"", &saveptr);
308 if (name != NULL && saveptr != NULL) {
309 name = strtok_r(NULL, "\"", &saveptr);
310 if (name != NULL) {
311 sstrncpy(memory_record->dimm_name, name,
312 sizeof(memory_record->dimm_name));
313 DEBUG("%s: Got DIMM NAME %s", MCELOG_PLUGIN,
314 memory_record->dimm_name);
315 }
316 }
317 }
318 if (!strncmp(buf, MCELOG_CORRECTED_ERR, strlen(MCELOG_CORRECTED_ERR))) {
319 /* Get next line*/
320 if (fgets(buf, sizeof(buf), p_file) != NULL) {
321 sscanf(buf, "\t%d total", &(memory_record->corrected_err_total));
322 DEBUG("%s: Got corrected error total %d", MCELOG_PLUGIN,
323 memory_record->corrected_err_total);
324 }
325 if (fgets(buf, sizeof(buf), p_file) != NULL) {
326 sscanf(buf, "\t%d in %s", &(memory_record->corrected_err_timed),
327 memory_record->corrected_err_timed_period);
328 DEBUG("%s: Got timed corrected errors %d in %s", MCELOG_PLUGIN,
329 memory_record->corrected_err_total,
330 memory_record->corrected_err_timed_period);
331 }
332 }
333 if (!strncmp(buf, MCELOG_UNCORRECTED_ERR, strlen(MCELOG_UNCORRECTED_ERR))) {
334 if (fgets(buf, sizeof(buf), p_file) != NULL) {
335 sscanf(buf, "\t%d total", &(memory_record->uncorrected_err_total));
336 DEBUG("%s: Got uncorrected error total %d", MCELOG_PLUGIN,
337 memory_record->uncorrected_err_total);
338 }
339 if (fgets(buf, sizeof(buf), p_file) != NULL) {
340 sscanf(buf, "\t%d in %s", &(memory_record->uncorrected_err_timed),
341 memory_record->uncorrected_err_timed_period);
342 DEBUG("%s: Got timed uncorrected errors %d in %s", MCELOG_PLUGIN,
343 memory_record->uncorrected_err_total,
344 memory_record->uncorrected_err_timed_period);
345 }
346 }
347 memset(buf, 0, sizeof(buf));
348 }
349 /* parsing definitely finished */
350 return 0;
351 }
353 static void poll_worker_cleanup(void *arg) {
354 mcelog_thread_running = 0;
355 FILE *p_file = *((FILE **)arg);
356 if (p_file != NULL)
357 fclose(p_file);
358 free(arg);
359 }
361 static int socket_receive(socket_adapter_t *self, FILE **pp_file) {
362 int res = -1;
363 pthread_rwlock_rdlock(&self->lock);
364 struct pollfd poll_fd = {
365 .fd = self->sock_fd, .events = POLLIN | POLLPRI,
366 };
368 if ((res = poll(&poll_fd, 1, MCELOG_POLL_TIMEOUT)) <= 0) {
369 if (res != 0 && errno != EINTR) {
370 char errbuf[MCELOG_BUFF_SIZE];
371 ERROR("mcelog: poll failed: %s",
372 sstrerror(errno, errbuf, sizeof(errbuf)));
373 }
374 pthread_rwlock_unlock(&self->lock);
375 return res;
376 }
378 if (poll_fd.revents & (POLLERR | POLLHUP | POLLNVAL)) {
379 /* connection is broken */
380 ERROR("%s: Connection to socket is broken", MCELOG_PLUGIN);
381 if (poll_fd.revents & (POLLERR | POLLHUP)) {
382 notification_t n = {
383 NOTIF_FAILURE, cdtime(), "", "", MCELOG_PLUGIN, "", "", "", NULL};
384 ssnprintf(n.message, sizeof(n.message),
385 "Connection to mcelog socket is broken.");
386 sstrncpy(n.type_instance, "mcelog_status", sizeof(n.type_instance));
387 mcelog_dispatch_notification(n);
388 }
389 pthread_rwlock_unlock(&self->lock);
390 return -1;
391 }
393 if (!(poll_fd.revents & (POLLIN | POLLPRI))) {
394 INFO("%s: No data to read", MCELOG_PLUGIN);
395 pthread_rwlock_unlock(&self->lock);
396 return 0;
397 }
399 if ((*pp_file = fdopen(dup(self->sock_fd), "r")) == NULL)
400 res = -1;
402 pthread_rwlock_unlock(&self->lock);
403 return res;
404 }
406 static void *poll_worker(__attribute__((unused)) void *arg) {
407 char errbuf[MCELOG_BUFF_SIZE];
408 mcelog_thread_running = 1;
409 FILE **pp_file = calloc(1, sizeof(*pp_file));
410 if (pp_file == NULL) {
411 ERROR("mcelog: memory allocation failed: %s",
412 sstrerror(errno, errbuf, sizeof(errbuf)));
413 pthread_exit((void *)1);
414 }
416 pthread_cleanup_push(poll_worker_cleanup, pp_file);
418 while (1) {
419 int res = 0;
420 /* blocking call */
421 res = socket_adapter.receive(&socket_adapter, pp_file);
422 if (res < 0) {
423 socket_adapter.close(&socket_adapter);
424 if (socket_adapter.reinit(&socket_adapter) != 0) {
425 socket_adapter.close(&socket_adapter);
426 usleep(MCELOG_POLL_TIMEOUT);
427 }
428 continue;
429 }
430 /* timeout or no data to read */
431 else if (res == 0)
432 continue;
434 if (*pp_file == NULL)
435 continue;
437 mcelog_memory_rec_t memory_record = {0};
438 while (parse_memory_info(*pp_file, &memory_record)) {
439 notification_t n = {NOTIF_OKAY, cdtime(), "", "", MCELOG_PLUGIN,
440 "", "", "", NULL};
441 ssnprintf(n.message, sizeof(n.message), "Got memory errors info.");
442 sstrncpy(n.type_instance, "memory_erros", sizeof(n.type_instance));
443 if (mcelog_prepare_notification(&n, memory_record) == 0)
444 mcelog_dispatch_notification(n);
445 if (mcelog_submit(memory_record) != 0)
446 ERROR("%s: Failed to submit memory errors", MCELOG_PLUGIN);
447 memset(&memory_record, 0, sizeof(memory_record));
448 }
450 fclose(*pp_file);
451 *pp_file = NULL;
452 }
454 mcelog_thread_running = 0;
455 pthread_cleanup_pop(1);
456 return NULL;
457 }
459 static int mcelog_init(void) {
460 if (socket_adapter.reinit(&socket_adapter) != 0) {
461 ERROR("%s: Cannot connect to client socket", MCELOG_PLUGIN);
462 return -1;
463 }
465 if (plugin_thread_create(&g_mcelog_config.tid, NULL, poll_worker, NULL,
466 NULL) != 0) {
467 ERROR("%s: Error creating poll thread.", MCELOG_PLUGIN);
468 return -1;
469 }
470 return 0;
471 }
473 static int get_memory_machine_checks(void) {
474 static const char dump[] = "dump all bios\n";
475 int ret = socket_adapter.write(&socket_adapter, dump, sizeof(dump));
476 if (ret != 0)
477 ERROR("%s: SENT DUMP REQUEST FAILED", MCELOG_PLUGIN);
478 else
479 DEBUG("%s: SENT DUMP REQUEST OK", MCELOG_PLUGIN);
480 return ret;
481 }
483 static int mcelog_read(__attribute__((unused)) user_data_t *ud) {
484 DEBUG("%s: %s", MCELOG_PLUGIN, __FUNCTION__);
486 if (get_memory_machine_checks() != 0)
487 ERROR("%s: MACHINE CHECK INFO NOT AVAILABLE", MCELOG_PLUGIN);
489 return 0;
490 }
492 static int mcelog_shutdown(void) {
493 int ret = 0;
494 if (mcelog_thread_running) {
495 pthread_cancel(g_mcelog_config.tid);
496 if (pthread_join(g_mcelog_config.tid, NULL) != 0) {
497 ERROR("%s: Stopping thread failed.", MCELOG_PLUGIN);
498 ret = -1;
499 }
500 }
502 ret = socket_adapter.close(&socket_adapter) || ret;
503 pthread_rwlock_destroy(&(socket_adapter.lock));
504 return -ret;
505 }
507 void module_register(void) {
508 plugin_register_complex_config(MCELOG_PLUGIN, mcelog_config);
509 plugin_register_init(MCELOG_PLUGIN, mcelog_init);
510 plugin_register_complex_read(NULL, MCELOG_PLUGIN, mcelog_read, 0, NULL);
511 plugin_register_shutdown(MCELOG_PLUGIN, mcelog_shutdown);
512 }