1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * snmp/lustre-snmp-trap.c
38 * Author: PJ Kirner <pjkirner@clusterfs.com>
42 * include important headers
45 #include <net-snmp/net-snmp-config.h>
46 #include <net-snmp/net-snmp-includes.h>
47 #include <net-snmp/agent/net-snmp-agent-includes.h>
53 #include <sys/types.h>
54 #if defined (__linux__)
62 #include "lustre-snmp-util.h"
64 /**************************************************************************
66 *************************************************************************/
68 #define DEFAULT_POLL_INTERVAL_SECONDS 60
69 #define POLL_INTERVAL_ENV_VAR "LSNMP_POLL_INTERVAL"
70 #define SNMP_HEALTH_CHECK_TEST_FILE "LSNMP_HEALTH_CHECK_TEST_FILE"
72 /**************************************************************************
74 *************************************************************************/
76 static oid objid_snmptrap[] =
77 { 1,3,6,1,6,3,1,1,4,1,0};
78 static oid lustre_portals_trap[] =
79 { 1,3,6,1,4,1,13140,2,1,0,1};
80 static oid lustre_portals_trap_string[]=
81 { 1,3,6,1,4,1,13140,2,1,0,2};
82 static oid lustre_unhealthy_trap[] =
83 { 1,3,6,1,4,1,13140,2,1,0,3};
84 static oid lustre_unhealthy_trap_device_name_string[]=
85 { 1,3,6,1,4,1,13140,2,1,0,4};
86 static oid lustre_unhealthy_trap_reason_string[]=
87 { 1,3,6,1,4,1,13140,2,1,0,5};
89 /**************************************************************************
91 *************************************************************************/
93 typedef struct obd_unhealthy_entry_struct{
95 /*1-if seen as part of the the is_unhealthy scan, otherwise 0*/
98 /*single linked list pointer*/
99 struct obd_unhealthy_entry_struct *next;
101 /*obdname - variable size*/
104 }obd_unhealthy_entry;
106 /**************************************************************************
108 *************************************************************************/
110 int get_poll_interval_seconds();
111 void health_poll_worker(unsigned int registration_number, void *clientarg);
112 void send_portals_catastrophe_trap(char *reason_string);
113 void send_obd_unhealthy_trap(char *obd_name,char *reason_string);
114 int is_obd_newly_unhealthy(const char* obd_name);
115 void obd_unhealthy_scan(void);
116 void health_entry_parser(void);
118 /**************************************************************************
120 *************************************************************************/
122 static int g_sent_portals_catastrophe = 0;
123 static obd_unhealthy_entry* g_obd_unhealthy_list = NULL;
124 static int g_poll_interval_seconds;
125 static unsigned int g_registration_handle;
126 static char *g_health_check_test_file = 0;
128 /*****************************************************************************
129 * Function: initilize_trap_handler
131 * Description: Initlized the trap poll haalder.
135 * Output: Global g_poll_interval_seconds is set.
137 ****************************************************************************/
139 void initilize_trap_handler(void)
141 g_poll_interval_seconds = get_poll_interval_seconds();
143 g_registration_handle = snmp_alarm_register(g_poll_interval_seconds, 0, health_poll_worker, NULL);
144 if (g_registration_handle == 0)
145 report("%s %s: line %d %s", __FILE__, __FUNCTION__, __LINE__,
146 "snmp_alarm_register failed");
148 DEBUGMSGTL(("lsnmpd","lsnmp alarm registered poll interval = %d seconds\n",g_poll_interval_seconds));
150 g_health_check_test_file = getenv(SNMP_HEALTH_CHECK_TEST_FILE);
151 if(g_health_check_test_file != 0)
152 DEBUGMSGTL(("lsnmpd","lsnmp health check test file set to \'%s\'\n",g_health_check_test_file));
155 /*****************************************************************************
156 * Function: terminate_trap_handler
158 * Description: Terminate the trap poll haalder.
162 * Output: Global g_poll_interval_seconds is set.
164 ****************************************************************************/
166 void terminate_trap_handler(void)
168 snmp_alarm_unregister(g_registration_handle);
171 /*****************************************************************************
172 * Function: get_poll_interval_seconds
174 * Description: This function used to get the poll period for timer, which
175 * is used to read throughput values periodically.
177 * Output: Alarm period, default value(if env var not set) otherwise.
178 ****************************************************************************/
180 int get_poll_interval_seconds()
183 int ret_val = DEFAULT_POLL_INTERVAL_SECONDS;
185 /* Get Alarm period for reading the Lustre client table. */
187 alarm_period = getenv(POLL_INTERVAL_ENV_VAR);
188 if (alarm_period != NULL) {
189 char *ptr = alarm_period;
190 while(isdigit(*ptr)) ptr++;
192 /* if we have only digits then conver it*/
194 int time = atoi(alarm_period);
196 ret_val = time; /* Alarm period in seconds */
202 /*****************************************************************************
203 * Function: health_poll_worker
205 * Description: This is the routine registered to system timer for updating
206 * the throughput values for all the clients and its respective osc(s).
208 * Input: 'registration_number` value obtained during the alarm registration
209 * 'clientarg' pointing to user defined data type.
211 *****************************************************************************/
213 void health_poll_worker(unsigned int registration_number, void *clientarg)
215 health_entry_parser();
217 /* Register the function again to call after lustre_alarm_period */
218 if (!snmp_alarm_register(g_poll_interval_seconds, 0, health_poll_worker, NULL)) {
219 report("%s %s:line %d %s", __FILE__, __FUNCTION__, __LINE__,
220 "snmp_alarm_register failed");
224 /*****************************************************************************
225 * Function: health_entry_parser
227 * Description: This routine is called to parse the health_check entry
231 *****************************************************************************/
233 void health_entry_parser(void)
236 char string[MAX_LINE_SIZE];
237 int b_seen_portals_catastrophe = 0;
238 const char *filename = g_health_check_test_file == 0 ?
239 LUSTRE_PATH FILENAME_SYSHEALTHCHECK :
240 g_health_check_test_file;
242 /*DEBUGMSGTL(("lsnmpd","health_entry_parser(%s)\n",filename));*/
244 /* Open the file. Use the test file env variable if
246 fptr = fopen(filename,"r");
248 /* If the path is not found do nothing */
252 while( NULL != fgets(string, sizeof(string), fptr)){
254 /*DEBUGMSGTL(("lsnmpd","health_entry_parser() looking at = \'%s\'\n",string));*/
257 * First handle the portals catastrophe
258 * Look for the string "LBUG"
260 if(0 == strncmp(string,"LBUG",4)){
262 * If we haven't sent the catastrophe message yet
263 * send it now. And keep track that we've sent it
265 if(!g_sent_portals_catastrophe){
266 send_portals_catastrophe_trap("LBUG");
267 g_sent_portals_catastrophe = 1;
269 b_seen_portals_catastrophe = 1;
273 * Now handle any of the OBD object failures
274 * look for "device <OBDNAME> reported unhealthy"
276 else if(0 == strncmp(string,"device ",7)){
277 char *obd_name = string+7;
278 char *space_after_obd_name;
281 * Now find the space after the obd name
282 * Again if there is no space we're in trouble
284 space_after_obd_name = strchr(obd_name,' ');
285 if(space_after_obd_name == 0)
289 * Null terminate the obd_name
291 *space_after_obd_name = 0;
293 DEBUGMSGTL(("lsnmpd","Looking at obd=%s\n",obd_name));
296 * If we haven't sent a trap for this one
299 if(is_obd_newly_unhealthy(obd_name))
300 send_obd_unhealthy_trap(obd_name,"unhealthy");
304 /* If we don't find it reset the catastrope flag*/
305 if(!b_seen_portals_catastrophe && g_sent_portals_catastrophe)
307 DEBUGMSGTL(("lsnmpd","LBUG has been cleared\n"));
308 g_sent_portals_catastrophe = 0;
312 * Any <OBDNAMES> that weren't queried above are now unhealthy.
313 * Scan through and cleanup the newly healthy obds
315 obd_unhealthy_scan();
320 /*****************************************************************************
321 * Function: send_portals_catastrophe_trap
323 * Description: Send the SNMP V2 trap
325 * Input: 'reason_string' the reason for the catastrope.
328 *****************************************************************************/
330 void send_portals_catastrophe_trap(char *reason_string)
333 * Setup the trap variables.
334 * It's a linked list of netsnmp_variable_list items.
336 netsnmp_variable_list var_trap[2];
338 DEBUGMSGTL(("lsnmpd","Sending portals catastrophe trap reason=%s\n",reason_string));
341 * Setup the first variable in the trap data.
342 * Have it chain to another variable.
344 var_trap[0].next_variable = &var_trap[1];
346 /*The "name" must be the standard snmp "trap" OID.*/
347 var_trap[0].name = objid_snmptrap;
348 var_trap[0].name_length = sizeof(objid_snmptrap) / sizeof(oid);
350 /*But the data contained in this variable, is an OID that is the trap OID.*/
351 var_trap[0].type = ASN_OBJECT_ID;
352 var_trap[0].val.objid = lustre_portals_trap;
353 var_trap[0].val_len = sizeof(lustre_portals_trap);
356 * Setup the second variable in the trap data.
357 * It is the last in the chain so set next to NULL
359 var_trap[1].next_variable = NULL;
361 /*The "name" is the OID of the portals trap reason strong*/
362 var_trap[1].name = lustre_portals_trap_string;
363 var_trap[1].name_length = sizeof(lustre_portals_trap_string) / sizeof(oid);
365 /*And the data is a octet string, that contains the actually reason string*/
366 var_trap[1].type = ASN_OCTET_STR;
367 var_trap[1].val.string = reason_string;
368 var_trap[1].val_len = strlen(reason_string);
370 /*And now send off the trap*/
371 send_v2trap(var_trap);
375 /*****************************************************************************
376 * Function: send_obd_unhealthy_trap
378 * Description: Send the SNMP V2 trap
380 * Input: 'obd_name' the name of the obd
381 * 'reason_string' the reason for the catastrope.
383 *****************************************************************************/
385 void send_obd_unhealthy_trap(char *obd_name,char *reason_string)
388 * Setup the trap variables.
389 * It's a linked list of netsnmp_variable_list items.
391 netsnmp_variable_list var_trap[3];
393 DEBUGMSGTL(("lsnmpd","Sending OBD unhealthy trap obd=%s reason=%s\n",obd_name,reason_string));
396 * Setup the first variable in the trap data.
397 * Have it chain to another variable.
399 var_trap[0].next_variable = &var_trap[1];
401 /*The "name" must be the standard snmp "trap" OID.*/
402 var_trap[0].name = objid_snmptrap;
403 var_trap[0].name_length = sizeof(objid_snmptrap) / sizeof(oid);
405 /*But the data contained in this variable, is an OID that is the trap OID.*/
406 var_trap[0].type = ASN_OBJECT_ID;
407 var_trap[0].val.objid = lustre_unhealthy_trap;
408 var_trap[0].val_len = sizeof(lustre_unhealthy_trap);
411 * Setup the second variable in the trap data.
412 * Have it chain to another variable.
414 var_trap[1].next_variable = &var_trap[2];;
416 /*The "name" is the OID of the portals trap reason strong*/
417 var_trap[1].name = lustre_unhealthy_trap_device_name_string;
418 var_trap[1].name_length = sizeof(lustre_unhealthy_trap_device_name_string) / sizeof(oid);
420 /*And the data is a octet string, that contains the actually reason strong*/
421 var_trap[1].type = ASN_OCTET_STR;
422 var_trap[1].val.string = obd_name;
423 var_trap[1].val_len = strlen(obd_name);
426 * Setup the third variable in the trap data.
427 * It is the last in the chain so set next to NULL
429 var_trap[2].next_variable = NULL;
431 /*The "name" is the OID of the portals trap reason strong*/
432 var_trap[2].name = lustre_unhealthy_trap_reason_string;
433 var_trap[2].name_length = sizeof(lustre_unhealthy_trap_reason_string) / sizeof(oid);
435 /*And the data is a octet string, that contains the actually reason strong*/
436 var_trap[2].type = ASN_OCTET_STR;
437 var_trap[2].val.string = reason_string;
438 var_trap[2].val_len = strlen(reason_string);
440 /*And now send off the trap*/
441 send_v2trap(var_trap);
445 /*****************************************************************************
446 * Function: is_obd_newly_unhealthy
448 * Description: Deterime if the obd is going from health->unhealth
449 * Also mark all unhealhy (new and old) as seen.
451 * Input: 'obd_name' the name of the obd
453 * Output: 1 if newly unhealthy 0 if previolsy unhealthy
454 *****************************************************************************/
456 int is_obd_newly_unhealthy(const char* obd_name)
458 /*for all elements in g_obd_unhealthy_list*/
459 obd_unhealthy_entry* walker;
460 obd_unhealthy_entry* entry;
463 for(walker = g_obd_unhealthy_list; walker != 0; walker = walker->next)
465 /*If the names match*/
466 if(0 == strcmp (walker->name,obd_name))
468 /* Commented out because it was just to noisy!
469 * DEBUGMSGTL(("lsnmpd","obd %s was already unhealthy\n",obd_name));
472 /*Mark the entry as seen, and return that it was previously unhealthy*/
478 DEBUGMSGTL(("lsnmpd","obd %s is now unhealthy\n",obd_name));
480 /*We didn't find an entry so we need to create a new one. */
481 /*Calculate the obd_name length*/
482 name_len = strlen(obd_name)+1;
484 /*Allocate a new entry*/
485 entry = malloc(sizeof(*entry) + name_len);
487 /*Put this element at the front of the list*/
488 entry->next = g_obd_unhealthy_list;
489 g_obd_unhealthy_list = entry;
491 /*Mark it initially as seen*/
494 /*And copy the entry name*/
495 memcpy(entry->name,obd_name,name_len);
497 /*return this obd as newly unhealthy.*/
502 /*****************************************************************************
503 * Function: obd_unhealthy_scan
505 * Description: Deterime if any obd is going from unhealthy->healthy
506 * Any of the obds that weren't "seen" by the
507 * is_obd_newly_unhealthy() pass are now health so
508 * remove them from the lists
509 * Also clear all "seen" flags.
513 *****************************************************************************/
515 void obd_unhealthy_scan(void)
517 /*fore all elements in g_obd_unhealthy_list*/
518 obd_unhealthy_entry* walker = g_obd_unhealthy_list;
519 obd_unhealthy_entry* prev = 0;
522 /*remove any that was not seen as unhealthy the last time*/
523 if(walker->seen == 0)
525 /*Remove element from the list, but first fix up the walker pointer*/
526 obd_unhealthy_entry* temp = walker;
528 DEBUGMSGTL(("lsnmpd","obd %s is now healthy\n",walker->name));
530 walker = walker->next;
532 /*Now adjust the pointers to effectively remove this entry*/
534 g_obd_unhealthy_list = walker;
538 /*And free the pointer. */
540 /*walker and prev are correctly setup so we can go around the loop again.*/
543 /*Mark all other entries as NOT seen for next pass through*/
547 /*Go onto the next entry*/
549 walker = walker->next;