4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
31 * This file is part of Lustre, http://www.lustre.org/
32 * Lustre is a trademark of Sun Microsystems, Inc.
34 * snmp/lustre-snmp-trap.c
36 * Author: PJ Kirner <pjkirner@clusterfs.com>
40 * include important headers
43 #include <net-snmp/net-snmp-config.h>
44 #include <net-snmp/net-snmp-includes.h>
45 #include <net-snmp/agent/net-snmp-agent-includes.h>
51 #include <sys/types.h>
59 #include "lustre-snmp-util.h"
61 /**************************************************************************
63 *************************************************************************/
65 #define DEFAULT_POLL_INTERVAL_SECONDS 60
66 #define POLL_INTERVAL_ENV_VAR "LSNMP_POLL_INTERVAL"
67 #define SNMP_HEALTH_CHECK_TEST_FILE "LSNMP_HEALTH_CHECK_TEST_FILE"
69 /**************************************************************************
71 *************************************************************************/
73 static oid objid_snmptrap[] =
74 { 1,3,6,1,6,3,1,1,4,1,0};
75 static oid lustre_portals_trap[] =
76 { 1,3,6,1,4,1,13140,2,1,0,1};
77 static oid lustre_portals_trap_string[]=
78 { 1,3,6,1,4,1,13140,2,1,0,2};
79 static oid lustre_unhealthy_trap[] =
80 { 1,3,6,1,4,1,13140,2,1,0,3};
81 static oid lustre_unhealthy_trap_device_name_string[]=
82 { 1,3,6,1,4,1,13140,2,1,0,4};
83 static oid lustre_unhealthy_trap_reason_string[]=
84 { 1,3,6,1,4,1,13140,2,1,0,5};
86 /**************************************************************************
88 *************************************************************************/
90 typedef struct obd_unhealthy_entry_struct{
92 /*1-if seen as part of the the is_unhealthy scan, otherwise 0*/
95 /*single linked list pointer*/
96 struct obd_unhealthy_entry_struct *next;
98 /*obdname - variable size*/
101 }obd_unhealthy_entry;
103 /**************************************************************************
105 *************************************************************************/
107 int get_poll_interval_seconds();
108 void health_poll_worker(unsigned int registration_number, void *clientarg);
109 void send_portals_catastrophe_trap(char *reason_string);
110 void send_obd_unhealthy_trap(char *obd_name,char *reason_string);
111 int is_obd_newly_unhealthy(const char* obd_name);
112 void obd_unhealthy_scan(void);
113 void health_entry_parser(void);
115 /**************************************************************************
117 *************************************************************************/
119 static int g_sent_portals_catastrophe = 0;
120 static obd_unhealthy_entry* g_obd_unhealthy_list = NULL;
121 static int g_poll_interval_seconds;
122 static unsigned int g_registration_handle;
123 static char *g_health_check_test_file = 0;
125 /*****************************************************************************
126 * Function: initialize_trap_handler
128 * Description: Initlized the trap poll haalder.
132 * Output: Global g_poll_interval_seconds is set.
134 ****************************************************************************/
136 void initialize_trap_handler(void)
138 g_poll_interval_seconds = get_poll_interval_seconds();
140 g_registration_handle = snmp_alarm_register(g_poll_interval_seconds, 0, health_poll_worker, NULL);
141 if (g_registration_handle == 0)
142 report("%s %s: line %d %s", __FILE__, __FUNCTION__, __LINE__,
143 "snmp_alarm_register failed");
145 DEBUGMSGTL(("lsnmpd","lsnmp alarm registered poll interval = %d seconds\n",g_poll_interval_seconds));
147 g_health_check_test_file = getenv(SNMP_HEALTH_CHECK_TEST_FILE);
148 if(g_health_check_test_file != 0)
149 DEBUGMSGTL(("lsnmpd","lsnmp health check test file set to \'%s\'\n",g_health_check_test_file));
152 /*****************************************************************************
153 * Function: terminate_trap_handler
155 * Description: Terminate the trap poll haalder.
159 * Output: Global g_poll_interval_seconds is set.
161 ****************************************************************************/
163 void terminate_trap_handler(void)
165 snmp_alarm_unregister(g_registration_handle);
168 /*****************************************************************************
169 * Function: get_poll_interval_seconds
171 * Description: This function used to get the poll period for timer, which
172 * is used to read throughput values periodically.
174 * Output: Alarm period, default value(if env var not set) otherwise.
175 ****************************************************************************/
177 int get_poll_interval_seconds()
180 int ret_val = DEFAULT_POLL_INTERVAL_SECONDS;
182 /* Get Alarm period for reading the Lustre client table. */
184 alarm_period = getenv(POLL_INTERVAL_ENV_VAR);
185 if (alarm_period != NULL) {
186 char *ptr = alarm_period;
187 while(isdigit(*ptr)) ptr++;
189 /* if we have only digits then conver it*/
191 int time = atoi(alarm_period);
193 ret_val = time; /* Alarm period in seconds */
199 /*****************************************************************************
200 * Function: health_poll_worker
202 * Description: This is the routine registered to system timer for updating
203 * the throughput values for all the clients and its respective osc(s).
205 * Input: 'registration_number` value obtained during the alarm registration
206 * 'clientarg' pointing to user defined data type.
208 *****************************************************************************/
210 void health_poll_worker(unsigned int registration_number, void *clientarg)
212 health_entry_parser();
214 /* Register the function again to call after lustre_alarm_period */
215 if (!snmp_alarm_register(g_poll_interval_seconds, 0, health_poll_worker, NULL)) {
216 report("%s %s:line %d %s", __FILE__, __FUNCTION__, __LINE__,
217 "snmp_alarm_register failed");
221 /*****************************************************************************
222 * Function: health_entry_parser
224 * Description: This routine is called to parse the health_check entry
228 *****************************************************************************/
230 void health_entry_parser(void)
233 char string[MAX_LINE_SIZE];
234 int b_seen_portals_catastrophe = 0;
235 const char *filename = g_health_check_test_file == 0 ?
236 LUSTRE_PATH FILENAME_SYSHEALTHCHECK :
237 g_health_check_test_file;
239 /*DEBUGMSGTL(("lsnmpd","health_entry_parser(%s)\n",filename));*/
241 /* Open the file. Use the test file env variable if
243 fptr = fopen(filename,"r");
245 /* If the path is not found do nothing */
249 while( NULL != fgets(string, sizeof(string), fptr)){
251 /*DEBUGMSGTL(("lsnmpd","health_entry_parser() looking at = \'%s\'\n",string));*/
254 * First handle the portals catastrophe
255 * Look for the string "LBUG"
257 if(0 == strncmp(string,"LBUG",4)){
259 * If we haven't sent the catastrophe message yet
260 * send it now. And keep track that we've sent it
262 if(!g_sent_portals_catastrophe){
263 send_portals_catastrophe_trap("LBUG");
264 g_sent_portals_catastrophe = 1;
266 b_seen_portals_catastrophe = 1;
270 * Now handle any of the OBD object failures
271 * look for "device <OBDNAME> reported unhealthy"
273 else if(0 == strncmp(string,"device ",7)){
274 char *obd_name = string+7;
275 char *space_after_obd_name;
278 * Now find the space after the obd name
279 * Again if there is no space we're in trouble
281 space_after_obd_name = strchr(obd_name,' ');
282 if(space_after_obd_name == 0)
286 * Null terminate the obd_name
288 *space_after_obd_name = 0;
290 DEBUGMSGTL(("lsnmpd","Looking at obd=%s\n",obd_name));
293 * If we haven't sent a trap for this one
296 if(is_obd_newly_unhealthy(obd_name))
297 send_obd_unhealthy_trap(obd_name,"unhealthy");
301 /* If we don't find it reset the catastrope flag*/
302 if(!b_seen_portals_catastrophe && g_sent_portals_catastrophe)
304 DEBUGMSGTL(("lsnmpd","LBUG has been cleared\n"));
305 g_sent_portals_catastrophe = 0;
309 * Any <OBDNAMES> that weren't queried above are now unhealthy.
310 * Scan through and cleanup the newly healthy obds
312 obd_unhealthy_scan();
317 /*****************************************************************************
318 * Function: send_portals_catastrophe_trap
320 * Description: Send the SNMP V2 trap
322 * Input: 'reason_string' the reason for the catastrope.
325 *****************************************************************************/
327 void send_portals_catastrophe_trap(char *reason_string)
330 * Setup the trap variables.
331 * It's a linked list of netsnmp_variable_list items.
333 netsnmp_variable_list var_trap[2];
335 DEBUGMSGTL(("lsnmpd","Sending portals catastrophe trap reason=%s\n",reason_string));
338 * Setup the first variable in the trap data.
339 * Have it chain to another variable.
341 var_trap[0].next_variable = &var_trap[1];
343 /*The "name" must be the standard snmp "trap" OID.*/
344 var_trap[0].name = objid_snmptrap;
345 var_trap[0].name_length = sizeof(objid_snmptrap) / sizeof(oid);
347 /*But the data contained in this variable, is an OID that is the trap OID.*/
348 var_trap[0].type = ASN_OBJECT_ID;
349 var_trap[0].val.objid = lustre_portals_trap;
350 var_trap[0].val_len = sizeof(lustre_portals_trap);
353 * Setup the second variable in the trap data.
354 * It is the last in the chain so set next to NULL
356 var_trap[1].next_variable = NULL;
358 /* The "name" is the OID of the portals trap reason string */
359 var_trap[1].name = lustre_portals_trap_string;
360 var_trap[1].name_length = sizeof(lustre_portals_trap_string) / sizeof(oid);
362 /* And the data is an octet string, that contains the actually reason
364 var_trap[1].type = ASN_OCTET_STR;
365 var_trap[1].val.string = (unsigned char *)reason_string;
366 var_trap[1].val_len = strlen(reason_string);
368 /*And now send off the trap*/
369 send_v2trap(var_trap);
373 /*****************************************************************************
374 * Function: send_obd_unhealthy_trap
376 * Description: Send the SNMP V2 trap
378 * Input: 'obd_name' the name of the obd
379 * 'reason_string' the reason for the catastrope.
381 *****************************************************************************/
383 void send_obd_unhealthy_trap(char *obd_name,char *reason_string)
386 * Setup the trap variables.
387 * It's a linked list of netsnmp_variable_list items.
389 netsnmp_variable_list var_trap[3];
391 DEBUGMSGTL(("lsnmpd","Sending OBD unhealthy trap obd=%s reason=%s\n",obd_name,reason_string));
394 * Setup the first variable in the trap data.
395 * Have it chain to another variable.
397 var_trap[0].next_variable = &var_trap[1];
399 /*The "name" must be the standard snmp "trap" OID.*/
400 var_trap[0].name = objid_snmptrap;
401 var_trap[0].name_length = sizeof(objid_snmptrap) / sizeof(oid);
403 /*But the data contained in this variable, is an OID that is the trap OID.*/
404 var_trap[0].type = ASN_OBJECT_ID;
405 var_trap[0].val.objid = lustre_unhealthy_trap;
406 var_trap[0].val_len = sizeof(lustre_unhealthy_trap);
409 * Setup the second variable in the trap data.
410 * Have it chain to another variable.
412 var_trap[1].next_variable = &var_trap[2];;
414 /* The "name" is the OID of the portals trap reason string */
415 var_trap[1].name = lustre_unhealthy_trap_device_name_string;
416 var_trap[1].name_length = sizeof(lustre_unhealthy_trap_device_name_string) / sizeof(oid);
418 /* And the data is an octet string, that contains the actual reason
420 var_trap[1].type = ASN_OCTET_STR;
421 var_trap[1].val.string = (unsigned char *)obd_name;
422 var_trap[1].val_len = strlen(obd_name);
425 * Setup the third variable in the trap data.
426 * It is the last in the chain so set next to NULL
428 var_trap[2].next_variable = NULL;
430 /* The "name" is the OID of the portals trap reason string */
431 var_trap[2].name = lustre_unhealthy_trap_reason_string;
432 var_trap[2].name_length = sizeof(lustre_unhealthy_trap_reason_string) / sizeof(oid);
434 /* And the data is an octet string, that contains the actual reason
436 var_trap[2].type = ASN_OCTET_STR;
437 var_trap[2].val.string = (unsigned char *)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;