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>
52 #if defined (__linux__)
61 #include "lustre-snmp-util.h"
63 /**************************************************************************
65 *************************************************************************/
67 #define DEFAULT_POLL_INTERVAL_SECONDS 60
68 #define POLL_INTERVAL_ENV_VAR "LSNMP_POLL_INTERVAL"
69 #define SNMP_HEALTH_CHECK_TEST_FILE "LSNMP_HEALTH_CHECK_TEST_FILE"
71 /**************************************************************************
73 *************************************************************************/
75 static oid objid_snmptrap[] =
76 { 1,3,6,1,6,3,1,1,4,1,0};
77 static oid lustre_portals_trap[] =
78 { 1,3,6,1,4,1,13140,2,1,0,1};
79 static oid lustre_portals_trap_string[]=
80 { 1,3,6,1,4,1,13140,2,1,0,2};
81 static oid lustre_unhealthy_trap[] =
82 { 1,3,6,1,4,1,13140,2,1,0,3};
83 static oid lustre_unhealthy_trap_device_name_string[]=
84 { 1,3,6,1,4,1,13140,2,1,0,4};
85 static oid lustre_unhealthy_trap_reason_string[]=
86 { 1,3,6,1,4,1,13140,2,1,0,5};
88 /**************************************************************************
90 *************************************************************************/
92 typedef struct obd_unhealthy_entry_struct{
94 /*1-if seen as part of the the is_unhealthy scan, otherwise 0*/
97 /*single linked list pointer*/
98 struct obd_unhealthy_entry_struct *next;
100 /*obdname - variable size*/
103 }obd_unhealthy_entry;
105 /**************************************************************************
107 *************************************************************************/
109 int get_poll_interval_seconds();
110 void health_poll_worker(unsigned int registration_number, void *clientarg);
111 void send_portals_catastrophe_trap(char *reason_string);
112 void send_obd_unhealthy_trap(char *obd_name,char *reason_string);
113 int is_obd_newly_unhealthy(const char* obd_name);
114 void obd_unhealthy_scan(void);
115 void health_entry_parser(void);
117 /**************************************************************************
119 *************************************************************************/
121 static int g_sent_portals_catastrophe = 0;
122 static obd_unhealthy_entry* g_obd_unhealthy_list = NULL;
123 static int g_poll_interval_seconds;
124 static unsigned int g_registration_handle;
125 static char *g_health_check_test_file = 0;
127 /*****************************************************************************
128 * Function: initilize_trap_handler
130 * Description: Initlized the trap poll haalder.
134 * Output: Global g_poll_interval_seconds is set.
136 ****************************************************************************/
138 void initilize_trap_handler(void)
140 g_poll_interval_seconds = get_poll_interval_seconds();
142 g_registration_handle = snmp_alarm_register(g_poll_interval_seconds, 0, health_poll_worker, NULL);
143 if (g_registration_handle == 0)
144 report("%s %s: line %d %s", __FILE__, __FUNCTION__, __LINE__,
145 "snmp_alarm_register failed");
147 DEBUGMSGTL(("lsnmpd","lsnmp alarm registered poll interval = %d seconds\n",g_poll_interval_seconds));
149 g_health_check_test_file = getenv(SNMP_HEALTH_CHECK_TEST_FILE);
150 if(g_health_check_test_file != 0)
151 DEBUGMSGTL(("lsnmpd","lsnmp health check test file set to \'%s\'\n",g_health_check_test_file));
154 /*****************************************************************************
155 * Function: terminate_trap_handler
157 * Description: Terminate the trap poll haalder.
161 * Output: Global g_poll_interval_seconds is set.
163 ****************************************************************************/
165 void terminate_trap_handler(void)
167 snmp_alarm_unregister(g_registration_handle);
170 /*****************************************************************************
171 * Function: get_poll_interval_seconds
173 * Description: This function used to get the poll period for timer, which
174 * is used to read throughput values periodically.
176 * Output: Alarm period, default value(if env var not set) otherwise.
177 ****************************************************************************/
179 int get_poll_interval_seconds()
182 int ret_val = DEFAULT_POLL_INTERVAL_SECONDS;
184 /* Get Alarm period for reading the Lustre client table. */
186 alarm_period = getenv(POLL_INTERVAL_ENV_VAR);
187 if (alarm_period != NULL) {
188 char *ptr = alarm_period;
189 while(isdigit(*ptr)) ptr++;
191 /* if we have only digits then conver it*/
193 int time = atoi(alarm_period);
195 ret_val = time; /* Alarm period in seconds */
201 /*****************************************************************************
202 * Function: health_poll_worker
204 * Description: This is the routine registered to system timer for updating
205 * the throughput values for all the clients and its respective osc(s).
207 * Input: 'registration_number` value obtained during the alarm registration
208 * 'clientarg' pointing to user defined data type.
210 *****************************************************************************/
212 void health_poll_worker(unsigned int registration_number, void *clientarg)
214 health_entry_parser();
216 /* Register the function again to call after lustre_alarm_period */
217 if (!snmp_alarm_register(g_poll_interval_seconds, 0, health_poll_worker, NULL)) {
218 report("%s %s:line %d %s", __FILE__, __FUNCTION__, __LINE__,
219 "snmp_alarm_register failed");
223 /*****************************************************************************
224 * Function: health_entry_parser
226 * Description: This routine is called to parse the health_check entry
230 *****************************************************************************/
232 void health_entry_parser(void)
235 char string[MAX_LINE_SIZE];
236 int b_seen_portals_catastrophe = 0;
237 const char *filename = g_health_check_test_file == 0 ?
238 LUSTRE_PATH FILENAME_SYSHEALTHCHECK :
239 g_health_check_test_file;
241 /*DEBUGMSGTL(("lsnmpd","health_entry_parser(%s)\n",filename));*/
243 /* Open the file. Use the test file env variable if
245 fptr = fopen(filename,"r");
247 /* If the path is not found do nothing */
251 while( NULL != fgets(string, sizeof(string), fptr)){
253 /*DEBUGMSGTL(("lsnmpd","health_entry_parser() looking at = \'%s\'\n",string));*/
256 * First handle the portals catastrophe
257 * Look for the string "LBUG"
259 if(0 == strncmp(string,"LBUG",4)){
261 * If we haven't sent the catastrophe message yet
262 * send it now. And keep track that we've sent it
264 if(!g_sent_portals_catastrophe){
265 send_portals_catastrophe_trap("LBUG");
266 g_sent_portals_catastrophe = 1;
268 b_seen_portals_catastrophe = 1;
272 * Now handle any of the OBD object failures
273 * look for "device <OBDNAME> reported unhealthy"
275 else if(0 == strncmp(string,"device ",7)){
276 char *obd_name = string+7;
277 char *space_after_obd_name;
280 * Now find the space after the obd name
281 * Again if there is no space we're in trouble
283 space_after_obd_name = strchr(obd_name,' ');
284 if(space_after_obd_name == 0)
288 * Null terminate the obd_name
290 *space_after_obd_name = 0;
292 DEBUGMSGTL(("lsnmpd","Looking at obd=%s\n",obd_name));
295 * If we haven't sent a trap for this one
298 if(is_obd_newly_unhealthy(obd_name))
299 send_obd_unhealthy_trap(obd_name,"unhealthy");
303 /* If we don't find it reset the catastrope flag*/
304 if(!b_seen_portals_catastrophe && g_sent_portals_catastrophe)
306 DEBUGMSGTL(("lsnmpd","LBUG has been cleared\n"));
307 g_sent_portals_catastrophe = 0;
311 * Any <OBDNAMES> that weren't queried above are now unhealthy.
312 * Scan through and cleanup the newly healthy obds
314 obd_unhealthy_scan();
319 /*****************************************************************************
320 * Function: send_portals_catastrophe_trap
322 * Description: Send the SNMP V2 trap
324 * Input: 'reason_string' the reason for the catastrope.
327 *****************************************************************************/
329 void send_portals_catastrophe_trap(char *reason_string)
332 * Setup the trap variables.
333 * It's a linked list of netsnmp_variable_list items.
335 netsnmp_variable_list var_trap[2];
337 DEBUGMSGTL(("lsnmpd","Sending portals catastrophe trap reason=%s\n",reason_string));
340 * Setup the first variable in the trap data.
341 * Have it chain to another variable.
343 var_trap[0].next_variable = &var_trap[1];
345 /*The "name" must be the standard snmp "trap" OID.*/
346 var_trap[0].name = objid_snmptrap;
347 var_trap[0].name_length = sizeof(objid_snmptrap) / sizeof(oid);
349 /*But the data contained in this variable, is an OID that is the trap OID.*/
350 var_trap[0].type = ASN_OBJECT_ID;
351 var_trap[0].val.objid = lustre_portals_trap;
352 var_trap[0].val_len = sizeof(lustre_portals_trap);
355 * Setup the second variable in the trap data.
356 * It is the last in the chain so set next to NULL
358 var_trap[1].next_variable = NULL;
360 /*The "name" is the OID of the portals trap reason strong*/
361 var_trap[1].name = lustre_portals_trap_string;
362 var_trap[1].name_length = sizeof(lustre_portals_trap_string) / sizeof(oid);
364 /*And the data is a octet string, that contains the actually reason string*/
365 var_trap[1].type = ASN_OCTET_STR;
366 var_trap[1].val.string = (unsigned char *)reason_string;
367 var_trap[1].val_len = strlen(reason_string);
369 /*And now send off the trap*/
370 send_v2trap(var_trap);
374 /*****************************************************************************
375 * Function: send_obd_unhealthy_trap
377 * Description: Send the SNMP V2 trap
379 * Input: 'obd_name' the name of the obd
380 * 'reason_string' the reason for the catastrope.
382 *****************************************************************************/
384 void send_obd_unhealthy_trap(char *obd_name,char *reason_string)
387 * Setup the trap variables.
388 * It's a linked list of netsnmp_variable_list items.
390 netsnmp_variable_list var_trap[3];
392 DEBUGMSGTL(("lsnmpd","Sending OBD unhealthy trap obd=%s reason=%s\n",obd_name,reason_string));
395 * Setup the first variable in the trap data.
396 * Have it chain to another variable.
398 var_trap[0].next_variable = &var_trap[1];
400 /*The "name" must be the standard snmp "trap" OID.*/
401 var_trap[0].name = objid_snmptrap;
402 var_trap[0].name_length = sizeof(objid_snmptrap) / sizeof(oid);
404 /*But the data contained in this variable, is an OID that is the trap OID.*/
405 var_trap[0].type = ASN_OBJECT_ID;
406 var_trap[0].val.objid = lustre_unhealthy_trap;
407 var_trap[0].val_len = sizeof(lustre_unhealthy_trap);
410 * Setup the second variable in the trap data.
411 * Have it chain to another variable.
413 var_trap[1].next_variable = &var_trap[2];;
415 /*The "name" is the OID of the portals trap reason strong*/
416 var_trap[1].name = lustre_unhealthy_trap_device_name_string;
417 var_trap[1].name_length = sizeof(lustre_unhealthy_trap_device_name_string) / sizeof(oid);
419 /*And the data is a octet string, that contains the actually reason strong*/
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 strong*/
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 a octet string, that contains the actually reason strong*/
435 var_trap[2].type = ASN_OCTET_STR;
436 var_trap[2].val.string = (unsigned char *)reason_string;
437 var_trap[2].val_len = strlen(reason_string);
439 /*And now send off the trap*/
440 send_v2trap(var_trap);
444 /*****************************************************************************
445 * Function: is_obd_newly_unhealthy
447 * Description: Deterime if the obd is going from health->unhealth
448 * Also mark all unhealhy (new and old) as seen.
450 * Input: 'obd_name' the name of the obd
452 * Output: 1 if newly unhealthy 0 if previolsy unhealthy
453 *****************************************************************************/
455 int is_obd_newly_unhealthy(const char* obd_name)
457 /*for all elements in g_obd_unhealthy_list*/
458 obd_unhealthy_entry* walker;
459 obd_unhealthy_entry* entry;
462 for(walker = g_obd_unhealthy_list; walker != 0; walker = walker->next)
464 /*If the names match*/
465 if(0 == strcmp (walker->name,obd_name))
467 /* Commented out because it was just to noisy!
468 * DEBUGMSGTL(("lsnmpd","obd %s was already unhealthy\n",obd_name));
471 /*Mark the entry as seen, and return that it was previously unhealthy*/
477 DEBUGMSGTL(("lsnmpd","obd %s is now unhealthy\n",obd_name));
479 /*We didn't find an entry so we need to create a new one. */
480 /*Calculate the obd_name length*/
481 name_len = strlen(obd_name)+1;
483 /*Allocate a new entry*/
484 entry = malloc(sizeof(*entry) + name_len);
486 /*Put this element at the front of the list*/
487 entry->next = g_obd_unhealthy_list;
488 g_obd_unhealthy_list = entry;
490 /*Mark it initially as seen*/
493 /*And copy the entry name*/
494 memcpy(entry->name,obd_name,name_len);
496 /*return this obd as newly unhealthy.*/
501 /*****************************************************************************
502 * Function: obd_unhealthy_scan
504 * Description: Deterime if any obd is going from unhealthy->healthy
505 * Any of the obds that weren't "seen" by the
506 * is_obd_newly_unhealthy() pass are now health so
507 * remove them from the lists
508 * Also clear all "seen" flags.
512 *****************************************************************************/
514 void obd_unhealthy_scan(void)
516 /*fore all elements in g_obd_unhealthy_list*/
517 obd_unhealthy_entry* walker = g_obd_unhealthy_list;
518 obd_unhealthy_entry* prev = 0;
521 /*remove any that was not seen as unhealthy the last time*/
522 if(walker->seen == 0)
524 /*Remove element from the list, but first fix up the walker pointer*/
525 obd_unhealthy_entry* temp = walker;
527 DEBUGMSGTL(("lsnmpd","obd %s is now healthy\n",walker->name));
529 walker = walker->next;
531 /*Now adjust the pointers to effectively remove this entry*/
533 g_obd_unhealthy_list = walker;
537 /*And free the pointer. */
539 /*walker and prev are correctly setup so we can go around the loop again.*/
542 /*Mark all other entries as NOT seen for next pass through*/
546 /*Go onto the next entry*/
548 walker = walker->next;