module.c revision 888e055994b8b0dc77b98c53dd97026237caec5d
/*
* CDDL HEADER START
*
* The contents of this file are subject to the terms of the
* Common Development and Distribution License (the "License").
* You may not use this file except in compliance with the License.
*
* You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
* or http://www.opensolaris.org/os/licensing.
* See the License for the specific language governing permissions
* and limitations under the License.
*
* When distributing Covered Code, include this CDDL HEADER in each
* file and include the License file at usr/src/OPENSOLARIS.LICENSE.
* If applicable, add the following below this CDDL HEADER, with the
* fields enclosed by brackets "[]" replaced with your own identifying
* information: Portions Copyright [yyyy] [name of copyright owner]
*
* CDDL HEADER END
*/
/*
* Copyright 2009 Sun Microsystems, Inc. All rights reserved.
* Use is subject to license terms.
*/
#include <fm/fmd_adm.h>
#include <fm/fmd_snmp.h>
#include <net-snmp/net-snmp-config.h>
#include <net-snmp/net-snmp-includes.h>
#include <net-snmp/agent/net-snmp-agent-includes.h>
#include <pthread.h>
#include <stddef.h>
#include <errno.h>
#include <libuutil.h>
#include "sunFM_impl.h"
#include "module.h"
static uu_avl_pool_t *mod_name_avl_pool;
static uu_avl_pool_t *mod_index_avl_pool;
static uu_avl_t *mod_name_avl;
static uu_avl_t *mod_index_avl;
#define VALID_AVL_STATE (mod_name_avl_pool != NULL && \
mod_index_avl_pool != NULL && mod_name_avl != NULL && \
mod_index_avl != NULL)
#define UPDATE_WAIT_MILLIS 10 /* poll interval in milliseconds */
/*
* Update types. Single-index and all are mutually exclusive.
*/
#define UCT_INDEX 0x1
#define UCT_ALL 0x2
#define UCT_FLAGS 0x3
#define MODULE_DATA_VALID(d) ((d)->d_valid == valid_stamp)
/*
* Locking rules are straightforward. There is only one updater thread
* for each table, and requests for update that are received while
* another update is in progress are ignored. The single per-table lock
* protects all the data for the table, the valid_stamp and max_index
* tags for new data, and - importantly - all the hidden static data
* used by the Net-SNMP library. The result return callbacks are always
* called in the master agent thread; holding the table lock is
* therefore sufficient since only one table's callback can be run at
* any one time. Finer-grained locking is possible here but
* substantially more difficult because nearly all Net-SNMP functions
* are unsafe.
*
* In practice this is more than adequate, since the purpose of
* threading out the update is to prevent excessively time-consuming
* data collection from bottlenecking the entire agent, not to improve
* result throughput (SNMP is not intended to be used for applications
* requiring high throughput anyway). If the agent itself ever becomes
* multithreaded, locking requirements become limited to our local
* per-table data (the tree, max_index, and valid_stamp), and the
* implementation could be revisited for better performance.
*/
static ulong_t max_index;
static int valid_stamp;
static pthread_mutex_t update_lock;
static pthread_cond_t update_cv;
static volatile enum { US_QUIET, US_NEEDED, US_INPROGRESS } update_status;
static Netsnmp_Node_Handler sunFmModuleTable_handler;
static sunFmModule_data_t *
key_build(const char *name, const ulong_t index)
{
static sunFmModule_data_t key;
key.d_index = index;
if (name)
(void) strlcpy(key.d_ami_name, name, sizeof (key.d_ami_name));
else
key.d_ami_name[0] = '\0';
return (&key);
}
/*
* If name is the name of a module we have previously seen and indexed, return
* data for it. Otherwise, return NULL. Note that the module may not be
* valid; that is, it may have been removed from the fault manager since its
* information was last updated.
*/
static sunFmModule_data_t *
module_lookup_name(const char *name)
{
sunFmModule_data_t *key;
key = key_build(name, 0);
return (uu_avl_find(mod_name_avl, key, NULL, NULL));
}
/*
* If index corresponds to a module we have previously seen and indexed, return
* data for it. Otherwise, return NULL. Note that the module may not be
* valid; that is, it may have been removed from the fault manager since its
* information was last updated.
*/
static sunFmModule_data_t *
module_lookup_index_exact(const ulong_t index)
{
sunFmModule_data_t *key;
key = key_build(NULL, index);
return (uu_avl_find(mod_index_avl, key, NULL, NULL));
}
/*
* If index corresponds to a valid (that is, extant as of latest information
* from the fault manager) fmd module, return the data for that module.
* Otherwise, return the data for the valid module whose index is as close as
* possible to index but not lower. This preserves the lexicographical
* ordering required for GETNEXT processing.
*/
static sunFmModule_data_t *
module_lookup_index_nextvalid(const ulong_t index)
{
sunFmModule_data_t *key, *data;
uu_avl_index_t idx;
key = key_build(NULL, index);
if ((data = uu_avl_find(mod_index_avl, key, NULL, &idx)) != NULL &&
MODULE_DATA_VALID(data))
return (data);
data = uu_avl_nearest_next(mod_index_avl, idx);
while (data != NULL && !MODULE_DATA_VALID(data))
data = uu_avl_next(mod_index_avl, data);
return (data);
}
/*
* Possible update the contents of a single module within the cache. This
* is our callback from fmd_module_iter.
*/
static int
modinfo_update_one(const fmd_adm_modinfo_t *modinfo, void *arg)
{
const sunFmModule_update_ctx_t *update_ctx =
(sunFmModule_update_ctx_t *)arg;
sunFmModule_data_t *data = module_lookup_name(modinfo->ami_name);
/*
* An fmd module we haven't seen before. We're obligated to index
* it and link it into our cache so that we can find it, but we're
* not obligated to fill it in completely unless we're doing a
* thorough update or this is the module we were asked for. This
* avoids unnecessary iteration and memory manipulation for data
* we're not going to return for this request.
*/
if (data == NULL) {
uu_avl_index_t idx;
DEBUGMSGTL((MODNAME_STR, "found new fmd module %s\n",
modinfo->ami_name));
if ((data = SNMP_MALLOC_TYPEDEF(sunFmModule_data_t)) == NULL) {
(void) snmp_log(LOG_ERR, MODNAME_STR ": Out of memory "
"for new module data at %s:%d\n", __FILE__,
__LINE__);
return (1);
}
/*
* We allocate indices sequentially and never reuse them.
* This ensures we can always return valid GETNEXT responses
* without having to reindex, and it provides the user a
* more consistent view of the fault manager.
*/
data->d_index = ++max_index;
DEBUGMSGTL((MODNAME_STR, "index %lu is %s@%p\n", data->d_index,
modinfo->ami_name, data));
(void) strlcpy(data->d_ami_name, modinfo->ami_name,
sizeof (data->d_ami_name));
uu_avl_node_init(data, &data->d_name_avl, mod_name_avl_pool);
(void) uu_avl_find(mod_name_avl, data, NULL, &idx);
uu_avl_insert(mod_name_avl, data, idx);
uu_avl_node_init(data, &data->d_index_avl, mod_index_avl_pool);
(void) uu_avl_find(mod_index_avl, data, NULL, &idx);
uu_avl_insert(mod_index_avl, data, idx);
DEBUGMSGTL((MODNAME_STR, "completed new module %lu/%s@%p\n",
data->d_index, data->d_ami_name, data));
}
data->d_valid = valid_stamp;
DEBUGMSGTL((MODNAME_STR, "timestamp updated for %lu/%s@%p: %lu\n",
data->d_index, data->d_ami_name, data, data->d_valid));
if ((update_ctx->uc_type & UCT_ALL) ||
update_ctx->uc_index == data->d_index) {
(void) strlcpy(data->d_ami_vers, modinfo->ami_vers,
sizeof (data->d_ami_vers));
(void) strlcpy(data->d_ami_desc, modinfo->ami_desc,
sizeof (data->d_ami_desc));
data->d_ami_flags = modinfo->ami_flags;
}
return (!(update_ctx->uc_type & UCT_ALL) &&
update_ctx->uc_index == data->d_index);
}
/*
* Update some or all module data from fmd. If thorough is set, all modules
* will be indexed and their data cached. Otherwise, updates will stop once
* the module matching index has been updated.
*
* Returns appropriate SNMP error codes.
*/
static int
modinfo_update(sunFmModule_update_ctx_t *update_ctx)
{
fmd_adm_t *adm;
ASSERT(update_ctx != NULL);
ASSERT((update_ctx->uc_type & (UCT_INDEX|UCT_ALL)) !=
(UCT_INDEX|UCT_ALL));
ASSERT((update_ctx->uc_type & ~UCT_FLAGS) == 0);
ASSERT(VALID_AVL_STATE);
if ((adm = fmd_adm_open(update_ctx->uc_host, update_ctx->uc_prog,
update_ctx->uc_version)) == NULL) {
(void) snmp_log(LOG_ERR, MODNAME_STR ": Communication with fmd "
"failed: %s\n", strerror(errno));
return (SNMP_ERR_RESOURCEUNAVAILABLE);
}
++valid_stamp;
if (fmd_adm_module_iter(adm, modinfo_update_one, update_ctx) != 0) {
(void) snmp_log(LOG_ERR, MODNAME_STR ": fmd module information "
"update failed: %s\n", fmd_adm_errmsg(adm));
fmd_adm_close(adm);
return (SNMP_ERR_RESOURCEUNAVAILABLE);
}
DEBUGMSGTL((MODNAME_STR, "module iteration completed\n"));
fmd_adm_close(adm);
return (SNMP_ERR_NOERROR);
}
/*ARGSUSED*/
static void
update_thread(void *arg)
{
/*
* The current modinfo_update implementation offers minimal savings
* for the use of index-only updates; therefore we always do a full
* update. If it becomes advantageous to limit updates to a single
* index, the contexts can be queued by the handler instead.
*/
sunFmModule_update_ctx_t uc;
uc.uc_host = NULL;
uc.uc_prog = FMD_ADM_PROGRAM;
uc.uc_version = FMD_ADM_VERSION;
uc.uc_index = 0;
uc.uc_type = UCT_ALL;
for (;;) {
(void) pthread_mutex_lock(&update_lock);
update_status = US_QUIET;
while (update_status == US_QUIET)
(void) pthread_cond_wait(&update_cv, &update_lock);
update_status = US_INPROGRESS;
(void) pthread_mutex_unlock(&update_lock);
(void) modinfo_update(&uc);
}
}
static void
request_update(void)
{
(void) pthread_mutex_lock(&update_lock);
if (update_status != US_QUIET) {
(void) pthread_mutex_unlock(&update_lock);
return;
}
update_status = US_NEEDED;
(void) pthread_cond_signal(&update_cv);
(void) pthread_mutex_unlock(&update_lock);
}
/*ARGSUSED*/
static int
module_compare_name(const void *l, const void *r, void *private)
{
sunFmModule_data_t *l_data = (sunFmModule_data_t *)l;
sunFmModule_data_t *r_data = (sunFmModule_data_t *)r;
ASSERT(l_data != NULL && r_data != NULL);
return (strcmp(l_data->d_ami_name, r_data->d_ami_name));
}
/*ARGSUSED*/
static int
module_compare_index(const void *l, const void *r, void *private)
{
sunFmModule_data_t *l_data = (sunFmModule_data_t *)l;
sunFmModule_data_t *r_data = (sunFmModule_data_t *)r;
ASSERT(l_data != NULL && r_data != NULL);
return (l_data->d_index < r_data->d_index ? -1 :
l_data->d_index > r_data->d_index ? 1 : 0);
}
int
sunFmModuleTable_init(void)
{
static oid sunFmModuleTable_oid[] = { SUNFMMODULETABLE_OID };
netsnmp_table_registration_info *table_info;
netsnmp_handler_registration *handler;
int err;
if ((err = pthread_mutex_init(&update_lock, NULL)) != 0) {
(void) snmp_log(LOG_ERR, MODNAME_STR ": mutex_init failure: "
"%s\n", strerror(err));
return (MIB_REGISTRATION_FAILED);
}
if ((err = pthread_cond_init(&update_cv, NULL)) != 0) {
(void) snmp_log(LOG_ERR, MODNAME_STR ": cond_init failure: "
"%s\n", strerror(err));
return (MIB_REGISTRATION_FAILED);
}
if ((err = pthread_create(NULL, NULL, (void *(*)(void *))update_thread,
NULL)) != 0) {
(void) snmp_log(LOG_ERR, MODNAME_STR ": error creating update "
"thread: %s\n", strerror(err));
return (MIB_REGISTRATION_FAILED);
}
if ((table_info =
SNMP_MALLOC_TYPEDEF(netsnmp_table_registration_info)) == NULL)
return (MIB_REGISTRATION_FAILED);
if ((handler = netsnmp_create_handler_registration("sunFmModuleTable",
sunFmModuleTable_handler, sunFmModuleTable_oid,
OID_LENGTH(sunFmModuleTable_oid), HANDLER_CAN_RONLY)) == NULL) {
SNMP_FREE(table_info);
return (MIB_REGISTRATION_FAILED);
}
/*
* The Net-SNMP template uses add_indexes here, but that
* function is unsafe because it does not check for failure.
*/
if (netsnmp_table_helper_add_index(table_info, ASN_UNSIGNED) == NULL) {
SNMP_FREE(table_info);
SNMP_FREE(handler);
return (MIB_REGISTRATION_FAILED);
}
table_info->min_column = SUNFMMODULE_COLMIN;
table_info->max_column = SUNFMMODULE_COLMAX;
if ((mod_name_avl_pool = uu_avl_pool_create("mod_name",
sizeof (sunFmModule_data_t),
offsetof(sunFmModule_data_t, d_name_avl), module_compare_name,
UU_AVL_DEBUG)) == NULL) {
snmp_free_varbind(table_info->indexes);
SNMP_FREE(table_info);
SNMP_FREE(handler);
}
if ((mod_name_avl = uu_avl_create(mod_name_avl_pool, NULL,
UU_AVL_DEBUG)) == NULL) {
(void) snmp_log(LOG_ERR, MODNAME_STR ": mod_name_avl creation "
"failed: %s\n", uu_strerror(uu_error()));
snmp_free_varbind(table_info->indexes);
SNMP_FREE(table_info);
SNMP_FREE(handler);
uu_avl_pool_destroy(mod_name_avl_pool);
return (MIB_REGISTRATION_FAILED);
}
if ((mod_index_avl_pool = uu_avl_pool_create("mod_index",
sizeof (sunFmModule_data_t),
offsetof(sunFmModule_data_t, d_index_avl),
module_compare_index, UU_AVL_DEBUG)) == NULL) {
snmp_free_varbind(table_info->indexes);
SNMP_FREE(table_info);
SNMP_FREE(handler);
uu_avl_destroy(mod_name_avl);
uu_avl_pool_destroy(mod_name_avl_pool);
}
if ((mod_index_avl = uu_avl_create(mod_index_avl_pool, NULL,
UU_AVL_DEBUG)) == NULL) {
(void) snmp_log(LOG_ERR, MODNAME_STR ": mod_index_avl creation "
"failed: %s\n", uu_strerror(uu_error()));
snmp_free_varbind(table_info->indexes);
SNMP_FREE(table_info);
SNMP_FREE(handler);
uu_avl_destroy(mod_name_avl);
uu_avl_pool_destroy(mod_name_avl_pool);
uu_avl_pool_destroy(mod_index_avl_pool);
return (MIB_REGISTRATION_FAILED);
}
if ((err = netsnmp_register_table(handler, table_info)) !=
MIB_REGISTERED_OK) {
snmp_free_varbind(table_info->indexes);
SNMP_FREE(table_info);
SNMP_FREE(handler);
uu_avl_destroy(mod_name_avl);
uu_avl_pool_destroy(mod_name_avl_pool);
uu_avl_destroy(mod_index_avl);
uu_avl_pool_destroy(mod_index_avl_pool);
return (err);
}
return (MIB_REGISTERED_OK);
}
/*
* These two functions form the core of GET/GETNEXT/GETBULK handling (the
* only kind we do). They perform two functions:
*
* - First, frob the request to set all the index variables to correspond
* to the value that's going to be returned. For GET, this is a nop;
* for GETNEXT/GETBULK it always requires some work.
* - Second, find and return the fmd module information corresponding to
* the (possibly updated) indices.
*
* These should be as fast as possible; they run in the agent thread.
*/
static sunFmModule_data_t *
sunFmModuleTable_nextmod(netsnmp_handler_registration *reginfo,
netsnmp_table_request_info *table_info)
{
sunFmModule_data_t *data;
netsnmp_variable_list *var;
ulong_t index;
/*
* If we have no index, we must make one.
*/
if (table_info->number_indexes < 1) {
oid tmpoid[MAX_OID_LEN];
index = 1;
DEBUGMSGTL((MODNAME_STR, "nextmod: no indexes given\n"));
var = SNMP_MALLOC_TYPEDEF(netsnmp_variable_list);
(void) snmp_set_var_typed_value(var, ASN_UNSIGNED,
(uchar_t *)&index, sizeof (index));
(void) memcpy(tmpoid, reginfo->rootoid,
reginfo->rootoid_len * sizeof (oid));
tmpoid[reginfo->rootoid_len] = 1; /* Entry is .1 */
tmpoid[reginfo->rootoid_len + 1] = table_info->colnum;
if (build_oid(&var->name, &var->name_length, tmpoid,
reginfo->rootoid_len + 2, var) != SNMPERR_SUCCESS) {
snmp_free_varbind(var);
return (NULL);
}
DEBUGMSGTL((MODNAME_STR, "nextmod: built fake index:\n"));
DEBUGMSGVAR((MODNAME_STR, var));
DEBUGMSG((MODNAME_STR, "\n"));
} else {
var = snmp_clone_varbind(table_info->indexes);
index = *var->val.integer;
DEBUGMSGTL((MODNAME_STR, "nextmod: received index:\n"));
DEBUGMSGVAR((MODNAME_STR, var));
DEBUGMSG((MODNAME_STR, "\n"));
index++;
}
snmp_free_varbind(table_info->indexes);
table_info->indexes = NULL;
table_info->number_indexes = 0;
if ((data = module_lookup_index_nextvalid(index)) == NULL) {
DEBUGMSGTL((MODNAME_STR, "nextmod: exact match not found for "
"index %lu; trying next column\n", index));
if (table_info->colnum >=
netsnmp_find_table_registration_info(reginfo)->max_column) {
snmp_free_varbind(var);
DEBUGMSGTL((MODNAME_STR, "nextmod: out of columns\n"));
return (NULL);
}
table_info->colnum++;
index = 1;
data = module_lookup_index_nextvalid(index);
}
if (data == NULL) {
DEBUGMSGTL((MODNAME_STR, "nextmod: exact match not found for "
"index %lu; stopping\n", index));
snmp_free_varbind(var);
return (NULL);
}
*var->val.integer = data->d_index;
table_info->indexes = var;
table_info->number_indexes = 1;
DEBUGMSGTL((MODNAME_STR, "matching data is %lu/%s@%p\n", data->d_index,
data->d_ami_name, data));
return (data);
}
/*ARGSUSED*/
static sunFmModule_data_t *
sunFmModuleTable_mod(netsnmp_handler_registration *reginfo,
netsnmp_table_request_info *table_info)
{
ASSERT(table_info->number_indexes == 1);
return (module_lookup_index_exact(table_info->index_oid[0]));
}
/*ARGSUSED*/
static void
sunFmModuleTable_return(unsigned int reg, void *arg)
{
netsnmp_delegated_cache *cache = (netsnmp_delegated_cache *)arg;
netsnmp_request_info *request;
netsnmp_agent_request_info *reqinfo;
netsnmp_handler_registration *reginfo;
netsnmp_table_request_info *table_info;
sunFmModule_data_t *data;
ulong_t modstate;
ASSERT(netsnmp_handler_check_cache(cache) != NULL);
(void) pthread_mutex_lock(&update_lock);
if (update_status != US_QUIET) {
struct timeval tv;
tv.tv_sec = UPDATE_WAIT_MILLIS / 1000;
tv.tv_usec = (UPDATE_WAIT_MILLIS % 1000) * 1000;
(void) snmp_alarm_register_hr(tv, 0, sunFmModuleTable_return,
cache);
(void) pthread_mutex_unlock(&update_lock);
return;
}
request = cache->requests;
reqinfo = cache->reqinfo;
reginfo = cache->reginfo;
table_info = netsnmp_extract_table_info(request);
request->delegated = 0;
ASSERT(table_info->colnum >= SUNFMMODULE_COLMIN);
ASSERT(table_info->colnum <= SUNFMMODULE_COLMAX);
/*
* table_info->colnum contains the column number requested.
* table_info->indexes contains a linked list of snmp variable
* bindings for the indexes of the table. Values in the list
* have been set corresponding to the indexes of the
* request. We have other guarantees as well:
*
* - The column number is always within range.
* - If we have no index data, table_info->index_oid_len is 0.
* - We will never receive requests outside our table nor
* those with the first subid anything other than 1 (Entry)
* nor those without a column number. This is true even
* for GETNEXT requests.
*/
switch (reqinfo->mode) {
case MODE_GET:
if ((data = sunFmModuleTable_mod(reginfo, table_info)) ==
NULL) {
netsnmp_free_delegated_cache(cache);
(void) pthread_mutex_unlock(&update_lock);
return;
}
break;
case MODE_GETNEXT:
case MODE_GETBULK:
if ((data = sunFmModuleTable_nextmod(reginfo, table_info)) ==
NULL) {
netsnmp_free_delegated_cache(cache);
(void) pthread_mutex_unlock(&update_lock);
return;
}
break;
default:
(void) snmp_log(LOG_ERR, MODNAME_STR ": Unsupported request "
"mode %d\n", reqinfo->mode);
netsnmp_free_delegated_cache(cache);
(void) pthread_mutex_unlock(&update_lock);
return;
}
switch (table_info->colnum) {
case SUNFMMODULE_COL_NAME:
(void) netsnmp_table_build_result(reginfo, request, table_info,
ASN_OCTET_STR, (uchar_t *)data->d_ami_name,
strlen(data->d_ami_name));
break;
case SUNFMMODULE_COL_VERSION:
(void) netsnmp_table_build_result(reginfo, request, table_info,
ASN_OCTET_STR, (uchar_t *)data->d_ami_vers,
strlen(data->d_ami_vers));
break;
case SUNFMMODULE_COL_STATUS:
modstate = (data->d_ami_flags & FMD_ADM_MOD_FAILED) ?
SUNFMMODULE_STATE_FAILED : SUNFMMODULE_STATE_ACTIVE;
(void) netsnmp_table_build_result(reginfo, request, table_info,
ASN_INTEGER, (uchar_t *)&modstate,
sizeof (modstate));
break;
case SUNFMMODULE_COL_DESCRIPTION:
(void) netsnmp_table_build_result(reginfo, request, table_info,
ASN_OCTET_STR, (uchar_t *)data->d_ami_desc,
strlen(data->d_ami_desc));
break;
default:
break;
}
netsnmp_free_delegated_cache(cache);
(void) pthread_mutex_unlock(&update_lock);
}
static int
sunFmModuleTable_handler(netsnmp_mib_handler *handler,
netsnmp_handler_registration *reginfo, netsnmp_agent_request_info *reqinfo,
netsnmp_request_info *requests)
{
netsnmp_request_info *request;
struct timeval tv;
tv.tv_sec = UPDATE_WAIT_MILLIS / 1000;
tv.tv_usec = (UPDATE_WAIT_MILLIS % 1000) * 1000;
request_update();
for (request = requests; request; request = request->next) {
if (request->processed != 0)
continue;
if (netsnmp_extract_table_info(request) == NULL)
continue;
request->delegated = 1;
(void) snmp_alarm_register_hr(tv, 0, sunFmModuleTable_return,
(void *) netsnmp_create_delegated_cache(handler, reginfo,
reqinfo, request, NULL));
}
return (SNMP_ERR_NOERROR);
}