diff options
Diffstat (limited to 'runtime')
56 files changed, 4244 insertions, 2096 deletions
diff --git a/runtime/Makefile.am b/runtime/Makefile.am index c1a15198..f7db3e35 100644 --- a/runtime/Makefile.am +++ b/runtime/Makefile.am @@ -9,6 +9,7 @@ librsyslog_la_SOURCES = \ rsyslog.h \ unicode-helper.h \ atomic.h \ + batch.h \ syslogd-types.h \ module-template.h \ obj-types.h \ @@ -20,6 +21,8 @@ librsyslog_la_SOURCES = \ conf.h \ parser.h \ parser.c \ + strgen.h \ + strgen.c \ msg.c \ msg.h \ linkedlist.c \ @@ -136,7 +139,10 @@ lmnet_la_LDFLAGS = -module -avoid-version lmnet_la_LIBADD = # network stream master class and stream factory -lmnetstrms_la_SOURCES = netstrms.c netstrms.h netstrm.c netstrm.h nssel.c nssel.h +lmnetstrms_la_SOURCES = netstrms.c netstrms.h \ + netstrm.c netstrm.h \ + nssel.c nssel.h \ + nspoll.c nspoll.h lmnetstrms_la_CPPFLAGS = $(PTHREADS_CFLAGS) $(RSRT_CFLAGS) lmnetstrms_la_LDFLAGS = -module -avoid-version lmnetstrms_la_LIBADD = @@ -152,7 +158,9 @@ lmstrmsrv_la_LIBADD = # plain tcp driver - main driver pkglib_LTLIBRARIES += lmnsd_ptcp.la -lmnsd_ptcp_la_SOURCES = nsd_ptcp.c nsd_ptcp.h nsdsel_ptcp.c nsdsel_ptcp.h +lmnsd_ptcp_la_SOURCES = nsd_ptcp.c nsd_ptcp.h \ + nsdsel_ptcp.c nsdsel_ptcp.h \ + nsdpoll_ptcp.c nsdpoll_ptcp.h lmnsd_ptcp_la_CPPFLAGS = $(PTHREADS_CFLAGS) $(RSRT_CFLAGS) lmnsd_ptcp_la_LDFLAGS = -module -avoid-version lmnsd_ptcp_la_LIBADD = diff --git a/runtime/apc.c b/runtime/apc.c index bc330e39..3c6b7ec4 100644 --- a/runtime/apc.c +++ b/runtime/apc.c @@ -40,9 +40,11 @@ #include "obj.h" #include "apc.h" #include "srUtils.h" +#include "datetime.h" /* static data */ DEFobjStaticHelpers +DEFobjCurrIf(datetime) /* following is a used to implement a monotonically increasing id for the apcs. That * ID can be used to cancel an apc request. Note that the ID is generated with modulo @@ -200,7 +202,7 @@ unlistCurrent(apc_list_t **ppList) DEFiRet; assert(ppList != NULL); - time(&tCurr); + datetime.GetTime(&tCurr); if(apcListRoot == NULL || apcListRoot->pApc->ttExec > tCurr) { *ppList = NULL; @@ -249,12 +251,11 @@ execScheduled(void) apc_list_t *pExecList; apc_list_t *pCurr; apc_list_t *pNext; - DEFVARS_mutexProtection_uncond; DEFiRet; - BEGIN_MTX_PROTECTED_OPERATIONS_UNCOND(&listMutex); + d_pthread_mutex_lock(&listMutex); iRet = unlistCurrent(&pExecList); - END_MTX_PROTECTED_OPERATIONS_UNCOND(&listMutex); + d_pthread_mutex_unlock(&listMutex); CHKiRet(iRet); if(pExecList != NULL) { @@ -290,14 +291,12 @@ ENDobjConstruct(apc) static rsRetVal apcConstructFinalize(apc_t *pThis, apc_id_t *pID) { - DEFVARS_mutexProtection_uncond; DEFiRet; ISOBJ_TYPE_assert(pThis, apc); assert(pID != NULL); - BEGIN_MTX_PROTECTED_OPERATIONS_UNCOND(&listMutex); + d_pthread_mutex_lock(&listMutex); insertApc(pThis, pID); - END_MTX_PROTECTED_OPERATIONS_UNCOND(&listMutex); -RUNLOG_STR("apcConstructFinalize post mutex unlock\n"); + d_pthread_mutex_unlock(&listMutex); RETiRet; } @@ -333,12 +332,10 @@ SetParam2(apc_t *pThis, void *param2) static rsRetVal CancelApc(apc_id_t id) { - DEFVARS_mutexProtection_uncond; - BEGINfunc - BEGIN_MTX_PROTECTED_OPERATIONS_UNCOND(&listMutex); + d_pthread_mutex_lock(&listMutex); deleteApc(id); - END_MTX_PROTECTED_OPERATIONS_UNCOND(&listMutex); + d_pthread_mutex_unlock(&listMutex); ENDfunc return RS_RET_OK; } @@ -380,7 +377,7 @@ ENDobjQueryInterface(apc) * rgerhards, 2009-04-06 */ BEGINObjClassExit(apc, OBJ_IS_CORE_MODULE) /* class, version */ - //objRelease(apcstk, CORE_COMPONENT); + objRelease(datetime, CORE_COMPONENT); pthread_mutex_destroy(&listMutex); ENDObjClassExit(apc) @@ -391,7 +388,7 @@ ENDObjClassExit(apc) */ BEGINObjClassInit(apc, 1, OBJ_IS_CORE_MODULE) /* class, version */ /* request objects we use */ - //CHKiRet(objUse(apcstk, CORE_COMPONENT)); + CHKiRet(objUse(datetime, CORE_COMPONENT)); /* set our own handlers */ OBJSetMethodHandler(objMethod_DEBUGPRINT, apcDebugPrint); diff --git a/runtime/atomic.h b/runtime/atomic.h index fc3e0b2d..da544c4b 100644 --- a/runtime/atomic.h +++ b/runtime/atomic.h @@ -39,16 +39,18 @@ * They simply came in too late. -- rgerhards, 2008-04-02 */ #ifdef HAVE_ATOMIC_BUILTINS +# define ATOMIC_SUB(data, val, phlpmut) __sync_fetch_and_sub(data, val) +# define ATOMIC_ADD(data, val) __sync_fetch_and_add(&(data), val) # define ATOMIC_INC(data, phlpmut) ((void) __sync_fetch_and_add(data, 1)) -# define ATOMIC_INC_AND_FETCH(data) __sync_fetch_and_add(&(data), 1) +# define ATOMIC_INC_AND_FETCH(data, phlpmut) __sync_fetch_and_add(data, 1) # define ATOMIC_DEC(data, phlpmut) ((void) __sync_sub_and_fetch(data, 1)) # define ATOMIC_DEC_AND_FETCH(data, phlpmut) __sync_sub_and_fetch(data, 1) -# define ATOMIC_FETCH_32BIT(data) ((unsigned) __sync_fetch_and_and(&(data), 0xffffffff)) +# define ATOMIC_FETCH_32BIT(data, phlpmut) ((unsigned) __sync_fetch_and_and(data, 0xffffffff)) # define ATOMIC_STORE_1_TO_32BIT(data) __sync_lock_test_and_set(&(data), 1) # define ATOMIC_STORE_0_TO_INT(data, phlpmut) __sync_fetch_and_and(data, 0) # define ATOMIC_STORE_1_TO_INT(data, phlpmut) __sync_fetch_and_or(data, 1) # define ATOMIC_STORE_INT_TO_INT(data, val) __sync_fetch_and_or(&(data), (val)) -# define ATOMIC_CAS(data, oldVal, newVal) __sync_bool_compare_and_swap(&(data), (oldVal), (newVal)); +# define ATOMIC_CAS(data, oldVal, newVal, phlpmut) __sync_bool_compare_and_swap(data, (oldVal), (newVal)) # define ATOMIC_CAS_VAL(data, oldVal, newVal, phlpmut) __sync_val_compare_and_swap(data, (oldVal), (newVal)); /* functions below are not needed if we have atomics */ @@ -77,18 +79,33 @@ } # define ATOMIC_STORE_0_TO_INT(data, hlpmut) { \ - pthread_mutex_lock(&hlpmut); \ + pthread_mutex_lock(hlpmut); \ *(data) = 0; \ - pthread_mutex_unlock(&hlpmut); \ + pthread_mutex_unlock(hlpmut); \ } # define ATOMIC_STORE_1_TO_INT(data, hlpmut) { \ - pthread_mutex_lock(&hlpmut); \ + pthread_mutex_lock(hlpmut); \ *(data) = 1; \ - pthread_mutex_unlock(&hlpmut); \ + pthread_mutex_unlock(hlpmut); \ } static inline int + ATOMIC_CAS(int *data, int oldVal, int newVal, pthread_mutex_t *phlpmut) { + int bSuccess; + pthread_mutex_lock(phlpmut); + if(*data == oldVal) { + *data = newVal; + bSuccess = 1; + } else { + bSuccess = 0; + } + pthread_mutex_unlock(phlpmut); + return(bSuccess); + } + + + static inline int ATOMIC_CAS_VAL(int *data, int oldVal, int newVal, pthread_mutex_t *phlpmut) { int val; pthread_mutex_lock(phlpmut); @@ -107,6 +124,15 @@ } static inline int + ATOMIC_INC_AND_FETCH(int *data, pthread_mutex_t *phlpmut) { + int val; + pthread_mutex_lock(phlpmut); + val = ++(*data); + pthread_mutex_unlock(phlpmut); + return(val); + } + + static inline int ATOMIC_DEC_AND_FETCH(int *data, pthread_mutex_t *phlpmut) { int val; pthread_mutex_lock(phlpmut); @@ -114,15 +140,30 @@ pthread_mutex_unlock(phlpmut); return(val); } + + static inline int + ATOMIC_FETCH_32BIT(int *data, pthread_mutex_t *phlpmut) { + int val; + pthread_mutex_lock(phlpmut); + val = (*data); + pthread_mutex_unlock(phlpmut); + return(val); + } + + static inline void + ATOMIC_SUB(int *data, int val, pthread_mutex_t *phlpmut) { + pthread_mutex_lock(phlpmut); + (*data) -= val; + pthread_mutex_unlock(phlpmut); + } #if 0 # warning "atomic builtins not available, using nul operations - rsyslogd will probably be racy!" # define ATOMIC_INC_AND_FETCH(data) (++(data)) -# define ATOMIC_FETCH_32BIT(data) (data) // TODO: del # define ATOMIC_STORE_1_TO_32BIT(data) (data) = 1 // TODO: del #endif # define DEF_ATOMIC_HELPER_MUT(x) pthread_mutex_t x # define INIT_ATOMIC_HELPER_MUT(x) pthread_mutex_init(&(x), NULL) -# define DESTROY_ATOMIC_HELPER_MUT(x) pthread_mutex_init(&(x), NULL) +# define DESTROY_ATOMIC_HELPER_MUT(x) pthread_mutex_destroy(&(x)) # define PREFER_ATOMIC_INC(data) ((void) ++data) diff --git a/runtime/batch.h b/runtime/batch.h new file mode 100644 index 00000000..68f48d8b --- /dev/null +++ b/runtime/batch.h @@ -0,0 +1,194 @@ +/* Definition of the batch_t data structure. + * I am not sure yet if this will become a full-blown object. For now, this header just + * includes the object definition and is not accompanied by code. + * + * Copyright 2009 by Rainer Gerhards and Adiscon GmbH. + * + * This file is part of the rsyslog runtime library. + * + * The rsyslog runtime library is free software: you can redistribute it and/or modify + * it under the terms of the GNU Lesser General Public License as published by + * the Free Software Foundation, either version 3 of the License, or + * (at your option) any later version. + * + * The rsyslog runtime library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public License + * along with the rsyslog runtime library. If not, see <http://www.gnu.org/licenses/>. + * + * A copy of the GPL can be found in the file "COPYING" in this distribution. + * A copy of the LGPL can be found in the file "COPYING.LESSER" in this distribution. + */ + +#ifndef BATCH_H_INCLUDED +#define BATCH_H_INCLUDED + +#include <string.h> +#include "msg.h" + +/* enum for batch states. Actually, we violate a layer here, in that we assume that a batch is used + * for action processing. So far, this seems acceptable, the status is simply ignored inside the + * main message queue. But over time, it could potentially be useful to split the two. + * rgerhad, 2009-05-12 + */ +typedef enum { + BATCH_STATE_RDY = 0, /* object ready for processing */ + BATCH_STATE_BAD = 1, /* unrecoverable failure while processing, do NOT resubmit to same action */ + BATCH_STATE_SUB = 2, /* message submitted for processing, outcome yet unknown */ + BATCH_STATE_COMM = 3, /* message successfully commited */ + BATCH_STATE_DISC = 4, /* discarded - processed OK, but do not submit to any other action */ +} batch_state_t; + + +/* an object inside a batch, including any information (state!) needed for it to "life". + */ +struct batch_obj_s { + obj_t *pUsrp; /* pointer to user object (most often message) */ + batch_state_t state; /* associated state */ + /* work variables for action processing; these are reused for each action (or block of + * actions) + */ + sbool bFilterOK; /* work area for filter processing (per action, reused!) */ + sbool bPrevWasSuspended; + void *staticActParams[CONF_OMOD_NUMSTRINGS_MAXSIZE]; + /* a cache to save malloc(), if not absolutely necessary */ + size_t staticLenParams[CONF_OMOD_NUMSTRINGS_MAXSIZE]; + /* and the same for the message length (if used) */ + /* end action work variables */ +}; + +/* the batch + * This object is used to dequeue multiple user pointers which are than handed over + * to processing. The size of elements is fixed after queue creation, but may be + * modified by config variables (better said: queue properties). + * Note that a "user pointer" in rsyslog context so far always is a message + * object. We stick to the more generic term because queues may potentially hold + * other types of objects, too. + * rgerhards, 2009-05-12 + * Note that nElem is not necessarily equal to nElemDeq. This is the case when we + * discard some elements (because of configuration) during dequeue processing. As + * all Elements are only deleted when the batch is processed, we can not immediately + * delete them. So we need to keep their number that we can delete them when the batch + * is completed (else, the whole process does not work correctly). + */ +struct batch_s { + int maxElem; /* maximum number of elements that this batch supports */ + int nElem; /* actual number of element in this entry */ + int nElemDeq; /* actual number of elements dequeued (and thus to be deleted) - see comment above! */ + int iDoneUpTo; /* all messages below this index have state other than RDY */ + qDeqID deqID; /* ID of dequeue operation that generated this batch */ + int *pbShutdownImmediate;/* end processing of this batch immediately if set to 1 */ + sbool bSingleRuleset; /* do all msgs of this batch use a single ruleset? */ + batch_obj_t *pElem; /* batch elements */ +}; + + +/* some inline functions (we may move this off to an object .. or not) */ +static inline void +batchSetSingleRuleset(batch_t *pBatch, sbool val) { + pBatch->bSingleRuleset = val; +} + +/* get the batches ruleset (if we have a single ruleset) */ +static inline ruleset_t* +batchGetRuleset(batch_t *pBatch) { + return (pBatch->nElem > 0) ? ((msg_t*) pBatch->pElem[0].pUsrp)->pRuleset : NULL; +} + +/* get the ruleset of a specifc element of the batch (index not verified!) */ +static inline ruleset_t* +batchElemGetRuleset(batch_t *pBatch, int i) { + return ((msg_t*) pBatch->pElem[i].pUsrp)->pRuleset; +} + +/* get number of msgs for this batch */ +static inline int +batchNumMsgs(batch_t *pBatch) { + return pBatch->nElem; +} + + +/* set the status of the i-th batch element. Note that once the status is + * DISC, it will never be reset. So this function can NOT be used to initialize + * the state table. -- rgerhards, 2010-06-10 + */ +static inline void +batchSetElemState(batch_t *pBatch, int i, batch_state_t newState) { + if(pBatch->pElem[i].state != BATCH_STATE_DISC) + pBatch->pElem[i].state = newState; +} + + +/* check if an element is a valid entry. We do NOT verify if the + * element index is valid. -- rgerhards, 2010-06-10 + */ +static inline int +batchIsValidElem(batch_t *pBatch, int i) { + return(pBatch->pElem[i].bFilterOK && pBatch->pElem[i].state != BATCH_STATE_DISC); +} + + +/* copy one batch element to another. + * This creates a complete duplicate in those cases where + * it is needed. Use duplication only when absolutely necessary! + * rgerhards, 2010-06-10 + */ +static inline void +batchCopyElem(batch_obj_t *pDest, batch_obj_t *pSrc) { + memcpy(pDest, pSrc, sizeof(batch_obj_t)); +} + + +/* free members of a batch "object". Note that we can not do the usual + * destruction as the object typically is allocated on the stack and so the + * object itself cannot be freed! -- rgerhards, 2010-06-15 + */ +static inline void +batchFree(batch_t *pBatch) { + int i; + int j; + for(i = 0 ; i < pBatch->maxElem ; ++i) { + for(j = 0 ; j < CONF_OMOD_NUMSTRINGS_MAXSIZE ; ++j) { + free(pBatch->pElem[i].staticActParams[j]); + } + } + free(pBatch->pElem); +} + + +/* initialiaze a batch "object". The record must already exist, + * we "just" initialize it. The max number of elements must be + * provided. -- rgerhards, 2010-06-15 + */ +static inline rsRetVal +batchInit(batch_t *pBatch, int maxElem) { + DEFiRet; + pBatch->maxElem = maxElem; + CHKmalloc(pBatch->pElem = calloc((size_t)maxElem, sizeof(batch_obj_t))); + // TODO: replace calloc by inidividual writes? +finalize_it: + RETiRet; +} + + +/* primarily a helper for debug purposes, get human-readble name of state */ +static inline char * +batchState2String(batch_state_t state) { + switch(state) { + case BATCH_STATE_RDY: + return "BATCH_STATE_RDY"; + case BATCH_STATE_BAD: + return "BATCH_STATE_BAD"; + case BATCH_STATE_SUB: + return "BATCH_STATE_SUB"; + case BATCH_STATE_COMM: + return "BATCH_STATE_COMM"; + case BATCH_STATE_DISC: + return "BATCH_STATE_DISC"; + } + return "ERROR, batch state not known!"; +} +#endif /* #ifndef BATCH_H_INCLUDED */ diff --git a/runtime/conf.c b/runtime/conf.c index e2d3a889..d41f2950 100644 --- a/runtime/conf.c +++ b/runtime/conf.c @@ -93,19 +93,18 @@ DEFobjCurrIf(net) DEFobjCurrIf(rule) DEFobjCurrIf(ruleset) -static int iNbrActions; /* number of actions the running config has. Needs to be init on ReInitConf() */ +static int iNbrActions = 0; /* number of currently defined actions */ -/* The following global variables are used for building +/* The following module-global variables are used for building * tag and host selector lines during startup and config reload. * This is stored as a global variable pool because of its ease. It is * also fairly compatible with multi-threading as the stratup code must - * be run in a single thread anyways. So there can be no race conditions. These - * variables are no longer used once the configuration has been loaded (except, - * of course, during a reload). rgerhards 2005-10-18 + * be run in a single thread anyways. So there can be no race conditions. + * rgerhards 2005-10-18 */ -EHostnameCmpMode eDfltHostnameCmpMode; -cstr_t *pDfltHostnameCmp; -cstr_t *pDfltProgNameCmp; +static EHostnameCmpMode eDfltHostnameCmpMode = HN_NO_COMP; +static cstr_t *pDfltHostnameCmp = NULL; +static cstr_t *pDfltProgNameCmp = NULL; /* process a directory and include all of its files into @@ -1041,7 +1040,7 @@ static rsRetVal cflineDoFilter(uchar **pp, rule_t *f) DEFiRet; ASSERT(pp != NULL); - ASSERT(f != NULL); + ISOBJ_TYPE_assert(f, rule); /* check which filter we need to pull... */ switch(**pp) { @@ -1063,6 +1062,7 @@ static rsRetVal cflineDoFilter(uchar **pp, rule_t *f) * and, if so, we copy them over. rgerhards, 2005-10-18 */ if(pDfltProgNameCmp != NULL) { +RUNLOG_STR("dflt ProgNameCmp != NULL, setting opCSProgNameComp"); CHKiRet(rsCStrConstructFromCStr(&(f->pCSProgNameComp), pDfltProgNameCmp)); } @@ -1105,7 +1105,7 @@ static rsRetVal cflineDoAction(uchar **p, action_t **ppAction) dbgprintf("module is incompatible with RepeatedMsgReduction - turned off\n"); pAction->f_ReduceRepeated = 0; } - pAction->bEnabled = 1; /* action is enabled */ + pAction->eState = ACT_STATE_RDY; /* action is enabled */ iNbrActions++; /* one more active action! */ } break; @@ -1207,21 +1207,6 @@ cfline(uchar *line, rule_t **pfCurr) } -/* Reinitialize the configuration subsystem. This is a "work-around" to the fact - * that we do not yet have actual config objects. This method is to be called - * whenever a totally new config is started (which means on startup and HUP). - * Note that it MUST NOT be called for an included config file. - * rgerhards, 2008-07-28 - */ -static rsRetVal -ReInitConf(void) -{ - DEFiRet; - iNbrActions = 0; /* this is what we created the function for ;) - action count is reset */ - RETiRet; -} - - /* return the current number of active actions * rgerhards, 2008-07-28 */ @@ -1255,7 +1240,6 @@ CODESTARTobjQueryInterface(conf) pIf->doIncludeLine = doIncludeLine; pIf->cfline = cfline; pIf->processConfFile = processConfFile; - pIf->ReInitConf = ReInitConf; pIf->GetNbrActActions = GetNbrActActions; finalize_it: @@ -1267,6 +1251,15 @@ ENDobjQueryInterface(conf) */ BEGINObjClassExit(conf, OBJ_IS_CORE_MODULE) /* CHANGE class also in END MACRO! */ CODESTARTObjClassExit(conf) + /* free no-longer needed module-global variables */ + if(pDfltHostnameCmp != NULL) { + rsCStrDestruct(&pDfltHostnameCmp); + } + + if(pDfltProgNameCmp != NULL) { + rsCStrDestruct(&pDfltProgNameCmp); + } + /* release objects we no longer need */ objRelease(expr, CORE_COMPONENT); objRelease(ctok, CORE_COMPONENT); diff --git a/runtime/conf.h b/runtime/conf.h index 25b887be..d85d1f82 100644 --- a/runtime/conf.h +++ b/runtime/conf.h @@ -37,20 +37,18 @@ BEGINinterface(conf) /* name must also be changed in ENDinterface macro! */ rsRetVal (*doIncludeLine)(uchar **pp, __attribute__((unused)) void* pVal); rsRetVal (*cfline)(uchar *line, rule_t **pfCurr); rsRetVal (*processConfFile)(uchar *pConfFile); - rsRetVal (*ReInitConf)(void); rsRetVal (*GetNbrActActions)(int *); ENDinterface(conf) -#define confCURR_IF_VERSION 2 /* increment whenever you change the interface structure! */ +#define confCURR_IF_VERSION 3 /* increment whenever you change the interface structure! */ +/* in Version 3, entry point "ReInitConf()" was removed, as we do not longer need + * to support restart-type HUP -- rgerhards, 2009-07-15 + */ /* prototypes */ PROTOTYPEObj(conf); -/* TODO: remove them below (means move the config init code) -- rgerhards, 2008-02-19 */ -extern EHostnameCmpMode eDfltHostnameCmpMode; -extern cstr_t *pDfltHostnameCmp; -extern cstr_t *pDfltProgNameCmp; /* TODO: the following 2 need to go in conf obj interface... */ rsRetVal cflineParseTemplateName(uchar** pp, omodStringRequest_t *pOMSR, int iEntry, int iTplOpts, uchar *dfltTplName); rsRetVal cflineParseFileName(uchar* p, uchar *pFileName, omodStringRequest_t *pOMSR, int iEntry, int iTplOpts, uchar *pszTpl); diff --git a/runtime/datetime.c b/runtime/datetime.c index eff72f91..de26762d 100644 --- a/runtime/datetime.c +++ b/runtime/datetime.c @@ -127,6 +127,24 @@ static void getCurrTime(struct syslogTime *t, time_t *ttSeconds) } +/* A fast alternative to getCurrTime() and time() that only obtains + * a timestamp like time() does. I was told that gettimeofday(), at + * least under Linux, is much faster than time() and I could confirm + * this testing. So I created that function as a replacement. + * rgerhards, 2009-11-12 + */ +static time_t +getTime(time_t *ttSeconds) +{ + struct timeval tp; + + if(gettimeofday(&tp, NULL) == -1) + return -1; + + if(ttSeconds != NULL) + *ttSeconds = tp.tv_sec; + return tp.tv_sec; +} /******************************************************************* @@ -842,6 +860,7 @@ CODESTARTobjQueryInterface(datetime) * of course, also affects the "if" above). */ pIf->getCurrTime = getCurrTime; + pIf->GetTime = getTime; pIf->ParseTIMESTAMP3339 = ParseTIMESTAMP3339; pIf->ParseTIMESTAMP3164 = ParseTIMESTAMP3164; pIf->formatTimestampToMySQL = formatTimestampToMySQL; diff --git a/runtime/datetime.h b/runtime/datetime.h index 9dcce3c5..82bd415b 100644 --- a/runtime/datetime.h +++ b/runtime/datetime.h @@ -41,8 +41,10 @@ BEGINinterface(datetime) /* name must also be changed in ENDinterface macro! */ int (*formatTimestamp3339)(struct syslogTime *ts, char* pBuf); int (*formatTimestamp3164)(struct syslogTime *ts, char* pBuf, int); int (*formatTimestampSecFrac)(struct syslogTime *ts, char* pBuf); + /* v3, 2009-11-12 */ + time_t (*GetTime)(time_t *ttSeconds); ENDinterface(datetime) -#define datetimeCURR_IF_VERSION 4 /* increment whenever you change the interface structure! */ +#define datetimeCURR_IF_VERSION 5 /* increment whenever you change the interface structure! */ /* interface changes: * 1 - initial version * 2 - not compatible to 1 - bugfix required ParseTIMESTAMP3164 to accept char ** as @@ -50,6 +52,7 @@ ENDinterface(datetime) * third-party module should call. -- rgerhards, 2008.-09-12 * 3 - taken by v5 branch! * 4 - formatTimestamp3164 takes a third int parameter + * 5 - merge of versions 3 + 4 (2010-03-09) */ /* prototypes */ diff --git a/runtime/debug.c b/runtime/debug.c index 0ada909b..64e251e5 100644 --- a/runtime/debug.c +++ b/runtime/debug.c @@ -46,6 +46,9 @@ #include <unistd.h> #include <fcntl.h> #include <sys/stat.h> +#if _POSIX_TIMERS <= 0 +#include <sys/time.h> +#endif #include "rsyslog.h" #include "debug.h" @@ -833,13 +836,12 @@ sigsegvHdlr(int signum) abort(); } -#if 1 -#pragma GCC diagnostic ignored "-Wempty-body" -/* write the debug message. This is a helper to dbgprintf and dbgoprint which - * contains common code. added 2008-09-26 rgerhards +/* actually write the debug message. This is a separate fuction because the cleanup_push/_pop + * interface otherwise is unsafe to use (generates compiler warnings at least). + * 2009-05-20 rgerhards */ -static void -dbgprint(obj_t *pObj, char *pszMsg, size_t lenMsg) +static inline void +do_dbgprint(uchar *pszObjName, char *pszMsg, size_t lenMsg) { static pthread_t ptLastThrdID = 0; static int bWasNL = 0; @@ -847,22 +849,9 @@ dbgprint(obj_t *pObj, char *pszMsg, size_t lenMsg) char pszWriteBuf[32*1024]; size_t lenWriteBuf; struct timespec t; - uchar *pszObjName = NULL; - int ret; - - /* we must get the object name before we lock the mutex, because the object - * potentially calls back into us. If we locked the mutex, we would deadlock - * ourselfs. On the other hand, the GetName call needs not to be protected, as - * this thread has a valid reference. If such an object is deleted by another - * thread, we are in much more trouble than just for dbgprint(). -- rgerhards, 2008-09-26 - */ - if(pObj != NULL) { - pszObjName = obj.GetName(pObj); - } - - ret = pthread_mutex_lock(&mutdbgprint); - assert(ret == 0); /* make sure mutex operation does not fail */ - pthread_cleanup_push(dbgMutexCancelCleanupHdlr, &mutdbgprint); +# if _POSIX_TIMERS <= 0 + struct timeval tv; +# endif /* The bWasNL handler does not really work. It works if no thread * switching occurs during non-NL messages. Else, things are messed @@ -888,7 +877,14 @@ dbgprint(obj_t *pObj, char *pszMsg, size_t lenMsg) if(bWasNL) { if(bPrintTime) { +# if _POSIX_TIMERS > 0 + /* this is the "regular" code */ clock_gettime(CLOCK_REALTIME, &t); +# else + gettimeofday(&tv, NULL); + t.tv_sec = tv.tv_sec; + t.tv_nsec = tv.tv_usec * 1000; +# endif lenWriteBuf = snprintf(pszWriteBuf, sizeof(pszWriteBuf), "%4.4ld.%9.9ld:", (long) (t.tv_sec % 10000), t.tv_nsec); if(stddbg != -1) write(stddbg, pszWriteBuf, lenWriteBuf); @@ -910,11 +906,35 @@ dbgprint(obj_t *pObj, char *pszMsg, size_t lenMsg) if(altdbg != -1) write(altdbg, pszMsg, lenMsg); bWasNL = (pszMsg[lenMsg - 1] == '\n') ? 1 : 0; +} + +#pragma GCC diagnostic ignored "-Wempty-body" +/* write the debug message. This is a helper to dbgprintf and dbgoprint which + * contains common code. added 2008-09-26 rgerhards + */ +static void +dbgprint(obj_t *pObj, char *pszMsg, size_t lenMsg) +{ + uchar *pszObjName = NULL; + + /* we must get the object name before we lock the mutex, because the object + * potentially calls back into us. If we locked the mutex, we would deadlock + * ourselfs. On the other hand, the GetName call needs not to be protected, as + * this thread has a valid reference. If such an object is deleted by another + * thread, we are in much more trouble than just for dbgprint(). -- rgerhards, 2008-09-26 + */ + if(pObj != NULL) { + pszObjName = obj.GetName(pObj); + } + + pthread_mutex_lock(&mutdbgprint); + pthread_cleanup_push(dbgMutexCancelCleanupHdlr, &mutdbgprint); + + do_dbgprint(pszObjName, pszMsg, lenMsg); pthread_cleanup_pop(1); } #pragma GCC diagnostic warning "-Wempty-body" -#endif /* print some debug output when an object is given * This is mostly a copy of dbgprintf, but I do not know how to combine it @@ -1075,7 +1095,9 @@ int dbgEntrFunc(dbgFuncDB_t **ppFuncDB, const char *file, const char *func, int /* when we reach this point, we have a fully-initialized FuncDB! */ PREFER_ATOMIC_INC(pFuncDB->nTimesCalled); if(bLogFuncFlow && dbgPrintNameIsInList((const uchar*)pFuncDB->file, printNameFileRoot)) - dbgprintf("%s:%d: %s: enter\n", pFuncDB->file, pFuncDB->line, pFuncDB->func); + if(strcmp(pFuncDB->file, "stringbuf.c")) { /* TODO: make configurable */ + dbgprintf("%s:%d: %s: enter\n", pFuncDB->file, pFuncDB->line, pFuncDB->func); + } if(pThrd->stackPtr >= (int) (sizeof(pThrd->callStack) / sizeof(dbgFuncDB_t*))) { dbgprintf("%s:%d: %s: debug module: call stack for this thread full, suspending call tracking\n", pFuncDB->file, pFuncDB->line, pFuncDB->func); @@ -1105,10 +1127,12 @@ void dbgExitFunc(dbgFuncDB_t *pFuncDB, int iStackPtrRestore, int iRet) dbgFuncDBPrintActiveMutexes(pFuncDB, "WARNING: mutex still owned by us as we exit function, mutex: ", pthread_self()); if(bLogFuncFlow && dbgPrintNameIsInList((const uchar*)pFuncDB->file, printNameFileRoot)) { - if(iRet == RS_RET_NO_IRET) - dbgprintf("%s:%d: %s: exit: (no iRet)\n", pFuncDB->file, pFuncDB->line, pFuncDB->func); - else - dbgprintf("%s:%d: %s: exit: %d\n", pFuncDB->file, pFuncDB->line, pFuncDB->func, iRet); + if(strcmp(pFuncDB->file, "stringbuf.c")) { /* TODO: make configurable */ + if(iRet == RS_RET_NO_IRET) + dbgprintf("%s:%d: %s: exit: (no iRet)\n", pFuncDB->file, pFuncDB->line, pFuncDB->func); + else + dbgprintf("%s:%d: %s: exit: %d\n", pFuncDB->file, pFuncDB->line, pFuncDB->func, iRet); + } } pThrd->stackPtr = iStackPtrRestore; if(pThrd->stackPtr < 0) { @@ -1254,6 +1278,20 @@ dbgPrintNameIsInList(const uchar *pName, dbgPrintName_t *pRoot) } +/* this is a special version of malloc that fills the alloced memory with + * HIGHVALUE, as this helps to identify bugs. -- rgerhards, 2009-10-22 + */ +void * +dbgmalloc(size_t size) +{ + void *pRet; + pRet = malloc(size); + if(pRet != NULL) + memset(pRet, 0xff, size); + return pRet; +} + + /* read in the runtime options * rgerhards, 2008-02-28 */ diff --git a/runtime/debug.h b/runtime/debug.h index cfdf819c..c011dd2d 100644 --- a/runtime/debug.h +++ b/runtime/debug.h @@ -3,7 +3,7 @@ * Definitions for the debug and run-time analysis support module. * Contains a lot of macros. * - * Copyright 2008 Rainer Gerhards and Adiscon GmbH. + * Copyright 2008, 2009 Rainer Gerhards and Adiscon GmbH. * * This file is part of the rsyslog runtime library. * @@ -104,6 +104,7 @@ void dbgExitFunc(dbgFuncDB_t *pFuncDB, int iStackPtrRestore, int iRet); void dbgSetExecLocation(int iStackPtr, int line); void dbgSetThrdName(uchar *pszName); void dbgPrintAllDebugInfo(void); +void *dbgmalloc(size_t size); /* macros */ #define DBGPRINTF(...) if(Debug) { dbgprintf(__VA_ARGS__); } @@ -131,6 +132,12 @@ void dbgPrintAllDebugInfo(void); # define RUNLOG_STR(str) #endif +#ifdef MEMCHECK +# define MALLOC(x) dbgmalloc(x) +#else +# define MALLOC(x) malloc(x) +#endif + /* mutex operations */ #define MUTOP_LOCKWAIT 1 #define MUTOP_LOCK 2 diff --git a/runtime/glbl.c b/runtime/glbl.c index 58558ed2..278bc4e1 100644 --- a/runtime/glbl.c +++ b/runtime/glbl.c @@ -39,6 +39,7 @@ #include "cfsysline.h" #include "glbl.h" #include "prop.h" +#include "atomic.h" /* some defaults */ #ifndef DFLT_NETSTRM_DRVR @@ -55,7 +56,7 @@ DEFobjCurrIf(prop) */ static uchar *pszWorkDir = NULL; static int bOptimizeUniProc = 1; /* enable uniprocessor optimizations */ -static int bHUPisRestart = 0; /* should SIGHUP cause a full system restart? */ +static int bParseHOSTNAMEandTAG = 1; /* parser modification (based on startup params!) */ static int bPreserveFQDN = 0; /* should FQDNs always be preserved? */ static int iMaxLine = 2048; /* maximum length of a syslog message */ static int iDefPFFamily = PF_UNSPEC; /* protocol family (IPv4, IPv6 or both) */ @@ -72,6 +73,10 @@ static uchar *pszDfltNetstrmDrvr = NULL; /* module name of default netstream dri static uchar *pszDfltNetstrmDrvrCAF = NULL; /* default CA file for the netstrm driver */ static uchar *pszDfltNetstrmDrvrKeyFile = NULL; /* default key file for the netstrm driver (server) */ static uchar *pszDfltNetstrmDrvrCertFile = NULL; /* default cert file for the netstrm driver (server) */ +static int bTerminateInputs = 0; /* global switch that inputs shall terminate ASAP (1=> terminate) */ +#ifndef HAVE_ATOMIC_BUILTINS +static DEF_ATOMIC_HELPER_MUT(mutTerminateInputs); +#endif #ifdef USE_UNLIMITED_SELECT static int iFdSetSize = howmany(FD_SETSIZE, __NFDBITS) * sizeof (fd_mask); /* size of select() bitmask in bytes */ #endif @@ -96,9 +101,9 @@ static dataType Get##nameFunc(void) \ return(nameVar); \ } +SIMP_PROP(ParseHOSTNAMEandTAG, bParseHOSTNAMEandTAG, int) SIMP_PROP(OptimizeUniProc, bOptimizeUniProc, int) SIMP_PROP(PreserveFQDN, bPreserveFQDN, int) -SIMP_PROP(HUPisRestart, bHUPisRestart, int) SIMP_PROP(MaxLine, iMaxLine, int) SIMP_PROP(DefPFFamily, iDefPFFamily, int) /* note that in the future we may check the family argument */ SIMP_PROP(DropMalPTRMsgs, bDropMalPTRMsgs, int) @@ -123,6 +128,24 @@ SIMP_PROP_SET(DfltNetstrmDrvrCertFile, pszDfltNetstrmDrvrCertFile, uchar*) /* TO #undef SIMP_PROP_GET +/* return global input termination status + * rgerhards, 2009-07-20 + */ +static int GetGlobalInputTermState(void) +{ + return ATOMIC_FETCH_32BIT(&bTerminateInputs, &mutTerminateInputs); +} + + +/* set global termiantion state to "terminate". Note that this is a + * "once in a lifetime" action which can not be undone. -- gerhards, 2009-07-20 + */ +static void SetGlobalInputTermination(void) +{ + ATOMIC_STORE_1_TO_INT(&bTerminateInputs, &mutTerminateInputs); +} + + /* return our local hostname. if it is not set, "[localhost]" is returned */ static uchar* @@ -171,6 +194,7 @@ finalize_it: RETiRet; } + /* return our local hostname as a string property */ static prop_t* @@ -247,13 +271,15 @@ CODESTARTobjQueryInterface(glbl) pIf->GetWorkDir = GetWorkDir; pIf->GenerateLocalHostNameProperty = GenerateLocalHostNameProperty; pIf->GetLocalHostNameProp = GetLocalHostNameProp; + pIf->SetGlobalInputTermination = SetGlobalInputTermination; + pIf->GetGlobalInputTermState = GetGlobalInputTermState; #define SIMP_PROP(name) \ pIf->Get##name = Get##name; \ pIf->Set##name = Set##name; SIMP_PROP(MaxLine); SIMP_PROP(OptimizeUniProc); + SIMP_PROP(ParseHOSTNAMEandTAG); SIMP_PROP(PreserveFQDN); - SIMP_PROP(HUPisRestart); SIMP_PROP(DefPFFamily); SIMP_PROP(DropMalPTRMsgs); SIMP_PROP(Option_DisallowWarning); @@ -302,7 +328,6 @@ static rsRetVal resetConfigVariables(uchar __attribute__((unused)) *pp, void __a } bDropMalPTRMsgs = 0; bOptimizeUniProc = 1; - bHUPisRestart = 0; bPreserveFQDN = 0; #ifdef USE_UNLIMITED_SELECT iFdSetSize = howmany(FD_SETSIZE, __NFDBITS) * sizeof (fd_mask); @@ -328,9 +353,10 @@ BEGINAbstractObjClassInit(glbl, 1, OBJ_IS_CORE_MODULE) /* class, version */ CHKiRet(regCfSysLineHdlr((uchar *)"defaultnetstreamdriverkeyfile", 0, eCmdHdlrGetWord, NULL, &pszDfltNetstrmDrvrKeyFile, NULL)); CHKiRet(regCfSysLineHdlr((uchar *)"defaultnetstreamdrivercertfile", 0, eCmdHdlrGetWord, NULL, &pszDfltNetstrmDrvrCertFile, NULL)); CHKiRet(regCfSysLineHdlr((uchar *)"optimizeforuniprocessor", 0, eCmdHdlrBinary, NULL, &bOptimizeUniProc, NULL)); - CHKiRet(regCfSysLineHdlr((uchar *)"hupisrestart", 0, eCmdHdlrBinary, NULL, &bHUPisRestart, NULL)); CHKiRet(regCfSysLineHdlr((uchar *)"preservefqdn", 0, eCmdHdlrBinary, NULL, &bPreserveFQDN, NULL)); CHKiRet(regCfSysLineHdlr((uchar *)"resetconfigvariables", 1, eCmdHdlrCustomHandler, resetConfigVariables, NULL, NULL)); + + INIT_ATOMIC_HELPER_MUT(mutTerminateInputs); ENDObjClassInit(glbl) @@ -353,6 +379,7 @@ BEGINObjClassExit(glbl, OBJ_IS_CORE_MODULE) /* class, version */ if(LocalFQDNName != NULL) free(LocalFQDNName); objRelease(prop, CORE_COMPONENT); + DESTROY_ATOMIC_HELPER_MUT(mutTerminateInputs); ENDObjClassExit(glbl) /* vi:set ai: diff --git a/runtime/glbl.h b/runtime/glbl.h index 6a332576..4b4bdf83 100644 --- a/runtime/glbl.h +++ b/runtime/glbl.h @@ -8,7 +8,7 @@ * Please note that there currently is no glbl.c file as we do not yet * have any implementations. * - * Copyright 2008 Rainer Gerhards and Adiscon GmbH. + * Copyright 2008, 2009 Rainer Gerhards and Adiscon GmbH. * * This file is part of the rsyslog runtime library. * @@ -44,7 +44,6 @@ BEGINinterface(glbl) /* name must also be changed in ENDinterface macro! */ rsRetVal (*Set##name)(dataType); SIMP_PROP(MaxLine, int) SIMP_PROP(OptimizeUniProc, int) - SIMP_PROP(HUPisRestart, int) SIMP_PROP(PreserveFQDN, int) SIMP_PROP(DefPFFamily, int) SIMP_PROP(DropMalPTRMsgs, int) @@ -62,6 +61,11 @@ BEGINinterface(glbl) /* name must also be changed in ENDinterface macro! */ /* added v3, 2009-06-30 */ rsRetVal (*GenerateLocalHostNameProperty)(void); prop_t* (*GetLocalHostNameProp)(void); + /* added v4, 2009-07-20 */ + int (*GetGlobalInputTermState)(void); + void (*SetGlobalInputTermination)(void); + /* added v5, 2009-11-03 */ + SIMP_PROP(ParseHOSTNAMEandTAG, int) /* note: v4, v5 are already used by more recent versions, so we need to skip them! */ /* added v6, 2009-11-16 as part of varmojfekoj's "unlimited select()" patch * Note that it must be always present, otherwise the interface would have different @@ -71,9 +75,11 @@ BEGINinterface(glbl) /* name must also be changed in ENDinterface macro! */ * at some later stage). */ SIMP_PROP(FdSetSize, int) + /* v7: was neeeded to mean v5+v6 - do NOT add anything else for that version! */ + /* next change is v8! */ #undef SIMP_PROP ENDinterface(glbl) -#define glblCURR_IF_VERSION 6 /* increment whenever you change the interface structure! */ +#define glblCURR_IF_VERSION 7 /* increment whenever you change the interface structure! */ /* version 2 had PreserveFQDN added - rgerhards, 2008-12-08 */ /* the remaining prototypes */ diff --git a/runtime/module-template.h b/runtime/module-template.h index 3e963199..d05ec23c 100644 --- a/runtime/module-template.h +++ b/runtime/module-template.h @@ -46,6 +46,12 @@ DEFobjCurrIf(obj) #define DEF_LMOD_STATIC_DATA \ DEF_MOD_STATIC_DATA +#define DEF_PMOD_STATIC_DATA \ + DEFobjCurrIf(obj) \ + DEF_MOD_STATIC_DATA +#define DEF_SMOD_STATIC_DATA \ + DEFobjCurrIf(obj) \ + DEF_MOD_STATIC_DATA /* Macro to define the module type. Each module can only have a single type. If @@ -65,6 +71,8 @@ static rsRetVal modGetType(eModType_t *modType) \ #define MODULE_TYPE_INPUT MODULE_TYPE(eMOD_IN) #define MODULE_TYPE_OUTPUT MODULE_TYPE(eMOD_OUT) +#define MODULE_TYPE_PARSER MODULE_TYPE(eMOD_PARSER) +#define MODULE_TYPE_STRGEN MODULE_TYPE(eMOD_STRGEN) #define MODULE_TYPE_LIB \ DEF_LMOD_STATIC_DATA \ MODULE_TYPE(eMOD_LIB) @@ -368,6 +376,17 @@ static rsRetVal queryEtryPt(uchar *name, rsRetVal (**pEtryPoint)())\ *pEtryPoint = endTransaction;\ } + +/* the following definition is a queryEtryPt block that must be added + * if a non-output module supports "isCompatibleWithFeature". + * rgerhards, 2009-07-20 + */ +#define CODEqueryEtryPt_IsCompatibleWithFeature_IF_OMOD_QUERIES \ + else if(!strcmp((char*) name, "isCompatibleWithFeature")) {\ + *pEtryPoint = isCompatibleWithFeature;\ + } + + /* the following definition is the standard block for queryEtryPt for INPUT * modules. This can be used if no specific handling (e.g. to cover version * differences) is needed. @@ -389,6 +408,30 @@ static rsRetVal queryEtryPt(uchar *name, rsRetVal (**pEtryPoint)())\ #define CODEqueryEtryPt_STD_LIB_QUERIES \ CODEqueryEtryPt_STD_MOD_QUERIES +/* the following definition is the standard block for queryEtryPt for PARSER + * modules. This can be used if no specific handling (e.g. to cover version + * differences) is needed. + */ +#define CODEqueryEtryPt_STD_PMOD_QUERIES \ + CODEqueryEtryPt_STD_MOD_QUERIES \ + else if(!strcmp((char*) name, "parse")) {\ + *pEtryPoint = parse;\ + } else if(!strcmp((char*) name, "GetParserName")) {\ + *pEtryPoint = GetParserName;\ + } + +/* the following definition is the standard block for queryEtryPt for Strgen + * modules. This can be used if no specific handling (e.g. to cover version + * differences) is needed. + */ +#define CODEqueryEtryPt_STD_SMOD_QUERIES \ + CODEqueryEtryPt_STD_MOD_QUERIES \ + else if(!strcmp((char*) name, "strgen")) {\ + *pEtryPoint = strgen;\ + } else if(!strcmp((char*) name, "GetName")) {\ + *pEtryPoint = GetStrgenName;\ + } + /* modInit() * This has an extra parameter, which is the specific name of the modInit * function. That is needed for built-in modules, which must have unique @@ -579,5 +622,58 @@ static rsRetVal doHUP(instanceData __attribute__((unused)) *pData)\ } +/* parse() - main entry point of parser modules + */ +#define BEGINparse \ +static rsRetVal parse(msg_t *pMsg)\ +{\ + DEFiRet; + +#define CODESTARTparse \ + assert(pMsg != NULL); + +#define ENDparse \ + RETiRet;\ +} + + +/* strgen() - main entry point of parser modules + */ +#define BEGINstrgen \ +static rsRetVal strgen(msg_t *pMsg, uchar **ppBuf, size_t *pLenBuf) \ +{\ + DEFiRet; + +#define CODESTARTstrgen \ + assert(pMsg != NULL); + +#define ENDstrgen \ + RETiRet;\ +} + + +/* function to specify the parser name. This is done via a single command which + * receives a ANSI string as parameter. + */ +#define PARSER_NAME(x) \ +static rsRetVal GetParserName(uchar **ppSz)\ +{\ + *ppSz = UCHAR_CONSTANT(x);\ + return RS_RET_OK;\ +} + + + +/* function to specify the strgen name. This is done via a single command which + * receives a ANSI string as parameter. + */ +#define STRGEN_NAME(x) \ +static rsRetVal GetStrgenName(uchar **ppSz)\ +{\ + *ppSz = UCHAR_CONSTANT(x);\ + return RS_RET_OK;\ +} + + /* vim:set ai: */ diff --git a/runtime/modules.c b/runtime/modules.c index 871f356a..d7362753 100644 --- a/runtime/modules.c +++ b/runtime/modules.c @@ -11,7 +11,7 @@ * * File begun on 2007-07-22 by RGerhards * - * Copyright 2007 Rainer Gerhards and Adiscon GmbH. + * Copyright 2007, 2009 Rainer Gerhards and Adiscon GmbH. * * This file is part of the rsyslog runtime library. * @@ -57,10 +57,14 @@ #include "cfsysline.h" #include "modules.h" #include "errmsg.h" +#include "parser.h" +#include "strgen.h" /* static data */ DEFobjStaticHelpers DEFobjCurrIf(errmsg) +DEFobjCurrIf(parser) +DEFobjCurrIf(strgen) /* we must ensure that only one thread at one time tries to load or unload * modules, otherwise we may see race conditions. This first came up with @@ -77,6 +81,26 @@ static modInfo_t *pLoadedModulesLast = NULL; /* tail-pointer */ uchar *pModDir = NULL; /* read-only after startup */ +/* we provide a set of dummy functions for modules that do not support the + * some interfaces. + * On the commit feature: As the modules do not support it, they commit each message they + * receive, and as such the dummies can always return RS_RET_OK without causing + * harm. This simplifies things as in action processing we do not need to check + * if the transactional entry points exist. + */ +static rsRetVal dummyBeginTransaction() +{ + return RS_RET_OK; +} +static rsRetVal dummyEndTransaction() +{ + return RS_RET_OK; +} +static rsRetVal dummyIsCompatibleWithFeature() +{ + return RS_RET_INCOMPATIBLE; +} + #ifdef DEBUG /* we add some home-grown support to track our users (and detect who does not free us). In * the long term, this should probably be migrated into debug.c (TODO). -- rgerhards, 2008-03-11 @@ -216,19 +240,38 @@ static void moduleDestruct(modInfo_t *pThis) } +/* This enables a module to query the core for specific features. + * rgerhards, 2009-04-22 + */ +static rsRetVal queryCoreFeatureSupport(int *pBool, unsigned uFeat) +{ + DEFiRet; + + if((pBool == NULL)) + ABORT_FINALIZE(RS_RET_PARAM_ERROR); + + *pBool = (uFeat & CORE_FEATURE_BATCHING) ? 1 : 0; + +finalize_it: + RETiRet; +} + + /* The following function is the queryEntryPoint for host-based entry points. * Modules may call it to get access to core interface functions. Please note * that utility functions can be accessed via shared libraries - at least this * is my current shool of thinking. * Please note that the implementation as a query interface allows to take * care of plug-in interface version differences. -- rgerhards, 2007-07-31 + * ... but often it better not to use a new interface. So we now add core + * functions here that a plugin may request. -- rgerhards, 2009-04-22 */ static rsRetVal queryHostEtryPt(uchar *name, rsRetVal (**pEtryPoint)()) { DEFiRet; if((name == NULL) || (pEtryPoint == NULL)) - return RS_RET_PARAM_ERROR; + ABORT_FINALIZE(RS_RET_PARAM_ERROR); if(!strcmp((char*) name, "regCfSysLineHdlr")) { *pEtryPoint = regCfSysLineHdlr; @@ -236,6 +279,8 @@ static rsRetVal queryHostEtryPt(uchar *name, rsRetVal (**pEtryPoint)()) *pEtryPoint = objGetObjInterface; } else if(!strcmp((char*) name, "OMSRgetSupportedTplOpts")) { *pEtryPoint = OMSRgetSupportedTplOpts; + } else if(!strcmp((char*) name, "queryCoreFeatureSupport")) { + *pEtryPoint = queryCoreFeatureSupport; } else { *pEtryPoint = NULL; /* to be on the safe side */ ABORT_FINALIZE(RS_RET_ENTRY_POINT_NOT_FOUND); @@ -361,10 +406,14 @@ finalize_it: static rsRetVal doModInit(rsRetVal (*modInit)(int, int*, rsRetVal(**)(), rsRetVal(*)(), modInfo_t*), uchar *name, void *pModHdlr) { - DEFiRet; rsRetVal localRet; modInfo_t *pNew = NULL; + uchar *pName; + parser_t *pParser; /* used for parser modules */ + strgen_t *pStrgen; /* used for strgen modules */ + rsRetVal (*GetName)(uchar**); rsRetVal (*modGetType)(eModType_t *pType); + DEFiRet; assert(modInit != NULL); @@ -392,6 +441,11 @@ doModInit(rsRetVal (*modInit)(int, int*, rsRetVal(**)(), rsRetVal(*)(), modInfo_ */ CHKiRet((*pNew->modQueryEtryPt)((uchar*)"modGetID", &pNew->modGetID)); CHKiRet((*pNew->modQueryEtryPt)((uchar*)"modExit", &pNew->modExit)); + localRet = (*pNew->modQueryEtryPt)((uchar*)"isCompatibleWithFeature", &pNew->isCompatibleWithFeature); + if(localRet == RS_RET_MODULE_ENTRY_POINT_NOT_FOUND) + pNew->isCompatibleWithFeature = dummyIsCompatibleWithFeature; + else if(localRet != RS_RET_OK) + ABORT_FINALIZE(localRet); /* ... and now the module-specific interfaces */ switch(pNew->eType) { @@ -406,15 +460,70 @@ doModInit(rsRetVal (*modInit)(int, int*, rsRetVal(**)(), rsRetVal(*)(), modInfo_ CHKiRet((*pNew->modQueryEtryPt)((uchar*)"dbgPrintInstInfo", &pNew->dbgPrintInstInfo)); CHKiRet((*pNew->modQueryEtryPt)((uchar*)"doAction", &pNew->mod.om.doAction)); CHKiRet((*pNew->modQueryEtryPt)((uchar*)"parseSelectorAct", &pNew->mod.om.parseSelectorAct)); - CHKiRet((*pNew->modQueryEtryPt)((uchar*)"isCompatibleWithFeature", &pNew->isCompatibleWithFeature)); CHKiRet((*pNew->modQueryEtryPt)((uchar*)"tryResume", &pNew->tryResume)); /* try load optional interfaces */ localRet = (*pNew->modQueryEtryPt)((uchar*)"doHUP", &pNew->doHUP); if(localRet != RS_RET_OK && localRet != RS_RET_MODULE_ENTRY_POINT_NOT_FOUND) ABORT_FINALIZE(localRet); + + localRet = (*pNew->modQueryEtryPt)((uchar*)"beginTransaction", &pNew->mod.om.beginTransaction); + if(localRet == RS_RET_MODULE_ENTRY_POINT_NOT_FOUND) + pNew->mod.om.beginTransaction = dummyBeginTransaction; + else if(localRet != RS_RET_OK) + ABORT_FINALIZE(localRet); + + localRet = (*pNew->modQueryEtryPt)((uchar*)"endTransaction", &pNew->mod.om.endTransaction); + if(localRet == RS_RET_MODULE_ENTRY_POINT_NOT_FOUND) { + pNew->mod.om.endTransaction = dummyEndTransaction; + } else if(localRet != RS_RET_OK) { + ABORT_FINALIZE(localRet); + } break; case eMOD_LIB: break; + case eMOD_PARSER: + /* first, we need to obtain the parser object. We could not do that during + * init as that would have caused class bootstrap issues which are not + * absolutely necessary. Note that we can call objUse() multiple times, it + * handles that. + */ + CHKiRet(objUse(parser, CORE_COMPONENT)); + /* here, we create a new parser object */ + CHKiRet((*pNew->modQueryEtryPt)((uchar*)"parse", &pNew->mod.pm.parse)); + CHKiRet((*pNew->modQueryEtryPt)((uchar*)"GetParserName", &GetName)); + CHKiRet(GetName(&pName)); + CHKiRet(parser.Construct(&pParser)); + + /* check some features */ + localRet = pNew->isCompatibleWithFeature(sFEATUREAutomaticSanitazion); + if(localRet == RS_RET_OK){ + CHKiRet(parser.SetDoSanitazion(pParser, TRUE)); + } + localRet = pNew->isCompatibleWithFeature(sFEATUREAutomaticPRIParsing); + if(localRet == RS_RET_OK){ + CHKiRet(parser.SetDoPRIParsing(pParser, TRUE)); + } + + CHKiRet(parser.SetName(pParser, pName)); + CHKiRet(parser.SetModPtr(pParser, pNew)); + CHKiRet(parser.ConstructFinalize(pParser)); + break; + case eMOD_STRGEN: + /* first, we need to obtain the strgen object. We could not do that during + * init as that would have caused class bootstrap issues which are not + * absolutely necessary. Note that we can call objUse() multiple times, it + * handles that. + */ + CHKiRet(objUse(strgen, CORE_COMPONENT)); + /* here, we create a new parser object */ + CHKiRet((*pNew->modQueryEtryPt)((uchar*)"strgen", &pNew->mod.sm.strgen)); + CHKiRet((*pNew->modQueryEtryPt)((uchar*)"GetName", &GetName)); + CHKiRet(GetName(&pName)); + CHKiRet(strgen.Construct(&pStrgen)); + CHKiRet(strgen.SetName(pStrgen, pName)); + CHKiRet(strgen.SetModPtr(pStrgen, pNew)); + CHKiRet(strgen.ConstructFinalize(pStrgen)); + break; } pNew->pszName = (uchar*) strdup((char*)name); /* we do not care if strdup() fails, we can accept that */ @@ -461,14 +570,49 @@ static void modPrintList(void) case eMOD_LIB: dbgprintf("library"); break; + case eMOD_PARSER: + dbgprintf("parser"); + break; + case eMOD_STRGEN: + dbgprintf("strgen"); + break; } dbgprintf(" module.\n"); dbgprintf("Entry points:\n"); dbgprintf("\tqueryEtryPt: 0x%lx\n", (unsigned long) pMod->modQueryEtryPt); - dbgprintf("\tdoAction: 0x%lx\n", (unsigned long) pMod->mod.om.doAction); - dbgprintf("\tparseSelectorAct: 0x%lx\n", (unsigned long) pMod->mod.om.parseSelectorAct); dbgprintf("\tdbgPrintInstInfo: 0x%lx\n", (unsigned long) pMod->dbgPrintInstInfo); dbgprintf("\tfreeInstance: 0x%lx\n", (unsigned long) pMod->freeInstance); + switch(pMod->eType) { + case eMOD_OUT: + dbgprintf("Output Module Entry Points:\n"); + dbgprintf("\tdoAction: 0x%lx\n", (unsigned long) pMod->mod.om.doAction); + dbgprintf("\tparseSelectorAct: 0x%lx\n", (unsigned long) pMod->mod.om.parseSelectorAct); + dbgprintf("\ttryResume: 0x%lx\n", (unsigned long) pMod->tryResume); + dbgprintf("\tdoHUP: 0x%lx\n", (unsigned long) pMod->doHUP); + dbgprintf("\tBeginTransaction: 0x%lx\n", (unsigned long) + ((pMod->mod.om.beginTransaction == dummyBeginTransaction) ? + 0 : pMod->mod.om.beginTransaction)); + dbgprintf("\tEndTransaction: 0x%lx\n", (unsigned long) + ((pMod->mod.om.endTransaction == dummyEndTransaction) ? + 0 : pMod->mod.om.endTransaction)); + break; + case eMOD_IN: + dbgprintf("Input Module Entry Points\n"); + dbgprintf("\trunInput: 0x%lx\n", (unsigned long) pMod->mod.im.runInput); + dbgprintf("\twillRun: 0x%lx\n", (unsigned long) pMod->mod.im.willRun); + dbgprintf("\tafterRun: 0x%lx\n", (unsigned long) pMod->mod.im.afterRun); + break; + case eMOD_LIB: + break; + case eMOD_PARSER: + dbgprintf("Parser Module Entry Points\n"); + dbgprintf("\tparse: 0x%lx\n", (unsigned long) pMod->mod.pm.parse); + break; + case eMOD_STRGEN: + dbgprintf("Strgen Module Entry Points\n"); + dbgprintf("\tstrgen: 0x%lx\n", (unsigned long) pMod->mod.sm.strgen); + break; + } dbgprintf("\n"); pMod = GetNxt(pMod); /* done, go next */ } @@ -807,6 +951,7 @@ BEGINObjClassExit(module, OBJ_IS_LOADABLE_MODULE) /* CHANGE class also in END MA CODESTARTObjClassExit(module) /* release objects we no longer need */ objRelease(errmsg, CORE_COMPONENT); + objRelease(parser, CORE_COMPONENT); /* We have a problem in our reference counting, which leads to this function * being called too early. This usually is no problem, but if we destroy * the mutex object, we get into trouble. So rather than finding the root cause, diff --git a/runtime/modules.h b/runtime/modules.h index 4d874019..49586e8d 100644 --- a/runtime/modules.h +++ b/runtime/modules.h @@ -12,7 +12,7 @@ * * File begun on 2007-07-22 by RGerhards * - * Copyright 2007, 2008 Rainer Gerhards and Adiscon GmbH. + * Copyright 2007-2009 Rainer Gerhards and Adiscon GmbH. * * This file is part of the rsyslog runtime library. * @@ -51,9 +51,11 @@ #define CURR_MOD_IF_VERSION 5 typedef enum eModType_ { - eMOD_IN, /* input module */ - eMOD_OUT, /* output module */ - eMOD_LIB /* library module - this module provides one or many interfaces */ + eMOD_IN = 0, /* input module */ + eMOD_OUT = 1, /* output module */ + eMOD_LIB = 2, /* library module */ + eMOD_PARSER = 3,/* parser module */ + eMOD_STRGEN = 4 /* strgen module */ } eModType_t; @@ -73,7 +75,7 @@ typedef enum eModLinkType_ { eMOD_LINK_ALL /* special: all linkage types, e.g. for unload */ } eModLinkType_t; -typedef struct modInfo_s { +struct modInfo_s { struct modInfo_s *pPrev; /* support for creating a double linked module list */ struct modInfo_s *pNext; /* support for creating a linked module list */ int iIFVers; /* Interface version of module */ @@ -111,11 +113,19 @@ typedef struct modInfo_s { struct {/* data for output modules */ /* below: perform the configured action */ + rsRetVal (*beginTransaction)(void*); rsRetVal (*doAction)(uchar**, unsigned, void*); + rsRetVal (*endTransaction)(void*); rsRetVal (*parseSelectorAct)(uchar**, void**,omodStringRequest_t**); } om; struct { /* data for library modules */ - } fm; + } lm; + struct { /* data for parser modules */ + rsRetVal (*parse)(msg_t*); + } pm; + struct { /* data for strgen modules */ + rsRetVal (*strgen)(msg_t*, uchar**, size_t *); + } sm; } mod; void *pModHdlr; /* handler to the dynamic library holding the module */ # ifdef DEBUG @@ -124,7 +134,8 @@ typedef struct modInfo_s { */ modUsr_t *pModUsrRoot; # endif -} modInfo_t; +}; + /* interfaces */ BEGINinterface(module) /* name must also be changed in ENDinterface macro! */ @@ -148,8 +159,4 @@ PROTOTYPEObj(module); /* TODO: remove them below (means move the config init code) -- rgerhards, 2008-02-19 */ extern uchar *pModDir; /* read-only after startup */ - #endif /* #ifndef MODULES_H_INCLUDED */ -/* - * vi:set ai: - */ diff --git a/runtime/msg.c b/runtime/msg.c index 57291def..1b188263 100644 --- a/runtime/msg.c +++ b/runtime/msg.c @@ -35,6 +35,8 @@ #include <string.h> #include <assert.h> #include <ctype.h> +#include <sys/socket.h> +#include <netdb.h> #if HAVE_MALLOC_H # include <malloc.h> #endif @@ -51,6 +53,7 @@ #include "unicode-helper.h" #include "ruleset.h" #include "prop.h" +#include "net.h" /* static data */ DEFobjStaticHelpers @@ -59,6 +62,7 @@ DEFobjCurrIf(datetime) DEFobjCurrIf(glbl) DEFobjCurrIf(regexp) DEFobjCurrIf(prop) +DEFobjCurrIf(net) static struct { uchar *pszName; @@ -274,8 +278,13 @@ static char *syslog_severity_names[8] = { "emerg", "alert", "crit", "err", "warn static char *syslog_number_names[24] = { "0", "1", "2", "3", "4", "5", "6", "7", "8", "9", "10", "11", "12", "13", "14", "15", "16", "17", "18", "19", "20", "21", "22", "23" }; +/* global variables */ +#if defined(HAVE_MALLOC_TRIM) && !defined(HAVE_ATOMIC_BUILTINS) +static pthread_mutex_t mutTrimCtr; /* mutex to handle malloc trim */ +#endif + /* some forward declarations */ -static int getAPPNAMELen(msg_t *pM, bool bLockMutex); +static int getAPPNAMELen(msg_t *pM, sbool bLockMutex); static inline int getProtocolVersion(msg_t *pM) @@ -284,11 +293,46 @@ static inline int getProtocolVersion(msg_t *pM) } +/* do a DNS reverse resolution, if not already done, reflect status + * rgerhards, 2009-11-16 + */ +static inline rsRetVal +resolveDNS(msg_t *pMsg) { + rsRetVal localRet; + prop_t *propFromHost = NULL; + prop_t *propFromHostIP = NULL; + uchar fromHost[NI_MAXHOST]; + uchar fromHostIP[NI_MAXHOST]; + uchar fromHostFQDN[NI_MAXHOST]; + DEFiRet; + + CHKiRet(objUse(net, CORE_COMPONENT)); + if(pMsg->msgFlags & NEEDS_DNSRESOL) { + localRet = net.cvthname(pMsg->rcvFrom.pfrominet, fromHost, fromHostFQDN, fromHostIP); + if(localRet == RS_RET_OK) { + MsgSetRcvFromStr(pMsg, fromHost, ustrlen(fromHost), &propFromHost); + CHKiRet(MsgSetRcvFromIPStr(pMsg, fromHostIP, ustrlen(fromHostIP), &propFromHostIP)); + } + } +finalize_it: + if(iRet != RS_RET_OK) { + /* best we can do: remove property */ + MsgSetRcvFromStr(pMsg, UCHAR_CONSTANT(""), 0, &propFromHost); + prop.Destruct(&propFromHost); + } + if(propFromHost != NULL) + prop.Destruct(&propFromHost); + if(propFromHostIP != NULL) + prop.Destruct(&propFromHostIP); + RETiRet; +} + + static inline void getInputName(msg_t *pM, uchar **ppsz, int *plen) { BEGINfunc - if(pM == NULL) { + if(pM == NULL || pM->pInputName == NULL) { *ppsz = UCHAR_CONSTANT(""); *plen = 0; } else { @@ -307,6 +351,7 @@ getRcvFromIP(msg_t *pM) if(pM == NULL) { psz = UCHAR_CONSTANT(""); } else { + resolveDNS(pM); /* make sure we have a resolved entry */ if(pM->pRcvFromIP == NULL) psz = UCHAR_CONSTANT(""); else @@ -626,13 +671,12 @@ static inline rsRetVal msgBaseConstruct(msg_t **ppThis) msg_t *pM; assert(ppThis != NULL); - CHKmalloc(pM = malloc(sizeof(msg_t))); + CHKmalloc(pM = MALLOC(sizeof(msg_t))); objConstructSetObjInfo(pM); /* intialize object helper entities */ /* initialize members in ORDER they appear in structure (think "cache line"!) */ pM->flowCtlType = 0; pM->bDoLock = 0; - pM->bParseHOSTNAME = 0; pM->iRefCount = 1; pM->iSeverity = -1; pM->iFacility = -1; @@ -661,7 +705,7 @@ static inline rsRetVal msgBaseConstruct(msg_t **ppThis) pM->pCSMSGID = NULL; pM->pInputName = NULL; pM->pRcvFromIP = NULL; - pM->pRcvFrom = NULL; + pM->rcvFrom.pRcvFrom = NULL; pM->pRuleset = NULL; memset(&pM->tRcvdAt, 0, sizeof(pM->tRcvdAt)); memset(&pM->tTIMESTAMP, 0, sizeof(pM->tTIMESTAMP)); @@ -745,6 +789,9 @@ static inline void freeHOSTNAME(msg_t *pThis) BEGINobjDestruct(msg) /* be sure to specify the object type also in END and CODESTART macros! */ int currRefCount; +# if HAVE_MALLOC_TRIM + int currCnt; +# endif CODESTARTobjDestruct(msg) /* DEV Debugging only ! dbgprintf("msgDestruct\t0x%lx, Ref now: %d\n", (unsigned long)pThis, pThis->iRefCount - 1); */ # ifdef HAVE_ATOMIC_BUILTINS @@ -762,8 +809,12 @@ CODESTARTobjDestruct(msg) freeHOSTNAME(pThis); if(pThis->pInputName != NULL) prop.Destruct(&pThis->pInputName); - if(pThis->pRcvFrom != NULL) - prop.Destruct(&pThis->pRcvFrom); + if((pThis->msgFlags & NEEDS_DNSRESOL) == 0) { + if(pThis->rcvFrom.pRcvFrom != NULL) + prop.Destruct(&pThis->rcvFrom.pRcvFrom); + } else { + free(pThis->rcvFrom.pfrominet); + } if(pThis->pRcvFromIP != NULL) prop.Destruct(&pThis->pRcvFromIP); free(pThis->pszRcvdAt3164); @@ -800,18 +851,10 @@ CODESTARTobjDestruct(msg) * that we trim too often when the counter wraps. */ static unsigned iTrimCtr = 1; -# ifdef HAVE_ATOMICS - if(ATOMIC_INC_AND_FETCH(iTrimCtr) % 100000 == 0) { + currCnt = ATOMIC_INC_AND_FETCH(&iTrimCtr, &mutTrimCtr); + if(currCnt % 100000 == 0) { malloc_trim(128*1024); } -# else -static pthread_mutex_t mutTrimCtr = PTHREAD_MUTEX_INITIALIZER; - d_pthread_mutex_lock(&mutTrimCtr); - if(iTrimCtr++ % 100000 == 0) { - malloc_trim(128*1024); - } - d_pthread_mutex_unlock(&mutTrimCtr); -# endif } # endif } else { @@ -858,6 +901,7 @@ ENDobjDestruct(msg) msg_t* MsgDup(msg_t* pOld) { msg_t* pNew; + rsRetVal localRet; assert(pOld != NULL); @@ -870,7 +914,6 @@ msg_t* MsgDup(msg_t* pOld) pNew->iRefCount = 1; pNew->iSeverity = pOld->iSeverity; pNew->iFacility = pOld->iFacility; - pNew->bParseHOSTNAME = pOld->bParseHOSTNAME; pNew->msgFlags = pOld->msgFlags; pNew->iProtocolVersion = pOld->iProtocolVersion; pNew->ttGenTime = pOld->ttGenTime; @@ -879,9 +922,19 @@ msg_t* MsgDup(msg_t* pOld) pNew->iLenMSG = pOld->iLenMSG; pNew->iLenTAG = pOld->iLenTAG; pNew->iLenHOSTNAME = pOld->iLenHOSTNAME; - if(pOld->pRcvFrom != NULL) { - pNew->pRcvFrom = pOld->pRcvFrom; - prop.AddRef(pNew->pRcvFrom); + if((pOld->msgFlags & NEEDS_DNSRESOL) == 1) { + localRet = msgSetFromSockinfo(pNew, pOld->rcvFrom.pfrominet); + if(localRet != RS_RET_OK) { + /* if something fails, we accept loss of this property, it is + * better than losing the whole message. + */ + pNew->msgFlags &= ~NEEDS_DNSRESOL; + } + } else { + if(pOld->rcvFrom.pRcvFrom != NULL) { + pNew->rcvFrom.pRcvFrom = pOld->rcvFrom.pRcvFrom; + prop.AddRef(pNew->rcvFrom.pRcvFrom); + } } if(pOld->pRcvFromIP != NULL) { pNew->pRcvFromIP = pOld->pRcvFromIP; @@ -944,7 +997,7 @@ msg_t* MsgDup(msg_t* pOld) * We do not serialize the cache properties. We re-create them when needed. * This saves us a lot of memory. Performance is no concern, as serializing * is a so slow operation that recration of the caches does not count. Also, - * we do not serialize bParseHOSTNAME, as this is only a helper variable + * we do not serialize --currently none--, as this is only a helper variable * during msg construction - and never again used later. * rgerhards, 2008-01-03 */ @@ -1145,15 +1198,21 @@ char *getProtocolVersionString(msg_t *pM) } -static char *getRawMsg(msg_t *pM) +static inline void +getRawMsg(msg_t *pM, uchar **pBuf, int *piLen) { - if(pM == NULL) - return ""; - else - if(pM->pszRawMsg == NULL) - return ""; - else - return (char*)pM->pszRawMsg; + if(pM == NULL) { + *pBuf= UCHAR_CONSTANT(""); + *piLen = 0; + } else { + if(pM->pszRawMsg == NULL) { + *pBuf= UCHAR_CONSTANT(""); + *piLen = 0; + } else { + *pBuf = pM->pszRawMsg; + *piLen = pM->iLenRawMsg; + } + } } @@ -1198,7 +1257,8 @@ static int getPRIi(msg_t *pM) /* Get PRI value in text form */ -static inline char *getPRI(msg_t *pM) +char * +getPRI(msg_t *pM) { /* PRI is a number in the range 0..191. Thus, we use a simple lookup table to obtain the * string value. It looks a bit clumpsy here in code ;) @@ -1213,7 +1273,8 @@ static inline char *getPRI(msg_t *pM) } -static inline char *getTimeReported(msg_t *pM, enum tplFormatTypes eFmt) +char * +getTimeReported(msg_t *pM, enum tplFormatTypes eFmt) { BEGINfunc if(pM == NULL) @@ -1234,7 +1295,7 @@ static inline char *getTimeReported(msg_t *pM, enum tplFormatTypes eFmt) case tplFmtMySQLDate: MsgLock(pM); if(pM->pszTIMESTAMP_MySQL == NULL) { - if((pM->pszTIMESTAMP_MySQL = malloc(15)) == NULL) { + if((pM->pszTIMESTAMP_MySQL = MALLOC(15)) == NULL) { MsgUnlock(pM); return ""; } @@ -1245,7 +1306,7 @@ static inline char *getTimeReported(msg_t *pM, enum tplFormatTypes eFmt) case tplFmtPgSQLDate: MsgLock(pM); if(pM->pszTIMESTAMP_PgSQL == NULL) { - if((pM->pszTIMESTAMP_PgSQL = malloc(21)) == NULL) { + if((pM->pszTIMESTAMP_PgSQL = MALLOC(21)) == NULL) { MsgUnlock(pM); return ""; } @@ -1286,7 +1347,7 @@ static inline char *getTimeGenerated(msg_t *pM, enum tplFormatTypes eFmt) case tplFmtDefault: MsgLock(pM); if(pM->pszRcvdAt3164 == NULL) { - if((pM->pszRcvdAt3164 = malloc(16)) == NULL) { + if((pM->pszRcvdAt3164 = MALLOC(16)) == NULL) { MsgUnlock(pM); return ""; } @@ -1297,7 +1358,7 @@ static inline char *getTimeGenerated(msg_t *pM, enum tplFormatTypes eFmt) case tplFmtMySQLDate: MsgLock(pM); if(pM->pszRcvdAt_MySQL == NULL) { - if((pM->pszRcvdAt_MySQL = malloc(15)) == NULL) { + if((pM->pszRcvdAt_MySQL = MALLOC(15)) == NULL) { MsgUnlock(pM); return ""; } @@ -1308,7 +1369,7 @@ static inline char *getTimeGenerated(msg_t *pM, enum tplFormatTypes eFmt) case tplFmtPgSQLDate: MsgLock(pM); if(pM->pszRcvdAt_PgSQL == NULL) { - if((pM->pszRcvdAt_PgSQL = malloc(21)) == NULL) { + if((pM->pszRcvdAt_PgSQL = MALLOC(21)) == NULL) { MsgUnlock(pM); return ""; } @@ -1320,7 +1381,7 @@ static inline char *getTimeGenerated(msg_t *pM, enum tplFormatTypes eFmt) case tplFmtRFC3164BuggyDate: MsgLock(pM); if(pM->pszRcvdAt3164 == NULL) { - if((pM->pszRcvdAt3164 = malloc(16)) == NULL) { + if((pM->pszRcvdAt3164 = MALLOC(16)) == NULL) { MsgUnlock(pM); return ""; } @@ -1332,7 +1393,7 @@ static inline char *getTimeGenerated(msg_t *pM, enum tplFormatTypes eFmt) case tplFmtRFC3339Date: MsgLock(pM); if(pM->pszRcvdAt3339 == NULL) { - if((pM->pszRcvdAt3339 = malloc(33)) == NULL) { + if((pM->pszRcvdAt3339 = MALLOC(33)) == NULL) { MsgUnlock(pM); return ""; } @@ -1497,7 +1558,7 @@ finalize_it: * This must be called WITHOUT the message lock being held. * rgerhards, 2009-06-26 */ -static inline void preparePROCID(msg_t *pM, bool bLockMutex) +static inline void preparePROCID(msg_t *pM, sbool bLockMutex) { if(pM->pCSPROCID == NULL) { if(bLockMutex == LOCK_MUTEX) @@ -1514,7 +1575,7 @@ static inline void preparePROCID(msg_t *pM, bool bLockMutex) #if 0 /* rgerhards, 2005-11-24 */ -static inline int getPROCIDLen(msg_t *pM, bool bLockMutex) +static inline int getPROCIDLen(msg_t *pM, sbool bLockMutex) { assert(pM != NULL); preparePROCID(pM, bLockMutex); @@ -1525,7 +1586,7 @@ static inline int getPROCIDLen(msg_t *pM, bool bLockMutex) /* rgerhards, 2005-11-24 */ -char *getPROCID(msg_t *pM, bool bLockMutex) +char *getPROCID(msg_t *pM, sbool bLockMutex) { ISOBJ_TYPE_assert(pM, msg); preparePROCID(pM, bLockMutex); @@ -1583,7 +1644,7 @@ void MsgSetTAG(msg_t *pMsg, uchar* pszBuf, size_t lenBuf) /* small enough: use fixed buffer (faster!) */ pBuf = pMsg->TAG.szBuf; } else { - if((pBuf = (uchar*) malloc(pMsg->iLenTAG + 1)) == NULL) { + if((pBuf = (uchar*) MALLOC(pMsg->iLenTAG + 1)) == NULL) { /* truncate message, better than completely loosing it... */ pBuf = pMsg->TAG.szBuf; pMsg->iLenTAG = CONF_TAG_BUFSIZE - 1; @@ -1604,7 +1665,7 @@ void MsgSetTAG(msg_t *pMsg, uchar* pszBuf, size_t lenBuf) * if there is a TAG and, if not, if it can emulate it. * rgerhards, 2005-11-24 */ -static inline void tryEmulateTAG(msg_t *pM, bool bLockMutex) +static inline void tryEmulateTAG(msg_t *pM, sbool bLockMutex) { size_t lenTAG; uchar bufTAG[CONF_TAG_MAXSIZE]; @@ -1632,7 +1693,7 @@ static inline void tryEmulateTAG(msg_t *pM, bool bLockMutex) } -static inline void +void getTAG(msg_t *pM, uchar **ppBuf, int *piLen) { if(pM == NULL) { @@ -1657,12 +1718,13 @@ int getHOSTNAMELen(msg_t *pM) if(pM == NULL) return 0; else - if(pM->pszHOSTNAME == NULL) - if(pM->pRcvFrom == NULL) + if(pM->pszHOSTNAME == NULL) { + resolveDNS(pM); + if(pM->rcvFrom.pRcvFrom == NULL) return 0; else - return prop.GetStringLen(pM->pRcvFrom); - else + return prop.GetStringLen(pM->rcvFrom.pRcvFrom); + } else return pM->iLenHOSTNAME; } @@ -1673,12 +1735,13 @@ char *getHOSTNAME(msg_t *pM) return ""; else if(pM->pszHOSTNAME == NULL) { - if(pM->pRcvFrom == NULL) { + resolveDNS(pM); + if(pM->rcvFrom.pRcvFrom == NULL) { return ""; } else { uchar *psz; int len; - prop.GetString(pM->pRcvFrom, &psz, &len); + prop.GetString(pM->rcvFrom.pRcvFrom, &psz, &len); return (char*) psz; } } else { @@ -1692,13 +1755,15 @@ uchar *getRcvFrom(msg_t *pM) uchar *psz; int len; BEGINfunc + if(pM == NULL) { psz = UCHAR_CONSTANT(""); } else { - if(pM->pRcvFrom == NULL) + resolveDNS(pM); + if(pM->rcvFrom.pRcvFrom == NULL) psz = UCHAR_CONSTANT(""); else - prop.GetString(pM->pRcvFrom, &psz, &len); + prop.GetString(pM->rcvFrom.pRcvFrom, &psz, &len); } ENDfunc return psz; @@ -1745,7 +1810,7 @@ static inline char *getStructuredData(msg_t *pM) /* check if we have a ProgramName, and, if not, try to aquire/emulate it. * rgerhards, 2009-06-26 */ -static inline void prepareProgramName(msg_t *pM, bool bLockMutex) +static inline void prepareProgramName(msg_t *pM, sbool bLockMutex) { if(pM->pCSProgName == NULL) { if(bLockMutex == LOCK_MUTEX) @@ -1764,7 +1829,7 @@ static inline void prepareProgramName(msg_t *pM, bool bLockMutex) /* get the length of the "programname" sz string * rgerhards, 2005-10-19 */ -int getProgramNameLen(msg_t *pM, bool bLockMutex) +int getProgramNameLen(msg_t *pM, sbool bLockMutex) { assert(pM != NULL); prepareProgramName(pM, bLockMutex); @@ -1775,10 +1840,10 @@ int getProgramNameLen(msg_t *pM, bool bLockMutex) /* get the "programname" as sz string * rgerhards, 2005-10-19 */ -char *getProgramName(msg_t *pM, bool bLockMutex) +uchar *getProgramName(msg_t *pM, sbool bLockMutex) { prepareProgramName(pM, bLockMutex); - return (pM->pCSProgName == NULL) ? "" : (char*) rsCStrGetSzStrNoNULL(pM->pCSProgName); + return (pM->pCSProgName == NULL) ? UCHAR_CONSTANT("") : rsCStrGetSzStrNoNULL(pM->pCSProgName); } @@ -1795,7 +1860,7 @@ static void tryEmulateAPPNAME(msg_t *pM) if(getProtocolVersion(pM) == 0) { /* only then it makes sense to emulate */ - MsgSetAPPNAME(pM, getProgramName(pM, MUTEX_ALREADY_LOCKED)); + MsgSetAPPNAME(pM, (char*)getProgramName(pM, MUTEX_ALREADY_LOCKED)); } } @@ -1805,7 +1870,7 @@ static void tryEmulateAPPNAME(msg_t *pM) * This must be called WITHOUT the message lock being held. * rgerhards, 2009-06-26 */ -static inline void prepareAPPNAME(msg_t *pM, bool bLockMutex) +static inline void prepareAPPNAME(msg_t *pM, sbool bLockMutex) { if(pM->pCSAPPNAME == NULL) { if(bLockMutex == LOCK_MUTEX) @@ -1822,7 +1887,7 @@ static inline void prepareAPPNAME(msg_t *pM, bool bLockMutex) /* rgerhards, 2005-11-24 */ -char *getAPPNAME(msg_t *pM, bool bLockMutex) +char *getAPPNAME(msg_t *pM, sbool bLockMutex) { assert(pM != NULL); prepareAPPNAME(pM, bLockMutex); @@ -1831,7 +1896,7 @@ char *getAPPNAME(msg_t *pM, bool bLockMutex) /* rgerhards, 2005-11-24 */ -static int getAPPNAMELen(msg_t *pM, bool bLockMutex) +static int getAPPNAMELen(msg_t *pM, sbool bLockMutex) { assert(pM != NULL); prepareAPPNAME(pM, bLockMutex); @@ -1854,6 +1919,28 @@ void MsgSetInputName(msg_t *pThis, prop_t *inputName) } +/* Set the pfrominet socket store, so that we can obtain the peer at some + * later time. Note that we do not check if pRcvFrom is already set, so this + * function must only be called during message creation. + * NOTE: msgFlags is NOT set. While this is somewhat a violation of layers, + * it is done because it gains us some performance. So the caller must make + * sure the message flags are properly maintained. For all current callers, + * this is always the case and without extra effort required. + * rgerhards, 2009-11-17 + */ +rsRetVal +msgSetFromSockinfo(msg_t *pThis, struct sockaddr_storage *sa){ + DEFiRet; + assert(pThis->rcvFrom.pRcvFrom == NULL); + + CHKmalloc(pThis->rcvFrom.pfrominet = malloc(sizeof(struct sockaddr_storage))); + memcpy(pThis->rcvFrom.pfrominet, sa, sizeof(struct sockaddr_storage)); + +finalize_it: + RETiRet; +} + + /* rgerhards 2008-09-10: set RcvFrom name in msg object. This calls AddRef() * on the property, because this must be done in all current cases and there * is no case expected where this may not be necessary. @@ -1864,9 +1951,15 @@ void MsgSetRcvFrom(msg_t *pThis, prop_t *new) assert(pThis != NULL); prop.AddRef(new); - if(pThis->pRcvFrom != NULL) - prop.Destruct(&pThis->pRcvFrom); - pThis->pRcvFrom = new; + if(pThis->msgFlags & NEEDS_DNSRESOL) { + if(pThis->rcvFrom.pfrominet != NULL) + free(pThis->rcvFrom.pfrominet); + pThis->msgFlags &= ~NEEDS_DNSRESOL; + } else { + if(pThis->rcvFrom.pRcvFrom != NULL) + prop.Destruct(&pThis->rcvFrom.pRcvFrom); + } + pThis->rcvFrom.pRcvFrom = new; } @@ -1948,7 +2041,7 @@ void MsgSetHOSTNAME(msg_t *pThis, uchar* pszHOSTNAME, int lenHOSTNAME) if(pThis->iLenHOSTNAME < CONF_HOSTNAME_BUFSIZE) { /* small enough: use fixed buffer (faster!) */ pThis->pszHOSTNAME = pThis->szHOSTNAME; - } else if((pThis->pszHOSTNAME = (uchar*) malloc(pThis->iLenHOSTNAME + 1)) == NULL) { + } else if((pThis->pszHOSTNAME = (uchar*) MALLOC(pThis->iLenHOSTNAME + 1)) == NULL) { /* truncate message, better than completely loosing it... */ pThis->pszHOSTNAME = pThis->szHOSTNAME; pThis->iLenHOSTNAME = CONF_HOSTNAME_BUFSIZE - 1; @@ -2000,7 +2093,7 @@ rsRetVal MsgReplaceMSG(msg_t *pThis, uchar* pszMSG, int lenMSG) lenNew = pThis->iLenRawMsg + lenMSG - pThis->iLenMSG; if(lenMSG > pThis->iLenMSG && lenNew >= CONF_RAWMSG_BUFSIZE) { /* we have lost our "bet" and need to alloc a new buffer ;) */ - CHKmalloc(bufNew = malloc(lenNew + 1)); + CHKmalloc(bufNew = MALLOC(lenNew + 1)); memcpy(bufNew, pThis->pszRawMsg, pThis->offMSG); if(pThis->pszRawMsg != pThis->szRawMsg) free(pThis->pszRawMsg); @@ -2033,7 +2126,7 @@ void MsgSetRawMsg(msg_t *pThis, char* pszRawMsg, size_t lenMsg) if(pThis->iLenRawMsg < CONF_RAWMSG_BUFSIZE) { /* small enough: use fixed buffer (faster!) */ pThis->pszRawMsg = pThis->szRawMsg; - } else if((pThis->pszRawMsg = (uchar*) malloc(pThis->iLenRawMsg + 1)) == NULL) { + } else if((pThis->pszRawMsg = (uchar*) MALLOC(pThis->iLenRawMsg + 1)) == NULL) { /* truncate message, better than completely loosing it... */ pThis->pszRawMsg = pThis->szRawMsg; pThis->iLenRawMsg = CONF_RAWMSG_BUFSIZE - 1; @@ -2089,7 +2182,7 @@ static uchar *getNOW(eNOWType eNow) uchar *pBuf; struct syslogTime t; - if((pBuf = (uchar*) malloc(sizeof(uchar) * tmpBUFSIZE)) == NULL) { + if((pBuf = (uchar*) MALLOC(sizeof(uchar) * tmpBUFSIZE)) == NULL) { return NULL; } @@ -2200,12 +2293,13 @@ uchar *MsgGetProp(msg_t *pMsg, struct templateEntry *pTpe, break; case PROP_HOSTNAME: pRes = (uchar*)getHOSTNAME(pMsg); + bufLen = getHOSTNAMELen(pMsg); break; case PROP_SYSLOGTAG: getTAG(pMsg, &pRes, &bufLen); break; case PROP_RAWMSG: - pRes = (uchar*)getRawMsg(pMsg); + getRawMsg(pMsg, &pRes, &bufLen); break; /* enable this, if someone actually uses UxTradMsg, delete after some time has * passed and nobody complained -- rgerhards, 2009-06-16 @@ -2226,7 +2320,7 @@ uchar *MsgGetProp(msg_t *pMsg, struct templateEntry *pTpe, pRes = (uchar*)getPRI(pMsg); break; case PROP_PRI_TEXT: - pBuf = malloc(20 * sizeof(uchar)); + pBuf = MALLOC(20 * sizeof(uchar)); if(pBuf == NULL) { RET_OUT_OF_MEMORY; } else { @@ -2236,6 +2330,7 @@ uchar *MsgGetProp(msg_t *pMsg, struct templateEntry *pTpe, break; case PROP_IUT: pRes = UCHAR_CONSTANT("1"); /* always 1 for syslog messages (a MonitorWare thing;)) */ + bufLen = 1; break; case PROP_SYSLOGFACILITY: pRes = (uchar*)getFacility(pMsg); @@ -2253,7 +2348,7 @@ uchar *MsgGetProp(msg_t *pMsg, struct templateEntry *pTpe, pRes = (uchar*)getTimeGenerated(pMsg, pTpe->data.field.eDateFormat); break; case PROP_PROGRAMNAME: - pRes = (uchar*)getProgramName(pMsg, LOCK_MUTEX); + pRes = getProgramName(pMsg, LOCK_MUTEX); break; case PROP_PROTOCOL_VERSION: pRes = (uchar*)getProtocolVersionString(pMsg); @@ -2381,7 +2476,7 @@ uchar *MsgGetProp(msg_t *pMsg, struct templateEntry *pTpe, /* we got our end pointer, now do the copy */ /* TODO: code copied from below, this is a candidate for a separate function */ iLen = pFldEnd - pFld + 1; /* the +1 is for an actual char, NOT \0! */ - pBufStart = pBuf = malloc((iLen + 1) * sizeof(char)); + pBufStart = pBuf = MALLOC((iLen + 1) * sizeof(char)); if(pBuf == NULL) { if(*pbMustBeFreed == 1) free(pRes); @@ -2427,7 +2522,7 @@ uchar *MsgGetProp(msg_t *pMsg, struct templateEntry *pTpe, ; /*DO NOTHING*/ } else { iLen = iTo - iFrom + 1; /* the +1 is for an actual char, NOT \0! */ - pBufStart = pBuf = malloc((iLen + 1) * sizeof(char)); + pBufStart = pBuf = MALLOC((iLen + 1) * sizeof(char)); if(pBuf == NULL) { if(*pbMustBeFreed == 1) free(pRes); @@ -2547,7 +2642,7 @@ uchar *MsgGetProp(msg_t *pMsg, struct templateEntry *pTpe, iLenBuf = pmatch[pTpe->data.field.iSubMatchToUse].rm_eo - pmatch[pTpe->data.field.iSubMatchToUse].rm_so; - pB = malloc((iLenBuf + 1) * sizeof(uchar)); + pB = MALLOC((iLenBuf + 1) * sizeof(uchar)); if (pB == NULL) { if (*pbMustBeFreed == 1) @@ -2604,7 +2699,7 @@ uchar *MsgGetProp(msg_t *pMsg, struct templateEntry *pTpe, uchar *pBStart; uchar *pB; uchar *pSrc; - pBStart = pB = malloc((bufLen + 1) * sizeof(char)); + pBStart = pB = MALLOC((bufLen + 1) * sizeof(char)); if(pB == NULL) { if(*pbMustBeFreed == 1) free(pRes); @@ -2650,7 +2745,7 @@ uchar *MsgGetProp(msg_t *pMsg, struct templateEntry *pTpe, } if(bDropped) { - pDst = pDstStart = malloc(iLenBuf + 1); + pDst = pDstStart = MALLOC(iLenBuf + 1); if(pDst == NULL) { if(*pbMustBeFreed == 1) free(pRes); @@ -2685,7 +2780,7 @@ uchar *MsgGetProp(msg_t *pMsg, struct templateEntry *pTpe, } else { if(bufLen == -1) bufLen = ustrlen(pRes); - pDst = pDstStart = malloc(bufLen + 1); + pDst = pDstStart = MALLOC(bufLen + 1); if(pDst == NULL) { if(*pbMustBeFreed == 1) free(pRes); @@ -2724,7 +2819,7 @@ uchar *MsgGetProp(msg_t *pMsg, struct templateEntry *pTpe, int i; iLenBuf += iNumCC * 4; - pBStart = pB = malloc((iLenBuf + 1) * sizeof(uchar)); + pBStart = pB = MALLOC((iLenBuf + 1) * sizeof(uchar)); if(pB == NULL) { if(*pbMustBeFreed == 1) free(pRes); @@ -2769,7 +2864,7 @@ uchar *MsgGetProp(msg_t *pMsg, struct templateEntry *pTpe, } if(bDropped) { - pDst = pDstStart = malloc(iLenBuf + 1); + pDst = pDstStart = MALLOC(iLenBuf + 1); if(pDst == NULL) { if(*pbMustBeFreed == 1) free(pRes); @@ -2804,7 +2899,7 @@ uchar *MsgGetProp(msg_t *pMsg, struct templateEntry *pTpe, } else { if(bufLen == -1) bufLen = ustrlen(pRes); - pDst = pDstStart = malloc(bufLen + 1); + pDst = pDstStart = MALLOC(bufLen + 1); if(pDst == NULL) { if(*pbMustBeFreed == 1) free(pRes); @@ -2860,7 +2955,7 @@ uchar *MsgGetProp(msg_t *pMsg, struct templateEntry *pTpe, /* check if we need to obtain a private copy */ if(*pbMustBeFreed == 0) { /* ok, original copy, need a private one */ - pB = malloc((iLn + 1) * sizeof(uchar)); + pB = MALLOC((iLn + 1) * sizeof(uchar)); if(pB == NULL) { RET_OUT_OF_MEMORY; } @@ -2889,7 +2984,7 @@ uchar *MsgGetProp(msg_t *pMsg, struct templateEntry *pTpe, bufLen = ustrlen(pRes); iBufLen = bufLen; /* the malloc may be optimized, we currently use the worst case... */ - pBStart = pDst = malloc((2 * iBufLen + 3) * sizeof(uchar)); + pBStart = pDst = MALLOC((2 * iBufLen + 3) * sizeof(uchar)); if(pDst == NULL) { if(*pbMustBeFreed == 1) free(pRes); @@ -3090,6 +3185,10 @@ BEGINObjClassInit(msg, 1, OBJ_IS_CORE_MODULE) funcUnlock = MsgLockingDummy; funcDeleteMutex = MsgLockingDummy; funcMsgPrepareEnqueue = MsgLockingDummy; + /* some more inits */ +# if HAVE_MALLOC_TRIM + INIT_ATOMIC_HELPER_MUT(mutTrimCtr); +# endif ENDObjClassInit(msg) /* vim:set ai: */ diff --git a/runtime/msg.h b/runtime/msg.h index cda206fc..d42f1de2 100644 --- a/runtime/msg.h +++ b/runtime/msg.h @@ -60,15 +60,8 @@ struct msg { flowControl_t flowCtlType; /**< type of flow control we can apply, for enqueueing, needs not to be persisted because once data has entered the queue, this property is no longer needed. */ pthread_mutex_t mut; - bool bDoLock; /* use the mutex? */ - bool bParseHOSTNAME; /* should the hostname be parsed from the message? */ + sbool bDoLock; /* use the mutex? */ short iRefCount; /* reference counter (0 = unused) */ - /* background: the hostname is not present on "regular" messages - * received via UNIX domain sockets from the same machine. However, - * it is available when we have a forwarder (e.g. rfc3195d) using local - * sockets. All in all, the parser would need parse templates, that would - * resolve all these issues... rgerhards, 2005-10-06 - */ short iSeverity; /* the severity 0..7 */ short iFacility; /* Facility code 0 .. 23*/ short offAfterPRI; /* offset, at which raw message WITHOUT PRI part starts in pszRawMsg */ @@ -96,8 +89,12 @@ struct msg { cstr_t *pCSPROCID; /* PROCID */ cstr_t *pCSMSGID; /* MSGID */ prop_t *pInputName; /* input name property */ - prop_t *pRcvFrom; /* name of system message was received from */ prop_t *pRcvFromIP; /* IP of system message was received from */ + union { + prop_t *pRcvFrom;/* name of system message was received from */ + struct sockaddr_storage *pfrominet; /* unresolved name */ + } rcvFrom; + ruleset_t *pRuleset; /* ruleset to be used for processing this message */ time_t ttGenTime; /* time msg object was generated, same as tRcvdAt, but a Unix timestamp. While this field looks redundant, it is required because a Unix timestamp @@ -115,8 +112,8 @@ struct msg { uchar *pszTAG; /* pointer to tag value */ uchar szBuf[CONF_TAG_BUFSIZE]; } TAG; - char pszTimestamp3164[16]; - char pszTimestamp3339[33]; + char pszTimestamp3164[CONST_LEN_TIMESTAMP_3164 + 1]; + char pszTimestamp3339[CONST_LEN_TIMESTAMP_3339 + 1]; char pszTIMESTAMP_SecFrac[7]; /* Note: a pointer is 64 bits/8 char, so this is actually fewer than a pointer! */ char pszRcvdAt_SecFrac[7]; /* same as above. Both are fractional seconds for their respective timestamp */ }; @@ -131,6 +128,8 @@ struct msg { #define MARK 0x008 /* this message is a mark */ #define NEEDS_PARSING 0x010 /* raw message, must be parsed before processing can be done */ #define PARSE_HOSTNAME 0x020 /* parse the hostname during message parsing */ +#define NEEDS_DNSRESOL 0x040 /* fromhost address is unresolved and must be locked up via DNS reverse lookup first */ +#define NEEDS_ACLCHK_U 0x080 /* check UDP ACLs after DNS resolution has been done in main queue consumer */ #define NO_PRI_IN_RAW 0x100 /* rawmsg does not include a PRI (Solaris!), but PRI is already set correctly in the msg object */ @@ -151,6 +150,7 @@ void MsgSetTAG(msg_t *pMsg, uchar* pszBuf, size_t lenBuf); void MsgSetRuleset(msg_t *pMsg, ruleset_t*); rsRetVal MsgSetFlowControlType(msg_t *pMsg, flowControl_t eFlowCtl); rsRetVal MsgSetStructuredData(msg_t *pMsg, char* pszStrucData); +rsRetVal msgSetFromSockinfo(msg_t *pThis, struct sockaddr_storage *sa); void MsgSetRcvFrom(msg_t *pMsg, prop_t*); void MsgSetRcvFromStr(msg_t *pMsg, uchar* pszRcvFrom, int, prop_t **); rsRetVal MsgSetRcvFromIP(msg_t *pMsg, prop_t*); @@ -167,19 +167,22 @@ char *textpri(char *pRes, size_t pResLen, int pri); rsRetVal msgGetMsgVar(msg_t *pThis, cstr_t *pstrPropName, var_t **ppVar); rsRetVal MsgEnableThreadSafety(void); uchar *getRcvFrom(msg_t *pM); +void getTAG(msg_t *pM, uchar **ppBuf, int *piLen); +char *getTimeReported(msg_t *pM, enum tplFormatTypes eFmt); +char *getPRI(msg_t *pMsg); /* TODO: remove these five (so far used in action.c) */ uchar *getMSG(msg_t *pM); char *getHOSTNAME(msg_t *pM); -char *getPROCID(msg_t *pM, bool bLockMutex); -char *getAPPNAME(msg_t *pM, bool bLockMutex); +char *getPROCID(msg_t *pM, sbool bLockMutex); +char *getAPPNAME(msg_t *pM, sbool bLockMutex); int getMSGLen(msg_t *pM); char *getHOSTNAME(msg_t *pM); int getHOSTNAMELen(msg_t *pM); -char *getProgramName(msg_t *pM, bool bLockMutex); -int getProgramNameLen(msg_t *pM, bool bLockMutex); +uchar *getProgramName(msg_t *pM, sbool bLockMutex); +int getProgramNameLen(msg_t *pM, sbool bLockMutex); uchar *getRcvFrom(msg_t *pM); rsRetVal propNameToID(cstr_t *pCSPropName, propid_t *pPropID); uchar *propIDToName(propid_t propID); @@ -211,6 +214,16 @@ MsgSetRawMsgSize(msg_t *pMsg, size_t newLen) } +/* get the ruleset that is associated with the ruleset. + * May be NULL. -- rgerhards, 2009-10-27 + */ +static inline ruleset_t* +MsgGetRuleset(msg_t *pMsg) +{ + return pMsg->pRuleset; +} + + #endif /* #ifndef MSG_H_INCLUDED */ /* vim:set ai: */ diff --git a/runtime/net.c b/runtime/net.c index 4eb5a3bb..7653ea1d 100644 --- a/runtime/net.c +++ b/runtime/net.c @@ -173,7 +173,7 @@ AddPermittedPeerWildcard(permittedPeers_t *pPeer, uchar* pszStr, size_t lenStr) /* alloc memory for the domain component. We may waste a byte or * two, but that's ok. */ - CHKmalloc(pNew->pszDomainPart = malloc(lenStr +1 )); + CHKmalloc(pNew->pszDomainPart = MALLOC(lenStr +1 )); } if(pszStr[0] == '*') { @@ -695,7 +695,7 @@ static rsRetVal AddAllowedSender(struct AllowedSenders **ppRoot, struct AllowedS case AF_INET: /* add IPv4 */ iSignificantBits = 32; allowIP.flags = 0; - if((allowIP.addr.NetAddr = malloc(res->ai_addrlen)) == NULL) { + if((allowIP.addr.NetAddr = MALLOC(res->ai_addrlen)) == NULL) { ABORT_FINALIZE(RS_RET_OUT_OF_MEMORY); } memcpy(allowIP.addr.NetAddr, res->ai_addr, res->ai_addrlen); @@ -710,7 +710,7 @@ static rsRetVal AddAllowedSender(struct AllowedSenders **ppRoot, struct AllowedS iSignificantBits = 32; allowIP.flags = 0; - if((allowIP.addr.NetAddr = malloc(sizeof(struct sockaddr_in))) + if((allowIP.addr.NetAddr = MALLOC(sizeof(struct sockaddr_in))) == NULL) { ABORT_FINALIZE(RS_RET_OUT_OF_MEMORY); } @@ -732,7 +732,7 @@ static rsRetVal AddAllowedSender(struct AllowedSenders **ppRoot, struct AllowedS iSignificantBits = 128; allowIP.flags = 0; - if((allowIP.addr.NetAddr = malloc(res->ai_addrlen)) == NULL) { + if((allowIP.addr.NetAddr = MALLOC(res->ai_addrlen)) == NULL) { ABORT_FINALIZE(RS_RET_OUT_OF_MEMORY); } memcpy(allowIP.addr.NetAddr, res->ai_addr, res->ai_addrlen); @@ -892,15 +892,18 @@ rsRetVal addAllowedSenderLine(char* pName, uchar** ppRestOfConfLine) * including IPv4/v6 as well as domain name wildcards. * This is a helper to isAllowedSender. As it is only called once, it is * declared inline. - * Returns 0 if they do not match, something else otherwise. - * contributed 1007-07-16 by mildew@gmail.com + * Returns 0 if they do not match, 1 if they match and 2 if a DNS name would have been required. + * contributed 2007-07-16 by mildew@gmail.com */ -static inline int MaskCmp(struct NetAddr *pAllow, uint8_t bits, struct sockaddr *pFrom, const char *pszFromHost) +static inline int +MaskCmp(struct NetAddr *pAllow, uint8_t bits, struct sockaddr *pFrom, const char *pszFromHost, int bChkDNS) { assert(pAllow != NULL); assert(pFrom != NULL); if(F_ISSET(pAllow->flags, ADDR_NAME)) { + if(bChkDNS == 0) + return 2; dbgprintf("MaskCmp: host=\"%s\"; pattern=\"%s\"\n", pszFromHost, pAllow->addr.HostWildcard); # if !defined(FNM_CASEFOLD) @@ -967,18 +970,22 @@ static inline int MaskCmp(struct NetAddr *pAllow, uint8_t bits, struct sockaddr /* check if a sender is allowed. The root of the the allowed sender. * list must be proveded by the caller. As such, this function can be * used to check both UDP and TCP allowed sender lists. - * returns 1, if the sender is allowed, 0 otherwise. + * returns 1, if the sender is allowed, 0 if not and 2 if we could not + * obtain a result because we would need a dns name, which we don't have + * (2 was added rgerhards, 2009-11-16). * rgerhards, 2005-09-26 */ -static int isAllowedSender(uchar *pszType, struct sockaddr *pFrom, const char *pszFromHost) +static int isAllowedSender2(uchar *pszType, struct sockaddr *pFrom, const char *pszFromHost, int bChkDNS) { struct AllowedSenders *pAllow; struct AllowedSenders *pAllowRoot; + int bNeededDNS = 0; /* partial check because we could not resolve DNS? */ + int ret; assert(pFrom != NULL); if(setAllowRoot(&pAllowRoot, pszType) != RS_RET_OK) - return 0; /* if something went wrong, we denie access - that's the better choice... */ + return 0; /* if something went wrong, we deny access - that's the better choice... */ if(pAllowRoot == NULL) return 1; /* checking disabled, everything is valid! */ @@ -990,10 +997,20 @@ static int isAllowedSender(uchar *pszType, struct sockaddr *pFrom, const char *p * that the sender is disallowed. */ for(pAllow = pAllowRoot ; pAllow != NULL ; pAllow = pAllow->pNext) { - if (MaskCmp (&(pAllow->allowedSender), pAllow->SignificantBits, pFrom, pszFromHost)) + ret = MaskCmp (&(pAllow->allowedSender), pAllow->SignificantBits, pFrom, pszFromHost, bChkDNS); + if(ret == 1) return 1; + else if(ret == 2) + bNeededDNS = 2; } - return 0; + return bNeededDNS; +} + + +/* legacy API, not to be used any longer */ +static int +isAllowedSender(uchar *pszType, struct sockaddr *pFrom, const char *pszFromHost) { + return isAllowedSender2(pszType, pFrom, pszFromHost, 1); } @@ -1306,7 +1323,7 @@ getLocalHostname(uchar **ppName) do { if(buf == NULL) { buf_len = 128; /* Initial guess */ - CHKmalloc(buf = malloc(buf_len)); + CHKmalloc(buf = MALLOC(buf_len)); } else { buf_len += buf_len; CHKmalloc(buf = realloc (buf, buf_len)); @@ -1370,7 +1387,7 @@ int *create_udp_socket(uchar *hostname, uchar *pszPort, int bIsServer) /* Count max number of sockets we may open */ for (maxs = 0, r = res; r != NULL ; r = r->ai_next, maxs++) /* EMPTY */; - socks = malloc((maxs+1) * sizeof(int)); + socks = MALLOC((maxs+1) * sizeof(int)); if (socks == NULL) { errmsg.LogError(0, NO_ERRCODE, "couldn't allocate memory for UDP sockets, suspending UDP message reception"); freeaddrinfo(res); @@ -1533,12 +1550,36 @@ static int CmpHost(struct sockaddr_storage *s1, struct sockaddr_storage* s2, siz ret = memcmp(s1, s2, socklen); } -dbgprintf("CmpHost returns %d\n", ret); finalize_it: return ret; } + +/* check if restrictions (ALCs) exists. The goal of this function is to disable the + * somewhat time-consuming ACL checks if no restrictions are defined (the usual case). + * This also permits to gain some speedup by using firewall-based ACLs instead of + * rsyslog ACLs (the recommended method. + * rgerhards, 2009-11-16 + */ +static rsRetVal +HasRestrictions(uchar *pszType, int *bHasRestrictions) { + struct AllowedSenders *pAllowRoot; + DEFiRet; + + CHKiRet(setAllowRoot(&pAllowRoot, pszType)); + + *bHasRestrictions = (pAllowRoot == NULL) ? 0 : 1; + +finalize_it: + if(iRet != RS_RET_OK) { + *bHasRestrictions = 1; /* in this case it is better to check individually */ + DBGPRINTF("Error %d trying to obtain ACL restriction state of '%s'\n", iRet, pszType); + } + RETiRet; +} + + /* queryInterface function * rgerhards, 2008-03-05 */ @@ -1562,12 +1603,14 @@ CODESTARTobjQueryInterface(net) pIf->create_udp_socket = create_udp_socket; pIf->closeUDPListenSockets = closeUDPListenSockets; pIf->isAllowedSender = isAllowedSender; + pIf->isAllowedSender2 = isAllowedSender2; pIf->should_use_so_bsdcompat = should_use_so_bsdcompat; pIf->getLocalHostname = getLocalHostname; pIf->AddPermittedPeer = AddPermittedPeer; pIf->DestructPermittedPeers = DestructPermittedPeers; pIf->PermittedPeerWildcardMatch = PermittedPeerWildcardMatch; pIf->CmpHost = CmpHost; + pIf->HasRestrictions = HasRestrictions; /* data members */ pIf->pACLAddHostnameOnFail = &ACLAddHostnameOnFail; pIf->pACLDontResolve = &ACLDontResolve; diff --git a/runtime/net.h b/runtime/net.h index ec364b1c..101ce79d 100644 --- a/runtime/net.h +++ b/runtime/net.h @@ -139,7 +139,7 @@ BEGINinterface(net) /* name must also be changed in ENDinterface macro! */ void (*debugListenInfo)(int fd, char *type); int *(*create_udp_socket)(uchar *hostname, uchar *LogPort, int bIsServer); void (*closeUDPListenSockets)(int *finet); - int (*isAllowedSender)(uchar *pszType, struct sockaddr *pFrom, const char *pszFromHost); + int (*isAllowedSender)(uchar *pszType, struct sockaddr *pFrom, const char *pszFromHost); /* deprecated! */ rsRetVal (*getLocalHostname)(uchar**); int (*should_use_so_bsdcompat)(void); /* permitted peer handling should be replaced by something better (see comments above) */ @@ -148,11 +148,14 @@ BEGINinterface(net) /* name must also be changed in ENDinterface macro! */ rsRetVal (*PermittedPeerWildcardMatch)(permittedPeers_t *pPeer, uchar *pszNameToMatch, int *pbIsMatching); /* v5 interface additions */ int (*CmpHost)(struct sockaddr_storage *, struct sockaddr_storage*, size_t); + /* v6 interface additions - 2009-11-16 */ + rsRetVal (*HasRestrictions)(uchar *, int *bHasRestrictions); + int (*isAllowedSender2)(uchar *pszType, struct sockaddr *pFrom, const char *pszFromHost, int bChkDNS); /* data members - these should go away over time... TODO */ int *pACLAddHostnameOnFail; /* add hostname to acl when DNS resolving has failed */ int *pACLDontResolve; /* add hostname to acl instead of resolving it to IP(s) */ ENDinterface(net) -#define netCURR_IF_VERSION 5 /* increment whenever you change the interface structure! */ +#define netCURR_IF_VERSION 6 /* increment whenever you change the interface structure! */ /* prototypes */ PROTOTYPEObj(net); diff --git a/runtime/netstrms.c b/runtime/netstrms.c index 6b28e7ea..e9ff2568 100644 --- a/runtime/netstrms.c +++ b/runtime/netstrms.c @@ -36,6 +36,7 @@ #include "nsd.h" #include "netstrm.h" #include "nssel.h" +#include "nspoll.h" #include "netstrms.h" MODULE_TYPE_LIB @@ -304,6 +305,7 @@ ENDObjClassInit(netstrms) BEGINmodExit CODESTARTmodExit nsselClassExit(); + nspollClassExit(); netstrmsClassExit(); netstrmClassExit(); /* we use this object, so we must exit it after we are finished */ ENDmodExit @@ -322,6 +324,7 @@ CODESTARTmodInit /* Initialize all classes that are in our module - this includes ourselfs */ CHKiRet(netstrmClassInit(pModInfo)); CHKiRet(nsselClassInit(pModInfo)); + CHKiRet(nspollClassInit(pModInfo)); CHKiRet(netstrmsClassInit(pModInfo)); ENDmodInit /* vi:set ai: diff --git a/runtime/nsd.h b/runtime/nsd.h index 8668c934..e5b9320b 100644 --- a/runtime/nsd.h +++ b/runtime/nsd.h @@ -87,4 +87,13 @@ BEGINinterface(nsdsel) /* name must also be changed in ENDinterface macro! */ ENDinterface(nsdsel) #define nsdselCURR_IF_VERSION 1 /* increment whenever you change the interface structure! */ +/* interface for the epoll call */ +BEGINinterface(nsdpoll) /* name must also be changed in ENDinterface macro! */ + rsRetVal (*Construct)(nsdpoll_t **ppThis); + rsRetVal (*Destruct)(nsdpoll_t **ppThis); + rsRetVal (*Ctl)(nsdpoll_t *pNsdpoll, nsd_t *pNsd, int id, void *pUsr, int mode, int op); + rsRetVal (*Wait)(nsdpoll_t *pNsdpoll, int timeout, int *idRdy, void **ppUsr); +ENDinterface(nsdpoll) +#define nsdpollCURR_IF_VERSION 1 /* increment whenever you change the interface structure! */ + #endif /* #ifndef INCLUDED_NSD_H */ diff --git a/runtime/nsd_gtls.c b/runtime/nsd_gtls.c index fb2e219d..0ee70e56 100644 --- a/runtime/nsd_gtls.c +++ b/runtime/nsd_gtls.c @@ -44,6 +44,7 @@ #include "stringbuf.h" #include "errmsg.h" #include "net.h" +#include "datetime.h" #include "nsd_ptcp.h" #include "nsdsel_gtls.h" #include "nsd_gtls.h" @@ -61,6 +62,7 @@ DEFobjStaticHelpers DEFobjCurrIf(errmsg) DEFobjCurrIf(glbl) DEFobjCurrIf(net) +DEFobjCurrIf(datetime) DEFobjCurrIf(nsd_ptcp) static int bGlblSrvrInitDone = 0; /**< 0 - server global init not yet done, 1 - already done */ @@ -129,7 +131,7 @@ readFile(uchar *pszFile, gnutls_datum_t *pBuf) ABORT_FINALIZE(RS_RET_FILE_TOO_LARGE); } - CHKmalloc(pBuf->data = malloc(stat_st.st_size)); + CHKmalloc(pBuf->data = MALLOC(stat_st.st_size)); pBuf->size = stat_st.st_size; if(read(fd, pBuf->data, stat_st.st_size) != stat_st.st_size) { errmsg.LogError(0, RS_RET_IO_ERROR, "error or incomplete read of file '%s'", pszFile); @@ -1016,7 +1018,7 @@ gtlsChkPeerCertValidity(nsd_gtls_t *pThis) } /* get current time for certificate validation */ - if(time(&ttNow) == -1) + if(datetime.GetTime(&ttNow) == -1) ABORT_FINALIZE(RS_RET_SYS_ERR); /* as it looks, we need to validate the expiration dates ourselves... @@ -1482,7 +1484,7 @@ Rcv(nsd_t *pNsd, uchar *pBuf, ssize_t *pLenBuf) if(pThis->pszRcvBuf == NULL) { /* we have no buffer, so we need to malloc one */ - CHKmalloc(pThis->pszRcvBuf = malloc(NSD_GTLS_MAX_RCVBUF)); + CHKmalloc(pThis->pszRcvBuf = MALLOC(NSD_GTLS_MAX_RCVBUF)); pThis->lenRcvBuf = -1; } @@ -1697,6 +1699,7 @@ CODESTARTObjClassExit(nsd_gtls) objRelease(nsd_ptcp, LM_NSD_PTCP_FILENAME); objRelease(net, LM_NET_FILENAME); objRelease(glbl, CORE_COMPONENT); + objRelease(datetime, CORE_COMPONENT); objRelease(errmsg, CORE_COMPONENT); ENDObjClassExit(nsd_gtls) @@ -1708,6 +1711,7 @@ ENDObjClassExit(nsd_gtls) BEGINObjClassInit(nsd_gtls, 1, OBJ_IS_LOADABLE_MODULE) /* class, version */ /* request objects we use */ CHKiRet(objUse(errmsg, CORE_COMPONENT)); + CHKiRet(objUse(datetime, CORE_COMPONENT)); CHKiRet(objUse(glbl, CORE_COMPONENT)); CHKiRet(objUse(net, LM_NET_FILENAME)); CHKiRet(objUse(nsd_ptcp, LM_NSD_PTCP_FILENAME)); diff --git a/runtime/nsd_ptcp.c b/runtime/nsd_ptcp.c index 54ee0666..ca00749c 100644 --- a/runtime/nsd_ptcp.c +++ b/runtime/nsd_ptcp.c @@ -48,6 +48,7 @@ #include "netstrms.h" #include "netstrm.h" #include "nsdsel_ptcp.h" +#include "nsdpoll_ptcp.h" #include "nsd_ptcp.h" MODULE_TYPE_LIB @@ -296,12 +297,12 @@ FillRemHost(nsd_ptcp_t *pThis, struct sockaddr *pAddr) * memory consumption) */ len = strlen((char*)szIP) + 1; /* +1 for \0 byte */ - if((pThis->pRemHostIP = malloc(len)) == NULL) + if((pThis->pRemHostIP = MALLOC(len)) == NULL) ABORT_FINALIZE(RS_RET_OUT_OF_MEMORY); memcpy(pThis->pRemHostIP, szIP, len); len = strlen((char*)szHname) + 1; /* +1 for \0 byte */ - if((pThis->pRemHostName = malloc(len)) == NULL) { + if((pThis->pRemHostName = MALLOC(len)) == NULL) { free(pThis->pRemHostIP); /* prevent leak */ pThis->pRemHostIP = NULL; ABORT_FINALIZE(RS_RET_OUT_OF_MEMORY); @@ -562,6 +563,7 @@ finalize_it: static rsRetVal Rcv(nsd_t *pNsd, uchar *pRcvBuf, ssize_t *pLenBuf) { + char errStr[1024]; DEFiRet; nsd_ptcp_t *pThis = (nsd_ptcp_t*) pNsd; ISOBJ_TYPE_assert(pThis, nsd_ptcp); @@ -571,7 +573,9 @@ Rcv(nsd_t *pNsd, uchar *pRcvBuf, ssize_t *pLenBuf) if(*pLenBuf == 0) { ABORT_FINALIZE(RS_RET_CLOSED); } else if (*pLenBuf < 0) { - ABORT_FINALIZE(RS_RET_ERR); + rs_strerror_r(errno, errStr, sizeof(errStr)); + dbgprintf("error during recv on NSD %p: %s\n", pNsd, errStr); + ABORT_FINALIZE(RS_RET_RCV_ERR); } finalize_it: @@ -821,6 +825,9 @@ ENDObjClassInit(nsd_ptcp) BEGINmodExit CODESTARTmodExit +# ifdef HAVE_EPOLL_CREATE /* module only available if epoll() is supported! */ + nsdpoll_ptcpClassExit(); +# endif nsdsel_ptcpClassExit(); nsd_ptcpClassExit(); ENDmodExit @@ -839,6 +846,9 @@ CODESTARTmodInit /* Initialize all classes that are in our module - this includes ourselfs */ CHKiRet(nsd_ptcpClassInit(pModInfo)); /* must be done after tcps_sess, as we use it */ CHKiRet(nsdsel_ptcpClassInit(pModInfo)); /* must be done after tcps_sess, as we use it */ +# ifdef HAVE_EPOLL_CREATE /* module only available if epoll() is supported! */ + CHKiRet(nsdpoll_ptcpClassInit(pModInfo)); /* must be done after tcps_sess, as we use it */ +# endif ENDmodInit /* vi:set ai: */ diff --git a/runtime/nsdpoll_ptcp.c b/runtime/nsdpoll_ptcp.c new file mode 100644 index 00000000..51006707 --- /dev/null +++ b/runtime/nsdpoll_ptcp.c @@ -0,0 +1,288 @@ +/* nsdpoll_ptcp.c + * + * An implementation of the nsd epoll() interface for plain tcp sockets. + * + * Copyright 2009 Rainer Gerhards and Adiscon GmbH. + * + * This file is part of the rsyslog runtime library. + * + * The rsyslog runtime library is free software: you can redistribute it and/or modify + * it under the terms of the GNU Lesser General Public License as published by + * the Free Software Foundation, either version 3 of the License, or + * (at your option) any later version. + * + * The rsyslog runtime library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public License + * along with the rsyslog runtime library. If not, see <http://www.gnu.org/licenses/>. + * + * A copy of the GPL can be found in the file "COPYING" in this distribution. + * A copy of the LGPL can be found in the file "COPYING.LESSER" in this distribution. + */ +#include "config.h" + +#ifdef HAVE_EPOLL_CREATE /* this module requires epoll! */ + +#include <stdlib.h> +#include <assert.h> +#include <errno.h> +#include <string.h> +#if HAVE_SYS_EPOLL_H +# include <sys/epoll.h> +#endif + +#include "rsyslog.h" +#include "module-template.h" +#include "obj.h" +#include "errmsg.h" +#include "srUtils.h" +#include "nspoll.h" +#include "nsd_ptcp.h" +#include "nsdpoll_ptcp.h" +#include "unlimited_select.h" + +/* static data */ +DEFobjStaticHelpers +DEFobjCurrIf(errmsg) +DEFobjCurrIf(glbl) + + +/* -START------------------------- helpers for event list ------------------------------------ */ + +/* add new entry to list. We assume that the fd is not already present and DO NOT check this! + * Returns newly created entry in pEvtLst. + * Note that we currently need to use level-triggered mode, because the upper layers do not work + * in parallel. As such, in edge-triggered mode we may not get notified, because new data comes + * in after we have read everything that was present. To use ET mode, we need to change the upper + * peers so that they immediately start a new wait before processing the data read. That obviously + * requires more elaborate redesign and we postpone this until the current more simplictic mode has + * been proven OK in practice. + * rgerhards, 2009-11-18 + */ +static inline rsRetVal +addEvent(nsdpoll_ptcp_t *pThis, int id, void *pUsr, int mode, nsd_ptcp_t *pSock, nsdpoll_epollevt_lst_t **pEvtLst) { + nsdpoll_epollevt_lst_t *pNew; + DEFiRet; + + CHKmalloc(pNew = (nsdpoll_epollevt_lst_t*) malloc(sizeof(nsdpoll_epollevt_lst_t))); + pNew->id = id; + pNew->pUsr = pUsr; + pNew->pSock = pSock; + pNew->event.events = 0; /* TODO: at some time we should be able to use EPOLLET */ + if(mode & NSDPOLL_IN) + pNew->event.events |= EPOLLIN; + if(mode & NSDPOLL_OUT) + pNew->event.events |= EPOLLOUT; + pNew->event.data.u64 = (uint64) pNew; + pNew->pNext = pThis->pRoot; + pThis->pRoot = pNew; + *pEvtLst = pNew; + +finalize_it: + RETiRet; +} + + +/* find and unlink the entry identified by id/pUsr from the list. + * rgerhards, 2009-11-23 + */ +static inline rsRetVal +unlinkEvent(nsdpoll_ptcp_t *pThis, int id, void *pUsr, nsdpoll_epollevt_lst_t **ppEvtLst) { + nsdpoll_epollevt_lst_t *pEvtLst; + nsdpoll_epollevt_lst_t *pPrev = NULL; + DEFiRet; + + pEvtLst = pThis->pRoot; + while(pEvtLst != NULL && !(pEvtLst->id == id && pEvtLst->pUsr == pUsr)) { + pPrev = pEvtLst; + pEvtLst = pEvtLst->pNext; + } + if(pEvtLst == NULL) + ABORT_FINALIZE(RS_RET_NOT_FOUND); + + *ppEvtLst = pEvtLst; + + /* unlink */ + if(pPrev == NULL) + pThis->pRoot = pEvtLst->pNext; + else + pPrev->pNext = pEvtLst->pNext; + +finalize_it: + RETiRet; +} + + +/* destruct the provided element. It must already be unlinked from the list. + * rgerhards, 2009-11-23 + */ +static inline rsRetVal +delEvent(nsdpoll_epollevt_lst_t **ppEvtLst) { + DEFiRet; + free(*ppEvtLst); + *ppEvtLst = NULL; + RETiRet; +} + + +/* -END--------------------------- helpers for event list ------------------------------------ */ + + +/* Standard-Constructor + */ +BEGINobjConstruct(nsdpoll_ptcp) /* be sure to specify the object type also in END macro! */ +# if defined(EPOLL_CLOEXEC) && defined(HAVE_EPOLL_CREATE1) + DBGPRINTF("nsdpoll_ptcp uses epoll_create1()\n"); + pThis->efd = epoll_create1(EPOLL_CLOEXEC); +# else + DBGPRINTF("nsdpoll_ptcp uses epoll_create()\n"); + pThis->efd = epoll_create(100); /* size is ignored in newer kernels, but 100 is not bad... */ +# endif + if(pThis->efd < 0) { + DBGPRINTF("epoll_create1() could not create fd\n"); + ABORT_FINALIZE(RS_RET_IO_ERROR); + } +finalize_it: +ENDobjConstruct(nsdpoll_ptcp) + + +/* destructor for the nsdpoll_ptcp object */ +BEGINobjDestruct(nsdpoll_ptcp) /* be sure to specify the object type also in END and CODESTART macros! */ +CODESTARTobjDestruct(nsdpoll_ptcp) +ENDobjDestruct(nsdpoll_ptcp) + + +/* Modify socket set */ +static rsRetVal +Ctl(nsdpoll_t *pNsdpoll, nsd_t *pNsd, int id, void *pUsr, int mode, int op) { + nsdpoll_ptcp_t *pThis = (nsdpoll_ptcp_t*) pNsdpoll; + nsd_ptcp_t *pSock = (nsd_ptcp_t*) pNsd; + nsdpoll_epollevt_lst_t *pEventLst; + int errSave; + char errStr[512]; + DEFiRet; + + if(op == NSDPOLL_ADD) { + dbgprintf("adding nsdpoll entry %d/%p, sock %d\n", id, pUsr, pSock->sock); + CHKiRet(addEvent(pThis, id, pUsr, mode, pSock, &pEventLst)); + if(epoll_ctl(pThis->efd, EPOLL_CTL_ADD, pSock->sock, &pEventLst->event) < 0) { + errSave = errno; + rs_strerror_r(errSave, errStr, sizeof(errStr)); + errmsg.LogError(errSave, RS_RET_ERR_EPOLL_CTL, + "epoll_ctl failed on fd %d, id %d/%p, op %d with %s\n", + pSock->sock, id, pUsr, mode, errStr); + } + } else if(op == NSDPOLL_DEL) { + dbgprintf("removing nsdpoll entry %d/%p, sock %d\n", id, pUsr, pSock->sock); + CHKiRet(unlinkEvent(pThis, id, pUsr, &pEventLst)); + if(epoll_ctl(pThis->efd, EPOLL_CTL_DEL, pSock->sock, &pEventLst->event) < 0) { + errSave = errno; + rs_strerror_r(errSave, errStr, sizeof(errStr)); + errmsg.LogError(errSave, RS_RET_ERR_EPOLL_CTL, + "epoll_ctl failed on fd %d, id %d/%p, op %d with %s\n", + pSock->sock, id, pUsr, mode, errStr); + ABORT_FINALIZE(RS_RET_ERR_EPOLL_CTL); + } + CHKiRet(delEvent(&pEventLst)); + } else { + dbgprintf("program error: invalid NSDPOLL_mode %d - ignoring request\n", op); + ABORT_FINALIZE(RS_RET_ERR); + } + +finalize_it: + RETiRet; +} + + +/* Wait for io to become ready. After the successful call, idRdy contains the + * id set by the caller for that i/o event, ppUsr is a pointer to a location + * where the user pointer shall be stored. + * TODO: this is a trivial implementation that only polls one event at a time. We + * may later extend it to poll for multiple events, what would cause less + * overhead. + * rgerhards, 2009-11-18 + */ +static rsRetVal +Wait(nsdpoll_t *pNsdpoll, int timeout, int *idRdy, void **ppUsr) { + nsdpoll_ptcp_t *pThis = (nsdpoll_ptcp_t*) pNsdpoll; + nsdpoll_epollevt_lst_t *pOurEvt; + struct epoll_event event; + int nfds; + DEFiRet; + + assert(idRdy != NULL); + assert(ppUsr != NULL); + + nfds = epoll_wait(pThis->efd, &event, 1, timeout); + if(nfds == -1) { + if(errno == EINTR) { + ABORT_FINALIZE(RS_RET_EINTR); + } else { + DBGPRINTF("epoll() returned with error code %d\n", errno); + ABORT_FINALIZE(RS_RET_ERR_EPOLL); + } + } else if(nfds == 0) { + ABORT_FINALIZE(RS_RET_TIMEOUT); + } + + /* we got a valid event, so tell the caller... */ + pOurEvt = (nsdpoll_epollevt_lst_t*) event.data.u64; + *idRdy = pOurEvt->id; + *ppUsr = pOurEvt->pUsr; + +finalize_it: + RETiRet; +} + + +/* ------------------------------ end support for the epoll() interface ------------------------------ */ + + +/* queryInterface function */ +BEGINobjQueryInterface(nsdpoll_ptcp) +CODESTARTobjQueryInterface(nsdpoll_ptcp) + if(pIf->ifVersion != nsdCURR_IF_VERSION) {/* check for current version, increment on each change */ + ABORT_FINALIZE(RS_RET_INTERFACE_NOT_SUPPORTED); + } + + /* ok, we have the right interface, so let's fill it + * Please note that we may also do some backwards-compatibility + * work here (if we can support an older interface version - that, + * of course, also affects the "if" above). + */ + pIf->Construct = (rsRetVal(*)(nsdpoll_t**)) nsdpoll_ptcpConstruct; + pIf->Destruct = (rsRetVal(*)(nsdpoll_t**)) nsdpoll_ptcpDestruct; + pIf->Ctl = Ctl; + pIf->Wait = Wait; +finalize_it: +ENDobjQueryInterface(nsdpoll_ptcp) + + +/* exit our class + */ +BEGINObjClassExit(nsdpoll_ptcp, OBJ_IS_CORE_MODULE) /* CHANGE class also in END MACRO! */ +CODESTARTObjClassExit(nsdpoll_ptcp) + /* release objects we no longer need */ + objRelease(glbl, CORE_COMPONENT); + objRelease(errmsg, CORE_COMPONENT); +ENDObjClassExit(nsdpoll_ptcp) + + +/* Initialize the nsdpoll_ptcp class. Must be called as the very first method + * before anything else is called inside this class. + * rgerhards, 2008-02-19 + */ +BEGINObjClassInit(nsdpoll_ptcp, 1, OBJ_IS_CORE_MODULE) /* class, version */ + /* request objects we use */ + CHKiRet(objUse(errmsg, CORE_COMPONENT)); + CHKiRet(objUse(glbl, CORE_COMPONENT)); + + /* set our own handlers */ +ENDObjClassInit(nsdpoll_ptcp) +#endif /* #ifdef HAVE_EPOLL_CREATE this module requires epoll! */ + +/* vi:set ai: + */ diff --git a/runtime/nsdpoll_ptcp.h b/runtime/nsdpoll_ptcp.h new file mode 100644 index 00000000..cea2823d --- /dev/null +++ b/runtime/nsdpoll_ptcp.h @@ -0,0 +1,60 @@ +/* An implementation of the nsd poll interface for plain tcp sockets. + * + * Copyright 2009 Rainer Gerhards and Adiscon GmbH. + * + * This file is part of the rsyslog runtime library. + * + * The rsyslog runtime library is free software: you can redistribute it and/or modify + * it under the terms of the GNU Lesser General Public License as published by + * the Free Software Foundation, either version 3 of the License, or + * (at your option) any later version. + * + * The rsyslog runtime library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public License + * along with the rsyslog runtime library. If not, see <http://www.gnu.org/licenses/>. + * + * A copy of the GPL can be found in the file "COPYING" in this distribution. + * A copy of the LGPL can be found in the file "COPYING.LESSER" in this distribution. + */ + +#ifndef INCLUDED_NSDPOLL_PTCP_H +#define INCLUDED_NSDPOLL_PTCP_H + +#include "nsd.h" +#if HAVE_SYS_EPOLL_H +# include <sys/epoll.h> +#endif +typedef nsdpoll_if_t nsdpoll_ptcp_if_t; /* we just *implement* this interface */ +/* a helper object to keep track of the epoll event records + * Note that we need to keep track of that list because we need to + * free the events when they are no longer needed. + */ +typedef struct nsdpoll_epollevt_lst_s nsdpoll_epollevt_lst_t; +struct nsdpoll_epollevt_lst_s { +#if HAVE_SYS_EPOLL_H + epoll_event_t event; +#endif + int id; + void *pUsr; + nsd_ptcp_t *pSock; /* our associated netstream driver data */ + nsdpoll_epollevt_lst_t *pNext; +}; + +/* the nsdpoll_ptcp object */ +struct nsdpoll_ptcp_s { + BEGINobjInstance; /* Data to implement generic object - MUST be the first data element! */ + int efd; /* file descriptor used by epoll */ + nsdpoll_epollevt_lst_t *pRoot; /* Root of the epoll event list */ +}; + +/* interface is defined in nsd.h, we just implement it! */ +#define nsdpoll_ptcpCURR_IF_VERSION nsdCURR_IF_VERSION + +/* prototypes */ +PROTOTYPEObj(nsdpoll_ptcp); + +#endif /* #ifndef INCLUDED_NSDPOLL_PTCP_H */ diff --git a/runtime/nspoll.c b/runtime/nspoll.c new file mode 100644 index 00000000..64927280 --- /dev/null +++ b/runtime/nspoll.c @@ -0,0 +1,195 @@ +/* nspoll.c + * + * This is an io waiter interface utilizing the much-more-efficient poll/epoll API. + * Note that it may not always be available for a given driver. If so, that is reported + * back to the upper peer which then should consult a nssel-based io waiter. + * + * Work on this module begun 2009-11-18 by Rainer Gerhards. + * + * Copyright 2009 Rainer Gerhards and Adiscon GmbH. + * + * This file is part of the rsyslog runtime library. + * + * The rsyslog runtime library is free software: you can redistribute it and/or modify + * it under the terms of the GNU Lesser General Public License as published by + * the Free Software Foundation, either version 3 of the License, or + * (at your option) any later version. + * + * The rsyslog runtime library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public License + * along with the rsyslog runtime library. If not, see <http://www.gnu.org/licenses/>. + * + * A copy of the GPL can be found in the file "COPYING" in this distribution. + * A copy of the LGPL can be found in the file "COPYING.LESSER" in this distribution. + */ +#include "config.h" + +#include "rsyslog.h" +#include <stdio.h> +#include <stdlib.h> +#include <assert.h> +#include <errno.h> +#include <string.h> + +#include "rsyslog.h" +#include "obj.h" +#include "module-template.h" +#include "netstrm.h" +#include "nspoll.h" + +/* static data */ +DEFobjStaticHelpers +DEFobjCurrIf(glbl) + + +/* load our low-level driver. This must be done before any + * driver-specific functions (allmost all...) can be carried + * out. Note that the driver's .ifIsLoaded is correctly + * initialized by calloc() and we depend on that. Please note that + * we do some name-mangeling. We know that each nsd driver also needs + * a nspoll driver. So we simply append "sel" to the nsd driver name: This, + * of course, means that the driver name must match these rules, but that + * shouldn't be a real problem. + * WARNING: this code is mostly identical to similar code in + * netstrms.c - TODO: abstract it and move it to some common place. + * rgerhards, 2008-04-28 + */ +static rsRetVal +loadDrvr(nspoll_t *pThis) +{ + DEFiRet; + uchar *pBaseDrvrName; + uchar szDrvrName[48]; /* 48 shall be large enough */ + + pBaseDrvrName = pThis->pBaseDrvrName; + if(pBaseDrvrName == NULL) /* if no drvr name is set, use system default */ + pBaseDrvrName = glbl.GetDfltNetstrmDrvr(); + if(snprintf((char*)szDrvrName, sizeof(szDrvrName), "lmnsdpoll_%s", pBaseDrvrName) == sizeof(szDrvrName)) + ABORT_FINALIZE(RS_RET_DRVRNAME_TOO_LONG); + CHKmalloc(pThis->pDrvrName = (uchar*) strdup((char*)szDrvrName)); + + pThis->Drvr.ifVersion = nsdCURR_IF_VERSION; + /* The pDrvrName+2 below is a hack to obtain the object name. It + * safes us to have yet another variable with the name without "lm" in + * front of it. If we change the module load interface, we may re-think + * about this hack, but for the time being it is efficient and clean + * enough. -- rgerhards, 2008-04-18 + */ + CHKiRet(obj.UseObj(__FILE__, szDrvrName+2, DONT_LOAD_LIB, (void*) &pThis->Drvr)); + +finalize_it: + if(iRet != RS_RET_OK) { + if(pThis->pDrvrName != NULL) + free(pThis->pDrvrName); + pThis->pDrvrName = NULL; + } + RETiRet; +} + + +/* Standard-Constructor */ +BEGINobjConstruct(nspoll) /* be sure to specify the object type also in END macro! */ +ENDobjConstruct(nspoll) + + +/* destructor for the nspoll object */ +BEGINobjDestruct(nspoll) /* be sure to specify the object type also in END and CODESTART macros! */ +CODESTARTobjDestruct(nspoll) + if(pThis->pDrvrData != NULL) + pThis->Drvr.Destruct(&pThis->pDrvrData); + + /* and now we must release our driver, if we got one. We use the presence of + * a driver name string as load indicator (because we also need that string + * to release the driver + */ + if(pThis->pDrvrName != NULL) { + obj.ReleaseObj(__FILE__, pThis->pDrvrName+2, DONT_LOAD_LIB, (void*) &pThis->Drvr); + free(pThis->pDrvrName); + } +ENDobjDestruct(nspoll) + + +/* ConstructionFinalizer */ +static rsRetVal +ConstructFinalize(nspoll_t *pThis) +{ + DEFiRet; + ISOBJ_TYPE_assert(pThis, nspoll); + CHKiRet(loadDrvr(pThis)); + CHKiRet(pThis->Drvr.Construct(&pThis->pDrvrData)); +finalize_it: + RETiRet; +} + + +/* Carries out the actual wait (all done in lower layers) + */ +static rsRetVal +Wait(nspoll_t *pThis, int timeout, int *idRdy, void **ppUsr) { + DEFiRet; + ISOBJ_TYPE_assert(pThis, nspoll); + assert(idRdy != NULL); + iRet = pThis->Drvr.Wait(pThis->pDrvrData, timeout, idRdy, ppUsr); + RETiRet; +} + + +/* semantics like the epoll_ctl() function, does the same thing. + * rgerhards, 2009-11-18 + */ +static rsRetVal +Ctl(nspoll_t *pThis, netstrm_t *pStrm, int id, void *pUsr, int mode, int op) { + DEFiRet; + ISOBJ_TYPE_assert(pThis, nspoll); + iRet = pThis->Drvr.Ctl(pThis->pDrvrData, pStrm->pDrvrData, id, pUsr, mode, op); + RETiRet; +} + + +/* queryInterface function */ +BEGINobjQueryInterface(nspoll) +CODESTARTobjQueryInterface(nspoll) + if(pIf->ifVersion != nspollCURR_IF_VERSION) {/* check for current version, increment on each change */ + ABORT_FINALIZE(RS_RET_INTERFACE_NOT_SUPPORTED); + } + + /* ok, we have the right interface, so let's fill it + * Please note that we may also do some backwards-compatibility + * work here (if we can support an older interface version - that, + * of course, also affects the "if" above). + */ + pIf->Construct = nspollConstruct; + pIf->ConstructFinalize = ConstructFinalize; + pIf->Destruct = nspollDestruct; + pIf->Wait = Wait; + pIf->Ctl = Ctl; +finalize_it: +ENDobjQueryInterface(nspoll) + + +/* exit our class + */ +BEGINObjClassExit(nspoll, OBJ_IS_LOADABLE_MODULE) /* CHANGE class also in END MACRO! */ +CODESTARTObjClassExit(nspoll) + /* release objects we no longer need */ + objRelease(glbl, CORE_COMPONENT); +ENDObjClassExit(nspoll) + + +/* Initialize the nspoll class. Must be called as the very first method + * before anything else is called inside this class. + * rgerhards, 2008-02-19 + */ +BEGINObjClassInit(nspoll, 1, OBJ_IS_CORE_MODULE) /* class, version */ + /* request objects we use */ + DBGPRINTF("doing nspollClassInit\n"); + CHKiRet(objUse(glbl, CORE_COMPONENT)); + + /* set our own handlers */ +ENDObjClassInit(nspoll) +/* vi:set ai: + */ diff --git a/runtime/nspoll.h b/runtime/nspoll.h new file mode 100644 index 00000000..a77759c0 --- /dev/null +++ b/runtime/nspoll.h @@ -0,0 +1,65 @@ +/* Definitions for the nspoll io activity waiter + * + * Copyright 2009 Rainer Gerhards and Adiscon GmbH. + * + * This file is part of the rsyslog runtime library. + * + * The rsyslog runtime library is free software: you can redistribute it and/or modify + * it under the terms of the GNU Lesser General Public License as published by + * the Free Software Foundation, either version 3 of the License, or + * (at your option) any later version. + * + * The rsyslog runtime library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public License + * along with the rsyslog runtime library. If not, see <http://www.gnu.org/licenses/>. + * + * A copy of the GPL can be found in the file "COPYING" in this distribution. + * A copy of the LGPL can be found in the file "COPYING.LESSER" in this distribution. + */ + +#ifndef INCLUDED_NSPOLL_H +#define INCLUDED_NSPOLL_H + +#include "netstrms.h" + +/* some operations to be portable when we do not have epoll() available */ +#define NSDPOLL_ADD 1 +#define NSDPOLL_DEL 2 + +/* and some mode specifiers for waiting on input/output */ +#define NSDPOLL_IN 1 /* EPOLLIN */ +#define NSDPOLL_OUT 2 /* EPOLLOUT */ +/* next is 4, 8, 16, ... - must be bit values, as they are ored! */ + +/* the nspoll object */ +struct nspoll_s { + BEGINobjInstance; /* Data to implement generic object - MUST be the first data element! */ + nsd_t *pDrvrData; /**< the driver's data elements */ + uchar *pBaseDrvrName; /**< nsd base driver name to use, or NULL if system default */ + uchar *pDrvrName; /**< full base driver name (set when driver is loaded) */ + nsdpoll_if_t Drvr; /**< our stream driver */ +}; + + +/* interface */ +BEGINinterface(nspoll) /* name must also be changed in ENDinterface macro! */ + rsRetVal (*Construct)(nspoll_t **ppThis); + rsRetVal (*ConstructFinalize)(nspoll_t *pThis); + rsRetVal (*Destruct)(nspoll_t **ppThis); + rsRetVal (*Wait)(nspoll_t *pNsdpoll, int timeout, int *idRdy, void **ppUsr); + rsRetVal (*Ctl)(nspoll_t *pNsdpoll, netstrm_t *pStrm, int id, void *pUsr, int mode, int op); + rsRetVal (*IsEPollSupported)(void); /* static method */ +ENDinterface(nspoll) +#define nspollCURR_IF_VERSION 1 /* increment whenever you change the interface structure! */ + +/* prototypes */ +PROTOTYPEObj(nspoll); + +/* the name of our library binary */ +#define LM_NSPOLL_FILENAME LM_NETSTRMS_FILENAME + +#endif /* #ifndef INCLUDED_NSPOLL_H */ diff --git a/runtime/nssel.c b/runtime/nssel.c index d11d5fe1..7c5be3a9 100644 --- a/runtime/nssel.c +++ b/runtime/nssel.c @@ -219,6 +219,7 @@ ENDObjClassExit(nssel) */ BEGINObjClassInit(nssel, 1, OBJ_IS_CORE_MODULE) /* class, version */ /* request objects we use */ + DBGPRINTF("doing nsselClassInit\n"); CHKiRet(objUse(glbl, CORE_COMPONENT)); /* set our own handlers */ diff --git a/runtime/obj.c b/runtime/obj.c index aebea332..45dac776 100644 --- a/runtime/obj.c +++ b/runtime/obj.c @@ -48,7 +48,7 @@ * * File begun on 2008-01-04 by RGerhards * - * Copyright 2008 Rainer Gerhards and Adiscon GmbH. + * Copyright 2008, 2009 Rainer Gerhards and Adiscon GmbH. * * This file is part of the rsyslog runtime library. * @@ -90,6 +90,7 @@ #include "cfsysline.h" #include "unicode-helper.h" #include "apc.h" +#include "datetime.h" /* static data */ DEFobjCurrIf(obj) /* we define our own interface, as this is expected by some macros! */ @@ -1129,7 +1130,7 @@ UseObj(char *srcFile, uchar *pObjName, uchar *pObjFile, interface_t *pIf) /* DEV debug only: dbgprintf("source file %s requests object '%s', ifIsLoaded %d\n", srcFile, pObjName, pIf->ifIsLoaded); */ - d_pthread_mutex_lock(&mutObjGlobalOp); + pthread_mutex_lock(&mutObjGlobalOp); if(pIf->ifIsLoaded == 1) { ABORT_FINALIZE(RS_RET_OK); /* we are already set */ @@ -1170,7 +1171,7 @@ UseObj(char *srcFile, uchar *pObjName, uchar *pObjFile, interface_t *pIf) pIf->ifIsLoaded = 1; /* we are happy */ finalize_it: - d_pthread_mutex_unlock(&mutObjGlobalOp); + pthread_mutex_unlock(&mutObjGlobalOp); if(pStr != NULL) rsCStrDestruct(&pStr); @@ -1193,7 +1194,7 @@ ReleaseObj(char *srcFile, uchar *pObjName, uchar *pObjFile, interface_t *pIf) /* dev debug only dbgprintf("source file %s releasing object '%s', ifIsLoaded %d\n", srcFile, pObjName, pIf->ifIsLoaded); */ - d_pthread_mutex_lock(&mutObjGlobalOp); + pthread_mutex_lock(&mutObjGlobalOp); if(pObjFile == NULL) FINALIZE; /* if it is not a lodable module, we do not need to do anything... */ @@ -1214,7 +1215,7 @@ ReleaseObj(char *srcFile, uchar *pObjName, uchar *pObjFile, interface_t *pIf) pIf->ifIsLoaded = 0; /* indicated "no longer valid" */ finalize_it: - d_pthread_mutex_unlock(&mutObjGlobalOp); + pthread_mutex_unlock(&mutObjGlobalOp); if(pStr != NULL) rsCStrDestruct(&pStr); @@ -1330,8 +1331,9 @@ objClassInit(modInfo_t *pModInfo) CHKiRet(objGetObjInterface(&obj)); /* get ourselves ;) */ /* init classes we use (limit to as few as possible!) */ - CHKiRet(apcClassInit(pModInfo)); CHKiRet(errmsgClassInit(pModInfo)); + CHKiRet(datetimeClassInit(pModInfo)); + CHKiRet(apcClassInit(pModInfo)); CHKiRet(cfsyslineInit()); CHKiRet(varClassInit(pModInfo)); CHKiRet(moduleClassInit(pModInfo)); diff --git a/runtime/objomsr.c b/runtime/objomsr.c index 8dddc4b8..1d442c61 100644 --- a/runtime/objomsr.c +++ b/runtime/objomsr.c @@ -67,31 +67,25 @@ rsRetVal OMSRconstruct(omodStringRequest_t **ppThis, int iNumEntries) assert(ppThis != NULL); assert(iNumEntries >= 0); - if((pThis = calloc(1, sizeof(omodStringRequest_t))) == NULL) { - iRet = RS_RET_OUT_OF_MEMORY; - goto abort_it; - } + CHKmalloc(pThis = calloc(1, sizeof(omodStringRequest_t))); /* got the structure, so fill it */ pThis->iNumEntries = iNumEntries; /* allocate string for template name array. The individual strings will be * allocated as the code progresses (we do not yet know the string sizes) */ - if((pThis->ppTplName = calloc(iNumEntries, sizeof(uchar*))) == NULL) { - OMSRdestruct(pThis); - pThis = NULL; - iRet = RS_RET_OUT_OF_MEMORY; - goto abort_it; - } + CHKmalloc(pThis->ppTplName = calloc(iNumEntries, sizeof(uchar*))); + /* allocate the template options array. */ - if((pThis->piTplOpts = calloc(iNumEntries, sizeof(int))) == NULL) { - OMSRdestruct(pThis); - pThis = NULL; - iRet = RS_RET_OUT_OF_MEMORY; - goto abort_it; - } + CHKmalloc(pThis->piTplOpts = calloc(iNumEntries, sizeof(int))); -abort_it: +finalize_it: + if(iRet != RS_RET_OK) { + if(pThis != NULL) { + OMSRdestruct(pThis); + pThis = NULL; + } + } *ppThis = pThis; RETiRet; } @@ -155,7 +149,7 @@ OMSRgetSupportedTplOpts(unsigned long *pOpts) { DEFiRet; assert(pOpts != NULL); - *pOpts = OMSR_RQD_TPL_OPT_SQL | OMSR_TPL_AS_ARRAY; + *pOpts = OMSR_RQD_TPL_OPT_SQL | OMSR_TPL_AS_ARRAY | OMSR_TPL_AS_MSG; RETiRet; } diff --git a/runtime/objomsr.h b/runtime/objomsr.h index 75ad0fb8..e59b774f 100644 --- a/runtime/objomsr.h +++ b/runtime/objomsr.h @@ -27,8 +27,12 @@ /* define flags for required template options */ #define OMSR_NO_RQD_TPL_OPTS 0 #define OMSR_RQD_TPL_OPT_SQL 1 +/* only one of OMSR_TPL_AS_ARRAY or _AS_MSG must be specified, if both are given + * results are unpredictable. + */ #define OMSR_TPL_AS_ARRAY 2 /* introduced in 4.1.6, 2009-04-03 */ -/* next option is 4, 8, 16, ... */ +#define OMSR_TPL_AS_MSG 4 /* introduced in 5.3.4, 2009-11-02 */ +/* next option is 8, 16, 32, ... */ struct omodStringRequest_s { /* strings requested by output module for doAction() */ int iNumEntries; /* number of array entries for data elements below */ diff --git a/runtime/parser.c b/runtime/parser.c index 810bf42b..40374ae1 100644 --- a/runtime/parser.c +++ b/runtime/parser.c @@ -37,7 +37,13 @@ #include "dirty.h" #include "msg.h" #include "obj.h" +#include "datetime.h" #include "errmsg.h" +#include "parser.h" +#include "ruleset.h" +#include "unicode-helper.h" +#include "dirty.h" +#include "cfsysline.h" /* some defines */ #define DEFUPRI (LOG_USER|LOG_NOTICE) @@ -46,25 +52,166 @@ DEFobjStaticHelpers DEFobjCurrIf(glbl) DEFobjCurrIf(errmsg) +DEFobjCurrIf(datetime) +DEFobjCurrIf(ruleset) /* static data */ +/* config data */ +static uchar cCCEscapeChar = '#';/* character to be used to start an escape sequence for control chars */ +static int bEscapeCCOnRcv = 1; /* escape control characters on reception: 0 - no, 1 - yes */ +static int bEscape8BitChars = 0; /* escape characters > 127 on reception: 0 - no, 1 - yes */ +static int bEscapeTab = 1; /* escape tab control character when doing CC escapes: 0 - no, 1 - yes */ +static int bDropTrailingLF = 1; /* drop trailing LF's on reception? */ + +/* This is the list of all parsers known to us. + * This is also used to unload all modules on shutdown. + */ +parserList_t *pParsLstRoot = NULL; + +/* this is the list of the default parsers, to be used if no others + * are specified. + */ +parserList_t *pDfltParsLst = NULL; + + +/* intialize (but NOT allocate) a parser list. Primarily meant as a hook + * which can be used to extend the list in the future. So far, just sets + * it to NULL. + */ +static rsRetVal +InitParserList(parserList_t **pListRoot) +{ + *pListRoot = NULL; + return RS_RET_OK; +} + + +/* destruct a parser list. The list elements are destroyed, but the parser objects + * themselves are not modified. (That is done at a late stage during rsyslogd + * shutdown and need not be considered here.) + */ +static rsRetVal +DestructParserList(parserList_t **ppListRoot) +{ + parserList_t *pParsLst; + parserList_t *pParsLstDel; + + pParsLst = *ppListRoot; + while(pParsLst != NULL) { + pParsLstDel = pParsLst; + pParsLst = pParsLst->pNext; + free(pParsLstDel); + } + *ppListRoot = NULL; + return RS_RET_OK; +} + -/* this is a dummy class init +/* Add a parser to the list. We use a VERY simple and ineffcient algorithm, + * but it is employed only for a few milliseconds during config processing. So + * I prefer to keep it very simple and with simple data structures. Unfortunately, + * we need to preserve the order, but I don't like to add a tail pointer as that + * would require a container object. So I do the extra work to skip to the tail + * when adding elements... + * rgerhards, 2009-11-03 */ -rsRetVal parserClassInit(void) +static rsRetVal +AddParserToList(parserList_t **ppListRoot, parser_t *pParser) { + parserList_t *pThis; + parserList_t *pTail; DEFiRet; - /* request objects we use */ - CHKiRet(objGetObjInterface(&obj)); /* this provides the root pointer for all other queries */ - CHKiRet(objUse(glbl, CORE_COMPONENT)); - CHKiRet(objUse(errmsg, CORE_COMPONENT)); -// TODO: free components! see action.c + CHKmalloc(pThis = MALLOC(sizeof(parserList_t))); + pThis->pParser = pParser; + pThis->pNext = NULL; + + if(*ppListRoot == NULL) { + pThis->pNext = *ppListRoot; + *ppListRoot = pThis; + } else { + /* find tail first */ + for(pTail = *ppListRoot ; pTail->pNext != NULL ; pTail = pTail->pNext) + /* just search, do nothing else */; + /* add at tail */ + pTail->pNext = pThis; + } + +finalize_it: + RETiRet; +} + + +/* find a parser based on the provided name */ +static rsRetVal +FindParser(parser_t **ppParser, uchar *pName) +{ + parserList_t *pThis; + DEFiRet; + + for(pThis = pParsLstRoot ; pThis != NULL ; pThis = pThis->pNext) { + if(ustrcmp(pThis->pParser->pName, pName) == 0) { + *ppParser = pThis->pParser; + FINALIZE; /* found it, iRet still eq. OK! */ + } + } + + iRet = RS_RET_PARSER_NOT_FOUND; + +finalize_it: + RETiRet; +} + + +/* --- END helper functions for parser list handling --- */ + +/* Add a an already existing parser to the default list. As usual, order + * of calls is important (most importantly, that means the legacy parser, + * which can process everything, MUST be added last!). + * rgerhards, 2009-11-04 + */ +static rsRetVal +AddDfltParser(uchar *pName) +{ + parser_t *pParser; + DEFiRet; + + CHKiRet(FindParser(&pParser, pName)); + CHKiRet(AddParserToList(&pDfltParsLst, pParser)); + dbgprintf("Parser '%s' added to default parser set.\n", pName); + +finalize_it: + RETiRet; +} + + + +BEGINobjConstruct(parser) /* be sure to specify the object type also in END macro! */ +ENDobjConstruct(parser) + +/* ConstructionFinalizer. The most important chore is to add the parser object + * to our global list of available parsers. + * rgerhards, 2009-11-03 + */ +rsRetVal parserConstructFinalize(parser_t *pThis) +{ + DEFiRet; + + ISOBJ_TYPE_assert(pThis, parser); + CHKiRet(AddParserToList(&pParsLstRoot, pThis)); + DBGPRINTF("Parser '%s' added to list of available parsers.\n", pThis->pName); + finalize_it: RETiRet; } +BEGINobjDestruct(parser) /* be sure to specify the object type also in END and CODESTART macros! */ +CODESTARTobjDestruct(parser) + dbgprintf("destructing parser '%s'\n", pThis->pName); + free(pThis->pName); +ENDobjDestruct(parser) + /* uncompress a received message if it is compressed. * pMsg->pszRawMsg buffer is updated. @@ -96,7 +243,7 @@ static inline rsRetVal uncompressMessage(msg_t *pMsg) */ int ret; iLenDefBuf = glbl.GetMaxLine(); - CHKmalloc(deflateBuf = malloc(sizeof(uchar) * (iLenDefBuf + 1))); + CHKmalloc(deflateBuf = MALLOC(sizeof(uchar) * (iLenDefBuf + 1))); ret = uncompress((uchar *) deflateBuf, &iLenDefBuf, (uchar *) pszMsg+1, lenMsg-1); DBGPRINTF("Compressed message uncompressed with status %d, length: new %ld, old %d.\n", ret, (long) iLenDefBuf, (int) (lenMsg-1)); @@ -153,7 +300,7 @@ finalize_it: * rgerhards, 2007-09-14 */ static inline rsRetVal -sanitizeMessage(msg_t *pMsg) +SanitizeMsg(msg_t *pMsg) { DEFiRet; uchar *pszMsg; @@ -163,16 +310,12 @@ sanitizeMessage(msg_t *pMsg) size_t iDst; size_t iMaxLine; size_t maxDest; - bool bUpdatedLen = FALSE; + sbool bUpdatedLen = FALSE; uchar szSanBuf[32*1024]; /* buffer used for sanitizing a string */ assert(pMsg != NULL); assert(pMsg->iLenRawMsg > 0); -# ifdef USE_NETZIP - CHKiRet(uncompressMessage(pMsg)); -# endif - pszMsg = pMsg->pszRawMsg; lenMsg = pMsg->iLenRawMsg; @@ -202,6 +345,11 @@ sanitizeMessage(msg_t *pMsg) * needs sanitation than to do the sanitation in any case. So we first do * this and terminate when it is not needed - which is expectedly the case * for the vast majority of messages. -- rgerhards, 2009-06-15 + * Note that we do NOT check here if tab characters are to be escaped or + * not. I expect this functionality to be seldomly used and thus I do not + * like to pay the performance penalty. So the penalty is only with those + * that actually use it, because we may call the sanitizer without actual + * need below (but it then still will work perfectly well!). -- rgerhards, 2009-11-27 */ int bNeedSanitize = 0; for(iSrc = 0 ; iSrc < lenMsg ; iSrc++) { @@ -210,6 +358,9 @@ sanitizeMessage(msg_t *pMsg) bNeedSanitize = 1; break; } + } else if(pszMsg[iSrc] > 127 && bEscape8BitChars) { + bNeedSanitize = 1; + break; } } @@ -227,23 +378,31 @@ sanitizeMessage(msg_t *pMsg) if(maxDest < sizeof(szSanBuf)) pDst = szSanBuf; else - CHKmalloc(pDst = malloc(sizeof(uchar) * (iMaxLine + 1))); + CHKmalloc(pDst = MALLOC(sizeof(uchar) * (iMaxLine + 1))); iSrc = iDst = 0; while(iSrc < lenMsg && iDst < maxDest - 3) { /* leave some space if last char must be escaped */ - if(iscntrl((int) pszMsg[iSrc])) { + if(iscntrl((int) pszMsg[iSrc]) && (pszMsg[iSrc] != '\t' || bEscapeTab)) { /* note: \0 must always be escaped, the rest of the code currently * can not handle it! -- rgerhards, 2009-08-26 */ if(pszMsg[iSrc] == '\0' || bEscapeCCOnRcv) { - /* we are configured to escape control characters. Please note - * that this most probably break non-western character sets like - * Japanese, Korean or Chinese. rgerhards, 2007-07-17 + /* we are configured to escape control characters. Please note + * that this most probably break non-western character sets like + * Japanese, Korean or Chinese. rgerhards, 2007-07-17 + */ + pDst[iDst++] = cCCEscapeChar; + pDst[iDst++] = '0' + ((pszMsg[iSrc] & 0300) >> 6); + pDst[iDst++] = '0' + ((pszMsg[iSrc] & 0070) >> 3); + pDst[iDst++] = '0' + ((pszMsg[iSrc] & 0007)); + } + } else if(pszMsg[iSrc] > 127 && bEscape8BitChars) { + /* In this case, we also do the conversion. Note that this most + * probably breaks European languages. -- rgerhards, 2010-01-27 */ pDst[iDst++] = cCCEscapeChar; pDst[iDst++] = '0' + ((pszMsg[iSrc] & 0300) >> 6); pDst[iDst++] = '0' + ((pszMsg[iSrc] & 0070) >> 3); pDst[iDst++] = '0' + ((pszMsg[iSrc] & 0007)); - } } else { pDst[iDst++] = pszMsg[iSrc]; } @@ -260,27 +419,18 @@ finalize_it: RETiRet; } - -/* Parse a received message. The object's rawmsg property is taken and - * parsed according to the relevant standards. This can later be - * extended to support configured parsers. - * rgerhards, 2008-10-09 +/* A standard parser to parse out the PRI. This is made available in + * this module as it is expected that allmost all parsers will need + * that functionality and so they do not need to implement it themsleves. */ -rsRetVal parseMsg(msg_t *pMsg) +static inline rsRetVal +ParsePRI(msg_t *pMsg) { - DEFiRet; - uchar *msg; int pri; + uchar *msg; int lenMsg; int iPriText; - - if(pMsg->iLenRawMsg == 0) - ABORT_FINALIZE(RS_RET_EMPTY_MSG); - - CHKiRet(sanitizeMessage(pMsg)); - - /* we needed to sanitize first, because we otherwise do not have a C-string we can print... */ - DBGPRINTF("msg parser: flags %x, from '%s', msg '%s'\n", pMsg->msgFlags, getRcvFrom(pMsg), pMsg->pszRawMsg); + DEFiRet; /* pull PRI */ lenMsg = pMsg->iLenRawMsg; @@ -309,31 +459,252 @@ rsRetVal parseMsg(msg_t *pMsg) pMsg->iFacility = LOG_FAC(pri); pMsg->iSeverity = LOG_PRI(pri); MsgSetAfterPRIOffs(pMsg, msg - pMsg->pszRawMsg); + RETiRet; +} + + +/* Parse a received message. The object's rawmsg property is taken and + * parsed according to the relevant standards. This can later be + * extended to support configured parsers. + * rgerhards, 2008-10-09 + */ +static rsRetVal +ParseMsg(msg_t *pMsg) +{ + rsRetVal localRet = RS_RET_ERR; + parserList_t *pParserList; + parser_t *pParser; + sbool bIsSanitized; + sbool bPRIisParsed; + static int iErrMsgRateLimiter = 0; + DEFiRet; + + if(pMsg->iLenRawMsg == 0) + ABORT_FINALIZE(RS_RET_EMPTY_MSG); - /* rger 2005-11-24 (happy thanksgiving!): we now need to check if we have - * a traditional syslog message or one formatted according to syslog-protocol. - * We need to apply different parsers depending on that. We use the - * -protocol VERSION field for the detection. +# ifdef USE_NETZIP + CHKiRet(uncompressMessage(pMsg)); +# endif + + /* we take the risk to print a non-sanitized string, because this is the best we can get + * (and that functionality is too important for debugging to drop it...). + */ + DBGPRINTF("msg parser: flags %x, from '%s', msg '%.60s'\n", pMsg->msgFlags, + (pMsg->msgFlags & NEEDS_DNSRESOL) ? UCHAR_CONSTANT("~NOTRESOLVED~") : getRcvFrom(pMsg), + pMsg->pszRawMsg); + + /* we now need to go through our list of parsers and see which one is capable of + * parsing the message. Note that the first parser that requires message sanitization + * will cause it to happen. After that, access to the unsanitized message is no + * loger possible. */ - if(msg[0] == '1' && msg[1] == ' ') { - dbgprintf("Message has syslog-protocol format.\n"); - setProtocolVersion(pMsg, 1); - if(parseRFCSyslogMsg(pMsg, pMsg->msgFlags) == 1) { - msgDestruct(&pMsg); - ABORT_FINALIZE(RS_RET_ERR); // TODO: we need to handle these cases! + pParserList = ruleset.GetParserList(pMsg); + if(pParserList == NULL) { + pParserList = pDfltParsLst; + } + DBGPRINTF("parse using parser list %p%s.\n", pParserList, + (pParserList == pDfltParsLst) ? " (the default list)" : ""); + + bIsSanitized = FALSE; + bPRIisParsed = FALSE; + while(pParserList != NULL) { + pParser = pParserList->pParser; + if(pParser->bDoSanitazion && bIsSanitized == FALSE) { + CHKiRet(SanitizeMsg(pMsg)); + if(pParser->bDoPRIParsing && bPRIisParsed == FALSE) { + CHKiRet(ParsePRI(pMsg)); + bPRIisParsed = TRUE; + } + bIsSanitized = TRUE; } - } else { /* we have legacy syslog */ - dbgprintf("Message has legacy syslog format.\n"); - setProtocolVersion(pMsg, 0); - if(parseLegacySyslogMsg(pMsg, pMsg->msgFlags) == 1) { - msgDestruct(&pMsg); - ABORT_FINALIZE(RS_RET_ERR); // TODO: we need to handle these cases! + localRet = pParser->pModule->mod.pm.parse(pMsg); + dbgprintf("Parser '%s' returned %d\n", pParser->pName, localRet); + if(localRet != RS_RET_COULD_NOT_PARSE) + break; + pParserList = pParserList->pNext; + } + + /* We need to log a warning message and drop the message if we did not find a parser. + * Note that we log at most the first 1000 message, as this may very well be a problem + * that causes a message generation loop. We do not synchronize that counter, it doesn't + * matter if we log a handful messages more than we should... + */ + if(localRet != RS_RET_OK) { + if(++iErrMsgRateLimiter > 1000) { + errmsg.LogError(0, localRet, "Error: one message could not be processed by " + "any parser, message is being discarded (start of raw msg: '%.50s')", + pMsg->pszRawMsg); } + DBGPRINTF("No parser could process the message (state %d), we need to discard it.\n", localRet); + ABORT_FINALIZE(localRet); } - /* finalize message object */ + /* "finalize" message object */ pMsg->msgFlags &= ~NEEDS_PARSING; /* this message is now parsed */ finalize_it: RETiRet; } + +/* set the parser name - string is copied over, call can continue to use it, + * but must free it if desired. + */ +static rsRetVal +SetName(parser_t *pThis, uchar *name) +{ + DEFiRet; + + ISOBJ_TYPE_assert(pThis, parser); + assert(name != NULL); + + if(pThis->pName != NULL) { + free(pThis->pName); + pThis->pName = NULL; + } + + CHKmalloc(pThis->pName = ustrdup(name)); + +finalize_it: + RETiRet; +} + + +/* set a pointer to "our" module. Note that no module + * pointer must already be set. + */ +static rsRetVal +SetModPtr(parser_t *pThis, modInfo_t *pMod) +{ + ISOBJ_TYPE_assert(pThis, parser); + assert(pMod != NULL); + assert(pThis->pModule == NULL); + pThis->pModule = pMod; + return RS_RET_OK; +} + + +/* Specify if we should do standard message sanitazion before we pass the data + * down to the parser. + */ +static rsRetVal +SetDoSanitazion(parser_t *pThis, int bDoIt) +{ + ISOBJ_TYPE_assert(pThis, parser); + pThis->bDoSanitazion = bDoIt; + return RS_RET_OK; +} + + +/* Specify if we should do standard PRI parsing before we pass the data + * down to the parser module. + */ +static rsRetVal +SetDoPRIParsing(parser_t *pThis, int bDoIt) +{ + ISOBJ_TYPE_assert(pThis, parser); + pThis->bDoPRIParsing = bDoIt; + return RS_RET_OK; +} + + +/* queryInterface function-- rgerhards, 2009-11-03 + */ +BEGINobjQueryInterface(parser) +CODESTARTobjQueryInterface(parser) + if(pIf->ifVersion != parserCURR_IF_VERSION) { /* check for current version, increment on each change */ + ABORT_FINALIZE(RS_RET_INTERFACE_NOT_SUPPORTED); + } + + /* ok, we have the right interface, so let's fill it + * Please note that we may also do some backwards-compatibility + * work here (if we can support an older interface version - that, + * of course, also affects the "if" above). + */ + pIf->Construct = parserConstruct; + pIf->ConstructFinalize = parserConstructFinalize; + pIf->Destruct = parserDestruct; + pIf->SetName = SetName; + pIf->SetModPtr = SetModPtr; + pIf->SetDoSanitazion = SetDoSanitazion; + pIf->SetDoPRIParsing = SetDoPRIParsing; + pIf->ParseMsg = ParseMsg; + pIf->SanitizeMsg = SanitizeMsg; + pIf->InitParserList = InitParserList; + pIf->DestructParserList = DestructParserList; + pIf->AddParserToList = AddParserToList; + pIf->AddDfltParser = AddDfltParser; + pIf->FindParser = FindParser; +finalize_it: +ENDobjQueryInterface(parser) + + + +/* Reset config variables to default values. + * rgerhards, 2007-07-17 + */ +static rsRetVal +resetConfigVariables(uchar __attribute__((unused)) *pp, void __attribute__((unused)) *pVal) +{ + cCCEscapeChar = '#'; + bEscapeCCOnRcv = 1; /* default is to escape control characters */ + bEscape8BitChars = 0; /* default is to escape control characters */ + bEscapeTab = 1; /* default is to escape control characters */ + bDropTrailingLF = 1; /* default is to drop trailing LF's on reception */ + + return RS_RET_OK; +} + +/* This destroys the master parserlist and all of its parser entries. MUST only be + * done when the module is shut down. Parser modules are NOT unloaded, rsyslog + * does that at a later stage for all dynamically loaded modules. + */ +static void +destroyMasterParserList(void) +{ + parserList_t *pParsLst; + parserList_t *pParsLstDel; + + pParsLst = pParsLstRoot; + while(pParsLst != NULL) { + parserDestruct(&pParsLst->pParser); + pParsLstDel = pParsLst; + pParsLst = pParsLst->pNext; + free(pParsLstDel); + } +} + +/* Exit our class. + * rgerhards, 2009-11-04 + */ +BEGINObjClassExit(parser, OBJ_IS_CORE_MODULE) /* class, version */ + DestructParserList(&pDfltParsLst); + destroyMasterParserList(); + objRelease(glbl, CORE_COMPONENT); + objRelease(errmsg, CORE_COMPONENT); + objRelease(datetime, CORE_COMPONENT); + objRelease(ruleset, CORE_COMPONENT); +ENDObjClassExit(parser) + + +/* Initialize the parser class. Must be called as the very first method + * before anything else is called inside this class. + * rgerhards, 2009-11-02 + */ +BEGINObjClassInit(parser, 1, OBJ_IS_CORE_MODULE) /* class, version */ + /* request objects we use */ + CHKiRet(objUse(glbl, CORE_COMPONENT)); + CHKiRet(objUse(errmsg, CORE_COMPONENT)); + CHKiRet(objUse(datetime, CORE_COMPONENT)); + CHKiRet(objUse(ruleset, CORE_COMPONENT)); + + CHKiRet(regCfSysLineHdlr((uchar *)"controlcharacterescapeprefix", 0, eCmdHdlrGetChar, NULL, &cCCEscapeChar, NULL)); + CHKiRet(regCfSysLineHdlr((uchar *)"droptrailinglfonreception", 0, eCmdHdlrBinary, NULL, &bDropTrailingLF, NULL)); + CHKiRet(regCfSysLineHdlr((uchar *)"escapecontrolcharactersonreceive", 0, eCmdHdlrBinary, NULL, &bEscapeCCOnRcv, NULL)); + CHKiRet(regCfSysLineHdlr((uchar *)"escape8bitcharactersonreceive", 0, eCmdHdlrBinary, NULL, &bEscape8BitChars, NULL)); + CHKiRet(regCfSysLineHdlr((uchar *)"escapecontrolcharactertab", 0, eCmdHdlrBinary, NULL, &bEscapeTab, NULL)); + CHKiRet(regCfSysLineHdlr((uchar *)"resetconfigvariables", 1, eCmdHdlrCustomHandler, resetConfigVariables, NULL, NULL)); + + InitParserList(&pParsLstRoot); + InitParserList(&pDfltParsLst); +ENDObjClassInit(parser) + diff --git a/runtime/parser.h b/runtime/parser.h index cec9c083..bdd572cb 100644 --- a/runtime/parser.h +++ b/runtime/parser.h @@ -1,8 +1,6 @@ /* header for parser.c - * This is not yet an object, but contains all those code necessary to - * parse syslog messages. * - * Copyright 2008 Rainer Gerhards and Adiscon GmbH. + * Copyright 2008,2009 Rainer Gerhards and Adiscon GmbH. * * This file is part of the rsyslog runtime library. * @@ -21,10 +19,53 @@ * * A copy of the LGPL can be found in the file "COPYING.LESSER" in this distribution. */ -#ifndef INCLUDED_PARSE_H -#define INCLUDED_PARSE_H +#ifndef INCLUDED_PARSER_H +#define INCLUDED_PARSER_H -extern rsRetVal parserClassInit(void); -extern rsRetVal parseMsg(msg_t*); -#endif /* #ifndef INCLUDED_PARSE_H */ +/* we create a small helper object, a list of parsers, that we can use to + * build a chain of them whereever this is needed (initially thought to be + * used in ruleset.c as well as ourselvs). + */ +struct parserList_s { + parser_t *pParser; + parserList_t *pNext; +}; + + +/* the parser object, a dummy because we have only static methods */ +struct parser_s { + BEGINobjInstance; /* Data to implement generic object - MUST be the first data element! */ + uchar *pName; /* name of this parser */ + modInfo_t *pModule; /* pointer to parser's module */ + sbool bDoSanitazion; /* do standard message sanitazion before calling parser? */ + sbool bDoPRIParsing; /* do standard PRI parsing before calling parser? */ +}; + +/* interfaces */ +BEGINinterface(parser) /* name must also be changed in ENDinterface macro! */ + INTERFACEObjDebugPrint(var); + rsRetVal (*Construct)(parser_t **ppThis); + rsRetVal (*ConstructFinalize)(parser_t *pThis); + rsRetVal (*Destruct)(parser_t **ppThis); + rsRetVal (*SetName)(parser_t *pThis, uchar *name); + rsRetVal (*SetModPtr)(parser_t *pThis, modInfo_t *pMod); + rsRetVal (*SetDoSanitazion)(parser_t *pThis, int); + rsRetVal (*SetDoPRIParsing)(parser_t *pThis, int); + rsRetVal (*FindParser)(parser_t **ppThis, uchar*name); + rsRetVal (*InitParserList)(parserList_t **pListRoot); + rsRetVal (*DestructParserList)(parserList_t **pListRoot); + rsRetVal (*AddParserToList)(parserList_t **pListRoot, parser_t *pParser); + /* static functions */ + rsRetVal (*ParseMsg)(msg_t *pMsg); + rsRetVal (*SanitizeMsg)(msg_t *pMsg); + rsRetVal (*AddDfltParser)(uchar *); +ENDinterface(parser) +#define parserCURR_IF_VERSION 1 /* increment whenever you change the interface above! */ + + +/* prototypes */ +PROTOTYPEObj(parser); + + +#endif /* #ifndef INCLUDED_PARSER_H */ diff --git a/runtime/prop.c b/runtime/prop.c index 7f2a56ff..d925bb43 100644 --- a/runtime/prop.c +++ b/runtime/prop.c @@ -85,7 +85,7 @@ static rsRetVal SetString(prop_t *pThis, uchar *psz, int len) if(len < CONF_PROP_BUFSIZE) { memcpy(pThis->szVal.sz, psz, len + 1); } else { - CHKmalloc(pThis->szVal.psz = malloc(len + 1)); + CHKmalloc(pThis->szVal.psz = MALLOC(len + 1)); memcpy(pThis->szVal.psz, psz, len + 1); } diff --git a/runtime/queue.c b/runtime/queue.c index bedefb77..60d17086 100644 --- a/runtime/queue.c +++ b/runtime/queue.c @@ -8,7 +8,11 @@ * (and in the web doc set on http://www.rsyslog.com/doc). Be sure to read it * if you are getting aquainted to the object. * - * Copyright 2008 Rainer Gerhards and Adiscon GmbH. + * NOTE: as of 2009-04-22, I have begin to remove the qqueue* prefix from static + * function names - this makes it really hard to read and does not provide much + * benefit, at least I (now) think so... + * + * Copyright 2008, 2009 Rainer Gerhards and Adiscon GmbH. * * This file is part of the rsyslog runtime library. * @@ -51,71 +55,175 @@ #include "wti.h" #include "msg.h" #include "atomic.h" +#include "errmsg.h" +#include "datetime.h" +#include "unicode-helper.h" +#include "msg.h" /* TODO: remove once we remove MsgAddRef() call */ #ifdef OS_SOLARIS # include <sched.h> -# define pthread_yield() sched_yield() #endif /* static data */ DEFobjStaticHelpers DEFobjCurrIf(glbl) DEFobjCurrIf(strm) +DEFobjCurrIf(errmsg) +DEFobjCurrIf(datetime) /* forward-definitions */ -static rsRetVal qqueueChkPersist(qqueue_t *pThis); -static rsRetVal qqueueSetEnqOnly(qqueue_t *pThis, int bEnqOnly, int bLockMutex); -static rsRetVal qqueueRateLimiter(qqueue_t *pThis); +static inline rsRetVal doEnqSingleObj(qqueue_t *pThis, flowControl_t flowCtlType, void *pUsr); +static rsRetVal qqueueChkPersist(qqueue_t *pThis, int nUpdates); +static rsRetVal RateLimiter(qqueue_t *pThis); static int qqueueChkStopWrkrDA(qqueue_t *pThis); -static int qqueueIsIdleDA(qqueue_t *pThis); -static rsRetVal qqueueConsumerDA(qqueue_t *pThis, wti_t *pWti, int iCancelStateSave); -static rsRetVal qqueueConsumerCancelCleanup(void *arg1, void *arg2); -static rsRetVal qqueueUngetObj(qqueue_t *pThis, obj_t *pUsr, int bLockMutex); +static rsRetVal GetDeqBatchSize(qqueue_t *pThis, int *pVal); +static rsRetVal ConsumerDA(qqueue_t *pThis, wti_t *pWti); +static rsRetVal batchProcessed(qqueue_t *pThis, wti_t *pWti); +static rsRetVal qqueueMultiEnqObjNonDirect(qqueue_t *pThis, multi_submit_t *pMultiSub); +static rsRetVal qqueueMultiEnqObjDirect(qqueue_t *pThis, multi_submit_t *pMultiSub); /* some constants for queuePersist () */ #define QUEUE_CHECKPOINT 1 #define QUEUE_NO_CHECKPOINT 0 +/*********************************************************************** + * we need a private data structure, the "to-delete" list. As C does + * not provide any partly private data structures, we implement this + * structure right here inside the module. + * Note that this list must always be kept sorted based on a unique + * dequeue ID (which is monotonically increasing). + * rgerhards, 2009-05-18 + ***********************************************************************/ + +/* generate next uniqueue dequeue ID. Note that uniqueness is only required + * on a per-queue basis and while this instance runs. So a stricly monotonically + * increasing counter is sufficient (if enough bits are used). + */ +static inline qDeqID getNextDeqID(qqueue_t *pQueue) +{ + ISOBJ_TYPE_assert(pQueue, qqueue); + return pQueue->deqIDAdd++; +} + + +/* return the top element of the to-delete list or NULL, if the + * list is empty. + */ +static inline toDeleteLst_t *tdlPeek(qqueue_t *pQueue) +{ + ISOBJ_TYPE_assert(pQueue, qqueue); + return pQueue->toDeleteLst; +} + + +/* remove the top element of the to-delete list. Nothing but the + * element itself is destroyed. Must not be called when the list + * is empty. + */ +static inline rsRetVal tdlPop(qqueue_t *pQueue) +{ + toDeleteLst_t *pRemove; + DEFiRet; + + ISOBJ_TYPE_assert(pQueue, qqueue); + assert(pQueue->toDeleteLst != NULL); + + pRemove = pQueue->toDeleteLst; + pQueue->toDeleteLst = pQueue->toDeleteLst->pNext; + free(pRemove); + + RETiRet; +} + + +/* Add a new to-delete list entry. The function allocates the data + * structure, populates it with the values provided and links the new + * element into the correct place inside the list. + */ +static inline rsRetVal tdlAdd(qqueue_t *pQueue, qDeqID deqID, int nElemDeq) +{ + toDeleteLst_t *pNew; + toDeleteLst_t *pPrev; + DEFiRet; + + ISOBJ_TYPE_assert(pQueue, qqueue); + assert(pQueue->toDeleteLst != NULL); + + CHKmalloc(pNew = MALLOC(sizeof(toDeleteLst_t))); + pNew->deqID = deqID; + pNew->nElemDeq = nElemDeq; + + /* now find right spot */ + for( pPrev = pQueue->toDeleteLst + ; pPrev != NULL && deqID > pPrev->deqID + ; pPrev = pPrev->pNext) { + /*JUST SEARCH*/; + } + + if(pPrev == NULL) { + pNew->pNext = pQueue->toDeleteLst; + pQueue->toDeleteLst = pNew; + } else { + pNew->pNext = pPrev->pNext; + pPrev->pNext = pNew; + } + +finalize_it: + RETiRet; +} + + /* methods */ -/* get the overall queue size, which includes ungotten objects. Must only be called +/* get the physical queue size. Must only be called * while mutex is locked! * rgerhards, 2008-01-29 */ static inline int -qqueueGetOverallQueueSize(qqueue_t *pThis) +getPhysicalQueueSize(qqueue_t *pThis) { -#if 0 /* leave a bit in for debugging -- rgerhards, 2008-01-30 */ -BEGINfunc -dbgoprint((obj_t*) pThis, "queue size: %d (regular %d, ungotten %d)\n", - pThis->iQueueSize + pThis->iUngottenObjs, pThis->iQueueSize, pThis->iUngottenObjs); -ENDfunc -#endif - return pThis->iQueueSize + pThis->iUngottenObjs; + return pThis->iQueueSize; } +/* get the logical queue size (that is store size minus logically dequeued elements). + * Must only be called while mutex is locked! + * rgerhards, 2009-05-19 + */ +static inline int +getLogicalQueueSize(qqueue_t *pThis) +{ + return pThis->iQueueSize - pThis->nLogDeq; +} + + + /* This function drains the queue in cases where this needs to be done. The most probable * reason is a HUP which needs to discard data (because the queue is configured to be lossy). * During a shutdown, this is typically not needed, as the OS frees up ressources and does * this much quicker than when we clean up ourselvs. -- rgerhards, 2008-10-21 * This function returns void, as it makes no sense to communicate an error back, even if * it happens. + * This functions works "around" the regular deque mechanism, because it is only used to + * clean up (in cases where message loss is acceptable). */ static inline void queueDrain(qqueue_t *pThis) { void *pUsr; - ASSERT(pThis != NULL); + BEGINfunc + DBGOPRINT((obj_t*) pThis, "queue (type %d) will lose %d messages, destroying...\n", pThis->qType, pThis->iQueueSize); /* iQueueSize is not decremented by qDel(), so we need to do it ourselves */ while(ATOMIC_DEC_AND_FETCH(&pThis->iQueueSize, &pThis->mutQueueSize) > 0) { - pThis->qDel(pThis, &pUsr); + pThis->qDeq(pThis, &pUsr); if(pUsr != NULL) { objDestruct(pUsr); } + pThis->qDel(pThis); } + ENDfunc } @@ -126,7 +234,8 @@ static inline void queueDrain(qqueue_t *pThis) * this point in time. The mutex must be locked when * ths function is called. -- rgerhards, 2008-01-25 */ -static inline rsRetVal qqueueAdviseMaxWorkers(qqueue_t *pThis) +static inline rsRetVal +qqueueAdviseMaxWorkers(qqueue_t *pThis) { DEFiRet; int iMaxWorkers; @@ -134,96 +243,17 @@ static inline rsRetVal qqueueAdviseMaxWorkers(qqueue_t *pThis) ISOBJ_TYPE_assert(pThis, qqueue); if(!pThis->bEnqOnly) { - if(pThis->bRunsDA) { - /* if we have not yet reached the high water mark, there is no need to start a - * worker. -- rgerhards, 2008-01-26 - */ - if(qqueueGetOverallQueueSize(pThis) >= pThis->iHighWtrMrk || pThis->bQueueStarted == 0) { - wtpAdviseMaxWorkers(pThis->pWtpDA, 1); /* disk queues have always one worker */ - } + if(pThis->bIsDA && getLogicalQueueSize(pThis) >= pThis->iHighWtrMrk) { + wtpAdviseMaxWorkers(pThis->pWtpDA, 1); /* disk queues have always one worker */ } else { - if(pThis->qType == QUEUETYPE_DISK || pThis->iMinMsgsPerWrkr == 0) { + if(getLogicalQueueSize(pThis) == 0) { + iMaxWorkers = 0; + } else if(pThis->qType == QUEUETYPE_DISK || pThis->iMinMsgsPerWrkr == 0) { iMaxWorkers = 1; } else { - iMaxWorkers = qqueueGetOverallQueueSize(pThis) / pThis->iMinMsgsPerWrkr + 1; + iMaxWorkers = getLogicalQueueSize(pThis) / pThis->iMinMsgsPerWrkr + 1; } - wtpAdviseMaxWorkers(pThis->pWtpReg, iMaxWorkers); /* disk queues have always one worker */ - } - } - - RETiRet; -} - - -/* wait until we have a fully initialized DA queue. Sometimes, we need to - * sync with it, as we expect it for some function. - * rgerhards, 2008-02-27 - */ -static rsRetVal -qqueueWaitDAModeInitialized(qqueue_t *pThis) -{ - DEFiRet; - - ISOBJ_TYPE_assert(pThis, qqueue); - ASSERT(pThis->bRunsDA); - - while(pThis->bRunsDA != 2) { - d_pthread_cond_wait(&pThis->condDAReady, pThis->mut); - } - - RETiRet; -} - - -/* Destruct DA queue. This is the last part of DA-to-normal-mode - * transistion. This is called asynchronously and some time quite a - * while after the actual transistion. The key point is that we need to - * do it at some later time, because we need to destruct the DA queue. That, - * however, can not be done in a thread that has been signalled - * This is to be called when we revert back to our own queue. - * This function must be called with the queue mutex locked (the wti - * class ensures this). - * rgerhards, 2008-01-15 - */ -static rsRetVal -qqueueTurnOffDAMode(qqueue_t *pThis) -{ - DEFiRet; - - ISOBJ_TYPE_assert(pThis, qqueue); - ASSERT(pThis->bRunsDA); - - /* at this point, we need a fully initialized DA queue. So if it isn't, we finally need - * to wait for its startup... -- rgerhards, 2008-01-25 - */ - qqueueWaitDAModeInitialized(pThis); - - /* if we need to pull any data that we still need from the (child) disk queue, - * now would be the time to do so. At present, we do not need this, but I'd like to - * keep that comment if future need arises. - */ - - /* we need to check if the DA queue is empty because the DA worker may simply have - * terminated do to no new messages arriving. That does not, however, mean that the - * DA queue is empty. If there is still data in that queue, we do nothing and leave - * that for a later incarnation of this function (it will be called multiple times - * during the lifetime of DA-mode, depending on how often the DA worker receives an - * inactivity timeout. -- rgerhards, 2008-01-25 - */ - if(pThis->pqDA->iQueueSize == 0) { - pThis->bRunsDA = 0; /* tell the world we are back in non-DA mode */ - /* we destruct the queue object, which will also shutdown the queue worker. As the queue is empty, - * this will be quick. - */ - qqueueDestruct(&pThis->pqDA); /* and now we are ready to destruct the DA queue */ - dbgoprint((obj_t*) pThis, "disk-assistance has been turned off, disk queue was empty (iRet %d)\n", - iRet); - /* now we need to check if the regular queue has some messages. This may be the case - * when it is waiting that the high water mark is reached again. If so, we need to start up - * a regular worker. -- rgerhards, 2008-01-26 - */ - if(qqueueGetOverallQueueSize(pThis) > 0) { - qqueueAdviseMaxWorkers(pThis); + wtpAdviseMaxWorkers(pThis->pWtpReg, iMaxWorkers); } } @@ -246,37 +276,26 @@ qqueueChkIsDA(qqueue_t *pThis) ISOBJ_TYPE_assert(pThis, qqueue); if(pThis->pszFilePrefix != NULL) { pThis->bIsDA = 1; - dbgoprint((obj_t*) pThis, "is disk-assisted, disk will be used on demand\n"); + DBGOPRINT((obj_t*) pThis, "is disk-assisted, disk will be used on demand\n"); } else { - dbgoprint((obj_t*) pThis, "is NOT disk-assisted\n"); + DBGOPRINT((obj_t*) pThis, "is NOT disk-assisted\n"); } RETiRet; } -/* Start disk-assisted queue mode. All internal settings are changed. This is supposed - * to be called from the DA worker, which must have been started before. The most important - * chore of this function is to create the DA queue object. If that function fails, - * the DA worker should return with an appropriate state, which in turn should lead to - * a re-set to non-DA mode in the Enq process. The queue mutex must be locked when this - * function is called, else a number of races will happen. - * Please note that this function may be called *while* we in DA mode. This is due to the - * fact that the DA worker calls it and the DA worker may be suspended (and restarted) due - * to inactivity timeouts. +/* Start disk-assisted queue mode. * rgerhards, 2008-01-15 */ static rsRetVal -qqueueStartDA(qqueue_t *pThis) +StartDA(qqueue_t *pThis) { DEFiRet; uchar pszDAQName[128]; ISOBJ_TYPE_assert(pThis, qqueue); - if(pThis->bRunsDA == 2) /* check if already in (fully initialized) DA mode... */ - FINALIZE; /* ... then we are already done! */ - /* create message queue */ CHKiRet(qqueueConstruct(&pThis->pqDA, QUEUETYPE_DISK , 1, 0, pThis->pConsumer)); @@ -298,39 +317,22 @@ qqueueStartDA(qqueue_t *pThis) CHKiRet(qqueueSetbSyncQueueFiles(pThis->pqDA, pThis->bSyncQueueFiles)); CHKiRet(qqueueSettoActShutdown(pThis->pqDA, pThis->toActShutdown)); CHKiRet(qqueueSettoEnq(pThis->pqDA, pThis->toEnq)); - CHKiRet(qqueueSetEnqOnly(pThis->pqDA, pThis->bDAEnqOnly, MUTEX_ALREADY_LOCKED)); CHKiRet(qqueueSetiDeqtWinFromHr(pThis->pqDA, pThis->iDeqtWinFromHr)); CHKiRet(qqueueSetiDeqtWinToHr(pThis->pqDA, pThis->iDeqtWinToHr)); + CHKiRet(qqueueSettoQShutdown(pThis->pqDA, pThis->toQShutdown)); CHKiRet(qqueueSetiHighWtrMrk(pThis->pqDA, 0)); CHKiRet(qqueueSetiDiscardMrk(pThis->pqDA, 0)); - if(pThis->toQShutdown == 0) { - CHKiRet(qqueueSettoQShutdown(pThis->pqDA, 0)); /* if the user really wants... */ - } else { - /* we use the shortest possible shutdown (0 is endless!) because when we run on disk AND - * have an obviously large backlog, we can't finish it in any case. So there is no point - * in holding shutdown longer than necessary. -- rgerhards, 2008-01-15 - */ - CHKiRet(qqueueSettoQShutdown(pThis->pqDA, 1)); - } iRet = qqueueStart(pThis->pqDA); /* file not found is expected, that means it is no previous QIF available */ - if(iRet != RS_RET_OK && iRet != RS_RET_FILE_NOT_FOUND) + if(iRet != RS_RET_OK && iRet != RS_RET_FILE_NOT_FOUND) { + errno = 0; /* else an errno is shown in errmsg! */ + errmsg.LogError(errno, iRet, "error starting up disk queue, using pure in-memory mode"); + pThis->bIsDA = 0; /* disable memory mode */ FINALIZE; /* something is wrong */ + } - /* as we are right now starting DA mode because we are so busy, it is - * extremely unlikely that any regular worker is sleeping on empty queue. HOWEVER, - * we want to be on the safe side, and so we awake anyone that is waiting - * on one. So even if the scheduler plays badly with us, things should be - * quite well. -- rgerhards, 2008-01-15 - */ - wtpWakeupWrkr(pThis->pWtpReg); /* awake all workers, but not ourselves ;) */ - - pThis->bRunsDA = 2; /* we are now in DA mode, but not fully initialized */ - pThis->bChildIsDone = 0;/* set to 1 when child's worker detect queue is finished */ - pthread_cond_broadcast(&pThis->condDAReady); /* signal we are now initialized and ready to go ;) */ - - dbgoprint((obj_t*) pThis, "is now running in disk assisted mode, disk queue 0x%lx\n", + DBGOPRINT((obj_t*) pThis, "DA queue initialized, disk queue 0x%lx\n", qqueueGetID(pThis->pqDA)); finalize_it: @@ -338,7 +340,7 @@ finalize_it: if(pThis->pqDA != NULL) { qqueueDestruct(&pThis->pqDA); } - dbgoprint((obj_t*) pThis, "error %d creating disk queue - giving up.\n", iRet); + DBGOPRINT((obj_t*) pThis, "error %d creating disk queue - giving up.\n", iRet); pThis->bIsDA = 0; } @@ -352,8 +354,8 @@ finalize_it: * If this function fails (should not happen), DA mode is not turned on. * rgerhards, 2008-01-16 */ -static inline rsRetVal -qqueueInitDA(qqueue_t *pThis, int bEnqOnly, int bLockMutex) +static rsRetVal +InitDA(qqueue_t *pThis, int bLockMutex) { DEFiRet; DEFVARS_mutexProtection; @@ -366,82 +368,30 @@ qqueueInitDA(qqueue_t *pThis, int bEnqOnly, int bLockMutex) * is intentional. We assume that when we need it once, we may also need it on another * occasion. Ressources used are quite minimal when no worker is running. * rgerhards, 2008-01-24 + * NOTE: this is the DA worker *pool*, not the DA queue! */ - if(pThis->pWtpDA == NULL) { - lenBuf = snprintf((char*)pszBuf, sizeof(pszBuf), "%s:DA", obj.GetName((obj_t*) pThis)); - CHKiRet(wtpConstruct (&pThis->pWtpDA)); - CHKiRet(wtpSetDbgHdr (pThis->pWtpDA, pszBuf, lenBuf)); - CHKiRet(wtpSetpfChkStopWrkr (pThis->pWtpDA, (rsRetVal (*)(void *pUsr, int)) qqueueChkStopWrkrDA)); - CHKiRet(wtpSetpfIsIdle (pThis->pWtpDA, (rsRetVal (*)(void *pUsr, int)) qqueueIsIdleDA)); - CHKiRet(wtpSetpfDoWork (pThis->pWtpDA, (rsRetVal (*)(void *pUsr, void *pWti, int)) qqueueConsumerDA)); - CHKiRet(wtpSetpfOnWorkerCancel (pThis->pWtpDA, (rsRetVal (*)(void *pUsr, void*pWti)) qqueueConsumerCancelCleanup)); - CHKiRet(wtpSetpfOnWorkerStartup (pThis->pWtpDA, (rsRetVal (*)(void *pUsr)) qqueueStartDA)); - CHKiRet(wtpSetpfOnWorkerShutdown(pThis->pWtpDA, (rsRetVal (*)(void *pUsr)) qqueueTurnOffDAMode)); - CHKiRet(wtpSetpmutUsr (pThis->pWtpDA, pThis->mut)); - CHKiRet(wtpSetpcondBusy (pThis->pWtpDA, &pThis->notEmpty)); - CHKiRet(wtpSetiNumWorkerThreads (pThis->pWtpDA, 1)); - CHKiRet(wtpSettoWrkShutdown (pThis->pWtpDA, pThis->toWrkShutdown)); - CHKiRet(wtpSetpUsr (pThis->pWtpDA, pThis)); - CHKiRet(wtpConstructFinalize (pThis->pWtpDA)); - } + lenBuf = snprintf((char*)pszBuf, sizeof(pszBuf), "%s:DAwpool", obj.GetName((obj_t*) pThis)); + CHKiRet(wtpConstruct (&pThis->pWtpDA)); + CHKiRet(wtpSetDbgHdr (pThis->pWtpDA, pszBuf, lenBuf)); + CHKiRet(wtpSetpfChkStopWrkr (pThis->pWtpDA, (rsRetVal (*)(void *pUsr, int)) qqueueChkStopWrkrDA)); + CHKiRet(wtpSetpfGetDeqBatchSize (pThis->pWtpDA, (rsRetVal (*)(void *pUsr, int*)) GetDeqBatchSize)); + CHKiRet(wtpSetpfDoWork (pThis->pWtpDA, (rsRetVal (*)(void *pUsr, void *pWti)) ConsumerDA)); + CHKiRet(wtpSetpfObjProcessed (pThis->pWtpDA, (rsRetVal (*)(void *pUsr, wti_t *pWti)) batchProcessed)); + CHKiRet(wtpSetpmutUsr (pThis->pWtpDA, pThis->mut)); + CHKiRet(wtpSetpcondBusy (pThis->pWtpDA, &pThis->notEmpty)); + CHKiRet(wtpSetiNumWorkerThreads (pThis->pWtpDA, 1)); + CHKiRet(wtpSettoWrkShutdown (pThis->pWtpDA, pThis->toWrkShutdown)); + CHKiRet(wtpSetpUsr (pThis->pWtpDA, pThis)); + CHKiRet(wtpConstructFinalize (pThis->pWtpDA)); /* if we reach this point, we have a "good" DA worker pool */ - /* indicate we now run in DA mode - this is reset by the DA worker if it fails */ - pThis->bRunsDA = 1; - pThis->bDAEnqOnly = bEnqOnly; - - /* now we must now adivse the wtp that we need one worker. If none is yet active, - * that will also start one up. If we forgot that step, everything would be stalled - * until the next enqueue request. - */ - wtpAdviseMaxWorkers(pThis->pWtpDA, 1); /* DA queues alsways have just one worker max */ - -finalize_it: - END_MTX_PROTECTED_OPERATIONS(pThis->mut); - RETiRet; -} - - -/* check if we need to start disk assisted mode and send some signals to - * keep it running if we are already in it. It also checks if DA mode is - * partially initialized, in which case it waits for initialization to - * complete. - * rgerhards, 2008-01-14 - */ -static inline rsRetVal -qqueueChkStrtDA(qqueue_t *pThis) -{ - DEFiRet; - - ISOBJ_TYPE_assert(pThis, qqueue); - - /* if we do not hit the high water mark, we have nothing to do */ - if(qqueueGetOverallQueueSize(pThis) != pThis->iHighWtrMrk) - ABORT_FINALIZE(RS_RET_OK); - - if(pThis->bRunsDA) { - /* then we need to signal that we are at the high water mark again. If that happens - * on our way down the queue, that doesn't matter, because then nobody is waiting - * on the condition variable. - * (Remember that a DA queue stops draining the queue once it has reached the low - * water mark and restarts it when the high water mark is reached again - this is - * what this code here is responsible for. Please note that all workers may have been - * terminated due to the inactivity timeout, thus we need to advise the pool that - * we need at least one). - */ - dbgoprint((obj_t*) pThis, "%d entries - passed high water mark in DA mode, send notify\n", - qqueueGetOverallQueueSize(pThis)); - qqueueAdviseMaxWorkers(pThis); - } else { - /* this is the case when we are currently not running in DA mode. So it is time - * to turn it back on. - */ - dbgoprint((obj_t*) pThis, "%d entries - passed high water mark for disk-assisted mode, initiating...\n", - qqueueGetOverallQueueSize(pThis)); - qqueueInitDA(pThis, QUEUE_MODE_ENQDEQ, MUTEX_ALREADY_LOCKED); /* initiate DA mode */ + /* now construct the actual queue (if it does not already exist) */ + if(pThis->pqDA == NULL) { + CHKiRet(StartDA(pThis)); } finalize_it: + END_MTX_PROTECTED_OPERATIONS(pThis->mut); RETiRet; } @@ -465,10 +415,11 @@ static rsRetVal qConstructFixedArray(qqueue_t *pThis) if(pThis->iMaxQueueSize == 0) ABORT_FINALIZE(RS_RET_QSIZE_ZERO); - if((pThis->tVars.farray.pBuf = malloc(sizeof(void *) * pThis->iMaxQueueSize)) == NULL) { + if((pThis->tVars.farray.pBuf = MALLOC(sizeof(void *) * pThis->iMaxQueueSize)) == NULL) { ABORT_FINALIZE(RS_RET_OUT_OF_MEMORY); } + pThis->tVars.farray.deqhead = 0; pThis->tVars.farray.head = 0; pThis->tVars.farray.tail = 0; @@ -486,9 +437,7 @@ static rsRetVal qDestructFixedArray(qqueue_t *pThis) ASSERT(pThis != NULL); queueDrain(pThis); /* discard any remaining queue entries */ - - if(pThis->tVars.farray.pBuf != NULL) - free(pThis->tVars.farray.pBuf); + free(pThis->tVars.farray.pBuf); RETiRet; } @@ -507,76 +456,37 @@ static rsRetVal qAddFixedArray(qqueue_t *pThis, void* in) RETiRet; } -static rsRetVal qDelFixedArray(qqueue_t *pThis, void **out) + +static rsRetVal qDeqFixedArray(qqueue_t *pThis, void **out) { DEFiRet; ASSERT(pThis != NULL); - *out = (void*) pThis->tVars.farray.pBuf[pThis->tVars.farray.head]; + *out = (void*) pThis->tVars.farray.pBuf[pThis->tVars.farray.deqhead]; - pThis->tVars.farray.head++; - if (pThis->tVars.farray.head == pThis->iMaxQueueSize) - pThis->tVars.farray.head = 0; + pThis->tVars.farray.deqhead++; + if (pThis->tVars.farray.deqhead == pThis->iMaxQueueSize) + pThis->tVars.farray.deqhead = 0; RETiRet; } -/* -------------------- linked list -------------------- */ - -/* first some generic functions which are also used for the unget linked list */ - -static inline rsRetVal qqueueAddLinkedList(qLinkedList_t **ppRoot, qLinkedList_t **ppLast, void* pUsr) +static rsRetVal qDelFixedArray(qqueue_t *pThis) { DEFiRet; - qLinkedList_t *pEntry; - - ASSERT(ppRoot != NULL); - ASSERT(ppLast != NULL); - - if((pEntry = (qLinkedList_t*) malloc(sizeof(qLinkedList_t))) == NULL) { - ABORT_FINALIZE(RS_RET_OUT_OF_MEMORY); - } - pEntry->pNext = NULL; - pEntry->pUsr = pUsr; + ASSERT(pThis != NULL); - if(*ppRoot == NULL) { - *ppRoot = *ppLast = pEntry; - } else { - (*ppLast)->pNext = pEntry; - *ppLast = pEntry; - } + pThis->tVars.farray.head++; + if (pThis->tVars.farray.head == pThis->iMaxQueueSize) + pThis->tVars.farray.head = 0; -finalize_it: RETiRet; } -static inline rsRetVal qqueueDelLinkedList(qLinkedList_t **ppRoot, qLinkedList_t **ppLast, obj_t **ppUsr) -{ - DEFiRet; - qLinkedList_t *pEntry; - - ASSERT(ppRoot != NULL); - ASSERT(ppLast != NULL); - ASSERT(ppUsr != NULL); - ASSERT(*ppRoot != NULL); - - pEntry = *ppRoot; - *ppUsr = pEntry->pUsr; - - if(*ppRoot == *ppLast) { - *ppRoot = NULL; - *ppLast = NULL; - } else { - *ppRoot = pEntry->pNext; - } - free(pEntry); - - RETiRet; -} -/* end generic functions which are also used for the unget linked list */ +/* -------------------- linked list -------------------- */ static rsRetVal qConstructLinkedList(qqueue_t *pThis) @@ -585,8 +495,9 @@ static rsRetVal qConstructLinkedList(qqueue_t *pThis) ASSERT(pThis != NULL); - pThis->tVars.linklist.pRoot = 0; - pThis->tVars.linklist.pLast = 0; + pThis->tVars.linklist.pDeqRoot = NULL; + pThis->tVars.linklist.pDelRoot = NULL; + pThis->tVars.linklist.pLast = NULL; qqueueChkIsDA(pThis); @@ -609,54 +520,59 @@ static rsRetVal qDestructLinkedList(qqueue_t __attribute__((unused)) *pThis) static rsRetVal qAddLinkedList(qqueue_t *pThis, void* pUsr) { - DEFiRet; - - iRet = qqueueAddLinkedList(&pThis->tVars.linklist.pRoot, &pThis->tVars.linklist.pLast, pUsr); -#if 0 qLinkedList_t *pEntry; + DEFiRet; - ASSERT(pThis != NULL); - if((pEntry = (qLinkedList_t*) malloc(sizeof(qLinkedList_t))) == NULL) { - ABORT_FINALIZE(RS_RET_OUT_OF_MEMORY); - } + CHKmalloc((pEntry = (qLinkedList_t*) MALLOC(sizeof(qLinkedList_t)))); pEntry->pNext = NULL; pEntry->pUsr = pUsr; - if(pThis->tVars.linklist.pRoot == NULL) { - pThis->tVars.linklist.pRoot = pThis->tVars.linklist.pLast = pEntry; + if(pThis->tVars.linklist.pDelRoot == NULL) { + pThis->tVars.linklist.pDelRoot = pThis->tVars.linklist.pDeqRoot = pThis->tVars.linklist.pLast = pEntry; } else { pThis->tVars.linklist.pLast->pNext = pEntry; pThis->tVars.linklist.pLast = pEntry; } + if(pThis->tVars.linklist.pDeqRoot == NULL) { + pThis->tVars.linklist.pDeqRoot = pEntry; + } + finalize_it: -#endif RETiRet; } -static rsRetVal qDelLinkedList(qqueue_t *pThis, obj_t **ppUsr) + +static rsRetVal qDeqLinkedList(qqueue_t *pThis, obj_t **ppUsr) { - DEFiRet; - iRet = qqueueDelLinkedList(&pThis->tVars.linklist.pRoot, &pThis->tVars.linklist.pLast, ppUsr); -#if 0 qLinkedList_t *pEntry; + DEFiRet; - ASSERT(pThis != NULL); - ASSERT(pThis->tVars.linklist.pRoot != NULL); - - pEntry = pThis->tVars.linklist.pRoot; + pEntry = pThis->tVars.linklist.pDeqRoot; + ISOBJ_TYPE_assert(pEntry->pUsr, msg); *ppUsr = pEntry->pUsr; + pThis->tVars.linklist.pDeqRoot = pEntry->pNext; + + RETiRet; +} + - if(pThis->tVars.linklist.pRoot == pThis->tVars.linklist.pLast) { - pThis->tVars.linklist.pRoot = NULL; - pThis->tVars.linklist.pLast = NULL; +static rsRetVal qDelLinkedList(qqueue_t *pThis) +{ + qLinkedList_t *pEntry; + DEFiRet; + + pEntry = pThis->tVars.linklist.pDelRoot; + + if(pThis->tVars.linklist.pDelRoot == pThis->tVars.linklist.pLast) { + pThis->tVars.linklist.pDelRoot = pThis->tVars.linklist.pDeqRoot = pThis->tVars.linklist.pLast = NULL; } else { - pThis->tVars.linklist.pRoot = pEntry->pNext; + pThis->tVars.linklist.pDelRoot = pEntry->pNext; } + free(pEntry); -#endif RETiRet; } @@ -676,44 +592,6 @@ finalize_it: } -/* This method checks if we have a QIF file for the current queue (no matter of - * queue mode). Returns RS_RET_OK if we have a QIF file or an error status otherwise. - * rgerhards, 2008-01-15 - */ -static rsRetVal -qqueueHaveQIF(qqueue_t *pThis) -{ - DEFiRet; - uchar pszQIFNam[MAXFNAME]; - size_t lenQIFNam; - struct stat stat_buf; - - ISOBJ_TYPE_assert(pThis, qqueue); - - if(pThis->pszFilePrefix == NULL) - ABORT_FINALIZE(RS_RET_NO_FILEPREFIX); - - /* Construct file name */ - lenQIFNam = snprintf((char*)pszQIFNam, sizeof(pszQIFNam) / sizeof(uchar), "%s/%s.qi", - (char*) glbl.GetWorkDir(), (char*)pThis->pszFilePrefix); - - /* check if the file exists */ - if(stat((char*) pszQIFNam, &stat_buf) == -1) { - if(errno == ENOENT) { - dbgoprint((obj_t*) pThis, "no .qi file found\n"); - ABORT_FINALIZE(RS_RET_FILE_NOT_FOUND); - } else { - dbgoprint((obj_t*) pThis, "error %d trying to access .qi file\n", errno); - ABORT_FINALIZE(RS_RET_IO_ERROR); - } - } - /* If we reach this point, we have a .qi file */ - -finalize_it: - RETiRet; -} - - /* The method loads the persistent queue information. * rgerhards, 2008-01-11 */ @@ -725,8 +603,6 @@ qqueueTryLoadPersistedInfo(qqueue_t *pThis) uchar pszQIFNam[MAXFNAME]; size_t lenQIFNam; struct stat stat_buf; - int iUngottenObjs; - obj_t *pUsr; ISOBJ_TYPE_assert(pThis, qqueue); @@ -737,10 +613,10 @@ qqueueTryLoadPersistedInfo(qqueue_t *pThis) /* check if the file exists */ if(stat((char*) pszQIFNam, &stat_buf) == -1) { if(errno == ENOENT) { - dbgoprint((obj_t*) pThis, "clean startup, no .qi file found\n"); + DBGOPRINT((obj_t*) pThis, "clean startup, no .qi file found\n"); ABORT_FINALIZE(RS_RET_FILE_NOT_FOUND); } else { - dbgoprint((obj_t*) pThis, "error %d trying to access .qi file\n", errno); + DBGOPRINT((obj_t*) pThis, "error %d trying to access .qi file\n", errno); ABORT_FINALIZE(RS_RET_IO_ERROR); } } @@ -756,25 +632,22 @@ qqueueTryLoadPersistedInfo(qqueue_t *pThis) /* first, we try to read the property bag for ourselfs */ CHKiRet(obj.DeserializePropBag((obj_t*) pThis, psQIF)); - /* then the ungotten object queue */ - iUngottenObjs = pThis->iUngottenObjs; - pThis->iUngottenObjs = 0; /* will be incremented when we add objects! */ - - while(iUngottenObjs > 0) { - /* fill the queue from disk */ - CHKiRet(obj.Deserialize((void*) &pUsr, (uchar*)"msg", psQIF, NULL, NULL)); - qqueueUngetObj(pThis, pUsr, MUTEX_ALREADY_LOCKED); - --iUngottenObjs; /* one less */ - } - - /* and now the stream objects (some order as when persisted!) */ + /* then the stream objects (same order as when persisted!) */ CHKiRet(obj.Deserialize(&pThis->tVars.disk.pWrite, (uchar*) "strm", psQIF, (rsRetVal(*)(obj_t*,void*))qqueueLoadPersStrmInfoFixup, pThis)); - CHKiRet(obj.Deserialize(&pThis->tVars.disk.pRead, (uchar*) "strm", psQIF, + CHKiRet(obj.Deserialize(&pThis->tVars.disk.pReadDel, (uchar*) "strm", psQIF, (rsRetVal(*)(obj_t*,void*))qqueueLoadPersStrmInfoFixup, pThis)); + /* create a duplicate for the read "pointer". + */ + + CHKiRet(strm.Dup(pThis->tVars.disk.pReadDel, &pThis->tVars.disk.pReadDeq)); + CHKiRet(strm.SetbDeleteOnClose(pThis->tVars.disk.pReadDeq, 0)); /* deq must NOT delete the files! */ + CHKiRet(strm.ConstructFinalize(pThis->tVars.disk.pReadDeq)); + CHKiRet(strm.SeekCurrOffs(pThis->tVars.disk.pWrite)); - CHKiRet(strm.SeekCurrOffs(pThis->tVars.disk.pRead)); + CHKiRet(strm.SeekCurrOffs(pThis->tVars.disk.pReadDel)); + CHKiRet(strm.SeekCurrOffs(pThis->tVars.disk.pReadDeq)); /* OK, we could successfully read the file, so we now can request that it be * deleted when we are done with the persisted information. @@ -786,7 +659,7 @@ finalize_it: strm.Destruct(&psQIF); if(iRet != RS_RET_OK) { - dbgoprint((obj_t*) pThis, "error %d reading .qi file - can not read persisted info (if any)\n", + DBGOPRINT((obj_t*) pThis, "error %d reading .qi file - can not read persisted info (if any)\n", iRet); } @@ -826,18 +699,26 @@ static rsRetVal qConstructDisk(qqueue_t *pThis) CHKiRet(strm.SetsType(pThis->tVars.disk.pWrite, STREAMTYPE_FILE_CIRCULAR)); CHKiRet(strm.ConstructFinalize(pThis->tVars.disk.pWrite)); - CHKiRet(strm.Construct(&pThis->tVars.disk.pRead)); - CHKiRet(strm.SetbSync(pThis->tVars.disk.pRead, pThis->bSyncQueueFiles)); - CHKiRet(strm.SetbDeleteOnClose(pThis->tVars.disk.pRead, 1)); - CHKiRet(strm.SetDir(pThis->tVars.disk.pRead, glbl.GetWorkDir(), strlen((char*)glbl.GetWorkDir()))); - CHKiRet(strm.SetiMaxFiles(pThis->tVars.disk.pRead, 10000000)); - CHKiRet(strm.SettOperationsMode(pThis->tVars.disk.pRead, STREAMMODE_READ)); - CHKiRet(strm.SetsType(pThis->tVars.disk.pRead, STREAMTYPE_FILE_CIRCULAR)); - CHKiRet(strm.ConstructFinalize(pThis->tVars.disk.pRead)); + CHKiRet(strm.Construct(&pThis->tVars.disk.pReadDeq)); + CHKiRet(strm.SetbDeleteOnClose(pThis->tVars.disk.pReadDeq, 0)); + CHKiRet(strm.SetDir(pThis->tVars.disk.pReadDeq, glbl.GetWorkDir(), strlen((char*)glbl.GetWorkDir()))); + CHKiRet(strm.SetiMaxFiles(pThis->tVars.disk.pReadDeq, 10000000)); + CHKiRet(strm.SettOperationsMode(pThis->tVars.disk.pReadDeq, STREAMMODE_READ)); + CHKiRet(strm.SetsType(pThis->tVars.disk.pReadDeq, STREAMTYPE_FILE_CIRCULAR)); + CHKiRet(strm.ConstructFinalize(pThis->tVars.disk.pReadDeq)); + CHKiRet(strm.Construct(&pThis->tVars.disk.pReadDel)); + CHKiRet(strm.SetbSync(pThis->tVars.disk.pReadDel, pThis->bSyncQueueFiles)); + CHKiRet(strm.SetbDeleteOnClose(pThis->tVars.disk.pReadDel, 1)); + CHKiRet(strm.SetDir(pThis->tVars.disk.pReadDel, glbl.GetWorkDir(), strlen((char*)glbl.GetWorkDir()))); + CHKiRet(strm.SetiMaxFiles(pThis->tVars.disk.pReadDel, 10000000)); + CHKiRet(strm.SettOperationsMode(pThis->tVars.disk.pReadDel, STREAMMODE_READ)); + CHKiRet(strm.SetsType(pThis->tVars.disk.pReadDel, STREAMTYPE_FILE_CIRCULAR)); + CHKiRet(strm.ConstructFinalize(pThis->tVars.disk.pReadDel)); - CHKiRet(strm.SetFName(pThis->tVars.disk.pWrite, pThis->pszFilePrefix, pThis->lenFilePrefix)); - CHKiRet(strm.SetFName(pThis->tVars.disk.pRead, pThis->pszFilePrefix, pThis->lenFilePrefix)); + CHKiRet(strm.SetFName(pThis->tVars.disk.pWrite, pThis->pszFilePrefix, pThis->lenFilePrefix)); + CHKiRet(strm.SetFName(pThis->tVars.disk.pReadDeq, pThis->pszFilePrefix, pThis->lenFilePrefix)); + CHKiRet(strm.SetFName(pThis->tVars.disk.pReadDel, pThis->pszFilePrefix, pThis->lenFilePrefix)); } /* now we set (and overwrite in case of a persisted restart) some parameters which @@ -846,7 +727,8 @@ static rsRetVal qConstructDisk(qqueue_t *pThis) * ability to read existing queue files. -- rgerhards, 2008-01-12 */ CHKiRet(strm.SetiMaxFileSize(pThis->tVars.disk.pWrite, pThis->iMaxFileSize)); - CHKiRet(strm.SetiMaxFileSize(pThis->tVars.disk.pRead, pThis->iMaxFileSize)); + CHKiRet(strm.SetiMaxFileSize(pThis->tVars.disk.pReadDeq, pThis->iMaxFileSize)); + CHKiRet(strm.SetiMaxFileSize(pThis->tVars.disk.pReadDel, pThis->iMaxFileSize)); finalize_it: RETiRet; @@ -858,11 +740,13 @@ static rsRetVal qDestructDisk(qqueue_t *pThis) DEFiRet; ASSERT(pThis != NULL); - - if (pThis->tVars.disk.pWrite != NULL) + + if(pThis->tVars.disk.pWrite != NULL) strm.Destruct(&pThis->tVars.disk.pWrite); - if (pThis->tVars.disk.pRead != NULL) - strm.Destruct(&pThis->tVars.disk.pRead); + if(pThis->tVars.disk.pReadDeq != NULL) + strm.Destruct(&pThis->tVars.disk.pReadDeq); + if(pThis->tVars.disk.pReadDel != NULL) + strm.Destruct(&pThis->tVars.disk.pReadDel); RETiRet; } @@ -887,23 +771,37 @@ static rsRetVal qAddDisk(qqueue_t *pThis, void* pUsr) */ objDestruct(pUsr); - dbgoprint((obj_t*) pThis, "write wrote %lld octets to disk, queue disk size now %lld octets\n", - nWriteCount, pThis->tVars.disk.sizeOnDisk); + DBGOPRINT((obj_t*) pThis, "write wrote %lld octets to disk, queue disk size now %lld octets, EnqOnly:%d\n", + nWriteCount, pThis->tVars.disk.sizeOnDisk, pThis->bEnqOnly); finalize_it: RETiRet; } -static rsRetVal qDelDisk(qqueue_t *pThis, void **ppUsr) + +static rsRetVal qDeqDisk(qqueue_t *pThis, void **ppUsr) { DEFiRet; + CHKiRet(obj.Deserialize(ppUsr, (uchar*) "msg", pThis->tVars.disk.pReadDeq, NULL, NULL)); + +finalize_it: + RETiRet; +} + + +static rsRetVal qDelDisk(qqueue_t *pThis) +{ + obj_t *pDummyObj; /* we need to deserialize it... */ + DEFiRet; + int64 offsIn; int64 offsOut; - CHKiRet(strm.GetCurrOffset(pThis->tVars.disk.pRead, &offsIn)); - CHKiRet(obj.Deserialize(ppUsr, (uchar*) "msg", pThis->tVars.disk.pRead, NULL, NULL)); - CHKiRet(strm.GetCurrOffset(pThis->tVars.disk.pRead, &offsOut)); + CHKiRet(strm.GetCurrOffset(pThis->tVars.disk.pReadDel, &offsIn)); + CHKiRet(obj.Deserialize(&pDummyObj, (uchar*) "msg", pThis->tVars.disk.pReadDel, NULL, NULL)); + objDestruct(pDummyObj); + CHKiRet(strm.GetCurrOffset(pThis->tVars.disk.pReadDel, &offsOut)); /* This time it is a bit tricky: we free disk space only upon file deletion. So we need * to keep track of what we have read until we get an out-offset that is lower than the @@ -915,7 +813,7 @@ static rsRetVal qDelDisk(qqueue_t *pThis, void **ppUsr) } else { pThis->tVars.disk.sizeOnDisk -= pThis->tVars.disk.bytesRead; pThis->tVars.disk.bytesRead = offsOut; - dbgoprint((obj_t*) pThis, "a file has been deleted, now %lld octets disk space used\n", pThis->tVars.disk.sizeOnDisk); + DBGOPRINT((obj_t*) pThis, "a file has been deleted, now %lld octets disk space used\n", pThis->tVars.disk.sizeOnDisk); /* awake possibly waiting enq process */ pthread_cond_signal(&pThis->notFull); /* we hold the mutex while we are in here! */ } @@ -924,6 +822,7 @@ finalize_it: RETiRet; } + /* -------------------- direct (no queueing) -------------------- */ static rsRetVal qConstructDirect(qqueue_t __attribute__((unused)) *pThis) { @@ -938,8 +837,11 @@ static rsRetVal qDestructDirect(qqueue_t __attribute__((unused)) *pThis) static rsRetVal qAddDirect(qqueue_t *pThis, void* pUsr) { + batch_t singleBatch; + batch_obj_t batchObj; DEFiRet; + //TODO: init batchObj (states _OK and new fields -- CHECK) ASSERT(pThis != NULL); /* calling the consumer is quite different here than it is from a worker thread */ @@ -947,70 +849,52 @@ static rsRetVal qAddDirect(qqueue_t *pThis, void* pUsr) * mode the consumer probably has a lot to convey (which get's lost in the other modes * because they are asynchronous. But direct mode is deliberately synchronous. * rgerhards, 2008-02-12 + * We use our knowledge about the batch_t structure below, but without that, we + * pay a too-large performance toll... -- rgerhards, 2009-04-22 */ - iRet = pThis->pConsumer(pThis->pUsr, pUsr); + memset(&batchObj, 0, sizeof(batch_obj_t)); + memset(&singleBatch, 0, sizeof(batch_t)); + batchObj.state = BATCH_STATE_RDY; + batchObj.pUsrp = (obj_t*) pUsr; + batchObj.bFilterOK = 1; + singleBatch.nElem = 1; /* there always is only one in direct mode */ + singleBatch.pElem = &batchObj; + iRet = pThis->pConsumer(pThis->pUsr, &singleBatch, &pThis->bShutdownImmediate); + objDestruct(pUsr); RETiRet; } -static rsRetVal qDelDirect(qqueue_t __attribute__((unused)) *pThis, __attribute__((unused)) void **out) -{ - return RS_RET_OK; -} - - -/* --------------- end type-specific handlers -------------------- */ - - -/* unget a user pointer that has been dequeued. This functionality is especially important - * for consumer cancel cleanup handlers. To support it, a short list of ungotten user pointers - * is maintened in memory. - * rgerhards, 2008-01-20 +/* "enqueue" a batch in direct mode. This is a shortcut which saves all the overhead + * otherwise incured. -- rgerhards, ~2010-06-23 */ -static rsRetVal -qqueueUngetObj(qqueue_t *pThis, obj_t *pUsr, int bLockMutex) +rsRetVal qqueueEnqObjDirectBatch(qqueue_t *pThis, batch_t *pBatch) { DEFiRet; - DEFVARS_mutexProtection; - ISOBJ_TYPE_assert(pThis, qqueue); - ISOBJ_assert(pUsr); /* TODO: we aborted right at this place at least 3 times -- race? 2008-02-28, -03-10, -03-15 - The second time I noticed it the queue was in destruction with NO worker threads - running. The pUsr ptr was totally off and provided no clue what it may be pointing - at (except that it looked like the static data pool). Both times, the abort happend - inside an action queue */ + ASSERT(pThis != NULL); - dbgoprint((obj_t*) pThis, "ungetting user object %s\n", obj.GetName(pUsr)); - BEGIN_MTX_PROTECTED_OPERATIONS(pThis->mut, bLockMutex); - iRet = qqueueAddLinkedList(&pThis->pUngetRoot, &pThis->pUngetLast, pUsr); - ++pThis->iUngottenObjs; /* indicate one more */ - END_MTX_PROTECTED_OPERATIONS(pThis->mut); + /* calling the consumer is quite different here than it is from a worker thread */ + /* we need to provide the consumer's return value back to the caller because in direct + * mode the consumer probably has a lot to convey (which get's lost in the other modes + * because they are asynchronous. But direct mode is deliberately synchronous. + * rgerhards, 2008-02-12 + * We use our knowledge about the batch_t structure below, but without that, we + * pay a too-large performance toll... -- rgerhards, 2009-04-22 + */ + iRet = pThis->pConsumer(pThis->pUsr, pBatch, &pThis->bShutdownImmediate); RETiRet; } -/* dequeues a user pointer from the ungotten queue. Pointers from there should always be - * dequeued first. - * - * This function must only be called when the mutex is locked! - * - * rgerhards, 2008-01-29 - */ -static rsRetVal -qqueueGetUngottenObj(qqueue_t *pThis, obj_t **ppUsr) +static rsRetVal qDelDirect(qqueue_t __attribute__((unused)) *pThis) { - DEFiRet; - - ISOBJ_TYPE_assert(pThis, qqueue); - ASSERT(ppUsr != NULL); + return RS_RET_OK; +} - iRet = qqueueDelLinkedList(&pThis->pUngetRoot, &pThis->pUngetLast, ppUsr); - --pThis->iUngottenObjs; /* indicate one less */ - dbgoprint((obj_t*) pThis, "dequeued ungotten user object %s\n", obj.GetName(*ppUsr)); - RETiRet; -} +/* --------------- end type-specific handlers -------------------- */ /* generic code to add a queue entry @@ -1029,7 +913,8 @@ qqueueAdd(qqueue_t *pThis, void *pUsr) if(pThis->qType != QUEUETYPE_DIRECT) { ATOMIC_INC(&pThis->iQueueSize, &pThis->mutQueueSize); - dbgoprint((obj_t*) pThis, "entry added, size now %d entries\n", pThis->iQueueSize); + DBGOPRINT((obj_t*) pThis, "entry added, size now log %d, phys %d entries\n", + getLogicalQueueSize(pThis), getPhysicalQueueSize(pThis)); } finalize_it: @@ -1037,12 +922,10 @@ finalize_it: } -/* generic code to remove a queue entry - * rgerhards, 2008-01-29: we must first see if there is any object in the - * ungotten list and, if so, dequeue it first. +/* generic code to dequeue a queue entry */ static rsRetVal -qqueueDel(qqueue_t *pThis, void *pUsr) +qqueueDeq(qqueue_t *pThis, void **ppUsr) { DEFiRet; @@ -1053,224 +936,249 @@ qqueueDel(qqueue_t *pThis, void *pUsr) * If we decrement, however, we may lose a message. But that is better than * losing the whole process because it loops... -- rgerhards, 2008-01-03 */ - if(pThis->iUngottenObjs > 0) { - iRet = qqueueGetUngottenObj(pThis, (obj_t**) pUsr); - } else { - iRet = pThis->qDel(pThis, pUsr); - ATOMIC_DEC(&pThis->iQueueSize, &pThis->mutQueueSize); - } + iRet = pThis->qDeq(pThis, ppUsr); + ATOMIC_INC(&pThis->nLogDeq, &pThis->mutLogDeq); - dbgoprint((obj_t*) pThis, "entry deleted, state %d, size now %d entries\n", - iRet, pThis->iQueueSize); +// DBGOPRINT((obj_t*) pThis, "entry deleted, size now log %d, phys %d entries\n", +// getLogicalQueueSize(pThis), getPhysicalQueueSize(pThis)); RETiRet; } -/* This function shuts down all worker threads and waits until they - * have terminated. If they timeout, they are cancelled. Parameters have been set - * before this function is called so that DA queues will be fully persisted to - * disk (if configured to do so). - * rgerhards, 2008-01-24 - * Please note that this function shuts down BOTH the parent AND the child queue - * in DA case. This is necessary because their timeouts are tightly coupled. Most - * importantly, the timeouts would be applied twice (or logic be extremely - * complex) if each would have its own shutdown. The function does not self check - * this condition - the caller must make sure it is not called with a parent. +/* Try to shut down regular and DA queue workers, within the queue timeout + * period. That means processing continues as usual. This is the expected + * usual case, where during shutdown those messages remaining are being + * processed. At this point, it is acceptable that the queue can not be + * fully depleted, that case is handled in the next step. During this phase, + * we first shut down the main queue DA worker to prevent new data to arrive + * at the DA queue, and then we ask the regular workers of both the Regular + * and DA queue to try complete processing. + * rgerhards, 2009-10-14 */ -static rsRetVal qqueueShutdownWorkers(qqueue_t *pThis) +static inline rsRetVal +tryShutdownWorkersWithinQueueTimeout(qqueue_t *pThis) { - DEFiRet; - DEFVARS_mutexProtection; struct timespec tTimeout; rsRetVal iRetLocal; + DEFiRet; ISOBJ_TYPE_assert(pThis, qqueue); ASSERT(pThis->pqParent == NULL); /* detect invalid calling sequence */ - dbgoprint((obj_t*) pThis, "initiating worker thread shutdown sequence\n"); + if(pThis->bIsDA) { + /* We need to lock the mutex, as otherwise we may have a race that prevents + * us from awaking the DA worker. */ + d_pthread_mutex_lock(pThis->mut); - /* we reduce the low water mark in any case. This is not absolutely necessary, but - * it is useful because we enable DA mode at several spots below and so we do not need - * to think about the low water mark each time. - */ - pThis->iHighWtrMrk = 1; /* if we do not do this, the DA queue will not stop! */ - pThis->iLowWtrMrk = 0; + /* tell regular queue DA worker to stop shuffling messages to DA queue... */ + DBGOPRINT((obj_t*) pThis, "setting EnqOnly mode for DA worker\n"); + pThis->pqDA->bEnqOnly = 1; + wtpSetState(pThis->pWtpDA, wtpState_SHUTDOWN_IMMEDIATE); + wtpAdviseMaxWorkers(pThis->pWtpDA, 1); + DBGOPRINT((obj_t*) pThis, "awoke DA worker, told it to shut down.\n"); - /* first try to shutdown the queue within the regular shutdown period */ - BEGIN_MTX_PROTECTED_OPERATIONS(pThis->mut, LOCK_MUTEX); /* some workers may be running in parallel! */ - if(qqueueGetOverallQueueSize(pThis) > 0) { - if(pThis->bRunsDA) { - /* We may have waited on the low water mark. As it may have changed, we - * see if we reactivate the worker. - */ - wtpAdviseMaxWorkers(pThis->pWtpDA, 1); - } + /* also tell the DA queue worker to shut down, so that it already knows... */ + wtpSetState(pThis->pqDA->pWtpReg, wtpState_SHUTDOWN); + wtpAdviseMaxWorkers(pThis->pqDA->pWtpReg, 1); /* awake its lone worker */ + DBGOPRINT((obj_t*) pThis, "awoke DA queue regular worker, told it to shut down when done.\n"); + + d_pthread_mutex_unlock(pThis->mut); } - END_MTX_PROTECTED_OPERATIONS(pThis->mut); - /* Now wait for the queue's workers to shut down. Note that we run into the code even if we just found - * out there are no active workers - that doesn't matter: the wtp knows about that and so will - * return immediately. - * We do not yet care about the DA worker - that will be handled down later in the process. - * Note that we must not request shutdown right now - that may introduce a race: if the regular queue - * still runs DA assisted and the DA worker gets scheduled first, it will terminate itself (if the DA - * queue happens to be empty at that instant). Then the regular worker enqueues messages, what will lead - * to a restart of the worker. Of course, everything will continue to run, but in a bit sub-optimal way - * (from a performance point of view). So we don't do anything right now. The DA queue will continue to - * process messages and shutdown itself in any case if there is nothing to do. So we don't loose anything - * by not requesting shutdown now. - * rgerhards, 2008-01-25 - */ + /* first calculate absolute timeout - we need the absolute value here, because we need to coordinate * shutdown of both the regular and DA queue on *the same* timeout. */ timeoutComp(&tTimeout, pThis->toQShutdown); - dbgoprint((obj_t*) pThis, "trying shutdown of regular workers\n"); + DBGOPRINT((obj_t*) pThis, "trying shutdown of regular workers\n"); iRetLocal = wtpShutdownAll(pThis->pWtpReg, wtpState_SHUTDOWN, &tTimeout); if(iRetLocal == RS_RET_TIMED_OUT) { - dbgoprint((obj_t*) pThis, "regular shutdown timed out on primary queue (this is OK)\n"); + DBGOPRINT((obj_t*) pThis, "regular shutdown timed out on primary queue (this is OK)\n"); } else { - /* OK, the regular queue is now shut down. So we can now wait for the DA queue (if running DA) */ - dbgoprint((obj_t*) pThis, "regular queue workers shut down.\n"); - BEGIN_MTX_PROTECTED_OPERATIONS(pThis->mut, LOCK_MUTEX); /* some workers may be running in parallel! */ - if(pThis->bRunsDA) { - END_MTX_PROTECTED_OPERATIONS(pThis->mut); - dbgoprint((obj_t*) pThis, "we have a DA queue (0x%lx), requesting its shutdown.\n", - qqueueGetID(pThis->pqDA)); - /* we use the same absolute timeout as above, so we do not use more than the configured - * timeout interval! - */ - dbgoprint((obj_t*) pThis, "trying shutdown of DA workers\n"); - iRetLocal = wtpShutdownAll(pThis->pWtpDA, wtpState_SHUTDOWN, &tTimeout); - if(iRetLocal == RS_RET_TIMED_OUT) { - dbgoprint((obj_t*) pThis, "shutdown timed out on DA queue (this is OK)\n"); - } - } else { - END_MTX_PROTECTED_OPERATIONS(pThis->mut); - } + DBGOPRINT((obj_t*) pThis, "regular queue workers shut down.\n"); } - /* when we reach this point, both queues are either empty or the regular queue shutdown timeout - * has expired. Now we need to check if we are configured to not loose messages. If so, we need - * to persist the queue to disk (this is only possible if the queue is DA-enabled). We must also - * set the primary queue to SHUTDOWN_IMMEDIATE, as it shall now terminate as soon as its consumer - * is done. This is especially important as we otherwise may interfere with queue order while the - * DA consumer is running. -- rgerhards, 2008-01-27 - * Note: there was a note that we should not wait eternally on the DA worker if we run in - * enqueue-only note. I have reviewed the code and think there is no need for this check. Howerver, - * I'd like to keep this note in here should we happen to run into some related trouble. - * rgerhards, 2008-01-28 - */ - wtpSetState(pThis->pWtpReg, wtpState_SHUTDOWN_IMMEDIATE); /* set primary queue to shutdown only */ - - /* at this stage, we need to have the DA worker properly initialized and running (if there is one) */ - if(pThis->bRunsDA) - qqueueWaitDAModeInitialized(pThis); - - BEGIN_MTX_PROTECTED_OPERATIONS(pThis->mut, LOCK_MUTEX); /* some workers may be running in parallel! */ - /* optimize parameters for shutdown of DA-enabled queues */ - if(pThis->bIsDA && qqueueGetOverallQueueSize(pThis) > 0 && pThis->bSaveOnShutdown) { - /* switch to enqueue-only mode so that no more actions happen */ - if(pThis->bRunsDA == 0) { - qqueueInitDA(pThis, QUEUE_MODE_ENQONLY, MUTEX_ALREADY_LOCKED); /* switch to DA mode */ + /* OK, the worker for the regular queue is processed, on the the DA queue regular worker. */ + if(pThis->pqDA != NULL) { + DBGOPRINT((obj_t*) pThis, "we have a DA queue (0x%lx), requesting its shutdown.\n", + qqueueGetID(pThis->pqDA)); + /* we use the same absolute timeout as above, so we do not use more than the configured + * timeout interval! + */ + DBGOPRINT((obj_t*) pThis, "trying shutdown of regular worker of DA queue\n"); + iRetLocal = wtpShutdownAll(pThis->pqDA->pWtpReg, wtpState_SHUTDOWN, &tTimeout); + if(iRetLocal == RS_RET_TIMED_OUT) { + DBGOPRINT((obj_t*) pThis, "shutdown timed out on DA queue worker (this is OK)\n"); } else { - /* TODO: RACE: we may reach this point when the DA worker has been initialized (state 1) - * but is not yet running (state 2). In this case, pThis->pqDA is NULL! rgerhards, 2008-02-27 - */ - qqueueSetEnqOnly(pThis->pqDA, QUEUE_MODE_ENQONLY, MUTEX_ALREADY_LOCKED); /* switch to enqueue-only mode */ - } - END_MTX_PROTECTED_OPERATIONS(pThis->mut); - /* make sure we do not timeout before we are done */ - dbgoprint((obj_t*) pThis, "bSaveOnShutdown configured, eternal timeout set\n"); - timeoutComp(&tTimeout, QUEUE_TIMEOUT_ETERNAL); - /* and run the primary queue's DA worker to drain the queue */ - iRetLocal = wtpShutdownAll(pThis->pWtpDA, wtpState_SHUTDOWN, &tTimeout); - if(iRetLocal != RS_RET_OK) { - dbgoprint((obj_t*) pThis, "unexpected iRet state %d after trying to shut down primary queue in disk save mode, " - "continuing, but results are unpredictable\n", iRetLocal); + DBGOPRINT((obj_t*) pThis, "DA queue worker shut down.\n"); } - } else { - END_MTX_PROTECTED_OPERATIONS(pThis->mut); } - /* now the primary queue is either empty, persisted to disk - or set to loose messages. So we - * can now request immediate shutdown of any remaining workers. Note that if bSaveOnShutdown was set, - * the queue is now empty. If regular workers are still running, and try to pull the next message, - * they will automatically terminate as there no longer is any message left to process. - */ - BEGIN_MTX_PROTECTED_OPERATIONS(pThis->mut, LOCK_MUTEX); /* some workers may be running in parallel! */ - if(qqueueGetOverallQueueSize(pThis) > 0) { - timeoutComp(&tTimeout, pThis->toActShutdown); - if(wtpGetCurNumWrkr(pThis->pWtpReg, LOCK_MUTEX) > 0) { - END_MTX_PROTECTED_OPERATIONS(pThis->mut); - dbgoprint((obj_t*) pThis, "trying immediate shutdown of regular workers\n"); - iRetLocal = wtpShutdownAll(pThis->pWtpReg, wtpState_SHUTDOWN_IMMEDIATE, &tTimeout); - if(iRetLocal == RS_RET_TIMED_OUT) { - dbgoprint((obj_t*) pThis, "immediate shutdown timed out on primary queue (this is acceptable and " - "triggers cancellation)\n"); - } else if(iRetLocal != RS_RET_OK) { - dbgoprint((obj_t*) pThis, "unexpected iRet state %d after trying immediate shutdown of the primary queue " - "in disk save mode. Continuing, but results are unpredictable\n", iRetLocal); - } - /* we need to re-aquire the mutex for the next check in this case! */ - BEGIN_MTX_PROTECTED_OPERATIONS(pThis->mut, LOCK_MUTEX); /* some workers may be running in parallel! */ + RETiRet; +} + + +/* Try to shut down regular and DA queue workers, within the action timeout + * period. This aborts processing, but at the end of the current action, in + * a well-defined manner. During this phase, we terminate all three worker + * pools, including the regular queue DA worker if it not yet has terminated. + * Not finishing processing all messages is OK (and expected) at this stage + * (they may be preserved later, depending * on bSaveOnShutdown setting). + * rgerhards, 2009-10-14 + */ +static rsRetVal +tryShutdownWorkersWithinActionTimeout(qqueue_t *pThis) +{ + struct timespec tTimeout; + rsRetVal iRetLocal; + DEFiRet; + +RUNLOG_STR("trying to shutdown workers within Action Timeout"); + ISOBJ_TYPE_assert(pThis, qqueue); + ASSERT(pThis->pqParent == NULL); /* detect invalid calling sequence */ + + /* instruct workers to finish ASAP, even if still work exists */ + DBGOPRINT((obj_t*) pThis, "setting EnqOnly mode\n"); + pThis->bEnqOnly = 1; + pThis->bShutdownImmediate = 1; + /* now DA queue */ + if(pThis->bIsDA) { + pThis->pqDA->bEnqOnly = 1; + pThis->pqDA->bShutdownImmediate = 1; + } + +// TODO: make sure we have at minimum a 10ms timeout - workers deserve a chance... + /* now give the queue workers a last chance to gracefully shut down (based on action timeout setting) */ + timeoutComp(&tTimeout, pThis->toActShutdown); + DBGOPRINT((obj_t*) pThis, "trying immediate shutdown of regular workers (if any)\n"); + iRetLocal = wtpShutdownAll(pThis->pWtpReg, wtpState_SHUTDOWN_IMMEDIATE, &tTimeout); + if(iRetLocal == RS_RET_TIMED_OUT) { + DBGOPRINT((obj_t*) pThis, "immediate shutdown timed out on primary queue (this is acceptable and " + "triggers cancellation)\n"); + } else if(iRetLocal != RS_RET_OK) { + DBGOPRINT((obj_t*) pThis, "unexpected iRet state %d after trying immediate shutdown of the primary queue " + "in disk save mode. Continuing, but results are unpredictable\n", iRetLocal); + } + + if(pThis->pqDA != NULL) { + /* and now the same for the DA queue */ + DBGOPRINT((obj_t*) pThis, "trying immediate shutdown of DA queue workers\n"); + iRetLocal = wtpShutdownAll(pThis->pqDA->pWtpReg, wtpState_SHUTDOWN_IMMEDIATE, &tTimeout); + if(iRetLocal == RS_RET_TIMED_OUT) { + DBGOPRINT((obj_t*) pThis, "immediate shutdown timed out on DA queue (this is acceptable " + "and triggers cancellation)\n"); + } else if(iRetLocal != RS_RET_OK) { + DBGOPRINT((obj_t*) pThis, "unexpected iRet state %d after trying immediate shutdown of the DA " + "queue in disk save mode. Continuing, but results are unpredictable\n", iRetLocal); } - if(pThis->bRunsDA && wtpGetCurNumWrkr(pThis->pWtpDA, LOCK_MUTEX) > 0) { - /* and now the same for the DA queue */ - END_MTX_PROTECTED_OPERATIONS(pThis->mut); - dbgoprint((obj_t*) pThis, "trying immediate shutdown of DA workers\n"); - iRetLocal = wtpShutdownAll(pThis->pWtpDA, wtpState_SHUTDOWN_IMMEDIATE, &tTimeout); - if(iRetLocal == RS_RET_TIMED_OUT) { - dbgoprint((obj_t*) pThis, "immediate shutdown timed out on DA queue (this is acceptable and " - "triggers cancellation)\n"); - } else if(iRetLocal != RS_RET_OK) { - dbgoprint((obj_t*) pThis, "unexpected iRet state %d after trying immediate shutdown of the DA queue " - "in disk save mode. Continuing, but results are unpredictable\n", iRetLocal); - } + + /* and now we need to terminate the DA worker itself. We always grant it a 100ms timeout, + * which should be sufficient and usually not be required (it is expected to have finished + * long before while we were processing the queue timeout in shutdown phase 1). + * rgerhards, 2009-10-14 + */ + timeoutComp(&tTimeout, 100); + DBGOPRINT((obj_t*) pThis, "trying regular shutdown of main queue DA worker pool\n"); + iRetLocal = wtpShutdownAll(pThis->pWtpDA, wtpState_SHUTDOWN_IMMEDIATE, &tTimeout); + if(iRetLocal == RS_RET_TIMED_OUT) { + DBGOPRINT((obj_t*) pThis, "shutdown timed out on main queue DA worker pool " + "(this is not good, but probably OK)\n"); } else { - END_MTX_PROTECTED_OPERATIONS(pThis->mut); + DBGOPRINT((obj_t*) pThis, "main queue DA worker pool shut down.\n"); } - } else { - END_MTX_PROTECTED_OPERATIONS(pThis->mut); } + RETiRet; +} + + +/* This function cancels all remaining regular workers for both the main and the DA + * queue. + * rgerhards, 2009-05-29 + */ +static rsRetVal +cancelWorkers(qqueue_t *pThis) +{ + rsRetVal iRetLocal; + DEFiRet; + /* Now queue workers should have terminated. If not, we need to cancel them as we have applied * all timeout setting. If any worker in any queue still executes, its consumer is possibly - * long-running and cancelling is the only way to get rid of it. Note that the - * cancellation handler will probably re-queue a user pointer, so the queue's enqueue - * function is still needed (what is no problem as we do not yet destroy the queue - but I - * thought it's a good idea to mention that fact). -- rgerhards, 2008-01-25 + * long-running and cancelling is the only way to get rid of it. */ - dbgoprint((obj_t*) pThis, "checking to see if we need to cancel any worker threads of the primary queue\n"); + DBGOPRINT((obj_t*) pThis, "checking to see if we need to cancel any worker threads of the primary queue\n"); iRetLocal = wtpCancelAll(pThis->pWtpReg); /* returns immediately if all threads already have terminated */ if(iRetLocal != RS_RET_OK) { - dbgoprint((obj_t*) pThis, "unexpected iRet state %d trying to cancel primary queue worker " + DBGOPRINT((obj_t*) pThis, "unexpected iRet state %d trying to cancel primary queue worker " "threads, continuing, but results are unpredictable\n", iRetLocal); } - - /* TODO: think: do we really need to do this here? Can't it happen on DA queue destruction? If we - * disable it, we get an assertion... I think this is OK, as we need to have a certain order and - * canceling the DA workers here ensures that order. But in any instant, we may have a look at this - * code after we have reaced the milestone. -- rgerhards, 2008-01-27 - */ /* ... and now the DA queue, if it exists (should always be after the primary one) */ if(pThis->pqDA != NULL) { - dbgoprint((obj_t*) pThis, "checking to see if we need to cancel any worker threads of the DA queue\n"); + DBGOPRINT((obj_t*) pThis, "checking to see if we need to cancel any worker threads of the DA queue\n"); iRetLocal = wtpCancelAll(pThis->pqDA->pWtpReg); /* returns immediately if all threads already have terminated */ if(iRetLocal != RS_RET_OK) { - dbgoprint((obj_t*) pThis, "unexpected iRet state %d trying to cancel DA queue worker " + DBGOPRINT((obj_t*) pThis, "unexpected iRet state %d trying to cancel DA queue worker " "threads, continuing, but results are unpredictable\n", iRetLocal); } + + /* finally, we cancel the main queue's DA worker pool, if it still is running. It may be + * restarted later to persist the queue. But we stop it, because otherwise we get into + * big trouble when resetting the logical dequeue pointer. This operation can only be + * done when *no* worker is running. So time for a shutdown... -- rgerhards, 2009-05-28 + */ + DBGOPRINT((obj_t*) pThis, "checking to see if main queue DA worker pool needs to be cancelled\n"); + iRetLocal = wtpCancelAll(pThis->pWtpDA); /* returns immediately if all threads already have terminated */ + } + + RETiRet; +} + + +/* This function shuts down all worker threads and waits until they + * have terminated. If they timeout, they are cancelled. + * rgerhards, 2008-01-24 + * Please note that this function shuts down BOTH the parent AND the child queue + * in DA case. This is necessary because their timeouts are tightly coupled. Most + * importantly, the timeouts would be applied twice (or logic be extremely + * complex) if each would have its own shutdown. The function does not self check + * this condition - the caller must make sure it is not called with a parent. + * rgerhards, 2009-05-26: we do NO longer persist the queue here if bSaveOnShutdown + * is set. This must be handled by the caller. Not doing that cleans up the queue + * shutdown considerably. Also, older engines had a potential hang condition when + * the DA queue was already started and the DA worker configured for infinite + * retries and the action was during retry processing. This was a design issue, + * which is solved as of now. Note that the shutdown now may take a little bit + * longer, because we no longer can persist the queue in parallel to waiting + * on worker timeouts. + */ +static rsRetVal +ShutdownWorkers(qqueue_t *pThis) +{ + DEFiRet; + + ISOBJ_TYPE_assert(pThis, qqueue); + ASSERT(pThis->pqParent == NULL); /* detect invalid calling sequence */ + + DBGOPRINT((obj_t*) pThis, "initiating worker thread shutdown sequence\n"); + + CHKiRet(tryShutdownWorkersWithinQueueTimeout(pThis)); + + if(getPhysicalQueueSize(pThis) > 0) { + CHKiRet(tryShutdownWorkersWithinActionTimeout(pThis)); } + CHKiRet(cancelWorkers(pThis)); + /* ... finally ... all worker threads have terminated :-) * Well, more precisely, they *are in termination*. Some cancel cleanup handlers - * may still be running. + * may still be running. Note that the main queue's DA worker may still be running. */ - dbgoprint((obj_t*) pThis, "worker threads terminated, remaining queue size %d.\n", qqueueGetOverallQueueSize(pThis)); + DBGOPRINT((obj_t*) pThis, "worker threads terminated, remaining queue size log %d, phys %d.\n", + getLogicalQueueSize(pThis), getPhysicalQueueSize(pThis)); +finalize_it: RETiRet; } @@ -1282,7 +1190,7 @@ static rsRetVal qqueueShutdownWorkers(qqueue_t *pThis) * to modify some parameters before the queue is actually started. */ rsRetVal qqueueConstruct(qqueue_t **ppThis, queueType_t qType, int iWorkerThreads, - int iMaxQueueSize, rsRetVal (*pConsumer)(void*,void*)) + int iMaxQueueSize, rsRetVal (*pConsumer)(void*, batch_t*,int*)) { DEFiRet; qqueue_t *pThis; @@ -1291,9 +1199,7 @@ rsRetVal qqueueConstruct(qqueue_t **ppThis, queueType_t qType, int iWorkerThread ASSERT(pConsumer != NULL); ASSERT(iWorkerThreads >= 0); - if((pThis = (qqueue_t *)calloc(1, sizeof(qqueue_t))) == NULL) { - ABORT_FINALIZE(RS_RET_OUT_OF_MEMORY); - } + CHKmalloc(pThis = (qqueue_t *)calloc(1, sizeof(qqueue_t))); /* we have an object, so let's fill the properties */ objConstructSetObjInfo(pThis); @@ -1304,13 +1210,15 @@ rsRetVal qqueueConstruct(qqueue_t **ppThis, queueType_t qType, int iWorkerThread pThis->iFullDlyMrk = iMaxQueueSize - (iMaxQueueSize / 100) * 3; /* default 97% */ pThis->iLightDlyMrk = iMaxQueueSize - (iMaxQueueSize / 100) * 30; /* default 70% */ - pThis->lenSpoolDir = strlen((char*)pThis->pszSpoolDir); + pThis->lenSpoolDir = ustrlen(pThis->pszSpoolDir); pThis->iMaxFileSize = 1024 * 1024; /* default is 1 MiB */ pThis->iQueueSize = 0; + pThis->nLogDeq = 0; pThis->iMaxQueueSize = iMaxQueueSize; pThis->pConsumer = pConsumer; pThis->iNumWorkerThreads = iWorkerThreads; pThis->iDeqtWinToHr = 25; /* disable time-windowed dequeuing by default */ + pThis->iDeqBatchSize = 8; /* conservative default, should still provide good performance */ pThis->pszFilePrefix = NULL; pThis->qType = qType; @@ -1321,19 +1229,25 @@ rsRetVal qqueueConstruct(qqueue_t **ppThis, queueType_t qType, int iWorkerThread pThis->qConstruct = qConstructFixedArray; pThis->qDestruct = qDestructFixedArray; pThis->qAdd = qAddFixedArray; + pThis->qDeq = qDeqFixedArray; pThis->qDel = qDelFixedArray; + pThis->MultiEnq = qqueueMultiEnqObjNonDirect; break; case QUEUETYPE_LINKEDLIST: pThis->qConstruct = qConstructLinkedList; pThis->qDestruct = qDestructLinkedList; pThis->qAdd = qAddLinkedList; - pThis->qDel = (rsRetVal (*)(qqueue_t*,void**)) qDelLinkedList; + pThis->qDeq = (rsRetVal (*)(qqueue_t*,void**)) qDeqLinkedList; + pThis->qDel = (rsRetVal (*)(qqueue_t*)) qDelLinkedList; + pThis->MultiEnq = qqueueMultiEnqObjNonDirect; break; case QUEUETYPE_DISK: pThis->qConstruct = qConstructDisk; pThis->qDestruct = qDestructDisk; pThis->qAdd = qAddDisk; + pThis->qDeq = qDeqDisk; pThis->qDel = qDelDisk; + pThis->MultiEnq = qqueueMultiEnqObjNonDirect; /* special handling */ pThis->iNumWorkerThreads = 1; /* we need exactly one worker */ break; @@ -1342,10 +1256,12 @@ rsRetVal qqueueConstruct(qqueue_t **ppThis, queueType_t qType, int iWorkerThread pThis->qDestruct = qDestructDirect; pThis->qAdd = qAddDirect; pThis->qDel = qDelDirect; + pThis->MultiEnq = qqueueMultiEnqObjDirect; break; } INIT_ATOMIC_HELPER_MUT(pThis->mutQueueSize); + INIT_ATOMIC_HELPER_MUT(pThis->mutLogDeq); finalize_it: OBJCONSTRUCT_CHECK_SUCCESS_AND_CLEANUP @@ -1353,40 +1269,10 @@ finalize_it: } -/* cancellation cleanup handler for queueWorker () - * Updates admin structure and frees ressources. - * Params: - * arg1 - user pointer (in this case a qqueue_t) - * arg2 - user data pointer (in this case a queue data element, any object [queue's pUsr ptr!]) - * Note that arg2 may be NULL, in which case no dequeued but unprocessed pUsr exists! - * rgerhards, 2008-01-16 - */ -static rsRetVal -qqueueConsumerCancelCleanup(void *arg1, void *arg2) -{ - DEFiRet; - - qqueue_t *pThis = (qqueue_t*) arg1; - obj_t *pUsr = (obj_t*) arg2; - - ISOBJ_TYPE_assert(pThis, qqueue); - - if(pUsr != NULL) { - /* make sure the data element is not lost */ - dbgoprint((obj_t*) pThis, "cancelation cleanup handler consumer called, we need to unget one user data element\n"); - CHKiRet(qqueueUngetObj(pThis, pUsr, LOCK_MUTEX)); - } - -finalize_it: - RETiRet; -} - - - /* This function checks if the provided message shall be discarded and does so, if needed. * In DA mode, we do not discard any messages as we assume the disk subsystem is fast enough to * provide real-time creation of spool files. - * Note: cached copies of iQueueSize and bRunsDA are provided so that no mutex locks are required. + * Note: cached copies of iQueueSize is provided so that no mutex locks are required. * The caller must have obtained them while the mutex was locked. Of course, these values may no * longer be current, but that is OK for the discard check. At worst, the message is either processed * or discarded when it should not have been. As discarding is in itself somewhat racy and erratic, @@ -1396,7 +1282,7 @@ finalize_it: * the return state! * rgerhards, 2008-01-24 */ -static int qqueueChkDiscardMsg(qqueue_t *pThis, int iQueueSize, int bRunsDA, void *pUsr) +static int qqueueChkDiscardMsg(qqueue_t *pThis, int iQueueSize, void *pUsr) { DEFiRet; rsRetVal iRetLocal; @@ -1405,15 +1291,15 @@ static int qqueueChkDiscardMsg(qqueue_t *pThis, int iQueueSize, int bRunsDA, voi ISOBJ_TYPE_assert(pThis, qqueue); ISOBJ_assert(pUsr); - if(pThis->iDiscardMrk > 0 && iQueueSize >= pThis->iDiscardMrk && bRunsDA == 0) { + if(pThis->iDiscardMrk > 0 && iQueueSize >= pThis->iDiscardMrk) { iRetLocal = objGetSeverity(pUsr, &iSeverity); if(iRetLocal == RS_RET_OK && iSeverity >= pThis->iDiscardSeverity) { - dbgoprint((obj_t*) pThis, "queue nearly full (%d entries), discarded severity %d message\n", + DBGOPRINT((obj_t*) pThis, "queue nearly full (%d entries), discarded severity %d message\n", iQueueSize, iSeverity); objDestruct(pUsr); ABORT_FINALIZE(RS_RET_QUEUE_FULL); } else { - dbgoprint((obj_t*) pThis, "queue nearly full (%d entries), but could not drop msg " + DBGOPRINT((obj_t*) pThis, "queue nearly full (%d entries), but could not drop msg " "(iRet: %d, severity %d)\n", iQueueSize, iRetLocal, iSeverity); } } @@ -1423,38 +1309,191 @@ finalize_it: } -/* dequeue the queued object for the queue consumers. - * rgerhards, 2008-10-21 +/* Finally remove n elements from the queue store. */ -static rsRetVal -qqueueDequeueConsumable(qqueue_t *pThis, wti_t *pWti, int iCancelStateSave) +static inline rsRetVal +DoDeleteBatchFromQStore(qqueue_t *pThis, int nElem) { + int i; DEFiRet; + + ISOBJ_TYPE_assert(pThis, qqueue); + + /* now send delete request to storage driver */ + for(i = 0 ; i < nElem ; ++i) { + pThis->qDel(pThis); + } + + /* iQueueSize is not decremented by qDel(), so we need to do it ourselves */ + ATOMIC_SUB(&pThis->iQueueSize, nElem, &pThis->mutQueueSize); + ATOMIC_SUB(&pThis->nLogDeq, nElem, &pThis->mutLogDeq); +dbgprintf("delete batch from store, new sizes: log %d, phys %d\n", getLogicalQueueSize(pThis), getPhysicalQueueSize(pThis)); + ++pThis->deqIDDel; /* one more batch dequeued */ + + RETiRet; +} + + +/* remove messages from the physical queue store that are fully processed. This is + * controlled via the to-delete list. + */ +static inline rsRetVal +DeleteBatchFromQStore(qqueue_t *pThis, batch_t *pBatch) +{ + toDeleteLst_t *pTdl; + qDeqID deqIDDel; + DEFiRet; + + ISOBJ_TYPE_assert(pThis, qqueue); + assert(pBatch != NULL); + + pTdl = tdlPeek(pThis); /* get current head element */ + if(pTdl == NULL) { /* to-delete list empty */ + DoDeleteBatchFromQStore(pThis, pBatch->nElem); + } else if(pBatch->deqID == pThis->deqIDDel) { + deqIDDel = pThis->deqIDDel; + pTdl = tdlPeek(pThis); + while(pTdl != NULL && deqIDDel == pTdl->deqID) { + DoDeleteBatchFromQStore(pThis, pTdl->nElemDeq); + tdlPop(pThis); + ++deqIDDel; + pTdl = tdlPeek(pThis); + } + /* old entries deleted, now delete current ones... */ + DoDeleteBatchFromQStore(pThis, pBatch->nElem); + } else { + /* can not delete, insert into to-delete list */ + dbgprintf("not at head of to-delete list, enqueue %d\n", (int) pBatch->deqID); + CHKiRet(tdlAdd(pThis, pBatch->deqID, pBatch->nElem)); + } + +finalize_it: + RETiRet; +} + + +/* Delete a batch of processed user objects from the queue, which includes + * destructing the objects themself. Any entries not marked as finally + * processed are enqueued again. The new enqueue is necessary because we have a + * rgerhards, 2009-05-13 + */ +static inline rsRetVal +DeleteProcessedBatch(qqueue_t *pThis, batch_t *pBatch) +{ + int i; void *pUsr; + int nEnqueued = 0; + rsRetVal localRet; + DEFiRet; + + ISOBJ_TYPE_assert(pThis, qqueue); + assert(pBatch != NULL); + + for(i = 0 ; i < pBatch->nElem ; ++i) { + pUsr = pBatch->pElem[i].pUsrp; + if( pBatch->pElem[i].state == BATCH_STATE_RDY + || pBatch->pElem[i].state == BATCH_STATE_SUB) { +dbgprintf("XXX: DeleteProcessedBatch re-enqueue %d of %d, state %d\n", i, pBatch->nElem, pBatch->pElem[i].state); + localRet = doEnqSingleObj(pThis, eFLOWCTL_NO_DELAY, + (obj_t*)MsgAddRef((msg_t*) pUsr)); + ++nEnqueued; + if(localRet != RS_RET_OK) { + DBGPRINTF("error %d re-enqueuing unprocessed data element - discarded\n", localRet); + } + } + objDestruct(pUsr); + } + + dbgprintf("we deleted %d objects and enqueued %d objects\n", i-nEnqueued, nEnqueued); + + if(nEnqueued > 0) + qqueueChkPersist(pThis, nEnqueued); + + iRet = DeleteBatchFromQStore(pThis, pBatch); + + pBatch->nElem = pBatch->nElemDeq = 0; /* reset batch */ // TODO: more fine init, new fields! 2010-06-14 + + RETiRet; +} + + +/* dequeue as many user pointers as are available, until we hit the configured + * upper limit of pointers. Note that this function also deletes all processed + * objects from the previous batch. However, it is perfectly valid that the + * previous batch contained NO objects at all. For example, this happens + * immediately after system startup or when a queue was exhausted and the queue + * worker needed to wait for new data. + * This must only be called when the queue mutex is LOOKED, otherwise serious + * malfunction will happen. + */ +static inline rsRetVal +DequeueConsumableElements(qqueue_t *pThis, wti_t *pWti, int *piRemainingQueueSize) +{ + int nDequeued; + int nDiscarded; + int nDeleted; int iQueueSize; - int bRunsDA; /* cache for early mutex release */ - - /* dequeue element (still protected from mutex) */ - iRet = qqueueDel(pThis, &pUsr); - qqueueChkPersist(pThis); - iQueueSize = qqueueGetOverallQueueSize(pThis); /* cache this for after mutex release */ - bRunsDA = pThis->bRunsDA; /* cache this for after mutex release */ - - /* We now need to save the user pointer for the cancel cleanup handler, BUT ONLY - * if we could successfully obtain a user pointer. Otherwise, we would bring the - * cancel cleanup handler into big troubles (and we did ;)). Note that we can - * NOT set the variable further below, as this may lead to an object leak. We - * may get cancelled before we reach that part of the code, so the only - * solution is to do it here. -- rgerhards, 2008-02-27 - */ - if(iRet == RS_RET_OK) { - pWti->pUsrp = pUsr; + void *pUsr; + rsRetVal localRet; + DEFiRet; + + nDeleted = pWti->batch.nElemDeq; + DeleteProcessedBatch(pThis, &pWti->batch); + + nDequeued = nDiscarded = 0; + while((iQueueSize = getLogicalQueueSize(pThis)) > 0 && nDequeued < pThis->iDeqBatchSize) { + CHKiRet(qqueueDeq(pThis, &pUsr)); + + /* check if we should discard this element */ + localRet = qqueueChkDiscardMsg(pThis, pThis->iQueueSize, pUsr); + if(localRet == RS_RET_QUEUE_FULL) { + ++nDiscarded; + continue; + } else if(localRet != RS_RET_OK) { + ABORT_FINALIZE(localRet); + } + + /* all well, use this element */ + pWti->batch.pElem[nDequeued].pUsrp = pUsr; + pWti->batch.pElem[nDequeued].state = BATCH_STATE_RDY; + pWti->batch.pElem[nDequeued].bFilterOK = 1; // TODO: think again if we can handle that with more performance + ++nDequeued; } + /* it is sufficient to persist only when the bulk of work is done */ + qqueueChkPersist(pThis, nDequeued+nDiscarded+nDeleted); + + pWti->batch.nElem = nDequeued; + pWti->batch.nElemDeq = nDequeued + nDiscarded; + pWti->batch.deqID = getNextDeqID(pThis); + *piRemainingQueueSize = iQueueSize; + +finalize_it: + RETiRet; +} + + +/* dequeue the queued object for the queue consumers. + * rgerhards, 2008-10-21 + * I made a radical change - we now dequeue multiple elements, and store these objects in + * an array of user pointers. We expect that this increases performance. + * rgerhards, 2009-04-22 + */ +static rsRetVal +DequeueConsumable(qqueue_t *pThis, wti_t *pWti) +{ + DEFiRet; + int iQueueSize = 0; /* keep the compiler happy... */ + + /* dequeue element batch (still protected from mutex) */ + iRet = DequeueConsumableElements(pThis, pWti, &iQueueSize); + /* awake some flow-controlled sources if we can do this right now */ /* TODO: this could be done better from a performance point of view -- do it only if * we have someone waiting for the condition (or only when we hit the watermark right * on the nail [exact value]) -- rgerhards, 2008-03-14 + * now that we dequeue batches of pointers, this is much less an issue... + * rgerhards, 2009-04-22 */ if(iQueueSize < pThis->iFullDlyMrk / 2) { pthread_cond_broadcast(&pThis->belowFullDlyWtrMrk); @@ -1464,37 +1503,15 @@ qqueueDequeueConsumable(qqueue_t *pThis, wti_t *pWti, int iCancelStateSave) pthread_cond_broadcast(&pThis->belowLightDlyWtrMrk); } - /* rgerhards, 2008-09-30: I reversed the order of cond_signal und mutex_unlock - * as of the pthreads recommendation on predictable scheduling behaviour. I don't see - * any problems caused by this, but I add this comment in case some will be seen - * in the next time. - */ + // TODO: MULTI: check physical queue size? pthread_cond_signal(&pThis->notFull); - d_pthread_mutex_unlock(pThis->mut); - pthread_setcancelstate(iCancelStateSave, NULL); /* WE ARE NO LONGER PROTECTED BY THE MUTEX */ - /* do actual processing (the lengthy part, runs in parallel) - * If we had a problem while dequeing, we do not call the consumer, - * but we otherwise ignore it. This is in the hopes that it will be - * self-healing. However, this is really not a good thing. - * rgerhards, 2008-01-03 - */ - if(iRet != RS_RET_OK) - FINALIZE; - - /* we are running in normal, non-disk-assisted mode do a quick check if we need to drain the queue. - * In DA mode, we do not discard any messages as we assume the disk subsystem is fast enough to - * provide real-time creation of spool files. - * Note: It is OK to use the cached iQueueSize here, because it does not hurt if it is slightly wrong. - */ - CHKiRet(qqueueChkDiscardMsg(pThis, iQueueSize, bRunsDA, pUsr)); - -finalize_it: if(iRet != RS_RET_OK && iRet != RS_RET_DISCARDMSG) { - dbgoprint((obj_t*) pThis, "error %d dequeueing element - ignoring, but strange things " + DBGOPRINT((obj_t*) pThis, "error %d dequeueing element - ignoring, but strange things " "may happen\n", iRet); } + RETiRet; } @@ -1537,7 +1554,7 @@ finalize_it: * but you get the idea from the code above. */ static rsRetVal -qqueueRateLimiter(qqueue_t *pThis) +RateLimiter(qqueue_t *pThis) { DEFiRet; int iDelay; @@ -1550,7 +1567,7 @@ qqueueRateLimiter(qqueue_t *pThis) iDelay = 0; if(pThis->iDeqtWinToHr != 25) { /* 25 means disabled */ /* time calls are expensive, so only do them when needed */ - time(&tCurr); + datetime.GetTime(&tCurr); localtime_r(&tCurr, &m); iHrCurr = m.tm_hour; @@ -1586,7 +1603,7 @@ qqueueRateLimiter(qqueue_t *pThis) } if(iDelay > 0) { - dbgoprint((obj_t*) pThis, "outside dequeue time window, delaying %d seconds\n", iDelay); + DBGOPRINT((obj_t*) pThis, "outside dequeue time window, delaying %d seconds\n", iDelay); srSleep(iDelay, 0); } @@ -1594,37 +1611,105 @@ qqueueRateLimiter(qqueue_t *pThis) } +/* This dequeues the next batch. Note that this function must not be + * cancelled, else it will leave back an inconsistent state. + * rgerhards, 2009-05-20 + */ +static inline rsRetVal +DequeueForConsumer(qqueue_t *pThis, wti_t *pWti) +{ + DEFiRet; + + ISOBJ_TYPE_assert(pThis, qqueue); + ISOBJ_TYPE_assert(pWti, wti); + + CHKiRet(DequeueConsumable(pThis, pWti)); + + if(pWti->batch.nElem == 0) + ABORT_FINALIZE(RS_RET_IDLE); + + +finalize_it: + RETiRet; +} + + +/* This is called when a batch is processed and the worker does not + * ask for another batch (e.g. because it is to be terminated) + * Note that we must not be terminated while we delete a processed + * batch. Otherwise, we may not complete it, and then the cancel + * handler also tries to delete the batch. But then it finds some of + * the messages already destructed. This was a bug we have seen, especially + * with disk mode, where a delete takes rather long. Anyhow, the coneptual + * problem exists in all queue modes. + * rgerhards, 2009-05-27 + */ +static rsRetVal +batchProcessed(qqueue_t *pThis, wti_t *pWti) +{ + DEFiRet; + + ISOBJ_TYPE_assert(pThis, qqueue); + ISOBJ_TYPE_assert(pWti, wti); + + int iCancelStateSave; + /* at this spot, we must not be cancelled */ + pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &iCancelStateSave); + DeleteProcessedBatch(pThis, &pWti->batch); + qqueueChkPersist(pThis, pWti->batch.nElemDeq); + pthread_setcancelstate(iCancelStateSave, NULL); + + RETiRet; +} + /* This is the queue consumer in the regular (non-DA) case. It is * protected by the queue mutex, but MUST release it as soon as possible. * rgerhards, 2008-01-21 */ static rsRetVal -qqueueConsumerReg(qqueue_t *pThis, wti_t *pWti, int iCancelStateSave) +ConsumerReg(qqueue_t *pThis, wti_t *pWti) { + int iCancelStateSave; DEFiRet; ISOBJ_TYPE_assert(pThis, qqueue); ISOBJ_TYPE_assert(pWti, wti); - CHKiRet(qqueueDequeueConsumable(pThis, pWti, iCancelStateSave)); - CHKiRet(pThis->pConsumer(pThis->pUsr, pWti->pUsrp)); + CHKiRet(DequeueForConsumer(pThis, pWti)); + + /* we now have a non-idle batch of work, so we can release the queue mutex and process it */ + d_pthread_mutex_unlock(pThis->mut); + + /* at this spot, we may be cancelled */ + pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, &iCancelStateSave); + + CHKiRet(pThis->pConsumer(pThis->pUsr, &pWti->batch, &pThis->bShutdownImmediate)); /* we now need to check if we should deliberately delay processing a bit * and, if so, do that. -- rgerhards, 2008-01-30 */ +//TODO: MULTIQUEUE: the following setting is no longer correct - need to think about how to do that... if(pThis->iDeqSlowdown) { - dbgoprint((obj_t*) pThis, "sleeping %d microseconds as requested by config params\n", + DBGOPRINT((obj_t*) pThis, "sleeping %d microseconds as requested by config params\n", pThis->iDeqSlowdown); srSleep(pThis->iDeqSlowdown / 1000000, pThis->iDeqSlowdown % 1000000); } + /* but now cancellation is no longer permitted */ + pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &iCancelStateSave); + + /* now we are done, but need to re-aquire the mutex */ + d_pthread_mutex_lock(pThis->mut); + finalize_it: + dbgprintf("regular consumer finished, iret=%d, szlog %d sz phys %d\n", iRet, + getLogicalQueueSize(pThis), getPhysicalQueueSize(pThis)); RETiRet; } -/* This is a special consumer to feed the disk-queue in disk-assited mode. +/* This is a special consumer to feed the disk-queue in disk-assisted mode. * When active, our own queue more or less acts as a memory buffer to the disk. * So this consumer just needs to drain the memory queue and submit entries * to the disk queue. The disk queue will then call the actual consumer from @@ -1634,151 +1719,92 @@ finalize_it: * rgerhards, 2008-01-14 */ static rsRetVal -qqueueConsumerDA(qqueue_t *pThis, wti_t *pWti, int iCancelStateSave) +ConsumerDA(qqueue_t *pThis, wti_t *pWti) { + int i; + int iCancelStateSave; DEFiRet; ISOBJ_TYPE_assert(pThis, qqueue); ISOBJ_TYPE_assert(pWti, wti); - CHKiRet(qqueueDequeueConsumable(pThis, pWti, iCancelStateSave)); - CHKiRet(qqueueEnqObj(pThis->pqDA, eFLOWCTL_NO_DELAY, pWti->pUsrp)); + CHKiRet(DequeueForConsumer(pThis, pWti)); + + /* we now have a non-idle batch of work, so we can release the queue mutex and process it */ + d_pthread_mutex_unlock(pThis->mut); + + /* at this spot, we may be cancelled */ + pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, &iCancelStateSave); + + /* iterate over returned results and enqueue them in DA queue */ + for(i = 0 ; i < pWti->batch.nElem && !pThis->bShutdownImmediate ; i++) { + /* TODO: we must add a generic "addRef" mechanism, because the disk queue enqueue destructs + * the message. So far, we simply assume we always have msg_t, what currently is always the case. + * rgerhards, 2009-05-28 + */ + CHKiRet(qqueueEnqObj(pThis->pqDA, eFLOWCTL_NO_DELAY, + (obj_t*)MsgAddRef((msg_t*)(pWti->batch.pElem[i].pUsrp)))); + pWti->batch.pElem[i].state = BATCH_STATE_COMM; /* commited to other queue! */ + } + + /* but now cancellation is no longer permitted */ + pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &iCancelStateSave); + + /* now we are done, but need to re-aquire the mutex */ + d_pthread_mutex_lock(pThis->mut); finalize_it: - dbgoprint((obj_t*) pThis, "DAConsumer returns with iRet %d\n", iRet); + DBGOPRINT((obj_t*) pThis, "DAConsumer returns with iRet %d\n", iRet); RETiRet; } /* must only be called when the queue mutex is locked, else results * are not stable! - * If we are a child, we have done our duty when the queue is empty. In that case, - * we can terminate. - * Version for the DA worker thread. NOTE: the pThis->bRunsDA is different from - * the DA queue */ -static int +static rsRetVal qqueueChkStopWrkrDA(qqueue_t *pThis) { - /* if our queue is in destruction, we drain to the DA queue and so we shall not terminate - * until we have done so. - */ - int bStopWrkr; - - BEGINfunc + DEFiRet; if(pThis->bEnqOnly) { - bStopWrkr = 1; - } else { - if(pThis->bRunsDA == 2) { - ASSERT(pThis->pqDA != NULL); - if( pThis->pqDA->bEnqOnly - && pThis->pqDA->sizeOnDiskMax > 0 - && pThis->pqDA->tVars.disk.sizeOnDisk > pThis->pqDA->sizeOnDiskMax) { - /* this queue can never grow, so we can give up... */ - bStopWrkr = 1; - } else if(qqueueGetOverallQueueSize(pThis) < pThis->iHighWtrMrk && pThis->bQueueStarted == 1) { - bStopWrkr = 1; - } else { - bStopWrkr = 0; - } - } else { - bStopWrkr = 1; - } + iRet = RS_RET_TERMINATE_WHEN_IDLE; } - ENDfunc - return bStopWrkr; + RETiRet; } /* must only be called when the queue mutex is locked, else results * are not stable! * If we are a child, we have done our duty when the queue is empty. In that case, - * we can terminate. - * Version for the regular worker thread. NOTE: the pThis->bRunsDA is different from - * the DA queue - */ -static int -qqueueChkStopWrkrReg(qqueue_t *pThis) -{ - return pThis->bEnqOnly || pThis->bRunsDA || (pThis->pqParent != NULL && qqueueGetOverallQueueSize(pThis) == 0); -} - - -/* must only be called when the queue mutex is locked, else results - * are not stable! DA queue version - */ -static int -qqueueIsIdleDA(qqueue_t *pThis) -{ - /* remember: iQueueSize is the DA queue size, not the main queue! */ - /* TODO: I think we need just a single function for DA and non-DA mode - but I leave it for now as is */ - return(qqueueGetOverallQueueSize(pThis) == 0 || (pThis->bRunsDA && qqueueGetOverallQueueSize(pThis) <= pThis->iLowWtrMrk)); -} -/* must only be called when the queue mutex is locked, else results - * are not stable! Regular queue version - */ -static int -qqueueIsIdleReg(qqueue_t *pThis) -{ -#if 0 /* enable for performance testing */ - int ret; - ret = qqueueGetOverallQueueSize(pThis) == 0 || (pThis->bRunsDA && qqueueGetOverallQueueSize(pThis) <= pThis->iLowWtrMrk); - if(ret) fprintf(stderr, "queue is idle\n"); - return ret; -#else - /* regular code! */ - return(qqueueGetOverallQueueSize(pThis) == 0 || (pThis->bRunsDA && qqueueGetOverallQueueSize(pThis) <= pThis->iLowWtrMrk)); -#endif -} - - -/* This function is called when a worker thread for the regular queue is shut down. - * If we are the primary queue, this is not really interesting to us. If, however, - * we are the DA (child) queue, that means the DA queue is empty. In that case, we - * need to signal the parent queue's DA worker, so that it can terminate DA mode. - * rgerhards, 2008-01-26 - * rgerhards, 2008-02-27: HOWEVER, in a shutdown condition, it may be that the parent's worker thread pool - * has already been terminated and destructed. This *is* a legal condition and happens - * from time to time in practice. So we need to signal only if there still is a - * parent DA worker queue. Please keep in mind that the the parent's DA worker - * pool is DIFFERENT from our (DA queue) regular worker pool. So when the parent's - * pWtpDA is destructed, there can still be some of our (DAq/wtp) threads be running. - * I am telling this, because I, too, always get confused by those... + * we can terminate. Version for the regular worker thread. */ static rsRetVal -qqueueRegOnWrkrShutdown(qqueue_t *pThis) +ChkStopWrkrReg(qqueue_t *pThis) { DEFiRet; - - ISOBJ_TYPE_assert(pThis, qqueue); - - if(pThis->pqParent != NULL) { - pThis->pqParent->bChildIsDone = 1; /* indicate we are done */ - if(pThis->pqParent->pWtpDA != NULL) { /* see comment in function header from 2008-02-27 */ - wtpAdviseMaxWorkers(pThis->pqParent->pWtpDA, 1); /* reactivate DA worker (always 1) */ - } + if(pThis->bEnqOnly) { + iRet = RS_RET_TERMINATE_NOW; + } else if(pThis->pqParent != NULL) { + iRet = RS_RET_TERMINATE_WHEN_IDLE; } RETiRet; } -/* The following function is called when a regular queue worker starts up. We need this - * hook to indicate in the parent queue (if we are a child) that we are not done yet. +/* return the configured "deq max at once" interval + * rgerhards, 2009-04-22 */ static rsRetVal -qqueueRegOnWrkrStartup(qqueue_t *pThis) +GetDeqBatchSize(qqueue_t *pThis, int *pVal) { DEFiRet; - - ISOBJ_TYPE_assert(pThis, qqueue); - - if(pThis->pqParent != NULL) { - pThis->pqParent->bChildIsDone = 0; - } - + assert(pVal != NULL); + *pVal = pThis->iDeqBatchSize; +if(pThis->pqParent != NULL) // TODO: check why we actually do this! + *pVal = 16; RETiRet; } @@ -1786,11 +1812,10 @@ qqueueRegOnWrkrStartup(qqueue_t *pThis) /* start up the queue - it must have been constructed and parameters defined * before. */ -rsRetVal qqueueStart(qqueue_t *pThis) /* this is the ConstructionFinalizer */ +rsRetVal +qqueueStart(qqueue_t *pThis) /* this is the ConstructionFinalizer */ { DEFiRet; - rsRetVal iRetLocal; - int bInitialized = 0; /* is queue already initialized? */ uchar pszBuf[64]; size_t lenBuf; @@ -1804,11 +1829,11 @@ rsRetVal qqueueStart(qqueue_t *pThis) /* this is the ConstructionFinalizer */ * influenced by properties which might have been set after queueConstruct () */ if(pThis->pqParent == NULL) { - pThis->mut = (pthread_mutex_t *) malloc (sizeof (pthread_mutex_t)); + pThis->mut = (pthread_mutex_t *) MALLOC (sizeof (pthread_mutex_t)); pthread_mutex_init(pThis->mut, NULL); } else { /* child queue, we need to use parent's mutex */ - dbgoprint((obj_t*) pThis, "I am a child\n"); + DBGOPRINT((obj_t*) pThis, "I am a child\n"); pThis->mut = pThis->pqParent->mut; } @@ -1822,28 +1847,26 @@ rsRetVal qqueueStart(qqueue_t *pThis) /* this is the ConstructionFinalizer */ /* call type-specific constructor */ CHKiRet(pThis->qConstruct(pThis)); /* this also sets bIsDA */ - dbgoprint((obj_t*) pThis, "type %d, enq-only %d, disk assisted %d, maxFileSz %lld, qsize %d, child %d, " - "full delay %d, light delay %d starting\n", + DBGOPRINT((obj_t*) pThis, "type %d, enq-only %d, disk assisted %d, maxFileSz %lld, lqsize %d, pqsize %d, child %d, " + "full delay %d, light delay %d, deq batch size %d starting\n", pThis->qType, pThis->bEnqOnly, pThis->bIsDA, pThis->iMaxFileSize, - qqueueGetOverallQueueSize(pThis), pThis->pqParent == NULL ? 0 : 1, - pThis->iFullDlyMrk, pThis->iLightDlyMrk); + getLogicalQueueSize(pThis), getPhysicalQueueSize(pThis), + pThis->pqParent == NULL ? 0 : 1, pThis->iFullDlyMrk, pThis->iLightDlyMrk, + pThis->iDeqBatchSize); if(pThis->qType == QUEUETYPE_DIRECT) FINALIZE; /* with direct queues, we are already finished... */ - /* create worker thread pools for regular operation. The DA pool is created on an as-needed - * basis, which potentially means never under most circumstances. + /* create worker thread pools for regular and DA operation. */ lenBuf = snprintf((char*)pszBuf, sizeof(pszBuf), "%s:Reg", obj.GetName((obj_t*) pThis)); CHKiRet(wtpConstruct (&pThis->pWtpReg)); CHKiRet(wtpSetDbgHdr (pThis->pWtpReg, pszBuf, lenBuf)); - CHKiRet(wtpSetpfRateLimiter (pThis->pWtpReg, (rsRetVal (*)(void *pUsr)) qqueueRateLimiter)); - CHKiRet(wtpSetpfChkStopWrkr (pThis->pWtpReg, (rsRetVal (*)(void *pUsr, int)) qqueueChkStopWrkrReg)); - CHKiRet(wtpSetpfIsIdle (pThis->pWtpReg, (rsRetVal (*)(void *pUsr, int)) qqueueIsIdleReg)); - CHKiRet(wtpSetpfDoWork (pThis->pWtpReg, (rsRetVal (*)(void *pUsr, void *pWti, int)) qqueueConsumerReg)); - CHKiRet(wtpSetpfOnWorkerCancel (pThis->pWtpReg, (rsRetVal (*)(void *pUsr, void*pWti))qqueueConsumerCancelCleanup)); - CHKiRet(wtpSetpfOnWorkerStartup (pThis->pWtpReg, (rsRetVal (*)(void *pUsr)) qqueueRegOnWrkrStartup)); - CHKiRet(wtpSetpfOnWorkerShutdown(pThis->pWtpReg, (rsRetVal (*)(void *pUsr)) qqueueRegOnWrkrShutdown)); + CHKiRet(wtpSetpfRateLimiter (pThis->pWtpReg, (rsRetVal (*)(void *pUsr)) RateLimiter)); + CHKiRet(wtpSetpfChkStopWrkr (pThis->pWtpReg, (rsRetVal (*)(void *pUsr, int)) ChkStopWrkrReg)); + CHKiRet(wtpSetpfGetDeqBatchSize (pThis->pWtpReg, (rsRetVal (*)(void *pUsr, int*)) GetDeqBatchSize)); + CHKiRet(wtpSetpfDoWork (pThis->pWtpReg, (rsRetVal (*)(void *pUsr, void *pWti)) ConsumerReg)); + CHKiRet(wtpSetpfObjProcessed (pThis->pWtpReg, (rsRetVal (*)(void *pUsr, wti_t *pWti)) batchProcessed)); CHKiRet(wtpSetpmutUsr (pThis->pWtpReg, pThis->mut)); CHKiRet(wtpSetpcondBusy (pThis->pWtpReg, &pThis->notEmpty)); CHKiRet(wtpSetiNumWorkerThreads (pThis->pWtpReg, pThis->iNumWorkerThreads)); @@ -1851,27 +1874,11 @@ rsRetVal qqueueStart(qqueue_t *pThis) /* this is the ConstructionFinalizer */ CHKiRet(wtpSetpUsr (pThis->pWtpReg, pThis)); CHKiRet(wtpConstructFinalize (pThis->pWtpReg)); - /* initialize worker thread instances */ - if(pThis->bIsDA) { - /* If we are disk-assisted, we need to check if there is a QIF file - * which we need to load. -- rgerhards, 2008-01-15 - */ - iRetLocal = qqueueHaveQIF(pThis); - if(iRetLocal == RS_RET_OK) { - dbgoprint((obj_t*) pThis, "on-disk queue present, needs to be reloaded\n"); - qqueueInitDA(pThis, QUEUE_MODE_ENQDEQ, LOCK_MUTEX); /* initiate DA mode */ - bInitialized = 1; /* we are done */ - } else { - /* TODO: use logerror? -- rgerhards, 2008-01-16 */ - dbgoprint((obj_t*) pThis, "error %d trying to access on-disk queue files, starting without them. " - "Some data may be lost\n", iRetLocal); - } - } + /* set up DA system if we have a disk-assisted queue */ + if(pThis->bIsDA) + InitDA(pThis, LOCK_MUTEX); /* initiate DA mode */ - if(!bInitialized) { - dbgoprint((obj_t*) pThis, "queue starts up without (loading) any DA disk state (this is normal for the DA " - "queue itself!)\n"); - } + DBGOPRINT((obj_t*) pThis, "queue finished initialization\n"); /* if the queue already contains data, we need to start the correct number of worker threads. This can be * the case when a disk queue has been loaded. If we did not start it here, it would never start. @@ -1897,12 +1904,11 @@ static rsRetVal qqueuePersist(qqueue_t *pThis, int bIsCheckpoint) strm_t *psQIF = NULL; /* Queue Info File */ uchar pszQIFNam[MAXFNAME]; size_t lenQIFNam; - obj_t *pUsr; ASSERT(pThis != NULL); if(pThis->qType != QUEUETYPE_DISK) { - if(qqueueGetOverallQueueSize(pThis) > 0) { + if(getPhysicalQueueSize(pThis) > 0) { /* This error code is OK, but we will probably not implement this any time * The reason is that persistence happens via DA queues. But I would like to * leave the code as is, as we so have a hook in case we need one. @@ -1913,20 +1919,20 @@ static rsRetVal qqueuePersist(qqueue_t *pThis, int bIsCheckpoint) FINALIZE; /* if the queue is empty, we are happy and done... */ } - dbgoprint((obj_t*) pThis, "persisting queue to disk, %d entries...\n", qqueueGetOverallQueueSize(pThis)); + DBGOPRINT((obj_t*) pThis, "persisting queue to disk, %d entries...\n", getPhysicalQueueSize(pThis)); /* Construct file name */ lenQIFNam = snprintf((char*)pszQIFNam, sizeof(pszQIFNam) / sizeof(uchar), "%s/%s.qi", (char*) glbl.GetWorkDir(), (char*)pThis->pszFilePrefix); - if((bIsCheckpoint != QUEUE_CHECKPOINT) && (qqueueGetOverallQueueSize(pThis) == 0)) { + if((bIsCheckpoint != QUEUE_CHECKPOINT) && (getPhysicalQueueSize(pThis) == 0)) { if(pThis->bNeedDelQIF) { unlink((char*)pszQIFNam); pThis->bNeedDelQIF = 0; } /* indicate spool file needs to be deleted */ - if (pThis->tVars.disk.pRead != NULL) - CHKiRet(strm.SetbDeleteOnClose(pThis->tVars.disk.pRead, 1)); + if(pThis->tVars.disk.pReadDel != NULL) /* may be NULL if we had a startup failure! */ + CHKiRet(strm.SetbDeleteOnClose(pThis->tVars.disk.pReadDel, 1)); FINALIZE; /* nothing left to do, so be happy */ } @@ -1945,31 +1951,19 @@ static rsRetVal qqueuePersist(qqueue_t *pThis, int bIsCheckpoint) */ CHKiRet(obj.BeginSerializePropBag(psQIF, (obj_t*) pThis)); objSerializeSCALAR(psQIF, iQueueSize, INT); - objSerializeSCALAR(psQIF, iUngottenObjs, INT); objSerializeSCALAR(psQIF, tVars.disk.sizeOnDisk, INT64); objSerializeSCALAR(psQIF, tVars.disk.bytesRead, INT64); CHKiRet(obj.EndSerialize(psQIF)); - /* now we must persist all objects on the ungotten queue - they can not go to - * to the regular files. -- rgerhards, 2008-01-29 - */ - while(pThis->iUngottenObjs > 0) { - CHKiRet(qqueueGetUngottenObj(pThis, &pUsr)); - CHKiRet((objSerialize(pUsr))(pUsr, psQIF)); - objDestruct(pUsr); - } - /* now persist the stream info */ - if (pThis->tVars.disk.pWrite != NULL) - CHKiRet(strm.Serialize(pThis->tVars.disk.pWrite, psQIF)); - if (pThis->tVars.disk.pRead != NULL) - CHKiRet(strm.Serialize(pThis->tVars.disk.pRead, psQIF)); + CHKiRet(strm.Serialize(pThis->tVars.disk.pWrite, psQIF)); + CHKiRet(strm.Serialize(pThis->tVars.disk.pReadDel, psQIF)); /* tell the input file object that it must not delete the file on close if the queue * is non-empty - but only if we are not during a simple checkpoint */ - if(bIsCheckpoint != QUEUE_CHECKPOINT && pThis->tVars.disk.pRead != NULL) { - CHKiRet(strm.SetbDeleteOnClose(pThis->tVars.disk.pRead, 0)); + if(bIsCheckpoint != QUEUE_CHECKPOINT) { + CHKiRet(strm.SetbDeleteOnClose(pThis->tVars.disk.pReadDel, 0)); } /* we have persisted the queue object. So whenever it comes to an empty queue, @@ -1986,36 +1980,86 @@ finalize_it: /* check if we need to persist the current queue info. If an - * error occurs, thus should be ignored by caller (but we still + * error occurs, this should be ignored by caller (but we still * abide to our regular call interface)... * rgerhards, 2008-01-13 + * nUpdates is the number of updates since the last call to this function. + * It may be > 1 due to batches. -- rgerhards, 2009-05-12 */ -static rsRetVal qqueueChkPersist(qqueue_t *pThis) +static rsRetVal qqueueChkPersist(qqueue_t *pThis, int nUpdates) { + DEFiRet; ISOBJ_TYPE_assert(pThis, qqueue); + assert(nUpdates >= 0); + + if(nUpdates == 0) + FINALIZE; - if(pThis->iPersistUpdCnt && ++pThis->iUpdsSincePersist >= pThis->iPersistUpdCnt) { + pThis->iUpdsSincePersist += nUpdates; + if(pThis->iPersistUpdCnt && pThis->iUpdsSincePersist >= pThis->iPersistUpdCnt) { qqueuePersist(pThis, QUEUE_CHECKPOINT); pThis->iUpdsSincePersist = 0; } - return RS_RET_OK; +finalize_it: + RETiRet; +} + + +/* persist a queue with all data elements to disk - this is used to handle + * bSaveOnShutdown. We utilize the DA worker to do this. This must only + * be called after all workers have been shut down and if bSaveOnShutdown + * is actually set. Note that this function may potentially run long, + * depending on the queue configuration (e.g. store on remote machine). + * rgerhards, 2009-05-26 + */ +static inline rsRetVal +DoSaveOnShutdown(qqueue_t *pThis) +{ + struct timespec tTimeout; + rsRetVal iRetLocal; + DEFiRet; + + ISOBJ_TYPE_assert(pThis, qqueue); + + /* we reduce the low water mark, otherwise the DA worker would terminate when + * it is reached. + */ + DBGOPRINT((obj_t*) pThis, "bSaveOnShutdown set, restarting DA worker...\n"); + pThis->bShutdownImmediate = 0; /* would termiante the DA worker! */ + pThis->iLowWtrMrk = 0; + wtpSetState(pThis->pWtpDA, wtpState_SHUTDOWN); /* shutdown worker (only) when done (was _IMMEDIATE!) */ + wtpAdviseMaxWorkers(pThis->pWtpDA, 1); /* restart DA worker */ + + DBGOPRINT((obj_t*) pThis, "waiting for DA worker to terminate...\n"); + timeoutComp(&tTimeout, QUEUE_TIMEOUT_ETERNAL); + /* and run the primary queue's DA worker to drain the queue */ + iRetLocal = wtpShutdownAll(pThis->pWtpDA, wtpState_SHUTDOWN, &tTimeout); + DBGOPRINT((obj_t*) pThis, "end queue persistence run, iRet %d, queue size log %d, phys %d\n", + iRetLocal, getLogicalQueueSize(pThis), getPhysicalQueueSize(pThis)); + if(iRetLocal != RS_RET_OK) { + DBGOPRINT((obj_t*) pThis, "unexpected iRet state %d after trying to shut down primary queue in disk save mode, " + "continuing, but results are unpredictable\n", iRetLocal); + } + + RETiRet; } /* destructor for the queue object */ BEGINobjDestruct(qqueue) /* be sure to specify the object type also in END and CODESTART macros! */ CODESTARTobjDestruct(qqueue) - pThis->bQueueInDestruction = 1; /* indicate we are in destruction (modifies some behaviour) */ - - /* shut down all workers (handles *all* of the persistence logic) - * See function head comment of queueShutdownWorkers () on why we don't call it - * We also do not need to shutdown workers when we are in enqueue-only mode or we are a + /* shut down all workers + * We do not need to shutdown workers when we are in enqueue-only mode or we are a * direct queue - because in both cases we have none... ;) * with a child! -- rgerhards, 2008-01-28 */ if(pThis->qType != QUEUETYPE_DIRECT && !pThis->bEnqOnly && pThis->pqParent == NULL) - qqueueShutdownWorkers(pThis); + ShutdownWorkers(pThis); + + if(pThis->bIsDA && getPhysicalQueueSize(pThis) > 0 && pThis->bSaveOnShutdown) { + CHKiRet(DoSaveOnShutdown(pThis)); + } /* finally destruct our (regular) worker thread pool * Note: currently pWtpReg is never NULL, but if we optimize our logic, this may happen, @@ -2051,7 +2095,7 @@ CODESTARTobjDestruct(qqueue) * if need arises (what I doubt...) -- rgerhards, 2008-01-25 */ CHKiRet_Hdlr(qqueuePersist(pThis, QUEUE_NO_CHECKPOINT)) { - dbgoprint((obj_t*) pThis, "error %d persisting queue - data lost!\n", iRet); + DBGOPRINT((obj_t*) pThis, "error %d persisting queue - data lost!\n", iRet); } /* finally, clean up some simple things... */ @@ -2068,15 +2112,13 @@ CODESTARTobjDestruct(qqueue) pthread_cond_destroy(&pThis->belowLightDlyWtrMrk); DESTROY_ATOMIC_HELPER_MUT(pThis->mutQueueSize); + DESTROY_ATOMIC_HELPER_MUT(pThis->mutLogDeq); /* type-specific destructor */ iRet = pThis->qDestruct(pThis); - if(pThis->pszFilePrefix != NULL) - free(pThis->pszFilePrefix); - - if(pThis->pszSpoolDir != NULL) - free(pThis->pszSpoolDir); + free(pThis->pszFilePrefix); + free(pThis->pszSpoolDir); ENDobjDestruct(qqueue) @@ -2090,13 +2132,13 @@ qqueueSetFilePrefix(qqueue_t *pThis, uchar *pszPrefix, size_t iLenPrefix) { DEFiRet; - if(pThis->pszFilePrefix != NULL) - free(pThis->pszFilePrefix); + free(pThis->pszFilePrefix); + pThis->pszFilePrefix = NULL; if(pszPrefix == NULL) /* just unset the prefix! */ ABORT_FINALIZE(RS_RET_OK); - if((pThis->pszFilePrefix = malloc(sizeof(uchar) * iLenPrefix + 1)) == NULL) + if((pThis->pszFilePrefix = MALLOC(sizeof(uchar) * iLenPrefix + 1)) == NULL) ABORT_FINALIZE(RS_RET_OUT_OF_MEMORY); memcpy(pThis->pszFilePrefix, pszPrefix, iLenPrefix + 1); pThis->lenFilePrefix = iLenPrefix; @@ -2126,111 +2168,7 @@ finalize_it: } -/* enqueue a new user data element - * Enqueues the new element and awakes worker thread. - */ -rsRetVal -qqueueEnqObj(qqueue_t *pThis, flowControl_t flowCtlType, void *pUsr) -{ - DEFiRet; - int iCancelStateSave; - struct timespec t; - - ISOBJ_TYPE_assert(pThis, qqueue); - - /* first check if we need to discard this message (which will cause CHKiRet() to exit) - * rgerhards, 2008-10-07: It is OK to do this outside of mutex protection. The iQueueSize - * and bRunsDA parameters may not reflect the correct settings here, but they are - * "good enough" in the sense that they can be used to drive the decision. Valgrind's - * threading tools may point this access to be an error, but this is done - * intentional. I do not see this causes problems to us. - */ - CHKiRet(qqueueChkDiscardMsg(pThis, pThis->iQueueSize, pThis->bRunsDA, pUsr)); - - /* Please note that this function is not cancel-safe and consequently - * sets the calling thread's cancelibility state to PTHREAD_CANCEL_DISABLE - * during its execution. If that is not done, race conditions occur if the - * thread is canceled (most important use case is input module termination). - * rgerhards, 2008-01-08 - */ - if(pThis->qType != QUEUETYPE_DIRECT) { - pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &iCancelStateSave); - d_pthread_mutex_lock(pThis->mut); - } - - /* then check if we need to add an assistance disk queue */ - if(pThis->bIsDA) - CHKiRet(qqueueChkStrtDA(pThis)); - - /* handle flow control - * There are two different flow control mechanisms: basic and advanced flow control. - * Basic flow control has always been implemented and protects the queue structures - * in that it makes sure no more data is enqueued than the queue is configured to - * support. Enhanced flow control is being added today. There are some sources which - * can easily be stopped, e.g. a file reader. This is the case because it is unlikely - * that blocking those sources will have negative effects (after all, the file is - * continued to be written). Other sources can somewhat be blocked (e.g. the kernel - * log reader or the local log stream reader): in general, nothing is lost if messages - * from these sources are not picked up immediately. HOWEVER, they can not block for - * an extended period of time, as this either causes message loss or - even worse - some - * other bad effects (e.g. unresponsive system in respect to the main system log socket). - * Finally, there are some (few) sources which can not be blocked at all. UDP syslog is - * a prime example. If a UDP message is not received, it is simply lost. So we can't - * do anything against UDP sockets that come in too fast. The core idea of advanced - * flow control is that we take into account the different natures of the sources and - * select flow control mechanisms that fit these needs. This also means, in the end - * result, that non-blockable sources like UDP syslog receive priority in the system. - * It's a side effect, but a good one ;) -- rgerhards, 2008-03-14 - */ - if(flowCtlType == eFLOWCTL_FULL_DELAY) { - while(pThis->iQueueSize >= pThis->iFullDlyMrk) { - dbgoprint((obj_t*) pThis, "enqueueMsg: FullDelay mark reached for full delayable message - blocking.\n"); - pthread_cond_wait(&pThis->belowFullDlyWtrMrk, pThis->mut); /* TODO error check? But what do then? */ - } - } else if(flowCtlType == eFLOWCTL_LIGHT_DELAY) { - if(pThis->iQueueSize >= pThis->iLightDlyMrk) { - dbgoprint((obj_t*) pThis, "enqueueMsg: LightDelay mark reached for light delayable message - blocking a bit.\n"); - timeoutComp(&t, 1000); /* 1000 millisconds = 1 second TODO: make configurable */ - pthread_cond_timedwait(&pThis->belowLightDlyWtrMrk, pThis->mut, &t); /* TODO error check? But what do then? */ - } - } - - /* from our regular flow control settings, we are now ready to enqueue the object. - * However, we now need to do a check if the queue permits to add more data. If that - * is not the case, basic flow control enters the field, which means we wait for - * the queue to become ready or drop the new message. -- rgerhards, 2008-03-14 - */ - while( (pThis->iMaxQueueSize > 0 && pThis->iQueueSize >= pThis->iMaxQueueSize) - || (pThis->qType == QUEUETYPE_DISK && pThis->sizeOnDiskMax != 0 - && pThis->tVars.disk.sizeOnDisk > pThis->sizeOnDiskMax)) { - dbgoprint((obj_t*) pThis, "enqueueMsg: queue FULL - waiting to drain.\n"); - timeoutComp(&t, pThis->toEnq); - if(pthread_cond_timedwait(&pThis->notFull, pThis->mut, &t) != 0) { - dbgoprint((obj_t*) pThis, "enqueueMsg: cond timeout, dropping message!\n"); - objDestruct(pUsr); - ABORT_FINALIZE(RS_RET_QUEUE_FULL); - } - } - - /* and finally enqueue the message */ - CHKiRet(qqueueAdd(pThis, pUsr)); - qqueueChkPersist(pThis); - -finalize_it: - if(pThis->qType != QUEUETYPE_DIRECT) { - /* make sure at least one worker is running. */ - qqueueAdviseMaxWorkers(pThis); - /* and release the mutex */ - d_pthread_mutex_unlock(pThis->mut); - pthread_setcancelstate(iCancelStateSave, NULL); - dbgoprint((obj_t*) pThis, "EnqueueMsg advised worker start\n"); - } - - RETiRet; -} - - -/* enqueue a single data object. This currently is a helper to qqueueMultiEnqObj. +/* enqueue a single data object. * Note that the queue mutex MUST already be locked when this function is called. * rgerhards, 2009-06-16 */ @@ -2242,12 +2180,8 @@ doEnqSingleObj(qqueue_t *pThis, flowControl_t flowCtlType, void *pUsr) /* first check if we need to discard this message (which will cause CHKiRet() to exit) */ - CHKiRet(qqueueChkDiscardMsg(pThis, pThis->iQueueSize, pThis->bRunsDA, pUsr)); + CHKiRet(qqueueChkDiscardMsg(pThis, pThis->iQueueSize, pUsr)); - /* then check if we need to add an assistance disk queue */ - if(pThis->bIsDA) - CHKiRet(qqueueChkStrtDA(pThis)); - /* handle flow control * There are two different flow control mechanisms: basic and advanced flow control. * Basic flow control has always been implemented and protects the queue structures @@ -2270,12 +2204,12 @@ doEnqSingleObj(qqueue_t *pThis, flowControl_t flowCtlType, void *pUsr) */ if(flowCtlType == eFLOWCTL_FULL_DELAY) { while(pThis->iQueueSize >= pThis->iFullDlyMrk) { - dbgoprint((obj_t*) pThis, "enqueueMsg: FullDelay mark reached for full delayable message - blocking.\n"); + DBGOPRINT((obj_t*) pThis, "enqueueMsg: FullDelay mark reached for full delayable message - blocking.\n"); pthread_cond_wait(&pThis->belowFullDlyWtrMrk, pThis->mut); /* TODO error check? But what do then? */ } } else if(flowCtlType == eFLOWCTL_LIGHT_DELAY) { if(pThis->iQueueSize >= pThis->iLightDlyMrk) { - dbgoprint((obj_t*) pThis, "enqueueMsg: LightDelay mark reached for light delayable message - blocking a bit.\n"); + DBGOPRINT((obj_t*) pThis, "enqueueMsg: LightDelay mark reached for light delayable message - blocking a bit.\n"); timeoutComp(&t, 1000); /* 1000 millisconds = 1 second TODO: make configurable */ pthread_cond_timedwait(&pThis->belowLightDlyWtrMrk, pThis->mut, &t); /* TODO error check? But what do then? */ } @@ -2289,10 +2223,11 @@ doEnqSingleObj(qqueue_t *pThis, flowControl_t flowCtlType, void *pUsr) while( (pThis->iMaxQueueSize > 0 && pThis->iQueueSize >= pThis->iMaxQueueSize) || (pThis->qType == QUEUETYPE_DISK && pThis->sizeOnDiskMax != 0 && pThis->tVars.disk.sizeOnDisk > pThis->sizeOnDiskMax)) { - dbgoprint((obj_t*) pThis, "enqueueMsg: queue FULL - waiting to drain.\n"); + DBGOPRINT((obj_t*) pThis, "enqueueMsg: queue FULL - waiting to drain.\n"); timeoutComp(&t, pThis->toEnq); +// TODO : handle enqOnly => discard! if(pthread_cond_timedwait(&pThis->notFull, pThis->mut, &t) != 0) { - dbgoprint((obj_t*) pThis, "enqueueMsg: cond timeout, dropping message!\n"); + DBGOPRINT((obj_t*) pThis, "enqueueMsg: cond timeout, dropping message!\n"); objDestruct(pUsr); ABORT_FINALIZE(RS_RET_QUEUE_FULL); } @@ -2300,12 +2235,12 @@ doEnqSingleObj(qqueue_t *pThis, flowControl_t flowCtlType, void *pUsr) /* and finally enqueue the message */ CHKiRet(qqueueAdd(pThis, pUsr)); - qqueueChkPersist(pThis); // TODO: optimize, do in outer function! (but we need parts from v5?) finalize_it: RETiRet; } +/* ------------------------------ multi-enqueue functions ------------------------------ */ /* enqueue multiple user data elements at once. The aim is to provide a faster interface * for object submission. Uses the multi_submit_t helper object. * Please note that this function is not cancel-safe and consequently @@ -2313,9 +2248,12 @@ finalize_it: * during its execution. If that is not done, race conditions occur if the * thread is canceled (most important use case is input module termination). * rgerhards, 2009-06-16 + * Note: there now exists multiple different functions implementing specially + * optimized algorithms for different config cases. -- rgerhards, 2010-06-09 */ -rsRetVal -qqueueMultiEnqObj(qqueue_t *pThis, multi_submit_t *pMultiSub) +/* now the function for all modes but direct */ +static rsRetVal +qqueueMultiEnqObjNonDirect(qqueue_t *pThis, multi_submit_t *pMultiSub) { int iCancelStateSave; int i; @@ -2324,78 +2262,89 @@ qqueueMultiEnqObj(qqueue_t *pThis, multi_submit_t *pMultiSub) ISOBJ_TYPE_assert(pThis, qqueue); assert(pMultiSub != NULL); - if(pThis->qType != QUEUETYPE_DIRECT) { - pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &iCancelStateSave); - d_pthread_mutex_lock(pThis->mut); - } - + pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &iCancelStateSave); + d_pthread_mutex_lock(pThis->mut); for(i = 0 ; i < pMultiSub->nElem ; ++i) { -dbgprintf("queueMultiEnq: %d\n", i); CHKiRet(doEnqSingleObj(pThis, pMultiSub->ppMsgs[i]->flowCtlType, (void*)pMultiSub->ppMsgs[i])); } + qqueueChkPersist(pThis, pMultiSub->nElem); finalize_it: - if(pThis->qType != QUEUETYPE_DIRECT) { - /* make sure at least one worker is running. */ - qqueueAdviseMaxWorkers(pThis); - /* and release the mutex */ - d_pthread_mutex_unlock(pThis->mut); - pthread_setcancelstate(iCancelStateSave, NULL); - dbgoprint((obj_t*) pThis, "MultiEnqObj advised worker start\n"); + /* make sure at least one worker is running. */ + qqueueAdviseMaxWorkers(pThis); + /* and release the mutex */ + d_pthread_mutex_unlock(pThis->mut); + pthread_setcancelstate(iCancelStateSave, NULL); + DBGOPRINT((obj_t*) pThis, "MultiEnqObj advised worker start\n"); + + RETiRet; +} + +/* now, the same function, but for direct mode */ +static rsRetVal +qqueueMultiEnqObjDirect(qqueue_t *pThis, multi_submit_t *pMultiSub) +{ + int i; + DEFiRet; + + ISOBJ_TYPE_assert(pThis, qqueue); + assert(pMultiSub != NULL); + + for(i = 0 ; i < pMultiSub->nElem ; ++i) { + CHKiRet(qAddDirect(pThis, (void*)pMultiSub->ppMsgs[i])); } +finalize_it: RETiRet; } +/* ------------------------------ END multi-enqueue functions ------------------------------ */ -/* set queue mode to enqueue only or not - * There is one subtle issue: this method may be called during queue - * construction or while it is running. In the former case, the queue - * mutex does not yet exist (it is NULL), while in the later case it - * must be locked. The function detects the state and operates as - * required. - * rgerhards, 2008-01-16 +/* enqueue a new user data element in direct mode + * NOTE/TODO: This is a TESTER/EXPERIEMENTAL, to be changed to better + * code later on (like multi submit!) 2010-06-10 + * Enqueues the new element and awakes worker thread. */ -static rsRetVal -qqueueSetEnqOnly(qqueue_t *pThis, int bEnqOnly, int bLockMutex) +rsRetVal +qqueueEnqObjDirect(qqueue_t *pThis, void *pUsr) { DEFiRet; - DEFVARS_mutexProtection; + ISOBJ_TYPE_assert(pThis, qqueue); + iRet = qAddDirect(pThis, pUsr); + RETiRet; +} + + +/* enqueue a new user data element + * Enqueues the new element and awakes worker thread. + */ +rsRetVal +qqueueEnqObj(qqueue_t *pThis, flowControl_t flowCtlType, void *pUsr) +{ + DEFiRet; + int iCancelStateSave; ISOBJ_TYPE_assert(pThis, qqueue); - /* for simplicity, we do one big mutex lock. This method is extremely seldom - * called, so that doesn't matter... -- rgerhards, 2008-01-16 - */ - if(pThis->mut != NULL) { - BEGIN_MTX_PROTECTED_OPERATIONS(pThis->mut, bLockMutex); + if(pThis->qType != QUEUETYPE_DIRECT) { + pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &iCancelStateSave); + d_pthread_mutex_lock(pThis->mut); } - if(bEnqOnly == pThis->bEnqOnly) - FINALIZE; /* no change, nothing to do */ - - if(pThis->bQueueStarted) { - /* we need to adjust queue operation only if we are not during initial param setup */ - if(bEnqOnly == 1) { - /* switch to enqueue-only mode */ - /* this means we need to terminate all workers - that's it... */ - dbgoprint((obj_t*) pThis, "switching to enqueue-only mode, terminating all worker threads\n"); - if(pThis->pWtpReg != NULL) - wtpWakeupAllWrkr(pThis->pWtpReg); - if(pThis->pWtpDA != NULL) - wtpWakeupAllWrkr(pThis->pWtpDA); - } else { - /* switch back to regular mode */ - ABORT_FINALIZE(RS_RET_NOT_IMPLEMENTED); /* we don't need this so far... */ - } - } + CHKiRet(doEnqSingleObj(pThis, flowCtlType, pUsr)); - pThis->bEnqOnly = bEnqOnly; + qqueueChkPersist(pThis, 1); finalize_it: - if(pThis->mut != NULL) { - END_MTX_PROTECTED_OPERATIONS(pThis->mut); + if(pThis->qType != QUEUETYPE_DIRECT) { + /* make sure at least one worker is running. */ + qqueueAdviseMaxWorkers(pThis); + /* and release the mutex */ + d_pthread_mutex_unlock(pThis->mut); + pthread_setcancelstate(iCancelStateSave, NULL); + DBGOPRINT((obj_t*) pThis, "EnqueueMsg advised worker start\n"); } + RETiRet; } @@ -2419,6 +2368,7 @@ DEFpropSetMeth(qqueue, iMinMsgsPerWrkr, int) DEFpropSetMeth(qqueue, bSaveOnShutdown, int) DEFpropSetMeth(qqueue, pUsr, void*) DEFpropSetMeth(qqueue, iDeqSlowdown, int) +DEFpropSetMeth(qqueue, iDeqBatchSize, int) DEFpropSetMeth(qqueue, sizeOnDiskMax, int64) @@ -2437,8 +2387,6 @@ static rsRetVal qqueueSetProperty(qqueue_t *pThis, var_t *pProp) if(isProp("iQueueSize")) { pThis->iQueueSize = pProp->val.num; - } else if(isProp("iUngottenObjs")) { - pThis->iUngottenObjs = pProp->val.num; } else if(isProp("tVars.disk.sizeOnDisk")) { pThis->tVars.disk.sizeOnDisk = pProp->val.num; } else if(isProp("tVars.disk.bytesRead")) { @@ -2464,6 +2412,8 @@ BEGINObjClassInit(qqueue, 1, OBJ_IS_CORE_MODULE) /* request objects we use */ CHKiRet(objUse(glbl, CORE_COMPONENT)); CHKiRet(objUse(strm, CORE_COMPONENT)); + CHKiRet(objUse(datetime, CORE_COMPONENT)); + CHKiRet(objUse(errmsg, CORE_COMPONENT)); /* now set our own handlers */ OBJSetMethodHandler(objMethod_SETPROPERTY, qqueueSetProperty); diff --git a/runtime/queue.h b/runtime/queue.h index aafdaa45..1c758134 100644 --- a/runtime/queue.h +++ b/runtime/queue.h @@ -27,8 +27,18 @@ #include <pthread.h> #include "obj.h" #include "wtp.h" +#include "batch.h" #include "stream.h" +/* support for the toDelete list */ +typedef struct toDeleteLst_s toDeleteLst_t; +struct toDeleteLst_s { + qDeqID deqID; + int nElemDeq; /* numbe of elements that were dequeued and as such must now be discarded */ + struct toDeleteLst_s *pNext; +}; + + /* queue types */ typedef enum { QUEUETYPE_FIXED_ARRAY = 0,/* a simple queue made out of a fixed (initially malloced) array fast but memoryhog */ @@ -44,24 +54,15 @@ typedef struct qLinkedList_S { } qLinkedList_t; -typedef struct qWrkThrd_s { - pthread_t thrdID; /* thread ID */ - qWrkCmd_t tCurrCmd; /* current command to be carried out by worker */ - obj_t *pUsr; /* current user object being processed (or NULL if none) */ - struct queue_s *pQueue; /* my queue (important if only the work thread instance is passed! */ - int iThrd; /* my worker thread array index */ - pthread_cond_t condInitDone; /* signaled when the thread startup is done (once per thread existance) */ - pthread_mutex_t mut; -} qWrkThrd_t; /* type for queue worker threads */ - /* the queue object */ -typedef struct queue_s { +struct queue_s { BEGINobjInstance; queueType_t qType; - bool bEnqOnly; /* does queue run in enqueue-only mode (1) or not (0)? */ - bool bSaveOnShutdown;/* persists everthing on shutdown (if DA!)? 1-yes, 0-no */ - bool bQueueStarted; /* has queueStart() been called on this queue? 1-yes, 0-no */ - bool bQueueInDestruction;/* 1 if queue is in destruction process, 0 otherwise */ + int nLogDeq; /* number of elements currently logically dequeued */ + int bShutdownImmediate; /* should all workers cease processing messages? */ + sbool bEnqOnly; /* does queue run in enqueue-only mode (1) or not (0)? */ + sbool bSaveOnShutdown;/* persists everthing on shutdown (if DA!)? 1-yes, 0-no */ + sbool bQueueStarted; /* has queueStart() been called on this queue? 1-yes, 0-no */ int iQueueSize; /* Current number of elements in the queue */ int iMaxQueueSize; /* how large can the queue grow? */ int iNumWorkerThreads;/* number of worker threads to use */ @@ -72,18 +73,20 @@ typedef struct queue_s { void *pUsr; /* a global, user-supplied pointer. Is passed back to consumer. */ int iUpdsSincePersist;/* nbr of queue updates since the last persist call */ int iPersistUpdCnt; /* persits queue info after this nbr of updates - 0 -> persist only on shutdown */ - bool bSyncQueueFiles;/* if working with files, sync them after each write? */ + sbool bSyncQueueFiles;/* if working with files, sync them after each write? */ int iHighWtrMrk; /* high water mark for disk-assisted memory queues */ int iLowWtrMrk; /* low water mark for disk-assisted memory queues */ int iDiscardMrk; /* if the queue is above this mark, low-severity messages are discarded */ int iFullDlyMrk; /* if the queue is above this mark, FULL_DELAYable message are put on hold */ int iLightDlyMrk; /* if the queue is above this mark, LIGHT_DELAYable message are put on hold */ int iDiscardSeverity;/* messages of this severity above are discarded on too-full queue */ - bool bNeedDelQIF; /* does the QIF file need to be deleted when queue becomes empty? */ + sbool bNeedDelQIF; /* does the QIF file need to be deleted when queue becomes empty? */ int toQShutdown; /* timeout for regular queue shutdown in ms */ int toActShutdown; /* timeout for long-running action shutdown in ms */ int toWrkShutdown; /* timeout for idle workers in ms, -1 means indefinite (0 is immediate) */ + toDeleteLst_t *toDeleteLst;/* this queue's to-delete list */ int toEnq; /* enqueue timeout */ + int iDeqBatchSize; /* max number of elements that shall be dequeued at once */ /* rate limiting settings (will be expanded) */ int iDeqSlowdown; /* slow down dequeue by specified nbr of microseconds */ /* end rate limiting */ @@ -97,19 +100,23 @@ typedef struct queue_s { * applied to detect user configuration errors (and tell me how should we detect what * the user really wanted...). -- rgerhards, 2008-04-02 */ - /* ane dequeue time window */ - rsRetVal (*pConsumer)(void *,void*); /* user-supplied consumer function for dequeued messages */ + /* end dequeue time window */ + rsRetVal (*pConsumer)(void *,batch_t*,int*); /* user-supplied consumer function for dequeued messages */ /* calling interface for pConsumer: arg1 is the global user pointer from this structure, arg2 is the - * user pointer that was dequeued (actual sample: for actions, arg1 is the pAction and arg2 is pointer - * to message) - * rgerhards, 2008-01-28 + * user pointer array that was dequeued (actual sample: for actions, arg1 is the pAction and arg2 + * is pointer to an array of message message pointers), arg3 is a pointer to an interger which is zero + * during normal operations and one if the consumer must urgently shut down. */ /* type-specific handlers (set during construction) */ rsRetVal (*qConstruct)(struct queue_s *pThis); rsRetVal (*qDestruct)(struct queue_s *pThis); rsRetVal (*qAdd)(struct queue_s *pThis, void *pUsr); - rsRetVal (*qDel)(struct queue_s *pThis, void **ppUsr); + rsRetVal (*qDeq)(struct queue_s *pThis, void **ppUsr); + rsRetVal (*qDel)(struct queue_s *pThis); /* end type-specific handler */ + /* public entry points (set during construction, permit to set best algorithm for params selected) */ + rsRetVal (*MultiEnq)(qqueue_t *pThis, multi_submit_t *pMultiSub); + /* end public entry points */ /* synchronization variables */ pthread_mutex_t mutThrdMgmt; /* mutex for the queue's thread management */ pthread_mutex_t *mut; /* mutex for enqueing and dequeueing messages */ @@ -117,7 +124,6 @@ typedef struct queue_s { pthread_cond_t belowFullDlyWtrMrk; /* below eFLOWCTL_FULL_DELAY watermark */ pthread_cond_t belowLightDlyWtrMrk; /* below eFLOWCTL_FULL_DELAY watermark */ pthread_cond_t condDAReady;/* signalled when the DA queue is fully initialized and ready for processing */ - int bChildIsDone; /* set to 1 when the child DA queue has finished processing, 0 otherwise */ int bThrdStateChanged; /* at least one thread state has changed if 1 */ /* end sync variables */ /* the following variables are always present, because they @@ -132,43 +138,35 @@ typedef struct queue_s { int iNumberFiles; /* how many files make up the queue? */ int64 iMaxFileSize; /* max size for a single queue file */ int64 sizeOnDiskMax; /* maximum size on disk allowed */ + qDeqID deqIDAdd; /* next dequeue ID to use during add to queue store */ + qDeqID deqIDDel; /* queue store delete position */ int bIsDA; /* is this queue disk assisted? */ - int bRunsDA; /* is this queue actually *running* disk assisted? */ struct queue_s *pqDA; /* queue for disk-assisted modes */ struct queue_s *pqParent;/* pointer to the parent (if this is a child queue) */ int bDAEnqOnly; /* EnqOnly setting for DA queue */ - /* some data elements for the queueUngetObj() functionality. This list should always be short - * and is always kept in memory - */ - qLinkedList_t *pUngetRoot; - qLinkedList_t *pUngetLast; - int iUngottenObjs; /* number of objects currently in the "ungotten" list */ /* now follow queueing mode specific data elements */ union { /* different data elements based on queue type (qType) */ struct { - long head, tail; + long deqhead, head, tail; void** pBuf; /* the queued user data structure */ } farray; struct { - qLinkedList_t *pRoot; + qLinkedList_t *pDeqRoot; + qLinkedList_t *pDelRoot; qLinkedList_t *pLast; } linklist; struct { int64 sizeOnDisk; /* current amount of disk space used */ int64 bytesRead; /* number of bytes read from current (undeleted!) file */ - strm_t *pWrite; /* current file to be written */ - strm_t *pRead; /* current file to be read */ + strm_t *pWrite; /* current file to be written */ + strm_t *pReadDeq; /* current file for dequeueing */ + strm_t *pReadDel; /* current file for deleting */ } disk; } tVars; DEF_ATOMIC_HELPER_MUT(mutQueueSize); -} qqueue_t; - -/* some symbolic constants for easier reference */ -#define QUEUE_MODE_ENQDEQ 0 -#define QUEUE_MODE_ENQONLY 1 + DEF_ATOMIC_HELPER_MUT(mutLogDeq); +}; -#define QUEUE_IDX_DA_WORKER 0 /* index for the DA worker (fixed) */ -#define QUEUE_PTR_DA_WORKER(x) (&((pThis)->pWrkThrds[0])) /* the define below is an "eternal" timeout for the timeout settings which require a value. * It is one day, which is not really eternal, but comes close to it if we think about @@ -179,13 +177,14 @@ typedef struct queue_s { /* prototypes */ rsRetVal qqueueDestruct(qqueue_t **ppThis); -rsRetVal qqueueMultiEnqObj(qqueue_t *pThis, multi_submit_t *pMultiSub); +rsRetVal qqueueEnqObjDirect(qqueue_t *pThis, void *pUsr); rsRetVal qqueueEnqObj(qqueue_t *pThis, flowControl_t flwCtlType, void *pUsr); rsRetVal qqueueStart(qqueue_t *pThis); rsRetVal qqueueSetMaxFileSize(qqueue_t *pThis, size_t iMaxFileSize); rsRetVal qqueueSetFilePrefix(qqueue_t *pThis, uchar *pszPrefix, size_t iLenPrefix); rsRetVal qqueueConstruct(qqueue_t **ppThis, queueType_t qType, int iWorkerThreads, - int iMaxQueueSize, rsRetVal (*pConsumer)(void*,void*)); + int iMaxQueueSize, rsRetVal (*pConsumer)(void*,batch_t*, int*)); +rsRetVal qqueueEnqObjDirectBatch(qqueue_t *pThis, batch_t *pBatch); PROTOTYPEObjClassInit(qqueue); PROTOTYPEpropSetMeth(qqueue, iPersistUpdCnt, int); PROTOTYPEpropSetMeth(qqueue, bSyncQueueFiles, int); @@ -204,6 +203,7 @@ PROTOTYPEpropSetMeth(qqueue, bSaveOnShutdown, int); PROTOTYPEpropSetMeth(qqueue, pUsr, void*); PROTOTYPEpropSetMeth(qqueue, iDeqSlowdown, int); PROTOTYPEpropSetMeth(qqueue, sizeOnDiskMax, int64); +PROTOTYPEpropSetMeth(qqueue, iDeqBatchSize, int); #define qqueueGetID(pThis) ((unsigned long) pThis) #endif /* #ifndef QUEUE_H_INCLUDED */ diff --git a/runtime/rsyslog.c b/runtime/rsyslog.c index c209ae30..a9794840 100644 --- a/runtime/rsyslog.c +++ b/runtime/rsyslog.c @@ -80,6 +80,8 @@ #include "prop.h" #include "rule.h" #include "ruleset.h" +#include "parser.h" +#include "strgen.h" #include "atomic.h" /* forward definitions */ @@ -152,8 +154,6 @@ rsrtInit(char **ppErrObj, obj_if_t *pObjIF) CHKiRet(propClassInit(NULL)); if(ppErrObj != NULL) *ppErrObj = "glbl"; CHKiRet(glblClassInit(NULL)); - if(ppErrObj != NULL) *ppErrObj = "datetime"; - CHKiRet(datetimeClassInit(NULL)); if(ppErrObj != NULL) *ppErrObj = "msg"; CHKiRet(msgClassInit(NULL)); if(ppErrObj != NULL) *ppErrObj = "ctok_token"; @@ -184,6 +184,10 @@ rsrtInit(char **ppErrObj, obj_if_t *pObjIF) CHKiRet(qqueueClassInit(NULL)); if(ppErrObj != NULL) *ppErrObj = "conf"; CHKiRet(confClassInit(NULL)); + if(ppErrObj != NULL) *ppErrObj = "parser"; + CHKiRet(parserClassInit(NULL)); + if(ppErrObj != NULL) *ppErrObj = "strgen"; + CHKiRet(strgenClassInit(NULL)); /* dummy "classes" */ if(ppErrObj != NULL) *ppErrObj = "str"; diff --git a/runtime/rsyslog.h b/runtime/rsyslog.h index 9dd45b38..34eaedca 100644 --- a/runtime/rsyslog.h +++ b/runtime/rsyslog.h @@ -27,6 +27,12 @@ #define INCLUDED_RSYSLOG_H /* ############################################################# * + * # Some constant values # * + * ############################################################# */ +#define CONST_LEN_TIMESTAMP_3164 15 /* number of chars (excluding \0!) in a RFC3164 timestamp */ +#define CONST_LEN_TIMESTAMP_3339 32 /* number of chars (excluding \0!) in a RFC3339 timestamp */ + +/* ############################################################# * * # Config Settings # * * ############################################################# */ #define RS_STRINGBUF_ALLOC_INCREMENT 128 @@ -40,7 +46,27 @@ #define CONF_TAG_BUFSIZE 32 #define CONF_HOSTNAME_BUFSIZE 32 #define CONF_PROP_BUFSIZE 16 /* should be close to sizeof(ptr) or lighly above it */ - +#define CONF_MIN_SIZE_FOR_COMPRESS 60 /* config param: minimum message size to try compression. The smaller + * the message, the less likely is any compression gain. We check for + * gain before we submit the message. But to do so we still need to + * do the (costly) compress() call. The following setting sets a size + * for which no call to compress() is done at all. This may result in + * a few more bytes being transmited but better overall performance. + * Note: I have not yet checked the minimum UDP packet size. It might be + * that we do not save anything by compressing very small messages, because + * UDP might need to pad ;) + * rgerhards, 2006-11-30 + */ + +#define CONF_OMOD_NUMSTRINGS_MAXSIZE 2 /* cache for pointers to output module buffer pointers. All + * rsyslog-provided plugins do NOT need more than two buffers. If + * more are needed (future developments, third-parties), rsyslog + * must be recompiled with a larger parameter. Hardcoding this + * saves us some overhead, both in runtime in code complexity. As + * it is doubtful if ever more than 2 parameters are needed, the + * approach taken here is considered appropriate. + * rgerhards, 2010-06-24 + */ /* ############################################################# * * # End Config Settings # * @@ -60,9 +86,29 @@ #endif +/* the rsyslog core provides information about present feature to plugins + * asking it. Below are feature-test macros which must be used to query + * features. Note that this must be powers of two, so that multiple queries + * can be combined. -- rgerhards, 2009-04-27 + */ +#define CORE_FEATURE_BATCHING 1 +/*#define CORE_FEATURE_whatever 2 ... and so on ... */ + +/* under Solaris (actually only SPARC), we need to redefine some types + * to be void, so that we get void* pointers. Otherwise, we will see + * alignment errors. + */ +/* some universal fixed size integer defines ... */ +typedef long long int64; +typedef long long unsigned uint64; +typedef int64 number_t; /* type to use for numbers - TODO: maybe an autoconf option? */ +typedef char intTiny; /* 0..127! */ +typedef unsigned char uintTiny; /* 0..255! */ + /* define some base data types */ typedef unsigned char uchar;/* get rid of the unhandy "unsigned char" */ +typedef struct aUsrp_s aUsrp_t; typedef struct thrdInfo thrdInfo_t; typedef struct obj_s obj_t; typedef struct ruleset_s ruleset_t; @@ -72,6 +118,7 @@ typedef struct NetAddr netAddr_t; typedef struct netstrms_s netstrms_t; typedef struct netstrm_s netstrm_t; typedef struct nssel_s nssel_t; +typedef struct nspoll_s nspoll_t; typedef enum nsdsel_waitOp_e nsdsel_waitOp_t; typedef struct nsd_ptcp_s nsd_ptcp_t; typedef struct nsd_gtls_s nsd_gtls_t; @@ -79,7 +126,10 @@ typedef struct nsd_gsspi_s nsd_gsspi_t; typedef struct nsd_nss_s nsd_nss_t; typedef struct nsdsel_ptcp_s nsdsel_ptcp_t; typedef struct nsdsel_gtls_s nsdsel_gtls_t; +typedef struct nsdpoll_ptcp_s nsdpoll_ptcp_t; +typedef struct wti_s wti_t; typedef struct msg msg_t; +typedef struct queue_s qqueue_t; typedef struct prop_s prop_t; typedef struct interface_s interface_t; typedef struct objInfo_s objInfo_t; @@ -92,7 +142,16 @@ typedef struct tcps_sess_s tcps_sess_t; typedef struct strmsrv_s strmsrv_t; typedef struct strms_sess_s strms_sess_t; typedef struct vmstk_s vmstk_t; +typedef struct batch_obj_s batch_obj_t; +typedef struct batch_s batch_t; +typedef struct wtp_s wtp_t; +typedef struct modInfo_s modInfo_t; +typedef struct parser_s parser_t; +typedef struct parserList_s parserList_t; +typedef struct strgen_s strgen_t; +typedef struct strgenList_s strgenList_t; typedef rsRetVal (*prsf_t)(struct vmstk_s*, int); /* pointer to a RainerScript function */ +typedef uint64 qDeqID; /* queue Dequeue order ID. 32 bits is considered dangerously few */ typedef struct tcpLstnPortList_s tcpLstnPortList_t; // TODO: rename? typedef struct strmLstnPortList_s strmLstnPortList_t; // TODO: rename? @@ -105,26 +164,23 @@ typedef struct strmLstnPortList_s strmLstnPortList_t; // TODO: rename? typedef void * obj_t_ptr; typedef void nsd_t; typedef void nsdsel_t; + typedef void nsdpoll_t; #else typedef obj_t *obj_t_ptr; typedef obj_t nsd_t; typedef obj_t nsdsel_t; + typedef obj_t nsdpoll_t; #endif -/* some universal 64 bit define... */ -typedef long long int64; -typedef long long unsigned uint64; -typedef int64 number_t; /* type to use for numbers - TODO: maybe an autoconf option? */ -typedef char intTiny; /* 0..127! */ -typedef uchar uintTiny; /* 0..255! */ - #ifdef __hpux typedef unsigned int u_int32_t; /* TODO: is this correct? */ typedef int socklen_t; #endif -typedef char bool; /* I intentionally use char, to keep it slim so that many fit into the CPU cache! */ +typedef struct epoll_event epoll_event_t; + +typedef char sbool; /* (small bool) I intentionally use char, to keep it slim so that many fit into the CPU cache! */ /* settings for flow control * TODO: is there a better place for them? -- rgerhards, 2008-03-14 @@ -364,7 +420,7 @@ enum rsRetVal_ /** return value. All methods return this if not specified oth RS_RET_RSCORE_TOO_OLD = -2120, /**< rsyslog core is too old for ... (eg this plugin) */ RS_RET_DEFER_COMMIT = -2121, /**< output plugin status: not yet committed (an OK state!) */ RS_RET_PREVIOUS_COMMITTED = -2122, /**< output plugin status: previous record was committed (an OK state!) */ - RS_RET_ACTION_FAILED = -2123, /**< action failed and is now suspended (consider this permanent for the time being) */ + RS_RET_ACTION_FAILED = -2123, /**< action failed and is now suspended */ RS_RET_NONFATAL_CONFIG_ERR = -2124, /**< non-fatal error during config processing */ RS_RET_NON_SIZELIMITCMD = -2125, /**< size limit for file defined, but no size limit command given */ RS_RET_SIZELIMITCMD_DIDNT_RESOLVE = -2126, /**< size limit command did not resolve situation */ @@ -377,6 +433,22 @@ enum rsRetVal_ /** return value. All methods return this if not specified oth RS_RET_ERR_OPEN_KLOG = -2145, /**< error opening the kernel log socket (primarily solaris) */ RS_RET_ERR_AQ_CONLOG = -2146, /**< error aquiring console log (on solaris) */ RS_RET_ERR_DOOR = -2147, /**< some problems with handling the Solaris door functionality */ + RS_RET_NO_SRCNAME_TPL = -2150, /**< sourcename template was not specified where one was needed (omudpspoof spoof addr) */ + RS_RET_HOST_NOT_SPECIFIED = -2151, /**< (target) host was not specified where it was needed */ + RS_RET_ERR_LIBNET_INIT = -2152, /**< error initializing libnet */ + RS_RET_FORCE_TERM = -2153, /**< thread was forced to terminate by bShallShutdown, a state, not an error */ + RS_RET_RULES_QUEUE_EXISTS = -2154,/**< we were instructed to create a new ruleset queue, but one already exists */ + RS_RET_NO_CURR_RULESET = -2155,/**< no current ruleset exists (but one is required) */ + RS_RET_NO_MSG_PASSING = -2156,/**< output module interface parameter passing mode "MSG" is not available but required */ + RS_RET_RULESET_NOT_FOUND = -2157,/**< a required ruleset could not be found */ + RS_RET_NO_RULESET= -2158,/**< no ruleset name as specified where one was needed */ + RS_RET_PARSER_NOT_FOUND = -2159,/**< parser with the specified name was not found */ + RS_RET_COULD_NOT_PARSE = -2160,/**< (this) parser could not parse the message (no error, means try next one) */ + RS_RET_EINTR = -2161, /**< EINTR occured during a system call (not necessarily an error) */ + RS_RET_ERR_EPOLL = -2162, /**< epoll() returned with an unexpected error code */ + RS_RET_ERR_EPOLL_CTL = -2163, /**< epol_ctll() returned with an unexpected error code */ + RS_RET_TIMEOUT = -2164, /**< timeout occured during operation */ + RS_RET_RCV_ERR = -2165, /**< error occured during socket rcv operation */ RS_RET_NO_SOCK_CONFIGURED = -2166, /**< no socket (name) was configured where one is required */ RS_RET_NO_LSTN_DEFINED = -2172, /**< no listener defined (e.g. inside an input module) */ RS_RET_EPOLL_CR_FAILED = -2173, /**< epoll_create() failed */ @@ -387,10 +459,12 @@ enum rsRetVal_ /** return value. All methods return this if not specified oth RS_RET_SYSVAR_NOT_FOUND = 1001, /**< system variable could not be found (maybe misspelled) */ /* some generic error/status codes */ + RS_RET_OK = 0, /**< operation successful */ RS_RET_OK_DELETE_LISTENTRY = 1, /**< operation successful, but callee requested the deletion of an entry (special state) */ RS_RET_TERMINATE_NOW = 2, /**< operation successful, function is requested to terminate (mostly used with threads) */ RS_RET_NO_RUN = 3, /**< operation successful, but function does not like to be executed */ - RS_RET_OK = 0 /**< operation successful */ + RS_RET_IDLE = 4, /**< operation successful, but callee is idle (e.g. because queue is empty) */ + RS_RET_TERMINATE_WHEN_IDLE = 5 /**< operation successful, function is requested to terminate when idle */ }; /* some helpful macros to work with srRetVals. diff --git a/runtime/rule.c b/runtime/rule.c index 4c2c9edb..42773768 100644 --- a/runtime/rule.c +++ b/runtime/rule.c @@ -39,8 +39,8 @@ #include "vm.h" #include "var.h" #include "srUtils.h" +#include "batch.h" #include "unicode-helper.h" -#include "dirty.h" /* for getFIOPName */ /* static data */ DEFobjStaticHelpers @@ -49,6 +49,35 @@ DEFobjCurrIf(expr) DEFobjCurrIf(var) DEFobjCurrIf(vm) + +/* support for simple textual representation of FIOP names + * rgerhards, 2005-09-27 + */ +static char* +getFIOPName(unsigned iFIOP) +{ + char *pRet; + switch(iFIOP) { + case FIOP_CONTAINS: + pRet = "contains"; + break; + case FIOP_ISEQUAL: + pRet = "isequal"; + break; + case FIOP_STARTSWITH: + pRet = "startswith"; + break; + case FIOP_REGEX: + pRet = "regex"; + break; + default: + pRet = "NOP"; + break; + } + return pRet; +} + + /* iterate over all actions, this is often needed, for example when HUP processing * must be done or a shutdown is pending. */ @@ -59,40 +88,20 @@ iterateAllActions(rule_t *pThis, rsRetVal (*pFunc)(void*, void*), void* pParam) } - /* helper to processMsg(), used to call the configured actions. It is * executed from within llExecFunc() of the action list. * rgerhards, 2007-08-02 */ -typedef struct processMsgDoActions_s { - int bPrevWasSuspended; /* was the previous action suspended? */ - msg_t *pMsg; -} processMsgDoActions_t; -DEFFUNC_llExecFunc(processMsgDoActions) +DEFFUNC_llExecFunc(processBatchDoActions) { DEFiRet; rsRetVal iRetMod; /* return value of module - we do not always pass that back */ action_t *pAction = (action_t*) pData; - processMsgDoActions_t *pDoActData = (processMsgDoActions_t*) pParam; - - assert(pAction != NULL); - - if((pAction->bExecWhenPrevSusp == 1) && (pDoActData->bPrevWasSuspended == 0)) { - dbgprintf("not calling action because the previous one is not suspended\n"); - ABORT_FINALIZE(RS_RET_OK); - } + batch_t *pBatch = (batch_t*) pParam; - iRetMod = actionCallAction(pAction, pDoActData->pMsg); - if(iRetMod == RS_RET_DISCARDMSG) { - ABORT_FINALIZE(RS_RET_DISCARDMSG); - } else if(iRetMod == RS_RET_SUSPENDED) { - /* indicate suspension for next module to be called */ - pDoActData->bPrevWasSuspended = 1; - } else { - pDoActData->bPrevWasSuspended = 0; - } + DBGPRINTF("Processing next action\n"); + iRetMod = pAction->submitToActQ(pAction, pBatch); -finalize_it: RETiRet; } @@ -101,7 +110,7 @@ finalize_it: * provided filter condition. */ static rsRetVal -shouldProcessThisMessage(rule_t *pRule, msg_t *pMsg, int *bProcessMsg) +shouldProcessThisMessage(rule_t *pRule, msg_t *pMsg, sbool *bProcessMsg) { DEFiRet; unsigned short pbMustBeFreed; @@ -250,26 +259,29 @@ finalize_it: -/* Process (consume) a received message. Calls the actions configured. +/* Process (consume) a batch of messages. Calls the actions configured. * rgerhards, 2005-10-13 */ static rsRetVal -processMsg(rule_t *pThis, msg_t *pMsg) +processBatch(rule_t *pThis, batch_t *pBatch) { - int bProcessMsg; - processMsgDoActions_t DoActData; + int i; DEFiRet; ISOBJ_TYPE_assert(pThis, rule); - assert(pMsg != NULL); - - /* first check the filters... */ - CHKiRet(shouldProcessThisMessage(pThis, pMsg, &bProcessMsg)); - if(bProcessMsg) { - DoActData.pMsg = pMsg; - DoActData.bPrevWasSuspended = 0; - CHKiRet(llExecFunc(&pThis->llActList, processMsgDoActions, (void*)&DoActData)); + assert(pBatch != NULL); + + /* first check the filters and reset status variables */ + for(i = 0 ; i < batchNumMsgs(pBatch) && !*(pBatch->pbShutdownImmediate) ; ++i) { + CHKiRet(shouldProcessThisMessage(pThis, (msg_t*)(pBatch->pElem[i].pUsrp), + &(pBatch->pElem[i].bFilterOK))); + // TODO: really abort on error? 2010-06-10 + if(pBatch->pElem[i].bFilterOK) { + /* re-init only when actually needed (cache write cost!) */ + pBatch->pElem[i].bPrevWasSuspended = 0; + } } + CHKiRet(llExecFunc(&pThis->llActList, processBatchDoActions, pBatch)); finalize_it: RETiRet; @@ -412,7 +424,7 @@ CODESTARTobjQueryInterface(rule) pIf->DebugPrint = ruleDebugPrint; pIf->IterateAllActions = iterateAllActions; - pIf->ProcessMsg = processMsg; + pIf->ProcessBatch = processBatch; pIf->SetAssRuleset = setAssRuleset; pIf->GetAssRuleset = getAssRuleset; finalize_it: diff --git a/runtime/rule.h b/runtime/rule.h index 99ac44e7..309a2ed8 100644 --- a/runtime/rule.h +++ b/runtime/rule.h @@ -47,7 +47,7 @@ struct rule_s { fiop_t operation; regex_t *regex_cache; /* cache for compiled REs, if such are used */ cstr_t *pCSCompValue; /* value to "compare" against */ - bool isNegated; + sbool isNegated; propid_t propID; /* ID of the requested property */ } prop; expr_t *f_expr; /* expression object */ @@ -64,11 +64,12 @@ BEGINinterface(rule) /* name must also be changed in ENDinterface macro! */ rsRetVal (*ConstructFinalize)(rule_t __attribute__((unused)) *pThis); rsRetVal (*Destruct)(rule_t **ppThis); rsRetVal (*IterateAllActions)(rule_t *pThis, rsRetVal (*pFunc)(void*, void*), void *pParam); - rsRetVal (*ProcessMsg)(rule_t *pThis, msg_t *pMsg); + rsRetVal (*ProcessBatch)(rule_t *pThis, batch_t *pBatch); rsRetVal (*SetAssRuleset)(rule_t *pThis, ruleset_t*); ruleset_t* (*GetAssRuleset)(rule_t *pThis); ENDinterface(rule) -#define ruleCURR_IF_VERSION 1 /* increment whenever you change the interface structure! */ +#define ruleCURR_IF_VERSION 2 /* increment whenever you change the interface structure! */ +/* change for v2: ProcessMsg replaced by ProcessBatch - 2010-06-10 */ /* prototypes */ diff --git a/runtime/ruleset.c b/runtime/ruleset.c index af61f24f..0584e8d6 100644 --- a/runtime/ruleset.c +++ b/runtime/ruleset.c @@ -40,22 +40,28 @@ #include "rsyslog.h" #include "obj.h" +#include "cfsysline.h" #include "msg.h" #include "ruleset.h" #include "rule.h" #include "errmsg.h" +#include "parser.h" +#include "batch.h" #include "unicode-helper.h" - -static rsRetVal debugPrintAll(void); // TODO: remove! +#include "dirty.h" /* for main ruleset queue creation */ /* static data */ DEFobjStaticHelpers DEFobjCurrIf(errmsg) DEFobjCurrIf(rule) +DEFobjCurrIf(parser) linkedList_t llRulesets; /* this is NOT a pointer - no typo here ;) */ ruleset_t *pCurrRuleset = NULL; /* currently "active" ruleset */ -ruleset_t *pDfltRuleset = NULL; /* currentl default ruleset, e.g. for binding to actions which have no other */ +ruleset_t *pDfltRuleset = NULL; /* current default ruleset, e.g. for binding to actions which have no other */ + +/* forward definitions */ +static rsRetVal processBatch(batch_t *pBatch); /* ---------- linked-list key handling functions ---------- */ @@ -132,39 +138,114 @@ finalize_it: -/* helper to processMsg(), used to call the configured actions. It is +/* helper to processBatch(), used to call the configured actions. It is * executed from within llExecFunc() of the action list. * rgerhards, 2007-08-02 */ -DEFFUNC_llExecFunc(processMsgDoRules) +DEFFUNC_llExecFunc(processBatchDoRules) { + rsRetVal iRet; ISOBJ_TYPE_assert(pData, rule); - return rule.ProcessMsg((rule_t*) pData, (msg_t*) pParam); + dbgprintf("Processing next rule\n"); + iRet = rule.ProcessBatch((rule_t*) pData, (batch_t*) pParam); +dbgprintf("ruleset: get iRet %d from rule.ProcessMsg()\n", iRet); + return iRet; } -/* Process (consume) a received message. Calls the actions configured. + +/* This function is similar to processBatch(), but works on a batch that + * contains rules from multiple rulesets. In this case, we can not push + * the whole batch through the ruleset. Instead, we examine it and + * partition it into sub-rulesets which we then push through the system. + * Note that when we evaluate which message must be processed, we do NOT need + * to look at bFilterOK, because this value is only set in a later processing + * stage. Doing so caused a bug during development ;) + * rgerhards, 2010-06-15 + */ +static inline rsRetVal +processBatchMultiRuleset(batch_t *pBatch) +{ + ruleset_t *currRuleset; + batch_t snglRuleBatch; + int i; + int iStart; /* start index of partial batch */ + int iNew; /* index for new (temporary) batch */ + DEFiRet; + + CHKiRet(batchInit(&snglRuleBatch, pBatch->nElem)); + snglRuleBatch.pbShutdownImmediate = pBatch->pbShutdownImmediate; + + while(1) { /* loop broken inside */ + /* search for first unprocessed element */ + for(iStart = 0 ; iStart < pBatch->nElem && pBatch->pElem[iStart].state == BATCH_STATE_DISC ; ++iStart) + /* just search, no action */; + + if(iStart == pBatch->nElem) + FINALIZE; /* everything processed */ + + /* prepare temporary batch */ + currRuleset = batchElemGetRuleset(pBatch, iStart); + iNew = 0; + for(i = iStart ; i < pBatch->nElem ; ++i) { + if(batchElemGetRuleset(pBatch, i) == currRuleset) { + batchCopyElem(&(snglRuleBatch.pElem[iNew++]), &(pBatch->pElem[i])); + /* We indicate the element also as done, so it will not be processed again */ + pBatch->pElem[i].state = BATCH_STATE_DISC; + } + } + snglRuleBatch.nElem = iNew; /* was left just right by the for loop */ + batchSetSingleRuleset(&snglRuleBatch, 1); + /* process temp batch */ + processBatch(&snglRuleBatch); + } + batchFree(&snglRuleBatch); + +finalize_it: + RETiRet; +} + +/* Process (consume) a batch of messages. Calls the actions configured. + * If the whole batch uses a singel ruleset, we can process the batch as + * a whole. Otherwise, we need to process it slower, on a message-by-message + * basis (what can be optimized to a per-ruleset basis) * rgerhards, 2005-10-13 */ static rsRetVal -processMsg(msg_t *pMsg) +processBatch(batch_t *pBatch) { ruleset_t *pThis; DEFiRet; - assert(pMsg != NULL); - - pThis = (pMsg->pRuleset == NULL) ? pDfltRuleset : pMsg->pRuleset; - ISOBJ_TYPE_assert(pThis, ruleset); - - CHKiRet(llExecFunc(&pThis->llRules, processMsgDoRules, pMsg)); + assert(pBatch != NULL); + +dbgprintf("ZZZ: processBatch: batch of %d elements must be processed\n", pBatch->nElem); + if(pBatch->bSingleRuleset) { + pThis = batchGetRuleset(pBatch); + if(pThis == NULL) + pThis = pDfltRuleset; + ISOBJ_TYPE_assert(pThis, ruleset); + CHKiRet(llExecFunc(&pThis->llRules, processBatchDoRules, pBatch)); + } else { + CHKiRet(processBatchMultiRuleset(pBatch)); + } finalize_it: - if(iRet == RS_RET_DISCARDMSG) - iRet = RS_RET_OK; - +dbgprintf("ruleset.ProcessMsg() returns %d\n", iRet); RETiRet; } + +/* return the ruleset-assigned parser list. NULL means use the default + * parser list. + * rgerhards, 2009-11-04 + */ +static parserList_t* +GetParserList(msg_t *pMsg) +{ + return (pMsg->pRuleset == NULL) ? pDfltRuleset->pParserLst : pMsg->pRuleset->pParserLst; +} + + /* Add a new rule to the end of the current rule set. We do a number * of checks and ignore the rule if it does not pass them. */ @@ -214,6 +295,19 @@ GetCurrent(void) } +/* get main queue associated with ruleset. If no ruleset-specifc main queue + * is set, the primary main message queue is returned. + * We use a non-standard calling interface, as nothing can go wrong and it + * is really much more natural to return the pointer directly. + */ +static qqueue_t* +GetRulesetQueue(ruleset_t *pThis) +{ + ISOBJ_TYPE_assert(pThis, ruleset); + return (pThis->pQueue == NULL) ? pMsgQueue : pThis->pQueue; +} + + /* Find the ruleset with the given name and return a pointer to its object. */ static rsRetVal @@ -319,6 +413,12 @@ finalize_it: BEGINobjDestruct(ruleset) /* be sure to specify the object type also in END and CODESTART macros! */ CODESTARTobjDestruct(ruleset) dbgprintf("destructing ruleset %p, name %p\n", pThis, pThis->pszName); + if(pThis->pQueue != NULL) { + qqueueDestruct(&pThis->pQueue); + } + if(pThis->pParserLst != NULL) { + parser.DestructParserList(&pThis->pParserLst); + } llDestroy(&pThis->llRules); free(pThis->pszName); ENDobjDestruct(ruleset) @@ -385,6 +485,81 @@ debugPrintAll(void) } +/* Create a ruleset-specific "main" queue for this ruleset. If one is already + * defined, an error message is emitted but nothing else is done. + * Note: we use the main message queue parameters for queue creation and access + * syslogd.c directly to obtain these. This is far from being perfect, but + * considered acceptable for the time being. + * rgerhards, 2009-10-27 + */ +static rsRetVal +rulesetCreateQueue(void __attribute__((unused)) *pVal, int *pNewVal) +{ + DEFiRet; + + if(pCurrRuleset == NULL) { + errmsg.LogError(0, RS_RET_NO_CURR_RULESET, "error: currently no specific ruleset specified, thus a " + "queue can not be added to it"); + ABORT_FINALIZE(RS_RET_NO_CURR_RULESET); + } + + if(pCurrRuleset->pQueue != NULL) { + errmsg.LogError(0, RS_RET_RULES_QUEUE_EXISTS, "error: ruleset already has a main queue, can not " + "add another one"); + ABORT_FINALIZE(RS_RET_RULES_QUEUE_EXISTS); + } + + if(pNewVal == 0) + FINALIZE; /* if it is turned off, we do not need to change anything ;) */ + + dbgprintf("adding a ruleset-specific \"main\" queue"); + CHKiRet(createMainQueue(&pCurrRuleset->pQueue, UCHAR_CONSTANT("ruleset"))); + +finalize_it: + RETiRet; +} + + +/* Add a ruleset specific parser to the ruleset. Note that adding the first + * parser automatically disables the default parsers. If they are needed as well, + * the must be added via explicit config directives. + * Note: this is the only spot in the code that requires the parser object. In order + * to solve some class init bootstrap sequence problems, we get the object handle here + * instead of during module initialization. Note that objUse() is capable of being + * called multiple times. + * rgerhards, 2009-11-04 + */ +static rsRetVal +rulesetAddParser(void __attribute__((unused)) *pVal, uchar *pName) +{ + parser_t *pParser; + DEFiRet; + + assert(pCurrRuleset != NULL); + + CHKiRet(objUse(parser, CORE_COMPONENT)); + iRet = parser.FindParser(&pParser, pName); + if(iRet == RS_RET_PARSER_NOT_FOUND) { + errmsg.LogError(0, RS_RET_PARSER_NOT_FOUND, "error: parser '%s' unknown at this time " + "(maybe defined too late in rsyslog.conf?)", pName); + ABORT_FINALIZE(RS_RET_NO_CURR_RULESET); + } else if(iRet != RS_RET_OK) { + errmsg.LogError(0, iRet, "error trying to find parser '%s'\n", pName); + FINALIZE; + } + + CHKiRet(parser.AddParserToList(&pCurrRuleset->pParserLst, pParser)); + + dbgprintf("added parser '%s' to ruleset '%s'\n", pName, pCurrRuleset->pszName); +RUNLOG_VAR("%p", pCurrRuleset->pParserLst); + +finalize_it: + d_free(pName); /* no longer needed */ + + RETiRet; +} + + /* queryInterface function * rgerhards, 2008-02-21 */ @@ -407,13 +582,15 @@ CODESTARTobjQueryInterface(ruleset) pIf->IterateAllActions = iterateAllActions; pIf->DestructAllActions = destructAllActions; pIf->AddRule = addRule; - pIf->ProcessMsg = processMsg; + pIf->ProcessBatch = processBatch; pIf->SetName = setName; pIf->DebugPrintAll = debugPrintAll; pIf->GetCurrent = GetCurrent; pIf->GetRuleset = GetRuleset; pIf->SetDefaultRuleset = SetDefaultRuleset; pIf->SetCurrRuleset = SetCurrRuleset; + pIf->GetRulesetQueue = GetRulesetQueue; + pIf->GetParserList = GetParserList; finalize_it: ENDobjQueryInterface(ruleset) @@ -425,6 +602,7 @@ BEGINObjClassExit(ruleset, OBJ_IS_CORE_MODULE) /* class, version */ llDestroy(&llRulesets); objRelease(errmsg, CORE_COMPONENT); objRelease(rule, CORE_COMPONENT); + objRelease(parser, CORE_COMPONENT); ENDObjClassExit(ruleset) @@ -443,6 +621,10 @@ BEGINObjClassInit(ruleset, 1, OBJ_IS_CORE_MODULE) /* class, version */ /* prepare global data */ CHKiRet(llInit(&llRulesets, rulesetDestructForLinkedList, keyDestruct, strcasecmp)); + + /* config file handlers */ + CHKiRet(regCfSysLineHdlr((uchar *)"rulesetparser", 0, eCmdHdlrGetWord, rulesetAddParser, NULL, NULL)); + CHKiRet(regCfSysLineHdlr((uchar *)"rulesetcreatemainqueue", 0, eCmdHdlrBinary, rulesetCreateQueue, NULL, NULL)); ENDObjClassInit(ruleset) /* vi:set ai: diff --git a/runtime/ruleset.h b/runtime/ruleset.h index 32571687..acebd17a 100644 --- a/runtime/ruleset.h +++ b/runtime/ruleset.h @@ -25,6 +25,7 @@ #ifndef INCLUDED_RULESET_H #define INCLUDED_RULESET_H +#include "queue.h" #include "linkedlist.h" /* the ruleset object */ @@ -32,6 +33,8 @@ struct ruleset_s { BEGINobjInstance; /* Data to implement generic object - MUST be the first data element! */ linkedList_t llRules; /* this is NOT a pointer - no typo here ;) */ uchar *pszName; /* name of our ruleset */ + qqueue_t *pQueue; /* "main" message queue, if the ruleset has its own (else NULL) */ + parserList_t *pParserLst;/* list of parsers to use for this ruleset */ }; /* interfaces */ @@ -45,13 +48,16 @@ BEGINinterface(ruleset) /* name must also be changed in ENDinterface macro! */ rsRetVal (*DestructAllActions)(void); rsRetVal (*AddRule)(ruleset_t *pThis, rule_t **ppRule); rsRetVal (*SetName)(ruleset_t *pThis, uchar *pszName); - rsRetVal (*ProcessMsg)(msg_t *pMsg); + rsRetVal (*ProcessBatch)(batch_t*); rsRetVal (*GetRuleset)(ruleset_t **ppThis, uchar*); rsRetVal (*SetDefaultRuleset)(uchar*); rsRetVal (*SetCurrRuleset)(uchar*); ruleset_t* (*GetCurrent)(void); + qqueue_t* (*GetRulesetQueue)(ruleset_t*); + /* v3, 2009-11-04 */ + parserList_t* (*GetParserList)(msg_t *); ENDinterface(ruleset) -#define rulesetCURR_IF_VERSION 1 /* increment whenever you change the interface structure! */ +#define rulesetCURR_IF_VERSION 4 /* increment whenever you change the interface structure! */ /* prototypes */ diff --git a/runtime/srUtils.h b/runtime/srUtils.h index 16766312..c4f73e16 100644 --- a/runtime/srUtils.h +++ b/runtime/srUtils.h @@ -110,30 +110,17 @@ rsRetVal getFileSize(uchar *pszName, off_t *pSize); /* some useful constants */ #define DEFVARS_mutexProtection\ - int iCancelStateSave; \ int bLockedOpIsLocked=0 #define BEGIN_MTX_PROTECTED_OPERATIONS(mut, bMustLock) \ if(bMustLock == LOCK_MUTEX) { \ - pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &iCancelStateSave); \ d_pthread_mutex_lock(mut); \ + assert(bLockedOpIsLocked == 0); \ bLockedOpIsLocked = 1; \ } #define END_MTX_PROTECTED_OPERATIONS(mut) \ if(bLockedOpIsLocked) { \ d_pthread_mutex_unlock(mut); \ - pthread_setcancelstate(iCancelStateSave, NULL); \ + bLockedOpIsLocked = 0; \ } -/* The unconditional versions of the macro always lock the mutex. They are preferred in - * complex scenarios, where the simple ones might get mixed up by multiple calls. - */ -#define DEFVARS_mutexProtection_uncond\ - int iCancelStateSave -#define BEGIN_MTX_PROTECTED_OPERATIONS_UNCOND(mut) \ - pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &iCancelStateSave); \ - d_pthread_mutex_lock(mut); -#define END_MTX_PROTECTED_OPERATIONS_UNCOND(mut) \ - d_pthread_mutex_unlock(mut); \ - pthread_setcancelstate(iCancelStateSave, NULL); - #endif diff --git a/runtime/srutils.c b/runtime/srutils.c index 1452c9b7..d357cd77 100644 --- a/runtime/srutils.c +++ b/runtime/srutils.c @@ -46,6 +46,9 @@ #include "srUtils.h" #include "obj.h" +#if _POSIX_TIMERS <= 0 +#include <sys/time.h> +#endif /* here we host some syslog specific names. There currently is no better place * to do it, but over here is also not ideal... -- rgerhards, 2008-02-14 @@ -158,7 +161,7 @@ uchar *srUtilStrDup(uchar *pOld, size_t len) assert(pOld != NULL); - if((pNew = malloc(len + 1)) != NULL) + if((pNew = MALLOC(len + 1)) != NULL) memcpy(pNew, pOld, len + 1); return pNew; @@ -197,7 +200,7 @@ int makeFileParentDirs(uchar *szFile, size_t lenFile, mode_t mode, assert(lenFile > 0); len = lenFile + 1; /* add one for '\0'-byte */ - if((pszWork = malloc(sizeof(uchar) * len)) == NULL) + if((pszWork = MALLOC(sizeof(uchar) * len)) == NULL) return -1; memcpy(pszWork, szFile, len); for(p = pszWork+1 ; *p ; p++) @@ -346,7 +349,7 @@ rsRetVal genFileName(uchar **ppName, uchar *pDirName, size_t lenDirName, uchar * } lenName = lenDirName + 1 + lenFName + lenBuf + 1; /* last +1 for \0 char! */ - if((pName = malloc(sizeof(uchar) * lenName)) == NULL) + if((pName = MALLOC(sizeof(uchar) * lenName)) == NULL) ABORT_FINALIZE(RS_RET_OUT_OF_MEMORY); /* got memory, now construct string */ @@ -392,10 +395,22 @@ int getNumberDigits(long lNum) rsRetVal timeoutComp(struct timespec *pt, long iTimeout) { +# if _POSIX_TIMERS <= 0 + struct timeval tv; +# endif + BEGINfunc assert(pt != NULL); /* compute timeout */ + +# if _POSIX_TIMERS > 0 + /* this is the "regular" code */ clock_gettime(CLOCK_REALTIME, pt); +# else + gettimeofday(&tv, NULL); + pt->tv_sec = tv.tv_sec; + pt->tv_nsec = tv.tv_usec * 1000; +# endif pt->tv_sec += iTimeout / 1000; pt->tv_nsec += (iTimeout % 1000) * 1000000; /* think INTEGER arithmetic! */ if(pt->tv_nsec > 999999999) { /* overrun? */ @@ -417,11 +432,21 @@ timeoutVal(struct timespec *pt) { struct timespec t; long iTimeout; - BEGINfunc +# if _POSIX_TIMERS <= 0 + struct timeval tv; +# endif + BEGINfunc assert(pt != NULL); /* compute timeout */ +# if _POSIX_TIMERS > 0 + /* this is the "regular" code */ clock_gettime(CLOCK_REALTIME, &t); +# else + gettimeofday(&tv, NULL); + t.tv_sec = tv.tv_sec; + t.tv_nsec = tv.tv_usec * 1000; +# endif iTimeout = (pt->tv_nsec - t.tv_nsec) / 1000000; iTimeout += (pt->tv_sec - t.tv_sec) * 1000; diff --git a/runtime/stream.c b/runtime/stream.c index 815176a2..6b3040d8 100644 --- a/runtime/stream.c +++ b/runtime/stream.c @@ -261,7 +261,7 @@ static rsRetVal strmOpenFile(strm_t *pThis) pThis->pszFName, pThis->lenFName, pThis->iCurrFNum, pThis->iFileNumDigits)); } else { if(pThis->pszDir == NULL) { - if((pThis->pszCurrFName = (uchar*) strdup((char*) pThis->pszFName)) == NULL) + if((pThis->pszCurrFName = ustrdup(pThis->pszFName)) == NULL) ABORT_FINALIZE(RS_RET_OUT_OF_MEMORY); } else { CHKiRet(genFileName(&pThis->pszCurrFName, pThis->pszDir, pThis->lenDir, @@ -626,7 +626,7 @@ static rsRetVal strmConstructFinalize(strm_t *pThis) * to make sure we can write out everything with a SINGLE api call! * We add another 128 bytes to take care of the gzip header and "all eventualities". */ - CHKmalloc(pThis->pZipBuf = (Bytef*) malloc(sizeof(uchar) * (pThis->sIOBufSize + 128))); + CHKmalloc(pThis->pZipBuf = (Bytef*) MALLOC(sizeof(uchar) * (pThis->sIOBufSize + 128))); } } @@ -658,7 +658,7 @@ static rsRetVal strmConstructFinalize(strm_t *pThis) pthread_cond_init(&pThis->isEmpty, 0); pThis->iCnt = pThis->iEnq = pThis->iDeq = 0; for(i = 0 ; i < STREAM_ASYNC_NUMBUFS ; ++i) { - CHKmalloc(pThis->asyncBuf[i].pBuf = (uchar*) malloc(sizeof(uchar) * pThis->sIOBufSize)); + CHKmalloc(pThis->asyncBuf[i].pBuf = (uchar*) MALLOC(sizeof(uchar) * pThis->sIOBufSize)); } pThis->pIOBuf = pThis->asyncBuf[0].pBuf; pThis->bStopWriter = 0; @@ -666,7 +666,7 @@ static rsRetVal strmConstructFinalize(strm_t *pThis) DBGPRINTF("ERROR: stream %p cold not create writer thread\n", pThis); } else { /* we work synchronously, so we need to alloc a fixed pIOBuf */ - CHKmalloc(pThis->pIOBuf = (uchar*) malloc(sizeof(uchar) * pThis->sIOBufSize)); + CHKmalloc(pThis->pIOBuf = (uchar*) MALLOC(sizeof(uchar) * pThis->sIOBufSize)); } finalize_it: @@ -924,7 +924,7 @@ asyncWriterThread(void *pPtr) { int iDeq; struct timespec t; - bool bTimedOut = 0; + sbool bTimedOut = 0; strm_t *pThis = (strm_t*) pPtr; ISOBJ_TYPE_assert(pThis, strm); @@ -937,7 +937,6 @@ asyncWriterThread(void *pPtr) while(1) { /* loop broken inside */ d_pthread_mutex_lock(&pThis->mut); -dbgprintf("XXX: asyncWriterThread iterating %s\n", pThis->pszFName); while(pThis->iCnt == 0) { if(pThis->bStopWriter) { pthread_cond_broadcast(&pThis->isEmpty); @@ -953,7 +952,6 @@ dbgprintf("XXX: asyncWriterThread iterating %s\n", pThis->pszFName); bTimedOut = 0; timeoutComp(&t, pThis->iFlushInterval * 2000); /* *1000 millisconds */ // TODO: check the 2000?!? if(pThis->bDoTimedWait) { -dbgprintf("asyncWriter thread going to timeout sleep\n"); if(pthread_cond_timedwait(&pThis->notEmpty, &pThis->mut, &t) != 0) { int err = errno; if(err == ETIMEDOUT) { @@ -967,16 +965,13 @@ dbgprintf("asyncWriter thread going to timeout sleep\n"); } } } else { -dbgprintf("asyncWriter thread going to eternal sleep\n"); d_pthread_cond_wait(&pThis->notEmpty, &pThis->mut); } -dbgprintf("asyncWriter woke up\n"); } bTimedOut = 0; /* we may have timed out, but there *is* work to do... */ iDeq = pThis->iDeq++ % STREAM_ASYNC_NUMBUFS; -dbgprintf("asyncWriter writes data\n"); doWriteInternal(pThis, pThis->asyncBuf[iDeq].pBuf, pThis->asyncBuf[iDeq].lenBuf); // TODO: error check????? 2009-07-06 @@ -1091,7 +1086,7 @@ doZipWrite(strm_t *pThis, uchar *pBuf, size_t lenBuf) { z_stream zstrm; int zRet; /* zlib return state */ - bool bzInitDone = FALSE; + sbool bzInitDone = FALSE; DEFiRet; assert(pThis != NULL); assert(pBuf != NULL); @@ -1385,7 +1380,7 @@ strmSetFName(strm_t *pThis, uchar *pszName, size_t iLenName) if(pThis->pszFName != NULL) free(pThis->pszFName); - if((pThis->pszFName = malloc(sizeof(uchar) * (iLenName + 1))) == NULL) + if((pThis->pszFName = MALLOC(sizeof(uchar) * (iLenName + 1))) == NULL) ABORT_FINALIZE(RS_RET_OUT_OF_MEMORY); memcpy(pThis->pszFName, pszName, iLenName + 1); /* always think about the \0! */ @@ -1412,7 +1407,7 @@ strmSetDir(strm_t *pThis, uchar *pszDir, size_t iLenDir) if(iLenDir < 1) ABORT_FINALIZE(RS_RET_FILE_PREFIX_MISSING); - CHKmalloc(pThis->pszDir = malloc(sizeof(uchar) * iLenDir + 1)); + CHKmalloc(pThis->pszDir = MALLOC(sizeof(uchar) * (iLenDir + 1))); memcpy(pThis->pszDir, pszDir, iLenDir + 1); /* always think about the \0! */ pThis->lenDir = iLenDir; @@ -1510,6 +1505,46 @@ finalize_it: } +/* duplicate a stream object excluding dynamic properties. This function is + * primarily meant to provide a duplicate that later on can be used to access + * the data. This is needed, for example, for a restart of the disk queue. + * Note that ConstructFinalize() is NOT called. So our caller may change some + * properties before finalizing things. + * rgerhards, 2009-05-26 + */ +rsRetVal +strmDup(strm_t *pThis, strm_t **ppNew) +{ + strm_t *pNew = NULL; + DEFiRet; + + ISOBJ_TYPE_assert(pThis, strm); + assert(ppNew != NULL); + + CHKiRet(strmConstruct(&pNew)); + pNew->sType = pThis->sType; + pNew->iCurrFNum = pThis->iCurrFNum; + CHKmalloc(pNew->pszFName = ustrdup(pThis->pszFName)); + pNew->lenFName = pThis->lenFName; + CHKmalloc(pNew->pszDir = ustrdup(pThis->pszDir)); + pNew->lenDir = pThis->lenDir; + pNew->tOperationsMode = pThis->tOperationsMode; + pNew->tOpenMode = pThis->tOpenMode; + pNew->iMaxFileSize = pThis->iMaxFileSize; + pNew->iMaxFiles = pThis->iMaxFiles; + pNew->iFileNumDigits = pThis->iFileNumDigits; + pNew->bDeleteOnClose = pThis->bDeleteOnClose; + pNew->iCurrOffs = pThis->iCurrOffs; + + *ppNew = pNew; + pNew = NULL; + +finalize_it: + if(pNew != NULL) + strmDestruct(&pNew); + + RETiRet; +} /* set a user write-counter. This counter is initialized to zero and * receives the number of bytes written. It is accurate only after a @@ -1625,6 +1660,7 @@ CODESTARTobjQueryInterface(strm) pIf->RecordEnd = strmRecordEnd; pIf->Serialize = strmSerialize; pIf->GetCurrOffset = strmGetCurrOffset; + pIf->Dup = strmDup; pIf->SetWCntr = strmSetWCntr; /* set methods */ pIf->SetbDeleteOnClose = strmSetbDeleteOnClose; diff --git a/runtime/stream.h b/runtime/stream.h index 369d5a0f..37e9d570 100644 --- a/runtime/stream.h +++ b/runtime/stream.h @@ -102,12 +102,12 @@ typedef struct strm_s { int64 iMaxFileSize;/* maximum size a file may grow to */ int iMaxFiles; /* maximum number of files if a circular mode is in use */ int iFileNumDigits;/* min number of digits to use in file number (only in circular mode) */ - bool bDeleteOnClose; /* set to 1 to auto-delete on close -- be careful with that setting! */ + sbool bDeleteOnClose; /* set to 1 to auto-delete on close -- be careful with that setting! */ int64 iCurrOffs;/* current offset */ int64 *pUsrWCntr; /* NULL or a user-provided counter that receives the nbr of bytes written since the last CntrSet() */ /* dynamic properties, valid only during file open, not to be persistet */ - bool bDisabled; /* should file no longer be written to? (currently set only if omfile file size limit fails) */ - bool bSync; /* sync this file after every write? */ + sbool bDisabled; /* should file no longer be written to? (currently set only if omfile file size limit fails) */ + sbool bSync; /* sync this file after every write? */ size_t sIOBufSize;/* size of IO buffer */ uchar *pszDir; /* Directory */ int lenDir; @@ -118,13 +118,13 @@ typedef struct strm_s { size_t iBufPtrMax; /* current max Ptr in Buffer (if partial read!) */ size_t iBufPtr; /* pointer into current buffer */ int iUngetC; /* char set via UngetChar() call or -1 if none set */ - bool bInRecord; /* if 1, indicates that we are currently writing a not-yet complete record */ + sbool bInRecord; /* if 1, indicates that we are currently writing a not-yet complete record */ int iZipLevel; /* zip level (0..9). If 0, zip is completely disabled */ Bytef *pZipBuf; /* support for async flush procesing */ - bool bAsyncWrite; /* do asynchronous writes (always if a flush interval is given) */ - bool bStopWriter; /* shall writer thread terminate? */ - bool bDoTimedWait; /* instruct writer thread to do a times wait to support flush timeouts */ + sbool bAsyncWrite; /* do asynchronous writes (always if a flush interval is given) */ + sbool bStopWriter; /* shall writer thread terminate? */ + sbool bDoTimedWait; /* instruct writer thread to do a times wait to support flush timeouts */ int iFlushInterval; /* flush in which interval - 0, no flushing */ apc_id_t apcID; /* id of current Apc request (used for cancelling) */ pthread_mutex_t mut;/* mutex for flush in async mode */ @@ -143,7 +143,7 @@ typedef struct strm_s { /* support for omfile size-limiting commands, special counters, NOT persisted! */ off_t iSizeLimit; /* file size limit, 0 = no limit */ uchar *pszSizeLimitCmd; /* command to carry out when size limit is reached */ - bool bIsTTY; /* is this a tty file? */ + sbool bIsTTY; /* is this a tty file? */ } strm_t; @@ -169,6 +169,7 @@ BEGINinterface(strm) /* name must also be changed in ENDinterface macro! */ rsRetVal (*Serialize)(strm_t *pThis, strm_t *pStrm); rsRetVal (*GetCurrOffset)(strm_t *pThis, int64 *pOffs); rsRetVal (*SetWCntr)(strm_t *pThis, number_t *pWCnt); + rsRetVal (*Dup)(strm_t *pThis, strm_t **ppNew); INTERFACEpropSetMeth(strm, bDeleteOnClose, int); INTERFACEpropSetMeth(strm, iMaxFileSize, int); INTERFACEpropSetMeth(strm, iMaxFiles, int); @@ -183,7 +184,7 @@ BEGINinterface(strm) /* name must also be changed in ENDinterface macro! */ INTERFACEpropSetMeth(strm, iFlushInterval, int); INTERFACEpropSetMeth(strm, pszSizeLimitCmd, uchar*); ENDinterface(strm) -#define strmCURR_IF_VERSION 2 /* increment whenever you change the interface structure! */ +#define strmCURR_IF_VERSION 5 /* increment whenever you change the interface structure! */ /* prototypes */ diff --git a/runtime/strgen.c b/runtime/strgen.c new file mode 100644 index 00000000..46be1236 --- /dev/null +++ b/runtime/strgen.c @@ -0,0 +1,279 @@ +/* strgen.c + * Module to handle string generators. These are C modules that receive + * the message object and return a custom-built string. The primary purpose + * for their existance is performance -- they do the same as template strings, but + * potentially faster (if well implmented). + * + * Module begun 2010-06-01 by Rainer Gerhards + * + * Copyright 2010 Rainer Gerhards and Adiscon GmbH. + * + * This file is part of the rsyslog runtime library. + * + * The rsyslog runtime library is free software: you can redistribute it and/or modify + * it under the terms of the GNU Lesser General Public License as published by + * the Free Software Foundation, either version 3 of the License, or + * (at your option) any later version. + * + * The rsyslog runtime library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public License + * along with the rsyslog runtime library. If not, see <http://www.gnu.org/licenses/>. + * + * A copy of the GPL can be found in the file "COPYING" in this distribution. + * A copy of the LGPL can be found in the file "COPYING.LESSER" in this distribution. + */ +#include "config.h" +#include <stdlib.h> +#include <string.h> +#include <assert.h> + +#include "rsyslog.h" +#include "msg.h" +#include "obj.h" +#include "errmsg.h" +#include "strgen.h" +#include "ruleset.h" +#include "unicode-helper.h" +#include "cfsysline.h" + +/* definitions for objects we access */ +DEFobjStaticHelpers +DEFobjCurrIf(glbl) +DEFobjCurrIf(errmsg) +DEFobjCurrIf(ruleset) + +/* static data */ + +/* config data */ + +/* This is the list of all strgens known to us. + * This is also used to unload all modules on shutdown. + */ +strgenList_t *pStrgenLstRoot = NULL; + + +/* intialize (but NOT allocate) a strgen list. Primarily meant as a hook + * which can be used to extend the list in the future. So far, just sets + * it to NULL. + */ +static rsRetVal +InitStrgenList(strgenList_t **pListRoot) +{ + *pListRoot = NULL; + return RS_RET_OK; +} + + +/* destruct a strgen list. The list elements are destroyed, but the strgen objects + * themselves are not modified. (That is done at a late stage during rsyslogd + * shutdown and need not be considered here.) + */ +static rsRetVal +DestructStrgenList(strgenList_t **ppListRoot) +{ + strgenList_t *pStrgenLst; + strgenList_t *pStrgenLstDel; + + pStrgenLst = *ppListRoot; + while(pStrgenLst != NULL) { + pStrgenLstDel = pStrgenLst; + pStrgenLst = pStrgenLst->pNext; + free(pStrgenLstDel); + } + *ppListRoot = NULL; + return RS_RET_OK; +} + + +/* Add a strgen to the list. We use a VERY simple and ineffcient algorithm, + * but it is employed only for a few milliseconds during config processing. So + * I prefer to keep it very simple and with simple data structures. Unfortunately, + * we need to preserve the order, but I don't like to add a tail pointer as that + * would require a container object. So I do the extra work to skip to the tail + * when adding elements... + */ +static rsRetVal +AddStrgenToList(strgenList_t **ppListRoot, strgen_t *pStrgen) +{ + strgenList_t *pThis; + strgenList_t *pTail; + DEFiRet; + + CHKmalloc(pThis = MALLOC(sizeof(strgenList_t))); + pThis->pStrgen = pStrgen; + pThis->pNext = NULL; + + if(*ppListRoot == NULL) { + pThis->pNext = *ppListRoot; + *ppListRoot = pThis; + } else { + /* find tail first */ + for(pTail = *ppListRoot ; pTail->pNext != NULL ; pTail = pTail->pNext) + /* just search, do nothing else */; + /* add at tail */ + pTail->pNext = pThis; + } + +finalize_it: + RETiRet; +} + + +/* find a strgen based on the provided name */ +static rsRetVal +FindStrgen(strgen_t **ppStrgen, uchar *pName) +{ + strgenList_t *pThis; + DEFiRet; + + for(pThis = pStrgenLstRoot ; pThis != NULL ; pThis = pThis->pNext) { + if(ustrcmp(pThis->pStrgen->pName, pName) == 0) { + *ppStrgen = pThis->pStrgen; + FINALIZE; /* found it, iRet still eq. OK! */ + } + } + + iRet = RS_RET_PARSER_NOT_FOUND; + +finalize_it: + RETiRet; +} + + +/* --- END helper functions for strgen list handling --- */ + + +BEGINobjConstruct(strgen) /* be sure to specify the object type also in END macro! */ +ENDobjConstruct(strgen) + +/* ConstructionFinalizer. The most important chore is to add the strgen object + * to our global list of available strgens. + * rgerhards, 2009-11-03 + */ +rsRetVal strgenConstructFinalize(strgen_t *pThis) +{ + DEFiRet; + + ISOBJ_TYPE_assert(pThis, strgen); + CHKiRet(AddStrgenToList(&pStrgenLstRoot, pThis)); + DBGPRINTF("Strgen '%s' added to list of available strgens.\n", pThis->pName); + +finalize_it: + RETiRet; +} + +BEGINobjDestruct(strgen) /* be sure to specify the object type also in END and CODESTART macros! */ +CODESTARTobjDestruct(strgen) + dbgprintf("destructing strgen '%s'\n", pThis->pName); + free(pThis->pName); +ENDobjDestruct(strgen) + +/* set the strgen name - string is copied over, call can continue to use it, + * but must free it if desired. + */ +static rsRetVal +SetName(strgen_t *pThis, uchar *name) +{ + DEFiRet; + + ISOBJ_TYPE_assert(pThis, strgen); + assert(name != NULL); + + if(pThis->pName != NULL) { + free(pThis->pName); + pThis->pName = NULL; + } + + CHKmalloc(pThis->pName = ustrdup(name)); + +finalize_it: + RETiRet; +} + + +/* set a pointer to "our" module. Note that no module + * pointer must already be set. + */ +static rsRetVal +SetModPtr(strgen_t *pThis, modInfo_t *pMod) +{ + ISOBJ_TYPE_assert(pThis, strgen); + assert(pMod != NULL); + assert(pThis->pModule == NULL); + pThis->pModule = pMod; + return RS_RET_OK; +} + + +/* queryInterface function-- rgerhards, 2009-11-03 + */ +BEGINobjQueryInterface(strgen) +CODESTARTobjQueryInterface(strgen) + if(pIf->ifVersion != strgenCURR_IF_VERSION) { /* check for current version, increment on each change */ + ABORT_FINALIZE(RS_RET_INTERFACE_NOT_SUPPORTED); + } + + /* ok, we have the right interface, so let's fill it + * Please note that we may also do some backwards-compatibility + * work here (if we can support an older interface version - that, + * of course, also affects the "if" above). + */ + pIf->Construct = strgenConstruct; + pIf->ConstructFinalize = strgenConstructFinalize; + pIf->Destruct = strgenDestruct; + pIf->SetName = SetName; + pIf->SetModPtr = SetModPtr; + pIf->InitStrgenList = InitStrgenList; + pIf->DestructStrgenList = DestructStrgenList; + pIf->AddStrgenToList = AddStrgenToList; + pIf->FindStrgen = FindStrgen; +finalize_it: +ENDobjQueryInterface(strgen) + + +/* This destroys the master strgenlist and all of its strgen entries. MUST only be + * done when the module is shut down. Strgen modules are NOT unloaded, rsyslog + * does that at a later stage for all dynamically loaded modules. + */ +static void +destroyMasterStrgenList(void) +{ + strgenList_t *pStrgenLst; + strgenList_t *pStrgenLstDel; + + pStrgenLst = pStrgenLstRoot; + while(pStrgenLst != NULL) { + strgenDestruct(&pStrgenLst->pStrgen); + pStrgenLstDel = pStrgenLst; + pStrgenLst = pStrgenLst->pNext; + free(pStrgenLstDel); + } +} + +/* Exit our class. + * rgerhards, 2009-11-04 + */ +BEGINObjClassExit(strgen, OBJ_IS_CORE_MODULE) /* class, version */ + destroyMasterStrgenList(); + objRelease(glbl, CORE_COMPONENT); + objRelease(errmsg, CORE_COMPONENT); + objRelease(ruleset, CORE_COMPONENT); +ENDObjClassExit(strgen) + + +/* Initialize the strgen class. Must be called as the very first method + * before anything else is called inside this class. + * rgerhards, 2009-11-02 + */ +BEGINObjClassInit(strgen, 1, OBJ_IS_CORE_MODULE) /* class, version */ + /* request objects we use */ + CHKiRet(objUse(glbl, CORE_COMPONENT)); + CHKiRet(objUse(errmsg, CORE_COMPONENT)); + CHKiRet(objUse(ruleset, CORE_COMPONENT)); + InitStrgenList(&pStrgenLstRoot); +ENDObjClassInit(strgen) + diff --git a/runtime/strgen.h b/runtime/strgen.h new file mode 100644 index 00000000..3819dccd --- /dev/null +++ b/runtime/strgen.h @@ -0,0 +1,60 @@ +/* header for strgen.c + * + * Copyright 2010 Rainer Gerhards and Adiscon GmbH. + * + * This file is part of the rsyslog runtime library. + * + * The rsyslog runtime library is free software: you can redistribute it and/or modify + * it under the terms of the GNU Lesser General Public License as published by + * the Free Software Foundation, either version 3 of the License, or + * (at your option) any later version. + * + * The rsyslog runtime library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public License + * along with the rsyslog runtime library. If not, see <http://www.gnu.org/licenses/>. + * + * A copy of the LGPL can be found in the file "COPYING.LESSER" in this distribution. + */ +#ifndef INCLUDED_STRGEN_H +#define INCLUDED_STRGEN_H + + +/* we create a small helper object, a list of strgens, that we can use to + * build a chain of them whereever this is needed. + */ +struct strgenList_s { + strgen_t *pStrgen; + strgenList_t *pNext; +}; + + +/* the strgen object, a dummy because we have only static methods */ +struct strgen_s { + BEGINobjInstance; /* Data to implement generic object - MUST be the first data element! */ + uchar *pName; /* name of this strgen */ + modInfo_t *pModule; /* pointer to strgen's module */ +}; + +/* interfaces */ +BEGINinterface(strgen) /* name must also be changed in ENDinterface macro! */ + rsRetVal (*Construct)(strgen_t **ppThis); + rsRetVal (*ConstructFinalize)(strgen_t *pThis); + rsRetVal (*Destruct)(strgen_t **ppThis); + rsRetVal (*SetName)(strgen_t *pThis, uchar *name); + rsRetVal (*SetModPtr)(strgen_t *pThis, modInfo_t *pMod); + rsRetVal (*FindStrgen)(strgen_t **ppThis, uchar*name); + rsRetVal (*InitStrgenList)(strgenList_t **pListRoot); + rsRetVal (*DestructStrgenList)(strgenList_t **pListRoot); + rsRetVal (*AddStrgenToList)(strgenList_t **pListRoot, strgen_t *pStrgen); +ENDinterface(strgen) +#define strgenCURR_IF_VERSION 1 /* increment whenever you change the interface above! */ + + +/* prototypes */ +PROTOTYPEObj(strgen); + +#endif /* #ifndef INCLUDED_STRGEN_H */ diff --git a/runtime/stringbuf.c b/runtime/stringbuf.c index 93995b38..ccf115c1 100644 --- a/runtime/stringbuf.c +++ b/runtime/stringbuf.c @@ -90,7 +90,7 @@ rsRetVal rsCStrConstructFromszStr(cstr_t **ppThis, uchar *sz) CHKiRet(rsCStrConstruct(&pThis)); pThis->iBufSize = pThis->iStrLen = strlen((char *) sz); - if((pThis->pBuf = (uchar*) malloc(sizeof(uchar) * pThis->iStrLen)) == NULL) { + if((pThis->pBuf = (uchar*) MALLOC(sizeof(uchar) * pThis->iStrLen)) == NULL) { RSFREEOBJ(pThis); ABORT_FINALIZE(RS_RET_OUT_OF_MEMORY); } @@ -119,7 +119,7 @@ rsRetVal rsCStrConstructFromCStr(cstr_t **ppThis, cstr_t *pFrom) CHKiRet(rsCStrConstruct(&pThis)); pThis->iBufSize = pThis->iStrLen = pFrom->iStrLen; - if((pThis->pBuf = (uchar*) malloc(sizeof(uchar) * pThis->iStrLen)) == NULL) { + if((pThis->pBuf = (uchar*) MALLOC(sizeof(uchar) * pThis->iStrLen)) == NULL) { RSFREEOBJ(pThis); ABORT_FINALIZE(RS_RET_OUT_OF_MEMORY); } @@ -267,7 +267,7 @@ rsRetVal rsCStrSetSzStr(cstr_t *pThis, uchar *pszNew) pThis->pszBuf = NULL; /* now save the new value */ - if((pThis->pBuf = (uchar*) malloc(sizeof(uchar) * pThis->iStrLen)) == NULL) { + if((pThis->pBuf = (uchar*) MALLOC(sizeof(uchar) * pThis->iStrLen)) == NULL) { RSFREEOBJ(pThis); return RS_RET_OUT_OF_MEMORY; } @@ -315,7 +315,7 @@ uchar* rsCStrGetSzStr(cstr_t *pThis) if(pThis->pBuf != NULL) if(pThis->pszBuf == NULL) { /* we do not yet have a usable sz version - so create it... */ - if((pThis->pszBuf = malloc((pThis->iStrLen + 1) * sizeof(uchar))) == NULL) { + if((pThis->pszBuf = MALLOC((pThis->iStrLen + 1) * sizeof(uchar))) == NULL) { /* TODO: think about what to do - so far, I have no bright * idea... rgerhards 2005-09-07 */ @@ -369,7 +369,7 @@ rsRetVal cstrConvSzStrAndDestruct(cstr_t *pThis, uchar **ppSz, int bRetNULL) if(pThis->pBuf == NULL) { if(bRetNULL == 0) { - CHKmalloc(pRetBuf = malloc(sizeof(uchar))); + CHKmalloc(pRetBuf = MALLOC(sizeof(uchar))); *pRetBuf = '\0'; } else { pRetBuf = NULL; diff --git a/runtime/strmsrv.c b/runtime/strmsrv.c index 3dc53a97..a122ca8a 100644 --- a/runtime/strmsrv.c +++ b/runtime/strmsrv.c @@ -165,7 +165,7 @@ addNewLstnPort(strmsrv_t *pThis, uchar *pszPort) ISOBJ_TYPE_assert(pThis, strmsrv); /* create entry */ - CHKmalloc(pEntry = malloc(sizeof(strmLstnPortList_t))); + CHKmalloc(pEntry = MALLOC(sizeof(strmLstnPortList_t))); pEntry->pszPort = pszPort; pEntry->pSrv = pThis; CHKmalloc(pEntry->pszInputName = ustrdup(pThis->pszInputName)); diff --git a/runtime/sync.c b/runtime/sync.c index a3053e28..15d5c80f 100644 --- a/runtime/sync.c +++ b/runtime/sync.c @@ -28,12 +28,13 @@ #include "rsyslog.h" #include "sync.h" +#include "debug.h" void SyncObjInit(pthread_mutex_t **mut) { - *mut = (pthread_mutex_t *) malloc (sizeof (pthread_mutex_t)); + *mut = (pthread_mutex_t *) MALLOC(sizeof (pthread_mutex_t)); pthread_mutex_init(*mut, NULL); } diff --git a/runtime/syslogd-types.h b/runtime/syslogd-types.h index 4a26f993..fcebd985 100644 --- a/runtime/syslogd-types.h +++ b/runtime/syslogd-types.h @@ -56,7 +56,10 @@ * applications I do not yet envision. -- rgerhards, 2007-07-24 */ typedef enum _syslogFeature { - sFEATURERepeatedMsgReduction = 1 + sFEATURERepeatedMsgReduction = 1, /* for output modules */ + sFEATURENonCancelInputTermination = 2, /* for input modules */ + sFEATUREAutomaticSanitazion = 3, /* for parser modules */ + sFEATUREAutomaticPRIParsing = 4 /* for parser modules */ } syslogFeature; /* we define our own facility and severities */ diff --git a/runtime/wti.c b/runtime/wti.c index 90bb14ed..9343f5c5 100644 --- a/runtime/wti.c +++ b/runtime/wti.c @@ -39,10 +39,6 @@ #include <pthread.h> #include <errno.h> -#ifdef OS_SOLARIS -# include <sched.h> -#endif - #include "rsyslog.h" #include "stringbuf.h" #include "srUtils.h" @@ -75,92 +71,72 @@ wtiGetDbgHdr(wti_t *pThis) } -/* get the current worker state. For simplicity and speed, we have - * NOT used our regular calling interface this time. I hope that won't - * bite in the long term... -- rgerhards, 2008-01-17 - * TODO: may be performance optimized by atomic operations +/* return the current worker processing state. For the sake of + * simplicity, we do not use the iRet interface. -- rgerhards, 2009-07-17 */ -qWrkCmd_t -wtiGetState(wti_t *pThis, int bLockMutex) +sbool +wtiGetState(wti_t *pThis) { - DEFVARS_mutexProtection; - qWrkCmd_t tCmd; + return ATOMIC_FETCH_32BIT(&pThis->bIsRunning, &pThis->mutIsRunning); +} - BEGINfunc - ISOBJ_TYPE_assert(pThis, wti); - BEGIN_MTX_PROTECTED_OPERATIONS(&pThis->mut, bLockMutex); - tCmd = pThis->tCurrCmd; - END_MTX_PROTECTED_OPERATIONS(&pThis->mut); +/* Set this thread to "always running" state (can not be unset) + * rgerhards, 2009-07-20 + */ +rsRetVal +wtiSetAlwaysRunning(wti_t *pThis) +{ + ISOBJ_TYPE_assert(pThis, wti); + pThis->bAlwaysRunning = TRUE; + return RS_RET_OK; +} - ENDfunc - return tCmd; +/* Set status (thread is running or not), actually an property of + * use for wtp, but we need to have it per thread instance (thus it + * is inside wti). -- rgerhards, 2009-07-17 + */ +rsRetVal +wtiSetState(wti_t *pThis, sbool bNewVal) +{ + ISOBJ_TYPE_assert(pThis, wti); + if(bNewVal) { + ATOMIC_STORE_1_TO_INT(&pThis->bIsRunning, &pThis->mutIsRunning); + } else { + ATOMIC_STORE_0_TO_INT(&pThis->bIsRunning, &pThis->mutIsRunning); + } + return RS_RET_OK; } -/* send a command to a specific thread - * bActiveOnly specifies if the command should be sent only when the worker is - * in an active state. -- rgerhards, 2008-01-20 +/* advise all workers to start by interrupting them. That should unblock all srSleep() + * calls. */ rsRetVal -wtiSetState(wti_t *pThis, qWrkCmd_t tCmd, int bActiveOnly, int bLockMutex) +wtiWakeupThrd(wti_t *pThis) { DEFiRet; - qWrkCmd_t tCurrCmd; - DEFVARS_mutexProtection; ISOBJ_TYPE_assert(pThis, wti); - assert(tCmd <= eWRKTHRD_SHUTDOWN_IMMEDIATE); - BEGIN_MTX_PROTECTED_OPERATIONS(&pThis->mut, bLockMutex); - - tCurrCmd = pThis->tCurrCmd; - /* all worker states must be followed sequentially, only termination can be set in any state */ - if( (bActiveOnly && (tCurrCmd < eWRKTHRD_RUN_CREATED)) - || (tCurrCmd > tCmd && !(tCmd == eWRKTHRD_TERMINATING || tCmd == eWRKTHRD_STOPPED))) { - DBGPRINTF("%s: command %d can not be accepted in current %d processing state - ignored\n", - wtiGetDbgHdr(pThis), tCmd, tCurrCmd); - } else { - DBGPRINTF("%s: receiving command %d\n", wtiGetDbgHdr(pThis), tCmd); - /* we could replace this with a simple if, but we leave the switch in in case we need - * to add something at a later stage. -- rgerhards, 2008-09-30 - */ - switch(tCmd) { - case eWRKTHRD_TERMINATING: - /* TODO: re-enable meaningful debug msg! (via function callback?) - dbgprintf("%s: thread terminating with %d entries left in queue, %d workers running.\n", - wtiGetDbgHdr(pThis->pQueue), pThis->pQueue->iQueueSize, - pThis->pQueue->iCurNumWrkThrd); - */ - pthread_cond_signal(&pThis->condExitDone); - dbgprintf("%s: worker terminating\n", wtiGetDbgHdr(pThis)); - break; - /* these cases just to satisfy the compiler, we do (yet) not act an them: */ - case eWRKTHRD_RUNNING: - case eWRKTHRD_STOPPED: - case eWRKTHRD_RUN_CREATED: - case eWRKTHRD_RUN_INIT: - case eWRKTHRD_SHUTDOWN: - case eWRKTHRD_SHUTDOWN_IMMEDIATE: - /* DO NOTHING */ - break; - } - /* apply the new state */ - unsigned val = ATOMIC_CAS_VAL((int*)&pThis->tCurrCmd, tCurrCmd, tCmd, &pThis->mutCurrCmd); - if(val != tCurrCmd) { - DBGPRINTF("wtiSetState PROBLEM, tCurrCmd %d overwritten with %d, wanted to set %d\n", tCurrCmd, val, tCmd); - } + if(wtiGetState(pThis)) { + /* we first try the cooperative "cancel" interface */ + pthread_kill(pThis->thrdID, SIGTTIN); + dbgprintf("sent SIGTTIN to worker thread 0x%x\n", (unsigned) pThis->thrdID); } - END_MTX_PROTECTED_OPERATIONS(&pThis->mut); RETiRet; } -/* Cancel the thread. If the thread is already cancelled or terminated, - * we do not again cancel it. But it is save and legal to call wtiCancelThrd() in - * such situations. +/* Cancel the thread. If the thread is not running. But it is save and legal to + * call wtiCancelThrd() in such situations. This function only returns when the + * thread has terminated. Else we may get race conditions all over the code... + * Note that when waiting for the thread to terminate, we do a busy wait, checking + * progress every 10ms. It is very unlikely that we will ever cancel a thread + * and, if so, it will only happen at the end of the rsyslog run. So doing this + * kind of non-optimal wait is considered preferable over using condition variables. * rgerhards, 2008-02-26 */ rsRetVal @@ -170,19 +146,24 @@ wtiCancelThrd(wti_t *pThis) ISOBJ_TYPE_assert(pThis, wti); - d_pthread_mutex_lock(&pThis->mut); - wtiProcessThrdChanges(pThis, MUTEX_ALREADY_LOCKED); /* process state change, so that we have current state vars */ + if(wtiGetState(pThis)) { + /* we first try the cooperative "cancel" interface */ + pthread_kill(pThis->thrdID, SIGTTIN); + dbgprintf("sent SIGTTIN to worker thread 0x%x, giving it a chance to terminate\n", (unsigned) pThis->thrdID); + srSleep(0, 10000); + } - if(pThis->tCurrCmd >= eWRKTHRD_TERMINATING) { - dbgoprint((obj_t*) pThis, "canceling worker thread, curr stat %d\n", pThis->tCurrCmd); + if(wtiGetState(pThis)) { + dbgprintf("cooperative worker termination failed, using cancellation...\n"); + dbgoprint((obj_t*) pThis, "canceling worker thread\n"); pthread_cancel(pThis->thrdID); - wtiSetState(pThis, eWRKTHRD_TERMINATING, 0, MUTEX_ALREADY_LOCKED); - wtpSetThrdStateChanged(pThis->pWtp, 1); /* indicate change, so harverster will be called */ + /* now wait until the thread terminates... */ + while(wtiGetState(pThis)) { + srSleep(0, 10000); + } } - d_pthread_mutex_unlock(&pThis->mut); - RETiRet; } @@ -190,26 +171,9 @@ wtiCancelThrd(wti_t *pThis) /* Destructor */ BEGINobjDestruct(wti) /* be sure to specify the object type also in END and CODESTART macros! */ CODESTARTobjDestruct(wti) - /* if we reach this point, we must make sure the associated worker has terminated. It is - * the callers duty to make sure the worker already knows it shall terminate. - * TODO: is it *really* the caller's duty? ...mmmhhhh.... smells bad... rgerhards, 2008-01-25 - */ - wtiProcessThrdChanges(pThis, LOCK_MUTEX); /* process state change one last time */ - - d_pthread_mutex_lock(&pThis->mut); - if(wtiGetState(pThis, MUTEX_ALREADY_LOCKED) != eWRKTHRD_STOPPED) { - dbgprintf("%s: WARNING: worker %p shall be destructed but is still running (might be OK) - joining it\n", - wtiGetDbgHdr(pThis), pThis); - /* let's hope the caller actually instructed it to shutdown... */ - pthread_cond_wait(&pThis->condExitDone, &pThis->mut); - wtiJoinThrd(pThis); - } - d_pthread_mutex_unlock(&pThis->mut); - /* actual destruction */ - pthread_cond_destroy(&pThis->condExitDone); - pthread_mutex_destroy(&pThis->mut); - DESTROY_ATOMIC_HELPER_MUT(pThis->mutCurrCmd); + batchFree(&pThis->batch); + DESTROY_ATOMIC_HELPER_MUT(pThis->mutIsRunning); free(pThis->pszDbgHdr); ENDobjDestruct(wti) @@ -218,9 +182,7 @@ ENDobjDestruct(wti) /* Standard-Constructor for the wti object */ BEGINobjConstruct(wti) /* be sure to specify the object type also in END macro! */ - pthread_cond_init(&pThis->condExitDone, NULL); - pthread_mutex_init(&pThis->mut, NULL); - INIT_ATOMIC_HELPER_MUT(pThis->mutCurrCmd); + INIT_ATOMIC_HELPER_MUT(pThis->mutIsRunning); ENDobjConstruct(wti) @@ -231,81 +193,28 @@ rsRetVal wtiConstructFinalize(wti_t *pThis) { DEFiRet; + int iDeqBatchSize; ISOBJ_TYPE_assert(pThis, wti); dbgprintf("%s: finalizing construction of worker instance data\n", wtiGetDbgHdr(pThis)); - /* initialize our thread instance descriptor */ - pThis->pUsrp = NULL; - pThis->tCurrCmd = eWRKTHRD_STOPPED; - - RETiRet; -} - - -/* join a specific worker thread - * we do not lock the mutex, because join will sync anyways... - */ -rsRetVal -wtiJoinThrd(wti_t *pThis) -{ - DEFiRet; - - ISOBJ_TYPE_assert(pThis, wti); - dbgprintf("waiting for worker %s termination, current state %d\n", wtiGetDbgHdr(pThis), pThis->tCurrCmd); - if (pThis->thrdID == 0) { - dbgprintf("worker %s was already stopped\n", wtiGetDbgHdr(pThis)); - } else { - pthread_join(pThis->thrdID, NULL); - wtiSetState(pThis, eWRKTHRD_STOPPED, 0, MUTEX_ALREADY_LOCKED); /* back to virgin... */ - pThis->thrdID = 0; /* invalidate the thread ID so that we do not accidently find reused ones */ - dbgprintf("worker %s has stopped\n", wtiGetDbgHdr(pThis)); - } - - RETiRet; -} - -/* check if we had a worker thread changes and, if so, act - * on it. At a minimum, terminated threads are harvested (joined). - */ -rsRetVal -wtiProcessThrdChanges(wti_t *pThis, int bLockMutex) -{ - DEFiRet; - DEFVARS_mutexProtection; + /* initialize our thread instance descriptor (no concurrency here) */ + pThis->bIsRunning = FALSE; - ISOBJ_TYPE_assert(pThis, wti); - - BEGIN_MTX_PROTECTED_OPERATIONS(&pThis->mut, bLockMutex); - switch(pThis->tCurrCmd) { - case eWRKTHRD_TERMINATING: - /* we need to at least temporarily release the mutex, because otherwise - * we may deadlock with the thread we intend to join (it aquires the mutex - * during termination processing). -- rgerhards, 2008-02-26 - */ - END_MTX_PROTECTED_OPERATIONS(&pThis->mut); - iRet = wtiJoinThrd(pThis); - BEGIN_MTX_PROTECTED_OPERATIONS(&pThis->mut, bLockMutex); - break; - /* these cases just to satisfy the compiler, we do not act an them: */ - case eWRKTHRD_STOPPED: - case eWRKTHRD_RUN_CREATED: - case eWRKTHRD_RUN_INIT: - case eWRKTHRD_RUNNING: - case eWRKTHRD_SHUTDOWN: - case eWRKTHRD_SHUTDOWN_IMMEDIATE: - /* DO NOTHING */ - break; - } - END_MTX_PROTECTED_OPERATIONS(&pThis->mut); + /* we now alloc the array for user pointers. We obtain the max from the queue itself. */ + CHKiRet(pThis->pWtp->pfGetDeqBatchSize(pThis->pWtp->pUsr, &iDeqBatchSize)); + CHKiRet(batchInit(&pThis->batch, iDeqBatchSize)); +finalize_it: RETiRet; } /* cancellation cleanup handler for queueWorker () - * Updates admin structure and frees ressources. + * Most importantly, it must bring back the batch into a consistent state. + * Keep in mind that cancellation is disabled if we run into + * the cancel cleanup handler (and have been cancelled). * rgerhards, 2008-01-16 */ static void @@ -313,7 +222,6 @@ wtiWorkerCancelCleanup(void *arg) { wti_t *pThis = (wti_t*) arg; wtp_t *pWtp; - int iCancelStateSave; BEGINfunc ISOBJ_TYPE_assert(pThis, wti); @@ -321,102 +229,108 @@ wtiWorkerCancelCleanup(void *arg) ISOBJ_TYPE_assert(pWtp, wtp); DBGPRINTF("%s: cancelation cleanup handler called.\n", wtiGetDbgHdr(pThis)); + pWtp->pfObjProcessed(pWtp->pUsr, pThis); + DBGPRINTF("%s: done cancelation cleanup handler.\n", wtiGetDbgHdr(pThis)); - /* call user supplied handler (that one e.g. requeues the element) */ - pWtp->pfOnWorkerCancel(pThis->pWtp->pUsr, pThis->pUsrp); + ENDfunc +} - pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &iCancelStateSave); - d_pthread_mutex_lock(&pWtp->mut); - wtiSetState(pThis, eWRKTHRD_TERMINATING, 0, MUTEX_ALREADY_LOCKED); - wtpSetThrdStateChanged(pWtp, 1); /* indicate change, so harverster will be called */ - d_pthread_mutex_unlock(&pWtp->mut); - pthread_setcancelstate(iCancelStateSave, NULL); +/* wait for queue to become non-empty or timeout + * helper to wtiWorker. Note the the predicate is + * re-tested by the caller, so it is OK to NOT do it here. + * rgerhards, 2009-05-20 + */ +static inline void +doIdleProcessing(wti_t *pThis, wtp_t *pWtp, int *pbInactivityTOOccured) +{ + struct timespec t; + + BEGINfunc + DBGPRINTF("%s: worker IDLE, waiting for work.\n", wtiGetDbgHdr(pThis)); + + if(pThis->bAlwaysRunning) { + /* never shut down any started worker */ + d_pthread_cond_wait(pWtp->pcondBusy, pWtp->pmutUsr); + } else { + timeoutComp(&t, pWtp->toWrkShutdown);/* get absolute timeout */ + if(d_pthread_cond_timedwait(pWtp->pcondBusy, pWtp->pmutUsr, &t) != 0) { + DBGPRINTF("%s: inactivity timeout, worker terminating...\n", wtiGetDbgHdr(pThis)); + *pbInactivityTOOccured = 1; /* indicate we had a timeout */ + } + } + dbgoprint((obj_t*) pThis, "worker awoke from idle processing\n"); ENDfunc } -/* generic worker thread framework +/* generic worker thread framework. Note that we prohibit cancellation + * during almost all times, because it can have very undesired side effects. + * However, we may need to cancel a thread if the consumer blocks for too + * long (during shutdown). So what we do is block cancellation, and every + * consumer must enable it during the periods where it is safe. */ #pragma GCC diagnostic ignored "-Wempty-body" rsRetVal wtiWorker(wti_t *pThis) { - DEFiRet; - DEFVARS_mutexProtection; - struct timespec t; wtp_t *pWtp; /* our worker thread pool */ int bInactivityTOOccured = 0; + rsRetVal localRet; + rsRetVal terminateRet; + int iCancelStateSave; + DEFiRet; ISOBJ_TYPE_assert(pThis, wti); pWtp = pThis->pWtp; /* shortcut */ ISOBJ_TYPE_assert(pWtp, wtp); dbgSetThrdName(pThis->pszDbgHdr); - pThis->pUsrp = NULL; pthread_cleanup_push(wtiWorkerCancelCleanup, pThis); - - BEGIN_MTX_PROTECTED_OPERATIONS(pWtp->pmutUsr, LOCK_MUTEX); - pWtp->pfOnWorkerStartup(pWtp->pUsr); - END_MTX_PROTECTED_OPERATIONS(pWtp->pmutUsr); + pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &iCancelStateSave); /* now we have our identity, on to real processing */ while(1) { /* loop will be broken below - need to do mutex locks */ - /* process any pending thread requests */ - wtpProcessThrdChanges(pWtp); - - /* if we have a rate-limiter set for this worker pool, let's call it. Please - * keep in mind that the rate-limiter may hold us for an extended period - * of time. -- rgerhards, 2008-04-02 - */ - if(pWtp->pfRateLimiter != NULL) { + if(pWtp->pfRateLimiter != NULL) { /* call rate-limiter, if defined */ pWtp->pfRateLimiter(pWtp->pUsr); } - wtpSetInactivityGuard(pThis->pWtp, 0, LOCK_MUTEX); /* must be set before usr mutex is locked! */ - BEGIN_MTX_PROTECTED_OPERATIONS(pWtp->pmutUsr, LOCK_MUTEX); - - if( (bInactivityTOOccured && pWtp->pfIsIdle(pWtp->pUsr, MUTEX_ALREADY_LOCKED)) - || wtpChkStopWrkr(pWtp, LOCK_MUTEX, MUTEX_ALREADY_LOCKED)) { - END_MTX_PROTECTED_OPERATIONS(pWtp->pmutUsr); - break; /* end worker thread run */ + d_pthread_mutex_lock(pWtp->pmutUsr); + + /* first check if we are in shutdown process (but evaluate a bit later) */ + terminateRet = wtpChkStopWrkr(pWtp, MUTEX_ALREADY_LOCKED); + if(terminateRet == RS_RET_TERMINATE_NOW) { + /* we now need to free the old batch */ + localRet = pWtp->pfObjProcessed(pWtp->pUsr, pThis); + dbgoprint((obj_t*) pThis, "terminating worker because of TERMINATE_NOW mode, del iRet %d\n", + localRet); + d_pthread_mutex_unlock(pWtp->pmutUsr); + break; } - bInactivityTOOccured = 0; /* reset for next run */ - /* if we reach this point, we are still protected by the mutex */ - - if(pWtp->pfIsIdle(pWtp->pUsr, MUTEX_ALREADY_LOCKED)) { - DBGPRINTF("%s: worker IDLE, waiting for work.\n", wtiGetDbgHdr(pThis)); - pWtp->pfOnIdle(pWtp->pUsr, MUTEX_ALREADY_LOCKED); - - if(pWtp->toWrkShutdown == -1) { - /* never shut down any started worker */ - d_pthread_cond_wait(pWtp->pcondBusy, pWtp->pmutUsr); - } else { - timeoutComp(&t, pWtp->toWrkShutdown);/* get absolute timeout */ - if(d_pthread_cond_timedwait(pWtp->pcondBusy, pWtp->pmutUsr, &t) != 0) { - DBGPRINTF("%s: inactivity timeout, worker terminating...\n", wtiGetDbgHdr(pThis)); - bInactivityTOOccured = 1; /* indicate we had a timeout */ - } + /* try to execute and process whatever we have */ + /* Note that this function releases and re-aquires the mutex. The returned + * information on idle state must be processed before releasing the mutex again. + */ + localRet = pWtp->pfDoWork(pWtp->pUsr, pThis); + + if(localRet == RS_RET_IDLE) { + if(terminateRet == RS_RET_TERMINATE_WHEN_IDLE || bInactivityTOOccured) { + d_pthread_mutex_unlock(pWtp->pmutUsr); + break; /* end of loop */ } - END_MTX_PROTECTED_OPERATIONS(pWtp->pmutUsr); + doIdleProcessing(pThis, pWtp, &bInactivityTOOccured); + d_pthread_mutex_unlock(pWtp->pmutUsr); continue; /* request next iteration */ } - /* if we reach this point, we have a non-empty queue (and are still protected by mutex) */ - pWtp->pfDoWork(pWtp->pUsr, pThis, iCancelStateSave); + d_pthread_mutex_unlock(pWtp->pmutUsr); + + bInactivityTOOccured = 0; /* reset for next run */ } /* indicate termination */ - pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &iCancelStateSave); - d_pthread_mutex_lock(&pThis->mut); pthread_cleanup_pop(0); /* remove cleanup handler */ - - pWtp->pfOnWorkerShutdown(pWtp->pUsr); - - wtiSetState(pThis, eWRKTHRD_TERMINATING, 0, MUTEX_ALREADY_LOCKED); - wtpSetThrdStateChanged(pWtp, 1); /* indicate change, so harverster will be called */ - d_pthread_mutex_unlock(&pThis->mut); pthread_setcancelstate(iCancelStateSave, NULL); RETiRet; @@ -445,10 +359,9 @@ wtiSetDbgHdr(wti_t *pThis, uchar *pszMsg, size_t lenMsg) if(pThis->pszDbgHdr != NULL) { free(pThis->pszDbgHdr); - pThis->pszDbgHdr = NULL; } - if((pThis->pszDbgHdr = malloc(sizeof(uchar) * lenMsg + 1)) == NULL) + if((pThis->pszDbgHdr = MALLOC(sizeof(uchar) * lenMsg + 1)) == NULL) ABORT_FINALIZE(RS_RET_OUT_OF_MEMORY); memcpy(pThis->pszDbgHdr, pszMsg, lenMsg + 1); /* always think about the \0! */ @@ -479,6 +392,5 @@ BEGINObjClassInit(wti, 1, OBJ_IS_CORE_MODULE) /* one is the object version (most CHKiRet(objUse(glbl, CORE_COMPONENT)); ENDObjClassInit(wti) -/* - * vi:set ai: +/* vi:set ai: */ diff --git a/runtime/wti.h b/runtime/wti.h index d81672f3..51ece4ef 100644 --- a/runtime/wti.h +++ b/runtime/wti.h @@ -1,6 +1,6 @@ /* Definition of the worker thread instance (wti) class. * - * Copyright 2008 Rainer Gerhards and Adiscon GmbH. + * Copyright 2008, 2009 by Rainer Gerhards and Adiscon GmbH. * * This file is part of the rsyslog runtime library. * @@ -27,22 +27,20 @@ #include <pthread.h> #include "wtp.h" #include "obj.h" +#include "batch.h" + /* the worker thread instance class */ -typedef struct wti_s { +struct wti_s { BEGINobjInstance; - pthread_t thrdID; /* thread ID */ - qWrkCmd_t tCurrCmd; /* current command to be carried out by worker */ - obj_t *pUsrp; /* pointer to an object meaningful for current user pointer (e.g. queue pUsr data elemt) */ + pthread_t thrdID; /* thread ID */ + int bIsRunning; /* is this thread currently running? (must be int for atomic op!) */ + sbool bAlwaysRunning; /* should this thread always run? */ wtp_t *pWtp; /* my worker thread pool (important if only the work thread instance is passed! */ - pthread_cond_t condExitDone; /* signaled when the thread exit is done (once per thread existance) */ - pthread_mutex_t mut; - bool bShutdownRqtd; /* shutdown for this thread requested? 0 - no , 1 - yes */ + batch_t batch; /* pointer to an object array meaningful for current user pointer (e.g. queue pUsr data elemt) */ uchar *pszDbgHdr; /* header string for debug messages */ - DEF_ATOMIC_HELPER_MUT(mutCurrCmd); -} wti_t; - -/* some symbolic constants for easier reference */ + DEF_ATOMIC_HELPER_MUT(mutIsRunning); +}; /* prototypes */ @@ -50,12 +48,12 @@ rsRetVal wtiConstruct(wti_t **ppThis); rsRetVal wtiConstructFinalize(wti_t *pThis); rsRetVal wtiDestruct(wti_t **ppThis); rsRetVal wtiWorker(wti_t *pThis); -rsRetVal wtiProcessThrdChanges(wti_t *pThis, int bLockMutex); rsRetVal wtiSetDbgHdr(wti_t *pThis, uchar *pszMsg, size_t lenMsg); -rsRetVal wtiSetState(wti_t *pThis, qWrkCmd_t tCmd, int bActiveOnly, int bLockMutex); -rsRetVal wtiJoinThrd(wti_t *pThis); rsRetVal wtiCancelThrd(wti_t *pThis); -qWrkCmd_t wtiGetState(wti_t *pThis, int bLockMutex); +rsRetVal wtiSetAlwaysRunning(wti_t *pThis); +rsRetVal wtiSetState(wti_t *pThis, sbool bNew); +rsRetVal wtiWakeupThrd(wti_t *pThis); +sbool wtiGetState(wti_t *pThis); PROTOTYPEObjClassInit(wti); PROTOTYPEpropSetMeth(wti, pszDbgHdr, uchar*); PROTOTYPEpropSetMeth(wti, pWtp, wtp_t*); diff --git a/runtime/wtp.c b/runtime/wtp.c index fff37c2f..ece80911 100644 --- a/runtime/wtp.c +++ b/runtime/wtp.c @@ -8,7 +8,7 @@ * (and in the web doc set on http://www.rsyslog.com/doc). Be sure to read it * if you are getting aquainted to the object. * - * Copyright 2008 Rainer Gerhards and Adiscon GmbH. + * Copyright 2008,2009 Rainer Gerhards and Adiscon GmbH. * * This file is part of the rsyslog runtime library. * @@ -44,9 +44,10 @@ # include <sys/prctl.h> #endif -#ifdef OS_SOLARIS -# include <sched.h> -#endif +/// TODO: check on solaris if this is any longer needed - I don't think so - rgerhards, 2009-09-20 +//#ifdef OS_SOLARIS +//# include <sched.h> +//#endif #include "rsyslog.h" #include "stringbuf.h" @@ -82,22 +83,21 @@ wtpGetDbgHdr(wtp_t *pThis) /* Not implemented dummy function for constructor */ -static rsRetVal NotImplementedDummy() { return RS_RET_OK; } +static rsRetVal NotImplementedDummy() { return RS_RET_NOT_IMPLEMENTED; } /* Standard-Constructor for the wtp object */ BEGINobjConstruct(wtp) /* be sure to specify the object type also in END macro! */ - pthread_mutex_init(&pThis->mut, NULL); - pthread_mutex_init(&pThis->mutThrdShutdwn, NULL); + pthread_mutex_init(&pThis->mutWtp, NULL); pthread_cond_init(&pThis->condThrdTrm, NULL); + pthread_attr_init(&pThis->attrThrd); + pthread_attr_setdetachstate(&pThis->attrThrd, PTHREAD_CREATE_DETACHED); /* set all function pointers to "not implemented" dummy so that we can safely call them */ pThis->pfChkStopWrkr = NotImplementedDummy; - pThis->pfIsIdle = NotImplementedDummy; + pThis->pfGetDeqBatchSize = NotImplementedDummy; pThis->pfDoWork = NotImplementedDummy; - pThis->pfOnIdle = NotImplementedDummy; - pThis->pfOnWorkerCancel = NotImplementedDummy; - pThis->pfOnWorkerStartup = NotImplementedDummy; - pThis->pfOnWorkerShutdown = NotImplementedDummy; - INIT_ATOMIC_HELPER_MUT(pThis->mutThrdStateChanged); + pThis->pfObjProcessed = NotImplementedDummy; + INIT_ATOMIC_HELPER_MUT(pThis->mutCurNumWrkThrd); + INIT_ATOMIC_HELPER_MUT(pThis->mutWtpState); ENDobjConstruct(wtp) @@ -115,13 +115,12 @@ wtpConstructFinalize(wtp_t *pThis) ISOBJ_TYPE_assert(pThis, wtp); - dbgprintf("%s: finalizing construction of worker thread pool\n", wtpGetDbgHdr(pThis)); + DBGPRINTF("%s: finalizing construction of worker thread pool\n", wtpGetDbgHdr(pThis)); /* alloc and construct workers - this can only be done in finalizer as we previously do * not know the max number of workers */ - if((pThis->pWrkr = malloc(sizeof(wti_t*) * pThis->iNumWorkerThreads)) == NULL) - ABORT_FINALIZE(RS_RET_OUT_OF_MEMORY); - + CHKmalloc(pThis->pWrkr = MALLOC(sizeof(wti_t*) * pThis->iNumWorkerThreads)); + for(i = 0 ; i < pThis->iNumWorkerThreads ; ++i) { CHKiRet(wtiConstruct(&pThis->pWrkr[i])); pWti = pThis->pWrkr[i]; @@ -141,8 +140,6 @@ finalize_it: BEGINobjDestruct(wtp) /* be sure to specify the object type also in END and CODESTART macros! */ int i; CODESTARTobjDestruct(wtp) - wtpProcessThrdChanges(pThis); /* process thread changes one last time */ - /* destruct workers */ for(i = 0 ; i < pThis->iNumWorkerThreads ; ++i) wtiDestruct(&pThis->pWrkr[i]); @@ -152,149 +149,70 @@ CODESTARTobjDestruct(wtp) /* actual destruction */ pthread_cond_destroy(&pThis->condThrdTrm); - pthread_mutex_destroy(&pThis->mut); - pthread_mutex_destroy(&pThis->mutThrdShutdwn); - DESTROY_ATOMIC_HELPER_MUT(pThis->mutThrdStateChanged); + pthread_mutex_destroy(&pThis->mutWtp); + pthread_attr_destroy(&pThis->attrThrd); + DESTROY_ATOMIC_HELPER_MUT(pThis->mutCurNumWrkThrd); + DESTROY_ATOMIC_HELPER_MUT(pThis->mutWtpState); free(pThis->pszDbgHdr); ENDobjDestruct(wtp) -/* wake up at least one worker thread. - * rgerhards, 2008-01-20 - */ -rsRetVal -wtpWakeupWrkr(wtp_t *pThis) -{ - DEFiRet; - - /* TODO; mutex? I think not needed, as we do not need predictable exec order -- rgerhards, 2008-01-28 */ - ISOBJ_TYPE_assert(pThis, wtp); - pthread_cond_signal(pThis->pcondBusy); - RETiRet; -} - -/* wake up all worker threads. - * rgerhards, 2008-01-16 - */ -rsRetVal -wtpWakeupAllWrkr(wtp_t *pThis) -{ - DEFiRet; - - ISOBJ_TYPE_assert(pThis, wtp); - pthread_cond_broadcast(pThis->pcondBusy); - RETiRet; -} - - -/* set the bThrdStateChanged in an atomic way. Note that - * val may only be 0 or 1. - */ -void -wtpSetThrdStateChanged(wtp_t *pThis, int val) -{ - if(val == 0) { - ATOMIC_STORE_0_TO_INT(&pThis->bThrdStateChanged, pThis->mutThrdStateChanged); - } else { - ATOMIC_STORE_1_TO_INT(&pThis->bThrdStateChanged, pThis->mutThrdStateChanged); - } -} - - -/* check if we had any worker thread changes and, if so, act - * on them. At a minimum, terminated threads are harvested (joined). - * This function MUST NEVER block on the queue mutex! - */ -rsRetVal -wtpProcessThrdChanges(wtp_t *pThis) -{ - DEFiRet; - int i; - - ISOBJ_TYPE_assert(pThis, wtp); - - if(pThis->bThrdStateChanged == 0) - FINALIZE; - - if(d_pthread_mutex_trylock(&(pThis->mutThrdShutdwn)) != 0) { - /* another thread is already in the loop */ - FINALIZE; - } - - /* Note: there is a left-over potential race condition below: - * pThis->bThrdStateChanged may be re-set by another thread while - * we work on it and thus the loop may terminate too early. However, - * there are no really bad effects from that so I perfer - for this - * version - to live with the problem as is. Not a good idea to - * introduce that large change into the stable branch without very - * good reason. -- rgerhards, 2009-04-02 - */ - do { - /* reset the change marker */ - wtpSetThrdStateChanged(pThis, 0); - /* go through all threads */ - for(i = 0 ; i < pThis->iNumWorkerThreads ; ++i) { - wtiProcessThrdChanges(pThis->pWrkr[i], LOCK_MUTEX); - } - /* restart if another change occured while we were processing the changes */ - } while(pThis->bThrdStateChanged != 0); - - d_pthread_mutex_unlock(&(pThis->mutThrdShutdwn)); - -finalize_it: - RETiRet; -} - - -/* Sent a specific state for the worker thread pool. - * rgerhards, 2008-01-21 +/* Sent a specific state for the worker thread pool. -- rgerhards, 2008-01-21 + * We do not need to do atomic instructions as set operations are only + * called when terminating the pool, and then in strict sequence. So we + * can never overwrite each other. On the other hand, it also doesn't + * matter if the read operation obtains an older value, as we then simply + * do one more iteration, what is perfectly legal (during shutdown + * they are awoken in any case). -- rgerhards, 2009-07-20 */ rsRetVal wtpSetState(wtp_t *pThis, wtpState_t iNewState) { - DEFiRet; - ISOBJ_TYPE_assert(pThis, wtp); - pThis->wtpState = iNewState; - /* TODO: must wakeup workers? seen to be not needed -- rgerhards, 2008-01-28 */ - - RETiRet; + pThis->wtpState = iNewState; // TODO: do we need a mutex here? 2010-04-26 + return RS_RET_OK; } /* check if the worker shall shutdown (1 = yes, 0 = no) - * TODO: check if we can use atomic operations to enhance performance * Note: there may be two mutexes locked, the bLockUsrMutex is the one in our "user" * (e.g. the queue clas) * rgerhards, 2008-01-21 */ rsRetVal -wtpChkStopWrkr(wtp_t *pThis, int bLockMutex, int bLockUsrMutex) +wtpChkStopWrkr(wtp_t *pThis, int bLockUsrMutex) { DEFiRet; - DEFVARS_mutexProtection; + wtpState_t wtpState; ISOBJ_TYPE_assert(pThis, wtp); + /* we need a consistent value, but it doesn't really matter if it is changed + * right after the fetch - then we simply do one more iteration in the worker + */ + wtpState = (wtpState_t) ATOMIC_FETCH_32BIT((int*)&pThis->wtpState, &pThis->mutWtpState); - BEGIN_MTX_PROTECTED_OPERATIONS(&pThis->mut, bLockMutex); - if( (pThis->wtpState == wtpState_SHUTDOWN_IMMEDIATE) - || ((pThis->wtpState == wtpState_SHUTDOWN) && pThis->pfIsIdle(pThis->pUsr, bLockUsrMutex))) - iRet = RS_RET_TERMINATE_NOW; - END_MTX_PROTECTED_OPERATIONS(&pThis->mut); + if(wtpState == wtpState_SHUTDOWN_IMMEDIATE) { + ABORT_FINALIZE(RS_RET_TERMINATE_NOW); + } else if(wtpState == wtpState_SHUTDOWN) { + ABORT_FINALIZE(RS_RET_TERMINATE_WHEN_IDLE); + } /* try customer handler if one was set and we do not yet have a definite result */ - if(iRet == RS_RET_OK && pThis->pfChkStopWrkr != NULL) { + if(pThis->pfChkStopWrkr != NULL) { iRet = pThis->pfChkStopWrkr(pThis->pUsr, bLockUsrMutex); } +finalize_it: RETiRet; } #pragma GCC diagnostic ignored "-Wempty-body" /* Send a shutdown command to all workers and see if they terminate. - * A timeout may be specified. + * A timeout may be specified. This function may also be called with + * the current number of workers being 0, in which case it does not + * shut down any worker. * rgerhards, 2008-01-14 */ rsRetVal @@ -302,72 +220,50 @@ wtpShutdownAll(wtp_t *pThis, wtpState_t tShutdownCmd, struct timespec *ptTimeout { DEFiRet; int bTimedOut; - int iCancelStateSave; + int i; ISOBJ_TYPE_assert(pThis, wtp); + /* lock mutex to prevent races (may otherwise happen during idle processing and such...) */ + d_pthread_mutex_lock(pThis->pmutUsr); wtpSetState(pThis, tShutdownCmd); - wtpWakeupAllWrkr(pThis); + pthread_cond_broadcast(pThis->pcondBusy); /* wake up all workers */ + /* awake workers in retry loop */ + for(i = 0 ; i < pThis->iNumWorkerThreads ; ++i) { + wtiWakeupThrd(pThis->pWrkr[i]); + } + d_pthread_mutex_unlock(pThis->pmutUsr); - /* see if we need to harvest (join) any terminated threads (even in timeout case, - * some may have terminated... - */ - wtpProcessThrdChanges(pThis); - - /* and wait for their termination */ - pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &iCancelStateSave); - d_pthread_mutex_lock(&pThis->mut); - pthread_cleanup_push(mutexCancelCleanup, &pThis->mut); - pthread_setcancelstate(iCancelStateSave, NULL); + /* wait for worker thread termination */ + d_pthread_mutex_lock(&pThis->mutWtp); + pthread_cleanup_push(mutexCancelCleanup, &pThis->mutWtp); bTimedOut = 0; while(pThis->iCurNumWrkThrd > 0 && !bTimedOut) { - dbgprintf("%s: waiting %ldms on worker thread termination, %d still running\n", - wtpGetDbgHdr(pThis), timeoutVal(ptTimeout), pThis->iCurNumWrkThrd); + DBGPRINTF("%s: waiting %ldms on worker thread termination, %d still running\n", + wtpGetDbgHdr(pThis), timeoutVal(ptTimeout), + ATOMIC_FETCH_32BIT(&pThis->iCurNumWrkThrd, &pThis->mutCurNumWrkThrd)); - if(d_pthread_cond_timedwait(&pThis->condThrdTrm, &pThis->mut, ptTimeout) != 0) { - dbgprintf("%s: timeout waiting on worker thread termination\n", wtpGetDbgHdr(pThis)); + if(d_pthread_cond_timedwait(&pThis->condThrdTrm, &pThis->mutWtp, ptTimeout) != 0) { + DBGPRINTF("%s: timeout waiting on worker thread termination\n", wtpGetDbgHdr(pThis)); bTimedOut = 1; /* we exit the loop on timeout */ } + + /* awake workers in retry loop */ + for(i = 0 ; i < pThis->iNumWorkerThreads ; ++i) { + wtiWakeupThrd(pThis->pWrkr[i]); + } + } pthread_cleanup_pop(1); if(bTimedOut) iRet = RS_RET_TIMED_OUT; - /* see if we need to harvest (join) any terminated threads (even in timeout case, - * some may have terminated... - */ - wtpProcessThrdChanges(pThis); - RETiRet; } #pragma GCC diagnostic warning "-Wempty-body" -/* indicate that a thread has terminated and awake anyone waiting on it - * rgerhards, 2008-01-23 - */ -rsRetVal wtpSignalWrkrTermination(wtp_t *pThis) -{ - DEFiRet; - /* I leave the mutex code here out as it gives us deadlocks. I think it is not really - * needed and we are on the safe side. I leave this comment in if practice proves us - * wrong. The whole thing should be removed after half a year or year if we see there - * actually is no issue (or revisit it from a theoretical POV). - * rgerhards, 2008-01-28 - * revisited 2008-09-30, still a bit unclear, leave in - */ - /*TODO: mutex or not mutex, that's the question ;)DEFVARS_mutexProtection;*/ - - ISOBJ_TYPE_assert(pThis, wtp); - - /*BEGIN_MTX_PROTECTED_OPERATIONS(&pThis->mut, LOCK_MUTEX);*/ - pthread_cond_signal(&pThis->condThrdTrm); /* activate anyone waiting on thread shutdown */ - /*END_MTX_PROTECTED_OPERATIONS(&pThis->mut);*/ - RETiRet; -} - - /* Unconditionally cancel all running worker threads. * rgerhards, 2008-01-14 */ @@ -379,12 +275,8 @@ wtpCancelAll(wtp_t *pThis) ISOBJ_TYPE_assert(pThis, wtp); - /* process any pending thread requests so that we know who actually is still running */ - wtpProcessThrdChanges(pThis); - /* go through all workers and cancel those that are active */ for(i = 0 ; i < pThis->iNumWorkerThreads ; ++i) { - dbgprintf("%s: try canceling worker thread %d\n", wtpGetDbgHdr(pThis), i); wtiCancelThrd(pThis->pWrkr[i]); } @@ -392,40 +284,57 @@ wtpCancelAll(wtp_t *pThis) } - -/* Set the Inactivity Guard - * rgerhards, 2008-01-21 +/* this function contains shared code for both regular worker shutdown as + * well as shutdown via cancellation. We can not simply use pthread_cleanup_pop(1) + * as this introduces a race in the debug system (RETiRet system). + * rgerhards, 2009-10-26 */ -rsRetVal -wtpSetInactivityGuard(wtp_t *pThis, int bNewState, int bLockMutex) +static inline void +wtpWrkrExecCleanup(wti_t *pWti) { - DEFiRet; - DEFVARS_mutexProtection; + wtp_t *pThis; - BEGIN_MTX_PROTECTED_OPERATIONS(&pThis->mut, bLockMutex); - pThis->bInactivityGuard = bNewState; - END_MTX_PROTECTED_OPERATIONS(&pThis->mut); + BEGINfunc + ISOBJ_TYPE_assert(pWti, wti); + pThis = pWti->pWtp; + ISOBJ_TYPE_assert(pThis, wtp); - RETiRet; + /* the order of the next two statements is important! */ + wtiSetState(pWti, WRKTHRD_STOPPED); + ATOMIC_DEC(&pThis->iCurNumWrkThrd, &pThis->mutCurNumWrkThrd); + + DBGPRINTF("%s: Worker thread %lx, terminated, um workers now %d\n", + wtpGetDbgHdr(pThis), (unsigned long) pWti, + ATOMIC_FETCH_32BIT(&pThis->iCurNumWrkThrd, &pThis->mutCurNumWrkThrd)); + + ENDfunc } -/* cancellation cleanup handler for executing worker - * decrements the worker counter - * rgerhards, 2008-01-20 +/* cancellation cleanup handler for executing worker decrements the worker counter. + * rgerhards, 2009-07-20 */ -void +static void wtpWrkrExecCancelCleanup(void *arg) { - wtp_t *pThis = (wtp_t*) arg; + wti_t *pWti = (wti_t*) arg; + wtp_t *pThis; BEGINfunc + ISOBJ_TYPE_assert(pWti, wti); + pThis = pWti->pWtp; ISOBJ_TYPE_assert(pThis, wtp); - pThis->iCurNumWrkThrd--; - wtpSignalWrkrTermination(pThis); + DBGPRINTF("%s: Worker thread %lx requested to be cancelled.\n", + wtpGetDbgHdr(pThis), (unsigned long) pWti); + + wtpWrkrExecCleanup(pWti); - dbgprintf("%s: thread CANCELED with %d workers running.\n", wtpGetDbgHdr(pThis), pThis->iCurNumWrkThrd); ENDfunc + /* NOTE: we must call ENDfunc FIRST, because otherwise the schedule may activate the main + * thread after the broadcast, which could destroy the debug class, resulting in a potential + * segfault. So we need to do the broadcast as actually the last action in our processing + */ + pthread_cond_broadcast(&pThis->condThrdTrm); /* activate anyone waiting on thread shutdown */ } @@ -437,8 +346,6 @@ wtpWrkrExecCancelCleanup(void *arg) static void * wtpWorker(void *arg) /* the arg is actually a wti object, even though we are in wtp! */ { - DEFiRet; - DEFVARS_mutexProtection; wti_t *pWti = (wti_t*) arg; wtp_t *pThis; sigset_t sigSet; @@ -447,13 +354,20 @@ wtpWorker(void *arg) /* the arg is actually a wti object, even though we are in uchar thrdName[32] = "rs:"; # endif + BEGINfunc ISOBJ_TYPE_assert(pWti, wti); pThis = pWti->pWtp; ISOBJ_TYPE_assert(pThis, wtp); + /* block all signals */ sigfillset(&sigSet); pthread_sigmask(SIG_BLOCK, &sigSet, NULL); + /* but ignore SIGTTN, which we (ab)use to signal the thread to shutdown -- rgerhards, 2009-07-20 */ + sigemptyset(&sigSet); + sigaddset(&sigSet, SIGTTIN); + pthread_sigmask(SIG_UNBLOCK, &sigSet, NULL); + # if HAVE_PRCTL && defined PR_SET_NAME /* set thread name - we ignore if the call fails, has no harsh consequences... */ pszDbgHdr = wtpGetDbgHdr(pThis); @@ -463,41 +377,17 @@ wtpWorker(void *arg) /* the arg is actually a wti object, even though we are in } # endif - BEGIN_MTX_PROTECTED_OPERATIONS(&pThis->mut, LOCK_MUTEX); - - /* do some late initialization */ - - pthread_cleanup_push(wtpWrkrExecCancelCleanup, pThis); - - /* finally change to RUNNING state. We need to check if we actually should still run, - * because someone may have requested us to shut down even before we got a chance to do - * our init. That would be a bad race... -- rgerhards, 2008-01-16 - */ - wtiSetState(pWti, eWRKTHRD_RUNNING, 0, MUTEX_ALREADY_LOCKED); /* we are running now! */ - - do { - END_MTX_PROTECTED_OPERATIONS(&pThis->mut); - - iRet = wtiWorker(pWti); /* just to make sure: this is NOT protected by the mutex! */ - - BEGIN_MTX_PROTECTED_OPERATIONS(&pThis->mut, LOCK_MUTEX); - } while(pThis->iCurNumWrkThrd == 1 && pThis->bInactivityGuard == 1); - /* inactivity guard prevents shutdown of all workers while one should be running due to race - * condition. It can lead to one more worker running than desired, but that is acceptable. After - * all, that worker will shutdown itself due to inactivity timeout. If, however, none were running - * when one was required, processing could come to a halt. -- rgerhards, 2008-01-21 - */ - + pthread_cleanup_push(wtpWrkrExecCancelCleanup, pWti); + wtiWorker(pWti); pthread_cleanup_pop(0); - pThis->iCurNumWrkThrd--; - wtpSignalWrkrTermination(pThis); - - dbgprintf("%s: Worker thread %lx, terminated, num workers now %d\n", - wtpGetDbgHdr(pThis), (unsigned long) pWti, pThis->iCurNumWrkThrd); - - END_MTX_PROTECTED_OPERATIONS(&pThis->mut); + wtpWrkrExecCleanup(pWti); ENDfunc + /* NOTE: we must call ENDfunc FIRST, because otherwise the schedule may activate the main + * thread after the broadcast, which could destroy the debug class, resulting in a potential + * segfault. So we need to do the broadcast as actually the last action in our processing + */ + pthread_cond_broadcast(&pThis->condThrdTrm); /* activate anyone waiting on thread shutdown */ pthread_exit(0); } #pragma GCC diagnostic warning "-Wempty-body" @@ -505,27 +395,20 @@ wtpWorker(void *arg) /* the arg is actually a wti object, even though we are in /* start a new worker */ static rsRetVal -wtpStartWrkr(wtp_t *pThis, int bLockMutex) +wtpStartWrkr(wtp_t *pThis) { - DEFiRet; - DEFVARS_mutexProtection; wti_t *pWti; int i; int iState; + DEFiRet; ISOBJ_TYPE_assert(pThis, wtp); - wtpProcessThrdChanges(pThis); // TODO: Performance: this causes a lot of FUTEX calls + d_pthread_mutex_lock(&pThis->mutWtp); - BEGIN_MTX_PROTECTED_OPERATIONS(&pThis->mut, bLockMutex); - - pThis->iCurNumWrkThrd++; - - /* find free spot in thread table. If we find at least one worker that is in initialization, - * we do NOT start a new one. Let's give the other one a chance, first. - */ + /* find free spot in thread table. */ for(i = 0 ; i < pThis->iNumWorkerThreads ; ++i) { - if(wtiGetState(pThis->pWrkr[i], LOCK_MUTEX) == eWRKTHRD_STOPPED) { + if(wtiGetState(pThis->pWrkr[i]) == WRKTHRD_STOPPED) { break; } } @@ -533,17 +416,21 @@ wtpStartWrkr(wtp_t *pThis, int bLockMutex) if(i == pThis->iNumWorkerThreads) ABORT_FINALIZE(RS_RET_NO_MORE_THREADS); + if(i == 0 || pThis->toWrkShutdown == -1) { + wtiSetAlwaysRunning(pThis->pWrkr[i]); + } + pWti = pThis->pWrkr[i]; - wtiSetState(pWti, eWRKTHRD_RUN_CREATED, 0, LOCK_MUTEX); - iState = pthread_create(&(pWti->thrdID), NULL, wtpWorker, (void*) pWti); - dbgprintf("%s: started with state %d, num workers now %d\n", - wtpGetDbgHdr(pThis), iState, pThis->iCurNumWrkThrd); + wtiSetState(pWti, WRKTHRD_RUNNING); + iState = pthread_create(&(pWti->thrdID), &pThis->attrThrd, wtpWorker, (void*) pWti); + ATOMIC_INC(&pThis->iCurNumWrkThrd, &pThis->mutCurNumWrkThrd); /* we got one more! */ - /* indicate we just started a worker and would like to see it running */ - wtpSetInactivityGuard(pThis, 1, MUTEX_ALREADY_LOCKED); + DBGPRINTF("%s: started with state %d, num workers now %d\n", + wtpGetDbgHdr(pThis), iState, + ATOMIC_FETCH_32BIT(&pThis->iCurNumWrkThrd, &pThis->mutCurNumWrkThrd)); finalize_it: - END_MTX_PROTECTED_OPERATIONS(&pThis->mut); + d_pthread_mutex_unlock(&pThis->mutWtp); RETiRet; } @@ -560,7 +447,6 @@ rsRetVal wtpAdviseMaxWorkers(wtp_t *pThis, int nMaxWrkr) { DEFiRet; - DEFVARS_mutexProtection; int nMissing; /* number workers missing to run */ int i; @@ -569,29 +455,24 @@ wtpAdviseMaxWorkers(wtp_t *pThis, int nMaxWrkr) if(nMaxWrkr == 0) FINALIZE; - BEGIN_MTX_PROTECTED_OPERATIONS(&pThis->mut, LOCK_MUTEX); - if(nMaxWrkr > pThis->iNumWorkerThreads) /* limit to configured maximum */ nMaxWrkr = pThis->iNumWorkerThreads; - nMissing = nMaxWrkr - pThis->iCurNumWrkThrd; + nMissing = nMaxWrkr - ATOMIC_FETCH_32BIT(&pThis->iCurNumWrkThrd, &pThis->mutCurNumWrkThrd); if(nMissing > 0) { - dbgprintf("%s: high activity - starting %d additional worker thread(s).\n", wtpGetDbgHdr(pThis), nMissing); + DBGPRINTF("%s: high activity - starting %d additional worker thread(s).\n", + wtpGetDbgHdr(pThis), nMissing); /* start the rqtd nbr of workers */ for(i = 0 ; i < nMissing ; ++i) { - CHKiRet(wtpStartWrkr(pThis, MUTEX_ALREADY_LOCKED)); - } - } else { - if(nMaxWrkr > 0) { - dbgprintf("wtpAdviseMaxWorkers signals busy\n"); - wtpWakeupWrkr(pThis); + CHKiRet(wtpStartWrkr(pThis)); } + } else { + pthread_cond_signal(pThis->pcondBusy); } finalize_it: - END_MTX_PROTECTED_OPERATIONS(&pThis->mut); RETiRet; } @@ -605,35 +486,9 @@ DEFpropSetMethPTR(wtp, pmutUsr, pthread_mutex_t) DEFpropSetMethPTR(wtp, pcondBusy, pthread_cond_t) DEFpropSetMethFP(wtp, pfChkStopWrkr, rsRetVal(*pVal)(void*, int)) DEFpropSetMethFP(wtp, pfRateLimiter, rsRetVal(*pVal)(void*)) -DEFpropSetMethFP(wtp, pfIsIdle, rsRetVal(*pVal)(void*, int)) -DEFpropSetMethFP(wtp, pfDoWork, rsRetVal(*pVal)(void*, void*, int)) -DEFpropSetMethFP(wtp, pfOnIdle, rsRetVal(*pVal)(void*, int)) -DEFpropSetMethFP(wtp, pfOnWorkerCancel, rsRetVal(*pVal)(void*, void*)) -DEFpropSetMethFP(wtp, pfOnWorkerStartup, rsRetVal(*pVal)(void*)) -DEFpropSetMethFP(wtp, pfOnWorkerShutdown, rsRetVal(*pVal)(void*)) - - -/* return the current number of worker threads. - * TODO: atomic operation would bring a nice performance - * enhancemcent - * rgerhards, 2008-01-27 - */ -int -wtpGetCurNumWrkr(wtp_t *pThis, int bLockMutex) -{ - DEFVARS_mutexProtection; - int iNumWrkr; - - BEGINfunc - ISOBJ_TYPE_assert(pThis, wtp); - - BEGIN_MTX_PROTECTED_OPERATIONS(&pThis->mut, bLockMutex); - iNumWrkr = pThis->iCurNumWrkThrd; - END_MTX_PROTECTED_OPERATIONS(&pThis->mut); - - ENDfunc - return iNumWrkr; -} +DEFpropSetMethFP(wtp, pfGetDeqBatchSize, rsRetVal(*pVal)(void*, int*)) +DEFpropSetMethFP(wtp, pfDoWork, rsRetVal(*pVal)(void*, void*)) +DEFpropSetMethFP(wtp, pfObjProcessed, rsRetVal(*pVal)(void*, wti_t*)) /* set the debug header message @@ -657,7 +512,7 @@ wtpSetDbgHdr(wtp_t *pThis, uchar *pszMsg, size_t lenMsg) pThis->pszDbgHdr = NULL; } - if((pThis->pszDbgHdr = malloc(sizeof(uchar) * lenMsg + 1)) == NULL) + if((pThis->pszDbgHdr = MALLOC(sizeof(uchar) * lenMsg + 1)) == NULL) ABORT_FINALIZE(RS_RET_OUT_OF_MEMORY); memcpy(pThis->pszDbgHdr, pszMsg, lenMsg + 1); /* always think about the \0! */ @@ -687,6 +542,5 @@ BEGINObjClassInit(wtp, 1, OBJ_IS_CORE_MODULE) CHKiRet(objUse(glbl, CORE_COMPONENT)); ENDObjClassInit(wtp) -/* - * vi:set ai: +/* vi:set ai: */ diff --git a/runtime/wtp.h b/runtime/wtp.h index 640c3320..7e6b4394 100644 --- a/runtime/wtp.h +++ b/runtime/wtp.h @@ -28,18 +28,9 @@ #include "obj.h" #include "atomic.h" -/* commands and states for worker threads. */ -typedef enum { - eWRKTHRD_STOPPED = 0, /* worker thread is not running (either actually never ran or was shut down) */ - eWRKTHRD_TERMINATING = 1,/* worker thread has shut down, but some finalzing is still needed */ - /* ALL active states MUST be numerically higher than eWRKTHRD_TERMINATED and NONE must be lower! */ - eWRKTHRD_RUN_CREATED = 2,/* worker thread has been created, but not yet begun initialization (prob. not yet scheduled) */ - eWRKTHRD_RUN_INIT = 3, /* worker thread is initializing, but not yet fully running */ - eWRKTHRD_RUNNING = 4, /* worker thread is up and running and shall continue to do so */ - eWRKTHRD_SHUTDOWN = 5, /* worker thread is running but shall terminate when wtp is empty */ - eWRKTHRD_SHUTDOWN_IMMEDIATE = 6/* worker thread is running but shall terminate even if wtp is full */ - /* SHUTDOWN_IMMEDIATE MUST alsways be the numerically highest state! */ -} qWrkCmd_t; +/* states for worker threads. */ +#define WRKTHRD_STOPPED FALSE +#define WRKTHRD_RUNNING TRUE /* possible states of a worker thread pool */ @@ -51,37 +42,33 @@ typedef enum { /* the worker thread pool (wtp) object */ -typedef struct wtp_s { +struct wtp_s { BEGINobjInstance; wtpState_t wtpState; int iNumWorkerThreads;/* number of worker threads to use */ int iCurNumWrkThrd;/* current number of active worker threads */ struct wti_s **pWrkr;/* array with control structure for the worker thread(s) associated with this wtp */ int toWrkShutdown; /* timeout for idle workers in ms, -1 means indefinite (0 is immediate) */ - bool bInactivityGuard;/* prevents inactivity due to race condition */ rsRetVal (*pConsumer)(void *); /* user-supplied consumer function for dewtpd messages */ /* synchronization variables */ - pthread_mutex_t mutThrdShutdwn; /* mutex to guard thread shutdown processing */ - pthread_mutex_t mut; /* mutex for the wtp's thread management */ + pthread_mutex_t mutWtp; /* mutex for the wtp's thread management */ pthread_cond_t condThrdTrm;/* signalled when threads terminate */ - int bThrdStateChanged; /* at least one thread state has changed if 1 */ /* end sync variables */ /* user objects */ - void *pUsr; /* pointer to user object */ + void *pUsr; /* pointer to user object (in this case, the queue the wtp belongs to) */ + pthread_attr_t attrThrd;/* attribute for new threads (created just once and cached here) */ pthread_mutex_t *pmutUsr; pthread_cond_t *pcondBusy; /* condition the user will signal "busy again, keep runing" on (awakes worker) */ rsRetVal (*pfChkStopWrkr)(void *pUsr, int); + rsRetVal (*pfGetDeqBatchSize)(void *pUsr, int*); /* obtains max dequeue count from queue config */ + rsRetVal (*pfObjProcessed)(void *pUsr, wti_t *pWti); /* indicate user object is processed */ rsRetVal (*pfRateLimiter)(void *pUsr); - rsRetVal (*pfIsIdle)(void *pUsr, int); - rsRetVal (*pfDoWork)(void *pUsr, void *pWti, int); - rsRetVal (*pfOnIdle)(void *pUsr, int); - rsRetVal (*pfOnWorkerCancel)(void *pUsr, void*pWti); - rsRetVal (*pfOnWorkerStartup)(void *pUsr); - rsRetVal (*pfOnWorkerShutdown)(void *pUsr); + rsRetVal (*pfDoWork)(void *pUsr, void *pWti); /* end user objects */ uchar *pszDbgHdr; /* header string for debug messages */ - DEF_ATOMIC_HELPER_MUT(mutThrdStateChanged); -} wtp_t; + DEF_ATOMIC_HELPER_MUT(mutCurNumWrkThrd); + DEF_ATOMIC_HELPER_MUT(mutWtpState); +}; /* some symbolic constants for easier reference */ @@ -92,26 +79,18 @@ rsRetVal wtpConstructFinalize(wtp_t *pThis); rsRetVal wtpDestruct(wtp_t **ppThis); rsRetVal wtpAdviseMaxWorkers(wtp_t *pThis, int nMaxWrkr); rsRetVal wtpProcessThrdChanges(wtp_t *pThis); -rsRetVal wtpSetInactivityGuard(wtp_t *pThis, int bNewState, int bLockMutex); -rsRetVal wtpChkStopWrkr(wtp_t *pThis, int bLockMutex, int bLockUsrMutex); +rsRetVal wtpChkStopWrkr(wtp_t *pThis, int bLockUsrMutex); rsRetVal wtpSetState(wtp_t *pThis, wtpState_t iNewState); -rsRetVal wtpWakeupWrkr(wtp_t *pThis); rsRetVal wtpWakeupAllWrkr(wtp_t *pThis); rsRetVal wtpCancelAll(wtp_t *pThis); rsRetVal wtpSetDbgHdr(wtp_t *pThis, uchar *pszMsg, size_t lenMsg); -rsRetVal wtpSignalWrkrTermination(wtp_t *pWtp); rsRetVal wtpShutdownAll(wtp_t *pThis, wtpState_t tShutdownCmd, struct timespec *ptTimeout); -int wtpGetCurNumWrkr(wtp_t *pThis, int bLockMutex); -void wtpSetThrdStateChanged(wtp_t *pThis, int val); PROTOTYPEObjClassInit(wtp); PROTOTYPEpropSetMethFP(wtp, pfChkStopWrkr, rsRetVal(*pVal)(void*, int)); PROTOTYPEpropSetMethFP(wtp, pfRateLimiter, rsRetVal(*pVal)(void*)); -PROTOTYPEpropSetMethFP(wtp, pfIsIdle, rsRetVal(*pVal)(void*, int)); -PROTOTYPEpropSetMethFP(wtp, pfDoWork, rsRetVal(*pVal)(void*, void*, int)); -PROTOTYPEpropSetMethFP(wtp, pfOnIdle, rsRetVal(*pVal)(void*, int)); -PROTOTYPEpropSetMethFP(wtp, pfOnWorkerCancel, rsRetVal(*pVal)(void*,void*)); -PROTOTYPEpropSetMethFP(wtp, pfOnWorkerStartup, rsRetVal(*pVal)(void*)); -PROTOTYPEpropSetMethFP(wtp, pfOnWorkerShutdown, rsRetVal(*pVal)(void*)); +PROTOTYPEpropSetMethFP(wtp, pfGetDeqBatchSize, rsRetVal(*pVal)(void*, int*)); +PROTOTYPEpropSetMethFP(wtp, pfDoWork, rsRetVal(*pVal)(void*, void*)); +PROTOTYPEpropSetMethFP(wtp, pfObjProcessed, rsRetVal(*pVal)(void*, wti_t*)); PROTOTYPEpropSetMeth(wtp, toWrkShutdown, long); PROTOTYPEpropSetMeth(wtp, wtpState, wtpState_t); PROTOTYPEpropSetMeth(wtp, iMaxWorkerThreads, int); |