/* cache.c - routines to maintain an in-core cache of entries */ /* $OpenLDAP$ */ /* * Copyright 1998-2003 The OpenLDAP Foundation, All Rights Reserved. * COPYING RESTRICTIONS APPLY, see COPYRIGHT file */ #include "portable.h" #include #include #include #include #include "slap.h" #include "back-bdb.h" static int bdb_cache_delete_entry_internal(Cache *cache, EntryInfo *e); #ifdef LDAP_DEBUG static void bdb_lru_print(Cache *cache); #endif static EntryInfo * bdb_cache_entryinfo_new( ) { EntryInfo *ei; ei = ch_calloc(1, sizeof(struct bdb_entry_info)); ldap_pvt_thread_mutex_init( &ei->bei_kids_mutex ); return ei; } /* Atomically release and reacquire a lock */ int bdb_cache_entry_db_relock( DB_ENV *env, u_int32_t locker, EntryInfo *ei, int rw, int tryOnly, DB_LOCK *lock ) { #ifdef NO_THREADS return 0; #else int rc; DBT lockobj; DB_LOCKREQ list[2]; lockobj.data = ei; lockobj.size = sizeof(ei->bei_parent) + sizeof(ei->bei_id); list[0].op = DB_LOCK_PUT; list[0].lock = *lock; list[1].op = DB_LOCK_GET; list[1].lock = *lock; list[1].mode = rw ? DB_LOCK_WRITE : DB_LOCK_READ; list[1].obj = &lockobj; rc = env->lock_vec(env, locker, tryOnly ? DB_LOCK_NOWAIT : 0, list, 2, NULL ); if (rc) { #ifdef NEW_LOGGING LDAP_LOG( CACHE, DETAIL1, "bdb_cache_entry_db_relock: entry %d, rw %d, rc %d\n", ei->bei_id, rw, rc ); #else Debug( LDAP_DEBUG_TRACE, "bdb_cache_entry_db_relock: entry %d, rw %d, rc %d\n", ei->bei_id, rw, rc ); #endif } else { *lock = list[1].lock; } return rc; #endif } int bdb_cache_entry_db_lock ( DB_ENV *env, u_int32_t locker, EntryInfo *ei, int rw, int tryOnly, DB_LOCK *lock ) { #ifdef NO_THREADS return 0; #else int rc; DBT lockobj; int db_rw; if (rw) db_rw = DB_LOCK_WRITE; else db_rw = DB_LOCK_READ; lockobj.data = ei; lockobj.size = sizeof(ei->bei_parent) + sizeof(ei->bei_id); rc = LOCK_GET(env, locker, tryOnly ? DB_LOCK_NOWAIT : 0, &lockobj, db_rw, lock); if (rc) { #ifdef NEW_LOGGING LDAP_LOG( CACHE, DETAIL1, "bdb_cache_entry_db_lock: entry %d, rw %d, rc %d\n", ei->bei_id, rw, rc ); #else Debug( LDAP_DEBUG_TRACE, "bdb_cache_entry_db_lock: entry %d, rw %d, rc %d\n", ei->bei_id, rw, rc ); #endif } return rc; #endif /* NO_THREADS */ } int bdb_cache_entry_db_unlock ( DB_ENV *env, DB_LOCK *lock ) { #ifdef NO_THREADS return 0; #else int rc; rc = LOCK_PUT ( env, lock ); return rc; #endif } static int bdb_cache_entryinfo_destroy( EntryInfo *e ) { ldap_pvt_thread_mutex_destroy( &e->bei_kids_mutex ); free( e->bei_nrdn.bv_val ); free( e ); return 0; } #define LRU_DELETE( cache, ei ) do { \ if ( (ei)->bei_lruprev != NULL ) { \ (ei)->bei_lruprev->bei_lrunext = (ei)->bei_lrunext; \ } else { \ (cache)->c_lruhead = (ei)->bei_lrunext; \ } \ if ( (ei)->bei_lrunext != NULL ) { \ (ei)->bei_lrunext->bei_lruprev = (ei)->bei_lruprev; \ } else { \ (cache)->c_lrutail = (ei)->bei_lruprev; \ } \ } while(0) #define LRU_ADD( cache, ei ) do { \ (ei)->bei_lrunext = (cache)->c_lruhead; \ if ( (ei)->bei_lrunext != NULL ) { \ (ei)->bei_lrunext->bei_lruprev = (ei); \ } \ (cache)->c_lruhead = (ei); \ (ei)->bei_lruprev = NULL; \ if ( (cache)->c_lrutail == NULL ) { \ (cache)->c_lrutail = (ei); \ } \ } while(0) /* Do a lexical sort on normalized RDNs */ static int bdb_rdn_cmp( const void *v_e1, const void *v_e2 ) { const EntryInfo *e1 = v_e1, *e2 = v_e2; int rc = strncmp( e1->bei_nrdn.bv_val, e2->bei_nrdn.bv_val, e1->bei_nrdn.bv_len ); if (rc == 0) rc = e1->bei_nrdn.bv_len - e2->bei_nrdn.bv_len; return rc; } static int bdb_id_cmp( const void *v_e1, const void *v_e2 ) { const EntryInfo *e1 = v_e1, *e2 = v_e2; return e1->bei_id - e2->bei_id; } /* Create an entryinfo in the cache. Caller must release the locks later. */ int bdb_entryinfo_add_internal( struct bdb_info *bdb, EntryInfo *eip, ID id, struct berval *nrdn, EntryInfo **res, u_int32_t locker ) { Cache *cache = &bdb->bi_cache; DB_ENV *env = bdb->bi_dbenv; EntryInfo *ei2 = NULL; int incr = 1; int addkid = 1; int rc; DB_LOCK lock; *res = NULL; ldap_pvt_thread_rdwr_wlock( &bdb->bi_cache.c_rwlock ); bdb_cache_entryinfo_lock( eip ); /* if parent was previously considered a leaf node, * it was on the LRU list. Now it's going to have * kids, take it off the LRU list. */ ldap_pvt_thread_mutex_lock( &cache->lru_mutex ); if ( eip->bei_id && !eip->bei_kids ) { LRU_DELETE( cache, eip ); incr = 0; } cache->c_cursize += incr; /* See if we're above the cache size limit */ if ( cache->c_cursize > cache->c_maxsize ) { EntryInfo *elru, *elprev; int i = 0; /* Look for an unused entry to remove */ for (elru = cache->c_lrutail; elru; elru = elprev, i++ ) { elprev = elru->bei_lruprev; /* Too many probes, not enough idle, give up */ if (i > 10) break; /* If we can successfully writelock it, then * the object is idle. */ if ( bdb_cache_entry_db_lock( env, locker, elru, 1, 1, &lock ) == 0 ) { /* Need to lock parent to delete child */ if ( ldap_pvt_thread_mutex_trylock( &elru->bei_parent->bei_kids_mutex )) { bdb_cache_entry_db_unlock( env, &lock ); continue; } bdb_cache_delete_entry_internal( cache, elru ); bdb_cache_entryinfo_unlock( elru->bei_parent ); elru->bei_e->e_private = NULL; bdb_entry_return( elru->bei_e ); bdb_cache_entry_db_unlock( env, &lock ); if (ei2) { bdb_cache_entryinfo_destroy( elru ); } else { /* re-use this one */ ch_free(elru->bei_nrdn.bv_val); elru->bei_nrdn.bv_val = NULL; elru->bei_e = NULL; elru->bei_kids = NULL; elru->bei_lrunext = NULL; elru->bei_lruprev = NULL; elru->bei_state = 0; ei2 = elru; } if (cache->c_cursize < cache->c_maxsize) break; } } } if (!ei2) { ei2 = bdb_cache_entryinfo_new(); } ei2->bei_id = id; ei2->bei_parent = eip; /* Add to cache ID tree */ if (avl_insert( &cache->c_idtree, ei2, bdb_id_cmp, avl_dup_error )) { EntryInfo *ei; ei = avl_find( cache->c_idtree, ei2, bdb_id_cmp ); bdb_cache_entryinfo_destroy( ei2 ); ei2 = ei; addkid = 0; cache->c_cursize -= incr; } else { LRU_ADD( cache, ei2 ); ber_dupbv( &ei2->bei_nrdn, nrdn ); } if ( addkid ) { avl_insert( &eip->bei_kids, ei2, bdb_rdn_cmp, avl_dup_error ); } ldap_pvt_thread_mutex_unlock( &cache->lru_mutex ); #if 0 /* caller must do these frees */ ldap_pvt_thread_rdwr_wunlock( &cache->c_rwlock ); bdb_cache_entryinfo_unlock( eip ); #endif *res = ei2; return 0; } /* Find the EntryInfo for the requested DN. If the DN cannot be found, return * the info for its closest ancestor. *res should be NULL to process a * complete DN starting from the tree root. Otherwise *res must be the * immediate parent of the requested DN, and only the RDN will be searched. * The EntryInfo is locked upon return and must be unlocked by the caller. */ int bdb_cache_find_entry_ndn2id( Backend *be, DB_TXN *txn, struct berval *ndn, EntryInfo **res, u_int32_t locker, void *ctx ) { struct bdb_info *bdb = (struct bdb_info *) be->be_private; EntryInfo ei, *eip, *ei2; ID id; int rc = 0; char *ptr; /* this function is always called with normalized DN */ if ( *res ) { /* we're doing a onelevel search for an RDN */ ei.bei_nrdn.bv_val = ndn->bv_val; ei.bei_nrdn.bv_len = dn_rdnlen( be, ndn ); eip = *res; } else { /* we're searching a full DN from the root */ ptr = ndn->bv_val + ndn->bv_len - be->be_nsuffix[0].bv_len; ei.bei_nrdn.bv_val = ptr; ei.bei_nrdn.bv_len = be->be_nsuffix[0].bv_len; eip = &bdb->bi_cache.c_dntree; } for ( bdb_cache_entryinfo_lock( eip ); eip; ) { ei2 = (EntryInfo *)avl_find( eip->bei_kids, &ei, bdb_rdn_cmp ); if ( !ei2 ) { int len = ei.bei_nrdn.bv_len; ei.bei_nrdn.bv_len = ndn->bv_len - (ei.bei_nrdn.bv_val - ndn->bv_val); bdb_cache_entryinfo_unlock( eip ); rc = bdb_dn2id( be, txn, &ei.bei_nrdn, &id, ctx ); if (rc) { bdb_cache_entryinfo_lock( eip ); *res = eip; return rc; } /* DN exists but needs to be added to cache */ ei.bei_nrdn.bv_len = len; rc = bdb_entryinfo_add_internal( bdb, eip, id, &ei.bei_nrdn, &ei2, locker ); /* add_internal left eip and c_rwlock locked */ ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock ); if ( rc ) { *res = eip; return rc; } } else if ( ei2->bei_state & CACHE_ENTRY_DELETED ) { /* In the midst of deleting? Give it a chance to * complete. */ bdb_cache_entryinfo_unlock( eip ); ldap_pvt_thread_yield(); bdb_cache_entryinfo_lock( eip ); *res = eip; return DB_NOTFOUND; } bdb_cache_entryinfo_unlock( eip ); bdb_cache_entryinfo_lock( ei2 ); eip = ei2; /* Advance to next lower RDN */ for (ptr = ei.bei_nrdn.bv_val - 2; ptr > ndn->bv_val && !DN_SEPARATOR(*ptr); ptr--); if ( ptr >= ndn->bv_val ) { if (DN_SEPARATOR(*ptr)) ptr++; ei.bei_nrdn.bv_len = ei.bei_nrdn.bv_val - ptr - 1; ei.bei_nrdn.bv_val = ptr; } if ( ptr < ndn->bv_val ) { *res = eip; break; } } return rc; } /* * cache_find_entry_id - find an entry in the cache, given id. * The entry is locked for Read upon return. Call with islocked TRUE if * the supplied *eip was already locked. */ int bdb_cache_find_entry_id( Backend *be, DB_TXN *tid, ID id, EntryInfo **eip, int islocked, u_int32_t locker, DB_LOCK *lock, void *ctx ) { struct bdb_info *bdb = (struct bdb_info *) be->be_private; Entry *ep = NULL; int rc = 0; EntryInfo ei; ei.bei_id = id; /* If we weren't given any info, see if we have it already cached */ if ( !*eip ) { ldap_pvt_thread_rdwr_rlock( &bdb->bi_cache.c_rwlock ); *eip = (EntryInfo *) avl_find( bdb->bi_cache.c_idtree, (caddr_t) &ei, bdb_id_cmp ); if ( *eip ) { bdb_cache_entryinfo_lock( *eip ); islocked = 1; } ldap_pvt_thread_rdwr_runlock( &bdb->bi_cache.c_rwlock ); } /* See if the ID exists in the database; add it to the cache if so */ if ( !*eip ) { rc = bdb_id2entry( be, tid, id, &ep ); if ( rc == 0 ) { rc = bdb_cache_find_entry_ndn2id( be, tid, &ep->e_nname, eip, locker, ctx ); if ( *eip ) islocked = 1; if ( rc ) { bdb_entry_return( ep ); ep = NULL; } } } /* Ok, we found the info, do we have the entry? */ if ( *eip && rc == 0 ) { if ( (*eip)->bei_state & CACHE_ENTRY_DELETED ) { rc = DB_NOTFOUND; } else if (!(*eip)->bei_e ) { if (!ep) { rc = bdb_id2entry( be, tid, id, &ep ); } if ( rc == 0 ) { bdb_cache_entry_db_lock( bdb->bi_dbenv, locker, *eip, 1, 0, lock ); (*eip)->bei_e = ep; ep->e_private = *eip; bdb_cache_entry_db_relock( bdb->bi_dbenv, locker, *eip, 0, 0, lock ); } } else { bdb_cache_entry_db_lock( bdb->bi_dbenv, locker, *eip, 0, 0, lock ); } } if ( rc == 0 && (*eip)->bei_kids == NULL ) { /* set lru mutex */ ldap_pvt_thread_mutex_lock( &bdb->bi_cache.lru_mutex ); LRU_DELETE( &bdb->bi_cache, *eip ); LRU_ADD( &bdb->bi_cache, *eip ); ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.lru_mutex ); } if ( islocked ) { bdb_cache_entryinfo_unlock( *eip ); } return rc; } /* Update the cache after a successful database Add. */ int bdb_cache_add( struct bdb_info *bdb, EntryInfo *ei, Entry *e, struct berval *nrdn, u_int32_t locker ) { EntryInfo *new; int rc; rc = bdb_entryinfo_add_internal( bdb, ei, e->e_id, nrdn, &new, locker ); new->bei_e = e; e->e_private = new; new->bei_state = CACHE_ENTRY_NO_KIDS; ei->bei_state &= ~CACHE_ENTRY_NO_KIDS; bdb_cache_entryinfo_unlock( ei ); ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock ); return rc; } int bdb_cache_modify( Entry *e, Attribute *newAttrs, DB_ENV *env, u_int32_t locker, DB_LOCK *lock ) { EntryInfo *ei = BEI(e); /* Get write lock on data */ bdb_cache_entry_db_relock( env, locker, ei, 1, 0, lock ); /* If we've done repeated mods on a cached entry, then e_attrs * is no longer contiguous with the entry, and must be freed. */ if ( (void *)e->e_attrs != (void *)(e+1) ) { attrs_free( e->e_attrs ); } e->e_attrs = newAttrs; return 0; } /* * Change the rdn in the entryinfo. Also move to a new parent if needed. */ int bdb_cache_modrdn( Entry *e, struct berval *nrdn, Entry *new, EntryInfo *ein, DB_ENV *env, u_int32_t locker, DB_LOCK *lock ) { EntryInfo *ei = BEI(e), *pei; int rc = 0; /* Get write lock on data */ bdb_cache_entry_db_relock( env, locker, ei, 1, 0, lock ); /* If we've done repeated mods on a cached entry, then e_attrs * is no longer contiguous with the entry, and must be freed. */ if ( (void *)e->e_attrs != (void *)(e+1) ) { attrs_free( e->e_attrs ); } e->e_attrs = new->e_attrs; #ifdef BDB_HIER ch_free(e->e_name.bv_val); #else if( e->e_nname.bv_val < e->e_bv.bv_val || e->e_nname.bv_val > e->e_bv.bv_val + e->e_bv.bv_len ) { ch_free(e->e_name.bv_val); ch_free(e->e_nname.bv_val); } #endif e->e_name = new->e_name; e->e_nname = new->e_nname; /* Lock the parent's kids AVL tree */ pei = ei->bei_parent; bdb_cache_entryinfo_lock( pei ); avl_delete( &pei->bei_kids, (caddr_t) ei, bdb_rdn_cmp ); free( ei->bei_nrdn.bv_val ); ber_dupbv( &ei->bei_nrdn, nrdn ); if (!ein) { ein = ei->bei_parent; } else { ei->bei_parent = ein; bdb_cache_entryinfo_unlock( pei ); bdb_cache_entryinfo_lock( ein ); } avl_insert( &ein->bei_kids, ei, bdb_rdn_cmp, avl_dup_error ); bdb_cache_entryinfo_unlock( ein ); return rc; } /* * cache_delete_entry - delete the entry e from the cache. * * returns: 0 e was deleted ok * 1 e was not in the cache * -1 something bad happened */ int bdb_cache_delete_entry( Cache *cache, Entry *e, DB_ENV *env, u_int32_t locker, DB_LOCK *lock ) { EntryInfo *ei = BEI(e); int rc; assert( e->e_private ); /* Set this early, warn off any queriers */ ei->bei_state |= CACHE_ENTRY_DELETED; /* Get write lock on the data */ bdb_cache_entry_db_relock( env, locker, ei, 1, 0, lock ); /* set cache write lock */ ldap_pvt_thread_rdwr_wlock( &cache->c_rwlock ); /* Lock the parent's kids tree */ bdb_cache_entryinfo_lock( ei->bei_parent ); #ifdef NEW_LOGGING LDAP_LOG( CACHE, ENTRY, "bdb_cache_delete_entry: delete %ld.\n", e->e_id, 0, 0 ); #else Debug( LDAP_DEBUG_TRACE, "====> bdb_cache_delete_entry( %ld )\n", e->e_id, 0, 0 ); #endif /* set lru mutex */ ldap_pvt_thread_mutex_lock( &cache->lru_mutex ); rc = bdb_cache_delete_entry_internal( cache, e->e_private ); /* free lru mutex */ ldap_pvt_thread_mutex_unlock( &cache->lru_mutex ); /* free cache write lock */ ldap_pvt_thread_rdwr_wunlock( &cache->c_rwlock ); bdb_cache_entryinfo_unlock( ei->bei_parent ); bdb_cache_entryinfo_destroy( ei ); e->e_private = NULL; return( rc ); } static int bdb_cache_delete_entry_internal( Cache *cache, EntryInfo *e ) { int rc = 0; /* return code */ /* dn tree */ if ( avl_delete( &e->bei_parent->bei_kids, (caddr_t) e, bdb_rdn_cmp ) == NULL ) { rc = -1; } /* If parent has no more kids, put in on LRU list */ if ( e->bei_parent->bei_kids == NULL ) { LRU_ADD( cache, e->bei_parent ); cache->c_cursize++; } /* id tree */ if ( avl_delete( &cache->c_idtree, (caddr_t) e, bdb_id_cmp ) == NULL ) { rc = -1; } if (rc != 0) { return rc; } /* lru */ LRU_DELETE( cache, e ); cache->c_cursize--; /* * flag entry to be freed later by a call to cache_return_entry() */ e->bei_state |= CACHE_ENTRY_DELETED; return( 0 ); } static void bdb_entryinfo_release( void *data ) { EntryInfo *ei = (EntryInfo *)data; avl_free( ei->bei_kids, NULL ); if ( ei->bei_e ) { ei->bei_e->e_private = NULL; bdb_entry_return( ei->bei_e ); } bdb_cache_entryinfo_destroy( ei ); } void bdb_cache_release_all( Cache *cache ) { /* set cache write lock */ ldap_pvt_thread_rdwr_wlock( &cache->c_rwlock ); /* set lru mutex */ ldap_pvt_thread_mutex_lock( &cache->lru_mutex ); #ifdef NEW_LOGGING LDAP_LOG( CACHE, ENTRY, "bdb_cache_release_all: enter\n", 0, 0, 0 ); #else Debug( LDAP_DEBUG_TRACE, "====> bdb_cache_release_all\n", 0, 0, 0 ); #endif avl_free( cache->c_dntree.bei_kids, NULL ); avl_free( cache->c_idtree, bdb_entryinfo_release ); cache->c_lruhead = NULL; cache->c_lrutail = NULL; /* free lru mutex */ ldap_pvt_thread_mutex_unlock( &cache->lru_mutex ); /* free cache write lock */ ldap_pvt_thread_rdwr_wunlock( &cache->c_rwlock ); } #ifdef LDAP_DEBUG static void bdb_lru_print( Cache *cache ) { EntryInfo *e; fprintf( stderr, "LRU queue (head to tail):\n" ); for ( e = cache->c_lruhead; e != NULL; e = e->bei_lrunext ) { fprintf( stderr, "\trdn \"%20s\" id %ld\n", e->bei_nrdn.bv_val, e->bei_id ); } fprintf( stderr, "LRU queue (tail to head):\n" ); for ( e = cache->c_lrutail; e != NULL; e = e->bei_lruprev ) { fprintf( stderr, "\trdn \"%20s\" id %ld\n", e->bei_nrdn.bv_val, e->bei_id ); } } #endif #ifdef BDB_REUSE_LOCKERS void bdb_locker_id_free( void *key, void *data ) { DB_ENV *env = key; int lockid = (int) data; XLOCK_ID_FREE( env, lockid ); } int bdb_locker_id( Operation *op, DB_ENV *env, int *locker ) { int i, rc, lockid; void *data; void *ctx; if ( !env || !locker ) return -1; /* If no op was provided, try to find the ctx anyway... */ if ( op ) { ctx = op->o_threadctx; } else { ctx = ldap_pvt_thread_pool_context(); } /* Shouldn't happen unless we're single-threaded */ if ( !ctx ) { *locker = 0; return 0; } if ( ldap_pvt_thread_pool_getkey( ctx, env, &data, NULL ) ) { for ( i=0, rc=1; rc != 0 && i<4; i++ ) { rc = XLOCK_ID( env, &lockid ); if (rc) ldap_pvt_thread_yield(); } if ( rc != 0) { return rc; } data = (void *)lockid; if ( ( rc = ldap_pvt_thread_pool_setkey( ctx, env, data, bdb_locker_id_free ) ) ) { XLOCK_ID_FREE( env, lockid ); #ifdef NEW_LOGGING LDAP_LOG( BACK_BDB, ERR, "bdb_locker_id: err %s(%d)\n", db_strerror(rc), rc, 0 ); #else Debug( LDAP_DEBUG_ANY, "bdb_locker_id: err %s(%d)\n", db_strerror(rc), rc, 0 ); #endif return rc; } } else { lockid = (int)data; } *locker = lockid; return 0; } #endif